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/sched/mm.h>
26 #include <linux/sched/coredump.h>
27 #include <linux/sched/task.h>
28 #include <linux/swap.h>
29 #include <linux/timex.h>
30 #include <linux/jiffies.h>
31 #include <linux/cpuset.h>
32 #include <linux/export.h>
33 #include <linux/notifier.h>
34 #include <linux/memcontrol.h>
35 #include <linux/mempolicy.h>
36 #include <linux/security.h>
37 #include <linux/ptrace.h>
38 #include <linux/freezer.h>
39 #include <linux/ftrace.h>
40 #include <linux/ratelimit.h>
41 #include <linux/kthread.h>
42 #include <linux/init.h>
43 #include <linux/mmu_notifier.h>
49 #define CREATE_TRACE_POINTS
50 #include <trace/events/oom.h>
52 int sysctl_panic_on_oom;
53 int sysctl_oom_kill_allocating_task;
54 int sysctl_oom_dump_tasks = 1;
57 * Serializes oom killer invocations (out_of_memory()) from all contexts to
58 * prevent from over eager oom killing (e.g. when the oom killer is invoked
59 * from different domains).
61 * oom_killer_disable() relies on this lock to stabilize oom_killer_disabled
64 DEFINE_MUTEX(oom_lock);
68 * has_intersects_mems_allowed() - check task eligiblity for kill
69 * @start: task struct of which task to consider
70 * @mask: nodemask passed to page allocator for mempolicy ooms
72 * Task eligibility is determined by whether or not a candidate task, @tsk,
73 * shares the same mempolicy nodes as current if it is bound by such a policy
74 * and whether or not it has the same set of allowed cpuset nodes.
76 static bool has_intersects_mems_allowed(struct task_struct *start,
77 const nodemask_t *mask)
79 struct task_struct *tsk;
83 for_each_thread(start, tsk) {
86 * If this is a mempolicy constrained oom, tsk's
87 * cpuset is irrelevant. Only return true if its
88 * mempolicy intersects current, otherwise it may be
91 ret = mempolicy_nodemask_intersects(tsk, mask);
94 * This is not a mempolicy constrained oom, so only
95 * check the mems of tsk's cpuset.
97 ret = cpuset_mems_allowed_intersects(current, tsk);
107 static bool has_intersects_mems_allowed(struct task_struct *tsk,
108 const nodemask_t *mask)
112 #endif /* CONFIG_NUMA */
115 * The process p may have detached its own ->mm while exiting or through
116 * use_mm(), but one or more of its subthreads may still have a valid
117 * pointer. Return p, or any of its subthreads with a valid ->mm, with
120 struct task_struct *find_lock_task_mm(struct task_struct *p)
122 struct task_struct *t;
126 for_each_thread(p, t) {
140 * order == -1 means the oom kill is required by sysrq, otherwise only
141 * for display purposes.
143 static inline bool is_sysrq_oom(struct oom_control *oc)
145 return oc->order == -1;
148 static inline bool is_memcg_oom(struct oom_control *oc)
150 return oc->memcg != NULL;
153 /* return true if the task is not adequate as candidate victim task. */
154 static bool oom_unkillable_task(struct task_struct *p,
155 struct mem_cgroup *memcg, const nodemask_t *nodemask)
157 if (is_global_init(p))
159 if (p->flags & PF_KTHREAD)
162 /* When mem_cgroup_out_of_memory() and p is not member of the group */
163 if (memcg && !task_in_mem_cgroup(p, memcg))
166 /* p may not have freeable memory in nodemask */
167 if (!has_intersects_mems_allowed(p, nodemask))
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(NR_SLAB_UNRECLAIMABLE) > 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
196 * @memcg: task's memory controller, if constrained
197 * @nodemask: nodemask passed to page allocator for mempolicy ooms
199 * The heuristic for determining which task to kill is made to be as simple and
200 * predictable as possible. The goal is to return the highest value for the
201 * task consuming the most memory to avoid subsequent oom failures.
