1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1998,2000 Rik van Riel
6 * Thanks go out to Claus Fischer for some serious inspiration and
7 * for goading me into coding this file...
8 * Copyright (C) 2010 Google, Inc.
9 * Rewritten by David Rientjes
11 * The routines in this file are used to kill a process when
12 * we're seriously out of memory. This gets called from __alloc_pages()
13 * in mm/page_alloc.c when we really run out of memory.
15 * Since we won't call these routines often (on a well-configured
16 * machine) this file will double as a 'coding guide' and a signpost
17 * for newbie kernel hackers. It features several pointers to major
18 * kernel subsystems and hints as to where to find out what things do.
21 #include <linux/oom.h>
23 #include <linux/err.h>
24 #include <linux/gfp.h>
25 #include <linux/sched.h>
26 #include <linux/sched/mm.h>
27 #include <linux/sched/coredump.h>
28 #include <linux/sched/task.h>
29 #include <linux/sched/debug.h>
30 #include <linux/swap.h>
31 #include <linux/timex.h>
32 #include <linux/jiffies.h>
33 #include <linux/cpuset.h>
34 #include <linux/export.h>
35 #include <linux/notifier.h>
36 #include <linux/memcontrol.h>
37 #include <linux/mempolicy.h>
38 #include <linux/security.h>
39 #include <linux/ptrace.h>
40 #include <linux/freezer.h>
41 #include <linux/ftrace.h>
42 #include <linux/ratelimit.h>
43 #include <linux/kthread.h>
44 #include <linux/init.h>
45 #include <linux/mmu_notifier.h>
51 #define CREATE_TRACE_POINTS
52 #include <trace/events/oom.h>
54 int sysctl_panic_on_oom;
55 int sysctl_oom_kill_allocating_task;
56 int sysctl_oom_dump_tasks = 1;
59 * Serializes oom killer invocations (out_of_memory()) from all contexts to
60 * prevent from over eager oom killing (e.g. when the oom killer is invoked
61 * from different domains).
63 * oom_killer_disable() relies on this lock to stabilize oom_killer_disabled
66 DEFINE_MUTEX(oom_lock);
67 /* Serializes oom_score_adj and oom_score_adj_min updates */
68 DEFINE_MUTEX(oom_adj_mutex);
70 static inline bool is_memcg_oom(struct oom_control *oc)
72 return oc->memcg != NULL;
77 * oom_cpuset_eligible() - check task eligiblity for kill
78 * @start: task struct of which task to consider
79 * @oc: pointer to struct oom_control
81 * Task eligibility is determined by whether or not a candidate task, @tsk,
82 * shares the same mempolicy nodes as current if it is bound by such a policy
83 * and whether or not it has the same set of allowed cpuset nodes.
85 * This function is assuming oom-killer context and 'current' has triggered
88 static bool oom_cpuset_eligible(struct task_struct *start,
89 struct oom_control *oc)
91 struct task_struct *tsk;
93 const nodemask_t *mask = oc->nodemask;
99 for_each_thread(start, tsk) {
102 * If this is a mempolicy constrained oom, tsk's
103 * cpuset is irrelevant. Only return true if its
104 * mempolicy intersects current, otherwise it may be
107 ret = mempolicy_nodemask_intersects(tsk, mask);
110 * This is not a mempolicy constrained oom, so only
111 * check the mems of tsk's cpuset.
113 ret = cpuset_mems_allowed_intersects(current, tsk);
123 static bool oom_cpuset_eligible(struct task_struct *tsk, struct oom_control *oc)
127 #endif /* CONFIG_NUMA */
130 * The process p may have detached its own ->mm while exiting or through
131 * kthread_use_mm(), but one or more of its subthreads may still have a valid
132 * pointer. Return p, or any of its subthreads with a valid ->mm, with
135 struct task_struct *find_lock_task_mm(struct task_struct *p)
137 struct task_struct *t;
141 for_each_thread(p, t) {
155 * order == -1 means the oom kill is required by sysrq, otherwise only
156 * for display purposes.
