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 /* return true if the task is not adequate as candidate victim task. */
136 static bool oom_unkillable_task(struct task_struct
*p
,
137 struct mem_cgroup
*memcg
, const nodemask_t
*nodemask
)
139 if (is_global_init(p
))
141 if (p
->flags
& PF_KTHREAD
)
144 /* When mem_cgroup_out_of_memory() and p is not member of the group */
145 if (memcg
&& !task_in_mem_cgroup(p
, memcg
))
148 /* p may not have freeable memory in nodemask */
149 if (!has_intersects_mems_allowed(p
, nodemask
))
156 * oom_badness - heuristic function to determine which candidate task to kill
157 * @p: task struct of which task we should calculate
158 * @totalpages: total present RAM allowed for page allocation
160 * The heuristic for determining which task to kill is made to be as simple and
161 * predictable as possible. The goal is to return the highest value for the
162 * task consuming the most memory to avoid subsequent oom failures.
164 unsigned long oom_badness(struct task_struct
*p
, struct mem_cgroup
*memcg
,
165 const nodemask_t
*nodemask
, unsigned long totalpages
)
170 if (oom_unkillable_task(p
, memcg
, nodemask
))
173 p
= find_lock_task_mm(p
);
178 * Do not even consider tasks which are explicitly marked oom
179 * unkillable or have been already oom reaped or the are in
180 * the middle of vfork
182 adj
= (long)p
->signal
->oom_score_adj
;
183 if (adj
== OOM_SCORE_ADJ_MIN
||
184 test_bit(MMF_OOM_REAPED
, &p
->mm
->flags
) ||
191 * The baseline for the badness score is the proportion of RAM that each
192 * task's rss, pagetable and swap space use.
194 points
= get_mm_rss(p
->mm
) + get_mm_counter(p
->mm
, MM_SWAPENTS
) +
195 atomic_long_read(&p
->mm
->nr_ptes
) + mm_nr_pmds(p
->mm
);
199 * Root processes get 3% bonus, just like the __vm_enough_memory()
200 * implementation used by LSMs.
202 if (has_capability_noaudit(p
, CAP_SYS_ADMIN
))
203 points
-= (points
* 3) / 100;
205 /* Normalize to oom_score_adj units */
206 adj
*= totalpages
/ 1000;
210 * Never return 0 for an eligible task regardless of the root bonus and
211 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
213 return points
> 0 ? points
: 1;
217 * Determine the type of allocation constraint.
220 static enum oom_constraint
constrained_alloc(struct oom_control
*oc
,
221 unsigned long *totalpages
)
225 enum zone_type high_zoneidx
= gfp_zone(oc
->gfp_mask
);
226 bool cpuset_limited
= false;
229 /* Default to all available memory */
230 *totalpages
= totalram_pages
+ total_swap_pages
;
233 return CONSTRAINT_NONE
;
235 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
236 * to kill current.We have to random task kill in this case.
237 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
239 if (oc
->gfp_mask
& __GFP_THISNODE
)
240 return CONSTRAINT_NONE
;
243 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
244 * the page allocator means a mempolicy is in effect. Cpuset policy
245 * is enforced in get_page_from_freelist().
248 !nodes_subset(node_states
[N_MEMORY
], *oc
->nodemask
)) {
249 *totalpages
= total_swap_pages
;
250 for_each_node_mask(nid
, *oc
->nodemask
)
251 *totalpages
+= node_spanned_pages(nid
);
252 return CONSTRAINT_MEMORY_POLICY
;
255 /* Check this allocation failure is caused by cpuset's wall function */
256 for_each_zone_zonelist_nodemask(zone
, z
, oc
->zonelist
,
257 high_zoneidx
, oc
->nodemask
)
258 if (!cpuset_zone_allowed(zone
, oc
->gfp_mask
))
259 cpuset_limited
= true;
261 if (cpuset_limited
) {
262 *totalpages
= total_swap_pages
;
263 for_each_node_mask(nid
, cpuset_current_mems_allowed
)
264 *totalpages
+= node_spanned_pages(nid
);
265 return CONSTRAINT_CPUSET
;
267 return CONSTRAINT_NONE
;
270 static enum oom_constraint
constrained_alloc(struct oom_control
*oc
,
271 unsigned long *totalpages
)
273 *totalpages
= totalram_pages
+ total_swap_pages
;
274 return CONSTRAINT_NONE
;
278 enum oom_scan_t
oom_scan_process_thread(struct oom_control
*oc
,
279 struct task_struct
*task
)
281 if (oom_unkillable_task(task
, NULL
, oc
->nodemask
))
282 return OOM_SCAN_CONTINUE
;
285 * This task already has access to memory reserves and is being killed.