203 unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
204 const nodemask_t *nodemask, unsigned long totalpages)
209 if (oom_unkillable_task(p, memcg, nodemask))
212 p = find_lock_task_mm(p);
217 * Do not even consider tasks which are explicitly marked oom
218 * unkillable or have been already oom reaped or the are in
219 * the middle of vfork
221 adj = (long)p->signal->oom_score_adj;
222 if (adj == OOM_SCORE_ADJ_MIN ||
223 test_bit(MMF_OOM_SKIP, &p->mm->flags) ||
230 * The baseline for the badness score is the proportion of RAM that each
231 * task's rss, pagetable and swap space use.
233 points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
234 mm_pgtables_bytes(p->mm) / PAGE_SIZE;
237 /* Normalize to oom_score_adj units */
238 adj *= totalpages / 1000;
242 * Never return 0 for an eligible task regardless of the root bonus and
243 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
245 return points > 0 ? points : 1;
248 enum oom_constraint {
251 CONSTRAINT_MEMORY_POLICY,
256 * Determine the type of allocation constraint.
258 static enum oom_constraint constrained_alloc(struct oom_control *oc)
262 enum zone_type high_zoneidx = gfp_zone(oc->gfp_mask);
263 bool cpuset_limited = false;
266 if (is_memcg_oom(oc)) {
267 oc->totalpages = mem_cgroup_get_max(oc->memcg) ?: 1;
268 return CONSTRAINT_MEMCG;
271 /* Default to all available memory */
272 oc->totalpages = totalram_pages + total_swap_pages;
274 if (!IS_ENABLED(CONFIG_NUMA))
275 return CONSTRAINT_NONE;
278 return CONSTRAINT_NONE;
280 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
281 * to kill current.We have to random task kill in this case.
282 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
284 if (oc->gfp_mask & __GFP_THISNODE)
285 return CONSTRAINT_NONE;
288 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
289 * the page allocator means a mempolicy is in effect. Cpuset policy
290 * is enforced in get_page_from_freelist().
293 !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) {
294 oc->totalpages = total_swap_pages;
295 for_each_node_mask(nid, *oc->nodemask)
296 oc->totalpages += node_spanned_pages(nid);
297 return CONSTRAINT_MEMORY_POLICY;
300 /* Check this allocation failure is caused by cpuset's wall function */
301 for_each_zone_zonelist_nodemask(zone, z, oc->zonelist,
302 high_zoneidx, oc->nodemask)
303 if (!cpuset_zone_allowed(zone, oc->gfp_mask))
304 cpuset_limited = true;
306 if (cpuset_limited) {
307 oc->totalpages = total_swap_pages;
308 for_each_node_mask(nid, cpuset_current_mems_allowed)
309 oc->totalpages += node_spanned_pages(nid);
310 return CONSTRAINT_CPUSET;
312 return CONSTRAINT_NONE;
315 static int oom_evaluate_task(struct task_struct *task, void *arg)
317 struct oom_control *oc = arg;
318 unsigned long points;
320 if (oom_unkillable_task(task, NULL, oc->nodemask))
324 * This task already has access to memory reserves and is being killed.
325 * Don't allow any other task to have access to the reserves unless
326 * the task has MMF_OOM_SKIP because chances that it would release
327 * any memory is quite low.
329 if (!is_sysrq_oom(oc) && tsk_is_oom_victim(task)) {
330 if (test_bit(MMF_OOM_SKIP, &task->signal->oom_mm->flags))
336 * If task is allocating a lot of memory and has been marked to be
337 * killed first if it triggers an oom, then select it.
339 if (oom_task_origin(task)) {
344 points = oom_badness(task, NULL, oc->nodemask, oc->totalpages);
345 if (!points || points < oc->chosen_points)
348 /* Prefer thread group leaders for display purposes */
349 if (points == oc->chosen_points && thread_group_leader(oc->chosen))
353 put_task_struct(oc->chosen);
354 get_task_struct(task);
356 oc->chosen_points = points;
361 put_task_struct(oc->chosen);
362 oc->chosen = (void *)-1UL;
367 * Simple selection loop. We choose the process with the highest number of
368 * 'points'. In case scan was aborted, oc->chosen is set to -1.
