oom_kill.c 19.7 KB
Newer Older
L
Linus Torvalds 已提交
1 2 3 4 5 6
/*
 *  linux/mm/oom_kill.c
 * 
 *  Copyright (C)  1998,2000  Rik van Riel
 *	Thanks go out to Claus Fischer for some serious inspiration and
 *	for goading me into coding this file...
D
David Rientjes 已提交
7 8
 *  Copyright (C)  2010  Google, Inc.
 *	Rewritten by David Rientjes
L
Linus Torvalds 已提交
9 10
 *
 *  The routines in this file are used to kill a process when
P
Paul Jackson 已提交
11 12
 *  we're seriously out of memory. This gets called from __alloc_pages()
 *  in mm/page_alloc.c when we really run out of memory.
L
Linus Torvalds 已提交
13 14 15 16 17 18 19
 *
 *  Since we won't call these routines often (on a well-configured
 *  machine) this file will double as a 'coding guide' and a signpost
 *  for newbie kernel hackers. It features several pointers to major
 *  kernel subsystems and hints as to where to find out what things do.
 */

20
#include <linux/oom.h>
L
Linus Torvalds 已提交
21
#include <linux/mm.h>
A
Alexey Dobriyan 已提交
22
#include <linux/err.h>
23
#include <linux/gfp.h>
L
Linus Torvalds 已提交
24 25 26 27
#include <linux/sched.h>
#include <linux/swap.h>
#include <linux/timex.h>
#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 40

#define CREATE_TRACE_POINTS
#include <trace/events/oom.h>
L
Linus Torvalds 已提交
41

42
int sysctl_panic_on_oom;
43
int sysctl_oom_kill_allocating_task;
44
int sysctl_oom_dump_tasks = 1;
D
David Rientjes 已提交
45
static DEFINE_SPINLOCK(zone_scan_lock);
L
Linus Torvalds 已提交
46

47 48 49
#ifdef CONFIG_NUMA
/**
 * has_intersects_mems_allowed() - check task eligiblity for kill
50
 * @start: task struct of which task to consider
51 52 53 54 55
 * @mask: nodemask passed to page allocator for mempolicy ooms
 *
 * Task eligibility is determined by whether or not a candidate task, @tsk,
 * shares the same mempolicy nodes as current if it is bound by such a policy
 * and whether or not it has the same set of allowed cpuset nodes.
56
 */
57
static bool has_intersects_mems_allowed(struct task_struct *start,
58
					const nodemask_t *mask)
59
{
60 61
	struct task_struct *tsk;
	bool ret = false;
62

63
	rcu_read_lock();
64
	for_each_thread(start, tsk) {
65 66 67 68 69 70 71
		if (mask) {
			/*
			 * If this is a mempolicy constrained oom, tsk's
			 * cpuset is irrelevant.  Only return true if its
			 * mempolicy intersects current, otherwise it may be
			 * needlessly killed.
			 */
72
			ret = mempolicy_nodemask_intersects(tsk, mask);
73 74 75 76 77
		} else {
			/*
			 * This is not a mempolicy constrained oom, so only
			 * check the mems of tsk's cpuset.
			 */
78
			ret = cpuset_mems_allowed_intersects(current, tsk);
79
		}
80 81
		if (ret)
			break;
82
	}
83
	rcu_read_unlock();
84

85
	return ret;
86 87 88 89 90 91
}
#else
static bool has_intersects_mems_allowed(struct task_struct *tsk,
					const nodemask_t *mask)
{
	return true;
92
}
93
#endif /* CONFIG_NUMA */
94

95 96 97 98 99 100
/*
 * The process p may have detached its own ->mm while exiting or through
 * use_mm(), but one or more of its subthreads may still have a valid
 * pointer.  Return p, or any of its subthreads with a valid ->mm, with
 * task_lock() held.
 */
101
struct task_struct *find_lock_task_mm(struct task_struct *p)
102
{
103
	struct task_struct *t;
104

105 106
	rcu_read_lock();

107
	for_each_thread(p, t) {
108 109
		task_lock(t);
		if (likely(t->mm))
110
			goto found;
111
		task_unlock(t);
112
	}
113 114 115
	t = NULL;
found:
	rcu_read_unlock();
116

117
	return t;
118 119
}

120
/* return true if the task is not adequate as candidate victim task. */
121
static bool oom_unkillable_task(struct task_struct *p,
122
		const struct mem_cgroup *memcg, const nodemask_t *nodemask)
123 124 125 126 127 128 129
{
	if (is_global_init(p))
		return true;
	if (p->flags & PF_KTHREAD)
		return true;

