{"id":"CVE-2026-99268","kind":"vulnerability","aliases":["CVE-2026-99268"],"related":[],"title":"Linux kernel: io_uring ring pollers can sleep through pending completions","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nio_uring: wake ring pollers for local task_work\n\nWith IORING_SETUP_DEFER_TASKRUN completions are queued as local work, and\nonly become CQEs when the submitter enters the kernel to wait for events.\nFor an application that finds out via poll() or epoll on the ring fd when\nit is time to do that, io_uring_poll() reports EPOLLIN if local work is\npending.\n\nThat only helps if the work is there at the time of the call. When\nDEFER_TASKRUN was added, io_req_local_work_add() did io_cqring_wake(),\nwhich took care of tasks sleeping in poll() as well. Then polling got a\nwait queue of its own, ->poll_wq, and the wake up in\nio_req_local_work_add() was replaced by waking the submitter directly.\nSince then ->poll_wq is only woken when CQEs are committed, which for\nsuch a ring is done by the very task that is waiting to be told to:\n\n  submitter                           completion\n  poll(ring_fd)\n    io_uring_poll()  // nothing yet\n    schedule()\n                                      io_req_local_work_add()\n                                        // ->cq_wait_nr: nobody waits\n  // sleeps on\n\nWake ->poll_wq when the first entry is put on the work list. Like for the\neventfd, that must not be done right away if the work is queued from a\nwake up handler, as the chain of wake ups could get back to the wait\nqueue whose lock is held. Use an RCU callback in that case, with a\nreference to the ring.","summary":"With IORING_SETUP_DEFER_TASKRUN, a process polling the io_uring ring fd is not woken when completion work is queued, so it can sleep past work that is ready. The sources describe a missed wakeup and do not state an attacker capability.","severity":"unknown","cvss":null,"epss":null,"exploited":false,"exploited_at":null,"products":[],"ecosystems":[],"tags":["io_uring","kernel","linux","runtime"],"not_affected_if":[],"how_to_check":null,"fix":null,"status":"open","published_at":"2026-11-03T12:17:33+00:00","first_seen_at":"2026-11-03T12:20:40+00:00","triaged_at":"2026-11-03T12:22:38+00:00","updated_at":"2026-11-03T12:22:38+00:00","category":"runtime","confidence":0.4,"triage":{"at":"2026-11-03T12:22:38Z","model":"claude-haiku-5-5","output":{"fix":"unknown","reach":"local","title":"Linux kernel: io_uring ring pollers can sleep through pending completions","impact":"low","summary":"With IORING_SETUP_DEFER_TASKRUN, a process polling the io_uring ring fd is not woken when completion work is queued, so it can sleep past work that is ready. The sources describe a missed wakeup and do not state an attacker capability.","audience":"businesses","category":"runtime","nicknames":[],"who_cares":["linux","kernel","io_uring"],"confidence":0.4,"fix_status":"unknown","needs_auth":"unknown","worst_case":"unknown","analyst_note":"Check the NVD entry and the upstream kernel commit for a CVSS score, affected kernel ranges and whether the missed wakeup has any security impact beyond a hang.","how_to_check":"unknown","plain_action":"","needs_analyst":true,"not_affected_if":[],"who_is_affected":"Systems running a Linux kernel that uses io_uring with IORING_SETUP_DEFER_TASKRUN and applications that wait on the ring fd with poll() or epoll. The sources describe a wakeup bug and do not say who can exploit it. unknown","user_interaction":"unknown","needs_special_setup":"yes","severity_if_unknown":"unknown"},"cost_usd":0.000682,"input_tokens":4535,"output_tokens":458,"severity_before":"unknown"},"analyst_status":null,"analyst_note":null,"search":"'-2026':13A,16A '-99268':14A,17A 'add':159C,164C,199C,251C 'applic':112C 'attack':64B 'away':282C 'back':301C 'becom':98C 'call':154C 'callback':313C 'capabl':65B 'care':171C 'case':316C 'chain':295C 'commit':217C 'complet':11A,40B,90C,237C 'could':299C 'cq':252C 'cqes':99C,215C 'cqring':167C 'cve':12A,15A 'defer':26B,88C,156C 'describ':55B 'direct':206C 'done':224C,280C 'enter':103C 'entri':265C 'epol':119C 'epollin':135C 'event':109C 'eventfd':275C 'fd':35B,123C,240C 'find':114C 'first':264C 'follow':71C 'get':300C 'got':181C 'handler':292C 'held':309C 'help':143C 'io':3A,18B,32B,76C,131C,160C,166C,195C,241C,247C 'ior':24B,86C 'kernel':2A,20B,69C,105C 'like':272C 'linux':1A,21B,68C 'list':271C 'local':82C,94C,137C,162C,197C,249C 'lock':307C 'miss':57B 'must':277C 'nobodi':255C 'noth':244C 'nr':254C 'past':48B 'pend':10A,140C 'poll':30B,117C,133C,176C,180C,188C,209C,238C,243C,260C 'poller':6A,80C 'process':29B 'put':267C 'queu':43B,92C,287C 'queue':184C,305C 'rcu':312C 'readi':52B 'refer':319C 'replac':201C 'report':134C 'req':161C,196C,248C 'resolv':75C 'right':281C 'ring':5A,34B,79C,122C,222C,239C,322C 'runtim':22B 'schedul':246C 'setup':25B,87C 'sinc':207C 'sleep':8A,47B,174C,257C 'sourc':54B 'state':62B 'submitt':102C,205C,236C 'task':83C,173C,228C 'taskrun':27B,89C,157C 'time':127C,151C 'told':234C 'took':170C 'up':298C 'ure':4A,19B,33B,77C,132C,242C 'use':310C 'via':116C 'vulner':72C 'wait':107C,183C,231C,253C,256C,304C 'wake':78C,168C,192C,203C,259C,290C,297C 'wakeup':58B 'well':178C 'whose':306C 'woken':38B,213C 'work':41B,49B,84C,95C,138C,146C,163C,198C,250C,270C,285C 'wq':189C,210C,261C 'yet':245C","nicknames":[],"mitigations":[],"who_affected":"Systems running a Linux kernel that uses io_uring with IORING_SETUP_DEFER_TASKRUN and applications that wait on the ring fd with poll() or epoll. The sources describe a wakeup bug and do not say who can exploit it. unknown","tier":"C","verified":{},"analyst_revisit_at":null,"fix_status":"unknown","audience":"businesses","plain_action":null,"pkeys":[],"ptokens":[],"labels":["iouring","kernel","linux","runtime","unknown"],"first_public_at":"2026-11-03T12:17:33+00:00","kev":null,"cvss_vector":null,"cvss_by":null,"poc_count":0,"sources":[{"source":"nvd","external_id":"CVE-2026-99268","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-99268","severity":null,"published_at":"2026-11-03T12:17:33+00:00","modified_at":"2026-11-03T12:17:33+00:00","first_seen_at":"2026-11-03T12:20:40+00:00","fetched_at":"2026-11-03T12:20:40+00:00","note":null}],"documents":[],"notice":"Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai"}