| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| In the Linux kernel, the following vulnerability has been resolved:
nvme-rdma: fix -EIO cleanup order in queue_rq
On -EIO, the RDMA queue_rq path reports a host path error and then
still cleans up the command and unmaps the SQE DMA. The path error
helper completes the request, so that is double cleanup and DMA unmap
after the request is already complete.
Unmap the SQE first, then report the host path error. Skip the outer
command cleanup on that path. |
| In the Linux kernel, the following vulnerability has been resolved:
bpf: Fix BPF_F_CPU validation for sparse CPU IDs
BPF_F_CPU stores the target CPU ID in the upper 32 bits of the map
operation flags. bpf_map_check_op_flags() currently compares that ID
with num_possible_cpus(), which is the number of possible CPUs rather
than a bound on CPU IDs.
On an arm64 QEMU guest with a CPU device-tree hole, the possible CPU
mask was 0,2-3. A userspace program using raw bpf() syscalls creates
a BPF_MAP_TYPE_PERCPU_ARRAY and performs update and lookup operations
for each CPU by setting BPF_F_CPU and the CPU ID in the flags.
With the old check, CPU 1 is incorrectly accepted while valid CPU 3 is
rejected with -ERANGE. The CPU 1 update then reaches the per-CPU map
access path and triggers:
Unable to handle kernel paging request at virtual address ...
pc : __pi_memcpy_generic+0x5c/0x22c
lr : bpf_percpu_array_update+0x2dc/0x2e8
Call trace:
__pi_memcpy_generic
bpf_map_update_value
map_update_elem
__sys_bpf
Check the CPU ID against nr_cpu_ids and cpu_possible() instead. This
rejects CPU IDs outside the valid range and CPUs absent from the
possible mask, while allowing valid sparse CPU IDs. |
| In the Linux kernel, the following vulnerability has been resolved:
accel: ethosu: Don't read the U65 rounding mode as a storage mode
Bits 15:14 of NPU_SET_{IFM,OFM}_PRECISION select the activation storage
mode on U85 only. On U65 the same field holds the rounding mode, and the
command stream parser has read it as a storage mode since the driver was
added.
That went unnoticed while unknown values fell through the switch, but
now that they are rejected, every U65 command stream that asks for
natural rounding (2) fails CMDSTREAM_BO_CREATE with -EINVAL. Mesa emits
it for average pooling, concatenation, split, unpack, strided slice, LUT
and argmax, which is 72 failures of the Teflon test suite on an i.MX93.
Truncating rounding (1) is misread as well: it picks the two-tile
address path and computes a bogus feature map size from tile bases the
command stream never set.
Read the field as a storage mode only on the hardware where it is one. |
| In the Linux kernel, the following vulnerability has been resolved:
sctp: fix a TOCTOU race in SCTP_CMD_TIMER_START
The SCTP_CMD_TIMER_START handler checks timer_pending() before calling
timer_reduce(). The timer can expire and detach between these operations,
causing timer_reduce() to rearm the timer without taking the association
reference required for the newly armed timer.
The timer callback later unconditionally drops its association reference,
which can leave the association reference count unbalanced and result in
use-after-free during association teardown.
Use the return value of timer_reduce() to determine whether the timer was
actually armed. Take the association reference only when timer_reduce()
successfully starts a new timer, closing the race between checking the
timer state and rearming it.
This issue was reported by Nico Yip (@_cyeaa_) working with TrendAI Zero
Day Initiative. |
| In the Linux kernel, the following vulnerability has been resolved:
vxlan: mdb: Fix use-after-free in vxlan_mdb_remote_src_del()
vxlan_mdb_is_valid_source(), which validates MDBE_ATTR_SOURCE and every
MDBE_ATTR_SRC_LIST member, accepts the all-zeros address.
A source list is only accepted on a (*, G) entry, whose source is the
all-zeros address, and for each member of the list an (S, G) entry is
derived from it by substituting the source. Entries are keyed by a plain
memcmp() of struct vxlan_mdb_entry_key, so if MDBE_ATTR_SOURCE is present
and holds the all-zeros address and the source list holds it as well, the
derived (S, G) key is byte-identical to the (*, G) key and resolves to the
same entry. Omitting MDBE_ATTR_SOURCE is not equivalent, as the key is
then left with a zero address family.
vxlan_mdb_remote_src_del() removes the forwarding entry of a source before
freeing the source entry:
vxlan_mdb_remote_src_fwd_del(vxlan, group, remote, &ent->addr);
vxlan_mdb_remote_src_entry_del(ent);
With the keys aliased, the first call deletes the remote of the entry that
owns 'ent' instead of a separate (S, G) entry, and frees 'ent'. The second
call then runs on the freed entry, and its hlist_del() reads ->pprev and
->next out of it and writes through them.
