The nodemask routines had mixed return values that provided potentially
signed return values that could never happen. This was leading to the
compiler getting confusing about the range of possible return values
(it was thinking things could be negative where they could not be). Fix
all the nodemask routines that should be returning unsigned
(or bool) values. Silences:
mm/swapfile.c: In function ‘setup_swap_info’:
mm/swapfile.c:2291:47: error: array subscript -1 is below array bounds of ‘struct plist_node[]’ [-Werror=array-bounds]
2291 | p->avail_lists[i].prio = 1;
| ~~~~~~~~~~~~~~^~~
In file included from mm/swapfile.c:16:
./include/linux/swap.h:292:27: note: while referencing ‘avail_lists’
292 | struct plist_node avail_lists[]; /*
| ^~~~~~~~~~~
We should not be including unused smb20 specific code when legacy
support is disabled (CONFIG_CIFS_ALLOW_INSECURE_LEGACY turned
off). For example smb2_operations and smb2_values aren't used
in that case. Over time we can move more and more SMB1/CIFS and SMB2.0
code into the insecure legacy ifdefs
Reviewed-by: Ronnie Sahlberg <lsahlber@redhat.com> Signed-off-by: Steve French <stfrench@microsoft.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
With large and many guest with storage keys it is possible to create
large latencies or stalls during initial key setting:
rcu: INFO: rcu_sched self-detected stall on CPU
rcu: 18-....: (2099 ticks this GP) idle=54e/1/0x4000000000000002 softirq=35598716/35598716 fqs=998
(t=2100 jiffies g=155867385 q=20879)
Task dump for CPU 18:
CPU 1/KVM R running task 0 1030947 256019 0x06000004
Call Trace:
sched_show_task
rcu_dump_cpu_stacks
rcu_sched_clock_irq
update_process_times
tick_sched_handle
tick_sched_timer
__hrtimer_run_queues
hrtimer_interrupt
do_IRQ
ext_int_handler
ptep_zap_key
The mmap lock is held during the page walking but since this is a
semaphore scheduling is still possible. Same for the kvm srcu.
To minimize overhead do this on every segment table entry or large page.
Signed-off-by: Christian Borntraeger <borntraeger@linux.ibm.com> Reviewed-by: Alexander Gordeev <agordeev@linux.ibm.com> Reviewed-by: Claudio Imbrenda <imbrenda@linux.ibm.com> Link: https://lore.kernel.org/r/20220530092706.11637-2-borntraeger@linux.ibm.com Signed-off-by: Christian Borntraeger <borntraeger@linux.ibm.com> Signed-off-by: Heiko Carstens <hca@linux.ibm.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
"qemu-ndb -d" will call ioctl 'NBD_DISCONNECT' first, however, following
message was found:
block nbd0: Send disconnect failed -32
Which indicate that something is wrong with the server. Then,
"qemu-nbd -d" will call ioctl 'NBD_CLEAR_SOCK', however ioctl can't clear
requests after commit 2516ab1543fd("nbd: only clear the queue on device
teardown"). And in the meantime, request can't complete through timeout
because nbd_xmit_timeout() will always return 'BLK_EH_RESET_TIMER', which
means such request will never be completed in this situation.
Now that the flag 'NBD_CMD_INFLIGHT' can make sure requests won't
complete multiple times, switch back to call nbd_clear_sock() in
nbd_clear_sock_ioctl(), so that inflight requests can be cleared.
When nbd module is being removing, nbd_alloc_config() may be
called concurrently by nbd_genl_connect(), although try_module_get()
will return false, but nbd_alloc_config() doesn't handle it.
The race may lead to the leak of nbd_config and its related
resources (e.g, recv_workq) and oops in nbd_read_stat() due
to the unload of nbd module as shown below:
Fixing it by checking the return value of try_module_get()
in nbd_alloc_config(). As nbd_alloc_config() may return ERR_PTR(-ENODEV),
assign nbd->config only when nbd_alloc_config() succeeds to ensure
the value of nbd->config is binary (valid or NULL).
Also adding a debug message to check the reference counter
of nbd_config during module removal.
As x86 uses the <asm-generic/bitops/instrumented-*.h> headers, the
regular forms of all bitops are instrumented with explicit calls to
KASAN and KCSAN checks. As these are explicit calls, these are not
suppressed by the noinstr function attribute.
This can result in calls to those check functions in noinstr code, which
objtool warns about:
vmlinux.o: warning: objtool: enter_from_user_mode+0x24: call to __kcsan_check_access() leaves .noinstr.text section
vmlinux.o: warning: objtool: syscall_enter_from_user_mode+0x28: call to __kcsan_check_access() leaves .noinstr.text section
vmlinux.o: warning: objtool: syscall_enter_from_user_mode_prepare+0x24: call to __kcsan_check_access() leaves .noinstr.text section
vmlinux.o: warning: objtool: irqentry_enter_from_user_mode+0x24: call to __kcsan_check_access() leaves .noinstr.text section
Prevent this by using the arch_*() bitops, which are the underlying
bitops without explciit instrumentation.
The return value of is_arm_mapping_symbol() is unpredictable when "$"
is passed in.
strchr(3) says:
The strchr() and strrchr() functions return a pointer to the matched
character or NULL if the character is not found. The terminating null
byte is considered part of the string, so that if c is specified as
'\0', these functions return a pointer to the terminator.
When str[1] is '\0', strchr("axtd", str[1]) is not NULL, and str[2] is
referenced (i.e. buffer overrun).
In radeon_fp_native_mode(), the return value of drm_mode_duplicate()
is assigned to mode, which will lead to a NULL pointer dereference
on failure of drm_mode_duplicate(). Add a check to avoid npd.
The failure status of drm_cvt_mode() on the other path is checked too.
Signed-off-by: Gong Yuanjun <ruc_gongyuanjun@163.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
`rctime' has been a pain point in cephfs due to its buggy
nature - inconsistent values reported and those sorts.
Fixing rctime is non-trivial needing an overall redesign
of the entire nested statistics infrastructure.
As a workaround, PR
http://github.com/ceph/ceph/pull/37938
allows this extended attribute to be manually set. This allows
users to "fixup" inconsistent rctime values. While this sounds
messy, its probably the wisest approach allowing users/scripts
to workaround buggy rctime values.
