4 * Copyright (C) 1998,2000 Rik van Riel
5 * Thanks go out to Claus Fischer for some serious inspiration and
6 * for goading me into coding this file...
7 * Copyright (C) 2010 Google, Inc.
8 * Rewritten by David Rientjes
10 * The routines in this file are used to kill a process when
11 * we're seriously out of memory. This gets called from __alloc_pages()
12 * in mm/page_alloc.c when we really run out of memory.
14 * Since we won't call these routines often (on a well-configured
15 * machine) this file will double as a 'coding guide' and a signpost
16 * for newbie kernel hackers. It features several pointers to major
17 * kernel subsystems and hints as to where to find out what things do.
20 #include <linux/oom.h>
22 #include <linux/err.h>
23 #include <linux/gfp.h>
24 #include <linux/sched.h>
25 #include <linux/swap.h>
26 #include <linux/timex.h>
27 #include <linux/jiffies.h>
28 #include <linux/cpuset.h>
29 #include <linux/export.h>
30 #include <linux/notifier.h>
31 #include <linux/memcontrol.h>
32 #include <linux/mempolicy.h>
33 #include <linux/security.h>
34 #include <linux/ptrace.h>
35 #include <linux/freezer.h>
36 #include <linux/ftrace.h>
37 #include <linux/ratelimit.h>
38 #include <linux/kthread.h>
39 #include <linux/init.h>
44 #define CREATE_TRACE_POINTS
45 #include <trace/events/oom.h>
47 int sysctl_panic_on_oom;
48 int sysctl_oom_kill_allocating_task;
49 int sysctl_oom_dump_tasks = 1;
51 DEFINE_MUTEX(oom_lock);
55 * has_intersects_mems_allowed() - check task eligiblity for kill
56 * @start: task struct of which task to consider
57 * @mask: nodemask passed to page allocator for mempolicy ooms
59 * Task eligibility is determined by whether or not a candidate task, @tsk,
60 * shares the same mempolicy nodes as current if it is bound by such a policy
61 * and whether or not it has the same set of allowed cpuset nodes.
63 static bool has_intersects_mems_allowed(struct task_struct *start,
64 const nodemask_t *mask)
66 struct task_struct *tsk;
70 for_each_thread(start, tsk) {
73 * If this is a mempolicy constrained oom, tsk's
74 * cpuset is irrelevant. Only return true if its
75 * mempolicy intersects current, otherwise it may be
78 ret = mempolicy_nodemask_intersects(tsk, mask);
81 * This is not a mempolicy constrained oom, so only
82 * check the mems of tsk's cpuset.
84 ret = cpuset_mems_allowed_intersects(current, tsk);
94 static bool has_intersects_mems_allowed(struct task_struct *tsk,
95 const nodemask_t *mask)
99 #endif /* CONFIG_NUMA */
102 * The process p may have detached its own ->mm while exiting or through
103 * use_mm(), but one or more of its subthreads may still have a valid
104 * pointer. Return p, or any of its subthreads with a valid ->mm, with
107 struct task_struct *find_lock_task_mm(struct task_struct *p)
109 struct task_struct *t;
113 for_each_thread(p, t) {
127 * order == -1 means the oom kill is required by sysrq, otherwise only
128 * for display purposes.
130 static inline bool is_sysrq_oom(struct oom_control *oc)
132 return oc->order == -1;
135 /* return true if the task is not adequate as candidate victim task. */
136 static bool oom_unkillable_task(struct task_struct *p,
137 struct mem_cgroup *memcg, const nodemask_t *nodemask)
139 if (is_global_init(p))
141 if (p->flags & PF_KTHREAD)
144 /* When mem_cgroup_out_of_memory() and p is not member of the group */
145 if (memcg && !task_in_mem_cgroup(p, memcg))
148 /* p may not have freeable memory in nodemask */
149 if (!has_intersects_mems_allowed(p, nodemask))
156 * oom_badness - heuristic function to determine which candidate task to kill
157 * @p: task struct of which task we should calculate
158 * @totalpages: total present RAM allowed for page allocation
160 * The heuristic for determining which task to kill is made to be as simple and
161 * predictable as possible. The goal is to return the highest value for the
162 * task consuming the most memory to avoid subsequent oom failures.