	/* When mem_cgroup_out_of_memory() and p is not member of the group */
130
	if (memcg && !task_in_mem_cgroup(p, memcg))
131 132 133 134 135 136 137 138 139
		return true;

	/* p may not have freeable memory in nodemask */
	if (!has_intersects_mems_allowed(p, nodemask))
		return true;

	return false;
}

L
Linus Torvalds 已提交
140
/**
D
David Rientjes 已提交
141
 * oom_badness - heuristic function to determine which candidate task to kill
L
Linus Torvalds 已提交
142
 * @p: task struct of which task we should calculate
D
David Rientjes 已提交
143
 * @totalpages: total present RAM allowed for page allocation
L
Linus Torvalds 已提交
144
 *
D
David Rientjes 已提交
145 146 147
 * The heuristic for determining which task to kill is made to be as simple and
 * predictable as possible.  The goal is to return the highest value for the
 * task consuming the most memory to avoid subsequent oom failures.
L
Linus Torvalds 已提交
148
 */
149 150
unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
			  const nodemask_t *nodemask, unsigned long totalpages)
L
Linus Torvalds 已提交
151
{
152
	long points;
153
	long adj;
154

155
	if (oom_unkillable_task(p, memcg, nodemask))
156
		return 0;
L
Linus Torvalds 已提交
157

158 159
	p = find_lock_task_mm(p);
	if (!p)
L
Linus Torvalds 已提交
160 161
		return 0;

162
	adj = (long)p->signal->oom_score_adj;
163
	if (adj == OOM_SCORE_ADJ_MIN) {
164 165 166 167
		task_unlock(p);
		return 0;
	}

L
Linus Torvalds 已提交
168
	/*
D
David Rientjes 已提交
169
	 * The baseline for the badness score is the proportion of RAM that each
170
	 * task's rss, pagetable and swap space use.
L
Linus Torvalds 已提交
171
	 */
172
	points = get_mm_rss(p->mm) + atomic_long_read(&p->mm->nr_ptes) +
173
		 get_mm_counter(p->mm, MM_SWAPENTS);
D
David Rientjes 已提交
174
	task_unlock(p);
L
Linus Torvalds 已提交
175 176

	/*
D
David Rientjes 已提交
177 178
	 * Root processes get 3% bonus, just like the __vm_enough_memory()
	 * implementation used by LSMs.
L
Linus Torvalds 已提交
179
	 */
D
David Rientjes 已提交
180
	if (has_capability_noaudit(p, CAP_SYS_ADMIN))
181
		points -= (points * 3) / 100;
L
Linus Torvalds 已提交
182

183 184 185
	/* Normalize to oom_score_adj units */
	adj *= totalpages / 1000;
	points += adj;
L
Linus Torvalds 已提交
186

187
	/*
188 189
	 * Never return 0 for an eligible task regardless of the root bonus and
	 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
190
	 */
191
	return points > 0 ? points : 1;
L
Linus Torvalds 已提交
192 193
}

194 195 196 197
/*
 * Determine the type of allocation constraint.
 */
#ifdef CONFIG_NUMA
198
static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
D
David Rientjes 已提交
199 200
				gfp_t gfp_mask, nodemask_t *nodemask,
				unsigned long *totalpages)
201
{
202
	struct zone *zone;
203
	struct zoneref *z;
204
	enum zone_type high_zoneidx = gfp_zone(gfp_mask);
D
David Rientjes 已提交
205 206
	bool cpuset_limited = false;
	int nid;
207

D
David Rientjes 已提交
208 209 210 211 212
	/* Default to all available memory */
	*totalpages = totalram_pages + total_swap_pages;

	if (!zonelist)
		return CONSTRAINT_NONE;
213 214 215 216 217 218 219
	/*
	 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
	 * to kill current.We have to random task kill in this case.
	 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
	 */
	if (gfp_mask & __GFP_THISNODE)
		return CONSTRAINT_NONE;
220

221
	/*
D
David Rientjes 已提交
222 223 224
	 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
	 * the page allocator means a mempolicy is in effect.  Cpuset policy
	 * is enforced in get_page_from_freelist().
225
	 */
226
	if (nodemask && !nodes_subset(node_states[N_MEMORY], *nodemask)) {
D
David Rientjes 已提交
227 228 229
		*totalpages = total_swap_pages;
		for_each_node_mask(nid, *nodemask)
			*totalpages += node_spanned_pages(nid);
230
		return CONSTRAINT_MEMORY_POLICY;
D
David Rientjes 已提交
231
	}
232 233 234 235 236