The above PR enables Ceph MDS to allow manually setting
rctime extended attribute with the corresponding user-land
changes. We may as well allow the same to be done via kclient
for parity.
A non-zero return value from pfkey_broadcast() does not necessarily mean
an error occurred as this function returns -ESRCH when no registered
listener received the message. In particular, a call with
BROADCAST_PROMISC_ONLY flag and null one_sk argument can never return
zero so that this commit in fact prevents processing any PF_KEY message.
One visible effect is that racoon daemon fails to find encryption
algorithms like aes and refuses to start.
Excluding -ESRCH return value would fix this but it's not obvious that
we really want to bail out here and most other callers of
pfkey_broadcast() also ignore the return value. Also, as pointed out by
Steffen Klassert, PF_KEY is kind of deprecated and newer userspace code
should use netlink instead so that we should only disturb the code for
really important fixes.
v2: add a comment explaining why is the return value ignored
When myrb_probe() fails the callback might not be set, so we need to
validate the 'disable_intr' callback in myrb_cleanup() to not cause a null
pointer exception. And while at it do not call myrb_cleanup() if we cannot
enable the PCI device at all.
Link: https://lore.kernel.org/r/20220523120244.99515-1-hare@suse.de Reported-by: Zheyu Ma <zheyuma97@gmail.com> Tested-by: Zheyu Ma <zheyuma97@gmail.com> Signed-off-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Generally, the md_unregister_thread is called with reconfig_mutex, but
raid_message in dm-raid doesn't hold reconfig_mutex to unregister thread,
so md_unregister_thread can be called simulitaneously from two call sites
in theory.
Then after previous commit which remove the protection of reconfig_mutex
for md_unregister_thread completely, the potential issue could be worse
than before.
Let's take pers_lock at the beginning of function to ensure reentrancy.
Reported-by: Donald Buczek <buczek@molgen.mpg.de> Signed-off-by: Guoqing Jiang <guoqing.jiang@linux.dev> Signed-off-by: Song Liu <song@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Executing reboot command several times on the machine "Dell
PowerEdge R740", UEFI security detection stopped machine
with the following prompt:
UEFI0082: The system was reset due to a timeout from the watchdog
timer. Check the System Event Log (SEL) or crash dumps from
Operating Sysstem to identify the source that triggered the
watchdog timer reset. Update the firmware or driver for the
identified device.
iDRAC has warning event: "The watchdog timer reset the system".
This patch fixes this issue by adding the reboot notifier.
Previously the protection of kernfs_pr_cont_buf was piggy backed by
rename_lock, which means that pr_cont() needs to be protected under
rename_lock. This can cause potential circular lock dependencies.
If there is an OOM, we have the following call hierarchy:
pr_cont_kernfs_name() will grab rename_lock and call printk. So we have
the following lock dependencies:
kernfs_rename_lock -> console_sem
Sometimes, printk does a wakeup before releasing console_sem, which has
the dependence chain:
console_sem -> p->pi_lock -> rq->lock
Now, imagine one wants to read cgroup_name under rq->lock, for example,
printing cgroup_name in a tracepoint in the scheduler code. They will
be holding rq->lock and take rename_lock:
rq->lock -> kernfs_rename_lock
Now they will deadlock.
A prevention to this circular lock dependency is to separate the
protection of pr_cont_buf from rename_lock. In principle, rename_lock
is to protect the integrity of cgroup name when copying to buf. Once
pr_cont_buf has got its content, rename_lock can be dropped. So it's
safe to drop rename_lock after kernfs_name_locked (and
kernfs_path_from_node_locked) and rely on a dedicated pr_cont_lock
to protect pr_cont_buf.
__msm_console_write() assumes that interrupts are disabled, but
with threaded console printers it is possible that the write()
callback of the console is called with interrupts enabled.
Explicitly disable interrupts using local_irq_save() to preserve
the assumed context.
Reported-by: Marek Szyprowski <m.szyprowski@samsung.com> Reviewed-by: Petr Mladek <pmladek@suse.com> Signed-off-by: John Ogness <john.ogness@linutronix.de> Link: https://lore.kernel.org/r/20220506213324.470461-1-john.ogness@linutronix.de Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Local variable mac created at:
r871xu_drv_init+0x1771/0x3070 drivers/staging/rtl8712/usb_intf.c:394
usb_probe_interface+0xf19/0x1600 drivers/usb/core/driver.c:396
Local variable data created at:
usb_read8+0x5d/0x130 drivers/staging/rtl8712/usb_ops.c:33
r8712_read8+0xa5/0xd0 drivers/staging/rtl8712/rtl8712_io.c:29
When a machine sports more than one SP804 timer instance, we only bring
up the first one, since multiple timers of the same kind are not useful
to Linux. As this is intentional behaviour, we should not return an
error message, as we do today:
===============
[ 0.000800] Failed to initialize '/bus@8000000/motherboard-bus@8000000/iofpga-bus@300000000/timer@120000': -22
===============
Replace the -EINVAL return with a debug message and return 0 instead.
Also we do not reach the init function anymore if the DT node is
disabled (as this is now handled by OF_DECLARE), so remove the explicit
check for that case.
This fixes a long standing bogus error when booting ARM's fastmodels.
Currently, someone can invoke the sysfs such as state_show()
intermittently before dev_set_drvdata() is done.
And it can be a cause of kernel Oops because of edev is Null at that time.
So modified the driver registration to after setting drviver data.
rtsx_usb_probe() doesn't call usb_set_intfdata() to null out the
interface pointer when probe fails. This leaves a stale pointer.
Noticed the missing usb_set_intfdata() while debugging an unrelated
invalid DMA mapping problem.
Fix it with a call to usb_set_intfdata(..., NULL).
UDC driver should not touch gadget's driver internals, especially it
should not reset driver->bus. This wasn't harmful so far, but since
commit fc274c1e9973 ("USB: gadget: Add a new bus for gadgets") gadget
subsystem got it's own bus and messing with ->bus triggers the
following NULL pointer dereference:
dwc2 12480000.hsotg: bound driver g_ether
8<--- cut here ---
Unable to handle kernel NULL pointer dereference at virtual address 00000000
[00000000] *pgd=00000000
Internal error: Oops: 5 [#1] SMP ARM
Modules linked in: ...
