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;
56 DEFINE_MUTEX(oom_lock);
60 * has_intersects_mems_allowed() - check task eligiblity for kill
61 * @start: task struct of which task to consider
62 * @mask: nodemask passed to page allocator for mempolicy ooms
64 * Task eligibility is determined by whether or not a candidate task, @tsk,
65 * shares the same mempolicy nodes as current if it is bound by such a policy
66 * and whether or not it has the same set of allowed cpuset nodes.
68 static bool has_intersects_mems_allowed(struct task_struct *start,
69 const nodemask_t *mask)
71 struct task_struct *tsk;
75 for_each_thread(start, tsk) {
78 * If this is a mempolicy constrained oom, tsk's
79 * cpuset is irrelevant. Only return true if its
80 * mempolicy intersects current, otherwise it may be
83 ret = mempolicy_nodemask_intersects(tsk, mask);
86 * This is not a mempolicy constrained oom, so only
87 * check the mems of tsk's cpuset.
89 ret = cpuset_mems_allowed_intersects(current, tsk);
99 static bool has_intersects_mems_allowed(struct task_struct *tsk,
100 const nodemask_t *mask)
104 #endif /* CONFIG_NUMA */
107 * The process p may have detached its own ->mm while exiting or through
108 * use_mm(), but one or more of its subthreads may still have a valid
109 * pointer. Return p, or any of its subthreads with a valid ->mm, with
112 struct task_struct *find_lock_task_mm(struct task_struct *p)
114 struct task_struct *t;
118 for_each_thread(p, t) {
132 * order == -1 means the oom kill is required by sysrq, otherwise only
133 * for display purposes.
135 static inline bool is_sysrq_oom(struct oom_control *oc)
137 return oc->order == -1;
140 static inline bool is_memcg_oom(struct oom_control *oc)
142 return oc->memcg != NULL;
145 /* return true if the task is not adequate as candidate victim task. */
146 static bool oom_unkillable_task(struct task_struct *p,
147 struct mem_cgroup *memcg, const nodemask_t *nodemask)
149 if (is_global_init(p))
151 if (p->flags & PF_KTHREAD)
154 /* When mem_cgroup_out_of_memory() and p is not member of the group */
155 if (memcg && !task_in_mem_cgroup(p, memcg))
158 /* p may not have freeable memory in nodemask */
159 if (!has_intersects_mems_allowed(p, nodemask))
166 * Print out unreclaimble slabs info when unreclaimable slabs amount is greater
167 * than all user memory (LRU pages)
169 static bool is_dump_unreclaim_slabs(void)
171 unsigned long nr_lru;
173 nr_lru = global_node_page_state(NR_ACTIVE_ANON) +
174 global_node_page_state(NR_INACTIVE_ANON) +
175 global_node_page_state(NR_ACTIVE_FILE) +
176 global_node_page_state(NR_INACTIVE_FILE) +
177 global_node_page_state(NR_ISOLATED_ANON) +
178 global_node_page_state(NR_ISOLATED_FILE) +
179 global_node_page_state(NR_UNEVICTABLE);
181 return (global_node_page_state(NR_SLAB_UNRECLAIMABLE) > nr_lru);
185 * oom_badness - heuristic function to determine which candidate task to kill
186 * @p: task struct of which task we should calculate
187 * @totalpages: total present RAM allowed for page allocation
188 * @memcg: task's memory controller, if constrained
189 * @nodemask: nodemask passed to page allocator for mempolicy ooms
191 * The heuristic for determining which task to kill is made to be as simple and
192 * predictable as possible. The goal is to return the highest value for the
193 * task consuming the most memory to avoid subsequent oom failures.
195 unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
196 const nodemask_t *nodemask, unsigned long totalpages)
201 if (oom_unkillable_task(p, memcg, nodemask))
204 p = find_lock_task_mm(p);
209 * Do not even consider tasks which are explicitly marked oom
210 * unkillable or have been already oom reaped or the are in
211 * the middle of vfork
213 adj = (long)p->signal->oom_score_adj;
214 if (adj == OOM_SCORE_ADJ_MIN ||
215 test_bit(MMF_OOM_SKIP, &p->mm->flags) ||
222 * The baseline for the badness score is the proportion of RAM that each
223 * task's rss, pagetable and swap space use.
