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>
37 int sysctl_panic_on_oom;
38 int sysctl_oom_kill_allocating_task;
39 int sysctl_oom_dump_tasks = 1;
40 static DEFINE_SPINLOCK(zone_scan_lock);
43 * compare_swap_oom_score_adj() - compare and swap current's oom_score_adj
44 * @old_val: old oom_score_adj for compare
45 * @new_val: new oom_score_adj for swap
47 * Sets the oom_score_adj value for current to @new_val iff its present value is
48 * @old_val. Usually used to reinstate a previous value to prevent racing with
49 * userspacing tuning the value in the interim.
51 void compare_swap_oom_score_adj(int old_val, int new_val)
53 struct sighand_struct *sighand = current->sighand;
55 spin_lock_irq(&sighand->siglock);
56 if (current->signal->oom_score_adj == old_val)
57 current->signal->oom_score_adj = new_val;
58 spin_unlock_irq(&sighand->siglock);
62 * test_set_oom_score_adj() - set current's oom_score_adj and return old value
63 * @new_val: new oom_score_adj value
65 * Sets the oom_score_adj value for current to @new_val with proper
66 * synchronization and returns the old value. Usually used to temporarily
67 * set a value, save the old value in the caller, and then reinstate it later.
69 int test_set_oom_score_adj(int new_val)
71 struct sighand_struct *sighand = current->sighand;
74 spin_lock_irq(&sighand->siglock);
75 old_val = current->signal->oom_score_adj;
76 current->signal->oom_score_adj = new_val;
77 spin_unlock_irq(&sighand->siglock);
84 * has_intersects_mems_allowed() - check task eligiblity for kill
85 * @tsk: task struct of which task to consider
86 * @mask: nodemask passed to page allocator for mempolicy ooms
88 * Task eligibility is determined by whether or not a candidate task, @tsk,
89 * shares the same mempolicy nodes as current if it is bound by such a policy
90 * and whether or not it has the same set of allowed cpuset nodes.
92 static bool has_intersects_mems_allowed(struct task_struct *tsk,
93 const nodemask_t *mask)
95 struct task_struct *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 if (mempolicy_nodemask_intersects(tsk, mask))
109 * This is not a mempolicy constrained oom, so only
110 * check the mems of tsk's cpuset.
112 if (cpuset_mems_allowed_intersects(current, tsk))
115 } while_each_thread(start, tsk);
120 static bool has_intersects_mems_allowed(struct task_struct *tsk,
121 const nodemask_t *mask)
125 #endif /* CONFIG_NUMA */
128 * The process p may have detached its own ->mm while exiting or through
129 * 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 = p;
142 } while_each_thread(p, t);
147 /* return true if the task is not adequate as candidate victim task. */
148 static bool oom_unkillable_task(struct task_struct *p,
149 const struct mem_cgroup *mem, const nodemask_t *nodemask)
151 if (is_global_init(p))
153 if (p->flags & PF_KTHREAD)
156 /* When mem_cgroup_out_of_memory() and p is not member of the group */
157 if (mem && !task_in_mem_cgroup(p, mem))
160 /* p may not have freeable memory in nodemask */
161 if (!has_intersects_mems_allowed(p, nodemask))
168 * oom_badness - heuristic function to determine which candidate task to kill
169 * @p: task struct of which task we should calculate
170 * @totalpages: total present RAM allowed for page allocation
172 * The heuristic for determining which task to kill is made to be as simple and
173 * predictable as possible. The goal is to return the highest value for the
174 * task consuming the most memory to avoid subsequent oom failures.
176 unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem,
177 const nodemask_t *nodemask, unsigned long totalpages)
181 if (oom_unkillable_task(p, mem, nodemask))
184 p = find_lock_task_mm(p);
188 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
194 * The memory controller may have a limit of 0 bytes, so avoid a divide
195 * by zero, if necessary.
201 * The baseline for the badness score is the proportion of RAM that each
202 * task's rss, pagetable and swap space use.
204 points = get_mm_rss(p->mm) + p->mm->nr_ptes;
205 points += get_mm_counter(p->mm, MM_SWAPENTS);
208 points /= totalpages;
212 * Root processes get 3% bonus, just like the __vm_enough_memory()
213 * implementation used by LSMs.
215 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
219 * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may
220 * either completely disable oom killing or always prefer a certain
223 points += p->signal->oom_score_adj;
226 * Never return 0 for an eligible task that may be killed since it's
227 * possible that no single user task uses more than 0.1% of memory and
228 * no single admin tasks uses more than 3.0%.