Adding the (*, G) entry with NLM_F_REPLACE and no source list marks the
all-zeros source for deletion and reaches this from the sweep at the end
of vxlan_mdb_remote_srcs_replace().
BUG: KASAN: slab-use-after-free in __vxlan_mdb_add+0x1cd/0xd70
Read of size 8 at addr ffff888102852500 by task poc/84
__vxlan_mdb_add+0x1cd/0xd70
vxlan_mdb_add+0xc0/0x140
rtnl_mdb_add+0x157/0x2a0
rtnetlink_rcv_msg+0x207/0x5a0
Allocated by task 84:
__kmalloc_cache_noprof+0x153/0x360
vxlan_mdb_remote_srcs_add+0x2eb/0x440
__vxlan_mdb_add+0x803/0xd70
Freed by task 84:
kfree+0x14c/0x3b0
vxlan_mdb_remote_del+0x129/0x1a0
__vxlan_mdb_del+0x4f/0xe0
vxlan_mdb_remote_src_fwd_del.isra.0+0x162/0x1b0
__vxlan_mdb_add+0x1c5/0xd70
The MDB operations are netns-scoped, so an unprivileged user can perform
them in a new user and network namespace.
Reject the all-zeros address in vxlan_mdb_is_valid_source(), which covers
both call sites. A (*, G) entry is expressed by omitting the source, so
nothing legitimate is refused.
Discovered by XBOW, triaged by Baul Lee <baul.lee@xbow.com> |
| In the Linux kernel, the following vulnerability has been resolved:
ALSA: pcm: Serialize PCM mmap with buffer reallocation to fix page UAF
snd_pcm_hw_params() and snd_pcm_hw_free() guard buffer reallocation
with an mmap_count check performed under the PCM stream lock, but the
lock is released long before the buffer is actually freed:
snd_pcm_sync_stop(), constraint refinement and do_free_pages() all
happen in between. snd_pcm_mmap_data(), on the other hand, takes no
lock at all: it validates against the old buffer's state and
dma_bytes, remaps its pages into the VMA, and only then increments
mmap_count.
A concurrent mmap() can therefore slip in between the check and the
free. remap_pfn_range() installs writable PTEs for the old buffer's
pages without taking page references, and the subsequent
do_free_pages() returns those pages to the page allocator while the
VMA still maps them. This leaves a stale, writable mapping of freed
pages: a page-level use-after-free that can be leveraged for local
privilege escalation.
Make snd_pcm_mmap_data() participate in the buffer-access scheme
introduced for hw_params/hw_free: acquire runtime->buffer_accessing
before validating and remapping, and release it afterwards. Buffer
reallocation already fails with -EBUSY while accessors are active,
and the mmap side now fails with -EBUSY while a reallocation is in
progress, so the validate/remap sequence and the check/free sequence
can no longer interleave.
A reproducer that turns this race into a stale writable mapping of
the freed DMA buffer pages is available on request. |
| In the Linux kernel, the following vulnerability has been resolved:
ksmbd: safely drain sessions during logoff
SMB3 multichannel allows requests for one session to run on multiple
connections. Wait for all channels bound to a session before freeing
shared session objects.
A deferred byte-range lock remains counted as a running request and only
wakes when its file closes. Wake blocked locks during the drain without
unpublishing or modifying their file objects. Synchronous CANCEL requests
must invoke their cancellation callback to wake pending operations, while
CHANGE_NOTIFY completion remains specific to the asynchronous path.
Serialize session teardown with channel registration and previous-session
cleanup, and use atomic work-state transitions so LOGOFF, CANCEL, and
connection teardown invoke cancellation callbacks only once. |
| In the Linux kernel, the following vulnerability has been resolved:
ksmbd: fix listener task lifetime on netdev events
The listener thread exits when its listening socket is shutdown. The
netdevice notifier shuts down the socket before calling kthread_stop(), so
the task_struct can be freed before kthread_stop() gets its reference.