370 static void select_bad_process(struct oom_control *oc)
372 if (is_memcg_oom(oc))
373 mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc);
375 struct task_struct *p;
379 if (oom_evaluate_task(p, oc))
384 oc->chosen_points = oc->chosen_points * 1000 / oc->totalpages;
388 * dump_tasks - dump current memory state of all system tasks
389 * @memcg: current's memory controller, if constrained
390 * @nodemask: nodemask passed to page allocator for mempolicy ooms
392 * Dumps the current memory state of all eligible tasks. Tasks not in the same
393 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
395 * State information includes task's pid, uid, tgid, vm size, rss,
396 * pgtables_bytes, swapents, oom_score_adj value, and name.
398 static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask)
400 struct task_struct *p;
401 struct task_struct *task;
403 pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name\n");
405 for_each_process(p) {
406 if (oom_unkillable_task(p, memcg, nodemask))
409 task = find_lock_task_mm(p);
412 * This is a kthread or all of p's threads have already
413 * detached their mm's. There's no need to report
414 * them; they can't be oom killed anyway.
419 pr_info("[%5d] %5d %5d %8lu %8lu %8ld %8lu %5hd %s\n",
420 task->pid, from_kuid(&init_user_ns, task_uid(task)),
421 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
422 mm_pgtables_bytes(task->mm),
423 get_mm_counter(task->mm, MM_SWAPENTS),
424 task->signal->oom_score_adj, task->comm);
430 static void dump_header(struct oom_control *oc, struct task_struct *p)
432 pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), nodemask=%*pbl, order=%d, oom_score_adj=%hd\n",
433 current->comm, oc->gfp_mask, &oc->gfp_mask,
434 nodemask_pr_args(oc->nodemask), oc->order,
435 current->signal->oom_score_adj);
436 if (!IS_ENABLED(CONFIG_COMPACTION) && oc->order)
437 pr_warn("COMPACTION is disabled!!!\n");
439 cpuset_print_current_mems_allowed();
441 if (is_memcg_oom(oc))
442 mem_cgroup_print_oom_info(oc->memcg, p);
444 show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask);
445 if (is_dump_unreclaim_slabs())
446 dump_unreclaimable_slab();
448 if (sysctl_oom_dump_tasks)
449 dump_tasks(oc->memcg, oc->nodemask);
453 * Number of OOM victims in flight
455 static atomic_t oom_victims = ATOMIC_INIT(0);
456 static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
458 static bool oom_killer_disabled __read_mostly;
460 #define K(x) ((x) << (PAGE_SHIFT-10))
463 * task->mm can be NULL if the task is the exited group leader. So to
464 * determine whether the task is using a particular mm, we examine all the
465 * task's threads: if one of those is using this mm then this task was also
468 bool process_shares_mm(struct task_struct *p, struct mm_struct *mm)
470 struct task_struct *t;
472 for_each_thread(p, t) {
473 struct mm_struct *t_mm = READ_ONCE(t->mm);
482 * OOM Reaper kernel thread which tries to reap the memory used by the OOM
483 * victim (if that is possible) to help the OOM killer to move on.
485 static struct task_struct *oom_reaper_th;
486 static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait);
487 static struct task_struct *oom_reaper_list;
488 static DEFINE_SPINLOCK(oom_reaper_lock);
490 void __oom_reap_task_mm(struct mm_struct *mm)
492 struct vm_area_struct *vma;
495 * Tell all users of get_user/copy_from_user etc... that the content
496 * is no longer stable. No barriers really needed because unmapping
497 * should imply barriers already and the reader would hit a page fault
498 * if it stumbled over a reaped memory.
