Vulnerabilities exploitable today
364,217in current view
Single score combining CVSS, KEV membership and EPSS. Every CVE with its own record — timeline from publication to active exploitation.
In KEV catalog1,674
New KEV · 24H0
Exploit Today ≥ 701,611
Distribution · last window
- Critical2,615
- High10,698
- Medium5,947
- Low568
Window
Severity
Flags
CVECVSSEPSSKEVRExploitTitleMod.
CVE-2026-74636——
———In the Linux kernel, the following vulnerability has been resolved:
tracing: Fix race between update_event_fields and, event_define_fields
The following sequence may leads race between event_define_fields()
and update_event_fields():
CPU0 (loads module A) CPU1 (loads module B)
=============================== ===============================
load_module(A) load_module(B)
notifier_call_chain notifier_call_chain
trace_module_notify trace_module_notify
mutex_lock(&event_mutex) trace_event_update_all()
trace_module_add_events(A) down_write(&trace_event_sem)
__register_event(call_A)
__add_event_to_tracers(call_A)
event_define_fields(call_A)
for each f: list_for_each_entry(field,
list_add(&f->link, &class->fields, link)
&class->fields) field = class->fields->next;
Where access to the class->fields is not protected by the event_mutex in
trace_event_update_all().
This produces the following panic:
Unable to handle kernel access ... at virtual address 0000000000000018
pc : update_event_fields+0xf8/0x368
Call trace:
update_event_fields+0xf8/0x368
trace_event_update_all+0x7c/0x2b4
trace_module_notify+0x4c/0x1dc
notifier_call_chain+0x84/0x168
blocking_notifier_call_chain_robust+0x64/0xd4
load_module+0x10c8/0x123c
__arm64_sys_finit_module+0x230/0x31c
Fix by taking event_mutex in trace_event_update_all() before
trace_event_sem.4hCVE-2026-74637——
———In the Linux kernel, the following vulnerability has been resolved:
perf/core: Fix group leader use-after-free after sibling detach
perf_group_detach() handles leader and sibling detach differently. When the
group leader is detached, all siblings are promoted to singleton events and
their group_leader pointer is reset to themselves. When a sibling is
detached, it is removed from the leader's sibling_list, but its
group_leader pointer is left pointing at the old leader.
That is harmless when the sibling is being closed and freed immediately, as
in the DETACH_DEAD path. It is not safe when the sibling is detached but
kept alive, such as during CPU hotplug with DETACH_GROUP. In that case the
sibling is removed from the context, while its file descriptor can still
keep it alive.
A typical failing sequence is:
- A group contains leader L and sibling S.
- CPU hot-unplug detaches S with DETACH_GROUP, removing it from
L->sibling_list but leaving S->group_leader == L.
- L is later closed and freed.
- A PERF_IOC_FLAG_GROUP ioctl on S follows S->group_leader and
dereferences the freed leader.
This was reproduced by running the perf event fuzzer, CPU hotplug, and a
stress workload concurrently:
Unable to handle kernel paging request at virtual address 006b6b6b6b6b6cdb
CPU: 2 PID: 12489 Comm: perf_fuzzer 6.18.7 PREEMPT
pc : perf_ioctl+0x34c/0xc68
x20: ffffff89a3fa2c70 x8 : 6b6b6b6b6b6b6b6b
Code: 943c4a0e 340047a0 f9404a94 f9411e88 (f940b908)
Call trace:
perf_ioctl+0x34c/0xc68 (P)
__arm64_sys_ioctl+0xa0/0xf4
invoke_syscall+0x58/0xe4
el0_svc_common+0xa8/0xdc
do_el0_svc+0x1c/0x28
el0_svc+0x40/0xc0
el0t_64_sync_handler+0x68/0xdc
el0t_64_sync+0x1c4/0x1c8
The fault happened in perf_ioctl(), where perf_event_for_each() follows
the stale group_leader pointer and perf_event_for_each_child() then
dereferences the freed leader's context.
Fix the use-after-free by promoting the detached sibling to a singleton.
Also fix __event_disable() cgroup accounting and event state change.4hCVE-2026-74638——
———In the Linux kernel, the following vulnerability has been resolved:
drm/v3d: Serialize the scheduler timeout handlers
V3D exposes several independent hardware queues (BIN, RENDER, TFU and
CSD) but has only a single, global reset. A timeout on any one queue
therefore has to stop, reset and restart the schedulers of every other
queue as well. That makes concurrent timeout handlers unsafe.
`reset_lock` was never able to make them safe, as a driver-side lock can
only cover the driver's &drm_sched_backend_ops.timedout_job callback.
The scheduler handles the timed out job and its pending list around that
callback, outside of the driver's control, so a global reset triggered
by one queue can still interfere with another queue that is in the
middle of handling a timeout of its own.
Consequently, if a reset happens in the CSD queue while a CL-intensive
application is running, the global reset stops and restarts the CL
queue's scheduler while that queue is handling a timeout of its own. As
drm_sched_stop() and drm_sched_start() subtract and add the credits of
every job sitting on the pending list of the scheduler they are called
on, and as the CL queue's handler concurrently takes its job off that
same list and puts it back, the stop and the start no longer see the
same set of jobs. The CL queue is left with more credits in flight than
its limit:
[ 327.302739] ------------[ cut here ]------------
[ 327.302744] WARNING: CPU: 2 PID: 43 at drivers/gpu/drm/scheduler/sched_main.c:102 drm_sched_run_job_work+0x238/0x4d0 [gpu_sched]
[ 327.302884] CPU: 2 UID: 0 PID: 43 Comm: kworker/u16:1 Not tainted 6.18.39-v8-16k+ #3 PREEMPT
[ 327.302889] Hardware name: Raspberry Pi 5 Model B Rev 1.0 (DT)
[ 327.302893] Workqueue: v3d_bin drm_sched_run_job_work [gpu_sched]
[ 327.302984] Call trace:
[ 327.302987] drm_sched_run_job_work+0x238/0x4d0 [gpu_sched] (P)
[ 327.302997] process_scheduled_works+0x180/0x3d0
[ 327.303010] worker_thread+0x268/0x3e8
[ 327.303016] kthread+0x140/0x250
[ 327.303022] ret_from_fork+0x10/0x20
[ 327.303031] ---[ end trace 0000000000000000 ]---
From that point on, the credit count of the CL queue is broken, causing
a complete GPU hang and UI freeze.
