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 a thread sharing p->mm that is OOM_SCORE_ADJ_MIN
166 * so the entire heuristic doesn't need to be executed for something
167 * that cannot be killed.
169 if (atomic_read(&p
->mm
->oom_disable_count
)) {
175 * When the PF_OOM_ORIGIN bit is set, it indicates the task should have
176 * priority for oom killing.
178 if (p
->flags
& PF_OOM_ORIGIN
) {
184 * The memory controller may have a limit of 0 bytes, so avoid a divide
185 * by zero, if necessary.
191 * The baseline for the badness score is the proportion of RAM that each
192 * task's rss and swap space use.
194 points
= (get_mm_rss(p
->mm
) + get_mm_counter(p
->mm
, MM_SWAPENTS
)) * 1000 /
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
))
206 * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may
207 * either completely disable oom killing or always prefer a certain
210 points
+= p
->signal
->oom_score_adj
;
213 * Never return 0 for an eligible task that may be killed since it's
214 * possible that no single user task uses more than 0.1% of memory and
215 * no single admin tasks uses more than 3.0%.
219 return (points
< 1000) ? points
: 1000;
223 * Determine the type of allocation constraint.
226 static enum oom_constraint
constrained_alloc(struct zonelist
*zonelist
,
227 gfp_t gfp_mask
, nodemask_t
*nodemask
,
228 unsigned long *totalpages
)
232 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
233 bool cpuset_limited
= false;
236 /* Default to all available memory */
237 *totalpages
= totalram_pages
+ total_swap_pages
;
240 return CONSTRAINT_NONE
;
242 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
243 * to kill current.We have to random task kill in this case.
244 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
246 if (gfp_mask
& __GFP_THISNODE
)
247 return CONSTRAINT_NONE
;
250 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
251 * the page allocator means a mempolicy is in effect. Cpuset policy
252 * is enforced in get_page_from_freelist().
254 if (nodemask
&& !nodes_subset(node_states
[N_HIGH_MEMORY
], *nodemask
)) {
255 *totalpages
= total_swap_pages
;
256 for_each_node_mask(nid
, *nodemask
)
257 *totalpages
+= node_spanned_pages(nid
);
258 return CONSTRAINT_MEMORY_POLICY
;
261 /* Check this allocation failure is caused by cpuset's wall function */
262 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
263 high_zoneidx
, nodemask
)
264 if (!cpuset_zone_allowed_softwall(zone
, gfp_mask
))
265 cpuset_limited
= true;
267 if (cpuset_limited
) {
268 *totalpages
= total_swap_pages
;
269 for_each_node_mask(nid
, cpuset_current_mems_allowed
)
270 *totalpages
+= node_spanned_pages(nid
);
271 return CONSTRAINT_CPUSET
;
273 return CONSTRAINT_NONE
;
276 static enum oom_constraint
constrained_alloc(struct zonelist
*zonelist
,
277 gfp_t gfp_mask
, nodemask_t
*nodemask
,
278 unsigned long *totalpages
)
280 *totalpages
= totalram_pages
+ total_swap_pages
;
281 return CONSTRAINT_NONE
;
286 * Simple selection loop. We chose the process with the highest
287 * number of 'points'. We expect the caller will lock the tasklist.
289 * (not docbooked, we don't want this one cluttering up the manual)
291 static struct task_struct
*select_bad_process(unsigned int *ppoints
,
292 unsigned long totalpages
, struct mem_cgroup
*mem
,
293 const nodemask_t
*nodemask
)
295 struct task_struct
*p
;
296 struct task_struct
*chosen
= NULL
;
299 for_each_process(p
) {
302 if (oom_unkillable_task(p
, mem
, nodemask
))
306 * This task already has access to memory reserves and is
307 * being killed. Don't allow any other task access to the
310 * Note: this may have a chance of deadlock if it gets
311 * blocked waiting for another task which itself is waiting
312 * for memory. Is there a better alternative?
314 if (test_tsk_thread_flag(p
, TIF_MEMDIE
))
315 return ERR_PTR(-1UL);
318 * This is in the process of releasing memory so wait for it
319 * to finish before killing some other task by mistake.
321 * However, if p is the current task, we allow the 'kill' to
322 * go ahead if it is exiting: this will simply set TIF_MEMDIE,
323 * which will allow it to gain access to memory reserves in
324 * the process of exiting and releasing its resources.