225 points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
226 mm_pgtables_bytes(p->mm) / PAGE_SIZE;
229 /* Normalize to oom_score_adj units */
230 adj *= totalpages / 1000;
234 * Never return 0 for an eligible task regardless of the root bonus and
235 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
237 return points > 0 ? points : 1;
240 enum oom_constraint {
243 CONSTRAINT_MEMORY_POLICY,
248 * Determine the type of allocation constraint.
250 static enum oom_constraint constrained_alloc(struct oom_control *oc)
254 enum zone_type high_zoneidx = gfp_zone(oc->gfp_mask);
255 bool cpuset_limited = false;
258 if (is_memcg_oom(oc)) {
259 oc->totalpages = mem_cgroup_get_max(oc->memcg) ?: 1;
260 return CONSTRAINT_MEMCG;
263 /* Default to all available memory */
264 oc->totalpages = totalram_pages + total_swap_pages;
266 if (!IS_ENABLED(CONFIG_NUMA))
267 return CONSTRAINT_NONE;
270 return CONSTRAINT_NONE;
272 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
273 * to kill current.We have to random task kill in this case.
274 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
276 if (oc->gfp_mask & __GFP_THISNODE)
277 return CONSTRAINT_NONE;
280 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
281 * the page allocator means a mempolicy is in effect. Cpuset policy
282 * is enforced in get_page_from_freelist().
285 !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) {
286 oc->totalpages = total_swap_pages;
287 for_each_node_mask(nid, *oc->nodemask)
288 oc->totalpages += node_spanned_pages(nid);
289 return CONSTRAINT_MEMORY_POLICY;
292 /* Check this allocation failure is caused by cpuset's wall function */
293 for_each_zone_zonelist_nodemask(zone, z, oc->zonelist,
294 high_zoneidx, oc->nodemask)
295 if (!cpuset_zone_allowed(zone, oc->gfp_mask))
296 cpuset_limited = true;
298 if (cpuset_limited) {
299 oc->totalpages = total_swap_pages;
300 for_each_node_mask(nid, cpuset_current_mems_allowed)
301 oc->totalpages += node_spanned_pages(nid);
302 return CONSTRAINT_CPUSET;
304 return CONSTRAINT_NONE;
307 static int oom_evaluate_task(struct task_struct *task, void *arg)
309 struct oom_control *oc = arg;
310 unsigned long points;
312 if (oom_unkillable_task(task, NULL, oc->nodemask))
316 * This task already has access to memory reserves and is being killed.
317 * Don't allow any other task to have access to the reserves unless
318 * the task has MMF_OOM_SKIP because chances that it would release
319 * any memory is quite low.
321 if (!is_sysrq_oom(oc) && tsk_is_oom_victim(task)) {
322 if (test_bit(MMF_OOM_SKIP, &task->signal->oom_mm->flags))
328 * If task is allocating a lot of memory and has been marked to be
329 * killed first if it triggers an oom, then select it.
331 if (oom_task_origin(task)) {
336 points = oom_badness(task, NULL, oc->nodemask, oc->totalpages);
337 if (!points || points < oc->chosen_points)
340 /* Prefer thread group leaders for display purposes */
341 if (points == oc->chosen_points && thread_group_leader(oc->chosen))
345 put_task_struct(oc->chosen);
346 get_task_struct(task);
348 oc->chosen_points = points;
353 put_task_struct(oc->chosen);
354 oc->chosen = (void *)-1UL;
359 * Simple selection loop. We choose the process with the highest number of
360 * 'points'. In case scan was aborted, oc->chosen is set to -1.