The DRM scheduler already provides a mechanism to serialize the timeout
handlers of different schedulers: an ordered workqueue passed as
drm_sched_init()'s @timeout_wq parameter. By default, each scheduler
queues its timeout work on the system workqueue, which runs the handlers
concurrently. Give all of the queues a shared ordered workqueue instead,
as recommended by the DRM scheduler documentation for hardware that has
distinct queues but resets globally.4hCVE-2026-74639——
———In the Linux kernel, the following vulnerability has been resolved:
ALSA: us144mkii: re-anchor capture URBs on resubmission
capture_urb_complete() resubmits each capture URB without anchoring it:
usb_get_urb(urb);
ret = usb_submit_urb(urb, GFP_ATOMIC);
Anchoring is a property of a submission, not of the URB. The giveback
path calls usb_unanchor_urb() before urb->complete(), so an URB
resubmitted from its own completion handler is off the anchor. The
capture URBs are anchored once, at stream start, so from the first
completion onward tascam->capture_anchor is empty.
tascam_free_urbs(), tascam_disconnect(), tascam_suspend() and the
stop-work path all call usb_kill_anchored_urbs(&tascam->capture_anchor)
to reap the capture URBs before anything is freed. With the anchor empty
those calls return immediately and the URBs stay queued on the host
controller.
tascam_free_urbs() then returns the capture transfer buffers with
usb_free_coherent(), and snd_card_free() releases the snd_card
allocation that embeds tascam (card->private_data). The controller
completes the queued URBs afterwards, writing device-supplied data into
the freed transfer buffer, and capture_urb_complete() dereferences the
freed driver object.
KASAN on 7.2.0-rc5 (arm64):
BUG: KASAN: slab-use-after-free in dummy_timer
Write of size 512 at addr ffff000015b62000
__asan_memcpy
dummy_timer
hrtimer_run_softirq
Allocated by task 64:
usb_alloc_coherent
tascam_alloc_urbs
tascam_probe
Freed by task 170:
usb_free_coherent
tascam_free_urbs
tascam_disconnect
usb_unbind_interface
BUG: KASAN: slab-use-after-free in capture_urb_complete
Read of size 4 at addr ffff0000170ee878
Freed by task 170:
release_card_device
snd_card_free
tascam_disconnect
Restore the usb_anchor_urb() between the reference count bump and the
resubmission. That also makes the handler's usb_unanchor_urb() failure
arm meaningful again and restores usb_kill_anchored_urbs() as a barrier
on the disconnect, suspend and stop-work paths.
The anchoring was removed on the premise that the URB is already anchored
from the initial submission, which does not hold once the first giveback
has run.
Discovered by XBOW, triaged by Baul Lee <baul.lee@xbow.com>4hCVE-2026-74640——
———In the Linux kernel, the following vulnerability has been resolved:
ALSA: FCP: fix OOB write in fcp_meter_ctl_get()
fcp_ioctl_set_meter_map() bounds the user-supplied Level Meter map size
by the driver's own limit of 255
if (map.map_size < 1 || map.map_size > 255 ||
map.meter_slots < 1 || map.meter_slots > 255)
return -EINVAL;
and passes it to fcp_add_new_ctl() as the control's channel count, where
it is stored as elem->channels.
Every control read writes into struct snd_ctl_elem_value, whose integer
array is declared long value[128], so the limit is 128, not 255.
fcp_meter_ctl_get() stores one 64-bit word per channel into that array
with no bound of its own:
for (i = 0; i < elem->channels; i++) {
int idx = private->meter_level_map[i];
int value = idx < 0 ? 0 : le32_to_cpu(resp[idx]);
ucontrol->value.integer.value[i] = value;
}
snd_ctl_elem_read_user() serves that object from
memdup_user(_control, sizeof(*control)), 1224 bytes on LP64 out of
kmalloc-2048. offsetof(struct snd_ctl_elem_value, value) is 72, so
element i is written at byte 72 + 8 * i and element 144 already lands
past the allocation. At map_size 255 the last store ends at byte 2112,
888 bytes past the object and 64 bytes into the adjacent slab object.
The stored words come from the device and meter_level_map[] selects
which word lands in which slot, so extent and contents are both
controlled.
The core does not catch this. snd_ctl_check_elem_info() is reached only
from __snd_ctl_elem_info(), which snd_ctl_elem_read() calls under
CONFIG_SND_CTL_DEBUG; without that option snd_ctl_skip_validation() is a
compile-time true. __snd_ctl_add_replace() validates kcontrol->count and
never inspects elem->channels.
Installing an oversized map needs CAP_SYS_RAWIO, but the control outlives
the hwdep descriptor that created it, so the out-of-bounds stores are
issued by any process able to read controls on /dev/snd/controlC0.
KASAN on 7.2.0-rc5 (arm64), triggered by an unprivileged control read:
BUG: KASAN: slab-out-of-bounds in fcp_meter_ctl_get
Write of size 8 at addr ffff000017af04c8 by task fcp_trigger/185
__asan_store8
fcp_meter_ctl_get
snd_ctl_elem_read
snd_ctl_ioctl
Allocated by task 185:
memdup_user
snd_ctl_ioctl
The buggy address is located 0 bytes to the right of
allocated 1224-byte region [ffff000017af0000, ffff000017af04c8)
Bound the map size by the ABI limit rather than by 255, and bound the
store loop at the sink so it cannot run past the value array whatever
elem->channels holds.
Discovered by XBOW, triaged by Baul Lee <baul.lee@xbow.com>4hCVE-2026-74641——
———In the Linux kernel, the following vulnerability has been resolved:
ALSA: usx2y: bound the hwdep mmap fault offset
snd_us428ctls_vm_fault() turns the faulting page offset into a kernel
address with no bound of any kind:
offset = vmf->pgoff << PAGE_SHIFT;
vaddr = (char *)(...)->us428ctls_sharedmem + offset;
page = virt_to_page(vaddr);
get_page(page);
vmf->page = page;
return 0;
snd_us428ctls_mmap() checks only the length of the mapping, never the
offset, and us428ctls_sharedmem is a single page from
alloc_pages_exact(). For a character device file_mmap_size_max()
returns ULONG_MAX, so the mm layer imposes no ceiling either. Every page
offset above zero resolves to a struct page outside the object, and the
handler installs it into the caller's address space read-write; the vma
is not marked read-only.
The caller picks the page frame with a single mmap() argument and gets
read-write access to a page of kernel memory it does not own; an offset
that lands in an unpopulated vmemmap region oopses instead.
A process that can open the hwdep node of an attached US-X2Y reaches
this after loading the FPGA image through the same node; no capability
check is involved.
On 7.2.0-rc5 (arm64), mmap() with a large offset:
Unable to handle kernel paging request at virtual address fffffdffc45d5ac8
pc : snd_us428ctls_vm_fault+0x68/0x140 [snd_usb_usx2y]
Call trace:
snd_us428ctls_vm_fault+0x68/0x140 [snd_usb_usx2y]
__do_fault
__handle_mm_fault
handle_mm_fault
el0_da
Reject any offset outside the shared region. The pcm hwdep handler in
usx2yhwdeppcm.c computes its address the same way and needs the same
bound.