CPU: 0 PID: 620 Comm: modprobe Not tainted 5.18.0-rc5-next-20220504 #11862
Hardware name: Samsung Exynos (Flattened Device Tree)
PC is at module_add_driver+0x44/0xe8
LR is at sysfs_do_create_link_sd+0x84/0xe0
...
Process modprobe (pid: 620, stack limit = 0x(ptrval))
...
module_add_driver from bus_add_driver+0xf4/0x1e4
bus_add_driver from driver_register+0x78/0x10c
driver_register from usb_gadget_register_driver_owner+0x40/0xb4
usb_gadget_register_driver_owner from do_one_initcall+0x44/0x1e0
do_one_initcall from do_init_module+0x44/0x1c8
do_init_module from load_module+0x19b8/0x1b9c
load_module from sys_finit_module+0xdc/0xfc
sys_finit_module from ret_fast_syscall+0x0/0x54
Exception stack(0xf1771fa8 to 0xf1771ff0)
...
dwc2 12480000.hsotg: new device is high-speed
---[ end trace 0000000000000000 ]---
The backtrace of current CPU also should be printed as it is active. This
change add stack trace for current CPU and print a hint for idle CPU for
the generic workqueue based printing. (x86 already does this)
The documentation for the freeze() method says that it "should quiesce
the device so that it doesn't generate IRQs or DMA". The unspoken
consequence of not doing this is that MSIs aimed at non-boot CPUs may
get fully lost if they're sent during the period where the target CPU is
offline.
The current callbacks for USB HCD do not fully quiesce interrupts,
specifically on XHCI. Change to use the full suspend/resume flow for
freeze/thaw to ensure interrupts are fully quiesced. This fixes issues
where USB devices fail to thaw during hibernation because XHCI misses
its interrupt and cannot recover.
We hold oxu->lock in position (1) of thread 1, and use
del_timer_sync() to wait timer to stop, but timer handler
also need oxu->lock in position (2) of thread 2. As a result,
oxu_bus_suspend() will block forever.
This patch extracts del_timer_sync() from the protection of
spin_lock_irq(), which could let timer handler to obtain
the needed lock.
We hold sport->port.lock in position (1) of thread 1 and
use del_timer_sync() to wait timer to stop, but timer handler
also need sport->port.lock in position (2) of thread 2. As a result,
sa1100_set_termios() will block forever.
This patch moves del_timer_sync() before spin_lock_irqsave()
in order to prevent the deadlock.
We hold ieee->beacon_lock in position (1) of thread 1 and
use del_timer_sync() to wait timer to stop, but timer handler
also need ieee->beacon_lock in position (2) of thread 2.
As a result, rtllib_beacons_stop() will block forever.
This patch extracts del_timer_sync() from the protection of
spin_lock_irqsave(), which could let timer handler to obtain
the needed lock.
We hold ieee->beacon_lock in position (1) of thread 1 and use
del_timer_sync() to wait timer to stop, but timer handler
also need ieee->beacon_lock in position (2) of thread 2.
As a result, ieee80211_beacons_stop() will block forever.
This patch extracts del_timer_sync() from the protection of
spin_lock_irqsave(), which could let timer handler to obtain
the needed lock.
To be sufficiently out of range for the usercopy test to see the lifetime
mismatch, expand the size of the "bad" buffer, which will let it be
beyond current_stack_pointer regardless of stack growth direction.
Paired with the recent addition of stack depth checking under
CONFIG_HARDENED_USERCOPY=y, this will correctly start tripping again.
Right now the (framework) mlock lock is (ab)used for multiple purposes:
1- protecting concurrent accesses over the odr local cache
2- avoid changing samplig frequency whilst buffer is running
Let's start by handling situation #1 with a local lock.
In r871xu_drv_init(), if r8712_init_drv_sw() fails, then the memory
allocated by r8712_alloc_io_queue() in r8712_usb_dvobj_init() is not
properly released as there is no action will be performed by
r8712_usb_dvobj_deinit().
To properly release it, we should call r8712_free_io_queue() in
r8712_usb_dvobj_deinit().
Besides, in r871xu_dev_remove(), r8712_usb_dvobj_deinit() will be called
by r871x_dev_unload() under condition `padapter->bup` and
r8712_free_io_queue() is called by r8712_free_drv_sw().
However, r8712_usb_dvobj_deinit() does not rely on `padapter->bup` and
calling r8712_free_io_queue() in r8712_free_drv_sw() is negative for
better understading the code.
So I move r8712_usb_dvobj_deinit() into r871xu_dev_remove(), and remove
r8712_free_io_queue() from r8712_free_drv_sw().
kstrdup() is also a memory allocation-related function, it returns NULL
when some memory errors happen. So it is better to check the return
value of it so to catch the memory error in time. Besides, there should
have a kfree() to clear up the allocation if we get a failure later in
this function to prevent memory leak.
Gcc-12 correctly warned about this code using a non-NULL pointer as a
truth value:
drivers/gpu/drm/imx/ipuv3-crtc.c: In function ‘ipu_crtc_disable_planes’:
drivers/gpu/drm/imx/ipuv3-crtc.c:72:21: error: the comparison will always evaluate as ‘true’ for the address of ‘plane’ will never be NULL [-Werror=address]
72 | if (&ipu_crtc->plane[1] && plane == &ipu_crtc->plane[1]->base)
| ^
due to the extraneous '&' address-of operator.
Philipp Zabel points out that The mistake had no adverse effect since
the following condition doesn't actually dereference the NULL pointer,
but the intent of the code was obviously to check for it, not to take
the address of the member.
Every iteration of for_each_child_of_node() decrements
the reference count of the previous node.
When break from a for_each_child_of_node() loop,
we need to explicitly call of_node_put() on the child node when
not need anymore.
Add missing of_node_put() to avoid refcount leak.
