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>
39 #define CREATE_TRACE_POINTS
40 #include <trace/events/oom.h>
42 int sysctl_panic_on_oom;
43 int sysctl_oom_kill_allocating_task;
44 int sysctl_oom_dump_tasks = 1;
46 DEFINE_MUTEX(oom_lock);
50 * has_intersects_mems_allowed() - check task eligiblity for kill
51 * @start: task struct of which task to consider
52 * @mask: nodemask passed to page allocator for mempolicy ooms
54 * Task eligibility is determined by whether or not a candidate task, @tsk,
55 * shares the same mempolicy nodes as current if it is bound by such a policy
56 * and whether or not it has the same set of allowed cpuset nodes.
58 static bool has_intersects_mems_allowed(struct task_struct *start,
59 const nodemask_t *mask)
61 struct task_struct *tsk;
65 for_each_thread(start, tsk) {
68 * If this is a mempolicy constrained oom, tsk's
69 * cpuset is irrelevant. Only return true if its
70 * mempolicy intersects current, otherwise it may be
73 ret = mempolicy_nodemask_intersects(tsk, mask);
76 * This is not a mempolicy constrained oom, so only
77 * check the mems of tsk's cpuset.
79 ret = cpuset_mems_allowed_intersects(current, tsk);
89 static bool has_intersects_mems_allowed(struct task_struct *tsk,
90 const nodemask_t *mask)
94 #endif /* CONFIG_NUMA */
97 * The process p may have detached its own ->mm while exiting or through
98 * use_mm(), but one or more of its subthreads may still have a valid
99 * pointer. Return p, or any of its subthreads with a valid ->mm, with
102 struct task_struct *find_lock_task_mm(struct task_struct *p)
104 struct task_struct *t;
108 for_each_thread(p, t) {
121 /* return true if the task is not adequate as candidate victim task. */
122 static bool oom_unkillable_task(struct task_struct *p,
123 struct mem_cgroup *memcg, const nodemask_t *nodemask)
125 if (is_global_init(p))
127 if (p->flags & PF_KTHREAD)
130 /* When mem_cgroup_out_of_memory() and p is not member of the group */
131 if (memcg && !task_in_mem_cgroup(p, memcg))
134 /* p may not have freeable memory in nodemask */
135 if (!has_intersects_mems_allowed(p, nodemask))
142 * oom_badness - heuristic function to determine which candidate task to kill
143 * @p: task struct of which task we should calculate
144 * @totalpages: total present RAM allowed for page allocation
146 * The heuristic for determining which task to kill is made to be as simple and
147 * predictable as possible. The goal is to return the highest value for the
148 * task consuming the most memory to avoid subsequent oom failures.
150 unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
151 const nodemask_t *nodemask, unsigned long totalpages)
156 if (oom_unkillable_task(p, memcg, nodemask))
159 p = find_lock_task_mm(p);
163 adj = (long)p->signal->oom_score_adj;
164 if (adj == OOM_SCORE_ADJ_MIN) {
170 * The baseline for the badness score is the proportion of RAM that each
171 * task's rss, pagetable and swap space use.
173 points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
174 atomic_long_read(&p->mm->nr_ptes) + mm_nr_pmds(p->mm);
178 * Root processes get 3% bonus, just like the __vm_enough_memory()
179 * implementation used by LSMs.
181 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
182 points -= (points * 3) / 100;
184 /* Normalize to oom_score_adj units */
185 adj *= totalpages / 1000;
189 * Never return 0 for an eligible task regardless of the root bonus and
190 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
192 return points > 0 ? points : 1;
196 * Determine the type of allocation constraint.
199 static enum oom_constraint constrained_alloc(struct oom_control *oc,
200 unsigned long *totalpages)
204 enum zone_type high_zoneidx = gfp_zone(oc->gfp_mask);
205 bool cpuset_limited = false;
208 /* Default to all available memory */
209 *totalpages = totalram_pages + total_swap_pages;
212 return CONSTRAINT_NONE;
214 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
215 * to kill current.We have to random task kill in this case.
216 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
218 if (oc->gfp_mask & __GFP_THISNODE)
219 return CONSTRAINT_NONE;
222 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
223 * the page allocator means a mempolicy is in effect. Cpuset policy
224 * is enforced in get_page_from_freelist().
