1/*
2 *  linux/mm/oom_kill.c
3 *
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
9 *
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.
13 *
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.
18 */
19
20#include <linux/oom.h>
21#include <linux/mm.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
39#define CREATE_TRACE_POINTS
40#include <trace/events/oom.h>
41
42int sysctl_panic_on_oom;
43int sysctl_oom_kill_allocating_task;
44int sysctl_oom_dump_tasks = 1;
45static DEFINE_SPINLOCK(zone_scan_lock);
46
47#ifdef CONFIG_NUMA
48/**
49 * has_intersects_mems_allowed() - check task eligiblity for kill
50 * @start: task struct of which task to consider
51 * @mask: nodemask passed to page allocator for mempolicy ooms
52 *
53 * Task eligibility is determined by whether or not a candidate task, @tsk,
54 * shares the same mempolicy nodes as current if it is bound by such a policy
55 * and whether or not it has the same set of allowed cpuset nodes.
56 */
57static bool has_intersects_mems_allowed(struct task_struct *start,
58					const nodemask_t *mask)
59{
60	struct task_struct *tsk;
61	bool ret = false;
62
63	rcu_read_lock();
64	for_each_thread(start, tsk) {
65		if (mask) {
66			/*
67			 * If this is a mempolicy constrained oom, tsk's
68			 * cpuset is irrelevant.  Only return true if its
69			 * mempolicy intersects current, otherwise it may be
70			 * needlessly killed.
71			 */
72			ret = mempolicy_nodemask_intersects(tsk, mask);
73		} else {
74			/*
75			 * This is not a mempolicy constrained oom, so only
76			 * check the mems of tsk's cpuset.
77			 */
78			ret = cpuset_mems_allowed_intersects(current, tsk);
79		}
80		if (ret)
81			break;
82	}
83	rcu_read_unlock();
84
85	return ret;
86}
87#else
88static bool has_intersects_mems_allowed(struct task_struct *tsk,
89					const nodemask_t *mask)
90{
91	return true;
92}
93#endif /* CONFIG_NUMA */
94
95/*
96 * The process p may have detached its own ->mm while exiting or through
97 * use_mm(), but one or more of its subthreads may still have a valid
98 * pointer.  Return p, or any of its subthreads with a valid ->mm, with
99 * task_lock() held.
100 */
101struct task_struct *find_lock_task_mm(struct task_struct *p)
102{
103	struct task_struct *t;
104
105	rcu_read_lock();
106
107	for_each_thread(p, t) {
108		task_lock(t);
109		if (likely(t->mm))
110			goto found;
111		task_unlock(t);
112	}
113	t = NULL;
114found:
115	rcu_read_unlock();
116
117	return t;
118}
119
120/* return true if the task is not adequate as candidate victim task. */
121static bool oom_unkillable_task(struct task_struct *p,
122		struct mem_cgroup *memcg, const nodemask_t *nodemask)
123{
124	if (is_global_init(p))
125		return true;
126	if (p->flags & PF_KTHREAD)
127		return true;
128
129	/* When mem_cgroup_out_of_memory() and p is not member of the group */
130	if (memcg && !task_in_mem_cgroup(p, memcg))
131		return true;
132
133	/* p may not have freeable memory in nodemask */
134	if (!has_intersects_mems_allowed(p, nodemask))
135		return true;
136
137	return false;
138}
139
140/**
141 * oom_badness - heuristic function to determine which candidate task to kill
142 * @p: task struct of which task we should calculate
143 * @totalpages: total present RAM allowed for page allocation
144 *
145 * The heuristic for determining which task to kill is made to be as simple and
146 * predictable as possible.  The goal is to return the highest value for the
147 * task consuming the most memory to avoid subsequent oom failures.
148 */
149unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
150			  const nodemask_t *nodemask, unsigned long totalpages)
151{
152	long points;
153	long adj;
154
155	if (oom_unkillable_task(p, memcg, nodemask))
156		return 0;
157
158	p = find_lock_task_mm(p);
159	if (!p)
160		return 0;
161
162	adj = (long)p->signal->oom_score_adj;
163	if (adj == OOM_SCORE_ADJ_MIN) {
164		task_unlock(p);
165		return 0;
166	}
167
168	/*
169	 * The baseline for the badness score is the proportion of RAM that each
170	 * task's rss, pagetable and swap space use.