500 set_bit(MMF_UNSTABLE, &mm->flags);
502 for (vma = mm->mmap ; vma; vma = vma->vm_next) {
503 if (!can_madv_dontneed_vma(vma))
507 * Only anonymous pages have a good chance to be dropped
508 * without additional steps which we cannot afford as we
511 * We do not even care about fs backed pages because all
512 * which are reclaimable have already been reclaimed and
513 * we do not want to block exit_mmap by keeping mm ref
514 * count elevated without a good reason.
516 if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) {
517 const unsigned long start = vma->vm_start;
518 const unsigned long end = vma->vm_end;
519 struct mmu_gather tlb;
521 tlb_gather_mmu(&tlb, mm, start, end);
522 mmu_notifier_invalidate_range_start(mm, start, end);
523 unmap_page_range(&tlb, vma, start, end, NULL);
524 mmu_notifier_invalidate_range_end(mm, start, end);
525 tlb_finish_mmu(&tlb, start, end);
530 static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
535 * We have to make sure to not race with the victim exit path
536 * and cause premature new oom victim selection:
537 * oom_reap_task_mm exit_mm
540 * atomic_dec_and_test
545 * # no TIF_MEMDIE task selects new victim
546 * unmap_page_range # frees some memory
548 mutex_lock(&oom_lock);
550 if (!down_read_trylock(&mm->mmap_sem)) {
552 trace_skip_task_reaping(tsk->pid);
557 * If the mm has invalidate_{start,end}() notifiers that could block,
558 * sleep to give the oom victim some more time.
559 * TODO: we really want to get rid of this ugly hack and make sure that
560 * notifiers cannot block for unbounded amount of time
562 if (mm_has_blockable_invalidate_notifiers(mm)) {
563 up_read(&mm->mmap_sem);
564 schedule_timeout_idle(HZ);
569 * MMF_OOM_SKIP is set by exit_mmap when the OOM reaper can't
570 * work on the mm anymore. The check for MMF_OOM_SKIP must run
571 * under mmap_sem for reading because it serializes against the
572 * down_write();up_write() cycle in exit_mmap().
574 if (test_bit(MMF_OOM_SKIP, &mm->flags)) {
575 up_read(&mm->mmap_sem);
576 trace_skip_task_reaping(tsk->pid);
580 trace_start_task_reaping(tsk->pid);
582 __oom_reap_task_mm(mm);
584 pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
585 task_pid_nr(tsk), tsk->comm,
586 K(get_mm_counter(mm, MM_ANONPAGES)),
587 K(get_mm_counter(mm, MM_FILEPAGES)),
588 K(get_mm_counter(mm, MM_SHMEMPAGES)));
589 up_read(&mm->mmap_sem);
591 trace_finish_task_reaping(tsk->pid);
593 mutex_unlock(&oom_lock);
597 #define MAX_OOM_REAP_RETRIES 10
598 static void oom_reap_task(struct task_struct *tsk)
601 struct mm_struct *mm = tsk->signal->oom_mm;
603 /* Retry the down_read_trylock(mmap_sem) a few times */
604 while (attempts++ < MAX_OOM_REAP_RETRIES && !oom_reap_task_mm(tsk, mm))
605 schedule_timeout_idle(HZ/10);
607 if (attempts <= MAX_OOM_REAP_RETRIES ||
608 test_bit(MMF_OOM_SKIP, &mm->flags))
611 pr_info("oom_reaper: unable to reap pid:%d (%s)\n",
612 task_pid_nr(tsk), tsk->comm);
613 debug_show_all_locks();
616 tsk->oom_reaper_list = NULL;
619 * Hide this mm from OOM killer because it has been either reaped or
620 * somebody can't call up_write(mmap_sem).