286 * Don't allow any other task to have access to the reserves.
288 if (!is_sysrq_oom(oc
) && atomic_read(&task
->signal
->oom_victims
))
289 return OOM_SCAN_ABORT
;
292 * If task is allocating a lot of memory and has been marked to be
293 * killed first if it triggers an oom, then select it.
295 if (oom_task_origin(task
))
296 return OOM_SCAN_SELECT
;
302 * Simple selection loop. We chose the process with the highest
303 * number of 'points'. Returns -1 on scan abort.
305 static struct task_struct
*select_bad_process(struct oom_control
*oc
,
306 unsigned int *ppoints
, unsigned long totalpages
)
308 struct task_struct
*p
;
309 struct task_struct
*chosen
= NULL
;
310 unsigned long chosen_points
= 0;
313 for_each_process(p
) {
316 switch (oom_scan_process_thread(oc
, p
)) {
317 case OOM_SCAN_SELECT
:
319 chosen_points
= ULONG_MAX
;
321 case OOM_SCAN_CONTINUE
:
325 return (struct task_struct
*)(-1UL);
329 points
= oom_badness(p
, NULL
, oc
->nodemask
, totalpages
);
330 if (!points
|| points
< chosen_points
)
334 chosen_points
= points
;
337 get_task_struct(chosen
);
340 *ppoints
= chosen_points
* 1000 / totalpages
;
345 * dump_tasks - dump current memory state of all system tasks
346 * @memcg: current's memory controller, if constrained
347 * @nodemask: nodemask passed to page allocator for mempolicy ooms
349 * Dumps the current memory state of all eligible tasks. Tasks not in the same
350 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
352 * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
353 * swapents, oom_score_adj value, and name.
355 static void dump_tasks(struct mem_cgroup
*memcg
, const nodemask_t
*nodemask
)
357 struct task_struct
*p
;
358 struct task_struct
*task
;
360 pr_info("[ pid ] uid tgid total_vm rss nr_ptes nr_pmds swapents oom_score_adj name\n");
362 for_each_process(p
) {
363 if (oom_unkillable_task(p
, memcg
, nodemask
))
366 task
= find_lock_task_mm(p
);
369 * This is a kthread or all of p's threads have already
370 * detached their mm's. There's no need to report
371 * them; they can't be oom killed anyway.
376 pr_info("[%5d] %5d %5d %8lu %8lu %7ld %7ld %8lu %5hd %s\n",
377 task
->pid
, from_kuid(&init_user_ns
, task_uid(task
)),
378 task
->tgid
, task
->mm
->total_vm
, get_mm_rss(task
->mm
),
379 atomic_long_read(&task
->mm
->nr_ptes
),
380 mm_nr_pmds(task
->mm
),
381 get_mm_counter(task
->mm
, MM_SWAPENTS
),
382 task
->signal
->oom_score_adj
, task
->comm
);
388 static void dump_header(struct oom_control
*oc
, struct task_struct
*p
)
390 pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), order=%d, oom_score_adj=%hd\n",
391 current
->comm
, oc
->gfp_mask
, &oc
->gfp_mask
, oc
->order
,
392 current
->signal
->oom_score_adj
);
394 cpuset_print_current_mems_allowed();
397 mem_cgroup_print_oom_info(oc
->memcg
, p
);
399 show_mem(SHOW_MEM_FILTER_NODES
);
400 if (sysctl_oom_dump_tasks
)
401 dump_tasks(oc
->memcg
, oc
->nodemask
);
405 * Number of OOM victims in flight
407 static atomic_t oom_victims
= ATOMIC_INIT(0);
408 static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait
);
410 bool oom_killer_disabled __read_mostly
;
412 #define K(x) ((x) << (PAGE_SHIFT-10))
415 * task->mm can be NULL if the task is the exited group leader. So to
416 * determine whether the task is using a particular mm, we examine all the
417 * task's threads: if one of those is using this mm then this task was also
420 bool process_shares_mm(struct task_struct
*p
, struct mm_struct
*mm
)
422 struct task_struct
*t
;
424 for_each_thread(p
, t
) {
425 struct mm_struct
*t_mm
= READ_ONCE(t
->mm
);
435 * OOM Reaper kernel thread which tries to reap the memory used by the OOM
436 * victim (if that is possible) to help the OOM killer to move on.
