| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| In the Linux kernel, the following vulnerability has been resolved:
ALSA: timer: don't re-enter an instance callback that is still running
The userspace-driven timer (utimer) TRIGGER ioctl calls
snd_timer_interrupt() directly with no serialization, so two threads
triggering the same utimer can run snd_timer_interrupt() on one
snd_timer concurrently.
snd_timer_process_callbacks() drops timer->lock around each instance
callback and marks the in-flight callback with the single
SNDRV_TIMER_IFLG_CALLBACK bit; snd_timer_close_locked() waits on that
bit to drain an in-flight callback before freeing the instance. The bit
cannot represent two concurrent callbacks: when a second interrupt
re-queues an instance whose callback is still running, both run at once,
the first to finish clears the bit, and the close-path drain then frees
the instance (and its callback_data) while the other callback is still
live - a use-after-free reachable by any user able to open
/dev/snd/timer, both via a user timer instance and via a sequencer queue
timer bound to the utimer.
snd_timer_interrupt() sets IFLG_CALLBACK before dropping timer->lock, so
a concurrent interrupt already observes it under the lock. Skip
re-queuing an instance (and its slaves) to the ack/sack list while its
callback is in flight; the accumulated pticks are delivered on the next
tick, so no event is lost. |
| In the Linux kernel, the following vulnerability has been resolved:
Bluetooth: hci_sync: Protect UUID list traversal
The hci_sync conversion moved class-of-device and EIR generation from an
HCI request built under hdev->lock to asynchronous command sync work.
The worker holds hdev->req_lock, but that lock does not serialize access
to hdev->uuids against add_uuid() and remove_uuid(), which update the
list under hdev->lock.
The following interleaving can therefore occur:
CPU0 (command sync work) CPU1 (management socket)
fetch uuid from the list
list_del(&uuid->list)
kfree(uuid)
read uuid->size
KASAN reports the resulting use-after-free:
BUG: KASAN: slab-use-after-free in eir_create+0xb8f/0xee0
Read of size 1 at addr ffff88810dbd8620 by task kworker/u17:0/87
Workqueue: hci0 hci_cmd_sync_work
Call Trace:
eir_create+0xb8f/0xee0
hci_update_eir_sync+0x1c0/0x330
hci_cmd_sync_work+0x13c/0x290
process_one_work+0x63a/0x1070
worker_thread+0x45b/0xd10
Allocated by task 86:
__kasan_kmalloc+0x8f/0xa0
add_uuid+0x18a/0x4b0
hci_sock_sendmsg+0x1033/0x1ea0
Freed by task 92:
__kasan_slab_free+0x43/0x70
kfree+0x131/0x3c0
remove_uuid+0x25e/0x560
hci_sock_sendmsg+0x1033/0x1ea0
Hold hdev->lock while generating and committing the class-of-device and
EIR snapshots. Release it before sending an HCI command, so controller
waits do not happen under the device lock. This protects all UUID list
walks in these paths and restores the serialization lost in the command
sync conversion. |
| In the Linux kernel, the following vulnerability has been resolved:
net: slip: serialize receive against buffer reallocation
sl_realloc_bufs() replaces rbuff and updates buffsize while holding
sl->lock. slip_receive_buf() reads those fields and writes through rbuff
without holding the lock.
An MTU change can therefore race with receive processing. An MTU shrink
can expose the new smaller rbuff with the old larger bound, causing an
out-of-bounds write. A receive callback which already loaded the old
rbuff can instead continue writing after that buffer has been freed.
Serialize receive processing with sl_realloc_bufs() by holding sl->lock
while consuming each receive batch. |
| In the Linux kernel, the following vulnerability has been resolved:
net/sched: serialize qdisc_rtab_list against concurrent get/put
qdisc_get_rtab() and qdisc_put_rtab() mutate the process-global singly
linked list qdisc_rtab_list and a plain non-atomic 'int refcnt' with no
lock. This was only safe because every caller historically held the RTNL
mutex, which serialized all rate-table lookups, inserts and frees.