232 return (points < 1000) ? points : 1000;
236 * Determine the type of allocation constraint.
239 static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
240 gfp_t gfp_mask, nodemask_t *nodemask,
241 unsigned long *totalpages)
245 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
246 bool cpuset_limited = false;
249 /* Default to all available memory */
250 *totalpages = totalram_pages + total_swap_pages;
253 return CONSTRAINT_NONE;
255 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
256 * to kill current.We have to random task kill in this case.
257 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
259 if (gfp_mask & __GFP_THISNODE)
260 return CONSTRAINT_NONE;
263 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
264 * the page allocator means a mempolicy is in effect. Cpuset policy
265 * is enforced in get_page_from_freelist().
267 if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) {
268 *totalpages = total_swap_pages;
269 for_each_node_mask(nid, *nodemask)
270 *totalpages += node_spanned_pages(nid);
271 return CONSTRAINT_MEMORY_POLICY;
274 /* Check this allocation failure is caused by cpuset's wall function */
275 for_each_zone_zonelist_nodemask(zone, z, zonelist,
276 high_zoneidx, nodemask)
277 if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
278 cpuset_limited = true;
280 if (cpuset_limited) {
281 *totalpages = total_swap_pages;
282 for_each_node_mask(nid, cpuset_current_mems_allowed)
283 *totalpages += node_spanned_pages(nid);
284 return CONSTRAINT_CPUSET;
286 return CONSTRAINT_NONE;
289 static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
290 gfp_t gfp_mask, nodemask_t *nodemask,
291 unsigned long *totalpages)
293 *totalpages = totalram_pages + total_swap_pages;
294 return CONSTRAINT_NONE;
299 * Simple selection loop. We chose the process with the highest
300 * number of 'points'. We expect the caller will lock the tasklist.
302 * (not docbooked, we don't want this one cluttering up the manual)
304 static struct task_struct *select_bad_process(unsigned int *ppoints,
305 unsigned long totalpages, struct mem_cgroup *mem,
306 const nodemask_t *nodemask)
308 struct task_struct *g, *p;
309 struct task_struct *chosen = NULL;
312 do_each_thread(g, p) {
317 if (oom_unkillable_task(p, mem, nodemask))
321 * This task already has access to memory reserves and is
322 * being killed. Don't allow any other task access to the
325 * Note: this may have a chance of deadlock if it gets
326 * blocked waiting for another task which itself is waiting
327 * for memory. Is there a better alternative?
329 if (test_tsk_thread_flag(p, TIF_MEMDIE)) {
330 if (unlikely(frozen(p)))
332 return ERR_PTR(-1UL);
337 if (p->flags & PF_EXITING) {
339 * If p is the current task and is in the process of
340 * releasing memory, we allow the "kill" to set
341 * TIF_MEMDIE, which will allow it to gain access to
342 * memory reserves. Otherwise, it may stall forever.
344 * The loop isn't broken here, however, in case other
345 * threads are found to have already been oom killed.
352 * If this task is not being ptraced on exit,
353 * then wait for it to finish before killing
354 * some other task unnecessarily.
356 if (!(p->group_leader->ptrace & PT_TRACE_EXIT))
357 return ERR_PTR(-1UL);
361 points = oom_badness(p, mem, nodemask, totalpages);
362 if (points > *ppoints) {
366 } while_each_thread(g, p);
372 * dump_tasks - dump current memory state of all system tasks
373 * @mem: current's memory controller, if constrained
374 * @nodemask: nodemask passed to page allocator for mempolicy ooms
376 * Dumps the current memory state of all eligible tasks. Tasks not in the same
377 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
379 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
380 * value, oom_score_adj value, and name.
382 * Call with tasklist_lock read-locked.
384 static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask)
386 struct task_struct *p;
387 struct task_struct *task;
389 pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n");
390 for_each_process(p) {
391 if (oom_unkillable_task(p, mem, nodemask))
394 task = find_lock_task_mm(p);
397 * This is a kthread or all of p's threads have already
398 * detached their mm's. There's no need to report
399 * them; they can't be oom killed anyway.