438 static struct task_struct
*oom_reaper_th
;
439 static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait
);
440 static struct task_struct
*oom_reaper_list
;
441 static DEFINE_SPINLOCK(oom_reaper_lock
);
443 static bool __oom_reap_task(struct task_struct
*tsk
)
445 struct mmu_gather tlb
;
446 struct vm_area_struct
*vma
;
447 struct mm_struct
*mm
= NULL
;
448 struct task_struct
*p
;
449 struct zap_details details
= {.check_swap_entries
= true,
450 .ignore_dirty
= true};
454 * We have to make sure to not race with the victim exit path
455 * and cause premature new oom victim selection:
456 * __oom_reap_task exit_mm
459 * atomic_dec_and_test
464 * # no TIF_MEMDIE task selects new victim
465 * unmap_page_range # frees some memory
467 mutex_lock(&oom_lock
);
470 * Make sure we find the associated mm_struct even when the particular
471 * thread has already terminated and cleared its mm.
472 * We might have race with exit path so consider our work done if there
475 p
= find_lock_task_mm(tsk
);
479 atomic_inc(&mm
->mm_count
);
482 if (!down_read_trylock(&mm
->mmap_sem
)) {
488 * increase mm_users only after we know we will reap something so
489 * that the mmput_async is called only when we have reaped something
490 * and delayed __mmput doesn't matter that much
492 if (!mmget_not_zero(mm
)) {
493 up_read(&mm
->mmap_sem
);
497 tlb_gather_mmu(&tlb
, mm
, 0, -1);
498 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
499 if (is_vm_hugetlb_page(vma
))
503 * mlocked VMAs require explicit munlocking before unmap.
504 * Let's keep it simple here and skip such VMAs.
506 if (vma
->vm_flags
& VM_LOCKED
)
510 * Only anonymous pages have a good chance to be dropped
511 * without additional steps which we cannot afford as we
514 * We do not even care about fs backed pages because all
515 * which are reclaimable have already been reclaimed and
516 * we do not want to block exit_mmap by keeping mm ref
517 * count elevated without a good reason.
519 if (vma_is_anonymous(vma
) || !(vma
->vm_flags
& VM_SHARED
))
520 unmap_page_range(&tlb
, vma
, vma
->vm_start
, vma
->vm_end
,
523 tlb_finish_mmu(&tlb
, 0, -1);
524 pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
525 task_pid_nr(tsk
), tsk
->comm
,
526 K(get_mm_counter(mm
, MM_ANONPAGES
)),
527 K(get_mm_counter(mm
, MM_FILEPAGES
)),
528 K(get_mm_counter(mm
, MM_SHMEMPAGES
)));
529 up_read(&mm
->mmap_sem
);
532 * This task can be safely ignored because we cannot do much more
533 * to release its memory.
535 set_bit(MMF_OOM_REAPED
, &mm
->flags
);
537 * Drop our reference but make sure the mmput slow path is called from a
538 * different context because we shouldn't risk we get stuck there and
539 * put the oom_reaper out of the way.
545 mutex_unlock(&oom_lock
);
549 #define MAX_OOM_REAP_RETRIES 10
550 static void oom_reap_task(struct task_struct
*tsk
)
554 /* Retry the down_read_trylock(mmap_sem) a few times */
555 while (attempts
++ < MAX_OOM_REAP_RETRIES
&& !__oom_reap_task(tsk
))
556 schedule_timeout_idle(HZ
/10);
558 if (attempts
> MAX_OOM_REAP_RETRIES
) {
559 pr_info("oom_reaper: unable to reap pid:%d (%s)\n",
560 task_pid_nr(tsk
), tsk
->comm
);
561 debug_show_all_locks();
565 * Clear TIF_MEMDIE because the task shouldn't be sitting on a
566 * reasonably reclaimable memory anymore or it is not a good candidate
567 * for the oom victim right now because it cannot release its memory
568 * itself nor by the oom reaper.
