{"id":"CVE-2026-99255","kind":"vulnerability","aliases":["CVE-2026-99255"],"related":[],"title":"Linux kernel: io_uring race can hang a task that is exiting","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nio_uring/io-wq: don't hide a link from cancellations\n\nA task that exits cancels its tracked requests and waits until they are\ngone. io_uring_cancel_generic() looks everywhere a request can be, and\ngoes to sleep if it found nothing to cancel. It is woken up each time\none of the requests of the task is completed by task work or freed.\n\nWhen io-wq is done with a request that has a link, there is a time in which\nthe next request of the link can't be found:\n\n  io_worker_handle_work()\n    io_assign_current_work(worker, NULL)\n    io_wq_free_work(work)\n      io_req_find_next()           // takes the link off the request\n      io_free_req()                // task work for the task\n    io_assign_current_work(worker, linked)\n      io_run_task_work()\n      cond_resched()\n      worker->cur_work = linked\n\nIt is not pending, it is not the current work of a worker, and it is not\nyet on the list that it is about to be parked on. If the pass that the\nexiting task makes because of io_free_req() falls into that time, it finds\nnothing and the task goes to sleep. The worker then runs the link, which\nIO_WQ_BIT_EXIT doesn't prevent for bounded work. If that parks the request\nnothing is completed or freed, and nobody wakes up the task. If it is the\nprocess that exits it can't be killed any more either.\n\n  NOP, IOSQE_ASYNC | IOSQE_IO_LINK\n  FUTEX_WAIT, FUTEX2_PRIVATE\n  exit of the thread, or of the process, up to 63us later\n\nWith four of these per round 5 to 8 of 10,000 rounds hang (4 CPUs, KVM):\n\n  io_uring_cancel_generic\n  do_exit\n\nCommit f8b632e89a10 (\"io_uring: tighten task exit cancellations\") is about\nthe same gap, and makes the task look again for whatever happens before\nio_free_req().\n\nFree the request once the link is the current work of the worker, or\npending. Whoever is woken up by that can find the link, until it is parked\nwhere cancellations look as well. To do so split io_wq_free_work() in two.","summary":"In the io_uring subsystem, a request link can briefly go untracked while io-wq finishes a request, so an exiting task can sleep forever and become impossible to kill. The sources describe a hang, not code execution or data access.","severity":"unknown","cvss":null,"epss":null,"exploited":false,"exploited_at":null,"products":[],"ecosystems":[],"tags":["io_uring","linux-kernel","runtime","self-hosted"],"not_affected_if":["io_uring-disabled","no-io-uring-workloads"],"how_to_check":"- `uname -r` to see the running kernel version\n- Check whether the kernel build includes the fix commit for io_uring/io-wq (the sources do not give the fixed version)","fix":null,"status":"open","published_at":"2026-11-03T12:17:32+00:00","first_seen_at":"2026-11-03T12:20:39+00:00","triaged_at":"2026-11-03T12:22:51+00:00","updated_at":"2026-11-03T12:22:51+00:00","category":"runtime","confidence":0.6,"triage":{"at":"2026-11-03T12:22:51Z","model":"claude-haiku-5-5","output":{"fix":"unknown","reach":"local","title":"Linux kernel: io_uring race can hang a task that is exiting","impact":"low","summary":"In the io_uring subsystem, a request link can briefly go untracked while io-wq finishes a request, so an exiting task can sleep forever and become impossible to kill. The sources describe a hang, not code execution or data access.","audience":"businesses","category":"runtime","nicknames":[],"who_cares":["linux-kernel","io_uring","self-hosted"],"confidence":0.6,"fix_status":"unknown","needs_auth":"unknown","worst_case":"Exiting processes hang and cannot be killed; a denial of service.","analyst_note":"Confirm the affected and fixed kernel versions from the NVD entry and the upstream stable commit referenced for f8b632e89a10 and this fix.","how_to_check":"- `uname -r` to see the