That invariant no longer holds. cls_flower sets
TCF_PROTO_OPS_DOIT_UNLOCKED, so tc_new_tfilter() keeps rtnl_held == false
for it and sets TCA_ACT_FLAGS_NO_RTNL. That flag propagates through
tcf_exts_validate_ex() -> tcf_action_init() -> tcf_action_init_1() ->
tcf_police_init(), which calls qdisc_get_rtab()/qdisc_put_rtab() with the
RTNL mutex NOT held. Two RTM_NEWTFILTER requests on different CPUs, each
adding a flower filter with a police action carrying the same rate, then
race on qdisc_rtab_list and on the non-atomic refcnt, leading to a
use-after-free / double-free of the kmalloc-2k struct qdisc_rate_table.
qdisc_rtab_list is a single global (not per-netns), so the corrupted
object is shared system-wide.
BUG: KASAN: slab-use-after-free in qdisc_put_rtab+0x12f/0x160
qdisc_put_rtab+0x12f/0x160
tcf_police_init+0xda9/0x1590
tcf_action_init_1+0x460/0x6b0
tcf_action_init+0x439/0xa40
tcf_exts_validate_ex+0x42d/0x550
fl_change+0xddd/0x7da0
tc_new_tfilter+0xaa7/0x2420
rtnetlink_rcv_msg+0x95e/0xe90
which belongs to the cache kmalloc-2k of size 2048
Protect qdisc_rtab_list and the refcount with a dedicated spinlock. The
(sleeping, GFP_KERNEL) allocation in qdisc_get_rtab() is performed before
taking the lock; if a concurrent inserter added an identical table in the
meantime the freshly allocated one is freed under the lock, so no
duplicate is leaked. qdisc_put_rtab() now decrements the refcount and
unlinks under the same lock. |
| In the Linux kernel, the following vulnerability has been resolved:
drm/amdgpu/vcn4: avoid rereading IB param length
Reuse the parameter length returned by
vcn_v4_0_enc_find_ib_param() instead of rereading it from
the IB.
This avoids a potential TOCTOU issue if the IB contents
change between reads.
(cherry picked from commit dbb02b4755f8c1f3773263f2d779872c1c0c073a) |
| In the Linux kernel, the following vulnerability has been resolved:
drm/amdgpu: fix division by zero with invalid uvd dimensions
When width or height is less than 16, width_in_mb or height_in_mb
becomes 0, leading to fs_in_mb being 0. This causes a division by
zero when calculating num_dpb_buffer in H264 and H264 Perf decode
paths.
Add validation to reject frames with width < 16 or height < 16
before performing any calculations that depend on these values.
V2: Format change - move up all vaiable definitions.
V3: Use warn_once to avoid spam.
(cherry picked from commit 3e41d26c70b0a459d041cc19482a226c4b7423cb) |
| In the Linux kernel, the following vulnerability has been resolved:
drm/amdgpu: reject mapping a reserved doorbell to a new queue
When creating an user-queue, the user space
provides a doorbell BO handle and an offset within
the bo to obtain a doorbell.
However current implementation using xa_store_irq()
to store a doorbell, which allows a later queue created
with the same BO and offset parameters to overwrite an
existing queue and doorbell mapping.
This can cause problems like misrouting fence IRQ
processing to a wrong queue, and mislead the cleanup
process of one queue erasing the mapping of another queue.
