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.
181 adj
= (long)p
->signal
->oom_score_adj
;
182 if (adj
== OOM_SCORE_ADJ_MIN
||
183 test_bit(MMF_OOM_REAPED
, &p
->mm
->flags
)) {
189 * The baseline for the badness score is the proportion of RAM that each
190 * task's rss, pagetable and swap space use.
192 points
= get_mm_rss(p
->mm
) + get_mm_counter(p
->mm
, MM_SWAPENTS
) +
193 atomic_long_read(&p
->mm
->nr_ptes
) + mm_nr_pmds(p
->mm
);
197 * Root processes get 3% bonus, just like the __vm_enough_memory()
198 * implementation used by LSMs.
200 if (has_capability_noaudit(p
, CAP_SYS_ADMIN
))
201 points
-= (points
* 3) / 100;
203 /* Normalize to oom_score_adj units */
204 adj
*= totalpages
/ 1000;
208 * Never return 0 for an eligible task regardless of the root bonus and
209 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
211 return points
> 0 ? points
: 1;
215 * Determine the type of allocation constraint.
218 static enum oom_constraint
constrained_alloc(struct oom_control
*oc
,
219 unsigned long *totalpages
)
223 enum zone_type high_zoneidx
= gfp_zone(oc
->gfp_mask
);
224 bool cpuset_limited
= false;
227 /* Default to all available memory */
228 *totalpages
= totalram_pages
+ total_swap_pages
;
231 return CONSTRAINT_NONE
;
233 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
234 * to kill current.We have to random task kill in this case.
235 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
237 if (oc
->gfp_mask
& __GFP_THISNODE
)
238 return CONSTRAINT_NONE
;
241 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
242 * the page allocator means a mempolicy is in effect. Cpuset policy
243 * is enforced in get_page_from_freelist().
246 !nodes_subset(node_states
[N_MEMORY
], *oc
->nodemask
)) {
247 *totalpages
= total_swap_pages
;
248 for_each_node_mask(nid
, *oc
->nodemask
)
249 *totalpages
+= node_spanned_pages(nid
);
250 return CONSTRAINT_MEMORY_POLICY
;
253 /* Check this allocation failure is caused by cpuset's wall function */
254 for_each_zone_zonelist_nodemask(zone
, z
, oc
->zonelist
,
255 high_zoneidx
, oc
->nodemask
)
256 if (!cpuset_zone_allowed(zone
, oc
->gfp_mask
))
257 cpuset_limited
= true;
259 if (cpuset_limited
) {
260 *totalpages
= total_swap_pages
;
261 for_each_node_mask(nid
, cpuset_current_mems_allowed
)
262 *totalpages
+= node_spanned_pages(nid
);
263 return CONSTRAINT_CPUSET
;
265 return CONSTRAINT_NONE
;
268 static enum oom_constraint
constrained_alloc(struct oom_control
*oc
,
269 unsigned long *totalpages
)
271 *totalpages
= totalram_pages
+ total_swap_pages
;
272 return CONSTRAINT_NONE
;
276 enum oom_scan_t
oom_scan_process_thread(struct oom_control
*oc
,
277 struct task_struct
*task
)
279 if (oom_unkillable_task(task
, NULL
, oc
->nodemask
))
280 return OOM_SCAN_CONTINUE
;
283 * This task already has access to memory reserves and is being killed.
284 * Don't allow any other task to have access to the reserves.
286 if (!is_sysrq_oom(oc
) && atomic_read(&task
->signal
->oom_victims
))
287 return OOM_SCAN_ABORT
;
290 * If task is allocating a lot of memory and has been marked to be
291 * killed first if it triggers an oom, then select it.
293 if (oom_task_origin(task
))
294 return OOM_SCAN_SELECT
;
300 * Simple selection loop. We chose the process with the highest
301 * number of 'points'. Returns -1 on scan abort.
