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/module.h>
30 #include <linux/notifier.h>
31 #include <linux/memcontrol.h>
32 #include <linux/mempolicy.h>
33 #include <linux/security.h>
35 int sysctl_panic_on_oom;
36 int sysctl_oom_kill_allocating_task;
37 int sysctl_oom_dump_tasks = 1;
38 static DEFINE_SPINLOCK(zone_scan_lock);
42 * has_intersects_mems_allowed() - check task eligiblity for kill
43 * @tsk: task struct of which task to consider
44 * @mask: nodemask passed to page allocator for mempolicy ooms
46 * Task eligibility is determined by whether or not a candidate task, @tsk,
47 * shares the same mempolicy nodes as current if it is bound by such a policy
48 * and whether or not it has the same set of allowed cpuset nodes.
50 static bool has_intersects_mems_allowed(struct task_struct *tsk,
51 const nodemask_t *mask)
53 struct task_struct *start = tsk;
58 * If this is a mempolicy constrained oom, tsk's
59 * cpuset is irrelevant. Only return true if its
60 * mempolicy intersects current, otherwise it may be
63 if (mempolicy_nodemask_intersects(tsk, mask))
67 * This is not a mempolicy constrained oom, so only
68 * check the mems of tsk's cpuset.
70 if (cpuset_mems_allowed_intersects(current, tsk))
73 } while_each_thread(start, tsk);
78 static bool has_intersects_mems_allowed(struct task_struct *tsk,
79 const nodemask_t *mask)
83 #endif /* CONFIG_NUMA */
86 * If this is a system OOM (not a memcg OOM) and the task selected to be
87 * killed is not already running at high (RT) priorities, speed up the
88 * recovery by boosting the dying task to the lowest FIFO priority.
89 * That helps with the recovery and avoids interfering with RT tasks.
91 static void boost_dying_task_prio(struct task_struct *p,
92 struct mem_cgroup *mem)
94 struct sched_param param = { .sched_priority = 1 };
100 sched_setscheduler_nocheck(p, SCHED_FIFO, ¶m);
104 * The process p may have detached its own ->mm while exiting or through
105 * use_mm(), but one or more of its subthreads may still have a valid
106 * pointer. Return p, or any of its subthreads with a valid ->mm, with
109 struct task_struct *find_lock_task_mm(struct task_struct *p)
111 struct task_struct *t = p;
118 } while_each_thread(p, t);
123 /* return true if the task is not adequate as candidate victim task. */
124 static bool oom_unkillable_task(struct task_struct *p,
125 const struct mem_cgroup *mem, const nodemask_t *nodemask)
127 if (is_global_init(p))
129 if (p->flags & PF_KTHREAD)
132 /* When mem_cgroup_out_of_memory() and p is not member of the group */
133 if (mem && !task_in_mem_cgroup(p, mem))
136 /* p may not have freeable memory in nodemask */
137 if (!has_intersects_mems_allowed(p, nodemask))
144 * oom_badness - heuristic function to determine which candidate task to kill
145 * @p: task struct of which task we should calculate
146 * @totalpages: total present RAM allowed for page allocation
148 * The heuristic for determining which task to kill is made to be as simple and
149 * predictable as possible. The goal is to return the highest value for the
150 * task consuming the most memory to avoid subsequent oom failures.
152 unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem,
153 const nodemask_t *nodemask, unsigned long totalpages)
157 if (oom_unkillable_task(p, mem, nodemask))
160 p = find_lock_task_mm(p);
165 * Shortcut check for OOM_SCORE_ADJ_MIN so the entire heuristic doesn't
166 * need to be executed for something that cannot be killed.
168 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
174 * When the PF_OOM_ORIGIN bit is set, it indicates the task should have
175 * priority for oom killing.
177 if (p->flags & PF_OOM_ORIGIN) {
183 * The memory controller may have a limit of 0 bytes, so avoid a divide
184 * by zero, if necessary.
190 * The baseline for the badness score is the proportion of RAM that each
191 * task's rss and swap space use.
193 points = (get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS)) * 1000 /
198 * Root processes get 3% bonus, just like the __vm_enough_memory()
199 * implementation used by LSMs.