Fixes: bbd2190ce96d ("Altera TSE: Add main and header file for Altera Ethernet Driver") Signed-off-by: Miaoqian Lin <linmq006@gmail.com> Link: https://lore.kernel.org/r/20220607041144.7553-1-linmq006@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
But was eventually revised more thoroughly:
- restrict the check to the only branch where needed, in an
uncommon GRE path that uses header_ops and calls skb_pull.
- test skb_transport_header, which is set along with csum_start
in skb_partial_csum_set in the normal header_ops datapath.
Turns out skbs can arrive in this branch without the transport
header set, e.g., through BPF redirection.
Revise the check back to check csum_start directly, and only if
CHECKSUM_PARTIAL. Do leave the check in the updated location.
Check field regardless of whether TUNNEL_CSUM is configured.
When combining two steering rules into one check
not only do they share the same actions but those
actions are also the same. This resolves an issue where
when creating two different rules with the same match
the actions are overwritten and one of the rules is deleted
a FW syndrome can be seen in dmesg.
mlx5_core 0000:03:00.0: mlx5_cmd_check:819:(pid 2105): DEALLOC_MODIFY_HEADER_CONTEXT(0x941) op_mod(0x0) failed, status bad resource state(0x9), syndrome (0x1ab444)
Fixes: 0d235c3fabb7 ("net/mlx5: Add hash table to search FTEs in a flow-group") Signed-off-by: Mark Bloch <mbloch@nvidia.com> Reviewed-by: Maor Gottlieb <maorg@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
The current design does not arm the tracer if traces are available before
the tracer string database is fully loaded, leading to an unfunctional tracer.
This fix will rearm the tracer every time the FW triggers tracer event
regardless of the tracer strings database status.
EXPORT_SYMBOL and __init is a bad combination because the .init.text
section is freed up after the initialization. Hence, modules cannot
use symbols annotated __init. The access to a freed symbol may end up
with kernel panic.
modpost used to detect it, but it has been broken for a decade.
Recently, I fixed modpost so it started to warn it again, then this
showed up in linux-next builds.
There are two ways to fix it:
- Remove __init
- Remove EXPORT_SYMBOL
I chose the latter for this case because the caller (net/ipv6/seg6.c)
and the callee (net/ipv6/seg6_hmac.c) belong to the same module.
It seems an internal function call in ipv6.ko.
Fixes: bf355b8d2c30 ("ipv6: sr: add core files for SR HMAC support") Reported-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Masahiro Yamada <masahiroy@kernel.org> Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
EXPORT_SYMBOL and __init is a bad combination because the .init.text
section is freed up after the initialization. Hence, modules cannot
use symbols annotated __init. The access to a freed symbol may end up
with kernel panic.
modpost used to detect it, but it has been broken for a decade.
Recently, I fixed modpost so it started to warn it again, then this
showed up in linux-next builds.
There are two ways to fix it:
- Remove __init
- Remove EXPORT_SYMBOL
I chose the latter for this case because the only in-tree call-site,
net/ipv4/xfrm4_policy.c is never compiled as modular.
(CONFIG_XFRM is boolean)
Fixes: 2f32b51b609f ("xfrm: Introduce xfrm_input_afinfo to access the the callbacks properly") Reported-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Masahiro Yamada <masahiroy@kernel.org> Acked-by: Steffen Klassert <steffen.klassert@secunet.com> Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
EXPORT_SYMBOL and __init is a bad combination because the .init.text
section is freed up after the initialization. Hence, modules cannot
use symbols annotated __init. The access to a freed symbol may end up
with kernel panic.
modpost used to detect it, but it has been broken for a decade.
Recently, I fixed modpost so it started to warn it again, then this
showed up in linux-next builds.
There are two ways to fix it:
- Remove __init
- Remove EXPORT_SYMBOL
I chose the latter for this case because the only in-tree call-site,
drivers/net/phy/phy_device.c is never compiled as modular.
(CONFIG_PHYLIB is boolean)
Fixes: 90eff9096c01 ("net: phy: Allow splitting MDIO bus/device support from PHYs") Reported-by: Stephen Rothwell <sfr@canb.auug.org.au> Signed-off-by: Masahiro Yamada <masahiroy@kernel.org> Reviewed-by: Florian Fainelli <f.fainelli@gmail.com> Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk> Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
I found that NFSD's new NFSv3 READDIRPLUS XDR encoder was screwing up
right at the end of the page array. xdr_get_next_encode_buffer() does
not compute the value of xdr->end correctly:
* The check to see if we're on the final available page in xdr->buf
needs to account for the space consumed by @nbytes.
* The new xdr->end value needs to account for the portion of @nbytes
that is to be encoded into the previous buffer.
Fixes: 2825a7f90753 ("nfsd4: allow encoding across page boundaries") Signed-off-by: Chuck Lever <chuck.lever@oracle.com> Reviewed-by: NeilBrown <neilb@suse.de> Reviewed-by: J. Bruce Fields <bfields@fieldses.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Every iteration of for_each_available_child_of_node() decrements
the reference count of the previous node.
when breaking early from a for_each_available_child_of_node() loop,
we need to explicitly call of_node_put() on the gphy_fw_np.
Add missing of_node_put() to avoid refcount leak.
Fixes: 14fceff4771e ("net: dsa: Add Lantiq / Intel DSA driver for vrx200") Signed-off-by: Miaoqian Lin <linmq006@gmail.com> Link: https://lore.kernel.org/r/20220605072335.11257-1-linmq006@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
syzbot reported an illegal copy_to_user() attempt
from bpf_prog_get_info_by_fd() [1]
There was no repro yet on this bug, but I think
that commit 0aef499f3172 ("mm/usercopy: Detect vmalloc overruns")
is exposing a prior bug in bpf arm64.
bpf_prog_get_info_by_fd() looks at prog->jited_len
to determine if the JIT image can be copied out to user space.
My theory is that syzbot managed to get a prog where prog->jited_len
has been set to 43, while prog->bpf_func has ben cleared.
It is not clear why copy_to_user(uinsns, NULL, ulen) is triggering
this particular warning.
I thought find_vma_area(NULL) would not find a vm_struct.
As we do not hold vmap_area_lock spinlock, it might be possible
that the found vm_struct was garbage.
unix_dgram_poll() calls unix_dgram_peer_wake_me() without `other`'s
lock held and check if its receive queue is full. Here we need to
use unix_recvq_full_lockless() instead of unix_recvq_full(), otherwise
KCSAN will report a data-race.