	/* Check this allocation failure is caused by cpuset's wall function */
	for_each_zone_zonelist_nodemask(zone, z, zonelist,
			high_zoneidx, nodemask)
		if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
D
David Rientjes 已提交
237
			cpuset_limited = true;
238

D
David Rientjes 已提交
239 240 241 242 243 244
	if (cpuset_limited) {
		*totalpages = total_swap_pages;
		for_each_node_mask(nid, cpuset_current_mems_allowed)
			*totalpages += node_spanned_pages(nid);
		return CONSTRAINT_CPUSET;
	}
245 246
	return CONSTRAINT_NONE;
}
247 248
#else
static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
D
David Rientjes 已提交
249 250
				gfp_t gfp_mask, nodemask_t *nodemask,
				unsigned long *totalpages)
251
{
D
David Rientjes 已提交
252
	*totalpages = totalram_pages + total_swap_pages;
253 254 255
	return CONSTRAINT_NONE;
}
#endif
256

257 258 259
enum oom_scan_t oom_scan_process_thread(struct task_struct *task,
		unsigned long totalpages, const nodemask_t *nodemask,
		bool force_kill)
260
{
261
	if (oom_unkillable_task(task, NULL, nodemask))
262 263 264 265 266 267 268 269 270 271 272 273 274 275 276
		return OOM_SCAN_CONTINUE;

	/*
	 * This task already has access to memory reserves and is being killed.
	 * Don't allow any other task to have access to the reserves.
	 */
	if (test_tsk_thread_flag(task, TIF_MEMDIE)) {
		if (unlikely(frozen(task)))
			__thaw_task(task);
		if (!force_kill)
			return OOM_SCAN_ABORT;
	}
	if (!task->mm)
		return OOM_SCAN_CONTINUE;

277 278 279 280 281 282 283
	/*
	 * If task is allocating a lot of memory and has been marked to be
	 * killed first if it triggers an oom, then select it.
	 */
	if (oom_task_origin(task))
		return OOM_SCAN_SELECT;

284
	if (task->flags & PF_EXITING && !force_kill) {
285
		/*
286 287
		 * If this task is not being ptraced on exit, then wait for it
		 * to finish before killing some other task unnecessarily.
288
		 */
289 290
		if (!(task->group_leader->ptrace & PT_TRACE_EXIT))
			return OOM_SCAN_ABORT;
291 292 293 294
	}
	return OOM_SCAN_OK;
}

L
Linus Torvalds 已提交
295 296
/*
 * Simple selection loop. We chose the process with the highest
297
 * number of 'points'.  Returns -1 on scan abort.
L
Linus Torvalds 已提交
298 299 300
 *
 * (not docbooked, we don't want this one cluttering up the manual)
 */
D
David Rientjes 已提交
301
static struct task_struct *select_bad_process(unsigned int *ppoints,
302 303
		unsigned long totalpages, const nodemask_t *nodemask,
		bool force_kill)
L
Linus Torvalds 已提交
304
{
305
	struct task_struct *g, *p;
L
Linus Torvalds 已提交
306
	struct task_struct *chosen = NULL;
307
	unsigned long chosen_points = 0;
L
Linus Torvalds 已提交
308

309
	rcu_read_lock();
310
	for_each_process_thread(g, p) {
D
David Rientjes 已提交
311
		unsigned int points;
P
Paul Jackson 已提交
312

313
		switch (oom_scan_process_thread(p, totalpages, nodemask,
314 315 316 317 318 319
						force_kill)) {
		case OOM_SCAN_SELECT:
			chosen = p;
			chosen_points = ULONG_MAX;
			/* fall through */
		case OOM_SCAN_CONTINUE:
320
			continue;
321
		case OOM_SCAN_ABORT:
322
			rcu_read_unlock();
323
			return (struct task_struct *)(-1UL);
324 325 326
		case OOM_SCAN_OK:
			break;
		};
327
		points = oom_badness(p, NULL, nodemask, totalpages);
328 329 330 331 332 333 334 335
		if (!points || points < chosen_points)
			continue;
		/* Prefer thread group leaders for display purposes */
		if (points == chosen_points && thread_group_leader(chosen))
			continue;

		chosen = p;
		chosen_points = points;
336
	}
337 338 339
	if (chosen)
		get_task_struct(chosen);
	rcu_read_unlock();
340