303 static struct task_struct
*select_bad_process(struct oom_control
*oc
,
304 unsigned int *ppoints
, unsigned long totalpages
)
306 struct task_struct
*p
;
307 struct task_struct
*chosen
= NULL
;
308 unsigned long chosen_points
= 0;
311 for_each_process(p
) {
314 switch (oom_scan_process_thread(oc
, p
)) {
315 case OOM_SCAN_SELECT
:
317 chosen_points
= ULONG_MAX
;
319 case OOM_SCAN_CONTINUE
:
323 return (struct task_struct
*)(-1UL);
327 points
= oom_badness(p
, NULL
, oc
->nodemask
, totalpages
);
328 if (!points
|| points
< chosen_points
)
332 chosen_points
= points
;
335 get_task_struct(chosen
);
338 *ppoints
= chosen_points
* 1000 / totalpages
;
343 * dump_tasks - dump current memory state of all system tasks
344 * @memcg: current's memory controller, if constrained
345 * @nodemask: nodemask passed to page allocator for mempolicy ooms
347 * Dumps the current memory state of all eligible tasks. Tasks not in the same
348 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
350 * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
351 * swapents, oom_score_adj value, and name.
353 static void dump_tasks(struct mem_cgroup
*memcg
, const nodemask_t
*nodemask
)
355 struct task_struct
*p
;
356 struct task_struct
*task
;
358 pr_info("[ pid ] uid tgid total_vm rss nr_ptes nr_pmds swapents oom_score_adj name\n");
360 for_each_process(p
) {
361 if (oom_unkillable_task(p
, memcg
, nodemask
))
364 task
= find_lock_task_mm(p
);
367 * This is a kthread or all of p's threads have already
368 * detached their mm's. There's no need to report
369 * them; they can't be oom killed anyway.
374 pr_info("[%5d] %5d %5d %8lu %8lu %7ld %7ld %8lu %5hd %s\n",
375 task
->pid
, from_kuid(&init_user_ns
, task_uid(task
)),
376 task
->tgid
, task
->mm
->total_vm
, get_mm_rss(task
->mm
),
377 atomic_long_read(&task
->mm
->nr_ptes
),
378 mm_nr_pmds(task
->mm
),
379 get_mm_counter(task
->mm
, MM_SWAPENTS
),
380 task
->signal
->oom_score_adj
, task
->comm
);
386 static void dump_header(struct oom_control
*oc
, struct task_struct
*p
)
388 pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), order=%d, oom_score_adj=%hd\n",
389 current
->comm
, oc
->gfp_mask
, &oc
->gfp_mask
, oc
->order
,
390 current
->signal
->oom_score_adj
);
392 cpuset_print_current_mems_allowed();
395 mem_cgroup_print_oom_info(oc
->memcg
, p
);
397 show_mem(SHOW_MEM_FILTER_NODES
);
398 if (sysctl_oom_dump_tasks
)
399 dump_tasks(oc
->memcg
, oc
->nodemask
);
403 * Number of OOM victims in flight
405 static atomic_t oom_victims
= ATOMIC_INIT(0);
406 static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait
);
408 bool oom_killer_disabled __read_mostly
;
410 #define K(x) ((x) << (PAGE_SHIFT-10))
413 * task->mm can be NULL if the task is the exited group leader. So to
414 * determine whether the task is using a particular mm, we examine all the
415 * task's threads: if one of those is using this mm then this task was also
418 static bool process_shares_mm(struct task_struct
*p
, struct mm_struct
*mm
)
420 struct task_struct
*t
;
422 for_each_thread(p
, t
) {
423 struct mm_struct
*t_mm
= READ_ONCE(t
->mm
);
433 * OOM Reaper kernel thread which tries to reap the memory used by the OOM
434 * victim (if that is possible) to help the OOM killer to move on.
436 static struct task_struct
*oom_reaper_th
;
437 static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait
);
438 static struct task_struct
*oom_reaper_list
;
439 static DEFINE_SPINLOCK(oom_reaper_lock
);
441 static bool __oom_reap_task(struct task_struct
*tsk
)
443 struct mmu_gather tlb
;
444 struct vm_area_struct
*vma
;
445 struct mm_struct
*mm
= NULL
;
446 struct task_struct
*p
;
447 struct zap_details details
= {.check_swap_entries
= true,
448 .ignore_dirty
= true};
452 * We have to make sure to not race with the victim exit path
453 * and cause premature new oom victim selection:
454 * __oom_reap_task exit_mm
457 * atomic_dec_and_test
462 * # no TIF_MEMDIE task selects new victim
463 * unmap_page_range # frees some memory
465 mutex_lock(&oom_lock
);
468 * Make sure we find the associated mm_struct even when the particular
469 * thread has already terminated and cleared its mm.
