1 // SPDX-License-Identifier: GPL-2.0-only
5 * Copyright (C) 1998,2000 Rik van Riel
6 * Thanks go out to Claus Fischer for some serious inspiration and
7 * for goading me into coding this file...
8 * Copyright (C) 2010 Google, Inc.
9 * Rewritten by David Rientjes
11 * The routines in this file are used to kill a process when
12 * we're seriously out of memory. This gets called from __alloc_pages()
13 * in mm/page_alloc.c when we really run out of memory.
15 * Since we won't call these routines often (on a well-configured
16 * machine) this file will double as a 'coding guide' and a signpost
17 * for newbie kernel hackers. It features several pointers to major
18 * kernel subsystems and hints as to where to find out what things do.
21 #include <linux/oom.h>
23 #include <linux/err.h>
24 #include <linux/gfp.h>
25 #include <linux/sched.h>
26 #include <linux/sched/mm.h>
27 #include <linux/sched/coredump.h>
28 #include <linux/sched/task.h>
29 #include <linux/sched/debug.h>
30 #include <linux/swap.h>
31 #include <linux/timex.h>
32 #include <linux/jiffies.h>
33 #include <linux/cpuset.h>
34 #include <linux/export.h>
35 #include <linux/notifier.h>
36 #include <linux/memcontrol.h>
37 #include <linux/mempolicy.h>
38 #include <linux/security.h>
39 #include <linux/ptrace.h>
40 #include <linux/freezer.h>
41 #include <linux/ftrace.h>
42 #include <linux/ratelimit.h>
43 #include <linux/kthread.h>
44 #include <linux/init.h>
45 #include <linux/mmu_notifier.h>
51 #define CREATE_TRACE_POINTS
52 #include <trace/events/oom.h>
54 int sysctl_panic_on_oom;
55 int sysctl_oom_kill_allocating_task;
56 int sysctl_oom_dump_tasks = 1;
59 * Serializes oom killer invocations (out_of_memory()) from all contexts to
60 * prevent from over eager oom killing (e.g. when the oom killer is invoked
61 * from different domains).
63 * oom_killer_disable() relies on this lock to stabilize oom_killer_disabled
66 DEFINE_MUTEX(oom_lock);
68 static inline bool is_memcg_oom(struct oom_control *oc)
70 return oc->memcg != NULL;
75 * oom_cpuset_eligible() - check task eligiblity for kill
76 * @start: task struct of which task to consider
77 * @oc: pointer to struct oom_control
79 * Task eligibility is determined by whether or not a candidate task, @tsk,
80 * shares the same mempolicy nodes as current if it is bound by such a policy
81 * and whether or not it has the same set of allowed cpuset nodes.
83 * This function is assuming oom-killer context and 'current' has triggered
86 static bool oom_cpuset_eligible(struct task_struct *start,
87 struct oom_control *oc)
89 struct task_struct *tsk;
91 const nodemask_t *mask = oc->nodemask;
97 for_each_thread(start, tsk) {
100 * If this is a mempolicy constrained oom, tsk's
101 * cpuset is irrelevant. Only return true if its
102 * mempolicy intersects current, otherwise it may be
105 ret = mempolicy_nodemask_intersects(tsk, mask);
108 * This is not a mempolicy constrained oom, so only
109 * check the mems of tsk's cpuset.
111 ret = cpuset_mems_allowed_intersects(current, tsk);
121 static bool oom_cpuset_eligible(struct task_struct *tsk, struct oom_control *oc)
125 #endif /* CONFIG_NUMA */
128 * The process p may have detached its own ->mm while exiting or through
129 * kthread_use_mm(), but one or more of its subthreads may still have a valid
130 * pointer. Return p, or any of its subthreads with a valid ->mm, with
133 struct task_struct *find_lock_task_mm(struct task_struct *p)
135 struct task_struct *t;
139 for_each_thread(p, t) {
153 * order == -1 means the oom kill is required by sysrq, otherwise only
154 * for display purposes.