164 unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
165 const nodemask_t *nodemask, unsigned long totalpages)
170 if (oom_unkillable_task(p, memcg, nodemask))
173 p = find_lock_task_mm(p);
178 * Do not even consider tasks which are explicitly marked oom
179 * unkillable or have been already oom reaped or the are in
180 * the middle of vfork
182 adj = (long)p->signal->oom_score_adj;
183 if (adj == OOM_SCORE_ADJ_MIN ||
184 test_bit(MMF_OOM_REAPED, &p->mm->flags) ||
191 * The baseline for the badness score is the proportion of RAM that each
192 * task's rss, pagetable and swap space use.
194 points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
195 atomic_long_read(&p->mm->nr_ptes) + mm_nr_pmds(p->mm);
199 * Root processes get 3% bonus, just like the __vm_enough_memory()
200 * implementation used by LSMs.
202 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
203 points -= (points * 3) / 100;
205 /* Normalize to oom_score_adj units */
206 adj *= totalpages / 1000;
210 * Never return 0 for an eligible task regardless of the root bonus and
211 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
213 return points > 0 ? points : 1;
217 * Determine the type of allocation constraint.
220 static enum oom_constraint constrained_alloc(struct oom_control *oc,
221 unsigned long *totalpages)
225 enum zone_type high_zoneidx = gfp_zone(oc->gfp_mask);
226 bool cpuset_limited = false;
229 /* Default to all available memory */
230 *totalpages = totalram_pages + total_swap_pages;
233 return CONSTRAINT_NONE;
235 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
236 * to kill current.We have to random task kill in this case.
237 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
239 if (oc->gfp_mask & __GFP_THISNODE)
240 return CONSTRAINT_NONE;
243 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
244 * the page allocator means a mempolicy is in effect. Cpuset policy
245 * is enforced in get_page_from_freelist().
248 !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) {
249 *totalpages = total_swap_pages;
250 for_each_node_mask(nid, *oc->nodemask)
251 *totalpages += node_spanned_pages(nid);
252 return CONSTRAINT_MEMORY_POLICY;
255 /* Check this allocation failure is caused by cpuset's wall function */
256 for_each_zone_zonelist_nodemask(zone, z, oc->zonelist,
257 high_zoneidx, oc->nodemask)
258 if (!cpuset_zone_allowed(zone, oc->gfp_mask))
259 cpuset_limited = true;
261 if (cpuset_limited) {
262 *totalpages = total_swap_pages;
263 for_each_node_mask(nid, cpuset_current_mems_allowed)
264 *totalpages += node_spanned_pages(nid);
265 return CONSTRAINT_CPUSET;
267 return CONSTRAINT_NONE;
270 static enum oom_constraint constrained_alloc(struct oom_control *oc,
271 unsigned long *totalpages)
273 *totalpages = totalram_pages + total_swap_pages;
274 return CONSTRAINT_NONE;
278 enum oom_scan_t oom_scan_process_thread(struct oom_control *oc,
279 struct task_struct *task)
281 if (oom_unkillable_task(task, NULL, oc->nodemask))
282 return OOM_SCAN_CONTINUE;
285 * This task already has access to memory reserves and is being killed.
286 * Don't allow any other task to have access to the reserves.
288 if (!is_sysrq_oom(oc) && atomic_read(&task->signal->oom_victims))
289 return OOM_SCAN_ABORT;
292 * If task is allocating a lot of memory and has been marked to be
293 * killed first if it triggers an oom, then select it.
