]> www.infradead.org Git - users/jedix/linux-maple.git/commitdiff
Merge tag 'for-linus-2024071601' of git://git.kernel.org/pub/scm/linux/kernel/git...
authorLinus Torvalds <torvalds@linux-foundation.org>
Thu, 18 Jul 2024 00:28:31 +0000 (17:28 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 18 Jul 2024 00:28:31 +0000 (17:28 -0700)
Pull HID updates from Benjamin Tissoires:

 - rewrite of the HID-BPF internal implementation to use bpf struct_ops
   instead of a tracing endpoint (Benjamin Tissoires)

 - add two new HID-BPF hooks to be able to intercept userspace calls
   targeting a HID device and filtering them (Benjamin Tissoires)

 - add support for various new devices through HID-BPF filters (Benjamin
   Tissoires)

 - add support for the magic keyboard backlight (Orlando Chamberlain)

 - add the missing MODULE_DESCRIPTION() macros in HID drivers (Jeff
   Johnson)

 - use of kvzalloc in case memory gets too fragmented (Hailong Liu)

 - retrieve the device firmware node in the child HID device (Danny
   Kaehn)

 - some hid-uclogic improvements (José Expósito)

 - some more typos, trivial fixes, kernel doctext and unused functions
   cleanups

* tag 'for-linus-2024071601' of git://git.kernel.org/pub/scm/linux/kernel/git/hid/hid: (60 commits)
  HID: hid-steam: Fix typo in goto label
  HID: mcp2221: Remove unnecessary semicolon
  HID: Fix spelling mistakes "Kensigton" -> "Kensington"
  HID: add more missing MODULE_DESCRIPTION() macros
  HID: samples: fix the 2 struct_ops definitions
  HID: fix for amples in for-6.11/bpf
  HID: apple: Add support for magic keyboard backlight on T2 Macs
  HID: bpf: Thrustmaster TCA Yoke Boeing joystick fix
  HID: bpf: Add Huion Dial 2 bpf fixup
  HID: bpf: Add support for the XP-PEN Deco Mini 4
  HID: bpf: move the BIT() macro to hid_bpf_helpers.h
  HID: bpf: add a driver for the Huion Inspiroy 2S (H641P)
  HID: bpf: Add a HID report composition helper macros
  HID: bpf: doc fixes for hid_hw_request() hooks
  HID: bpf: doc fixes for hid_hw_request() hooks
  HID: bpf: fix gcc warning and unify __u64 into u64
  selftests/hid: ensure CKI can compile our new tests on old kernels
  selftests/hid: add an infinite loop test for hid_bpf_try_input_report
  selftests/hid: add another test for injecting an event from an event hook
  HID: bpf: allow hid_device_event hooks to inject input reports on self
  ...

