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 static inline bool is_memcg_oom(struct oom_control *oc)
137 return oc->memcg != NULL;
140 /* return true if the task is not adequate as candidate victim task. */
141 static bool oom_unkillable_task(struct task_struct *p,
142 struct mem_cgroup *memcg, const nodemask_t *nodemask)
144 if (is_global_init(p))
146 if (p->flags & PF_KTHREAD)
149 /* When mem_cgroup_out_of_memory() and p is not member of the group */
150 if (memcg && !task_in_mem_cgroup(p, memcg))
153 /* p may not have freeable memory in nodemask */
154 if (!has_intersects_mems_allowed(p, nodemask))
161 * oom_badness - heuristic function to determine which candidate task to kill
162 * @p: task struct of which task we should calculate
163 * @totalpages: total present RAM allowed for page allocation
165 * The heuristic for determining which task to kill is made to be as simple and
166 * predictable as possible. The goal is to return the highest value for the
167 * task consuming the most memory to avoid subsequent oom failures.
169 unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
170 const nodemask_t *nodemask, unsigned long totalpages)
175 if (oom_unkillable_task(p, memcg, nodemask))
178 p = find_lock_task_mm(p);
183 * Do not even consider tasks which are explicitly marked oom
184 * unkillable or have been already oom reaped or the are in
185 * the middle of vfork
187 adj = (long)p->signal->oom_score_adj;
188 if (adj == OOM_SCORE_ADJ_MIN ||
189 test_bit(MMF_OOM_SKIP, &p->mm->flags) ||
196 * The baseline for the badness score is the proportion of RAM that each
197 * task's rss, pagetable and swap space use.
199 points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
200 atomic_long_read(&p->mm->nr_ptes) + mm_nr_pmds(p->mm);
204 * Root processes get 3% bonus, just like the __vm_enough_memory()
205 * implementation used by LSMs.
207 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
208 points -= (points * 3) / 100;
210 /* Normalize to oom_score_adj units */
211 adj *= totalpages / 1000;
215 * Never return 0 for an eligible task regardless of the root bonus and
216 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
218 return points > 0 ? points : 1;
221 enum oom_constraint {
224 CONSTRAINT_MEMORY_POLICY,
229 * Determine the type of allocation constraint.
231 static enum oom_constraint constrained_alloc(struct oom_control *oc)
235 enum zone_type high_zoneidx = gfp_zone(oc->gfp_mask);
236 bool cpuset_limited = false;
239 if (is_memcg_oom(oc)) {
240 oc->totalpages = mem_cgroup_get_limit(oc->memcg) ?: 1;
241 return CONSTRAINT_MEMCG;
244 /* Default to all available memory */
245 oc->totalpages = totalram_pages + total_swap_pages;
247 if (!IS_ENABLED(CONFIG_NUMA))
248 return CONSTRAINT_NONE;
251 return CONSTRAINT_NONE;
253 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
254 * to kill current.We have to random task kill in this case.
255 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
257 if (oc->gfp_mask & __GFP_THISNODE)
258 return CONSTRAINT_NONE;
261 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
262 * the page allocator means a mempolicy is in effect. Cpuset policy
263 * is enforced in get_page_from_freelist().
266 !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) {
267 oc->totalpages = total_swap_pages;
268 for_each_node_mask(nid, *oc->nodemask)
269 oc->totalpages += node_spanned_pages(nid);
270 return CONSTRAINT_MEMORY_POLICY;
273 /* Check this allocation failure is caused by cpuset's wall function */
274 for_each_zone_zonelist_nodemask(zone, z, oc->zonelist,
275 high_zoneidx, oc->nodemask)
276 if (!cpuset_zone_allowed(zone, oc->gfp_mask))
277 cpuset_limited = true;
279 if (cpuset_limited) {
280 oc->totalpages = total_swap_pages;
281 for_each_node_mask(nid, cpuset_current_mems_allowed)
282 oc->totalpages += node_spanned_pages(nid);
283 return CONSTRAINT_CPUSET;
285 return CONSTRAINT_NONE;
288 static int oom_evaluate_task(struct task_struct *task, void *arg)
290 struct oom_control *oc = arg;
291 unsigned long points;
293 if (oom_unkillable_task(task, NULL, oc->nodemask))
297 * This task already has access to memory reserves and is being killed.