295 if (oom_task_origin(task))
296 return OOM_SCAN_SELECT;
302 * Simple selection loop. We chose the process with the highest
303 * number of 'points'. Returns -1 on scan abort.
305 static struct task_struct *select_bad_process(struct oom_control *oc,
306 unsigned int *ppoints, unsigned long totalpages)
308 struct task_struct *p;
309 struct task_struct *chosen = NULL;
310 unsigned long chosen_points = 0;
313 for_each_process(p) {
316 switch (oom_scan_process_thread(oc, p)) {
317 case OOM_SCAN_SELECT:
319 chosen_points = ULONG_MAX;
321 case OOM_SCAN_CONTINUE:
325 return (struct task_struct *)(-1UL);
329 points = oom_badness(p, NULL, oc->nodemask, totalpages);
330 if (!points || points < chosen_points)
334 chosen_points = points;
337 get_task_struct(chosen);
340 *ppoints = chosen_points * 1000 / totalpages;
345 * dump_tasks - dump current memory state of all system tasks
346 * @memcg: current's memory controller, if constrained
347 * @nodemask: nodemask passed to page allocator for mempolicy ooms
349 * Dumps the current memory state of all eligible tasks. Tasks not in the same
350 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
352 * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
353 * swapents, oom_score_adj value, and name.
355 static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask)
357 struct task_struct *p;
358 struct task_struct *task;
360 pr_info("[ pid ] uid tgid total_vm rss nr_ptes nr_pmds swapents oom_score_adj name\n");
362 for_each_process(p) {
363 if (oom_unkillable_task(p, memcg, nodemask))
366 task = find_lock_task_mm(p);
369 * This is a kthread or all of p's threads have already
370 * detached their mm's. There's no need to report
371 * them; they can't be oom killed anyway.
376 pr_info("[%5d] %5d %5d %8lu %8lu %7ld %7ld %8lu %5hd %s\n",
377 task->pid, from_kuid(&init_user_ns, task_uid(task)),
378 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
379 atomic_long_read(&task->mm->nr_ptes),
380 mm_nr_pmds(task->mm),
381 get_mm_counter(task->mm, MM_SWAPENTS),
382 task->signal->oom_score_adj, task->comm);
388 static void dump_header(struct oom_control *oc, struct task_struct *p)
390 pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), order=%d, oom_score_adj=%hd\n",
391 current->comm, oc->gfp_mask, &oc->gfp_mask, oc->order,
392 current->signal->oom_score_adj);
394 cpuset_print_current_mems_allowed();
397 mem_cgroup_print_oom_info(oc->memcg, p);
399 show_mem(SHOW_MEM_FILTER_NODES);
400 if (sysctl_oom_dump_tasks)
401 dump_tasks(oc->memcg, oc->nodemask);
405 * Number of OOM victims in flight
407 static atomic_t oom_victims = ATOMIC_INIT(0);
408 static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
410 bool oom_killer_disabled __read_mostly;
412 #define K(x) ((x) << (PAGE_SHIFT-10))
415 * task->mm can be NULL if the task is the exited group leader. So to
416 * determine whether the task is using a particular mm, we examine all the
417 * task's threads: if one of those is using this mm then this task was also
420 bool process_shares_mm(struct task_struct *p, struct mm_struct *mm)
422 struct task_struct *t;
424 for_each_thread(p, t) {
425 struct mm_struct *t_mm = READ_ONCE(t->mm);
435 * OOM Reaper kernel thread which tries to reap the memory used by the OOM
436 * victim (if that is possible) to help the OOM killer to move on.