622 set_bit(MMF_OOM_SKIP, &mm->flags);
624 /* Drop a reference taken by wake_oom_reaper */
625 put_task_struct(tsk);
628 static int oom_reaper(void *unused)
631 struct task_struct *tsk = NULL;
633 wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL);
634 spin_lock(&oom_reaper_lock);
635 if (oom_reaper_list != NULL) {
636 tsk = oom_reaper_list;
637 oom_reaper_list = tsk->oom_reaper_list;
639 spin_unlock(&oom_reaper_lock);
648 static void wake_oom_reaper(struct task_struct *tsk)
650 /* tsk is already queued? */
651 if (tsk == oom_reaper_list || tsk->oom_reaper_list)
654 get_task_struct(tsk);
656 spin_lock(&oom_reaper_lock);
657 tsk->oom_reaper_list = oom_reaper_list;
658 oom_reaper_list = tsk;
659 spin_unlock(&oom_reaper_lock);
660 trace_wake_reaper(tsk->pid);
661 wake_up(&oom_reaper_wait);
664 static int __init oom_init(void)
666 oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper");
669 subsys_initcall(oom_init)
671 static inline void wake_oom_reaper(struct task_struct *tsk)
674 #endif /* CONFIG_MMU */
677 * mark_oom_victim - mark the given task as OOM victim
680 * Has to be called with oom_lock held and never after
681 * oom has been disabled already.
683 * tsk->mm has to be non NULL and caller has to guarantee it is stable (either
684 * under task_lock or operate on the current).
686 static void mark_oom_victim(struct task_struct *tsk)
688 struct mm_struct *mm = tsk->mm;
690 WARN_ON(oom_killer_disabled);
691 /* OOM killer might race with memcg OOM */
692 if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
695 /* oom_mm is bound to the signal struct life time. */
696 if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm)) {
697 mmgrab(tsk->signal->oom_mm);
698 set_bit(MMF_OOM_VICTIM, &mm->flags);
702 * Make sure that the task is woken up from uninterruptible sleep
703 * if it is frozen because OOM killer wouldn't be able to free
704 * any memory and livelock. freezing_slow_path will tell the freezer
705 * that TIF_MEMDIE tasks should be ignored.
708 atomic_inc(&oom_victims);
709 trace_mark_victim(tsk->pid);
713 * exit_oom_victim - note the exit of an OOM victim
715 void exit_oom_victim(void)
717 clear_thread_flag(TIF_MEMDIE);
719 if (!atomic_dec_return(&oom_victims))
720 wake_up_all(&oom_victims_wait);
724 * oom_killer_enable - enable OOM killer
726 void oom_killer_enable(void)
728 oom_killer_disabled = false;
729 pr_info("OOM killer enabled.\n");
733 * oom_killer_disable - disable OOM killer
734 * @timeout: maximum timeout to wait for oom victims in jiffies
736 * Forces all page allocations to fail rather than trigger OOM killer.
737 * Will block and wait until all OOM victims are killed or the given
740 * The function cannot be called when there are runnable user tasks because
741 * the userspace would see unexpected allocation failures as a result. Any
742 * new usage of this function should be consulted with MM people.
744 * Returns true if successful and false if the OOM killer cannot be
747 bool oom_killer_disable(signed long timeout)
752 * Make sure to not race with an ongoing OOM killer. Check that the
753 * current is not killed (possibly due to sharing the victim's memory).
755 if (mutex_lock_killable(&oom_lock))
757 oom_killer_disabled = true;
758 mutex_unlock(&oom_lock);
760 ret = wait_event_interruptible_timeout(oom_victims_wait,
761 !atomic_read(&oom_victims), timeout);
766 pr_info("OOM killer disabled.\n");
771 static inline bool __task_will_free_mem(struct task_struct *task)
773 struct signal_struct *sig = task->signal;
776 * A coredumping process may sleep for an extended period in exit_mm(),
777 * so the oom killer cannot assume that the process will promptly exit
778 * and release memory.
780 if (sig->flags & SIGNAL_GROUP_COREDUMP)
783 if (sig->flags & SIGNAL_GROUP_EXIT)
786 if (thread_group_empty(task) && (task->flags & PF_EXITING))
793 * Checks whether the given task is dying or exiting and likely to
794 * release its address space. This means that all threads and processes
795 * sharing the same mm have to be killed or exiting.
796 * Caller has to make sure that task->mm is stable (hold task_lock or
797 * it operates on the current).