298 * Don't allow any other task to have access to the reserves unless
299 * the task has MMF_OOM_SKIP because chances that it would release
300 * any memory is quite low.
302 if (!is_sysrq_oom(oc) && tsk_is_oom_victim(task)) {
303 if (test_bit(MMF_OOM_SKIP, &task->signal->oom_mm->flags))
309 * If task is allocating a lot of memory and has been marked to be
310 * killed first if it triggers an oom, then select it.
312 if (oom_task_origin(task)) {
317 points = oom_badness(task, NULL, oc->nodemask, oc->totalpages);
318 if (!points || points < oc->chosen_points)
321 /* Prefer thread group leaders for display purposes */
322 if (points == oc->chosen_points && thread_group_leader(oc->chosen))
326 put_task_struct(oc->chosen);
327 get_task_struct(task);
329 oc->chosen_points = points;
334 put_task_struct(oc->chosen);
335 oc->chosen = (void *)-1UL;
340 * Simple selection loop. We choose the process with the highest number of
341 * 'points'. In case scan was aborted, oc->chosen is set to -1.
343 static void select_bad_process(struct oom_control *oc)
345 if (is_memcg_oom(oc))
346 mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc);
348 struct task_struct *p;
352 if (oom_evaluate_task(p, oc))
357 oc->chosen_points = oc->chosen_points * 1000 / oc->totalpages;
361 * dump_tasks - dump current memory state of all system tasks
362 * @memcg: current's memory controller, if constrained
363 * @nodemask: nodemask passed to page allocator for mempolicy ooms
365 * Dumps the current memory state of all eligible tasks. Tasks not in the same
366 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
368 * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
369 * swapents, oom_score_adj value, and name.
371 static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask)
373 struct task_struct *p;
374 struct task_struct *task;
376 pr_info("[ pid ] uid tgid total_vm rss nr_ptes nr_pmds swapents oom_score_adj name\n");
378 for_each_process(p) {
379 if (oom_unkillable_task(p, memcg, nodemask))
382 task = find_lock_task_mm(p);
385 * This is a kthread or all of p's threads have already
386 * detached their mm's. There's no need to report
387 * them; they can't be oom killed anyway.
392 pr_info("[%5d] %5d %5d %8lu %8lu %7ld %7ld %8lu %5hd %s\n",
393 task->pid, from_kuid(&init_user_ns, task_uid(task)),
394 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
395 atomic_long_read(&task->mm->nr_ptes),
396 mm_nr_pmds(task->mm),
397 get_mm_counter(task->mm, MM_SWAPENTS),
398 task->signal->oom_score_adj, task->comm);
404 static void dump_header(struct oom_control *oc, struct task_struct *p)
406 pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), order=%d, oom_score_adj=%hd\n",
407 current->comm, oc->gfp_mask, &oc->gfp_mask, oc->order,
408 current->signal->oom_score_adj);
410 cpuset_print_current_mems_allowed();
413 mem_cgroup_print_oom_info(oc->memcg, p);
415 show_mem(SHOW_MEM_FILTER_NODES);
416 if (sysctl_oom_dump_tasks)
417 dump_tasks(oc->memcg, oc->nodemask);
421 * Number of OOM victims in flight
423 static atomic_t oom_victims = ATOMIC_INIT(0);
424 static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
426 static bool oom_killer_disabled __read_mostly;
428 #define K(x) ((x) << (PAGE_SHIFT-10))
431 * task->mm can be NULL if the task is the exited group leader. So to
432 * determine whether the task is using a particular mm, we examine all the
433 * task's threads: if one of those is using this mm then this task was also
436 bool process_shares_mm(struct task_struct *p, struct mm_struct *mm)
438 struct task_struct *t;
440 for_each_thread(p, t) {
441 struct mm_struct *t_mm = READ_ONCE(t->mm);
451 * OOM Reaper kernel thread which tries to reap the memory used by the OOM
452 * victim (if that is possible) to help the OOM killer to move on.