158 static inline bool is_sysrq_oom(struct oom_control *oc)
160 return oc->order == -1;
163 /* return true if the task is not adequate as candidate victim task. */
164 static bool oom_unkillable_task(struct task_struct *p)
166 if (is_global_init(p))
168 if (p->flags & PF_KTHREAD)
174 * Print out unreclaimble slabs info when unreclaimable slabs amount is greater
175 * than all user memory (LRU pages)
177 static bool is_dump_unreclaim_slabs(void)
179 unsigned long nr_lru;
181 nr_lru = global_node_page_state(NR_ACTIVE_ANON) +
182 global_node_page_state(NR_INACTIVE_ANON) +
183 global_node_page_state(NR_ACTIVE_FILE) +
184 global_node_page_state(NR_INACTIVE_FILE) +
185 global_node_page_state(NR_ISOLATED_ANON) +
186 global_node_page_state(NR_ISOLATED_FILE) +
187 global_node_page_state(NR_UNEVICTABLE);
189 return (global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B) > nr_lru);
193 * oom_badness - heuristic function to determine which candidate task to kill
194 * @p: task struct of which task we should calculate
195 * @totalpages: total present RAM allowed for page allocation
197 * The heuristic for determining which task to kill is made to be as simple and
198 * predictable as possible. The goal is to return the highest value for the
199 * task consuming the most memory to avoid subsequent oom failures.
201 long oom_badness(struct task_struct *p, unsigned long totalpages)
206 if (oom_unkillable_task(p))
209 p = find_lock_task_mm(p);
214 * Do not even consider tasks which are explicitly marked oom
215 * unkillable or have been already oom reaped or the are in
216 * the middle of vfork
218 adj = (long)p->signal->oom_score_adj;
219 if (adj == OOM_SCORE_ADJ_MIN ||
220 test_bit(MMF_OOM_SKIP, &p->mm->flags) ||
227 * The baseline for the badness score is the proportion of RAM that each
228 * task's rss, pagetable and swap space use.
230 points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
231 mm_pgtables_bytes(p->mm) / PAGE_SIZE;
234 /* Normalize to oom_score_adj units */
235 adj *= totalpages / 1000;
241 static const char * const oom_constraint_text[] = {
242 [CONSTRAINT_NONE] = "CONSTRAINT_NONE",
243 [CONSTRAINT_CPUSET] = "CONSTRAINT_CPUSET",
244 [CONSTRAINT_MEMORY_POLICY] = "CONSTRAINT_MEMORY_POLICY",
245 [CONSTRAINT_MEMCG] = "CONSTRAINT_MEMCG",
249 * Determine the type of allocation constraint.
251 static enum oom_constraint constrained_alloc(struct oom_control *oc)
255 enum zone_type highest_zoneidx = gfp_zone(oc->gfp_mask);
256 bool cpuset_limited = false;
259 if (is_memcg_oom(oc)) {
260 oc->totalpages = mem_cgroup_get_max(oc->memcg) ?: 1;
261 return CONSTRAINT_MEMCG;
264 /* Default to all available memory */
265 oc->totalpages = totalram_pages() + total_swap_pages;
267 if (!IS_ENABLED(CONFIG_NUMA))
268 return CONSTRAINT_NONE;
271 return CONSTRAINT_NONE;
273 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
274 * to kill current.We have to random task kill in this case.
275 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
277 if (oc->gfp_mask & __GFP_THISNODE)
278 return CONSTRAINT_NONE;
281 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
282 * the page allocator means a mempolicy is in effect. Cpuset policy
283 * is enforced in get_page_from_freelist().