404 pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n",
405 task->pid, task_uid(task), task->tgid,
406 task->mm->total_vm, get_mm_rss(task->mm),
407 task_cpu(task), task->signal->oom_adj,
408 task->signal->oom_score_adj, task->comm);
413 static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
414 struct mem_cgroup *mem, const nodemask_t *nodemask)
417 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
418 "oom_adj=%d, oom_score_adj=%d\n",
419 current->comm, gfp_mask, order, current->signal->oom_adj,
420 current->signal->oom_score_adj);
421 cpuset_print_task_mems_allowed(current);
422 task_unlock(current);
424 mem_cgroup_print_oom_info(mem, p);
425 show_mem(SHOW_MEM_FILTER_NODES);
426 if (sysctl_oom_dump_tasks)
427 dump_tasks(mem, nodemask);
430 #define K(x) ((x) << (PAGE_SHIFT-10))
431 static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem)
433 struct task_struct *q;
434 struct mm_struct *mm;
436 p = find_lock_task_mm(p);
440 /* mm cannot be safely dereferenced after task_unlock(p) */
443 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
444 task_pid_nr(p), p->comm, K(p->mm->total_vm),
445 K(get_mm_counter(p->mm, MM_ANONPAGES)),
446 K(get_mm_counter(p->mm, MM_FILEPAGES)));
450 * Kill all user processes sharing p->mm in other thread groups, if any.
451 * They don't get access to memory reserves or a higher scheduler
452 * priority, though, to avoid depletion of all memory or task
453 * starvation. This prevents mm->mmap_sem livelock when an oom killed
454 * task cannot exit because it requires the semaphore and its contended
455 * by another thread trying to allocate memory itself. That thread will
456 * now get access to memory reserves since it has a pending fatal
460 if (q->mm == mm && !same_thread_group(q, p) &&
461 !(q->flags & PF_KTHREAD)) {
462 if (q->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
465 task_lock(q); /* Protect ->comm from prctl() */
466 pr_err("Kill process %d (%s) sharing same memory\n",
467 task_pid_nr(q), q->comm);
469 force_sig(SIGKILL, q);
472 set_tsk_thread_flag(p, TIF_MEMDIE);
473 force_sig(SIGKILL, p);
479 static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
480 unsigned int points, unsigned long totalpages,
481 struct mem_cgroup *mem, nodemask_t *nodemask,
484 struct task_struct *victim = p;
485 struct task_struct *child;
486 struct task_struct *t = p;
487 unsigned int victim_points = 0;
489 if (printk_ratelimit())
490 dump_header(p, gfp_mask, order, mem, nodemask);
493 * If the task is already exiting, don't alarm the sysadmin or kill
494 * its children or threads, just set TIF_MEMDIE so it can die quickly
496 if (p->flags & PF_EXITING) {
497 set_tsk_thread_flag(p, TIF_MEMDIE);
502 pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
503 message, task_pid_nr(p), p->comm, points);
507 * If any of p's children has a different mm and is eligible for kill,
508 * the one with the highest oom_badness() score is sacrificed for its
509 * parent. This attempts to lose the minimal amount of work done while
510 * still freeing memory.
513 list_for_each_entry(child, &t->children, sibling) {
514 unsigned int child_points;
516 if (child->mm == p->mm)
519 * oom_badness() returns 0 if the thread is unkillable
521 child_points = oom_badness(child, mem, nodemask,
523 if (child_points > victim_points) {
525 victim_points = child_points;
528 } while_each_thread(p, t);
530 return oom_kill_task(victim, mem);
534 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
536 static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
537 int order, const nodemask_t *nodemask)
539 if (likely(!sysctl_panic_on_oom))
541 if (sysctl_panic_on_oom != 2) {
543 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
544 * does not panic for cpuset, mempolicy, or memcg allocation
547 if (constraint != CONSTRAINT_NONE)
550 read_lock(&tasklist_lock);
551 dump_header(NULL, gfp_mask, order, NULL, nodemask);
552 read_unlock(&tasklist_lock);
553 panic("Out of memory: %s panic_on_oom is enabled\n",
554 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
557 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
558 void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
561 unsigned int points = 0;
562 struct task_struct *p;
565 * If current has a pending SIGKILL, then automatically select it. The
566 * goal is to allow it to allocate so that it may quickly exit and free
569 if (fatal_signal_pending(current)) {
570 set_thread_flag(TIF_MEMDIE);
574 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL);
575 limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT;
576 read_lock(&tasklist_lock);
578 p = select_bad_process(&points, limit, mem, NULL);
579 if (!p || PTR_ERR(p) == -1UL)
582 if (oom_kill_process(p, gfp_mask, 0, points, limit, mem, NULL,
583 "Memory cgroup out of memory"))
586 read_unlock(&tasklist_lock);
590 static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
592 int register_oom_notifier(struct notifier_block *nb)
594 return blocking_notifier_chain_register(&oom_notify_list, nb);
596 EXPORT_SYMBOL_GPL(register_oom_notifier);
598 int unregister_oom_notifier(struct notifier_block *nb)
600 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
602 EXPORT_SYMBOL_GPL(unregister_oom_notifier);
605 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
606 * if a parallel OOM killing is already taking place that includes a zone in
607 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
609 int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
615 spin_lock(&zone_scan_lock);
616 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
617 if (zone_is_oom_locked(zone)) {
623 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
625 * Lock each zone in the zonelist under zone_scan_lock so a
626 * parallel invocation of try_set_zonelist_oom() doesn't succeed
629 zone_set_flag(zone, ZONE_OOM_LOCKED);
633 spin_unlock(&zone_scan_lock);
638 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
639 * allocation attempts with zonelists containing them may now recall the OOM
640 * killer, if necessary.
