CVE-2026-99796: io_uring/poll: cancel all matching requests in one pass
In the Linux kernel, the following vulnerability has been resolved:
iouring/poll: cancel all matching requests in one pass
With IORINGASYNCCANCELALL, ioasynccancel() calls iotrycancel() until nothing is found, and iopollcancel() cancels one request per call, starting each lookup at the beginning of the hash. A cancelled request stays hashed until its task work runs, so every lookup walks past the ones already cancelled. Cancelling K of N pending polls visits about KN/2 requests in one iouringenter(), with uringlock held.
A single CANCELFD|ALL on 80000 polls on one eventfd takes 17.06s, 0.02s with this change. With voluntary preemption the soft lockup watchdog fires:
watchdog: BUG: soft lockup - CPU#1 stuck for 26s! [repro:100] RIP: 0010:iocancelreqmatch+0x19/0xa0 iopollcancel+0x73/0x170 iotrycancel+0x7b/0xf0 ioasynccancel+0x41/0x110
Walk the hash once and cancel every match, as iocancelremove() does for futex and waitid. Cancel by userdata only needs its own bucket. ioasynccancel() still loops until nothing is found, so a request that io-wq parks in the poll hash meanwhile is still cancelled.
A cancel that matches one request now walks the whole hash twice instead of stopping at the match.
Affected Software
Remediation
Recommended actions to resolve this vulnerability, in priority order.
- Upgrade
Upgrade
linux/Kernelto a version that resolves this vulnerability.Fixed in 6.12.118 - Upgrade
Upgrade
linux/Kernelto a version that resolves this vulnerability.Fixed in 6.18.64 - Upgrade
Upgrade
linux/Kernelto a version that resolves this vulnerability.Fixed in 7.3.5 - Upgrade
Upgrade
debian/linuxto a version that resolves this vulnerability.Fixed in 7.3.5-1 - Compensating control
Change io_uring poll cancellation to walk the hash once and cancel every matching request in one pass, rather than repeatedly restarting lookups at the beginning of the hash.
Frequently Asked Questions
What workload is exposed to the excessive CPU hold time?
The issue is triggered when many pending io_uring poll requests match a single asynchronous cancellation operation, such as CANCEL_FD|ALL against 80,000 polls on one eventfd. The cancellation runs within one io_uring_enter() call while uring_lock is held.
What must an attacker or tenant be able to do to trigger the problem?
They need the ability to create a large number of pending io_uring polls and issue an IORING_ASYNC_CANCEL_ALL cancellation that matches them. The described proof of concept uses 80,000 polls associated with one eventfd.
How can administrators recognize a system affected during exploitation or reproduction?
On systems with voluntary preemption, the soft-lockup watchdog may report a CPU stuck for an extended period. The reported stack includes io_cancel_req_match, io_poll_cancel, io_try_cancel, and __io_async_cancel.