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_REAPED, &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_REAPED because chances that it would release
300 * any memory is quite low.
302 if (!is_sysrq_oom(oc) && atomic_read(&task->signal->oom_victims)) {
303 struct task_struct *p = find_lock_task_mm(task);
307 reaped = test_bit(MMF_OOM_REAPED, &p->mm->flags);
316 * If task is allocating a lot of memory and has been marked to be
317 * killed first if it triggers an oom, then select it.
319 if (oom_task_origin(task)) {
324 points = oom_badness(task, NULL, oc->nodemask, oc->totalpages);
325 if (!points || points < oc->chosen_points)
328 /* Prefer thread group leaders for display purposes */
329 if (points == oc->chosen_points && thread_group_leader(oc->chosen))
333 put_task_struct(oc->chosen);
334 get_task_struct(task);
336 oc->chosen_points = points;
341 put_task_struct(oc->chosen);
342 oc->chosen = (void *)-1UL;
347 * Simple selection loop. We choose the process with the highest number of
348 * 'points'. In case scan was aborted, oc->chosen is set to -1.
350 static void select_bad_process(struct oom_control *oc)
352 if (is_memcg_oom(oc))
353 mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc);
355 struct task_struct *p;
359 if (oom_evaluate_task(p, oc))
364 oc->chosen_points = oc->chosen_points * 1000 / oc->totalpages;
368 * dump_tasks - dump current memory state of all system tasks
369 * @memcg: current's memory controller, if constrained
370 * @nodemask: nodemask passed to page allocator for mempolicy ooms
372 * Dumps the current memory state of all eligible tasks. Tasks not in the same
373 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
375 * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
376 * swapents, oom_score_adj value, and name.
378 static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask)
380 struct task_struct *p;
381 struct task_struct *task;
383 pr_info("[ pid ] uid tgid total_vm rss nr_ptes nr_pmds swapents oom_score_adj name\n");
385 for_each_process(p) {
386 if (oom_unkillable_task(p, memcg, nodemask))
389 task = find_lock_task_mm(p);
392 * This is a kthread or all of p's threads have already
393 * detached their mm's. There's no need to report
394 * them; they can't be oom killed anyway.
399 pr_info("[%5d] %5d %5d %8lu %8lu %7ld %7ld %8lu %5hd %s\n",
400 task->pid, from_kuid(&init_user_ns, task_uid(task)),
401 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
402 atomic_long_read(&task->mm->nr_ptes),
403 mm_nr_pmds(task->mm),
404 get_mm_counter(task->mm, MM_SWAPENTS),
405 task->signal->oom_score_adj, task->comm);
411 static void dump_header(struct oom_control *oc, struct task_struct *p)
413 pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), order=%d, oom_score_adj=%hd\n",
414 current->comm, oc->gfp_mask, &oc->gfp_mask, oc->order,
415 current->signal->oom_score_adj);
417 cpuset_print_current_mems_allowed();
420 mem_cgroup_print_oom_info(oc->memcg, p);
422 show_mem(SHOW_MEM_FILTER_NODES);
423 if (sysctl_oom_dump_tasks)
424 dump_tasks(oc->memcg, oc->nodemask);
428 * Number of OOM victims in flight
430 static atomic_t oom_victims = ATOMIC_INIT(0);
431 static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
433 static bool oom_killer_disabled __read_mostly;
435 #define K(x) ((x) << (PAGE_SHIFT-10))
438 * task->mm can be NULL if the task is the exited group leader. So to
439 * determine whether the task is using a particular mm, we examine all the
440 * task's threads: if one of those is using this mm then this task was also
443 bool process_shares_mm(struct task_struct *p, struct mm_struct *mm)
445 struct task_struct *t;
447 for_each_thread(p, t) {
448 struct mm_struct *t_mm = READ_ONCE(t->mm);
458 * OOM Reaper kernel thread which tries to reap the memory used by the OOM
459 * victim (if that is possible) to help the OOM killer to move on.