171	 */
172	points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
173		atomic_long_read(&p->mm->nr_ptes) + mm_nr_pmds(p->mm);
174	task_unlock(p);
175
176	/*
177	 * Root processes get 3% bonus, just like the __vm_enough_memory()
178	 * implementation used by LSMs.
179	 */
180	if (has_capability_noaudit(p, CAP_SYS_ADMIN))
181		points -= (points * 3) / 100;
182
183	/* Normalize to oom_score_adj units */
184	adj *= totalpages / 1000;
185	points += adj;
186
187	/*
188	 * Never return 0 for an eligible task regardless of the root bonus and
189	 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
190	 */
191	return points > 0 ? points : 1;
192}
193
194/*
195 * Determine the type of allocation constraint.
196 */
197#ifdef CONFIG_NUMA
198static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
199				gfp_t gfp_mask, nodemask_t *nodemask,
200				unsigned long *totalpages)
201{
202	struct zone *zone;
203	struct zoneref *z;
204	enum zone_type high_zoneidx = gfp_zone(gfp_mask);
205	bool cpuset_limited = false;
206	int nid;
207
208	/* Default to all available memory */
209	*totalpages = totalram_pages + total_swap_pages;
210
211	if (!zonelist)
212		return CONSTRAINT_NONE;
213	/*
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.
217	 */
218	if (gfp_mask & __GFP_THISNODE)
219		return CONSTRAINT_NONE;
220
221	/*
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().
225	 */
226	if (nodemask && !nodes_subset(node_states[N_MEMORY], *nodemask)) {
227		*totalpages = total_swap_pages;
228		for_each_node_mask(nid, *nodemask)
229			*totalpages += node_spanned_pages(nid);
230		return CONSTRAINT_MEMORY_POLICY;
231	}
232
233	/* Check this allocation failure is caused by cpuset's wall function */
234	for_each_zone_zonelist_nodemask(zone, z, zonelist,
235			high_zoneidx, nodemask)
236		if (!cpuset_zone_allowed(zone, gfp_mask))
237			cpuset_limited = true;
238
239	if (cpuset_limited) {
240		*totalpages = total_swap_pages;
241		for_each_node_mask(nid, cpuset_current_mems_allowed)
242			*totalpages += node_spanned_pages(nid);
243		return CONSTRAINT_CPUSET;
244	}
245	return CONSTRAINT_NONE;
246}
247#else
248static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
249				gfp_t gfp_mask, nodemask_t *nodemask,
250				unsigned long *totalpages)
251{
252	*totalpages = totalram_pages + total_swap_pages;
253	return CONSTRAINT_NONE;
254}
255#endif
256
257enum oom_scan_t oom_scan_process_thread(struct task_struct *task,
258		unsigned long totalpages, const nodemask_t *nodemask,
259		bool force_kill)
260{
261	if (oom_unkillable_task(task, NULL, nodemask))
262		return OOM_SCAN_CONTINUE;
263
264	/*
265	 * This task already has access to memory reserves and is being killed.
266	 * Don't allow any other task to have access to the reserves.
267	 */
268	if (test_tsk_thread_flag(task, TIF_MEMDIE)) {
269		if (!force_kill)
270			return OOM_SCAN_ABORT;
271	}
272	if (!task->mm)
273		return OOM_SCAN_CONTINUE;
274
275	/*
276	 * If task is allocating a lot of memory and has been marked to be
277	 * killed first if it triggers an oom, then select it.
278	 */
279	if (oom_task_origin(task))
280		return OOM_SCAN_SELECT;
281
282	if (task_will_free_mem(task) && !force_kill)
283		return OOM_SCAN_ABORT;
284
285	return OOM_SCAN_OK;
286}
287
288/*
289 * Simple selection loop. We chose the process with the highest
290 * number of 'points'.  Returns -1 on scan abort.