470 * We might have race with exit path so consider our work done if there
473 p
= find_lock_task_mm(tsk
);
477 atomic_inc(&mm
->mm_count
);
480 if (!down_read_trylock(&mm
->mmap_sem
)) {
486 * increase mm_users only after we know we will reap something so
487 * that the mmput_async is called only when we have reaped something
488 * and delayed __mmput doesn't matter that much
490 if (!mmget_not_zero(mm
)) {
491 up_read(&mm
->mmap_sem
);
495 tlb_gather_mmu(&tlb
, mm
, 0, -1);
496 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
497 if (is_vm_hugetlb_page(vma
))
501 * mlocked VMAs require explicit munlocking before unmap.
502 * Let's keep it simple here and skip such VMAs.
504 if (vma
->vm_flags
& VM_LOCKED
)
508 * Only anonymous pages have a good chance to be dropped
509 * without additional steps which we cannot afford as we
512 * We do not even care about fs backed pages because all
513 * which are reclaimable have already been reclaimed and
514 * we do not want to block exit_mmap by keeping mm ref
515 * count elevated without a good reason.
517 if (vma_is_anonymous(vma
) || !(vma
->vm_flags
& VM_SHARED
))
518 unmap_page_range(&tlb
, vma
, vma
->vm_start
, vma
->vm_end
,
521 tlb_finish_mmu(&tlb
, 0, -1);
522 pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
523 task_pid_nr(tsk
), tsk
->comm
,
524 K(get_mm_counter(mm
, MM_ANONPAGES
)),
525 K(get_mm_counter(mm
, MM_FILEPAGES
)),
526 K(get_mm_counter(mm
, MM_SHMEMPAGES
)));
527 up_read(&mm
->mmap_sem
);
530 * This task can be safely ignored because we cannot do much more
531 * to release its memory.
533 set_bit(MMF_OOM_REAPED
, &mm
->flags
);
535 * Drop our reference but make sure the mmput slow path is called from a
536 * different context because we shouldn't risk we get stuck there and
537 * put the oom_reaper out of the way.
543 mutex_unlock(&oom_lock
);
547 #define MAX_OOM_REAP_RETRIES 10
548 static void oom_reap_task(struct task_struct
*tsk
)
552 /* Retry the down_read_trylock(mmap_sem) a few times */
553 while (attempts
++ < MAX_OOM_REAP_RETRIES
&& !__oom_reap_task(tsk
))
554 schedule_timeout_idle(HZ
/10);
556 if (attempts
> MAX_OOM_REAP_RETRIES
) {
557 pr_info("oom_reaper: unable to reap pid:%d (%s)\n",
558 task_pid_nr(tsk
), tsk
->comm
);
559 debug_show_all_locks();
563 * Clear TIF_MEMDIE because the task shouldn't be sitting on a
564 * reasonably reclaimable memory anymore or it is not a good candidate
565 * for the oom victim right now because it cannot release its memory
566 * itself nor by the oom reaper.
568 tsk
->oom_reaper_list
= NULL
;
569 exit_oom_victim(tsk
);
571 /* Drop a reference taken by wake_oom_reaper */
572 put_task_struct(tsk
);
575 static int oom_reaper(void *unused
)
580 struct task_struct
*tsk
= NULL
;
582 wait_event_freezable(oom_reaper_wait
, oom_reaper_list
!= NULL
);
583 spin_lock(&oom_reaper_lock
);
584 if (oom_reaper_list
!= NULL
) {
585 tsk
= oom_reaper_list
;
586 oom_reaper_list
= tsk
->oom_reaper_list
;
588 spin_unlock(&oom_reaper_lock
);
597 static void wake_oom_reaper(struct task_struct
*tsk
)
602 /* tsk is already queued? */
603 if (tsk
== oom_reaper_list
|| tsk
->oom_reaper_list
)
606 get_task_struct(tsk
);
608 spin_lock(&oom_reaper_lock
);
609 tsk
->oom_reaper_list
= oom_reaper_list
;
610 oom_reaper_list
= tsk
;
611 spin_unlock(&oom_reaper_lock
);
612 wake_up(&oom_reaper_wait
);
615 /* Check if we can reap the given task. This has to be called with stable
618 void try_oom_reaper(struct task_struct
*tsk
)
620 struct mm_struct
*mm
= tsk
->mm
;
621 struct task_struct
*p
;
627 * There might be other threads/processes which are either not
628 * dying or even not killable.