201 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
205 * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may
206 * either completely disable oom killing or always prefer a certain
209 points += p->signal->oom_score_adj;
212 * Never return 0 for an eligible task that may be killed since it's
213 * possible that no single user task uses more than 0.1% of memory and
214 * no single admin tasks uses more than 3.0%.
218 return (points < 1000) ? points : 1000;
222 * Determine the type of allocation constraint.
225 static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
226 gfp_t gfp_mask, nodemask_t *nodemask,
227 unsigned long *totalpages)
231 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
232 bool cpuset_limited = false;
235 /* Default to all available memory */
236 *totalpages = totalram_pages + total_swap_pages;
239 return CONSTRAINT_NONE;
241 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
242 * to kill current.We have to random task kill in this case.
243 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
245 if (gfp_mask & __GFP_THISNODE)
246 return CONSTRAINT_NONE;
249 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
250 * the page allocator means a mempolicy is in effect. Cpuset policy
251 * is enforced in get_page_from_freelist().
253 if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) {
254 *totalpages = total_swap_pages;
255 for_each_node_mask(nid, *nodemask)
256 *totalpages += node_spanned_pages(nid);
257 return CONSTRAINT_MEMORY_POLICY;
260 /* Check this allocation failure is caused by cpuset's wall function */
261 for_each_zone_zonelist_nodemask(zone, z, zonelist,
262 high_zoneidx, nodemask)
263 if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
264 cpuset_limited = true;
266 if (cpuset_limited) {
267 *totalpages = total_swap_pages;
268 for_each_node_mask(nid, cpuset_current_mems_allowed)
269 *totalpages += node_spanned_pages(nid);
270 return CONSTRAINT_CPUSET;
272 return CONSTRAINT_NONE;
275 static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
276 gfp_t gfp_mask, nodemask_t *nodemask,
277 unsigned long *totalpages)
279 *totalpages = totalram_pages + total_swap_pages;
280 return CONSTRAINT_NONE;
285 * Simple selection loop. We chose the process with the highest
286 * number of 'points'. We expect the caller will lock the tasklist.
288 * (not docbooked, we don't want this one cluttering up the manual)
290 static struct task_struct *select_bad_process(unsigned int *ppoints,
291 unsigned long totalpages, struct mem_cgroup *mem,
292 const nodemask_t *nodemask)
294 struct task_struct *p;
295 struct task_struct *chosen = NULL;
298 for_each_process(p) {
301 if (oom_unkillable_task(p, mem, nodemask))
305 * This task already has access to memory reserves and is
306 * being killed. Don't allow any other task access to the
309 * Note: this may have a chance of deadlock if it gets
310 * blocked waiting for another task which itself is waiting
311 * for memory. Is there a better alternative?
313 if (test_tsk_thread_flag(p, TIF_MEMDIE))
314 return ERR_PTR(-1UL);
317 * This is in the process of releasing memory so wait for it
318 * to finish before killing some other task by mistake.
320 * However, if p is the current task, we allow the 'kill' to
321 * go ahead if it is exiting: this will simply set TIF_MEMDIE,
322 * which will allow it to gain access to memory reserves in
323 * the process of exiting and releasing its resources.
324 * Otherwise we could get an easy OOM deadlock.
326 if (thread_group_empty(p) && (p->flags & PF_EXITING) && p->mm) {
328 return ERR_PTR(-1UL);
334 points = oom_badness(p, mem, nodemask, totalpages);
335 if (points > *ppoints) {
345 * dump_tasks - dump current memory state of all system tasks
346 * @mem: 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, cpu, oom_adj
353 * value, oom_score_adj value, and name.
355 * Call with tasklist_lock read-locked.
357 static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask)
359 struct task_struct *p;
360 struct task_struct *task;
362 pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n");
363 for_each_process(p) {
364 if (oom_unkillable_task(p, mem, nodemask))
367 task = find_lock_task_mm(p);
370 * This is a kthread or all of p's threads have already
371 * detached their mm's. There's no need to report
372 * them; they can't be oom killed anyway.