362 static void select_bad_process(struct oom_control *oc)
364 if (is_memcg_oom(oc))
365 mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc);
367 struct task_struct *p;
371 if (oom_evaluate_task(p, oc))
376 oc->chosen_points = oc->chosen_points * 1000 / oc->totalpages;
380 * dump_tasks - dump current memory state of all system tasks
381 * @memcg: current's memory controller, if constrained
382 * @nodemask: nodemask passed to page allocator for mempolicy ooms
384 * Dumps the current memory state of all eligible tasks. Tasks not in the same
385 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
387 * State information includes task's pid, uid, tgid, vm size, rss,
388 * pgtables_bytes, swapents, oom_score_adj value, and name.
390 static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask)
392 struct task_struct *p;
393 struct task_struct *task;
395 pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name\n");
397 for_each_process(p) {
398 if (oom_unkillable_task(p, memcg, nodemask))
401 task = find_lock_task_mm(p);
404 * This is a kthread or all of p's threads have already
405 * detached their mm's. There's no need to report
406 * them; they can't be oom killed anyway.
411 pr_info("[%5d] %5d %5d %8lu %8lu %8ld %8lu %5hd %s\n",
412 task->pid, from_kuid(&init_user_ns, task_uid(task)),
413 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
414 mm_pgtables_bytes(task->mm),
415 get_mm_counter(task->mm, MM_SWAPENTS),
416 task->signal->oom_score_adj, task->comm);
422 static void dump_header(struct oom_control *oc, struct task_struct *p)
424 pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), nodemask=%*pbl, order=%d, oom_score_adj=%hd\n",
425 current->comm, oc->gfp_mask, &oc->gfp_mask,
426 nodemask_pr_args(oc->nodemask), oc->order,
427 current->signal->oom_score_adj);
428 if (!IS_ENABLED(CONFIG_COMPACTION) && oc->order)
429 pr_warn("COMPACTION is disabled!!!\n");
431 cpuset_print_current_mems_allowed();
433 if (is_memcg_oom(oc))
434 mem_cgroup_print_oom_info(oc->memcg, p);
436 show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask);
437 if (is_dump_unreclaim_slabs())
438 dump_unreclaimable_slab();
440 if (sysctl_oom_dump_tasks)
441 dump_tasks(oc->memcg, oc->nodemask);
445 * Number of OOM victims in flight
447 static atomic_t oom_victims = ATOMIC_INIT(0);
448 static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
450 static bool oom_killer_disabled __read_mostly;
452 #define K(x) ((x) << (PAGE_SHIFT-10))
455 * task->mm can be NULL if the task is the exited group leader. So to
456 * determine whether the task is using a particular mm, we examine all the
457 * task's threads: if one of those is using this mm then this task was also
460 bool process_shares_mm(struct task_struct *p, struct mm_struct *mm)
462 struct task_struct *t;
464 for_each_thread(p, t) {
465 struct mm_struct *t_mm = READ_ONCE(t->mm);
474 * OOM Reaper kernel thread which tries to reap the memory used by the OOM
475 * victim (if that is possible) to help the OOM killer to move on.
477 static struct task_struct *oom_reaper_th;
478 static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait);
479 static struct task_struct *oom_reaper_list;
480 static DEFINE_SPINLOCK(oom_reaper_lock);
482 void __oom_reap_task_mm(struct mm_struct *mm)
484 struct vm_area_struct *vma;
487 * Tell all users of get_user/copy_from_user etc... that the content
488 * is no longer stable. No barriers really needed because unmapping
489 * should imply barriers already and the reader would hit a page fault
490 * if it stumbled over a reaped memory.
492 set_bit(MMF_UNSTABLE, &mm->flags);
494 for (vma = mm->mmap ; vma; vma = vma->vm_next) {
495 if (!can_madv_dontneed_vma(vma))
499 * Only anonymous pages have a good chance to be dropped
500 * without additional steps which we cannot afford as we
503 * We do not even care about fs backed pages because all
504 * which are reclaimable have already been reclaimed and
505 * we do not want to block exit_mmap by keeping mm ref
506 * count elevated without a good reason.