630 if (atomic_read(&mm
->mm_users
) > 1) {
632 for_each_process(p
) {
633 if (!process_shares_mm(p
, mm
))
635 if (fatal_signal_pending(p
))
639 * If the task is exiting make sure the whole thread group
640 * is exiting and cannot acces mm anymore.
642 if (signal_group_exit(p
->signal
))
652 wake_oom_reaper(tsk
);
655 static int __init
oom_init(void)
657 oom_reaper_th
= kthread_run(oom_reaper
, NULL
, "oom_reaper");
658 if (IS_ERR(oom_reaper_th
)) {
659 pr_err("Unable to start OOM reaper %ld. Continuing regardless\n",
660 PTR_ERR(oom_reaper_th
));
661 oom_reaper_th
= NULL
;
665 subsys_initcall(oom_init
)
667 static void wake_oom_reaper(struct task_struct
*tsk
)
673 * mark_oom_victim - mark the given task as OOM victim
676 * Has to be called with oom_lock held and never after
677 * oom has been disabled already.
679 void mark_oom_victim(struct task_struct
*tsk
)
681 WARN_ON(oom_killer_disabled
);
682 /* OOM killer might race with memcg OOM */
683 if (test_and_set_tsk_thread_flag(tsk
, TIF_MEMDIE
))
685 atomic_inc(&tsk
->signal
->oom_victims
);
687 * Make sure that the task is woken up from uninterruptible sleep
688 * if it is frozen because OOM killer wouldn't be able to free
689 * any memory and livelock. freezing_slow_path will tell the freezer
690 * that TIF_MEMDIE tasks should be ignored.
693 atomic_inc(&oom_victims
);
697 * exit_oom_victim - note the exit of an OOM victim
699 void exit_oom_victim(struct task_struct
*tsk
)
701 if (!test_and_clear_tsk_thread_flag(tsk
, TIF_MEMDIE
))
703 atomic_dec(&tsk
->signal
->oom_victims
);
705 if (!atomic_dec_return(&oom_victims
))
706 wake_up_all(&oom_victims_wait
);
710 * oom_killer_disable - disable OOM killer
712 * Forces all page allocations to fail rather than trigger OOM killer.
713 * Will block and wait until all OOM victims are killed.
715 * The function cannot be called when there are runnable user tasks because
716 * the userspace would see unexpected allocation failures as a result. Any
717 * new usage of this function should be consulted with MM people.
719 * Returns true if successful and false if the OOM killer cannot be
722 bool oom_killer_disable(void)
725 * Make sure to not race with an ongoing OOM killer. Check that the
726 * current is not killed (possibly due to sharing the victim's memory).
728 if (mutex_lock_killable(&oom_lock
))
730 oom_killer_disabled
= true;
731 mutex_unlock(&oom_lock
);
733 wait_event(oom_victims_wait
, !atomic_read(&oom_victims
));
739 * oom_killer_enable - enable OOM killer
741 void oom_killer_enable(void)
743 oom_killer_disabled
= false;
747 * Must be called while holding a reference to p, which will be released upon
750 void oom_kill_process(struct oom_control
*oc
, struct task_struct
*p
,
751 unsigned int points
, unsigned long totalpages
,
754 struct task_struct
*victim
= p
;
755 struct task_struct
*child
;
756 struct task_struct
*t
;
757 struct mm_struct
*mm
;
758 unsigned int victim_points
= 0;
759 static DEFINE_RATELIMIT_STATE(oom_rs
, DEFAULT_RATELIMIT_INTERVAL
,
760 DEFAULT_RATELIMIT_BURST
);
761 bool can_oom_reap
= true;
764 * If the task is already exiting, don't alarm the sysadmin or kill
765 * its children or threads, just set TIF_MEMDIE so it can die quickly
768 if (p
->mm
&& task_will_free_mem(p
)) {
777 if (__ratelimit(&oom_rs
))
780 pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n",
781 message
, task_pid_nr(p
), p
->comm
, points
);
784 * If any of p's children has a different mm and is eligible for kill,
785 * the one with the highest oom_badness() score is sacrificed for its
786 * parent. This attempts to lose the minimal amount of work done while
787 * still freeing memory.