227 !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) {
228 *totalpages = total_swap_pages;
229 for_each_node_mask(nid, *oc->nodemask)
230 *totalpages += node_spanned_pages(nid);
231 return CONSTRAINT_MEMORY_POLICY;
234 /* Check this allocation failure is caused by cpuset's wall function */
235 for_each_zone_zonelist_nodemask(zone, z, oc->zonelist,
236 high_zoneidx, oc->nodemask)
237 if (!cpuset_zone_allowed(zone, oc->gfp_mask))
238 cpuset_limited = true;
240 if (cpuset_limited) {
241 *totalpages = total_swap_pages;
242 for_each_node_mask(nid, cpuset_current_mems_allowed)
243 *totalpages += node_spanned_pages(nid);
244 return CONSTRAINT_CPUSET;
246 return CONSTRAINT_NONE;
249 static enum oom_constraint constrained_alloc(struct oom_control *oc,
250 unsigned long *totalpages)
252 *totalpages = totalram_pages + total_swap_pages;
253 return CONSTRAINT_NONE;
257 enum oom_scan_t oom_scan_process_thread(struct oom_control *oc,
258 struct task_struct *task, unsigned long totalpages)
260 if (oom_unkillable_task(task, NULL, oc->nodemask))
261 return OOM_SCAN_CONTINUE;
264 * This task already has access to memory reserves and is being killed.
265 * Don't allow any other task to have access to the reserves.
267 if (test_tsk_thread_flag(task, TIF_MEMDIE)) {
269 return OOM_SCAN_ABORT;
272 return OOM_SCAN_CONTINUE;
275 * If task is allocating a lot of memory and has been marked to be
276 * killed first if it triggers an oom, then select it.
278 if (oom_task_origin(task))
279 return OOM_SCAN_SELECT;
281 if (task_will_free_mem(task) && oc->order != -1)
282 return OOM_SCAN_ABORT;
288 * Simple selection loop. We chose the process with the highest
289 * number of 'points'. Returns -1 on scan abort.
291 static struct task_struct *select_bad_process(struct oom_control *oc,
292 unsigned int *ppoints, unsigned long totalpages)
294 struct task_struct *g, *p;
295 struct task_struct *chosen = NULL;
296 unsigned long chosen_points = 0;
299 for_each_process_thread(g, p) {
302 switch (oom_scan_process_thread(oc, p, totalpages)) {
303 case OOM_SCAN_SELECT:
305 chosen_points = ULONG_MAX;
307 case OOM_SCAN_CONTINUE:
311 return (struct task_struct *)(-1UL);
315 points = oom_badness(p, NULL, oc->nodemask, totalpages);
316 if (!points || points < chosen_points)
318 /* Prefer thread group leaders for display purposes */
319 if (points == chosen_points && thread_group_leader(chosen))
323 chosen_points = points;
326 get_task_struct(chosen);
329 *ppoints = chosen_points * 1000 / totalpages;
334 * dump_tasks - dump current memory state of all system tasks
335 * @memcg: current's memory controller, if constrained
336 * @nodemask: nodemask passed to page allocator for mempolicy ooms
338 * Dumps the current memory state of all eligible tasks. Tasks not in the same
339 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
341 * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
342 * swapents, oom_score_adj value, and name.
344 static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask)
346 struct task_struct *p;
347 struct task_struct *task;
349 pr_info("[ pid ] uid tgid total_vm rss nr_ptes nr_pmds swapents oom_score_adj name\n");
351 for_each_process(p) {
352 if (oom_unkillable_task(p, memcg, nodemask))
355 task = find_lock_task_mm(p);
358 * This is a kthread or all of p's threads have already
359 * detached their mm's. There's no need to report
360 * them; they can't be oom killed anyway.
365 pr_info("[%5d] %5d %5d %8lu %8lu %7ld %7ld %8lu %5hd %s\n",
366 task->pid, from_kuid(&init_user_ns, task_uid(task)),
367 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
368 atomic_long_read(&task->mm->nr_ptes),
369 mm_nr_pmds(task->mm),
370 get_mm_counter(task->mm, MM_SWAPENTS),
371 task->signal->oom_score_adj, task->comm);
377 static void dump_header(struct oom_control *oc, struct task_struct *p,
378 struct mem_cgroup *memcg)
381 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
382 "oom_score_adj=%hd\n",
383 current->comm, oc->gfp_mask, oc->order,
384 current->signal->oom_score_adj);
385 cpuset_print_task_mems_allowed(current);
386 task_unlock(current);
389 mem_cgroup_print_oom_info(memcg, p);
391 show_mem(SHOW_MEM_FILTER_NODES);
392 if (sysctl_oom_dump_tasks)
393 dump_tasks(memcg, oc->nodemask);
397 * Number of OOM victims in flight
399 static atomic_t oom_victims = ATOMIC_INIT(0);
400 static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
402 bool oom_killer_disabled __read_mostly;
405 * mark_oom_victim - mark the given task as OOM victim
408 * Has to be called with oom_lock held and never after
409 * oom has been disabled already.
