• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
42 int sysctl_panic_on_oom;
43 int sysctl_oom_kill_allocating_task;
44 int sysctl_oom_dump_tasks = 1;
45 static 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  */
has_intersects_mems_allowed(struct task_struct * start,const nodemask_t * mask)57 static 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
has_intersects_mems_allowed(struct task_struct * tsk,const nodemask_t * mask)88 static 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  */
find_lock_task_mm(struct task_struct * p)101 struct 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;
114 found:
115 	rcu_read_unlock();
116 
117 	return t;
118 }
119 
120 /* return true if the task is not adequate as candidate victim task. */
oom_unkillable_task(struct task_struct * p,const struct mem_cgroup * memcg,const nodemask_t * nodemask)121 static bool oom_unkillable_task(struct task_struct *p,
122 		const 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  */
oom_badness(struct task_struct * p,struct mem_cgroup * memcg,const nodemask_t * nodemask,unsigned long totalpages)149 unsigned 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) + atomic_long_read(&p->mm->nr_ptes) +
173 		 get_mm_counter(p->mm, MM_SWAPENTS);
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
constrained_alloc(struct zonelist * zonelist,gfp_t gfp_mask,nodemask_t * nodemask,unsigned long * totalpages)198 static 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_softwall(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
constrained_alloc(struct zonelist * zonelist,gfp_t gfp_mask,nodemask_t * nodemask,unsigned long * totalpages)248 static 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 
oom_scan_process_thread(struct task_struct * task,unsigned long totalpages,const nodemask_t * nodemask,bool force_kill)257 enum 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 (unlikely(frozen(task)))
270 			__thaw_task(task);
271 		if (!force_kill)
272 			return OOM_SCAN_ABORT;
273 	}
274 	if (!task->mm)
275 		return OOM_SCAN_CONTINUE;
276 
277 	/*
278 	 * If task is allocating a lot of memory and has been marked to be
279 	 * killed first if it triggers an oom, then select it.
280 	 */
281 	if (oom_task_origin(task))
282 		return OOM_SCAN_SELECT;
283 
284 	if (task->flags & PF_EXITING && !force_kill) {
285 		/*
286 		 * If this task is not being ptraced on exit, then wait for it
287 		 * to finish before killing some other task unnecessarily.
288 		 */
289 		if (!(task->group_leader->ptrace & PT_TRACE_EXIT))
290 			return OOM_SCAN_ABORT;
291 	}
292 	return OOM_SCAN_OK;
293 }
294 
295 /*
296  * Simple selection loop. We chose the process with the highest
297  * number of 'points'.  Returns -1 on scan abort.
298  *
299  * (not docbooked, we don't want this one cluttering up the manual)
300  */
select_bad_process(unsigned int * ppoints,unsigned long totalpages,const nodemask_t * nodemask,bool force_kill)301 static struct task_struct *select_bad_process(unsigned int *ppoints,
302 		unsigned long totalpages, const nodemask_t *nodemask,
303 		bool force_kill)
304 {
305 	struct task_struct *g, *p;
306 	struct task_struct *chosen = NULL;
307 	unsigned long chosen_points = 0;
308 
309 	rcu_read_lock();
310 	for_each_process_thread(g, p) {
311 		unsigned int points;
312 
313 		switch (oom_scan_process_thread(p, totalpages, nodemask,
314 						force_kill)) {
315 		case OOM_SCAN_SELECT:
316 			chosen = p;
317 			chosen_points = ULONG_MAX;
318 			/* fall through */
319 		case OOM_SCAN_CONTINUE:
320 			continue;
321 		case OOM_SCAN_ABORT:
322 			rcu_read_unlock();
323 			return (struct task_struct *)(-1UL);
324 		case OOM_SCAN_OK:
325 			break;
326 		};
327 		points = oom_badness(p, NULL, nodemask, totalpages);
328 		if (!points || points < chosen_points)
329 			continue;
330 		/* Prefer thread group leaders for display purposes */
331 		if (points == chosen_points && thread_group_leader(chosen))
332 			continue;
333 
334 		chosen = p;
335 		chosen_points = points;
336 	}
337 	if (chosen)
338 		get_task_struct(chosen);
339 	rcu_read_unlock();
340 
341 	*ppoints = chosen_points * 1000 / totalpages;
342 	return chosen;
343 }
344 
345 /**
346  * dump_tasks - dump current memory state of all system tasks
347  * @memcg: current's memory controller, if constrained
348  * @nodemask: nodemask passed to page allocator for mempolicy ooms
349  *
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
352  * are not shown.