325 * Otherwise we could get an easy OOM deadlock.
327 if (thread_group_empty(p
) && (p
->flags
& PF_EXITING
) && p
->mm
) {
329 return ERR_PTR(-1UL);
335 points
= oom_badness(p
, mem
, nodemask
, totalpages
);
336 if (points
> *ppoints
) {
346 * dump_tasks - dump current memory state of all system tasks
347 * @mem: current's memory controller, if constrained
348 * @nodemask: nodemask passed to page allocator for mempolicy ooms
350 * Dumps the current memory state of all eligible tasks. Tasks not in the same
351 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
353 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
354 * value, oom_score_adj value, and name.
356 * Call with tasklist_lock read-locked.
358 static void dump_tasks(const struct mem_cgroup
*mem
, const nodemask_t
*nodemask
)
360 struct task_struct
*p
;
361 struct task_struct
*task
;
363 pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n");
364 for_each_process(p
) {
365 if (oom_unkillable_task(p
, mem
, nodemask
))
368 task
= find_lock_task_mm(p
);
371 * This is a kthread or all of p's threads have already
372 * detached their mm's. There's no need to report
373 * them; they can't be oom killed anyway.
378 pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n",
379 task
->pid
, task_uid(task
), task
->tgid
,
380 task
->mm
->total_vm
, get_mm_rss(task
->mm
),
381 task_cpu(task
), task
->signal
->oom_adj
,
382 task
->signal
->oom_score_adj
, task
->comm
);
387 static void dump_header(struct task_struct
*p
, gfp_t gfp_mask
, int order
,
388 struct mem_cgroup
*mem
, const nodemask_t
*nodemask
)
391 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
392 "oom_adj=%d, oom_score_adj=%d\n",
393 current
->comm
, gfp_mask
, order
, current
->signal
->oom_adj
,
394 current
->signal
->oom_score_adj
);
395 cpuset_print_task_mems_allowed(current
);
396 task_unlock(current
);
398 mem_cgroup_print_oom_info(mem
, p
);
400 if (sysctl_oom_dump_tasks
)
401 dump_tasks(mem
, nodemask
);
404 #define K(x) ((x) << (PAGE_SHIFT-10))
405 static int oom_kill_task(struct task_struct
*p
, struct mem_cgroup
*mem
)
407 struct task_struct
*q
;
408 struct mm_struct
*mm
;
410 p
= find_lock_task_mm(p
);
414 /* mm cannot be safely dereferenced after task_unlock(p) */
417 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
418 task_pid_nr(p
), p
->comm
, K(p
->mm
->total_vm
),
419 K(get_mm_counter(p
->mm
, MM_ANONPAGES
)),
420 K(get_mm_counter(p
->mm
, MM_FILEPAGES
)));
424 * Kill all processes sharing p->mm in other thread groups, if any.
425 * They don't get access to memory reserves or a higher scheduler
426 * priority, though, to avoid depletion of all memory or task
427 * starvation. This prevents mm->mmap_sem livelock when an oom killed
428 * task cannot exit because it requires the semaphore and its contended
429 * by another thread trying to allocate memory itself. That thread will
430 * now get access to memory reserves since it has a pending fatal
434 if (q
->mm
== mm
&& !same_thread_group(q
, p
)) {
435 task_lock(q
); /* Protect ->comm from prctl() */
436 pr_err("Kill process %d (%s) sharing same memory\n",
437 task_pid_nr(q
), q
->comm
);
439 force_sig(SIGKILL
, q
);
442 set_tsk_thread_flag(p
, TIF_MEMDIE
);
443 force_sig(SIGKILL
, p
);
446 * We give our sacrificial lamb high priority and access to
447 * all the memory it needs. That way it should be able to
448 * exit() and clear out its resources quickly...
450 boost_dying_task_prio(p
, mem
);
456 static int oom_kill_process(struct task_struct
*p
, gfp_t gfp_mask
, int order
,
457 unsigned int points
, unsigned long totalpages
,
458 struct mem_cgroup
*mem
, nodemask_t
*nodemask
,
461 struct task_struct
*victim
= p
;
462 struct task_struct
*child
;
463 struct task_struct
*t
= p
;
464 unsigned int victim_points
= 0;
466 if (printk_ratelimit())
467 dump_header(p
, gfp_mask
, order
, mem
, nodemask
);
470 * If the task is already exiting, don't alarm the sysadmin or kill
471 * its children or threads, just set TIF_MEMDIE so it can die quickly
473 if (p
->flags
& PF_EXITING
) {
474 set_tsk_thread_flag(p
, TIF_MEMDIE
);
475 boost_dying_task_prio(p
, mem
);
480 pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
481 message
, task_pid_nr(p
), p
->comm
, points
);
485 * If any of p's children has a different mm and is eligible for kill,
486 * the one with the highest badness() score is sacrificed for its
487 * parent. This attempts to lose the minimal amount of work done while
488 * still freeing memory.