411 void mark_oom_victim(struct task_struct *tsk)
413 WARN_ON(oom_killer_disabled);
414 /* OOM killer might race with memcg OOM */
415 if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
418 * Make sure that the task is woken up from uninterruptible sleep
419 * if it is frozen because OOM killer wouldn't be able to free
420 * any memory and livelock. freezing_slow_path will tell the freezer
421 * that TIF_MEMDIE tasks should be ignored.
424 atomic_inc(&oom_victims);
428 * exit_oom_victim - note the exit of an OOM victim
430 void exit_oom_victim(void)
432 clear_thread_flag(TIF_MEMDIE);
434 if (!atomic_dec_return(&oom_victims))
435 wake_up_all(&oom_victims_wait);
439 * oom_killer_disable - disable OOM killer
441 * Forces all page allocations to fail rather than trigger OOM killer.
442 * Will block and wait until all OOM victims are killed.
444 * The function cannot be called when there are runnable user tasks because
445 * the userspace would see unexpected allocation failures as a result. Any
446 * new usage of this function should be consulted with MM people.
448 * Returns true if successful and false if the OOM killer cannot be
451 bool oom_killer_disable(void)
454 * Make sure to not race with an ongoing OOM killer
455 * and that the current is not the victim.
457 mutex_lock(&oom_lock);
458 if (test_thread_flag(TIF_MEMDIE)) {
459 mutex_unlock(&oom_lock);
463 oom_killer_disabled = true;
464 mutex_unlock(&oom_lock);
466 wait_event(oom_victims_wait, !atomic_read(&oom_victims));
472 * oom_killer_enable - enable OOM killer
474 void oom_killer_enable(void)
476 oom_killer_disabled = false;
479 #define K(x) ((x) << (PAGE_SHIFT-10))
481 * Must be called while holding a reference to p, which will be released upon
484 void oom_kill_process(struct oom_control *oc, struct task_struct *p,
485 unsigned int points, unsigned long totalpages,
486 struct mem_cgroup *memcg, const char *message)
488 struct task_struct *victim = p;
489 struct task_struct *child;
490 struct task_struct *t;
491 struct mm_struct *mm;
492 unsigned int victim_points = 0;
493 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
494 DEFAULT_RATELIMIT_BURST);
497 * If the task is already exiting, don't alarm the sysadmin or kill
498 * its children or threads, just set TIF_MEMDIE so it can die quickly
501 if (p->mm && task_will_free_mem(p)) {
509 if (__ratelimit(&oom_rs))
510 dump_header(oc, p, memcg);
513 pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n",
514 message, task_pid_nr(p), p->comm, points);
518 * If any of p's children has a different mm and is eligible for kill,
519 * the one with the highest oom_badness() score is sacrificed for its
520 * parent. This attempts to lose the minimal amount of work done while
521 * still freeing memory.
523 read_lock(&tasklist_lock);
524 for_each_thread(p, t) {
525 list_for_each_entry(child, &t->children, sibling) {
526 unsigned int child_points;
528 if (child->mm == p->mm)
531 * oom_badness() returns 0 if the thread is unkillable
533 child_points = oom_badness(child, memcg, oc->nodemask,
535 if (child_points > victim_points) {
536 put_task_struct(victim);
538 victim_points = child_points;
539 get_task_struct(victim);
543 read_unlock(&tasklist_lock);
545 p = find_lock_task_mm(victim);
547 put_task_struct(victim);
549 } else if (victim != p) {
551 put_task_struct(victim);
555 /* mm cannot safely be dereferenced after task_unlock(victim) */
557 mark_oom_victim(victim);
558 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
559 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
560 K(get_mm_counter(victim->mm, MM_ANONPAGES)),
561 K(get_mm_counter(victim->mm, MM_FILEPAGES)));
565 * Kill all user processes sharing victim->mm in other thread groups, if
566 * any. They don't get access to memory reserves, though, to avoid
567 * depletion of all memory. This prevents mm->mmap_sem livelock when an
568 * oom killed thread cannot exit because it requires the semaphore and
569 * its contended by another thread trying to allocate memory itself.
