From f3b7777c448d223a6ce571ad241b05c2630ad144 Mon Sep 17 00:00:00 2001 From: Nick Alcock Date: Fri, 30 Sep 2011 18:55:09 +0100 Subject: [PATCH 1/1] Initial commit of DTrace. Build via a simple 'make' if you're already running this kernel, or via 'make KERNELDIR=/path/to/kernel/top/level' otherwise. Installed via 'make install'. Signed-off-by: Nick Alcock --- dtrace/Kbuild | 34 + dtrace/Makefile | 12 + dtrace/dt_test.h | 14 + dtrace/dt_test_dev.c | 96 + dtrace/dt_test_mod.c | 40 + dtrace/dtrace_actdesc.c | 73 + dtrace/dtrace_anon.c | 130 ++ dtrace/dtrace_asm.S | 328 +++ dtrace/dtrace_buffer.c | 454 ++++ dtrace/dtrace_dev.c | 1317 +++++++++++ dtrace/dtrace_dev.h | 17 + dtrace/dtrace_dif.c | 4473 +++++++++++++++++++++++++++++++++++++ dtrace/dtrace_dof.c | 1053 +++++++++ dtrace/dtrace_ecb.c | 845 +++++++ dtrace/dtrace_enable.c | 414 ++++ dtrace/dtrace_fmt.c | 84 + dtrace/dtrace_hash.c | 203 ++ dtrace/dtrace_helper.c | 175 ++ dtrace/dtrace_isa.c | 547 +++++ dtrace/dtrace_match.c | 325 +++ dtrace/dtrace_mod.c | 34 + dtrace/dtrace_predicate.c | 66 + dtrace/dtrace_priv.c | 105 + dtrace/dtrace_probe.c | 1228 ++++++++++ dtrace/dtrace_probe_ctx.c | 538 +++++ dtrace/dtrace_ptofapi.c | 417 ++++ dtrace/dtrace_spec.c | 426 ++++ dtrace/dtrace_state.c | 1057 +++++++++ dtrace/dtrace_util.c | 215 ++ dtrace/fasttrap.h | 7 + dtrace/fasttrap_dev.c | 58 + dtrace/fasttrap_mod.c | 25 + dtrace/fbt.h | 7 + dtrace/fbt_dev.c | 58 + dtrace/fbt_mod.c | 25 + dtrace/lockstat.h | 7 + dtrace/lockstat_dev.c | 58 + dtrace/lockstat_mod.c | 25 + dtrace/profile.h | 15 + dtrace/profile_dev.c | 348 +++ dtrace/profile_mod.c | 40 + dtrace/sdt.h | 7 + dtrace/sdt_dev.c | 58 + dtrace/sdt_impl.h | 60 + dtrace/sdt_mod.c | 618 +++++ dtrace/sdt_subr.c | 1171 ++++++++++ dtrace/systrace_dev.c | 211 ++ dtrace/systrace_mod.c | 40 + 48 files changed, 17558 insertions(+) create mode 100644 dtrace/Kbuild create mode 100644 dtrace/Makefile create mode 100644 dtrace/dt_test.h create mode 100644 dtrace/dt_test_dev.c create mode 100644 dtrace/dt_test_mod.c create mode 100644 dtrace/dtrace_actdesc.c create mode 100644 dtrace/dtrace_anon.c create mode 100644 dtrace/dtrace_asm.S create mode 100644 dtrace/dtrace_buffer.c create mode 100644 dtrace/dtrace_dev.c create mode 100644 dtrace/dtrace_dev.h create mode 100644 dtrace/dtrace_dif.c create mode 100644 dtrace/dtrace_dof.c create mode 100644 dtrace/dtrace_ecb.c create mode 100644 dtrace/dtrace_enable.c create mode 100644 dtrace/dtrace_fmt.c create mode 100644 dtrace/dtrace_hash.c create mode 100644 dtrace/dtrace_helper.c create mode 100644 dtrace/dtrace_isa.c create mode 100644 dtrace/dtrace_match.c create mode 100644 dtrace/dtrace_mod.c create mode 100644 dtrace/dtrace_predicate.c create mode 100644 dtrace/dtrace_priv.c create mode 100644 dtrace/dtrace_probe.c create mode 100644 dtrace/dtrace_probe_ctx.c create mode 100644 dtrace/dtrace_ptofapi.c create mode 100644 dtrace/dtrace_spec.c create mode 100644 dtrace/dtrace_state.c create mode 100644 dtrace/dtrace_util.c create mode 100644 dtrace/fasttrap.h create mode 100644 dtrace/fasttrap_dev.c create mode 100644 dtrace/fasttrap_mod.c create mode 100644 dtrace/fbt.h create mode 100644 dtrace/fbt_dev.c create mode 100644 dtrace/fbt_mod.c create mode 100644 dtrace/lockstat.h create mode 100644 dtrace/lockstat_dev.c create mode 100644 dtrace/lockstat_mod.c create mode 100644 dtrace/profile.h create mode 100644 dtrace/profile_dev.c create mode 100644 dtrace/profile_mod.c create mode 100644 dtrace/sdt.h create mode 100644 dtrace/sdt_dev.c create mode 100644 dtrace/sdt_impl.h create mode 100644 dtrace/sdt_mod.c create mode 100644 dtrace/sdt_subr.c create mode 100644 dtrace/systrace_dev.c create mode 100644 dtrace/systrace_mod.c diff --git a/dtrace/Kbuild b/dtrace/Kbuild new file mode 100644 index 000000000000..93aeb0dd2391 --- /dev/null +++ b/dtrace/Kbuild @@ -0,0 +1,34 @@ +# +# Makefile for DTrace +# + +GCOV_PROFILE := y + +EXTRA_CFLAGS := -Ikernel/dtrace + +obj-$(CONFIG_DT_CORE) += dtrace.o +obj-$(CONFIG_DT_FASTTRAP) += fasttrap.o +obj-$(CONFIG_DT_FBT) += fbt.o +obj-$(CONFIG_DT_LOCKSTAT) += lockstat.o +obj-$(CONFIG_DT_PROFILE) += profile.o +obj-$(CONFIG_DT_SDT) += sdt.o +obj-$(CONFIG_DT_SYSTRACE) += systrace.o +obj-$(CONFIG_DT_DT_TEST) += dt_test.o + +dtrace-y := dtrace_mod.o dtrace_dev.o \ + dtrace_asm.o dtrace_isa.o \ + dtrace_actdesc.o dtrace_anon.o \ + dtrace_buffer.o dtrace_dif.o dtrace_dof.o \ + dtrace_ecb.o dtrace_enable.o \ + dtrace_fmt.o dtrace_hash.o dtrace_helper.o \ + dtrace_match.o dtrace_priv.o \ + dtrace_probe.o dtrace_probe_ctx.o \ + dtrace_ptofapi.o dtrace_predicate.o \ + dtrace_spec.o dtrace_state.o dtrace_util.o +fasttrap-y := fasttrap_mod.o fasttrap_dev.o +fbt-y := fbt_mod.o fbt_dev.o +lockstat-y := lockstat_mod.o lockstat_dev.o +profile-y := profile_mod.o profile_dev.o +sdt-y := sdt_mod.o sdt_dev.o sdt_subr.o +systrace-y := systrace_mod.o systrace_dev.o +dt_test-y := dt_test_mod.o dt_test_dev.o diff --git a/dtrace/Makefile b/dtrace/Makefile new file mode 100644 index 000000000000..0adc58f76630 --- /dev/null +++ b/dtrace/Makefile @@ -0,0 +1,12 @@ +# +# Makefile for DTrace +# + +KERNELDIR := /lib/modules/`uname -r`/build + +modules:: + +install:: modules_install + +%:: + $(MAKE) -C $(KERNELDIR) M=`pwd` $@ diff --git a/dtrace/dt_test.h b/dtrace/dt_test.h new file mode 100644 index 000000000000..86bd55abacfc --- /dev/null +++ b/dtrace/dt_test.h @@ -0,0 +1,14 @@ +#ifndef _DT_TEST_H_ +#define _DT_TEST_H_ + +extern void dt_test_provide(void *, const dtrace_probedesc_t *); +extern int dt_test_enable(void *arg, dtrace_id_t, void *); +extern void dt_test_disable(void *arg, dtrace_id_t, void *); +extern void dt_test_destroy(void *, dtrace_id_t, void *); + +extern dtrace_provider_id_t dt_test_id; + +extern int dt_test_dev_init(void); +extern void dt_test_dev_exit(void); + +#endif /* _DT_TEST_H_ */ diff --git a/dtrace/dt_test_dev.c b/dtrace/dt_test_dev.c new file mode 100644 index 000000000000..55998ec3d00d --- /dev/null +++ b/dtrace/dt_test_dev.c @@ -0,0 +1,96 @@ +/* + * FILE: dt_test_dev.c + * DESCRIPTION: DTrace Test Probe: device file handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include +#include +#include + +#include "dtrace.h" +#include "dtrace_dev.h" +#include "dt_test.h" + +static dtrace_id_t pid = 0; +static int enabled = 0; + +void dt_test_provide(void *arg, const dtrace_probedesc_t *desc) +{ + if (dtrace_probe_lookup(dt_test_id, "dt_test", NULL, "test") != 0) + return; + + pid = dtrace_probe_create(dt_test_id, + "dt_test", NULL, "test", 0, NULL); +} + +int dt_test_enable(void *arg, dtrace_id_t id, void *parg) +{ + enabled = 1; + + return 0; +} + +void dt_test_disable(void *arg, dtrace_id_t id, void *parg) +{ + enabled = 0; +} + +void dt_test_destroy(void *arg, dtrace_id_t id, void *parg) +{ +} + +static long dt_test_ioctl(struct file *file, + unsigned int cmd, unsigned long arg) +{ + if (enabled) { + dtrace_probe(pid, cmd, arg, 2, 3, 4); + + return 0; + } + + return -EAGAIN; +} + +static int dt_test_open(struct inode *inode, struct file *file) +{ + return 0; +} + +static int dt_test_close(struct inode *inode, struct file *file) +{ + return 0; +} + +static const struct file_operations dt_test_fops = { + .owner = THIS_MODULE, + .unlocked_ioctl = dt_test_ioctl, + .open = dt_test_open, + .release = dt_test_close, +}; + +static struct miscdevice dt_test_dev = { + .minor = DT_DEV_DT_TEST_MINOR, + .name = "dt_test", + .nodename = "dtrace/provider/dt_test", + .fops = &dt_test_fops, +}; + +int dt_test_dev_init(void) +{ + int ret = 0; + + ret = misc_register(&dt_test_dev); + if (ret) + pr_err("%s: Can't register misc device %d\n", + dt_test_dev.name, dt_test_dev.minor); + + return ret; +} + +void dt_test_dev_exit(void) +{ + misc_deregister(&dt_test_dev); +} diff --git a/dtrace/dt_test_mod.c b/dtrace/dt_test_mod.c new file mode 100644 index 000000000000..822df2be71f3 --- /dev/null +++ b/dtrace/dt_test_mod.c @@ -0,0 +1,40 @@ +/* + * FILE: dt_test_mod.c + * DESCRIPTION: DTrace Test Probe: module handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace.h" +#include "dtrace_dev.h" +#include "dt_test.h" + +MODULE_AUTHOR("Kris Van Hees (kris.van.hees@oracle.com)"); +MODULE_DESCRIPTION("DTrace Test Probe"); +MODULE_VERSION("v0.1"); +MODULE_LICENSE("Proprietary"); + +static const dtrace_pattr_t dt_test_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pops_t dt_test_pops = { + dt_test_provide, + NULL, + dt_test_enable, + dt_test_disable, + NULL, + NULL, + NULL, + NULL, + NULL, + dt_test_destroy +}; + +DT_PROVIDER_MODULE(dt_test, DTRACE_PRIV_USER); diff --git a/dtrace/dtrace_actdesc.c b/dtrace/dtrace_actdesc.c new file mode 100644 index 000000000000..5b84cf86676d --- /dev/null +++ b/dtrace/dtrace_actdesc.c @@ -0,0 +1,73 @@ +/* + * FILE: dtrace_actdesc.c + * DESCRIPTION: Dynamic Tracing: action description functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace.h" + +dtrace_actdesc_t *dtrace_actdesc_create(dtrace_actkind_t kind, uint32_t ntuple, + uint64_t uarg, uint64_t arg) +{ + dtrace_actdesc_t *act; + +#ifdef FIXME + ASSERT(!DTRACEACT_ISPRINTFLIKE(kind) || + (arg != 0 && (uintptr_t)arg >= KERNELBASE) || + (arg == 0 && kind == DTRACEACT_PRINTA)); +#else + ASSERT(!DTRACEACT_ISPRINTFLIKE(kind) || + (arg != 0) || + (arg == 0 && kind == DTRACEACT_PRINTA)); +#endif + + act = kzalloc(sizeof (dtrace_actdesc_t), GFP_KERNEL); + act->dtad_kind = kind; + act->dtad_ntuple = ntuple; + act->dtad_uarg = uarg; + act->dtad_arg = arg; + act->dtad_refcnt = 1; + + return act; +} + +void dtrace_actdesc_hold(dtrace_actdesc_t *act) +{ + ASSERT(act->dtad_refcnt >= 1); + + act->dtad_refcnt++; +} + +void dtrace_actdesc_release(dtrace_actdesc_t *act, dtrace_vstate_t *vstate) +{ + dtrace_actkind_t kind = act->dtad_kind; + dtrace_difo_t *dp; + + ASSERT(act->dtad_refcnt >= 1); + + if (--act->dtad_refcnt != 0) + return; + + if ((dp = act->dtad_difo) != NULL) + dtrace_difo_release(dp, vstate); + + if (DTRACEACT_ISPRINTFLIKE(kind)) { + char *str = (char *)(uintptr_t)act->dtad_arg; + +#ifdef FIXME + ASSERT((str != NULL && (uintptr_t)str >= KERNELBASE) || + (str == NULL && act->dtad_kind == DTRACEACT_PRINTA)); +#else + ASSERT((str != NULL) || + (str == NULL && act->dtad_kind == DTRACEACT_PRINTA)); +#endif + + if (str != NULL) + kfree(str); + } + + kfree(act); +} diff --git a/dtrace/dtrace_anon.c b/dtrace/dtrace_anon.c new file mode 100644 index 000000000000..57de888eff0a --- /dev/null +++ b/dtrace/dtrace_anon.c @@ -0,0 +1,130 @@ +/* + * FILE: dtrace_anon.c + * DESCRIPTION: Dynamic Tracing: anonymous enabling functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include "dtrace.h" + +dtrace_anon_t dtrace_anon; + +dtrace_state_t *dtrace_anon_grab(void) +{ + dtrace_state_t *state; + + ASSERT(mutex_is_locked(&dtrace_lock)); + + if ((state = dtrace_anon.dta_state) == NULL) { + ASSERT(dtrace_anon.dta_enabling == NULL); + + return NULL; + } + + ASSERT(dtrace_anon.dta_enabling != NULL); + ASSERT(dtrace_retained != NULL); + + dtrace_enabling_destroy(dtrace_anon.dta_enabling); + dtrace_anon.dta_enabling = NULL; + dtrace_anon.dta_state = NULL; + + return state; +} + +void dtrace_anon_property(void) +{ + int i, rv; + dtrace_state_t *state; + dof_hdr_t *dof; + char c[32]; /* enough for "dof-data-" + digits */ + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(mutex_is_locked(&cpu_lock)); + + for (i = 0; ; i++) { + snprintf(c, sizeof (c), "dof-data-%d", i); + + dtrace_err_verbose = 1; + + if ((dof = dtrace_dof_property(c)) == NULL) { + dtrace_err_verbose = 0; + break; + } + +#ifdef FIXME + /* + * We want to create anonymous state, so we need to transition + * the kernel debugger to indicate that DTrace is active. If + * this fails (e.g. because the debugger has modified text in + * some way), we won't continue with the processing. + */ + if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) { + pr_info("kernel debugger active; " + "anonymous enabling ignored."); + dtrace_dof_destroy(dof); + break; + } +#endif + + /* + * If we haven't allocated an anonymous state, we'll do so now. + */ + if ((state = dtrace_anon.dta_state) == NULL) { + state = dtrace_state_create(NULL); + dtrace_anon.dta_state = state; + + if (state == NULL) { + /* + * This basically shouldn't happen: there is no + * failure mode from dtrace_state_create(). + * Still, the interface allows for a failure + * mode, and we want to fail as gracefully as + * possible: we'll emit an error message and + * cease processing anonymous state in this + * case. + */ + pr_warning("failed to create anonymous state"); + dtrace_dof_destroy(dof); + break; + } + } + + rv = dtrace_dof_slurp(dof, &state->dts_vstate, current_cred(), + &dtrace_anon.dta_enabling, 0, TRUE); + + if (rv == 0) + rv = dtrace_dof_options(dof, state); + + dtrace_err_verbose = 0; + dtrace_dof_destroy(dof); + + if (rv != 0) { + /* + * This is malformed DOF; chuck any anonymous state + * that we created. + */ + ASSERT(dtrace_anon.dta_enabling == NULL); + dtrace_state_destroy(state); + dtrace_anon.dta_state = NULL; + break; + } + + ASSERT(dtrace_anon.dta_enabling != NULL); + } + + if (dtrace_anon.dta_enabling != NULL) { + int rval; + + /* + * dtrace_enabling_retain() can only fail because we are + * trying to retain more enablings than are allowed -- but + * we only have one anonymous enabling, and we are guaranteed + * to be allowed at least one retained enabling; we assert + * that dtrace_enabling_retain() returns success. + */ + rval = dtrace_enabling_retain(dtrace_anon.dta_enabling); + ASSERT(rval == 0); + + dtrace_enabling_dump(dtrace_anon.dta_enabling); + } +} diff --git a/dtrace/dtrace_asm.S b/dtrace/dtrace_asm.S new file mode 100644 index 000000000000..3aeb2c53ebb3 --- /dev/null +++ b/dtrace/dtrace_asm.S @@ -0,0 +1,328 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#include + +#define CPU_DTRACE_BADADDR 0x0004 /* DTrace fault: bad address */ + +#if defined(__x86_64__) + + ENTRY(dtrace_getfp) + movq %rbp, %rax + ret + ENDPROC(dtrace_getfp) + +#elif defined(__i386__) + + ENTRY(dtrace_getfp) + movl %ebp, %eax + ret + ENDPROC(dtrace_getfp) + +#endif /* __i386__ */ + + +#if defined(__x86_64__) + + ENTRY(dtrace_cas32) + movl %esi, %eax + lock + cmpxchgl %edx, (%rdi) + ret + ENDPROC(dtrace_cas32) + + ENTRY(dtrace_casptr) + movq %rsi, %rax + lock + cmpxchgq %rdx, (%rdi) + ret + ENDPROC(dtrace_casptr) + +#elif defined(__i386__) + + ENTRY(dtrace_cas32) + ENTRY(dtrace_casptr) + movl 4(%esp), %edx + movl 8(%esp), %eax + movl 12(%esp), %ecx + lock + cmpxchgl %ecx, (%edx) + ret + ENDPROC(dtrace_casptr) + ENDPROC(dtrace_cas32) + +#endif /* __i386__ */ + +#if defined(__x86_64__) + ENTRY(dtrace_caller) + movq $-1, %rax + ret + ENDPROC(dtrace_caller) + +#elif defined(__i386__) + + ENTRY(dtrace_caller) + movl $-1, %eax + ret + ENDPROC(dtrace_caller) + +#endif /* __i386__ */ + +#if defined(__x86_64__) + + ENTRY(dtrace_copy) + pushq %rbp + movq %rsp, %rbp + + xchgq %rdi, %rsi # make %rsi source, %rdi dest + movq %rdx, %rcx # load count + repz # repeat for count ... + smovb # move from %ds:rsi to %ed:rdi + leave + ret + ENDPROC(dtrace_copy) + +#elif defined(__i386__) + + ENTRY(dtrace_copy) + pushl %ebp + movl %esp, %ebp + pushl %esi + pushl %edi + + movl 8(%ebp), %esi # Load source address + movl 12(%ebp), %edi # Load destination address + movl 16(%ebp), %ecx # Load count + repz # Repeat for count... + smovb # move from %ds:si to %es:di + + popl %edi + popl %esi + movl %ebp, %esp + popl %ebp + ret + ENDPROC(dtrace_copy) + +#endif /* __i386__ */ + +#if defined(__x86_64__) + + ENTRY(dtrace_copystr) + pushq %rbp + movq %rsp, %rbp + +0: + movb (%rdi), %al # load from source + movb %al, (%rsi) # store to destination + addq $1, %rdi # increment source pointer + addq $1, %rsi # increment destination pointer + subq $1, %rdx # decrement remaining count + cmpb $0, %al + je 2f + testq $0xfff, %rdx # test if count is 4k-aligned + jnz 1f # if not, continue with copying + testq $CPU_DTRACE_BADADDR, (%rcx) # load and test dtrace flags + jnz 2f +1: + cmpq $0, %rdx + jne 0b +2: + leave + ret + + ENDPROC(dtrace_copystr) + +#elif defined(__i386__) + + ENTRY(dtrace_copystr) + + pushl %ebp # Setup stack frame + movl %esp, %ebp + pushl %ebx # Save registers + + movl 8(%ebp), %ebx # Load source address + movl 12(%ebp), %edx # Load destination address + movl 16(%ebp), %ecx # Load count + +0: + movb (%ebx), %al # Load from source + movb %al, (%edx) # Store to destination + incl %ebx # Increment source pointer + incl %edx # Increment destination pointer + decl %ecx # Decrement remaining count + cmpb $0, %al + je 2f + testl $0xfff, %ecx # Check if count is 4k-aligned + jnz 1f + movl 20(%ebp), %eax # load flags pointer + testl $CPU_DTRACE_BADADDR, (%eax) # load and test dtrace flags + jnz 2f +1: + cmpl $0, %ecx + jne 0b + +2: + popl %ebx + movl %ebp, %esp + popl %ebp + ret + + ENDPROC(dtrace_copystr) + +#endif /* __i386__ */ + +#if defined(__x86_64__) + + ENTRY(dtrace_fulword) + movq (%rdi), %rax + ret + ENDPROC(dtrace_fulword) + +#elif defined(__i386__) + + ENTRY(dtrace_fulword) + movl 4(%esp), %ecx + xorl %eax, %eax + movl (%ecx), %eax + ret + ENDPROC(dtrace_fulword) + +#endif /* __i386__ */ + +#if defined(__x86_64__) + + ENTRY(dtrace_fuword8_nocheck) + xorq %rax, %rax + movb (%rdi), %al + ret + ENDPROC(dtrace_fuword8_nocheck) + +#elif defined(__i386__) + + ENTRY(dtrace_fuword8_nocheck) + movl 4(%esp), %ecx + xorl %eax, %eax + movzbl (%ecx), %eax + ret + ENDPROC(dtrace_fuword8_nocheck) + +#endif /* __i386__ */ + +#if defined(__x86_64__) + + ENTRY(dtrace_fuword16_nocheck) + xorq %rax, %rax + movw (%rdi), %ax + ret + ENDPROC(dtrace_fuword16_nocheck) + +#elif defined(__i386__) + + ENTRY(dtrace_fuword16_nocheck) + movl 4(%esp), %ecx + xorl %eax, %eax + movzwl (%ecx), %eax + ret + ENDPROC(dtrace_fuword16_nocheck) + +#endif /* __i386__ */ + +#if defined(__x86_64__) + + ENTRY(dtrace_fuword32_nocheck) + xorq %rax, %rax + movl (%rdi), %eax + ret + ENDPROC(dtrace_fuword32_nocheck) + +#elif defined(__i386__) + + ENTRY(dtrace_fuword32_nocheck) + movl 4(%esp), %ecx + xorl %eax, %eax + movl (%ecx), %eax + ret + ENDPROC(dtrace_fuword32_nocheck) + +#endif /* __i386__ */ + +#if defined(__x86_64__) + + ENTRY(dtrace_fuword64_nocheck) + movq (%rdi), %rax + ret + ENDPROC(dtrace_fuword64_nocheck) + +#elif defined(__i386__) + + ENTRY(dtrace_fuword64_nocheck) + movl 4(%esp), %ecx + xorl %eax, %eax + xorl %edx, %edx + movl (%ecx), %eax + movl 4(%ecx), %edx + ret + ENDPROC(dtrace_fuword64_nocheck) + +#endif /* __i386__ */ + +#if defined(__x86_64__) + + ENTRY(dtrace_probe_error) + pushq %rbp + movq %rsp, %rbp + subq $0x8, %rsp + movq %r9, (%rsp) + movq %r8, %r9 + movq %rcx, %r8 + movq %rdx, %rcx + movq %rsi, %rdx + movq %rdi, %rsi + movl dtrace_probeid_error(%rip), %edi + call dtrace_probe + addq $0x8, %rsp + leave + ret + ENDPROC(dtrace_probe_error) + +#elif defined(__i386__) + + ENTRY(dtrace_probe_error) + pushl %ebp + movl %esp, %ebp + pushl 0x1c(%ebp) + pushl 0x18(%ebp) + pushl 0x14(%ebp) + pushl 0x10(%ebp) + pushl 0xc(%ebp) + pushl 0x8(%ebp) + pushl dtrace_probeid_error + call dtrace_probe + movl %ebp, %esp + popl %ebp + ret + ENDPROC(dtrace_probe_error) + +#endif /* __i386__ */ diff --git a/dtrace/dtrace_buffer.c b/dtrace/dtrace_buffer.c new file mode 100644 index 000000000000..073cd59f3b38 --- /dev/null +++ b/dtrace/dtrace_buffer.c @@ -0,0 +1,454 @@ +/* + * FILE: dtrace_buffer.c + * DESCRIPTION: Dynamic Tracing: buffer functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace.h" + +dtrace_optval_t dtrace_nonroot_maxsize = (16 * 1024 * 1024); + +/* + * Note: called from cross call context. This function switches the two + * buffers on a given CPU. The atomicity of this operation is assured by + * disabling interrupts while the actual switch takes place; the disabling of + * interrupts serializes the execution with any execution of dtrace_probe() on + * the same CPU. + */ +void dtrace_buffer_switch(dtrace_buffer_t *buf) +{ + caddr_t tomax = buf->dtb_tomax; + caddr_t xamot = buf->dtb_xamot; + dtrace_icookie_t cookie; + + ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); + ASSERT(!(buf->dtb_flags & DTRACEBUF_RING)); + + local_irq_save(cookie); + + buf->dtb_tomax = xamot; + buf->dtb_xamot = tomax; + buf->dtb_xamot_drops = buf->dtb_drops; + buf->dtb_xamot_offset = buf->dtb_offset; + buf->dtb_xamot_errors = buf->dtb_errors; + buf->dtb_xamot_flags = buf->dtb_flags; + buf->dtb_offset = 0; + buf->dtb_drops = 0; + buf->dtb_errors = 0; + buf->dtb_flags &= ~(DTRACEBUF_ERROR | DTRACEBUF_DROPPED); + + local_irq_restore(cookie); +} + +/* + * Note: called from cross call context. This function activates a buffer + * on a CPU. As with dtrace_buffer_switch(), the atomicity of the operation + * is guaranteed by the disabling of interrupts. + */ +void dtrace_buffer_activate(dtrace_state_t *state) +{ + dtrace_buffer_t *buf; + dtrace_icookie_t cookie; + + local_irq_save(cookie); + + buf = &state->dts_buffer[smp_processor_id()]; + + if (buf->dtb_tomax != NULL) { + /* + * We might like to assert that the buffer is marked inactive, + * but this isn't necessarily true: the buffer for the CPU + * that processes the BEGIN probe has its buffer activated + * manually. In this case, we take the (harmless) action + * re-clearing the bit INACTIVE bit. + */ + buf->dtb_flags &= ~DTRACEBUF_INACTIVE; + } + + local_irq_restore(cookie); +} + +int dtrace_buffer_alloc(dtrace_buffer_t *bufs, size_t size, int flags, + processorid_t cpuid) +{ + processorid_t cpu; + dtrace_buffer_t *buf; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(mutex_is_locked(&cpu_lock)); + +#ifdef FIXME + if (size > dtrace_nonroot_maxsize && + !PRIV_POLICY_CHOICE(current_cred(), PRIV_ALL, FALSE)) + return -EFBIG; +#endif + + for_each_online_cpu(cpu) { + if (cpuid != DTRACE_CPUALL && cpuid != cpu) + continue; + + buf = &bufs[cpu]; + + /* + * If there is already a buffer allocated for this CPU, it + * is only possible that this is a DR event. In this case, + * the buffer size must match our specified size. + */ + if (buf->dtb_tomax != NULL) { + ASSERT(buf->dtb_size == size); + continue; + } + + ASSERT(buf->dtb_xamot == NULL); + + if ((buf->dtb_tomax = dtrace_vzalloc_try(size)) == NULL) + goto err; + + buf->dtb_size = size; + buf->dtb_flags = flags; + buf->dtb_offset = 0; + buf->dtb_drops = 0; + + if (flags & DTRACEBUF_NOSWITCH) + continue; + + if ((buf->dtb_xamot = dtrace_vzalloc_try(size)) == NULL) + goto err; + } + + return 0; + +err: + for_each_online_cpu(cpu) { + if (cpuid != DTRACE_CPUALL && cpuid != cpu) + continue; + + buf = &bufs[cpu]; + + if (buf->dtb_xamot != NULL) { + ASSERT(buf->dtb_tomax != NULL); + ASSERT(buf->dtb_size == size); + vfree(buf->dtb_xamot); + } + + if (buf->dtb_tomax != NULL) { + ASSERT(buf->dtb_size == size); + vfree(buf->dtb_tomax); + } + + buf->dtb_tomax = NULL; + buf->dtb_xamot = NULL; + buf->dtb_size = 0; + } + + return -ENOMEM; +} +void dtrace_buffer_drop(dtrace_buffer_t *buf) +{ + buf->dtb_drops++; +} + +intptr_t dtrace_buffer_reserve(dtrace_buffer_t *buf, size_t needed, + size_t align, dtrace_state_t *state, + dtrace_mstate_t *mstate) +{ + intptr_t offs = buf->dtb_offset, soffs; + intptr_t woffs; + caddr_t tomax; + size_t total; + + if (buf->dtb_flags & DTRACEBUF_INACTIVE) + return -1; + + if ((tomax = buf->dtb_tomax) == NULL) { + dtrace_buffer_drop(buf); + return -1; + } + + if (!(buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL))) { + while (offs & (align - 1)) { + /* + * Assert that our alignment is off by a number which + * is itself sizeof (uint32_t) aligned. + */ + ASSERT(!((align - (offs & (align - 1))) & + (sizeof (uint32_t) - 1))); + DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE); + offs += sizeof (uint32_t); + } + + if ((soffs = offs + needed) > buf->dtb_size) { + dtrace_buffer_drop(buf); + return -1; + } + + if (mstate == NULL) + return (offs); + + mstate->dtms_scratch_base = (uintptr_t)tomax + soffs; + mstate->dtms_scratch_size = buf->dtb_size - soffs; + mstate->dtms_scratch_ptr = mstate->dtms_scratch_base; + + return offs; + } + + if (buf->dtb_flags & DTRACEBUF_FILL) { + if (state->dts_activity != DTRACE_ACTIVITY_COOLDOWN && + (buf->dtb_flags & DTRACEBUF_FULL)) + return -1; + + goto out; + } + + total = needed + (offs & (align - 1)); + + /* + * For a ring buffer, life is quite a bit more complicated. Before + * we can store any padding, we need to adjust our wrapping offset. + * (If we've never before wrapped or we're not about to, no adjustment + * is required.) + */ + if ((buf->dtb_flags & DTRACEBUF_WRAPPED) || + offs + total > buf->dtb_size) { + woffs = buf->dtb_xamot_offset; + + if (offs + total > buf->dtb_size) { + /* + * We can't fit in the end of the buffer. First, a + * sanity check that we can fit in the buffer at all. + */ + if (total > buf->dtb_size) { + dtrace_buffer_drop(buf); + return -1; + } + + /* + * We're going to be storing at the top of the buffer, + * so now we need to deal with the wrapped offset. We + * only reset our wrapped offset to 0 if it is + * currently greater than the current offset. If it + * is less than the current offset, it is because a + * previous allocation induced a wrap -- but the + * allocation didn't subsequently take the space due + * to an error or false predicate evaluation. In this + * case, we'll just leave the wrapped offset alone: if + * the wrapped offset hasn't been advanced far enough + * for this allocation, it will be adjusted in the + * lower loop. + */ + if (buf->dtb_flags & DTRACEBUF_WRAPPED) { + if (woffs >= offs) + woffs = 0; + } else + woffs = 0; + + /* + * Now we know that we're going to be storing to the + * top of the buffer and that there is room for us + * there. We need to clear the buffer from the current + * offset to the end (there may be old gunk there). + */ + while (offs < buf->dtb_size) + tomax[offs++] = 0; + + /* + * We need to set our offset to zero. And because we + * are wrapping, we need to set the bit indicating as + * much. We can also adjust our needed space back + * down to the space required by the ECB -- we know + * that the top of the buffer is aligned. + */ + offs = 0; + total = needed; + buf->dtb_flags |= DTRACEBUF_WRAPPED; + } else { + /* + * There is room for us in the buffer, so we simply + * need to check the wrapped offset. + */ + if (woffs < offs) { + /* + * The wrapped offset is less than the offset. + * This can happen if we allocated buffer space + * that induced a wrap, but then we didn't + * subsequently take the space due to an error + * or false predicate evaluation. This is + * okay; we know that _this_ allocation isn't + * going to induce a wrap. We still can't + * reset the wrapped offset to be zero, + * however: the space may have been trashed in + * the previous failed probe attempt. But at + * least the wrapped offset doesn't need to + * be adjusted at all... + */ + goto out; + } + } + + while (offs + total > woffs) { + dtrace_epid_t epid = *(uint32_t *)(tomax + woffs); + size_t size; + + if (epid == DTRACE_EPIDNONE) + size = sizeof (uint32_t); + else { + ASSERT(epid <= state->dts_necbs); + ASSERT(state->dts_ecbs[epid - 1] != NULL); + + size = state->dts_ecbs[epid - 1]->dte_size; + } + + ASSERT(woffs + size <= buf->dtb_size); + ASSERT(size != 0); + + if (woffs + size == buf->dtb_size) { + /* + * We've reached the end of the buffer; we want + * to set the wrapped offset to 0 and break + * out. However, if the offs is 0, then we're + * in a strange edge-condition: the amount of + * space that we want to reserve plus the size + * of the record that we're overwriting is + * space but subsequently don't consume it (due + * to a failed predicate or error) the wrapped + * offset will be 0 -- yet the EPID at offset 0 + * will not be committed. This situation is + * relatively easy to deal with: if we're in + * this case, the buffer is indistinguishable + * from one that hasn't wrapped; we need only + * finish the job by clearing the wrapped bit, + * explicitly setting the offset to be 0, and + * zero'ing out the old data in the buffer. + */ + if (offs == 0) { + buf->dtb_flags &= ~DTRACEBUF_WRAPPED; + buf->dtb_offset = 0; + woffs = total; + + while (woffs < buf->dtb_size) + tomax[woffs++] = 0; + } + + woffs = 0; + break; + } + + woffs += size; + } + + /* + * We have a wrapped offset. It may be that the wrapped offset + * has become zero -- that's okay. + */ + buf->dtb_xamot_offset = woffs; + } + +out: + /* + * Now we can plow the buffer with any necessary padding. + */ + while (offs & (align - 1)) { + /* + * Assert that our alignment is off by a number which + * is itself sizeof (uint32_t) aligned. + */ + ASSERT(!((align - (offs & (align - 1))) & + (sizeof (uint32_t) - 1))); + DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE); + offs += sizeof (uint32_t); + } + + if (buf->dtb_flags & DTRACEBUF_FILL) { + if (offs + needed > buf->dtb_size - state->dts_reserve) { + buf->dtb_flags |= DTRACEBUF_FULL; + return -1; + } + } + + if (mstate == NULL) + return offs; + + /* + * For ring buffers and fill buffers, the scratch space is always + * the inactive buffer. + */ + mstate->dtms_scratch_base = (uintptr_t)buf->dtb_xamot; + mstate->dtms_scratch_size = buf->dtb_size; + mstate->dtms_scratch_ptr = mstate->dtms_scratch_base; + + return offs; +} + +void dtrace_buffer_polish(dtrace_buffer_t *buf) +{ + ASSERT(buf->dtb_flags & DTRACEBUF_RING); + ASSERT(mutex_is_locked(&dtrace_lock)); + + if (!(buf->dtb_flags & DTRACEBUF_WRAPPED)) + return; + + /* + * We need to polish the ring buffer. There are three cases: + * + * - The first (and presumably most common) is that there is no gap + * between the buffer offset and the wrapped offset. In this case, + * there is nothing in the buffer that isn't valid data; we can + * mark the buffer as polished and return. + * + * - The second (less common than the first but still more common + * than the third) is that there is a gap between the buffer offset + * and the wrapped offset, and the wrapped offset is larger than the + * buffer offset. This can happen because of an alignment issue, or + * can happen because of a call to dtrace_buffer_reserve() that + * didn't subsequently consume the buffer space. In this case, + * we need to zero the data from the buffer offset to the wrapped + * offset. + * + * - The third (and least common) is that there is a gap between the + * buffer offset and the wrapped offset, but the wrapped offset is + * _less_ than the buffer offset. This can only happen because a + * call to dtrace_buffer_reserve() induced a wrap, but the space + * was not subsequently consumed. In this case, we need to zero the + * space from the offset to the end of the buffer _and_ from the + * top of the buffer to the wrapped offset. + */ + if (buf->dtb_offset < buf->dtb_xamot_offset) + memset(buf->dtb_tomax + buf->dtb_offset, 0, + buf->dtb_xamot_offset - buf->dtb_offset); + + if (buf->dtb_offset > buf->dtb_xamot_offset) { + memset(buf->dtb_tomax + buf->dtb_offset, 0, + buf->dtb_size - buf->dtb_offset); + memset(buf->dtb_tomax, 0, buf->dtb_xamot_offset); + } +} + +void dtrace_buffer_free(dtrace_buffer_t *bufs) +{ + int cpu; + + for_each_online_cpu(cpu) { + dtrace_buffer_t *buf = &bufs[cpu]; + + if (buf->dtb_tomax == NULL) { + ASSERT(buf->dtb_xamot == NULL); + ASSERT(buf->dtb_size == 0); + + continue; + } + + if (buf->dtb_xamot != NULL) { + ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); + + vfree(buf->dtb_xamot); + buf->dtb_xamot = NULL; + } + + vfree(buf->dtb_tomax); + buf->dtb_size = 0; + buf->dtb_tomax = NULL; + } +} diff --git a/dtrace/dtrace_dev.c b/dtrace/dtrace_dev.c new file mode 100644 index 000000000000..9dfe442e0c42 --- /dev/null +++ b/dtrace/dtrace_dev.c @@ -0,0 +1,1317 @@ +/* + * FILE: dtrace_dev.c + * DESCRIPTION: Dynamic Tracing: device file handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include +#include +#include +#include +#include +#include + +#include "ctf_api.h" +#include "dtrace.h" +#include "dtrace_dev.h" +#include "dtrace_ioctl.h" + +uint32_t dtrace_helptrace_next = 0; +uint32_t dtrace_helptrace_nlocals; +char *dtrace_helptrace_buffer; +int dtrace_helptrace_bufsize = 512 * 1024; + +#ifdef CONFIG_DT_DEBUG +int dtrace_helptrace_enabled = 1; +#else +int dtrace_helptrace_enabled = 0; +#endif + +int dtrace_opens; +int dtrace_err_verbose; + +dtrace_pops_t dtrace_provider_ops = { + (void (*)(void *, const dtrace_probedesc_t *))dtrace_nullop, + (void (*)(void *, struct module *))dtrace_nullop, + (int (*)(void *, dtrace_id_t, void *))dtrace_enable_nullop, + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, + NULL, + NULL, + NULL, + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop +}; + +dtrace_toxrange_t *dtrace_toxrange; +int dtrace_toxranges; + +static size_t dtrace_retain_max = 1024; + +static int dtrace_toxranges_max; + +static dtrace_pattr_t dtrace_provider_attr = { +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +}; + +void (*dtrace_modload)(struct module *); +void (*dtrace_modunload)(struct module *); + +void dtrace_nullop(void) +{ +} + +int dtrace_enable_nullop(void) +{ + return 0; +} + +static long dtrace_ioctl(struct file *file, + unsigned int cmd, unsigned long arg) +{ + dtrace_state_t *state = (dtrace_state_t *)file->private_data; + int rval; + void __user *argp = (void __user *)arg; + + if (state->dts_anon) { + ASSERT(dtrace_anon.dta_state == NULL); + state = state->dts_anon; + } + + switch (cmd) { + case DTRACEIOC_PROVIDER: { + dtrace_providerdesc_t pvd; + dtrace_provider_t *pvp; + +printk(KERN_INFO "IOCTL provider\n"); + if (copy_from_user(&pvd, argp, sizeof(pvd)) != 0) + return -EFAULT; + + pvd.dtvd_name[DTRACE_PROVNAMELEN - 1] = '\0'; + mutex_lock(&dtrace_provider_lock); + + for (pvp = dtrace_provider; pvp != NULL; pvp = pvp->dtpv_next) { + if (strcmp(pvp->dtpv_name, pvd.dtvd_name) == 0) + break; + } + + mutex_unlock(&dtrace_provider_lock); + + if (pvp == NULL) + return -ESRCH; + + memcpy(&pvd.dtvd_priv, &pvp->dtpv_priv, + sizeof(dtrace_ppriv_t)); + memcpy(&pvd.dtvd_attr, &pvp->dtpv_attr, + sizeof(dtrace_pattr_t)); + + if (copy_to_user(argp, &pvd, sizeof(pvd)) != 0) + return -EFAULT; + + return 0; + } + + case DTRACEIOC_EPROBE: { + dtrace_eprobedesc_t epdesc; + dtrace_ecb_t *ecb; + dtrace_action_t *act; + void *buf; + size_t size; + uint8_t *dest; + int nrecs; + +printk(KERN_INFO "IOCTL eprobe\n"); + if (copy_from_user(&epdesc, argp, sizeof(epdesc)) != 0) + return -EFAULT; + + mutex_lock(&dtrace_lock); + +printk(KERN_INFO " Looking for ECB %ld\n", epdesc.dtepd_epid); + if ((ecb = dtrace_epid2ecb(state, epdesc.dtepd_epid)) == NULL) { + mutex_unlock(&dtrace_lock); +printk(KERN_INFO " ECB not found\n"); + return -EINVAL; + } + + if (ecb->dte_probe == NULL) { + mutex_unlock(&dtrace_lock); +printk(KERN_INFO " ECB has no probe\n"); + return -EINVAL; + } + +printk(KERN_INFO " ECB has probe %ld\n", ecb->dte_probe->dtpr_id); + epdesc.dtepd_probeid = ecb->dte_probe->dtpr_id; + epdesc.dtepd_uarg = ecb->dte_uarg; + epdesc.dtepd_size = ecb->dte_size; + + nrecs = epdesc.dtepd_nrecs; + epdesc.dtepd_nrecs = 0; + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple) + continue; + + epdesc.dtepd_nrecs++; + } + + /* + * Now that we have the size, we need to allocate a temporary + * buffer in which to store the complete description. We need + * the temporary buffer to be able to drop dtrace_lock() + * across the copy_to_user(), below. + */ + size = sizeof(dtrace_eprobedesc_t) + + (epdesc.dtepd_nrecs * sizeof(dtrace_recdesc_t)); + + buf = kmalloc(size, GFP_KERNEL); + dest = buf; + + memcpy(dest, &epdesc, sizeof(epdesc)); + dest += offsetof(dtrace_eprobedesc_t, dtepd_rec[0]); + + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple) + continue; + + if (nrecs-- == 0) + break; + + memcpy(dest, &act->dta_rec, sizeof(dtrace_recdesc_t)); + dest += sizeof(dtrace_recdesc_t); + } + + mutex_unlock(&dtrace_lock); + + if (copy_to_user(argp, buf, + (uintptr_t)(dest - (uint8_t *)buf)) != 0) { + kfree(buf); + return -EFAULT; + } + + kfree(buf); + return 0; + } + + case DTRACEIOC_AGGDESC: { + dtrace_aggdesc_t aggdesc; + dtrace_action_t *act; + dtrace_aggregation_t *agg; + int nrecs; + uint32_t offs; + dtrace_recdesc_t *lrec; + void *buf; + size_t size; + uint8_t *dest; + +printk(KERN_INFO "IOCTL aggdesc\n"); + if (copy_from_user(&aggdesc, argp, sizeof(aggdesc)) != 0) + return -EFAULT; + + mutex_lock(&dtrace_lock); + + if ((agg = dtrace_aggid2agg(state, aggdesc.dtagd_id)) == NULL) { + mutex_unlock(&dtrace_lock); + return -EINVAL; + } +printk(KERN_INFO "IOCTL aggdesc: Found agg %lu, ECB %lu\n", agg->dtag_id, agg->dtag_ecb->dte_epid); + + aggdesc.dtagd_epid = agg->dtag_ecb->dte_epid; + + nrecs = aggdesc.dtagd_nrecs; + aggdesc.dtagd_nrecs = 0; + + offs = agg->dtag_base; + lrec = &agg->dtag_action.dta_rec; + aggdesc.dtagd_size = lrec->dtrd_offset + lrec->dtrd_size - + offs; + + for (act = agg->dtag_first; ; act = act->dta_next) { + ASSERT(act->dta_intuple || + DTRACEACT_ISAGG(act->dta_kind)); + + /* + * If this action has a record size of zero, it + * denotes an argument to the aggregating action. + * Because the presence of this record doesn't (or + * shouldn't) affect the way the data is interpreted, + * we don't copy it out to save user-level the + * confusion of dealing with a zero-length record. + */ + if (act->dta_rec.dtrd_size == 0) { + ASSERT(agg->dtag_hasarg); + continue; + } + + aggdesc.dtagd_nrecs++; + + if (act == &agg->dtag_action) + break; + } + + /* + * Now that we have the size, we need to allocate a temporary + * buffer in which to store the complete description. We need + * the temporary buffer to be able to drop dtrace_lock() + * across the copyout(), below. + */ + size = sizeof(dtrace_aggdesc_t) + + (aggdesc.dtagd_nrecs * sizeof(dtrace_recdesc_t)); + + buf = kmalloc(size, GFP_KERNEL); + dest = buf; + + memcpy(dest, &aggdesc, sizeof(aggdesc)); + dest += offsetof(dtrace_aggdesc_t, dtagd_rec[0]); + + for (act = agg->dtag_first; ; act = act->dta_next) { + dtrace_recdesc_t rec = act->dta_rec; + + /* + * See the comment in the above loop for why we pass + * over zero-length records. + */ + if (rec.dtrd_size == 0) { + ASSERT(agg->dtag_hasarg); + continue; + } + + if (nrecs-- == 0) + break; + + rec.dtrd_offset -= offs; + memcpy(dest, &rec, sizeof(rec)); + dest += sizeof(dtrace_recdesc_t); + + if (act == &agg->dtag_action) + break; + } + + mutex_unlock(&dtrace_lock); + + if (copy_to_user(argp, buf, + (uintptr_t)(dest - (uint8_t *)buf)) != 0) { + kfree(buf); + return -EFAULT; + } + + kfree(buf); + return 0; + } + + case DTRACEIOC_ENABLE: { + dof_hdr_t *dof; + dtrace_enabling_t *enab = NULL; + dtrace_vstate_t *vstate; + int err = 0; + int rv; + +printk(KERN_INFO "IOCTL enable\n"); + rv = 0; + + /* + * If a NULL argument has been passed, we take this as our + * cue to reevaluate our enablings. + */ + if (argp == NULL) { + dtrace_enabling_matchall(); + + return 0; + } + + if ((dof = dtrace_dof_copyin(argp, &rval)) == NULL) + return rval; + + mutex_lock(&cpu_lock); + mutex_lock(&dtrace_lock); + vstate = &state->dts_vstate; + + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) { + mutex_unlock(&dtrace_lock); + mutex_unlock(&cpu_lock); + dtrace_dof_destroy(dof); + return -EBUSY; + } + + if (dtrace_dof_slurp(dof, vstate, file->f_cred, &enab, 0, + TRUE) != 0) { + mutex_unlock(&dtrace_lock); + mutex_unlock(&cpu_lock); + dtrace_dof_destroy(dof); + return -EINVAL; + } + + if ((rval = dtrace_dof_options(dof, state)) != 0) { + dtrace_enabling_destroy(enab); + mutex_unlock(&dtrace_lock); + mutex_unlock(&cpu_lock); + dtrace_dof_destroy(dof); + return rval; + } + + if ((err = dtrace_enabling_match(enab, &rv)) == 0) + err = dtrace_enabling_retain(enab); + else + dtrace_enabling_destroy(enab); + + mutex_unlock(&dtrace_lock); + mutex_unlock(&cpu_lock); + dtrace_dof_destroy(dof); + + return err == 0 ? rv : err; + } + + case DTRACEIOC_REPLICATE: { + dtrace_repldesc_t desc; + dtrace_probedesc_t *match = &desc.dtrpd_match; + dtrace_probedesc_t *create = &desc.dtrpd_create; + int err; + +printk(KERN_INFO "IOCTL replicate\n"); + if (copy_from_user(&desc, argp, sizeof(desc)) != 0) + return -EFAULT; + + match->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; + match->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; + match->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; + match->dtpd_name[DTRACE_NAMELEN - 1] = '\0'; + + create->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; + create->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; + create->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; + create->dtpd_name[DTRACE_NAMELEN - 1] = '\0'; + + mutex_lock(&dtrace_lock); + err = dtrace_enabling_replicate(state, match, create); + mutex_unlock(&dtrace_lock); + + return err; + } + + case DTRACEIOC_PROBEMATCH: + case DTRACEIOC_PROBES: { + dtrace_probe_t *probe = NULL; + dtrace_probedesc_t desc; + dtrace_probekey_t pkey; + uint32_t priv; + uid_t uid; + +printk(KERN_INFO "IOCTL %s\n", cmd == DTRACEIOC_PROBEMATCH ? "probematch" : "probes"); + if (copy_from_user(&desc, argp, sizeof(desc)) != 0) + return -EFAULT; + + desc.dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; + desc.dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; + desc.dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; + desc.dtpd_name[DTRACE_NAMELEN - 1] = '\0'; + + /* + * Before we attempt to match this probe, we want to give + * all providers the opportunity to provide it. + */ + if (desc.dtpd_id == DTRACE_IDNONE) { + mutex_lock(&dtrace_provider_lock); + dtrace_probe_provide(&desc, NULL); + mutex_unlock(&dtrace_provider_lock); + } + + if (cmd == DTRACEIOC_PROBEMATCH) { + dtrace_probekey(&desc, &pkey); + pkey.dtpk_id = DTRACE_IDNONE; + } +printk(KERN_INFO "IOCTL %s: id[%d] p[%s] m[%s] f[%s] n[%s]\n", cmd == DTRACEIOC_PROBEMATCH ? "probematch" : "probes", desc.dtpd_id, desc.dtpd_provider, desc.dtpd_mod, desc.dtpd_func, desc.dtpd_name); + + dtrace_cred2priv(file->f_cred, &priv, &uid); + + mutex_lock(&dtrace_lock); + + if (cmd == DTRACEIOC_PROBEMATCH) { + int m = 0; + + while ((probe = dtrace_probe_get_next(desc.dtpd_id)) + != NULL) { + if ((m = dtrace_match_probe( + probe, &pkey, priv, uid))) + break; + + desc.dtpd_id = probe->dtpr_id + 1; + } + + if (m < 0) { + mutex_unlock(&dtrace_lock); + return -EINVAL; + } + } else { + while ((probe = dtrace_probe_get_next(desc.dtpd_id)) + != NULL) { + if (dtrace_match_priv(probe, priv, uid)) + break; + + desc.dtpd_id = probe->dtpr_id + 1; + } + } + + if (probe == NULL) { + mutex_unlock(&dtrace_lock); + return -ESRCH; + } + + dtrace_probe_description(probe, &desc); + mutex_unlock(&dtrace_lock); + +printk(KERN_INFO "Returning probe [%s]\n", desc.dtpd_name); + if (copy_to_user(argp, &desc, sizeof(desc)) != 0) + return -EFAULT; + + return 0; + } + + case DTRACEIOC_PROBEARG: { + dtrace_argdesc_t desc; + dtrace_probe_t *probe; + dtrace_provider_t *prov; + +printk(KERN_INFO "IOCTL probearg\n"); + if (copy_from_user(&desc, argp, sizeof(desc)) != 0) + return -EFAULT; + + if (desc.dtargd_id == DTRACE_IDNONE) + return -EINVAL; + + if (desc.dtargd_ndx == DTRACE_ARGNONE) + return -EINVAL; + + mutex_lock(&dtrace_provider_lock); +// mutex_lock(&module_mutex); /* FIXME */ + mutex_lock(&dtrace_lock); + + probe = dtrace_probe_lookup_id(desc.dtargd_id); + if (probe == NULL) { + mutex_unlock(&dtrace_lock); +// mutex_unlock(&module_mutex); /* FIXME */ + mutex_unlock(&dtrace_provider_lock); + + return -EINVAL; + } + + mutex_unlock(&dtrace_lock); + + prov = probe->dtpr_provider; + + if (prov->dtpv_pops.dtps_getargdesc == NULL) { + /* + * There isn't any typed information for this probe. + * Set the argument number to DTRACE_ARGNONE. + */ + desc.dtargd_ndx = DTRACE_ARGNONE; + } else { + desc.dtargd_native[0] = '\0'; + desc.dtargd_xlate[0] = '\0'; + desc.dtargd_mapping = desc.dtargd_ndx; + + prov->dtpv_pops.dtps_getargdesc( + prov->dtpv_arg, probe->dtpr_id, + probe->dtpr_arg, &desc); + } + +// mutex_unlock(&module_mutex); /* FIXME */ + mutex_unlock(&dtrace_provider_lock); + + if (copy_to_user(argp, &desc, sizeof(desc)) != 0) + return -EFAULT; + + return 0; + } + + case DTRACEIOC_GO: { + processorid_t cpuid; + +printk(KERN_INFO "IOCTL go\n"); + rval = dtrace_state_go(state, &cpuid); + + if (rval != 0) + return rval; + + if (copy_to_user(argp, &cpuid, sizeof(cpuid)) != 0) + return -EFAULT; + + return 0; + } + + case DTRACEIOC_STOP: { + processorid_t cpuid; + +printk(KERN_INFO "IOCTL stop\n"); + mutex_lock(&dtrace_lock); + rval = dtrace_state_stop(state, &cpuid); + mutex_unlock(&dtrace_lock); + + if (rval != 0) + return rval; + + if (copy_to_user(argp, &cpuid, sizeof(cpuid)) != 0) + return -EFAULT; + + return 0; + } + + case DTRACEIOC_DOFGET: { + dof_hdr_t hdr, *dof; + uint64_t len; + +printk(KERN_INFO "IOCTL dofget\n"); + if (copy_from_user(&hdr, argp, sizeof(hdr)) != 0) + return -EFAULT; + + mutex_lock(&dtrace_lock); + dof = dtrace_dof_create(state); + mutex_unlock(&dtrace_lock); + + len = min(hdr.dofh_loadsz, dof->dofh_loadsz); + rval = copy_to_user(argp, dof, len); + dtrace_dof_destroy(dof); + + return rval == 0 ? 0 : -EFAULT; + } + + case DTRACEIOC_AGGSNAP: + case DTRACEIOC_BUFSNAP: { + dtrace_bufdesc_t desc; + caddr_t cached; + dtrace_buffer_t *buf; + +printk(KERN_INFO "IOCTL %s\n", cmd == DTRACEIOC_AGGSNAP ? "aggsnap" : "bufsnap"); + if (copy_from_user(&desc, argp, sizeof(desc)) != 0) + return -EFAULT; + + if (desc.dtbd_cpu < 0 || desc.dtbd_cpu >= NR_CPUS) + return -EINVAL; + + mutex_lock(&dtrace_lock); + + if (cmd == DTRACEIOC_BUFSNAP) + buf = &state->dts_buffer[desc.dtbd_cpu]; + else + buf = &state->dts_aggbuffer[desc.dtbd_cpu]; + + if (buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL)) { + size_t sz = buf->dtb_offset; + + if (state->dts_activity != DTRACE_ACTIVITY_STOPPED) { + mutex_unlock(&dtrace_lock); + return -EBUSY; + } + + /* + * If this buffer has already been consumed, we're + * going to indicate that there's nothing left here + * to consume. + */ + if (buf->dtb_flags & DTRACEBUF_CONSUMED) { + mutex_unlock(&dtrace_lock); + + desc.dtbd_size = 0; + desc.dtbd_drops = 0; + desc.dtbd_errors = 0; + desc.dtbd_oldest = 0; + sz = sizeof(desc); + + if (copy_to_user(argp, &desc, sz) != 0) + return -EFAULT; + + return 0; + } + + /* + * If this is a ring buffer that has wrapped, we want + * to copy the whole thing out. + */ + if (buf->dtb_flags & DTRACEBUF_WRAPPED) { + dtrace_buffer_polish(buf); + sz = buf->dtb_size; + } + + if (copy_to_user(desc.dtbd_data, buf->dtb_tomax, + sz) != 0) { + mutex_unlock(&dtrace_lock); + return -EFAULT; + } + + desc.dtbd_size = sz; + desc.dtbd_drops = buf->dtb_drops; + desc.dtbd_errors = buf->dtb_errors; + desc.dtbd_oldest = buf->dtb_xamot_offset; + + mutex_unlock(&dtrace_lock); + + if (copy_to_user(argp, &desc, sizeof(desc)) != 0) + return -EFAULT; + + buf->dtb_flags |= DTRACEBUF_CONSUMED; + + return 0; + } + + if (buf->dtb_tomax == NULL) { + ASSERT(buf->dtb_xamot == NULL); + mutex_unlock(&dtrace_lock); + return -ENOENT; + } + + cached = buf->dtb_tomax; + + dtrace_xcall(desc.dtbd_cpu, + (dtrace_xcall_t)dtrace_buffer_switch, buf); + + state->dts_errors += buf->dtb_xamot_errors; + + /* + * If the buffers did not actually switch, then the cross call + * did not take place -- presumably because the given CPU is + * not in the ready set. If this is the case, we'll return + * ENOENT. + */ + if (buf->dtb_tomax == cached) { + ASSERT(buf->dtb_xamot != cached); + mutex_unlock(&dtrace_lock); + return -ENOENT; + } + + ASSERT(cached == buf->dtb_xamot); + + /* + * We have our snapshot; now copy it out. + */ + if (copy_to_user(desc.dtbd_data, buf->dtb_xamot, + buf->dtb_xamot_offset) != 0) { + mutex_unlock(&dtrace_lock); + return -EFAULT; + } + + desc.dtbd_size = buf->dtb_xamot_offset; + desc.dtbd_drops = buf->dtb_xamot_drops; + desc.dtbd_errors = buf->dtb_xamot_errors; + desc.dtbd_oldest = 0; + + mutex_unlock(&dtrace_lock); + + /* + * Finally, copy out the buffer description. + */ + if (copy_to_user(argp, &desc, sizeof(desc)) != 0) + return -EFAULT; + + return 0; + } + + case DTRACEIOC_CONF: { + dtrace_conf_t conf; + +printk(KERN_INFO "IOCTL conf\n"); + memset(&conf, 0, sizeof(conf)); + conf.dtc_difversion = DIF_VERSION; + conf.dtc_difintregs = DIF_DIR_NREGS; + conf.dtc_diftupregs = DIF_DTR_NREGS; + conf.dtc_ctfmodel = CTF_MODEL_NATIVE; + conf.dtc_maxbufs = nr_cpu_ids; + + if (copy_to_user(argp, &conf, sizeof(conf)) != 0) + return -EFAULT; + + return 0; + } + + case DTRACEIOC_STATUS: { + dtrace_status_t stat; + dtrace_dstate_t *dstate; + int i, j; + uint64_t nerrs; + +printk(KERN_INFO "IOCTL status\n"); + /* + * See the comment in dtrace_state_deadman() for the reason + * for setting dts_laststatus to INT64_MAX before setting + * it to the correct value. + */ + state->dts_laststatus = ns_to_ktime(INT64_MAX); + dtrace_membar_producer(); + state->dts_laststatus = dtrace_gethrtime(); + + memset(&stat, 0, sizeof(stat)); + + mutex_lock(&dtrace_lock); + + if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) { + mutex_unlock(&dtrace_lock); + return -ENOENT; + } + + if (state->dts_activity == DTRACE_ACTIVITY_DRAINING) + stat.dtst_exiting = 1; + + nerrs = state->dts_errors; + dstate = &state->dts_vstate.dtvs_dynvars; + + for (i = 0; i < NR_CPUS; i++) { + dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[i]; + + stat.dtst_dyndrops += dcpu->dtdsc_drops; + stat.dtst_dyndrops_dirty += dcpu->dtdsc_dirty_drops; + stat.dtst_dyndrops_rinsing += dcpu->dtdsc_rinsing_drops; + + if (state->dts_buffer[i].dtb_flags & DTRACEBUF_FULL) + stat.dtst_filled++; + + nerrs += state->dts_buffer[i].dtb_errors; + + for (j = 0; j < state->dts_nspeculations; j++) { + dtrace_speculation_t *spec; + dtrace_buffer_t *buf; + + spec = &state->dts_speculations[j]; + buf = &spec->dtsp_buffer[i]; + stat.dtst_specdrops += buf->dtb_xamot_drops; + } + } + + stat.dtst_specdrops_busy = state->dts_speculations_busy; + stat.dtst_specdrops_unavail = state->dts_speculations_unavail; + stat.dtst_stkstroverflows = state->dts_stkstroverflows; + stat.dtst_dblerrors = state->dts_dblerrors; + stat.dtst_killed = (state->dts_activity == + DTRACE_ACTIVITY_KILLED); + stat.dtst_errors = nerrs; + + mutex_unlock(&dtrace_lock); + + if (copy_to_user(argp, &stat, sizeof(stat)) != 0) + return -EFAULT; + + return 0; + } + + case DTRACEIOC_FORMAT: { + dtrace_fmtdesc_t fmt; + char *str; + int len; + +printk(KERN_INFO "IOCTL format\n"); + if (copy_from_user(&fmt, argp, sizeof (fmt)) != 0) + return -EFAULT; + + mutex_lock(&dtrace_lock); + + if (fmt.dtfd_format == 0 || + fmt.dtfd_format > state->dts_nformats) { + mutex_unlock(&dtrace_lock); + return -EINVAL; + } + + /* + * Format strings are allocated contiguously and they are + * never freed; if a format index is less than the number + * of formats, we can assert that the format map is non-NULL + * and that the format for the specified index is non-NULL. + */ + ASSERT(state->dts_formats != NULL); + str = state->dts_formats[fmt.dtfd_format - 1]; + ASSERT(str != NULL); + + len = strlen(str) + 1; + + if (len > fmt.dtfd_length) { + fmt.dtfd_length = len; + + if (copy_to_user(argp, &fmt, sizeof (fmt)) != 0) { + mutex_unlock(&dtrace_lock); + return -EINVAL; + } + } else { + if (copy_to_user(fmt.dtfd_string, str, len) != 0) { + mutex_unlock(&dtrace_lock); + return -EINVAL; + } + } + + mutex_unlock(&dtrace_lock); + + return 0; + } + + default: + break; + } + + return -ENOTTY; +} + +static int dtrace_open(struct inode *inode, struct file *file) +{ + dtrace_state_t *state; + uint32_t priv; + uid_t uid; + + dtrace_cred2priv(file->f_cred, &priv, &uid); + if (priv == DTRACE_PRIV_NONE) + return -EACCES; + + mutex_lock(&dtrace_provider_lock); + dtrace_probe_provide(NULL, NULL); + mutex_unlock(&dtrace_provider_lock); + + mutex_lock(&cpu_lock); + mutex_lock(&dtrace_lock); + dtrace_opens++; + dtrace_membar_producer(); + +#ifdef FIXME + /* + * Is this relevant for Linux? Is there an equivalent? + */ + if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) { + dtrace_opens--; + mutex_unlock(&cpu_lock); + mutex_unlock(&dtrace_lock); + return -EBUSY; + } +#endif + + state = dtrace_state_create(file); + mutex_unlock(&cpu_lock); + + if (state == NULL) { +#ifdef FIXME + if (--dtrace_opens == 0 && dtrace_anon.dta_enabling == NULL) + (void)kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); +#endif + + mutex_unlock(&dtrace_lock); + + return -EAGAIN; + } + + file->private_data = state; + + mutex_unlock(&dtrace_lock); + + return 0; +} + +static int dtrace_close(struct inode *inode, struct file *file) +{ + dtrace_state_t *state; + + mutex_lock(&cpu_lock); + mutex_lock(&dtrace_lock); + + /* + * If there is anonymous state, destroy that first. + */ + state = file->private_data; + if (state->dts_anon) { + ASSERT(dtrace_anon.dta_state == NULL); + + dtrace_state_destroy(state->dts_anon); + } + + dtrace_state_destroy(state); + ASSERT(dtrace_opens > 0); + +#ifdef FIXME + if (--dtrace_opens == 0 && dtrace_anon.dta_enabling == NULL) + (void)kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); +#else + --dtrace_opens; +#endif + + mutex_unlock(&dtrace_lock); + mutex_unlock(&cpu_lock); + + return 0; +} + +static const struct file_operations dtrace_fops = { + .owner = THIS_MODULE, + .unlocked_ioctl = dtrace_ioctl, + .open = dtrace_open, + .release = dtrace_close, +}; + +static struct miscdevice dtrace_dev = { + .minor = DT_DEV_DTRACE_MINOR, + .name = "dtrace", + .nodename = "dtrace/dtrace", + .fops = &dtrace_fops, +}; + +static void +dtrace_module_loaded(struct module *module) +{ + dtrace_provider_t *prv; + + mutex_lock(&dtrace_provider_lock); + /* FIXME: mutex_lock(&mod_lock); */ + + //ASSERT(ctl->mod_busy); + + /* + * We're going to call each providers per-module provide operation + * specifying only this module. + */ + for (prv = dtrace_provider; prv != NULL; prv = prv->dtpv_next) + prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, module); + + /* FIXME: mutex_unlock(&mod_lock); */ + mutex_unlock(&dtrace_provider_lock); + + /* + * If we have any retained enablings, we need to match against them. + * Enabling probes requires that cpu_lock be held, and we cannot hold + * cpu_lock here -- it is legal for cpu_lock to be held when loading a + * module. (In particular, this happens when loading scheduling + * classes.) So if we have any retained enablings, we need to dispatch + * our task queue to do the match for us. + */ + mutex_lock(&dtrace_lock); + + if (dtrace_retained == NULL) { + mutex_unlock(&dtrace_lock); + return; + } + +#ifdef FIXME + /* FIXME: maybe convert to a Linux workqueue */ + (void) taskq_dispatch(dtrace_taskq, + (task_func_t *)dtrace_enabling_matchall, NULL, TQ_SLEEP); +#else + dtrace_enabling_matchall(); +#endif + + mutex_unlock(&dtrace_lock); + + /* + * And now, for a little heuristic sleaze: in general, we want to + * match modules as soon as they load. However, we cannot guarantee + * this, because it would lead us to the lock ordering violation + * outlined above. The common case, of course, is that cpu_lock is + * _not_ held -- so we delay here for a clock tick, hoping that that's + * long enough for the task queue to do its work. If it's not, it's + * not a serious problem -- it just means that the module that we + * just loaded may not be immediately instrumentable. + */ + udelay(jiffies_to_usecs(1)); +} + +static void +dtrace_module_unloaded(struct module *module) +{ + dtrace_probe_t template, *probe, *first, *next; + dtrace_provider_t *prov; + + template.dtpr_mod = module->name; + + mutex_lock(&dtrace_provider_lock); + /* FIXME: mutex_lock(&mod_lock); */ + mutex_lock(&dtrace_lock); + + if (dtrace_bymod == NULL) { + /* + * The