491 list_for_each_entry(child
, &t
->children
, sibling
) {
492 unsigned int child_points
;
495 * oom_badness() returns 0 if the thread is unkillable
497 child_points
= oom_badness(child
, mem
, nodemask
,
499 if (child_points
> victim_points
) {
501 victim_points
= child_points
;
504 } while_each_thread(p
, t
);
506 return oom_kill_task(victim
, mem
);
510 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
512 static void check_panic_on_oom(enum oom_constraint constraint
, gfp_t gfp_mask
,
513 int order
, const nodemask_t
*nodemask
)
515 if (likely(!sysctl_panic_on_oom
))
517 if (sysctl_panic_on_oom
!= 2) {
519 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
520 * does not panic for cpuset, mempolicy, or memcg allocation
523 if (constraint
!= CONSTRAINT_NONE
)
526 read_lock(&tasklist_lock
);
527 dump_header(NULL
, gfp_mask
, order
, NULL
, nodemask
);
528 read_unlock(&tasklist_lock
);
529 panic("Out of memory: %s panic_on_oom is enabled\n",
530 sysctl_panic_on_oom
== 2 ? "compulsory" : "system-wide");
533 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
534 void mem_cgroup_out_of_memory(struct mem_cgroup
*mem
, gfp_t gfp_mask
)
537 unsigned int points
= 0;
538 struct task_struct
*p
;
540 check_panic_on_oom(CONSTRAINT_MEMCG
, gfp_mask
, 0, NULL
);
541 limit
= mem_cgroup_get_limit(mem
) >> PAGE_SHIFT
;
542 read_lock(&tasklist_lock
);
544 p
= select_bad_process(&points
, limit
, mem
, NULL
);
545 if (!p
|| PTR_ERR(p
) == -1UL)
548 if (oom_kill_process(p
, gfp_mask
, 0, points
, limit
, mem
, NULL
,
549 "Memory cgroup out of memory"))
552 read_unlock(&tasklist_lock
);
556 static BLOCKING_NOTIFIER_HEAD(oom_notify_list
);
558 int register_oom_notifier(struct notifier_block
*nb
)
560 return blocking_notifier_chain_register(&oom_notify_list
, nb
);
562 EXPORT_SYMBOL_GPL(register_oom_notifier
);
564 int unregister_oom_notifier(struct notifier_block
*nb
)
566 return blocking_notifier_chain_unregister(&oom_notify_list
, nb
);
568 EXPORT_SYMBOL_GPL(unregister_oom_notifier
);
571 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
572 * if a parallel OOM killing is already taking place that includes a zone in
573 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
575 int try_set_zonelist_oom(struct zonelist
*zonelist
, gfp_t gfp_mask
)
581 spin_lock(&zone_scan_lock
);
582 for_each_zone_zonelist(zone
, z
, zonelist
, gfp_zone(gfp_mask
)) {
583 if (zone_is_oom_locked(zone
)) {
589 for_each_zone_zonelist(zone
, z
, zonelist
, gfp_zone(gfp_mask
)) {
591 * Lock each zone in the zonelist under zone_scan_lock so a
592 * parallel invocation of try_set_zonelist_oom() doesn't succeed
595 zone_set_flag(zone
, ZONE_OOM_LOCKED
);
599 spin_unlock(&zone_scan_lock
);
604 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
605 * allocation attempts with zonelists containing them may now recall the OOM
606 * killer, if necessary.