353  * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
354  * swapents, oom_score_adj value, and name.
355  */
dump_tasks(const struct mem_cgroup * memcg,const nodemask_t * nodemask)356 static void dump_tasks(const struct mem_cgroup *memcg, const nodemask_t *nodemask)
357 {
358 	struct task_struct *p;
359 	struct task_struct *task;
360 
361 	pr_info("[ pid ]   uid  tgid total_vm      rss nr_ptes swapents oom_score_adj name\n");
362 	rcu_read_lock();
363 	for_each_process(p) {
364 		if (oom_unkillable_task(p, memcg, nodemask))
365 			continue;
366 
367 		task = find_lock_task_mm(p);
368 		if (!task) {
369 			/*
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.
373 			 */
374 			continue;
375 		}
376 
377 		pr_info("[%5d] %5d %5d %8lu %8lu %7ld %8lu         %5hd %s\n",
378 			task->pid, from_kuid(&init_user_ns, task_uid(task)),
379 			task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
380 			atomic_long_read(&task->mm->nr_ptes),
381 			get_mm_counter(task->mm, MM_SWAPENTS),
382 			task->signal->oom_score_adj, task->comm);
383 		task_unlock(task);
384 	}
385 	rcu_read_unlock();
386 }
387 
dump_header(struct task_struct * p,gfp_t gfp_mask,int order,struct mem_cgroup * memcg,const nodemask_t * nodemask)388 static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
389 			struct mem_cgroup *memcg, const nodemask_t *nodemask)
390 {
391 	task_lock(current);
392 	pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
393 		"oom_score_adj=%hd\n",
394 		current->comm, gfp_mask, order,
395 		current->signal->oom_score_adj);
396 	cpuset_print_task_mems_allowed(current);
397 	task_unlock(current);
398 	dump_stack();
399 	if (memcg)
400 		mem_cgroup_print_oom_info(memcg, p);
401 	else
402 		show_mem(SHOW_MEM_FILTER_NODES);
403 	if (sysctl_oom_dump_tasks)
404 		dump_tasks(memcg, nodemask);
405 }
406 
407 /*
408  * Number of OOM killer invocations (including memcg OOM killer).
409  * Primarily used by PM freezer to check for potential races with
410  * OOM killed frozen task.
411  */
412 static atomic_t oom_kills = ATOMIC_INIT(0);
413 
oom_kills_count(void)414 int oom_kills_count(void)
415 {
416 	return atomic_read(&oom_kills);
417 }
418 
note_oom_kill(void)419 void note_oom_kill(void)
420 {
421 	atomic_inc(&oom_kills);
422 }
423 
424 #define K(x) ((x) << (PAGE_SHIFT-10))
425 /*
426  * Must be called while holding a reference to p, which will be released upon
427  * returning.
428  */
oom_kill_process(struct task_struct * p,gfp_t gfp_mask,int order,unsigned int points,unsigned long totalpages,struct mem_cgroup * memcg,nodemask_t * nodemask,const char * message)429 void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
430 		      unsigned int points, unsigned long totalpages,
431 		      struct mem_cgroup *memcg, nodemask_t *nodemask,
432 		      const char *message)
433 {
434 	struct task_struct *victim = p;
435 	struct task_struct *child;
436 	struct task_struct *t;
437 	struct mm_struct *mm;
438 	unsigned int victim_points = 0;
439 	static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
440 					      DEFAULT_RATELIMIT_BURST);
441 
442 	/*
443 	 * If the task is already exiting, don't alarm the sysadmin or kill
444 	 * its children or threads, just set TIF_MEMDIE so it can die quickly
445 	 */
446 	if (p->flags & PF_EXITING) {
447 		set_tsk_thread_flag(p, TIF_MEMDIE);
448 		put_task_struct(p);
449 		return;
450 	}
451 
452 	if (__ratelimit(&oom_rs))
453 		dump_header(p, gfp_mask, order, memcg, nodemask);
454 
455 	task_lock(p);
456 	pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
457 		message, task_pid_nr(p), p->comm, points);
458 	task_unlock(p);
459 
460 	/*
461 	 * If any of p's children has a different mm and is eligible for kill,
462 	 * the one with the highest oom_badness() score is sacrificed for its
463 	 * parent.  This attempts to lose the minimal amount of work done while
464 	 * still freeing memory.
