oom_kill.c 29.7 KB
Newer Older
Linus Torvalds's avatar
Linus Torvalds committed
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...
7 8
 *  Copyright (C)  2010  Google, Inc.
 *	Rewritten by David Rientjes
Linus Torvalds's avatar
Linus Torvalds committed
9 10
 *
 *  The routines in this file are used to kill a process when
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.
Linus Torvalds's avatar
Linus Torvalds committed
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>
Linus Torvalds's avatar
Linus Torvalds committed
21
#include <linux/mm.h>
Alexey Dobriyan's avatar
Alexey Dobriyan committed
22
#include <linux/err.h>
23
#include <linux/gfp.h>
Linus Torvalds's avatar
Linus Torvalds committed
24
#include <linux/sched.h>
25
#include <linux/sched/mm.h>
26
#include <linux/sched/coredump.h>
27
#include <linux/sched/task.h>
Linus Torvalds's avatar
Linus Torvalds committed
28 29 30
#include <linux/swap.h>
#include <linux/timex.h>
#include <linux/jiffies.h>
31
#include <linux/cpuset.h>
32
#include <linux/export.h>
33
#include <linux/notifier.h>
34
#include <linux/memcontrol.h>
35
#include <linux/mempolicy.h>
36
#include <linux/security.h>
37
#include <linux/ptrace.h>
38
#include <linux/freezer.h>
39
#include <linux/ftrace.h>
40
#include <linux/ratelimit.h>
Michal Hocko's avatar
Michal Hocko committed
41 42
#include <linux/kthread.h>
#include <linux/init.h>
43
#include <linux/mmu_notifier.h>
Michal Hocko's avatar
Michal Hocko committed
44 45 46

#include <asm/tlb.h>
#include "internal.h"
47
#include "slab.h"
48 49 50

#define CREATE_TRACE_POINTS
#include <trace/events/oom.h>
Linus Torvalds's avatar
Linus Torvalds committed
51

52
int sysctl_panic_on_oom;
53
int sysctl_oom_kill_allocating_task;
54
int sysctl_oom_dump_tasks = 1;
55 56

DEFINE_MUTEX(oom_lock);
Linus Torvalds's avatar
Linus Torvalds committed
57

58 59 60
#ifdef CONFIG_NUMA
/**
 * has_intersects_mems_allowed() - check task eligiblity for kill
61
 * @start: task struct of which task to consider
62 63 64 65 66
 * @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.
67
 */
68
static bool has_intersects_mems_allowed(struct task_struct *start,
69
					const nodemask_t *mask)
70
{
71 72
	struct task_struct *tsk;
	bool ret = false;
73

74
	rcu_read_lock();
75
	for_each_thread(start, tsk) {
76 77 78 79 80 81 82
		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.
			 */
83
			ret = mempolicy_nodemask_intersects(tsk, mask);
84 85 86 87 88
		} else {
			/*
			 * This is not a mempolicy constrained oom, so only
			 * check the mems of tsk's cpuset.
			 */
89
			ret = cpuset_mems_allowed_intersects(current, tsk);
90
		}
91 92
		if (ret)
			break;
93
	}
94
	rcu_read_unlock();
95

96
	return ret;
97 98 99 100 101 102
}
#else
static bool has_intersects_mems_allowed(struct task_struct *tsk,
					const nodemask_t *mask)
{
	return true;
103
}
104
#endif /* CONFIG_NUMA */
105

106 107 108 109 110 111
/*
 * 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.
 */
112
struct task_struct *find_lock_task_mm(struct task_struct *p)
113
{
114
	struct task_struct *t;
115

116 117
	rcu_read_lock();

118
	for_each_thread(p, t) {
119 120
		task_lock(t);
		if (likely(t->mm))
121
			goto found;
122
		task_unlock(t);
123
	}
124 125 126
	t = NULL;
found:
	rcu_read_unlock();
127

128
	return t;
129 130
}

131 132 133 134 135 136 137 138 139
/*
 * order == -1 means the oom kill is required by sysrq, otherwise only
 * for display purposes.
 */
static inline bool is_sysrq_oom(struct oom_control *oc)
{
	return oc->order == -1;
}

140 141 142 143 144
static inline bool is_memcg_oom(struct oom_control *oc)
{
	return oc->memcg != NULL;
}

145
/* return true if the task is not adequate as candidate victim task. */
146
static bool oom_unkillable_task(struct task_struct *p,
147
		struct mem_cgroup *memcg, const nodemask_t *nodemask)
148 149 150 151 152 153 154
{
	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 */
155
	if (memcg && !task_in_mem_cgroup(p, memcg))
156 157 158 159 160 161 162 163 164
		return true;

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

	return false;
}

165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183
/*
 * Print out unreclaimble slabs info when unreclaimable slabs amount is greater
 * than all user memory (LRU pages)
 */
static bool is_dump_unreclaim_slabs(void)
{
	unsigned long nr_lru;

	nr_lru = global_node_page_state(NR_ACTIVE_ANON) +
		 global_node_page_state(NR_INACTIVE_ANON) +
		 global_node_page_state(NR_ACTIVE_FILE) +
		 global_node_page_state(NR_INACTIVE_FILE) +
		 global_node_page_state(NR_ISOLATED_ANON) +
		 global_node_page_state(NR_ISOLATED_FILE) +
		 global_node_page_state(NR_UNEVICTABLE);

	return (global_node_page_state(NR_SLAB_UNRECLAIMABLE) > nr_lru);
}

Linus Torvalds's avatar
Linus Torvalds committed
184
/**
185
 * oom_badness - heuristic function to determine which candidate task to kill
Linus Torvalds's avatar
Linus Torvalds committed
186
 * @p: task struct of which task we should calculate
187
 * @totalpages: total present RAM allowed for page allocation
Linus Torvalds's avatar
Linus Torvalds committed
188
 *
189 190 191
 * 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.
Linus Torvalds's avatar
Linus Torvalds committed
192
 */
193 194
unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
			  const nodemask_t *nodemask, unsigned long totalpages)
Linus Torvalds's avatar
Linus Torvalds committed
195
{
196
	long points;
197
	long adj;
198