EXPORT_SYMBOL and __init is a bad combination because the .init.text
section is freed up after the initialization. Hence, modules cannot
use symbols annotated __init. The access to a freed symbol may end up
with kernel panic.
modpost used to detect it, but it has been broken for a decade.
Recently, I fixed modpost so it started to warn it again, then this
showed up in linux-next builds.
There are two ways to fix it:
- Remove __init
- Remove EXPORT_SYMBOL
I chose the latter for this case because none of the in-tree call-sites
(arch/arm/xen/enlighten.c, arch/x86/xen/grant-table.c) is compiled as
modular.
If user requests for NFT_CHAIN_HW_OFFLOAD, then check if either device
provides the .ndo_setup_tc interface or there is an indirect flow block
that has been registered. Otherwise, bail out early from the preparation
phase. Moreover, validate that family == NFPROTO_NETDEV and hook is
NF_NETDEV_INGRESS.
of_find_device_by_node() takes reference, we should use put_device()
to release it when not need anymore.
Add missing put_device() to avoid refcount leak.
Fixes: 43f01da0f279 ("MIPS/OCTEON/ata: Convert pata_octeon_cf.c to use device tree.") Signed-off-by: Miaoqian Lin <linmq006@gmail.com> Reviewed-by: Sergey Shtylyov <s.shtylyov@omp.ru> Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Remove inactive bool field in nft_hook object that was introduced in abadb2f865d7 ("netfilter: nf_tables: delete devices from flowtable").
Move stale flowtable hooks to transaction list instead.
Deleting twice the same device does not result in ENOENT.
Fixes: abadb2f865d7 ("netfilter: nf_tables: delete devices from flowtable") Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
In pxa3xx_gcu_probe(), the sequence of error lable is wrong, it will
leads some resource leaked, so adjust the sequence to handle the error
correctly, and if pxa3xx_gcu_add_buffer() fails, pxa3xx_gcu_free_buffers()
need be called.
In pxa3xx_gcu_remove(), add missing clk_disable_unpreprare().
This patch fixes a bug where GEN1 VMs doesn't allow resolutions greater
than 64 MB size (eg 7680x4320). Unnecessary PCI check limits Gen1 VRAM
to legacy PCI BAR size only (ie 64MB). Thus any, resolution requesting
greater then 64MB (eg 7680x4320) would fail. MMIO region assigning this
memory shouldn't be limited by PCI bar size.
When doing layoutget as part of the open() compound, we have to be
careful to release the layout locks before we can call any further RPC
calls, such as setattr(). The reason is that those calls could trigger
a recall, which could deadlock.
The arguments desc_size and ZYNQMP_DMA_NUM_DESCS were 32 bit. Though
this overflow condition is not observed but it is a potential problem
in the case of 32-bit multiplication. Hence fix it by changing the
desc_size data type to size_t.
In addition to coverity fix it also reuse ZYNQMP_DMA_DESC_SIZE macro in
dma_alloc_coherent API argument.
When configuring a nommu classic m68k system enabling the uboot parameter
passing support (CONFIG_UBOOT) will produce the following compile error:
m68k-linux-ld: arch/m68k/kernel/uboot.o: in function `process_uboot_commandline':
uboot.c:(.init.text+0x32): undefined reference to `_init_sp'
The logic to support this option is only used on ColdFire based platforms
(in its head.S startup code). So make the selection of this option
depend on building for a ColdFire based platform.
The non-MMU m68k pagetable ZERO_PAGE() macro is being set to the
somewhat non-sensical value of "virt_to_page(0)". The zeroth page
is not in any way guaranteed to be a page full of "0". So the result
is that ZERO_PAGE() will almost certainly contain random values.
We already allocate a real "empty_zero_page" in the mm setup code shared
between MMU m68k and non-MMU m68k. It is just not hooked up to the
ZERO_PAGE() macro for the non-MMU m68k case.
Fix ZERO_PAGE() to use the allocated "empty_zero_page" pointer.
I am not aware of any specific issues caused by the old code.
Timeout as 1 second sets an upper limit on the length
of the transfer executed, but there is no maximum length
of a write or read message set in i2c_adapter_quirks for
this controller.
This upper limit affects devices that require sending
large firmware blobs over I2C.
To remove that limitation, calculate the minimal time
necessary, plus some wiggle room, for every message and
use it instead of the default one second, if more than
one second.
Signed-off-by: Lucas Tanure <tanureal@opensource.cirrus.com> Acked-by: Michal Simek <michal.simek@xilinx.com> Signed-off-by: Wolfram Sang <wsa@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Syzbot triggers two WARNs in f2fs_is_valid_blkaddr and
__is_bitmap_valid. For example, in f2fs_is_valid_blkaddr,
if type is DATA_GENERIC_ENHANCE or DATA_GENERIC_ENHANCE_READ,
it invokes WARN_ON if blkaddr is not in the right range.
The call trace is as follows:
It will cause null-ptr-deref when using 'res', if platform_get_resource()
returns NULL, so move using 'res' after devm_ioremap_resource() that
will check it to avoid null-ptr-deref.
And use devm_platform_get_and_ioremap_resource() to simplify code.
To prepare for support asynchronous tracer_init_tracefs initcall,
avoid calling create_trace_option_files before __update_tracer_options.
Otherwise, create_trace_option_files will show warning because
some tracers in trace_types list are already in tr->topts.
For example, hwlat_tracer call register_tracer in late_initcall,
and global_trace.dir is already created in tracing_init_dentry,
hwlat_tracer will be put into tr->topts.
Then if the __update_tracer_options is executed after hwlat_tracer
registered, create_trace_option_files find that hwlat_tracer is
already in tr->topts.
When setting bootparams="trace_event=initcall:initcall_start tp_printk=1" in the
cmdline, the output_printk() was called, and the spin_lock_irqsave() was called in the
atomic and irq disable interrupt context suitation. On the PREEMPT_RT kernel,
these locks are replaced with sleepable rt-spinlock, so the stack calltrace will
be triggered.