286 !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) {
287 oc->totalpages = total_swap_pages;
288 for_each_node_mask(nid, *oc->nodemask)
289 oc->totalpages += node_present_pages(nid);
290 return CONSTRAINT_MEMORY_POLICY;
293 /* Check this allocation failure is caused by cpuset's wall function */
294 for_each_zone_zonelist_nodemask(zone, z, oc->zonelist,
295 highest_zoneidx, oc->nodemask)
296 if (!cpuset_zone_allowed(zone, oc->gfp_mask))
297 cpuset_limited = true;
299 if (cpuset_limited) {
300 oc->totalpages = total_swap_pages;
301 for_each_node_mask(nid, cpuset_current_mems_allowed)
302 oc->totalpages += node_present_pages(nid);
303 return CONSTRAINT_CPUSET;
305 return CONSTRAINT_NONE;
308 static int oom_evaluate_task(struct task_struct *task, void *arg)
310 struct oom_control *oc = arg;
313 if (oom_unkillable_task(task))
316 /* p may not have freeable memory in nodemask */
317 if (!is_memcg_oom(oc) && !oom_cpuset_eligible(task, oc))
321 * This task already has access to memory reserves and is being killed.
322 * Don't allow any other task to have access to the reserves unless
323 * the task has MMF_OOM_SKIP because chances that it would release
324 * any memory is quite low.
326 if (!is_sysrq_oom(oc) && tsk_is_oom_victim(task)) {
327 if (test_bit(MMF_OOM_SKIP, &task->signal->oom_mm->flags))
333 * If task is allocating a lot of memory and has been marked to be
334 * killed first if it triggers an oom, then select it.
336 if (oom_task_origin(task)) {
341 points = oom_badness(task, oc->totalpages);
342 if (points == LONG_MIN || points < oc->chosen_points)
347 put_task_struct(oc->chosen);
348 get_task_struct(task);
350 oc->chosen_points = points;
355 put_task_struct(oc->chosen);
356 oc->chosen = (void *)-1UL;
361 * Simple selection loop. We choose the process with the highest number of
362 * 'points'. In case scan was aborted, oc->chosen is set to -1.
364 static void select_bad_process(struct oom_control *oc)
366 oc->chosen_points = LONG_MIN;
368 if (is_memcg_oom(oc))
369 mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc);
371 struct task_struct *p;
375 if (oom_evaluate_task(p, oc))
381 static int dump_task(struct task_struct *p, void *arg)
383 struct oom_control *oc = arg;
384 struct task_struct *task;
386 if (oom_unkillable_task(p))
389 /* p may not have freeable memory in nodemask */
390 if (!is_memcg_oom(oc) && !oom_cpuset_eligible(p, oc))
393 task = find_lock_task_mm(p);
396 * This is a kthread or all of p's threads have already
397 * detached their mm's. There's no need to report
398 * them; they can't be oom killed anyway.
403 pr_info("[%7d] %5d %5d %8lu %8lu %8ld %8lu %5hd %s\n",
404 task->pid, from_kuid(&init_user_ns, task_uid(task)),
405 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
406 mm_pgtables_bytes(task->mm),
407 get_mm_counter(task->mm, MM_SWAPENTS),
408 task->signal->oom_score_adj, task->comm);
415 * dump_tasks - dump current memory state of all system tasks
416 * @oc: pointer to struct oom_control
418 * Dumps the current memory state of all eligible tasks. Tasks not in the same
419 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
421 * State information includes task's pid, uid, tgid, vm size, rss,
422 * pgtables_bytes, swapents, oom_score_adj value, and name.