Create the listener in a stopped state and hold an extra task_struct
reference until kthread_stop_put() completes. Also stop and release
listeners before freeing their interface records during TCP teardown. |
| In the Linux kernel, the following vulnerability has been resolved:
net/rds: acquire the fastpath locks in rds_conn_shutdown()
rds_conn_shutdown() quiesces the transmit and receive-refill paths by
waiting for RDS_IN_XMIT and RDS_RECV_REFILL to be sampled clear, and
then runs the transport shutdown and rds_conn_path_reset(). Sampling
the bits clear is not the same as owning them: the moment after the
wait_event() returns, rds_send_xmit() can re-acquire RDS_IN_XMIT (or
rds_ib_recv_refill() can re-acquire RDS_RECV_REFILL) and run
concurrently with the teardown.
The sender does recheck the connection state after taking the lock,
but that recheck is a classic store-buffering pattern: teardown writes
the state and reads the bit while the sender writes the bit and reads
the state. acquire_in_xmit() is only an acquire operation, so on
weakly ordered architectures both sides can miss each other's write,
and the transmit path then runs while the transport zeroes its rings
(e.g. rds_ib_ring_init()) and rds_send_path_reset() rewrites the
transmit state under it.
Oracle UEK fixed the same class of crashes - a 14-year tail of
BUG_ON()s in rds_ib_sub_signaled(), unexpected op-codes and NULL
dereferences in rds_ib_send_cqe_handler() during failover testing -
by making the teardown path *acquire* the fastpath bit locks instead
of testing them ("rds: Make sure transmit path and connection
tear-down does not run concurrently"). Ownership of a single word is
decided by RMW atomicity, so no cross-variable ordering is needed.
Do the same here: take both locks before calling the transport
shutdown, hold them across rds_conn_path_reset(), and release them
explicitly with a wake-up afterwards. Both are released with
clear_bit_unlock(), so that the ring re-initialization done by the
transport shutdown and the transmit state rewritten by
rds_send_path_reset() are ordered before either bit is seen clear by
the next acquire_in_xmit() or acquire_refill().
The fastpath users of these bits - rds_send_xmit() and
rds_ib_recv_refill() - are trylock style and back off while teardown
owns the locks, so no new lock dependency is introduced for them.
rds_tcp_reset_callbacks() is different: since the previous patch it
acquires RDS_IN_XMIT as well, and it blocks doing so, so its wait now
spans the teardown instead of at most one send batch. That waiter
runs from rds_tcp_accept_one() on the single-threaded krdsd workqueue
and holds rds_tcp_accept_lock and t_conn_path_lock while it waits, so
a duelling SYN accepted while its path is being torn down parks
accept processing for the duration of the teardown - for TCP bounded
by the (up to 5 s) drain loop in rds_tcp_conn_path_shutdown(). An IB
path's drain in rds_ib_conn_path_shutdown() has no round cap, but no
blocking waiter either: rds_tcp_reset_callbacks() is the only blocking
acquirer of these bits and waits only on its own TCP path, and the
fastpaths are trylock-and-back-off on both transports, so a long IB
drain lengthens only that path's own quiesce. The
window is narrow: the accept-side state check has to pass before the
teardown moves the path to RDS_CONN_DISCONNECTING.
Because krdsd is a single global workqueue, everything else queued
there - accept processing for other connections and network
namespaces, and the flush_workqueue(rds_wq) in rds_tcp_listen_stop()
during namespace teardown - waits behind the parked accept worker for
that time. It cannot deadlock, although the waits do point at each
other: the teardown blocks until the bit's holder releases it, and
the holder may be that krdsd accept worker. The holder finishes
without needing anything the teardown owns: the sync cancels
rds_tcp_reset_callbacks() issues target cp_send_w and cp_recv_w on
the path's ordered cp_wq, whose only execution slot is occupied by
the blocked cp_down_w itself, so they are pending at most and cancel
without flushing - a reliance on cp_wq being ordered that is now
noted next to those cancels (on
---truncated--- |
| In the Linux kernel, the following vulnerability has been resolved:
eth: nfp: bound the ntuple rule dump by the caller's buffer size
nfp_net_get_fs_loc() dumps every entry of nn->fs.list into rule_locs[]
without consulting cmd->rule_cnt, which is how many entries the caller
had room for. ETHTOOL_GRXCLSRLALL requires no CAP_NET_ADMIN and the
ioctl sizes the buffer from the rule_cnt userspace passes in, so once an
admin has installed flow steering rules any user can ask for fewer slots
than there are rules and run off the end of the allocation. A rule_cnt
of 0 leaves the buffer pointer NULL and the walk dereferences it.