Fix it by raw_spin_lock_irqsave when PREEMPT_RT and "trace_event=initcall:initcall_start
tp_printk=1" enabled.
Even though idxd driver has always supported interrupt, it never actually
set the DMA_INTERRUPT cap bit. Rectify this mistake so the interrupt
capability is advertised.
The function percent_rmt_hitm_cmp() wrongly uses local HITMs for
sorting remote HITMs.
Since this function is to sort cache lines for remote HITMs, this patch
changes to use 'rmt_hitm' field for correct sorting.
Fixes: 9cb3500afc0980c5 ("perf c2c report: Add hitm/store percent related sort keys") Signed-off-by: Leo Yan <leo.yan@linaro.org> Acked-by: Namhyung Kim <namhyung@kernel.org> Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Jiri Olsa <jolsa@kernel.org> Cc: Joe Mario <jmario@redhat.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Peter Zijlstra <peterz@infradead.org> Link: https://lore.kernel.org/r/20220530084253.750190-1-leo.yan@linaro.org Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
Mounting NFS rootfs was timing out when deferred_probe_timeout was
non-zero [1]. This was because ip_auto_config() initcall times out
waiting for the network interfaces to show up when
deferred_probe_timeout was non-zero. While ip_auto_config() calls
wait_for_device_probe() to make sure any currently running deferred
probe work or asynchronous probe finishes, that wasn't sufficient to
account for devices being deferred until deferred_probe_timeout.
Commit 35a672363ab3 ("driver core: Ensure wait_for_device_probe() waits
until the deferred_probe_timeout fires") tried to fix that by making
sure wait_for_device_probe() waits for deferred_probe_timeout to expire
before returning.
However, if wait_for_device_probe() is called from the kernel_init()
context:
- Before deferred_probe_initcall() [2], it causes the boot process to
hang due to a deadlock.
- After deferred_probe_initcall() [3], it blocks kernel_init() from
continuing till deferred_probe_timeout expires and beats the point of
deferred_probe_timeout that's trying to wait for userspace to load
modules.
Neither of this is good. So revert the changes to
wait_for_device_probe().
If sd_probe() sees an early error before sdkp->device is initialized,
sd_zbc_release_disk() is called. This causes a NULL pointer dereference
when sd_is_zoned() is called inside that function. Avoid this by removing
the call to sd_zbc_release_disk() in sd_probe() error path.
This change is safe and does not result in zone information memory leakage
because the zone information for a zoned disk is allocated only when
sd_revalidate_disk() is called, at which point sdkp->disk_dev is fully set,
resulting in sd_disk_release() being called when needed to cleanup a disk
zone information using sd_zbc_release_disk().
Link: https://lore.kernel.org/r/20220601062544.905141-2-damien.lemoal@opensource.wdc.com Fixes: 89d947561077 ("sd: Implement support for ZBC devices") Reported-by: Dongliang Mu <mudongliangabcd@gmail.com> Suggested-by: Christoph Hellwig <hch@lst.de> Reviewed-by: Christoph Hellwig <hch@lst.de> Signed-off-by: Damien Le Moal <damien.lemoal@opensource.wdc.com> Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
In AFS, a directory is handled as a file that the client downloads and
parses locally for the purposes of performing lookup and getdents
operations. The in-kernel afs filesystem has a number of functions that
do this.
A directory file is arranged as a series of 2K blocks divided into
32-byte slots, where a directory entry occupies one or more slots, plus
each block starts with one or more metadata blocks.
When parsing a block, if the last slots are occupied by a dirent that
occupies more than a single slot and the file position points at a slot
that's not the initial one, the logic in afs_dir_iterate_block() that
skips over it won't advance the file pointer to the end of it. This
will cause an infinite loop in getdents() as it will keep retrying that
block and failing to advance beyond the final entry.
Fix this by advancing the file pointer if the next entry will be beyond
it when we skip a block.
This was found by the generic/676 xfstest but can also be triggered with
something like:
1) A new passive FastOpen TCP socket is created.
This FO socket waits for an ACK coming from client to be a complete
ESTABLISHED one.
2) A socket operation on this socket goes through lock_sock()
release_sock() dance.
3) While the socket is owned by the user in step 2),
a retransmit of the SYN is received and stored in socket backlog.
4) At release_sock() time, the socket backlog is processed while
in process context.
5) A SYNACK packet is cooked in response of the SYN retransmit.
6) -> tcp_rtx_synack() is called in process context.
Before blamed commit, tcp_rtx_synack() was always called from BH handler,
from a timer handler.
Fix this by using TCP_INC_STATS() & NET_INC_STATS()
which do not assume caller is in non preemptible context.
In qdisc_run_end(), the spin_unlock() only has store-release semantic,
which guarantees all earlier memory access are visible before it. But
the subsequent test_bit() has no barrier semantics so may be reordered
ahead of the spin_unlock(). The store-load reordering may cause a packet
stuck problem.
The concurrent operations can be described as below,
CPU 0 | CPU 1
qdisc_run_end() | qdisc_run_begin()
. | .
----> /* may be reorderd here */ | .
| . | .
| spin_unlock() | set_bit()
| . | smp_mb__after_atomic()
---- test_bit() | spin_trylock()
. | .
Consider the following sequence of events:
CPU 0 reorder test_bit() ahead and see MISSED = 0
CPU 1 calls set_bit()
CPU 1 calls spin_trylock() and return fail
CPU 0 executes spin_unlock()
At the end of the sequence, CPU 0 calls spin_unlock() and does nothing
because it see MISSED = 0. The skb on CPU 1 has beed enqueued but no one
take it, until the next cpu pushing to the qdisc (if ever ...) will
notice and dequeue it.
This patch fix this by adding one explicit barrier. As spin_unlock() and
test_bit() ordering is a store-load ordering, a full memory barrier
smp_mb() is needed here.
Fixes: a90c57f2cedd ("net: sched: fix packet stuck problem for lockless qdisc") Signed-off-by: Guoju Fang <gjfang@linux.alibaba.com> Link: https://lore.kernel.org/r/20220528101628.120193-1-gjfang@linux.alibaba.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
Some features (LRO, HW GRO) conflict with XDP. If there is an attempt to
enable such features while XDP is active, they will be set to `off
[requested on]`. In order to activate these features after XDP is turned
off, the driver needs to call netdev_update_features(). This commit adds
this missing call after XDP state changes.