This commit fixes this issue by replacing xa_store_irq with
xa_insert_irq, which rejects mapping a reserved
doorbell to a newly created queue
(cherry picked from commit 6244eae22966350db52faf9c1369d3b2ffc5de4e) |
| In the Linux kernel, the following vulnerability has been resolved:
xfs: resample the data fork mapping after cycling ILOCK
xfs_reflink_fill_{cow_hole,delalloc} are both presented with an inode,
a data fork mapping, and a cow fork mapping. Unfortunately, these two
helpers cycle the ILOCK to grab a transaction, which means that the
mappings are stale as soon as we reacquire the ILOCK. Currently we
refresh the cow fork mapping by re-calling xfs_find_trim_cow_extent, but
we don't refresh the data fork mapping beforehand, which means that the
xfs_bmap_trim_cow in that function queries the refcount btree about the
wrong physical blocks and returns an inaccurate value in *shared.
If *shared is now false, the directio write proceeds with a stale data
fork mapping. Fix this by querying the data fork mapping if the
sequence counter changes across the ILOCK cycle. |
| In the Linux kernel, the following vulnerability has been resolved:
xfrm: fix sk_dst_cache double-free in xfrm_user_policy()
xfrm_user_policy() clears the socket dst cache with __sk_dst_reset(),
i.e. the non-atomic __sk_dst_set(sk, NULL): it reads sk_dst_cache with
rcu_dereference_protected(), stores NULL and dst_release()s the old dst.
That is only safe if no other thread modifies sk_dst_cache concurrently.
For a connected UDP socket that does not hold: the transmit fast path
(udp_sendmsg -> sk_dst_check -> sk_dst_reset) resets the cache locklessly
with an atomic xchg(). A per-socket policy change racing a send can make
both sides observe the same old dst and each dst_release() it, dropping
the socket's single reference twice and freeing the xfrm_dst bundle while
it is still referenced:
BUG: KASAN: slab-use-after-free in dst_release
Write of size 4 at addr ffff88801897b6c0 by task exploit/155
Call Trace:
...
dst_release (... ./include/linux/rcuref.h:109)
xfrm_user_policy (./include/net/sock.h:2239 ./include/net/sock.h:2256 net/xfrm/xfrm_state.c:3053)
do_ip_setsockopt (net/ipv4/ip_sockglue.c:1347)
ip_setsockopt (net/ipv4/ip_sockglue.c:1417)
do_sock_setsockopt (net/socket.c:2368)
__sys_setsockopt (net/socket.c:2393)
__x64_sys_setsockopt (net/socket.c:2396)
do_syscall_64 (arch/x86/entry/syscall_64.c:94)
entry_SYSCALL_64_after_hwframe (arch/x86/entry/entry_64.S:121)
Reachable by an unprivileged user via a user+network namespace.
Use the atomic sk_dst_reset() so the cache is cleared and released with a
single xchg(): whichever side wins releases the dst once, the other sees
NULL and does nothing. Behaviour is otherwise unchanged. |
| In the Linux kernel, the following vulnerability has been resolved:
KVM: x86: Check for invalid/obsolete root *after* making MMU pages available
Check for a "stale" page fault, i.e. for an invalid and/or obsolete root,
after making MMU pages available for the shadow MMU. If reclaiming shadow
pages zaps an in-use root, i.e. marks it invalid, then KVM will attempt to
map memory into an invalid root. On its own, populating an invalid root is
"fine", but because child shadow pages inherit their parent's role, any
children created during the map/fetch will be created as invalid pages,
thus violating KVM's invariant that invalid pages are never on the list of
active MMU pages.
Note, the underlying flaw has existed since KVM first started tracking
invalid roots in 2008 (commit 2e53d63acba7, "KVM: MMU: ignore zapped root
pagetables"), but the true badness only came along in 2020 (Linux 5.9)
with the invariant that invalid shadow pages can't be on the list of
active pages.