438 static struct task_struct *oom_reaper_th;
439 static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait);
440 static struct task_struct *oom_reaper_list;
441 static DEFINE_SPINLOCK(oom_reaper_lock);
443 static bool __oom_reap_task(struct task_struct *tsk)
445 struct mmu_gather tlb;
446 struct vm_area_struct *vma;
447 struct mm_struct *mm = NULL;
448 struct task_struct *p;
449 struct zap_details details = {.check_swap_entries = true,
450 .ignore_dirty = true};
454 * We have to make sure to not race with the victim exit path
455 * and cause premature new oom victim selection:
456 * __oom_reap_task exit_mm
459 * atomic_dec_and_test
464 * # no TIF_MEMDIE task selects new victim
465 * unmap_page_range # frees some memory
467 mutex_lock(&oom_lock);
470 * Make sure we find the associated mm_struct even when the particular
471 * thread has already terminated and cleared its mm.
472 * We might have race with exit path so consider our work done if there
475 p = find_lock_task_mm(tsk);
479 atomic_inc(&mm->mm_count);
482 if (!down_read_trylock(&mm->mmap_sem)) {
488 * increase mm_users only after we know we will reap something so
489 * that the mmput_async is called only when we have reaped something
490 * and delayed __mmput doesn't matter that much
492 if (!mmget_not_zero(mm)) {
493 up_read(&mm->mmap_sem);
497 tlb_gather_mmu(&tlb, mm, 0, -1);
498 for (vma = mm->mmap ; vma; vma = vma->vm_next) {
499 if (is_vm_hugetlb_page(vma))
503 * mlocked VMAs require explicit munlocking before unmap.
504 * Let's keep it simple here and skip such VMAs.
506 if (vma->vm_flags & VM_LOCKED)
510 * Only anonymous pages have a good chance to be dropped
511 * without additional steps which we cannot afford as we
514 * We do not even care about fs backed pages because all
515 * which are reclaimable have already been reclaimed and
516 * we do not want to block exit_mmap by keeping mm ref
517 * count elevated without a good reason.
519 if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED))
520 unmap_page_range(&tlb, vma, vma->vm_start, vma->vm_end,
523 tlb_finish_mmu(&tlb, 0, -1);
524 pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
525 task_pid_nr(tsk), tsk->comm,
526 K(get_mm_counter(mm, MM_ANONPAGES)),
527 K(get_mm_counter(mm, MM_FILEPAGES)),
528 K(get_mm_counter(mm, MM_SHMEMPAGES)));
529 up_read(&mm->mmap_sem);
532 * This task can be safely ignored because we cannot do much more
533 * to release its memory.
535 set_bit(MMF_OOM_REAPED, &mm->flags);
537 * Drop our reference but make sure the mmput slow path is called from a
538 * different context because we shouldn't risk we get stuck there and
539 * put the oom_reaper out of the way.
545 mutex_unlock(&oom_lock);
549 #define MAX_OOM_REAP_RETRIES 10
550 static void oom_reap_task(struct task_struct *tsk)
554 /* Retry the down_read_trylock(mmap_sem) a few times */
555 while (attempts++ < MAX_OOM_REAP_RETRIES && !__oom_reap_task(tsk))
556 schedule_timeout_idle(HZ/10);
558 if (attempts > MAX_OOM_REAP_RETRIES) {
559 pr_info("oom_reaper: unable to reap pid:%d (%s)\n",
560 task_pid_nr(tsk), tsk->comm);
561 debug_show_all_locks();
565 * Clear TIF_MEMDIE because the task shouldn't be sitting on a
566 * reasonably reclaimable memory anymore or it is not a good candidate
567 * for the oom victim right now because it cannot release its memory
568 * itself nor by the oom reaper.