454 static struct task_struct *oom_reaper_th;
455 static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait);
456 static struct task_struct *oom_reaper_list;
457 static DEFINE_SPINLOCK(oom_reaper_lock);
459 static bool __oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
461 struct mmu_gather tlb;
462 struct vm_area_struct *vma;
463 struct zap_details details = {.check_swap_entries = true,
464 .ignore_dirty = true};
468 * We have to make sure to not race with the victim exit path
469 * and cause premature new oom victim selection:
470 * __oom_reap_task_mm exit_mm
473 * atomic_dec_and_test
478 * # no TIF_MEMDIE task selects new victim
479 * unmap_page_range # frees some memory
481 mutex_lock(&oom_lock);
483 if (!down_read_trylock(&mm->mmap_sem)) {
489 * increase mm_users only after we know we will reap something so
490 * that the mmput_async is called only when we have reaped something
491 * and delayed __mmput doesn't matter that much
493 if (!mmget_not_zero(mm)) {
494 up_read(&mm->mmap_sem);
498 tlb_gather_mmu(&tlb, mm, 0, -1);
499 for (vma = mm->mmap ; vma; vma = vma->vm_next) {
500 if (is_vm_hugetlb_page(vma))
504 * mlocked VMAs require explicit munlocking before unmap.
505 * Let's keep it simple here and skip such VMAs.
507 if (vma->vm_flags & VM_LOCKED)
511 * Only anonymous pages have a good chance to be dropped
512 * without additional steps which we cannot afford as we
515 * We do not even care about fs backed pages because all
516 * which are reclaimable have already been reclaimed and
517 * we do not want to block exit_mmap by keeping mm ref
518 * count elevated without a good reason.
520 if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED))
521 unmap_page_range(&tlb, vma, vma->vm_start, vma->vm_end,
524 tlb_finish_mmu(&tlb, 0, -1);
525 pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
526 task_pid_nr(tsk), tsk->comm,
527 K(get_mm_counter(mm, MM_ANONPAGES)),
528 K(get_mm_counter(mm, MM_FILEPAGES)),
529 K(get_mm_counter(mm, MM_SHMEMPAGES)));
530 up_read(&mm->mmap_sem);
533 * Drop our reference but make sure the mmput slow path is called from a
534 * different context because we shouldn't risk we get stuck there and
535 * put the oom_reaper out of the way.
539 mutex_unlock(&oom_lock);
543 #define MAX_OOM_REAP_RETRIES 10
544 static void oom_reap_task(struct task_struct *tsk)
547 struct mm_struct *mm = tsk->signal->oom_mm;
549 /* Retry the down_read_trylock(mmap_sem) a few times */
550 while (attempts++ < MAX_OOM_REAP_RETRIES && !__oom_reap_task_mm(tsk, mm))
551 schedule_timeout_idle(HZ/10);
553 if (attempts <= MAX_OOM_REAP_RETRIES)
557 pr_info("oom_reaper: unable to reap pid:%d (%s)\n",
558 task_pid_nr(tsk), tsk->comm);
559 debug_show_all_locks();
562 tsk->oom_reaper_list = NULL;
565 * Hide this mm from OOM killer because it has been either reaped or
566 * somebody can't call up_write(mmap_sem).