156 static inline bool is_sysrq_oom(struct oom_control *oc)
158 return oc->order == -1;
161 /* return true if the task is not adequate as candidate victim task. */
162 static bool oom_unkillable_task(struct task_struct *p)
164 if (is_global_init(p))
166 if (p->flags & PF_KTHREAD)
172 * Print out unreclaimble slabs info when unreclaimable slabs amount is greater
173 * than all user memory (LRU pages)
175 static bool is_dump_unreclaim_slabs(void)
177 unsigned long nr_lru;
179 nr_lru = global_node_page_state(NR_ACTIVE_ANON) +
180 global_node_page_state(NR_INACTIVE_ANON) +
181 global_node_page_state(NR_ACTIVE_FILE) +
182 global_node_page_state(NR_INACTIVE_FILE) +
183 global_node_page_state(NR_ISOLATED_ANON) +
184 global_node_page_state(NR_ISOLATED_FILE) +
185 global_node_page_state(NR_UNEVICTABLE);
187 return (global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B) > nr_lru);
191 * oom_badness - heuristic function to determine which candidate task to kill
192 * @p: task struct of which task we should calculate
193 * @totalpages: total present RAM allowed for page allocation
195 * The heuristic for determining which task to kill is made to be as simple and
196 * predictable as possible. The goal is to return the highest value for the
197 * task consuming the most memory to avoid subsequent oom failures.
199 long oom_badness(struct task_struct *p, unsigned long totalpages)
204 if (oom_unkillable_task(p))
207 p = find_lock_task_mm(p);
212 * Do not even consider tasks which are explicitly marked oom
213 * unkillable or have been already oom reaped or the are in
214 * the middle of vfork
216 adj = (long)p->signal->oom_score_adj;
217 if (adj == OOM_SCORE_ADJ_MIN ||
218 test_bit(MMF_OOM_SKIP, &p->mm->flags) ||
225 * The baseline for the badness score is the proportion of RAM that each
226 * task's rss, pagetable and swap space use.
228 points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
229 mm_pgtables_bytes(p->mm) / PAGE_SIZE;
232 /* Normalize to oom_score_adj units */
233 adj *= totalpages / 1000;
239 static const char * const oom_constraint_text[] = {
240 [CONSTRAINT_NONE] = "CONSTRAINT_NONE",
241 [CONSTRAINT_CPUSET] = "CONSTRAINT_CPUSET",
242 [CONSTRAINT_MEMORY_POLICY] = "CONSTRAINT_MEMORY_POLICY",
243 [CONSTRAINT_MEMCG] = "CONSTRAINT_MEMCG",
247 * Determine the type of allocation constraint.
249 static enum oom_constraint constrained_alloc(struct oom_control *oc)
253 enum zone_type highest_zoneidx = gfp_zone(oc->gfp_mask);
254 bool cpuset_limited = false;
257 if (is_memcg_oom(oc)) {
258 oc->totalpages = mem_cgroup_get_max(oc->memcg) ?: 1;
259 return CONSTRAINT_MEMCG;
262 /* Default to all available memory */
263 oc->totalpages = totalram_pages() + total_swap_pages;
265 if (!IS_ENABLED(CONFIG_NUMA))
266 return CONSTRAINT_NONE;
269 return CONSTRAINT_NONE;
271 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
272 * to kill current.We have to random task kill in this case.
273 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
275 if (oc->gfp_mask & __GFP_THISNODE)
276 return CONSTRAINT_NONE;
279 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
280 * the page allocator means a mempolicy is in effect. Cpuset policy
281 * is enforced in get_page_from_freelist().
284 !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) {
285 oc->totalpages = total_swap_pages;
286 for_each_node_mask(nid, *oc->nodemask)
287 oc->totalpages += node_present_pages(nid);
288 return CONSTRAINT_MEMORY_POLICY;
291 /* Check this allocation failure is caused by cpuset's wall function */
292 for_each_zone_zonelist_nodemask(zone, z, oc->zonelist,
293 highest_zoneidx, oc->nodemask)
294 if (!cpuset_zone_allowed(zone, oc->gfp_mask))
295 cpuset_limited = true;
297 if (cpuset_limited) {
298 oc->totalpages = total_swap_pages;
299 for_each_node_mask(nid, cpuset_current_mems_allowed)
300 oc->totalpages += node_present_pages(nid);
301 return CONSTRAINT_CPUSET;
303 return CONSTRAINT_NONE;
306 static int oom_evaluate_task(struct task_struct *task, void *arg)
308 struct oom_control *oc = arg;
311 if (oom_unkillable_task(task))
314 /* p may not have freeable memory in nodemask */
315 if (!is_memcg_oom(oc) && !oom_cpuset_eligible(task, oc))
319 * This task already has access to memory reserves and is being killed.