570 tsk->oom_reaper_list = NULL;
571 exit_oom_victim(tsk);
573 /* Drop a reference taken by wake_oom_reaper */
574 put_task_struct(tsk);
577 static int oom_reaper(void *unused)
582 struct task_struct *tsk = NULL;
584 wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL);
585 spin_lock(&oom_reaper_lock);
586 if (oom_reaper_list != NULL) {
587 tsk = oom_reaper_list;
588 oom_reaper_list = tsk->oom_reaper_list;
590 spin_unlock(&oom_reaper_lock);
599 void wake_oom_reaper(struct task_struct *tsk)
604 /* tsk is already queued? */
605 if (tsk == oom_reaper_list || tsk->oom_reaper_list)
608 get_task_struct(tsk);
610 spin_lock(&oom_reaper_lock);
611 tsk->oom_reaper_list = oom_reaper_list;
612 oom_reaper_list = tsk;
613 spin_unlock(&oom_reaper_lock);
614 wake_up(&oom_reaper_wait);
617 static int __init oom_init(void)
619 oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper");
620 if (IS_ERR(oom_reaper_th)) {
621 pr_err("Unable to start OOM reaper %ld. Continuing regardless\n",
622 PTR_ERR(oom_reaper_th));
623 oom_reaper_th = NULL;
627 subsys_initcall(oom_init)
631 * mark_oom_victim - mark the given task as OOM victim
634 * Has to be called with oom_lock held and never after
635 * oom has been disabled already.
637 void mark_oom_victim(struct task_struct *tsk)
639 WARN_ON(oom_killer_disabled);
640 /* OOM killer might race with memcg OOM */
641 if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
643 atomic_inc(&tsk->signal->oom_victims);
645 * Make sure that the task is woken up from uninterruptible sleep
646 * if it is frozen because OOM killer wouldn't be able to free
647 * any memory and livelock. freezing_slow_path will tell the freezer
648 * that TIF_MEMDIE tasks should be ignored.
651 atomic_inc(&oom_victims);
655 * exit_oom_victim - note the exit of an OOM victim
657 void exit_oom_victim(struct task_struct *tsk)
659 if (!test_and_clear_tsk_thread_flag(tsk, TIF_MEMDIE))
661 atomic_dec(&tsk->signal->oom_victims);
663 if (!atomic_dec_return(&oom_victims))
664 wake_up_all(&oom_victims_wait);
668 * oom_killer_disable - disable OOM killer
670 * Forces all page allocations to fail rather than trigger OOM killer.
671 * Will block and wait until all OOM victims are killed.
673 * The function cannot be called when there are runnable user tasks because
674 * the userspace would see unexpected allocation failures as a result. Any
675 * new usage of this function should be consulted with MM people.
677 * Returns true if successful and false if the OOM killer cannot be
680 bool oom_killer_disable(void)
683 * Make sure to not race with an ongoing OOM killer. Check that the
684 * current is not killed (possibly due to sharing the victim's memory).
686 if (mutex_lock_killable(&oom_lock))
688 oom_killer_disabled = true;
689 mutex_unlock(&oom_lock);
691 wait_event(oom_victims_wait, !atomic_read(&oom_victims));
697 * oom_killer_enable - enable OOM killer
699 void oom_killer_enable(void)
701 oom_killer_disabled = false;
704 static inline bool __task_will_free_mem(struct task_struct *task)
706 struct signal_struct *sig = task->signal;
709 * A coredumping process may sleep for an extended period in exit_mm(),
710 * so the oom killer cannot assume that the process will promptly exit
711 * and release memory.
713 if (sig->flags & SIGNAL_GROUP_COREDUMP)
716 if (sig->flags & SIGNAL_GROUP_EXIT)
719 if (thread_group_empty(task) && (task->flags & PF_EXITING))
726 * Checks whether the given task is dying or exiting and likely to
727 * release its address space. This means that all threads and processes
728 * sharing the same mm have to be killed or exiting.