508 if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) {
509 const unsigned long start = vma->vm_start;
510 const unsigned long end = vma->vm_end;
511 struct mmu_gather tlb;
513 tlb_gather_mmu(&tlb, mm, start, end);
514 mmu_notifier_invalidate_range_start(mm, start, end);
515 unmap_page_range(&tlb, vma, start, end, NULL);
516 mmu_notifier_invalidate_range_end(mm, start, end);
517 tlb_finish_mmu(&tlb, start, end);
522 static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
527 * We have to make sure to not race with the victim exit path
528 * and cause premature new oom victim selection:
529 * oom_reap_task_mm exit_mm
532 * atomic_dec_and_test
537 * # no TIF_MEMDIE task selects new victim
538 * unmap_page_range # frees some memory
540 mutex_lock(&oom_lock);
542 if (!down_read_trylock(&mm->mmap_sem)) {
544 trace_skip_task_reaping(tsk->pid);
549 * If the mm has invalidate_{start,end}() notifiers that could block,
550 * sleep to give the oom victim some more time.
551 * TODO: we really want to get rid of this ugly hack and make sure that
552 * notifiers cannot block for unbounded amount of time
554 if (mm_has_blockable_invalidate_notifiers(mm)) {
555 up_read(&mm->mmap_sem);
556 schedule_timeout_idle(HZ);
561 * MMF_OOM_SKIP is set by exit_mmap when the OOM reaper can't
562 * work on the mm anymore. The check for MMF_OOM_SKIP must run
563 * under mmap_sem for reading because it serializes against the
564 * down_write();up_write() cycle in exit_mmap().
566 if (test_bit(MMF_OOM_SKIP, &mm->flags)) {
567 up_read(&mm->mmap_sem);
568 trace_skip_task_reaping(tsk->pid);
572 trace_start_task_reaping(tsk->pid);
574 __oom_reap_task_mm(mm);
576 pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
577 task_pid_nr(tsk), tsk->comm,
578 K(get_mm_counter(mm, MM_ANONPAGES)),
579 K(get_mm_counter(mm, MM_FILEPAGES)),
580 K(get_mm_counter(mm, MM_SHMEMPAGES)));
581 up_read(&mm->mmap_sem);
583 trace_finish_task_reaping(tsk->pid);
585 mutex_unlock(&oom_lock);
589 #define MAX_OOM_REAP_RETRIES 10
590 static void oom_reap_task(struct task_struct *tsk)
593 struct mm_struct *mm = tsk->signal->oom_mm;
595 /* Retry the down_read_trylock(mmap_sem) a few times */
596 while (attempts++ < MAX_OOM_REAP_RETRIES && !oom_reap_task_mm(tsk, mm))
597 schedule_timeout_idle(HZ/10);
599 if (attempts <= MAX_OOM_REAP_RETRIES ||
600 test_bit(MMF_OOM_SKIP, &mm->flags))
603 pr_info("oom_reaper: unable to reap pid:%d (%s)\n",
604 task_pid_nr(tsk), tsk->comm);
605 debug_show_all_locks();
608 tsk->oom_reaper_list = NULL;
611 * Hide this mm from OOM killer because it has been either reaped or
612 * somebody can't call up_write(mmap_sem).
614 set_bit(MMF_OOM_SKIP, &mm->flags);
616 /* Drop a reference taken by wake_oom_reaper */
617 put_task_struct(tsk);
620 static int oom_reaper(void *unused)
623 struct task_struct *tsk = NULL;
625 wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL);
626 spin_lock(&oom_reaper_lock);
627 if (oom_reaper_list != NULL) {
628 tsk = oom_reaper_list;
629 oom_reaper_list = tsk->oom_reaper_list;
631 spin_unlock(&oom_reaper_lock);
640 static void wake_oom_reaper(struct task_struct *tsk)
642 /* tsk is already queued? */
643 if (tsk == oom_reaper_list || tsk->oom_reaper_list)
646 get_task_struct(tsk);
648 spin_lock(&oom_reaper_lock);
649 tsk->oom_reaper_list = oom_reaper_list;
650 oom_reaper_list = tsk;
651 spin_unlock(&oom_reaper_lock);
652 trace_wake_reaper(tsk->pid);
653 wake_up(&oom_reaper_wait);
656 static int __init oom_init(void)
658 oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper");
661 subsys_initcall(oom_init)
663 static inline void wake_oom_reaper(struct task_struct *tsk)
666 #endif /* CONFIG_MMU */
669 * mark_oom_victim - mark the given task as OOM victim
672 * Has to be called with oom_lock held and never after
673 * oom has been disabled already.