461 static struct task_struct *oom_reaper_th;
462 static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait);
463 static struct task_struct *oom_reaper_list;
464 static DEFINE_SPINLOCK(oom_reaper_lock);
466 static bool __oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
468 struct mmu_gather tlb;
469 struct vm_area_struct *vma;
470 struct zap_details details = {.check_swap_entries = true,
471 .ignore_dirty = true};
475 * We have to make sure to not race with the victim exit path
476 * and cause premature new oom victim selection:
477 * __oom_reap_task_mm exit_mm
480 * atomic_dec_and_test
485 * # no TIF_MEMDIE task selects new victim
486 * unmap_page_range # frees some memory
488 mutex_lock(&oom_lock);
490 if (!down_read_trylock(&mm->mmap_sem)) {
496 * increase mm_users only after we know we will reap something so
497 * that the mmput_async is called only when we have reaped something
498 * and delayed __mmput doesn't matter that much
500 if (!mmget_not_zero(mm)) {
501 up_read(&mm->mmap_sem);
505 tlb_gather_mmu(&tlb, mm, 0, -1);
506 for (vma = mm->mmap ; vma; vma = vma->vm_next) {
507 if (is_vm_hugetlb_page(vma))
511 * mlocked VMAs require explicit munlocking before unmap.
512 * Let's keep it simple here and skip such VMAs.
514 if (vma->vm_flags & VM_LOCKED)
518 * Only anonymous pages have a good chance to be dropped
519 * without additional steps which we cannot afford as we
522 * We do not even care about fs backed pages because all
523 * which are reclaimable have already been reclaimed and
524 * we do not want to block exit_mmap by keeping mm ref
525 * count elevated without a good reason.
527 if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED))
528 unmap_page_range(&tlb, vma, vma->vm_start, vma->vm_end,
531 tlb_finish_mmu(&tlb, 0, -1);
532 pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
533 task_pid_nr(tsk), tsk->comm,
534 K(get_mm_counter(mm, MM_ANONPAGES)),
535 K(get_mm_counter(mm, MM_FILEPAGES)),
536 K(get_mm_counter(mm, MM_SHMEMPAGES)));
537 up_read(&mm->mmap_sem);
540 * This task can be safely ignored because we cannot do much more
541 * to release its memory.
543 set_bit(MMF_OOM_REAPED, &mm->flags);
545 * Drop our reference but make sure the mmput slow path is called from a
546 * different context because we shouldn't risk we get stuck there and
547 * put the oom_reaper out of the way.
551 mutex_unlock(&oom_lock);
555 #define MAX_OOM_REAP_RETRIES 10
556 static void oom_reap_task(struct task_struct *tsk)
559 struct mm_struct *mm = NULL;
560 struct task_struct *p = find_lock_task_mm(tsk);
563 * Make sure we find the associated mm_struct even when the particular
564 * thread has already terminated and cleared its mm.
565 * We might have race with exit path so consider our work done if there
571 atomic_inc(&mm->mm_count);
574 /* Retry the down_read_trylock(mmap_sem) a few times */
575 while (attempts++ < MAX_OOM_REAP_RETRIES && !__oom_reap_task_mm(tsk, mm))
576 schedule_timeout_idle(HZ/10);
578 if (attempts <= MAX_OOM_REAP_RETRIES)
581 pr_info("oom_reaper: unable to reap pid:%d (%s)\n",
582 task_pid_nr(tsk), tsk->comm);
585 * If we've already tried to reap this task in the past and
586 * failed it probably doesn't make much sense to try yet again
587 * so hide the mm from the oom killer so that it can move on
588 * to another task with a different mm struct.
590 if (test_and_set_bit(MMF_OOM_NOT_REAPABLE, &mm->flags)) {
591 pr_info("oom_reaper: giving up pid:%d (%s)\n",
592 task_pid_nr(tsk), tsk->comm);
593 set_bit(MMF_OOM_REAPED, &mm->flags);
595 debug_show_all_locks();
599 * Clear TIF_MEMDIE because the task shouldn't be sitting on a
600 * reasonably reclaimable memory anymore or it is not a good candidate
601 * for the oom victim right now because it cannot release its memory
602 * itself nor by the oom reaper.