730 bool task_will_free_mem(struct task_struct *task)
732 struct mm_struct *mm;
733 struct task_struct *p;
736 if (!__task_will_free_mem(task))
740 * If the process has passed exit_mm we have to skip it because
741 * we have lost a link to other tasks sharing this mm, we do not
742 * have anything to reap and the task might then get stuck waiting
743 * for parent as zombie and we do not want it to hold TIF_MEMDIE
745 p = find_lock_task_mm(task);
750 if (atomic_read(&mm->mm_users) <= 1) {
755 /* pin the mm to not get freed and reused */
756 atomic_inc(&mm->mm_count);
760 * This is really pessimistic but we do not have any reliable way
761 * to check that external processes share with our mm
764 for_each_process(p) {
765 if (!process_shares_mm(p, mm))
767 if (same_thread_group(task, p))
769 ret = __task_will_free_mem(p);
780 * Must be called while holding a reference to p, which will be released upon
783 void oom_kill_process(struct oom_control *oc, struct task_struct *p,
784 unsigned int points, unsigned long totalpages,
787 struct task_struct *victim = p;
788 struct task_struct *child;
789 struct task_struct *t;
790 struct mm_struct *mm;
791 unsigned int victim_points = 0;
792 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
793 DEFAULT_RATELIMIT_BURST);
794 bool can_oom_reap = true;
797 * If the task is already exiting, don't alarm the sysadmin or kill
798 * its children or threads, just set TIF_MEMDIE so it can die quickly
800 if (task_will_free_mem(p)) {
807 if (__ratelimit(&oom_rs))
810 pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n",
811 message, task_pid_nr(p), p->comm, points);
814 * If any of p's children has a different mm and is eligible for kill,
815 * the one with the highest oom_badness() score is sacrificed for its
816 * parent. This attempts to lose the minimal amount of work done while
817 * still freeing memory.
819 read_lock(&tasklist_lock);
820 for_each_thread(p, t) {
821 list_for_each_entry(child, &t->children, sibling) {
822 unsigned int child_points;
824 if (process_shares_mm(child, p->mm))
827 * oom_badness() returns 0 if the thread is unkillable
829 child_points = oom_badness(child,
830 oc->memcg, oc->nodemask, totalpages);
831 if (child_points > victim_points) {
832 put_task_struct(victim);
834 victim_points = child_points;
835 get_task_struct(victim);
839 read_unlock(&tasklist_lock);
841 p = find_lock_task_mm(victim);
843 put_task_struct(victim);
845 } else if (victim != p) {
847 put_task_struct(victim);
851 /* Get a reference to safely compare mm after task_unlock(victim) */
853 atomic_inc(&mm->mm_count);
855 * We should send SIGKILL before setting TIF_MEMDIE in order to prevent
856 * the OOM victim from depleting the memory reserves from the user
857 * space under its control.
859 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
860 mark_oom_victim(victim);
861 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
862 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
863 K(get_mm_counter(victim->mm, MM_ANONPAGES)),
864 K(get_mm_counter(victim->mm, MM_FILEPAGES)),
865 K(get_mm_counter(victim->mm, MM_SHMEMPAGES)));
869 * Kill all user processes sharing victim->mm in other thread groups, if
870 * any. They don't get access to memory reserves, though, to avoid
871 * depletion of all memory. This prevents mm->mmap_sem livelock when an
872 * oom killed thread cannot exit because it requires the semaphore and
873 * its contended by another thread trying to allocate memory itself.
874 * That thread will now get access to memory reserves since it has a
875 * pending fatal signal.
878 for_each_process(p) {
879 if (!process_shares_mm(p, mm))
881 if (same_thread_group(p, victim))
883 if (unlikely(p->flags & PF_KTHREAD) || is_global_init(p)) {
885 * We cannot use oom_reaper for the mm shared by this
886 * process because it wouldn't get killed and so the
887 * memory might be still used.