Bail out with -EMSGSIZE when the buffer fills up, the way the other
ntuple capable drivers do, and report how many locations were filled so
a shrinking rule list does not leave the caller reading stale slots. |
| In the Linux kernel, the following vulnerability has been resolved:
vxlan: reject dynamic fdb entries that reference a nexthop id
The commit cited in the Fixes tag allowed VXLAN FDB entries to point to
FDB nexthops so that overlay traffic could be load balanced across
multiple VTEPs. Such entries can only be configured from user space,
cannot be learned and cannot roam. They only make sense with a user space
control plane such as E-VPN where data plane learning is disabled.
Despite that, the VXLAN driver does not currently prevent such entries
from being configured with the "dynamic" flag. The per-nexthop FDB list
is only protected by the per-device hash lock, which is not sufficient
when two VXLAN devices point to the same FDB nexthop and therefore share
the list. Aging runs in softirq context without RTNL, so an entry deleted
by one device can race with an addition or deletion from the other,
leading to list corruption:
list_del corruption. next->prev should be ffff8881069d9548, but was
dead000000000122. (next=ffff8881069d9448)
WARNING: CPU: 0 PID: 90 at lib/list_debug.c:65
__list_del_entry_valid_or_report+0x1aa/0x210
...
vxlan_fdb_destroy+0x5b8/0xad0
vxlan_cleanup+0x328/0x450
call_timer_fn+0x2a/0x1c0
run_timer_softirq+0x18c/0x210
BUG: KASAN: slab-use-after-free in vxlan_fdb_destroy
Fix this by rejecting the bogus configuration of dynamic FDB entries that
point to FDB nexthops, both when created and when an existing entry is
updated. As such, the per-nexthop FDB list is only ever mutated under the
RTNL lock. Add test cases to make sure that this does not regress in the
future. |
| In the Linux kernel, the following vulnerability has been resolved:
vdpa_sim_blk: reject out-of-range sector starts
vdpasim_blk_check_range() logs an invalid start sector but continues
validating the request. The subsequent unsigned capacity subtraction can
underflow and let an out-of-range buffer offset reach the data path.
The invalid offset is used by three request paths. VIRTIO_BLK_T_OUT
copies guest data to blk->buffer + offset through
vringh_iov_pull_iotlb(), causing an out-of-bounds write in
_copy_from_iter() or memcpy(). VIRTIO_BLK_T_IN copies from
blk->buffer + offset to the guest through vringh_iov_push_iotlb(),
causing an out-of-bounds read in _copy_to_iter().
VIRTIO_BLK_T_WRITE_ZEROES passes blk->buffer + offset to memset(),
causing an out-of-bounds write.
Reject starts at or beyond the capacity before the subtraction. Treat the
capacity boundary as invalid because the IN and OUT paths round byte counts
down to sectors for validation but later copy the original byte counts. A
sub-sector request at the capacity boundary would otherwise still access
past the end of the buffer.
I found this bug myself, though the patch was written with AI assistance. |
| In the Linux kernel, the following vulnerability has been resolved:
vdpa_sim_net: check TX pull result before RX copy
vringh_iov_pull_iotlb() returns a signed byte count. A failed TX pull is
currently added to the unsigned byte counter and then passed as a size_t
length to receive_filter() and vringh_iov_push_iotlb(). A negative error
can therefore become a large length in the RX path.
Handle non-positive pull results before every length use. Count the TX
error and complete the consumed TX descriptor with zero bytes.
I found this bug myself, though the patch was written with AI assistance. |
| In the Linux kernel, the following vulnerability has been resolved:
ipv6: fix fib6 walker UAF on seq stop
ipv6_route_iter_active() treats a walker in FWS_U at the table root as
already unlinked. fib6_del_route() can move a still-linked walker into
that same state when the current leaf is the last route at the root,
so ipv6_route_native_seq_stop() skips fib6_walker_unlink(). The seq
private object can then be freed while it remains on
net->ipv6.fib6_walkers. A later route deletion walks the dangling list
and uses the freed walker.
Use the list head as membership state and reinitialize it when
unlinking. Keep the existing w->node check so a never-started iterator
with a zeroed private object is not treated as linked.