604 tsk->oom_reaper_list = NULL;
605 exit_oom_victim(tsk);
607 /* Drop a reference taken by wake_oom_reaper */
608 put_task_struct(tsk);
609 /* Drop a reference taken above. */
614 static int oom_reaper(void *unused)
619 struct task_struct *tsk = NULL;
621 wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL);
622 spin_lock(&oom_reaper_lock);
623 if (oom_reaper_list != NULL) {
624 tsk = oom_reaper_list;
625 oom_reaper_list = tsk->oom_reaper_list;
627 spin_unlock(&oom_reaper_lock);
636 static void wake_oom_reaper(struct task_struct *tsk)
641 /* tsk is already queued? */
642 if (tsk == oom_reaper_list || tsk->oom_reaper_list)
645 get_task_struct(tsk);
647 spin_lock(&oom_reaper_lock);
648 tsk->oom_reaper_list = oom_reaper_list;
649 oom_reaper_list = tsk;
650 spin_unlock(&oom_reaper_lock);
651 wake_up(&oom_reaper_wait);
654 static int __init oom_init(void)
656 oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper");
657 if (IS_ERR(oom_reaper_th)) {
658 pr_err("Unable to start OOM reaper %ld. Continuing regardless\n",
659 PTR_ERR(oom_reaper_th));
660 oom_reaper_th = NULL;
664 subsys_initcall(oom_init)
666 static inline void wake_oom_reaper(struct task_struct *tsk)
669 #endif /* CONFIG_MMU */
672 * mark_oom_victim - mark the given task as OOM victim
675 * Has to be called with oom_lock held and never after
676 * oom has been disabled already.
678 static void mark_oom_victim(struct task_struct *tsk)
680 WARN_ON(oom_killer_disabled);
681 /* OOM killer might race with memcg OOM */
682 if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
684 atomic_inc(&tsk->signal->oom_victims);
686 * Make sure that the task is woken up from uninterruptible sleep
687 * if it is frozen because OOM killer wouldn't be able to free
688 * any memory and livelock. freezing_slow_path will tell the freezer
689 * that TIF_MEMDIE tasks should be ignored.
692 atomic_inc(&oom_victims);
696 * exit_oom_victim - note the exit of an OOM victim
698 void exit_oom_victim(struct task_struct *tsk)
700 if (!test_and_clear_tsk_thread_flag(tsk, TIF_MEMDIE))
702 atomic_dec(&tsk->signal->oom_victims);
704 if (!atomic_dec_return(&oom_victims))
705 wake_up_all(&oom_victims_wait);
709 * oom_killer_disable - disable OOM killer
711 * Forces all page allocations to fail rather than trigger OOM killer.
712 * Will block and wait until all OOM victims are killed.
714 * The function cannot be called when there are runnable user tasks because
715 * the userspace would see unexpected allocation failures as a result. Any
716 * new usage of this function should be consulted with MM people.
718 * Returns true if successful and false if the OOM killer cannot be
721 bool oom_killer_disable(void)
724 * Make sure to not race with an ongoing OOM killer. Check that the
725 * current is not killed (possibly due to sharing the victim's memory).
727 if (mutex_lock_killable(&oom_lock))
729 oom_killer_disabled = true;
730 mutex_unlock(&oom_lock);
732 wait_event(oom_victims_wait, !atomic_read(&oom_victims));
738 * oom_killer_enable - enable OOM killer
740 void oom_killer_enable(void)
742 oom_killer_disabled = false;
745 static inline bool __task_will_free_mem(struct task_struct *task)
747 struct signal_struct *sig = task->signal;
750 * A coredumping process may sleep for an extended period in exit_mm(),
751 * so the oom killer cannot assume that the process will promptly exit
752 * and release memory.
754 if (sig->flags & SIGNAL_GROUP_COREDUMP)
757 if (sig->flags & SIGNAL_GROUP_EXIT)
760 if (thread_group_empty(task) && (task->flags & PF_EXITING))
767 * Checks whether the given task is dying or exiting and likely to
768 * release its address space. This means that all threads and processes
769 * sharing the same mm have to be killed or exiting.
770 * Caller has to make sure that task->mm is stable (hold task_lock or
771 * it operates on the current).