642 void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
647 spin_lock(&zone_scan_lock);
648 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
649 zone_clear_flag(zone, ZONE_OOM_LOCKED);
651 spin_unlock(&zone_scan_lock);
655 * Try to acquire the oom killer lock for all system zones. Returns zero if a
656 * parallel oom killing is taking place, otherwise locks all zones and returns
659 static int try_set_system_oom(void)
664 spin_lock(&zone_scan_lock);
665 for_each_populated_zone(zone)
666 if (zone_is_oom_locked(zone)) {
670 for_each_populated_zone(zone)
671 zone_set_flag(zone, ZONE_OOM_LOCKED);
673 spin_unlock(&zone_scan_lock);
678 * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation
679 * attempts or page faults may now recall the oom killer, if necessary.
681 static void clear_system_oom(void)
685 spin_lock(&zone_scan_lock);
686 for_each_populated_zone(zone)
687 zone_clear_flag(zone, ZONE_OOM_LOCKED);
688 spin_unlock(&zone_scan_lock);
692 * out_of_memory - kill the "best" process when we run out of memory
693 * @zonelist: zonelist pointer
694 * @gfp_mask: memory allocation flags
695 * @order: amount of memory being requested as a power of 2
696 * @nodemask: nodemask passed to page allocator
698 * If we run out of memory, we have the choice between either
699 * killing a random task (bad), letting the system crash (worse)
700 * OR try to be smart about which process to kill. Note that we
701 * don't have to be perfect here, we just have to be good.
703 void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
704 int order, nodemask_t *nodemask)
706 const nodemask_t *mpol_mask;
707 struct task_struct *p;
708 unsigned long totalpages;
709 unsigned long freed = 0;
711 enum oom_constraint constraint = CONSTRAINT_NONE;
714 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
716 /* Got some memory back in the last second. */
720 * If current has a pending SIGKILL, then automatically select it. The
721 * goal is to allow it to allocate so that it may quickly exit and free
724 if (fatal_signal_pending(current)) {
725 set_thread_flag(TIF_MEMDIE);
730 * Check if there were limitations on the allocation (only relevant for
731 * NUMA) that may require different handling.
733 constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
735 mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
736 check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
738 read_lock(&tasklist_lock);
739 if (sysctl_oom_kill_allocating_task &&
740 !oom_unkillable_task(current, NULL, nodemask) &&
743 * oom_kill_process() needs tasklist_lock held. If it returns
744 * non-zero, current could not be killed so we must fallback to
747 if (!oom_kill_process(current, gfp_mask, order, 0, totalpages,
749 "Out of memory (oom_kill_allocating_task)"))
754 p = select_bad_process(&points, totalpages, NULL, mpol_mask);
755 if (PTR_ERR(p) == -1UL)
758 /* Found nothing?!?! Either we hang forever, or we panic. */
760 dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
761 read_unlock(&tasklist_lock);
762 panic("Out of memory and no killable processes...\n");
765 if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
766 nodemask, "Out of memory"))
770 read_unlock(&tasklist_lock);
773 * Give "p" a good chance of killing itself before we
774 * retry to allocate memory unless "p" is current
776 if (killed && !test_thread_flag(TIF_MEMDIE))
777 schedule_timeout_uninterruptible(1);
781 * The pagefault handler calls here because it is out of memory, so kill a
782 * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel
783 * oom killing is already in progress so do nothing. If a task is found with
784 * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit.
786 void pagefault_out_of_memory(void)
788 if (try_set_system_oom()) {
789 out_of_memory(NULL, 0, 0, NULL);
792 if (!test_thread_flag(TIF_MEMDIE))
793 schedule_timeout_uninterruptible(1);