465 	 */
466 	read_lock(&tasklist_lock);
467 	for_each_thread(p, t) {
468 		list_for_each_entry(child, &t->children, sibling) {
469 			unsigned int child_points;
470 
471 			if (child->mm == p->mm)
472 				continue;
473 			/*
474 			 * oom_badness() returns 0 if the thread is unkillable
475 			 */
476 			child_points = oom_badness(child, memcg, nodemask,
477 								totalpages);
478 			if (child_points > victim_points) {
479 				put_task_struct(victim);
480 				victim = child;
481 				victim_points = child_points;
482 				get_task_struct(victim);
483 			}
484 		}
485 	}
486 	read_unlock(&tasklist_lock);
487 
488 	p = find_lock_task_mm(victim);
489 	if (!p) {
490 		put_task_struct(victim);
491 		return;
492 	} else if (victim != p) {
493 		get_task_struct(p);
494 		put_task_struct(victim);
495 		victim = p;
496 	}
497 
498 	/* mm cannot safely be dereferenced after task_unlock(victim) */
499 	mm = victim->mm;
500 	pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
501 		task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
502 		K(get_mm_counter(victim->mm, MM_ANONPAGES)),
503 		K(get_mm_counter(victim->mm, MM_FILEPAGES)));
504 	task_unlock(victim);
505 
506 	/*
507 	 * Kill all user processes sharing victim->mm in other thread groups, if
508 	 * any.  They don't get access to memory reserves, though, to avoid
509 	 * depletion of all memory.  This prevents mm->mmap_sem livelock when an
510 	 * oom killed thread cannot exit because it requires the semaphore and
511 	 * its contended by another thread trying to allocate memory itself.
512 	 * That thread will now get access to memory reserves since it has a
513 	 * pending fatal signal.
514 	 */
515 	rcu_read_lock();
516 	for_each_process(p)
517 		if (p->mm == mm && !same_thread_group(p, victim) &&
518 		    !(p->flags & PF_KTHREAD)) {
519 			if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
520 				continue;
521 
522 			task_lock(p);	/* Protect ->comm from prctl() */
523 			pr_err("Kill process %d (%s) sharing same memory\n",
524 				task_pid_nr(p), p->comm);
525 			task_unlock(p);
526 			do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
527 		}
528 	rcu_read_unlock();
529 
530 	set_tsk_thread_flag(victim, TIF_MEMDIE);
531 	do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
532 	put_task_struct(victim);
533 }
534 #undef K
535 
536 /*
537  * Determines whether the kernel must panic because of the panic_on_oom sysctl.
538  */
check_panic_on_oom(enum oom_constraint constraint,gfp_t gfp_mask,int order,const nodemask_t * nodemask)539 void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
540 			int order, const nodemask_t *nodemask)
541 {
542 	if (likely(!sysctl_panic_on_oom))
543 		return;
544 	if (sysctl_panic_on_oom != 2) {
545 		/*
546 		 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
547 		 * does not panic for cpuset, mempolicy, or memcg allocation
548 		 * failures.
549 		 */
550 		if (constraint != CONSTRAINT_NONE)
551 			return;
552 	}
553 	dump_header(NULL, gfp_mask, order, NULL, nodemask);
554 	panic("Out of memory: %s panic_on_oom is enabled\n",
555 		sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
556 }
557 
558 static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
559 
register_oom_notifier(struct notifier_block * nb)560 int register_oom_notifier(struct notifier_block *nb)
561 {
562 	return blocking_notifier_chain_register(&oom_notify_list, nb);
563 }
564 EXPORT_SYMBOL_GPL(register_oom_notifier);
565 
unregister_oom_notifier(struct notifier_block * nb)566 int unregister_oom_notifier(struct notifier_block *nb)
567 {
568 	return blocking_notifier_chain_unregister(&oom_notify_list, nb);
569 }
570 EXPORT_SYMBOL_GPL(unregister_oom_notifier);
571 
572 /*
573  * Try to acquire the OOM killer lock for the zones in zonelist.  Returns zero
574  * if a parallel OOM killing is already taking place that includes a zone in
575  * the zonelist.  Otherwise, locks all zones in the zonelist and returns 1.
576  */
oom_zonelist_trylock(struct zonelist * zonelist,gfp_t gfp_mask)577 bool oom_zonelist_trylock(struct zonelist *zonelist, gfp_t gfp_mask)
578 {
579 	struct zoneref *z;
580 	struct zone *zone;
581 	bool ret = true;
582 
583 	spin_lock(&zone_scan_lock);
584 	for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask))
585 		if (test_bit(ZONE_OOM_LOCKED, &zone->flags)) {
586 			ret = false;
587 			goto out;
588 		}
589 
590 	/*
591 	 * Lock each zone in the zonelist under zone_scan_lock so a parallel
592 	 * call to oom_zonelist_trylock() doesn't succeed when it shouldn't.