789 read_lock(&tasklist_lock
);
790 for_each_thread(p
, t
) {
791 list_for_each_entry(child
, &t
->children
, sibling
) {
792 unsigned int child_points
;
794 if (process_shares_mm(child
, p
->mm
))
797 * oom_badness() returns 0 if the thread is unkillable
799 child_points
= oom_badness(child
,
800 oc
->memcg
, oc
->nodemask
, totalpages
);
801 if (child_points
> victim_points
) {
802 put_task_struct(victim
);
804 victim_points
= child_points
;
805 get_task_struct(victim
);
809 read_unlock(&tasklist_lock
);
811 p
= find_lock_task_mm(victim
);
813 put_task_struct(victim
);
815 } else if (victim
!= p
) {
817 put_task_struct(victim
);
821 /* Get a reference to safely compare mm after task_unlock(victim) */
823 atomic_inc(&mm
->mm_count
);
825 * We should send SIGKILL before setting TIF_MEMDIE in order to prevent
826 * the OOM victim from depleting the memory reserves from the user
827 * space under its control.
829 do_send_sig_info(SIGKILL
, SEND_SIG_FORCED
, victim
, true);
830 mark_oom_victim(victim
);
831 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
832 task_pid_nr(victim
), victim
->comm
, K(victim
->mm
->total_vm
),
833 K(get_mm_counter(victim
->mm
, MM_ANONPAGES
)),
834 K(get_mm_counter(victim
->mm
, MM_FILEPAGES
)),
835 K(get_mm_counter(victim
->mm
, MM_SHMEMPAGES
)));
839 * Kill all user processes sharing victim->mm in other thread groups, if
840 * any. They don't get access to memory reserves, though, to avoid
841 * depletion of all memory. This prevents mm->mmap_sem livelock when an
842 * oom killed thread cannot exit because it requires the semaphore and
843 * its contended by another thread trying to allocate memory itself.
844 * That thread will now get access to memory reserves since it has a
845 * pending fatal signal.
848 for_each_process(p
) {
849 if (!process_shares_mm(p
, mm
))
851 if (same_thread_group(p
, victim
))
853 if (unlikely(p
->flags
& PF_KTHREAD
) || is_global_init(p
) ||
854 p
->signal
->oom_score_adj
== OOM_SCORE_ADJ_MIN
) {
856 * We cannot use oom_reaper for the mm shared by this
857 * process because it wouldn't get killed and so the
858 * memory might be still used.
860 can_oom_reap
= false;
863 do_send_sig_info(SIGKILL
, SEND_SIG_FORCED
, p
, true);
868 wake_oom_reaper(victim
);
871 put_task_struct(victim
);
876 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
878 void check_panic_on_oom(struct oom_control
*oc
, enum oom_constraint constraint
)
880 if (likely(!sysctl_panic_on_oom
))
882 if (sysctl_panic_on_oom
!= 2) {
884 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
885 * does not panic for cpuset, mempolicy, or memcg allocation
888 if (constraint
!= CONSTRAINT_NONE
)
891 /* Do not panic for oom kills triggered by sysrq */
892 if (is_sysrq_oom(oc
))
894 dump_header(oc
, NULL
);
895 panic("Out of memory: %s panic_on_oom is enabled\n",
896 sysctl_panic_on_oom
== 2 ? "compulsory" : "system-wide");
899 static BLOCKING_NOTIFIER_HEAD(oom_notify_list
);
901 int register_oom_notifier(struct notifier_block
*nb
)
903 return blocking_notifier_chain_register(&oom_notify_list
, nb
);
905 EXPORT_SYMBOL_GPL(register_oom_notifier
);
907 int unregister_oom_notifier(struct notifier_block
*nb
)
909 return blocking_notifier_chain_unregister(&oom_notify_list
, nb
);
911 EXPORT_SYMBOL_GPL(unregister_oom_notifier
);
914 * out_of_memory - kill the "best" process when we run out of memory
915 * @oc: pointer to struct oom_control
917 * If we run out of memory, we have the choice between either
918 * killing a random task (bad), letting the system crash (worse)
919 * OR try to be smart about which process to kill. Note that we
920 * don't have to be perfect here, we just have to be good.