570 tsk
->oom_reaper_list
= NULL
;
571 exit_oom_victim(tsk
);
573 /* Drop a reference taken by wake_oom_reaper */
574 put_task_struct(tsk
);
577 static int oom_reaper(void *unused
)
582 struct task_struct
*tsk
= NULL
;
584 wait_event_freezable(oom_reaper_wait
, oom_reaper_list
!= NULL
);
585 spin_lock(&oom_reaper_lock
);
586 if (oom_reaper_list
!= NULL
) {
587 tsk
= oom_reaper_list
;
588 oom_reaper_list
= tsk
->oom_reaper_list
;
590 spin_unlock(&oom_reaper_lock
);
599 static void wake_oom_reaper(struct task_struct
*tsk
)
604 /* tsk is already queued? */
605 if (tsk
== oom_reaper_list
|| tsk
->oom_reaper_list
)
608 get_task_struct(tsk
);
610 spin_lock(&oom_reaper_lock
);
611 tsk
->oom_reaper_list
= oom_reaper_list
;
612 oom_reaper_list
= tsk
;
613 spin_unlock(&oom_reaper_lock
);
614 wake_up(&oom_reaper_wait
);
617 /* Check if we can reap the given task. This has to be called with stable
620 void try_oom_reaper(struct task_struct
*tsk
)
622 struct mm_struct
*mm
= tsk
->mm
;
623 struct task_struct
*p
;
629 * There might be other threads/processes which are either not
630 * dying or even not killable.
632 if (atomic_read(&mm
->mm_users
) > 1) {
634 for_each_process(p
) {
635 if (!process_shares_mm(p
, mm
))
637 if (fatal_signal_pending(p
))
641 * If the task is exiting make sure the whole thread group
642 * is exiting and cannot acces mm anymore.
644 if (signal_group_exit(p
->signal
))
654 wake_oom_reaper(tsk
);
657 static int __init
oom_init(void)
659 oom_reaper_th
= kthread_run(oom_reaper
, NULL
, "oom_reaper");
660 if (IS_ERR(oom_reaper_th
)) {
661 pr_err("Unable to start OOM reaper %ld. Continuing regardless\n",
662 PTR_ERR(oom_reaper_th
));
663 oom_reaper_th
= NULL
;
667 subsys_initcall(oom_init
)
669 static void wake_oom_reaper(struct task_struct
*tsk
)
675 * mark_oom_victim - mark the given task as OOM victim
678 * Has to be called with oom_lock held and never after
679 * oom has been disabled already.
681 void mark_oom_victim(struct task_struct
*tsk
)
683 WARN_ON(oom_killer_disabled
);
684 /* OOM killer might race with memcg OOM */
685 if (test_and_set_tsk_thread_flag(tsk
, TIF_MEMDIE
))
687 atomic_inc(&tsk
->signal
->oom_victims
);
689 * Make sure that the task is woken up from uninterruptible sleep
690 * if it is frozen because OOM killer wouldn't be able to free
691 * any memory and livelock. freezing_slow_path will tell the freezer
692 * that TIF_MEMDIE tasks should be ignored.
695 atomic_inc(&oom_victims
);
699 * exit_oom_victim - note the exit of an OOM victim
701 void exit_oom_victim(struct task_struct
*tsk
)
703 if (!test_and_clear_tsk_thread_flag(tsk
, TIF_MEMDIE
))
705 atomic_dec(&tsk
->signal
->oom_victims
);
707 if (!atomic_dec_return(&oom_victims
))
708 wake_up_all(&oom_victims_wait
);
712 * oom_killer_disable - disable OOM killer
714 * Forces all page allocations to fail rather than trigger OOM killer.
715 * Will block and wait until all OOM victims are killed.
717 * The function cannot be called when there are runnable user tasks because
718 * the userspace would see unexpected allocation failures as a result. Any
719 * new usage of this function should be consulted with MM people.
721 * Returns true if successful and false if the OOM killer cannot be
724 bool oom_killer_disable(void)
727 * Make sure to not race with an ongoing OOM killer. Check that the
728 * current is not killed (possibly due to sharing the victim's memory).