320 * Don't allow any other task to have access to the reserves unless
321 * the task has MMF_OOM_SKIP because chances that it would release
322 * any memory is quite low.
324 if (!is_sysrq_oom(oc) && tsk_is_oom_victim(task)) {
325 if (test_bit(MMF_OOM_SKIP, &task->signal->oom_mm->flags))
331 * If task is allocating a lot of memory and has been marked to be
332 * killed first if it triggers an oom, then select it.
334 if (oom_task_origin(task)) {
339 points = oom_badness(task, oc->totalpages);
340 if (points == LONG_MIN || points < oc->chosen_points)
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 oc->chosen_points = LONG_MIN;
366 if (is_memcg_oom(oc))
367 mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc);
369 struct task_struct *p;
373 if (oom_evaluate_task(p, oc))
379 static int dump_task(struct task_struct *p, void *arg)
381 struct oom_control *oc = arg;
382 struct task_struct *task;
384 if (oom_unkillable_task(p))
387 /* p may not have freeable memory in nodemask */
388 if (!is_memcg_oom(oc) && !oom_cpuset_eligible(p, oc))
391 task = find_lock_task_mm(p);
394 * This is a kthread or all of p's threads have already
395 * detached their mm's. There's no need to report
396 * them; they can't be oom killed anyway.
401 pr_info("[%7d] %5d %5d %8lu %8lu %8ld %8lu %5hd %s\n",
402 task->pid, from_kuid(&init_user_ns, task_uid(task)),
403 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
404 mm_pgtables_bytes(task->mm),
405 get_mm_counter(task->mm, MM_SWAPENTS),
406 task->signal->oom_score_adj, task->comm);
413 * dump_tasks - dump current memory state of all system tasks
414 * @oc: pointer to struct oom_control
416 * Dumps the current memory state of all eligible tasks. Tasks not in the same
417 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
419 * State information includes task's pid, uid, tgid, vm size, rss,
420 * pgtables_bytes, swapents, oom_score_adj value, and name.
422 static void dump_tasks(struct oom_control *oc)
424 pr_info("Tasks state (memory values in pages):\n");
425 pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name\n");
427 if (is_memcg_oom(oc))
428 mem_cgroup_scan_tasks(oc->memcg, dump_task, oc);
430 struct task_struct *p;
439 static void dump_oom_summary(struct oom_control *oc, struct task_struct *victim)
441 /* one line summary of the oom killer context. */
442 pr_info("oom-kill:constraint=%s,nodemask=%*pbl",
443 oom_constraint_text[oc->constraint],
444 nodemask_pr_args(oc->nodemask));
445 cpuset_print_current_mems_allowed();
446 mem_cgroup_print_oom_context(oc->memcg, victim);
447 pr_cont(",task=%s,pid=%d,uid=%d\n", victim->comm, victim->pid,
448 from_kuid(&init_user_ns, task_uid(victim)));
451 static void dump_header(struct oom_control *oc, struct task_struct *p)
453 pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), order=%d, oom_score_adj=%hd\n",
454 current->comm, oc->gfp_mask, &oc->gfp_mask, oc->order,
455 current->signal->oom_score_adj);
456 if (!IS_ENABLED(CONFIG_COMPACTION) && oc->order)
457 pr_warn("COMPACTION is disabled!!!\n");
460 if (is_memcg_oom(oc))
461 mem_cgroup_print_oom_meminfo(oc->memcg);
463 show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask);
464 if (is_dump_unreclaim_slabs())
465 dump_unreclaimable_slab();
467 if (sysctl_oom_dump_tasks)
470 dump_oom_summary(oc, p);
474 * Number of OOM victims in flight
476 static atomic_t oom_victims = ATOMIC_INIT(0);
477 static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
479 static bool oom_killer_disabled __read_mostly;
481 #define K(x) ((x) << (PAGE_SHIFT-10))
484 * task->mm can be NULL if the task is the exited group leader. So to
485 * determine whether the task is using a particular mm, we examine all the
486 * task's threads: if one of those is using this mm then this task was also
489 bool process_shares_mm(struct task_struct *p, struct mm_struct *mm)
491 struct task_struct *t;
493 for_each_thread(p, t) {
494 struct mm_struct *t_mm = READ_ONCE(t->mm);
503 * OOM Reaper kernel thread which tries to reap the memory used by the OOM
504 * victim (if that is possible) to help the OOM killer to move on.