291 *
292 * (not docbooked, we don't want this one cluttering up the manual)
293 */
294static struct task_struct *select_bad_process(unsigned int *ppoints,
295		unsigned long totalpages, const nodemask_t *nodemask,
296		bool force_kill)
297{
298	struct task_struct *g, *p;
299	struct task_struct *chosen = NULL;
300	unsigned long chosen_points = 0;
301
302	rcu_read_lock();
303	for_each_process_thread(g, p) {
304		unsigned int points;
305
306		switch (oom_scan_process_thread(p, totalpages, nodemask,
307						force_kill)) {
308		case OOM_SCAN_SELECT:
309			chosen = p;
310			chosen_points = ULONG_MAX;
311			/* fall through */
312		case OOM_SCAN_CONTINUE:
313			continue;
314		case OOM_SCAN_ABORT:
315			rcu_read_unlock();
316			return (struct task_struct *)(-1UL);
317		case OOM_SCAN_OK:
318			break;
319		};
320		points = oom_badness(p, NULL, nodemask, totalpages);
321		if (!points || points < chosen_points)
322			continue;
323		/* Prefer thread group leaders for display purposes */
324		if (points == chosen_points && thread_group_leader(chosen))
325			continue;
326
327		chosen = p;
328		chosen_points = points;
329	}
330	if (chosen)
331		get_task_struct(chosen);
332	rcu_read_unlock();
333
334	*ppoints = chosen_points * 1000 / totalpages;
335	return chosen;
336}
337
338/**
339 * dump_tasks - dump current memory state of all system tasks
340 * @memcg: current's memory controller, if constrained
341 * @nodemask: nodemask passed to page allocator for mempolicy ooms
342 *
343 * Dumps the current memory state of all eligible tasks.  Tasks not in the same
344 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
345 * are not shown.
346 * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
347 * swapents, oom_score_adj value, and name.
348 */
349static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask)
350{
351	struct task_struct *p;
352	struct task_struct *task;
353
354	pr_info("[ pid ]   uid  tgid total_vm      rss nr_ptes nr_pmds swapents oom_score_adj name\n");
355	rcu_read_lock();
356	for_each_process(p) {
357		if (oom_unkillable_task(p, memcg, nodemask))
358			continue;
359
360		task = find_lock_task_mm(p);
361		if (!task) {
362			/*
363			 * This is a kthread or all of p's threads have already
364			 * detached their mm's.  There's no need to report
365			 * them; they can't be oom killed anyway.
366			 */
367			continue;
368		}
369
370		pr_info("[%5d] %5d %5d %8lu %8lu %7ld %7ld %8lu         %5hd %s\n",
371			task->pid, from_kuid(&init_user_ns, task_uid(task)),
372			task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
373			atomic_long_read(&task->mm->nr_ptes),
374			mm_nr_pmds(task->mm),
375			get_mm_counter(task->mm, MM_SWAPENTS),
376			task->signal->oom_score_adj, task->comm);
377		task_unlock(task);
378	}
379	rcu_read_unlock();
380}
381
382static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
383			struct mem_cgroup *memcg, const nodemask_t *nodemask)
384{
385	task_lock(current);
386	pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
387		"oom_score_adj=%hd\n",
388		current->comm, gfp_mask, order,
389		current->signal->oom_score_adj);
390	cpuset_print_task_mems_allowed(current);
391	task_unlock(current);
392	dump_stack();
393	if (memcg)
394		mem_cgroup_print_oom_info(memcg, p);
395	else
396		show_mem(SHOW_MEM_FILTER_NODES);
397	if (sysctl_oom_dump_tasks)
398		dump_tasks(memcg, nodemask);
399}
400
401/*
402 * Number of OOM victims in flight
403 */
404static atomic_t oom_victims = ATOMIC_INIT(0);
405static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
406
407bool oom_killer_disabled __read_mostly;
408static DECLARE_RWSEM(oom_sem);
409
410/**
411 * mark_tsk_oom_victim - marks the given task as OOM victim.
412 * @tsk: task to mark
413 *
414 * Has to be called with oom_sem taken for read and never after
415 * oom has been disabled already.
416 */
417void mark_tsk_oom_victim(struct task_struct *tsk)
418{
419	WARN_ON(oom_killer_disabled);
420	/* OOM killer might race with memcg OOM */
421	if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
422		return;
423	/*
424	 * Make sure that the task is woken up from uninterruptible sleep
425	 * if it is frozen because OOM killer wouldn't be able to free
426	 * any memory and livelock. freezing_slow_path will tell the freezer
427	 * that TIF_MEMDIE tasks should be ignored.
428	 */
429	__thaw_task(tsk);
430	atomic_inc(&oom_victims);
431}
432
433/**
434 * unmark_oom_victim - unmarks the current task as OOM victim.