730 if (mutex_lock_killable(&oom_lock
))
732 oom_killer_disabled
= true;
733 mutex_unlock(&oom_lock
);
735 wait_event(oom_victims_wait
, !atomic_read(&oom_victims
));
741 * oom_killer_enable - enable OOM killer
743 void oom_killer_enable(void)
745 oom_killer_disabled
= false;
749 * Must be called while holding a reference to p, which will be released upon
752 void oom_kill_process(struct oom_control
*oc
, struct task_struct
*p
,
753 unsigned int points
, unsigned long totalpages
,
756 struct task_struct
*victim
= p
;
757 struct task_struct
*child
;
758 struct task_struct
*t
;
759 struct mm_struct
*mm
;
760 unsigned int victim_points
= 0;
761 static DEFINE_RATELIMIT_STATE(oom_rs
, DEFAULT_RATELIMIT_INTERVAL
,
762 DEFAULT_RATELIMIT_BURST
);
763 bool can_oom_reap
= true;
766 * If the task is already exiting, don't alarm the sysadmin or kill
767 * its children or threads, just set TIF_MEMDIE so it can die quickly
770 if (p
->mm
&& task_will_free_mem(p
)) {
779 if (__ratelimit(&oom_rs
))
782 pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n",
783 message
, task_pid_nr(p
), p
->comm
, points
);
786 * If any of p's children has a different mm and is eligible for kill,
787 * the one with the highest oom_badness() score is sacrificed for its
788 * parent. This attempts to lose the minimal amount of work done while
789 * still freeing memory.
791 read_lock(&tasklist_lock
);
792 for_each_thread(p
, t
) {
793 list_for_each_entry(child
, &t
->children
, sibling
) {
794 unsigned int child_points
;
796 if (process_shares_mm(child
, p
->mm
))
799 * oom_badness() returns 0 if the thread is unkillable
801 child_points
= oom_badness(child
,
802 oc
->memcg
, oc
->nodemask
, totalpages
);
803 if (child_points
> victim_points
) {
804 put_task_struct(victim
);
806 victim_points
= child_points
;
807 get_task_struct(victim
);
811 read_unlock(&tasklist_lock
);
813 p
= find_lock_task_mm(victim
);
815 put_task_struct(victim
);
817 } else if (victim
!= p
) {
819 put_task_struct(victim
);
823 /* Get a reference to safely compare mm after task_unlock(victim) */
825 atomic_inc(&mm
->mm_count
);
827 * We should send SIGKILL before setting TIF_MEMDIE in order to prevent
828 * the OOM victim from depleting the memory reserves from the user
829 * space under its control.
831 do_send_sig_info(SIGKILL
, SEND_SIG_FORCED
, victim
, true);
832 mark_oom_victim(victim
);
833 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
834 task_pid_nr(victim
), victim
->comm
, K(victim
->mm
->total_vm
),
835 K(get_mm_counter(victim
->mm
, MM_ANONPAGES
)),
836 K(get_mm_counter(victim
->mm
, MM_FILEPAGES
)),
837 K(get_mm_counter(victim
->mm
, MM_SHMEMPAGES
)));
841 * Kill all user processes sharing victim->mm in other thread groups, if
842 * any. They don't get access to memory reserves, though, to avoid
843 * depletion of all memory. This prevents mm->mmap_sem livelock when an
844 * oom killed thread cannot exit because it requires the semaphore and
845 * its contended by another thread trying to allocate memory itself.
846 * That thread will now get access to memory reserves since it has a
847 * pending fatal signal.
850 for_each_process(p
) {
851 if (!process_shares_mm(p
, mm
))
853 if (same_thread_group(p
, victim
))
855 if (unlikely(p
->flags
& PF_KTHREAD
) || is_global_init(p
) ||
856 p
->signal
->oom_score_adj
== OOM_SCORE_ADJ_MIN
) {
858 * We cannot use oom_reaper for the mm shared by this
859 * process because it wouldn't get killed and so the
860 * memory might be still used.