Note #2, inheriting role.invalid when creating child shadow pages is also
far from ideal; that flaw will be addressed separately. |
| In the Linux kernel, the following vulnerability has been resolved:
posix-cpu-timers: Prevent UAF caused by non-leader exec() race
Wongi and Jungwoo decoded and reported a non-leader exec() related race
which can result in an UAF:
sys_timer_delete() exec()
posix_cpu_timer_del()
// Observes old leader
p = pid_task(pid, pid_type); de_thread()
switch_leader();
release_task(old_leader)
__exit_signal(old_leader)
sighand = lock(old_leader, sighand);
posix_cpu_timers*_exit();
sighand = lock_task_sighand(p) unhash_task(old_leader);
sh = lock(p, sighand) old_leader->sighand = NULL;
unlock(sighand);
(p->sighand == NULL)
unlock(sh)
return NULL;
// Returns without action
if(!sighand)
return 0;
free_posix_timer();
This is "harmless" unless the deleted timer was armed and enqueued in
p->signal because on exec() a TGID targeted timer is inherited.
As sys_timer_delete() freed the underlying posix timer object
run_posix_cpu_timers() or any timerqueue related add/delete operations on
other timers will access the freed object's timerqueue node, which results
in an UAF.
There is a similar problem vs. posix_cpu_timer_set(). For regular posix
timers it just transiently returns -ESRCH to user space, but for the use
case in do_cpu_nanosleep() it's the same UAF just that the k_itimer is
allocated on the stack.
Also posix_cpu_timer_rearm() fails to rearm the timer, which means it stops
to expire.
While debating solutions Frederic pointed out another problem:
posix_cpu_timer_del(tmr)
__exit_signal(p)
posix_cpu_timers*_exit(p);
unhash_task(p);
p->sighand = NULL;
sh = lock_task_sighand(p)
sighand = p->sighand;
if (!sighand)
return NULL;
lock(sighand);
if (!sh)
WARN_ON_ONCE(timer_queued(tmr));
On weakly ordered architectures it is not guaranteed that
posix_cpu_timer_del() will observe the stores in posix_cpu_timers*_exit()
when p->sighand is observed as NULL, which means the WARN() can be a false
positive.
Solve these issues by:
1) Changing the store in __exit_signal() to smp_store_release().
2) Adding a smp_acquire__after_ctrl_dep() into the !sighand path
of lock_task_sighand().
3) Creating a helper function for looking up the task and locking sighand
which does not return when sighand == NULL. Instead it retries the
task lookup and only if that fails it gives up.
4) Using that helper in the three affected functions.
#1/#2 ensures that the reader side which observes sighand == NULL also
observes all preceeding stores, i.e. the stores in posix_cpu_timers*_exit()
and the ones in unhash_task().
#3 ensures that the above described non-leader exec() situation is handled
gracefully. When the task lookup returns the old leader, but sighand ==
NULL then it retries. In the non-leader exec() case the subsequent task
lookup will observe the new leader due to #1/#2. In normal exit() scenarios
the subsequent lookup fails.
When the task lookup fails, the function also checks whether the timer is
still enqueued and issues a warning if that's the case. Unfortunately there
is nothing which can be done about it, but as the task is already not
longer visible the timer should not be accessed anymore. This check also
requires memory ordering, which is not provided when the first lookup
fails. To achieve that the check is preceeded by a smp_rmb() which pairs
with the smp_wmb() in write_seqlock() in __exit_signal(). That ensures that
the stores in posix_cpu_timers*_exit() are visible.
The history of the non-leader exec() issue goes back to the early days of
posix CPU timers, which stored a pointer to the group leader task in the
timer. That obviously fails when a non-leader exec() switches the leader.
commit e0a70217107e ("posix-cpu-timers: workaround to suppress the problems
with mt exec") added a temporary workaround for that in 2010 which surv
---truncated--- |
| In the Linux kernel, the following vulnerability has been resolved:
net/smc: fix UAF in smc_cdc_rx_handler() by pinning the socket
smc_cdc_rx_handler() looks up the connection by token under the link
group's conns_lock, drops the lock, and then dereferences conn and the
smc_sock derived from it, ending in sock_hold(&smc->sk) inside
smc_cdc_msg_recv(). No reference is held across the lock release.