The same stop helper is used by /proc/net/ipv6_route and by the BPF
ipv6_route iterator. The BPF show path only widens the race. |
| In the Linux kernel, the following vulnerability has been resolved:
ufs: validate cylinder group metadata before caching it
ufs_read_cylinder() copies the cylinder group index and the rotor
positions straight from the on-disk group and caches them without any
check:
ucpi->c_cgx = fs32_to_cpu(sb, ucg->cg_cgx);
ucpi->c_rotor = fs32_to_cpu(sb, ucg->cg_rotor);
ucpi->c_frotor = fs32_to_cpu(sb, ucg->cg_frotor);
ucpi->c_irotor = fs32_to_cpu(sb, ucg->cg_irotor);
They are then used as indices during allocation and free:
- c_cgx indexes the cylinder summary array as
UFS_SB(sb)->fs_cs(ucpi->c_cgx), so a value past s_ncg writes a 32
bit count outside the s_csp allocation.
- c_frotor becomes a bitmap scan start, start = c_frotor >> 3, and
then length = ((s_fpg + 7) >> 3) - start. A start beyond the block
bitmap wraps the unsigned length to a huge value, so ubh_scanc()
walks far past the cylinder group buffers. c_irotor drives the
inode bitmap the same way.
A crafted image can set any of these freely, turning an ordinary
allocation into an out of bounds access.
Reject a cylinder group whose recorded index does not match the group
being read, or whose rotors fall outside the group, before the metadata
is cached. Valid filesystems keep cg_cgx equal to the group number and
the rotors within the group, so only malformed images are rejected. |
| In the Linux kernel, the following vulnerability has been resolved:
accel: ethosu: Ensure SRAM region size matches job
It is possible for userspace to set the job SRAM size to 0, but then still
have SRAM accesses in the command stream. When the job SRAM size is 0,
setting the region base register is skipped and a stale base address from
a prior job is used.
Check the region size against the job's SRAM size instead of just the size
of the SRAM. The job's SRAM size was already checked against the total SRAM
size. |
| In the Linux kernel, the following vulnerability has been resolved:
ASoC: sprd: validate compress buffer sizes against fixed allocations
sprd_platform_compr_open() allocates the stage 0 IRAM buffer (32K data
area) and the stage 1 DDR buffer (2M data area) with fixed sizes, but
sprd_platform_compr_copy() derives all copy lengths from the user
controlled runtime->fragment_size and the write() count, never
comparing them against the physical buffer sizes. The compress core
only checks fragment_size * fragments for an u32 overflow in
snd_compress_check_input(), so a local user can configure a logical
buffer of up to ~4GB via SNDRV_COMPRESS_SET_PARAMS, far exceeding the
fixed allocations.
A fragment_size larger than the 32K IRAM data area makes the stage 0
copy_from_user() overflow past the IRAM allocation, and a buffer_size
larger than the 2M DDR buffer makes the wrapping copy at the end of
sprd_platform_compr_copy() write fully user controlled data past the
buffer. No SNDRV_PCM_TRIGGER_START is needed, a write() in SETUP
state reaches the copy callback directly.
Reject parameters that do not fit into the fixed buffers in
set_params(), and fix the advertised max fragment size: 128K never
fitted into the 32K IRAM buffer. The caps values may have been carried over
from the qdsp6 driver, which allocates its buffers according to the
advertised maxima, unlike this driver. With 32K as max fragment size
the advertised limits are self-consistent: 32K * 64 = 2M equals the
DDR buffer size.
Discovered by Atuin - Automated Vulnerability Discovery Engine. |
| In the Linux kernel, the following vulnerability has been resolved:
exit: hold a reference to thread_pid across proc_flush_pid
Commit 0a36bad01731 ("release_task: kill the no longer needed
get/put_pid(thread_pid)") removed the reference around proc_flush_pid().
It assumed that free_pids(post.pids) at the end of release_task() would
keep thread_pid alive until then.
That assumption is wrong. __change_pid() only records a detached PID in
post.pids when pid_has_task() is false for every PIDTYPE. If another task
still uses the exiting task's PID as its process group or session ID,
__unhash_process() removes the exiting task's PIDTYPE_PID link but leaves
the PID out of post.pids. release_task() therefore holds no reference to
it after dropping tasklist_lock.
The other task can then remove the remaining PIDTYPE links. Its
free_pids() call schedules delayed_put_pid(), and the RCU callback can free
the PID before the first release_task() reaches proc_flush_pid().