506 static struct task_struct *oom_reaper_th;
507 static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait);
508 static struct task_struct *oom_reaper_list;
509 static DEFINE_SPINLOCK(oom_reaper_lock);
511 bool __oom_reap_task_mm(struct mm_struct *mm)
513 struct vm_area_struct *vma;
517 * Tell all users of get_user/copy_from_user etc... that the content
518 * is no longer stable. No barriers really needed because unmapping
519 * should imply barriers already and the reader would hit a page fault
520 * if it stumbled over a reaped memory.
522 set_bit(MMF_UNSTABLE, &mm->flags);
524 for (vma = mm->mmap ; vma; vma = vma->vm_next) {
525 if (!can_madv_lru_vma(vma))
529 * Only anonymous pages have a good chance to be dropped
530 * without additional steps which we cannot afford as we
533 * We do not even care about fs backed pages because all
534 * which are reclaimable have already been reclaimed and
535 * we do not want to block exit_mmap by keeping mm ref
536 * count elevated without a good reason.
538 if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) {
539 struct mmu_notifier_range range;
540 struct mmu_gather tlb;
542 mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0,
543 vma, mm, vma->vm_start,
545 tlb_gather_mmu(&tlb, mm, range.start, range.end);
546 if (mmu_notifier_invalidate_range_start_nonblock(&range)) {
547 tlb_finish_mmu(&tlb, range.start, range.end);
551 unmap_page_range(&tlb, vma, range.start, range.end, NULL);
552 mmu_notifier_invalidate_range_end(&range);
553 tlb_finish_mmu(&tlb, range.start, range.end);
561 * Reaps the address space of the give task.
563 * Returns true on success and false if none or part of the address space
564 * has been reclaimed and the caller should retry later.
566 static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
570 if (!mmap_read_trylock(mm)) {
571 trace_skip_task_reaping(tsk->pid);
576 * MMF_OOM_SKIP is set by exit_mmap when the OOM reaper can't
577 * work on the mm anymore. The check for MMF_OOM_SKIP must run
578 * under mmap_lock for reading because it serializes against the
579 * mmap_write_lock();mmap_write_unlock() cycle in exit_mmap().
581 if (test_bit(MMF_OOM_SKIP, &mm->flags)) {
582 trace_skip_task_reaping(tsk->pid);
586 trace_start_task_reaping(tsk->pid);
588 /* failed to reap part of the address space. Try again later */
589 ret = __oom_reap_task_mm(mm);
593 pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
594 task_pid_nr(tsk), tsk->comm,
595 K(get_mm_counter(mm, MM_ANONPAGES)),
596 K(get_mm_counter(mm, MM_FILEPAGES)),
597 K(get_mm_counter(mm, MM_SHMEMPAGES)));
599 trace_finish_task_reaping(tsk->pid);
601 mmap_read_unlock(mm);
606 #define MAX_OOM_REAP_RETRIES 10
607 static void oom_reap_task(struct task_struct *tsk)
610 struct mm_struct *mm = tsk->signal->oom_mm;
612 /* Retry the mmap_read_trylock(mm) a few times */
613 while (attempts++ < MAX_OOM_REAP_RETRIES && !oom_reap_task_mm(tsk, mm))
614 schedule_timeout_idle(HZ/10);
616 if (attempts <= MAX_OOM_REAP_RETRIES ||
617 test_bit(MMF_OOM_SKIP, &mm->flags))
620 pr_info("oom_reaper: unable to reap pid:%d (%s)\n",
621 task_pid_nr(tsk), tsk->comm);
622 sched_show_task(tsk);
623 debug_show_all_locks();
626 tsk->oom_reaper_list = NULL;
629 * Hide this mm from OOM killer because it has been either reaped or
630 * somebody can't call mmap_write_unlock(mm).