The only reference pinning the socket while the connection is
discoverable in the link group is taken in smc_lgr_register_conn()
(sock_hold) and dropped in __smc_lgr_unregister_conn() (sock_put), both
under conns_lock. Once the handler drops conns_lock, a concurrent
close() -> smc_release() -> smc_conn_free() -> smc_lgr_unregister_conn()
can drop that reference and free the smc_sock, so the handler's later
sock_hold() runs on freed memory:
WARNING: lib/refcount.c:25 at refcount_warn_saturate
Workqueue: rxe_wq do_work
refcount_warn_saturate (lib/refcount.c:25)
smc_cdc_msg_recv (net/smc/smc_cdc.c:430)
smc_cdc_rx_handler (net/smc/smc_cdc.c:502)
smc_wr_rx_tasklet_fn (net/smc/smc_wr.c:445)
tasklet_action_common (kernel/softirq.c:938)
handle_softirqs (kernel/softirq.c:622)
Kernel panic - not syncing: panic_on_warn set
Only SMC-R is affected. The SMC-D receive tasklet is stopped by
tasklet_kill(&conn->rx_tsklet) in smc_conn_free() before the connection
is unregistered, so it cannot run concurrently with the free.
Take the socket reference while still holding conns_lock, so the
registration reference can no longer be the last one, and drop it once
the handler is done. |
| In the Linux kernel, the following vulnerability has been resolved:
iio: event: Fix event FIFO reset race
`iio_event_getfd()` creates the event file descriptor with
`anon_inode_getfd()`, which allocates a new fd, creates the anonymous
file and installs it in the process fd table before returning to the
caller.
The IIO code resets the event FIFO after `anon_inode_getfd()` has returned,
but before `IIO_GET_EVENT_FD_IOCTL` has copied the fd number to userspace.
But since fd tables are shared between threads, another thread can guess
the newly allocated fd number and issue a `read()` on it as soon as the fd
has been installed.
This means the `kfifo_to_user()` in `iio_event_chrdev_read()` can run in
parallel with the `kfifo_reset_out()` in `iio_event_getfd()`.
The kfifo documentation says that `kfifo_reset_out()` is only safe when it
is called from the reader thread and there is only one concurrent reader.
Otherwise it is dangerous and must be handled in the same way as
`kfifo_reset()`.
If that happens, `kfifo_to_user()` can advance the FIFO `out` index based
on state from before the reset, after the reset has already moved the `out`
index to the current `in` index. That can leave the FIFO with an `out`
index past the `in` index. A later `read()` can then see an underflowed
FIFO length and copy more data than the event FIFO buffer contains. This
can result in an out-of-bounds read and leak adjacent kernel memory to
userspace.
Move the FIFO reset before `anon_inode_getfd()`. At that point the event fd is
marked busy, but the new fd has not been installed yet, so userspace cannot
access it while the FIFO is reset. |
| In the Linux kernel, the following vulnerability has been resolved:
proc: protect ptrace_may_access() with exec_update_lock (FD links)
proc_pid_get_link() and proc_pid_readlink() currently look up the task from
the pid once, then do the ptrace access check on that task, then look up
the task from the pid a second time to do the actual access.
That's racy in several ways.
To fix it, pass the task to the ->proc_get_link() handler, and instead of
proc_fd_access_allowed(), introduce a new helper call_proc_get_link() that
looks up and locks the task, does the access check, and calls
->proc_get_link(). |
| In the Linux kernel, the following vulnerability has been resolved:
KVM: x86: Ensure vendor's exit handler runs before fastpath userspace exits
Move the handling of fastpath userspace exits into vendor code to ensure
KVM runs vendor specific operations that need to run before userspace gains
control of the vCPU. E.g. for VMX (and soon to be for SVM as well), KVM
needs to flush the PML buffer prior to exiting to userspace, otherwise any
memory written by the final KVM_RUN might never be flagged as dirty.