Discovered by XBOW, triaged by Baul Lee <baul.lee@xbow.com>4hCVE-2026-74642——
———In the Linux kernel, the following vulnerability has been resolved:
ALSA: usb: Fix UAF at delayed release of MIDI2 EPs
The recent fix for UAF in ump_to_endpoint() caused another UAF because
it tries to dereference the UMP endpoint object, but this might be
executed at a delayed context where the endpoint has been already
released.
Add private_free to clear the associated data for avoiding the further
dereference for delayed releases.4hCVE-2026-74643——
———In the Linux kernel, the following vulnerability has been resolved:
samples/damon/mtier: error out for zero quota goal target values
Patch series "mm/damon: avoid division by zero from damos_quota_score()".
DAMON_SAMPLE_MTIER and DAMON_LRU_SORT allow the user to trigger division
by zero in damos_quota_score(). Avoid it by adding parameters validation
checks.
This patch (of 2):
damos_quota_score() can trigger division by zero if the target_value is
zero. DAMON_SAMPLE_MTIER lets users set the target_value via
node0_mem_{used,free}_bp parameters. It doesn't guard zero value case,
though. As a result, users can trigger division by zero. Fix the issue
by returning an error when the user tries to start DAMON with zero
node0_mem_{used,free}_bp parameter values.
DAMON_SAMPLE_MTIER is just a sample module, but the consequence is quite
bad. Also the zero node0_mem_free_bp parameter might look like a
reasonable setup to some users. Hence, the issue might really happen in
the real world.
One reliable way to reproduce the issue is like below:
# cd /sys/module/damon_sample_mtier/parameters
# echo 4096 > node0_start_addr
# echo 8192 > node0_end_addr
# echo 8192 > node1_start_addr
# echo 81920 > node1_end_addr
# echo 0 > node0_mem_free_bp
# echo Y > enabled
# dmesg -w
[...]
[18792.235916] Oops: divide error: 0000 [#1] SMP NOPTI
[...]
[18792.242787] RIP: 0010:damos_quota_score+0x6f/0x480
[...]
This issue was discovered [1] by Sashiko.4hCVE-2026-74644——
———In the Linux kernel, the following vulnerability has been resolved:
mm/damon/ops-common: putback folios on invalid migrate nid
damon_pa_migrate() and damos_va_migrate() isolate folios into a local list
and then call damon_migrate_pages(). When target_nid is invalid
(including the scheme default NUMA_NO_NODE / -1), damon_migrate_pages()
returns early without putting the folios back to the LRU.
Callers then discard the list head while those folios remain isolated with
an extra reference taken by folio_isolate_lru(). The pages stay off the
LRU for as long as the mapping exists (anon active+inactive counts drop
while RSS does not), and the leftover references can pin the pages after
the mapping is gone.
Put the folios back on the invalid-nid path so ignored migration requests
still return them to the LRU.4hCVE-2026-74645——
———In the Linux kernel, the following vulnerability has been resolved:
mm/damon/lru_sort: error out for >10000 active_mem_bp
damos_quota_score() can trigger division by zero if the target value is
zero. DAMON_LRU_SORT lets users set the target value for the hot memory
scheme via active_mem_bp parameter. It avoids setting it as the target
value if the parameter value is zero. However, it also sets the cold
memory scheme with a target value that is calculated as '10000 -
active_mem_bp + 2'. Hence, if a user sets active_mem_bp 10002, the cold
memory scheme's quota goal target value can be zero. As a result,
division by zero can be triggered. Fix by returning an error when the
user tries to start DAMON with >10000 active_mem_bp parameter value.
It makes no sense to set active_mem_bp with 10002. It also requires
module parameters write permission to reproduce the issue. That said, the
consequence is quite bad.
One reliable way to reproduce the issue is like below:
# cd /sys/module/damon_lru_sort/parameters
# echo 1000 > wmarks_high
# echo 995 > wmarks_mid
# echo 0 > wmarks_low
# echo 10002 > active_mem_bp
# echo Y > enabled
# dmesg -w
[...]
[ 597.421247] Oops: divide error: 0000 [#1] SMP NOPTI
[ 597.428848] RIP: 0010:damos_quota_score+0x6f/0x480
This issue was discovered [1] by Sashiko.4hCVE-2026-74646——
———In the Linux kernel, the following vulnerability has been resolved:
misc: fastrpc: take fl->lock when moving mmaps on interrupted invoke
When an invoke is interrupted by a signal,
wait_for_completion_interruptible() returns -ERESTARTSYS and
fastrpc_internal_invoke() moves every buffer from fl->mmaps onto
cctx->invoke_interrupted_mmaps. This list_del()/list_add_tail() walk
runs without holding fl->lock, the lock that serialises fl->mmaps in
fastrpc_req_mmap() and fastrpc_req_munmap() everywhere else.
Take fl->lock around the move, matching every other fl->mmaps accessor.4hCVE-2026-74647——
———In the Linux kernel, the following vulnerability has been resolved:
misc: fastrpc: Remove buffer from list prior to unmap operation
fastrpc_req_munmap_impl() is called to unmap any buffer. The buffer is
getting removed from the list after it is unmapped from DSP. This can
create potential race conditions if multiple threads invoke unmap
concurrently, where one thread may remove the entry from the list while
another thread's unmap operation is still ongoing.
Fix this by removing the buffer entry from the list before calling the
unmap operation. If the unmap fails, the entry is re-added to the list
so that userspace can retry the unmap, or alternatively, the buffer
will be cleaned up during device release when the DSP process is torn
down and all DSP-side mappings are freed along with remaining buffers
in the list.4hCVE-2026-74648——
———In the Linux kernel, the following vulnerability has been resolved:
staging: rtl8723bs: validate monitor transmit frame lengths
rtw_cfg80211_monitor_if_xmit_entry() removes the radiotap header and
then reads the 802.11 frame control field without checking that a base
802.11 header remains.
The data path also pulls the calculated 802.11, QoS and SNAP header
span before confirming that the skb contains it. A truncated frame can
therefore cause out-of-bounds reads or leave insufficient data for the
Ethernet address writes.
Reject frames that do not contain the base 802.11 header and data
frames that do not contain their complete calculated header span.4hCVE-2026-74649——
———In the Linux kernel, the following vulnerability has been resolved:
staging: rtl8723bs: fix missing shared-key auth challenge length check
The WEP shared-key authentication handler uses the challenge-text
element's attacker-controlled length without checking it against the
fixed 128-byte chg_txt buffer.