570 * That thread will now get access to memory reserves since it has a
571 * pending fatal signal.
575 if (p->mm == mm && !same_thread_group(p, victim) &&
576 !(p->flags & PF_KTHREAD)) {
577 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
580 task_lock(p); /* Protect ->comm from prctl() */
581 pr_err("Kill process %d (%s) sharing same memory\n",
582 task_pid_nr(p), p->comm);
584 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
588 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
589 put_task_struct(victim);
594 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
596 void check_panic_on_oom(struct oom_control *oc, enum oom_constraint constraint,
597 struct mem_cgroup *memcg)
599 if (likely(!sysctl_panic_on_oom))
601 if (sysctl_panic_on_oom != 2) {
603 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
604 * does not panic for cpuset, mempolicy, or memcg allocation
607 if (constraint != CONSTRAINT_NONE)
610 /* Do not panic for oom kills triggered by sysrq */
613 dump_header(oc, NULL, memcg);
614 panic("Out of memory: %s panic_on_oom is enabled\n",
615 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
618 static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
620 int register_oom_notifier(struct notifier_block *nb)
622 return blocking_notifier_chain_register(&oom_notify_list, nb);
624 EXPORT_SYMBOL_GPL(register_oom_notifier);
626 int unregister_oom_notifier(struct notifier_block *nb)
628 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
630 EXPORT_SYMBOL_GPL(unregister_oom_notifier);
633 * out_of_memory - kill the "best" process when we run out of memory
634 * @oc: pointer to struct oom_control
636 * If we run out of memory, we have the choice between either
637 * killing a random task (bad), letting the system crash (worse)
638 * OR try to be smart about which process to kill. Note that we
639 * don't have to be perfect here, we just have to be good.
641 bool out_of_memory(struct oom_control *oc)
643 struct task_struct *p;
644 unsigned long totalpages;
645 unsigned long freed = 0;
646 unsigned int uninitialized_var(points);
647 enum oom_constraint constraint = CONSTRAINT_NONE;
649 if (oom_killer_disabled)
652 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
654 /* Got some memory back in the last second. */
658 * If current has a pending SIGKILL or is exiting, then automatically
659 * select it. The goal is to allow it to allocate so that it may
660 * quickly exit and free its memory.
662 * But don't select if current has already released its mm and cleared
663 * TIF_MEMDIE flag at exit_mm(), otherwise an OOM livelock may occur.
666 (fatal_signal_pending(current) || task_will_free_mem(current))) {
667 mark_oom_victim(current);
672 * Check if there were limitations on the allocation (only relevant for
673 * NUMA) that may require different handling.
675 constraint = constrained_alloc(oc, &totalpages);
676 if (constraint != CONSTRAINT_MEMORY_POLICY)
678 check_panic_on_oom(oc, constraint, NULL);
680 if (sysctl_oom_kill_allocating_task && current->mm &&
681 !oom_unkillable_task(current, NULL, oc->nodemask) &&
682 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
683 get_task_struct(current);
684 oom_kill_process(oc, current, 0, totalpages, NULL,
685 "Out of memory (oom_kill_allocating_task)");
689 p = select_bad_process(oc, &points, totalpages);
690 /* Found nothing?!?! Either we hang forever, or we panic. */
691 if (!p && oc->order != -1) {
692 dump_header(oc, NULL, NULL);
693 panic("Out of memory and no killable processes...\n");
695 if (p && p != (void *)-1UL) {
696 oom_kill_process(oc, p, points, totalpages, NULL,
699 * Give the killed process a good chance to exit before trying
700 * to allocate memory again.
702 schedule_timeout_killable(1);
708 * The pagefault handler calls here because it is out of memory, so kill a
709 * memory-hogging task. If any populated zone has ZONE_OOM_LOCKED set, a
710 * parallel oom killing is already in progress so do nothing.
712 void pagefault_out_of_memory(void)
714 struct oom_control oc = {
721 if (mem_cgroup_oom_synchronize(true))
724 if (!mutex_trylock(&oom_lock))
727 if (!out_of_memory(&oc)) {
729 * There shouldn't be any user tasks runnable while the
730 * OOM killer is disabled, so the current task has to
731 * be a racing OOM victim for which oom_killer_disable()
734 WARN_ON(test_thread_flag(TIF_MEMDIE));
737 mutex_unlock(&oom_lock);