435 *
436 * Wakes up all waiters in oom_killer_disable()
437 */
438void unmark_oom_victim(void)
439{
440	if (!test_and_clear_thread_flag(TIF_MEMDIE))
441		return;
442
443	down_read(&oom_sem);
444	/*
445	 * There is no need to signal the lasst oom_victim if there
446	 * is nobody who cares.
447	 */
448	if (!atomic_dec_return(&oom_victims) && oom_killer_disabled)
449		wake_up_all(&oom_victims_wait);
450	up_read(&oom_sem);
451}
452
453/**
454 * oom_killer_disable - disable OOM killer
455 *
456 * Forces all page allocations to fail rather than trigger OOM killer.
457 * Will block and wait until all OOM victims are killed.
458 *
459 * The function cannot be called when there are runnable user tasks because
460 * the userspace would see unexpected allocation failures as a result. Any
461 * new usage of this function should be consulted with MM people.
462 *
463 * Returns true if successful and false if the OOM killer cannot be
464 * disabled.
465 */
466bool oom_killer_disable(void)
467{
468	/*
469	 * Make sure to not race with an ongoing OOM killer
470	 * and that the current is not the victim.
471	 */
472	down_write(&oom_sem);
473	if (test_thread_flag(TIF_MEMDIE)) {
474		up_write(&oom_sem);
475		return false;
476	}
477
478	oom_killer_disabled = true;
479	up_write(&oom_sem);
480
481	wait_event(oom_victims_wait, !atomic_read(&oom_victims));
482
483	return true;
484}
485
486/**
487 * oom_killer_enable - enable OOM killer
488 */
489void oom_killer_enable(void)
490{
491	down_write(&oom_sem);
492	oom_killer_disabled = false;
493	up_write(&oom_sem);
494}
495
496#define K(x) ((x) << (PAGE_SHIFT-10))
497/*
498 * Must be called while holding a reference to p, which will be released upon
499 * returning.
500 */
501void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
502		      unsigned int points, unsigned long totalpages,
503		      struct mem_cgroup *memcg, nodemask_t *nodemask,
504		      const char *message)
505{
506	struct task_struct *victim = p;
507	struct task_struct *child;
508	struct task_struct *t;
509	struct mm_struct *mm;
510	unsigned int victim_points = 0;
511	static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
512					      DEFAULT_RATELIMIT_BURST);
513
514	/*
515	 * If the task is already exiting, don't alarm the sysadmin or kill
516	 * its children or threads, just set TIF_MEMDIE so it can die quickly
517	 */
518	task_lock(p);
519	if (p->mm && task_will_free_mem(p)) {
520		mark_tsk_oom_victim(p);
521		task_unlock(p);
522		put_task_struct(p);
523		return;
524	}
525	task_unlock(p);
526
527	if (__ratelimit(&oom_rs))
528		dump_header(p, gfp_mask, order, memcg, nodemask);
529
530	task_lock(p);
531	pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
532		message, task_pid_nr(p), p->comm, points);
533	task_unlock(p);
534
535	/*
536	 * If any of p's children has a different mm and is eligible for kill,
537	 * the one with the highest oom_badness() score is sacrificed for its
538	 * parent.  This attempts to lose the minimal amount of work done while
539	 * still freeing memory.
540	 */
541	read_lock(&tasklist_lock);
542	for_each_thread(p, t) {
543		list_for_each_entry(child, &t->children, sibling) {
544			unsigned int child_points;
545
546			if (child->mm == p->mm)
547				continue;
548			/*
549			 * oom_badness() returns 0 if the thread is unkillable
550			 */
551			child_points = oom_badness(child, memcg, nodemask,
552								totalpages);
553			if (child_points > victim_points) {
554				put_task_struct(victim);
555				victim = child;
556				victim_points = child_points;
557				get_task_struct(victim);
558			}
559		}
560	}
561	read_unlock(&tasklist_lock);
562
563	p = find_lock_task_mm(victim);
564	if (!p) {
565		put_task_struct(victim);
566		return;
567	} else if (victim != p) {
568		get_task_struct(p);
569		put_task_struct(victim);
570		victim = p;
571	}
572
573	/* mm cannot safely be dereferenced after task_unlock(victim) */
574	mm = victim->mm;
575	mark_tsk_oom_victim(victim);
576	pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
577		task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
578		K(get_mm_counter(victim->mm, MM_ANONPAGES)),
579		K(get_mm_counter(victim->mm, MM_FILEPAGES)));
580	task_unlock(victim);
581
582	/*
583	 * Kill all user processes sharing victim->mm in other thread groups, if
584	 * any.  They don't get access to memory reserves, though, to avoid
585	 * depletion of all memory.  This prevents mm->mmap_sem livelock when an
586	 * oom killed thread cannot exit because it requires the semaphore and
587	 * its contended by another thread trying to allocate memory itself.