889 can_oom_reap = false;
892 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
897 wake_oom_reaper(victim);
900 put_task_struct(victim);
905 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
907 void check_panic_on_oom(struct oom_control *oc, enum oom_constraint constraint)
909 if (likely(!sysctl_panic_on_oom))
911 if (sysctl_panic_on_oom != 2) {
913 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
914 * does not panic for cpuset, mempolicy, or memcg allocation
917 if (constraint != CONSTRAINT_NONE)
920 /* Do not panic for oom kills triggered by sysrq */
921 if (is_sysrq_oom(oc))
923 dump_header(oc, NULL);
924 panic("Out of memory: %s panic_on_oom is enabled\n",
925 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
928 static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
930 int register_oom_notifier(struct notifier_block *nb)
932 return blocking_notifier_chain_register(&oom_notify_list, nb);
934 EXPORT_SYMBOL_GPL(register_oom_notifier);
936 int unregister_oom_notifier(struct notifier_block *nb)
938 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
940 EXPORT_SYMBOL_GPL(unregister_oom_notifier);
943 * out_of_memory - kill the "best" process when we run out of memory
944 * @oc: pointer to struct oom_control
946 * If we run out of memory, we have the choice between either
947 * killing a random task (bad), letting the system crash (worse)
948 * OR try to be smart about which process to kill. Note that we
949 * don't have to be perfect here, we just have to be good.
951 bool out_of_memory(struct oom_control *oc)
953 struct task_struct *p;
954 unsigned long totalpages;
955 unsigned long freed = 0;
956 unsigned int uninitialized_var(points);
957 enum oom_constraint constraint = CONSTRAINT_NONE;
959 if (oom_killer_disabled)
962 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
964 /* Got some memory back in the last second. */
968 * If current has a pending SIGKILL or is exiting, then automatically
969 * select it. The goal is to allow it to allocate so that it may
970 * quickly exit and free its memory.
972 * But don't select if current has already released its mm and cleared
973 * TIF_MEMDIE flag at exit_mm(), otherwise an OOM livelock may occur.
975 if (current->mm && task_will_free_mem(current)) {
976 mark_oom_victim(current);
977 wake_oom_reaper(current);
982 * The OOM killer does not compensate for IO-less reclaim.
983 * pagefault_out_of_memory lost its gfp context so we have to
984 * make sure exclude 0 mask - all other users should have at least
985 * ___GFP_DIRECT_RECLAIM to get here.
987 if (oc->gfp_mask && !(oc->gfp_mask & (__GFP_FS|__GFP_NOFAIL)))
991 * Check if there were limitations on the allocation (only relevant for
992 * NUMA) that may require different handling.
994 constraint = constrained_alloc(oc, &totalpages);
995 if (constraint != CONSTRAINT_MEMORY_POLICY)
997 check_panic_on_oom(oc, constraint);
999 if (sysctl_oom_kill_allocating_task && current->mm &&
1000 !oom_unkillable_task(current, NULL, oc->nodemask) &&
1001 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
1002 get_task_struct(current);
1003 oom_kill_process(oc, current, 0, totalpages,
1004 "Out of memory (oom_kill_allocating_task)");
1008 p = select_bad_process(oc, &points, totalpages);
1009 /* Found nothing?!?! Either we hang forever, or we panic. */
1010 if (!p && !is_sysrq_oom(oc)) {
1011 dump_header(oc, NULL);
1012 panic("Out of memory and no killable processes...\n");
1014 if (p && p != (void *)-1UL) {
1015 oom_kill_process(oc, p, points, totalpages, "Out of memory");
1017 * Give the killed process a good chance to exit before trying
1018 * to allocate memory again.
1020 schedule_timeout_killable(1);
1026 * The pagefault handler calls here because it is out of memory, so kill a
1027 * memory-hogging task. If oom_lock is held by somebody else, a parallel oom
1028 * killing is already in progress so do nothing.
1030 void pagefault_out_of_memory(void)
1032 struct oom_control oc = {
1040 if (mem_cgroup_oom_synchronize(true))
1043 if (!mutex_trylock(&oom_lock))
1046 if (!out_of_memory(&oc)) {
1048 * There shouldn't be any user tasks runnable while the
1049 * OOM killer is disabled, so the current task has to
1050 * be a racing OOM victim for which oom_killer_disable()
1053 WARN_ON(test_thread_flag(TIF_MEMDIE));
1056 mutex_unlock(&oom_lock);