In OnAuthClient() the length from rtw_get_ie() - up to 255 - is used
to perform memcpy() into the 128-byte pmlmeinfo->chg_txt, so a
malicious AP sending a malformed WLAN_EID_CHALLENGE element can
overflow/underfill chg_txt by up to 127 bytes. It is reachable over the
air, before association, during shared-key authentication. In the case
of an overflow, the driver can write out of bounds. In the case of an
underfill, the driver can echo stale buffer memory.
The challenge text is defined to be exactly 128 octets, which is
already provided as the WLAN_AUTH_CHALLENGE_LEN define; require the
element to be exactly that length before use.4hCVE-2026-74650——
———In the Linux kernel, the following vulnerability has been resolved:
staging: rtl8723bs: fix OOB read in WMM_param_handler()
WMM_param_handler() copies a fixed-size WMM parameter element out of a
received information element without checking that the element is long
enough, causing an out-of-bounds read for a short WMM IE.
The handler reads sizeof(struct WMM_para_element) (18) bytes at
pIE->data + 6, so it requires pIE->length to be at least 24
(WLAN_WMM_LEN), but it never validates the length. Two of its three
callers reach it after matching only the WMM OUI: OnAssocRsp() in
rtw_mlme_ext.c matches a 6-byte OUI, and join_cmd_hdl() matches a
4-byte OUI, before calling the handler. A vendor-specific IE carrying
the WMM OUI but a length between 6 and 23, placed in an association
response or in the IE blob handed to join_cmd_hdl(), passes the OUI
check and then makes the memcmp() and memcpy() at pIE->data + 6 read
past the end of the element. OnAssocRsp() parses a frame received from
the AP, so this is reachable from a remote peer.
The remaining caller in rtw_wlan_util.c already guards the handler with
"pIE->length == WLAN_WMM_LEN". Move the equivalent check into the
handler itself so every caller is covered; the sibling IE handlers in
the same parsing loop (HT_caps_handler(), HT_info_handler(),
ERP_IE_handler()) likewise bound their accesses by pIE->length.4hCVE-2026-74651——
———In the Linux kernel, the following vulnerability has been resolved:
staging: rtl8723bs: fix OOB read in rtw_get_wpa_ie()
rtw_get_wpa_ie() reads bytes at fixed offsets into a vendor-specific
information element without checking that the element is long enough,
causing an out-of-bounds read for a short trailing IE.
The function locates a vendor-specific IE (EID 221) with rtw_get_ie()
and then compares a 4-byte OUI+type at pbuf + 2 and reads a 2-byte
version word at pbuf + 6. Those accesses require the IE body to be at
least 6 bytes, but rtw_get_ie() only guarantees that the element fits
within the buffer; it does not enforce a minimum body length. A
vendor-specific IE whose length byte is 0 to 5, placed at the end of
the buffer, therefore makes these reads run past the end of the IE and
past the end of the buffer itself.
The buffer holds information elements taken from received management
frames and from the IE blob passed to rtw_cfg80211_set_wpa_ie(), which
is kmemdup'd to its exact length, so the read can run off the end of
the allocation.
The sibling helpers rtw_get_sec_ie(), rtw_get_wapi_ie() and
rtw_get_wps_ie() in this file already reject too-short vendor-specific
IEs before their OUI memcmp(); rtw_get_wpa_ie() was never brought in
line with them, and needs a minimum of 6 rather than 4 bytes because
of the version word. Add the missing length check.4hCVE-2026-74652——
———In the Linux kernel, the following vulnerability has been resolved:
serial: amba-pl011: cancel RS485 hrtimers after freeing IRQ
The RS485 trigger hrtimers are embedded in the devm-managed port and can
fire after it is freed. The IRQ handler can arm a timer, so free the IRQ
first and then cancel both timers.
Complete the RS485 stop without arming a timer, and cancel the timers
in remove() for the suspend-then-unbind path, where shutdown is not
called.
This issue was found by an in-house static analysis tool.4hCVE-2026-74653——
———In the Linux kernel, the following vulnerability has been resolved:
serial: 8250_of: clear stuck empty-FIFO RX-timeout on LPC32xx
The NXP LPC32xx UART (PORT_LPC3220) can latch an RX character-timeout
interrupt while the RX FIFO is empty: IIR reports UART_IIR_RX_TIMEOUT
(0x0c) but LSR.DR is clear. A character timeout is only cleared by
reading RHR, but serial8250_rx_chars() reads RHR only when LSR.DR is
set, so nothing ever clears the condition. The interrupt is
level-triggered and re-fires immediately, so on a single-core ARM926
the resulting interrupt storm livelocks the CPU.
It is reproducible when userspace repeatedly opens the front-panel port
(ttyS1): serial8250_do_set_termios() re-enables interrupts on unlock and
the handler then spins forever with iir=0xcc lsr=0x60 ier=0x05, tripping
the soft-lockup detector in serial8250_handle_irq_locked().
LPC32xx has no dedicated 8250 glue driver, it's driven by the generic
8250_of. Add a hardware specific handle_irq for PORT_LPC3220, wired up
in of_platform_serial_setup() the same way fsl8250_handle_irq is
installed. The handler follows dw8250_handle_irq(): on an RX timeout
with an empty FIFO (LSR.DR and LSR.BI clear) it does one throwaway RHR
read to clear the condition, then calls serial8250_handle_irq_locked().
No real received data is ever discarded, and it is a no-op on healthy
UARTs which never report a timeout with DR clear.
This is the same class of bug already worked around in other 8250 drivers;
see commit 424d79183af0 ("serial: 8250_dw: Avoid "too much work" from bogus rx timeout interrupt")
which reports the identical iir=0xcc/lsr=0x60. See also
UART_RX_TIMEOUT_QUIRK in 8250_omap, and the note in 8250_bcm7271.4hCVE-2026-74654——
———In the Linux kernel, the following vulnerability has been resolved:
serial: 8250_dma: Clear stale RX state on shutdown
serial8250_release_dma() terminates RX DMA and releases the channel, but
leaves rx_running set. If the port is closed while an RX transfer is
active, the stale state remains while rxchan is NULL until the channel is
requested again on the next open.
The DesignWare BUSY workaround added by commit a7b9ce39fbe4
("serial: 8250_dw: Ensure BUSY is deasserted") calls
serial8250_rx_dma_flush() from the LCR write path during startup. This
happens before serial8250_request_dma() obtains a new RX channel. On
reopen, the stale rx_running state therefore makes the flush path pass a
NULL channel to dmaengine_pause(), causing a kernel Oops.