608 void clear_zonelist_oom(struct zonelist
*zonelist
, gfp_t gfp_mask
)
613 spin_lock(&zone_scan_lock
);
614 for_each_zone_zonelist(zone
, z
, zonelist
, gfp_zone(gfp_mask
)) {
615 zone_clear_flag(zone
, ZONE_OOM_LOCKED
);
617 spin_unlock(&zone_scan_lock
);
621 * Try to acquire the oom killer lock for all system zones. Returns zero if a
622 * parallel oom killing is taking place, otherwise locks all zones and returns
625 static int try_set_system_oom(void)
630 spin_lock(&zone_scan_lock
);
631 for_each_populated_zone(zone
)
632 if (zone_is_oom_locked(zone
)) {
636 for_each_populated_zone(zone
)
637 zone_set_flag(zone
, ZONE_OOM_LOCKED
);
639 spin_unlock(&zone_scan_lock
);
644 * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation
645 * attempts or page faults may now recall the oom killer, if necessary.
647 static void clear_system_oom(void)
651 spin_lock(&zone_scan_lock
);
652 for_each_populated_zone(zone
)
653 zone_clear_flag(zone
, ZONE_OOM_LOCKED
);
654 spin_unlock(&zone_scan_lock
);
658 * out_of_memory - kill the "best" process when we run out of memory
659 * @zonelist: zonelist pointer
660 * @gfp_mask: memory allocation flags
661 * @order: amount of memory being requested as a power of 2
662 * @nodemask: nodemask passed to page allocator
664 * If we run out of memory, we have the choice between either
665 * killing a random task (bad), letting the system crash (worse)
666 * OR try to be smart about which process to kill. Note that we
667 * don't have to be perfect here, we just have to be good.
669 void out_of_memory(struct zonelist
*zonelist
, gfp_t gfp_mask
,
670 int order
, nodemask_t
*nodemask
)
672 const nodemask_t
*mpol_mask
;
673 struct task_struct
*p
;
674 unsigned long totalpages
;
675 unsigned long freed
= 0;
677 enum oom_constraint constraint
= CONSTRAINT_NONE
;
680 blocking_notifier_call_chain(&oom_notify_list
, 0, &freed
);
682 /* Got some memory back in the last second. */
686 * If current has a pending SIGKILL, then automatically select it. The
687 * goal is to allow it to allocate so that it may quickly exit and free
690 if (fatal_signal_pending(current
)) {
691 set_thread_flag(TIF_MEMDIE
);
692 boost_dying_task_prio(current
, NULL
);
697 * Check if there were limitations on the allocation (only relevant for
698 * NUMA) that may require different handling.
700 constraint
= constrained_alloc(zonelist
, gfp_mask
, nodemask
,
702 mpol_mask
= (constraint
== CONSTRAINT_MEMORY_POLICY
) ? nodemask
: NULL
;
703 check_panic_on_oom(constraint
, gfp_mask
, order
, mpol_mask
);
705 read_lock(&tasklist_lock
);
706 if (sysctl_oom_kill_allocating_task
&&
707 !oom_unkillable_task(current
, NULL
, nodemask
) &&
708 current
->mm
&& !atomic_read(¤t
->mm
->oom_disable_count
)) {
710 * oom_kill_process() needs tasklist_lock held. If it returns
711 * non-zero, current could not be killed so we must fallback to
714 if (!oom_kill_process(current
, gfp_mask
, order
, 0, totalpages
,
716 "Out of memory (oom_kill_allocating_task)"))
721 p
= select_bad_process(&points
, totalpages
, NULL
, mpol_mask
);
722 if (PTR_ERR(p
) == -1UL)
725 /* Found nothing?!?! Either we hang forever, or we panic. */
727 dump_header(NULL
, gfp_mask
, order
, NULL
, mpol_mask
);
728 read_unlock(&tasklist_lock
);
729 panic("Out of memory and no killable processes...\n");
732 if (oom_kill_process(p
, gfp_mask
, order
, points
, totalpages
, NULL
,
733 nodemask
, "Out of memory"))
737 read_unlock(&tasklist_lock
);
740 * Give "p" a good chance of killing itself before we
741 * retry to allocate memory unless "p" is current
743 if (killed
&& !test_thread_flag(TIF_MEMDIE
))
744 schedule_timeout_uninterruptible(1);
748 * The pagefault handler calls here because it is out of memory, so kill a
749 * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel
750 * oom killing is already in progress so do nothing. If a task is found with
751 * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit.
753 void pagefault_out_of_memory(void)
755 if (try_set_system_oom()) {
756 out_of_memory(NULL
, 0, 0, NULL
);
759 if (!test_thread_flag(TIF_MEMDIE
))
760 schedule_timeout_uninterruptible(1);