341
	*ppoints = chosen_points * 1000 / totalpages;
L
Linus Torvalds 已提交
342 343 344
	return chosen;
}

345
/**
R
Randy Dunlap 已提交
346
 * dump_tasks - dump current memory state of all system tasks
W
Wanpeng Li 已提交
347
 * @memcg: current's memory controller, if constrained
348
 * @nodemask: nodemask passed to page allocator for mempolicy ooms
R
Randy Dunlap 已提交
349
 *
350 351 352
 * Dumps the current memory state of all eligible tasks.  Tasks not in the same
 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
 * are not shown.
353 354
 * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
 * swapents, oom_score_adj value, and name.
355
 */
356
static void dump_tasks(const struct mem_cgroup *memcg, const nodemask_t *nodemask)
357
{
358 359
	struct task_struct *p;
	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

367 368
		task = find_lock_task_mm(p);
		if (!task) {
369
			/*
370 371
			 * This is a kthread or all of p's threads have already
			 * detached their mm's.  There's no need to report
372
			 * them; they can't be oom killed anyway.
373 374 375
			 */
			continue;
		}
376

377
		pr_info("[%5d] %5d %5d %8lu %8lu %7ld %8lu         %5hd %s\n",
378 379
			task->pid, from_kuid(&init_user_ns, task_uid(task)),
			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),
D
David Rientjes 已提交
382
			task->signal->oom_score_adj, task->comm);
383 384
		task_unlock(task);
	}
385
	rcu_read_unlock();
386 387
}

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",
D
Davidlohr Bueso 已提交
394
		current->comm, gfp_mask, order,
D
David Rientjes 已提交
395
		current->signal->oom_score_adj);
396 397 398
	cpuset_print_task_mems_allowed(current);
	task_unlock(current);
	dump_stack();
399 400 401 402
	if (memcg)
		mem_cgroup_print_oom_info(memcg, p);
	else
		show_mem(SHOW_MEM_FILTER_NODES);
403
	if (sysctl_oom_dump_tasks)
404
		dump_tasks(memcg, nodemask);
405 406
}

407
#define K(x) ((x) << (PAGE_SHIFT-10))
408 409 410 411
/*
 * Must be called while holding a reference to p, which will be released upon
 * returning.
 */
412 413 414 415
void 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)
L
Linus Torvalds 已提交
416
{
417
	struct task_struct *victim = p;
418
	struct task_struct *child;
419
	struct task_struct *t;
420
	struct mm_struct *mm;
421
	unsigned int victim_points = 0;
422 423
	static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
					      DEFAULT_RATELIMIT_BURST);
L
Linus Torvalds 已提交
424

425 426 427 428
	/*
	 * If the task is already exiting, don't alarm the sysadmin or kill
	 * its children or threads, just set TIF_MEMDIE so it can die quickly
	 */
429
	if (p->flags & PF_EXITING) {
430
		set_tsk_thread_flag(p, TIF_MEMDIE);
431
		put_task_struct(p);
432
		return;
433 434
	}

435
	if (__ratelimit(&oom_rs))
436 437
		dump_header(p, gfp_mask, order, memcg, nodemask);

438
	task_lock(p);
D
David Rientjes 已提交
439
	pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
440 441
		message, task_pid_nr(p), p->comm, points);
	task_unlock(p);
N
Nick Piggin 已提交
442

443 444
	/*
	 * If any of p's children has a different mm and is eligible for kill,
445
	 * the one with the highest oom_badness() score is sacrificed for its
446 447 448
	 * parent.  This attempts to lose the minimal amount of work done while
	 * still freeing memory.
	 */
449
	read_lock(&tasklist_lock);
450
	for_each_thread(p, t) {
451
		list_for_each_entry(child, &t->children, sibling) {
D
David Rientjes 已提交
452
			unsigned int child_points;
453

454 455
			if (child->mm == p->mm)
				continue;
D
David Rientjes 已提交
456 457 458
			/*
			 * oom_badness() returns 0 if the thread is unkillable
			 */
459
			child_points = oom_badness(child, memcg, nodemask,
D
David Rientjes 已提交
460
								totalpages);
461
			if (child_points > victim_points) {
462
				put_task_struct(victim);
463 464
				victim = child;
				victim_points = child_points;
465
				get_task_struct(victim);
466
			}
467
		}
468
	}
469
	read_unlock(&tasklist_lock);
470

