{"id":"CVE-2026-98530","published":"2026-10-18T16:17:04.530","lastModified":"2026-10-19T10:19:08.410","description":"In the Linux kernel, the following vulnerability has been resolved:\n\nkprobes: Fix permanent hang when flushing the kprobe optimizer\n\nWriting 0 to /proc/sys/debug/kprobes-optimization while a kprobe is\njump-optimized never returns. The writer sleeps in D state forever with\nkprobe_sysctl_mutex held, so any later read or write of that sysctl\nhangs as well. For example, with vfs_read+9 as an optimizable address\nin this build:\n\n  # cd /sys/kernel/tracing\n  # echo 'p:myprobe vfs_read+9' >> kprobe_events\n  # echo 1 > events/kprobes/myprobe/enable\n  # # wait until /sys/kernel/debug/kprobes/list shows [OPTIMIZED]\n  # echo 0 > /proc/sys/debug/kprobes-optimization\n\n  INFO: task sh:246 blocked for more than 10 seconds.\n  Call Trace:\n   <TASK>\n   __schedule+0x1176/0x4f70\n   schedule+0xdc/0x2c0\n   schedule_timeout+0x17b/0x260\n   wait_for_completion+0x173/0x3c0\n   wait_for_kprobe_optimizer_locked+0xbc/0x130\n   proc_kprobes_optimization_handler+0x156/0x1b0\n   proc_sys_call_handler+0x324/0x490\n   vfs_write+0x52d/0xfe0\n   ksys_write+0xff/0x200\n   do_syscall_64+0x106/0x630\n   entry_SYSCALL_64_after_hwframe+0x77/0x7f\n   </TASK>\n  ...\n  INFO: task cat:265 is blocked on a mutex likely owned by task sh:246.\n\nwait_for_kprobe_optimizer_locked() reinitializes optimizer_completion,\nasks the optimizer thread to flush and sleeps in wait_for_completion().\nThe thread drains the (un)optimizing lists, but calls complete() only\nif completion_done() is true, i.e. if the completion is already done,\nwhich never happens while someone waits. disarm_all_kprobes() and\nkprobe_trace_self_tests_init() wait the same way.\n\nCalling complete() unconditionally would not be enough: the waiter\ndrops kprobe_mutex while it sleeps, and nothing else serializes the\nsysctl handler against the debugfs \"enabled\" file. A second flusher\nthat still finds the lists non-empty, e.g. because a disabled probe is\nqueued for unoptimizing, reinitializes the completion under the first:\n\n  sysctl write                      debugfs \"enabled\" write\n  unoptimize_all_kprobes()\n    wait_for_kprobe_optimizer_locked()\n      init_completion(c)\n      mutex_unlock(&kprobe_mutex)\n      wait_for_completion(c)\n                                    disarm_all_kprobes()\n                                      wait_for_kprobe_optimizer_locked()\n                                        init_completion(c)\n                                          // c->wait is reset, the first\n                                          // waiter is off the queue\n                                        mutex_unlock(&kprobe_mutex)\n                                        wait_for_completion(c)\n  kprobe_optimizer()\n    complete(c)\n      // wakes the debugfs writer only\n\nwhere c is &optimizer_completion. Lining up the two writes during an\noptimizer pass loses the sysctl writer this way.\n\nReplace the completion with a counter of optimizer passes, bumped at the\nend of each pass and signalled with wake_up_var_locked(), both under\nkprobe_mutex. A flusher samples the count and waits with\nwait_var_event_mutex(), which drops kprobe_mutex only while sleeping, so\na new count means a whole pass ran in the meantime. Nothing is\nreinitialized, so several flushers can sleep in the wait at once.","cvssScore":null,"cvssSeverity":null,"cvssVector":null,"cwes":[],"vendors":[],"products":[],"references":[{"url":"https://git.kernel.org/stable/c/5bfa9f1a9dcb6ecb607adbc1c0226605c972935b","tags":[]},{"url":"https://git.kernel.org/stable/c/f05a01fc7d4f731b4c100d2aef1e85ad77728862","tags":[]}],"exploitRefs":[],"hasPoc":false,"ai":null}