424 static void dump_tasks(struct oom_control *oc)
426 pr_info("Tasks state (memory values in pages):\n");
427 pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name\n");
429 if (is_memcg_oom(oc))
430 mem_cgroup_scan_tasks(oc->memcg, dump_task, oc);
432 struct task_struct *p;
441 static void dump_oom_summary(struct oom_control *oc, struct task_struct *victim)
443 /* one line summary of the oom killer context. */
444 pr_info("oom-kill:constraint=%s,nodemask=%*pbl",
445 oom_constraint_text[oc->constraint],
446 nodemask_pr_args(oc->nodemask));
447 cpuset_print_current_mems_allowed();
448 mem_cgroup_print_oom_context(oc->memcg, victim);
449 pr_cont(",task=%s,pid=%d,uid=%d\n", victim->comm, victim->pid,
450 from_kuid(&init_user_ns, task_uid(victim)));
453 static void dump_header(struct oom_control *oc, struct task_struct *p)
455 pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), order=%d, oom_score_adj=%hd\n",
456 current->comm, oc->gfp_mask, &oc->gfp_mask, oc->order,
457 current->signal->oom_score_adj);
458 if (!IS_ENABLED(CONFIG_COMPACTION) && oc->order)
459 pr_warn("COMPACTION is disabled!!!\n");
462 if (is_memcg_oom(oc))
463 mem_cgroup_print_oom_meminfo(oc->memcg);
465 show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask);
466 if (is_dump_unreclaim_slabs())
467 dump_unreclaimable_slab();
469 if (sysctl_oom_dump_tasks)
472 dump_oom_summary(oc, p);
476 * Number of OOM victims in flight
478 static atomic_t oom_victims = ATOMIC_INIT(0);
479 static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
481 static bool oom_killer_disabled __read_mostly;
483 #define K(x) ((x) << (PAGE_SHIFT-10))
486 * task->mm can be NULL if the task is the exited group leader. So to
487 * determine whether the task is using a particular mm, we examine all the
488 * task's threads: if one of those is using this mm then this task was also
491 bool process_shares_mm(struct task_struct *p, struct mm_struct *mm)
493 struct task_struct *t;
495 for_each_thread(p, t) {
496 struct mm_struct *t_mm = READ_ONCE(t->mm);
505 * OOM Reaper kernel thread which tries to reap the memory used by the OOM
506 * victim (if that is possible) to help the OOM killer to move on.
508 static struct task_struct *oom_reaper_th;
509 static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait);
510 static struct task_struct *oom_reaper_list;
511 static DEFINE_SPINLOCK(oom_reaper_lock);
513 bool __oom_reap_task_mm(struct mm_struct *mm)
515 struct vm_area_struct *vma;
519 * Tell all users of get_user/copy_from_user etc... that the content
520 * is no longer stable. No barriers really needed because unmapping
521 * should imply barriers already and the reader would hit a page fault
522 * if it stumbled over a reaped memory.
524 set_bit(MMF_UNSTABLE, &mm->flags);
526 for (vma = mm->mmap ; vma; vma = vma->vm_next) {
527 if (!can_madv_lru_vma(vma))
531 * Only anonymous pages have a good chance to be dropped
532 * without additional steps which we cannot afford as we
535 * We do not even care about fs backed pages because all
536 * which are reclaimable have already been reclaimed and
537 * we do not want to block exit_mmap by keeping mm ref
538 * count elevated without a good reason.
540 if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) {
541 struct mmu_notifier_range range;
542 struct mmu_gather tlb;
544 mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0,
545 vma, mm, vma->vm_start,
547 tlb_gather_mmu(&tlb, mm, range.start, range.end);
548 if (mmu_notifier_invalidate_range_start_nonblock(&range)) {
549 tlb_finish_mmu(&tlb, range.start, range.end);
553 unmap_page_range(&tlb, vma, range.start, range.end, NULL);
554 mmu_notifier_invalidate_range_end(&range);
555 tlb_finish_mmu(&tlb, range.start, range.end);
563 * Reaps the address space of the give task.
565 * Returns true on success and false if none or part of the address space
566 * has been reclaimed and the caller should retry later.
568 static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
572 if (!mmap_read_trylock(mm)) {
573 trace_skip_task_reaping(tsk->pid);
578 * MMF_OOM_SKIP is set by exit_mmap when the OOM reaper can't
579 * work on the mm anymore. The check for MMF_OOM_SKIP must run
580 * under mmap_lock for reading because it serializes against the
581 * mmap_write_lock();mmap_write_unlock() cycle in exit_mmap().