675 * tsk->mm has to be non NULL and caller has to guarantee it is stable (either
676 * under task_lock or operate on the current).
678 static void mark_oom_victim(struct task_struct *tsk)
680 struct mm_struct *mm = tsk->mm;
682 WARN_ON(oom_killer_disabled);
683 /* OOM killer might race with memcg OOM */
684 if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
687 /* oom_mm is bound to the signal struct life time. */
688 if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm)) {
689 mmgrab(tsk->signal->oom_mm);
690 set_bit(MMF_OOM_VICTIM, &mm->flags);
694 * Make sure that the task is woken up from uninterruptible sleep
695 * if it is frozen because OOM killer wouldn't be able to free
696 * any memory and livelock. freezing_slow_path will tell the freezer
697 * that TIF_MEMDIE tasks should be ignored.
700 atomic_inc(&oom_victims);
701 trace_mark_victim(tsk->pid);
705 * exit_oom_victim - note the exit of an OOM victim
707 void exit_oom_victim(void)
709 clear_thread_flag(TIF_MEMDIE);
711 if (!atomic_dec_return(&oom_victims))
712 wake_up_all(&oom_victims_wait);
716 * oom_killer_enable - enable OOM killer
718 void oom_killer_enable(void)
720 oom_killer_disabled = false;
721 pr_info("OOM killer enabled.\n");
725 * oom_killer_disable - disable OOM killer
726 * @timeout: maximum timeout to wait for oom victims in jiffies
728 * Forces all page allocations to fail rather than trigger OOM killer.
729 * Will block and wait until all OOM victims are killed or the given
732 * The function cannot be called when there are runnable user tasks because
733 * the userspace would see unexpected allocation failures as a result. Any
734 * new usage of this function should be consulted with MM people.
736 * Returns true if successful and false if the OOM killer cannot be
739 bool oom_killer_disable(signed long timeout)
744 * Make sure to not race with an ongoing OOM killer. Check that the
745 * current is not killed (possibly due to sharing the victim's memory).
747 if (mutex_lock_killable(&oom_lock))
749 oom_killer_disabled = true;
750 mutex_unlock(&oom_lock);
752 ret = wait_event_interruptible_timeout(oom_victims_wait,
753 !atomic_read(&oom_victims), timeout);
758 pr_info("OOM killer disabled.\n");
763 static inline bool __task_will_free_mem(struct task_struct *task)
765 struct signal_struct *sig = task->signal;
768 * A coredumping process may sleep for an extended period in exit_mm(),
769 * so the oom killer cannot assume that the process will promptly exit
770 * and release memory.
772 if (sig->flags & SIGNAL_GROUP_COREDUMP)
775 if (sig->flags & SIGNAL_GROUP_EXIT)
778 if (thread_group_empty(task) && (task->flags & PF_EXITING))
785 * Checks whether the given task is dying or exiting and likely to
786 * release its address space. This means that all threads and processes
787 * sharing the same mm have to be killed or exiting.
788 * Caller has to make sure that task->mm is stable (hold task_lock or
789 * it operates on the current).
791 static bool task_will_free_mem(struct task_struct *task)
793 struct mm_struct *mm = task->mm;
794 struct task_struct *p;
798 * Skip tasks without mm because it might have passed its exit_mm and
799 * exit_oom_victim. oom_reaper could have rescued that but do not rely
800 * on that for now. We can consider find_lock_task_mm in future.