Clear rx_running after terminating RX DMA, matching the TX cleanup. Also
make the flush helper return if the DMA object or RX channel is not
available so startup and teardown paths cannot pass a NULL channel to the
DMAengine API.4hCVE-2026-74655——
———In the Linux kernel, the following vulnerability has been resolved:
serial: qcom-geni: fix TX DMA buffer flush
When transmit flushing a qcom-geni UART during an ongoing TX DMA, the
UART gets stuck infinitely repeating corrupted TX DMA frames.
The DMA-mode uart_ops does not provide a flush_buffer callback, so an
in-flight transfer can complete after serial core has reset the transmit
kfifo, underflowing its length and resubmitting page-sized transfers
indefinitely. Add one that stops the transfer and clears tx_remaining
and tx_queued.
The stop path was also broken: it unmapped the buffer while the serial
engine could still read it, and never reset the TX DMA state machine.
Cancel the main sequencer command first, then reset the state machine
and wait for it before unmapping. Drop the early return so a pending
mapping is also cleaned up when the main command is inactive.
The bug can be triggered from userspace with a large write immediately
followed by TCOFLUSH. A following tcdrain will hang forever. The bug was
reproduced and this fix was validated on Arduino Uno Q (QRB2210)
using /dev/ttyHS1.4hCVE-2026-74656——
———In the Linux kernel, the following vulnerability has been resolved:
ipv4: fix use-after-free in fib_nhc_update_mtu()
fib_nhc_update_mtu() walks the nexthop exception table under RTNL, but
RTNL does not serialize this walk with PMTU exception updates. The walk
uses rcu_dereference_protected() with a constant true condition without
holding fnhe_lock.
The following interleaving can therefore occur:
CPU 0 CPU 1
fib_nhc_update_mtu() update_or_create_fnhe()
load fnhe spin_lock_bh(&fnhe_lock)
fnhe_remove_oldest()
unlink fnhe
kfree_rcu(fnhe, rcu)
<quiescent state>
access fnhe after grace period
KASAN reported:
BUG: KASAN: slab-use-after-free in fib_nhc_update_mtu+0x3df/0x410
Read of size 8 at addr ffff888107d49000 by task poc/90
Call Trace:
fib_nhc_update_mtu+0x3df/0x410
fib_sync_mtu+0x7a/0xd0
fib_netdev_event+0x229/0x3f0
netif_set_mtu_ext+0x33a/0x570
dev_set_mtu+0x88/0x120
The same walk updates fnhe_pmtu and fnhe_mtu_locked. These fields form a
pair and other writers serialize them with fnhe_lock. RCU alone prevents
reclamation, but would still allow concurrent writers to leave a mixed
pair.
Walk the table under RCU and acquire fnhe_lock only while updating each
exception. RCU keeps the current entry alive while the short critical
section serializes its paired PMTU fields. This avoids holding the global
lock while scanning all 2048 buckets for every nexthop.4hCVE-2026-74657——
———In the Linux kernel, the following vulnerability has been resolved:
ipv4: Fix fib_nlmsg_size() for RTA_VIA nexthops
fib_nlmsg_size() still estimates nexthop space as if every gateway is
encoded as an IPv4 RTA_GATEWAY attribute. IPv4 routes can also carry an
IPv6 gateway, which fib_nexthop_info() dumps as RTA_VIA.
As a result, route notifications can allocate an skb that is too small.
fib_dump_info() then fails with -EMSGSIZE and rtmsg_fib() hits the
WARN_ON() that marks such failures as a fib_nlmsg_size() bug. With
panic_on_warn set, this becomes a kernel panic.
Mirror the actual nexthop dump layout in fib_nlmsg_size(): account for
IPv6 nexthop gateways dumped as RTA_VIA, for the no-header rtnexthop
layout used inside RTA_MULTIPATH, and for RTA_FLOW only when it is
actually present.4hCVE-2026-74658——
———In the Linux kernel, the following vulnerability has been resolved:
futex: Prevent robust futex exit race some more
A robust futex unlock stores 0 over the whole futex value - wiping
FUTEX_WAITERS - and wakes a single waiter. That wakeup is a one-shot
notification: the protocol relies on its recipient to either acquire the
futex (and eventually unlock while aware of the remaining contention) or
re-arm FUTEX_WAITERS before sleeping again. If the woken waiter is killed
before it can do either, the kernel must jump in and wake the next task
down the line.
This is a known complication of the futex protocol with a previous
partial fix in commit ca16d5bee598 ("futex: Prevent robust futex exit
race"). Unfortunately, that fix is insufficient.
If a third task re-acquired the futex through the uncontended fast
path in the meantime, the notification is lost: robust exit processing
sees that it is owned by another task and does nothing, while the new
owner sees no FUTEX_WAITERS when it unlocks and wakes nobody.
The remaining waiters sleep forever behind a free futex:
A owns the futex, B and C sleep in FUTEX_WAIT
uval == A | FUTEX_WAITERS
A robust unlock: store 0, FUTEX_WAKE(1) wakes B
uval == 0
D fast path acquire: cmpxchg(0 -> D)
uval == D, no FUTEX_WAITERS
B killed before acting on the wakeup
B exit walk, pending op: owner D != B -> no action
D unlock: no FUTEX_WAITERS -> no wake
C sleeps forever
This is clearly a shortcoming in the implementation, which fails to keep
the FUTEX_WAITERS bit consistent.
Work around this by augmenting the robust list exit processing to also
perform the extra wakeup if the futex word is owned by another thread but
FUTEX_WAITERS is not set.
This does not fix the problem of a non-contended take over/release and free
sequence, which has been discussed for years and has been addressed by
commit 3ca9595d9fb6 ("futex: Add support for unlocking robust futexes") and
subsequent changes, but failed to take the problem described above into
account.
A more complete solution which is based on the in kernel unlock of
contended robust futexes has been discussed in the context of this change
and should show up in mainline sooner than later.
[ tglx: Amend change log slightly and fixup coding style ]4hCVE-2026-74659——
———In the Linux kernel, the following vulnerability has been resolved:
net: bridge: mrp: fix uninitialised bytes on the wire
br_mrp_alloc_test_skb() builds MRP test frames on an skb from
dev_alloc_skb(), which does not clear the linear data area. On the MRA
ring-role branch the sub-option TLV header is appended with
sub_tlv = skb_put(skb, sizeof(*sub_tlv));
sub_tlv->type = BR_MRP_SUB_TLV_HEADER_TEST_AUTO_MGR;
so sub_tlv->length is never written, and the two trailing alignment bytes
are appended with a bare skb_put() that does not clear them either. The
neighbouring oui and sub_opt regions are explicitly zeroed, so three
uninitialised bytes are left in every MRA MRP_Test frame that goes out.