593 	 */
594 	for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask))
595 		set_bit(ZONE_OOM_LOCKED, &zone->flags);
596 
597 out:
598 	spin_unlock(&zone_scan_lock);
599 	return ret;
600 }
601 
602 /*
603  * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
604  * allocation attempts with zonelists containing them may now recall the OOM
605  * killer, if necessary.
606  */
oom_zonelist_unlock(struct zonelist * zonelist,gfp_t gfp_mask)607 void oom_zonelist_unlock(struct zonelist *zonelist, gfp_t gfp_mask)
608 {
609 	struct zoneref *z;
610 	struct zone *zone;
611 
612 	spin_lock(&zone_scan_lock);
613 	for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask))
614 		clear_bit(ZONE_OOM_LOCKED, &zone->flags);
615 	spin_unlock(&zone_scan_lock);
616 }
617 
618 /**
619  * out_of_memory - kill the "best" process when we run out of memory
620  * @zonelist: zonelist pointer
621  * @gfp_mask: memory allocation flags
622  * @order: amount of memory being requested as a power of 2
623  * @nodemask: nodemask passed to page allocator
624  * @force_kill: true if a task must be killed, even if others are exiting
625  *
626  * If we run out of memory, we have the choice between either
627  * killing a random task (bad), letting the system crash (worse)
628  * OR try to be smart about which process to kill. Note that we
629  * don't have to be perfect here, we just have to be good.
630  */
out_of_memory(struct zonelist * zonelist,gfp_t gfp_mask,int order,nodemask_t * nodemask,bool force_kill)631 void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
632 		int order, nodemask_t *nodemask, bool force_kill)
633 {
634 	const nodemask_t *mpol_mask;
635 	struct task_struct *p;
636 	unsigned long totalpages;
637 	unsigned long freed = 0;
638 	unsigned int uninitialized_var(points);
639 	enum oom_constraint constraint = CONSTRAINT_NONE;
640 	int killed = 0;
641 
642 	blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
643 	if (freed > 0)
644 		/* Got some memory back in the last second. */
645 		return;
646 
647 	/*
648 	 * If current has a pending SIGKILL or is exiting, then automatically
649 	 * select it.  The goal is to allow it to allocate so that it may
650 	 * quickly exit and free its memory.
651 	 */
652 	if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
653 		set_thread_flag(TIF_MEMDIE);
654 		return;
655 	}
656 
657 	/*
658 	 * Check if there were limitations on the allocation (only relevant for
659 	 * NUMA) that may require different handling.
660 	 */
661 	constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
662 						&totalpages);
663 	mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
664 	check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
665 
666 	if (sysctl_oom_kill_allocating_task && current->mm &&
667 	    !oom_unkillable_task(current, NULL, nodemask) &&
668 	    current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
669 		get_task_struct(current);
670 		oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL,
671 				 nodemask,
672 				 "Out of memory (oom_kill_allocating_task)");
673 		goto out;
674 	}
675 
676 	p = select_bad_process(&points, totalpages, mpol_mask, force_kill);
677 	/* Found nothing?!?! Either we hang forever, or we panic. */
678 	if (!p) {
679 		dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
680 		panic("Out of memory and no killable processes...\n");
681 	}
682 	if (p != (void *)-1UL) {
683 		oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
684 				 nodemask, "Out of memory");
685 		killed = 1;
686 	}
687 out:
688 	/*
689 	 * Give the killed threads a good chance of exiting before trying to
690 	 * allocate memory again.
691 	 */
692 	if (killed)
693 		schedule_timeout_killable(1);
694 }
695 
696 /*
697  * The pagefault handler calls here because it is out of memory, so kill a
698  * memory-hogging task.  If any populated zone has ZONE_OOM_LOCKED set, a
699  * parallel oom killing is already in progress so do nothing.
700  */
pagefault_out_of_memory(void)701 void pagefault_out_of_memory(void)
702 {
703 	struct zonelist *zonelist;
704 
705 	if (mem_cgroup_oom_synchronize(true))
706 		return;
707 
708 	zonelist = node_zonelist(first_memory_node, GFP_KERNEL);
709 	if (oom_zonelist_trylock(zonelist, GFP_KERNEL)) {
710 		out_of_memory(NULL, 0, 0, NULL, false);
711 		oom_zonelist_unlock(zonelist, GFP_KERNEL);
712 	}
713 }
714