799 static bool task_will_free_mem(struct task_struct *task)
801 struct mm_struct *mm = task->mm;
802 struct task_struct *p;
806 * Skip tasks without mm because it might have passed its exit_mm and
807 * exit_oom_victim. oom_reaper could have rescued that but do not rely
808 * on that for now. We can consider find_lock_task_mm in future.
813 if (!__task_will_free_mem(task))
817 * This task has already been drained by the oom reaper so there are
818 * only small chances it will free some more
820 if (test_bit(MMF_OOM_SKIP, &mm->flags))
823 if (atomic_read(&mm->mm_users) <= 1)
827 * Make sure that all tasks which share the mm with the given tasks
828 * are dying as well to make sure that a) nobody pins its mm and
829 * b) the task is also reapable by the oom reaper.
832 for_each_process(p) {
833 if (!process_shares_mm(p, mm))
835 if (same_thread_group(task, p))
837 ret = __task_will_free_mem(p);
846 static void oom_kill_process(struct oom_control *oc, const char *message)
848 struct task_struct *p = oc->chosen;
849 unsigned int points = oc->chosen_points;
850 struct task_struct *victim = p;
851 struct task_struct *child;
852 struct task_struct *t;
853 struct mm_struct *mm;
854 unsigned int victim_points = 0;
855 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
856 DEFAULT_RATELIMIT_BURST);
857 bool can_oom_reap = true;
860 * If the task is already exiting, don't alarm the sysadmin or kill
861 * its children or threads, just give it access to memory reserves
862 * so it can die quickly
865 if (task_will_free_mem(p)) {
874 if (__ratelimit(&oom_rs))
877 pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n",
878 message, task_pid_nr(p), p->comm, points);
881 * If any of p's children has a different mm and is eligible for kill,
882 * the one with the highest oom_badness() score is sacrificed for its
883 * parent. This attempts to lose the minimal amount of work done while
884 * still freeing memory.
886 read_lock(&tasklist_lock);
887 for_each_thread(p, t) {
888 list_for_each_entry(child, &t->children, sibling) {
889 unsigned int child_points;
891 if (process_shares_mm(child, p->mm))
894 * oom_badness() returns 0 if the thread is unkillable
896 child_points = oom_badness(child,
897 oc->memcg, oc->nodemask, oc->totalpages);
898 if (child_points > victim_points) {
899 put_task_struct(victim);
901 victim_points = child_points;
902 get_task_struct(victim);
906 read_unlock(&tasklist_lock);
908 p = find_lock_task_mm(victim);
910 put_task_struct(victim);
912 } else if (victim != p) {
914 put_task_struct(victim);
918 /* Get a reference to safely compare mm after task_unlock(victim) */
922 /* Raise event before sending signal: task reaper must see this */
923 count_vm_event(OOM_KILL);
924 memcg_memory_event_mm(mm, MEMCG_OOM_KILL);
927 * We should send SIGKILL before granting access to memory reserves
928 * in order to prevent the OOM victim from depleting the memory
929 * reserves from the user space under its control.
931 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
932 mark_oom_victim(victim);
933 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
934 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
935 K(get_mm_counter(victim->mm, MM_ANONPAGES)),
936 K(get_mm_counter(victim->mm, MM_FILEPAGES)),
937 K(get_mm_counter(victim->mm, MM_SHMEMPAGES)));
941 * Kill all user processes sharing victim->mm in other thread groups, if
942 * any. They don't get access to memory reserves, though, to avoid
943 * depletion of all memory. This prevents mm->mmap_sem livelock when an
944 * oom killed thread cannot exit because it requires the semaphore and
945 * its contended by another thread trying to allocate memory itself.
946 * That thread will now get access to memory reserves since it has a
947 * pending fatal signal.