Put the sub-option TLV header and the alignment padding in a single
skb_put_zero(), which clears both. The AUTO_MGR sub-TLV carries no
payload, so the zeroed length field is already the value it should have.4hCVE-2026-74660——
———In the Linux kernel, the following vulnerability has been resolved:
netfilter: ebt_nflog: pin the NFLOG backend
nf_log_unregister() runs after the per-net teardown so its final RCU
grace period also drains readers that obtained the logger from a per-net
binding. However, ebt_nflog passes an explicit ULOG log type to
nf_log_packet() without holding a reference on the selected logger module,
unlike the xt_NFLOG and nft_log frontends.
An ebtables nflog rule can therefore remain callable while nfnetlink_log
is unloaded. The resulting interleaving is:
CPU 0 CPU 1
nfnetlink_log_fini()
unregister_pernet_subsys()
kfree(nfnl_log_pernet(net))
ebt_nflog_tg()
nf_log_packet()
nfulnl_log_packet()
instance_lookup_get_rcu()
The global ULOG logger is still registered at this point, so CPU 1
dereferences the per-net state after CPU 0 has freed it. KASAN reported:
BUG: KASAN: slab-use-after-free in instance_lookup_get_rcu
Read of size 8 at addr ff110001052e6210 by task poc/92
Call Trace:
instance_lookup_get_rcu+0x1ce/0x1f0 [nfnetlink_log]
nfulnl_log_packet+0x248/0x2fb0 [nfnetlink_log]
nf_log_packet+0x204/0x300
ebt_nflog_tg+0x351/0x550
ebt_do_table+0xedf/0x22b0
Allocated by task 90:
__kmalloc_noprof+0x186/0x470
ops_init+0x6d/0x420
register_pernet_operations+0x2f6/0x670
register_pernet_subsys+0x23/0x40
Freed by task 93:
kfree+0x131/0x3c0
ops_undo_list+0x3e3/0x700
unregister_pernet_operations+0x232/0x490
unregister_pernet_subsys+0x1c/0x30
nfnetlink_log_fini+0x34/0x450 [nfnetlink_log]
Acquire the ULOG logger module reference when an ebt_nflog rule is
validated and release it when the rule is destroyed. Request the NFLOG
backend for legacy callers when needed, matching xt_NFLOG. This prevents
module teardown until all ebt_nflog rules have stopped using the logger.4hCVE-2026-74661——
———In the Linux kernel, the following vulnerability has been resolved:
mac802154: fix netdev use-after-free in beacon worker
mac802154_beacon_worker() reads local->beacon_req under RCU and derives
the sub-interface from the request, but then drops the RCU read lock and
continues to use both sdata and the embedded wpan_dev.
mac802154_stop_beacons_locked() cancels only pending beacon work, clears
local->beacon_req and frees the request. A beacon worker that is already
running can therefore continue after interface teardown and dereference
the freed netdev private area.
The scan worker already pins the netdev before leaving RCU. Apply the
same lifetime rule to the beacon worker: take a netdev reference while
the request is still protected by RCU, and release it on all paths that
continue after the reference is acquired.4hCVE-2026-74662——
———In the Linux kernel, the following vulnerability has been resolved:
inet: frags: publish queues before arming timer
inet_frag_create() arms the fragment queue timer before inserting the
queue into the fqdir rhashtable. If the namespace fragment timeout is
zero or negative, the timer can run before the queue is published.
The timer callback then marks the queue complete, tries to remove a node
that is not in the hash table yet, and drops the anticipated hash
reference. Creation can subsequently publish the completed queue without
restoring that reference, leaving a stale hash node after the caller drops
the remaining reference.
Publish the queue first and arm the timer while holding the queue lock.
This makes timer expiry wait until the queue is visible in the hash table,
so inet_frag_kill() can remove the node and balance the hash reference.4hCVE-2026-74663——
———In the Linux kernel, the following vulnerability has been resolved:
net/sched: reject overly deep qdisc hierarchies
Deep qdisc hierarchies can lead to excessive recursion in qdisc tree
walkers and exhaust the kernel stack. The existing loop check does not
cover the create-and-graft path, so a hierarchy can still be extended by
creating a new child qdisc below an already deep parent.
Store the hierarchy depth in struct Qdisc and update it when qdiscs are
grafted. Reject new child qdiscs once the parent is already at the maximum
allowed depth.4hCVE-2026-74664——
———In the Linux kernel, the following vulnerability has been resolved:
net: openvswitch: reallocate update replies for mismatched IDs
ovs_flow_cmd_new() preallocates the optional reply skb before it takes
ovs_mutex and before it knows which existing flow will be updated.
That is normally fine because the skb is sized from the request flow
identifier. That identifier also becomes the inserted flow's identifier.
For updates, however, a request with a UFID may miss the UFID lookup and
then fall back to the flow key lookup. That lookup can legitimately find
an existing key-identified flow. UFIDs are optional and the flow key is
the primary identifier.
For echoed replies, ovs_flow_cmd_fill_info() writes the matched flow's
identifier, not the request identifier used for the preallocation. A short
request UFID can therefore leave too little room for the key identifier.
The fill can then fail with -EMSGSIZE and hit the BUG_ON(error < 0) in the
update path.
Once the update target has been resolved, reallocate the reply skb if the
matched flow needs a larger reply than the request identifier allowed. Do
this before replacing the actions so the request can still fail cleanly if
the rare extra allocation fails.4hCVE-2026-74665——
———In the Linux kernel, the following vulnerability has been resolved:
net: fix skb length accounting after generic XDP frag adjustment
Generic XDP exposes non-linear skb fragments through an xdp_buff. If an
XDP program adjusts the fragment area, bpf_prog_run_generic_xdp() copies
xdp_frags_size back to skb->data_len but leaves skb->len containing the
old fragment contribution.
After a fragment shrink, this makes skb_headlen() larger than the actual
linear area. In the reproduced UDP receive path, __skb_datagram_iter()
copied 1024 bytes past the actual linear tail to userspace, starting at
struct skb_shared_info. The copied bytes included the affected skb's
nr_frags, xdp_frags_size and a kernel pointer from
skb_shinfo(skb)->frags[0]. Real packet data was displaced by the same
amount and truncated at the end.
Subtract the old data_len before replacing it and add the new data_len
afterwards, keeping skb->len and skb->data_len synchronized.