568 set_bit(MMF_OOM_SKIP, &mm->flags);
570 /* Drop a reference taken by wake_oom_reaper */
571 put_task_struct(tsk);
574 static int oom_reaper(void *unused)
577 struct task_struct *tsk = NULL;
579 wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL);
580 spin_lock(&oom_reaper_lock);
581 if (oom_reaper_list != NULL) {
582 tsk = oom_reaper_list;
583 oom_reaper_list = tsk->oom_reaper_list;
585 spin_unlock(&oom_reaper_lock);
594 static void wake_oom_reaper(struct task_struct *tsk)
599 /* tsk is already queued? */
600 if (tsk == oom_reaper_list || tsk->oom_reaper_list)
603 get_task_struct(tsk);
605 spin_lock(&oom_reaper_lock);
606 tsk->oom_reaper_list = oom_reaper_list;
607 oom_reaper_list = tsk;
608 spin_unlock(&oom_reaper_lock);
609 wake_up(&oom_reaper_wait);
612 static int __init oom_init(void)
614 oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper");
615 if (IS_ERR(oom_reaper_th)) {
616 pr_err("Unable to start OOM reaper %ld. Continuing regardless\n",
617 PTR_ERR(oom_reaper_th));
618 oom_reaper_th = NULL;
622 subsys_initcall(oom_init)
624 static inline void wake_oom_reaper(struct task_struct *tsk)
627 #endif /* CONFIG_MMU */
630 * mark_oom_victim - mark the given task as OOM victim
633 * Has to be called with oom_lock held and never after
634 * oom has been disabled already.
636 * tsk->mm has to be non NULL and caller has to guarantee it is stable (either
637 * under task_lock or operate on the current).
639 static void mark_oom_victim(struct task_struct *tsk)
641 struct mm_struct *mm = tsk->mm;
643 WARN_ON(oom_killer_disabled);
644 /* OOM killer might race with memcg OOM */
645 if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
648 /* oom_mm is bound to the signal struct life time. */
649 if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm))
650 atomic_inc(&tsk->signal->oom_mm->mm_count);
653 * Make sure that the task is woken up from uninterruptible sleep
654 * if it is frozen because OOM killer wouldn't be able to free
655 * any memory and livelock. freezing_slow_path will tell the freezer
656 * that TIF_MEMDIE tasks should be ignored.
659 atomic_inc(&oom_victims);
663 * exit_oom_victim - note the exit of an OOM victim
665 void exit_oom_victim(struct task_struct *tsk)
667 if (!test_and_clear_tsk_thread_flag(tsk, TIF_MEMDIE))
670 if (!atomic_dec_return(&oom_victims))
671 wake_up_all(&oom_victims_wait);
675 * oom_killer_enable - enable OOM killer
677 void oom_killer_enable(void)
679 oom_killer_disabled = false;
683 * oom_killer_disable - disable OOM killer
684 * @timeout: maximum timeout to wait for oom victims in jiffies
686 * Forces all page allocations to fail rather than trigger OOM killer.
687 * Will block and wait until all OOM victims are killed or the given
690 * The function cannot be called when there are runnable user tasks because
691 * the userspace would see unexpected allocation failures as a result. Any
692 * new usage of this function should be consulted with MM people.
694 * Returns true if successful and false if the OOM killer cannot be
697 bool oom_killer_disable(signed long timeout)
702 * Make sure to not race with an ongoing OOM killer. Check that the
703 * current is not killed (possibly due to sharing the victim's memory).
705 if (mutex_lock_killable(&oom_lock))
707 oom_killer_disabled = true;
708 mutex_unlock(&oom_lock);
710 ret = wait_event_interruptible_timeout(oom_victims_wait,
711 !atomic_read(&oom_victims), timeout);
720 static inline bool __task_will_free_mem(struct task_struct *task)
722 struct signal_struct *sig = task->signal;
725 * A coredumping process may sleep for an extended period in exit_mm(),
726 * so the oom killer cannot assume that the process will promptly exit
727 * and release memory.
729 if (sig->flags & SIGNAL_GROUP_COREDUMP)
732 if (sig->flags & SIGNAL_GROUP_EXIT)
735 if (thread_group_empty(task) && (task->flags & PF_EXITING))
742 * Checks whether the given task is dying or exiting and likely to
743 * release its address space. This means that all threads and processes
744 * sharing the same mm have to be killed or exiting.
745 * Caller has to make sure that task->mm is stable (hold task_lock or
746 * it operates on the current).
748 static bool task_will_free_mem(struct task_struct *task)
750 struct mm_struct *mm = task->mm;
751 struct task_struct *p;
755 * Skip tasks without mm because it might have passed its exit_mm and
756 * exit_oom_victim. oom_reaper could have rescued that but do not rely
757 * on that for now. We can consider find_lock_task_mm in future.