471 472 473
	p = find_lock_task_mm(victim);
	if (!p) {
		put_task_struct(victim);
474
		return;
475 476 477 478 479
	} else if (victim != p) {
		get_task_struct(p);
		put_task_struct(victim);
		victim = p;
	}
480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497

	/* mm cannot safely be dereferenced after task_unlock(victim) */
	mm = victim->mm;
	pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
		task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
		K(get_mm_counter(victim->mm, MM_ANONPAGES)),
		K(get_mm_counter(victim->mm, MM_FILEPAGES)));
	task_unlock(victim);

	/*
	 * Kill all user processes sharing victim->mm in other thread groups, if
	 * any.  They don't get access to memory reserves, though, to avoid
	 * depletion of all memory.  This prevents mm->mmap_sem livelock when an
	 * oom killed thread cannot exit because it requires the semaphore and
	 * its contended by another thread trying to allocate memory itself.
	 * That thread will now get access to memory reserves since it has a
	 * pending fatal signal.
	 */
498
	rcu_read_lock();
499 500 501 502 503 504 505 506 507 508
	for_each_process(p)
		if (p->mm == mm && !same_thread_group(p, victim) &&
		    !(p->flags & PF_KTHREAD)) {
			if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
				continue;

			task_lock(p);	/* Protect ->comm from prctl() */
			pr_err("Kill process %d (%s) sharing same memory\n",
				task_pid_nr(p), p->comm);
			task_unlock(p);
509
			do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
510
		}
511
	rcu_read_unlock();
512 513

	set_tsk_thread_flag(victim, TIF_MEMDIE);
514
	do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
515
	put_task_struct(victim);
L
Linus Torvalds 已提交
516
}
517
#undef K
L
Linus Torvalds 已提交
518

519 520 521
/*
 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
 */
522 523
void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
			int order, const nodemask_t *nodemask)
524 525 526 527 528 529 530 531 532 533 534 535
{
	if (likely(!sysctl_panic_on_oom))
		return;
	if (sysctl_panic_on_oom != 2) {
		/*
		 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
		 * does not panic for cpuset, mempolicy, or memcg allocation
		 * failures.
		 */
		if (constraint != CONSTRAINT_NONE)
			return;
	}
536
	dump_header(NULL, gfp_mask, order, NULL, nodemask);
537 538 539 540
	panic("Out of memory: %s panic_on_oom is enabled\n",
		sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
}

541 542 543 544 545 546 547 548 549 550 551 552 553 554
static BLOCKING_NOTIFIER_HEAD(oom_notify_list);

int register_oom_notifier(struct notifier_block *nb)
{
	return blocking_notifier_chain_register(&oom_notify_list, nb);
}
EXPORT_SYMBOL_GPL(register_oom_notifier);

int unregister_oom_notifier(struct notifier_block *nb)
{
	return blocking_notifier_chain_unregister(&oom_notify_list, nb);
}
EXPORT_SYMBOL_GPL(unregister_oom_notifier);

D
David Rientjes 已提交
555 556 557 558 559
/*
 * Try to acquire the OOM killer lock for the zones in zonelist.  Returns zero
 * if a parallel OOM killing is already taking place that includes a zone in
 * the zonelist.  Otherwise, locks all zones in the zonelist and returns 1.
 */
560
bool oom_zonelist_trylock(struct zonelist *zonelist, gfp_t gfp_mask)
D
David Rientjes 已提交
561
{
562 563
	struct zoneref *z;
	struct zone *zone;
564
	bool ret = true;
D
David Rientjes 已提交
565

D
David Rientjes 已提交
566
	spin_lock(&zone_scan_lock);
567
	for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask))
J
Johannes Weiner 已提交
568
		if (test_bit(ZONE_OOM_LOCKED, &zone->flags)) {
569
			ret = false;
D
David Rientjes 已提交
570 571
			goto out;
		}
572

573 574 575 576 577
	/*
	 * Lock each zone in the zonelist under zone_scan_lock so a parallel
	 * call to oom_zonelist_trylock() doesn't succeed when it shouldn't.
	 */
	for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask))
J
Johannes Weiner 已提交
578
		set_bit(ZONE_OOM_LOCKED, &zone->flags);
D
David Rientjes 已提交
579 580

out:
D
David Rientjes 已提交
581
	spin_unlock(&zone_scan_lock);
D
David Rientjes 已提交
582 583 584 585 586 587 588 589
	return ret;
}