632 set_bit(MMF_OOM_SKIP, &mm->flags);
634 /* Drop a reference taken by wake_oom_reaper */
635 put_task_struct(tsk);
638 static int oom_reaper(void *unused)
641 struct task_struct *tsk = NULL;
643 wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL);
644 spin_lock(&oom_reaper_lock);
645 if (oom_reaper_list != NULL) {
646 tsk = oom_reaper_list;
647 oom_reaper_list = tsk->oom_reaper_list;
649 spin_unlock(&oom_reaper_lock);
658 static void wake_oom_reaper(struct task_struct *tsk)
660 /* mm is already queued? */
661 if (test_and_set_bit(MMF_OOM_REAP_QUEUED, &tsk->signal->oom_mm->flags))
664 get_task_struct(tsk);
666 spin_lock(&oom_reaper_lock);
667 tsk->oom_reaper_list = oom_reaper_list;
668 oom_reaper_list = tsk;
669 spin_unlock(&oom_reaper_lock);
670 trace_wake_reaper(tsk->pid);
671 wake_up(&oom_reaper_wait);
674 static int __init oom_init(void)
676 oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper");
679 subsys_initcall(oom_init)
681 static inline void wake_oom_reaper(struct task_struct *tsk)
684 #endif /* CONFIG_MMU */
687 * mark_oom_victim - mark the given task as OOM victim
690 * Has to be called with oom_lock held and never after
691 * oom has been disabled already.
693 * tsk->mm has to be non NULL and caller has to guarantee it is stable (either
694 * under task_lock or operate on the current).
696 static void mark_oom_victim(struct task_struct *tsk)
698 struct mm_struct *mm = tsk->mm;
700 WARN_ON(oom_killer_disabled);
701 /* OOM killer might race with memcg OOM */
702 if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
705 /* oom_mm is bound to the signal struct life time. */
706 if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm)) {
707 mmgrab(tsk->signal->oom_mm);
708 set_bit(MMF_OOM_VICTIM, &mm->flags);
712 * Make sure that the task is woken up from uninterruptible sleep
713 * if it is frozen because OOM killer wouldn't be able to free
714 * any memory and livelock. freezing_slow_path will tell the freezer
715 * that TIF_MEMDIE tasks should be ignored.
718 atomic_inc(&oom_victims);
719 trace_mark_victim(tsk->pid);
723 * exit_oom_victim - note the exit of an OOM victim
725 void exit_oom_victim(void)
727 clear_thread_flag(TIF_MEMDIE);
729 if (!atomic_dec_return(&oom_victims))
730 wake_up_all(&oom_victims_wait);
734 * oom_killer_enable - enable OOM killer
736 void oom_killer_enable(void)
738 oom_killer_disabled = false;
739 pr_info("OOM killer enabled.\n");
743 * oom_killer_disable - disable OOM killer
744 * @timeout: maximum timeout to wait for oom victims in jiffies
746 * Forces all page allocations to fail rather than trigger OOM killer.
747 * Will block and wait until all OOM victims are killed or the given
750 * The function cannot be called when there are runnable user tasks because
751 * the userspace would see unexpected allocation failures as a result. Any
752 * new usage of this function should be consulted with MM people.
754 * Returns true if successful and false if the OOM killer cannot be
757 bool oom_killer_disable(signed long timeout)
762 * Make sure to not race with an ongoing OOM killer. Check that the
763 * current is not killed (possibly due to sharing the victim's memory).
765 if (mutex_lock_killable(&oom_lock))
767 oom_killer_disabled = true;
768 mutex_unlock(&oom_lock);
770 ret = wait_event_interruptible_timeout(oom_victims_wait,
771 !atomic_read(&oom_victims), timeout);
776 pr_info("OOM killer disabled.\n");
781 static inline bool __task_will_free_mem(struct task_struct *task)
783 struct signal_struct *sig = task->signal;
786 * A coredumping process may sleep for an extended period in exit_mm(),
787 * so the oom killer cannot assume that the process will promptly exit
788 * and release memory.
790 if (sig->flags & SIGNAL_GROUP_COREDUMP)
793 if (sig->flags & SIGNAL_GROUP_EXIT)
796 if (thread_group_empty(task) && (task->flags & PF_EXITING))
803 * Checks whether the given task is dying or exiting and likely to
804 * release its address space. This means that all threads and processes
805 * sharing the same mm have to be killed or exiting.
806 * Caller has to make sure that task->mm is stable (hold task_lock or
807 * it operates on the current).
809 static bool task_will_free_mem(struct task_struct *task)
811 struct mm_struct *mm = task->mm;
812 struct task_struct *p;
816 * Skip tasks without mm because it might have passed its exit_mm and
817 * exit_oom_victim. oom_reaper could have rescued that but do not rely
818 * on that for now. We can consider find_lock_task_mm in future.