583 if (test_bit(MMF_OOM_SKIP, &mm->flags)) {
584 trace_skip_task_reaping(tsk->pid);
588 trace_start_task_reaping(tsk->pid);
590 /* failed to reap part of the address space. Try again later */
591 ret = __oom_reap_task_mm(mm);
595 pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
596 task_pid_nr(tsk), tsk->comm,
597 K(get_mm_counter(mm, MM_ANONPAGES)),
598 K(get_mm_counter(mm, MM_FILEPAGES)),
599 K(get_mm_counter(mm, MM_SHMEMPAGES)));
601 trace_finish_task_reaping(tsk->pid);
603 mmap_read_unlock(mm);
608 #define MAX_OOM_REAP_RETRIES 10
609 static void oom_reap_task(struct task_struct *tsk)
612 struct mm_struct *mm = tsk->signal->oom_mm;
614 /* Retry the mmap_read_trylock(mm) a few times */
615 while (attempts++ < MAX_OOM_REAP_RETRIES && !oom_reap_task_mm(tsk, mm))
616 schedule_timeout_idle(HZ/10);
618 if (attempts <= MAX_OOM_REAP_RETRIES ||
619 test_bit(MMF_OOM_SKIP, &mm->flags))
622 pr_info("oom_reaper: unable to reap pid:%d (%s)\n",
623 task_pid_nr(tsk), tsk->comm);
624 sched_show_task(tsk);
625 debug_show_all_locks();
628 tsk->oom_reaper_list = NULL;
631 * Hide this mm from OOM killer because it has been either reaped or
632 * somebody can't call mmap_write_unlock(mm).
634 set_bit(MMF_OOM_SKIP, &mm->flags);
636 /* Drop a reference taken by queue_oom_reaper */
637 put_task_struct(tsk);
640 static int oom_reaper(void *unused)
643 struct task_struct *tsk = NULL;
645 wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL);
646 spin_lock_irq(&oom_reaper_lock);
647 if (oom_reaper_list != NULL) {
648 tsk = oom_reaper_list;
649 oom_reaper_list = tsk->oom_reaper_list;
651 spin_unlock_irq(&oom_reaper_lock);
660 static void wake_oom_reaper(struct timer_list *timer)
662 struct task_struct *tsk = container_of(timer, struct task_struct,
664 struct mm_struct *mm = tsk->signal->oom_mm;
667 /* The victim managed to terminate on its own - see exit_mmap */
668 if (test_bit(MMF_OOM_SKIP, &mm->flags)) {
669 put_task_struct(tsk);
673 spin_lock_irqsave(&oom_reaper_lock, flags);
674 tsk->oom_reaper_list = oom_reaper_list;
675 oom_reaper_list = tsk;
676 spin_unlock_irqrestore(&oom_reaper_lock, flags);
677 trace_wake_reaper(tsk->pid);
678 wake_up(&oom_reaper_wait);
682 * Give the OOM victim time to exit naturally before invoking the oom_reaping.
683 * The timers timeout is arbitrary... the longer it is, the longer the worst
684 * case scenario for the OOM can take. If it is too small, the oom_reaper can
685 * get in the way and release resources needed by the process exit path.
686 * e.g. The futex robust list can sit in Anon|Private memory that gets reaped
687 * before the exit path is able to wake the futex waiters.
689 #define OOM_REAPER_DELAY (2*HZ)
690 static void queue_oom_reaper(struct task_struct *tsk)
692 /* mm is already queued? */
693 if (test_and_set_bit(MMF_OOM_REAP_QUEUED, &tsk->signal->oom_mm->flags))
696 get_task_struct(tsk);
697 timer_setup(&tsk->oom_reaper_timer, wake_oom_reaper, 0);
698 tsk->oom_reaper_timer.expires = jiffies + OOM_REAPER_DELAY;
699 add_timer(&tsk->oom_reaper_timer);
702 static int __init oom_init(void)
704 oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper");
707 subsys_initcall(oom_init)
709 static inline void queue_oom_reaper(struct task_struct *tsk)
712 #endif /* CONFIG_MMU */
715 * mark_oom_victim - mark the given task as OOM victim
718 * Has to be called with oom_lock held and never after
719 * oom has been disabled already.
721 * tsk->mm has to be non NULL and caller has to guarantee it is stable (either
722 * under task_lock or operate on the current).