Fixes: cf6e34c8c22f ("net/mlx5e: Properly block LRO when XDP is enabled") Fixes: b0617e7b3500 ("net/mlx5e: Properly block HW GRO when XDP is enabled") Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com> Reviewed-by: Tariq Toukan <tariqt@nvidia.com> Signed-off-by: Saeed Mahameed <saeedm@nvidia.com> Signed-off-by: Sasha Levin <sashal@kernel.org>
tx_channel_offset is calculated in efx_allocate_msix_channels, but it is
also calculated again in efx_set_channels because it was originally done
there, and when efx_allocate_msix_channels was introduced it was
forgotten to be removed from efx_set_channels.
Moreover, the old calculation is wrong when using
efx_separate_tx_channels because now we can have XDP channels after the
TX channels, so n_channels - n_tx_channels doesn't point to the first TX
channel.
Remove the old calculation from efx_set_channels, and add the
initialization of this variable if MSI or legacy interrupts are used,
next to the initialization of the rest of the related variables, where
it was missing.
Fixes: 3990a8fffbda ("sfc: allocate channels for XDP tx queues") Reported-by: Tianhao Zhao <tizhao@redhat.com> Signed-off-by: Íñigo Huguet <ihuguet@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
Normally, all channels have RX and TX queues, but this is not true if
modparam efx_separate_tx_channels=1 is used. In that cases, some
channels only have RX queues and others only TX queues (or more
preciselly, they have them allocated, but not initialized).
Fix efx_channel_has_tx_queues to return the correct value for this case
too.
Messages shown at probe time before the fix:
sfc 0000:03:00.0 ens6f0np0: MC command 0x82 inlen 544 failed rc=-22 (raw=0) arg=0
------------[ cut here ]------------
netdevice: ens6f0np0: failed to initialise TXQ -1
WARNING: CPU: 1 PID: 626 at drivers/net/ethernet/sfc/ef10.c:2393 efx_ef10_tx_init+0x201/0x300 [sfc]
[...] stripped
RIP: 0010:efx_ef10_tx_init+0x201/0x300 [sfc]
[...] stripped
Call Trace:
efx_init_tx_queue+0xaa/0xf0 [sfc]
efx_start_channels+0x49/0x120 [sfc]
efx_start_all+0x1f8/0x430 [sfc]
efx_net_open+0x5a/0xe0 [sfc]
__dev_open+0xd0/0x190
__dev_change_flags+0x1b3/0x220
dev_change_flags+0x21/0x60
[...] stripped
Messages shown at remove time before the fix:
sfc 0000:03:00.0 ens6f0np0: failed to flush 10 queues
sfc 0000:03:00.0 ens6f0np0: failed to flush queues
Fixes: 8700aff08984 ("sfc: fix channel allocation with brute force") Reported-by: Tianhao Zhao <tizhao@redhat.com> Signed-off-by: Martin Habets <habetsm.xilinx@gmail.com> Tested-by: Íñigo Huguet <ihuguet@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
"struct smc_cdc_tx_pend **" can not directly convert
to "struct smc_wr_tx_pend_priv *".
Fixes: 2bced6aefa3d ("net/smc: put slot when connection is killed") Signed-off-by: Guangguan Wang <guangguan.wang@linux.alibaba.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
I think there is something wrong with BPF_PROBE_MEM in ___bpf_prog_run()
in big-endian machine. Let's make a test and see what will happen if we
want to load a 'u16' with BPF_PROBE_MEM.
Let's make the src value '0x0001', the value of dest register will become
0x0001000000000000, as the value will be loaded to the first 2 byte of
DST with following code:
The code of the test case is simply, it will load sk->sk_family to the
register with BPF_PROBE_MEM and check if it is AF_INET6. With this patch,
now the test case 'bpf_iter' can pass:
$10 bpf_iter:OK
Fixes: 2a02759ef5f8 ("bpf: Add support for BTF pointers to interpreter") Signed-off-by: Menglong Dong <imagedong@tencent.com> Signed-off-by: Daniel Borkmann <daniel@iogearbox.net> Reviewed-by: Jiang Biao <benbjiang@tencent.com> Reviewed-by: Hao Peng <flyingpeng@tencent.com> Cc: Ilya Leoshkevich <iii@linux.ibm.com> Link: https://lore.kernel.org/bpf/20220524021228.533216-1-imagedong@tencent.com Signed-off-by: Sasha Levin <sashal@kernel.org>
There at least 6 PEBs reserved on UBI device:
1. EBA_RESERVED_PEBS[1]
2. WL_RESERVED_PEBS[1]
3. UBI_LAYOUT_VOLUME_EBS[2]
4. MIN_FASTMAP_RESERVED_PEBS[2]
When all ubi volumes take all their PEBs, there are 3 (EBA_RESERVED_PEBS +
WL_RESERVED_PEBS + MIN_FASTMAP_RESERVED_PEBS - MIN_FASTMAP_TAKEN_PEBS[1])
free PEBs. Since commit f9c34bb529975fe ("ubi: Fix producing anchor PEBs")
and commit 4b68bf9a69d22dd ("ubi: Select fastmap anchor PEBs considering
wear level rules") applied, there is only 1 (3 - FASTMAP_ANCHOR_PEBS[1] -
FASTMAP_NEXT_ANCHOR_PEBS[1]) free PEB to fill pool and wl_pool, after
filling pool, wl_pool is always empty. So, UBI could be stuck in an
infinite loop:
, which cause high cpu usage of ubi_bgt:
top - 12:10:42 up 5 min, 2 users, load average: 1.76, 0.68, 0.27
Tasks: 123 total, 3 running, 54 sleeping, 0 stopped, 0 zombie
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
1589 root 20 0 0 0 0 R 45.0 0.0 0:38.86 ubi_bgt0d
319 root 20 0 0 0 0 I 15.2 0.0 0:15.29 kworker/0:3-eve
371 root 20 0 0 0 0 I 14.9 0.0 0:12.85 kworker/3:3-eve
20 root 20 0 0 0 0 I 11.3 0.0 0:05.33 kworker/1:0-eve
202 root 20 0 0 0 0 I 11.3 0.0 0:04.93 kworker/2:3-eve
In commit 4b68bf9a69d22dd ("ubi: Select fastmap anchor PEBs considering
wear level rules"), there are three key changes:
1) Choose the fastmap anchor when the most free PEBs are available.