823 if (!__task_will_free_mem(task))
827 * This task has already been drained by the oom reaper so there are
828 * only small chances it will free some more
830 if (test_bit(MMF_OOM_SKIP, &mm->flags))
833 if (atomic_read(&mm->mm_users) <= 1)
837 * Make sure that all tasks which share the mm with the given tasks
838 * are dying as well to make sure that a) nobody pins its mm and
839 * b) the task is also reapable by the oom reaper.
842 for_each_process(p) {
843 if (!process_shares_mm(p, mm))
845 if (same_thread_group(task, p))
847 ret = __task_will_free_mem(p);
856 static void __oom_kill_process(struct task_struct *victim, const char *message)
858 struct task_struct *p;
859 struct mm_struct *mm;
860 bool can_oom_reap = true;
862 p = find_lock_task_mm(victim);
864 pr_info("%s: OOM victim %d (%s) is already exiting. Skip killing the task\n",
865 message, task_pid_nr(victim), victim->comm);
866 put_task_struct(victim);
868 } else if (victim != p) {
870 put_task_struct(victim);
874 /* Get a reference to safely compare mm after task_unlock(victim) */
878 /* Raise event before sending signal: task reaper must see this */
879 count_vm_event(OOM_KILL);
880 memcg_memory_event_mm(mm, MEMCG_OOM_KILL);
883 * We should send SIGKILL before granting access to memory reserves
884 * in order to prevent the OOM victim from depleting the memory
885 * reserves from the user space under its control.
887 do_send_sig_info(SIGKILL, SEND_SIG_PRIV, victim, PIDTYPE_TGID);
888 mark_oom_victim(victim);
889 pr_err("%s: Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB, UID:%u pgtables:%lukB oom_score_adj:%hd\n",
890 message, task_pid_nr(victim), victim->comm, K(mm->total_vm),
891 K(get_mm_counter(mm, MM_ANONPAGES)),
892 K(get_mm_counter(mm, MM_FILEPAGES)),
893 K(get_mm_counter(mm, MM_SHMEMPAGES)),
894 from_kuid(&init_user_ns, task_uid(victim)),
895 mm_pgtables_bytes(mm) >> 10, victim->signal->oom_score_adj);
899 * Kill all user processes sharing victim->mm in other thread groups, if
900 * any. They don't get access to memory reserves, though, to avoid
901 * depletion of all memory. This prevents mm->mmap_lock livelock when an
902 * oom killed thread cannot exit because it requires the semaphore and
903 * its contended by another thread trying to allocate memory itself.
904 * That thread will now get access to memory reserves since it has a
905 * pending fatal signal.
908 for_each_process(p) {
909 if (!process_shares_mm(p, mm))
911 if (same_thread_group(p, victim))
913 if (is_global_init(p)) {
914 can_oom_reap = false;
915 set_bit(MMF_OOM_SKIP, &mm->flags);
916 pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n",
917 task_pid_nr(victim), victim->comm,
918 task_pid_nr(p), p->comm);
922 * No kthead_use_mm() user needs to read from the userspace so
923 * we are ok to reap it.
925 if (unlikely(p->flags & PF_KTHREAD))
927 do_send_sig_info(SIGKILL, SEND_SIG_PRIV, p, PIDTYPE_TGID);
932 wake_oom_reaper(victim);
935 put_task_struct(victim);
940 * Kill provided task unless it's secured by setting
941 * oom_score_adj to OOM_SCORE_ADJ_MIN.
943 static int oom_kill_memcg_member(struct task_struct *task, void *message)
945 if (task->signal->oom_score_adj != OOM_SCORE_ADJ_MIN &&
946 !is_global_init(task)) {
947 get_task_struct(task);
948 __oom_kill_process(task, message);
953 static void oom_kill_process(struct oom_control *oc, const char *message)
955 struct task_struct *victim = oc->chosen;
956 struct mem_cgroup *oom_group;
957 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
958 DEFAULT_RATELIMIT_BURST);
961 * If the task is already exiting, don't alarm the sysadmin or kill
962 * its children or threads, just give it access to memory reserves
963 * so it can die quickly
966 if (task_will_free_mem(victim)) {
967 mark_oom_victim(victim);
968 wake_oom_reaper(victim);
970 put_task_struct(victim);
975 if (__ratelimit(&oom_rs))
976 dump_header(oc, victim);
979 * Do we need to kill the entire memory cgroup?
980 * Or even one of the ancestor memory cgroups?
981 * Check this out before killing the victim task.