724 static void mark_oom_victim(struct task_struct *tsk)
726 struct mm_struct *mm = tsk->mm;
728 WARN_ON(oom_killer_disabled);
729 /* OOM killer might race with memcg OOM */
730 if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
733 /* oom_mm is bound to the signal struct life time. */
734 if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm)) {
735 mmgrab(tsk->signal->oom_mm);
736 set_bit(MMF_OOM_VICTIM, &mm->flags);
740 * Make sure that the task is woken up from uninterruptible sleep
741 * if it is frozen because OOM killer wouldn't be able to free
742 * any memory and livelock. freezing_slow_path will tell the freezer
743 * that TIF_MEMDIE tasks should be ignored.
746 atomic_inc(&oom_victims);
747 trace_mark_victim(tsk->pid);
751 * exit_oom_victim - note the exit of an OOM victim
753 void exit_oom_victim(void)
755 clear_thread_flag(TIF_MEMDIE);
757 if (!atomic_dec_return(&oom_victims))
758 wake_up_all(&oom_victims_wait);
762 * oom_killer_enable - enable OOM killer
764 void oom_killer_enable(void)
766 oom_killer_disabled = false;
767 pr_info("OOM killer enabled.\n");
771 * oom_killer_disable - disable OOM killer
772 * @timeout: maximum timeout to wait for oom victims in jiffies
774 * Forces all page allocations to fail rather than trigger OOM killer.
775 * Will block and wait until all OOM victims are killed or the given
778 * The function cannot be called when there are runnable user tasks because
779 * the userspace would see unexpected allocation failures as a result. Any
780 * new usage of this function should be consulted with MM people.
782 * Returns true if successful and false if the OOM killer cannot be
785 bool oom_killer_disable(signed long timeout)
790 * Make sure to not race with an ongoing OOM killer. Check that the
791 * current is not killed (possibly due to sharing the victim's memory).
793 if (mutex_lock_killable(&oom_lock))
795 oom_killer_disabled = true;
796 mutex_unlock(&oom_lock);
798 ret = wait_event_interruptible_timeout(oom_victims_wait,
799 !atomic_read(&oom_victims), timeout);
804 pr_info("OOM killer disabled.\n");
809 static inline bool __task_will_free_mem(struct task_struct *task)
811 struct signal_struct *sig = task->signal;
814 * A coredumping process may sleep for an extended period in exit_mm(),
815 * so the oom killer cannot assume that the process will promptly exit
816 * and release memory.
818 if (sig->flags & SIGNAL_GROUP_COREDUMP)
821 if (sig->flags & SIGNAL_GROUP_EXIT)
824 if (thread_group_empty(task) && (task->flags & PF_EXITING))
831 * Checks whether the given task is dying or exiting and likely to
832 * release its address space. This means that all threads and processes
833 * sharing the same mm have to be killed or exiting.
834 * Caller has to make sure that task->mm is stable (hold task_lock or
835 * it operates on the current).
837 static bool task_will_free_mem(struct task_struct *task)
839 struct mm_struct *mm = task->mm;
840 struct task_struct *p;
844 * Skip tasks without mm because it might have passed its exit_mm and
845 * exit_oom_victim. oom_reaper could have rescued that but do not rely
846 * on that for now. We can consider find_lock_task_mm in future.
851 if (!__task_will_free_mem(task))
855 * This task has already been drained by the oom reaper so there are
856 * only small chances it will free some more
858 if (test_bit(MMF_OOM_SKIP, &mm->flags))
861 if (atomic_read(&mm->mm_users) <= 1)
865 * Make sure that all tasks which share the mm with the given tasks
866 * are dying as well to make sure that a) nobody pins its mm and
867 * b) the task is also reapable by the oom reaper.
870 for_each_process(p) {
871 if (!process_shares_mm(p, mm))
873 if (same_thread_group(task, p))
875 ret = __task_will_free_mem(p);
884 static void __oom_kill_process(struct task_struct *victim, const char *message)
886 struct task_struct *p;
887 struct mm_struct *mm;
888 bool can_oom_reap = true;
890 p = find_lock_task_mm(victim);
892 pr_info("%s: OOM victim %d (%s) is already exiting. Skip killing the task\n",
893 message, task_pid_nr(victim), victim->comm);
894 put_task_struct(victim);
896 } else if (victim != p) {
898 put_task_struct(victim);
902 /* Get a reference to safely compare mm after task_unlock(victim) */
906 /* Raise event before sending signal: task reaper must see this */
907 count_vm_event(OOM_KILL);
908 memcg_memory_event_mm(mm, MEMCG_OOM_KILL);
911 * We should send SIGKILL before granting access to memory reserves
912 * in order to prevent the OOM victim from depleting the memory
913 * reserves from the user space under its control.