An unprivileged reproducer races wait4(-1) against setsid() to trigger this
ordering. Three of three fresh v7.2 KASAN boots reported:
BUG: KASAN: slab-use-after-free in
proc_invalidate_siblings_dcache+0x3e2/0x3f0
Read of size 8 by task h7_pid_reaper/1921
Call Trace:
proc_invalidate_siblings_dcache
release_task
wait_consider_task
__do_wait
do_wait
kernel_wait4
Freed by task 0:
kmem_cache_free
put_pid
delayed_put_pid
rcu_core
Last potentially related work creation:
__call_rcu_common
free_pids
ksys_setsid
KASAN identified a 144-byte object from the pid cache and located the bad
read 80 bytes into the freed object, matching pid->inodes. With an
explicit reference, three of three fresh boots completed without a KASAN
report. The concurrent RCU callback dropped its reference while
proc_flush_pid() was protected, and the balancing put_pid() performed the
final free afterward.
Take a reference before __unhash_process() clears p->thread_pid and release
it after proc_flush_pid() completes.
A tested source reproducer is available privately on request. No
controlled read or write, information leak, or privilege escalation is
claimed. The mainline patch applies directly to v6.19.y and newer;
v6.16.y through v6.18.y need a context-adjusted backport. |
| In the Linux kernel, the following vulnerability has been resolved:
net: mpls: clear inner_protocol when the last label is popped
skb_mpls_push() records the pre-encapsulation network header once, gated
on !skb->inner_protocol. skb_mpls_pop() never clears that record, so it
outlives the encapsulation it describes.
Open vSwitch can then re-push MPLS onto a packet whose
inner_network_header still points at the older, deeper offset: push a
label, pop every label, recirculate (ovs_flow_key_update() re-derives
key->eth.type and resets network_header, but leaves inner_*), then push
again. ovs_fragment() trusts the record:
skb->network_header = skb->inner_network_header;
so skb_network_offset() goes negative. The bound check is signed:
if (skb_network_offset(skb) > MAX_L2_LEN)
a negative offset passes it, and prepare_frag() widens the value:
unsigned int hlen = skb_network_offset(skb);
memcpy(&data->l2_data, skb->data, hlen);
which is a ~4GiB memcpy out of a 30-byte per-CPU buffer.
Reproduced on v7.3-rc1. RDX is the truncated length, (unsigned int)(-8):
BUG: unable to handle page fault for address: ffffe8ffffc16000
#PF: supervisor write access in kernel mode
Oops: 0002 [#1] SMP KASAN NOPTI
RIP: 0010:memcpy+0x8/0x20
RDX: 00000000fffffff8 RSI: ffff888105d732db RDI: ffffe8ffffc16000
prepare_frag+0x3df/0x4e0
ovs_fragment+0x589/0x7e0
do_output+0x4ce/0x5e0
do_execute_actions+0x55d2/0x7b30
ovs_execute_actions+0xea/0x450
Same root-cause shape as commit 975b5b067f52 ("ipv6: sr: restore network
header before routing and forwarding"): a stale network header offset
reaching a consumer that widens it. Here it originates in the MPLS
push/pop path.
Clear inner_protocol once the packet is no longer MPLS, so a later push
re-records the current header. net/sched/act_mpls.c is the only other
skb_mpls_pop() caller and gets the same fix; sch_frag.c saves and
restores inner_protocol around fragmentation in the same way OVS does. |
| In the Linux kernel, the following vulnerability has been resolved:
net: openvswitch: fix use-after-free of the flow table mask array
tbl_mask_array_realloc() retires the old mask_array before it stops being
reachable:
old = ovsl_dereference(tbl->mask_array);
if (old) {
...
call_rcu(&old->rcu, mask_array_rcu_cb);
}
rcu_assign_pointer(tbl->mask_array, new);
call_rcu() only waits for read-side critical sections already in flight.
tbl->mask_array still points at old between the call_rcu() and the
rcu_assign_pointer(), so a reader entering ovs_flow_tbl_lookup_stats() in
that window picks up old in a fresh critical section that the pending
grace period does not cover.
tbl_mask_array_realloc() runs in process context under ovs_mutex, so the
window is preemptible and can outlast the grace period. Then
mask_array_rcu_cb() frees old before the swap runs:
BUG: KASAN: slab-use-after-free in flow_lookup.constprop.0+0x2bf/0x2f0
Read of size 8 at addr ffff888020b3e018 by task poc/741
flow_lookup.constprop.0+0x2bf/0x2f0
ovs_flow_tbl_lookup_stats+0x4a3/0x5c0
ovs_dp_process_packet+0x19c/0x710
ovs_vport_receive+0x243/0x390
internal_dev_xmit+0x81/0x170
Freed by task 728:
kfree+0x16a/0x4e0
rcu_core+0x853/0x1030
Publish the new array before retiring the old one. The kfree_rcu() that
call_rcu() replaced ran after the swap. |