588	 * That thread will now get access to memory reserves since it has a
589	 * pending fatal signal.
590	 */
591	rcu_read_lock();
592	for_each_process(p)
593		if (p->mm == mm && !same_thread_group(p, victim) &&
594		    !(p->flags & PF_KTHREAD)) {
595			if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
596				continue;
597
598			task_lock(p);	/* Protect ->comm from prctl() */
599			pr_err("Kill process %d (%s) sharing same memory\n",
600				task_pid_nr(p), p->comm);
601			task_unlock(p);
602			do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
603		}
604	rcu_read_unlock();
605
606	do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
607	put_task_struct(victim);
608}
609#undef K
610
611/*
612 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
613 */
614void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
615			int order, const nodemask_t *nodemask,
616			struct mem_cgroup *memcg)
617{
618	if (likely(!sysctl_panic_on_oom))
619		return;
620	if (sysctl_panic_on_oom != 2) {
621		/*
622		 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
623		 * does not panic for cpuset, mempolicy, or memcg allocation
624		 * failures.
625		 */
626		if (constraint != CONSTRAINT_NONE)
627			return;
628	}
629	dump_header(NULL, gfp_mask, order, memcg, nodemask);
630	panic("Out of memory: %s panic_on_oom is enabled\n",
631		sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
632}
633
634static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
635
636int register_oom_notifier(struct notifier_block *nb)
637{
638	return blocking_notifier_chain_register(&oom_notify_list, nb);
639}
640EXPORT_SYMBOL_GPL(register_oom_notifier);
641
642int unregister_oom_notifier(struct notifier_block *nb)
643{
644	return blocking_notifier_chain_unregister(&oom_notify_list, nb);
645}
646EXPORT_SYMBOL_GPL(unregister_oom_notifier);
647
648/*
649 * Try to acquire the OOM killer lock for the zones in zonelist.  Returns zero
650 * if a parallel OOM killing is already taking place that includes a zone in
651 * the zonelist.  Otherwise, locks all zones in the zonelist and returns 1.
652 */
653bool oom_zonelist_trylock(struct zonelist *zonelist, gfp_t gfp_mask)
654{
655	struct zoneref *z;
656	struct zone *zone;
657	bool ret = true;
658
659	spin_lock(&zone_scan_lock);
660	for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask))
661		if (test_bit(ZONE_OOM_LOCKED, &zone->flags)) {
662			ret = false;
663			goto out;
664		}
665
666	/*
667	 * Lock each zone in the zonelist under zone_scan_lock so a parallel
668	 * call to oom_zonelist_trylock() doesn't succeed when it shouldn't.
669	 */
670	for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask))
671		set_bit(ZONE_OOM_LOCKED, &zone->flags);
672
673out:
674	spin_unlock(&zone_scan_lock);
675	return ret;
676}
677
678/*
679 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
680 * allocation attempts with zonelists containing them may now recall the OOM
681 * killer, if necessary.
682 */
683void oom_zonelist_unlock(struct zonelist *zonelist, gfp_t gfp_mask)
684{
685	struct zoneref *z;
686	struct zone *zone;
687
688	spin_lock(&zone_scan_lock);
689	for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask))
690		clear_bit(ZONE_OOM_LOCKED, &zone->flags);
691	spin_unlock(&zone_scan_lock);
692}
693
694/**
695 * __out_of_memory - kill the "best" process when we run out of memory
696 * @zonelist: zonelist pointer
697 * @gfp_mask: memory allocation flags
698 * @order: amount of memory being requested as a power of 2
699 * @nodemask: nodemask passed to page allocator
700 * @force_kill: true if a task must be killed, even if others are exiting
701 *
702 * If we run out of memory, we have the choice between either
703 * killing a random task (bad), letting the system crash (worse)
704 * OR try to be smart about which process to kill. Note that we
705 * don't have to be perfect here, we just have to be good.