Note, waiting to snapshot CR0 and CR3 until svm_handle_exit() is flawed in
general, as that risks consuming stale state in a fastpath handler. That
will be addressed in a future change. |
| In the Linux kernel, the following vulnerability has been resolved:
fuse: re-lock request before returning from fuse_ref_folio()
fuse_ref_folio() unlocks the request but does not re-lock it before
returning. fuse_chan_abort() can end the request and the async end
callback (eg fuse_writepage_free()) can free the args while the
subsequent copy chain logic after fuse_ref_folio() accesses them,
leading to use-after-free issues.
Fix this by locking the request in fuse_ref_folio() before returning. |
| In the Linux kernel, the following vulnerability has been resolved:
net/mlx5e: xsk: Fix unlocked writing to ICOSQ
During napi poll, when the affinity changes and there's still XSK work
to be done, we trigger an ICOSQ interrupt on the new CPU. However, this
triggering on the ICOSQ is done unprotected.
There are 2 such races:
A) mlx5e_trigger_irq() is called while mlx5e_xsk_alloc_rx_mpwqe() is
running from a different CPU due to affinity change. This can happen
because IRQ triggering is done after napi_complete_done(). At this point
the NAPI can be scheduled on a different CPU. Like this:
CPU A (old affinity, NAPI tail) CPU B (new affinity, fresh NAPI)
------------------------------- --------------------------------
napi_complete_done() clears SCHED
mlx5e_cq_arm(...)
napi_schedule_prep() sets SCHED
mlx5e_napi_poll()
mlx5e_xsk_alloc_rx_mpwqe()
mlx5e_icosq_sync_lock() // noop
memcpy 640 B UMR body
advance sq->pc by 10
mlx5e_trigger_irq(&c->icosq)
wqe_info[pi] = {NOP, 1}
mlx5e_post_nop() advances sq->pc
B) mlx5e_trigger_irq() is called on the ICOSQ when
mlx5e_trigger_napi_icosq() is running.
The obvious fix would be to lock the ICOSQ. But ICOSQ has an optimized
locking scheme that doesn't work for this scenario. Kick the async ICOSQ
instead which is always locked.
This issue was noticed in the wild with the following splat:
netdevice: ge-0-0-1: Bad OP in ICOSQ CQE: 0xd
WARNING: drivers/net/ethernet/mellanox/mlx5/core/en_rx.c:826 [...]
[...]
Call Trace:
<IRQ>
mlx5e_napi_poll+0x11d/0x7f0 [mlx5_core]
__napi_poll+0x30/0x200
? skb_defer_free_flush+0x9c/0xc0
net_rx_action+0x2fe/0x3f0
handle_softirqs+0xd8/0x340
__irq_exit_rcu+0xbc/0xe0
common_interrupt+0x85/0xa0
</IRQ>
<TASK>
asm_common_interrupt+0x26/0x40
[...]
---[ end trace 0000000000000000 ]---
mlx5_core 0000:08:00.0 ge-0-0-1: Error cqe on cqn 0x548, ci 0x2022, qn 0x8f4,
opcode 0xd, syndrome 0x2, vendor syndrome 0x68
00000000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00000020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00000030: 00 00 00 00 01 00 68 02 01 00 08 f4 de 14 59 d2
WQE DUMP: WQ size 16384 WQ cur size 0, WQE index 0x1e14, len: 64
00000000: 00 00 00 01 d9 ed 80 02 00 00 00 01 d9 ed 90 02
00000010: 00 00 00 01 d9 ed a0 02 00 00 00 01 d9 ed b0 02
00000020: 00 00 00 01 d9 ed c0 02 00 00 00 01 d9 ed d0 02
00000030: 00 00 00 01 d9 ed e0 02 00 00 00 01 d9 ed f0 02
mlx5_core 0000:08:00.0 ge-0-0-1: Error cqe on cqn 0x548, ci 0x2023, qn 0x8f4,
opcode 0xd, syndrome 0x5, vendor syndrome 0xf9
00000000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00000020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00000030: 00 00 00 00 01 00 f9 05 01 00 08 f4 de 15 cf d2 |
| In the Linux kernel, the following vulnerability has been resolved:
Bluetooth: L2CAP: cancel pending_rx_work before taking conn->lock
l2cap_conn_del() takes conn->lock and then calls cancel_work_sync() for
pending_rx_work. process_pending_rx() takes the same mutex, so teardown
can deadlock against the worker it is flushing.