805 if (!__task_will_free_mem(task))
809 * This task has already been drained by the oom reaper so there are
810 * only small chances it will free some more
812 if (test_bit(MMF_OOM_SKIP, &mm->flags))
815 if (atomic_read(&mm->mm_users) <= 1)
819 * Make sure that all tasks which share the mm with the given tasks
820 * are dying as well to make sure that a) nobody pins its mm and
821 * b) the task is also reapable by the oom reaper.
824 for_each_process(p) {
825 if (!process_shares_mm(p, mm))
827 if (same_thread_group(task, p))
829 ret = __task_will_free_mem(p);
838 static void oom_kill_process(struct oom_control *oc, const char *message)
840 struct task_struct *p = oc->chosen;
841 unsigned int points = oc->chosen_points;
842 struct task_struct *victim = p;
843 struct task_struct *child;
844 struct task_struct *t;
845 struct mm_struct *mm;
846 unsigned int victim_points = 0;
847 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
848 DEFAULT_RATELIMIT_BURST);
849 bool can_oom_reap = true;
852 * If the task is already exiting, don't alarm the sysadmin or kill
853 * its children or threads, just give it access to memory reserves
854 * so it can die quickly
857 if (task_will_free_mem(p)) {
866 if (__ratelimit(&oom_rs))
869 pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n",
870 message, task_pid_nr(p), p->comm, points);
873 * If any of p's children has a different mm and is eligible for kill,
874 * the one with the highest oom_badness() score is sacrificed for its
875 * parent. This attempts to lose the minimal amount of work done while
876 * still freeing memory.
878 read_lock(&tasklist_lock);
879 for_each_thread(p, t) {
880 list_for_each_entry(child, &t->children, sibling) {
881 unsigned int child_points;
883 if (process_shares_mm(child, p->mm))
886 * oom_badness() returns 0 if the thread is unkillable
888 child_points = oom_badness(child,
889 oc->memcg, oc->nodemask, oc->totalpages);
890 if (child_points > victim_points) {
891 put_task_struct(victim);
893 victim_points = child_points;
894 get_task_struct(victim);
898 read_unlock(&tasklist_lock);
900 p = find_lock_task_mm(victim);
902 put_task_struct(victim);
904 } else if (victim != p) {
906 put_task_struct(victim);
910 /* Get a reference to safely compare mm after task_unlock(victim) */
914 /* Raise event before sending signal: task reaper must see this */
915 count_vm_event(OOM_KILL);
916 memcg_memory_event_mm(mm, MEMCG_OOM_KILL);
919 * We should send SIGKILL before granting access to memory reserves
920 * in order to prevent the OOM victim from depleting the memory
921 * reserves from the user space under its control.
923 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
924 mark_oom_victim(victim);
925 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
926 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
927 K(get_mm_counter(victim->mm, MM_ANONPAGES)),
928 K(get_mm_counter(victim->mm, MM_FILEPAGES)),
929 K(get_mm_counter(victim->mm, MM_SHMEMPAGES)));
933 * Kill all user processes sharing victim->mm in other thread groups, if
934 * any. They don't get access to memory reserves, though, to avoid
935 * depletion of all memory. This prevents mm->mmap_sem livelock when an
936 * oom killed thread cannot exit because it requires the semaphore and
937 * its contended by another thread trying to allocate memory itself.
938 * That thread will now get access to memory reserves since it has a
939 * pending fatal signal.