running kernel version\n- Check whether the kernel build includes the fix commit for io_uring/io-wq (the sources do not give the fixed version)","plain_action":"","needs_analyst":true,"not_affected_if":["io_uring-disabled","no-io-uring-workloads"],"who_is_affected":"Systems running a Linux kernel with io_uring enabled that run workloads using io_uring with linked requests such as IOSQE_IO_LINK and IOSQE_ASYNC, where processes exit while those requests are in flight. Systems that do not use io_uring are not affected. The sources do not state the affected kernel version range.","user_interaction":"no","needs_special_setup":"yes","severity_if_unknown":"unknown"},"cost_usd":0.000751,"input_tokens":4743,"output_tokens":554,"severity_before":"unknown"},"analyst_status":null,"analyst_note":null,"search":"'-2026':14A,17A '-99255':15A,18A '000':357C '10':356C '4':360C '5':352C '63us':344C '8':354C 'access':69B 'assign':177C,206C 'async':326C 'becom':55B 'bit':285C 'bound':291C 'briefli':37B 'cancel':88C,93C,105C,122C,365C,376C,425C 'code':65B 'commit':369C 'complet':137C,300C 'cond':215C 'cpus':361C 'cur':218C 'current':178C,207C,229C,403C 'cve':13A,16A 'data':68B 'describ':61B 'doesn':287C 'done':148C 'either':323C 'everywher':108C 'execut':66B 'exit':12A,49B,92C,255C,286C,315C,334C,368C,375C 'f8b632e89a10':370C 'fall':263C 'find':189C,268C,417C 'finish':44B 'follow':75C 'forev':53B 'found':119C,171C 'four':347C 'free':184C,198C,261C,393C,395C,435C 'freed':142C,302C 'futex':330C 'futex2':332C 'gap':381C 'generic':106C,366C 'go':38B 'goe':114C,273C 'gone':102C 'handl':174C 'hang':7A,63B,359C 'happen':390C 'hide':84C 'host':27B 'imposs':56B 'io':3A,19B,30B,42B,80C,103C,145C,172C,176C,182C,187C,197C,205C,211C,260C,283C,328C,363C,371C,392C,433C 'io-wq':41B,144C 'iosq':325C,327C 'kernel':2A,23B,73C 'kill':58B,320C 'kvm':362C 'later':345C 'link':35B,86C,155C,167C,193C,210C,220C,281C,329C,400C,419C 'linux':1A,22B,72C 'linux-kernel':21B 'list':241C 'look':107C,386C,426C 'make':257C,383C 'next':163C,190C 'nobodi':304C 'nop':324C 'noth':120C,269C,298C 'null':181C 'one':129C 'park':248C,295C,423C 'pass':252C 'pend':224C,409C 'per':350C 'prevent':289C 'privat':333C 'process':313C,341C 'race':5A 'req':188C,199C,262C,394C 'request':34B,46B,96C,110C,132C,151C,164C,196C,297C,397C 'resch':216C 'resolv':79C 'round':351C,358C 'run':212C,279C 'runtim':24B 'self':26B 'self-host':25B 'sleep':52B,116C,275C 'sourc':60B 'split':432C 'subsystem':32B 'take':191C 'task':9A,50B,90C,135C,139C,200C,204C,213C,256C,272C,308C,374C,385C 'thread':337C 'tighten':373C 'time':128C,159C,266C 'track':95C 'two':438C 'untrack':39B 'ure':4A,20B,31B,104C,364C,372C 'uring/io-wq':81C 'vulner':76C 'wait':98C,331C 'wake':305C 'well':428C 'whatev':389C 'whoever':410C 'woken':125C,412C 'work':140C,175C,179C,185C,186C,201C,208C,214C,219C,230C,292C,404C,436C 'worker':173C,180C,209C,217C,233C,277C,407C 'wq':43B,146C,183C,284C,434C 'yet':238C","nicknames":[],"mitigations":[],"who_affected":"Systems running a Linux kernel with io_uring enabled that run workloads using io_uring with linked requests such as IOSQE_IO_LINK and IOSQE_ASYNC, where processes exit while those requests are in flight. Systems that do not use io_uring are not affected. The sources do not state the affected kernel version range.","tier":"C","verified":{},"analyst_revisit_at":null,"fix_status":"unknown","audience":"businesses","plain_action":null,"pkeys":[],"ptokens":[],"labels":["iouring","linuxkernel","runtime","selfhosted","unknown"],"first_public_at":"2026-11-03T12:17:32+00:00","kev":null,"cvss_vector":null,"cvss_by":null,"poc_count":0,"sources":[{"source":"nvd","external_id":"CVE-2026-99255","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-99255","severity":null,"published_at":"2026-11-03T12:17:32+00:00","modified_at":"2026-11-03T12:17:32+00:00","first_seen_at":"2026-11-03T12:20:39+00:00","fetched_at":"2026-11-03T12:20:39+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"}