377 pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n",
378 task->pid, task_uid(task), task->tgid,
379 task->mm->total_vm, get_mm_rss(task->mm),
380 task_cpu(task), task->signal->oom_adj,
381 task->signal->oom_score_adj, task->comm);
386 static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
387 struct mem_cgroup *mem, const nodemask_t *nodemask)
390 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
391 "oom_adj=%d, oom_score_adj=%d\n",
392 current->comm, gfp_mask, order, current->signal->oom_adj,
393 current->signal->oom_score_adj);
394 cpuset_print_task_mems_allowed(current);
395 task_unlock(current);
397 mem_cgroup_print_oom_info(mem, p);
399 if (sysctl_oom_dump_tasks)
400 dump_tasks(mem, nodemask);
403 #define K(x) ((x) << (PAGE_SHIFT-10))
404 static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem)
406 p = find_lock_task_mm(p);
410 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
411 task_pid_nr(p), p->comm, K(p->mm->total_vm),
412 K(get_mm_counter(p->mm, MM_ANONPAGES)),
413 K(get_mm_counter(p->mm, MM_FILEPAGES)));
417 set_tsk_thread_flag(p, TIF_MEMDIE);
418 force_sig(SIGKILL, p);
421 * We give our sacrificial lamb high priority and access to
422 * all the memory it needs. That way it should be able to
423 * exit() and clear out its resources quickly...
425 boost_dying_task_prio(p, mem);
431 static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
432 unsigned int points, unsigned long totalpages,
433 struct mem_cgroup *mem, nodemask_t *nodemask,
436 struct task_struct *victim = p;
437 struct task_struct *child;
438 struct task_struct *t = p;
439 unsigned int victim_points = 0;
441 if (printk_ratelimit())
442 dump_header(p, gfp_mask, order, mem, nodemask);
445 * If the task is already exiting, don't alarm the sysadmin or kill
446 * its children or threads, just set TIF_MEMDIE so it can die quickly
448 if (p->flags & PF_EXITING) {
449 set_tsk_thread_flag(p, TIF_MEMDIE);
450 boost_dying_task_prio(p, mem);
455 pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
456 message, task_pid_nr(p), p->comm, points);
460 * If any of p's children has a different mm and is eligible for kill,
461 * the one with the highest badness() score is sacrificed for its
462 * parent. This attempts to lose the minimal amount of work done while
463 * still freeing memory.
466 list_for_each_entry(child, &t->children, sibling) {
467 unsigned int child_points;
470 * oom_badness() returns 0 if the thread is unkillable
472 child_points = oom_badness(child, mem, nodemask,
474 if (child_points > victim_points) {
476 victim_points = child_points;
479 } while_each_thread(p, t);
481 return oom_kill_task(victim, mem);
485 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
487 static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
488 int order, const nodemask_t *nodemask)
490 if (likely(!sysctl_panic_on_oom))
492 if (sysctl_panic_on_oom != 2) {
494 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
495 * does not panic for cpuset, mempolicy, or memcg allocation
498 if (constraint != CONSTRAINT_NONE)
501 read_lock(&tasklist_lock);
502 dump_header(NULL, gfp_mask, order, NULL, nodemask);
503 read_unlock(&tasklist_lock);
504 panic("Out of memory: %s panic_on_oom is enabled\n",
505 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
508 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
509 void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
512 unsigned int points = 0;
513 struct task_struct *p;
515 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL);
516 limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT;
517 read_lock(&tasklist_lock);
519 p = select_bad_process(&points, limit, mem, NULL);
520 if (!p || PTR_ERR(p) == -1UL)
523 if (oom_kill_process(p, gfp_mask, 0, points, limit, mem, NULL,
524 "Memory cgroup out of memory"))
527 read_unlock(&tasklist_lock);
531 static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
533 int register_oom_notifier(struct notifier_block *nb)
535 return blocking_notifier_chain_register(&oom_notify_list, nb);
537 EXPORT_SYMBOL_GPL(register_oom_notifier);
539 int unregister_oom_notifier(struct notifier_block *nb)
541 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
543 EXPORT_SYMBOL_GPL(unregister_oom_notifier);
546 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
547 * if a parallel OOM killing is already taking place that includes a zone in
548 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
550 int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
556 spin_lock(&zone_scan_lock);
557 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
558 if (zone_is_oom_locked(zone)) {
564 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
566 * Lock each zone in the zonelist under zone_scan_lock so a
567 * parallel invocation of try_set_zonelist_oom() doesn't succeed
570 zone_set_flag(zone, ZONE_OOM_LOCKED);
574 spin_unlock(&zone_scan_lock);
579 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
580 * allocation attempts with zonelists containing them may now recall the OOM
581 * killer, if necessary.