DTrace module is loaded (obviously) but not attached; + * we don't have any work to do. + */ + mutex_unlock(&dtrace_provider_lock); + /* FIXME: mutex_unlock(&mod_lock); */ + mutex_unlock(&dtrace_lock); + return; + } + + for (probe = first = dtrace_hash_lookup(dtrace_bymod, &template); + probe != NULL; probe = probe->dtpr_nextmod) { + if (probe->dtpr_ecb != NULL) { + mutex_unlock(&dtrace_provider_lock); + /* FIXME: mutex_unlock(&mod_lock); */ + mutex_unlock(&dtrace_lock); + + /* + * This shouldn't _actually_ be possible -- we're + * unloading a module that has an enabled probe in it. + * (It's normally up to the provider to make sure that + * this can't happen.) However, because dtps_enable() + * doesn't have a failure mode, there can be an + * enable/unload race. Upshot: we don't want to + * assert, but we're not going to disable the + * probe, either. + */ + if (dtrace_err_verbose) { + pr_warning("unloaded module '%s' had " + "enabled probes", module->name); + } + + return; + } + } + + probe = first; + + for (first = NULL; probe != NULL; probe = next) { +//TBD ASSERT(dtrace_probes[probe->dtpr_id - 1] == probe); + +//TBD dtrace_probes[probe->dtpr_id - 1] = NULL; + + next = probe->dtpr_nextmod; + dtrace_hash_remove(dtrace_bymod, probe); + dtrace_hash_remove(dtrace_byfunc, probe); + dtrace_hash_remove(dtrace_byname, probe); + + if (first == NULL) { + first = probe; + probe->dtpr_nextmod = NULL; + } else { + probe->dtpr_nextmod = first; + first = probe; + } + } + + /* + * We've removed all of the module's probes from the hash chains and + * from the probe array. Now issue a dtrace_sync() to be sure that + * everyone has cleared out from any probe array processing. + */ + dtrace_sync(); + + for (probe = first; probe != NULL; probe = first) { + first = probe->dtpr_nextmod; + prov = probe->dtpr_provider; + prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, probe->dtpr_id, + probe->dtpr_arg); + kfree(probe->dtpr_mod); + kfree(probe->dtpr_func); + kfree(probe->dtpr_name); +//FIXME vmem_free(dtrace_arena, (void *)(uintptr_t)probe->dtpr_id, 1); + kfree(probe); + } + + mutex_unlock(&dtrace_lock); + /* FIXME: mutex_unlock(&mod_lock); */ + mutex_unlock(&dtrace_provider_lock); +} + +/* + * Register a toxic range. + */ +static void dtrace_toxrange_add(uintptr_t base, uintptr_t limit) +{ + if (dtrace_toxranges >= dtrace_toxranges_max) { + int osize, nsize; + dtrace_toxrange_t *range; + + osize = dtrace_toxranges_max * sizeof(dtrace_toxrange_t); + + if (osize == 0) { + ASSERT(dtrace_toxrange == NULL); + ASSERT(dtrace_toxranges_max == 0); + + dtrace_toxranges_max = 1; + } else + dtrace_toxranges_max <<= 1; + + nsize = dtrace_toxranges_max * sizeof(dtrace_toxrange_t); + range = kzalloc(nsize, GFP_KERNEL); + + if (dtrace_toxrange != NULL) { + ASSERT(osize != 0); + + memcpy(range, dtrace_toxrange, osize); + kfree(dtrace_toxrange); + } + + dtrace_toxrange = range; + } + + ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_base == (uintptr_t)NULL); + ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_limit == (uintptr_t)NULL); + + dtrace_toxrange[dtrace_toxranges].dtt_base = base; + dtrace_toxrange[dtrace_toxranges].dtt_limit = limit; + dtrace_toxranges++; +} + +/* + * Check if an address falls within a toxic region. + */ +int dtrace_istoxic(uintptr_t kaddr, size_t size) +{ + uintptr_t taddr, tsize; + int i; + + for (i = 0; i < dtrace_toxranges; i++) { + taddr = dtrace_toxrange[i].dtt_base; + tsize = dtrace_toxrange[i].dtt_limit - taddr; + + if (kaddr - taddr < tsize) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[smp_processor_id()].cpuc_dtrace_illval = kaddr; + return 1; + } + + if (taddr - kaddr < size) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[smp_processor_id()].cpuc_dtrace_illval = taddr; + return 1; + } + } + + return 0; +} + +/* + * Initialize the DTrace core. + * + * Equivalent to: dtrace_attach() + */ +int dtrace_dev_init(void) +{ + dtrace_provider_id_t id; + int rc = 0; + + mutex_lock(&cpu_lock); + mutex_lock(&dtrace_provider_lock); + mutex_lock(&dtrace_lock); + + /* + * Register the device for the DTrace core. + */ + rc = misc_register(&dtrace_dev); + if (rc) { + pr_err("%s: Can't register misc device %d\n", + dtrace_dev.name, dtrace_dev.minor); + + mutex_unlock(&cpu_lock); + mutex_unlock(&dtrace_provider_lock); + mutex_unlock(&dtrace_lock); + + return rc; + } + + dtrace_modload = dtrace_module_loaded; + dtrace_modunload = dtrace_module_unloaded; +#ifdef FIXME + dtrace_cpu_init = dtrace_cpu_setup_initial; + dtrace_helpers_cleanup = dtrace_helpers_destroy; + dtrace_helpers_fork = dtrace_helpers_duplicate; + dtrace_cpustart_init = dtrace_suspend; + dtrace_cpustart_fini = dtrace_resume; + dtrace_debugger_init = dtrace_suspend; + dtrace_debugger_fini = dtrace_resume; + + register_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL); +#endif + + dtrace_probe_init(); + +#ifdef FIXME + dtrace_taskq = taskq_create("dtrace_taskq", 1, maxclsyspri, 1, INT_MAX, + 0); +#endif + + dtrace_state_cache = kmem_cache_create("dtrace_state_cache", + sizeof(dtrace_dstate_percpu_t) * NR_CPUS, + __alignof__(dtrace_dstate_percpu_t), + SLAB_PANIC, NULL); + + /* + * Create the probe hashtables. + */ + dtrace_bymod = dtrace_hash_create( + offsetof(dtrace_probe_t, dtpr_mod), + offsetof(dtrace_probe_t, dtpr_nextmod), + offsetof(dtrace_probe_t, dtpr_prevmod)); + dtrace_byfunc = dtrace_hash_create( + offsetof(dtrace_probe_t, dtpr_func), + offsetof(dtrace_probe_t, dtpr_nextfunc), + offsetof(dtrace_probe_t, dtpr_prevfunc)); + dtrace_byname = dtrace_hash_create( + offsetof(dtrace_probe_t, dtpr_name), + offsetof(dtrace_probe_t, dtpr_nextname), + offsetof(dtrace_probe_t, dtpr_prevname)); + + /* + * Ensure that the X configuration parameter has a legal value. + */ + if (dtrace_retain_max < 1) { + pr_warning("Illegal value (%lu) for dtrace_retain_max; " + "setting to 1", (unsigned long)dtrace_retain_max); + + dtrace_retain_max = 1; + } + + /* + * Discover our toxic ranges. + */ + dtrace_toxic_ranges(dtrace_toxrange_add); + + /* + * Register ourselves as a provider. + */ + dtrace_register("dtrace", &dtrace_provider_attr, DTRACE_PRIV_NONE, 0, + &dtrace_provider_ops, NULL, &id); + + ASSERT(dtrace_provider != NULL); + ASSERT((dtrace_provider_id_t)dtrace_provider == id); + + /* + * Create BEGIN, END, and ERROR probes. + */ + dtrace_probeid_begin = dtrace_probe_create( + (dtrace_provider_id_t)dtrace_provider, NULL, + NULL, "BEGIN", 0, NULL); + dtrace_probeid_end = dtrace_probe_create( + (dtrace_provider_id_t)dtrace_provider, NULL, + NULL, "END", 0, NULL); + dtrace_probeid_error = dtrace_probe_create( + (dtrace_provider_id_t)dtrace_provider, NULL, + NULL, "ERROR", 1, NULL); + + dtrace_anon_property(); + mutex_unlock(&cpu_lock); + + /* + * If DTrace helper tracing is enabled, we need to allocate a trace + * buffer. + */ + if (dtrace_helptrace_enabled) { + ASSERT(dtrace_helptrace_buffer == NULL); + + dtrace_helptrace_buffer = dtrace_vzalloc( + dtrace_helptrace_bufsize); + dtrace_helptrace_next = 0; + } + +#ifdef FIXME + /* + * There is usually code here to handle the case where there already + * are providers when we get to this code. On Linux, that does not + * seem to be possible since the DTrace core module (this code) is + * loaded as a dependency for each provider, and thus this + * initialization code is executed prior to the initialization code of + * the first provider causing the core to be loaded. + */ +#endif + + mutex_unlock(&dtrace_provider_lock); + mutex_unlock(&dtrace_lock); + + return 0; +} + +void dtrace_dev_exit(void) +{ + kmem_cache_destroy(dtrace_state_cache); + misc_deregister(&dtrace_dev); + + dtrace_probe_exit(); +} diff --git a/dtrace/dtrace_dev.h b/dtrace/dtrace_dev.h new file mode 100644 index 000000000000..4e614f3123eb --- /dev/null +++ b/dtrace/dtrace_dev.h @@ -0,0 +1,17 @@ +#ifndef _DTRACE_DEV_H_ +#define _DTRACE_DEV_H_ + +#define DT_DEV_DTRACE_MINOR (16) +#define DT_DEV_HELPER_MINOR (DT_DEV_DTRACE_MINOR + 1) +#define DT_DEV_PROFILE_MINOR (DT_DEV_HELPER_MINOR + 1) +#define DT_DEV_SYSTRACE_MINOR (DT_DEV_PROFILE_MINOR + 1) +#define DT_DEV_FBT_MINOR (DT_DEV_SYSTRACE_MINOR + 1) +#define DT_DEV_SDT_MINOR (DT_DEV_FBT_MINOR + 1) +#define DT_DEV_FASTTRAP_MINOR (DT_DEV_SDT_MINOR + 1) +#define DT_DEV_LOCKSTAT_MINOR (DT_DEV_FASTTRAP_MINOR + 1) +#define DT_DEV_DT_TEST_MINOR (DT_DEV_LOCKSTAT_MINOR + 1) + +extern int dtrace_dev_init(void); +extern void dtrace_dev_exit(void); + +#endif /* _DTRACE_DEV_H_ */ diff --git a/dtrace/dtrace_dif.c b/dtrace/dtrace_dif.c new file mode 100644 index 000000000000..2bb4c8420e78 --- /dev/null +++ b/dtrace/dtrace_dif.c @@ -0,0 +1,4473 @@ +/* + * FILE: dtrace_dif.c + * DESCRIPTION: Dynamic Tracing: DIF object functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include +#include +#include +#include +#include +#include + +#include "dtrace.h" + +size_t dtrace_global_maxsize = 16 * 1024; + +static uint64_t dtrace_vtime_references; + +static int dtrace_difo_err(uint_t pc, const char *format, ...) +{ + char buf[256]; + + if (dtrace_err_verbose) { + va_list alist; + size_t len = strlen(format); + + pr_err("dtrace DIF object error: [%u]: ", pc); + + if (len >= 256 - sizeof(KERN_ERR)) { + pr_err(""); + return 1; + } + + memcpy(buf, KERN_ERR, sizeof(KERN_ERR)); + memcpy(buf + sizeof(KERN_ERR), format, len); + + va_start(alist, format); + vprintk(buf, alist); + va_end(alist); + } + + return 1; +} + +/* + * Validate a DTrace DIF object by checking the IR instructions. The following + * rules are currently enforced by dtrace_difo_validate(): + * + * 1. Each instruction must have a valid opcode + * 2. Each register, string, variable, or subroutine reference must be valid + * 3. No instruction can modify register %r0 (must be zero) + * 4. All instruction reserved bits must be set to zero + * 5. The last instruction must be a "ret" instruction + * 6. All branch targets must reference a valid instruction _after_ the branch + */ +int dtrace_difo_validate(dtrace_difo_t *dp, dtrace_vstate_t *vstate, + uint_t nregs, const cred_t *cr) +{ + int err = 0, i; + int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err; + int kcheckload = 0; + uint_t pc; + + kcheckload = cr == NULL || + (vstate->dtvs_state->dts_cred.dcr_visible & + DTRACE_CRV_KERNEL) == 0; + + dp->dtdo_destructive = 0; + + for (pc = 0; pc < dp->dtdo_len && err == 0; pc++) { + dif_instr_t instr = dp->dtdo_buf[pc]; + uint_t r1 = DIF_INSTR_R1(instr); + uint_t r2 = DIF_INSTR_R2(instr); + uint_t rd = DIF_INSTR_RD(instr); + uint_t rs = DIF_INSTR_RS(instr); + uint_t label = DIF_INSTR_LABEL(instr); + uint_t v = DIF_INSTR_VAR(instr); + uint_t subr = DIF_INSTR_SUBR(instr); + uint_t type = DIF_INSTR_TYPE(instr); + uint_t op = DIF_INSTR_OP(instr); + + switch (op) { + case DIF_OP_OR: + case DIF_OP_XOR: + case DIF_OP_AND: + case DIF_OP_SLL: + case DIF_OP_SRL: + case DIF_OP_SRA: + case DIF_OP_SUB: + case DIF_OP_ADD: + case DIF_OP_MUL: + case DIF_OP_SDIV: + case DIF_OP_UDIV: + case DIF_OP_SREM: + case DIF_OP_UREM: + case DIF_OP_COPYS: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_NOT: + case DIF_OP_MOV: + case DIF_OP_ALLOCS: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_LDSB: + case DIF_OP_LDSH: + case DIF_OP_LDSW: + case DIF_OP_LDUB: + case DIF_OP_LDUH: + case DIF_OP_LDUW: + case DIF_OP_LDX: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + if (kcheckload) + dp->dtdo_buf[pc] = DIF_INSTR_LOAD( + op + DIF_OP_RLDSB - + DIF_OP_LDSB, + r1, rd); + break; + case DIF_OP_RLDSB: + case DIF_OP_RLDSH: + case DIF_OP_RLDSW: + case DIF_OP_RLDUB: + case DIF_OP_RLDUH: + case DIF_OP_RLDUW: + case DIF_OP_RLDX: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_ULDSB: + case DIF_OP_ULDSH: + case DIF_OP_ULDSW: + case DIF_OP_ULDUB: + case DIF_OP_ULDUH: + case DIF_OP_ULDUW: + case DIF_OP_ULDX: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_STB: + case DIF_OP_STH: + case DIF_OP_STW: + case DIF_OP_STX: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to 0 address\n"); + break; + case DIF_OP_CMP: + case DIF_OP_SCMP: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rd != 0) + err += efunc(pc, "non-zero reserved bits\n"); + break; + case DIF_OP_TST: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0 || rd != 0) + err += efunc(pc, "non-zero reserved bits\n"); + break; + case DIF_OP_BA: + case DIF_OP_BE: + case DIF_OP_BNE: + case DIF_OP_BG: + case DIF_OP_BGU: + case DIF_OP_BGE: + case DIF_OP_BGEU: + case DIF_OP_BL: + case DIF_OP_BLU: + case DIF_OP_BLE: + case DIF_OP_BLEU: + if (label >= dp->dtdo_len) + err += efunc(pc, "invalid branch target %u\n", + label); + if (label <= pc) + err += efunc(pc, "backward branch to %u\n", + label); + break; + case DIF_OP_RET: + if (r1 != 0 || r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + break; + case DIF_OP_NOP: + case DIF_OP_POPTS: + case DIF_OP_FLUSHTS: + if (r1 != 0 || r2 != 0 || rd != 0) + err += efunc(pc, "non-zero reserved bits\n"); + break; + case DIF_OP_SETX: + if (DIF_INSTR_INTEGER(instr) >= dp->dtdo_intlen) + err += efunc(pc, "invalid integer ref %u\n", + DIF_INSTR_INTEGER(instr)); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_SETS: + if (DIF_INSTR_STRING(instr) >= dp->dtdo_strlen) + err += efunc(pc, "invalid string ref %u\n", + DIF_INSTR_STRING(instr)); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_LDGA: + case DIF_OP_LDTA: + if (r1 > DIF_VAR_ARRAY_MAX) + err += efunc(pc, "invalid array %u\n", r1); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_LDGS: + case DIF_OP_LDTS: + case DIF_OP_LDLS: + case DIF_OP_LDGAA: + case DIF_OP_LDTAA: + if (v < DIF_VAR_OTHER_MIN || v > DIF_VAR_OTHER_MAX) + err += efunc(pc, "invalid variable %u\n", v); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_STGS: + case DIF_OP_STTS: + case DIF_OP_STLS: + case DIF_OP_STGAA: + case DIF_OP_STTAA: + if (v < DIF_VAR_OTHER_UBASE || v > DIF_VAR_OTHER_MAX) + err += efunc(pc, "invalid variable %u\n", v); + if (rs >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + break; + case DIF_OP_CALL: + if (subr > DIF_SUBR_MAX) + err += efunc(pc, "invalid subr %u\n", subr); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + + if (subr == DIF_SUBR_COPYOUT || + subr == DIF_SUBR_COPYOUTSTR) + dp->dtdo_destructive = 1; + break; + case DIF_OP_PUSHTR: + if (type != DIF_TYPE_STRING && type != DIF_TYPE_CTF) + err += efunc(pc, "invalid ref type %u\n", type); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rs >= nregs) + err += efunc(pc, "invalid register %u\n", rs); + break; + case DIF_OP_PUSHTV: + if (type != DIF_TYPE_CTF) + err += efunc(pc, "invalid val type %u\n", type); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rs >= nregs) + err += efunc(pc, "invalid register %u\n", rs); + break; + default: + err += efunc(pc, "invalid opcode %u\n", + DIF_INSTR_OP(instr)); + } + } + + if (dp->dtdo_len != 0 && + DIF_INSTR_OP(dp->dtdo_buf[dp->dtdo_len - 1]) != DIF_OP_RET) { + err += efunc(dp->dtdo_len - 1, + "expected 'ret' as last DIF instruction\n"); + } + + if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) { + /* + * If we're not returning by reference, the size must be either + * 0 or the size of one of the base types. + */ + switch (dp->dtdo_rtype.dtdt_size) { + case 0: + case sizeof(uint8_t): + case sizeof(uint16_t): + case sizeof(uint32_t): + case sizeof(uint64_t): + break; + + default: + err += efunc(dp->dtdo_len - 1, "bad return size\n"); + } + } + + for (i = 0; i < dp->dtdo_varlen && err == 0; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i], + *existing = NULL; + dtrace_diftype_t *vt, *et; + uint_t id, ndx; + + if (v->dtdv_scope != DIFV_SCOPE_GLOBAL && + v->dtdv_scope != DIFV_SCOPE_THREAD && + v->dtdv_scope != DIFV_SCOPE_LOCAL) { + err += efunc(i, "unrecognized variable scope %d\n", + v->dtdv_scope); + break; + } + + if (v->dtdv_kind != DIFV_KIND_ARRAY && + v->dtdv_kind != DIFV_KIND_SCALAR) { + err += efunc(i, "unrecognized variable type %d\n", + v->dtdv_kind); + break; + } + + if ((id = v->dtdv_id) > DIF_VARIABLE_MAX) { + err += efunc(i, "%d exceeds variable id limit\n", id); + break; + } + + if (id < DIF_VAR_OTHER_UBASE) + continue; + + /* + * For user-defined variables, we need to check that this + * definition is identical to any previous definition that we + * encountered. + */ + ndx = id - DIF_VAR_OTHER_UBASE; + + switch (v->dtdv_scope) { + case DIFV_SCOPE_GLOBAL: + if (ndx < vstate->dtvs_nglobals) { + dtrace_statvar_t *svar; + + if ((svar = vstate->dtvs_globals[ndx]) != NULL) + existing = &svar->dtsv_var; + } + + break; + + case DIFV_SCOPE_THREAD: + if (ndx < vstate->dtvs_ntlocals) + existing = &vstate->dtvs_tlocals[ndx]; + break; + + case DIFV_SCOPE_LOCAL: + if (ndx < vstate->dtvs_nlocals) { + dtrace_statvar_t *svar; + + if ((svar = vstate->dtvs_locals[ndx]) != NULL) + existing = &svar->dtsv_var; + } + + break; + } + + vt = &v->dtdv_type; + + if (vt->dtdt_flags & DIF_TF_BYREF) { + if (vt->dtdt_size == 0) { + err += efunc(i, "zero-sized variable\n"); + break; + } + + if (v->dtdv_scope == DIFV_SCOPE_GLOBAL && + vt->dtdt_size > dtrace_global_maxsize) { + err += efunc(i, "oversized by-ref global\n"); + break; + } + } + + if (existing == NULL || existing->dtdv_id == 0) + continue; + + ASSERT(existing->dtdv_id == v->dtdv_id); + ASSERT(existing->dtdv_scope == v->dtdv_scope); + + if (existing->dtdv_kind != v->dtdv_kind) + err += efunc(i, "%d changed variable kind\n", id); + + et = &existing->dtdv_type; + + if (vt->dtdt_flags != et->dtdt_flags) { + err += efunc(i, "%d changed variable type flags\n", id); + break; + } + + if (vt->dtdt_size != 0 && vt->dtdt_size != et->dtdt_size) { + err += efunc(i, "%d changed variable type size\n", id); + break; + } + } + + return err; +} + +/* + * Returns 1 if the expression in the DIF object can be cached on a per-thread + * basis; 0 if not. + */ +int dtrace_difo_cacheable(dtrace_difo_t *dp) +{ + int i; + + if (dp == NULL) + return 0; + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + + if (v->dtdv_scope != DIFV_SCOPE_GLOBAL) + continue; + + switch (v->dtdv_id) { + case DIF_VAR_CURTHREAD: + case DIF_VAR_PID: + case DIF_VAR_TID: + case DIF_VAR_EXECNAME: + case DIF_VAR_ZONENAME: + break; + + default: + return 0; + } + } + + /* + * This DIF object may be cacheable. Now we need to look for any + * array loading instructions, any memory loading instructions, or + * any stores to thread-local variables. + */ + for (i = 0; i < dp->dtdo_len; i++) { + uint_t op = DIF_INSTR_OP(dp->dtdo_buf[i]); + + if ((op >= DIF_OP_LDSB && op <= DIF_OP_LDX) || + (op >= DIF_OP_ULDSB && op <= DIF_OP_ULDX) || + (op >= DIF_OP_RLDSB && op <= DIF_OP_RLDX) || + op == DIF_OP_LDGA || op == DIF_OP_STTS) + return 0; + } + + return 1; +} + +/* + * This routine calculates the dynamic variable chunksize for a given DIF + * object. The calculation is not fool-proof, and can probably be tricked by + * malicious DIF -- but it works for all compiler-generated DIF. Because this + * calculation is likely imperfect, dtrace_dynvar() is able to gracefully fail + * if a dynamic variable size exceeds the chunksize. + */ +static void dtrace_difo_chunksize(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ + uint64_t sval; + dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* + thread + id */ + const dif_instr_t *text = dp->dtdo_buf; + uint_t pc, srd = 0; + uint_t ttop = 0; + size_t size, ksize; + uint_t id, i; + + for (pc = 0; pc < dp->dtdo_len; pc++) { + dif_instr_t instr = text[pc]; + uint_t op = DIF_INSTR_OP(instr); + uint_t rd = DIF_INSTR_RD(instr); + uint_t r1 = DIF_INSTR_R1(instr); + uint_t nkeys = 0; + uchar_t scope; + dtrace_key_t *key = tupregs; + + switch (op) { + case DIF_OP_SETX: + sval = dp->dtdo_inttab[DIF_INSTR_INTEGER(instr)]; + srd = rd; + continue; + + case DIF_OP_STTS: + key = &tupregs[DIF_DTR_NREGS]; + key[0].dttk_size = 0; + key[1].dttk_size = 0; + nkeys = 2; + scope = DIFV_SCOPE_THREAD; + break; + + case DIF_OP_STGAA: + case DIF_OP_STTAA: + nkeys = ttop; + + if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) + key[nkeys++].dttk_size = 0; + + key[nkeys++].dttk_size = 0; + + if (op == DIF_OP_STTAA) + scope = DIFV_SCOPE_THREAD; + else + scope = DIFV_SCOPE_GLOBAL; + + break; + + case DIF_OP_PUSHTR: + if (ttop == DIF_DTR_NREGS) + return; + + /* + * If the register for the size of the "pushtr" is %r0 + * (or the value is 0) and the type is a string, we'll + * use the system-wide default string size. + */ + if ((srd == 0 || sval == 0) && r1 == DIF_TYPE_STRING) + tupregs[ttop++].dttk_size = + dtrace_strsize_default; + else { + if (srd == 0) + return; + + tupregs[ttop++].dttk_size = sval; + } + + break; + + case DIF_OP_PUSHTV: + if (ttop == DIF_DTR_NREGS) + return; + + tupregs[ttop++].dttk_size = 0; + break; + + case DIF_OP_FLUSHTS: + ttop = 0; + break; + + case DIF_OP_POPTS: + if (ttop != 0) + ttop--; + break; + } + + sval = 0; + srd = 0; + + if (nkeys == 0) + continue; + + /* + * We have a dynamic variable allocation; calculate its size. + */ + for (ksize = 0, i = 0; i < nkeys; i++) + ksize += P2ROUNDUP(key[i].dttk_size, sizeof(uint64_t)); + + size = sizeof(dtrace_dynvar_t); + size += sizeof(dtrace_key_t) * (nkeys - 1); + size += ksize; + + /* + * Now we need to determine the size of the stored data. + */ + id = DIF_INSTR_VAR(instr); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + + if (v->dtdv_id == id && v->dtdv_scope == scope) { + size += v->dtdv_type.dtdt_size; + break; + } + } + + if (i == dp->dtdo_varlen) + return; + + /* + * We have the size. If this is larger than the chunk size + * for our dynamic variable state, reset the chunk size. + */ + size = P2ROUNDUP(size, sizeof(uint64_t)); + + if (size > vstate->dtvs_dynvars.dtds_chunksize) + vstate->dtvs_dynvars.dtds_chunksize = size; + } +} + +void dtrace_difo_hold(dtrace_difo_t *dp) +{ + int i; + + dp->dtdo_refcnt++; + ASSERT(dp->dtdo_refcnt != 0); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + + if (v->dtdv_id != DIF_VAR_VTIMESTAMP) + continue; + + if (dtrace_vtime_references++ == 0) + dtrace_vtime_enable(); + } +} + +void dtrace_difo_init(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ + int i, oldsvars, osz, nsz, otlocals, ntlocals; + uint_t id; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(dp->dtdo_buf != NULL && dp->dtdo_len != 0); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + dtrace_statvar_t *svar, ***svarp; + size_t dsize = 0; + uint8_t scope = v->dtdv_scope; + int *np; + + if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE) + continue; + + id -= DIF_VAR_OTHER_UBASE; + + switch (scope) { + case DIFV_SCOPE_THREAD: + while (id >= (otlocals = vstate->dtvs_ntlocals)) { + dtrace_difv_t *tlocals; + + if ((ntlocals = (otlocals << 1)) == 0) + ntlocals = 1; + + osz = otlocals * sizeof(dtrace_difv_t); + nsz = ntlocals * sizeof(dtrace_difv_t); + + tlocals = kzalloc(nsz, GFP_KERNEL); + + if (osz != 0) { + memcpy(tlocals, vstate->dtvs_tlocals, + osz); + kfree(vstate->dtvs_tlocals); + } + + vstate->dtvs_tlocals = tlocals; + vstate->dtvs_ntlocals = ntlocals; + } + + vstate->dtvs_tlocals[id] = *v; + continue; + + case DIFV_SCOPE_LOCAL: + np = &vstate->dtvs_nlocals; + svarp = &vstate->dtvs_locals; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) + dsize = NR_CPUS * + (v->dtdv_type.dtdt_size + + sizeof(uint64_t)); + else + dsize = NR_CPUS * sizeof(uint64_t); + + break; + + case DIFV_SCOPE_GLOBAL: + np = &vstate->dtvs_nglobals; + svarp = &vstate->dtvs_globals; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) + dsize = v->dtdv_type.dtdt_size + + sizeof(uint64_t); + + break; + + default: + ASSERT(0); + continue; /* not reached */ + } + + while (id >= (oldsvars = *np)) { + dtrace_statvar_t **statics; + int newsvars, oldsize, newsize; + + if ((newsvars = (oldsvars << 1)) == 0) + newsvars = 1; + + oldsize = oldsvars * sizeof(dtrace_statvar_t *); + newsize = newsvars * sizeof(dtrace_statvar_t *); + + statics = kzalloc(newsize, GFP_KERNEL); + + if (oldsize != 0) { + memcpy(statics, *svarp, oldsize); + kfree(*svarp); + } + + *svarp = statics; + *np = newsvars; + } + + if ((svar = (*svarp)[id]) == NULL) { + svar = kzalloc(sizeof(dtrace_statvar_t), GFP_KERNEL); + svar->dtsv_var = *v; + + if ((svar->dtsv_size = dsize) != 0) { + svar->dtsv_data = + (uint64_t)(uintptr_t)kzalloc( + dsize, GFP_KERNEL); + } + + (*svarp)[id] = svar; + } + + svar->dtsv_refcnt++; + } + + dtrace_difo_chunksize(dp, vstate); + dtrace_difo_hold(dp); +} + +void dtrace_difo_destroy(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ + int i; + + ASSERT(dp->dtdo_refcnt == 0); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + dtrace_statvar_t *svar, **svarp; + uint_t id; + uint8_t scope = v->dtdv_scope; + int *np; + + switch (scope) { + case DIFV_SCOPE_THREAD: + continue; + + case DIFV_SCOPE_LOCAL: + np = &vstate->dtvs_nlocals; + svarp = vstate->dtvs_locals; + break; + + case DIFV_SCOPE_GLOBAL: + np = &vstate->dtvs_nglobals; + svarp = vstate->dtvs_globals; + break; + + default: + BUG(); + } + + if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE) + continue; + + id -= DIF_VAR_OTHER_UBASE; + ASSERT(id < *np); + + svar = svarp[id]; + ASSERT(svar != NULL); + ASSERT(svar->dtsv_refcnt > 0); + + if (--svar->dtsv_refcnt > 0) + continue; + + if (svar->dtsv_size != 0) { + ASSERT((void *)(uintptr_t)svar->dtsv_data != NULL); + kfree((void *)(uintptr_t)svar->dtsv_data); + } + + kfree(svar); + svarp[id] = NULL; + } + + kfree(dp->dtdo_buf); + kfree(dp->dtdo_inttab); + kfree(dp->dtdo_strtab); + kfree(dp->dtdo_vartab); + kfree(dp); +} + +void dtrace_difo_release(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ + int i; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(dp->dtdo_refcnt != 0); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + + if (v->dtdv_id != DIF_VAR_VTIMESTAMP) + continue; + + ASSERT(dtrace_vtime_references > 0); + + if (--dtrace_vtime_references == 0) + dtrace_vtime_disable(); + } + + if (--dp->dtdo_refcnt == 0) + dtrace_difo_destroy(dp, vstate); +} + +/* + * The key for a thread-local variable consists of the lower 63 bits of the + * task pid, prefixed by a bit indicating whether an interrupt is active (1) or + * not (0). + * We add DIF_VARIABLE_MAX to the pid to assure that the thread key is never + * equal to a variable identifier. This is necessary (but not sufficient) to + * assure that global associative arrays never collide with thread-local + * variables. To guarantee that they cannot collide, we must also define the + * order for keying dynamic variables. That order is: + * + * [ key0 ] ... [ keyn ] [ variable-key ] [ tls-key ] + * + * Because the variable-key and the tls-key are in orthogonal spaces, there is + * no way for a global variable key signature to match a thread-local key + * signature. + */ +#define DTRACE_TLS_THRKEY(where) \ + { \ + uint_t intr = in_interrupt() ? 1 : 0; \ + \ + (where) = ((current->pid + DIF_VARIABLE_MAX) & \ + (((uint64_t)1 << 63) - 1)) | \ + ((uint64_t)intr << 63); \ + } + +#ifndef FIXME +# define DTRACE_ALIGNCHECK(addr, size, flags) +#endif + +/* + * Test whether a range of memory starting at testaddr of size testsz falls + * within the range of memory described by addr, sz. We take care to avoid + * problems with overflow and underflow of the unsigned quantities, and + * disallow all negative sizes. Ranges of size 0 are allowed. + */ +#define DTRACE_INRANGE(testaddr, testsz, baseaddr, basesz) \ + ((testaddr) - (baseaddr) < (basesz) && \ + (testaddr) + (testsz) - (baseaddr) <= (basesz) && \ + (testaddr) + (testsz) >= (testaddr)) + +#define DTRACE_LOADFUNC(bits) \ + uint##bits##_t dtrace_load##bits(uintptr_t addr) \ + { \ + size_t size = bits / NBBY; \ + uint##bits##_t rval; \ + int i; \ + int cpu = smp_processor_id(); \ + volatile uint16_t *flags = (volatile uint16_t *) \ + &cpu_core[cpu].cpuc_dtrace_flags; \ + \ + /* \ + * Deviation from the OpenSolaris code... Protect \ + * against dereferencing the NULL pointer since that \ + * really causes us a lot of grief (crash). \ + */ \ + if (addr == 0) { \ + *flags |= CPU_DTRACE_BADADDR; \ + cpu_core[cpu].cpuc_dtrace_illval = addr; \ + return 0; \ + } \ + \ + DTRACE_ALIGNCHECK(addr, size, flags); \ + \ + for (i = 0; i < dtrace_toxranges; i++) { \ + if (addr >= dtrace_toxrange[i].dtt_limit) \ + continue; \ + \ + if (addr + size <= dtrace_toxrange[i].dtt_base) \ + continue; \ + \ + /* \ + * This address falls within a toxic region. \ + */ \ + *flags |= CPU_DTRACE_BADADDR; \ + cpu_core[cpu].cpuc_dtrace_illval = addr; \ + return 0; \ + } \ + \ + *flags |= CPU_DTRACE_NOFAULT; \ + rval = *((volatile uint##bits##_t *)addr); \ + *flags &= ~CPU_DTRACE_NOFAULT; \ + \ + return !(*flags & CPU_DTRACE_FAULT) ? rval : 0; \ + } + +#ifdef CONFIG_64BIT +# define dtrace_loadptr dtrace_load64 +#else +# define dtrace_loadptr dtrace_load32 +#endif + +/* + * Use the DTRACE_LOADFUNC macro to define functions for each of loading a + * uint8_t, a uint16_t, a uint32_t and a uint64_t. + */ +DTRACE_LOADFUNC(8) +DTRACE_LOADFUNC(16) +DTRACE_LOADFUNC(32) +DTRACE_LOADFUNC(64) + +#define DT_BSWAP_8(x) ((x) & 0xff) +#define DT_BSWAP_16(x) ((DT_BSWAP_8(x) << 8) | DT_BSWAP_8((x) >> 8)) +#define DT_BSWAP_32(x) ((DT_BSWAP_16(x) << 16) | DT_BSWAP_16((x) >> 16)) +#define DT_BSWAP_64(x) ((DT_BSWAP_32(x) << 32) | DT_BSWAP_32((x) >> 32)) + +static int dtrace_inscratch(uintptr_t dest, size_t size, + dtrace_mstate_t *mstate) +{ + if (dest < mstate->dtms_scratch_base) + return 0; + + if (dest + size < dest) + return 0; + + if (dest + size > mstate->dtms_scratch_ptr) + return 0; + + return 1; +} + +static int dtrace_canstore_statvar(uint64_t addr, size_t sz, + dtrace_statvar_t **svars, int nsvars) +{ + int i; + + for (i = 0; i < nsvars; i++) { + dtrace_statvar_t *svar = svars[i]; + + if (svar == NULL || svar->dtsv_size == 0) + continue; + + if (DTRACE_INRANGE(addr, sz, svar->dtsv_data, svar->dtsv_size)) + return 1; + } + + return 0; +} + +/* + * Check to see if the address is within a memory region to which a store may + * be issued. This includes the DTrace scratch areas, and any DTrace variable + * region. The caller of dtrace_canstore() is responsible for performing any + * alignment checks that are needed before stores are actually executed. + */ +static int dtrace_canstore(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate) +{ + /* + * First, check to see if the address is in scratch space... + */ + if (DTRACE_INRANGE(addr, sz, mstate->dtms_scratch_base, + mstate->dtms_scratch_size)) + return 1; + + /* + * Now check to see if it's a dynamic variable. This check will pick + * up both thread-local variables and any global dynamically-allocated + * variables. + */ + if (DTRACE_INRANGE(addr, sz, (uintptr_t)vstate->dtvs_dynvars.dtds_base, + vstate->dtvs_dynvars.dtds_size)) { + dtrace_dstate_t *dstate = &vstate->dtvs_dynvars; + uintptr_t base = (uintptr_t)dstate->dtds_base + + (dstate->dtds_hashsize * + sizeof(dtrace_dynhash_t)); + uintptr_t chunkoffs; + uint64_t num; + + /* + * Before we assume that we can store here, we need to make + * sure that it isn't in our metadata -- storing to our + * dynamic variable metadata would corrupt our state. For + * the range to not include any dynamic variable metadata, + * it must: + * + * (1) Start above the hash table that is at the base of + * the dynamic variable space + * + * (2) Have a starting chunk offset that is beyond the + * dtrace_dynvar_t that is at the base of every chunk + * + * (3) Not span a chunk boundary + */ + if (addr < base) + return 0; + + num = addr - base; + chunkoffs = do_div(num, dstate->dtds_chunksize); + + if (chunkoffs < sizeof(dtrace_dynvar_t)) + return 0; + + if (chunkoffs + sz > dstate->dtds_chunksize) + return 0; + + return 1; + } + + /* + * Finally, check the static local and global variables. These checks + * take the longest, so we perform them last. + */ + if (dtrace_canstore_statvar(addr, sz, vstate->dtvs_locals, + vstate->dtvs_nlocals)) + return 1; + + if (dtrace_canstore_statvar(addr, sz, vstate->dtvs_globals, + vstate->dtvs_nglobals)) + return 1; + + return 0; +} + +/* + * Convenience routine to check to see if the address is within a memory + * region in which a load may be issued given the user's privilege level; + * if not, it sets the appropriate error flags and loads 'addr' into the + * illegal value slot. + * + * DTrace subroutines (DIF_SUBR_*) should use this helper to implement + * appropriate memory access protection. + */ +static int +dtrace_canload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate) +{ + int cpu = smp_processor_id(); + volatile uintptr_t *illval = &cpu_core[cpu].cpuc_dtrace_illval; + + /* + * If we hold the privilege to read from kernel memory, then + * everything is readable. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) + return 1; + + /* + * You can obviously read that which you can store. + */ + if (dtrace_canstore(addr, sz, mstate, vstate)) + return 1; + + /* + * We're allowed to read from our own string table. + */ + if (DTRACE_INRANGE(addr, sz, (uintptr_t)mstate->dtms_difo->dtdo_strtab, + mstate->dtms_difo->dtdo_strlen)) + return 1; + + DTRACE_CPUFLAG_SET(CPU_DTRACE_KPRIV); + *illval = addr; + + return 0; +} + +/* + * Convenience routine to check to see if a given string is within a memory + * region in which a load may be issued given the user's privilege level; + * this exists so that we don't need to issue unnecessary dtrace_strlen() + * calls in the event that the user has all privileges. + */ +static int +dtrace_strcanload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate) +{ + size_t strsz; + + /* + * If we hold the privilege to read from kernel memory, then + * everything is readable. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) + return 1; + + strsz = 1 + dtrace_strlen((char *)(uintptr_t)addr, sz); + if (dtrace_canload(addr, strsz, mstate, vstate)) + return 1; + + return 0; +} + +/* + * Convenience routine to check to see if a given variable is within a memory + * region in which a load may be issued given the user's privilege level. + */ +int dtrace_vcanload(void *src, dtrace_diftype_t *type, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate) +{ + size_t sz; + + ASSERT(type->dtdt_flags & DIF_TF_BYREF); + + /* + * If we hold the privilege to read from kernel memory, then + * everything is readable. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) + return 1; + + if (type->dtdt_kind == DIF_TYPE_STRING) + sz = dtrace_strlen( + src, + vstate->dtvs_state->dts_options[DTRACEOPT_STRSIZE] + ) + 1; + else + sz = type->dtdt_size; + + return dtrace_canload((uintptr_t)src, sz, mstate, vstate); +} + +/* + * Copy src to dst using safe memory accesses. The src is assumed to be unsafe + * memory specified by the DIF program. The dst is assumed to be safe memory + * that we can store to directly because it is managed by DTrace. As with + * standard bcopy, overlapping copies are handled properly. + */ +static void dtrace_bcopy(const void *src, void *dst, size_t len) +{ + if (len != 0) { + uint8_t *s1 = dst; + const uint8_t *s2 = src; + + if (s1 <= s2) { + do { + *s1++ = dtrace_load8((uintptr_t)s2++); + } while (--len != 0); + } else { + s2 += len; + s1 += len; + + do { + *--s1 = dtrace_load8((uintptr_t)--s2); + } while (--len != 0); + } + } +} + +/* + * Copy src to dst using safe memory accesses, up to either the specified + * length, or the point that a nul byte is encountered. The src is assumed to + * be unsafe memory specified by the DIF program. The dst is assumed to be + * safe memory that we can store to directly because it is managed by DTrace. + * Unlike dtrace_bcopy(), overlapping regions are not handled. + */ +static void dtrace_strcpy(const void *src, void *dst, size_t len) +{ + if (len != 0) { + uint8_t *s1 = dst, c; + const uint8_t *s2 = src; + + do { + *s1++ = c = dtrace_load8((uintptr_t)s2++); + } while (--len != 0 && c != '\0'); + } +} +/* + * Copy src to dst, deriving the size and type from the specified (BYREF) + * variable type. The src is assumed to be unsafe memory specified by the DIF + * program. The dst is assumed to be DTrace variable memory that is of the + * specified type; we assume that we can store to directly. + */ +static void dtrace_vcopy(void *src, void *dst, dtrace_diftype_t *type) +{ + ASSERT(type->dtdt_flags & DIF_TF_BYREF); + + if (type->dtdt_kind == DIF_TYPE_STRING) + dtrace_strcpy(src, dst, type->dtdt_size); + else + dtrace_bcopy(src, dst, type->dtdt_size); +} + +/* + * Compare s1 to s2 using safe memory accesses. The s1 data is assumed to be + * unsafe memory specified by the DIF program. The s2 data is assumed to be + * safe memory that we can access directly because it is managed by DTrace. + */ +static int dtrace_bcmp(const void *s1, const void *s2, size_t len) +{ + volatile uint16_t *flags; + + flags = (volatile uint16_t *)&cpu_core[ + smp_processor_id()].cpuc_dtrace_flags; + + if (s1 == s2) + return 0; + + if (s1 == NULL || s2 == NULL) + return 1; + + if (s1 != s2 && len != 0) { + const uint8_t *ps1 = s1; + const uint8_t *ps2 = s2; + + do { + if (dtrace_load8((uintptr_t)ps1++) != *ps2++) + return 1; + } while (--len != 0 && !(*flags & CPU_DTRACE_FAULT)); + } + + return 0; +} + +/* + * Zero the specified region using a simple byte-by-byte loop. Note that this + * is for safe DTrace-managed memory only. + */ +void dtrace_bzero(void *dst, size_t len) +{ + uchar_t *cp; + + for (cp = dst; len != 0; len--) + *cp++ = 0; +} + +#define DTRACE_DYNHASH_FREE 0 +#define DTRACE_DYNHASH_SINK 1 +#define DTRACE_DYNHASH_VALID 2 + +/* + * Depending on the value of the op parameter, this function looks-up, + * allocates or deallocates an arbitrarily-keyed dynamic variable. If an + * allocation is requested, this function will return a pointer to a + * dtrace_dynvar_t corresponding to the allocated variable -- or NULL if no + * variable can be allocated. If NULL is returned, the appropriate counter + * will be incremented. + */ +static dtrace_dynvar_t *dtrace_dynvar(dtrace_dstate_t *dstate, uint_t nkeys, + dtrace_key_t *key, size_t dsize, + dtrace_dynvar_op_t op, + dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate) +{ + uint64_t hashval = DTRACE_DYNHASH_VALID; + dtrace_dynhash_t *hash = dstate->dtds_hash; + dtrace_dynvar_t *free, *new_free, *next, *dvar, *start, + *prev = NULL; + processorid_t me = smp_processor_id(), cpu = me; + dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[me]; + size_t bucket, ksize; + size_t chunksize = dstate->dtds_chunksize; + uintptr_t kdata, lock, nstate; + uint_t i; + + ASSERT(nkeys != 0); + + /* + * Hash the key. As with aggregations, we use Jenkins' "One-at-a-time" + * algorithm. For the by-value portions, we perform the algorithm in + * 16-bit chunks (as opposed to 8-bit chunks). This speeds things up a + * bit, and seems to have only a minute effect on distribution. For + * the by-reference data, we perform "One-at-a-time" iterating (safely) + * over each referenced byte. It's painful to do this, but it's much + * better than pathological hash distribution. The efficacy of the + * hashing algorithm (and a comparison with other algorithms) may be + * found by running the ::dtrace_dynstat MDB dcmd. + */ + for (i = 0; i < nkeys; i++) { + if (key[i].dttk_size == 0) { + uint64_t val = key[i].dttk_value; + + hashval += (val >> 48) & 0xffff; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + + hashval += (val >> 32) & 0xffff; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + + hashval += (val >> 16) & 0xffff; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + + hashval += val & 0xffff; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + } else { + /* + * This is incredibly painful, but it beats the hell + * out of the alternative. + */ + uint64_t j, size = key[i].dttk_size; + uintptr_t base = (uintptr_t)key[i].dttk_value; + + if (!dtrace_canload(base, size, mstate, vstate)) + break; + + for (j = 0; j < size; j++) { + hashval += dtrace_load8(base + j); + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + } + } + } + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT)) + return NULL; + + hashval += (hashval << 3); + hashval ^= (hashval >> 11); + hashval += (hashval << 15); + + /* + * There is a remote chance (ideally, 1 in 2^31) that our hashval + * comes out to be one of our two sentinel hash values. If this + * actually happens, we set the hashval to be a value known to be a + * non-sentinel value. + */ + if (hashval == DTRACE_DYNHASH_FREE || hashval == DTRACE_DYNHASH_SINK) + hashval = DTRACE_DYNHASH_VALID; + + /* + * Yes, it's painful to do a divide here. If the cycle count becomes + * important here, tricks can be pulled to reduce it. (However, it's + * critical that hash collisions be kept to an absolute minimum; + * they're much more painful than a divide.) It's better to have a + * solution that generates few collisions and still keeps things + * relatively simple. + * + * Linux cannot do a straight 64-bit divide without gcc requiring + * linking in code that the kernel doesn't link, so we need to use an + * alternative. + * + * bucket = hashval % dstate->dtds_hashsize; + */ + { + uint64_t num; + + num = hashval; + bucket = do_div(num, dstate->dtds_hashsize); + } + + if (op == DTRACE_DYNVAR_DEALLOC) { + volatile uintptr_t *lockp = &hash[bucket].dtdh_lock; + + for (;;) { + while ((lock = *lockp) & 1) + continue; + + if (cmpxchg(lockp, lock, (lock + 1)) == lock) + break; + } + + dtrace_membar_producer(); + } + +top: + prev = NULL; + lock = hash[bucket].dtdh_lock; + + dtrace_membar_consumer(); + + start = hash[bucket].dtdh_chain; + ASSERT(start != NULL && (start->dtdv_hashval == DTRACE_DYNHASH_SINK || + start->dtdv_hashval != DTRACE_DYNHASH_FREE || + op != DTRACE_DYNVAR_DEALLOC)); + + for (dvar = start; dvar != NULL; dvar = dvar->dtdv_next) { + dtrace_tuple_t *dtuple = &dvar->dtdv_tuple; + dtrace_key_t *dkey = &dtuple->dtt_key[0]; + + if (dvar->dtdv_hashval != hashval) { + if (dvar->dtdv_hashval == DTRACE_DYNHASH_SINK) { + /* + * We've reached the sink, and therefore the + * end of the hash chain; we can kick out of + * the loop knowing that we have seen a valid + * snapshot of state. + */ + ASSERT(dvar->dtdv_next == NULL); + ASSERT(dvar == &dtrace_dynhash_sink); + break; + } + + if (dvar->dtdv_hashval == DTRACE_DYNHASH_FREE) { + /* + * We've gone off the rails: somewhere along + * the line, one of the members of this hash + * chain was deleted. Note that we could also + * detect this by simply letting this loop run + * to completion, as we would eventually hit + * the end of the dirty list. However, we + * want to avoid running the length of the + * dirty list unnecessarily (it might be quite + * long), so we catch this as early as + * possible by detecting the hash marker. In + * this case, we simply set dvar to NULL and + * break; the conditional after the loop will + * send us back to top. + */ + dvar = NULL; + break; + } + + goto next; + } + + if (dtuple->dtt_nkeys != nkeys) + goto next; + + for (i = 0; i < nkeys; i++, dkey++) { + if (dkey->dttk_size != key[i].dttk_size) + goto next; /* size or type mismatch */ + + if (dkey->dttk_size != 0) { + if (dtrace_bcmp( + (void *)(uintptr_t)key[i].dttk_value, + (void *)(uintptr_t)dkey->dttk_value, + dkey->dttk_size)) + goto next; + } else { + if (dkey->dttk_value != key[i].dttk_value) + goto next; + } + } + + if (op != DTRACE_DYNVAR_DEALLOC) + return dvar; + + ASSERT(dvar->dtdv_next == NULL || + dvar->dtdv_next->dtdv_hashval != DTRACE_DYNHASH_FREE); + + if (prev != NULL) { + ASSERT(hash[bucket].dtdh_chain != dvar); + ASSERT(start != dvar); + ASSERT(prev->dtdv_next == dvar); + prev->dtdv_next = dvar->dtdv_next; + } else { + if (cmpxchg(&hash[bucket].dtdh_chain, start, + dvar->dtdv_next) != start) { + /* + * We have failed to atomically swing the + * hash table head pointer, presumably because + * of a conflicting allocation on another CPU. + * We need to reread the hash chain and try + * again. + */ + goto top; + } + } + + dtrace_membar_producer(); + + /* + * Now set the hash value to indicate that it's free. + */ + ASSERT(hash[bucket].dtdh_chain != dvar); + dvar->dtdv_hashval = DTRACE_DYNHASH_FREE; + + dtrace_membar_producer(); + + /* + * Set the next pointer to point at the dirty list, and + * atomically swing the dirty pointer to the newly freed dvar. + */ + do { + next = dcpu->dtdsc_dirty; + dvar->dtdv_next = next; + } while (cmpxchg(&dcpu->dtdsc_dirty, next, dvar) != next); + + /* + * Finally, unlock this hash bucket. + */ + ASSERT(hash[bucket].dtdh_lock == lock); + ASSERT(lock & 1); + hash[bucket].dtdh_lock++; + + return NULL; +next: + prev = dvar; + continue; + } + + if (dvar == NULL) { + /* + * If dvar is NULL, it is because we went off the rails: + * one of the elements that we traversed in the hash chain + * was deleted while we were traversing it. In this case, + * we assert that we aren't doing a dealloc (deallocs lock + * the hash bucket to prevent themselves from racing with + * one another), and retry the hash chain traversal. + */ + ASSERT(op != DTRACE_DYNVAR_DEALLOC); + goto top; + } + + if (op != DTRACE_DYNVAR_ALLOC) { + /* + * If we are not to allocate a new variable, we want to + * return NULL now. Before we return, check that the value + * of the lock word hasn't changed. If it has, we may have + * seen an inconsistent snapshot. + */ + if (op == DTRACE_DYNVAR_NOALLOC) { + if (hash[bucket].dtdh_lock != lock) + goto top; + } else { + ASSERT(op == DTRACE_DYNVAR_DEALLOC); + ASSERT(hash[bucket].dtdh_lock == lock); + ASSERT(lock & 1); + hash[bucket].dtdh_lock++; + } + + return NULL; + } + + /* + * We need to allocate a new dynamic variable. The size we need is the + * size of dtrace_dynvar plus the size of nkeys dtrace_key_t's plus the + * size of any auxiliary key data (rounded up to 8-byte alignment) plus + * the size of any referred-to data (dsize). We then round the final + * size up to the chunksize for allocation. + */ + for (ksize = 0, i = 0; i < nkeys; i++) + ksize += P2ROUNDUP(key[i].dttk_size, sizeof(uint64_t)); + + /* + * This should be pretty much impossible, but could happen if, say, + * strange DIF specified the tuple. Ideally, this should be an + * assertion and not an error condition -- but that requires that the + * chunksize calculation in dtrace_difo_chunksize() be absolutely + * bullet-proof. (That is, it must not be able to be fooled by + * malicious DIF.) Given the lack of backwards branches in DIF, + * solving this would presumably not amount to solving the Halting + * Problem -- but it still seems awfully hard. + */ + if (sizeof(dtrace_dynvar_t) + sizeof(dtrace_key_t) * (nkeys - 1) + + ksize + dsize > chunksize) { + dcpu->dtdsc_drops++; + return NULL; + } + + nstate = DTRACE_DSTATE_EMPTY; + + do { +retry: + free = dcpu->dtdsc_free; + + if (free == NULL) { + dtrace_dynvar_t *clean = dcpu->dtdsc_clean; + void *rval; + + if (clean == NULL) { + /* + * We're out of dynamic variable space on + * this CPU. Unless we have tried all CPUs, + * we'll try to allocate from a different + * CPU. + */ + switch (dstate->dtds_state) { + case DTRACE_DSTATE_CLEAN: { + uintptr_t *sp = + (uintptr_t *) + &dstate->dtds_state; + + if (++cpu >= NR_CPUS) + cpu = 0; + + if (dcpu->dtdsc_dirty != NULL && + nstate == DTRACE_DSTATE_EMPTY) + nstate = DTRACE_DSTATE_DIRTY; + + if (dcpu->dtdsc_rinsing != NULL) + nstate = DTRACE_DSTATE_RINSING; + + dcpu = &dstate->dtds_percpu[cpu]; + + if (cpu != me) + goto retry; + + cmpxchg(sp, DTRACE_DSTATE_CLEAN, + nstate); + + /* + * To increment the correct bean + * counter, take another lap. + */ + goto retry; + } + + case DTRACE_DSTATE_DIRTY: + dcpu->dtdsc_dirty_drops++; + break; + + case DTRACE_DSTATE_RINSING: + dcpu->dtdsc_rinsing_drops++; + break; + + case DTRACE_DSTATE_EMPTY: + dcpu->dtdsc_drops++; + break; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_DROP); + return NULL; + } + + /* + * The clean list appears to be non-empty. We want to + * move the clean list to the free list; we start by + * moving the clean pointer aside. + */ + if (cmpxchg(&dcpu->dtdsc_clean, clean, NULL) != clean) + /* + * We are in one of two situations: + * + * (a) The clean list was switched to the + * free list by another CPU. + * + * (b) The clean list was added to by the + * cleansing cyclic. + * + * In either of these situations, we can + * just reattempt the free list allocation. + */ + goto retry; + + ASSERT(clean->dtdv_hashval == DTRACE_DYNHASH_FREE); + + /* + * Now we'll move the clean list to the free list. + * It's impossible for this to fail: the only way + * the free list can be updated is through this + * code path, and only one CPU can own the clean list. + * Thus, it would only be possible for this to fail if + * this code were racing with dtrace_dynvar_clean(). + * (That is, if dtrace_dynvar_clean() updated the clean + * list, and we ended up racing to update the free + * list.) This race is prevented by the dtrace_sync() + * in dtrace_dynvar_clean() -- which flushes the + * owners of the clean lists out before resetting + * the clean lists. + */ + rval = cmpxchg(&dcpu->dtdsc_free, NULL, clean); + ASSERT(rval == NULL); + + goto retry; + } + + dvar = free; + new_free = dvar->dtdv_next; + } while (cmpxchg(&dcpu->dtdsc_free, free, new_free) != free); + + /* + * We have now allocated a new chunk. We copy the tuple keys into the + * tuple array and copy any referenced key data into the data space + * following the tuple array. As we do this, we relocate dttk_value + * in the final tuple to point to the key data address in the chunk. + */ + kdata = (uintptr_t)&dvar->dtdv_tuple.dtt_key[nkeys]; + dvar->dtdv_data = (void *)(kdata + ksize); + dvar->dtdv_tuple.dtt_nkeys = nkeys; + + for (i = 0; i < nkeys; i++) { + dtrace_key_t *dkey = &dvar->dtdv_tuple.dtt_key[i]; + size_t kesize = key[i].dttk_size; + + if (kesize != 0) { + dtrace_bcopy( + (const void *)(uintptr_t)key[i].dttk_value, + (void *)kdata, kesize); + dkey->dttk_value = kdata; + kdata += P2ROUNDUP(kesize, sizeof(uint64_t)); + } else + dkey->dttk_value = key[i].dttk_value; + + dkey->dttk_size = kesize; + } + + ASSERT(dvar->dtdv_hashval == DTRACE_DYNHASH_FREE); + dvar->dtdv_hashval = hashval; + dvar->dtdv_next = start; + + if (cmpxchg(&hash[bucket].dtdh_chain, start, dvar) == start) + return dvar; + + /* + * The cas has failed. Either another CPU is adding an element to + * this hash chain, or another CPU is deleting an element from this + * hash chain. The simplest way to deal with both of these cases + * (though not necessarily the most efficient) is to free our + * allocated block and tail-call ourselves. Note that the free is + * to the dirty list and _not_ to the free list. This is to prevent + * races with allocators, above. + */ + dvar->dtdv_hashval = DTRACE_DYNHASH_FREE; + + dtrace_membar_producer(); + + do { + free = dcpu->dtdsc_dirty; + dvar->dtdv_next = free; + } while (cmpxchg(&dcpu->dtdsc_dirty, free, dvar) != free); + + return dtrace_dynvar(dstate, nkeys, key, dsize, op, mstate, vstate); +} + +/* + * Return a string. In the event that the user lacks the privilege to access + * arbitrary kernel memory, we copy the string out to scratch memory so that we + * don't fail access checking. + * + * dtrace_dif_variable() uses this routine as a helper for various + * builtin values such as 'execname' and 'probefunc.' + */ +static uintptr_t dtrace_dif_varstr(uintptr_t addr, dtrace_state_t *state, + dtrace_mstate_t *mstate) +{ + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t ret; + size_t strsz; + + /* + * The easy case: this probe is allowed to read all of memory, so + * we can just return this as a vanilla pointer. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) + return addr; + + /* + * This is the tougher case: we copy the string in question from + * kernel memory into scratch memory and return it that way: this + * ensures that we won't trip up when access checking tests the + * BYREF return value. + */ + strsz = dtrace_strlen((char *)addr, size) + 1; + + if (mstate->dtms_scratch_ptr + strsz > + mstate->dtms_scratch_base + mstate->dtms_scratch_size) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + return (uintptr_t)NULL; + } + + dtrace_strcpy((const void *)addr, (void *)mstate->dtms_scratch_ptr, + strsz); + ret = mstate->dtms_scratch_ptr; + mstate->dtms_scratch_ptr += strsz; + + return ret; +} + +/* + * This function implements the DIF emulator's variable lookups. The emulator + * passes a reserved variable identifier and optional built-in array index. + */ +static uint64_t dtrace_dif_variable(dtrace_mstate_t *mstate, + dtrace_state_t *state, uint64_t v, + uint64_t ndx) +{ + /* + * If we're accessing one of the uncached arguments, we'll turn this + * into a reference in the args array. + */ + if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) { + ndx = v - DIF_VAR_ARG0; + v = DIF_VAR_ARGS; + } + + switch (v) { + case DIF_VAR_ARGS: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_ARGS); + + if (ndx >= + sizeof(mstate->dtms_arg) / sizeof(mstate->dtms_arg[0])) { + int aframes = + mstate->dtms_probe->dtpr_aframes + 2; + dtrace_provider_t *pv; + uint64_t val; + + pv = mstate->dtms_probe->dtpr_provider; + if (pv->dtpv_pops.dtps_getargval != NULL) + val = pv->dtpv_pops.dtps_getargval( + pv->dtpv_arg, + mstate->dtms_probe->dtpr_id, + mstate->dtms_probe->dtpr_arg, + ndx, aframes); + else + val = dtrace_getarg(ndx, aframes); + + /* + * This is regrettably required to keep the compiler + * from tail-optimizing the call to dtrace_getarg(). + * The condition always evaluates to true, but the + * compiler has no way of figuring that out a priori. + * (None of this would be necessary if the compiler + * could be relied upon to _always_ tail-optimize + * the call to dtrace_getarg() -- but it can't.) + */ + if (mstate->dtms_probe != NULL) + return val; + + ASSERT(0); + } + + return mstate->dtms_arg[ndx]; + + case DIF_VAR_UREGS: { + struct pt_regs *regs = task_pt_regs(current); + + if (!dtrace_priv_proc(state)) + return 0; + + return dtrace_getreg(regs, ndx); + } + + case DIF_VAR_CURTHREAD: + if (!dtrace_priv_kernel(state)) + return 0; + + return (uint64_t)(uintptr_t)current; + + case DIF_VAR_TIMESTAMP: + if (!(mstate->dtms_present & DTRACE_MSTATE_TIMESTAMP)) { + mstate->dtms_timestamp = dtrace_gethrtime(); + mstate->dtms_present |= DTRACE_MSTATE_TIMESTAMP; + } + + return ktime_to_ns(mstate->dtms_timestamp); + + case DIF_VAR_VTIMESTAMP: + ASSERT(dtrace_vtime_references != 0); + + return ktime_to_ns(current->dtrace_vtime); + + case DIF_VAR_WALLTIMESTAMP: + return ktime_to_ns(dtrace_getwalltime()); + + case DIF_VAR_IPL: + if (!dtrace_priv_kernel(state)) + return 0; + + if (!(mstate->dtms_present & DTRACE_MSTATE_IPL)) { + mstate->dtms_ipl = dtrace_getipl(); + mstate->dtms_present |= DTRACE_MSTATE_IPL; + } + + return mstate->dtms_ipl; + + case DIF_VAR_EPID: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_EPID); + ASSERT(mstate->dtms_present & DTRACE_MSTATE_EPID); + + return mstate->dtms_epid; + + case DIF_VAR_ID: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + return mstate->dtms_probe->dtpr_id; + + case DIF_VAR_STACKDEPTH: + if (!dtrace_priv_kernel(state)) + return 0; + if (!(mstate->dtms_present & DTRACE_MSTATE_STACKDEPTH)) { + int aframes = mstate->dtms_probe->dtpr_aframes + 2; + + mstate->dtms_stackdepth = dtrace_getstackdepth(aframes); + mstate->dtms_present |= DTRACE_MSTATE_STACKDEPTH; + } + + return mstate->dtms_stackdepth; + + case DIF_VAR_USTACKDEPTH: + if (!dtrace_priv_proc(state)) + return 0; + + if (!