773 static bool task_will_free_mem(struct task_struct *task)
775 struct mm_struct *mm = task->mm;
776 struct task_struct *p;
780 * Skip tasks without mm because it might have passed its exit_mm and
781 * exit_oom_victim. oom_reaper could have rescued that but do not rely
782 * on that for now. We can consider find_lock_task_mm in future.
787 if (!__task_will_free_mem(task))
791 * This task has already been drained by the oom reaper so there are
792 * only small chances it will free some more
794 if (test_bit(MMF_OOM_REAPED, &mm->flags))
797 if (atomic_read(&mm->mm_users) <= 1)
801 * Make sure that all tasks which share the mm with the given tasks
802 * are dying as well to make sure that a) nobody pins its mm and
803 * b) the task is also reapable by the oom reaper.
806 for_each_process(p) {
807 if (!process_shares_mm(p, mm))
809 if (same_thread_group(task, p))
811 ret = __task_will_free_mem(p);
820 static void oom_kill_process(struct oom_control *oc, const char *message)
822 struct task_struct *p = oc->chosen;
823 unsigned int points = oc->chosen_points;
824 struct task_struct *victim = p;
825 struct task_struct *child;
826 struct task_struct *t;
827 struct mm_struct *mm;
828 unsigned int victim_points = 0;
829 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
830 DEFAULT_RATELIMIT_BURST);
831 bool can_oom_reap = true;
834 * If the task is already exiting, don't alarm the sysadmin or kill
835 * its children or threads, just set TIF_MEMDIE so it can die quickly
838 if (task_will_free_mem(p)) {
847 if (__ratelimit(&oom_rs))
850 pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n",
851 message, task_pid_nr(p), p->comm, points);
854 * If any of p's children has a different mm and is eligible for kill,
855 * the one with the highest oom_badness() score is sacrificed for its
856 * parent. This attempts to lose the minimal amount of work done while
857 * still freeing memory.
859 read_lock(&tasklist_lock);
860 for_each_thread(p, t) {
861 list_for_each_entry(child, &t->children, sibling) {
862 unsigned int child_points;
864 if (process_shares_mm(child, p->mm))
867 * oom_badness() returns 0 if the thread is unkillable
869 child_points = oom_badness(child,
870 oc->memcg, oc->nodemask, oc->totalpages);
871 if (child_points > victim_points) {
872 put_task_struct(victim);
874 victim_points = child_points;
875 get_task_struct(victim);
879 read_unlock(&tasklist_lock);
881 p = find_lock_task_mm(victim);
883 put_task_struct(victim);
885 } else if (victim != p) {
887 put_task_struct(victim);
891 /* Get a reference to safely compare mm after task_unlock(victim) */
893 atomic_inc(&mm->mm_count);
895 * We should send SIGKILL before setting TIF_MEMDIE in order to prevent
896 * the OOM victim from depleting the memory reserves from the user
897 * space under its control.
899 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
900 mark_oom_victim(victim);
901 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
902 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
903 K(get_mm_counter(victim->mm, MM_ANONPAGES)),
904 K(get_mm_counter(victim->mm, MM_FILEPAGES)),
905 K(get_mm_counter(victim->mm, MM_SHMEMPAGES)));
909 * Kill all user processes sharing victim->mm in other thread groups, if
910 * any. They don't get access to memory reserves, though, to avoid
911 * depletion of all memory. This prevents mm->mmap_sem livelock when an
912 * oom killed thread cannot exit because it requires the semaphore and
913 * its contended by another thread trying to allocate memory itself.
914 * That thread will now get access to memory reserves since it has a
915 * pending fatal signal.
918 for_each_process(p) {
919 if (!process_shares_mm(p, mm))
921 if (same_thread_group(p, victim))
923 if (unlikely(p->flags & PF_KTHREAD) || is_global_init(p)) {
925 * We cannot use oom_reaper for the mm shared by this
926 * process because it wouldn't get killed and so the
927 * memory might be still used. Hide the mm from the oom
928 * killer to guarantee OOM forward progress.