762 if (!__task_will_free_mem(task))
766 * This task has already been drained by the oom reaper so there are
767 * only small chances it will free some more
769 if (test_bit(MMF_OOM_SKIP, &mm->flags))
772 if (atomic_read(&mm->mm_users) <= 1)
776 * Make sure that all tasks which share the mm with the given tasks
777 * are dying as well to make sure that a) nobody pins its mm and
778 * b) the task is also reapable by the oom reaper.
781 for_each_process(p) {
782 if (!process_shares_mm(p, mm))
784 if (same_thread_group(task, p))
786 ret = __task_will_free_mem(p);
795 static void oom_kill_process(struct oom_control *oc, const char *message)
797 struct task_struct *p = oc->chosen;
798 unsigned int points = oc->chosen_points;
799 struct task_struct *victim = p;
800 struct task_struct *child;
801 struct task_struct *t;
802 struct mm_struct *mm;
803 unsigned int victim_points = 0;
804 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
805 DEFAULT_RATELIMIT_BURST);
806 bool can_oom_reap = true;
809 * If the task is already exiting, don't alarm the sysadmin or kill
810 * its children or threads, just set TIF_MEMDIE so it can die quickly
813 if (task_will_free_mem(p)) {
822 if (__ratelimit(&oom_rs))
825 pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n",
826 message, task_pid_nr(p), p->comm, points);
829 * If any of p's children has a different mm and is eligible for kill,
830 * the one with the highest oom_badness() score is sacrificed for its
831 * parent. This attempts to lose the minimal amount of work done while
832 * still freeing memory.
834 read_lock(&tasklist_lock);
835 for_each_thread(p, t) {
836 list_for_each_entry(child, &t->children, sibling) {
837 unsigned int child_points;
839 if (process_shares_mm(child, p->mm))
842 * oom_badness() returns 0 if the thread is unkillable
844 child_points = oom_badness(child,
845 oc->memcg, oc->nodemask, oc->totalpages);
846 if (child_points > victim_points) {
847 put_task_struct(victim);
849 victim_points = child_points;
850 get_task_struct(victim);
854 read_unlock(&tasklist_lock);
856 p = find_lock_task_mm(victim);
858 put_task_struct(victim);
860 } else if (victim != p) {
862 put_task_struct(victim);
866 /* Get a reference to safely compare mm after task_unlock(victim) */
868 atomic_inc(&mm->mm_count);
870 * We should send SIGKILL before setting TIF_MEMDIE in order to prevent
871 * the OOM victim from depleting the memory reserves from the user
872 * space under its control.
874 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
875 mark_oom_victim(victim);
876 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
877 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
878 K(get_mm_counter(victim->mm, MM_ANONPAGES)),
879 K(get_mm_counter(victim->mm, MM_FILEPAGES)),
880 K(get_mm_counter(victim->mm, MM_SHMEMPAGES)));
884 * Kill all user processes sharing victim->mm in other thread groups, if
885 * any. They don't get access to memory reserves, though, to avoid
886 * depletion of all memory. This prevents mm->mmap_sem livelock when an
887 * oom killed thread cannot exit because it requires the semaphore and
888 * its contended by another thread trying to allocate memory itself.
889 * That thread will now get access to memory reserves since it has a
890 * pending fatal signal.
893 for_each_process(p) {
894 if (!process_shares_mm(p, mm))
896 if (same_thread_group(p, victim))
898 if (unlikely(p->flags & PF_KTHREAD) || is_global_init(p)) {
900 * We cannot use oom_reaper for the mm shared by this
901 * process because it wouldn't get killed and so the
902 * memory might be still used. Hide the mm from the oom
903 * killer to guarantee OOM forward progress.