915 do_send_sig_info(SIGKILL, SEND_SIG_PRIV, victim, PIDTYPE_TGID);
916 mark_oom_victim(victim);
917 pr_err("%s: Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB, UID:%u pgtables:%lukB oom_score_adj:%hd\n",
918 message, task_pid_nr(victim), victim->comm, K(mm->total_vm),
919 K(get_mm_counter(mm, MM_ANONPAGES)),
920 K(get_mm_counter(mm, MM_FILEPAGES)),
921 K(get_mm_counter(mm, MM_SHMEMPAGES)),
922 from_kuid(&init_user_ns, task_uid(victim)),
923 mm_pgtables_bytes(mm) >> 10, victim->signal->oom_score_adj);
927 * Kill all user processes sharing victim->mm in other thread groups, if
928 * any. They don't get access to memory reserves, though, to avoid
929 * depletion of all memory. This prevents mm->mmap_lock livelock when an
930 * oom killed thread cannot exit because it requires the semaphore and
931 * its contended by another thread trying to allocate memory itself.
932 * That thread will now get access to memory reserves since it has a
933 * pending fatal signal.
936 for_each_process(p) {
937 if (!process_shares_mm(p, mm))
939 if (same_thread_group(p, victim))
941 if (is_global_init(p)) {
942 can_oom_reap = false;
943 set_bit(MMF_OOM_SKIP, &mm->flags);
944 pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n",
945 task_pid_nr(victim), victim->comm,
946 task_pid_nr(p), p->comm);
950 * No kthead_use_mm() user needs to read from the userspace so
951 * we are ok to reap it.
953 if (unlikely(p->flags & PF_KTHREAD))
955 do_send_sig_info(SIGKILL, SEND_SIG_PRIV, p, PIDTYPE_TGID);
960 queue_oom_reaper(victim);
963 put_task_struct(victim);
968 * Kill provided task unless it's secured by setting
969 * oom_score_adj to OOM_SCORE_ADJ_MIN.
971 static int oom_kill_memcg_member(struct task_struct *task, void *message)
973 if (task->signal->oom_score_adj != OOM_SCORE_ADJ_MIN &&
974 !is_global_init(task)) {
975 get_task_struct(task);
976 __oom_kill_process(task, message);
981 static void oom_kill_process(struct oom_control *oc, const char *message)
983 struct task_struct *victim = oc->chosen;
984 struct mem_cgroup *oom_group;
985 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
986 DEFAULT_RATELIMIT_BURST);
989 * If the task is already exiting, don't alarm the sysadmin or kill
990 * its children or threads, just give it access to memory reserves
991 * so it can die quickly
994 if (task_will_free_mem(victim)) {
995 mark_oom_victim(victim);
996 queue_oom_reaper(victim);
998 put_task_struct(victim);
1001 task_unlock(victim);
1003 if (__ratelimit(&oom_rs))
1004 dump_header(oc, victim);
1007 * Do we need to kill the entire memory cgroup?
1008 * Or even one of the ancestor memory cgroups?
1009 * Check this out before killing the victim task.
1011 oom_group = mem_cgroup_get_oom_group(victim, oc->memcg);
1013 __oom_kill_process(victim, message);
1016 * If necessary, kill all tasks in the selected memory cgroup.