A 60000-byte UDP datagram on a veth pair with MTU 64000 was shortened by
1024 bytes from its fragment area. Before the fix, all 10 runs produced
corrupted payloads. After the fix, all 10 runs matched the expected
payload exactly.4hCVE-2026-74666——
———In the Linux kernel, the following vulnerability has been resolved:
packet: synchronize pressure clearing with ring reconfiguration
packet_set_ring() updates the RX ring state under sk_receive_queue.lock,
but used to publish the tpacket receive mode through po->prot_hook.func
after releasing that lock. packet_poll() and packet_recvmsg() can then
run the pressure clearing path after the ring has been cleared while
still seeing tpacket_rcv, causing __packet_rcv_has_room() to dereference
stale or NULL ring storage.
Move the existing receive hook assignment into the same
sk_receive_queue.lock section as the ring state update. Keep the
assignment otherwise unchanged, including on TX ring reconfiguration, to
avoid adding behavior changes that are not required for the fix.
Serialize packet_recvmsg() pressure clearing with the same queue lock
only after PACKET_SOCK_PRESSURE has been observed. If the flag is clear
and the socket has moved away from tpacket_rcv, packet_set_ring() has
already detached the socket and waited for synchronize_net(), so no new
packet input can set the flag again.
packet_poll() already holds sk_receive_queue.lock, so it uses the new
unlocked helper directly.4hCVE-2026-74667——
———In the Linux kernel, the following vulnerability has been resolved:
net/packet: reset the MAC header on the packet-socket transmit path
packet_parse_headers() resets the MAC header only for a SOCK_RAW frame
whose socket did not bind a protocol. A protocol-bound SOCK_RAW socket,
any SOCK_DGRAM frame, and the legacy SOCK_PACKET path therefore leave
skb->mac_header unset here.
For frames sent via __dev_queue_xmit() this is harmless: it resets the
MAC header unconditionally. But the packet-socket PACKET_QDISC_BYPASS
path uses dev_direct_xmit(), which does not, so the frame reaches
ndo_start_xmit() with the MAC header unset. A driver that reads
eth_hdr(skb) on transmit then dereferences skb->head + (u16)~0, an
out-of-bounds access ~64 KiB past the head -- the same class fixed for
one consumer in commit f5089008f90c ("macsec: do not read an unset MAC
header in macsec_encrypt()").
packet_parse_headers() runs only on the transmit path, where skb->data
points at the start of the L2 header for every packet-socket type
regardless of its length: SOCK_RAW and SOCK_PACKET carry a user-supplied
header and SOCK_DGRAM has one built by dev_hard_header(). Reset the MAC
header unconditionally, mirroring __dev_queue_xmit(), so the frame is
anchored on the bypass path too.
Found by 0sec (https://0sec.ai) using automated source analysis;
verified against source and matched to the macsec KASAN report in
f5089008f90c. Compile-tested.4hCVE-2026-74668——
———In the Linux kernel, the following vulnerability has been resolved:
packet: use consistent hard_header_len in TX_RING send path
tpacket_snd() reads dev->hard_header_len independently for skb
allocation and header construction in tpacket_fill_skb(). Concurrent
netdevice reconfiguration can therefore make the reserved headroom
smaller than the amount later pushed, or make copylen - hard_header_len
negative.
Snapshot hard_header_len once before processing ring frames and use it
for the frame limit, headroom allocation, copy length, and skb
construction. Pass the snapshot to tpacket_fill_skb().
The separate SOCK_DGRAM consistency problem between hard_header_len and
header_ops->create is not addressed here.4hCVE-2026-74669——
———In the Linux kernel, the following vulnerability has been resolved:
ipvs: clear IPv4 options after rebasing tunnel ICMP errors
ip_vs_in_icmp() rebases an skb from the outer ICMP packet to the
quoted original request before passing it to icmp_send(). However,
IPCB(skb)->opt still describes the outer IPv4 header.
A timestamp option in the outer header can therefore leave an offset
that points into the quoted transport header after the rebase.
__ip_options_echo() treats a byte at that stale location as the option
length and copies it into the fixed-size option storage on the
__icmp_send() stack, causing a stack out-of-bounds write.
Clear the stale option metadata after resetting the network header.
Keep the remaining control block fields, including the ingress
interface used by the ICMP response path.4hCVE-2026-74670——
———In the Linux kernel, the following vulnerability has been resolved:
ipvs: stop estimator after disabled calc phase
IPVS estimator kthread 0 starts with zeroed chain and tick limits until
its initial calculation phase completes. If network namespace teardown
clears ipvs->enable during that phase, ip_vs_est_calc_phase() can return
without installing positive limits.
The kthread can then continue into its main loop and drain
est_temp_list with zero chain_max, tick_max and est_max_count values.
Each enqueue consumes one available tick row, but est_count never
reaches the zero est_max_count value. After all rows are consumed, the
row lookup returns IPVS_EST_NTICKS and ip_vs_enqueue_estimator() writes
past the ticks and tick_len arrays.
Exit kthread 0 after the calculation phase if the kthread is stopping or
IPVS has been disabled. That keeps temporary estimators from being
drained after the limits failed to initialize.
Estimator kthreads can now self-exit before teardown or reload stops
kd->task. Keep an extra task reference after creation and release it
with kthread_stop_put(), so kd->task remains valid until the stop paths
consume that reference.4hCVE-2026-74671——
———In the Linux kernel, the following vulnerability has been resolved:
ima: fix out-of-bounds read in xattr_verify()
The digest-length check in xattr_verify() mixes int and size_t:
if (xattr_len - sizeof(xattr_value->type) - hash_start >=
iint->ima_hash->length)
sizeof() yields size_t, so the usual arithmetic conversions promote
the whole left-hand side to unsigned 64-bit before the subtraction
runs. For a truncated xattr this underflows instead of going negative:
a 1-byte IMA_XATTR_DIGEST_NG xattr (xattr_len == 1, hash_start == 1)
turns "1 - 1 - 1" into SIZE_MAX, which is trivially >= ima_hash->length.
The check then passes and the following memcmp() reads
iint->ima_hash->length bytes starting past the end of the buffer
vfs_getxattr_alloc() allocated for it.
Nothing upstream clamps xattr_len back into a safe range first:
ima_get_hash_algo() only special-cases xattr_len < 2 to pick a default
algorithm, and evm_verifyxattr() returns INTEGRITY_UNKNOWN rather than
failing when no HMAC key is loaded, so a truncated security.ima value
reaches the length check as-is.
Rewrite the comparison so every operand stays a signed int and no
implicit conversion to size_t can occur.4hCVE-2026-74672——
———In the Linux kernel, the following vulnerability has been resolved:
mm/vmalloc: acquire init_mm lock on huge vmap to avoid ptdump UAF
Patch series "mm: fix UAF caused by race between ptdump and vmap pgtable
freeing", v6.
Kernel page table walkers fall into two broad categories - those ranges
where no exclusion is required via walk_kernel_page_table_range_lockless()
and those where exclusion is required via walk_kernel_page_table_range()
or walk_page_range_debug().