2) Enable anchor move within the anchor area again as it is useful
for distributing wear.
3) Import a candidate fm anchor and check this PEB's erase count during
wear leveling. If the wear leveling limit is exceeded, use the used
anchor area PEB with the lowest erase count to replace it.
The anchor candidate can be removed, we can check fm_anchor PEB's erase
count during wear leveling. Fix it by:
1) Removing 'fm_next_anchor' and check 'fm_anchor' during wear leveling.
2) Preferentially filling one free peb into fm_wl_pool in condition of
ubi->free_count > ubi->beb_rsvd_pebs, then try to reserve enough
free count for fastmap non anchor pebs after the above prerequisites
are met.
Then, there are at least 1 PEB in pool and 1 PEB in wl_pool after calling
ubi_refill_pools() with all erase works done.
With the `-z unique-symbol` linker flag or any similar mechanism,
it is possible to trigger the following:
ERROR: modpost: "param_set_uint.0" [vmlinux] is a static EXPORT_SYMBOL
The reason is that for now the condition from remove_dot():
if (m && (s[n + m] == '.' || s[n + m] == 0))
which was designed to test if it's a dot or a '\0' after the suffix
is never satisfied.
This is due to that `s[n + m]` always points to the last digit of a
numeric suffix, not on the symbol next to it (from a custom debug
print added to modpost):
param_set_uint.0, s[n + m] is '0', s[n + m + 1] is '\0'
So it's off-by-one and was like that since 2014.
Fix this for the sake of any potential upcoming features, but don't
bother stable-backporting, as it's well hidden -- apart from that
LD flag, it can be triggered only with GCC LTO which never landed
upstream.
of_get_child_by_name() returns a node pointer with refcount
incremented, we should use of_node_put() on it when done.
mv88e6xxx_mdio_register() pass the device node to of_mdiobus_register().
We don't need the device node after it.
Add missing of_node_put() to avoid refcount leak.
Fixes: a3c53be55c95 ("net: dsa: mv88e6xxx: Support multiple MDIO busses") Signed-off-by: Miaoqian Lin <linmq006@gmail.com> Reviewed-by: Marek Behún <kabel@kernel.org> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
of_get_child_by_name() returns a node pointer with refcount
incremented, we should use of_node_put() on it when not need anymore.
am65_cpsw_init_cpts() and am65_cpsw_nuss_probe() don't release
the refcount in error case.
Add missing of_node_put() to avoid refcount leak.
Fixes: b1f66a5bee07 ("net: ethernet: ti: am65-cpsw-nuss: enable packet timestamping support") Fixes: 93a76530316a ("net: ethernet: ti: introduce am65x/j721e gigabit eth subsystem driver") Signed-off-by: Miaoqian Lin <linmq006@gmail.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
The "fsp->location" variable comes from user via ethtool_get_rxnfc().
Check that it is valid to prevent an out of bounds read.
Fixes: 7aab747e5563 ("net: ethernet: mediatek: add ethtool functions to configure RX flows of HW LRO") Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com> Signed-off-by: David S. Miller <davem@davemloft.net> Signed-off-by: Sasha Levin <sashal@kernel.org>
In qdisc_run_begin(), smp_mb__before_atomic() used before test_bit()
does not provide any ordering guarantee as test_bit() is not an atomic
operation. This, added to the fact that the spin_trylock() call at
the beginning of qdisc_run_begin() does not guarantee acquire
semantics if it does not grab the lock, makes it possible for the
following statement :
if (test_bit(__QDISC_STATE_MISSED, &qdisc->state))
to be executed before an enqueue operation called before
qdisc_run_begin().
In the above scenario, CPU 1 and CPU 2 both try to grab the
qdisc->seqlock at the same time. Only CPU 2 succeeds and enters the
bypass code path, where it emits its skb then calls __qdisc_run().
CPU1 fails, sets MISSED and goes down the traditionnal enqueue() +
dequeue() code path. But when executing qdisc_run_begin() for the
second time, after enqueuing its skbuff, it sees the MISSED bit still
set (by itself) and consequently chooses to exit early without setting
it again nor trying to grab the spinlock again.
Meanwhile CPU2 has seen MISSED = 1, cleared it, checked the queue
and found it empty, so it returned.
At the end of the sequence, we end up with skb1 enqueued in the
backlog, both CPUs out of __dev_xmit_skb(), the MISSED bit not set,
and no __netif_schedule() called made. skb1 will now linger in the
qdisc until somebody later performs a full __qdisc_run(). Associated
to the bypass capacity of the qdisc, and the ability of the TCP layer
to avoid resending packets which it knows are still in the qdisc, this
can lead to serious traffic "holes" in a TCP connection.
We fix this by replacing the smp_mb__before_atomic() / test_bit() /
set_bit() / smp_mb__after_atomic() sequence inside qdisc_run_begin()
by a single test_and_set_bit() call, which is more concise and
enforces the needed memory barriers.
Fixes: 89837eb4b246 ("net: sched: add barrier to ensure correct ordering for lockless qdisc") Signed-off-by: Vincent Ray <vray@kalrayinc.com> Signed-off-by: Eric Dumazet <edumazet@google.com> Link: https://lore.kernel.org/r/20220526001746.2437669-1-eric.dumazet@gmail.com Signed-off-by: Jakub Kicinski <kuba@kernel.org> Signed-off-by: Sasha Levin <sashal@kernel.org>
The argument of scatterwalk_unmap() is supposed to be the void* that was
returned by the previous scatterwalk_map() call.
The s390 AES-GCM implementation was instead passing the pointer to the
struct scatter_walk.
This doesn't actually break anything because scatterwalk_unmap() only uses
its argument under CONFIG_HIGHMEM and ARCH_HAS_FLUSH_ON_KUNMAP.
projects / users / dwmw2 / linux.git / log