1  2 
drivers/hid/bpf/hid_bpf_struct_ops.c

index 0000000000000000000000000000000000000000,b03b44ad3458fe4a63ba3f8540128ea9cda490cf..f59cce6e437f7720f028d988d79b23a5810dfedd
mode 000000,100644..100644
--- /dev/null
@@@ -1,0 -1,307 +1,307 @@@
 -static int hid_bpf_reg(void *kdata)
+ // SPDX-License-Identifier: GPL-2.0-only
+ /*
+  *  HID-BPF support for Linux
+  *
+  *  Copyright (c) 2024 Benjamin Tissoires
+  */
+ #include <linux/bitops.h>
+ #include <linux/bpf_verifier.h>
+ #include <linux/bpf.h>
+ #include <linux/btf.h>
+ #include <linux/btf_ids.h>
+ #include <linux/filter.h>
+ #include <linux/hid.h>
+ #include <linux/hid_bpf.h>
+ #include <linux/init.h>
+ #include <linux/module.h>
+ #include <linux/stddef.h>
+ #include <linux/workqueue.h>
+ #include "hid_bpf_dispatch.h"
+ static struct btf *hid_bpf_ops_btf;
+ static int hid_bpf_ops_init(struct btf *btf)
+ {
+       hid_bpf_ops_btf = btf;
+       return 0;
+ }
+ static bool hid_bpf_ops_is_valid_access(int off, int size,
+                                         enum bpf_access_type type,
+                                         const struct bpf_prog *prog,
+                                         struct bpf_insn_access_aux *info)
+ {
+       return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
+ }
+ static int hid_bpf_ops_check_member(const struct btf_type *t,
+                                     const struct btf_member *member,
+                                     const struct bpf_prog *prog)
+ {
+       u32 moff = __btf_member_bit_offset(t, member) / 8;
+       switch (moff) {
+       case offsetof(struct hid_bpf_ops, hid_rdesc_fixup):
+       case offsetof(struct hid_bpf_ops, hid_hw_request):
+       case offsetof(struct hid_bpf_ops, hid_hw_output_report):
+               break;
+       default:
+               if (prog->sleepable)
+                       return -EINVAL;
+       }
+       return 0;
+ }
+ struct hid_bpf_offset_write_range {
+       const char *struct_name;
+       u32 struct_length;
+       u32 start;
+       u32 end;
+ };
+ static int hid_bpf_ops_btf_struct_access(struct bpf_verifier_log *log,
+                                          const struct bpf_reg_state *reg,
+                                          int off, int size)
+ {
+ #define WRITE_RANGE(_name, _field, _is_string)                                        \
+       {                                                                       \
+               .struct_name = #_name,                                          \
+               .struct_length = sizeof(struct _name),                          \
+               .start = offsetof(struct _name, _field),                        \
+               .end = offsetofend(struct _name, _field) - !!(_is_string),      \
+       }
+       const struct hid_bpf_offset_write_range write_ranges[] = {
+               WRITE_RANGE(hid_bpf_ctx, retval, false),
+               WRITE_RANGE(hid_device, name, true),
+               WRITE_RANGE(hid_device, uniq, true),
+               WRITE_RANGE(hid_device, phys, true),
+       };
+ #undef WRITE_RANGE
+       const struct btf_type *state = NULL;
+       const struct btf_type *t;
+       const char *cur = NULL;
+       int i;
+       t = btf_type_by_id(reg->btf, reg->btf_id);
+       for (i = 0; i < ARRAY_SIZE(write_ranges); i++) {
+               const struct hid_bpf_offset_write_range *write_range = &write_ranges[i];
+               s32 type_id;
+               /* we already found a writeable struct, but there is a
+                * new one, let's break the loop.
+                */
+               if (t == state && write_range->struct_name != cur)
+                       break;
+               /* new struct to look for */
+               if (write_range->struct_name != cur) {
+                       type_id = btf_find_by_name_kind(reg->btf, write_range->struct_name,
+                                                       BTF_KIND_STRUCT);
+                       if (type_id < 0)
+                               return -EINVAL;
+                       state = btf_type_by_id(reg->btf, type_id);
+               }
+               /* this is not the struct we are looking for */
+               if (t != state) {
+                       cur = write_range->struct_name;
+                       continue;
+               }
+               /* first time we see this struct, check for out of bounds */
+               if (cur != write_range->struct_name &&
+                   off + size > write_range->struct_length) {
+                       bpf_log(log, "write access for struct %s at off %d with size %d\n",
+                               write_range->struct_name, off, size);
+                       return -EACCES;
+               }
+               /* now check if we are in our boundaries */
+               if (off >= write_range->start && off + size <= write_range->end)
+                       return NOT_INIT;
+               cur = write_range->struct_name;
+       }
+       if (t != state)
+               bpf_log(log, "write access to this struct is not supported\n");
+       else