/*
 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
 * allocation attempts with zonelists containing them may now recall the OOM
 * killer, if necessary.
 */
590
void oom_zonelist_unlock(struct zonelist *zonelist, gfp_t gfp_mask)
D
David Rientjes 已提交
591
{
592 593
	struct zoneref *z;
	struct zone *zone;
D
David Rientjes 已提交
594

D
David Rientjes 已提交
595
	spin_lock(&zone_scan_lock);
596
	for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask))
J
Johannes Weiner 已提交
597
		clear_bit(ZONE_OOM_LOCKED, &zone->flags);
D
David Rientjes 已提交
598
	spin_unlock(&zone_scan_lock);
D
David Rientjes 已提交
599 600
}

L
Linus Torvalds 已提交
601
/**
602
 * out_of_memory - kill the "best" process when we run out of memory
R
Randy Dunlap 已提交
603 604 605
 * @zonelist: zonelist pointer
 * @gfp_mask: memory allocation flags
 * @order: amount of memory being requested as a power of 2
606
 * @nodemask: nodemask passed to page allocator
607
 * @force_kill: true if a task must be killed, even if others are exiting
L
Linus Torvalds 已提交
608 609 610 611 612 613
 *
 * If we run out of memory, we have the choice between either
 * killing a random task (bad), letting the system crash (worse)
 * OR try to be smart about which process to kill. Note that we
 * don't have to be perfect here, we just have to be good.
 */
614
void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
615
		int order, nodemask_t *nodemask, bool force_kill)
L
Linus Torvalds 已提交
616
{
617
	const nodemask_t *mpol_mask;
618
	struct task_struct *p;
D
David Rientjes 已提交
619
	unsigned long totalpages;
620
	unsigned long freed = 0;
621
	unsigned int uninitialized_var(points);
622
	enum oom_constraint constraint = CONSTRAINT_NONE;
K
KOSAKI Motohiro 已提交
623
	int killed = 0;
624 625 626 627 628

	blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
	if (freed > 0)
		/* Got some memory back in the last second. */
		return;
L
Linus Torvalds 已提交
629

630
	/*
631 632 633
	 * If current has a pending SIGKILL or is exiting, then automatically
	 * select it.  The goal is to allow it to allocate so that it may
	 * quickly exit and free its memory.
634
	 */
635
	if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
636 637 638 639
		set_thread_flag(TIF_MEMDIE);
		return;
	}

640 641 642 643
	/*
	 * Check if there were limitations on the allocation (only relevant for
	 * NUMA) that may require different handling.
	 */
D
David Rientjes 已提交
644 645
	constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
						&totalpages);
646 647
	mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
	check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
648

649
	if (sysctl_oom_kill_allocating_task && current->mm &&
650
	    !oom_unkillable_task(current, NULL, nodemask) &&
651
	    current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
652
		get_task_struct(current);
653 654 655 656
		oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL,
				 nodemask,
				 "Out of memory (oom_kill_allocating_task)");
		goto out;
657 658
	}

659
	p = select_bad_process(&points, totalpages, mpol_mask, force_kill);
660 661
	/* Found nothing?!?! Either we hang forever, or we panic. */
	if (!p) {
662
		dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
663 664
		panic("Out of memory and no killable processes...\n");
	}
665
	if (p != (void *)-1UL) {
666 667 668 669
		oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
				 nodemask, "Out of memory");
		killed = 1;
	}
K
KOSAKI Motohiro 已提交
670
out:
L
Linus Torvalds 已提交
671
	/*
672 673
	 * Give the killed threads a good chance of exiting before trying to
	 * allocate memory again.
L
Linus Torvalds 已提交
674
	 */
675 676
	if (killed)
		schedule_timeout_killable(1);
L
Linus Torvalds 已提交
677
}
678 679 680

/*
 * The pagefault handler calls here because it is out of memory, so kill a
681 682
 * memory-hogging task.  If any populated zone has ZONE_OOM_LOCKED set, a
 * parallel oom killing is already in progress so do nothing.
683 684 685
 */
void pagefault_out_of_memory(void)
{
686
	struct zonelist *zonelist;
687

688
	if (mem_cgroup_oom_synchronize(true))
689 690
		return;

691
	zonelist = node_zonelist(first_memory_node, GFP_KERNEL);
692
	if (oom_zonelist_trylock(zonelist, GFP_KERNEL)) {
693
		out_of_memory(NULL, 0, 0, NULL, false);
694
		oom_zonelist_unlock(zonelist, GFP_KERNEL);
695 696
	}
}