950 for_each_process(p) {
951 if (!process_shares_mm(p, mm))
953 if (same_thread_group(p, victim))
955 if (is_global_init(p)) {
956 can_oom_reap = false;
957 set_bit(MMF_OOM_SKIP, &mm->flags);
958 pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n",
959 task_pid_nr(victim), victim->comm,
960 task_pid_nr(p), p->comm);
964 * No use_mm() user needs to read from the userspace so we are
967 if (unlikely(p->flags & PF_KTHREAD))
969 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
974 wake_oom_reaper(victim);
977 put_task_struct(victim);
982 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
984 static void check_panic_on_oom(struct oom_control *oc,
985 enum oom_constraint constraint)
987 if (likely(!sysctl_panic_on_oom))
989 if (sysctl_panic_on_oom != 2) {
991 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
992 * does not panic for cpuset, mempolicy, or memcg allocation
995 if (constraint != CONSTRAINT_NONE)
998 /* Do not panic for oom kills triggered by sysrq */
999 if (is_sysrq_oom(oc))
1001 dump_header(oc, NULL);
1002 panic("Out of memory: %s panic_on_oom is enabled\n",
1003 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
1006 static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
1008 int register_oom_notifier(struct notifier_block *nb)
1010 return blocking_notifier_chain_register(&oom_notify_list, nb);
1012 EXPORT_SYMBOL_GPL(register_oom_notifier);
1014 int unregister_oom_notifier(struct notifier_block *nb)
1016 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
1018 EXPORT_SYMBOL_GPL(unregister_oom_notifier);
1021 * out_of_memory - kill the "best" process when we run out of memory
1022 * @oc: pointer to struct oom_control
1024 * If we run out of memory, we have the choice between either
1025 * killing a random task (bad), letting the system crash (worse)
1026 * OR try to be smart about which process to kill. Note that we
1027 * don't have to be perfect here, we just have to be good.
1029 bool out_of_memory(struct oom_control *oc)
1031 unsigned long freed = 0;
1032 enum oom_constraint constraint = CONSTRAINT_NONE;
1034 if (oom_killer_disabled)
1037 if (!is_memcg_oom(oc)) {
1038 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
1040 /* Got some memory back in the last second. */
1045 * If current has a pending SIGKILL or is exiting, then automatically
1046 * select it. The goal is to allow it to allocate so that it may
1047 * quickly exit and free its memory.
1049 if (task_will_free_mem(current)) {
1050 mark_oom_victim(current);
1051 wake_oom_reaper(current);
1056 * The OOM killer does not compensate for IO-less reclaim.
1057 * pagefault_out_of_memory lost its gfp context so we have to
1058 * make sure exclude 0 mask - all other users should have at least
1059 * ___GFP_DIRECT_RECLAIM to get here.
1061 if (oc->gfp_mask && !(oc->gfp_mask & __GFP_FS))
1065 * Check if there were limitations on the allocation (only relevant for
1066 * NUMA and memcg) that may require different handling.
1068 constraint = constrained_alloc(oc);
1069 if (constraint != CONSTRAINT_MEMORY_POLICY)
1070 oc->nodemask = NULL;
1071 check_panic_on_oom(oc, constraint);
1073 if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task &&
1074 current->mm && !oom_unkillable_task(current, NULL, oc->nodemask) &&
1075 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
1076 get_task_struct(current);
1077 oc->chosen = current;
1078 oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)");
1082 select_bad_process(oc);
1083 /* Found nothing?!?! Either we hang forever, or we panic. */
1084 if (!oc->chosen && !is_sysrq_oom(oc) && !is_memcg_oom(oc)) {
1085 dump_header(oc, NULL);
1086 panic("Out of memory and no killable processes...\n");
1088 if (oc->chosen && oc->chosen != (void *)-1UL)
1089 oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" :
1090 "Memory cgroup out of memory");
1091 return !!oc->chosen;
1095 * The pagefault handler calls here because it is out of memory, so kill a
1096 * memory-hogging task. If oom_lock is held by somebody else, a parallel oom
1097 * killing is already in progress so do nothing.
1099 void pagefault_out_of_memory(void)
1101 struct oom_control oc = {
1109 if (mem_cgroup_oom_synchronize(true))
1112 if (!mutex_trylock(&oom_lock))
1115 mutex_unlock(&oom_lock);