706 */
707static void __out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
708		int order, nodemask_t *nodemask, bool force_kill)
709{
710	const nodemask_t *mpol_mask;
711	struct task_struct *p;
712	unsigned long totalpages;
713	unsigned long freed = 0;
714	unsigned int uninitialized_var(points);
715	enum oom_constraint constraint = CONSTRAINT_NONE;
716	int killed = 0;
717
718	blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
719	if (freed > 0)
720		/* Got some memory back in the last second. */
721		return;
722
723	/*
724	 * If current has a pending SIGKILL or is exiting, then automatically
725	 * select it.  The goal is to allow it to allocate so that it may
726	 * quickly exit and free its memory.
727	 *
728	 * But don't select if current has already released its mm and cleared
729	 * TIF_MEMDIE flag at exit_mm(), otherwise an OOM livelock may occur.
730	 */
731	if (current->mm &&
732	    (fatal_signal_pending(current) || task_will_free_mem(current))) {
733		mark_tsk_oom_victim(current);
734		return;
735	}
736
737	/*
738	 * Check if there were limitations on the allocation (only relevant for
739	 * NUMA) that may require different handling.
740	 */
741	constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
742						&totalpages);
743	mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
744	check_panic_on_oom(constraint, gfp_mask, order, mpol_mask, NULL);
745
746	if (sysctl_oom_kill_allocating_task && current->mm &&
747	    !oom_unkillable_task(current, NULL, nodemask) &&
748	    current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
749		get_task_struct(current);
750		oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL,
751				 nodemask,
752				 "Out of memory (oom_kill_allocating_task)");
753		goto out;
754	}
755
756	p = select_bad_process(&points, totalpages, mpol_mask, force_kill);
757	/* Found nothing?!?! Either we hang forever, or we panic. */
758	if (!p) {
759		dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
760		panic("Out of memory and no killable processes...\n");
761	}
762	if (p != (void *)-1UL) {
763		oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
764				 nodemask, "Out of memory");
765		killed = 1;
766	}
767out:
768	/*
769	 * Give the killed threads a good chance of exiting before trying to
770	 * allocate memory again.
771	 */
772	if (killed)
773		schedule_timeout_killable(1);
774}
775
776/**
777 * out_of_memory -  tries to invoke OOM killer.
778 * @zonelist: zonelist pointer
779 * @gfp_mask: memory allocation flags
780 * @order: amount of memory being requested as a power of 2
781 * @nodemask: nodemask passed to page allocator
782 * @force_kill: true if a task must be killed, even if others are exiting
783 *
784 * invokes __out_of_memory if the OOM is not disabled by oom_killer_disable()
785 * when it returns false. Otherwise returns true.
786 */
787bool out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
788		int order, nodemask_t *nodemask, bool force_kill)
789{
790	bool ret = false;
791
792	down_read(&oom_sem);
793	if (!oom_killer_disabled) {
794		__out_of_memory(zonelist, gfp_mask, order, nodemask, force_kill);
795		ret = true;
796	}
797	up_read(&oom_sem);
798
799	return ret;
800}
801
802/*
803 * The pagefault handler calls here because it is out of memory, so kill a
804 * memory-hogging task.  If any populated zone has ZONE_OOM_LOCKED set, a
805 * parallel oom killing is already in progress so do nothing.
806 */
807void pagefault_out_of_memory(void)
808{
809	struct zonelist *zonelist;
810
811	down_read(&oom_sem);
812	if (mem_cgroup_oom_synchronize(true))
813		goto unlock;
814
815	zonelist = node_zonelist(first_memory_node, GFP_KERNEL);
816	if (oom_zonelist_trylock(zonelist, GFP_KERNEL)) {
817		if (!oom_killer_disabled)
818			__out_of_memory(NULL, 0, 0, NULL, false);
819		else
820			/*
821			 * There shouldn't be any user tasks runable while the
822			 * OOM killer is disabled so the current task has to
823			 * be a racing OOM victim for which oom_killer_disable()
824			 * is waiting for.
825			 */
826			WARN_ON(test_thread_flag(TIF_MEMDIE));
827
828		oom_zonelist_unlock(zonelist, GFP_KERNEL);
829	}
830unlock:
831	up_read(&oom_sem);
832}
833