922 bool out_of_memory(struct oom_control
*oc
)
924 struct task_struct
*p
;
925 unsigned long totalpages
;
926 unsigned long freed
= 0;
927 unsigned int uninitialized_var(points
);
928 enum oom_constraint constraint
= CONSTRAINT_NONE
;
930 if (oom_killer_disabled
)
933 blocking_notifier_call_chain(&oom_notify_list
, 0, &freed
);
935 /* Got some memory back in the last second. */
939 * If current has a pending SIGKILL or is exiting, then automatically
940 * select it. The goal is to allow it to allocate so that it may
941 * quickly exit and free its memory.
943 * But don't select if current has already released its mm and cleared
944 * TIF_MEMDIE flag at exit_mm(), otherwise an OOM livelock may occur.
947 (fatal_signal_pending(current
) || task_will_free_mem(current
))) {
948 mark_oom_victim(current
);
949 try_oom_reaper(current
);
954 * The OOM killer does not compensate for IO-less reclaim.
955 * pagefault_out_of_memory lost its gfp context so we have to
956 * make sure exclude 0 mask - all other users should have at least
957 * ___GFP_DIRECT_RECLAIM to get here.
959 if (oc
->gfp_mask
&& !(oc
->gfp_mask
& (__GFP_FS
|__GFP_NOFAIL
)))
963 * Check if there were limitations on the allocation (only relevant for
964 * NUMA) that may require different handling.
966 constraint
= constrained_alloc(oc
, &totalpages
);
967 if (constraint
!= CONSTRAINT_MEMORY_POLICY
)
969 check_panic_on_oom(oc
, constraint
);
971 if (sysctl_oom_kill_allocating_task
&& current
->mm
&&
972 !oom_unkillable_task(current
, NULL
, oc
->nodemask
) &&
973 current
->signal
->oom_score_adj
!= OOM_SCORE_ADJ_MIN
) {
974 get_task_struct(current
);
975 oom_kill_process(oc
, current
, 0, totalpages
,
976 "Out of memory (oom_kill_allocating_task)");
980 p
= select_bad_process(oc
, &points
, totalpages
);
981 /* Found nothing?!?! Either we hang forever, or we panic. */
982 if (!p
&& !is_sysrq_oom(oc
)) {
983 dump_header(oc
, NULL
);
984 panic("Out of memory and no killable processes...\n");
986 if (p
&& p
!= (void *)-1UL) {
987 oom_kill_process(oc
, p
, points
, totalpages
, "Out of memory");
989 * Give the killed process a good chance to exit before trying
990 * to allocate memory again.
992 schedule_timeout_killable(1);
998 * The pagefault handler calls here because it is out of memory, so kill a
999 * memory-hogging task. If oom_lock is held by somebody else, a parallel oom
1000 * killing is already in progress so do nothing.
1002 void pagefault_out_of_memory(void)
1004 struct oom_control oc
= {
1012 if (mem_cgroup_oom_synchronize(true))
1015 if (!mutex_trylock(&oom_lock
))
1018 if (!out_of_memory(&oc
)) {
1020 * There shouldn't be any user tasks runnable while the
1021 * OOM killer is disabled, so the current task has to
1022 * be a racing OOM victim for which oom_killer_disable()
1025 WARN_ON(test_thread_flag(TIF_MEMDIE
));
1028 mutex_unlock(&oom_lock
);