(mstate->dtms_present & DTRACE_MSTATE_USTACKDEPTH)) { + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && + in_interrupt()) + mstate->dtms_ustackdepth = 0; + else { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + mstate->dtms_ustackdepth = + dtrace_getustackdepth(); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + } + + mstate->dtms_present |= DTRACE_MSTATE_USTACKDEPTH; + } + + return mstate->dtms_ustackdepth; + + case DIF_VAR_CALLER: + if (!dtrace_priv_kernel(state)) + return 0; + + if (!(mstate->dtms_present & DTRACE_MSTATE_CALLER)) { + int aframes = mstate->dtms_probe->dtpr_aframes + 2; + + if (!DTRACE_ANCHORED(mstate->dtms_probe)) { + /* + * If this is an unanchored probe, we are + * required to go through the slow path: + * dtrace_caller() only guarantees correct + * results for anchored probes. + */ + uint64_t caller[2]; + + dtrace_getpcstack(caller, 2, aframes, + (uint32_t *)(uintptr_t) + mstate->dtms_arg[0]); + mstate->dtms_caller = caller[1]; + } else if ((mstate->dtms_caller = + dtrace_caller(aframes)) == -1) { + /* + * We have failed to do this the quick way; + * we must resort to the slower approach of + * calling dtrace_getpcstack(). + */ + uint64_t caller; + + dtrace_getpcstack(&caller, 1, aframes, NULL); + mstate->dtms_caller = caller; + } + + mstate->dtms_present |= DTRACE_MSTATE_CALLER; + } + + return mstate->dtms_caller; + + case DIF_VAR_UCALLER: + if (!dtrace_priv_proc(state)) + return 0; + + if (!(mstate->dtms_present & DTRACE_MSTATE_UCALLER)) { + uint64_t ustack[3]; + + /* + * dtrace_getupcstack() fills in the first uint64_t + * with the current PID. The second uint64_t will + * be the program counter at user-level. The third + * uint64_t will contain the caller, which is what + * we're after. + */ + ustack[2] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_getupcstack(ustack, 3); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + mstate->dtms_ucaller = ustack[2]; + mstate->dtms_present |= DTRACE_MSTATE_UCALLER; + } + + return mstate->dtms_ucaller; + + case DIF_VAR_PROBEPROV: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + + return dtrace_dif_varstr( + (uintptr_t)mstate->dtms_probe->dtpr_provider->dtpv_name, + state, mstate); + + case DIF_VAR_PROBEMOD: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + return dtrace_dif_varstr( + (uintptr_t)mstate->dtms_probe->dtpr_mod, state, + mstate); + + case DIF_VAR_PROBEFUNC: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + + return dtrace_dif_varstr( + (uintptr_t)mstate->dtms_probe->dtpr_func, state, + mstate); + + case DIF_VAR_PROBENAME: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + + return dtrace_dif_varstr( + (uintptr_t)mstate->dtms_probe->dtpr_name, state, + mstate); + + case DIF_VAR_PID: + if (!dtrace_priv_proc(state)) + return 0; + + /* + * Note that we are assuming that an unanchored probe is + * always due to a high-level interrupt. (And we're assuming + * that there is only a single high level interrupt.) + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && in_interrupt()) + return init_task.pid; + + /* + * It is always safe to dereference current, it always points + * to a valid task_struct. + */ + return (uint64_t)current->pid; + + case DIF_VAR_PPID: + if (!dtrace_priv_proc(state)) + return 0; + + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && in_interrupt()) + return init_task.real_parent->pid; + + /* + * It is always safe to dereference current, it always points + * to a valid task_struct. + * + * Additionally, it is safe to dereference one's parent, since + * it is never NULL after process birth. + */ + return (uint64_t)current->real_parent->pid; + + case DIF_VAR_TID: + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && in_interrupt()) + return init_task.pid; + + return (uint64_t)current->pid; + + case DIF_VAR_EXECNAME: + if (!dtrace_priv_proc(state)) + return 0; + + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && in_interrupt()) + return (uint64_t)(uintptr_t)init_task.comm; + + /* + * It is always safe to dereference current, it always points + * to a valid task_struct. + */ + return dtrace_dif_varstr((uintptr_t)current->comm, state, + mstate); + + case DIF_VAR_ZONENAME: + return 0; + + case DIF_VAR_UID: + if (!dtrace_priv_proc(state)) + return 0; + + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && in_interrupt()) + return (uint64_t)init_task.real_cred->uid; + + /* + * It is always safe to dereference current, it always points + * to a valid task_struct. + * + * Additionally, it is safe to dereference one's own process + * credential, since this is never NULL after process birth. + */ + return (uint64_t)current->real_cred->uid; + + case DIF_VAR_GID: + if (!dtrace_priv_proc(state)) + return 0; + + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && in_interrupt()) + return (uint64_t)init_task.real_cred->gid; + + /* + * It is always safe to dereference current, it always points + * to a valid task_struct. + * + * Additionally, it is safe to dereference one's own process + * credential, since this is never NULL after process birth. + */ + return (uint64_t)current->real_cred->gid; + + case DIF_VAR_ERRNO: + if (!dtrace_priv_proc(state)) + return 0; + + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && in_interrupt()) + return 0; + + /* + * It is always safe to dereference current, it always points + * to a valid task_struct. + */ + return (uint64_t)current->thread.error_code; + + default: + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return 0; + } +} + +#define DTRACE_V4MAPPED_OFFSET (sizeof (uint32_t) * 3) + +/* + * Emulate the execution of DTrace ID subroutines invoked by the call opcode. + * Notice that we don't bother validating the proper number of arguments or + * their types in the tuple stack. This isn't needed because all argument + * interpretation is safe because of our load safety -- the worst that can + * happen is that a bogus program can obtain bogus results. + */ +static void dtrace_dif_subr(uint_t subr, uint_t rd, uint64_t *regs, + dtrace_key_t *tupregs, int nargs, + dtrace_mstate_t *mstate, dtrace_state_t *state) +{ + volatile uint16_t *flags = &cpu_core[ + smp_processor_id() + ].cpuc_dtrace_flags; + volatile uintptr_t *illval = &cpu_core[ + smp_processor_id() + ].cpuc_dtrace_illval; + dtrace_vstate_t *vstate = &state->dts_vstate; + + union { + struct mutex mi; + uint64_t mx; + } m; + + union { + rwlock_t ri; + uintptr_t rw; + } r; + + switch (subr) { + case DIF_SUBR_RAND: + regs[rd] = ktime_to_ns(dtrace_gethrtime()) * 2416 + 374441; + regs[rd] = do_div(regs[rd], 1771875); + break; + + case DIF_SUBR_MUTEX_OWNED: + if (!dtrace_canload(tupregs[0].dttk_value, + sizeof(struct mutex), mstate, vstate)) { + regs[rd] = 0; + break; + } + + m.mx = dtrace_load64(tupregs[0].dttk_value); +#ifdef CONFIG_SMP + regs[rd] = m.mi.owner != NULL; +#else + regs[rd] = mutex_is_locked(&m.mi); +#endif + break; + + case DIF_SUBR_MUTEX_OWNER: + if (!dtrace_canload(tupregs[0].dttk_value, + sizeof(struct mutex), mstate, vstate)) { + regs[rd] = 0; + break; + } + + m.mx = dtrace_load64(tupregs[0].dttk_value); +#ifdef CONFIG_SMP + regs[rd] = (uintptr_t)m.mi.owner; +#else + regs[rd] = 0; +#endif + break; + + case DIF_SUBR_MUTEX_TYPE_ADAPTIVE: + if (!dtrace_canload(tupregs[0].dttk_value, + sizeof(struct mutex), mstate, vstate)) { + regs[rd] = 0; + break; + } + + m.mx = dtrace_load64(tupregs[0].dttk_value); + /* + * On Linux, all mutexes are adaptive. + */ + regs[rd] = 1; + break; + + case DIF_SUBR_MUTEX_TYPE_SPIN: + if (!dtrace_canload(tupregs[0].dttk_value, + sizeof(struct mutex), mstate, vstate)) { + regs[rd] = 0; + break; + } + + m.mx = dtrace_load64(tupregs[0].dttk_value); + /* + * On Linux, all mutexes are adaptive. + */ + regs[rd] = 0; + break; + + case DIF_SUBR_RW_READ_HELD: { + if (!dtrace_canload(tupregs[0].dttk_value, sizeof(uintptr_t), + mstate, vstate)) { + regs[rd] = 0; + break; + } + + r.rw = dtrace_loadptr(tupregs[0].dttk_value); + regs[rd] = !write_can_lock(&r.ri) && read_can_lock(&r.ri); + break; + } + + case DIF_SUBR_RW_WRITE_HELD: + if (!dtrace_canload(tupregs[0].dttk_value, sizeof(rwlock_t), + mstate, vstate)) { + regs[rd] = 0; + break; + } + + r.rw = dtrace_loadptr(tupregs[0].dttk_value); + regs[rd] = !write_can_lock(&r.ri); + break; + + case DIF_SUBR_RW_ISWRITER: + if (!dtrace_canload(tupregs[0].dttk_value, sizeof(rwlock_t), + mstate, vstate)) { + regs[rd] = 0; + break; + } + + r.rw = dtrace_loadptr(tupregs[0].dttk_value); + /* + * On Linux there is no way to determine whether someone is + * trying to acquire a write lock. + */ + regs[rd] = !write_can_lock(&r.ri); + break; + + case DIF_SUBR_BCOPY: { + /* + * We need to be sure that the destination is in the scratch + * region -- no other region is allowed. + */ + uintptr_t src = tupregs[0].dttk_value; + uintptr_t dest = tupregs[1].dttk_value; + size_t size = tupregs[2].dttk_value; + + if (!dtrace_inscratch(dest, size, mstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + + if (!dtrace_canload(src, size, mstate, vstate)) { + regs[rd] = 0; + break; + } + + dtrace_bcopy((void *)src, (void *)dest, size); + break; + } + + case DIF_SUBR_ALLOCA: + case DIF_SUBR_COPYIN: { + uintptr_t dest = P2ROUNDUP(mstate->dtms_scratch_ptr, 8); + uint64_t size = tupregs[ + subr == DIF_SUBR_ALLOCA ? 0 : 1 + ].dttk_value; + size_t scratch_size = (dest - + mstate->dtms_scratch_ptr) + + size; + + /* + * This action doesn't require any credential checks since + * probes will not activate in user contexts to which the + * enabling user does not have permissions. + */ + + /* + * Rounding up the user allocation size could have overflowed + * a large, bogus allocation (like -1ULL) to 0. + */ + if (scratch_size < size || + !DTRACE_INSCRATCH(mstate, scratch_size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + if (subr == DIF_SUBR_COPYIN) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyin(tupregs[0].dttk_value, dest, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + } + + mstate->dtms_scratch_ptr += scratch_size; + regs[rd] = dest; + break; + } + + case DIF_SUBR_COPYINTO: { + uint64_t size = tupregs[1].dttk_value; + uintptr_t dest = tupregs[2].dttk_value; + + /* + * This action doesn't require any credential checks since + * probes will not activate in user contexts to which the + * enabling user does not have permissions. + */ + if (!dtrace_inscratch(dest, size, mstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyin(tupregs[0].dttk_value, dest, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + break; + } + + case DIF_SUBR_COPYINSTR: { + uintptr_t dest = mstate->dtms_scratch_ptr; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + + if (nargs > 1 && tupregs[1].dttk_value < size) + size = tupregs[1].dttk_value + 1; + + /* + * This action doesn't require any credential checks since + * probes will not activate in user contexts to which the + * enabling user does not have permissions. + */ + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyinstr(tupregs[0].dttk_value, dest, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + ((char *)dest)[size - 1] = '\0'; + mstate->dtms_scratch_ptr += size; + regs[rd] = dest; + break; + } + +#if 0 /* FIXME */ + case DIF_SUBR_MSGSIZE: + case DIF_SUBR_MSGDSIZE: { + uintptr_t baddr = tupregs[0].dttk_value, daddr; + uintptr_t wptr, rptr; + size_t count = 0; + int cont = 0; + + while (baddr != NULL && !(*flags & CPU_DTRACE_FAULT)) { + + if (!dtrace_canload(baddr, sizeof(mblk_t), mstate, + vstate)) { + regs[rd] = 0; + break; + } + + wptr = dtrace_loadptr(baddr + + offsetof(mblk_t, b_wptr)); + + rptr = dtrace_loadptr(baddr + + offsetof(mblk_t, b_rptr)); + + if (wptr < rptr) { + *flags |= CPU_DTRACE_BADADDR; + *illval = tupregs[0].dttk_value; + break; + } + + daddr = dtrace_loadptr(baddr + + offsetof(mblk_t, b_datap)); + + baddr = dtrace_loadptr(baddr + + offsetof(mblk_t, b_cont)); + + /* + * We want to prevent against denial-of-service here, + * so we're only going to search the list for + * dtrace_msgdsize_max mblks. + */ + if (cont++ > dtrace_msgdsize_max) { + *flags |= CPU_DTRACE_ILLOP; + break; + } + + if (subr == DIF_SUBR_MSGDSIZE) { + if (dtrace_load8(daddr + + offsetof(dblk_t, db_type)) != M_DATA) + continue; + } + + count += wptr - rptr; + } + + if (!(*flags & CPU_DTRACE_FAULT)) + regs[rd] = count; + + break; + } +#endif + + case DIF_SUBR_PROGENYOF: { + pid_t pid = tupregs[0].dttk_value; + struct task_struct *p; + int rval = 0; + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + + for (p = current; p != NULL; p = p->real_parent) { + if (p->pid == pid) { + rval = 1; + break; + } + } + + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + regs[rd] = rval; + break; + } + + case DIF_SUBR_SPECULATION: + regs[rd] = dtrace_speculation(state); + break; + + case DIF_SUBR_COPYOUT: { + uintptr_t kaddr = tupregs[0].dttk_value; + uintptr_t uaddr = tupregs[1].dttk_value; + uint64_t size = tupregs[2].dttk_value; + + if (!dtrace_destructive_disallow && + dtrace_priv_proc_control(state) && + !dtrace_istoxic(kaddr, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyout(kaddr, uaddr, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + } + break; + } + + case DIF_SUBR_COPYOUTSTR: { + uintptr_t kaddr = tupregs[0].dttk_value; + uintptr_t uaddr = tupregs[1].dttk_value; + uint64_t size = tupregs[2].dttk_value; + + if (!dtrace_destructive_disallow && + dtrace_priv_proc_control(state) && + !dtrace_istoxic(kaddr, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyoutstr(kaddr, uaddr, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + } + break; + } + + case DIF_SUBR_STRLEN: { + size_t sz; + uintptr_t addr = (uintptr_t)tupregs[0].dttk_value; + + sz = dtrace_strlen((char *)addr, + state->dts_options[DTRACEOPT_STRSIZE]); + + if (!dtrace_canload(addr, sz + 1, mstate, vstate)) { + regs[rd] = 0; + break; + } + + regs[rd] = sz; + + break; + } + + case DIF_SUBR_STRCHR: + case DIF_SUBR_STRRCHR: { + /* + * We're going to iterate over the string looking for the + * specified character. We will iterate until we have reached + * the string length or we have found the character. If this + * is DIF_SUBR_STRRCHR, we will look for the last occurrence + * of the specified character instead of the first. + */ + uintptr_t saddr = tupregs[0].dttk_value; + uintptr_t addr = tupregs[0].dttk_value; + uintptr_t limit = addr + + state->dts_options[DTRACEOPT_STRSIZE]; + char c, target = (char)tupregs[1].dttk_value; + + for (regs[rd] = 0; addr < limit; addr++) { + if ((c = dtrace_load8(addr)) == target) { + regs[rd] = addr; + + if (subr == DIF_SUBR_STRCHR) + break; + } + + if (c == '\0') + break; + } + + if (!dtrace_canload(saddr, addr - saddr, mstate, vstate)) { + regs[rd] = 0; + break; + } + + break; + } + + case DIF_SUBR_STRSTR: + case DIF_SUBR_INDEX: + case DIF_SUBR_RINDEX: { + /* + * We're going to iterate over the string looking for the + * specified string. We will iterate until we have reached + * the string length or we have found the string. (Yes, this + * is done in the most naive way possible -- but considering + * that the string we're searching for is likely to be + * relatively short, the complexity of Rabin-Karp or similar + * hardly seems merited.) + */ + char *addr = (char *)(uintptr_t) + tupregs[0].dttk_value; + char *substr = (char *)(uintptr_t) + tupregs[1].dttk_value; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + size_t len = dtrace_strlen(addr, size); + size_t sublen = dtrace_strlen(substr, size); + char *limit = addr + len, *orig = addr; + int notfound = subr == DIF_SUBR_STRSTR ? 0 : -1; + int inc = 1; + + regs[rd] = notfound; + + if (!dtrace_canload((uintptr_t)addr, len + 1, mstate, vstate)) { + regs[rd] = 0; + break; + } + + if (!dtrace_canload((uintptr_t)substr, sublen + 1, mstate, + vstate)) { + regs[rd] = 0; + break; + } + + /* + * strstr() and index()/rindex() have similar semantics if + * both strings are the empty string: strstr() returns a + * pointer to the (empty) string, and index() and rindex() + * both return index 0 (regardless of any position argument). + */ + if (sublen == 0 && len == 0) { + if (subr == DIF_SUBR_STRSTR) + regs[rd] = (uintptr_t)addr; + else + regs[rd] = 0; + break; + } + + if (subr != DIF_SUBR_STRSTR) { + if (subr == DIF_SUBR_RINDEX) { + limit = orig - 1; + addr += len; + inc = -1; + } + + /* + * Both index() and rindex() take an optional position + * argument that denotes the starting position. + */ + if (nargs == 3) { + int64_t pos = (int64_t)tupregs[2].dttk_value; + + /* + * If the position argument to index() is + * negative, Perl implicitly clamps it at + * zero. This semantic is a little surprising + * given the special meaning of negative + * positions to similar Perl functions like + * substr(), but it appears to reflect a + * notion that index() can start from a + * negative index and increment its way up to + * the string. Given this notion, Perl's + * rindex() is at least self-consistent in + * that it implicitly clamps positions greater + * than the string length to be the string + * length. Where Perl completely loses + * coherence, however, is when the specified + * substring is the empty string (""). In + * this case, even if the position is + * negative, rindex() returns 0 -- and even if + * the position is greater than the length, + * index() returns the string length. These + * semantics violate the notion that index() + * should never return a value less than the + * specified position and that rindex() should + * never return a value greater than the + * specified position. (One assumes that + * these semantics are artifacts of Perl's + * implementation and not the results of + * deliberate design -- it beggars belief that + * even Larry Wall could desire such oddness.) + * While in the abstract one would wish for + * consistent position semantics across + * substr(), index() and rindex() -- or at the + * very least self-consistent position + * semantics for index() and rindex() -- we + * instead opt to keep with the extant Perl + * semantics, in all their broken glory. (Do + * we have more desire to maintain Perl's + * semantics than Perl does? Probably.) + */ + if (subr == DIF_SUBR_RINDEX) { + if (pos < 0) { + if (sublen == 0) + regs[rd] = 0; + break; + } + + if (pos > len) + pos = len; + } else { + if (pos < 0) + pos = 0; + + if (pos >= len) { + if (sublen == 0) + regs[rd] = len; + break; + } + } + + addr = orig + pos; + } + } + + for (regs[rd] = notfound; addr != limit; addr += inc) { + if (dtrace_strncmp(addr, substr, sublen) == 0) { + if (subr != DIF_SUBR_STRSTR) { + /* + * As D index() and rindex() are + * modeled on Perl (and not on awk), + * we return a zero-based (and not a + * one-based) index. (For you Perl + * weenies: no, we're not going to add + * $[ -- and shouldn't you be at a con + * or something?) + */ + regs[rd] = (uintptr_t)(addr - orig); + break; + } + + ASSERT(subr == DIF_SUBR_STRSTR); + regs[rd] = (uintptr_t)addr; + break; + } + } + + break; + } + + case DIF_SUBR_STRTOK: { + uintptr_t addr = tupregs[0].dttk_value; + uintptr_t tokaddr = tupregs[1].dttk_value; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t limit, toklimit = tokaddr + size; + uint8_t c, tokmap[32]; /* 256 / 8 */ + char *dest = (char *)mstate->dtms_scratch_ptr; + int i; + + /* + * Check both the token buffer and (later) the input buffer, + * since both could be non-scratch addresses. + */ + if (!dtrace_strcanload(tokaddr, size, mstate, vstate)) { + regs[rd] = 0; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + if (addr == (uintptr_t)NULL) { + /* + * If the address specified is NULL, we use our saved + * strtok pointer from the mstate. Note that this + * means that the saved strtok pointer is _only_ + * valid within multiple enablings of the same probe -- + * it behaves like an implicit clause-local variable. + */ + addr = mstate->dtms_strtok; + } else { + /* + * If the user-specified address is non-NULL we must + * access check it. This is the only time we have + * a chance to do so, since this address may reside + * in the string table of this clause-- future calls + * (when we fetch addr from mstate->dtms_strtok) + * would fail this access check. + */ + if (!dtrace_strcanload(addr, size, mstate, vstate)) { + regs[rd] = 0; + break; + } + } + + /* + * First, zero the token map, and then process the token + * string -- setting a bit in the map for every character + * found in the token string. + */ + for (i = 0; i < sizeof(tokmap); i++) + tokmap[i] = 0; + + for (; tokaddr < toklimit; tokaddr++) { + if ((c = dtrace_load8(tokaddr)) == '\0') + break; + + ASSERT((c >> 3) < sizeof(tokmap)); + tokmap[c >> 3] |= (1 << (c & 0x7)); + } + + for (limit = addr + size; addr < limit; addr++) { + /* + * We're looking for a character that is _not_ contained + * in the token string. + */ + if ((c = dtrace_load8(addr)) == '\0') + break; + + if (!(tokmap[c >> 3] & (1 << (c & 0x7)))) + break; + } + + if (c == '\0') { + /* + * We reached the end of the string without finding + * any character that was not in the token string. + * We return NULL in this case, and we set the saved + * address to NULL as well. + */ + regs[rd] = 0; + mstate->dtms_strtok = (uintptr_t)NULL; + break; + } + + /* + * From here on, we're copying into the destination string. + */ + for (i = 0; addr < limit && i < size - 1; addr++) { + if ((c = dtrace_load8(addr)) == '\0') + break; + + if (tokmap[c >> 3] & (1 << (c & 0x7))) + break; + + ASSERT(i < size); + dest[i++] = c; + } + + ASSERT(i < size); + dest[i] = '\0'; + regs[rd] = (uintptr_t)dest; + mstate->dtms_scratch_ptr += size; + mstate->dtms_strtok = addr; + break; + } + + case DIF_SUBR_SUBSTR: { + uintptr_t s = tupregs[0].dttk_value; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + char *d = (char *)mstate->dtms_scratch_ptr; + int64_t index = (int64_t)tupregs[1].dttk_value; + int64_t remaining = (int64_t)tupregs[2].dttk_value; + size_t len = dtrace_strlen((char *)s, size); + int64_t i = 0; + + if (!dtrace_canload(s, len + 1, mstate, vstate)) { + regs[rd] = 0; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + if (nargs <= 2) + remaining = (int64_t)size; + + if (index < 0) { + index += len; + + if (index < 0 && index + remaining > 0) { + remaining += index; + index = 0; + } + } + + if (index >= len || index < 0) { + remaining = 0; + } else if (remaining < 0) { + remaining += len - index; + } else if (index + remaining > size) { + remaining = size - index; + } + + for (i = 0; i < remaining; i++) { + if ((d[i] = dtrace_load8(s + index + i)) == '\0') + break; + } + + d[i] = '\0'; + + mstate->dtms_scratch_ptr += size; + regs[rd] = (uintptr_t)d; + break; + } + + case DIF_SUBR_GETMAJOR: + regs[rd] = MAJOR(tupregs[0].dttk_value); + break; + + case DIF_SUBR_GETMINOR: + regs[rd] = MINOR(tupregs[0].dttk_value); + break; + +#if 0 /* FIXME */ + case DIF_SUBR_DDI_PATHNAME: { + /* + * This one is a galactic mess. We are going to roughly + * emulate ddi_pathname(), but it's made more complicated + * by the fact that we (a) want to include the minor name and + * (b) must proceed iteratively instead of recursively. + */ + uintptr_t dest = mstate->dtms_scratch_ptr; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + char *start = (char *)dest, *end = start + size - 1; + uintptr_t daddr = tupregs[0].dttk_value; + int64_t minor = (int64_t)tupregs[1].dttk_value; + char *s; + int i, len, depth = 0; + + /* + * Due to all the pointer jumping we do and context we must + * rely upon, we just mandate that the user must have kernel + * read privileges to use this routine. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) == 0) { + *flags |= CPU_DTRACE_KPRIV; + *illval = daddr; + regs[rd] = 0; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + *end = '\0'; + + /* + * We want to have a name for the minor. In order to do this, + * we need to walk the minor list from the devinfo. We want + * to be sure that we don't infinitely walk a circular list, + * so we check for circularity by sending a scout pointer + * ahead two elements for every element that we iterate over; + * if the list is circular, these will ultimately point to the + * same element. You may recognize this little trick as the + * answer to a stupid interview question -- one that always + * seems to be asked by those who had to have it laboriously + * explained to them, and who can't even concisely describe + * the conditions under which one would be forced to resort to + * this technique. Needless to say, those conditions are + * found here -- and probably only here. Is this the only use + * of this infamous trick in shipping, production code? If it + * isn't, it probably should be... + */ + if (minor != -1) { + uintptr_t maddr = dtrace_loadptr(daddr + + offsetof(struct dev_info, devi_minor)); + + uintptr_t next = offsetof(struct ddi_minor_data, next); + uintptr_t name = offsetof(struct ddi_minor_data, + d_minor) + offsetof(struct ddi_minor, name); + uintptr_t dev = offsetof(struct ddi_minor_data, + d_minor) + offsetof(struct ddi_minor, dev); + uintptr_t scout; + + if (maddr != NULL) + scout = dtrace_loadptr(maddr + next); + + while (maddr != NULL && !(*flags & CPU_DTRACE_FAULT)) { + uint64_t m; +#ifdef _LP64 + m = dtrace_load64(maddr + dev) & MAXMIN64; +#else + m = dtrace_load32(maddr + dev) & MAXMIN; +#endif + if (m != minor) { + maddr = dtrace_loadptr(maddr + next); + + if (scout == NULL) + continue; + + scout = dtrace_loadptr(scout + next); + + if (scout == NULL) + continue; + + scout = dtrace_loadptr(scout + next); + + if (scout == NULL) + continue; + + if (scout == maddr) { + *flags |= CPU_DTRACE_ILLOP; + break; + } + + continue; + } + + /* + * We have the minor data. Now we need to + * copy the minor's name into the end of the + * pathname. + */ + s = (char *)dtrace_loadptr(maddr + name); + len = dtrace_strlen(s, size); + + if (*flags & CPU_DTRACE_FAULT) + break; + + if (len != 0) { + if ((end -= (len + 1)) < start) + break; + + *end = ':'; + } + + for (i = 1; i <= len; i++) + end[i] = dtrace_load8((uintptr_t)s++); + break; + } + } + + while (daddr != NULL && !(*flags & CPU_DTRACE_FAULT)) { + ddi_node_state_t devi_state; + + devi_state = dtrace_load32(daddr + + offsetof(struct dev_info, devi_node_state)); + + if (*flags & CPU_DTRACE_FAULT) + break; + + if (devi_state >= DS_INITIALIZED) { + s = (char *)dtrace_loadptr(daddr + + offsetof(struct dev_info, devi_addr)); + len = dtrace_strlen(s, size); + + if (*flags & CPU_DTRACE_FAULT) + break; + + if (len != 0) { + if ((end -= (len + 1)) < start) + break; + + *end = '@'; + } + + for (i = 1; i <= len; i++) + end[i] = dtrace_load8((uintptr_t)s++); + } + + /* + * Now for the node name... + */ + s = (char *)dtrace_loadptr(daddr + + offsetof(struct dev_info, devi_node_name)); + + daddr = dtrace_loadptr(daddr + + offsetof(struct dev_info, devi_parent)); + + /* + * If our parent is NULL (that is, if we're the root + * node), we're going to use the special path + * "devices". + */ + if (daddr == NULL) + s = "devices"; + + len = dtrace_strlen(s, size); + if (*flags & CPU_DTRACE_FAULT) + break; + + if ((end -= (len + 1)) < start) + break; + + for (i = 1; i <= len; i++) + end[i] = dtrace_load8((uintptr_t)s++); + *end = '/'; + + if (depth++ > dtrace_devdepth_max) { + *flags |= CPU_DTRACE_ILLOP; + break; + } + } + + if (end < start) + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + + if (daddr == NULL) { + regs[rd] = (uintptr_t)end; + mstate->dtms_scratch_ptr += size; + } + + break; + } +#endif + + case DIF_SUBR_STRJOIN: { + char *d = (char *)mstate->dtms_scratch_ptr; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t s1 = tupregs[0].dttk_value; + uintptr_t s2 = tupregs[1].dttk_value; + int i = 0; + + if (!dtrace_strcanload(s1, size, mstate, vstate) || + !dtrace_strcanload(s2, size, mstate, vstate)) { + regs[rd] = 0; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + for (;;) { + if (i >= size) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + if ((d[i++] = dtrace_load8(s1++)) == '\0') { + i--; + break; + } + } + + for (;;) { + if (i >= size) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + if ((d[i++] = dtrace_load8(s2++)) == '\0') + break; + } + + if (i < size) { + mstate->dtms_scratch_ptr += i; + regs[rd] = (uintptr_t)d; + } + + break; + } + + case DIF_SUBR_LLTOSTR: { + int64_t i = (int64_t)tupregs[0].dttk_value; + int64_t val = i < 0 ? i * -1 : i; + uint64_t size = 22; /* room for 2^64 in dec */ + char *end = (char *)mstate->dtms_scratch_ptr + size + - 1; + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + /* + * GCC on Linux introduces calls to functions that are not + * linked into the kernel image, so we need to use the do_div() + * function instead. It modifies the first argument in place + * (replaces it with the quotient), and returns the remainder. + * + * Was: + * for (*end-- = '\0'; val; val /= 10) + * *end-- = '0' + (val % 10); + */ + for (*end-- = '\0'; val; ) + *end-- = '0' + do_div(val, 10); + + if (i == 0) + *end-- = '0'; + + if (i < 0) + *end-- = '-'; + + regs[rd] = (uintptr_t)end + 1; + mstate->dtms_scratch_ptr += size; + break; + } + + case DIF_SUBR_HTONS: + case DIF_SUBR_NTOHS: +#ifdef _BIG_ENDIAN + regs[rd] = (uint16_t)tupregs[0].dttk_value; +#else + regs[rd] = DT_BSWAP_16((uint16_t)tupregs[0].dttk_value); +#endif + break; + + + case DIF_SUBR_HTONL: + case DIF_SUBR_NTOHL: +#ifdef _BIG_ENDIAN + regs[rd] = (uint32_t)tupregs[0].dttk_value; +#else + regs[rd] = DT_BSWAP_32((uint32_t)tupregs[0].dttk_value); +#endif + break; + + + case DIF_SUBR_HTONLL: + case DIF_SUBR_NTOHLL: +#ifdef _BIG_ENDIAN + regs[rd] = (uint64_t)tupregs[0].dttk_value; +#else + regs[rd] = DT_BSWAP_64((uint64_t)tupregs[0].dttk_value); +#endif + break; + + + case DIF_SUBR_DIRNAME: + case DIF_SUBR_BASENAME: { + char *dest = (char *)mstate->dtms_scratch_ptr; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t src = tupregs[0].dttk_value; + int i, j, len = dtrace_strlen((char *)src, size); + int lastbase = -1, firstbase = -1, lastdir = -1; + int start, end; + + if (!dtrace_canload(src, len + 1, mstate, vstate)) { + regs[rd] = 0; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + /* + * The basename and dirname for a zero-length string is + * defined to be "." + */ + if (len == 0) { + len = 1; + src = (uintptr_t)"."; + } + + /* + * Start from the back of the string, moving back toward the + * front until we see a character that isn't a slash. That + * character is the last character in the basename. + */ + for (i = len - 1; i >= 0; i--) { + if (dtrace_load8(src + i) != '/') + break; + } + + if (i >= 0) + lastbase = i; + + /* + * Starting from the last character in the basename, move + * towards the front until we find a slash. The character + * that we processed immediately before that is the first + * character in the basename. + */ + for (; i >= 0; i--) { + if (dtrace_load8(src + i) == '/') + break; + } + + if (i >= 0) + firstbase = i + 1; + + /* + * Now keep going until we find a non-slash character. That + * character is the last character in the dirname. + */ + for (; i >= 0; i--) { + if (dtrace_load8(src + i) != '/') + break; + } + + if (i >= 0) + lastdir = i; + + ASSERT(!(lastbase == -1 && firstbase != -1)); + ASSERT(!(firstbase == -1 && lastdir != -1)); + + if (lastbase == -1) { + /* + * We didn't find a non-slash character. We know that + * the length is non-zero, so the whole string must be + * slashes. In either the dirname or the basename + * case, we return '/'. + */ + ASSERT(firstbase == -1); + firstbase = lastbase = lastdir = 0; + } + + if (firstbase == -1) { + /* + * The entire string consists only of a basename + * component. If we're looking for dirname, we need + * to change our string to be just "."; if we're + * looking for a basename, we'll just set the first + * character of the basename to be 0. + */ + if (subr == DIF_SUBR_DIRNAME) { + ASSERT(lastdir == -1); + src = (uintptr_t)"."; + lastdir = 0; + } else { + firstbase = 0; + } + } + + if (subr == DIF_SUBR_DIRNAME) { + if (lastdir == -1) { + /* + * We know that we have a slash in the name -- + * or lastdir would be set to 0, above. And + * because lastdir is -1, we know that this + * slash must be the first character. (That + * is, the full string must be of the form + * "/basename".) In this case, the last + * character of the directory name is 0. + */ + lastdir = 0; + } + + start = 0; + end = lastdir; + } else { + ASSERT(subr == DIF_SUBR_BASENAME); + ASSERT(firstbase != -1 && lastbase != -1); + start = firstbase; + end = lastbase; + } + + for (i = start, j = 0; i <= end && j < size - 1; i++, j++) + dest[j] = dtrace_load8(src + i); + + dest[j] = '\0'; + regs[rd] = (uintptr_t)dest; + mstate->dtms_scratch_ptr += size; + break; + } + + case DIF_SUBR_CLEANPATH: { + char *dest = (char *)mstate->dtms_scratch_ptr, c; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t src = tupregs[0].dttk_value; + int i = 0, j = 0; + + if (!dtrace_strcanload(src, size, mstate, vstate)) { + regs[rd] = 0; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + /* + * Move forward, loading each character. + */ + do { + c = dtrace_load8(src + i++); +next: + if (j + 5 >= size) /* 5 = strlen("/..c\0") */ + break; + + if (c != '/') { + dest[j++] = c; + continue; + } + + c = dtrace_load8(src + i++); + + if (c == '/') { + /* + * We have two slashes -- we can just advance + * to the next character. + */ + goto next; + } + + if (c != '.') { + /* + * This is not "." and it's not ".." -- we can + * just store the "/" and this character and + * drive on. + */ + dest[j++] = '/'; + dest[j++] = c; + continue; + } + + c = dtrace_load8(src + i++); + + if (c == '/') { + /* + * This is a "/./" component. We're not going + * to store anything in the destination buffer; + * we're just going to go to the next component. + */ + goto next; + } + + if (c != '.') { + /* + * This is not ".." -- we can just store the + * "/." and this character and continue + * processing. + */ + dest[j++] = '/'; + dest[j++] = '.'; + dest[j++] = c; + continue; + } + + c = dtrace_load8(src + i++); + + if (c != '/' && c != '\0') { + /* + * This is not ".." -- it's "..[mumble]". + * We'll store the "/.." and this character + * and continue processing. + */ + dest[j++] = '/'; + dest[j++] = '.'; + dest[j++] = '.'; + dest[j++] = c; + continue; + } + + /* + * This is "/../" or "/..\0". We need to back up + * our destination pointer until we find a "/". + */ + i--; + while (j != 0 && dest[--j] != '/') + continue; + + if (c == '\0') + dest[++j] = '/'; + } while (c != '\0'); + + dest[j] = '\0'; + regs[rd] = (uintptr_t)dest; + mstate->dtms_scratch_ptr += size; + break; + } + + case DIF_SUBR_INET_NTOA: + case DIF_SUBR_INET_NTOA6: + case DIF_SUBR_INET_NTOP: { + size_t size; + int af, argi, i; + char *base, *end; + + if (subr == DIF_SUBR_INET_NTOP) { + af = (int)tupregs[0].dttk_value; + argi = 1; + } else { + af = subr == DIF_SUBR_INET_NTOA ? AF_INET: AF_INET6; + argi = 0; + } + + if (af == AF_INET) { + ipaddr_t ip4; + uint8_t *ptr8, val; + + /* + * Safely load the IPv4 address. + */ + ip4 = dtrace_load32(tupregs[argi].dttk_value); + + /* + * Check an IPv4 string will fit in scratch. + */ + size = INET_ADDRSTRLEN; + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + base = (char *)mstate->dtms_scratch_ptr; + end = (char *)mstate->dtms_scratch_ptr + size - 1; + + /* + * Stringify as a dotted decimal quad. + */ + *end-- = '\0'; + ptr8 = (uint8_t *)&ip4; + for (i = 3; i >= 0; i--) { + val = ptr8[i]; + + if (val == 0) { + *end-- = '0'; + } else { + for (; val; val /= 10) { + *end-- = '0' + (val % 10); + } + } + + if (i > 0) + *end-- = '.'; + } + ASSERT(end + 1 >= base); +#ifdef CONFIG_IPV6 + } else if (af == AF_INET6) { + struct in6_addr ip6; + int firstzero, tryzero, numzero, v6end; + uint16_t val; + const char digits[] = "0123456789abcdef"; + + /* + * Stringify using RFC 1884 convention 2 - 16 bit + * hexadecimal values with a zero-run compression. + * Lower case hexadecimal digits are used. + * eg, fe80::214:4fff:fe0b:76c8. + * The IPv4 embedded form is returned for inet_ntop, + * just the IPv4 string is returned for inet_ntoa6. + */ + + /* + * Safely load the IPv6 address. + */ + dtrace_bcopy( + (void *)(uintptr_t)tupregs[argi].dttk_value, + (void *)(uintptr_t)&ip6, sizeof(struct in6_addr)); + + /* + * Check an IPv6 string will fit in scratch. + */ + size = INET6_ADDRSTRLEN; + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + base = (char *)mstate->dtms_scratch_ptr; + end = (char *)mstate->dtms_scratch_ptr + size - 1; + *end-- = '\0'; + + /* + * Find the longest run of 16 bit zero values + * for the single allowed zero compression - "::". + */ + firstzero = -1; + tryzero = -1; + numzero = 1; + for (i = 0; i < sizeof(struct in6_addr); i++) { + if (ip6.s6_addr[i] == 0 && + tryzero == -1 && i % 2 == 0) { + tryzero = i; + continue; + } + + if (tryzero != -1 && + (ip6.s6_addr[i] != 0 || + i == sizeof(struct in6_addr) - 1)) { + + if (i - tryzero <= numzero) { + tryzero = -1; + continue; + } + + firstzero = tryzero; + numzero = i - i % 2 - tryzero; + tryzero = -1; + + if (ip6.s6_addr[i] == 0 && + i == sizeof(struct in6_addr) - 1) + numzero += 2; + } + } + ASSERT(firstzero + numzero <= sizeof(struct in6_addr)); + + /* + * Check for an IPv4 embedded address. + */ + v6end = sizeof(struct in6_addr) - 2; + if (ipv6_addr_type(&ip6) | + (IPV6_ADDR_COMPATv4 | IPV6_ADDR_MAPPED)) { + for (i = sizeof(struct in6_addr) - 1; + i >= DTRACE_V4MAPPED_OFFSET; i--) { + ASSERT(end >= base); + + val = ip6.s6_addr[i]; + + if (val == 0) { + *end-- = '0'; + } else { + for (; val; val /= 10) { + *end-- = '0' + val % 10; + } + } + + if (i > DTRACE_V4MAPPED_OFFSET) + *end-- = '.'; + } + + if (subr == DIF_SUBR_INET_NTOA6) + goto inetout; + + /* + * Set v6end to skip the IPv4 address that + * we have already stringified. + */ + v6end = 10; + } + + /* + * Build the IPv6 string by working through the + * address in reverse. + */ + for (i = v6end; i >= 0; i -= 2) { + ASSERT(end >= base); + + if (i == firstzero + numzero - 2) { + *end-- = ':'; + *end-- = ':'; + i -= numzero - 2; + continue; + } + + if (i < 14 && i != firstzero - 2) + *end-- = ':'; + + val = (ip6.s6_addr[i] << 8) + + ip6.s6_addr[i + 1]; + + if (val == 0) { + *end-- = '0'; + } else { + for (; val; val /= 16) { + *end-- = digits[val % 16]; + } + } + } + ASSERT(end + 1 >= base); +#endif + } else { + /* + * The user didn't use AH_INET or AH_INET6. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + regs[rd] = 0; + break; + } + +#ifdef CONFIG_IPV6 +inetout: +#endif + regs[rd] = (uintptr_t)end + 1; + mstate->dtms_scratch_ptr += size; + break; + } + + } +} + +/* + * Emulate the execution of DTrace IR instructions specified by the given DIF + * object. This function is deliberately void fo assertions as all of the + * necessary checks are handled by a call to dtrace_difo_validate(). + */ +uint64_t dtrace_dif_emulate(dtrace_difo_t *difo, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate, dtrace_state_t *state) +{ + const dif_instr_t *text = difo->dtdo_buf; + const uint_t textlen = difo->dtdo_len; + const char *strtab = difo->dtdo_strtab; + const uint64_t *inttab = difo->dtdo_inttab; + + int cpu = smp_processor_id(); + uint64_t rval = 0; + dtrace_statvar_t *svar; + dtrace_dstate_t *dstate = &vstate->dtvs_dynvars; + dtrace_difv_t *v; + volatile uint16_t *flags = &cpu_core[cpu].cpuc_dtrace_flags; + volatile uintptr_t *illval = &cpu_core[cpu].cpuc_dtrace_illval; + + dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; + /* +2 for thread and id */ + uint64_t regs[DIF_DIR_NREGS]; + uint64_t *tmp; + + uint8_t cc_n = 0, cc_z = 0, cc_v = 0, cc_c = 0; + int64_t cc_r; + uint_t pc = 0, id, opc = 0; + uint8_t ttop = 0; + dif_instr_t instr; + uint_t r1, r2, rd; + + /* + * We stash the current DIF object into the machine state: we need it + * for subsequent access checking. + */ + mstate->dtms_difo = difo; + + regs[DIF_REG_R0] = 0; /* %r0 is fixed at zero */ + + while (pc < textlen && !(*flags & CPU_DTRACE_FAULT)) { + opc = pc; + + instr = text[pc++]; + r1 = DIF_INSTR_R1(instr); + r2 = DIF_INSTR_R2(instr); + rd = DIF_INSTR_RD(instr); + + switch (DIF_INSTR_OP(instr)) { + case DIF_OP_OR: + regs[rd] = regs[r1] | regs[r2]; + break; + case DIF_OP_XOR: + regs[rd] = regs[r1] ^ regs[r2]; + break; + case DIF_OP_AND: + regs[rd] = regs[r1] & regs[r2]; + break; + case DIF_OP_SLL: + regs[rd] = regs[r1] << regs[r2]; + break; + case DIF_OP_SRL: + regs[rd] = regs[r1] >> regs[r2]; + break; + case DIF_OP_SUB: + regs[rd] = regs[r1] - regs[r2]; + break; + case DIF_OP_ADD: + regs[rd] = regs[r1] + regs[r2]; + break; + case DIF_OP_MUL: + regs[rd] = regs[r1] * regs[r2]; + break; + case DIF_OP_SDIV: + if (regs[r2] == 0) { + regs[rd] = 0; + *flags |= CPU_DTRACE_DIVZERO; + } else { + /* + * We cannot simply do a 64-bit division, since + * gcc translates it into a call to a function + * that is not linked into the kernel. + * + * regs[rd] = (int64_t)regs[r1] / + * (int64_t)regs[r2]; + */ + regs[rd] = (int64_t)regs[r1]; + do_div(regs[rd], (int64_t)regs[r2]); + } + break; + + case DIF_OP_UDIV: + if (regs[r2] == 0) { + regs[rd] = 0; + *flags |= CPU_DTRACE_DIVZERO; + } else { + /* + * We cannot simply do a 64-bit division, since + * gcc translates it into a call to a function + * that is not linked into the kernel. + * + * regs[rd] = regs[r1] / regs[r2]; + */ + regs[rd] = regs[r1]; + do_div(regs[rd], regs[r2]); + } + break; + + case DIF_OP_SREM: + if (regs[r2] == 0) { + regs[rd] = 0; + *flags |= CPU_DTRACE_DIVZERO; + } else { + /* + * We cannot simply do a 64-bit division, since + * gcc translates it into a call to a function + * that is not linked into the kernel. + * + * regs[rd] = (int64_t)regs[r1] % + * (int64_t)regs[r2]; + */ + regs[rd] = (int64_t)regs[r1]; + regs[rd] = do_div(regs[rd], (int64_t)regs[r2]); + } + break; + + case DIF_OP_UREM: + if (regs[r2] == 0) { + regs[rd] = 0; + *flags |= CPU_DTRACE_DIVZERO; + } else { + /* + * We cannot simply do a 64-bit division, since + * gcc translates it into a call to a function + * that is not linked into the kernel. + * + * regs[rd] = regs[r1] % regs[r2]; + */ + regs[rd] = regs[r1]; + regs[rd] = do_div(regs[rd], regs[r2]); + } + break; + + case DIF_OP_NOT: + regs[rd] = ~regs[r1]; + break; + case DIF_OP_MOV: + regs[rd] = regs[r1]; + break; + case DIF_OP_CMP: + cc_r = regs[r1] - regs[r2]; + cc_n = cc_r < 0; + cc_z = cc_r == 0; + cc_v = 0; + cc_c = regs[r1] < regs[r2]; + break; + case DIF_OP_TST: + cc_n = cc_v = cc_c = 0; + cc_z = regs[r1] == 0; + break; + case DIF_OP_BA: + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BE: + if (cc_z) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BNE: + if (cc_z == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BG: + if ((cc_z | (cc_n ^ cc_v)) == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BGU: + if ((cc_c | cc_z) == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BGE: + if ((cc_n ^ cc_v) == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BGEU: + if (cc_c == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BL: + if (cc_n ^ cc_v) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BLU: + if (cc_c) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BLE: + if (cc_z | (cc_n ^ cc_v)) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BLEU: + if (cc_c | cc_z) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_RLDSB: +#ifdef FIXME_OPENSOLARIS_BUG + if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) { +#else + if (!dtrace_canload(regs[r1], 1, mstate, vstate)) { +#endif + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDSB: + regs[rd] = (int8_t)dtrace_load8(regs[r1]); + break; + case DIF_OP_RLDSH: + if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDSH: + regs[rd] = (int16_t)dtrace_load16(regs[r1]); + break; + case DIF_OP_RLDSW: + if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDSW: + regs[rd] = (int32_t)dtrace_load32(regs[r1]); + break; + case DIF_OP_RLDUB: + if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDUB: + regs[rd] = dtrace_load8(regs[r1]); + break; + case DIF_OP_RLDUH: + if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDUH: + regs[rd] = dtrace_load16(regs[r1]); + break; + case DIF_OP_RLDUW: + if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDUW: + regs[rd] = dtrace_load32(regs[r1]); + break; + case DIF_OP_RLDX: + if (!dtrace_canstore(regs[r1], 8, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + /*FALLTHROUGH*/ + case DIF_OP_LDX: + regs[rd] = dtrace_load64(regs[r1]); + break; + case DIF_OP_ULDSB: + regs[rd] = (int8_t)dtrace_fuword8( + (void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDSH: + regs[rd] = (int16_t)dtrace_fuword16( + (void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDSW: + regs[rd] = (int32_t)dtrace_fuword32( + (void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDUB: + regs[rd] = dtrace_fuword8((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDUH: + regs[rd] = dtrace_fuword16( + (void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDUW: + regs[rd] = dtrace_fuword32( + (void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDX: + regs[rd] = dtrace_fuword64( + (void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_RET: + rval = regs[rd]; + pc = textlen; + break; + case DIF_OP_NOP: + break; + case DIF_OP_SETX: + regs[rd] = inttab[DIF_INSTR_INTEGER(instr)]; + break; + case DIF_OP_SETS: + regs[rd] = (uint64_t)(uintptr_t) + (strtab + DIF_INSTR_STRING(instr)); + break; + case DIF_OP_SCMP: { + size_t sz = state->dts_options[ + DTRACEOPT_STRSIZE]; + uintptr_t s1 = regs[r1]; + uintptr_t s2 = regs[r2]; + + if (s1 != (uintptr_t)NULL && + !dtrace_strcanload(s1, sz, mstate, vstate)) + break; + if (s2 != (uintptr_t)NULL && + !dtrace_strcanload(s2, sz, mstate, vstate)) + break; + + cc_r = dtrace_strncmp((char *)s1, (char *)s2, sz); + + cc_n = cc_r < 0; + cc_z = cc_r == 0; + cc_v = cc_c = 0; + break; + } + case DIF_OP_LDGA: + regs[rd] = dtrace_dif_variable(mstate, state, r1, + regs[r2]); + break; + case DIF_OP_LDGS: + id = DIF_INSTR_VAR(instr); + + if (id >= DIF_VAR_OTHER_UBASE) { + uintptr_t a; + + id -= DIF_VAR_OTHER_UBASE; + svar = vstate->dtvs_globals[id]; + ASSERT(svar != NULL); + v = &svar->dtsv_var; + + if (!(v->dtdv_type.dtdt_flags & DIF_TF_BYREF)) { + regs[rd] = svar->dtsv_data; + break; + } + + a = (uintptr_t)svar->dtsv_data; + + /* + * If the 0th byte is set to UINT8_MAX then + * this is to be treated as a reference to a + * NULL variable. + */ + if (*(uint8_t *)a == UINT8_MAX) + regs[rd] = 0; + else + regs[rd] = a + sizeof(uint64_t); + + break; + } + + regs[rd] = dtrace_dif_variable(mstate, state, id, 0); + break; + + case DIF_OP_STGS: + id = DIF_INSTR_VAR(instr); + + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + + svar = vstate->dtvs_globals[id]; + ASSERT(svar != NULL); + v = &svar->dtsv_var; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + uintptr_t a = (uintptr_t)svar->dtsv_data; + + ASSERT(a != 0); + ASSERT(svar->dtsv_size != 0); + + if (regs[rd] == 0) { + *(uint8_t *)a = UINT8_MAX; + break; + } else { + *(uint8_t *)a = 0; + a += sizeof(uint64_t); + } + + if (!dtrace_vcanload( + (void *)(uintptr_t)regs[rd], + &v->dtdv_type, mstate, vstate)) + break; + + dtrace_vcopy((void *)(uintptr_t)regs[rd], + (void *)a, &v->dtdv_type); + break; + } + + svar->dtsv_data = regs[rd]; + break; + + case DIF_OP_LDTA: + /* + * There are no DTrace built-in thread-local arrays at + * present. This opcode is saved for future work. + */ + *flags |= CPU_DTRACE_ILLOP; + regs[rd] = 0; + break; + + case DIF_OP_LDLS: + id = DIF_INSTR_VAR(instr); + + if (id < DIF_VAR_OTHER_UBASE) { + /* + * For now, this has no meaning. + */ + regs[rd] = 0; + break; + } + + id -= DIF_VAR_OTHER_UBASE; + + ASSERT(id < vstate->dtvs_nlocals); + ASSERT(vstate->dtvs_locals != NULL); + + svar = vstate->dtvs_locals[id]; + ASSERT(svar != NULL); + v = &svar->dtsv_var; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + uintptr_t a = (uintptr_t)svar->dtsv_data; + size_t sz = v->dtdv_type.dtdt_size; + + sz += sizeof(uint64_t); + ASSERT(svar->dtsv_size == NR_CPUS * sz); + a += smp_processor_id() * sz; + + if (*(uint8_t *)a == UINT8_MAX) { + /* + * If the 0th byte is set to UINT8_MAX + * then this is to be treated as a + * reference to a NULL variable. + */ + regs[rd] = 0; + } else + regs[rd] = a + sizeof(uint64_t); + + break; + } + + ASSERT(svar->dtsv_size == NR_CPUS * sizeof(uint64_t)); + tmp = (uint64_t *)(uintptr_t)svar->dtsv_data; + regs[rd] = tmp[smp_processor_id()]; + break; + + case DIF_OP_STLS: + id = DIF_INSTR_VAR(instr); + + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + ASSERT(id < vstate->dtvs_nlocals); + + ASSERT(vstate->dtvs_locals != NULL); + svar = vstate->dtvs_locals[id]; + ASSERT(svar != NULL); + v = &svar->dtsv_var; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + uintptr_t a = (uintptr_t)svar->dtsv_data; + size_t sz = v->dtdv_type.dtdt_size; + + sz += sizeof(uint64_t); + ASSERT(svar->dtsv_size == NR_CPUS * sz); + a += smp_processor_id() * sz; + + if (regs[rd] == 0) { + *(uint8_t *)a = UINT8_MAX; + break; + } else { + *(uint8_t *)a = 0; + a += sizeof(uint64_t); + } + + if (!dtrace_vcanload( + (void *)(uintptr_t)regs[rd], + &v->dtdv_type, mstate, vstate)) + break; + + dtrace_vcopy((void *)(uintptr_t)regs[rd], + (void *)a, &v->dtdv_type); + break; + } + + ASSERT(svar->dtsv_size == NR_CPUS * sizeof(uint64_t)); + tmp = (uint64_t *)(uintptr_t)svar->dtsv_data; + tmp[smp_processor_id()] = regs[rd]; + break; + + case DIF_OP_LDTS: { + dtrace_dynvar_t *dvar; + dtrace_key_t *key; + + id = DIF_INSTR_VAR(instr); + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + v = &vstate->dtvs_tlocals[id]; + + key = &tupregs[DIF_DTR_NREGS]; + key[0].dttk_value = (uint64_t)id; + key[0].dttk_size = 0; + DTRACE_TLS_THRKEY(key[1].dttk_value); + key[1].dttk_size = 0; + + dvar = dtrace_dynvar(dstate, 2, key, sizeof(uint64_t), + DTRACE_DYNVAR_NOALLOC, mstate, + vstate); + + if (dvar == NULL) { + regs[rd] = 0; + break; + } + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) + regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data; + else + regs[rd] = *((uint64_t *)dvar->dtdv_data); + + break; + } + + case DIF_OP_STTS: { + dtrace_dynvar_t *dvar; + dtrace_key_t *key; + + id = DIF_INSTR_VAR(instr); + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + + key = &tupregs[DIF_DTR_NREGS]; + key[0].dttk_value = (uint64_t)id; + key[0].dttk_size = 0; + DTRACE_TLS_THRKEY(key[1].dttk_value); + key[1].dttk_size = 0; + v = &vstate->dtvs_tlocals[id]; + + dvar = dtrace_dynvar(dstate, 2, key, + v->dtdv_type.dtdt_size > sizeof(uint64_t) + ? v->dtdv_type.dtdt_size + : sizeof(uint64_t), + regs[rd] + ? DTRACE_DYNVAR_ALLOC + : DTRACE_DYNVAR_DEALLOC, + mstate, vstate); + + /* + * Given that we're storing to thread-local data, + * we need to flush our predicate cache. + */ + current->predcache = 0; + + if (dvar == NULL) + break; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + if (!dtrace_vcanload( + (void *)(uintptr_t)regs[rd], + &v->dtdv_type, mstate, vstate)) + break; + + dtrace_vcopy((void *)(uintptr_t)regs[rd], + dvar->dtdv_data, &v->dtdv_type); + } else + *((uint64_t *)dvar->dtdv_data) = regs[rd]; + + break; + } + + case DIF_OP_SRA: + regs[rd] = (int64_t)regs[r1] >> regs[r2]; + break; + + case DIF_OP_CALL: + dtrace_dif_subr(DIF_INSTR_SUBR(instr), rd, regs, + tupregs, ttop, mstate, state); + break; + + case DIF_OP_PUSHTR: + if (ttop == DIF_DTR_NREGS) { + *flags |= CPU_DTRACE_TUPOFLOW; + break; + } + + if (r1 == DIF_TYPE_STRING) + /* + * If this is a string type and the size is 0, + * we'll use the system-wide default string + * size. Note that we are _not_ looking at + * the value of the DTRACEOPT_STRSIZE option; + * had this been set, we would expect to have + * a non-zero size value in the "pushtr". + */ + tupregs[ttop].dttk_size = + dtrace_strlen( + (char *)(uintptr_t)regs[rd], + regs[r2] + ? regs[r2] + : dtrace_strsize_default + ) + 1; + else + tupregs[ttop].dttk_size = regs[r2]; + + tupregs[ttop++].dttk_value = regs[rd]; + break; + + case DIF_OP_PUSHTV: + if (ttop == DIF_DTR_NREGS) { + *flags |= CPU_DTRACE_TUPOFLOW; + break; + } + + tupregs[ttop].dttk_value = regs[rd]; + tupregs[ttop++].dttk_size = 0; + break; + + case DIF_OP_POPTS: + if (ttop != 0) + ttop--; + break; + + case DIF_OP_FLUSHTS: + ttop = 0; + break; + + case DIF_OP_LDGAA: + case DIF_OP_LDTAA: { + dtrace_dynvar_t *dvar; + dtrace_key_t *key = tupregs; + uint_t nkeys = ttop; + + id = DIF_INSTR_VAR(instr); + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + + key[nkeys].dttk_value = (uint64_t)id; + key[nkeys++].dttk_size = 0; + + if (DIF_INSTR_OP(instr) == DIF_OP_LDTAA) { + DTRACE_TLS_THRKEY(key[nkeys].dttk_value); + key[nkeys++].dttk_size = 0; + v = &vstate->dtvs_tlocals[id]; + } else + v = &vstate->dtvs_globals[id]->dtsv_var; + + dvar = dtrace_dynvar(dstate, nkeys, key, + v->dtdv_type.dtdt_size > sizeof(uint64_t) ? + v->dtdv_type.dtdt_size : sizeof(uint64_t), + DTRACE_DYNVAR_NOALLOC, mstate, vstate); + + if (dvar == NULL) { + regs[rd] = 0; + break; + } + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) + regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data; + else + regs[rd] = *((uint64_t *)dvar->dtdv_data); + + break; + } + + case DIF_OP_STGAA: + case DIF_OP_STTAA: { + dtrace_dynvar_t *dvar; + dtrace_key_t *key = tupregs; + uint_t nkeys = ttop; + + id = DIF_INSTR_VAR(instr); + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + + key[nkeys].dttk_value = (uint64_t)id; + key[nkeys++].dttk_size = 0; + + if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) { + DTRACE_TLS_THRKEY(key[nkeys].dttk_value); + key[nkeys++].dttk_size = 0; + v = &vstate->dtvs_tlocals[id]; + } else + v = &vstate->dtvs_globals[id]->dtsv_var; + + dvar = dtrace_dynvar(dstate, nkeys, key, + v->dtdv_type.dtdt_size > sizeof(uint64_t) + ? v->dtdv_type.dtdt_size + : sizeof(uint64_t), + regs[rd] ? DTRACE_DYNVAR_ALLOC + : DTRACE_DYNVAR_DEALLOC, + mstate, vstate); + + if (dvar == NULL) + break; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + if (!dtrace_vcanload( + (void *)(uintptr_t)regs[rd], + &v->dtdv_type, mstate, vstate)) + break; + + dtrace_vcopy((void *)(uintptr_t)regs[rd], + dvar->dtdv_data, &v->dtdv_type); + } else + *((uint64_t *)dvar->dtdv_data) = regs[rd]; + + break; + } + + case DIF_OP_ALLOCS: { + uintptr_t ptr = + P2ROUNDUP(mstate->dtms_scratch_ptr, 8); + size_t size = ptr - mstate->dtms_scratch_ptr + + regs[r1]; + + /* + * Rounding up the user allocation size could have + * overflowed large, bogus allocations (like -1ULL) to + * 0. + */ + if (size < regs[r1] || + !DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = 0; + break; + } + + dtrace_bzero((void *) mstate->dtms_scratch_ptr, size); + mstate->dtms_scratch_ptr += size; + regs[rd] = ptr; + break; + } + + case DIF_OP_COPYS: + if (!dtrace_canstore(regs[rd], regs[r2], mstate, + vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + + if (!dtrace_canload(regs[r1], regs[r2], mstate, vstate)) + break; + + dtrace_bcopy((void *)(uintptr_t)regs[r1], + (void *)(uintptr_t)regs[rd], + (size_t)regs[r2]); + break; + + case DIF_OP_STB: + if (!dtrace_canstore(regs[rd], 1, mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + + *((uint8_t *)(uintptr_t)regs[rd]) = (uint8_t)regs[r1]; + break; + + case DIF_OP_STH: + if (!dtrace_canstore(regs[rd], 2, mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + + if (regs[rd] & 1) { + *flags |= CPU_DTRACE_BADALIGN; + *illval = regs[rd]; + break; + } + + *((uint16_t *)(uintptr_t)regs[rd]) = (uint16_t)regs[r1]; + break; + + case DIF_OP_STW: + if (!dtrace_canstore(regs[rd], 4, mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + + if (regs[rd] & 3) { + *flags |= CPU_DTRACE_BADALIGN; + *illval = regs[rd]; + break; + } + + *((uint32_t *)(uintptr_t)regs[rd]) = (uint32_t)regs[r1]; + break; + + case DIF_OP_STX: + if (!dtrace_canstore(regs[rd], 8, mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + + if (regs[rd] & 7) { + *flags |= CPU_DTRACE_BADALIGN; + *illval = regs[rd]; + break; + } + + *((uint64_t *)(uintptr_t)regs[rd]) = regs[r1]; + break; + } + } + + if (!