862 can_oom_reap
= false;
865 do_send_sig_info(SIGKILL
, SEND_SIG_FORCED
, p
, true);
870 wake_oom_reaper(victim
);
873 put_task_struct(victim
);
878 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
880 void check_panic_on_oom(struct oom_control
*oc
, enum oom_constraint constraint
)
882 if (likely(!sysctl_panic_on_oom
))
884 if (sysctl_panic_on_oom
!= 2) {
886 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
887 * does not panic for cpuset, mempolicy, or memcg allocation
890 if (constraint
!= CONSTRAINT_NONE
)
893 /* Do not panic for oom kills triggered by sysrq */
894 if (is_sysrq_oom(oc
))
896 dump_header(oc
, NULL
);
897 panic("Out of memory: %s panic_on_oom is enabled\n",
898 sysctl_panic_on_oom
== 2 ? "compulsory" : "system-wide");
901 static BLOCKING_NOTIFIER_HEAD(oom_notify_list
);
903 int register_oom_notifier(struct notifier_block
*nb
)
905 return blocking_notifier_chain_register(&oom_notify_list
, nb
);
907 EXPORT_SYMBOL_GPL(register_oom_notifier
);
909 int unregister_oom_notifier(struct notifier_block
*nb
)
911 return blocking_notifier_chain_unregister(&oom_notify_list
, nb
);
913 EXPORT_SYMBOL_GPL(unregister_oom_notifier
);
916 * out_of_memory - kill the "best" process when we run out of memory
917 * @oc: pointer to struct oom_control
919 * If we run out of memory, we have the choice between either
920 * killing a random task (bad), letting the system crash (worse)
921 * OR try to be smart about which process to kill. Note that we
922 * don't have to be perfect here, we just have to be good.
924 bool out_of_memory(struct oom_control
*oc
)
926 struct task_struct
*p
;
927 unsigned long totalpages
;
928 unsigned long freed
= 0;
929 unsigned int uninitialized_var(points
);
930 enum oom_constraint constraint
= CONSTRAINT_NONE
;
932 if (oom_killer_disabled
)
935 blocking_notifier_call_chain(&oom_notify_list
, 0, &freed
);
937 /* Got some memory back in the last second. */
941 * If current has a pending SIGKILL or is exiting, then automatically
942 * select it. The goal is to allow it to allocate so that it may
943 * quickly exit and free its memory.
945 * But don't select if current has already released its mm and cleared
946 * TIF_MEMDIE flag at exit_mm(), otherwise an OOM livelock may occur.
949 (fatal_signal_pending(current
) || task_will_free_mem(current
))) {
950 mark_oom_victim(current
);
951 try_oom_reaper(current
);
956 * The OOM killer does not compensate for IO-less reclaim.
957 * pagefault_out_of_memory lost its gfp context so we have to
958 * make sure exclude 0 mask - all other users should have at least
959 * ___GFP_DIRECT_RECLAIM to get here.
961 if (oc
->gfp_mask
&& !(oc
->gfp_mask
& (__GFP_FS
|__GFP_NOFAIL
)))
965 * Check if there were limitations on the allocation (only relevant for
966 * NUMA) that may require different handling.
968 constraint
= constrained_alloc(oc
, &totalpages
);
969 if (constraint
!= CONSTRAINT_MEMORY_POLICY
)
971 check_panic_on_oom(oc
, constraint
);
973 if (sysctl_oom_kill_allocating_task
&& current
->mm
&&
974 !oom_unkillable_task(current
, NULL
, oc
->nodemask
) &&
975 current
->signal
->oom_score_adj
!= OOM_SCORE_ADJ_MIN
) {
976 get_task_struct(current
);
977 oom_kill_process(oc
, current
, 0, totalpages
,
978 "Out of memory (oom_kill_allocating_task)");
982 p
= select_bad_process(oc
, &points
, totalpages
);
983 /* Found nothing?!?! Either we hang forever, or we panic. */
984 if (!p
&& !is_sysrq_oom(oc
)) {
985 dump_header(oc
, NULL
);
986 panic("Out of memory and no killable processes...\n");
988 if (p
&& p
!= (void *)-1UL) {
989 oom_kill_process(oc
, p
, points
, totalpages
, "Out of memory");
991 * Give the killed process a good chance to exit before trying
992 * to allocate memory again.
994 schedule_timeout_killable(1);
1000 * The pagefault handler calls here because it is out of memory, so kill a
1001 * memory-hogging task. If oom_lock is held by somebody else, a parallel oom
1002 * killing is already in progress so do nothing.
1004 void pagefault_out_of_memory(void)
1006 struct oom_control oc
= {
1014 if (mem_cgroup_oom_synchronize(true))
1017 if (!mutex_trylock(&oom_lock
))
1020 if (!out_of_memory(&oc
)) {
1022 * There shouldn't be any user tasks runnable while the
1023 * OOM killer is disabled, so the current task has to
1024 * be a racing OOM victim for which oom_killer_disable()
1027 WARN_ON(test_thread_flag(TIF_MEMDIE
));
1030 mutex_unlock(&oom_lock
);