905 can_oom_reap = false;
906 set_bit(MMF_OOM_SKIP, &mm->flags);
907 pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n",
908 task_pid_nr(victim), victim->comm,
909 task_pid_nr(p), p->comm);
912 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
917 wake_oom_reaper(victim);
920 put_task_struct(victim);
925 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
927 static void check_panic_on_oom(struct oom_control *oc,
928 enum oom_constraint constraint)
930 if (likely(!sysctl_panic_on_oom))
932 if (sysctl_panic_on_oom != 2) {
934 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
935 * does not panic for cpuset, mempolicy, or memcg allocation
938 if (constraint != CONSTRAINT_NONE)
941 /* Do not panic for oom kills triggered by sysrq */
942 if (is_sysrq_oom(oc))
944 dump_header(oc, NULL);
945 panic("Out of memory: %s panic_on_oom is enabled\n",
946 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
949 static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
951 int register_oom_notifier(struct notifier_block *nb)
953 return blocking_notifier_chain_register(&oom_notify_list, nb);
955 EXPORT_SYMBOL_GPL(register_oom_notifier);
957 int unregister_oom_notifier(struct notifier_block *nb)
959 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
961 EXPORT_SYMBOL_GPL(unregister_oom_notifier);
964 * out_of_memory - kill the "best" process when we run out of memory
965 * @oc: pointer to struct oom_control
967 * If we run out of memory, we have the choice between either
968 * killing a random task (bad), letting the system crash (worse)
969 * OR try to be smart about which process to kill. Note that we
970 * don't have to be perfect here, we just have to be good.
972 bool out_of_memory(struct oom_control *oc)
974 unsigned long freed = 0;
975 enum oom_constraint constraint = CONSTRAINT_NONE;
977 if (oom_killer_disabled)
980 if (!is_memcg_oom(oc)) {
981 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
983 /* Got some memory back in the last second. */
988 * If current has a pending SIGKILL or is exiting, then automatically
989 * select it. The goal is to allow it to allocate so that it may
990 * quickly exit and free its memory.
992 if (task_will_free_mem(current)) {
993 mark_oom_victim(current);
994 wake_oom_reaper(current);
999 * The OOM killer does not compensate for IO-less reclaim.
1000 * pagefault_out_of_memory lost its gfp context so we have to
1001 * make sure exclude 0 mask - all other users should have at least
1002 * ___GFP_DIRECT_RECLAIM to get here.
1004 if (oc->gfp_mask && !(oc->gfp_mask & (__GFP_FS|__GFP_NOFAIL)))
1008 * Check if there were limitations on the allocation (only relevant for
1009 * NUMA and memcg) that may require different handling.
1011 constraint = constrained_alloc(oc);
1012 if (constraint != CONSTRAINT_MEMORY_POLICY)
1013 oc->nodemask = NULL;
1014 check_panic_on_oom(oc, constraint);
1016 if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task &&
1017 current->mm && !oom_unkillable_task(current, NULL, oc->nodemask) &&
1018 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
1019 get_task_struct(current);
1020 oc->chosen = current;
1021 oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)");
1025 select_bad_process(oc);
1026 /* Found nothing?!?! Either we hang forever, or we panic. */
1027 if (!oc->chosen && !is_sysrq_oom(oc) && !is_memcg_oom(oc)) {
1028 dump_header(oc, NULL);
1029 panic("Out of memory and no killable processes...\n");
1031 if (oc->chosen && oc->chosen != (void *)-1UL) {
1032 oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" :
1033 "Memory cgroup out of memory");
1035 * Give the killed process a good chance to exit before trying
1036 * to allocate memory again.
1038 schedule_timeout_killable(1);
1040 return !!oc->chosen;
1044 * The pagefault handler calls here because it is out of memory, so kill a
1045 * memory-hogging task. If oom_lock is held by somebody else, a parallel oom
1046 * killing is already in progress so do nothing.
1048 void pagefault_out_of_memory(void)
1050 struct oom_control oc = {
1058 if (mem_cgroup_oom_synchronize(true))
1061 if (!mutex_trylock(&oom_lock))
1064 if (!out_of_memory(&oc)) {
1066 * There shouldn't be any user tasks runnable while the
1067 * OOM killer is disabled, so the current task has to
1068 * be a racing OOM victim for which oom_killer_disable()
1071 WARN_ON(test_thread_flag(TIF_MEMDIE));
1074 mutex_unlock(&oom_lock);