583 void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
588 spin_lock(&zone_scan_lock);
589 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
590 zone_clear_flag(zone, ZONE_OOM_LOCKED);
592 spin_unlock(&zone_scan_lock);
596 * Try to acquire the oom killer lock for all system zones. Returns zero if a
597 * parallel oom killing is taking place, otherwise locks all zones and returns
600 static int try_set_system_oom(void)
605 spin_lock(&zone_scan_lock);
606 for_each_populated_zone(zone)
607 if (zone_is_oom_locked(zone)) {
611 for_each_populated_zone(zone)
612 zone_set_flag(zone, ZONE_OOM_LOCKED);
614 spin_unlock(&zone_scan_lock);
619 * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation
620 * attempts or page faults may now recall the oom killer, if necessary.
622 static void clear_system_oom(void)
626 spin_lock(&zone_scan_lock);
627 for_each_populated_zone(zone)
628 zone_clear_flag(zone, ZONE_OOM_LOCKED);
629 spin_unlock(&zone_scan_lock);
633 * out_of_memory - kill the "best" process when we run out of memory
634 * @zonelist: zonelist pointer
635 * @gfp_mask: memory allocation flags
636 * @order: amount of memory being requested as a power of 2
637 * @nodemask: nodemask passed to page allocator
639 * If we run out of memory, we have the choice between either
640 * killing a random task (bad), letting the system crash (worse)
641 * OR try to be smart about which process to kill. Note that we
642 * don't have to be perfect here, we just have to be good.
644 void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
645 int order, nodemask_t *nodemask)
647 const nodemask_t *mpol_mask;
648 struct task_struct *p;
649 unsigned long totalpages;
650 unsigned long freed = 0;
652 enum oom_constraint constraint = CONSTRAINT_NONE;
655 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
657 /* Got some memory back in the last second. */
661 * If current has a pending SIGKILL, then automatically select it. The
662 * goal is to allow it to allocate so that it may quickly exit and free
665 if (fatal_signal_pending(current)) {
666 set_thread_flag(TIF_MEMDIE);
667 boost_dying_task_prio(current, NULL);
672 * Check if there were limitations on the allocation (only relevant for
673 * NUMA) that may require different handling.
675 constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
677 mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
678 check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
680 read_lock(&tasklist_lock);
681 if (sysctl_oom_kill_allocating_task &&
682 !oom_unkillable_task(current, NULL, nodemask) &&
683 (current->signal->oom_adj != OOM_DISABLE)) {
685 * oom_kill_process() needs tasklist_lock held. If it returns
686 * non-zero, current could not be killed so we must fallback to
689 if (!oom_kill_process(current, gfp_mask, order, 0, totalpages,
691 "Out of memory (oom_kill_allocating_task)"))
696 p = select_bad_process(&points, totalpages, NULL, mpol_mask);
697 if (PTR_ERR(p) == -1UL)
700 /* Found nothing?!?! Either we hang forever, or we panic. */
702 dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
703 read_unlock(&tasklist_lock);
704 panic("Out of memory and no killable processes...\n");
707 if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
708 nodemask, "Out of memory"))
712 read_unlock(&tasklist_lock);
715 * Give "p" a good chance of killing itself before we
716 * retry to allocate memory unless "p" is current
718 if (killed && !test_thread_flag(TIF_MEMDIE))
719 schedule_timeout_uninterruptible(1);
723 * The pagefault handler calls here because it is out of memory, so kill a
724 * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel
725 * oom killing is already in progress so do nothing. If a task is found with
726 * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit.
728 void pagefault_out_of_memory(void)
730 if (try_set_system_oom()) {
731 out_of_memory(NULL, 0, 0, NULL);
734 if (!test_thread_flag(TIF_MEMDIE))
735 schedule_timeout_uninterruptible(1);