983 oom_group = mem_cgroup_get_oom_group(victim, oc->memcg);
985 __oom_kill_process(victim, message);
988 * If necessary, kill all tasks in the selected memory cgroup.
991 mem_cgroup_print_oom_group(oom_group);
992 mem_cgroup_scan_tasks(oom_group, oom_kill_memcg_member,
994 mem_cgroup_put(oom_group);
999 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
1001 static void check_panic_on_oom(struct oom_control *oc)
1003 if (likely(!sysctl_panic_on_oom))
1005 if (sysctl_panic_on_oom != 2) {
1007 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
1008 * does not panic for cpuset, mempolicy, or memcg allocation
1011 if (oc->constraint != CONSTRAINT_NONE)
1014 /* Do not panic for oom kills triggered by sysrq */
1015 if (is_sysrq_oom(oc))
1017 dump_header(oc, NULL);
1018 panic("Out of memory: %s panic_on_oom is enabled\n",
1019 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
1022 static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
1024 int register_oom_notifier(struct notifier_block *nb)
1026 return blocking_notifier_chain_register(&oom_notify_list, nb);
1028 EXPORT_SYMBOL_GPL(register_oom_notifier);
1030 int unregister_oom_notifier(struct notifier_block *nb)
1032 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
1034 EXPORT_SYMBOL_GPL(unregister_oom_notifier);
1037 * out_of_memory - kill the "best" process when we run out of memory
1038 * @oc: pointer to struct oom_control
1040 * If we run out of memory, we have the choice between either
1041 * killing a random task (bad), letting the system crash (worse)
1042 * OR try to be smart about which process to kill. Note that we
1043 * don't have to be perfect here, we just have to be good.
1045 bool out_of_memory(struct oom_control *oc)
1047 unsigned long freed = 0;
1049 if (oom_killer_disabled)
1052 if (!is_memcg_oom(oc)) {
1053 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
1055 /* Got some memory back in the last second. */
1060 * If current has a pending SIGKILL or is exiting, then automatically
1061 * select it. The goal is to allow it to allocate so that it may
1062 * quickly exit and free its memory.
1064 if (task_will_free_mem(current)) {
1065 mark_oom_victim(current);
1066 wake_oom_reaper(current);
1071 * The OOM killer does not compensate for IO-less reclaim.
1072 * pagefault_out_of_memory lost its gfp context so we have to
1073 * make sure exclude 0 mask - all other users should have at least
1074 * ___GFP_DIRECT_RECLAIM to get here. But mem_cgroup_oom() has to
1075 * invoke the OOM killer even if it is a GFP_NOFS allocation.
1077 if (oc->gfp_mask && !(oc->gfp_mask & __GFP_FS) && !is_memcg_oom(oc))
1081 * Check if there were limitations on the allocation (only relevant for
1082 * NUMA and memcg) that may require different handling.
1084 oc->constraint = constrained_alloc(oc);
1085 if (oc->constraint != CONSTRAINT_MEMORY_POLICY)
1086 oc->nodemask = NULL;
1087 check_panic_on_oom(oc);
1089 if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task &&
1090 current->mm && !oom_unkillable_task(current) &&
1091 oom_cpuset_eligible(current, oc) &&
1092 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
1093 get_task_struct(current);
1094 oc->chosen = current;
1095 oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)");
1099 select_bad_process(oc);
1100 /* Found nothing?!?! */
1102 dump_header(oc, NULL);
1103 pr_warn("Out of memory and no killable processes...\n");
1105 * If we got here due to an actual allocation at the
1106 * system level, we cannot survive this and will enter
1107 * an endless loop in the allocator. Bail out now.
1109 if (!is_sysrq_oom(oc) && !is_memcg_oom(oc))
1110 panic("System is deadlocked on memory\n");
1112 if (oc->chosen && oc->chosen != (void *)-1UL)
1113 oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" :
1114 "Memory cgroup out of memory");
1115 return !!oc->chosen;
1119 * The pagefault handler calls here because it is out of memory, so kill a
1120 * memory-hogging task. If oom_lock is held by somebody else, a parallel oom
1121 * killing is already in progress so do nothing.
1123 void pagefault_out_of_memory(void)
1125 struct oom_control oc = {
1133 if (mem_cgroup_oom_synchronize(true))
1136 if (!mutex_trylock(&oom_lock))
1139 mutex_unlock(&oom_lock);