This issue was found by our static analysis tool and then manually
reviewed against the current tree.
The grounded PoC kept the l2cap_conn_ready() -> queue_work(...,
&conn->pending_rx_work) submit path, the l2cap_conn_del() ->
cancel_work_sync(&conn->pending_rx_work) teardown path, and the
process_pending_rx() -> mutex_lock(&conn->lock) worker edge. Lockdep
WARNING: possible circular locking dependency detected
process_pending_rx+0x21/0x2a [vuln_msv]
l2cap_conn_del.constprop.0+0x3f/0x4e [vuln_msv]
*** DEADLOCK ***
Cancel pending_rx_work before taking conn->lock, matching the existing
lock-before-drain ordering used for the two delayed works in the same
teardown path. The pending_rx queue is still purged after the work has
been cancelled and conn->lock has been acquired. |
| In the Linux kernel, the following vulnerability has been resolved:
netfs: Fix potential for tearing in ->remote_i_size and ->zero_point
Fix potential tearing in using ->remote_i_size and ->zero_point by copying
i_size_read() and i_size_write() and using the same seqcount as for i_size.
We need to make sure that netfslib and the filesystems that use it always
hold i_lock whilst updating any of the sizes to prevent i_size_seqcount
from getting corrupted. |
| In the Linux kernel, the following vulnerability has been resolved:
rbd: eliminate a race in lock_dwork draining on unmap
Given how rbd_lock_add_request() and rbd_img_exclusive_lock() are
written, lock_dwork may be (re)queued more than it's actually needed:
for example in case a new I/O request comes in while we are in the
middle of rbd_acquire_lock() on behalf of another I/O request. This is
expected and with rbd_release_lock() preemptively canceling lock_dwork
is benign under normal operation.
A more problematic example is maybe_kick_acquire():
if (have_requests || delayed_work_pending(&rbd_dev->lock_dwork)) {
dout("%s rbd_dev %p kicking lock_dwork\n", __func__, rbd_dev);
mod_delayed_work(rbd_dev->task_wq, &rbd_dev->lock_dwork, 0);
}
It's not unrealistic for lock_dwork to get canceled right after
delayed_work_pending() returns true and for mod_delayed_work() to
requeue it right there anyway. This is a classic TOCTOU race.
When it comes to unmapping the image, there is an implicit assumption
of no self-initiated exclusive lock activity past the point of return
from rbd_dev_image_unlock() which unlocks the lock if it happens to be
held. This unlock is assumed to be final and lock_dwork (as well as
all other exclusive lock tasks, really) isn't expected to get queued
again. However, lock_dwork is canceled only in cancel_tasks_sync()
(i.e. later in the unmap sequence) and on top of that the cancellation
can get in effect nullified by maybe_kick_acquire(). This may result
in rbd_acquire_lock() executing after rbd_dev_device_release() and
rbd_dev_image_release() run and free and/or reset a bunch of things.
One of the possible failure modes then is a violated
rbd_assert(rbd_image_format_valid(rbd_dev->image_format));
in rbd_dev_header_info() which is called via rbd_dev_refresh() from
rbd_post_acquire_action().
Redo exclusive lock task draining to provide saner semantics and try
to meet the assumptions around rbd_dev_image_unlock(). |