1019 mem_cgroup_print_oom_group(oom_group);
1020 mem_cgroup_scan_tasks(oom_group, oom_kill_memcg_member,
1022 mem_cgroup_put(oom_group);
1027 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
1029 static void check_panic_on_oom(struct oom_control *oc)
1031 if (likely(!sysctl_panic_on_oom))
1033 if (sysctl_panic_on_oom != 2) {
1035 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
1036 * does not panic for cpuset, mempolicy, or memcg allocation
1039 if (oc->constraint != CONSTRAINT_NONE)
1042 /* Do not panic for oom kills triggered by sysrq */
1043 if (is_sysrq_oom(oc))
1045 dump_header(oc, NULL);
1046 panic("Out of memory: %s panic_on_oom is enabled\n",
1047 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
1050 static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
1052 int register_oom_notifier(struct notifier_block *nb)
1054 return blocking_notifier_chain_register(&oom_notify_list, nb);
1056 EXPORT_SYMBOL_GPL(register_oom_notifier);
1058 int unregister_oom_notifier(struct notifier_block *nb)
1060 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
1062 EXPORT_SYMBOL_GPL(unregister_oom_notifier);
1065 * out_of_memory - kill the "best" process when we run out of memory
1066 * @oc: pointer to struct oom_control
1068 * If we run out of memory, we have the choice between either
1069 * killing a random task (bad), letting the system crash (worse)
1070 * OR try to be smart about which process to kill. Note that we
1071 * don't have to be perfect here, we just have to be good.
1073 bool out_of_memory(struct oom_control *oc)
1075 unsigned long freed = 0;
1077 if (oom_killer_disabled)
1080 if (!is_memcg_oom(oc)) {
1081 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
1083 /* Got some memory back in the last second. */
1088 * If current has a pending SIGKILL or is exiting, then automatically
1089 * select it. The goal is to allow it to allocate so that it may
1090 * quickly exit and free its memory.
1092 if (task_will_free_mem(current)) {
1093 mark_oom_victim(current);
1094 queue_oom_reaper(current);
1099 * The OOM killer does not compensate for IO-less reclaim.
1100 * pagefault_out_of_memory lost its gfp context so we have to
1101 * make sure exclude 0 mask - all other users should have at least
1102 * ___GFP_DIRECT_RECLAIM to get here. But mem_cgroup_oom() has to
1103 * invoke the OOM killer even if it is a GFP_NOFS allocation.
1105 if (oc->gfp_mask && !(oc->gfp_mask & __GFP_FS) && !is_memcg_oom(oc))
1109 * Check if there were limitations on the allocation (only relevant for
1110 * NUMA and memcg) that may require different handling.
1112 oc->constraint = constrained_alloc(oc);
1113 if (oc->constraint != CONSTRAINT_MEMORY_POLICY)
1114 oc->nodemask = NULL;
1115 check_panic_on_oom(oc);
1117 if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task &&
1118 current->mm && !oom_unkillable_task(current) &&
1119 oom_cpuset_eligible(current, oc) &&
1120 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
1121 get_task_struct(current);
1122 oc->chosen = current;
1123 oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)");
1127 select_bad_process(oc);
1128 /* Found nothing?!?! */
1130 dump_header(oc, NULL);
1131 pr_warn("Out of memory and no killable processes...\n");
1133 * If we got here due to an actual allocation at the
1134 * system level, we cannot survive this and will enter
1135 * an endless loop in the allocator. Bail out now.
1137 if (!is_sysrq_oom(oc) && !is_memcg_oom(oc))
1138 panic("System is deadlocked on memory\n");
1140 if (oc->chosen && oc->chosen != (void *)-1UL)
1141 oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" :
1142 "Memory cgroup out of memory");
1143 return !!oc->chosen;
1147 * The pagefault handler calls here because some allocation has failed. We have
1148 * to take care of the memcg OOM here because this is the only safe context without
1149 * any locks held but let the oom killer triggered from the allocation context care
1150 * about the global OOM.
1152 void pagefault_out_of_memory(void)
1154 static DEFINE_RATELIMIT_STATE(pfoom_rs, DEFAULT_RATELIMIT_INTERVAL,
1155 DEFAULT_RATELIMIT_BURST);
1157 if (mem_cgroup_oom_synchronize(true))
1160 if (fatal_signal_pending(current))
1163 if (__ratelimit(&pfoom_rs))
1164 pr_warn("Huh VM_FAULT_OOM leaked out to the #PF handler. Retrying PF\n");