930 can_oom_reap = false;
931 set_bit(MMF_OOM_REAPED, &mm->flags);
932 pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n",
933 task_pid_nr(victim), victim->comm,
934 task_pid_nr(p), p->comm);
937 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
942 wake_oom_reaper(victim);
945 put_task_struct(victim);
950 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
952 static void check_panic_on_oom(struct oom_control *oc,
953 enum oom_constraint constraint)
955 if (likely(!sysctl_panic_on_oom))
957 if (sysctl_panic_on_oom != 2) {
959 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
960 * does not panic for cpuset, mempolicy, or memcg allocation
963 if (constraint != CONSTRAINT_NONE)
966 /* Do not panic for oom kills triggered by sysrq */
967 if (is_sysrq_oom(oc))
969 dump_header(oc, NULL);
970 panic("Out of memory: %s panic_on_oom is enabled\n",
971 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
974 static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
976 int register_oom_notifier(struct notifier_block *nb)
978 return blocking_notifier_chain_register(&oom_notify_list, nb);
980 EXPORT_SYMBOL_GPL(register_oom_notifier);
982 int unregister_oom_notifier(struct notifier_block *nb)
984 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
986 EXPORT_SYMBOL_GPL(unregister_oom_notifier);
989 * out_of_memory - kill the "best" process when we run out of memory
990 * @oc: pointer to struct oom_control
992 * If we run out of memory, we have the choice between either
993 * killing a random task (bad), letting the system crash (worse)
994 * OR try to be smart about which process to kill. Note that we
995 * don't have to be perfect here, we just have to be good.
997 bool out_of_memory(struct oom_control *oc)
999 unsigned long freed = 0;
1000 enum oom_constraint constraint = CONSTRAINT_NONE;
1002 if (oom_killer_disabled)
1005 if (!is_memcg_oom(oc)) {
1006 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
1008 /* Got some memory back in the last second. */
1013 * If current has a pending SIGKILL or is exiting, then automatically
1014 * select it. The goal is to allow it to allocate so that it may
1015 * quickly exit and free its memory.
1017 if (task_will_free_mem(current)) {
1018 mark_oom_victim(current);
1019 wake_oom_reaper(current);
1024 * The OOM killer does not compensate for IO-less reclaim.
1025 * pagefault_out_of_memory lost its gfp context so we have to
1026 * make sure exclude 0 mask - all other users should have at least
1027 * ___GFP_DIRECT_RECLAIM to get here.
1029 if (oc->gfp_mask && !(oc->gfp_mask & (__GFP_FS|__GFP_NOFAIL)))
1033 * Check if there were limitations on the allocation (only relevant for
1034 * NUMA and memcg) that may require different handling.
1036 constraint = constrained_alloc(oc);
1037 if (constraint != CONSTRAINT_MEMORY_POLICY)
1038 oc->nodemask = NULL;
1039 check_panic_on_oom(oc, constraint);
1041 if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task &&
1042 current->mm && !oom_unkillable_task(current, NULL, oc->nodemask) &&
1043 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
1044 get_task_struct(current);
1045 oc->chosen = current;
1046 oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)");
1050 select_bad_process(oc);
1051 /* Found nothing?!?! Either we hang forever, or we panic. */
1052 if (!oc->chosen && !is_sysrq_oom(oc) && !is_memcg_oom(oc)) {
1053 dump_header(oc, NULL);
1054 panic("Out of memory and no killable processes...\n");
1056 if (oc->chosen && oc->chosen != (void *)-1UL) {
1057 oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" :
1058 "Memory cgroup out of memory");
1060 * Give the killed process a good chance to exit before trying
1061 * to allocate memory again.
1063 schedule_timeout_killable(1);
1065 return !!oc->chosen;
1069 * The pagefault handler calls here because it is out of memory, so kill a
1070 * memory-hogging task. If oom_lock is held by somebody else, a parallel oom
1071 * killing is already in progress so do nothing.
1073 void pagefault_out_of_memory(void)
1075 struct oom_control oc = {
1083 if (mem_cgroup_oom_synchronize(true))
1086 if (!mutex_trylock(&oom_lock))
1089 if (!out_of_memory(&oc)) {
1091 * There shouldn't be any user tasks runnable while the
1092 * OOM killer is disabled, so the current task has to
1093 * be a racing OOM victim for which oom_killer_disable()
1096 WARN_ON(test_thread_flag(TIF_MEMDIE));
1099 mutex_unlock(&oom_lock);