(*flags & CPU_DTRACE_FAULT)) + return rval; + + mstate->dtms_fltoffs = opc * sizeof(dif_instr_t); + mstate->dtms_present |= DTRACE_MSTATE_FLTOFFS; + + return 0; +} diff --git a/dtrace/dtrace_dof.c b/dtrace/dtrace_dof.c new file mode 100644 index 000000000000..805b2dca30ef --- /dev/null +++ b/dtrace/dtrace_dof.c @@ -0,0 +1,1053 @@ +/* + * FILE: dtrace_dof.c + * DESCRIPTION: Dynamic Tracing: DOF object functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include +#include + +#include "dtrace.h" + +size_t dtrace_difo_maxsize = 256 * 1024; +dtrace_optval_t dtrace_dof_maxsize = 256 * 1024; +size_t dtrace_actions_max = 16 * 1024; + +static void dtrace_dof_error(dof_hdr_t *dof, const char *str) +{ + if (dtrace_err_verbose) + pr_warning("failed to process DOF: %s", str); + +#ifdef DTRACE_ERRDEBUG + dtrace_errdebug(str); +#endif +} + +/* + * Create DOF out of a currently enabled state. Right now, we only create + * DOF containing the run-time options -- but this could be expanded to create + * complete DOF representing the enabled state. + */ +dof_hdr_t *dtrace_dof_create(dtrace_state_t *state) +{ + dof_hdr_t *dof; + dof_sec_t *sec; + dof_optdesc_t *opt; + int i, len = sizeof(dof_hdr_t) + + roundup(sizeof(dof_sec_t), sizeof(uint64_t)) + + sizeof(dof_optdesc_t) * DTRACEOPT_MAX; + + ASSERT(mutex_is_locked(&dtrace_lock)); + + dof = kmalloc(len, GFP_KERNEL); + dof->dofh_ident[DOF_ID_MAG0] = DOF_MAG_MAG0; + dof->dofh_ident[DOF_ID_MAG1] = DOF_MAG_MAG1; + dof->dofh_ident[DOF_ID_MAG2] = DOF_MAG_MAG2; + dof->dofh_ident[DOF_ID_MAG3] = DOF_MAG_MAG3; + + dof->dofh_ident[DOF_ID_MODEL] = DOF_MODEL_NATIVE; + dof->dofh_ident[DOF_ID_ENCODING] = DOF_ENCODE_NATIVE; + dof->dofh_ident[DOF_ID_VERSION] = DOF_VERSION; + dof->dofh_ident[DOF_ID_DIFVERS] = DIF_VERSION; + dof->dofh_ident[DOF_ID_DIFIREG] = DIF_DIR_NREGS; + dof->dofh_ident[DOF_ID_DIFTREG] = DIF_DTR_NREGS; + + dof->dofh_flags = 0; + dof->dofh_hdrsize = sizeof(dof_hdr_t); + dof->dofh_secsize = sizeof(dof_sec_t); + dof->dofh_secnum = 1; /* only DOF_SECT_OPTDESC */ + dof->dofh_secoff = sizeof(dof_hdr_t); + dof->dofh_loadsz = len; + dof->dofh_filesz = len; + dof->dofh_pad = 0; + + /* + * Fill in the option section header... + */ + sec = (dof_sec_t *)((uintptr_t)dof + sizeof(dof_hdr_t)); + sec->dofs_type = DOF_SECT_OPTDESC; + sec->dofs_align = sizeof(uint64_t); + sec->dofs_flags = DOF_SECF_LOAD; + sec->dofs_entsize = sizeof(dof_optdesc_t); + + opt = (dof_optdesc_t *)((uintptr_t)sec + + roundup(sizeof(dof_sec_t), sizeof(uint64_t))); + + sec->dofs_offset = (uintptr_t)opt - (uintptr_t)dof; + sec->dofs_size = sizeof(dof_optdesc_t) * DTRACEOPT_MAX; + + for (i = 0; i < DTRACEOPT_MAX; i++) { + opt[i].dofo_option = i; + opt[i].dofo_strtab = DOF_SECIDX_NONE; + opt[i].dofo_value = state->dts_options[i]; + } + + return dof; +} + +dof_hdr_t *dtrace_dof_copyin(void __user *argp, int *errp) +{ + dof_hdr_t hdr, *dof; + +#ifdef FIXME + /* This seems to be unnecessary and actually wrong). */ + ASSERT(!mutex_is_locked(&dtrace_lock)); +#endif + + /* + * First, we're going to copyin() the sizeof(dof_hdr_t). + */ + if (copy_from_user(&hdr, argp, sizeof(hdr)) != 0) { + dtrace_dof_error(NULL, "failed to copyin DOF header"); + *errp = -EFAULT; + return NULL; + } + + /* + * Now we'll allocate the entire DOF and copy it in -- provided + * that the length isn't outrageous. + */ + if (hdr.dofh_loadsz >= dtrace_dof_maxsize) { + dtrace_dof_error(&hdr, "load size exceeds maximum"); + *errp = -E2BIG; + return NULL; + } + + if (hdr.dofh_loadsz < sizeof(hdr)) { + dtrace_dof_error(&hdr, "invalid load size"); + *errp = -EINVAL; + return NULL; + } + + dof = kmalloc(hdr.dofh_loadsz, GFP_KERNEL); + + if (copy_from_user(dof, argp, hdr.dofh_loadsz) != 0 || + dof->dofh_loadsz != hdr.dofh_loadsz) { + kfree(dof); + *errp = -EFAULT; + return NULL; + } + + return dof; +} + +dof_hdr_t *dtrace_dof_property(const char *name) +{ + uchar_t *buf; + uint64_t loadsz; + unsigned int len, i; + dof_hdr_t *dof; + + /* + * Unfortunately, array of values in .conf files are always (and + * only) interpreted to be integer arrays. We must read our DOF + * as an integer array, and then squeeze it into a byte array. + */ +#ifdef FIXME + if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dtrace_devi, 0, + (char *)name, (int **)&buf, &len) != + DDI_PROP_SUCCESS) + return NULL; +#else + return NULL; +#endif + + for (i = 0; i < len; i++) + buf[i] = (uchar_t)(((int *)buf)[i]); + + if (len < sizeof(dof_hdr_t)) { +#ifdef FIXME + ddi_prop_free(buf); +#endif + dtrace_dof_error(NULL, "truncated header"); + return NULL; + } + + if (len < (loadsz = ((dof_hdr_t *)buf)->dofh_loadsz)) { +#ifdef FIXME + ddi_prop_free(buf); +#endif + dtrace_dof_error(NULL, "truncated DOF"); + return NULL; + } + + if (loadsz >= dtrace_dof_maxsize) { +#ifdef FIXME + ddi_prop_free(buf); +#endif + dtrace_dof_error(NULL, "oversized DOF"); + return NULL; + } + + dof = kmalloc(loadsz, GFP_KERNEL); + memcpy(dof, buf, loadsz); +#ifdef FIXME + ddi_prop_free(buf); +#endif + + return dof; +} + +void dtrace_dof_destroy(dof_hdr_t *dof) +{ + kfree(dof); +} + +/* + * Return the dof_sec_t pointer corresponding to a given section index. If the + * index is not valid, dtrace_dof_error() is called and NULL is returned. If + * a type other than DOF_SECT_NONE is specified, the header is checked against + * this type and NULL is returned if the types do not match. + */ +static dof_sec_t *dtrace_dof_sect(dof_hdr_t *dof, uint32_t type, + dof_secidx_t i) +{ + dof_sec_t *sec = (dof_sec_t *)(uintptr_t)((uintptr_t)dof + + dof->dofh_secoff + + i * dof->dofh_secsize); + + if (i >= dof->dofh_secnum) { + dtrace_dof_error(dof, "referenced section index is invalid"); + return NULL; + } + + if (!(sec->dofs_flags & DOF_SECF_LOAD)) { + dtrace_dof_error(dof, "referenced section is not loadable"); + return NULL; + } + + if (type != DOF_SECT_NONE && type != sec->dofs_type) { + dtrace_dof_error(dof, "referenced section is the wrong type"); + return NULL; + } + + return sec; +} + +static dtrace_probedesc_t *dtrace_dof_probedesc(dof_hdr_t *dof, dof_sec_t *sec, + dtrace_probedesc_t *desc) +{ + dof_probedesc_t *probe; + dof_sec_t *strtab; + uintptr_t daddr = (uintptr_t)dof; + uintptr_t str; + size_t size; + + if (sec->dofs_type != DOF_SECT_PROBEDESC) { + dtrace_dof_error(dof, "invalid probe section"); + return NULL; + } + + if (sec->dofs_align != sizeof(dof_secidx_t)) { + dtrace_dof_error(dof, "bad alignment in probe description"); + return NULL; + } + + if (sec->dofs_offset + sizeof(dof_probedesc_t) > dof->dofh_loadsz) { + dtrace_dof_error(dof, "truncated probe description"); + return NULL; + } + + probe = (dof_probedesc_t *)(uintptr_t)(daddr + sec->dofs_offset); + strtab = dtrace_dof_sect(dof, DOF_SECT_STRTAB, probe->dofp_strtab); + + if (strtab == NULL) + return NULL; + + str = daddr + strtab->dofs_offset; + size = strtab->dofs_size; + + if (probe->dofp_provider >= strtab->dofs_size) { + dtrace_dof_error(dof, "corrupt probe provider"); + return NULL; + } + + strncpy(desc->dtpd_provider, (char *)(str + probe->dofp_provider), + min((size_t)DTRACE_PROVNAMELEN - 1, + size - probe->dofp_provider)); + + if (probe->dofp_mod >= strtab->dofs_size) { + dtrace_dof_error(dof, "corrupt probe module"); + return NULL; + } + + strncpy(desc->dtpd_mod, (char *)(str + probe->dofp_mod), + min((size_t)DTRACE_MODNAMELEN - 1, size - probe->dofp_mod)); + + if (probe->dofp_func >= strtab->dofs_size) { + dtrace_dof_error(dof, "corrupt probe function"); + return NULL; + } + + strncpy(desc->dtpd_func, (char *)(str + probe->dofp_func), + min((size_t)DTRACE_FUNCNAMELEN - 1, size - probe->dofp_func)); + + if (probe->dofp_name >= strtab->dofs_size) { + dtrace_dof_error(dof, "corrupt probe name"); + return NULL; + } + + strncpy(desc->dtpd_name, (char *)(str + probe->dofp_name), + min((size_t)DTRACE_NAMELEN - 1, size - probe->dofp_name)); + + return desc; +} + +static dtrace_difo_t *dtrace_dof_difo(dof_hdr_t *dof, dof_sec_t *sec, + dtrace_vstate_t *vstate, + const cred_t *cr) +{ + dtrace_difo_t *dp; + size_t ttl = 0; + dof_difohdr_t *dofd; + uintptr_t daddr = (uintptr_t)dof; + size_t max = dtrace_difo_maxsize; + int i, l, n; + + static const struct { + int section; + int bufoffs; + int lenoffs; + int entsize; + int align; + const char *msg; + } difo[] = { + { + DOF_SECT_DIF, + offsetof(dtrace_difo_t, dtdo_buf), + offsetof(dtrace_difo_t, dtdo_len), + sizeof(dif_instr_t), + sizeof(dif_instr_t), + "multiple DIF sections" + }, + { + DOF_SECT_INTTAB, + offsetof(dtrace_difo_t, dtdo_inttab), + offsetof(dtrace_difo_t, dtdo_intlen), + sizeof(uint64_t), + sizeof(uint64_t), + "multiple integer tables" + }, + { + DOF_SECT_STRTAB, + offsetof(dtrace_difo_t, dtdo_strtab), + offsetof(dtrace_difo_t, dtdo_strlen), + 0, + sizeof(char), + "multiple string tables" + }, + { + DOF_SECT_VARTAB, + offsetof(dtrace_difo_t, dtdo_vartab), + offsetof(dtrace_difo_t, dtdo_varlen), + sizeof(dtrace_difv_t), + sizeof(uint_t), + "multiple variable tables" + }, + { + DOF_SECT_NONE, + 0, + 0, + 0, + 0, + NULL + } + }; + + if (sec->dofs_type != DOF_SECT_DIFOHDR) { + dtrace_dof_error(dof, "invalid DIFO header section"); + return NULL; + } + + if (sec->dofs_align != sizeof(dof_secidx_t)) { + dtrace_dof_error(dof, "bad alignment in DIFO header"); + return NULL; + } + + if (sec->dofs_size < sizeof(dof_difohdr_t) || + sec->dofs_size % sizeof(dof_secidx_t)) { + dtrace_dof_error(dof, "bad size in DIFO header"); + return NULL; + } + + dofd = (dof_difohdr_t *)(uintptr_t)(daddr + sec->dofs_offset); + n = (sec->dofs_size - sizeof(*dofd)) / sizeof(dof_secidx_t) + 1; + + dp = kzalloc(sizeof(dtrace_difo_t), GFP_KERNEL); + dp->dtdo_rtype = dofd->dofd_rtype; + + for (l = 0; l < n; l++) { + dof_sec_t *subsec; + void **bufp; + uint32_t *lenp; + + if ((subsec = + dtrace_dof_sect(dof, DOF_SECT_NONE, dofd->dofd_links[l])) + == NULL) + goto err; /* invalid section link */ + + if (ttl + subsec->dofs_size > max) { + dtrace_dof_error(dof, "exceeds maximum size"); + goto err; + } + + ttl += subsec->dofs_size; + + for (i = 0; difo[i].section != DOF_SECT_NONE; i++) { + if (subsec->dofs_type != difo[i].section) + continue; + + if (!(subsec->dofs_flags & DOF_SECF_LOAD)) { + dtrace_dof_error(dof, "section not loaded"); + goto err; + } + + if (subsec->dofs_align != difo[i].align) { + dtrace_dof_error(dof, "bad alignment"); + goto err; + } + + bufp = (void **)((uintptr_t)dp + difo[i].bufoffs); + lenp = (uint32_t *)((uintptr_t)dp + difo[i].lenoffs); + + if (*bufp != NULL) { + dtrace_dof_error(dof, difo[i].msg); + goto err; + } + + if (difo[i].entsize != subsec->dofs_entsize) { + dtrace_dof_error(dof, "entry size mismatch"); + goto err; + } + + if (subsec->dofs_entsize != 0) { + uint64_t n = subsec->dofs_size; + + if (do_div(n, subsec->dofs_entsize) != 0) { + dtrace_dof_error(dof, + "corrupt entry size"); + goto err; + } + } + + *lenp = subsec->dofs_size; + *bufp = kmalloc(subsec->dofs_size, GFP_KERNEL); + memcpy(*bufp, + (char *)(uintptr_t)(daddr + subsec->dofs_offset), + subsec->dofs_size); + + if (subsec->dofs_entsize != 0) + *lenp /= subsec->dofs_entsize; + + break; + } + + /* + * If we encounter a loadable DIFO sub-section that is not + * known to us, assume this is a broken program and fail. + */ + if (difo[i].section == DOF_SECT_NONE && + (subsec->dofs_flags & DOF_SECF_LOAD)) { + dtrace_dof_error(dof, "unrecognized DIFO subsection"); + goto err; + } + } + + if (dp->dtdo_buf == NULL) { + /* + * We can't have a DIF object without DIF text. + */ + dtrace_dof_error(dof, "missing DIF text"); + goto err; + } + + /* + * Before we validate the DIF object, run through the variable table + * looking for the strings -- if any of their size are under, we'll set + * their size to be the system-wide default string size. Note that + * this should _not_ happen if the "strsize" option has been set -- + * in this case, the compiler should have set the size to reflect the + * setting of the option. + */ + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + dtrace_diftype_t *t = &v->dtdv_type; + + if (v->dtdv_id < DIF_VAR_OTHER_UBASE) + continue; + + if (t->dtdt_kind == DIF_TYPE_STRING && t->dtdt_size == 0) + t->dtdt_size = dtrace_strsize_default; + } + + if (dtrace_difo_validate(dp, vstate, DIF_DIR_NREGS, cr) != 0) + goto err; + + dtrace_difo_init(dp, vstate); + return dp; + +err: + kfree(dp->dtdo_buf); + kfree(dp->dtdo_inttab); + kfree(dp->dtdo_strtab); + kfree(dp->dtdo_vartab); + + kfree(dp); + return NULL; +} + +static dtrace_predicate_t *dtrace_dof_predicate(dof_hdr_t *dof, dof_sec_t *sec, + dtrace_vstate_t *vstate, + const cred_t *cr) +{ + dtrace_difo_t *dp; + + if ((dp = dtrace_dof_difo(dof, sec, vstate, cr)) == NULL) + return NULL; + + return dtrace_predicate_create(dp); +} + +static dtrace_actdesc_t *dtrace_dof_actdesc(dof_hdr_t *dof, dof_sec_t *sec, + dtrace_vstate_t *vstate, + const cred_t *cr) +{ + dtrace_actdesc_t *act, *first = NULL, *last = NULL, *next; + dof_actdesc_t *desc; + dof_sec_t *difosec; + size_t offs; + uintptr_t daddr = (uintptr_t)dof; + uint64_t arg; + dtrace_actkind_t kind; + + if (sec->dofs_type != DOF_SECT_ACTDESC) { + dtrace_dof_error(dof, "invalid action section"); + return NULL; + } + + if (sec->dofs_offset + sizeof(dof_actdesc_t) > dof->dofh_loadsz) { + dtrace_dof_error(dof, "truncated action description"); + return NULL; + } + + if (sec->dofs_align != sizeof(uint64_t)) { + dtrace_dof_error(dof, "bad alignment in action description"); + return NULL; + } + + if (sec->dofs_size < sec->dofs_entsize) { + dtrace_dof_error(dof, "section entry size exceeds total size"); + return NULL; + } + + if (sec->dofs_entsize != sizeof(dof_actdesc_t)) { + dtrace_dof_error(dof, "bad entry size in action description"); + return NULL; + } + + /* + * Was: sec->dofs_size / sec->dofs_entsize > dtrace_actions_max + * but it is safer to simply avoid the division (it requires use of + * a macro in Linux to cover 64-bit division in a 32-bit kernel. + */ + if (sec->dofs_size > sec->dofs_entsize * dtrace_actions_max) { + dtrace_dof_error(dof, "actions exceed dtrace_actions_max"); + return NULL; + } + + for (offs = 0; offs < sec->dofs_size; offs += sec->dofs_entsize) { + desc = (dof_actdesc_t *)(daddr + + (uintptr_t)sec->dofs_offset + offs); + kind = (dtrace_actkind_t)desc->dofa_kind; + + if (DTRACEACT_ISPRINTFLIKE(kind) && + (kind != DTRACEACT_PRINTA || + desc->dofa_strtab != DOF_SECIDX_NONE)) { + dof_sec_t *strtab; + char *str, *fmt; + uint64_t i; + + /* + * The printf()-like actions must have a format string. + */ + if ((strtab = + dtrace_dof_sect(dof, DOF_SECT_STRTAB, + desc->dofa_strtab)) == NULL) + goto err; + + str = (char *)((uintptr_t)dof + + (uintptr_t)strtab->dofs_offset); + + for (i = desc->dofa_arg; i < strtab->dofs_size; i++) { + if (str[i] == '\0') + break; + } + + if (i >= strtab->dofs_size) { + dtrace_dof_error(dof, "bogus format string"); + goto err; + } + + if (i == desc->dofa_arg) { + dtrace_dof_error(dof, "empty format string"); + goto err; + } + + i -= desc->dofa_arg; + fmt = kmalloc(i + 1, GFP_KERNEL); + memcpy(fmt, &str[desc->dofa_arg], i + 1); + arg = (uint64_t)(uintptr_t)fmt; + } else { + if (kind == DTRACEACT_PRINTA) { + ASSERT(desc->dofa_strtab == DOF_SECIDX_NONE); + arg = 0; + } else + arg = desc->dofa_arg; + } + + act = dtrace_actdesc_create(kind, desc->dofa_ntuple, + desc->dofa_uarg, arg); + + if (last != NULL) + last->dtad_next = act; + else + first = act; + + last = act; + + if (desc->dofa_difo == DOF_SECIDX_NONE) + continue; + + if ((difosec = dtrace_dof_sect(dof, DOF_SECT_DIFOHDR, + desc->dofa_difo)) == NULL) + goto err; + + act->dtad_difo = dtrace_dof_difo(dof, difosec, vstate, cr); + + if (act->dtad_difo == NULL) + goto err; + } + + ASSERT(first != NULL); + return first; + +err: + for (act = first; act != NULL; act = next) { + next = act->dtad_next; + dtrace_actdesc_release(act, vstate); + } + + return NULL; +} + +static dtrace_ecbdesc_t *dtrace_dof_ecbdesc(dof_hdr_t *dof, dof_sec_t *sec, + dtrace_vstate_t *vstate, + const cred_t *cr) +{ + dtrace_ecbdesc_t *ep; + dof_ecbdesc_t *ecb; + dtrace_probedesc_t *desc; + dtrace_predicate_t *pred = NULL; + + if (sec->dofs_size < sizeof(dof_ecbdesc_t)) { + dtrace_dof_error(dof, "truncated ECB description"); + return NULL; + } + + if (sec->dofs_align != sizeof(uint64_t)) { + dtrace_dof_error(dof, "bad alignment in ECB description"); + return NULL; + } + + ecb = (dof_ecbdesc_t *)((uintptr_t)dof + (uintptr_t)sec->dofs_offset); + sec = dtrace_dof_sect(dof, DOF_SECT_PROBEDESC, ecb->dofe_probes); + + if (sec == NULL) + return NULL; + + ep = kzalloc(sizeof(dtrace_ecbdesc_t), GFP_KERNEL); + ep->dted_uarg = ecb->dofe_uarg; + desc = &ep->dted_probe; + + if (dtrace_dof_probedesc(dof, sec, desc) == NULL) + goto err; + + if (ecb->dofe_pred != DOF_SECIDX_NONE) { + if ((sec = dtrace_dof_sect(dof, DOF_SECT_DIFOHDR, + ecb->dofe_pred)) == NULL) + goto err; + + if ((pred = dtrace_dof_predicate(dof, sec, vstate, cr)) == NULL) + goto err; + + ep->dted_pred.dtpdd_predicate = pred; + } + + if (ecb->dofe_actions != DOF_SECIDX_NONE) { + if ((sec = dtrace_dof_sect(dof, DOF_SECT_ACTDESC, + ecb->dofe_actions)) == NULL) + goto err; + + ep->dted_action = dtrace_dof_actdesc(dof, sec, vstate, cr); + + if (ep->dted_action == NULL) + goto err; + } + + return ep; + +err: + if (pred != NULL) + dtrace_predicate_release(pred, vstate); + kfree(ep); + return NULL; +} + +/* + * Apply the relocations from the specified 'sec' (a DOF_SECT_URELHDR) to the + * specified DOF. At present, this amounts to simply adding 'ubase' to the + * site of any user SETX relocations to account for load object base address. + * In the future, if we need other relocations, this function can be extended. + */ +static int dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase) +{ + uintptr_t daddr = (uintptr_t)dof; + dof_relohdr_t *dofr = (dof_relohdr_t *)(uintptr_t)(daddr + + sec->dofs_offset); + dof_sec_t *ss, *rs, *ts; + dof_relodesc_t *r; + uint_t i, n; + + if (sec->dofs_size < sizeof(dof_relohdr_t) || + sec->dofs_align != sizeof(dof_secidx_t)) { + dtrace_dof_error(dof, "invalid relocation header"); + return -1; + } + + ss = dtrace_dof_sect(dof, DOF_SECT_STRTAB, dofr->dofr_strtab); + rs = dtrace_dof_sect(dof, DOF_SECT_RELTAB, dofr->dofr_relsec); + ts = dtrace_dof_sect(dof, DOF_SECT_NONE, dofr->dofr_tgtsec); + + if (ss == NULL || rs == NULL || ts == NULL) + return -1; /* dtrace_dof_error() has been called already */ + + if (rs->dofs_entsize < sizeof(dof_relodesc_t) || + rs->dofs_align != sizeof(uint64_t)) { + dtrace_dof_error(dof, "invalid relocation section"); + return -1; + } + + r = (dof_relodesc_t *)(uintptr_t)(daddr + rs->dofs_offset); + /* + * Was: n = rs->dofs_size / rs->dofs_entsize; + * but on Linux we need to use a macro for the division to handle the + * possible case of 64-bit division on a 32-bit kernel. + */ + n = rs->dofs_size; + do_div(n, rs->dofs_entsize); + + for (i = 0; i < n; i++) { + uintptr_t taddr = daddr + ts->dofs_offset + + r->dofr_offset; + + switch (r->dofr_type) { + case DOF_RELO_NONE: + break; + case DOF_RELO_SETX: + if (r->dofr_offset >= ts->dofs_size || + r->dofr_offset + sizeof(uint64_t) > + ts->dofs_size) { + dtrace_dof_error(dof, "bad relocation offset"); + return -1; + } + + if (!IS_ALIGNED(taddr, sizeof(uint64_t))) { + dtrace_dof_error(dof, "misaligned setx relo"); + return -1; + } + + *(uint64_t *)taddr += ubase; + break; + default: + dtrace_dof_error(dof, "invalid relocation type"); + return -1; + } + + r = (dof_relodesc_t *)((uintptr_t)r + rs->dofs_entsize); + } + + return 0; +} + +/* + * The dof_hdr_t passed to dtrace_dof_slurp() should be a partially validated + * header: it should be at the front of a memory region that is at least + * sizeof(dof_hdr_t) in size -- and then at least dof_hdr.dofh_loadsz in + * size. It need not be validated in any other way. + */ +int dtrace_dof_slurp(dof_hdr_t *dof, dtrace_vstate_t *vstate, const cred_t *cr, + dtrace_enabling_t **enabp, uint64_t ubase, int noprobes) +{ + uint64_t len = dof->dofh_loadsz, seclen; + uintptr_t daddr = (uintptr_t)dof; + dtrace_ecbdesc_t *ep; + dtrace_enabling_t *enab; + uint_t i; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(dof->dofh_loadsz >= sizeof(dof_hdr_t)); + + /* + * Check the DOF header identification bytes. In addition to checking + * valid settings, we also verify that unused bits/bytes are zeroed so + * we can use them later without fear of regressing existing binaries. + */ + if (memcmp(&dof->dofh_ident[DOF_ID_MAG0], DOF_MAG_STRING, + DOF_MAG_STRLEN) != 0) { + dtrace_dof_error(dof, "DOF magic string mismatch"); + return -1; + } + + if (dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_ILP32 && + dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_LP64) { + dtrace_dof_error(dof, "DOF has invalid data model"); + return -1; + } + + if (dof->dofh_ident[DOF_ID_ENCODING] != DOF_ENCODE_NATIVE) { + dtrace_dof_error(dof, "DOF encoding mismatch"); + return -1; + } + + if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 && + dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_2) { + dtrace_dof_error(dof, "DOF version mismatch"); + return -1; + } + + if (dof->dofh_ident[DOF_ID_DIFVERS] != DIF_VERSION_2) { + dtrace_dof_error(dof, "DOF uses unsupported instruction set"); + return -1; + } + + if (dof->dofh_ident[DOF_ID_DIFIREG] > DIF_DIR_NREGS) { + dtrace_dof_error(dof, "DOF uses too many integer registers"); + return -1; + } + + if (dof->dofh_ident[DOF_ID_DIFTREG] > DIF_DTR_NREGS) { + dtrace_dof_error(dof, "DOF uses too many tuple registers"); + return -1; + } + + for (i = DOF_ID_PAD; i < DOF_ID_SIZE; i++) { + if (dof->dofh_ident[i] != 0) { + dtrace_dof_error(dof, "DOF has invalid ident byte set"); return -1; + } + } + + if (dof->dofh_flags & ~DOF_FL_VALID) { + dtrace_dof_error(dof, "DOF has invalid flag bits set"); + return -1; + } + + if (dof->dofh_secsize == 0) { + dtrace_dof_error(dof, "zero section header size"); + return -1; + } + + /* + * Check that the section headers don't exceed the amount of DOF + * data. Note that we cast the section size and number of sections + * to uint64_t's to prevent possible overflow in the multiplication. + */ + seclen = (uint64_t)dof->dofh_secnum * (uint64_t)dof->dofh_secsize; + + if (dof->dofh_secoff > len || seclen > len || + dof->dofh_secoff + seclen > len) { + dtrace_dof_error(dof, "truncated section headers"); + return -1; + } + + if (!IS_ALIGNED(dof->dofh_secoff, sizeof(uint64_t))) { + dtrace_dof_error(dof, "misaligned section headers"); + return -1; + } + + if (!IS_ALIGNED(dof->dofh_secsize, sizeof(uint64_t))) { + dtrace_dof_error(dof, "misaligned section size"); + return -1; + } + + /* + * Take an initial pass through the section headers to be sure that + * the headers don't have stray offsets. If the 'noprobes' flag is + * set, do not permit sections relating to providers, probes, or args. + */ + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = + (dof_sec_t *)(daddr + + (uintptr_t)dof->dofh_secoff + + i * dof->dofh_secsize); + + if (noprobes) { + switch (sec->dofs_type) { + case DOF_SECT_PROVIDER: + case DOF_SECT_PROBES: + case DOF_SECT_PRARGS: + case DOF_SECT_PROFFS: + dtrace_dof_error( + dof, "illegal sections for enabling"); + return -1; + } + } + + if (DOF_SEC_ISLOADABLE(sec->dofs_type) && + !(sec->dofs_flags & DOF_SECF_LOAD)) { + dtrace_dof_error( + dof, "loadable section with load flag unset"); + return -1; + } + + /* + * Just ignore non-loadable sections. + */ + if (!(sec->dofs_flags & DOF_SECF_LOAD)) + continue; + + if (sec->dofs_align & (sec->dofs_align - 1)) { + dtrace_dof_error(dof, "bad section alignment"); + return -1; + } + + if (sec->dofs_offset & (sec->dofs_align - 1)) { + dtrace_dof_error(dof, "misaligned section"); + return -1; + } + + if (sec->dofs_offset > len || sec->dofs_size > len || + sec->dofs_offset + sec->dofs_size > len) { + dtrace_dof_error(dof, "corrupt section header"); + return -1; + } + + if (sec->dofs_type == DOF_SECT_STRTAB && *((char *)daddr + + sec->dofs_offset + sec->dofs_size - 1) != '\0') { + dtrace_dof_error(dof, "non-terminating string table"); + return -1; + } + } + + /* + * Take a second pass through the sections and locate and perform any + * relocations that are present. We do this after the first pass to + * be sure that all sections have had their headers validated. + */ + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = + (dof_sec_t *)(daddr + + (uintptr_t)dof->dofh_secoff + + i * dof->dofh_secsize); + + /* + * Skip sections that are not loadable. + */ + if (!(sec->dofs_flags & DOF_SECF_LOAD)) + continue; + + switch (sec->dofs_type) { + case DOF_SECT_URELHDR: + if (dtrace_dof_relocate(dof, sec, ubase) != 0) + return -1; + break; + } + } + + if ((enab = *enabp) == NULL) + enab = *enabp = dtrace_enabling_create(vstate); + + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = + (dof_sec_t *)(daddr + + (uintptr_t)dof->dofh_secoff + + i * dof->dofh_secsize); + + if (sec->dofs_type != DOF_SECT_ECBDESC) + continue; + + if ((ep = dtrace_dof_ecbdesc(dof, sec, vstate, cr)) == NULL) { + dtrace_enabling_destroy(enab); + *enabp = NULL; + return -1; + } + + dtrace_enabling_add(enab, ep); + } + + return 0; +} + +/* + * Process DOF for any options. This should be called after the DOF has been + * processed by dtrace_dof_slurp(). + */ +int dtrace_dof_options(dof_hdr_t *dof, dtrace_state_t *state) +{ + int i, rval; + uint32_t entsize; + size_t offs; + dof_optdesc_t *desc; + + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)((uintptr_t)dof + + (uintptr_t)dof->dofh_secoff + + i * dof->dofh_secsize); + + if (sec->dofs_type != DOF_SECT_OPTDESC) + continue; + + if (sec->dofs_align != sizeof(uint64_t)) { + dtrace_dof_error( + dof, "bad alignment in option description"); + return -EINVAL; + } + + if ((entsize = sec->dofs_entsize) == 0) { + dtrace_dof_error(dof, "zeroed option entry size"); + return -EINVAL; + } + + if (entsize < sizeof(dof_optdesc_t)) { + dtrace_dof_error(dof, "bad option entry size"); + return -EINVAL; + } + + for (offs = 0; offs < sec->dofs_size; offs += entsize) { + desc = (dof_optdesc_t *)((uintptr_t)dof + + (uintptr_t)sec->dofs_offset + + offs); + + if (desc->dofo_strtab != DOF_SECIDX_NONE) { + dtrace_dof_error( + dof, "non-zero option string"); + return -EINVAL; + } + + if (desc->dofo_value == DTRACEOPT_UNSET) { + dtrace_dof_error(dof, "unset option"); + return -EINVAL; + } + + if ((rval = dtrace_state_option( + state, desc->dofo_option, + desc->dofo_value)) != 0) { + dtrace_dof_error(dof, "rejected option"); + return rval; + } + } + } + + return 0; +} diff --git a/dtrace/dtrace_ecb.c b/dtrace/dtrace_ecb.c new file mode 100644 index 000000000000..a9950cd56485 --- /dev/null +++ b/dtrace/dtrace_ecb.c @@ -0,0 +1,845 @@ +/* + * FILE: dtrace_ecb.c + * DESCRIPTION: Dynamic Tracing: ECB functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace.h" + +dtrace_ecb_t *dtrace_ecb_create_cache; + +static dtrace_action_t *dtrace_ecb_aggregation_create(dtrace_ecb_t *ecb, + dtrace_actdesc_t *desc) +{ + dtrace_aggregation_t *agg; + size_t size = sizeof(uint64_t); + int ntuple = desc->dtad_ntuple; + dtrace_action_t *act; + dtrace_recdesc_t *frec; + dtrace_aggid_t aggid; + dtrace_state_t *state = ecb->dte_state; + int err; + + agg = kzalloc(sizeof(dtrace_aggregation_t), GFP_KERNEL); + agg->dtag_ecb = ecb; + + ASSERT(DTRACEACT_ISAGG(desc->dtad_kind)); + + switch (desc->dtad_kind) { + case DTRACEAGG_MIN: + agg->dtag_initial = UINT64_MAX; + agg->dtag_aggregate = dtrace_aggregate_min; + break; + + case DTRACEAGG_MAX: + agg->dtag_initial = UINT64_MIN; + agg->dtag_aggregate = dtrace_aggregate_max; + break; + + case DTRACEAGG_COUNT: + agg->dtag_aggregate = dtrace_aggregate_count; + break; + + case DTRACEAGG_QUANTIZE: + agg->dtag_aggregate = dtrace_aggregate_quantize; + size = (((sizeof(uint64_t) * NBBY) - 1) * 2 + 1) * + sizeof(uint64_t); + break; + + case DTRACEAGG_LQUANTIZE: { + uint16_t step = DTRACE_LQUANTIZE_STEP(desc->dtad_arg); + uint16_t levels = + DTRACE_LQUANTIZE_LEVELS(desc->dtad_arg); + + agg->dtag_initial = desc->dtad_arg; + agg->dtag_aggregate = dtrace_aggregate_lquantize; + + if (step == 0 || levels == 0) + goto err; + + size = levels * sizeof(uint64_t) + 3 * sizeof(uint64_t); + break; + } + + case DTRACEAGG_AVG: + agg->dtag_aggregate = dtrace_aggregate_avg; + size = sizeof(uint64_t) * 2; + break; + + case DTRACEAGG_STDDEV: + agg->dtag_aggregate = dtrace_aggregate_stddev; + size = sizeof(uint64_t) * 4; + break; + + case DTRACEAGG_SUM: + agg->dtag_aggregate = dtrace_aggregate_sum; + break; + + default: + goto err; + } + + agg->dtag_action.dta_rec.dtrd_size = size; + + if (ntuple == 0) + goto err; + + for (act = ecb->dte_action_last; act != NULL; act = act->dta_prev) { + if (DTRACEACT_ISAGG(act->dta_kind)) + break; + + if (--ntuple == 0) { + agg->dtag_first = act; + goto success; + } + } + + ASSERT(ntuple != 0); +err: + kfree(agg); + return NULL; + +success: + ASSERT(ecb->dte_action_last != NULL); + act = ecb->dte_action_last; + + if (act->dta_kind == DTRACEACT_DIFEXPR) { + ASSERT(act->dta_difo != NULL); + + if (act->dta_difo->dtdo_rtype.dtdt_size == 0) + agg->dtag_hasarg = 1; + } + + /* + * Get an ID for the aggregation (add it to the idr). + */ + mutex_unlock(&dtrace_lock); +again: + idr_pre_get(&state->dts_agg_idr, __GFP_NOFAIL); + mutex_lock(&dtrace_lock); + + err = idr_get_new(&state->dts_agg_idr, agg, &aggid); + if (err == -EAGAIN) { + mutex_unlock(&dtrace_lock); + goto again; + } + + agg->dtag_id = aggid; +printk(KERN_INFO "New aggregation: %u (for ECB %u)\n", aggid, ecb->dte_epid); + + frec = &agg->dtag_first->dta_rec; + if (frec->dtrd_alignment < sizeof(dtrace_aggid_t)) + frec->dtrd_alignment = sizeof(dtrace_aggid_t); + + for (act = agg->dtag_first; act != NULL; act = act->dta_next) { + ASSERT(!act->dta_intuple); + + act->dta_intuple = 1; + } + + return &agg->dtag_action; +} + +void dtrace_ecb_aggregation_destroy(dtrace_ecb_t *ecb, dtrace_action_t *act) +{ + dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act; + dtrace_state_t *state = ecb->dte_state; + + ASSERT(DTRACEACT_ISAGG(act->dta_kind)); + + idr_remove(&state->dts_agg_idr, agg->dtag_id); + + kfree(agg); +} + +static int dtrace_ecb_action_add(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc) +{ + dtrace_action_t *action, *last; + dtrace_difo_t *dp = desc->dtad_difo; + uint32_t size = 0, align = sizeof(uint8_t), mask; + uint16_t format = 0; + dtrace_recdesc_t *rec; + dtrace_state_t *state = ecb->dte_state; + dtrace_optval_t *opt = state->dts_options, nframes, strsize; + uint64_t arg = desc->dtad_arg; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(ecb->dte_action == NULL || ecb->dte_action->dta_refcnt == 1); + + if (DTRACEACT_ISAGG(desc->dtad_kind)) { + dtrace_action_t *act; + + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + if (act->dta_kind == DTRACEACT_COMMIT) + return -EINVAL; + + if (act->dta_kind == DTRACEACT_SPECULATE) + return -EINVAL; + } + + action = dtrace_ecb_aggregation_create(ecb, desc); + if (action == NULL) + return -EINVAL; + } else { + if (DTRACEACT_ISDESTRUCTIVE(desc->dtad_kind) || + (desc->dtad_kind == DTRACEACT_DIFEXPR && + dp != NULL && dp->dtdo_destructive)) + state->dts_destructive = 1; + + switch (desc->dtad_kind) { + case DTRACEACT_PRINTF: + case DTRACEACT_PRINTA: + case DTRACEACT_SYSTEM: + case DTRACEACT_FREOPEN: + if ((void *)(uintptr_t)arg == NULL) { + ASSERT(desc->dtad_kind == DTRACEACT_PRINTA); + + format = 0; + } else { + ASSERT((void *)(uintptr_t)arg != NULL); +#ifdef FIXME + ASSERT(arg > KERNELBASE); +#endif + + format = dtrace_format_add( + state, (char *)(uintptr_t)arg); + } + + case DTRACEACT_LIBACT: + case DTRACEACT_DIFEXPR: + if (dp == NULL) + return -EINVAL; + + if ((size = dp->dtdo_rtype.dtdt_size) != 0) + break; + + if (dp->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) { + if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) + return -EINVAL; + + size = opt[DTRACEOPT_STRSIZE]; + } + + break; + + case DTRACEACT_STACK: + if ((nframes = arg) == 0) { + nframes = opt[DTRACEOPT_STACKFRAMES]; + + ASSERT(nframes > 0); + + arg = nframes; + } + + size = nframes * sizeof(uint64_t); + break; + + case DTRACEACT_JSTACK: + if ((strsize = DTRACE_USTACK_STRSIZE(arg)) == 0) + strsize = opt[DTRACEOPT_JSTACKSTRSIZE]; + + if ((nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) + nframes = opt[DTRACEOPT_JSTACKFRAMES]; + + arg = DTRACE_USTACK_ARG(nframes, strsize); + + case DTRACEACT_USTACK: + if (desc->dtad_kind != DTRACEACT_JSTACK && + (nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) { + strsize = DTRACE_USTACK_STRSIZE(arg); + nframes = opt[DTRACEOPT_USTACKFRAMES]; + + ASSERT(nframes > 0); + + arg = DTRACE_USTACK_ARG(nframes, strsize); + } + + size = (nframes + 1) * sizeof(uint64_t); + size += DTRACE_USTACK_STRSIZE(arg); + size = P2ROUNDUP(size, (uint32_t)(sizeof(uintptr_t))); + + break; + + case DTRACEACT_SYM: + case DTRACEACT_MOD: + if (dp == NULL || ((size = dp->dtdo_rtype.dtdt_size) != + sizeof(uint64_t)) || + (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) + return -EINVAL; + + break; + + case DTRACEACT_USYM: + case DTRACEACT_UMOD: + case DTRACEACT_UADDR: + if (dp == NULL || + (dp->dtdo_rtype.dtdt_size != sizeof(uint64_t)) || + (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) + return -EINVAL; + + size = 2 * sizeof(uint64_t); + + break; + + case DTRACEACT_STOP: + case DTRACEACT_BREAKPOINT: + case DTRACEACT_PANIC: + break; + + case DTRACEACT_CHILL: + case DTRACEACT_DISCARD: + case DTRACEACT_RAISE: + if (dp == NULL) + return -EINVAL; + + break; + + case DTRACEACT_EXIT: + if (dp == NULL || (size = dp->dtdo_rtype.dtdt_size) != + sizeof(int) || + (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) + return -EINVAL; + + break; + + case DTRACEACT_SPECULATE: + if (ecb->dte_size > sizeof(dtrace_epid_t)) + return -EINVAL; + + if (dp == NULL) + return -EINVAL; + + state->dts_speculates = 1; + + break; + + case DTRACEACT_COMMIT: { + dtrace_action_t *act = ecb->dte_action; + + for (; act != NULL; act = act->dta_next) { + if (act->dta_kind == DTRACEACT_COMMIT) + return -EINVAL; + } + + if (dp == NULL) + return -EINVAL; + + break; + } + + default: + return -EINVAL; + } + + if (size != 0 || desc->dtad_kind == DTRACEACT_SPECULATE) { + dtrace_action_t *act = ecb->dte_action; + + for (; act != NULL; act = act->dta_next) { + if (act->dta_kind == DTRACEACT_COMMIT) + return -EINVAL; + } + } + + action = kzalloc(sizeof(dtrace_action_t), GFP_KERNEL); + action->dta_rec.dtrd_size = size; + } + + action->dta_refcnt = 1; + rec = &action->dta_rec; + size = rec->dtrd_size; + + for (mask = sizeof(uint64_t) - 1; size != 0 && mask > 0; mask >>= 1) { + if (!(size & mask)) { + align = mask + 1; + + break; + } + } + + action->dta_kind = desc->dtad_kind; + + if ((action->dta_difo = dp) != NULL) + dtrace_difo_hold(dp); + + rec->dtrd_action = action->dta_kind; + rec->dtrd_arg = arg; + rec->dtrd_uarg = desc->dtad_uarg; + rec->dtrd_alignment = (uint16_t)align; + rec->dtrd_format = format; + + if ((last = ecb->dte_action_last) != NULL) { + ASSERT(ecb->dte_action != NULL); + + action->dta_prev = last; + last->dta_next = action; + } else { + ASSERT(ecb->dte_action == NULL); + + ecb->dte_action = action; + } + + ecb->dte_action_last = action; + + return 0; +} + +static void dtrace_ecb_action_remove(dtrace_ecb_t *ecb) +{ + dtrace_action_t *act = ecb->dte_action, *next; + dtrace_vstate_t *vstate = &ecb->dte_state->dts_vstate; + dtrace_difo_t *dp; + uint16_t format; + + if (act != NULL && act->dta_refcnt > 1) { + ASSERT(act->dta_next == NULL || act->dta_next->dta_refcnt == 1); + + act->dta_refcnt--; + } else { + for (; act != NULL; act = next) { + next = act->dta_next; + ASSERT(next != NULL || act == ecb->dte_action_last); + ASSERT(act->dta_refcnt == 1); + + if ((format = act->dta_rec.dtrd_format) != 0) + dtrace_format_remove(ecb->dte_state, format); + + if ((dp = act->dta_difo) != NULL) + dtrace_difo_release(dp, vstate); + + if (DTRACEACT_ISAGG(act->dta_kind)) + dtrace_ecb_aggregation_destroy(ecb, act); + else + kfree(act); + } + } + + ecb->dte_action = NULL; + ecb->dte_action_last = NULL; + ecb->dte_size = sizeof(dtrace_epid_t); +} + +/* + * Disable the ECB by removing it from its probe. + */ +void dtrace_ecb_disable(dtrace_ecb_t *ecb) +{ + dtrace_ecb_t *pecb, *prev = NULL; + dtrace_probe_t *probe = ecb->dte_probe; + + ASSERT(mutex_is_locked(&dtrace_lock)); + + if (probe == NULL) + return; + + for (pecb = probe->dtpr_ecb; pecb != NULL; pecb = pecb->dte_next) { + if (pecb == ecb) + break; + + prev = pecb; + } + + ASSERT(pecb != NULL); + + if (prev == NULL) + probe->dtpr_ecb = ecb->dte_next; + else + prev->dte_next = ecb->dte_next; + + if (ecb == probe->dtpr_ecb_last) { + ASSERT(ecb->dte_next == NULL); + probe->dtpr_ecb_last = prev; + } + + /* + * The ECB has been disconnected from the probe; now sync to assure + * that all CPUs have seen the change before returning. + */ + dtrace_sync(); + + if (probe->dtpr_ecb == NULL) { + /* + * That was the last ECB on the probe; clear the predicate + * cache ID for the probe, disable it and sync one more time + * to assure that we'll never hit it again. + */ + dtrace_provider_t *prov = probe->dtpr_provider; + + ASSERT(ecb->dte_next == NULL); + ASSERT(probe->dtpr_ecb_last == NULL); + + probe->dtpr_predcache = DTRACE_CACHEIDNONE; + prov->dtpv_pops.dtps_disable(prov->dtpv_arg, + probe->dtpr_id, probe->dtpr_arg); + + dtrace_sync(); + } else { + /* + * There is at least one ECB remaining on the probe. If there + * is _exactly_ one, set the probe's predicate cache ID to be + * the predicate cache ID of the remaining ECB. + */ + ASSERT(probe->dtpr_ecb_last != NULL); + ASSERT(probe->dtpr_predcache == DTRACE_CACHEIDNONE); + + if (probe->dtpr_ecb == probe->dtpr_ecb_last) { + dtrace_predicate_t *p = + probe->dtpr_ecb->dte_predicate; + + ASSERT(probe->dtpr_ecb->dte_next == NULL); + + if (p != NULL) + probe->dtpr_predcache = p->dtp_cacheid; + } + + ecb->dte_next = NULL; + } +} + +static dtrace_ecb_t *dtrace_ecb_add(dtrace_state_t *state, + dtrace_probe_t *probe) +{ + dtrace_ecb_t *ecb; + dtrace_epid_t epid; + + ASSERT(mutex_is_locked(&dtrace_lock)); + + ecb = kzalloc(sizeof(dtrace_ecb_t), GFP_KERNEL); + ecb->dte_predicate = NULL; + ecb->dte_probe = probe; + ecb->dte_size = ecb->dte_needed = sizeof(dtrace_epid_t); + ecb->dte_alignment = sizeof(dtrace_epid_t); + + epid = state->dts_epid++; + + if (epid - 1 >= state->dts_necbs) { + dtrace_ecb_t **oecbs = state->dts_ecbs, **ecbs; + int necbs = state->dts_necbs << 1; + + ASSERT(epid == state->dts_necbs + 1); + + if (necbs == 0) { + ASSERT(oecbs == NULL); + + necbs = 1; + } + + ecbs = kcalloc(necbs, sizeof(*ecbs), GFP_KERNEL); + if (oecbs != NULL) + memcpy(ecbs, oecbs, state->dts_necbs * sizeof(*ecbs)); + + dtrace_membar_producer(); + + state->dts_ecbs = ecbs; + + if (oecbs != NULL) { + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) + dtrace_sync(); + + kfree(oecbs); + } + + dtrace_membar_producer(); + + state->dts_necbs = necbs; + } + + ecb->dte_state = state; + + ASSERT(state->dts_ecbs[epid - 1] == NULL); + + dtrace_membar_producer(); + + state->dts_ecbs[(ecb->dte_epid = epid) - 1] = ecb; + +printk(KERN_INFO "ecb_add: ECB %u for probe ID %u\n", epid, probe ? probe->dtpr_id : 0xffff); + return ecb; +} + +static dtrace_ecb_t *dtrace_ecb_create(dtrace_state_t *state, + dtrace_probe_t *probe, + dtrace_enabling_t *enab) +{ + dtrace_ecb_t *ecb; + dtrace_predicate_t *pred; + dtrace_actdesc_t *act; + dtrace_provider_t *prov; + dtrace_ecbdesc_t *desc = enab->dten_current; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(state != NULL); + + ecb = dtrace_ecb_add(state, probe); + ecb->dte_uarg = desc->dted_uarg; + + if ((pred = desc->dted_pred.dtpdd_predicate) != NULL) { + dtrace_predicate_hold(pred); + ecb->dte_predicate = pred; + } + + if (probe != NULL) { + prov = probe->dtpr_provider; + + if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLPROC) && + (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER)) + ecb->dte_cond |= DTRACE_COND_OWNER; + + if (!(state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) && + (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_KERNEL)) + ecb->dte_cond |= DTRACE_COND_USERMODE; + } + + if (dtrace_ecb_create_cache != NULL) { + dtrace_ecb_t *cached = dtrace_ecb_create_cache; + dtrace_action_t *act = cached->dte_action; + + if (act != NULL) { + ASSERT(act->dta_refcnt > 0); + + act->dta_refcnt++; + ecb->dte_action = act; + ecb->dte_action_last = cached->dte_action_last; + ecb->dte_needed = cached->dte_needed; + ecb->dte_size = cached->dte_size; + ecb->dte_alignment = cached->dte_alignment; + } + + return ecb; + } + + for (act = desc->dted_action; act != NULL; act = act->dtad_next) { + if ((enab->dten_error = dtrace_ecb_action_add(ecb, act)) != 0) { + dtrace_ecb_destroy(ecb); + return NULL; + } + } + + dtrace_ecb_resize(ecb); + + return (dtrace_ecb_create_cache = ecb); +} + +int dtrace_ecb_create_enable(dtrace_probe_t *probe, void *arg) +{ + dtrace_ecb_t *ecb; + dtrace_enabling_t *enab = arg; + dtrace_state_t *state = enab->dten_vstate->dtvs_state; + + ASSERT(state != NULL); + + if (probe != NULL && probe->dtpr_gen < enab->dten_probegen) + return DTRACE_MATCH_NEXT; + + if ((ecb = dtrace_ecb_create(state, probe, enab)) == NULL) + return DTRACE_MATCH_DONE; + + if (dtrace_ecb_enable(ecb) < 0) + return DTRACE_MATCH_FAIL; + + return DTRACE_MATCH_NEXT; +} + +void dtrace_ecb_destroy(dtrace_ecb_t *ecb) +{ + dtrace_state_t *state = ecb->dte_state; + dtrace_vstate_t *vstate = &state->dts_vstate; + dtrace_predicate_t *pred; + dtrace_epid_t epid = ecb->dte_epid; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(ecb->dte_next == NULL); + ASSERT(ecb->dte_probe == NULL || ecb->dte_probe->dtpr_ecb != ecb); + + if ((pred = ecb->dte_predicate) != NULL) + dtrace_predicate_release(pred, vstate); + + dtrace_ecb_action_remove(ecb); + + ASSERT(state->dts_ecbs[epid - 1] == ecb); + state->dts_ecbs[epid - 1] = NULL; + + kfree(ecb); +} + +void dtrace_ecb_resize(dtrace_ecb_t *ecb) +{ + uint32_t maxalign = sizeof(dtrace_epid_t); + uint32_t align = sizeof(uint8_t), offs, diff; + dtrace_action_t *act; + int wastuple = 0; + uint32_t aggbase = UINT32_MAX; + dtrace_state_t *state = ecb->dte_state; + + /* + * If we record anything, we always record the epid. (And we always + * record it first.) + */ + offs = sizeof(dtrace_epid_t); + ecb->dte_size = ecb->dte_needed = sizeof(dtrace_epid_t); + + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + dtrace_recdesc_t *rec = &act->dta_rec; + + if ((align = rec->dtrd_alignment) > maxalign) + maxalign = align; + + if (!wastuple && act->dta_intuple) { + /* + * This is the first record in a tuple. Align the + * offset to be at offset 4 in an 8-byte aligned + * block. + */ + diff = offs + sizeof(dtrace_aggid_t); + + if ((diff = (diff & (sizeof(uint64_t) - 1)))) + offs += sizeof(uint64_t) - diff; + + aggbase = offs - sizeof(dtrace_aggid_t); + ASSERT(!(aggbase & (sizeof(uint64_t) - 1))); + } + + if (rec->dtrd_size != 0 && (diff = (offs & (align - 1)))) { + /* + * The current offset is not properly aligned; align it. + */ + offs += align - diff; + } + + rec->dtrd_offset = offs; + + if (offs + rec->dtrd_size > ecb->dte_needed) { + ecb->dte_needed = offs + rec->dtrd_size; + + if (ecb->dte_needed > state->dts_needed) + state->dts_needed = ecb->dte_needed; + } + + if (DTRACEACT_ISAGG(act->dta_kind)) { + dtrace_aggregation_t *agg = + (dtrace_aggregation_t *)act; + dtrace_action_t *first = agg->dtag_first, + *prev; + + ASSERT(rec->dtrd_size != 0 && first != NULL); + ASSERT(wastuple); + ASSERT(aggbase != UINT32_MAX); + + agg->dtag_base = aggbase; + + while ((prev = first->dta_prev) != NULL && + DTRACEACT_ISAGG(prev->dta_kind)) { + agg = (dtrace_aggregation_t *)prev; + first = agg->dtag_first; + } + + if (prev != NULL) { + offs = prev->dta_rec.dtrd_offset + + prev->dta_rec.dtrd_size; + } else + offs = sizeof(dtrace_epid_t); + + wastuple = 0; + } else { + if (!act->dta_intuple) + ecb->dte_size = offs + rec->dtrd_size; + + offs += rec->dtrd_size; + } + + wastuple = act->dta_intuple; + } + + if ((act = ecb->dte_action) != NULL && + !(act->dta_kind == DTRACEACT_SPECULATE && act->dta_next == NULL) && + ecb->dte_size == sizeof(dtrace_epid_t)) { + /* + * If the size is still sizeof(dtrace_epid_t), then all + * actions store no data; set the size to 0. + */ + ecb->dte_alignment = maxalign; + ecb->dte_size = 0; + + /* + * If the needed space is still sizeof(dtrace_epid_t), then + * all actions need no additional space; set the needed + * size to 0. + */ + if (ecb->dte_needed == sizeof(dtrace_epid_t)) + ecb->dte_needed = 0; + + return; + } + + /* + * Set our alignment, and make sure that the dte_size and dte_needed + * are aligned to the size of an EPID. + */ + ecb->dte_alignment = maxalign; + ecb->dte_size = (ecb->dte_size + (sizeof(dtrace_epid_t) - 1)) & + ~(sizeof(dtrace_epid_t) - 1); + ecb->dte_needed = (ecb->dte_needed + (sizeof(dtrace_epid_t) - 1)) & + ~(sizeof(dtrace_epid_t) - 1); + ASSERT(ecb->dte_size <= ecb->dte_needed); +} + +int dtrace_ecb_enable(dtrace_ecb_t *ecb) +{ + dtrace_probe_t *probe = ecb->dte_probe; + + ASSERT(mutex_is_locked(&cpu_lock)); + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(ecb->dte_next == NULL); + + if (probe == NULL) + return 0; + + if (probe->dtpr_ecb == NULL) { + dtrace_provider_t *prov = probe->dtpr_provider; + + probe->dtpr_ecb = probe->dtpr_ecb_last = ecb; + + + if (ecb->dte_predicate != NULL) + probe->dtpr_predcache = ecb->dte_predicate->dtp_cacheid; + + return prov->dtpv_pops.dtps_enable(prov->dtpv_arg, + probe->dtpr_id, + probe->dtpr_arg); + } else { + ASSERT(probe->dtpr_ecb_last != NULL); + + probe->dtpr_ecb_last->dte_next = ecb; + probe->dtpr_ecb_last = ecb; + probe->dtpr_predcache = 0; + + dtrace_sync(); + + return 0; + } +} + +dtrace_ecb_t *dtrace_epid2ecb(dtrace_state_t *state, dtrace_epid_t id) +{ + dtrace_ecb_t *ecb; + + ASSERT(mutex_is_locked(&dtrace_lock)); + + if (id == 0 || id > state->dts_necbs) + return NULL; + + ASSERT(state->dts_necbs > 0 && state->dts_ecbs != NULL); + ecb = state->dts_ecbs[id - 1]; + ASSERT(ecb == NULL || ecb->dte_epid == id); + + return ecb; +} + +dtrace_aggregation_t *dtrace_aggid2agg(dtrace_state_t *state, + dtrace_aggid_t id) +{ + ASSERT(mutex_is_locked(&dtrace_lock)); + + return idr_find(&state->dts_agg_idr, id); +} diff --git a/dtrace/dtrace_enable.c b/dtrace/dtrace_enable.c new file mode 100644 index 000000000000..38355c2ab8aa --- /dev/null +++ b/dtrace/dtrace_enable.c @@ -0,0 +1,414 @@ +/* + * FILE: dtrace_enable.c + * DESCRIPTION: Dynamic Tracing: enabling functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include +#include + +#include "dtrace.h" + +size_t dtrace_retain_max = 1024; +dtrace_enabling_t *dtrace_retained; +dtrace_genid_t dtrace_retained_gen; + +dtrace_enabling_t *dtrace_enabling_create(dtrace_vstate_t *vstate) +{ + dtrace_enabling_t *enab; + + enab = kzalloc(sizeof (dtrace_enabling_t), GFP_KERNEL); + enab->dten_vstate = vstate; + + return enab; +} + +void dtrace_enabling_add(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb) +{ + dtrace_ecbdesc_t **ndesc; + size_t osize, nsize; + + /* + * We can't add to enablings after we've enabled them, or after we've + * retained them. + */ + ASSERT(enab->dten_probegen == 0); + ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL); + + if (enab->dten_ndesc < enab->dten_maxdesc) { + enab->dten_desc[enab->dten_ndesc++] = ecb; + return; + } + + osize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *); + + if (enab->dten_maxdesc == 0) + enab->dten_maxdesc = 1; + else + enab->dten_maxdesc <<= 1; + + ASSERT(enab->dten_ndesc < enab->dten_maxdesc); + + nsize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *); + ndesc = kzalloc(nsize, GFP_KERNEL); + memcpy(ndesc, enab->dten_desc, osize); + kfree(enab->dten_desc); + + enab->dten_desc = ndesc; + enab->dten_desc[enab->dten_ndesc++] = ecb; +} + +static void dtrace_enabling_addlike(dtrace_enabling_t *enab, + dtrace_ecbdesc_t *ecb, + dtrace_probedesc_t *pd) +{ + dtrace_ecbdesc_t *new; + dtrace_predicate_t *pred; + dtrace_actdesc_t *act; + + /* + * We're going to create a new ECB description that matches the + * specified ECB in every way, but has the specified probe description. + */ + new = kzalloc(sizeof (dtrace_ecbdesc_t), GFP_KERNEL); + + if ((pred = ecb->dted_pred.dtpdd_predicate) != NULL) + dtrace_predicate_hold(pred); + + for (act = ecb->dted_action; act != NULL; act = act->dtad_next) + dtrace_actdesc_hold(act); + + new->dted_action = ecb->dted_action; + new->dted_pred = ecb->dted_pred; + new->dted_probe = *pd; + new->dted_uarg = ecb->dted_uarg; + + dtrace_enabling_add(enab, new); +} + +void dtrace_enabling_dump(dtrace_enabling_t *enab) +{ + int i; + + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_probedesc_t *desc = + &enab->dten_desc[i]->dted_probe; + + pr_info("enabling probe %d (%s:%s:%s:%s)", + i, desc->dtpd_provider, desc->dtpd_mod, + desc->dtpd_func, desc->dtpd_name); + } +} + +void dtrace_enabling_destroy(dtrace_enabling_t *enab) +{ + int i; + dtrace_ecbdesc_t *ep; + dtrace_vstate_t *vstate = enab->dten_vstate; + + ASSERT(mutex_is_locked(&dtrace_lock)); + + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_actdesc_t *act, *next; + dtrace_predicate_t *pred; + + ep = enab->dten_desc[i]; + + if ((pred = ep->dted_pred.dtpdd_predicate) != NULL) + dtrace_predicate_release(pred, vstate); + + for (act = ep->dted_action; act != NULL; act = next) { + next = act->dtad_next; + dtrace_actdesc_release(act, vstate); + } + + kfree(ep); + } + + kfree(enab->dten_desc); + + /* + * If this was a retained enabling, decrement the dts_nretained count + * and remove it from the dtrace_retained list. + */ + if (enab->dten_prev != NULL || enab->dten_next != NULL || + dtrace_retained == enab) { + ASSERT(enab->dten_vstate->dtvs_state != NULL); + ASSERT(enab->dten_vstate->dtvs_state->dts_nretained > 0); + enab->dten_vstate->dtvs_state->dts_nretained--; + dtrace_retained_gen++; + } + + if (enab->dten_prev == NULL) { + if (dtrace_retained == enab) { + dtrace_retained = enab->dten_next; + + if (dtrace_retained != NULL) + dtrace_retained->dten_prev = NULL; + } + } else { + ASSERT(enab != dtrace_retained); + ASSERT(dtrace_retained != NULL); + enab->dten_prev->dten_next = enab->dten_next; + } + + if (enab->dten_next != NULL) { + ASSERT(dtrace_retained != NULL); + enab->dten_next->dten_prev = enab->dten_prev; + } + + kfree(enab); +} + +int dtrace_enabling_retain(dtrace_enabling_t *enab) +{ + dtrace_state_t *state; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL); + ASSERT(enab->dten_vstate != NULL); + + state = enab->dten_vstate->dtvs_state; + ASSERT(state != NULL); + + /* + * We only allow each state to retain dtrace_retain_max enablings. + */ + if (state->dts_nretained >= dtrace_retain_max) + return -ENOSPC; + + state->dts_nretained++; + dtrace_retained_gen++; + + if (dtrace_retained == NULL) { + dtrace_retained = enab; + return 0; + } + + enab->dten_next = dtrace_retained; + dtrace_retained->dten_prev = enab; + dtrace_retained = enab; + + return 0; +} + +int dtrace_enabling_replicate(dtrace_state_t *state, dtrace_probedesc_t *match, + dtrace_probedesc_t *create) +{ + dtrace_enabling_t *new, *enab; + int found = 0, err = -ENOENT; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(strlen(match->dtpd_provider) < DTRACE_PROVNAMELEN); + ASSERT(strlen(match->dtpd_mod) < DTRACE_MODNAMELEN); + ASSERT(strlen(match->dtpd_func) < DTRACE_FUNCNAMELEN); + ASSERT(strlen(match->dtpd_name) < DTRACE_NAMELEN); + + new = dtrace_enabling_create(&state->dts_vstate); + + /* + * Iterate over all retained enablings, looking for enablings that + * match the specified state. + */ + for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { + int i; + + /* + * dtvs_state can only be NULL for helper enablings -- and + * helper enablings can't be retained. + */ + ASSERT(enab->dten_vstate->dtvs_state != NULL); + + if (enab->dten_vstate->dtvs_state != state) + continue; + + /* + * Now iterate over each probe description; we're looking for + * an exact match to the specified probe description. + */ + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_ecbdesc_t *ep = enab->dten_desc[i]; + dtrace_probedesc_t *pd = &ep->dted_probe; + + if (strcmp(pd->dtpd_provider, match->dtpd_provider)) + continue; + + if (strcmp(pd->dtpd_mod, match->dtpd_mod)) + continue; + + if (strcmp(pd->dtpd_func, match->dtpd_func)) + continue; + + if (strcmp(pd->dtpd_name, match->dtpd_name)) + continue; + + /* + * We have a winning probe! Add it to our growing + * enabling. + */ + found = 1; + dtrace_enabling_addlike(new, ep, create); + } + } + + if (!found || (err = dtrace_enabling_retain(new)) != 0) { + dtrace_enabling_destroy(new); + return err; + } + + return 0; +} + +void dtrace_enabling_retract(dtrace_state_t *state) +{ + dtrace_enabling_t *enab, *next; + + ASSERT(mutex_is_locked(&dtrace_lock)); + + /* + * Iterate over all retained enablings, destroy the enablings retained + * for the specified state. + */ + for (enab = dtrace_retained; enab != NULL; enab = next) { + next = enab->dten_next; + + /* + * dtvs_state can only be NULL for helper enablings, and helper + * enablings can't be retained. + */ + ASSERT(enab->dten_vstate->dtvs_state != NULL); + + if (enab->dten_vstate->dtvs_state == state) { + ASSERT(state->dts_nretained > 0); + dtrace_enabling_destroy(enab); + } + } + + ASSERT(state->dts_nretained == 0); +} + +int dtrace_enabling_match(dtrace_enabling_t *enab, int *nmatched) +{ + int i; + int total_matched = 0, matched = 0; + + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_ecbdesc_t *ep = enab->dten_desc[i]; + + enab->dten_current = ep; + enab->dten_error = 0; + + if ((matched = dtrace_probe_enable(&ep->dted_probe, enab)) < 0) + return -EBUSY; + + total_matched += matched; + + if (enab->dten_error != 0) { + if (nmatched == NULL) + pr_warning("dtrace_enabling_match() error on %p: %d\n", (void *)ep, enab->dten_error); + + return enab->dten_error; + } + } + + enab->dten_probegen = dtrace_probegen; + if (nmatched != NULL) + *nmatched = total_matched; + + return 0; +} + +void dtrace_enabling_matchall(void) +{ + dtrace_enabling_t *enab; + + mutex_lock(&cpu_lock); + mutex_lock(&dtrace_lock); + + for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) + (void) dtrace_enabling_match(enab, NULL); + + mutex_unlock(&dtrace_lock); + mutex_unlock(&cpu_lock); +} + +/* + * If an enabling is to be enabled without having matched probes (that is, if + * dtrace_state_go() is to be called on the underlying dtrace_state_t), the + * enabling must be _primed_ by creating an ECB for every ECB description. + * This must be done to assure that we know the number of speculations, the + * number of aggregations, the minimum buffer size needed, etc. before we + * transition out of DTRACE_ACTIVITY_INACTIVE. To do this without actually + * enabling any probes, we create ECBs for every ECB decription, but with a + * NULL probe -- which is exactly what this function does. + */ +void dtrace_enabling_prime(dtrace_state_t *state) +{ + dtrace_enabling_t *enab; + int i; + + for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { + ASSERT(enab->dten_vstate->dtvs_state != NULL); + + if (enab->dten_vstate->dtvs_state != state) + continue; + + /* + * We don't want to prime an enabling more than once, lest + * we allow a malicious user to induce resource exhaustion. + * (The ECBs that result from priming an enabling aren't + * leaked -- but they also aren't deallocated until the + * consumer state is destroyed.) + */ + if (enab->dten_primed) + continue; + + for (i = 0; i < enab->dten_ndesc; i++) { + enab->dten_current = enab->dten_desc[i]; + dtrace_probe_enable(NULL, enab); + } + + enab->dten_primed = 1; + } +} + +void dtrace_enabling_provide(dtrace_provider_t *prv) +{ + int all = 0; + dtrace_genid_t gen; + + if (prv == NULL) { + all = 1; + prv = dtrace_provider; + } + + do { + dtrace_enabling_t *enab; + void *parg = prv->dtpv_arg; + +retry: + gen = dtrace_retained_gen; + for (enab = dtrace_retained; enab != NULL; + enab = enab->dten_next) { + int i; + + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_probedesc_t desc; + + desc = enab->dten_desc[i]->dted_probe; + mutex_unlock(&dtrace_lock); + prv->dtpv_pops.dtps_provide(parg, &desc); + mutex_lock(&dtrace_lock); + + if (gen != dtrace_retained_gen) + goto retry; + } + } + } while (all && (prv = prv->dtpv_next) != NULL); + + mutex_unlock(&dtrace_lock); + dtrace_probe_provide(NULL, all ? NULL : prv); + mutex_lock(&dtrace_lock); +} diff --git a/dtrace/dtrace_fmt.c b/dtrace/dtrace_fmt.c new file mode 100644 index 000000000000..a4bef75eaa50 --- /dev/null +++ b/dtrace/dtrace_fmt.c @@ -0,0 +1,84 @@ +/* + * FILE: dtrace_fmt.c + * DESCRIPTION: Dynamic Tracing: format functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace.h" + +uint16_t dtrace_format_add(dtrace_state_t *state, char *str) +{ + char *fmt, **new; + uint16_t ndx; + + fmt = dtrace_strdup(str); + + for (ndx = 0; ndx < state->dts_nformats; ndx++) { + if (state->dts_formats[ndx] == NULL) { + state->dts_formats[ndx] = fmt; + + return ndx + 1; + } + } + + if (state->dts_nformats == UINT16_MAX) { + kfree(fmt); + + return 0; + } + + ndx = state->dts_nformats++; + new = kmalloc((ndx + 1) * sizeof (char *), GFP_KERNEL); + + if (state->dts_formats != NULL) { + ASSERT(ndx != 0); + memcpy(new, state->dts_formats, ndx * sizeof (char *)); + kfree(state->dts_formats); + } + + state->dts_formats = new; + state->dts_formats[ndx] = fmt; + + return ndx + 1; +} + +void dtrace_format_remove(dtrace_state_t *state, uint16_t format) +{ + char *fmt; + + ASSERT(state->dts_formats != NULL); + ASSERT(format <= state->dts_nformats); + ASSERT(state->dts_formats[format - 1] != NULL); + + fmt = state->dts_formats[format - 1]; + kfree(fmt); + state->dts_formats[format - 1] = NULL; +} + +void dtrace_format_destroy(dtrace_state_t *state) +{ + int i; + + if (state->dts_nformats == 0) { + ASSERT(state->dts_formats == NULL); + return; + } + + ASSERT(state->dts_formats != NULL); + + for (i = 0; i < state->dts_nformats; i++) { + char *fmt = state->dts_formats[i]; + + if (fmt == NULL) + continue; + + kfree(fmt); + } + + kfree(state->dts_formats); + state->dts_nformats = 0; + state->dts_formats = NULL; +} diff --git a/dtrace/dtrace_hash.c b/dtrace/dtrace_hash.c new file mode 100644 index 000000000000..9becd58f7a09 --- /dev/null +++ b/dtrace/dtrace_hash.c @@ -0,0 +1,203 @@ +/* + * FILE: dtrace_hash.c + * DESCRIPTION: Dynamic Tracing: probe hashing functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace.h" + +#define DTRACE_HASHSTR(hash, probe) \ + dtrace_hash_str(*((char **)((uintptr_t)(probe) + (hash)->dth_stroffs))) +#define DTRACE_HASHEQ(hash, lhs, rhs) \ + (strcmp(*((char **)((uintptr_t)(lhs) + (hash)->dth_stroffs)), \ + *((char **)((uintptr_t)(rhs) + (hash)->dth_stroffs))) == 0) + +static uint_t dtrace_hash_str(char *p) +{ + uint_t g; + uint_t hval = 0; + + while (*p) { + hval = (hval << 4) + *p++; + if ((g = (hval & 0xf0000000)) != 0) + hval ^= g >> 24; + + hval &= ~g; + } + + return hval; +} + +dtrace_hash_t *dtrace_hash_create(uintptr_t stroffs, uintptr_t nextoffs, + uintptr_t prevoffs) +{ + dtrace_hash_t *hash = kzalloc(sizeof (dtrace_hash_t), GFP_KERNEL); + + hash->dth_stroffs = stroffs; + hash->dth_nextoffs = nextoffs; + hash->dth_prevoffs = prevoffs; + + hash->dth_size = 1; + hash->dth_mask = hash->dth_size - 1; + + hash->dth_tab = kzalloc(hash->dth_size * + sizeof (dtrace_hashbucket_t *), GFP_KERNEL); + + return hash; +} + +static void dtrace_hash_resize(dtrace_hash_t *hash) +{ + int size = hash->dth_size, i, ndx; + int new_size = hash->dth_size << 1; + int new_mask = new_size - 1; + dtrace_hashbucket_t **new_tab, *bucket, *next; + + ASSERT((new_size & new_mask) == 0); + + new_tab = kzalloc(new_size * sizeof (void *), GFP_KERNEL); + + for (i = 0; i < size; i++) { + for (bucket = hash->dth_tab[i]; bucket != NULL; + bucket = next) { + dtrace_probe_t *probe = bucket->dthb_chain; + + ASSERT(probe != NULL); + ndx = DTRACE_HASHSTR(hash, probe) & new_mask; + + next = bucket->dthb_next; + bucket->dthb_next = new_tab[ndx]; + new_tab[ndx] = bucket; + } + } + + kfree(hash->dth_tab); + hash->dth_tab = new_tab; + hash->dth_size = new_size; + hash->dth_mask = new_mask; +} + +void dtrace_hash_add(dtrace_hash_t *hash, dtrace_probe_t *new) +{ + int hashval = DTRACE_HASHSTR(hash, new); + int ndx = hashval & hash->dth_mask; + dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; + dtrace_probe_t **nextp, **prevp; + + for (; bucket != NULL; bucket = bucket->dthb_next) { + if (DTRACE_HASHEQ(hash, bucket->dthb_chain, new)) + goto add; + } + + if ((hash->dth_nbuckets >> 1) > hash->dth_size) { + dtrace_hash_resize(hash); + dtrace_hash_add(hash, new); + return; + } + + bucket = kzalloc(sizeof (dtrace_hashbucket_t), GFP_KERNEL); + bucket->dthb_next = hash->dth_tab[ndx]; + hash->dth_tab[ndx] = bucket; + hash->dth_nbuckets++; + +add: + nextp = DTRACE_HASHNEXT(hash, new); + + ASSERT(*nextp == NULL && *(DTRACE_HASHPREV(hash, new)) == NULL); + + *nextp = bucket->dthb_chain; + + if (bucket->dthb_chain != NULL) { + prevp = DTRACE_HASHPREV(hash, bucket->dthb_chain); + + ASSERT(*prevp == NULL); + + *prevp = new; + } + + bucket->dthb_chain = new; + bucket->dthb_len++; +} + +dtrace_probe_t *dtrace_hash_lookup(dtrace_hash_t *hash, + dtrace_probe_t *template) +{ + int hashval = DTRACE_HASHSTR(hash, template); + int ndx = hashval & hash->dth_mask; + dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; + + for (; bucket != NULL; bucket = bucket->dthb_next) { + if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template)) + return bucket->dthb_chain; + } + + return NULL; +} + +int dtrace_hash_collisions(dtrace_hash_t *hash, dtrace_probe_t *template) +{ + int hashval = DTRACE_HASHSTR(hash, template); + int ndx = hashval & hash->dth_mask; + dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; + + for (; bucket != NULL; bucket = bucket->dthb_next) { + if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template)) + return bucket->dthb_len; + } + + return 0; +} + +void dtrace_hash_remove(dtrace_hash_t *hash, dtrace_probe_t *probe) +{ + int ndx = DTRACE_HASHSTR(hash, probe) & + hash->dth_mask; + dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; + dtrace_probe_t **prevp = DTRACE_HASHPREV(hash, probe); + dtrace_probe_t **nextp = DTRACE_HASHNEXT(hash, probe); + + for (; bucket != NULL; bucket = bucket->dthb_next) { + if (DTRACE_HASHEQ(hash, bucket->dthb_chain, probe)) + break; + } + + ASSERT(bucket != NULL); + + if (*prevp == NULL) { + if (*nextp == NULL) { + /* + * This is the last probe in the bucket; we can remove + * the bucket. + */ + dtrace_hashbucket_t *b = hash->dth_tab[ndx]; + + ASSERT(bucket->dthb_chain == probe); + ASSERT(b != NULL); + + if (b == bucket) + hash->dth_tab[ndx] = bucket->dthb_next; + else { + while (b->dthb_next != bucket) + b = b->dthb_next; + + b->dthb_next = bucket->dthb_next; + } + + ASSERT(hash->dth_nbuckets > 0); + + hash->dth_nbuckets--; + kfree(bucket); + + return; + } + + bucket->dthb_chain = *nextp; + } else + *(DTRACE_HASHNEXT(hash, *prevp)) = *nextp; + + if (*nextp != NULL) + *(DTRACE_HASHPREV(hash, *nextp)) = *prevp; +} diff --git a/dtrace/dtrace_helper.c b/dtrace/dtrace_helper.c new file mode 100644 index 000000000000..75060f4ed257 --- /dev/null +++ b/dtrace/dtrace_helper.c @@ -0,0 +1,175 @@ +/* + * FILE: dtrace_helper.c + * DESCRIPTION: Dynamic Tracing: helper functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include "dtrace.h" + +static uint32_t dtrace_helptrace_next = 0; +static uint32_t dtrace_helptrace_nlocals; +static char *dtrace_helptrace_buffer; +static int dtrace_helptrace_bufsize = 512 * 1024; + +#ifdef CONFIG_DT_DEBUG +static int dtrace_helptrace_enabled = 1; +#else +static int dtrace_helptrace_enabled = 0; +#endif + +static void dtrace_helper_trace(dtrace_helper_action_t *helper, + dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate, int where) +{ + uint32_t size, next, nnext, i; + dtrace_helptrace_t *ent; + uint16_t flags = cpu_core[ + smp_processor_id() + ].cpuc_dtrace_flags; + + if (!dtrace_helptrace_enabled) + return; + + ASSERT(vstate->dtvs_nlocals <= dtrace_helptrace_nlocals); + + /* + * What would a tracing framework be without its own tracing + * framework? (Well, a hell of a lot simpler, for starters...) + */ + size = sizeof(dtrace_helptrace_t) + dtrace_helptrace_nlocals * + sizeof(uint64_t) - sizeof(uint64_t); + + /* + * Iterate until we can allocate a slot in the trace buffer. + */ + do { + next = dtrace_helptrace_next; + + if (next + size < dtrace_helptrace_bufsize) + nnext = next + size; + else + nnext = size; + } while (cmpxchg(&dtrace_helptrace_next, next, nnext) != next); + + /* + * We have our slot; fill it in. + */ + if (nnext == size) + next = 0; + + ent = (dtrace_helptrace_t *)&dtrace_helptrace_buffer[next]; + ent->dtht_helper = helper; + ent->dtht_where = where; + ent->dtht_nlocals = vstate->dtvs_nlocals; + + ent->dtht_fltoffs = (mstate->dtms_present & DTRACE_MSTATE_FLTOFFS) + ? mstate->dtms_fltoffs + : -1; + ent->dtht_fault = DTRACE_FLAGS2FLT(flags); + ent->dtht_illval = cpu_core[smp_processor_id()].cpuc_dtrace_illval; + + for (i = 0; i < vstate->dtvs_nlocals; i++) { + dtrace_statvar_t *svar; + + if ((svar = vstate->dtvs_locals[i]) == NULL) + continue; + + ASSERT(svar->dtsv_size >= NR_CPUS * sizeof(uint64_t)); + ent->dtht_locals[i] = + ((uint64_t *)(uintptr_t)svar->dtsv_data)[ + smp_processor_id()]; + } +} + +uint64_t dtrace_helper(int which, dtrace_mstate_t *mstate, + dtrace_state_t *state, uint64_t arg0, uint64_t arg1) +{ + uint16_t *flags = &cpu_core[ + smp_processor_id() + ].cpuc_dtrace_flags; + uint64_t sarg0 = mstate->dtms_arg[0]; + uint64_t sarg1 = mstate->dtms_arg[1]; + uint64_t rval = 0; + dtrace_helpers_t *helpers = current->dtrace_helpers; + dtrace_helper_action_t *helper; + dtrace_vstate_t *vstate; + dtrace_difo_t *pred; + int i, trace = dtrace_helptrace_enabled; + + ASSERT(which >= 0 && which < DTRACE_NHELPER_ACTIONS); + + if (helpers == NULL) + return 0; + + if ((helper = helpers->dthps_actions[which]) == NULL) + return 0; + + vstate = &helpers->dthps_vstate; + mstate->dtms_arg[0] = arg0; + mstate->dtms_arg[1] = arg1; + + /* + * Now iterate over each helper. If its predicate evaluates to 'true', + * we'll call the corresponding actions. Note that the below calls + * to dtrace_dif_emulate() may set faults in machine state. This is + * okay: our caller (the outer dtrace_dif_emulate()) will simply plow + * the stored DIF offset with its own (which is the desired behavior). + * Also, note the calls to dtrace_dif_emulate() may allocate scratch + * from machine state; this is okay, too. + */ + for (; helper != NULL; helper = helper->dtha_next) { + if ((pred = helper->dtha_predicate) != NULL) { + if (trace) + dtrace_helper_trace(helper, mstate, vstate, 0); + + if (!dtrace_dif_emulate(pred, mstate, vstate, state)) + goto next; + + if (*flags & CPU_DTRACE_FAULT) + goto err; + } + + for (i = 0; i < helper->dtha_nactions; i++) { + if (trace) + dtrace_helper_trace(helper, mstate, vstate, + i + 1); + + rval = dtrace_dif_emulate(helper->dtha_actions[i], + mstate, vstate, state); + + if (*flags & CPU_DTRACE_FAULT) + goto err; + } + +next: + if (trace) + dtrace_helper_trace(helper, mstate, vstate, + DTRACE_HELPTRACE_NEXT); + } + + if (trace) + dtrace_helper_trace(helper, mstate, vstate, + DTRACE_HELPTRACE_DONE); + + /* + * Restore the arg0 that we saved upon entry. + */ + mstate->dtms_arg[0] = sarg0; + mstate->dtms_arg[1] = sarg1; + + return rval; + +err: + if (trace) + dtrace_helper_trace(helper, mstate, vstate, + DTRACE_HELPTRACE_ERR); + + /* + * Restore the arg0 that we saved upon entry. + */ + mstate->dtms_arg[0] = sarg0; + mstate->dtms_arg[1] = sarg1; + + return 0; +} diff --git a/dtrace/dtrace_isa.c b/dtrace/dtrace_isa.c new file mode 100644 index 000000000000..8fd14fa72081 --- /dev/null +++ b/dtrace/dtrace_isa.c @@ -0,0 +1,547 @@ +/* + * FILE: dtrace_isa.c + * DESCRIPTION: Dynamic Tracing: architecture specific support functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include +#include + +#include "dtrace.h" + +/* FIXME */ +uintptr_t _userlimit = 0x00007fffffffffffLL; +uintptr_t kernelbase = 0xffff880000000000LL; + +cpu_core_t cpu_core[NR_CPUS]; +EXPORT_SYMBOL(cpu_core); +EXPORT_SYMBOL(dtrace_getfp); + +DEFINE_MUTEX(cpu_lock); +EXPORT_SYMBOL(cpu_lock); + +extern void dtrace_copy(uintptr_t, uintptr_t, size_t); +extern void dtrace_copystr(uintptr_t, uintptr_t, size_t, + volatile uint16_t *); + +static int dtrace_copycheck(uintptr_t uaddr, uintptr_t kaddr, size_t size) +{ +#ifdef FIXME + ASSERT(kaddr >= kernelbase && kaddr + size >= kaddr); +#else + if (kaddr < kernelbase || kaddr + size < kaddr) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[smp_processor_id()].cpuc_dtrace_illval = kaddr; + return 0; + } +#endif + + if (uaddr + size >= kernelbase || uaddr + size < uaddr) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[smp_processor_id()].cpuc_dtrace_illval = uaddr; + return 0; + } + + return 1; +} + +void dtrace_copyin(uintptr_t uaddr, uintptr_t kaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copy(uaddr, kaddr, size); +} + +void dtrace_copyout(uintptr_t uaddr, uintptr_t kaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copy(kaddr, uaddr, size); +} + +void dtrace_copyinstr(uintptr_t uaddr, uintptr_t kaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copystr(uaddr, kaddr, size, flags); +} + +void dtrace_copyoutstr(uintptr_t uaddr, uintptr_t kaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copystr(kaddr, uaddr, size, flags); +} + +#define DTRACE_FUWORD(bits) \ + uint##bits##_t dtrace_fuword##bits(void *uaddr) \ + { \ + extern uint##bits##_t dtrace_fuword##bits##_nocheck(void *);\ + \ + if ((uintptr_t)uaddr > _userlimit) { \ + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); \ + cpu_core[smp_processor_id()].cpuc_dtrace_illval = \ + (uintptr_t)uaddr; \ + return 0; \ + } \ + \ + return dtrace_fuword##bits##_nocheck(uaddr); \ + } + +DTRACE_FUWORD(8) +DTRACE_FUWORD(16) +DTRACE_FUWORD(32) +DTRACE_FUWORD(64) + +struct frame { + struct frame *fr_savfp; + unsigned long fr_savpc; +} __attribute__((packed)); + +static void dtrace_invop_callsite(void) +{ +} + +uint64_t dtrace_getarg(int arg, int aframes) +{ + struct frame *fp = (struct frame *)dtrace_getfp(); + uintptr_t *stack; + int i; + uint64_t val; +#ifdef __i386__ + int regmap[] = { + REG_EAX, + REG_EDX, + REG_ECX + }; +#else + int regmap[] = { + REG_RDI, + REG_RSI, + REG_RDX, + REG_RCX, + REG_R8, + REG_R9 + }; +#endif + int nreg = sizeof(regmap) / sizeof(regmap[0]) - 1; + +printk(KERN_INFO "getarg(%d, %d) -> nreg = %d\n", arg, aframes, nreg); + for (i = 1; i <= aframes; i++) { + fp = fp->fr_savfp; + + if (fp->fr_savpc == (uintptr_t)dtrace_invop_callsite) { +#ifdef __i386__ + /* FIXME */ +#else + /* FIXME */ +#endif + + goto load; + } + } + + /* + * We know that we did not get here through a trap to get into the + * dtrace_probe() function, so this was a straight call into it from + * a provider. In that case, we need to shift the argument that we + * are looking for, because the probe ID will be the first argument to + * dtrace_probe(). + */ + arg++; + +#ifndef __i386__ +printk(KERN_INFO "getarg(%d, %d) [!i386]-> nreg = %d\n", arg, aframes, nreg); + if (arg <= nreg) { + /* + * This should not happen. If the argument was passed in a + * register then it should have been, ...passed in a reguster. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return 0; + } + + arg -= nreg + 1; +#endif + + stack = (uintptr_t *)&fp[1]; + +load: + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + val = stack[arg]; + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + return val; +} + +int dtrace_getipl(void) +{ + return in_interrupt(); +} + +ulong_t dtrace_getreg(struct pt_regs *rp, uint_t reg) +{ +#ifdef __i386__ + if (reg > REG_SS) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return 0; + } + + switch (reg) { + case REG_GS: + case REG_FS: + case REG_ES: + case REG_DS: + case REG_CS: + return rp->cs; + case REG_EDI: + return rp->di; + case REG_ESI: + return rp->si; + case REG_EBP: + return rp->bp; + case REG_ESP: + case REG_UESP: + return rp->sp; + case REG_EBX: + return rp->bx; + case REG_EDX: + return rp->dx; + case REG_ECX: + return rp->cx; + case REG_EAX: + return rp->ax; + case REG_TRAPNO: + return rp->orig_ax; + case REG_ERR: + return rp->di; + case REG_EIP: + return rp->ip; + case REG_EFL: + return rp->flags; + case REG_SS: + return rp->ss; + default: + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return 0; + } +#else + int regmap[] = { + REG_GS, /* 0 -> GS */ + REG_FS, /* 1 -> FS */ + REG_ES, /* 2 -> ES */ + REG_DS, /* 3 -> DS */ + REG_RDI, /* 4 -> EDI */ + REG_RSI, /* 5 -> ESI */ + REG_RBP, /* 6 -> EBP */ + REG_RSP, /* 7 -> ESP */ + REG_RBX, /* 8 -> EBX */ + REG_RDX, /* 9 -> EDX */ + REG_RCX, /* 10 -> ECX */ + REG_RAX, /* 11 -> EAX */ + REG_TRAPNO, /* 12 -> TRAPNO */ + REG_ERR, /* 13 -> ERR */ + REG_RIP, /* 14 -> EIP */ + REG_CS, /* 15 -> CS */ + REG_RFL, /* 16 -> EFL */ + REG_RSP, /* 17 -> UESP */ + REG_SS, /* 18 -> SS */ + }; + + if (reg <= REG_SS) { + if (reg >= sizeof(regmap) / sizeof(int)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return 0; + } + + reg = regmap[reg]; + } else + reg -= REG_SS + 1; + + switch (reg) { + case REG_RDI: + return rp->di; + case REG_RSI: + return rp->si; + case REG_RDX: + return rp->dx; + case REG_RCX: + return rp->cx; + case REG_R8: + return rp->r8; + case REG_R9: + return rp->r9; + case REG_RAX: + return rp->ax; + case REG_RBX: + return rp->bx; + case REG_RBP: + return rp->bp; + case REG_R10: + return rp->r10; + case REG_R11: + return rp->r11; + case REG_R12: + return rp->r12; + case REG_R13: + return rp->r13; + case REG_R14: + return rp->r14; + case REG_R15: + return rp->r15; + case REG_CS: + case REG_DS: + case REG_ES: + case REG_FS: + case REG_GS: + return rp->cs; + case REG_TRAPNO: + return rp->orig_ax; + case REG_ERR: + return rp->di; + case REG_RIP: + return rp->ip; + case REG_SS: + return rp->ss; + case REG_RFL: + return rp->flags; + case REG_RSP: + return rp->sp; + default: + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return 0; + } +#endif +} + +static void dtrace_sync_func(void) +{ +} + +void dtrace_sync(void) +{ + dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)dtrace_sync_func, NULL); +} + +void dtrace_xcall(processorid_t cpu, dtrace_xcall_t func, void *arg) +{ + if (cpu == DTRACE_CPUALL) { + smp_call_function(func, arg, 1); + } else + smp_call_function_single(cpu, func, arg, 1); +} + +void dtrace_toxic_ranges(void (*func)(uintptr_t, uintptr_t)) +{ + /* FIXME */ +} + +ktime_t dtrace_gethrestime(void) +{ + return dtrace_gethrtime(); +} + +#define STACKTRACE_KERNEL 0x01 +#define STACKTRACE_USER 0x02 +#define STACKTRACE_SKIP 0x10 + +struct stacktrace_state { + uint64_t *pcs; + uint64_t *fps; + int limit; + int depth; + int flags; +}; + +static void dtrace_stacktrace_warning(void *data, char *msg) +{ +} + +static void dtrace_stacktrace_warning_symbol(void *data, char *msg, + unsigned long symbol) +{ +} + +static int dtrace_stacktrace_stack(void *data, char *name) +{ + struct stacktrace_state *st = (struct stacktrace_state *)data; + + /* + * We do not skip anything for non-user stack analysis. + */ + if (!(st->flags & STACKTRACE_USER)) + return 0; + + if (name != NULL && strlen(name) > 3) { + /* + * Sadly, the dump stack code calls us with both and EOI. + * Consistency would be much nicer. + */ + if ((name[0] == '<' && name[1] == 'E' && name[2] == 'O') || + (name[0] == 'E' && name[2] == 'O')) + st->flags &= ~STACKTRACE_SKIP; + } + + return 0; +} + +static void dtrace_stacktrace_address(void *data, unsigned long addr, + int reliable) +{ + struct stacktrace_state *st = (struct stacktrace_state *)data; + + if (st->flags & STACKTRACE_SKIP) + return; + + if (reliable == 2) { + if (st->fps) + st->fps[st->depth] = addr; + } else { + if (st->pcs != NULL) { + if (st->depth < st->limit) + st->pcs[st->depth++] = addr; + } else + st->depth++; + } +} + +static inline int valid_sp(struct thread_info *tinfo, void *p, + unsigned int size, void *end) +{ + void *t = tinfo; + + if (end) { + if (p < end && p >= (end - THREAD_SIZE)) + return 1; + else + return 0; + } + + return p > t && p < t + THREAD_SIZE - size; +} + +static unsigned long dtrace_stacktrace_walk_stack( + struct thread_info *tinfo, + unsigned long *stack, + unsigned long bp, + const struct stacktrace_ops *ops, + void *data, unsigned long *end, + int *graph) +{ + struct frame *fr = (struct frame *)bp; + unsigned long *pcp = &(fr->fr_savpc); + + while (valid_sp(tinfo, pcp, sizeof(*pcp), end)) { + unsigned long addr = *pcp; + + fr = fr->fr_savfp; + ops->address(data, (unsigned long)fr, 2); + ops->address(data, addr, 1); + pcp = &(fr->fr_savpc); + } + + return (unsigned long)fr; +} + +static const struct stacktrace_ops dtrace_tracetrace_ops = { + .warning = dtrace_stacktrace_warning, + .warning_symbol = dtrace_stacktrace_warning_symbol, + .stack = dtrace_stacktrace_stack, + .address = dtrace_stacktrace_address, + .walk_stack = print_context_stack +}; + +void dtrace_getpcstack(uint64_t *pcstack, int pcstack_limit, int aframes, + uint32_t *intrpc) +{ + struct stacktrace_state st = { + pcstack, + NULL, + pcstack_limit, + 0, + STACKTRACE_KERNEL + }; + + dump_trace(NULL, NULL, NULL, 0, &dtrace_tracetrace_ops, &st); + + while (st.depth < st.limit) + pcstack[st.depth++] = 0; +} + +void dtrace_getupcstack(uint64_t *pcstack, int pcstack_limit) +{ + struct stacktrace_state st = { + pcstack, + NULL, + pcstack_limit, + 0, + STACKTRACE_USER + }; + + dump_trace(NULL, NULL, NULL, 0, &dtrace_tracetrace_ops, &st); + + while (st.depth < st.limit) + pcstack[st.depth++] = 0; +} + +static const struct stacktrace_ops dtrace_tracetrace_ops_alt = { + .warning = dtrace_stacktrace_warning, + .warning_symbol = dtrace_stacktrace_warning_symbol, + .stack = dtrace_stacktrace_stack, + .address = dtrace_stacktrace_address, + .walk_stack = dtrace_stacktrace_walk_stack +}; + +void dtrace_getufpstack(uint64_t *pcstack, uint64_t *fpstack, + int pcstack_limit) +{ + struct stacktrace_state st = { + pcstack, + fpstack, + pcstack_limit, + 0, + STACKTRACE_USER + }; + + dump_trace(NULL, NULL, NULL, 0, &dtrace_tracetrace_ops_alt, &st); + + while (st.depth < st.limit) { + fpstack[st.depth] = 0; + pcstack[st.depth++] = 0; + } +} + +int dtrace_getstackdepth(int aframes) +{ + struct stacktrace_state st = { + NULL, + NULL, + 0, + 0, + STACKTRACE_KERNEL + }; + + dump_trace(NULL, NULL, NULL, 0, &dtrace_tracetrace_ops, &st); + + if (st.depth <= aframes) + return 0; + + return st.depth - aframes; +} + +int dtrace_getustackdepth(void) +{ + struct stacktrace_state st = { + NULL, + NULL, + 0, + 0, + STACKTRACE_USER + }; + + dump_trace(NULL, NULL, NULL, 0, &dtrace_tracetrace_ops, &st); + + return st.depth; +} diff --git a/dtrace/dtrace_match.c b/dtrace/dtrace_match.c new file mode 100644 index 000000000000..f70024f416bb --- /dev/null +++ b/dtrace/dtrace_match.c @@ -0,0 +1,325 @@ +/* + * FILE: dtrace_match.c + * DESCRIPTION: Dynamic Tracing: matching functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include "dtrace.h" + +dtrace_hash_t *dtrace_bymod; +dtrace_hash_t *dtrace_byfunc; +dtrace_hash_t *dtrace_byname; + +int dtrace_match_priv(const dtrace_probe_t *prp, uint32_t priv, uid_t uid) +{ + if (priv != DTRACE_PRIV_ALL) { + uint32_t ppriv = + prp->dtpr_provider->dtpv_priv.dtpp_flags; + uint32_t match = priv & ppriv; + + if ((priv & (DTRACE_PRIV_PROC | DTRACE_PRIV_USER | + DTRACE_PRIV_KERNEL)) == 0) + return 0; + + if (match == 0 && ppriv != 0) + return 0; + + if (((ppriv & ~match) & DTRACE_PRIV_OWNER) != 0 && + uid != prp->dtpr_provider->dtpv_priv.dtpp_uid) + return 0; + } + + return 1; +} + +int dtrace_match_probe(const dtrace_probe_t *prp, const dtrace_probekey_t *pkp, + uint32_t priv, uid_t uid) +{ + dtrace_provider_t *pvp = prp->dtpr_provider; + int rv; + + if (pvp->dtpv_defunct) + return 0; + + if ((rv = pkp->dtpk_pmatch(pvp->dtpv_name, pkp->dtpk_prov, 0)) <= 0) + return rv; + + if ((rv = pkp->dtpk_mmatch(prp->dtpr_mod, pkp->dtpk_mod, 0)) <= 0) + return rv; + + if ((rv = pkp->dtpk_fmatch(prp->dtpr_func, pkp->dtpk_func, 0)) <= 0) + return rv; + + if ((rv = pkp->dtpk_nmatch(prp->dtpr_name, pkp->dtpk_name, 0)) <= 0) + return rv; + + if (dtrace_match_priv(prp, priv, uid) == 0) + return 0; + + return rv; +} + +int dtrace_match_glob(const char *s, const char *p, int depth) +{ + const char *olds; + char s1, c; + int gs; + + if (depth > DTRACE_PROBEKEY_MAXDEPTH) + return -1; + + if (s == NULL) + s = ""; + +top: + olds = s; + s1 = *s++; + + if (p == NULL) + return 0; + + if ((c = *p++) == '\0') + return s1 == '\0'; + + switch (c) { + case '[': + { + int ok = 0, notflag = 0; + char lc = '\0'; + + if (s1 == '\0') + return 0; + + if (*p == '!') { + notflag = 1; + p++; + } + + if ((c = *p++) == '\0') + return 0; + + do { + if (c == '-' && lc != '\0' && *p != ']') { + if ((c = *p++) == '\0') + return 0; + if (c == '\\' && (c = *p++) == '\0') + return 0; + + if (notflag) { + if (s1 < lc || s1 > c) + ok++; + else + return 0; + } else if (lc <= s1 && s1 <= c) + ok++; + } else if (c == '\\' && (c = *p++) == '\0') + return 0; + + lc = c; + + if (notflag) { + if (s1 != c) + ok++; + else + return 0; + } else if (s1 == c) + ok++; + + if ((c = *p++) == '\0') + return 0; + } while (c != ']'); + + if (ok) + goto top; + + return 0; + } + + case '\\': + if ((c = *p++) == '\0') + return 0; + + default: + if (c != s1) + return 0; + + case '?': + if (s1 != '\0') + goto top; + + return 0; + + case '*': + while (*p == '*') + p++; + + if (*p == '\0') + return 1; + + for (s = olds; *s != '\0'; s++) { + if ((gs = dtrace_match_glob(s, p, depth + 1)) != 0) + return gs; + } + + return 0; + } +} + +int dtrace_match_string(const char *s, const char *p, int depth) +{ + return s != NULL && strcmp(s, p) == 0; +} + +int dtrace_match_nul(const char *s, const char *p, int depth) +{ + return 1; +} + +int dtrace_match_nonzero(const char *s, const char *p, int depth) +{ + return s != NULL && s[0] != '\0'; +} + +struct probe_match { + const dtrace_probekey_t *pkp; + uint32_t priv; + uid_t uid; + int (*matched)(dtrace_probe_t *, void *); + void *arg; + int nmatched; +}; + +static int dtrace_match_one(int id, void *p, void *data) +{ + struct probe_match *pbm = (struct probe_match *)data; + dtrace_probe_t *probe = (dtrace_probe_t *)p; + int rc; + + if (dtrace_match_probe(probe, pbm->pkp, pbm->priv, pbm->uid) <= 0) + return 0; + + pbm->nmatched++; + + if ((rc = (pbm->matched)(probe, pbm->arg)) != DTRACE_MATCH_NEXT) { + if (rc == DTRACE_MATCH_FAIL) + return DTRACE_MATCH_FAIL; + } + + return 0; +} + +int dtrace_match(const dtrace_probekey_t *pkp, uint32_t priv, uid_t uid, + int (*matched)(dtrace_probe_t *, void *), void *arg) +{ + dtrace_probe_t template, *probe; + dtrace_hash_t *hash = NULL; + int len, rc, best = INT_MAX, nmatched = 0; + + if (pkp->dtpk_id != DTRACE_IDNONE) { + if ((probe = dtrace_probe_lookup_id(pkp->dtpk_id)) != NULL && + dtrace_match_probe(probe, pkp, priv, uid) > 0) { + if ((*matched)(probe, arg) == DTRACE_MATCH_FAIL) + return DTRACE_MATCH_FAIL; + + nmatched++; + } + + return nmatched; + } + + template.dtpr_mod = (char *)pkp->dtpk_mod; + template.dtpr_func = (char *)pkp->dtpk_func; + template.dtpr_name = (char *)pkp->dtpk_name; + + if (pkp->dtpk_mmatch == &dtrace_match_string && + (len = dtrace_hash_collisions(dtrace_bymod, &template)) < best) { + best = len; + hash = dtrace_bymod; + } + + if (pkp->dtpk_fmatch == &dtrace_match_string && + (len = dtrace_hash_collisions(dtrace_byfunc, &template)) < best) { + best = len; + hash = dtrace_byfunc; + } + + if (pkp->dtpk_nmatch == &dtrace_match_string && + (len = dtrace_hash_collisions(dtrace_byname, &template)) < best) { + best = len; + hash = dtrace_byname; + } + + if (hash == NULL) { + struct probe_match pbm; + + pbm.pkp = pkp; + pbm.priv = priv; + pbm.uid = uid; + pbm.matched = matched; + pbm.arg = arg; + pbm.nmatched = 0; + + rc = dtrace_probe_for_each(dtrace_match_one, &pbm); + if (rc == DTRACE_MATCH_FAIL) + return DTRACE_MATCH_FAIL; + + return pbm.nmatched; + } + + for (probe = dtrace_hash_lookup(hash, &template); probe != NULL; + probe = *(DTRACE_HASHNEXT(hash, probe))) { + if (dtrace_match_probe(probe, pkp, priv, uid) <= 0) + continue; + + nmatched++; + + if ((rc = (*matched)(probe, arg)) != DTRACE_MATCH_NEXT) { + if (rc == DTRACE_MATCH_FAIL) + return DTRACE_MATCH_FAIL; + + break; + } + } + + return nmatched; +} + +static dtrace_probekey_f *dtrace_probekey_func(const char *p) +{ + char c; + + if (p == NULL || *p == '\0') + return &dtrace_match_nul; + + while ((c = *p++) != '\0') { + if (c == '[' || c == '?' || c == '*' || c == '\\') + return &dtrace_match_glob; + } + + return &dtrace_match_string; +} + +void dtrace_probekey(const dtrace_probedesc_t *pdp, dtrace_probekey_t *pkp) +{ + pkp->dtpk_prov = pdp->dtpd_provider; + pkp->dtpk_pmatch = dtrace_probekey_func(pdp->dtpd_provider); + + pkp->dtpk_mod = pdp->dtpd_mod; + pkp->dtpk_mmatch = dtrace_probekey_func(pdp->dtpd_mod); + + pkp->dtpk_func = pdp->dtpd_func; + pkp->dtpk_fmatch = dtrace_probekey_func(pdp->dtpd_func); + + pkp->dtpk_name = pdp->dtpd_name; + pkp->dtpk_nmatch = dtrace_probekey_func(pdp->dtpd_name); + + pkp->dtpk_id = pdp->dtpd_id; + + if (pkp->dtpk_id == DTRACE_IDNONE && + pkp->dtpk_pmatch == &dtrace_match_nul && + pkp->dtpk_mmatch == &dtrace_match_nul && + pkp->dtpk_fmatch == &dtrace_match_nul && + pkp->dtpk_nmatch == &dtrace_match_nul) + pkp->dtpk_fmatch = &dtrace_match_nonzero; +} diff --git a/dtrace/dtrace_mod.c b/dtrace/dtrace_mod.c new file mode 100644 index 000000000000..9f36a0a476d0 --- /dev/null +++ b/dtrace/dtrace_mod.c @@ -0,0 +1,34 @@ +/* + * FILE: dtrace_mod.c + * DESCRIPTION: Dynamic Tracing: module handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace_dev.h" + +MODULE_AUTHOR("Kris Van Hees (kris.van.hees@oracle.com)"); +MODULE_DESCRIPTION("Dynamic Tracing"); +MODULE_VERSION("v0.1"); +MODULE_LICENSE("Proprietary"); + +/* + * Initialize the module. + */ +static int __init dtrace_init(void) +{ + return dtrace_dev_init(); +} + +/* + * Perform cleanup before the module is removed. + */ +static void __exit dtrace_exit(void) +{ + dtrace_dev_exit(); +} + +module_init(dtrace_init); +module_exit(dtrace_exit); diff --git a/dtrace/dtrace_predicate.c b/dtrace/dtrace_predicate.c new file mode 100644 index 000000000000..30c1512cfd53 --- /dev/null +++ b/dtrace/dtrace_predicate.c @@ -0,0 +1,66 @@ +/* + * FILE: dtrace_predicate.c + * DESCRIPTION: Dynamic Tracing: predicate functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace.h" + +static dtrace_cacheid_t dtrace_predcache_id = DTRACE_CACHEIDNONE + 1; + +dtrace_predicate_t *dtrace_predicate_create(dtrace_difo_t *dp) +{ + dtrace_predicate_t *pred; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(dp->dtdo_refcnt != 0); + + pred = kzalloc(sizeof (dtrace_predicate_t), GFP_KERNEL); + pred->dtp_difo = dp; + pred->dtp_refcnt = 1; + + if (!dtrace_difo_cacheable(dp)) + return pred; + + /* + * This is only theoretically possible -- we have had 2^32 cacheable + * predicates on this machine. We cannot allow any more predicates to + * become cacheable: as unlikely as it is, there may be a thread + * caching a (now stale) predicate cache ID. (N.B.: the temptation is + * being successfully resisted to have this cmn_err() "Holy shit -- we + * executed this code!") + */ + if (dtrace_predcache_id == DTRACE_CACHEIDNONE) + return pred; + + pred->dtp_cacheid = dtrace_predcache_id++; + + return pred; +} + +void dtrace_predicate_hold(dtrace_predicate_t *pred) +{ + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(pred->dtp_difo != NULL && pred->dtp_difo->dtdo_refcnt != 0); + ASSERT(pred->dtp_refcnt > 0); + + pred->dtp_refcnt++; +} + +void dtrace_predicate_release(dtrace_predicate_t *pred, + dtrace_vstate_t *vstate) +{ + dtrace_difo_t *dp = pred->dtp_difo; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(dp != NULL && dp->dtdo_refcnt != 0); + ASSERT(pred->dtp_refcnt > 0); + + if (--pred->dtp_refcnt == 0) { + dtrace_difo_release(dp, vstate); + kfree(pred); + } +} diff --git a/dtrace/dtrace_priv.c b/dtrace/dtrace_priv.c new file mode 100644 index 000000000000..57a09950ee82 --- /dev/null +++ b/dtrace/dtrace_priv.c @@ -0,0 +1,105 @@ +/* + * FILE: dtrace_priv.c + * DESCRIPTION: Dynamic Tracing: privilege check functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include "dtrace.h" + +/* + * This privilege check should be used by actions and subroutines to + * verify that the user credentials of the process that enabled the + * invoking ECB match the target credentials + */ +int dtrace_priv_proc_common_user(dtrace_state_t *state) +{ + const cred_t *cr, *s_cr = state->dts_cred.dcr_cred; + + /* + * We should always have a non-NULL state cred here, since if cred + * is null (anonymous tracing), we fast-path bypass this routine. + */ + ASSERT(s_cr != NULL); + + if ((cr = current_cred()) != NULL && + s_cr->euid == cr->euid && + s_cr->euid == cr->uid && + s_cr->euid == cr->suid && + s_cr->egid == cr->egid && + s_cr->egid == cr->gid && + s_cr->egid == cr->sgid) + return 1; + + return 0; +} + +/* + * This privilege check should be used by actions and subroutines to + * verify that the process has not setuid or changed credentials. + */ +int dtrace_priv_proc_common_nocd(void) +{ +#ifdef FIXME + proc_t *proc; + + if ((proc = ttoproc(curthread)) != NULL && !(proc->p_flag & SNOCD)) + return 1; +#endif + + return 0; +} + +int dtrace_priv_proc_destructive(dtrace_state_t *state) +{ + int action = state->dts_cred.dcr_action; + + if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER) == 0) && + dtrace_priv_proc_common_user(state) == 0) + goto bad; + + if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG) == 0) && + dtrace_priv_proc_common_nocd() == 0) + goto bad; + + return 1; + +bad: + DTRACE_CPUFLAG_SET(CPU_DTRACE_UPRIV); + + return 0; +} + +int dtrace_priv_proc_control(dtrace_state_t *state) +{ + if (state->dts_cred.dcr_action & DTRACE_CRA_PROC_CONTROL) + return 1; + + if (dtrace_priv_proc_common_user(state) && + dtrace_priv_proc_common_nocd()) + return 1; + + DTRACE_CPUFLAG_SET(CPU_DTRACE_UPRIV); + + return 0; +} + +int dtrace_priv_proc(dtrace_state_t *state) +{ + if (state->dts_cred.dcr_action & DTRACE_CRA_PROC) + return 1; + + DTRACE_CPUFLAG_SET(CPU_DTRACE_UPRIV); + + return 0; +} + +int dtrace_priv_kernel(dtrace_state_t *state) +{ + if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL) + return 1; + + DTRACE_CPUFLAG_SET(CPU_DTRACE_KPRIV); + + return 0; +} diff --git a/dtrace/dtrace_probe.c b/dtrace/dtrace_probe.c new file mode 100644 index 000000000000..eb185d9e19a1 --- /dev/null +++ b/dtrace/dtrace_probe.c @@ -0,0 +1,1228 @@ +/* + * FILE: dtrace_probe.c + * DESCRIPTION: Dynamic Tracing: probe management functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include +#include +#include +#include + +#include "dtrace.h" + +ktime_t dtrace_chill_interval = + KTIME_INIT(1, 0); +ktime_t dtrace_chill_max = + KTIME_INIT(0, + 500 * (NANOSEC / MILLISEC)); + +dtrace_genid_t dtrace_probegen; + +static struct idr dtrace_probe_idr; +static uint64_t dtrace_vtime_references; + +static struct task_struct *dtrace_panicked; + +/* + * Create a new probe. + */ +dtrace_id_t dtrace_probe_create(dtrace_provider_id_t prov, const char *mod, + const char *func, const char *name, + int aframes, void *arg) +{ + dtrace_probe_t *probe; + dtrace_provider_t *provider = (dtrace_provider_t *)prov; + dtrace_id_t id; + int err; + + probe = kzalloc(sizeof(dtrace_probe_t), __GFP_NOFAIL); + + /* + * The idr_pre_get() function should be called without holding locks. + * When the provider is the DTrace core itself, dtrace_lock will be + * held when we enter this function. + */ + if (provider == dtrace_provider) { + ASSERT(mutex_is_locked(&dtrace_lock)); + mutex_unlock(&dtrace_lock); + } + +again: + idr_pre_get(&dtrace_probe_idr, __GFP_NOFAIL); + + mutex_lock(&dtrace_lock); + err = idr_get_new(&dtrace_probe_idr, probe, &id); + if (err == -EAGAIN) { + mutex_unlock(&dtrace_lock); + goto again; + } + + probe->dtpr_id = id; + probe->dtpr_gen = dtrace_probegen++; + probe->dtpr_mod = dtrace_strdup(mod); + probe->dtpr_func = dtrace_strdup(func); + probe->dtpr_name = dtrace_strdup(name); + probe->dtpr_arg = arg; + probe->dtpr_aframes = aframes; + probe->dtpr_provider = provider; + + dtrace_hash_add(dtrace_bymod, probe); + dtrace_hash_add(dtrace_byfunc, probe); + dtrace_hash_add(dtrace_byname, probe); + + if (provider != dtrace_provider) + mutex_unlock(&dtrace_lock); + +printk(KERN_INFO "probe_create(%s, %s, %s, %s) -> %d\n", provider->dtpv_name, mod, func, name, id); + return id; +} +EXPORT_SYMBOL(dtrace_probe_create); + +int dtrace_probe_enable(const dtrace_probedesc_t *desc, dtrace_enabling_t *enab) +{ + dtrace_probekey_t pkey; + uint32_t priv; + uid_t uid; + + dtrace_ecb_create_cache = NULL; + + if (desc == NULL) { + (void) dtrace_ecb_create_enable(NULL, enab); + + return 0; + } + + dtrace_probekey(desc, &pkey); + dtrace_cred2priv(enab->dten_vstate->dtvs_state->dts_cred.dcr_cred, + &priv, &uid); +printk(KERN_INFO "probe_enable(%d, %s, %s, %s, %s)\n", pkey.dtpk_id, pkey.dtpk_prov, pkey.dtpk_mod, pkey.dtpk_func, pkey.dtpk_name); + return dtrace_match(&pkey, priv, uid, dtrace_ecb_create_enable, enab); +} + +/* + * Return the probe argument associated with the specified probe. + */ +void *dtrace_probe_arg(dtrace_provider_id_t id, dtrace_id_t pid) +{ + dtrace_probe_t *probe; + void *rval = NULL; + + mutex_lock(&dtrace_lock); + + if ((probe = dtrace_probe_lookup_id(pid)) != NULL && + probe->dtpr_provider == (dtrace_provider_t *)id) + rval = probe->dtpr_arg; + + mutex_unlock(&dtrace_lock); + + return rval; +} +EXPORT_SYMBOL(dtrace_probe_arg); + +/* + * Copy a probe into a probe description. + */ +void dtrace_probe_description(const dtrace_probe_t *prp, + dtrace_probedesc_t *pdp) +{ + memset(pdp, 0, sizeof(dtrace_probedesc_t)); + pdp->dtpd_id = prp->dtpr_id; + + strncpy(pdp->dtpd_provider, prp->dtpr_provider->dtpv_name, + DTRACE_PROVNAMELEN - 1); + + strncpy(pdp->dtpd_mod, prp->dtpr_mod, DTRACE_MODNAMELEN - 1); + strncpy(pdp->dtpd_func, prp->dtpr_func, DTRACE_FUNCNAMELEN - 1); + strncpy(pdp->dtpd_name, prp->dtpr_name, DTRACE_NAMELEN - 1); +} + +void dtrace_probe_provide(dtrace_probedesc_t *desc, dtrace_provider_t *prv) +{ + struct module *mod; + int all = 0; + + if (prv == NULL) { + all = 1; + prv = dtrace_provider; + } + + do { + prv->dtpv_pops.dtps_provide(prv->dtpv_arg, desc); + +#ifdef FIXME +/* + * This needs work because (so far) I have not found a way to get access to the + * list of modules in Linux. + */ + mutex_lock(&module_mutex); + + list_for_each_entry(mod, &modules, list) { + if (mod->state != MODULE_STATE_LIVE) + continue; + + prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, mod); + } + + mutex_unlock(&module_mutex); +#endif + } while (all && (prv = prv->dtpv_next) != NULL); +} + +/* + * Atomically increment a specified error counter from probe context. + */ +static void dtrace_error(uint32_t *counter) +{ + /* + * Most counters stored to in probe context are per-CPU counters. + * However, there are some error conditions that are sufficiently + * arcane that they don't merit per-CPU storage. If these counters + * are incremented concurrently on different CPUs, scalability will be + * adversely affected -- but we don't expect them to be white-hot in a + * correctly constructed enabling... + */ + uint32_t oval, nval; + + do { + oval = *counter; + + if ((nval = oval + 1) == 0) { + /* + * If the counter would wrap, set it to 1 -- assuring + * that the counter is never zero when we have seen + * errors. (The counter must be 32-bits because we + * aren't guaranteed a 64-bit compare&swap operation.) + * To save this code both the infamy of being fingered + * by a priggish news story and the indignity of being + * the target of a neo-puritan witch trial, we're + * carefully avoiding any colorful description of the + * likelihood of this condition -- but suffice it to + * say that it is only slightly more likely than the + * overflow of predicate cache IDs, as discussed in + * dtrace_predicate_create(). + */ + nval = 1; + } + } while (cmpxchg(counter, oval, nval) != oval); +} + +static int dtrace_priv_kernel_destructive(dtrace_state_t *state) +{ + if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL_DESTRUCTIVE) + return 1; + + DTRACE_CPUFLAG_SET(CPU_DTRACE_KPRIV); + + return 0; +} + +static void dtrace_action_breakpoint(dtrace_ecb_t *ecb) +{ + dtrace_probe_t *probe = ecb->dte_probe; + dtrace_provider_t *prov = probe->dtpr_provider; + char c[DTRACE_FULLNAMELEN + 80], *str; + char *msg = "dtrace: breakpoint action at probe "; + char *ecbmsg = " (ecb "; + uintptr_t mask = (0xf << (sizeof(uintptr_t) * NBBY / 4)); + uintptr_t val = (uintptr_t)ecb; + int shift = (sizeof(uintptr_t) * NBBY) - 4, i = 0; + + if (dtrace_destructive_disallow) + return; + + /* + * It's impossible to be taking action on the NULL probe. + */ + ASSERT(probe != NULL); + + /* + * This is a poor man's (destitute man's?) sprintf(): we want to + * print the provider name, module name, function name and name of + * the probe, along with the hex address of the ECB with the breakpoint + * action -- all of which we must place in the character buffer by + * hand. + */ + while (*msg != '\0') + c[i++] = *msg++; + + for (str = prov->dtpv_name; *str != '\0'; str++) + c[i++] = *str; + c[i++] = ':'; + + for (str = probe->dtpr_mod; *str != '\0'; str++) + c[i++] = *str; + c[i++] = ':'; + + for (str = probe->dtpr_func; *str != '\0'; str++) + c[i++] = *str; + c[i++] = ':'; + + for (str = probe->dtpr_name; *str != '\0'; str++) + c[i++] = *str; + + while (*ecbmsg != '\0') + c[i++] = *ecbmsg++; + + while (shift >= 0) { + mask = (uintptr_t)0xf << shift; + + if (val >= ((uintptr_t)1 << shift)) + c[i++] = "0123456789abcdef"[(val & mask) >> shift]; + + shift -= 4; + } + + c[i++] = ')'; + c[i] = '\0'; + +// debug_enter(c); /* FIXME */ +} + +static void dtrace_action_panic(dtrace_ecb_t *ecb) +{ + dtrace_probe_t *probe = ecb->dte_probe; + + /* + * It's impossible to be taking action on the NULL probe. + */ + ASSERT(probe != NULL); + + if (dtrace_destructive_disallow) + return; + + if (dtrace_panicked != NULL) + return; + + if (cmpxchg(&dtrace_panicked, NULL, current) != NULL) + return; + + /* + * We won the right to panic. (We want to be sure that only one + * thread calls panic() from dtrace_probe(), and that panic() is + * called exactly once.) + */ + dtrace_panic(KERN_EMERG + "dtrace: panic action at probe %s:%s:%s:%s (ecb %p)", + probe->dtpr_provider->dtpv_name, probe->dtpr_mod, + probe->dtpr_func, probe->dtpr_name, (void *)ecb); +} + +static void dtrace_action_raise(uint64_t sig) +{ + if (dtrace_destructive_disallow) + return; + + if (sig >= _NSIG) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return; + } + + /* + * raise() has a queue depth of 1 -- we ignore all subsequent + * invocations of the raise() action. + */ + if (current->dtrace_sig == 0) + current->dtrace_sig = (uint8_t)sig; + +// current->sig_check = 1; /* FIXME */ +// aston(current); /* FIXME */ +} + +static void dtrace_action_stop(void) +{ + if (dtrace_destructive_disallow) + return; + + if (!current->dtrace_stop) { + current->dtrace_stop = 1; +// current->sig_check = 1; /* FIXME */ +// aston(current); /* FIXME */ + } +} + +static void dtrace_action_chill(dtrace_mstate_t *mstate, ktime_t val) +{ + ktime_t now; + volatile uint16_t *flags; + cpu_core_t *cpu = &cpu_core[smp_processor_id()]; + + if (dtrace_destructive_disallow) + return; + + flags = (volatile uint16_t *)&cpu->cpuc_dtrace_flags; + + now = dtrace_gethrtime(); + + if (ktime_gt(ktime_sub(now, cpu->cpu_dtrace_chillmark), + dtrace_chill_interval)) { + /* + * We need to advance the mark to the current time. + */ + cpu->cpu_dtrace_chillmark = now; + cpu->cpu_dtrace_chilled = ktime_set(0, 0); + } + + /* + * Now check to see if the requested chill time would take us over + * the maximum amount of time allowed in the chill interval. (Or + * worse, if the calculation itself induces overflow.) + */ + if (ktime_gt(ktime_add(cpu->cpu_dtrace_chilled, val), + dtrace_chill_max) || + ktime_lt(ktime_add(cpu->cpu_dtrace_chilled, val), + cpu->cpu_dtrace_chilled)) { + *flags |= CPU_DTRACE_ILLOP; + return; + } + + while (ktime_lt(ktime_sub(dtrace_gethrtime(), now), val)) + continue; + + /* + * Normally, we assure that the value of the variable "timestamp" does + * not change within an ECB. The presence of chill() represents an + * exception to this rule, however. + */ + mstate->dtms_present &= ~DTRACE_MSTATE_TIMESTAMP; + cpu->cpu_dtrace_chilled = ktime_add(cpu->cpu_dtrace_chilled, val); +} + +static void dtrace_action_ustack(dtrace_mstate_t *mstate, + dtrace_state_t *state, uint64_t *buf, + uint64_t arg) +{ + int nframes = DTRACE_USTACK_NFRAMES(arg); + int strsize = DTRACE_USTACK_STRSIZE(arg); + uint64_t *pcs = &buf[1], *fps; + char *str = (char *)&pcs[nframes]; + int size, offs = 0, i, j; + uintptr_t old = mstate->dtms_scratch_ptr, saved; + uint16_t *flags = &cpu_core[ + smp_processor_id() + ].cpuc_dtrace_flags; + char *sym; + + /* + * Should be taking a faster path if string space has not been + * allocated. + */ + ASSERT(strsize != 0); + + /* + * We will first allocate some temporary space for the frame pointers. + */ + fps = (uint64_t *)P2ROUNDUP(mstate->dtms_scratch_ptr, 8); + size = (uintptr_t)fps - mstate->dtms_scratch_ptr + + (nframes * sizeof (uint64_t)); + + if (!DTRACE_INSCRATCH(mstate, size)) { + /* + * Not enough room for our frame pointers -- need to indicate + * that we ran out of scratch space. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + return; + } + + mstate->dtms_scratch_ptr += size; + saved = mstate->dtms_scratch_ptr; + + /* + * Now get a stack with both program counters and frame pointers. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_getufpstack(buf, fps, nframes + 1); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + /* + * If that faulted, we're cooked. + */ + if (*flags & CPU_DTRACE_FAULT) + goto out; + + /* + * Now we want to walk up the stack, calling the USTACK helper. For + * each iteration, we restore the scratch pointer. + */ + for (i = 0; i < nframes; i++) { + mstate->dtms_scratch_ptr = saved; + + if (offs >= strsize) + break; + + sym = (char *)(uintptr_t)dtrace_helper( + DTRACE_HELPER_ACTION_USTACK, + mstate, state, pcs[i], fps[i]); + + /* + * If we faulted while running the helper, we're going to + * clear the fault and null out the corresponding string. + */ + if (*flags & CPU_DTRACE_FAULT) { + *flags &= ~CPU_DTRACE_FAULT; + str[offs++] = '\0'; + continue; + } + + if (sym == NULL) { + str[offs++] = '\0'; + continue; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + + /* + * Now copy in the string that the helper returned to us. + */ + for (j = 0; offs + j < strsize; j++) { + if ((str[offs + j] = sym[j]) == '\0') + break; + } + + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + offs += j + 1; + } + + /* + * If we didn't have room for all of the strings, we don't abort + * processing -- this needn't be a fatal error -- but we still want + * to increment a counter (dts_stkstroverflows) to allow this condition + * to be warned about. (If this is from a jstack() action, it is + * easily tuned via jstackstrsize.) + */ + if (offs >= strsize) + dtrace_error(&state->dts_stkstroverflows); + + while (offs < strsize) + str[offs++] = '\0'; + +out: + mstate->dtms_scratch_ptr = old; +} + +void dtrace_probe(dtrace_id_t id, uintptr_t arg0, uintptr_t arg1, + uintptr_t arg2, uintptr_t arg3, uintptr_t arg4) +{ + processorid_t cpuid; + dtrace_icookie_t cookie; + dtrace_probe_t *probe; + dtrace_mstate_t mstate; + dtrace_ecb_t *ecb; + dtrace_action_t *act; + intptr_t offs; + size_t size; + int vtime, onintr; + volatile uint16_t *flags; + ktime_t now; + +#ifdef FIXME + /* + * Kick out immediately if this CPU is still being born (in which case + * curthread will be set to -1) or the current thread can't allow + * probes in its current context. + */ + if (((uintptr_t)curthread & 1) || (curthread->t_flag & T_DONTDTRACE)) + return; +#endif + + local_irq_save(cookie); + probe = dtrace_probe_lookup_id(id); + cpuid = smp_processor_id(); + onintr = in_interrupt(); + + if (!onintr && probe->dtpr_predcache != DTRACE_CACHEIDNONE && + probe->dtpr_predcache == current->predcache) { + /* + * We have hit in the predicate cache; we know that + * this predicate would evaluate to be false. + */ + local_irq_restore(cookie); + return; + } + +#ifdef FIXME + if (panic_quiesce) { + /* + * We don't trace anything if we're panicking. + */ + local_irq_restore(cookie); + return; + } +#endif + + now = dtrace_gethrtime(); + vtime = dtrace_vtime_references != 0; + + if (vtime && ktime_nz(current->dtrace_start)) + current->dtrace_vtime = + ktime_add(current->dtrace_vtime, + ktime_sub(now,current->dtrace_start)); + + mstate.dtms_difo = NULL; + mstate.dtms_probe = probe; + mstate.dtms_strtok = (uintptr_t)NULL; + mstate.dtms_arg[0] = arg0; + mstate.dtms_arg[1] = arg1; + mstate.dtms_arg[2] = arg2; + mstate.dtms_arg[3] = arg3; + mstate.dtms_arg[4] = arg4; + + flags = (volatile uint16_t *)&cpu_core[cpuid].cpuc_dtrace_flags; + + for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) { + dtrace_predicate_t *pred = ecb->dte_predicate; + dtrace_state_t *state = ecb->dte_state; + dtrace_buffer_t *buf = &state->dts_buffer[cpuid]; + dtrace_buffer_t *aggbuf = &state->dts_aggbuffer[cpuid]; + dtrace_vstate_t *vstate = &state->dts_vstate; + dtrace_provider_t *prov = probe->dtpr_provider; + int committed = 0; + caddr_t tomax; + + /* + * A little subtlety with the following (seemingly innocuous) + * declaration of the automatic 'val': by looking at the + * code, you might think that it could be declared in the + * action processing loop, below. (That is, it's only used in + * the action processing loop.) However, it must be declared + * out of that scope because in the case of DIF expression + * arguments to aggregating actions, one iteration of the + * action loop will use the last iteration's value. + */ + uint64_t val = 0; + + mstate.dtms_present = DTRACE_MSTATE_ARGS | DTRACE_MSTATE_PROBE; + *flags &= ~CPU_DTRACE_ERROR; + + if (prov == dtrace_provider) { + /* + * If dtrace itself is the provider of this probe, + * we're only going to continue processing the ECB if + * arg0 (the dtrace_state_t) is equal to the ECB's + * creating state. (This prevents disjoint consumers + * from seeing one another's metaprobes.) + */ + if (arg0 != (uint64_t)(uintptr_t)state) + continue; + } + + if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE) { + /* + * We're not currently active. If our provider isn't + * the dtrace pseudo provider, we're not interested. + */ + if (prov != dtrace_provider) + continue; + + /* + * Now we must further check if we are in the BEGIN + * probe. If we are, we will only continue orocessing + * if we're still in WARMUP -- if one BEGIN enabling + * has invoked the exit() action, we don't want to + * evaluate subsequent BEGIN enablings. + */ + if (probe->dtpr_id == dtrace_probeid_begin && + state->dts_activity != DTRACE_ACTIVITY_WARMUP) { + ASSERT(state->dts_activity == + DTRACE_ACTIVITY_DRAINING); + continue; + } + } + + if (ecb->dte_cond) { + /* + * If the dte_cond bits indicate that this + * consumer is only allowed to see user-mode firings + * of this probe, call the provider's dtps_usermode() + * entry point to check that the probe was fired + * while in a user context. Skip this ECB if that's + * not the case. + */ + if ((ecb->dte_cond & DTRACE_COND_USERMODE) && + prov->dtpv_pops.dtps_usermode( + prov->dtpv_arg, probe->dtpr_id, probe->dtpr_arg + ) == 0) + continue; + + /* + * This is more subtle than it looks. We have to be + * absolutely certain that current_cred() isn't going + * to change out from under us so it's only legit to + * examine that structure if we're in constrained + * situations. Currently, the only times we'll use this + * check is if a non-super-user has enabled the + * profile or syscall providers -- providers that + * allow visibility of all processes. For the + * profile case, the check above will ensure that + * we're examining a user context. + */ + if (ecb->dte_cond & DTRACE_COND_OWNER) { + const cred_t *cr; + const cred_t *s_cr = + ecb->dte_state->dts_cred.dcr_cred; + + ASSERT(s_cr != NULL); + + if ((cr = current_cred()) == NULL || + s_cr->euid != cr->euid || + s_cr->euid != cr->uid || + s_cr->euid != cr->suid || + s_cr->egid != cr->egid || + s_cr->egid != cr->gid || + s_cr->egid != cr->sgid) + continue; + } + } + + if (ktime_gt(ktime_sub(now, state->dts_alive), + dtrace_deadman_timeout)) { + /* + * We seem to be dead. Unless we (a) have kernel + * destructive permissions (b) have expicitly enabled + * destructive actions and (c) destructive actions have + * not been disabled, we're going to transition into + * the KILLED state, from which no further processing + * on this state will be performed. + */ + if (!dtrace_priv_kernel_destructive(state) || + !state->dts_cred.dcr_destructive || + dtrace_destructive_disallow) { + dtrace_activity_t *activity = + &state->dts_activity; + dtrace_activity_t curr; + + do { + curr = state->dts_activity; + } while (cmpxchg(activity, curr, + DTRACE_ACTIVITY_KILLED) != curr); + + continue; + } + } + + if ((offs = dtrace_buffer_reserve(buf, ecb->dte_needed, + ecb->dte_alignment, state, + &mstate)) < 0) + continue; + + tomax = buf->dtb_tomax; + ASSERT(tomax != NULL); + + if (ecb->dte_size != 0) + DTRACE_STORE(uint32_t, tomax, offs, ecb->dte_epid); + + mstate.dtms_epid = ecb->dte_epid; + mstate.dtms_present |= DTRACE_MSTATE_EPID; + + if (state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) + mstate.dtms_access = DTRACE_ACCESS_KERNEL; + else + mstate.dtms_access = 0; + + if (pred != NULL) { + dtrace_difo_t *dp = pred->dtp_difo; + int rval; + + rval = dtrace_dif_emulate(dp, &mstate, vstate, state); + + if (!(*flags & CPU_DTRACE_ERROR) && !rval) { + dtrace_cacheid_t cid = + probe->dtpr_predcache; + + if (cid != DTRACE_CACHEIDNONE && !onintr) { + /* + * Update the predicate cache... + */ + ASSERT(cid == pred->dtp_cacheid); + current->predcache = cid; + } + + continue; + } + } + + for (act = ecb->dte_action; + !