942 for_each_process(p) {
943 if (!process_shares_mm(p, mm))
945 if (same_thread_group(p, victim))
947 if (is_global_init(p)) {
948 can_oom_reap = false;
949 set_bit(MMF_OOM_SKIP, &mm->flags);
950 pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n",
951 task_pid_nr(victim), victim->comm,
952 task_pid_nr(p), p->comm);
956 * No use_mm() user needs to read from the userspace so we are
959 if (unlikely(p->flags & PF_KTHREAD))
961 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
966 wake_oom_reaper(victim);
969 put_task_struct(victim);
974 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
976 static void check_panic_on_oom(struct oom_control *oc,
977 enum oom_constraint constraint)
979 if (likely(!sysctl_panic_on_oom))
981 if (sysctl_panic_on_oom != 2) {
983 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
984 * does not panic for cpuset, mempolicy, or memcg allocation
987 if (constraint != CONSTRAINT_NONE)
990 /* Do not panic for oom kills triggered by sysrq */
991 if (is_sysrq_oom(oc))
993 dump_header(oc, NULL);
994 panic("Out of memory: %s panic_on_oom is enabled\n",
995 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
998 static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
1000 int register_oom_notifier(struct notifier_block *nb)
1002 return blocking_notifier_chain_register(&oom_notify_list, nb);
1004 EXPORT_SYMBOL_GPL(register_oom_notifier);
1006 int unregister_oom_notifier(struct notifier_block *nb)
1008 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
1010 EXPORT_SYMBOL_GPL(unregister_oom_notifier);
1013 * out_of_memory - kill the "best" process when we run out of memory
1014 * @oc: pointer to struct oom_control
1016 * If we run out of memory, we have the choice between either
1017 * killing a random task (bad), letting the system crash (worse)
1018 * OR try to be smart about which process to kill. Note that we
1019 * don't have to be perfect here, we just have to be good.
1021 bool out_of_memory(struct oom_control *oc)
1023 unsigned long freed = 0;
1024 enum oom_constraint constraint = CONSTRAINT_NONE;
1026 if (oom_killer_disabled)
1029 if (!is_memcg_oom(oc)) {
1030 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
1032 /* Got some memory back in the last second. */
1037 * If current has a pending SIGKILL or is exiting, then automatically
1038 * select it. The goal is to allow it to allocate so that it may
1039 * quickly exit and free its memory.
1041 if (task_will_free_mem(current)) {
1042 mark_oom_victim(current);
1043 wake_oom_reaper(current);
1048 * The OOM killer does not compensate for IO-less reclaim.
1049 * pagefault_out_of_memory lost its gfp context so we have to
1050 * make sure exclude 0 mask - all other users should have at least
1051 * ___GFP_DIRECT_RECLAIM to get here.
1053 if (oc->gfp_mask && !(oc->gfp_mask & __GFP_FS))
1057 * Check if there were limitations on the allocation (only relevant for
1058 * NUMA and memcg) that may require different handling.
1060 constraint = constrained_alloc(oc);
1061 if (constraint != CONSTRAINT_MEMORY_POLICY)
1062 oc->nodemask = NULL;
1063 check_panic_on_oom(oc, constraint);
1065 if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task &&
1066 current->mm && !oom_unkillable_task(current, NULL, oc->nodemask) &&
1067 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
1068 get_task_struct(current);
1069 oc->chosen = current;
1070 oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)");
1074 select_bad_process(oc);
1075 /* Found nothing?!?! Either we hang forever, or we panic. */
1076 if (!oc->chosen && !is_sysrq_oom(oc) && !is_memcg_oom(oc)) {
1077 dump_header(oc, NULL);
1078 panic("Out of memory and no killable processes...\n");
1080 if (oc->chosen && oc->chosen != (void *)-1UL) {
1081 oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" :
1082 "Memory cgroup out of memory");
1084 * Give the killed process a good chance to exit before trying
1085 * to allocate memory again.
1087 schedule_timeout_killable(1);
1089 return !!oc->chosen;
1093 * The pagefault handler calls here because it is out of memory, so kill a
1094 * memory-hogging task. If oom_lock is held by somebody else, a parallel oom
1095 * killing is already in progress so do nothing.
1097 void pagefault_out_of_memory(void)
1099 struct oom_control oc = {
1107 if (mem_cgroup_oom_synchronize(true))
1110 if (!mutex_trylock(&oom_lock))
1113 mutex_unlock(&oom_lock);