+               bpf_log(log,
+                       "write access at off %d with size %d on read-only part of %s\n",
+                       off, size, cur);
+       return -EACCES;
+ }
+ static const struct bpf_verifier_ops hid_bpf_verifier_ops = {
+       .get_func_proto = bpf_base_func_proto,
+       .is_valid_access = hid_bpf_ops_is_valid_access,
+       .btf_struct_access = hid_bpf_ops_btf_struct_access,
+ };
+ static int hid_bpf_ops_init_member(const struct btf_type *t,
+                                const struct btf_member *member,
+                                void *kdata, const void *udata)
+ {
+       const struct hid_bpf_ops *uhid_bpf_ops;
+       struct hid_bpf_ops *khid_bpf_ops;
+       u32 moff;
+       uhid_bpf_ops = (const struct hid_bpf_ops *)udata;
+       khid_bpf_ops = (struct hid_bpf_ops *)kdata;
+       moff = __btf_member_bit_offset(t, member) / 8;
+       switch (moff) {
+       case offsetof(struct hid_bpf_ops, hid_id):
+               /* For hid_id and flags fields, this function has to copy it
+                * and return 1 to indicate that the data has been handled by
+                * the struct_ops type, or the verifier will reject the map if
+                * the value of those fields is not zero.
+                */
+               khid_bpf_ops->hid_id = uhid_bpf_ops->hid_id;
+               return 1;
+       case offsetof(struct hid_bpf_ops, flags):
+               if (uhid_bpf_ops->flags & ~BPF_F_BEFORE)
+                       return -EINVAL;
+               khid_bpf_ops->flags = uhid_bpf_ops->flags;
+               return 1;
+       }
+       return 0;
+ }
 -static void hid_bpf_unreg(void *kdata)
++static int hid_bpf_reg(void *kdata, struct bpf_link *link)
+ {
+       struct hid_bpf_ops *ops = kdata;
+       struct hid_device *hdev;
+       int count, err = 0;
+       hdev = hid_get_device(ops->hid_id);
+       if (IS_ERR(hdev))
+               return PTR_ERR(hdev);
+       ops->hdev = hdev;
+       mutex_lock(&hdev->bpf.prog_list_lock);
+       count = list_count_nodes(&hdev->bpf.prog_list);
+       if (count >= HID_BPF_MAX_PROGS_PER_DEV) {
+               err = -E2BIG;
+               goto out_unlock;
+       }
+       if (ops->hid_rdesc_fixup) {
+               if (hdev->bpf.rdesc_ops) {
+                       err = -EINVAL;
+                       goto out_unlock;
+               }
+               hdev->bpf.rdesc_ops = ops;
+       }
+       if (ops->hid_device_event) {
+               err = hid_bpf_allocate_event_data(hdev);
+               if (err)
+                       goto out_unlock;
+       }
+       if (ops->flags & BPF_F_BEFORE)
+               list_add_rcu(&ops->list, &hdev->bpf.prog_list);
+       else
+               list_add_tail_rcu(&ops->list, &hdev->bpf.prog_list);
+       synchronize_srcu(&hdev->bpf.srcu);
+ out_unlock:
+       mutex_unlock(&hdev->bpf.prog_list_lock);
+       if (err) {
+               if (hdev->bpf.rdesc_ops == ops)
+                       hdev->bpf.rdesc_ops = NULL;
+               hid_put_device(hdev);
+       } else if (ops->hid_rdesc_fixup) {
+               hid_bpf_reconnect(hdev);
+       }
+       return err;
+ }
++static void hid_bpf_unreg(void *kdata, struct bpf_link *link)
+ {
+       struct hid_bpf_ops *ops = kdata;
+       struct hid_device *hdev;
+       bool reconnect = false;
+       hdev = ops->hdev;
+       /* check if __hid_bpf_ops_destroy_device() has been called */
+       if (!hdev)
+               return;
+       mutex_lock(&hdev->bpf.prog_list_lock);
+       list_del_rcu(&ops->list);
+       synchronize_srcu(&hdev->bpf.srcu);
+       reconnect = hdev->bpf.rdesc_ops == ops;
+       if (reconnect)
+               hdev->bpf.rdesc_ops = NULL;
+       mutex_unlock(&hdev->bpf.prog_list_lock);
+       if (reconnect)
+               hid_bpf_reconnect(hdev);
+       hid_put_device(hdev);
+ }
+ static int __hid_bpf_device_event(struct hid_bpf_ctx *ctx, enum hid_report_type type, u64 source)
+ {
+       return 0;
+ }
+ static int __hid_bpf_rdesc_fixup(struct hid_bpf_ctx *ctx)
+ {
+       return 0;
+ }
+ static struct hid_bpf_ops __bpf_hid_bpf_ops = {
+       .hid_device_event = __hid_bpf_device_event,
+       .hid_rdesc_fixup = __hid_bpf_rdesc_fixup,
+ };
+ static struct bpf_struct_ops bpf_hid_bpf_ops = {
+       .verifier_ops = &hid_bpf_verifier_ops,
+       .init = hid_bpf_ops_init,
+       .check_member = hid_bpf_ops_check_member,
+       .init_member = hid_bpf_ops_init_member,
+       .reg = hid_bpf_reg,
+       .unreg = hid_bpf_unreg,
+       .name = "hid_bpf_ops",
+       .cfi_stubs = &__bpf_hid_bpf_ops,
+       .owner = THIS_MODULE,
+ };
+ void __hid_bpf_ops_destroy_device(struct hid_device *hdev)
+ {
+       struct hid_bpf_ops *e;
+       rcu_read_lock();
+       list_for_each_entry_rcu(e, &hdev->bpf.prog_list, list) {
+               hid_put_device(hdev);
+               e->hdev = NULL;
+       }
+       rcu_read_unlock();
+ }
+ static int __init hid_bpf_struct_ops_init(void)
+ {
+       return register_bpf_struct_ops(&bpf_hid_bpf_ops, hid_bpf_ops);
+ }
+ late_initcall(hid_bpf_struct_ops_init);