{"id":"CVE-2026-90183","published":"2026-09-17T17:17:12.710","lastModified":"2026-09-17T17:17:12.710","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nblk-iolatency: clear delay state when freeing policy data\n\nio.latency can throttle a group which has no latency target of its own.\nWhen a sibling misses its target, check_scale_change() scales down its\npeers, and a peer that reaches queue depth one gets blkcg_use_delay()\ncalled on it on every further scale-down, even with min_lat_nsec == 0.\n\niolatency_pd_offline() resets the target through\niolatency_set_min_lat_nsec(), which clears the delay only on a nonzero\nto zero transition, so it never clears such a peer.  Freeing the policy\ndata then leaves blkg->use_delay set and blkcg->congestion_count\nelevated with nothing left that can drop it.\n\nblk_cgroup_congested() then returns true for every task in that cgroup\nand its descendants for as long as the cgroup lives: page_cache_sync_ra()\ncuts readahead to a single page, page_cache_async_ra() skips it\naltogether, and __folio_throttle_swaprate() takes swap_avail_lock and\nschedules a throttle on anonymous folio allocation.\n\nClear the delay in iolatency_pd_free().  By then bio-held blkg\nreferences have drained, or the queue is frozen for policy\ndeactivation, so check_scale_change() cannot re-arm it.  The free\ncallback can also see policy data which was never attached to a blkg,\nhence the pd->blkg check.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/02f1b90ab33ef99c89dd96a385b432ff57abc64d","tags":[]},{"url":"https://git.kernel.org/stable/c/8935bf22c0a0db517a7f72f7097300e05dd852f5","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}