(*flags & CPU_DTRACE_ERROR) && act != NULL; + act = act->dta_next) { + size_t valoffs; + dtrace_difo_t *dp; + dtrace_recdesc_t *rec = &act->dta_rec; + + size = rec->dtrd_size; + valoffs = offs + rec->dtrd_offset; + + if (DTRACEACT_ISAGG(act->dta_kind)) { + uint64_t v = 0xbad; + dtrace_aggregation_t *agg; + + agg = (dtrace_aggregation_t *)act; + + if ((dp = act->dta_difo) != NULL) + v = dtrace_dif_emulate(dp, &mstate, + vstate, state); + + if (*flags & CPU_DTRACE_ERROR) + continue; + + /* + * Note that we always pass the expression + * value from the previous iteration of the + * action loop. This value will only be used + * if there is an expression argument to the + * aggregating action, denoted by the + * dtag_hasarg field. + */ + dtrace_aggregate(agg, buf, offs, aggbuf, v, + val); + continue; + } + + switch (act->dta_kind) { + case DTRACEACT_STOP: + if (dtrace_priv_proc_destructive(state)) + dtrace_action_stop(); + continue; + + case DTRACEACT_BREAKPOINT: + if (dtrace_priv_kernel_destructive(state)) + dtrace_action_breakpoint(ecb); + continue; + + case DTRACEACT_PANIC: + if (dtrace_priv_kernel_destructive(state)) + dtrace_action_panic(ecb); + continue; + + case DTRACEACT_STACK: + if (!dtrace_priv_kernel(state)) + continue; + + dtrace_getpcstack( + (uint64_t *)(tomax + valoffs), + size / sizeof(pc_t), + probe->dtpr_aframes, + DTRACE_ANCHORED(probe) + ? NULL + : (uint32_t *)arg0); + + continue; + + case DTRACEACT_JSTACK: + case DTRACEACT_USTACK: + if (!dtrace_priv_proc(state)) + continue; + + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate.dtms_probe) && + in_interrupt()) { + int depth = DTRACE_USTACK_NFRAMES( + rec->dtrd_arg) + 1; + + dtrace_bzero((void *)(tomax + valoffs), + DTRACE_USTACK_STRSIZE( + rec->dtrd_arg) + + depth * sizeof(uint64_t)); + + continue; + } + + if (DTRACE_USTACK_STRSIZE(rec->dtrd_arg) != 0 && + current->dtrace_helpers != NULL) { + /* + * This is the slow path -- we have + * allocated string space, and we're + * getting the stack of a process that + * has helpers. Call into a separate + * routine to perform this processing. + */ + dtrace_action_ustack( + &mstate, state, + (uint64_t *)(tomax + valoffs), + rec->dtrd_arg); + continue; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_getupcstack( + (uint64_t *)(tomax + valoffs), + DTRACE_USTACK_NFRAMES(rec->dtrd_arg) + + 1); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + continue; + + default: + break; + } + + dp = act->dta_difo; + ASSERT(dp != NULL); + + val = dtrace_dif_emulate(dp, &mstate, vstate, state); + + if (*flags & CPU_DTRACE_ERROR) + continue; + + switch (act->dta_kind) { + case DTRACEACT_SPECULATE: + ASSERT(buf == &state->dts_buffer[cpuid]); + buf = dtrace_speculation_buffer(state, cpuid, + val); + + if (buf == NULL) { + *flags |= CPU_DTRACE_DROP; + continue; + } + + offs = dtrace_buffer_reserve(buf, + ecb->dte_needed, + ecb->dte_alignment, + state, NULL); + + if (offs < 0) { + *flags |= CPU_DTRACE_DROP; + continue; + } + + tomax = buf->dtb_tomax; + ASSERT(tomax != NULL); + + if (ecb->dte_size != 0) + DTRACE_STORE(uint32_t, tomax, offs, + ecb->dte_epid); + + continue; + + case DTRACEACT_CHILL: + if (dtrace_priv_kernel_destructive(state)) + dtrace_action_chill(&mstate, + ns_to_ktime(val)); + + continue; + + case DTRACEACT_RAISE: + if (dtrace_priv_proc_destructive(state)) + dtrace_action_raise(val); + + continue; + + case DTRACEACT_COMMIT: + ASSERT(!committed); + + /* + * We need to commit our buffer state. + */ + if (ecb->dte_size) + buf->dtb_offset = offs + ecb->dte_size; + + buf = &state->dts_buffer[cpuid]; + dtrace_speculation_commit(state, cpuid, val); + committed = 1; + continue; + + case DTRACEACT_DISCARD: + dtrace_speculation_discard(state, cpuid, val); + continue; + + case DTRACEACT_DIFEXPR: + case DTRACEACT_LIBACT: + case DTRACEACT_PRINTF: + case DTRACEACT_PRINTA: + case DTRACEACT_SYSTEM: + case DTRACEACT_FREOPEN: + break; + + case DTRACEACT_SYM: + case DTRACEACT_MOD: + if (!dtrace_priv_kernel(state)) + continue; + break; + + case DTRACEACT_USYM: + case DTRACEACT_UMOD: + case DTRACEACT_UADDR: { + pid_t pid = current->pid; + + if (!dtrace_priv_proc(state)) + continue; + + DTRACE_STORE(uint64_t, tomax, valoffs, + (uint64_t)pid); + DTRACE_STORE(uint64_t, tomax, + valoffs + sizeof(uint64_t), val); + + continue; + } + + case DTRACEACT_EXIT: { + /* + * For the exit action, we are going to attempt + * to atomically set our activity to be + * draining. If this fails (either because + * another CPU has beat us to the exit action, + * or because our current activity is something + * other than ACTIVE or WARMUP), we will + * continue. This assures that the exit action + * can be successfully recorded at most once + * when we're in the ACTIVE state. If we're + * encountering the exit() action while in + * COOLDOWN, however, we want to honor the new + * status code. (We know that we're the only + * thread in COOLDOWN, so there is no race.) + */ + dtrace_activity_t *activity = + &state->dts_activity; + dtrace_activity_t curr = + state->dts_activity; + + if (curr == DTRACE_ACTIVITY_COOLDOWN) + break; + + if (curr != DTRACE_ACTIVITY_WARMUP) + curr = DTRACE_ACTIVITY_ACTIVE; + + if (cmpxchg(activity, curr, + DTRACE_ACTIVITY_DRAINING) != curr) { + *flags |= CPU_DTRACE_DROP; + continue; + } + + break; + } + + default: + ASSERT(0); + } + + if (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF) { + uintptr_t end = valoffs + size; + + if (!dtrace_vcanload((void *)(uintptr_t)val, + &dp->dtdo_rtype, &mstate, + vstate)) + continue; + + /* + * If this is a string, we're going to only + * load until we find the zero byte -- after + * which we'll store zero bytes. + */ + if (dp->dtdo_rtype.dtdt_kind == + DIF_TYPE_STRING) { + char c = '\0' + 1; + int intuple = act->dta_intuple; + size_t s; + + for (s = 0; s < size; s++) { + if (c != '\0') + c = dtrace_load8(val++); + + DTRACE_STORE(uint8_t, tomax, + valoffs++, c); + + if (c == '\0' && intuple) + break; + } + + continue; + } + + while (valoffs < end) + DTRACE_STORE(uint8_t, tomax, valoffs++, + dtrace_load8(val++)); + + continue; + } + + switch (size) { + case 0: + break; + case sizeof(uint8_t): + DTRACE_STORE(uint8_t, tomax, valoffs, val); + break; + case sizeof(uint16_t): + DTRACE_STORE(uint16_t, tomax, valoffs, val); + break; + case sizeof(uint32_t): + DTRACE_STORE(uint32_t, tomax, valoffs, val); + break; + case sizeof(uint64_t): + DTRACE_STORE(uint64_t, tomax, valoffs, val); + break; + default: + /* + * Any other size should have been returned by + * reference, not by value. + */ + ASSERT(0); + break; + } + } + + if (*flags & CPU_DTRACE_DROP) + continue; + + if (*flags & CPU_DTRACE_FAULT) { + int ndx; + dtrace_action_t *err; + + buf->dtb_errors++; + + if (probe->dtpr_id == dtrace_probeid_error) { + /* + * There's nothing we can do -- we had an + * error on the error probe. We bump an + * error counter to at least indicate that + * this condition happened. + */ + dtrace_error(&state->dts_dblerrors); + continue; + } + + if (vtime) + /* + * Before recursing on dtrace_probe(), we + * need to explicitly clear out our start + * time to prevent it from being accumulated + * into t_dtrace_vtime. + */ + current->dtrace_start = ktime_set(0, 0); + + /* + * Iterate over the actions to figure out which action + * we were processing when we experienced the error. + * Note that act points _past_ the faulting action; if + * act is ecb->dte_action, the fault was in the + * predicate, if it's ecb->dte_action->dta_next it's + * in action #1, and so on. + */ + for (err = ecb->dte_action, ndx = 0; + err != act; err = err->dta_next, ndx++) + continue; + + dtrace_probe_error( + state, ecb->dte_epid, ndx, + (mstate.dtms_present & DTRACE_MSTATE_FLTOFFS) + ? mstate.dtms_fltoffs + : -1, + DTRACE_FLAGS2FLT(*flags), + cpu_core[cpuid].cpuc_dtrace_illval); + + continue; + } + + if (!committed) + buf->dtb_offset = offs + ecb->dte_size; + } + + if (vtime) + current->dtrace_start = dtrace_gethrtime(); + + local_irq_restore(cookie); +} +EXPORT_SYMBOL(dtrace_probe); + +void dtrace_probe_init(void) +{ + dtrace_id_t id; + int err; + + idr_init(&dtrace_probe_idr); + + /* + * We create a ID 0 entry as a sentinel, so we can always depend on it + * being the very first entry. This is used in functionality that runs + * through the list of probes. + * + * We need to drop our locks when calling idr_pre_get(), so we try to + * get them back right after. + */ +again: + mutex_unlock(&cpu_lock); + mutex_unlock(&dtrace_provider_lock); + mutex_unlock(&dtrace_lock); + + idr_pre_get(&dtrace_probe_idr, __GFP_NOFAIL); + + mutex_lock(&dtrace_lock); + mutex_lock(&dtrace_provider_lock); + mutex_lock(&cpu_lock); + + err = idr_get_new(&dtrace_probe_idr, NULL, &id); + if (err == -EAGAIN) + goto again; + + ASSERT(id == 0); +} + +void dtrace_probe_exit(void) +{ + idr_remove_all(&dtrace_probe_idr); + idr_destroy(&dtrace_probe_idr); +} + +void dtrace_probe_remove_id(dtrace_id_t id) +{ + return idr_remove(&dtrace_probe_idr, id); +} + +dtrace_probe_t *dtrace_probe_lookup_id(dtrace_id_t id) +{ + return idr_find(&dtrace_probe_idr, id); +} + +static int dtrace_probe_lookup_match(dtrace_probe_t *probe, void *arg) +{ + *((dtrace_id_t *)arg) = probe->dtpr_id; + + return DTRACE_MATCH_DONE; +} + +dtrace_id_t dtrace_probe_lookup(dtrace_provider_id_t prid, const char *mod, + const char *func, const char *name) +{ + dtrace_probekey_t pkey; + dtrace_id_t id; + int match; + + /* FIXME: Maybe? We really should protect against this. */ + if (mutex_is_locked(&dtrace_lock)) { + WARN(1, "dtrace_probe_lookup() called with dtrace_lock held!"); + return 0; + } + + pkey.dtpk_prov = ((dtrace_provider_t *)prid)->dtpv_name; + pkey.dtpk_pmatch = &dtrace_match_string; + pkey.dtpk_mod = mod; + pkey.dtpk_mmatch = mod ? &dtrace_match_string : &dtrace_match_nul; + pkey.dtpk_func = func; + pkey.dtpk_fmatch = func ? &dtrace_match_string : &dtrace_match_nul; + pkey.dtpk_name = name; + pkey.dtpk_nmatch = name ? &dtrace_match_string : &dtrace_match_nul; + pkey.dtpk_id = DTRACE_IDNONE; + + mutex_lock(&dtrace_lock); + match = dtrace_match(&pkey, DTRACE_PRIV_ALL, 0, + dtrace_probe_lookup_match, &id); + mutex_unlock(&dtrace_lock); + + ASSERT(match == 1 || match == 0); + + return match ? id : 0; +} +EXPORT_SYMBOL(dtrace_probe_lookup); + +dtrace_probe_t *dtrace_probe_get_next(dtrace_id_t id) +{ + return idr_get_next(&dtrace_probe_idr, &id); +} + +int dtrace_probe_for_each(int (*fn)(int id, void *p, void *data), void *data) +{ + return idr_for_each(&dtrace_probe_idr, fn, data); +} diff --git a/dtrace/dtrace_probe_ctx.c b/dtrace/dtrace_probe_ctx.c new file mode 100644 index 000000000000..1fa96cdebad0 --- /dev/null +++ b/dtrace/dtrace_probe_ctx.c @@ -0,0 +1,538 @@ +/* + * FILE: dtrace_probe_ctx.c + * DESCRIPTION: Dynamic Tracing: probe context functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include "dtrace.h" + +void dtrace_panic(const char *fmt, ...) +{ + va_list alist; + + va_start(alist, fmt); + vprintk(fmt, alist); + va_end(alist); + + BUG(); +} + +int dtrace_assfail(const char *a, const char *f, int l) +{ + dtrace_panic(KERN_EMERG "assertion failed: %s, file: %s, line: %d", + a, f, l); + + /* + * FIXME: We can do better than this. The OpenSolaris DTrace source + * states that this cannot be optimized away. + */ + return a[(uintptr_t)f]; +} +EXPORT_SYMBOL(dtrace_assfail); + +#define DT_MASK_LO 0x00000000FFFFFFFFULL + +static void dtrace_add_128(uint64_t *addend1, uint64_t *addend2, uint64_t *sum) +{ + uint64_t result[2]; + + result[0] = addend1[0] + addend2[0]; + result[1] = addend1[1] + addend2[1] + + (result[0] < addend1[0] || result[0] < addend2[0] ? 1 : 0); + + sum[0] = result[0]; + sum[1] = result[1]; +} + +static void dtrace_shift_128(uint64_t *a, int b) +{ + uint64_t mask; + + if (b == 0) + return; + + if (b < 0) { + b = -b; + + if (b >= 64) { + a[0] = a[1] >> (b - 64); + a[1] = 0; + } else { + a[0] >>= b; + mask = 1LL << (64 - b); + mask -= 1; + a[0] |= ((a[1] & mask) << (64 - b)); + a[1] >>= b; + } + } else { + if (b >= 64) { + a[1] = a[0] << (b - 64); + a[0] = 0; + } else { + a[1] <<= b; + mask = a[0] >> (64 - b); + a[1] |= mask; + a[0] <<= b; + } + } +} + +static void dtrace_multiply_128(uint64_t factor1, uint64_t factor2, + uint64_t *product) +{ + uint64_t hi1, hi2, lo1, lo2; + uint64_t tmp[2]; + + hi1 = factor1 >> 32; + hi2 = factor2 >> 32; + + lo1 = factor1 & DT_MASK_LO; + lo2 = factor2 & DT_MASK_LO; + + product[0] = lo1 * lo2; + product[1] = hi1 * hi2; + + tmp[0] = hi1 * lo2; + tmp[1] = 0; + dtrace_shift_128(tmp, 32); + dtrace_add_128(product, tmp, product); + + tmp[0] = hi2 * lo1; + tmp[1] = 0; + dtrace_shift_128(tmp, 32); + dtrace_add_128(product, tmp, product); +} + +void dtrace_aggregate_min(uint64_t *oval, uint64_t nval, uint64_t arg) +{ + if ((int64_t)nval < (int64_t)*oval) + *oval = nval; +} + +void dtrace_aggregate_max(uint64_t *oval, uint64_t nval, uint64_t arg) +{ + if ((int64_t)nval > (int64_t)*oval) + *oval = nval; +} + +void dtrace_aggregate_quantize(uint64_t *quanta, uint64_t nval, uint64_t incr) +{ + int i, zero = DTRACE_QUANTIZE_ZEROBUCKET; + int64_t val = (int64_t)nval; + + if (val < 0) { + for (i = 0; i < zero; i++) { + if (val <= DTRACE_QUANTIZE_BUCKETVAL(i)) { + quanta[i] += incr; + + return; + } + } + } else { + for (i = zero + 1; i < DTRACE_QUANTIZE_NBUCKETS; i++) { + if (val < DTRACE_QUANTIZE_BUCKETVAL(i)) { + quanta[i - 1] += incr; + + return; + } + } + + quanta[DTRACE_QUANTIZE_NBUCKETS - 1] += incr; + + return; + } + + ASSERT(0); +} + +void dtrace_aggregate_lquantize(uint64_t *lquanta, uint64_t nval, + uint64_t incr) +{ + uint64_t arg = *lquanta++; + int32_t base = DTRACE_LQUANTIZE_BASE(arg); + uint16_t step = DTRACE_LQUANTIZE_STEP(arg); + uint16_t levels = DTRACE_LQUANTIZE_LEVELS(arg); + int32_t val = (int32_t)nval, level; + + ASSERT(step != 0); + ASSERT(levels != 0); + + if (val < base) { + lquanta[0] += incr; + + return; + } + + level = (val - base) / step; + + if (level < levels) { + lquanta[level + 1] += incr; + + return; + } + + lquanta[levels + 1] += incr; +} + +void dtrace_aggregate_avg(uint64_t *data, uint64_t nval, uint64_t arg) +{ + data[0]++; + data[1] += nval; +} + +void dtrace_aggregate_stddev(uint64_t *data, uint64_t nval, uint64_t arg) +{ + int64_t snval = (int64_t)nval; + uint64_t tmp[2]; + + data[0]++; + data[1] += nval; + + if (snval < 0) + snval = -snval; + + dtrace_multiply_128((uint64_t)snval, (uint64_t)snval, tmp); + dtrace_add_128(data + 2, tmp, data + 2); +} + +void dtrace_aggregate_count(uint64_t *oval, uint64_t nval, uint64_t arg) +{ + *oval = *oval + 1; +} + +void dtrace_aggregate_sum(uint64_t *oval, uint64_t nval, uint64_t arg) +{ + *oval += nval; +} + +/* + * DTrace Aggregation Buffers + * + * Aggregation buffers use much of the same mechanism as described above + * ("DTrace Buffers"). However, because an aggregation is fundamentally a + * hash, there exists dynamic metadata associated with an aggregation buffer + * that is not associated with other kinds of buffers. This aggregation + * metadata is _only_ relevant for the in-kernel implementation of + * aggregations; it is not actually relevant to user-level consumers. To do + * this, we allocate dynamic aggregation data (hash keys and hash buckets) + * starting below the _limit_ of the buffer, and we allocate data from the + * _base_ of the buffer. When the aggregation buffer is copied out, _only_ the + * data is copied out; the metadata is simply discarded. Schematically, + * aggregation buffers look like: + * + * base of data buffer ---> +-------+------+-----------+-------+ + * | aggid | key | value | aggid | + * +-------+------+-----------+-------+ + * | key | + * +-------+-------+-----+------------+ + * | value | aggid | key | value | + * +-------+------++-----+------+-----+ + * | aggid | key | value | | + * +-------+------+-------------+ | + * | || | + * | || | + * | \/ | + * : : + * . . + * . . + * . . + * : : + * | /\ | + * | || +------------+ + * | || | | + * +---------------------+ | + * | hash keys | + * | (dtrace_aggkey structures) | + * | | + * +----------------------------------+ + * | hash buckets | + * | (dtrace_aggbuffer structure) | + * | | + * limit of data buffer ---> +----------------------------------+ + * + * As implied above, just as we assure that ECBs always store a constant + * amount of data, we assure that a given aggregation -- identified by its + * aggregation ID -- always stores data of a constant quantity and type. + * As with EPIDs, this allows the aggregation ID to serve as the metadata for a + * given record. + * + * Note that the size of the dtrace_aggkey structure must be sizeof (uintptr_t) + * aligned. (If this the structure changes such that this becomes false, an + * assertion will fail in dtrace_aggregate().) + */ +#define DTRACE_AGGHASHSIZE_SLEW 17 + +typedef struct dtrace_aggkey { + uint32_t dtak_hashval; /* hash value */ + uint32_t dtak_action:4; /* action -- 4 bits */ + uint32_t dtak_size:28; /* size -- 28 bits */ + caddr_t dtak_data; /* data pointer */ + struct dtrace_aggkey *dtak_next; /* next in hash chain */ +} dtrace_aggkey_t; + +typedef struct dtrace_aggbuffer { + uintptr_t dtagb_hashsize; /* number of buckets */ + uintptr_t dtagb_free; /* free list of keys */ + dtrace_aggkey_t **dtagb_hash; /* hash table */ +} dtrace_aggbuffer_t; + +#define DTRACEACT_ISSTRING(act) \ + ((act)->dta_kind == DTRACEACT_DIFEXPR && \ + (act)->dta_difo->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) + +/* + * Aggregate given the tuple in the principal data buffer, and the aggregating + * action denoted by the specified dtrace_aggregation_t. The aggregation + * buffer is specified as the buf parameter. This routine does not return + * failure; if there is no space in the aggregation buffer, the data will be + * dropped, and a corresponding counter incremented. + */ +void dtrace_aggregate(dtrace_aggregation_t *agg, dtrace_buffer_t *dbuf, + intptr_t offset, dtrace_buffer_t *buf, uint64_t expr, + uint64_t arg) +{ + dtrace_recdesc_t *rec = &agg->dtag_action.dta_rec; + uint32_t i, ndx, size, fsize; + uint32_t align = sizeof (uint64_t) - 1; + dtrace_aggbuffer_t *agb; + dtrace_aggkey_t *key; + uint32_t hashval = 0, limit, isstr; + caddr_t tomax, data, kdata; + dtrace_actkind_t action; + dtrace_action_t *act; + uintptr_t offs; + + if (buf == NULL) + return; + + if (!agg->dtag_hasarg) + /* + * Currently, only quantize() and lquantize() take additional + * arguments, and they have the same semantics: an increment + * value that defaults to 1 when not present. If additional + * aggregating actions take arguments, the setting of the + * default argument value will presumably have to become more + * sophisticated... + */ + arg = 1; + + action = agg->dtag_action.dta_kind - DTRACEACT_AGGREGATION; + size = rec->dtrd_offset - agg->dtag_base; + fsize = size + rec->dtrd_size; + + ASSERT(dbuf->dtb_tomax != NULL); + data = dbuf->dtb_tomax + offset + agg->dtag_base; + + if ((tomax = buf->dtb_tomax) == NULL) { + dtrace_buffer_drop(buf); + return; + } + + /* + * The metastructure is always at the bottom of the buffer. + */ + agb = (dtrace_aggbuffer_t *)(tomax + buf->dtb_size - + sizeof (dtrace_aggbuffer_t)); + + if (buf->dtb_offset == 0) { + /* + * We just kludge up approximately 1/8th of the size to be + * buckets. If this guess ends up being routinely + * off-the-mark, we may need to dynamically readjust this + * based on past performance. + */ + uintptr_t hashsize = (buf->dtb_size >> 3) / + sizeof (uintptr_t); + + if ((uintptr_t)agb - hashsize * sizeof (dtrace_aggkey_t *) < + (uintptr_t)tomax || hashsize == 0) { + /* + * We've been given a ludicrously small buffer; + * increment our drop count and leave. + */ + dtrace_buffer_drop(buf); + return; + } + + /* + * And now, a pathetic attempt to try to get a an odd (or + * perchance, a prime) hash size for better hash distribution. + */ + if (hashsize > (DTRACE_AGGHASHSIZE_SLEW << 3)) + hashsize -= DTRACE_AGGHASHSIZE_SLEW; + + agb->dtagb_hashsize = hashsize; + agb->dtagb_hash = (dtrace_aggkey_t **)((uintptr_t)agb - + agb->dtagb_hashsize * sizeof (dtrace_aggkey_t *)); + agb->dtagb_free = (uintptr_t)agb->dtagb_hash; + + for (i = 0; i < agb->dtagb_hashsize; i++) + agb->dtagb_hash[i] = NULL; + } + + ASSERT(agg->dtag_first != NULL); + ASSERT(agg->dtag_first->dta_intuple); + + /* + * Calculate the hash value based on the key. Note that we _don't_ + * include the aggid in the hashing (but we will store it as part of + * the key). The hashing algorithm is Bob Jenkins' "One-at-a-time" + * algorithm: a simple, quick algorithm that has no known funnels, and + * gets good distribution in practice. The efficacy of the hashing + * algorithm (and a comparison with other algorithms) may be found by + * running the ::dtrace_aggstat MDB dcmd. + */ + for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) { + i = act->dta_rec.dtrd_offset - agg->dtag_base; + limit = i + act->dta_rec.dtrd_size; + ASSERT(limit <= size); + isstr = DTRACEACT_ISSTRING(act); + + for (; i < limit; i++) { + hashval += data[i]; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + + if (isstr && data[i] == '\0') + break; + } + } + + hashval += (hashval << 3); + hashval ^= (hashval >> 11); + hashval += (hashval << 15); + + /* + * Yes, the divide here is expensive -- but it's generally the least + * of the performance issues given the amount of data that we iterate + * over to compute hash values, compare data, etc. + */ + ndx = hashval % agb->dtagb_hashsize; + + for (key = agb->dtagb_hash[ndx]; key != NULL; key = key->dtak_next) { + ASSERT((caddr_t)key >= tomax); + ASSERT((caddr_t)key < tomax + buf->dtb_size); + + if (hashval != key->dtak_hashval || key->dtak_size != size) + continue; + + kdata = key->dtak_data; + ASSERT(kdata >= tomax && kdata < tomax + buf->dtb_size); + + for (act = agg->dtag_first; act->dta_intuple; + act = act->dta_next) { + i = act->dta_rec.dtrd_offset - agg->dtag_base; + limit = i + act->dta_rec.dtrd_size; + ASSERT(limit <= size); + isstr = DTRACEACT_ISSTRING(act); + + for (; i < limit; i++) { + if (kdata[i] != data[i]) + goto next; + + if (isstr && data[i] == '\0') + break; + } + } + + if (action != key->dtak_action) { + /* + * We are aggregating on the same value in the same + * aggregation with two different aggregating actions. + * (This should have been picked up in the compiler, + * so we may be dealing with errant or devious DIF.) + * This is an error condition; we indicate as much, + * and return. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return; + } + + /* + * This is a hit: we need to apply the aggregator to + * the value at this key. + */ + agg->dtag_aggregate((uint64_t *)(kdata + size), expr, arg); + return; +next: + continue; + } + + /* + * We didn't find it. We need to allocate some zero-filled space, + * link it into the hash table appropriately, and apply the aggregator + * to the (zero-filled) value. + */ + offs = buf->dtb_offset; + while (offs & (align - 1)) + offs += sizeof (uint32_t); + + /* + * If we don't have enough room to both allocate a new key _and_ + * its associated data, increment the drop count and return. + */ + if ((uintptr_t)tomax + offs + fsize > + agb->dtagb_free - sizeof (dtrace_aggkey_t)) { + dtrace_buffer_drop(buf); + return; + } + + ASSERT(!(sizeof (dtrace_aggkey_t) & (sizeof (uintptr_t) - 1))); + key = (dtrace_aggkey_t *)(agb->dtagb_free - sizeof (dtrace_aggkey_t)); + agb->dtagb_free -= sizeof (dtrace_aggkey_t); + + key->dtak_data = kdata = tomax + offs; + buf->dtb_offset = offs + fsize; + + /* + * Now copy the data across. + */ + *((dtrace_aggid_t *)kdata) = agg->dtag_id; + + for (i = sizeof (dtrace_aggid_t); i < size; i++) + kdata[i] = data[i]; + + /* + * Because strings are not zeroed out by default, we need to iterate + * looking for actions that store strings, and we need to explicitly + * pad these strings out with zeroes. + */ + for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) { + int nul; + + if (!DTRACEACT_ISSTRING(act)) + continue; + + i = act->dta_rec.dtrd_offset - agg->dtag_base; + limit = i + act->dta_rec.dtrd_size; + ASSERT(limit <= size); + + for (nul = 0; i < limit; i++) { + if (nul) { + kdata[i] = '\0'; + continue; + } + + if (data[i] != '\0') + continue; + + nul = 1; + } + } + + for (i = size; i < fsize; i++) + kdata[i] = 0; + + key->dtak_hashval = hashval; + key->dtak_size = size; + key->dtak_action = action; + key->dtak_next = agb->dtagb_hash[ndx]; + agb->dtagb_hash[ndx] = key; + + /* + * Finally, apply the aggregator. + */ + *((uint64_t *)(key->dtak_data + size)) = agg->dtag_initial; + agg->dtag_aggregate((uint64_t *)(key->dtak_data + size), expr, arg); +} diff --git a/dtrace/dtrace_ptofapi.c b/dtrace/dtrace_ptofapi.c new file mode 100644 index 000000000000..9e42af963791 --- /dev/null +++ b/dtrace/dtrace_ptofapi.c @@ -0,0 +1,417 @@ +/* + * FILE: dtrace_ptofapi.c + * DESCRIPTION: Dynamic Tracing: provider-to-framework API + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include +#include + +#include "dtrace.h" + +dtrace_provider_t *dtrace_provider; + +DEFINE_MUTEX(dtrace_lock); +DEFINE_MUTEX(dtrace_provider_lock); +DEFINE_MUTEX(dtrace_meta_lock); + +/* + * Register the calling provider with the DTrace core. This should generally + * be called by providers during module initialization. + */ +int dtrace_register(const char *name, const dtrace_pattr_t *pap, uint32_t priv, + cred_t *cr, const dtrace_pops_t *pops, void *arg, + dtrace_provider_id_t *idp) +{ + dtrace_provider_t *provider; + + if (name == NULL || pap == NULL || pops == NULL || idp == NULL) { + pr_warning("Failed to register provider %s: invalid args\n", + name ? name : ""); + return -EINVAL; + } + + if (name[0] == '\0' || dtrace_badname(name)) { + pr_warning("Failed to register provider %s: invalid name\n", + name); + return -EINVAL; + } + + if ((pops->dtps_provide == NULL && pops->dtps_provide_module == NULL) || + pops->dtps_enable == NULL || pops->dtps_disable == NULL || + pops->dtps_destroy == NULL || + ((pops->dtps_resume == NULL) != (pops->dtps_suspend == NULL))) { + pr_warning("Failed to register provider %s: invalid ops\n", + name); + return -EINVAL; + } + + if (dtrace_badattr(&pap->dtpa_provider) || + dtrace_badattr(&pap->dtpa_mod) || + dtrace_badattr(&pap->dtpa_func) || + dtrace_badattr(&pap->dtpa_name) || + dtrace_badattr(&pap->dtpa_args)) { + pr_warning("Failed to register provider %s: invalid " + "attributes\n", name); + return -EINVAL; + } + + if (priv & ~DTRACE_PRIV_ALL) { + pr_warning("Failed to register provider %s: invalid privilege " + "attributes\n", name); + return -EINVAL; + } + + if ((priv & DTRACE_PRIV_KERNEL) && + (priv & (DTRACE_PRIV_USER | DTRACE_PRIV_OWNER)) && + pops->dtps_usermode == NULL) { + pr_warning("Failed to register provider %s: need " + "dtps_usermode() op for given privilege " + "attributes\n", name); + return -EINVAL; + } + + provider = kzalloc(sizeof (dtrace_provider_t), GFP_KERNEL); + provider->dtpv_name = dtrace_strdup(name); + provider->dtpv_attr = *pap; + provider->dtpv_priv.dtpp_flags = priv; + + if (cr != NULL) { + provider->dtpv_priv.dtpp_uid = get_cred(cr)->uid; + put_cred(cr); + } + + provider->dtpv_pops = *pops; + + if (pops->dtps_provide == NULL) { + ASSERT(pops->dtps_provide_module != NULL); + provider->dtpv_pops.dtps_provide = + (void (*)(void *, const dtrace_probedesc_t *))dtrace_nullop; + } + + if (pops->dtps_provide_module == NULL) { + ASSERT(pops->dtps_provide != NULL); + provider->dtpv_pops.dtps_provide_module = + (void (*)(void *, struct module *))dtrace_nullop; + } + + if (pops->dtps_suspend == NULL) { + ASSERT(pops->dtps_resume == NULL); + provider->dtpv_pops.dtps_suspend = + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop; + provider->dtpv_pops.dtps_resume = + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop; + } + + provider->dtpv_arg = arg; + *idp = (dtrace_provider_id_t)provider; + + if (pops == &dtrace_provider_ops) { + ASSERT(mutex_is_locked(&dtrace_provider_lock)); + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(dtrace_anon.dta_enabling == NULL); + + /* + * The DTrace provider must be at the head of the provider + * chain. + */ + provider->dtpv_next = dtrace_provider; + dtrace_provider = provider; + + return 0; + } + + mutex_lock(&dtrace_provider_lock); + mutex_lock(&dtrace_lock); + + /* + * If there is at least one provider registered, we'll add this new one + * after the first provider. + */ + if (dtrace_provider != NULL) { + provider->dtpv_next = dtrace_provider->dtpv_next; + dtrace_provider->dtpv_next = provider; + } else + dtrace_provider = provider; + + if (dtrace_retained != NULL) { + dtrace_enabling_provide(provider); + + /* + * We must now call dtrace_enabling_matchall() which needs to + * acquire cpu_lock and dtrace_lock. We therefore need to drop + * our locks before calling it. + */ + mutex_unlock(&dtrace_lock); + mutex_unlock(&dtrace_provider_lock); + dtrace_enabling_matchall(); + + return 0; + } + + mutex_unlock(&dtrace_lock); + mutex_unlock(&dtrace_provider_lock); + + return 0; +} +EXPORT_SYMBOL(dtrace_register); + +struct unreg_state { + dtrace_provider_t *prov; + dtrace_probe_t *first; +}; + +/* + * Check whether the given probe is still enabled for the given provider. + */ +static int dtrace_unregister_check(int id, void *p, void *data) +{ + dtrace_probe_t *probe = (dtrace_probe_t *)p; + struct unreg_state *st = (struct unreg_state *)data; + + if (probe->dtpr_provider != st->prov) + return 0; + + if (probe->dtpr_ecb == NULL) + return 0; + + return -EBUSY; +} + +/* + * Remove the given probe from the hash tables and the probe IDR. The probes + * are chained for further processing. + */ +static int dtrace_unregister_probe(int id, void *p, void *data) +{ + dtrace_probe_t *probe = (dtrace_probe_t *)p; + struct unreg_state *st = (struct unreg_state *)data; + + if (probe->dtpr_provider != st->prov) + return 0; + + dtrace_hash_remove(dtrace_bymod, probe); + dtrace_hash_remove(dtrace_byfunc, probe); + dtrace_hash_remove(dtrace_byname, probe); + + if (st->first == NULL) { + st->first = probe; + probe->dtpr_nextmod = NULL; + } else { + probe->dtpr_nextmod = st->first; + st->first = probe; + } + + return 0; +} + +/* + * Unregister the specified provider from the DTrace core. This should be + * called by provider during module cleanup. + */ +int dtrace_unregister(dtrace_provider_id_t id) +{ + dtrace_provider_t *old = (dtrace_provider_t *)id; + dtrace_provider_t *prev = NULL; + int err, self = 0; + dtrace_probe_t *probe; + struct unreg_state st = { + old, + NULL + }; + +printk(KERN_INFO "unregister(%p)\n", old); + if (old->dtpv_pops.dtps_enable == + (int (*)(void *, dtrace_id_t, void *))dtrace_enable_nullop) { + /* + * When the provider is the DTrace core itself, we're called + * with locks already held. + */ + ASSERT(old == dtrace_provider); + ASSERT(mutex_is_locked(&dtrace_provider_lock)); + ASSERT(mutex_is_locked(&dtrace_lock)); + + self = 1; + + if (dtrace_provider->dtpv_next != NULL) { + /* + * We cannot and should not remove the DTrace provider + * if there is any other provider left. + */ + return -EBUSY; + } + } else { + mutex_lock(&dtrace_provider_lock); + /* FIXME: mutex_lock(&mod_lock); */ + mutex_lock(&dtrace_lock); + } + + /* + * If /dev/dtrace/dtrace is still held open by a process, or if there + * are anonymous probes that are still enabled, we refuse to deregister + * providers, unless the provider has been invalidated explicitly. + */ +printk(KERN_INFO "unregister(%p) %d opens\n", old, dtrace_opens); + if (!old->dtpv_defunct && + (dtrace_opens || (dtrace_anon.dta_state != NULL && + dtrace_anon.dta_state->dts_necbs > 0))) { + if (!self) { + mutex_unlock(&dtrace_lock); + /* FIXME: mutex_unlock(&mod_lock); */ + mutex_unlock(&dtrace_provider_lock); + } + + return -EBUSY; + } + + /* + * Check whether any of the probes associated with this provider are + * still enabled (having at least one ECB). If any are found, we + * cannot remove this provider. + */ + st.prov = old; + err = dtrace_probe_for_each(dtrace_unregister_check, &st); +printk(KERN_INFO "unregister(%p) check -> %d\n", old, err); + if (err < 0) { + if (!self) { + mutex_unlock(&dtrace_lock); + /* FIXME: mutex_unlock(&mod_lock); */ + mutex_unlock(&dtrace_provider_lock); + } + + return err; + } + + /* + * All the probes associated with this provider are disabled. We can + * safely remove these probes from the hashtables and the probe array. + * We chain all the probes together for further processing. + */ + dtrace_probe_for_each(dtrace_unregister_probe, &st); +printk(KERN_INFO "unregister(%p) unregister_probe done\n", old); + + /* + * The probes associated with the provider have been removed. Ensure + * synchronization on probe IDR processing. + */ + dtrace_sync(); + + /* + * Now get rid of the actual probes. + */ + for (probe = st.first; probe != NULL; probe = st.first) { + int probe_id = probe->dtpr_id; + + st.first = probe->dtpr_nextmod; + +printk(KERN_INFO "unregister(%p) Destroying probe %d\n", old, probe_id); + old->dtpv_pops.dtps_destroy(old->dtpv_arg, probe_id, + probe->dtpr_arg); + + kfree(probe->dtpr_mod); + kfree(probe->dtpr_func); + kfree(probe->dtpr_name); + kfree(probe); + + dtrace_probe_remove_id(probe_id); + } + + if ((prev = dtrace_provider) == old) { + /* + * We are removing the provider at the head of the chain. + */ + ASSERT(self); + ASSERT(old->dtpv_next == NULL); + + dtrace_provider = old->dtpv_next; + } else { + while (prev != NULL && prev->dtpv_next != old) + prev = prev->dtpv_next; + + if (prev == NULL) { + pr_err("Attempt to unregister non-existent DTrace " + "provider %p\n", (void *)id); + BUG(); + } + + prev->dtpv_next = old->dtpv_next; + } + + if (!self) { + mutex_unlock(&dtrace_lock); + /* FIXME: mutex_unlock(&mod_lock); */ + mutex_unlock(&dtrace_provider_lock); + } + + kfree(old->dtpv_name); + kfree(old); + + return 0; +} +EXPORT_SYMBOL(dtrace_unregister); + +/* + * Invalidate the specified provider. All subsequent probe lookups for the + * specified provider will fail, but the probes will not be removed. + */ +void dtrace_invalidate(dtrace_provider_id_t id) +{ + dtrace_provider_t *pvp = (dtrace_provider_t *)id; + + ASSERT(pvp->dtpv_pops.dtps_enable != + (int (*)(void *, dtrace_id_t, void *))dtrace_enable_nullop); + + mutex_lock(&dtrace_provider_lock); + mutex_lock(&dtrace_lock); + + pvp->dtpv_defunct = 1; + + mutex_unlock(&dtrace_lock); + mutex_unlock(&dtrace_provider_lock); +} +EXPORT_SYMBOL(dtrace_invalidate); + +/* + * Indicate whether or not DTrace has attached. + */ +int dtrace_attached(void) +{ + /* + * dtrace_provider will be non-NULL iff the DTrace driver has + * attached. (It's non-NULL because DTrace is always itself a + * provider.) + */ + return dtrace_provider != NULL; +} +EXPORT_SYMBOL(dtrace_attached); + +/* + * Remove all the unenabled probes for the given provider. This function is + * not unlike dtrace_unregister(), except that it doesn't remove the provider + * -- just as many of its associated probes as it can. + */ +int dtrace_condense(dtrace_provider_id_t id) +{ + dtrace_provider_t *prov = (dtrace_provider_t *)id; + int i; + dtrace_probe_t *probe; + + /* + * Make sure this isn't the DTrace provider itself. + */ + ASSERT(prov->dtpv_pops.dtps_enable != + (int (*)(void *, dtrace_id_t, void *))dtrace_enable_nullop); + + mutex_lock(&dtrace_provider_lock); + mutex_lock(&dtrace_lock); + + /* FIXME - INCOMPLETE */ + + mutex_unlock(&dtrace_lock); + mutex_unlock(&dtrace_provider_lock); + + return 0; +} +EXPORT_SYMBOL(dtrace_condense); diff --git a/dtrace/dtrace_spec.c b/dtrace/dtrace_spec.c new file mode 100644 index 000000000000..4b30d2fc1e75 --- /dev/null +++ b/dtrace/dtrace_spec.c @@ -0,0 +1,426 @@ +/* + * FILE: dtrace_spec.c + * DESCRIPTION: Dynamic Tracing: speculation functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include + +#include "dtrace.h" + +/* + * Given consumer state, this routine finds a speculation in the INACTIVE + * state and transitions it into the ACTIVE state. If there is no speculation + * in the INACTIVE state, 0 is returned. In this case, no error counter is + * incremented -- it is up to the caller to take appropriate action. + */ +int dtrace_speculation(dtrace_state_t *state) +{ + int i = 0; + dtrace_speculation_state_t curr; + uint32_t *stat = + &state->dts_speculations_unavail, + count; + + while (i < state->dts_nspeculations) { + dtrace_speculation_t *spec = &state->dts_speculations[i]; + + curr = spec->dtsp_state; + + if (curr != DTRACESPEC_INACTIVE) { + if (curr == DTRACESPEC_COMMITTINGMANY || + curr == DTRACESPEC_COMMITTING || + curr == DTRACESPEC_DISCARDING) + stat = &state->dts_speculations_busy; + + i++; + continue; + } + + if (cmpxchg((uint32_t *)&spec->dtsp_state, curr, + DTRACESPEC_ACTIVE) == curr) + return i + 1; + } + + /* + * We couldn't find a speculation. If we found as much as a single + * busy speculation buffer, we'll attribute this failure as "busy" + * instead of "unavail". + */ + do { + count = *stat; + } while (cmpxchg(stat, count, count + 1) != count); + + return 0; +} + +/* + * This routine commits an active speculation. If the specified speculation + * is not in a valid state to perform a commit(), this routine will silently do + * nothing. The state of the specified speculation is transitioned according + * to the state transition diagram outlined in + */ +void dtrace_speculation_commit(dtrace_state_t *state, processorid_t cpu, + dtrace_specid_t which) +{ + dtrace_speculation_t *spec; + dtrace_buffer_t *src, *dest; + uintptr_t daddr, saddr, dlimit; + dtrace_speculation_state_t curr, new = 0; + intptr_t offs; + + if (which == 0) + return; + +printk(KERN_INFO "spec-commit: CPU#%d, which %d\n", cpu, which); + if (which > state->dts_nspeculations) { + cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return; + } + + spec = &state->dts_speculations[which - 1]; + src = &spec->dtsp_buffer[cpu]; + dest = &state->dts_buffer[cpu]; + + do { +printk(KERN_INFO "spec-commit: a) CPU#%d, which %d, spec-state %d\n", cpu, which, spec->dtsp_state); + curr = spec->dtsp_state; + + if (curr == DTRACESPEC_COMMITTINGMANY) + break; + + switch (curr) { + case DTRACESPEC_INACTIVE: + case DTRACESPEC_DISCARDING: + return; + + case DTRACESPEC_COMMITTING: + /* + * This is only possible if we are (a) commit()'ing + * without having done a prior speculate() on this CPU + * and (b) racing with another commit() on a different + * CPU. There's nothing to do -- we just assert that + * our offset is 0. + */ + ASSERT(src->dtb_offset == 0); + return; + + case DTRACESPEC_ACTIVE: + new = DTRACESPEC_COMMITTING; + break; + + case DTRACESPEC_ACTIVEONE: + /* + * This speculation is active on one CPU. If our + * buffer offset is non-zero, we know that the one CPU + * must be us. Otherwise, we are committing on a + * different CPU from the speculate(), and we must + * rely on being asynchronously cleaned. + */ + if (src->dtb_offset != 0) { + new = DTRACESPEC_COMMITTING; + break; + } + /*FALLTHROUGH*/ + + case DTRACESPEC_ACTIVEMANY: + new = DTRACESPEC_COMMITTINGMANY; + break; + + default: + ASSERT(0); + } +printk(KERN_INFO "spec-commit: a) CPU#%d, which %d, spec-state %d -> %d\n", cpu, which, curr, new); + } while (cmpxchg((uint32_t *)&spec->dtsp_state, curr, new) != + curr); + + /* + * We have set the state to indicate that we are committing this + * speculation. Now reserve the necessary space in the destination + * buffer. + */ + offs = dtrace_buffer_reserve(dest, src->dtb_offset, sizeof (uint64_t), + state, NULL); + if (offs < 0) { + dtrace_buffer_drop(dest); + goto out; + } + + /* + * We have the space; copy the buffer across. (Note that this is a + * highly subobtimal bcopy(); in the unlikely event that this becomes + * a serious performance issue, a high-performance DTrace-specific + * bcopy() should obviously be invented.) + */ + daddr = (uintptr_t)dest->dtb_tomax + offs; + dlimit = daddr + src->dtb_offset; + saddr = (uintptr_t)src->dtb_tomax; + + /* + * First, the aligned portion. + */ + while (dlimit - daddr >= sizeof (uint64_t)) { + *((uint64_t *)daddr) = *((uint64_t *)saddr); + *((uint64_t *)daddr) = *((uint64_t *)saddr); + + daddr += sizeof (uint64_t); + saddr += sizeof (uint64_t); + } + + /* + * Now any left-over bit... + */ + while (dlimit - daddr) + *((uint8_t *)daddr++) = *((uint8_t *)saddr++); + + /* + * Finally, commit the reserved space in the destination buffer. + */ + dest->dtb_offset = offs + src->dtb_offset; + +out: + /* + * If we're lucky enough to be the only active CPU on this speculation + * buffer, we can just set the state back to DTRACESPEC_INACTIVE. + */ + if (curr == DTRACESPEC_ACTIVE || + (curr == DTRACESPEC_ACTIVEONE && new == DTRACESPEC_COMMITTING)) { + /* + * Will cause unused warning if DEBUG is not defined. + */ + uint32_t rval = + cmpxchg((uint32_t *)&spec->dtsp_state, + DTRACESPEC_COMMITTING, + DTRACESPEC_INACTIVE); + + ASSERT(rval == DTRACESPEC_COMMITTING); + rval = 0; /* Avoid warning about unused variable if !DEBUG */ + } + + src->dtb_offset = 0; + src->dtb_xamot_drops += src->dtb_drops; + src->dtb_drops = 0; +} + +/* + * This routine discards an active speculation. If the specified speculation + * is not in a valid state to perform a discard(), this routine will silently + * do nothing. The state of the specified speculation is transitioned + * according to the state transition diagram outlined in + */ +void dtrace_speculation_discard(dtrace_state_t *state, processorid_t cpu, + dtrace_specid_t which) +{ + dtrace_speculation_t *spec; + dtrace_speculation_state_t curr, new = 0; + dtrace_buffer_t *buf; + + if (which == 0) + return; + + if (which > state->dts_nspeculations) { + cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return; + } + + spec = &state->dts_speculations[which - 1]; + buf = &spec->dtsp_buffer[cpu]; + + do { + curr = spec->dtsp_state; + + switch (curr) { + case DTRACESPEC_INACTIVE: + case DTRACESPEC_COMMITTINGMANY: + case DTRACESPEC_COMMITTING: + case DTRACESPEC_DISCARDING: + return; + + case DTRACESPEC_ACTIVE: + case DTRACESPEC_ACTIVEMANY: + new = DTRACESPEC_DISCARDING; + break; + + case DTRACESPEC_ACTIVEONE: + if (buf->dtb_offset != 0) + new = DTRACESPEC_INACTIVE; + else + new = DTRACESPEC_DISCARDING; + + break; + + default: + ASSERT(0); + } + } while (cmpxchg((uint32_t *)&spec->dtsp_state, curr, new) != curr); + + buf->dtb_offset = 0; + buf->dtb_drops = 0; +} + +/* + * Note: not called from probe context. This function is called + * asynchronously from cross call context to clean any speculations that are + * in the COMMITTINGMANY or DISCARDING states. These speculations may not be + * transitioned back to the INACTIVE state until all CPUs have cleaned the + * speculation. + */ +void dtrace_speculation_clean_here(dtrace_state_t *state) +{ + dtrace_icookie_t cookie; + processorid_t cpu = smp_processor_id(); + dtrace_buffer_t *dest = &state->dts_buffer[cpu]; + dtrace_specid_t i; + + local_irq_save(cookie); + + if (dest->dtb_tomax == NULL) { + local_irq_restore(cookie); + return; + } + + for (i = 0; i < state->dts_nspeculations; i++) { + dtrace_speculation_t *spec = &state->dts_speculations[i]; + dtrace_buffer_t *src = &spec->dtsp_buffer[cpu]; + + if (src->dtb_tomax == NULL) + continue; + + if (spec->dtsp_state == DTRACESPEC_DISCARDING) { + src->dtb_offset = 0; + continue; + } + + if (spec->dtsp_state != DTRACESPEC_COMMITTINGMANY) + continue; + + if (src->dtb_offset == 0) + continue; + + dtrace_speculation_commit(state, cpu, i + 1); + } + + local_irq_restore(cookie); +} + +void dtrace_speculation_clean(dtrace_state_t *state) +{ + int work = 0, rv; + dtrace_specid_t i; + + for (i = 0; i < state->dts_nspeculations; i++) { + dtrace_speculation_t *spec = &state->dts_speculations[i]; + + ASSERT(!spec->dtsp_cleaning); + + if (spec->dtsp_state != DTRACESPEC_DISCARDING && + spec->dtsp_state != DTRACESPEC_COMMITTINGMANY) + continue; + + work++; + spec->dtsp_cleaning = 1; + } + + if (!work) + return; + + dtrace_xcall(DTRACE_CPUALL, + (dtrace_xcall_t)dtrace_speculation_clean_here, state); + + /* + * We now know that all CPUs have committed or discarded their + * speculation buffers, as appropriate. We can now set the state + * to inactive. + */ + for (i = 0; i < state->dts_nspeculations; i++) { + dtrace_speculation_t *spec = + &state->dts_speculations[i]; + dtrace_speculation_state_t curr, new; + + if (!spec->dtsp_cleaning) + continue; + + curr= spec->dtsp_state; + ASSERT(curr == DTRACESPEC_DISCARDING || + curr == DTRACESPEC_COMMITTINGMANY); + + new = DTRACESPEC_INACTIVE; + + rv = cmpxchg((uint32_t *)&spec->dtsp_state, curr, new); + ASSERT(rv == curr); + spec->dtsp_cleaning = 0; + } +} + +/* + * Called as part of a speculate() to get the speculative buffer associated + * with a given speculation. Returns NULL if the specified speculation is not + * in an ACTIVE state. If the speculation is in the ACTIVEONE state -- and + * the active CPU is not the specified CPU -- the speculation will be + * atomically transitioned into the ACTIVEMANY state. + */ +dtrace_buffer_t *dtrace_speculation_buffer(dtrace_state_t *state, + processorid_t cpuid, + dtrace_specid_t which) +{ + dtrace_speculation_t *spec; + dtrace_speculation_state_t curr, new = 0; + dtrace_buffer_t *buf; + + if (which == 0) + return NULL; + + if (which > state->dts_nspeculations) { + cpu_core[cpuid].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return NULL; + } + + spec = &state->dts_speculations[which - 1]; + buf = &spec->dtsp_buffer[cpuid]; + + do { + curr = spec->dtsp_state; + + switch (curr) { + case DTRACESPEC_INACTIVE: + case DTRACESPEC_COMMITTINGMANY: + case DTRACESPEC_DISCARDING: + return NULL; + + case DTRACESPEC_COMMITTING: + ASSERT(buf->dtb_offset == 0); + return NULL; + + case DTRACESPEC_ACTIVEONE: + /* + * This speculation is currently active on one CPU. + * Check the offset in the buffer; if it's non-zero, + * that CPU must be us (and we leave the state alone). + * If it's zero, assume that we're starting on a new + * CPU -- and change the state to indicate that the + * speculation is active on more than one CPU. + */ + if (buf->dtb_offset != 0) + return buf; + + new = DTRACESPEC_ACTIVEMANY; + break; + + case DTRACESPEC_ACTIVEMANY: + return buf; + + case DTRACESPEC_ACTIVE: + new = DTRACESPEC_ACTIVEONE; + break; + + default: + ASSERT(0); + } + } while (cmpxchg((uint32_t *)&spec->dtsp_state, curr, new) != curr); + + ASSERT(new == DTRACESPEC_ACTIVEONE || new == DTRACESPEC_ACTIVEMANY); + + return buf; +} diff --git a/dtrace/dtrace_state.c b/dtrace/dtrace_state.c new file mode 100644 index 000000000000..11d8819506e7 --- /dev/null +++ b/dtrace/dtrace_state.c @@ -0,0 +1,1057 @@ +/* + * FILE: dtrace_state.c + * DESCRIPTION: Dynamic Tracing: consumer state functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include +#include +#include +#include +#include + +#include "cyclic.h" +#include "dtrace.h" + +struct kmem_cache *dtrace_state_cache; +int dtrace_destructive_disallow = 0; +dtrace_optval_t dtrace_nspec_default = 1; +dtrace_optval_t dtrace_specsize_default = 32 * 1024; +dtrace_optval_t dtrace_dstate_defsize = 1 * 1024 * 1024; +size_t dtrace_strsize_default = 256; +dtrace_optval_t dtrace_stackframes_default = 20; +dtrace_optval_t dtrace_ustackframes_default = 20; +dtrace_optval_t dtrace_cleanrate_default = 9900990; +dtrace_optval_t dtrace_cleanrate_min = 20000; +dtrace_optval_t dtrace_cleanrate_max = (uint64_t)60 * NANOSEC; +dtrace_optval_t dtrace_aggrate_default = NANOSEC; +dtrace_optval_t dtrace_switchrate_default = NANOSEC; +dtrace_optval_t dtrace_statusrate_default = NANOSEC; +dtrace_optval_t dtrace_statusrate_max = (uint64_t)10 * NANOSEC; +dtrace_optval_t dtrace_jstackframes_default = 50; +dtrace_optval_t dtrace_jstackstrsize_default = 512; +ktime_t dtrace_deadman_interval = KTIME_INIT(1, 0); +ktime_t dtrace_deadman_timeout = KTIME_INIT(10, 0); +ktime_t dtrace_deadman_user = KTIME_INIT(30, 0); + +dtrace_id_t dtrace_probeid_begin; +dtrace_id_t dtrace_probeid_end; +dtrace_id_t dtrace_probeid_error; + +dtrace_dynvar_t dtrace_dynhash_sink; + +#define DTRACE_DYNHASH_FREE 0 +#define DTRACE_DYNHASH_SINK 1 +#define DTRACE_DYNHASH_VALID 2 + +#define DTRACE_DYNVAR_CHUNKSIZE 256 + +static void dtrace_dynvar_clean(dtrace_dstate_t *dstate) +{ + dtrace_dynvar_t *dirty; + dtrace_dstate_percpu_t *dcpu; + int i, work = 0; + + for (i = 0; i < NR_CPUS; i++) { + dcpu = &dstate->dtds_percpu[i]; + + ASSERT(dcpu->dtdsc_rinsing == NULL); + + /* + * If the dirty list is NULL, there is no dirty work to do. + */ + if (dcpu->dtdsc_dirty == NULL) + continue; + + /* + * If the clean list is non-NULL, then we're not going to do + * any work for this CPU -- it means that there has not been + * a dtrace_dynvar() allocation on this CPU (or from this CPU) + * since the last time we cleaned house. + */ + if (dcpu->dtdsc_clean != NULL) + continue; + + work = 1; + + /* + * Atomically move the dirty list aside. + */ + do { + dirty = dcpu->dtdsc_dirty; + + /* + * Before we zap the dirty list, set the rinsing list. + * (This allows for a potential assertion in + * dtrace_dynvar(): if a free dynamic variable appears + * on a hash chain, either the dirty list or the + * rinsing list for some CPU must be non-NULL.) + */ + dcpu->dtdsc_rinsing = dirty; + dtrace_membar_producer(); + } while (cmpxchg(&dcpu->dtdsc_dirty, dirty, NULL) != dirty); + } + + /* + * No work to do; return. + */ + if (!work) + return; + + dtrace_sync(); + + for (i = 0; i < NR_CPUS; i++) { + dcpu = &dstate->dtds_percpu[i]; + + if (dcpu->dtdsc_rinsing == NULL) + continue; + + /* + * We are now guaranteed that no hash chain contains a pointer + * into this dirty list; we can make it clean. + */ + ASSERT(dcpu->dtdsc_clean == NULL); + dcpu->dtdsc_clean = dcpu->dtdsc_rinsing; + dcpu->dtdsc_rinsing = NULL; + } + + /* + * Before we actually set the state to be DTRACE_DSTATE_CLEAN, make + * sure that all CPUs have seen all of the dtdsc_clean pointers. + * This prevents a race whereby a CPU incorrectly decides that + * the state should be something other than DTRACE_DSTATE_CLEAN + * after dtrace_dynvar_clean() has completed. + */ + dtrace_sync(); + + dstate->dtds_state = DTRACE_DSTATE_CLEAN; +} + +int dtrace_dstate_init(dtrace_dstate_t *dstate, size_t size) +{ + size_t hashsize, maxper, min, + chunksize = dstate->dtds_chunksize; + void *base; + uintptr_t limit; + dtrace_dynvar_t *dvar, *next, *start; + int i; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(dstate->dtds_base == NULL && dstate->dtds_percpu == NULL); + + memset(dstate, 0, sizeof (dtrace_dstate_t)); + + if ((dstate->dtds_chunksize = chunksize) == 0) + dstate->dtds_chunksize = DTRACE_DYNVAR_CHUNKSIZE; + + if (size < (min = dstate->dtds_chunksize + sizeof (dtrace_dynhash_t))) + size = min; + + if ((base = dtrace_vzalloc(size)) == NULL) + return -ENOMEM; + + dstate->dtds_size = size; + dstate->dtds_base = base; + dstate->dtds_percpu = kmem_cache_alloc(dtrace_state_cache, GFP_KERNEL); + memset(dstate->dtds_percpu, 0, + NR_CPUS * sizeof (dtrace_dstate_percpu_t)); + + hashsize = size / (dstate->dtds_chunksize + sizeof (dtrace_dynhash_t)); + + if (hashsize != 1 && (hashsize & 1)) + hashsize--; + + dstate->dtds_hashsize = hashsize; + dstate->dtds_hash = dstate->dtds_base; + + /* + * Set all of our hash buckets to point to the single sink, and (if + * it hasn't already been set), set the sink's hash value to be the + * sink sentinel value. The sink is needed for dynamic variable + * lookups to know that they have iterated over an entire, valid hash + * chain. + */ + for (i = 0; i < hashsize; i++) + dstate->dtds_hash[i].dtdh_chain = &dtrace_dynhash_sink; + + if (dtrace_dynhash_sink.dtdv_hashval != DTRACE_DYNHASH_SINK) + dtrace_dynhash_sink.dtdv_hashval = DTRACE_DYNHASH_SINK; + + /* + * Determine number of active CPUs. Divide free list evenly among + * active CPUs. + */ + start = (dtrace_dynvar_t *)((uintptr_t)base + + hashsize * sizeof (dtrace_dynhash_t)); + limit = (uintptr_t)base + size; + + maxper = (limit - (uintptr_t)start) / NR_CPUS; + maxper = (maxper / dstate->dtds_chunksize) * dstate->dtds_chunksize; + + for (i = 0; i < NR_CPUS; i++) { + dstate->dtds_percpu[i].dtdsc_free = dvar = start; + + /* + * If we don't even have enough chunks to make it once through + * NCPUs, we're just going to allocate everything to the first + * CPU. And if we're on the last CPU, we're going to allocate + * whatever is left over. In either case, we set the limit to + * be the limit of the dynamic variable space. + */ + if (maxper == 0 || i == NR_CPUS - 1) { + limit = (uintptr_t)base + size; + start = NULL; + } else { + limit = (uintptr_t)start + maxper; + start = (dtrace_dynvar_t *)limit; + } + + ASSERT(limit <= (uintptr_t)base + size); + + for (;;) { + next = (dtrace_dynvar_t *)((uintptr_t)dvar + + dstate->dtds_chunksize); + + if ((uintptr_t)next + dstate->dtds_chunksize >= limit) + break; + + dvar->dtdv_next = next; + dvar = next; + } + + if (maxper == 0) + break; + } + + return 0; +} + +void dtrace_dstate_fini(dtrace_dstate_t *dstate) +{ + ASSERT(mutex_is_locked(&cpu_lock)); + + if (dstate->dtds_base == NULL) + return; + + vfree(dstate->dtds_base); + kmem_cache_free(dtrace_state_cache, dstate->dtds_percpu); +} + +void dtrace_vstate_fini(dtrace_vstate_t *vstate) +{ + /* + * If only there was a logical XOR operator... + */ + ASSERT((vstate->dtvs_nglobals == 0) ^ (vstate->dtvs_globals != NULL)); + + if (vstate->dtvs_nglobals > 0) + kfree(vstate->dtvs_globals); + + if (vstate->dtvs_ntlocals > 0) + kfree(vstate->dtvs_tlocals); + + ASSERT((vstate->dtvs_nlocals == 0) ^ (vstate->dtvs_locals != NULL)); + + if (vstate->dtvs_nlocals > 0) + kfree(vstate->dtvs_locals); +} + +static void dtrace_state_clean(dtrace_state_t *state) +{ + if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) + return; + + dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars); + dtrace_speculation_clean(state); +} + +static void dtrace_state_deadman(dtrace_state_t *state) +{ + ktime_t now; + +#ifdef FIXME + /* + * This may not be needed for Linux - we'll see. + */ + dtrace_sync(); +#endif + + now = dtrace_gethrtime(); + + if (state != dtrace_anon.dta_state && + ktime_ge(ktime_sub(now, state->dts_laststatus), + dtrace_deadman_user)) + return; + + /* + * We must be sure that dts_alive never appears to be less than the + * value upon entry to dtrace_state_deadman(), and because we lack a + * dtrace_cas64(), we cannot store to it atomically. We thus instead + * store KTIME_MAX to it, followed by a memory barrier, followed by + * the new value. This assures that dts_alive never appears to be + * less than its true value, regardless of the order in which the + * stores to the underlying storage are issued. + */ + state->dts_alive = ktime_set(KTIME_SEC_MAX, 0); + dtrace_membar_producer(); + state->dts_alive = now; +} + +dtrace_state_t *dtrace_state_create(struct file *file) +{ + dtrace_state_t *state; + dtrace_optval_t *opt; + int bufsize = NR_CPUS * sizeof (dtrace_buffer_t), i; + const cred_t *cr = file->f_cred; + int err; + dtrace_aggid_t aggid; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(mutex_is_locked(&cpu_lock)); + + state = kzalloc(sizeof (dtrace_state_t), GFP_KERNEL); + state->dts_epid = DTRACE_EPIDNONE + 1; + /* state->dts_dev = NULL; -- FIXME: Do we even need this? */ + state->dts_buffer = kzalloc(NR_CPUS * sizeof (dtrace_buffer_t), + GFP_KERNEL); + state->dts_buffer = kzalloc(bufsize, GFP_KERNEL); + state->dts_aggbuffer = kzalloc(bufsize, GFP_KERNEL); + idr_init(&state->dts_agg_idr); + state->dts_cleaner = 0; + state->dts_deadman = 0; + state->dts_vstate.dtvs_state = state; + + /* + * Create a first entry in the aggregation IDR, so that ID 0 is used as + * that gets used as meaning 'none'. + */ +again: + mutex_unlock(&cpu_lock); + mutex_unlock(&dtrace_lock); + + idr_pre_get(&state->dts_agg_idr, __GFP_NOFAIL); + + mutex_lock(&dtrace_lock); + mutex_lock(&cpu_lock); + + err = idr_get_new(&state->dts_agg_idr, NULL, &aggid); + if (err == -EAGAIN) + goto again; + + ASSERT(aggid == 0); + + for (i = 0; i < DTRACEOPT_MAX; i++) + state->dts_options[i] = DTRACEOPT_UNSET; + + /* + * Set the default options. + */ + opt = state->dts_options; + opt[DTRACEOPT_BUFPOLICY] = DTRACEOPT_BUFPOLICY_SWITCH; + opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_AUTO; + opt[DTRACEOPT_NSPEC] = dtrace_nspec_default; + opt[DTRACEOPT_SPECSIZE] = dtrace_specsize_default; + opt[DTRACEOPT_CPU] = (dtrace_optval_t)DTRACE_CPUALL; + opt[DTRACEOPT_STRSIZE] = dtrace_strsize_default; + opt[DTRACEOPT_STACKFRAMES] = dtrace_stackframes_default; + opt[DTRACEOPT_USTACKFRAMES] = dtrace_ustackframes_default; + opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_default; + opt[DTRACEOPT_AGGRATE] = dtrace_aggrate_default; + opt[DTRACEOPT_SWITCHRATE] = dtrace_switchrate_default; + opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_default; + opt[DTRACEOPT_JSTACKFRAMES] = dtrace_jstackframes_default; + opt[DTRACEOPT_JSTACKSTRSIZE] = dtrace_jstackstrsize_default; + + state->dts_activity = DTRACE_ACTIVITY_INACTIVE; + +#ifdef FIXME + /* + * Set probe visibility and destructiveness based on user credential + * information. For actual anonymous tracing or if all privileges are + * set, checks are bypassed. + */ + if (cr == NULL || + PRIV_POLICY_ONLY(cr, PRIV_ALL, FALSE)) { + state->dts_cred.dcr_visible = DTRACE_CRV_ALL; + state->dts_cred.dcr_action = DTRACE_CRA_ALL; + } else { + state->dts_cred.dcr_cred = get_cred(cr); + + /* + * CRA_PROC means "we have *some* privilege for dtrace" and + * it unlocks the use of variables like pid, etc. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, FALSE) || + PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, FALSE)) + state->dts_cred.dcr_action |= DTRACE_CRA_PROC; + + /* + * The DTRACE_USER privilege allows the use of syscall and + * profile providers. If the user also has PROC_OWNER, we + * extend the scope to include additional visibility and + * destructive power. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, FALSE)) { + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, FALSE)) + state->dts_cred.dcr_visible |= + DTRACE_CRV_ALLPROC; + + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; + } + + /* + * Holding the DTRACE_KERNEL privilege also implies that + * the user has the DTRACE_USER privilege from a visibility + * perspective. But without further privileges, some + * destructive actions are not available. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, FALSE)) { + /* + * Make all probes in all zones visible. However, + * this doesn't mean that all actions become available + * to all zones. + */ + state->dts_cred.dcr_visible |= DTRACE_CRV_KERNEL | + DTRACE_CRV_ALLPROC; + state->dts_cred.dcr_action |= DTRACE_CRA_KERNEL | + DTRACE_CRA_PROC; + + /* + * Holding PROC_OWNER means that destructive actions + * are allowed. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, FALSE)) + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; + } + + /* + * Holding the DTRACE_PROC privilege gives control over the + * fasttrap and pid providers. We need to grant wider + * destructive privileges in the event that the user has + * PROC_OWNER . + */ + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, FALSE)) { + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, FALSE)) + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; + } + } +#else + state->dts_cred.dcr_visible = DTRACE_CRV_ALLPROC; + state->dts_cred.dcr_action = DTRACE_CRA_ALL; +#endif + + return state; +} + +static int dtrace_state_buffer(dtrace_state_t *state, dtrace_buffer_t *buf, + int which) +{ + dtrace_optval_t *opt = state->dts_options, size; + processorid_t cpu = DTRACE_CPUALL; + int flags = 0, rval; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(mutex_is_locked(&cpu_lock)); + ASSERT(which < DTRACEOPT_MAX); + ASSERT(state->dts_activity == DTRACE_ACTIVITY_INACTIVE || + (state == dtrace_anon.dta_state && + state->dts_activity == DTRACE_ACTIVITY_ACTIVE)); +printk(KERN_INFO "state_buffer(%p, %p, %d) opt[%d] = %lld\n", state, buf, which, which, opt[which]); + + if (opt[which] == DTRACEOPT_UNSET || opt[which] == 0) + return 0; + + if (opt[DTRACEOPT_CPU] != DTRACEOPT_UNSET) + cpu = opt[DTRACEOPT_CPU]; + + if (which == DTRACEOPT_SPECSIZE) + flags |= DTRACEBUF_NOSWITCH; + + if (which == DTRACEOPT_BUFSIZE) { +printk(KERN_INFO "state_buffer(%p, %p, %d) opt[CPU] = %lld, opt[BUFPOLICY] = %lld, sta_state = %p\n", state, buf, which, opt[DTRACEOPT_CPU], opt[DTRACEOPT_BUFPOLICY], dtrace_anon.dta_state); + if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_RING) + flags |= DTRACEBUF_RING; + + if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_FILL) + flags |= DTRACEBUF_FILL; + + if (state != dtrace_anon.dta_state || + state->dts_activity != DTRACE_ACTIVITY_ACTIVE) + flags |= DTRACEBUF_INACTIVE; +printk(KERN_INFO "state_buffer(%p, %p, %d) flags = %08x\n", state, buf, which, flags); + } + + for (size = opt[which]; size >= sizeof (uint64_t); size >>= 1) { + /* + * The size must be 8-byte aligned. If the size is not 8-byte + * aligned, drop it down by the difference. + */ + if (size & (sizeof (uint64_t) - 1)) + size -= size & (sizeof (uint64_t) - 1); + + if (size < state->dts_reserve) { + /* + * Buffers always must be large enough to accommodate + * their prereserved space. We return -E2BIG instead + * of ENOMEM in this case to allow for user-level + * software to differentiate the cases. + */ + return -E2BIG; + } + + rval = dtrace_buffer_alloc(buf, size, flags, cpu); +printk(KERN_INFO "state_buffer: Alloc %d buffer: tomax %p xamot %p\n", which, buf->dtb_tomax, buf->dtb_xamot); + + if (rval != -ENOMEM) { + opt[which] = size; + return rval; + } + + if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL) + return rval; + } + + return -ENOMEM; +} + +static int dtrace_state_buffers(dtrace_state_t *state) +{ + dtrace_speculation_t *spec = state->dts_speculations; + int rval, i; + + if ((rval = dtrace_state_buffer(state, state->dts_buffer, + DTRACEOPT_BUFSIZE)) != 0) + return rval; + + if ((rval = dtrace_state_buffer(state, state->dts_aggbuffer, + DTRACEOPT_AGGSIZE)) != 0) + return rval; + + for (i = 0; i < state->dts_nspeculations; i++) { + if ((rval = dtrace_state_buffer(state, spec[i].dtsp_buffer, + DTRACEOPT_SPECSIZE)) != 0) + return rval; + } + + return 0; +} + +static void dtrace_state_prereserve(dtrace_state_t *state) +{ + dtrace_ecb_t *ecb; + dtrace_probe_t *probe; + + state->dts_reserve = 0; + + if (state->dts_options[DTRACEOPT_BUFPOLICY] != DTRACEOPT_BUFPOLICY_FILL) + return; + + /* + * If our buffer policy is a "fill" buffer policy, we need to set the + * prereserved space to be the space required by the END probes. + */ + probe = dtrace_probe_lookup_id(dtrace_probeid_end); + ASSERT(probe != NULL); + + for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) { + if (ecb->dte_state != state) + continue; + + state->dts_reserve += ecb->dte_needed + ecb->dte_alignment; + } +} + +int dtrace_state_go(dtrace_state_t *state, processorid_t *cpu) +{ + dtrace_optval_t *opt = state->dts_options, sz, nspec; + dtrace_speculation_t *spec; + dtrace_buffer_t *buf; + cyc_handler_t hdlr; + cyc_time_t when; + int rval = 0, i, + bufsize = NR_CPUS * sizeof (dtrace_buffer_t); + dtrace_icookie_t cookie; + + mutex_lock(&cpu_lock); + mutex_lock(&dtrace_lock); + +printk(KERN_INFO "state_go(1): dts_activity = %d (vs warmup %d or draining %d)\n", state->dts_activity, DTRACE_ACTIVITY_WARMUP, DTRACE_ACTIVITY_DRAINING); + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) { + rval = -EBUSY; + goto out; + } + + /* + * Before we can perform any checks, we must prime all of the + * retained enablings that correspond to this state. + */ + dtrace_enabling_prime(state); + + if (state->dts_destructive && !state->dts_cred.dcr_destructive) { + rval = -EACCES; + goto out; + } + + dtrace_state_prereserve(state); + + /* + * Now we want to do is try to allocate our speculations. + * We do not automatically resize the number of speculations; if + * this fails, we will fail the operation. + */ + nspec = opt[DTRACEOPT_NSPEC]; + ASSERT(nspec != DTRACEOPT_UNSET); + + if (nspec > INT_MAX) { + rval = -ENOMEM; + goto out; + } + + spec = kzalloc(nspec * sizeof(dtrace_speculation_t), GFP_KERNEL); +printk(KERN_INFO "state_go: nspec = %d, spec = %p (%lld bytes)\n", (int)nspec, spec, nspec * sizeof(dtrace_speculation_t)); + if (spec == NULL) { + rval = -ENOMEM; + goto out; + } + + state->dts_speculations = spec; + state->dts_nspeculations = (int)nspec; + + for (i = 0; i < nspec; i++) { + if ((buf = kzalloc(bufsize, GFP_KERNEL)) == NULL) { + rval = -ENOMEM; + goto err; + } + +printk(KERN_INFO "state_go: spec[%d].buf = %p (%d bytes)\n", i, buf, bufsize); + spec[i].dtsp_buffer = buf; + } + + if (opt[DTRACEOPT_GRABANON] != DTRACEOPT_UNSET) { + if (dtrace_anon.dta_state == NULL) { + rval = -ENOENT; + goto out; + } + + if (state->dts_necbs != 0) { + rval = -EALREADY; + goto out; + } + + state->dts_anon = dtrace_anon_grab(); + ASSERT(state->dts_anon != NULL); + state = state->dts_anon; + + /* + * We want "grabanon" to be set in the grabbed state, so we'll + * copy that option value from the grabbing state into the + * grabbed state. + */ + state->dts_options[DTRACEOPT_GRABANON] = + opt[DTRACEOPT_GRABANON]; + + *cpu = dtrace_anon.dta_beganon; + + /* + * If the anonymous state is active (as it almost certainly + * is if the anonymous enabling ultimately matched anything), + * we don't allow any further option processing -- but we + * don't return failure. + */ + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) + goto out; + } + + if (opt[DTRACEOPT_AGGSIZE] != DTRACEOPT_UNSET && + opt[DTRACEOPT_AGGSIZE] != 0) { + if (idr_empty(&state->dts_agg_idr)) { + /* + * We're not going to create an aggregation buffer + * because we don't have any ECBs that contain + * aggregations -- set this option to 0. + */ + opt[DTRACEOPT_AGGSIZE] = 0; + } else { + /* + * If we have an aggregation buffer, we must also have + * a buffer to use as scratch. + */ + if (opt[DTRACEOPT_BUFSIZE] == DTRACEOPT_UNSET || + opt[DTRACEOPT_BUFSIZE] < state->dts_needed) + opt[DTRACEOPT_BUFSIZE] = state->dts_needed; + } + } + + if (opt[DTRACEOPT_SPECSIZE] != DTRACEOPT_UNSET && + opt[DTRACEOPT_SPECSIZE] != 0) { + /* + * We are not going to create speculation buffers if we do not + * have any ECBs that actually speculate. + */ + if (!state->dts_speculates) + opt[DTRACEOPT_SPECSIZE] = 0; + } + + /* + * The bare minimum size for any buffer that we're actually going to + * do anything to is sizeof (uint64_t). + */ + sz = sizeof (uint64_t); + + if ((state->dts_needed != 0 && opt[DTRACEOPT_BUFSIZE] < sz) || + (state->dts_speculates && opt[DTRACEOPT_SPECSIZE] < sz) || + (!idr_empty(&state->dts_agg_idr) && opt[DTRACEOPT_AGGSIZE] < sz)) { + /* + * A buffer size has been explicitly set to 0 (or to a size + * that will be adjusted to 0) and we need the space -- we + * need to return failure. We return -ENOSPC to differentiate + * it from failing to allocate a buffer due to failure to meet + * the reserve (for which we return -E2BIG). + */ + rval = -ENOSPC; + goto out; + } + + if ((rval = dtrace_state_buffers(state)) != 0) + goto err; + + if ((sz = opt[DTRACEOPT_DYNVARSIZE]) == DTRACEOPT_UNSET) + sz = dtrace_dstate_defsize; + + do { + rval = dtrace_dstate_init(&state->dts_vstate.dtvs_dynvars, sz); + + if (rval == 0) + break; + + if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL) + goto err; + } while (sz >>= 1); + + opt[DTRACEOPT_DYNVARSIZE] = sz; + + if (rval != 0) + goto err; + + if (opt[DTRACEOPT_STATUSRATE] > dtrace_statusrate_max) + opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_max; + + if (opt[DTRACEOPT_CLEANRATE] == 0) + opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max; + + if (opt[DTRACEOPT_CLEANRATE] < dtrace_cleanrate_min) + opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_min; + + if (opt[DTRACEOPT_CLEANRATE] > dtrace_cleanrate_max) + opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max; + + hdlr.cyh_func = (cyc_func_t)dtrace_state_clean; + hdlr.cyh_arg = state; + hdlr.cyh_level = CY_LOW_LEVEL; + + when.cyt_when = ktime_set(0, 0); + when.cyt_interval = ns_to_ktime(opt[DTRACEOPT_CLEANRATE]); + + state->dts_cleaner = cyclic_add(&hdlr, &when); + + hdlr.cyh_func = (cyc_func_t)dtrace_state_deadman; + hdlr.cyh_arg = state; + hdlr.cyh_level = CY_LOW_LEVEL; + + when.cyt_when = ktime_set(0, 0); + when.cyt_interval = dtrace_deadman_interval; + + state->dts_alive = state->dts_laststatus = dtrace_gethrtime(); + state->dts_deadman = cyclic_add(&hdlr, &when); + + state->dts_activity = DTRACE_ACTIVITY_WARMUP; +printk(KERN_INFO "state_go(2): dts_activity = %d (vs warmup %d or draining %d)\n", state->dts_activity, DTRACE_ACTIVITY_WARMUP, DTRACE_ACTIVITY_DRAINING); + + /* + * Now it's time to actually fire the BEGIN probe. We need to disable + * interrupts here both to record the CPU on which we fired the BEGIN + * probe (the data from this CPU will be processed first at user + * level) and to manually activate the buffer for this CPU. + */ + local_irq_save(cookie); + *cpu = smp_processor_id(); +// ASSERT(state->dts_buffer[*cpu].dtb_flags & DTRACEBUF_INACTIVE); + state->dts_buffer[*cpu].dtb_flags &= ~DTRACEBUF_INACTIVE; + + dtrace_probe(dtrace_probeid_begin, (uint64_t)(uintptr_t)state, 0, 0, 0, + 0); + local_irq_restore(cookie); + + /* + * We may have had an exit action from a BEGIN probe; only change our + * state to ACTIVE if we're still in WARMUP. + */ +printk(KERN_INFO "state_go(3): dts_activity = %d (vs warmup %d or draining %d)\n", state->dts_activity, DTRACE_ACTIVITY_WARMUP, DTRACE_ACTIVITY_DRAINING); + ASSERT(state->dts_activity == DTRACE_ACTIVITY_WARMUP || + state->dts_activity == DTRACE_ACTIVITY_DRAINING); + + if (state->dts_activity == DTRACE_ACTIVITY_WARMUP) + state->dts_activity = DTRACE_ACTIVITY_ACTIVE; + + /* + * Regardless of whether or not now we're in ACTIVE or DRAINING, we + * want each CPU to transition its principal buffer out of the + * INACTIVE state. Doing this assures that no CPU will suddenly begin + * processing an ECB halfway down a probe's ECB chain; all CPUs will + * atomically transition from processing none of a state's ECBs to + * processing all of them. + */ + dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)dtrace_buffer_activate, + state); + goto out; + +err: + dtrace_buffer_free(state->dts_buffer); + dtrace_buffer_free(state->dts_aggbuffer); + + if ((nspec = state->dts_nspeculations) == 0) { + ASSERT(state->dts_speculations == NULL); + goto out; + } + + spec = state->dts_speculations; + ASSERT(spec != NULL); + + for (i = 0; i < state->dts_nspeculations; i++) { + if ((buf = spec[i].dtsp_buffer) == NULL) + break; + + dtrace_buffer_free(buf); + kfree(buf); + } + + kfree(spec); + state->dts_nspeculations = 0; + state->dts_speculations = NULL; + +out: + mutex_unlock(&dtrace_lock); + mutex_unlock(&cpu_lock); + + return rval; +} + +int dtrace_state_stop(dtrace_state_t *state, processorid_t *cpu) +{ + dtrace_icookie_t cookie; + + ASSERT(mutex_is_locked(&dtrace_lock)); + + if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE && + state->dts_activity != DTRACE_ACTIVITY_DRAINING) + return -EINVAL; + + /* + * We'll set the activity to DTRACE_ACTIVITY_DRAINING, and issue a sync + * to be sure that every CPU has seen it. See below for the details + * on why this is done. + */ + state->dts_activity = DTRACE_ACTIVITY_DRAINING; + dtrace_sync(); + + /* + * By this point, it is impossible for any CPU to be still processing + * with DTRACE_ACTIVITY_ACTIVE. We can thus set our activity to + * DTRACE_ACTIVITY_COOLDOWN and know that we're not racing with any + * other CPU in dtrace_buffer_reserve(). This allows dtrace_probe() + * and callees to know that the activity is DTRACE_ACTIVITY_COOLDOWN + * iff we're in the END probe. + */ + state->dts_activity = DTRACE_ACTIVITY_COOLDOWN; + dtrace_sync(); + ASSERT(state->dts_activity == DTRACE_ACTIVITY_COOLDOWN); + + /* + * Finally, we can release the reserve and call the END probe. We + * disable interrupts across calling the END probe to allow us to + * return the CPU on which we actually called the END probe. This + * allows user-land to be sure that this CPU's principal buffer is + * processed last. + */ + state->dts_reserve = 0; + + local_irq_save(cookie); + *cpu = smp_processor_id(); + dtrace_probe(dtrace_probeid_end, (uint64_t)(uintptr_t)state, 0, 0, 0, + 0); + local_irq_restore(cookie); + + state->dts_activity = DTRACE_ACTIVITY_STOPPED; + dtrace_sync(); + + return 0; +} + +int dtrace_state_option(dtrace_state_t *state, dtrace_optid_t option, + dtrace_optval_t val) +{ + ASSERT(mutex_is_locked(&dtrace_lock)); + + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) + return -EBUSY; + + if (option >= DTRACEOPT_MAX) + return -EINVAL; + + if (option != DTRACEOPT_CPU && val < 0) + return -EINVAL; + + switch (option) { + case DTRACEOPT_DESTRUCTIVE: + if (dtrace_destructive_disallow) + return -EACCES; + + state->dts_cred.dcr_destructive = 1; + break; + + case DTRACEOPT_BUFSIZE: + case DTRACEOPT_DYNVARSIZE: + case DTRACEOPT_AGGSIZE: + case DTRACEOPT_SPECSIZE: + case DTRACEOPT_STRSIZE: + if (val < 0) + return -EINVAL; + + /* + * If this is an otherwise negative value, set it to the + * highest multiple of 128m less than LONG_MAX. Technically, + * we're adjusting the size without regard to the buffer + * resizing policy, but in fact, this has no effect -- if we + * set the buffer size to ~LONG_MAX and the buffer policy is + * ultimately set to be "manual", the buffer allocation is + * guaranteed to fail, if only because the allocation requires + * two buffers. (We set the the size to the highest multiple + * of 128m because it ensures that the size will remain a + * multiple of a megabyte when repeatedly halved -- all the + * way down to 15m.) + */ + if (val >= LONG_MAX) + val = LONG_MAX - (1 << 27) + 1; + } + + state->dts_options[option] = val; + + return 0; +} + +void dtrace_state_destroy(dtrace_state_t *state) +{ + dtrace_ecb_t *ecb; + dtrace_vstate_t *vstate = &state->dts_vstate; + int i; + dtrace_speculation_t *spec = state->dts_speculations; + int nspec = state->dts_nspeculations; + uint32_t match; + + ASSERT(mutex_is_locked(&dtrace_lock)); + ASSERT(mutex_is_locked(&cpu_lock)); + + /* + * First, retract any retained enablings for this state. + */ + dtrace_enabling_retract(state); + ASSERT(state->dts_nretained == 0); + + if (state->dts_activity == DTRACE_ACTIVITY_ACTIVE || + state->dts_activity == DTRACE_ACTIVITY_DRAINING) { + /* + * We have managed to come into dtrace_state_destroy() on a + * hot enabling -- almost certainly because of a disorderly + * shutdown of a consumer. (That is, a consumer that is + * exiting without having called dtrace_stop().) In this case, + * we're going to set our activity to be KILLED, and then + * issue a sync to be sure that everyone is out of probe + * context before we start blowing away ECBs. + */ +printk(KERN_INFO "state_destroy: Setting dts_activity from %d to %d (KILLED)\n", state->dts_activity, DTRACE_ACTIVITY_KILLED); + state->dts_activity = DTRACE_ACTIVITY_KILLED; + dtrace_sync(); + } + + /* + * Release the credential hold we took in dtrace_state_create(). + */ + if (state->dts_cred.dcr_cred != NULL) + put_cred(state->dts_cred.dcr_cred); + + /* + * Now we can safely disable and destroy any enabled probes. Because + * any DTRACE_PRIV_KERNEL probes may actually be slowing our progress + * (especially if they're all enabled), we take two passes through the + * ECBs: in the first, we disable just DTRACE_PRIV_KERNEL probes, and + * in the second we disable whatever is left over. + */ + for (match = DTRACE_PRIV_KERNEL; ; match = 0) { + for (i = 0; i < state->dts_necbs; i++) { + if ((ecb = state->dts_ecbs[i]) == NULL) + continue; + + if (match && ecb->dte_probe != NULL) { + dtrace_probe_t *probe = + ecb->dte_probe; + dtrace_provider_t *prov = + probe->dtpr_provider; + + if (!(prov->dtpv_priv.dtpp_flags & match)) + continue; + } + + dtrace_ecb_disable(ecb); + dtrace_ecb_destroy(ecb); + } + + if (!match) + break; + } + + /* + * Before we free the buffers, perform one more sync to assure that + * every CPU is out of probe context. + */ + dtrace_sync(); + + dtrace_buffer_free(state->dts_buffer); + dtrace_buffer_free(state->dts_aggbuffer); + + for (i = 0; i < nspec; i++) + dtrace_buffer_free(spec[i].dtsp_buffer); + + if (state->dts_cleaner != CYCLIC_NONE) + cyclic_remove(state->dts_cleaner); + + if (state->dts_deadman != CYCLIC_NONE) + cyclic_remove(state->dts_deadman); + + dtrace_dstate_fini(&vstate->dtvs_dynvars); + dtrace_vstate_fini(vstate); + kfree(state->dts_ecbs); + + /* + * If there were aggregations allocated, they should have been cleaned + * up by now, so we can get rid of the idr. + */ + if (!idr_empty(&state->dts_agg_idr)) { + idr_remove_all(&state->dts_agg_idr); + idr_destroy(&state->dts_agg_idr); + } + + kfree(state->dts_buffer); + kfree(state->dts_aggbuffer); + + for (i = 0; i < nspec; i++) + kfree(spec[i].dtsp_buffer); + + kfree(spec); + + dtrace_format_destroy(state); +} diff --git a/dtrace/dtrace_util.c b/dtrace/dtrace_util.c new file mode 100644 index 000000000000..6b8720717904 --- /dev/null +++ b/dtrace/dtrace_util.c @@ -0,0 +1,215 @@ +/* + * FILE: dtrace_util.c + * DESCRIPTION: DTrace utility functions + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace.h" + +dtrace_vtime_state_t dtrace_vtime_active = 0; + +int dtrace_badattr(const dtrace_attribute_t *a) +{ + return a->dtat_name > DTRACE_STABILITY_MAX || + a->dtat_data > DTRACE_STABILITY_MAX || + a->dtat_class > DTRACE_CLASS_MAX; +} + +/* + * Allocate a chunk of virtual memory in kernel space, and zero it out. This + * allocation might fail (which will report a backtrace in the kernel log, yet + * it is harmless). + */ +void *dtrace_vzalloc_try(unsigned long size) +{ + return __vmalloc(size, + __GFP_IO | __GFP_FS | __GFP_NORETRY | __GFP_ZERO, + PAGE_KERNEL); +} + +/* + * Allocate a chunk of virtual memory in kernel space, and zero it out. + */ +void *dtrace_vzalloc(unsigned long size) +{ + void *ptr = vmalloc(size); + + if (ptr == NULL) + return NULL; + + return memset(ptr, 0, size); +} + +/* + * Return a duplicate copy of a string. If the specified string is NULL, this + * function returs a zero-length string. + */ +char *dtrace_strdup(const char *str) +{ + return kstrdup(str ? str : "", GFP_KERNEL); +} + +/* + * Compare two strings using safe loads. + */ +int dtrace_strncmp(char *s1, char *s2, size_t limit) +{ + uint8_t c1, c2; + volatile uint16_t *flags; + + if (s1 == s2 || limit == 0) + return 0; + + flags = (volatile uint16_t *) + &cpu_core[smp_processor_id()].cpuc_dtrace_flags; + + do { + if (s1 == NULL) + c1 = '\0'; + else + c1 = dtrace_load8((uintptr_t)s1++); + + if (s2 == NULL) + c2 = '\0'; + else + c2 = dtrace_load8((uintptr_t)s2++); + + if (c1 != c2) + return (c1 - c2); + } while (--limit && c1 != '\0' && !(*flags & CPU_DTRACE_FAULT)); + + return 0; +} + +/* + * Compute strlen(s) for a string using safe memory accesses. The additional + * len parameter is used to specify a maximum length to ensure completion. + */ +size_t dtrace_strlen(const char *s, size_t lim) +{ + uint_t len; + + for (len = 0; len != lim; len++) { + if (dtrace_load8((uintptr_t)s++) == '\0') + break; + } + + return len; +} + +#define DTRACE_ISALPHA(c) (((c) >= 'a' && (c) <= 'z') || \ + ((c) >= 'A' && (c) <= 'Z')) +int dtrace_badname(const char *s) +{ + char c; + + if (s == NULL || (c = *s++) == '\0') + return 0; + + if (!DTRACE_ISALPHA(c) && c != '-' && c!= '_' && c != '.') + return 1; + + while ((c = *s++) != '\0') { + if (!DTRACE_ISALPHA(c) && (c < '0' || c > '9') && + c != '-' && c!= '_' && c != '.' && c != '`') + return 1; + } + + return 0; +} + +void dtrace_cred2priv(const cred_t *cr, uint32_t *privp, uid_t *uidp) +{ +#ifdef FIXME +/* + * This should probably be rewritten based on capabilities in the cred_t struct. + */ + uint32_t priv; + + if (cr == NULL) + priv = DTRACE_PRIV_ALL; + else { + const cred_t *lcr = get_cred(cr); + + if (PRIV_POLICY_ONLY(lcr, PRIV_ALL, FALSE)) + priv = DTRACE_PRIV_ALL; + else { + *uidp = lcr->uid; + priv = 0; + + if (PRIV_POLICY_ONLY(lcr, PRIV_DTRACE_KERNEL, FALSE)) + priv |= DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER; + else if (PRIV_POLICY_ONLY(lcr, PRIV_DTRACE_USER, + FALSE)) + priv |= DTRACE_PRIV_USER; + + if (PRIV_POLICY_ONLY(lcr, PRIV_DTRACE_PROC, FALSE)) + priv |= DTRACE_PRIV_PROC; + if (PRIV_POLICY_ONLY(lcr, PRIV_PROC_OWNER, FALSE)) + priv |= DTRACE_PRIV_OWNER; + } + + put_cred(cr); + } + + *privp = priv; +#else + *privp = DTRACE_PRIV_ALL; + + if (cr != NULL) { + const cred_t *lcr = get_cred(cr); + + *uidp = lcr->uid; + put_cred(cr); + } +#endif +} + +void dtrace_vtime_enable(void) +{ + dtrace_vtime_state_t state, nstate = 0; + + do { + state = dtrace_vtime_active; + + switch (state) { + case DTRACE_VTIME_INACTIVE: + nstate = DTRACE_VTIME_ACTIVE; + break; + case DTRACE_VTIME_INACTIVE_TNF: + nstate = DTRACE_VTIME_ACTIVE_TNF; + break; + case DTRACE_VTIME_ACTIVE: + case DTRACE_VTIME_ACTIVE_TNF: + panic("DTrace virtual time already enabled"); + /*NOTREACHED*/ + } + } while (cmpxchg((uint32_t *)&dtrace_vtime_active, state, nstate) != + state); +} + +void dtrace_vtime_disable(void) +{ + dtrace_vtime_state_t state, nstate = 0; + + do { + state = dtrace_vtime_active; + + switch (state) { + case DTRACE_VTIME_ACTIVE: + nstate = DTRACE_VTIME_INACTIVE; + break; + case DTRACE_VTIME_ACTIVE_TNF: + nstate = DTRACE_VTIME_INACTIVE_TNF; + break; + case DTRACE_VTIME_INACTIVE: + case DTRACE_VTIME_INACTIVE_TNF: + panic("DTrace virtual time already disabled"); + /*NOTREACHED*/ + } + } while (cmpxchg((uint32_t *)&dtrace_vtime_active, state, nstate) != + state); +} diff --git a/dtrace/fasttrap.h b/dtrace/fasttrap.h new file mode 100644 index 000000000000..665a26c506e0 --- /dev/null +++ b/dtrace/fasttrap.h @@ -0,0 +1,7 @@ +#ifndef _FASTTRAP_H_ +#define _FASTTRAP_H_ + +extern int fasttrap_dev_init(void); +extern void fasttrap_dev_exit(void); + +#endif /* _FASTTRAP_H_ */ diff --git a/dtrace/fasttrap_dev.c b/dtrace/fasttrap_dev.c new file mode 100644 index 000000000000..2ad91b926865 --- /dev/null +++ b/dtrace/fasttrap_dev.c @@ -0,0 +1,58 @@ +/* + * FILE: fasttrap_dev.c + * DESCRIPTION: Fasttrap Tracing: device file handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include + +#include "dtrace_dev.h" + +static long fasttrap_ioctl(struct file *file, + unsigned int cmd, unsigned long arg) +{ + return -EAGAIN; +} + +static int fasttrap_open(struct inode *inode, struct file *file) +{ + return -EAGAIN; +} + +static int fasttrap_close(struct inode *inode, struct file *file) +{ + return 0; +} + +static const struct file_operations fasttrap_fops = { + .owner = THIS_MODULE, + .unlocked_ioctl = fasttrap_ioctl, + .open = fasttrap_open, + .release = fasttrap_close, +}; + +static struct miscdevice fasttrap_dev = { + .minor = DT_DEV_FASTTRAP_MINOR, + .name = "fasttrap", + .nodename = "dtrace/provider/fasttrap", + .fops = &fasttrap_fops, +}; + +int fasttrap_dev_init(void) +{ + int ret = 0; + + ret = misc_register(&fasttrap_dev); + if (ret) + pr_err("%s: Can't register misc device %d\n", + fasttrap_dev.name, fasttrap_dev.minor); + + return ret; +} + +void fasttrap_dev_exit(void) +{ + misc_deregister(&fasttrap_dev); +} diff --git a/dtrace/fasttrap_mod.c b/dtrace/fasttrap_mod.c new file mode 100644 index 000000000000..44e7340d132b --- /dev/null +++ b/dtrace/fasttrap_mod.c @@ -0,0 +1,25 @@ +/* + * FILE: fasttrap_mod.c + * DESCRIPTION: Fasttrap Tracing: module handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace.h" +#include "dtrace_dev.h" +#include "fasttrap.h" + +MODULE_AUTHOR("Kris Van Hees (kris.van.hees@oracle.com)"); +MODULE_DESCRIPTION("Fasttrap Tracing"); +MODULE_VERSION("v0.1"); +MODULE_LICENSE("Proprietary"); + +static const dtrace_pattr_t fasttrap_attr = { +}; + +static dtrace_pops_t fasttrap_pops = { +}; + +DT_PROVIDER_MODULE(fasttrap, DTRACE_PRIV_PROC | DTRACE_PRIV_OWNER) diff --git a/dtrace/fbt.h b/dtrace/fbt.h new file mode 100644 index 000000000000..6b6134943d17 --- /dev/null +++ b/dtrace/fbt.h @@ -0,0 +1,7 @@ +#ifndef _FBT_H_ +#define _FBT_H_ + +extern int fbt_dev_init(void); +extern void fbt_dev_exit(void); + +#endif /* _FBT_H_ */ diff --git a/dtrace/fbt_dev.c b/dtrace/fbt_dev.c new file mode 100644 index 000000000000..17f3b8823e53 --- /dev/null +++ b/dtrace/fbt_dev.c @@ -0,0 +1,58 @@ +/* + * FILE: fbt_dev.c + * DESCRIPTION: Function Boundary Tracing: device file handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include + +#include "dtrace_dev.h" + +static long fbt_ioctl(struct file *file, + unsigned int cmd, unsigned long arg) +{ + return -EAGAIN; +} + +static int fbt_open(struct inode *inode, struct file *file) +{ + return -EAGAIN; +} + +static int fbt_close(struct inode *inode, struct file *file) +{ + return 0; +} + +static const struct file_operations fbt_fops = { + .owner = THIS_MODULE, + .unlocked_ioctl = fbt_ioctl, + .open = fbt_open, + .release = fbt_close, +}; + +static struct miscdevice fbt_dev = { + .minor = DT_DEV_FBT_MINOR, + .name = "fbt", + .nodename = "dtrace/provider/fbt", + .fops = &fbt_fops, +}; + +int fbt_dev_init(void) +{ + int ret = 0; + + ret = misc_register(&fbt_dev); + if (ret) + pr_err("%s: Can't register misc device %d\n", + fbt_dev.name, fbt_dev.minor); + + return ret; +} + +void fbt_dev_exit(void) +{ + misc_deregister(&fbt_dev); +} diff --git a/dtrace/fbt_mod.c b/dtrace/fbt_mod.c new file mode 100644 index 000000000000..b41d2b1b5cd9 --- /dev/null +++ b/dtrace/fbt_mod.c @@ -0,0 +1,25 @@ +/* + * FILE: fbt_dmod.c + * DESCRIPTION: Function Boundary Tracing: module handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace.h" +#include "dtrace_dev.h" +#include "fbt.h" + +MODULE_AUTHOR("Kris Van Hees (kris.van.hees@oracle.com)"); +MODULE_DESCRIPTION("Function Boundary Tracing"); +MODULE_VERSION("v0.1"); +MODULE_LICENSE("Proprietary"); + +static const dtrace_pattr_t fbt_attr = { +}; + +static dtrace_pops_t fbt_pops = { +}; + +DT_PROVIDER_MODULE(fbt, DTRACE_PRIV_KERNEL); diff --git a/dtrace/lockstat.h b/dtrace/lockstat.h new file mode 100644 index 000000000000..156ce7323328 --- /dev/null +++ b/dtrace/lockstat.h @@ -0,0 +1,7 @@ +#ifndef _LOCKSTAT_H_ +#define _LOCKSTAT_H_ + +extern int lockstat_dev_init(void); +extern void lockstat_dev_exit(void); + +#endif /* _LOCKSTAT_H_ */ diff --git a/dtrace/lockstat_dev.c b/dtrace/lockstat_dev.c new file mode 100644 index 000000000000..a4525bf7004c --- /dev/null +++ b/dtrace/lockstat_dev.c @@ -0,0 +1,58 @@ +/* + * FILE: lockstat_dev.c + * DESCRIPTION: Lock Statistics: device file handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include + +#include "dtrace_dev.h" + +static long lockstat_ioctl(struct file *file, + unsigned int cmd, unsigned long arg) +{ + return -EAGAIN; +} + +static int lockstat_open(struct inode *inode, struct file *file) +{ + return -EAGAIN; +} + +static int lockstat_close(struct inode *inode, struct file *file) +{ + return 0; +} + +static const struct file_operations lockstat_fops = { + .owner = THIS_MODULE, + .unlocked_ioctl = lockstat_ioctl, + .open = lockstat_open, + .release = lockstat_close, +}; + +static struct miscdevice lockstat_dev = { + .minor = DT_DEV_LOCKSTAT_MINOR, + .name = "dtrace", + .nodename = "dtrace/provider/lockstat", + .fops = &lockstat_fops, +}; + +int lockstat_dev_init(void) +{ + int ret = 0; + + ret = misc_register(&lockstat_dev); + if (ret) + pr_err("%s: Can't register misc device %d\n", + lockstat_dev.name, lockstat_dev.minor); + + return ret; +} + +void lockstat_dev_exit(void) +{ + misc_deregister(&lockstat_dev); +} diff --git a/dtrace/lockstat_mod.c b/dtrace/lockstat_mod.c new file mode 100644 index 000000000000..0b16ea949146 --- /dev/null +++ b/dtrace/lockstat_mod.c @@ -0,0 +1,25 @@ +/* + * FILE: lockstat_mod.c + * DESCRIPTION: Lock Statistics: module handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace.h" +#include "dtrace_dev.h" +#include "lockstat.h" + +MODULE_AUTHOR("Kris Van Hees (kris.van.hees@oracle.com)"); +MODULE_DESCRIPTION("Lock Statistics"); +MODULE_VERSION("v0.1"); +MODULE_LICENSE("Proprietary"); + +static const dtrace_pattr_t lockstat_attr = { +}; + +static dtrace_pops_t lockstat_pops = { +}; + +DT_PROVIDER_MODULE(lockstat, DTRACE_PRIV_KERNEL) diff --git a/dtrace/profile.h b/dtrace/profile.h new file mode 100644 index 000000000000..8f96a8acb2dd --- /dev/null +++ b/dtrace/profile.h @@ -0,0 +1,15 @@ +#ifndef _PROFILE_H_ +#define _PROFILE_H_ + +extern void profile_provide(void *, const dtrace_probedesc_t *); +extern int profile_enable(void *, dtrace_id_t, void *); +extern void profile_disable(void *, dtrace_id_t, void *); +extern int profile_usermode(void *, dtrace_id_t, void *); +extern void profile_destroy(void *, dtrace_id_t, void *); + +extern dtrace_provider_id_t profile_id; + +extern int profile_dev_init(void); +extern void profile_dev_exit(void); + +#endif /* _PROFILE_H_ */ diff --git a/dtrace/profile_dev.c b/dtrace/profile_dev.c new file mode 100644 index 000000000000..4f98957aa91d --- /dev/null +++ b/dtrace/profile_dev.c @@ -0,0 +1,348 @@ +/* + * FILE: profile_dev.c + * DESCRIPTION: Profile Interrupt Tracing: device file handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include + +#include "dtrace.h" +#include "dtrace_dev.h" +#include "profile.h" + +#define PROF_NAMELEN 15 +#define PROF_PROFILE 0 +#define PROF_TICK 1 +#define PROF_PREFIX_PROFILE "profile-" +#define PROF_PREFIX_TICK "tick-" + +typedef struct profile_probe { + char prof_name[PROF_NAMELEN]; + dtrace_id_t prof_id; + int prof_kind; + ktime_t prof_interval; + cyclic_id_t prof_cyclic; +} profile_probe_t; + +typedef struct profile_probe_percpu { + ktime_t profc_expected; + ktime_t profc_interval; + profile_probe_t *profc_probe; +} profile_probe_percpu_t; + +static ktime_t profile_interval_min = KTIME_INIT(0, NANOSEC / 5000); +static int profile_aframes = 0; + +static int profile_rates[] = { + 97, 199, 499, 997, 1999, + 4001, 4999, 0, 0, 0, + 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, + }; +static int profile_ticks[] = { + 1, 10, 100, 500, 1000, + 5000, 0, 0, 0, 0, + 0, 0, 0, 0, 0, + }; + +/* + * profile_max defines the upper bound on the number of profile probes that + * can exist (this is to prevent malicious or clumsy users from exhausing + * system resources by creating a slew of profile probes). At mod load time, + * this gets its value from PROFILE_MAX_DEFAULT or profile-max-probes if it's + * present as module parameter. + * FIXME: module parameter yet to be implemented. + */ +#define PROFILE_MAX_DEFAULT 1000 /* default max. number of probes */ + +static int profile_max; /* maximum number of profile probes */ +static atomic_t profile_total; /* current number of profile probes */ + +static void profile_tick(void *arg) +{ + profile_probe_t *prof = arg; + + dtrace_probe(prof->prof_id, 0, 0, 0, 0, 0); /* FIXME */ +} + +static void profile_create(ktime_t interval, const char *name, int kind) +{ + profile_probe_t *prof; + int nr_frames = 0; /* FIXME */ + + if (profile_aframes) + nr_frames = profile_aframes; + + if (ktime_lt(interval, profile_interval_min)) + return; + + if (dtrace_probe_lookup(profile_id, NULL, NULL, name) != 0) + return; + + atomic_inc(&profile_total); + if (atomic_read(&profile_total) > profile_max) { + atomic_dec(&profile_total); + return; + } + + prof = kzalloc(sizeof(profile_probe_t), GFP_KERNEL); + strcpy(prof->prof_name, name); + prof->prof_interval = interval; + prof->prof_cyclic = CYCLIC_NONE; + prof->prof_kind = kind; + prof->prof_id = dtrace_probe_create(profile_id, NULL, NULL, name, + nr_frames, prof); +} + +void profile_provide(void *arg, const dtrace_probedesc_t *desc) +{ + int i, j, rate, kind; + long val = 0, mult = 1, mult_s = 0, mult_ns = 0, len; + ktime_t interval; + const char *name, *suffix = NULL; + const struct { + char *prefix; + int kind; + } types[] = { + { PROF_PREFIX_PROFILE, PROF_PROFILE }, + { PROF_PREFIX_TICK, PROF_TICK }, + { NULL, 0 }, + }; + + const struct { + char *name; + long mult_s; + long mult_ns; + } suffixes[] = { + { "ns", 0, 1 }, + { "nsec", 0, 1 }, + { "us", 0, NANOSEC / MICROSEC }, + { "usec", 0, NANOSEC / MICROSEC }, + { "ms", 0, NANOSEC / MILLISEC }, + { "msec", 0, NANOSEC / MILLISEC }, + { "s", 1, 0 }, + { "sec", 1, 0 }, + { "m", 60, 0 }, + { "min", 60, 0 }, + { "h", 60 * 60, 0 }, + { "hour", 60 * 60, 0 }, + { "d", 24 * 60 * 60, 0 }, + { "day", 24 * 60 * 60, 0 }, + { "hz", 0, 0 }, + { NULL, }, + }; + + if (desc == NULL) { + char n[PROF_NAMELEN]; + + /* + * If no description was provided, provide all of our probes. + */ + for (i = 0; i < sizeof(profile_rates) / sizeof(int); i++) { + if ((rate = profile_rates[i]) == 0) + continue; + + snprintf(n, PROF_NAMELEN, "%s%d", + PROF_PREFIX_PROFILE, rate); + profile_create(ktime_set(0, NANOSEC / rate), + n, PROF_PROFILE); + } + + for (i = 0; i < sizeof(profile_ticks) / sizeof(int); i++) { + if ((rate = profile_ticks[i]) == 0) + continue; + + snprintf(n, PROF_NAMELEN, "%s%d", + PROF_PREFIX_TICK, rate); + profile_create(ktime_set(0, NANOSEC / rate), + n, PROF_TICK); + } + + return; + } + + name = desc->dtpd_name; + + for (i = 0; types[i].prefix != NULL; i++) { + len = strlen(types[i].prefix); + + if (strncmp(name, types[i].prefix, len) != 0) + continue; + + break; + } + + if (types[i].prefix == NULL) + return; + + kind = types[i].kind; + + /* + * We need to start before any time suffix. + */ + for (j = strlen(name); j >= len; j--) { + if (name[j] >= '0' && name[j] <= '9') + break; + + suffix = &name[j]; + } + + ASSERT(suffix != NULL); + + /* + * Now determine the numerical value present in the probe name. + */ + for (; j >= len; j--) { + if (name[j] < '0' || name[j] > '9') + return; + + val += (name[j] - '0') * mult; + mult *= 10; + } + + if (val == 0) + return; + + /* + * Look up the suffix to determine the multiplier. + */ + for (i = 0; suffixes[i].name != NULL; i++) { + if (strcasecmp(suffixes[i].name, suffix) == 0) { + mult_s = suffixes[i].mult_s; + mult_ns = suffixes[i].mult_ns; + break; + } + } + + if (suffixes[i].name == NULL && *suffix != '\0') + return; + + if (mult_s == 0 && mult_ns == 0) { + /* + * The default is frequency-per-second. + */ + interval = ns_to_ktime((int64_t)NANOSEC / val); + } else { + long sec; + long nsec = val * mult_ns; + + sec = nsec / NANOSEC; + nsec %= NANOSEC; + + interval = ktime_set(val * mult_s + sec, nsec); + } + + + profile_create(interval, name, kind); +} + +int profile_enable(void *arg, dtrace_id_t id, void *parg) +{ + profile_probe_t *prof = parg; +#ifdef FIXME + cyc_omni_handler_t omni; +#endif + cyc_handler_t hdlr; + cyc_time_t when; + + ASSERT(ktime_nz(prof->prof_interval)); + ASSERT(mutex_is_locked(&cpu_lock)); + + if (prof->prof_kind == PROF_TICK) { + hdlr.cyh_func = profile_tick; + hdlr.cyh_arg = prof; + hdlr.cyh_level = CY_HIGH_LEVEL; + + when.cyt_interval = prof->prof_interval; + when.cyt_when = ktime_set(0, 0); + + prof->prof_cyclic = cyclic_add(&hdlr, &when); + } else { + ASSERT(prof->prof_kind == PROF_PROFILE); + +#ifdef FIXME + omni.cyo_online = profile_online; + omni.cyo_offline = profile_offline; + omni.cyo_arg = prof; + + prof->prof_cyclic = cyclic_add_omni(&omni); +#else + prof->prof_cyclic = CYCLIC_NONE; + return -ENOTSUPP; +#endif + } + + return 0; +} + +void profile_disable(void *arg, dtrace_id_t id, void *parg) +{ + profile_probe_t *prof = parg; + +if (prof->prof_cyclic == CYCLIC_NONE) return; + ASSERT(prof->prof_cyclic != CYCLIC_NONE); + ASSERT(mutex_is_locked(&cpu_lock)); + + cyclic_remove(prof->prof_cyclic); + prof->prof_cyclic = CYCLIC_NONE; +} + +int profile_usermode(void *arg, dtrace_id_t id, void *parg) +{ + return 1; /* FIXME */ +} + +void profile_destroy(void *arg, dtrace_id_t id, void *parg) +{ + profile_probe_t *prof = parg; + + ASSERT(prof->prof_cyclic == CYCLIC_NONE); + kfree(prof); + + ASSERT(atomic_read(&profile_total) >= 1); + atomic_dec(&profile_total); +} + +static int profile_open(struct inode *inode, struct file *file) +{ + return 0; +} + +static int profile_close(struct inode *inode, struct file *file) +{ + return 0; +} + +static const struct file_operations profile_fops = { + .owner = THIS_MODULE, + .open = profile_open, + .release = profile_close, +}; + +static struct miscdevice profile_dev = { + .minor = DT_DEV_PROFILE_MINOR, + .name = "profile", + .nodename = "dtrace/provider/profile", + .fops = &profile_fops, +}; + +int profile_dev_init(void) +{ + int ret = 0; + + ret = misc_register(&profile_dev); + if (ret) + pr_err("%s: Can't register misc device %d\n", + profile_dev.name, profile_dev.minor); + + profile_max = PROFILE_MAX_DEFAULT; + + return ret; +} + +void profile_dev_exit(void) +{ + misc_deregister(&profile_dev); +} diff --git a/dtrace/profile_mod.c b/dtrace/profile_mod.c new file mode 100644 index 000000000000..0f0be15fb342 --- /dev/null +++ b/dtrace/profile_mod.c @@ -0,0 +1,40 @@ +/* + * FILE: profile_mod.c + * DESCRIPTION: Profile Interrupt Tracing: module handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace.h" +#include "dtrace_dev.h" +#include "profile.h" + +MODULE_AUTHOR("Kris Van Hees (kris.van.hees@oracle.com)"); +MODULE_DESCRIPTION("Profile Interrupt Tracing"); +MODULE_VERSION("v0.1"); +MODULE_LICENSE("Proprietary"); + +static const dtrace_pattr_t profile_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_UNSTABLE, DTRACE_STABILITY_UNSTABLE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +}; + +static dtrace_pops_t profile_pops = { + profile_provide, + NULL, + profile_enable, + profile_disable, + NULL, + NULL, + NULL, + NULL, + profile_usermode, + profile_destroy, +}; + +DT_PROVIDER_MODULE(profile, DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER) diff --git a/dtrace/sdt.h b/dtrace/sdt.h new file mode 100644 index 000000000000..6ad3c8013398 --- /dev/null +++ b/dtrace/sdt.h @@ -0,0 +1,7 @@ +#ifndef _SDT_H_ +#define _SDT_H_ + +extern int sdt_dev_init(void); +extern void sdt_dev_exit(void); + +#endif /* _SDT_H_ */ diff --git a/dtrace/sdt_dev.c b/dtrace/sdt_dev.c new file mode 100644 index 000000000000..6c3cbb8977a5 --- /dev/null +++ b/dtrace/sdt_dev.c @@ -0,0 +1,58 @@ +/* + * FILE: sdt_dev.c + * DESCRIPTION: Statically Defined Tracing: device file handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include + +#include "dtrace_dev.h" + +static long sdt_ioctl(struct file *file, + unsigned int cmd, unsigned long arg) +{ + return -EAGAIN; +} + +static int sdt_open(struct inode *inode, struct file *file) +{ + return -EAGAIN; +} + +static int sdt_close(struct inode *inode, struct file *file) +{ + return 0; +} + +static const struct file_operations sdt_fops = { + .owner = THIS_MODULE, + .unlocked_ioctl = sdt_ioctl, + .open = sdt_open, + .release = sdt_close, +}; + +static struct miscdevice sdt_dev = { + .minor = DT_DEV_SDT_MINOR, + .name = "sdt", + .nodename = "dtrace/provider/sdt", + .fops = &sdt_fops, +}; + +int sdt_dev_init(void) +{ + int ret = 0; + + ret = misc_register(&sdt_dev); + if (ret) + pr_err("%s: Can't register misc device %d\n", + sdt_dev.name, sdt_dev.minor); + + return ret; +} + +void sdt_dev_exit(void) +{ + misc_deregister(&sdt_dev); +} diff --git a/dtrace/sdt_impl.h b/dtrace/sdt_impl.h new file mode 100644 index 000000000000..e10a096e2a96 --- /dev/null +++ b/dtrace/sdt_impl.h @@ -0,0 +1,60 @@ +/* + * Copyright 2004 Oracle America, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _SYS_SDT_IMPL_H +#define _SYS_SDT_IMPL_H + +#ifdef __cplusplus +extern "C" { +#endif + +#include "dtrace.h" + +#if defined(__i386__) || defined(__x86_64__) +typedef uint8_t sdt_instr_t; +#else +typedef uint32_t sdt_instr_t; +#endif + +typedef struct sdt_provider { + char *sdtp_name; /* name of provider */ + char *sdtp_prefix; /* prefix for probe names */ + dtrace_pattr_t *sdtp_attr; /* stability attributes */ + dtrace_provider_id_t sdtp_id; /* provider ID */ +} sdt_provider_t; + +extern sdt_provider_t sdt_providers[]; /* array of providers */ + +typedef struct sdt_probe { + sdt_provider_t *sdp_provider; /* provider */ + char *sdp_name; /* name of probe */ + int sdp_namelen; /* length of allocated name */ + dtrace_id_t sdp_id; /* probe ID */ + struct module *sdp_ctl; /* module ptr */ + int sdp_loadcnt; /* load count for module */ + int sdp_primary; /* non-zero if primary mod */ + sdt_instr_t *sdp_patchpoint; /* patch point */ + sdt_instr_t sdp_patchval; /* instruction to patch */ + sdt_instr_t sdp_savedval; /* saved instruction value */ + struct sdt_probe *sdp_next; /* next probe */ + struct sdt_probe *sdp_hashnext; /* next on hash */ +} sdt_probe_t; + +typedef struct sdt_argdesc { + const char *sda_provider; /* provider for arg */ + const char *sda_name; /* name of probe */ + const int sda_ndx; /* argument index */ + const int sda_mapping; /* mapping of argument */ + const char *sda_native; /* native type of argument */ + const char *sda_xlate; /* translated type of arg */ +} sdt_argdesc_t; + +extern void sdt_getargdesc(void *, dtrace_id_t, void *, dtrace_argdesc_t *); + +#ifdef __cplusplus +} +#endif + +#endif /* _SYS_SDT_IMPL_H */ diff --git a/dtrace/sdt_mod.c b/dtrace/sdt_mod.c new file mode 100644 index 000000000000..5989f0da615c --- /dev/null +++ b/dtrace/sdt_mod.c @@ -0,0 +1,618 @@ +/* + * FILE: sdt_mod.c + * DESCRIPTION: Statically Defined Tracing: module handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include +#include +#include +#include +#include + +#include "dtrace.h" +#include "dtrace_dev.h" +#include "sdt.h" +#include "sdt_impl.h" + +#define SDT_PATCHVAL 0xf0 +#define SDT_ADDR2NDX(addr) ((((uintptr_t)(addr)) >> 4) & sdt_probetab_mask) +#define SDT_PROBETAB_SIZE 0x1000 /* 4k entries -- 16K total */ + +///static dev_info_t *sdt_devi; +static int sdt_verbose = 0; +static sdt_probe_t **sdt_probetab; +static int sdt_probetab_size; +static int sdt_probetab_mask; + +struct frame { /* TBD: move to header file, used in dtrace_isa.c also */ + struct frame *fr_savfp; + unsigned long fr_savpc; +} __attribute__((packed)); + +static inline bool mod_loaded(struct module *mod) +{ + return mod->state == MODULE_STATE_LIVE; +} + +static unsigned int mod_loadcnt(struct module *mod) +{ +#ifdef CONFIG_MODULE_UNLOAD + return module_refcount(mod); +#else + return 1; +#endif +} + +#if ELF_CLASS == ELFCLASS32 +typedef Elf32_Sym Sym; +#else +typedef Elf64_Sym Sym; +#endif + +char * +kernel_searchsym(struct module *mp, uintptr_t value, ulong_t *offset) +{ + Sym *symtabptr; + char *strtabptr; + int symnum; + Sym *sym; + Sym *cursym; + uintptr_t curval; + + *offset = (ulong_t)-1l; /* assume not found */ + cursym = NULL; + + if (!within_module_core((unsigned long)value, mp) && + !within_module_init((unsigned long)value, mp)) + return NULL; /* not in this module */ + + strtabptr = NULL; ///FIXME: mp->strings; + symtabptr = (Sym *)mp->symtab; + + /* + * Scan the module's symbol table for a symbol <= value + */ + for (symnum = 1, sym = symtabptr + 1; + symnum < mp->num_symtab; symnum++, sym = (Sym *) + ((uintptr_t)sym /* FIXME: + mp->symhdr->sh_entsize*/)) { + if (ELF_ST_BIND(sym->st_info) != STB_GLOBAL) { + if (ELF_ST_BIND(sym->st_info) != STB_LOCAL) + continue; + if (ELF_ST_TYPE(sym->st_info) != STT_OBJECT && + ELF_ST_TYPE(sym->st_info) != STT_FUNC) + continue; + } + + curval = (uintptr_t)sym->st_value; + + if (curval > value) + continue; + + /* + * If one or both are functions... + */ + if (ELF_ST_TYPE(sym->st_info) == STT_FUNC || (cursym != NULL && + ELF_ST_TYPE(cursym->st_info) == STT_FUNC)) { + /* Ignore if the address is out of the bounds */ + if (value - sym->st_value >= sym->st_size) + continue; + + if (cursym != NULL && + ELF_ST_TYPE(cursym->st_info) == STT_FUNC) { + /* Prefer the function to the non-function */ + if (ELF_ST_TYPE(sym->st_info) != STT_FUNC) + continue; + + /* Prefer the larger of the two functions */ + if (sym->st_size <= cursym->st_size) + continue; + } + } else if (value - curval >= *offset) { + continue; + } + + *offset = (ulong_t)(value - curval); + cursym = sym; + } + if (cursym == NULL) + return (NULL); + + return (strtabptr + cursym->st_name); +} + +static void dtrace_invop_callsite(void) /* TBD: copied from dtrace_isa.c */ +{ +} + +/*ARGSUSED*/ +static int +sdt_invop(uintptr_t addr, uintptr_t *stack, uintptr_t eax) +{ + uintptr_t stack0, stack1, stack2, stack3, stack4; + int i = 0; + sdt_probe_t *sdt = sdt_probetab[SDT_ADDR2NDX(addr)]; + +#ifdef __x86_64__ + /* + * On amd64, stack[0] contains the dereferenced stack pointer, + * stack[1] contains savfp, stack[2] contains savpc. We want + * to step over these entries. + */ + i += 3; +#endif + + for (; sdt != NULL; sdt = sdt->sdp_hashnext) { + if ((uintptr_t)sdt->sdp_patchpoint == addr) { + /* + * When accessing the arguments on the stack, we must + * protect against accessing beyond the stack. We can + * safely set NOFAULT here -- we know that interrupts + * are already disabled. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + stack0 = stack[i++]; + stack1 = stack[i++]; + stack2 = stack[i++]; + stack3 = stack[i++]; + stack4 = stack[i++]; + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT | + CPU_DTRACE_BADADDR); + + dtrace_probe(sdt->sdp_id, stack0, stack1, + stack2, stack3, stack4); + + return (DTRACE_INVOP_NOP); + } + } + + return (0); +} + +/*ARGSUSED*/ +static void +sdt_provide_module(void *arg, struct module *module) +{ + char *modname = module->name; + sdt_probedesc_t *sdpd; + sdt_probe_t *sdp, *old; + sdt_provider_t *prov; + int len; + + /* + * One for all, and all for one: if we haven't yet registered all of + * our providers, we'll refuse to provide anything. + */ + for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) { + if (prov->sdtp_id == DTRACE_PROVNONE) + return; + } + + if (module->sdt_nprobes != 0 || (sdpd = module->sdt_probes) == NULL) + return; + + for (sdpd = module->sdt_probes; sdpd != NULL; sdpd = sdpd->sdpd_next) { + char *name = sdpd->sdpd_name, *func, *nname; + int i, j; + sdt_provider_t *prov; + ulong_t offs; + dtrace_id_t id; + + for (prov = sdt_providers; prov->sdtp_prefix != NULL; prov++) { + char *prefix = prov->sdtp_prefix; + + if (strncmp(name, prefix, strlen(prefix)) == 0) { + name += strlen(prefix); + break; + } + } + + nname = kmalloc(len = strlen(name) + 1, GFP_KERNEL); + if (!nname) { + pr_warning("sdt cannot allocate memory for module: %s", + module->name); + return; + } + + for (i = 0, j = 0; name[j] != '\0'; i++) { + if (name[j] == '_' && name[j + 1] == '_') { + nname[i] = '-'; + j += 2; + } else { + nname[i] = name[j++]; + } + } + + nname[i] = '\0'; + + sdp = kzalloc(sizeof(sdt_probe_t), GFP_KERNEL); + if (!sdp) { + pr_warning("sdt cannot allocate probe for module: %s", + module->name); + kfree(nname); + return; + } +#ifdef CONFIG_MODULE_UNLOAD + sdp->sdp_loadcnt = module_refcount(module); // TBD: not loadcnt +#else + sdp->sdp_loadcnt = 1; +#endif + sdp->sdp_ctl = module; + sdp->sdp_name = nname; + sdp->sdp_namelen = len; + sdp->sdp_provider = prov; + + func = kernel_searchsym(module, sdpd->sdpd_offset, &offs); + + if (func == NULL) + func = ""; + + /* + * We have our provider. Now create the probe. + */ + if ((id = dtrace_probe_lookup(prov->sdtp_id, modname, + func, nname)) != DTRACE_IDNONE) { + old = dtrace_probe_arg(prov->sdtp_id, id); + ASSERT(old != NULL); + + sdp->sdp_next = old->sdp_next; + sdp->sdp_id = id; + old->sdp_next = sdp; + } else { + sdp->sdp_id = dtrace_probe_create(prov->sdtp_id, + modname, func, nname, 3, sdp); + + module->sdt_nprobes++; + } + + sdp->sdp_hashnext = + sdt_probetab[SDT_ADDR2NDX(sdpd->sdpd_offset)]; + sdt_probetab[SDT_ADDR2NDX(sdpd->sdpd_offset)] = sdp; + + sdp->sdp_patchval = SDT_PATCHVAL; + sdp->sdp_patchpoint = (uint8_t *)sdpd->sdpd_offset; + sdp->sdp_savedval = *sdp->sdp_patchpoint; + } +} + +/*ARGSUSED*/ +static void +sdt_destroy(void *arg, dtrace_id_t id, void *parg) +{ + sdt_probe_t *sdp = parg, *old, *last, *hash; + struct module *module = sdp->sdp_ctl; + int ndx; + + if (module != NULL && mod_loadcnt(module) == sdp->sdp_loadcnt) { + if ((mod_loadcnt(module) == sdp->sdp_loadcnt && + mod_loaded(module))) { + module->sdt_nprobes--; + } + } + + while (sdp != NULL) { + old = sdp; + + /* + * Now we need to remove this probe from the sdt_probetab. + */ + ndx = SDT_ADDR2NDX(sdp->sdp_patchpoint); + last = NULL; + hash = sdt_probetab[ndx]; + + while (hash != sdp) { + ASSERT(hash != NULL); + last = hash; + hash = hash->sdp_hashnext; + } + + if (last != NULL) { + last->sdp_hashnext = sdp->sdp_hashnext; + } else { + sdt_probetab[ndx] = sdp->sdp_hashnext; + } + + kfree(sdp->sdp_name); + sdp = sdp->sdp_next; + kfree(old); + } +} + +/*ARGSUSED*/ +static int +sdt_enable(void *arg, dtrace_id_t id, void *parg) +{ + sdt_probe_t *sdp = parg; + struct module *module = sdp->sdp_ctl; + + module->mod_nenabled++; + + /* + * If this module has disappeared since we discovered its probes, + * refuse to enable it. + */ + if (!mod_loaded(module)) { + if (sdt_verbose) { + pr_warning( "sdt is failing for probe %s " + "(module %s unloaded)", + sdp->sdp_name, module->name); + } + goto err; + } + + /* + * Now check that our module has the expected load count. If it + * doesn't, this module must have been unloaded and reloaded -- and + * we're not going to touch it. + */ + if (mod_loadcnt(module) != sdp->sdp_loadcnt) { + if (sdt_verbose) { + pr_warning("sdt is failing for probe %s " + "(module %s reloaded)", + sdp->sdp_name, module->name); + } + goto err; + } + + while (sdp != NULL) { + *sdp->sdp_patchpoint = sdp->sdp_patchval; + sdp = sdp->sdp_next; + } +err: + return (0); +} + +/*ARGSUSED*/ +static void +sdt_disable(void *arg, dtrace_id_t id, void *parg) +{ + sdt_probe_t *sdp = parg; + struct module *module = sdp->sdp_ctl; + + module->mod_nenabled--; + + if (!mod_loaded(module) || mod_loadcnt(module) != sdp->sdp_loadcnt) + goto err; + + while (sdp != NULL) { + *sdp->sdp_patchpoint = sdp->sdp_savedval; + sdp = sdp->sdp_next; + } + +err: + ; +} + +/*ARGSUSED*/ +uint64_t +sdt_getarg(void *arg, dtrace_id_t id, void *parg, int argno, int aframes) +{ + uintptr_t val; + struct frame *fp = (struct frame *)dtrace_getfp(); + uintptr_t *stack; + int i; +#if defined(__x86_64__) + /* + * A total of 6 arguments are passed via registers; any argument with + * index of 5 or lower is therefore in a register. + */ + int inreg = 5; +#endif + + for (i = 1; i <= aframes; i++) { + fp = (struct frame *)(fp->fr_savfp); + + if (fp->fr_savpc == (pc_t)dtrace_invop_callsite) { +#if !defined(__x86_64__) + /* + * If we pass through the invalid op handler, we will + * use the pointer that it passed to the stack as the + * second argument to dtrace_invop() as the pointer to + * the stack. + */ + stack = ((uintptr_t **)&fp[1])[1]; +#else + /* + * In the case of amd64, we will use the pointer to the + * regs structure that was pushed when we took the + * trap. To get this structure, we must increment + * beyond the frame structure. If the argument that + * we're seeking is passed on the stack, we'll pull + * the true stack pointer out of the saved registers + * and decrement our argument by the number of + * arguments passed in registers; if the argument + * we're seeking is passed in regsiters, we can just + * load it directly. + */ + struct pt_regs *rp = (struct pt_regs *)((uintptr_t)&fp[1] + + sizeof(uintptr_t)); /* TBD: CHECK */ + + if (argno <= inreg) { + stack = (uintptr_t *)&rp->di; + } else { + stack = (uintptr_t *)(rp->sp); + argno -= (inreg + 1); + } +#endif + goto load; + } + } + + /* + * We know that we did not come through a trap to get into + * dtrace_probe() -- the provider simply called dtrace_probe() + * directly. As this is the case, we need to shift the argument + * that we're looking for: the probe ID is the first argument to + * dtrace_probe(), so the argument n will actually be found where + * one would expect to find argument (n + 1). + */ + argno++; + +#if defined(__x86_64__) + if (argno <= inreg) { + /* + * This shouldn't happen. If the argument is passed in a + * register then it should have been, well, passed in a + * register... + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } + + argno -= (inreg + 1); +#endif + stack = (uintptr_t *)&fp[1]; + +load: + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + val = stack[argno]; + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + return (val); +} + +static dtrace_pops_t sdt_pops = { + NULL, + sdt_provide_module, + sdt_enable, + sdt_disable, + NULL, + NULL, + sdt_getargdesc, + sdt_getarg, + NULL, + sdt_destroy +}; + +#if 0 +/*ARGSUSED*/ +static int +sdt_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +{ + sdt_provider_t *prov; + + if (ddi_create_minor_node(devi, "sdt", S_IFCHR, + 0, DDI_PSEUDO, NULL) == DDI_FAILURE) { + pr_warning("/dev/sdt couldn't create minor node"); + ddi_remove_minor_node(devi, NULL); + return (DDI_FAILURE); + } + + ddi_report_dev(devi); + ///sdt_devi = devi; + + if (sdt_probetab_size == 0) + sdt_probetab_size = SDT_PROBETAB_SIZE; + + sdt_probetab_mask = sdt_probetab_size - 1; + sdt_probetab = + kzalloc(sdt_probetab_size * sizeof(sdt_probe_t *), GFP_KERNEL); + if (!sdt_probetab) { + pr_warning("sdt cannot allocate/register provider: %s", + prov->sdtp_name); + ddi_remove_minor_node(devi, NULL); + return (DDI_FAILURE); + } + + dtrace_invop_add(sdt_invop); + + for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) { + if (dtrace_register(prov->sdtp_name, prov->sdtp_attr, + DTRACE_PRIV_KERNEL, NULL, + &sdt_pops, prov, &prov->sdtp_id) != 0) { + pr_warning("failed to register sdt provider %s", + prov->sdtp_name); + } + } + + return (DDI_SUCCESS); +} + +/*ARGSUSED*/ +static int +sdt_detach(dev_info_t *dip, ddi_detach_cmd_t cmd) +{ + sdt_provider_t *prov; + + switch (cmd) { + case DDI_DETACH: + break; + + case DDI_SUSPEND: + return (DDI_SUCCESS); + + default: + return (DDI_FAILURE); + } + + for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) { + if (prov->sdtp_id != DTRACE_PROVNONE) { + if (dtrace_unregister(prov->sdtp_id) != 0) + return (DDI_FAILURE); + + prov->sdtp_id = DTRACE_PROVNONE; + } + } + + dtrace_invop_remove(sdt_invop); + kmem_free(sdt_probetab, sdt_probetab_size * sizeof(sdt_probe_t *)); + + return (DDI_SUCCESS); +} + +/*ARGSUSED*/ +static int +sdt_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *)sdt_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} +#endif + +/*ARGSUSED*/ +static int +sdt_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +{ + return (0); +} + +#if 0 +int +sdt_init(void) +{ + return 0; +} + +void +sdt_exit(void) +{ +} +#endif + +static const dtrace_pattr_t sdt_attr = { +}; + +extern int sdt_dev_init(void); +extern void sdt_dev_exit(void); + +DT_PROVIDER_MODULE(sdt, DTRACE_PRIV_KERNEL) + +MODULE_AUTHOR("Kris Van Hees (kris.van.hees@oracle.com)"); +MODULE_DESCRIPTION("Statically Defined Tracing"); +MODULE_VERSION("v0.1"); +MODULE_LICENSE("Proprietary"); diff --git a/dtrace/sdt_subr.c b/dtrace/sdt_subr.c new file mode 100644 index 000000000000..dd7ea5bfe230 --- /dev/null +++ b/dtrace/sdt_subr.c @@ -0,0 +1,1171 @@ +/* + * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved. + */ + +#include "sdt_impl.h" + +static dtrace_pattr_t vtrace_attr = { +{ DTRACE_STABILITY_UNSTABLE, DTRACE_STABILITY_UNSTABLE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_UNSTABLE, DTRACE_STABILITY_UNSTABLE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t info_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t fc_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t fpu_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_CPU }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t fsinfo_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t stab_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t sdt_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t xpv_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_PLATFORM }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_PLATFORM }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_PLATFORM }, +}; + +static dtrace_pattr_t iscsi_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +}; + +sdt_provider_t sdt_providers[] = { + { "vtrace", "__vtrace_", &vtrace_attr, 0 }, + { "sysinfo", "__cpu_sysinfo_", &info_attr, 0 }, + { "vminfo", "__cpu_vminfo_", &info_attr, 0 }, + { "fpuinfo", "__fpuinfo_", &fpu_attr, 0 }, + { "sched", "__sched_", &stab_attr, 0 }, + { "proc", "__proc_", &stab_attr, 0 }, + { "io", "__io_", &stab_attr, 0 }, + { "ip", "__ip_", &stab_attr, 0 }, + { "tcp", "__tcp_", &stab_attr, 0 }, + { "udp", "__udp_", &stab_attr, 0 }, + { "mib", "__mib_", &stab_attr, 0 }, + { "fsinfo", "__fsinfo_", &fsinfo_attr, 0 }, + { "iscsi", "__iscsi_", &iscsi_attr, 0 }, + { "nfsv3", "__nfsv3_", &stab_attr, 0 }, + { "nfsv4", "__nfsv4_", &stab_attr, 0 }, + { "xpv", "__xpv_", &xpv_attr, 0 }, + { "fc", "__fc_", &fc_attr, 0 }, + { "srp", "__srp_", &fc_attr, 0 }, + { "sysevent", "__sysevent_", &stab_attr, 0 }, + { "sdt", NULL, &sdt_attr, 0 }, + { NULL } +}; + +sdt_argdesc_t sdt_args[] = { + { "sched", "wakeup", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "wakeup", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "dequeue", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "dequeue", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "dequeue", 2, 1, "disp_t *", "cpuinfo_t *" }, + { "sched", "enqueue", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "enqueue", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "enqueue", 2, 1, "disp_t *", "cpuinfo_t *" }, + { "sched", "enqueue", 3, 2, "int" }, + { "sched", "off-cpu", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "off-cpu", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "tick", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "tick", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "change-pri", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "change-pri", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "change-pri", 2, 1, "pri_t" }, + { "sched", "schedctl-nopreempt", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "schedctl-nopreempt", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "schedctl-nopreempt", 2, 1, "int" }, + { "sched", "schedctl-preempt", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "schedctl-preempt", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "schedctl-yield", 0, 0, "int" }, + { "sched", "surrender", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "surrender", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "cpucaps-sleep", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "cpucaps-sleep", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "cpucaps-wakeup", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "cpucaps-wakeup", 1, 0, "kthread_t *", "psinfo_t *" }, + + { "proc", "create", 0, 0, "proc_t *", "psinfo_t *" }, + { "proc", "exec", 0, 0, "string" }, + { "proc", "exec-failure", 0, 0, "int" }, + { "proc", "exit", 0, 0, "int" }, + { "proc", "fault", 0, 0, "int" }, + { "proc", "fault", 1, 1, "siginfo_t *" }, + { "proc", "lwp-create", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "proc", "lwp-create", 1, 0, "kthread_t *", "psinfo_t *" }, + { "proc", "signal-clear", 0, 0, "int" }, + { "proc", "signal-clear", 1, 1, "siginfo_t *" }, + { "proc", "signal-discard", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "proc", "signal-discard", 1, 1, "proc_t *", "psinfo_t *" }, + { "proc", "signal-discard", 2, 2, "int" }, + { "proc", "signal-handle", 0, 0, "int" }, + { "proc", "signal-handle", 1, 1, "siginfo_t *" }, + { "proc", "signal-handle", 2, 2, "void (*)(void)" }, + { "proc", "signal-send", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "proc", "signal-send", 1, 0, "kthread_t *", "psinfo_t *" }, + { "proc", "signal-send", 2, 1, "int" }, + + { "io", "start", 0, 0, "buf_t *", "bufinfo_t *" }, + { "io", "start", 1, 0, "buf_t *", "devinfo_t *" }, + { "io", "start", 2, 0, "buf_t *", "fileinfo_t *" }, + { "io", "done", 0, 0, "buf_t *", "bufinfo_t *" }, + { "io", "done", 1, 0, "buf_t *", "devinfo_t *" }, + { "io", "done", 2, 0, "buf_t *", "fileinfo_t *" }, + { "io", "wait-start", 0, 0, "buf_t *", "bufinfo_t *" }, + { "io", "wait-start", 1, 0, "buf_t *", "devinfo_t *" }, + { "io", "wait-start", 2, 0, "buf_t *", "fileinfo_t *" }, + { "io", "wait-done", 0, 0, "buf_t *", "bufinfo_t *" }, + { "io", "wait-done", 1, 0, "buf_t *", "devinfo_t *" }, + { "io", "wait-done", 2, 0, "buf_t *", "fileinfo_t *" }, + + { "mib", NULL, 0, 0, "int" }, + + { "fsinfo", NULL, 0, 0, "vnode_t *", "fileinfo_t *" }, + { "fsinfo", NULL, 1, 1, "int", "int" }, + + { "iscsi", "async-send", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "async-send", 1, 1, "iscsi_async_evt_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "login-command", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "login-command", 1, 1, "iscsi_login_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "login-response", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "login-response", 1, 1, "iscsi_login_rsp_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "logout-command", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "logout-command", 1, 1, "iscsi_logout_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "logout-response", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "logout-response", 1, 1, "iscsi_logout_rsp_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "data-request", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "data-request", 1, 1, "iscsi_rtt_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "data-send", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "data-send", 1, 1, "iscsi_data_rsp_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "data-receive", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "data-receive", 1, 1, "iscsi_data_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "nop-send", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "nop-send", 1, 1, "iscsi_nop_in_hdr_t *", "iscsiinfo_t *" }, + { "iscsi", "nop-receive", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "nop-receive", 1, 1, "iscsi_nop_out_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "scsi-command", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "scsi-command", 1, 1, "iscsi_scsi_cmd_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "scsi-command", 2, 2, "scsi_task_t *", "scsicmd_t *" }, + { "iscsi", "scsi-response", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "scsi-response", 1, 1, "iscsi_scsi_rsp_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "task-command", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "task-command", 1, 1, "iscsi_scsi_task_mgt_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "task-response", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "task-response", 1, 1, "iscsi_scsi_task_mgt_rsp_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "text-command", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "text-command", 1, 1, "iscsi_text_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "text-response", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "text-response", 1, 1, "iscsi_text_rsp_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "xfer-start", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "xfer-start", 1, 0, "idm_conn_t *", "iscsiinfo_t *" }, + { "iscsi", "xfer-start", 2, 1, "uintptr_t", "xferinfo_t *" }, + { "iscsi", "xfer-start", 3, 2, "uint32_t"}, + { "iscsi", "xfer-start", 4, 3, "uintptr_t"}, + { "iscsi", "xfer-start", 5, 4, "uint32_t"}, + { "iscsi", "xfer-start", 6, 5, "uint32_t"}, + { "iscsi", "xfer-start", 7, 6, "uint32_t"}, + { "iscsi", "xfer-start", 8, 7, "int"}, + { "iscsi", "xfer-done", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "xfer-done", 1, 0, "idm_conn_t *", "iscsiinfo_t *" }, + { "iscsi", "xfer-done", 2, 1, "uintptr_t", "xferinfo_t *" }, + { "iscsi", "xfer-done", 3, 2, "uint32_t"}, + { "iscsi", "xfer-done", 4, 3, "uintptr_t"}, + { "iscsi", "xfer-done", 5, 4, "uint32_t"}, + { "iscsi", "xfer-done", 6, 5, "uint32_t"}, + { "iscsi", "xfer-done", 7, 6, "uint32_t"}, + { "iscsi", "xfer-done", 8, 7, "int"}, + + { "nfsv3", "op-getattr-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-getattr-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-getattr-start", 2, 3, "GETATTR3args *" }, + { "nfsv3", "op-getattr-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-getattr-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-getattr-done", 2, 3, "GETATTR3res *" }, + { "nfsv3", "op-setattr-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-setattr-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-setattr-start", 2, 3, "SETATTR3args *" }, + { "nfsv3", "op-setattr-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-setattr-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-setattr-done", 2, 3, "SETATTR3res *" }, + { "nfsv3", "op-lookup-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-lookup-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-lookup-start", 2, 3, "LOOKUP3args *" }, + { "nfsv3", "op-lookup-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-lookup-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-lookup-done", 2, 3, "LOOKUP3res *" }, + { "nfsv3", "op-access-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-access-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-access-start", 2, 3, "ACCESS3args *" }, + { "nfsv3", "op-access-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-access-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-access-done", 2, 3, "ACCESS3res *" }, + { "nfsv3", "op-commit-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-commit-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-commit-start", 2, 3, "COMMIT3args *" }, + { "nfsv3", "op-commit-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-commit-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-commit-done", 2, 3, "COMMIT3res *" }, + { "nfsv3", "op-create-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-create-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-create-start", 2, 3, "CREATE3args *" }, + { "nfsv3", "op-create-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-create-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-create-done", 2, 3, "CREATE3res *" }, + { "nfsv3", "op-fsinfo-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-fsinfo-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-fsinfo-start", 2, 3, "FSINFO3args *" }, + { "nfsv3", "op-fsinfo-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-fsinfo-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-fsinfo-done", 2, 3, "FSINFO3res *" }, + { "nfsv3", "op-fsstat-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-fsstat-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-fsstat-start", 2, 3, "FSSTAT3args *" }, + { "nfsv3", "op-fsstat-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-fsstat-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-fsstat-done", 2, 3, "FSSTAT3res *" }, + { "nfsv3", "op-link-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-link-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-link-start", 2, 3, "LINK3args *" }, + { "nfsv3", "op-link-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-link-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-link-done", 2, 3, "LINK3res *" }, + { "nfsv3", "op-mkdir-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-mkdir-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-mkdir-start", 2, 3, "MKDIR3args *" }, + { "nfsv3", "op-mkdir-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-mkdir-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-mkdir-done", 2, 3, "MKDIR3res *" }, + { "nfsv3", "op-mknod-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-mknod-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-mknod-start", 2, 3, "MKNOD3args *" }, + { "nfsv3", "op-mknod-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-mknod-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-mknod-done", 2, 3, "MKNOD3res *" }, + { "nfsv3", "op-null-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-null-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-null-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-null-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-pathconf-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-pathconf-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-pathconf-start", 2, 3, "PATHCONF3args *" }, + { "nfsv3", "op-pathconf-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-pathconf-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-pathconf-done", 2, 3, "PATHCONF3res *" }, + { "nfsv3", "op-read-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-read-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-read-start", 2, 3, "READ3args *" }, + { "nfsv3", "op-read-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-read-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-read-done", 2, 3, "READ3res *" }, + { "nfsv3", "op-readdir-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readdir-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readdir-start", 2, 3, "READDIR3args *" }, + { "nfsv3", "op-readdir-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readdir-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readdir-done", 2, 3, "READDIR3res *" }, + { "nfsv3", "op-readdirplus-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readdirplus-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readdirplus-start", 2, 3, "READDIRPLUS3args *" }, + { "nfsv3", "op-readdirplus-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readdirplus-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readdirplus-done", 2, 3, "READDIRPLUS3res *" }, + { "nfsv3", "op-readlink-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readlink-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readlink-start", 2, 3, "READLINK3args *" }, + { "nfsv3", "op-readlink-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readlink-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readlink-done", 2, 3, "READLINK3res *" }, + { "nfsv3", "op-remove-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-remove-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-remove-start", 2, 3, "REMOVE3args *" }, + { "nfsv3", "op-remove-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-remove-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-remove-done", 2, 3, "REMOVE3res *" }, + { "nfsv3", "op-rename-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-rename-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-rename-start", 2, 3, "RENAME3args *" }, + { "nfsv3", "op-rename-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-rename-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-rename-done", 2, 3, "RENAME3res *" }, + { "nfsv3", "op-rmdir-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-rmdir-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-rmdir-start", 2, 3, "RMDIR3args *" }, + { "nfsv3", "op-rmdir-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-rmdir-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-rmdir-done", 2, 3, "RMDIR3res *" }, + { "nfsv3", "op-setattr-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-setattr-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-setattr-start", 2, 3, "SETATTR3args *" }, + { "nfsv3", "op-setattr-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-setattr-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-setattr-done", 2, 3, "SETATTR3res *" }, + { "nfsv3", "op-symlink-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-symlink-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-symlink-start", 2, 3, "SYMLINK3args *" }, + { "nfsv3", "op-symlink-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-symlink-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-symlink-done", 2, 3, "SYMLINK3res *" }, + { "nfsv3", "op-write-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-write-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-write-start", 2, 3, "WRITE3args *" }, + { "nfsv3", "op-write-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-write-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-write-done", 2, 3, "WRITE3res *" }, + + { "nfsv4", "null-start", 0, 0, "struct svc_req *", "conninfo_t *" }, + { "nfsv4", "null-done", 0, 0, "struct svc_req *", "conninfo_t *" }, + { "nfsv4", "compound-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "compound-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "compound-start", 2, 1, "COMPOUND4args *" }, + { "nfsv4", "compound-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "compound-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "compound-done", 2, 1, "COMPOUND4res *" }, + { "nfsv4", "op-access-start", 0, 0, "struct compound_state *", + "conninfo_t *"}, + { "nfsv4", "op-access-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-access-start", 2, 1, "ACCESS4args *" }, + { "nfsv4", "op-access-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-access-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-access-done", 2, 1, "ACCESS4res *" }, + { "nfsv4", "op-close-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-close-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-close-start", 2, 1, "CLOSE4args *" }, + { "nfsv4", "op-close-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-close-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-close-done", 2, 1, "CLOSE4res *" }, + { "nfsv4", "op-commit-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-commit-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-commit-start", 2, 1, "COMMIT4args *" }, + { "nfsv4", "op-commit-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-commit-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-commit-done", 2, 1, "COMMIT4res *" }, + { "nfsv4", "op-create-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-create-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-create-start", 2, 1, "CREATE4args *" }, + { "nfsv4", "op-create-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-create-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-create-done", 2, 1, "CREATE4res *" }, + { "nfsv4", "op-delegpurge-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-delegpurge-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-delegpurge-start", 2, 1, "DELEGPURGE4args *" }, + { "nfsv4", "op-delegpurge-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-delegpurge-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-delegpurge-done", 2, 1, "DELEGPURGE4res *" }, + { "nfsv4", "op-delegreturn-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-delegreturn-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-delegreturn-start", 2, 1, "DELEGRETURN4args *" }, + { "nfsv4", "op-delegreturn-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-delegreturn-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-delegreturn-done", 2, 1, "DELEGRETURN4res *" }, + { "nfsv4", "op-getattr-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-getattr-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-getattr-start", 2, 1, "GETATTR4args *" }, + { "nfsv4", "op-getattr-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-getattr-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-getattr-done", 2, 1, "GETATTR4res *" }, + { "nfsv4", "op-getfh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-getfh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-getfh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-getfh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-getfh-done", 2, 1, "GETFH4res *" }, + { "nfsv4", "op-link-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-link-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-link-start", 2, 1, "LINK4args *" }, + { "nfsv4", "op-link-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-link-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-link-done", 2, 1, "LINK4res *" }, + { "nfsv4", "op-lock-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lock-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lock-start", 2, 1, "LOCK4args *" }, + { "nfsv4", "op-lock-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lock-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lock-done", 2, 1, "LOCK4res *" }, + { "nfsv4", "op-lockt-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lockt-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lockt-start", 2, 1, "LOCKT4args *" }, + { "nfsv4", "op-lockt-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lockt-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lockt-done", 2, 1, "LOCKT4res *" }, + { "nfsv4", "op-locku-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-locku-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-locku-start", 2, 1, "LOCKU4args *" }, + { "nfsv4", "op-locku-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-locku-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-locku-done", 2, 1, "LOCKU4res *" }, + { "nfsv4", "op-lookup-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lookup-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lookup-start", 2, 1, "LOOKUP4args *" }, + { "nfsv4", "op-lookup-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lookup-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lookup-done", 2, 1, "LOOKUP4res *" }, + { "nfsv4", "op-lookupp-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lookupp-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lookupp-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lookupp-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lookupp-done", 2, 1, "LOOKUPP4res *" }, + { "nfsv4", "op-nverify-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-nverify-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-nverify-start", 2, 1, "NVERIFY4args *" }, + { "nfsv4", "op-nverify-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-nverify-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-nverify-done", 2, 1, "NVERIFY4res *" }, + { "nfsv4", "op-open-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-start", 2, 1, "OPEN4args *" }, + { "nfsv4", "op-open-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-done", 2, 1, "OPEN4res *" }, + { "nfsv4", "op-open-confirm-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-confirm-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-confirm-start", 2, 1, "OPEN_CONFIRM4args *" }, + { "nfsv4", "op-open-confirm-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-confirm-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-confirm-done", 2, 1, "OPEN_CONFIRM4res *" }, + { "nfsv4", "op-open-downgrade-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-downgrade-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-downgrade-start", 2, 1, "OPEN_DOWNGRADE4args *" }, + { "nfsv4", "op-open-downgrade-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-downgrade-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-downgrade-done", 2, 1, "OPEN_DOWNGRADE4res *" }, + { "nfsv4", "op-openattr-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-openattr-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-openattr-start", 2, 1, "OPENATTR4args *" }, + { "nfsv4", "op-openattr-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-openattr-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-openattr-done", 2, 1, "OPENATTR4res *" }, + { "nfsv4", "op-putfh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putfh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putfh-start", 2, 1, "PUTFH4args *" }, + { "nfsv4", "op-putfh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putfh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putfh-done", 2, 1, "PUTFH4res *" }, + { "nfsv4", "op-putpubfh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putpubfh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putpubfh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putpubfh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putpubfh-done", 2, 1, "PUTPUBFH4res *" }, + { "nfsv4", "op-putrootfh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putrootfh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putrootfh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putrootfh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putrootfh-done", 2, 1, "PUTROOTFH4res *" }, + { "nfsv4", "op-read-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-read-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-read-start", 2, 1, "READ4args *" }, + { "nfsv4", "op-read-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-read-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-read-done", 2, 1, "READ4res *" }, + { "nfsv4", "op-readdir-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-readdir-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-readdir-start", 2, 1, "READDIR4args *" }, + { "nfsv4", "op-readdir-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-readdir-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-readdir-done", 2, 1, "READDIR4res *" }, + { "nfsv4", "op-readlink-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-readlink-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-readlink-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-readlink-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-readlink-done", 2, 1, "READLINK4res *" }, + { "nfsv4", "op-release-lockowner-start", 0, 0, + "struct compound_state *", "conninfo_t *" }, + { "nfsv4", "op-release-lockowner-start", 1, 0, + "struct compound_state *", "nfsv4opinfo_t *" }, + { "nfsv4", "op-release-lockowner-start", 2, 1, + "RELEASE_LOCKOWNER4args *" }, + { "nfsv4", "op-release-lockowner-done", 0, 0, + "struct compound_state *", "conninfo_t *" }, + { "nfsv4", "op-release-lockowner-done", 1, 0, + "struct compound_state *", "nfsv4opinfo_t *" }, + { "nfsv4", "op-release-lockowner-done", 2, 1, + "RELEASE_LOCKOWNER4res *" }, + { "nfsv4", "op-remove-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-remove-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-remove-start", 2, 1, "REMOVE4args *" }, + { "nfsv4", "op-remove-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-remove-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-remove-done", 2, 1, "REMOVE4res *" }, + { "nfsv4", "op-rename-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-rename-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-rename-start", 2, 1, "RENAME4args *" }, + { "nfsv4", "op-rename-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-rename-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-rename-done", 2, 1, "RENAME4res *" }, + { "nfsv4", "op-renew-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-renew-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-renew-start", 2, 1, "RENEW4args *" }, + { "nfsv4", "op-renew-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-renew-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-renew-done", 2, 1, "RENEW4res *" }, + { "nfsv4", "op-restorefh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-restorefh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-restorefh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-restorefh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-restorefh-done", 2, 1, "RESTOREFH4res *" }, + { "nfsv4", "op-savefh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-savefh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-savefh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-savefh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-savefh-done", 2, 1, "SAVEFH4res *" }, + { "nfsv4", "op-secinfo-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-secinfo-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-secinfo-start", 2, 1, "SECINFO4args *" }, + { "nfsv4", "op-secinfo-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-secinfo-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-secinfo-done", 2, 1, "SECINFO4res *" }, + { "nfsv4", "op-setattr-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-setattr-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-setattr-start", 2, 1, "SETATTR4args *" }, + { "nfsv4", "op-setattr-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-setattr-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-setattr-done", 2, 1, "SETATTR4res *" }, + { "nfsv4", "op-setclientid-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-setclientid-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-setclientid-start", 2, 1, "SETCLIENTID4args *" }, + { "nfsv4", "op-setclientid-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-setclientid-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-setclientid-done", 2, 1, "SETCLIENTID4res *" }, + { "nfsv4", "op-setclientid-confirm-start", 0, 0, + "struct compound_state *", "conninfo_t *" }, + { "nfsv4", "op-setclientid-confirm-start", 1, 0, + "struct compound_state *", "nfsv4opinfo_t *" }, + { "nfsv4", "op-setclientid-confirm-start", 2, 1, + "SETCLIENTID_CONFIRM4args *" }, + { "nfsv4", "op-setclientid-confirm-done", 0, 0, + "struct compound_state *", "conninfo_t *" }, + { "nfsv4", "op-setclientid-confirm-done", 1, 0, + "struct compound_state *", "nfsv4opinfo_t *" }, + { "nfsv4", "op-setclientid-confirm-done", 2, 1, + "SETCLIENTID_CONFIRM4res *" }, + { "nfsv4", "op-verify-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-verify-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-verify-start", 2, 1, "VERIFY4args *" }, + { "nfsv4", "op-verify-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-verify-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-verify-done", 2, 1, "VERIFY4res *" }, + { "nfsv4", "op-write-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-write-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-write-start", 2, 1, "WRITE4args *" }, + { "nfsv4", "op-write-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-write-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-write-done", 2, 1, "WRITE4res *" }, + { "nfsv4", "cb-recall-start", 0, 0, "rfs4_client_t *", + "conninfo_t *" }, + { "nfsv4", "cb-recall-start", 1, 1, "rfs4_deleg_state_t *", + "nfsv4cbinfo_t *" }, + { "nfsv4", "cb-recall-start", 2, 2, "CB_RECALL4args *" }, + { "nfsv4", "cb-recall-done", 0, 0, "rfs4_client_t *", + "conninfo_t *" }, + { "nfsv4", "cb-recall-done", 1, 1, "rfs4_deleg_state_t *", + "nfsv4cbinfo_t *" }, + { "nfsv4", "cb-recall-done", 2, 2, "CB_RECALL4res *" }, + + { "ip", "send", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "ip", "send", 1, 1, "conn_t *", "csinfo_t *" }, + { "ip", "send", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "ip", "send", 3, 3, "__dtrace_ipsr_ill_t *", "ifinfo_t *" }, + { "ip", "send", 4, 4, "ipha_t *", "ipv4info_t *" }, + { "ip", "send", 5, 5, "ip6_t *", "ipv6info_t *" }, + { "ip", "send", 6, 6, "int" }, /* used by __dtrace_ipsr_ill_t */ + { "ip", "receive", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "ip", "receive", 1, 1, "conn_t *", "csinfo_t *" }, + { "ip", "receive", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "ip", "receive", 3, 3, "__dtrace_ipsr_ill_t *", "ifinfo_t *" }, + { "ip", "receive", 4, 4, "ipha_t *", "ipv4info_t *" }, + { "ip", "receive", 5, 5, "ip6_t *", "ipv6info_t *" }, + { "ip", "receive", 6, 6, "int" }, /* used by __dtrace_ipsr_ill_t */ + + { "tcp", "connect-established", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "connect-established", 1, 1, "ip_xmit_attr_t *", + "csinfo_t *" }, + { "tcp", "connect-established", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "tcp", "connect-established", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "connect-established", 4, 4, "tcph_t *", "tcpinfo_t *" }, + { "tcp", "connect-refused", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "connect-refused", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "connect-refused", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "tcp", "connect-refused", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "connect-refused", 4, 4, "tcph_t *", "tcpinfo_t *" }, + { "tcp", "connect-request", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "connect-request", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "connect-request", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "tcp", "connect-request", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "connect-request", 4, 4, "tcph_t *", "tcpinfo_t *" }, + { "tcp", "accept-established", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "accept-established", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "accept-established", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "tcp", "accept-established", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "accept-established", 4, 4, "tcph_t *", "tcpinfo_t *" }, + { "tcp", "accept-refused", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "accept-refused", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "accept-refused", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "tcp", "accept-refused", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "accept-refused", 4, 4, "tcph_t *", "tcpinfo_t *" }, + { "tcp", "state-change", 0, 0, "void", "void" }, + { "tcp", "state-change", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "state-change", 2, 2, "void", "void" }, + { "tcp", "state-change", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "state-change", 4, 4, "void", "void" }, + { "tcp", "state-change", 5, 5, "int32_t", "tcplsinfo_t *" }, + { "tcp", "send", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "send", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "send", 2, 2, "__dtrace_tcp_void_ip_t *", "ipinfo_t *" }, + { "tcp", "send", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "send", 4, 4, "__dtrace_tcp_tcph_t *", "tcpinfo_t *" }, + { "tcp", "receive", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "receive", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "receive", 2, 2, "__dtrace_tcp_void_ip_t *", "ipinfo_t *" }, + { "tcp", "receive", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "receive", 4, 4, "__dtrace_tcp_tcph_t *", "tcpinfo_t *" }, + + { "udp", "send", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "udp", "send", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "udp", "send", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "udp", "send", 3, 3, "udp_t *", "udpsinfo_t *" }, + { "udp", "send", 4, 4, "udpha_t *", "udpinfo_t *" }, + { "udp", "receive", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "udp", "receive", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "udp", "receive", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "udp", "receive", 3, 3, "udp_t *", "udpsinfo_t *" }, + { "udp", "receive", 4, 4, "udpha_t *", "udpinfo_t *" }, + + { "sysevent", "post", 0, 0, "evch_bind_t *", "syseventchaninfo_t *" }, + { "sysevent", "post", 1, 1, "sysevent_impl_t *", "syseventinfo_t *" }, + + { "xpv", "add-to-physmap-end", 0, 0, "int" }, + { "xpv", "add-to-physmap-start", 0, 0, "domid_t" }, + { "xpv", "add-to-physmap-start", 1, 1, "uint_t" }, + { "xpv", "add-to-physmap-start", 2, 2, "ulong_t" }, + { "xpv", "add-to-physmap-start", 3, 3, "ulong_t" }, + { "xpv", "decrease-reservation-end", 0, 0, "int" }, + { "xpv", "decrease-reservation-start", 0, 0, "domid_t" }, + { "xpv", "decrease-reservation-start", 1, 1, "ulong_t" }, + { "xpv", "decrease-reservation-start", 2, 2, "uint_t" }, + { "xpv", "decrease-reservation-start", 3, 3, "ulong_t *" }, + { "xpv", "dom-create-start", 0, 0, "xen_domctl_t *" }, + { "xpv", "dom-destroy-start", 0, 0, "domid_t" }, + { "xpv", "dom-pause-start", 0, 0, "domid_t" }, + { "xpv", "dom-unpause-start", 0, 0, "domid_t" }, + { "xpv", "dom-create-end", 0, 0, "int" }, + { "xpv", "dom-destroy-end", 0, 0, "int" }, + { "xpv", "dom-pause-end", 0, 0, "int" }, + { "xpv", "dom-unpause-end", 0, 0, "int" }, + { "xpv", "evtchn-op-end", 0, 0, "int" }, + { "xpv", "evtchn-op-start", 0, 0, "int" }, + { "xpv", "evtchn-op-start", 1, 1, "void *" }, + { "xpv", "increase-reservation-end", 0, 0, "int" }, + { "xpv", "increase-reservation-start", 0, 0, "domid_t" }, + { "xpv", "increase-reservation-start", 1, 1, "ulong_t" }, + { "xpv", "increase-reservation-start", 2, 2, "uint_t" }, + { "xpv", "increase-reservation-start", 3, 3, "ulong_t *" }, + { "xpv", "mmap-end", 0, 0, "int" }, + { "xpv", "mmap-entry", 0, 0, "ulong_t" }, + { "xpv", "mmap-entry", 1, 1, "ulong_t" }, + { "xpv", "mmap-entry", 2, 2, "ulong_t" }, + { "xpv", "mmap-start", 0, 0, "domid_t" }, + { "xpv", "mmap-start", 1, 1, "int" }, + { "xpv", "mmap-start", 2, 2, "privcmd_mmap_entry_t *" }, + { "xpv", "mmapbatch-end", 0, 0, "int" }, + { "xpv", "mmapbatch-end", 1, 1, "struct seg *" }, + { "xpv", "mmapbatch-end", 2, 2, "caddr_t" }, + { "xpv", "mmapbatch-start", 0, 0, "domid_t" }, + { "xpv", "mmapbatch-start", 1, 1, "int" }, + { "xpv", "mmapbatch-start", 2, 2, "caddr_t" }, + { "xpv", "mmu-ext-op-end", 0, 0, "int" }, + { "xpv", "mmu-ext-op-start", 0, 0, "int" }, + { "xpv", "mmu-ext-op-start", 1, 1, "struct mmuext_op *" }, + { "xpv", "mmu-update-start", 0, 0, "int" }, + { "xpv", "mmu-update-start", 1, 1, "int" }, + { "xpv", "mmu-update-start", 2, 2, "mmu_update_t *" }, + { "xpv", "mmu-update-end", 0, 0, "int" }, + { "xpv", "populate-physmap-end", 0, 0, "int" }, + { "xpv", "populate-physmap-start", 0, 0, "domid_t" }, + { "xpv", "populate-physmap-start", 1, 1, "ulong_t" }, + { "xpv", "populate-physmap-start", 2, 2, "ulong_t *" }, + { "xpv", "set-memory-map-end", 0, 0, "int" }, + { "xpv", "set-memory-map-start", 0, 0, "domid_t" }, + { "xpv", "set-memory-map-start", 1, 1, "int" }, + { "xpv", "set-memory-map-start", 2, 2, "struct xen_memory_map *" }, + { "xpv", "setvcpucontext-end", 0, 0, "int" }, + { "xpv", "setvcpucontext-start", 0, 0, "domid_t" }, + { "xpv", "setvcpucontext-start", 1, 1, "vcpu_guest_context_t *" }, + + { "srp", "service-up", 0, 0, "srpt_session_t *", "conninfo_t *" }, + { "srp", "service-up", 1, 0, "srpt_session_t *", "srp_portinfo_t *" }, + { "srp", "service-down", 0, 0, "srpt_session_t *", "conninfo_t *" }, + { "srp", "service-down", 1, 0, "srpt_session_t *", + "srp_portinfo_t *" }, + { "srp", "login-command", 0, 0, "srpt_session_t *", "conninfo_t *" }, + { "srp", "login-command", 1, 0, "srpt_session_t *", + "srp_portinfo_t *" }, + { "srp", "login-command", 2, 1, "srp_login_req_t *", + "srp_logininfo_t *" }, + { "srp", "login-response", 0, 0, "srpt_session_t *", "conninfo_t *" }, + { "srp", "login-response", 1, 0, "srpt_session_t *", + "srp_portinfo_t *" }, + { "srp", "login-response", 2, 1, "srp_login_rsp_t *", + "srp_logininfo_t *" }, + { "srp", "login-response", 3, 2, "srp_login_rej_t *" }, + { "srp", "logout-command", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "logout-command", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "task-command", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "task-command", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "task-command", 2, 1, "srp_cmd_req_t *", "srp_taskinfo_t *" }, + { "srp", "task-response", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "task-response", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "task-response", 2, 1, "srp_rsp_t *", "srp_taskinfo_t *" }, + { "srp", "task-response", 3, 2, "scsi_task_t *" }, + { "srp", "task-response", 4, 3, "int8_t" }, + { "srp", "scsi-command", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "scsi-command", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "scsi-command", 2, 1, "scsi_task_t *", "scsicmd_t *" }, + { "srp", "scsi-command", 3, 2, "srp_cmd_req_t *", "srp_taskinfo_t *" }, + { "srp", "scsi-response", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "scsi-response", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "scsi-response", 2, 1, "srp_rsp_t *", "srp_taskinfo_t *" }, + { "srp", "scsi-response", 3, 2, "scsi_task_t *" }, + { "srp", "scsi-response", 4, 3, "int8_t" }, + { "srp", "xfer-start", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "xfer-start", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "xfer-start", 2, 1, "ibt_wr_ds_t *", "xferinfo_t *" }, + { "srp", "xfer-start", 3, 2, "srpt_iu_t *", "srp_taskinfo_t *" }, + { "srp", "xfer-start", 4, 3, "ibt_send_wr_t *"}, + { "srp", "xfer-start", 5, 4, "uint32_t" }, + { "srp", "xfer-start", 6, 5, "uint32_t" }, + { "srp", "xfer-start", 7, 6, "uint32_t" }, + { "srp", "xfer-start", 8, 7, "uint32_t" }, + { "srp", "xfer-done", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "xfer-done", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "xfer-done", 2, 1, "ibt_wr_ds_t *", "xferinfo_t *" }, + { "srp", "xfer-done", 3, 2, "srpt_iu_t *", "srp_taskinfo_t *" }, + { "srp", "xfer-done", 4, 3, "ibt_send_wr_t *"}, + { "srp", "xfer-done", 5, 4, "uint32_t" }, + { "srp", "xfer-done", 6, 5, "uint32_t" }, + { "srp", "xfer-done", 7, 6, "uint32_t" }, + { "srp", "xfer-done", 8, 7, "uint32_t" }, + + { "fc", "link-up", 0, 0, "fct_i_local_port_t *", "conninfo_t *" }, + { "fc", "link-down", 0, 0, "fct_i_local_port_t *", "conninfo_t *" }, + { "fc", "fabric-login-start", 0, 0, "fct_i_local_port_t *", + "conninfo_t *" }, + { "fc", "fabric-login-start", 1, 0, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "fabric-login-end", 0, 0, "fct_i_local_port_t *", + "conninfo_t *" }, + { "fc", "fabric-login-end", 1, 0, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-login-start", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "rport-login-start", 1, 1, "fct_local_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-login-start", 2, 2, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-login-start", 3, 3, "int", "int" }, + { "fc", "rport-login-end", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "rport-login-end", 1, 1, "fct_local_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-login-end", 2, 2, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-login-end", 3, 3, "int", "int" }, + { "fc", "rport-login-end", 4, 4, "int", "int" }, + { "fc", "rport-logout-start", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "rport-logout-start", 1, 1, "fct_local_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-logout-start", 2, 2, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-logout-start", 3, 3, "int", "int" }, + { "fc", "rport-logout-end", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "rport-logout-end", 1, 1, "fct_local_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-logout-end", 2, 2, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-logout-end", 3, 3, "int", "int" }, + { "fc", "scsi-command", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "scsi-command", 1, 1, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "scsi-command", 2, 2, "scsi_task_t *", + "scsicmd_t *" }, + { "fc", "scsi-command", 3, 3, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "scsi-response", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "scsi-response", 1, 1, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "scsi-response", 2, 2, "scsi_task_t *", + "scsicmd_t *" }, + { "fc", "scsi-response", 3, 3, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "xfer-start", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "xfer-start", 1, 1, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "xfer-start", 2, 2, "scsi_task_t *", + "scsicmd_t *" }, + { "fc", "xfer-start", 3, 3, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "xfer-start", 4, 4, "stmf_data_buf_t *", + "fc_xferinfo_t *" }, + { "fc", "xfer-done", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "xfer-done", 1, 1, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "xfer-done", 2, 2, "scsi_task_t *", + "scsicmd_t *" }, + { "fc", "xfer-done", 3, 3, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "xfer-done", 4, 4, "stmf_data_buf_t *", + "fc_xferinfo_t *" }, + { "fc", "rscn-receive", 0, 0, "fct_i_local_port_t *", + "conninfo_t *" }, + { "fc", "rscn-receive", 1, 1, "int", "int"}, + { "fc", "abts-receive", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "abts-receive", 1, 1, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "abts-receive", 2, 2, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + + + { NULL } +}; + +/*ARGSUSED*/ +void +sdt_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc) +{ + sdt_probe_t *sdp = parg; + int i; + + desc->dtargd_native[0] = '\0'; + desc->dtargd_xlate[0] = '\0'; + + for (i = 0; sdt_args[i].sda_provider != NULL; i++) { + sdt_argdesc_t *a = &sdt_args[i]; + + if (strcmp(sdp->sdp_provider->sdtp_name, a->sda_provider) != 0) + continue; + + if (a->sda_name != NULL && + strcmp(sdp->sdp_name, a->sda_name) != 0) + continue; + + if (desc->dtargd_ndx != a->sda_ndx) + continue; + + if (a->sda_native != NULL) + (void) strcpy(desc->dtargd_native, a->sda_native); + + if (a->sda_xlate != NULL) + (void) strcpy(desc->dtargd_xlate, a->sda_xlate); + + desc->dtargd_mapping = a->sda_mapping; + return; + } + + desc->dtargd_ndx = DTRACE_ARGNONE; +} diff --git a/dtrace/systrace_dev.c b/dtrace/systrace_dev.c new file mode 100644 index 000000000000..964cb0270a51 --- /dev/null +++ b/dtrace/systrace_dev.c @@ -0,0 +1,211 @@ +/* + * FILE: systrace_dev.c + * DESCRIPTION: System Call Tracing: device file handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include +#include +#if 0 +#include +#endif +#include + +#include "dtrace.h" +#include "dtrace_dev.h" +#include "systrace.h" + +#define SYSTRACE_ARTIFICIAL_FRAMES 1 + +#define SYSTRACE_SHIFT 16 +#define SYSTRACE_ENTRY(id) ((1 << SYSTRACE_SHIFT) | (id)) +#define SYSTRACE_RETURN(id) (id) +#define SYSTRACE_SYSNUM(x) ((int)(x) & ((1 << SYSTRACE_SHIFT) - 1)) +#define SYSTRACE_ISENTRY(x) ((int)(x) >> SYSTRACE_SHIFT) + +#if ((1 << SYSTRACE_SHIFT) <= NR_syscalls) +# error 1 << SYSTRACE_SHIFT must exceed number of system calls +#endif + +static systrace_info_t *systrace_info = NULL; + +void systrace_provide(void *arg, const dtrace_probedesc_t *desc) +{ + int i; + + ASSERT(systrace_info != NULL); + + if (desc != NULL) + return; + + for (i = 0; i < NR_syscalls; i++) { +#if 0 + struct syscall_metadata *sm = syscall_nr_to_meta(i); + const char *nm; + + if (sm == NULL) + continue; +#else + const char *nm = systrace_info->sysent[i].name; + int sz; +#endif +printk(KERN_INFO "systrace_provide: [%d] = %s\n", i, nm); + if (nm == NULL) + continue; + + if (systrace_info->sysent[i].stsy_underlying == NULL) + continue; + +#if 0 + nm = sm->name; +#endif + sz = strlen(nm); + if (sz > 4 && memcmp(nm, "sys_", 4) == 0) + nm += 4; + else if (sz > 5 && memcmp(nm, "stub_", 5) == 0) + nm += 5; + + if (dtrace_probe_lookup(syscall_id, NULL, nm, "entry") != 0) + continue; + + dtrace_probe_create(syscall_id, NULL, nm, "entry", + SYSTRACE_ARTIFICIAL_FRAMES, + (void *)((uintptr_t)SYSTRACE_ENTRY(i))); + dtrace_probe_create(syscall_id, NULL, nm, "return", + SYSTRACE_ARTIFICIAL_FRAMES, + (void *)((uintptr_t)SYSTRACE_RETURN(i))); + + systrace_info->sysent[i].stsy_entry = DTRACE_IDNONE; + systrace_info->sysent[i].stsy_return = DTRACE_IDNONE; + } +} + +int systrace_enable(void *arg, dtrace_id_t id, void *parg) +{ + int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); + int enabled = + systrace_info->sysent[sysnum].stsy_entry != DTRACE_IDNONE || + systrace_info->sysent[sysnum].stsy_return != DTRACE_IDNONE; + +#if 0 + if (SYSTRACE_ISENTRY((uintptr_t)parg)) + systrace_info->sysent[sysnum].stsy_entry = id; + else + systrace_info->sysent[sysnum].stsy_return = id; + + if (enabled) { + ASSERT((void *)*(systrace_info->sysent[sysnum].stsy_tblent) == + (void *)systrace_info->syscall); + + return 0; + } + + (void)cmpxchg(systrace_info->sysent[sysnum].stsy_tblent, + systrace_info->sysent[sysnum].stsy_underlying, + systrace_info->syscall); +#else + if (!enabled) { + if (cmpxchg(systrace_info->sysent[sysnum].stsy_tblent, + systrace_info->sysent[sysnum].stsy_underlying, + systrace_info->syscall) != + systrace_info->sysent[sysnum].stsy_underlying) + return 0; + } else + ASSERT((void *)*(systrace_info->sysent[sysnum].stsy_tblent) == + (void *)systrace_info->syscall); + + if (SYSTRACE_ISENTRY((uintptr_t)parg)) + systrace_info->sysent[sysnum].stsy_entry = id; + else + systrace_info->sysent[sysnum].stsy_return = id; +#endif + + return 0; +} + +void systrace_disable(void *arg, dtrace_id_t id, void *parg) +{ + int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); + int enabled = + systrace_info->sysent[sysnum].stsy_entry != DTRACE_IDNONE || + systrace_info->sysent[sysnum].stsy_return != DTRACE_IDNONE; + + if (enabled) + (void)cmpxchg(systrace_info->sysent[sysnum].stsy_tblent, + systrace_info->syscall, + systrace_info->sysent[sysnum].stsy_underlying); + + if (SYSTRACE_ISENTRY((uintptr_t)parg)) + systrace_info->sysent[sysnum].stsy_entry = DTRACE_IDNONE; + else + systrace_info->sysent[sysnum].stsy_return = DTRACE_IDNONE; +} + +void systrace_destroy(void *arg, dtrace_id_t id, void *parg) +{ + int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); + + /* + * Nothing to be done here - just ensure our probe has been disabled. + */ + if (SYSTRACE_ISENTRY((uintptr_t)parg)) + ASSERT(systrace_info->sysent[sysnum].stsy_entry == + DTRACE_IDNONE); + else + ASSERT(systrace_info->sysent[sysnum].stsy_return == + DTRACE_IDNONE); +} + +static long systrace_ioctl(struct file *file, + unsigned int cmd, unsigned long arg) +{ + return -EAGAIN; +} + +static int systrace_open(struct inode *inode, struct file *file) +{ + return -EAGAIN; +} + +static int systrace_close(struct inode *inode, struct file *file) +{ + return 0; +} + +static const struct file_operations systrace_fops = { + .owner = THIS_MODULE, + .unlocked_ioctl = systrace_ioctl, + .open = systrace_open, + .release = systrace_close, +}; + +static struct miscdevice systrace_dev = { + .minor = DT_DEV_SYSTRACE_MINOR, + .name = "systrace", + .nodename = "dtrace/provider/systrace", + .fops = &systrace_fops, +}; + +int syscall_dev_init(void) +{ + int ret = 0; + + systrace_info = dtrace_syscalls_init(); + + ret = misc_register(&systrace_dev); + if (ret) + pr_err("%s: Can't register misc device %d\n", + systrace_dev.name, systrace_dev.minor); + + *(systrace_info->probep) = (dtrace_systrace_probe_t)dtrace_probe; + + return ret; +} + +void syscall_dev_exit(void) +{ + *(systrace_info->probep) = systrace_info->stub; + + misc_deregister(&systrace_dev); +} diff --git a/dtrace/systrace_mod.c b/dtrace/systrace_mod.c new file mode 100644 index 000000000000..6f75b1897701 --- /dev/null +++ b/dtrace/systrace_mod.c @@ -0,0 +1,40 @@ +/* + * FILE: systrace_mod.c + * DESCRIPTION: System Call Tracing: module handling + * + * Copyright (C) 2010 Oracle Corporation + */ + +#include + +#include "dtrace.h" +#include "dtrace_dev.h" +#include "systrace.h" + +MODULE_AUTHOR("Kris Van Hees (kris.van.hees@oracle.com)"); +MODULE_DESCRIPTION("System Call Tracing"); +MODULE_VERSION("v0.1"); +MODULE_LICENSE("Proprietary"); + +static const dtrace_pattr_t syscall_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pops_t syscall_pops = { + systrace_provide, + NULL, + systrace_enable, + systrace_disable, + NULL, + NULL, + NULL, + NULL, + NULL, + systrace_destroy +}; + +DT_PROVIDER_MODULE(syscall, DTRACE_PRIV_USER); -- 2.50.1