The former category is used only by arm64 arch code operating on ranges it
both wholly owns and does not concurrently write.
The latter category consists of kernel page table walkers operating on
ranges that are wholly owned (but which need exclusion against concurrent
writers).
The lock used for exclusion is the mmap lock, and for kernel ranges this
is the mmap lock on init_mm.
ptdump is a special case being both the only user of
walk_page_range_debug(), and the only case in which it walks ranges it
does not own.
This presents a problem, as page tables may be freed under ptdump. And
indeed there is a use-after-free bug in the kernel as a result, which this
series addresses.
vmap promotes page tables to huge leaf entries where possible, freeing the
lower page table when it does. It does this with no meaningful locks held
against concurrent ptdump walks.
As a result, use-after-free can currently occur. This series addresses
the issue by having the vmap huge promotion logic acquire the mmap read
lock while both setting the huge page table entry and freeing the prior
leaf page table.
The ptdump code already acquires the mmap write lock, so by doing so we
ensure that the ptdump walker only ever observes either the huge page
table entry or the existing page table entry, and nothing is freed
underneath it.
A mitigation for this issue was already applied for arm64 in commit
fa93b45fd397 ("arm64: Enable vmalloc-huge with ptdump"), which this series
has to deal with carefully.
This mitigation resolves the issue by acquiring the mmap read lock on
init_mm on vmap page table free if a ptdump is in progress.
However the fix in this series would cause a deadlock if we were to simply
apply it for arm64 without also reverting the change.
This is because vmap may acquire the read lock before ptdump attempts to
acquire the write lock, which then gets queued, and rwsem starvation rules
mean that the (unacknowledged) nested mmap read lock in the arm64 code
would also block, meaning the original read lock is never released and
thus deadlock.
This series works around this by #ifndef CONFIG_ARM64'ing the mmap read
lock in vmap logic, then partially reverting commit fa93b45fd397 ("arm64:
Enable vmalloc-huge with ptdump"), keeping the enablement of huge vmap
support, and removing the ifdeffery with the partial revert patch.
There are related issues that are also addressed in this series:
* x86 page attribute logic, specifically Change Page Attributes (CPA),
implements a feature whereby huge ranges can be collapsed into huge leaf
entries. This can similarly cause a UAF when done in parallel with a
ptdump walk, so similarly acquire the init_mm mmap lock to avoid this.
* The CPA logic allows concurrent page table manipulation and CPA
collapse, meaning the former risks accessing a page table the latter
frees. Fix this by acquiring mmap write lock on init_mm across the
whole CPA collapse operation and read lock on the page table
manipulation.
* x86 and arm64 permit walks of non-kernel mm's (both allowing efi mm
walks, and in x86's case arbitrary mm's), so we ensure kernel mappings
remain stable by locking the init_mm as well as the mm being walked.
The ordering of patches is established for both strict dependencies (the
arm64 partial revert in particular has to be done after the vmap changes)
and logical ones (the non-kernel mm fix only makes sense once the vmap/CPA
fixes are in place).
This patch (of 3):
Currently there is a nasty ra
---truncated---4hCVE-2026-74673——
———In the Linux kernel, the following vulnerability has been resolved:
Input: evdev - fix information leak in evdev_pass_values()
In evdev_pass_values(), the input_event structure is allocated on the
kernel stack and populated field-by-field. However, it is never fully
initialized. On architectures where struct input_event contains explicit
or implicit padding (such as the 32-bit __pad field on SPARC64), these
padding bytes are left uninitialized.
When this event structure is subsequently passed to the client buffer
and later copied to userspace, the uninitialized padding bytes leak
kernel stack memory, potentially exposing sensitive information.
Similar issues exist in __evdev_queue_syn_dropped and __pass_event.
Fix this by explicitly zeroing the entire event structure with memset()
before populating its fields. This ensures all padding bytes are cleared
before the data crosses the security boundary.4hCVE-2026-74674——
———In the Linux kernel, the following vulnerability has been resolved:
mm: fix incorrect flush address in direct page table reclaim
When zap_pte_range reclaims a page table, it does:
pte_free_tlb(tlb, pmd_pgtable(pmdval), addr);
and this is unconditionally wrong: if this code executes, addr *always*
points one past the end of the range covered by the table. The addr
parameter is used to flush the TLB (really the paging-structure-cache)
to drop references to the to-be-freed table, and any architecture that
cares about the parameter will flush the wrong address. (But they'll
still free the correct page).
I think it's worth contemplating why the kernel works at all.
If we hit the offending line of code, we will first clear the PMD entry
(line 1954, zap_empty_pte_table), then we will issue pending flushes if
force_flush is set (tlb_flush_mmu_tlbonly(tlb)), then we will skip the
retry on line 1979 (phew!), and then we will do the offending
pte_free_tlb call. *Or* we will clear the PMD entry immediately before
pte_free_tlb (line 1983, zap_pte_table_if_empty).
If we have any pending flushes (i.e. we actually zapped any last-level
entries) at the time we clear the PMD entry, then the flush really ought
to flush all references to the table (Linus certainly seems to think it
will on all architectures [0]).
The condition under which we have no accumulated flushes at the time of
the clear is very complex (the whole zap_pte_range function has absurdly
complex control flow). If we do hit the bad case, then we will end up
clearing the PMD entry after the last time the range is flushed, and any
CPU is free to cache a reference to the (empty) page table. If this
happens due to an ordinary read or write, it would segfault, so it would
be rare. But the cache could be speculatively filled as well. Then
we'll flush the wrong address and then free and possibly reuse the
table.
On x86, even flushing the wrong address works on non-KPTI Intel systems
because INVLPG flushes *all* paging-structure-caches, not just the ones
for the target address. But INVPCID does not, and flush_tlb_one_user
will use INVPCID if it's available. And then we're toast. AMD systems
are more susceptible: we set the EFER.TCE bit, which makes even INVLPG
only flush the target address.
I think this might fix an issue in ripgrep reported here:
https://github.com/BurntSushi/ripgrep/issues/3494
[0] https://lore.kernel.org/all/CA+55aFzBggoXtNXQeng5d_mRoDnaMBE5Y+URs+PHR67nUpMtaw@mail.gmail.com/T/#u4hCVE-2026-74675——
———In the Linux kernel, the following vulnerability has been resolved:
vt: stabilize tty reference in kbd_keycode with tty_port_tty_get
kbd_keycode() reads vc->port.tty without acquiring a tty reference,
racing against con_shutdown() which clears port.tty under a different
lock. Use tty_port_tty_get()/tty_kref_put() to hold a proper reference
for the duration the tty pointer is needed.4h