199
	if (oom_unkillable_task(p, memcg, nodemask))
200
		return 0;
Linus Torvalds's avatar
Linus Torvalds committed
201

202 203
	p = find_lock_task_mm(p);
	if (!p)
Linus Torvalds's avatar
Linus Torvalds committed
204 205
		return 0;

206 207
	/*
	 * Do not even consider tasks which are explicitly marked oom
208 209
	 * unkillable or have been already oom reaped or the are in
	 * the middle of vfork
210
	 */
211
	adj = (long)p->signal->oom_score_adj;
212
	if (adj == OOM_SCORE_ADJ_MIN ||
213
			test_bit(MMF_OOM_SKIP, &p->mm->flags) ||
214
			in_vfork(p)) {
215 216 217 218
		task_unlock(p);
		return 0;
	}

Linus Torvalds's avatar
Linus Torvalds committed
219
	/*
220
	 * The baseline for the badness score is the proportion of RAM that each
221
	 * task's rss, pagetable and swap space use.
Linus Torvalds's avatar
Linus Torvalds committed
222
	 */
223
	points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
224
		mm_pgtables_bytes(p->mm) / PAGE_SIZE;
225
	task_unlock(p);
Linus Torvalds's avatar
Linus Torvalds committed
226 227

	/*
228 229
	 * Root processes get 3% bonus, just like the __vm_enough_memory()
	 * implementation used by LSMs.
Linus Torvalds's avatar
Linus Torvalds committed
230
	 */
231
	if (has_capability_noaudit(p, CAP_SYS_ADMIN))
232
		points -= (points * 3) / 100;
Linus Torvalds's avatar
Linus Torvalds committed
233

234 235 236
	/* Normalize to oom_score_adj units */
	adj *= totalpages / 1000;
	points += adj;
Linus Torvalds's avatar
Linus Torvalds committed
237

238
	/*
239 240
	 * 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).
241
	 */
242
	return points > 0 ? points : 1;
Linus Torvalds's avatar
Linus Torvalds committed
243 244
}

245 246 247 248 249 250 251
enum oom_constraint {
	CONSTRAINT_NONE,
	CONSTRAINT_CPUSET,
	CONSTRAINT_MEMORY_POLICY,
	CONSTRAINT_MEMCG,
};

252 253 254
/*
 * Determine the type of allocation constraint.
 */
255
static enum oom_constraint constrained_alloc(struct oom_control *oc)
256
{
257
	struct zone *zone;
258
	struct zoneref *z;
259
	enum zone_type high_zoneidx = gfp_zone(oc->gfp_mask);
260 261
	bool cpuset_limited = false;
	int nid;
262

263 264 265 266 267
	if (is_memcg_oom(oc)) {
		oc->totalpages = mem_cgroup_get_limit(oc->memcg) ?: 1;
		return CONSTRAINT_MEMCG;
	}

268
	/* Default to all available memory */
269 270 271 272
	oc->totalpages = totalram_pages + total_swap_pages;

	if (!IS_ENABLED(CONFIG_NUMA))
		return CONSTRAINT_NONE;
273

274
	if (!oc->zonelist)
275
		return CONSTRAINT_NONE;
276 277 278 279 280
	/*
	 * 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.
	 */
281
	if (oc->gfp_mask & __GFP_THISNODE)
282
		return CONSTRAINT_NONE;
283

284
	/*
285 286 287
	 * 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().
288
	 */
289 290
	if (oc->nodemask &&
	    !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) {
291
		oc->totalpages = total_swap_pages;
292
		for_each_node_mask(nid, *oc->nodemask)
293
			oc->totalpages += node_spanned_pages(nid);
294
		return CONSTRAINT_MEMORY_POLICY;
295
	}
296 297

	/* Check this allocation failure is caused by cpuset's wall function */
298 299 300
	for_each_zone_zonelist_nodemask(zone, z, oc->zonelist,
			high_zoneidx, oc->nodemask)
		if (!cpuset_zone_allowed(zone, oc->gfp_mask))
301
			cpuset_limited = true;
302

303
	if (cpuset_limited) {
304
		oc->totalpages = total_swap_pages;
305
		for_each_node_mask(nid, cpuset_current_mems_allowed)
306
			oc->totalpages += node_spanned_pages(nid);
307 308
		return CONSTRAINT_CPUSET;
	}
309 310 311
	return CONSTRAINT_NONE;
}

312
static int oom_evaluate_task(struct task_struct *task, void *arg)
313
{
314 315 316
	struct oom_control *oc = arg;
	unsigned long points;

317
	if (oom_unkillable_task(task, NULL, oc->nodemask))
318
		goto next;
319 320 321

	/*
	 * This task already has access to memory reserves and is being killed.
322
	 * Don't allow any other task to have access to the reserves unless
323
	 * the task has MMF_OOM_SKIP because chances that it would release
324
	 * any memory is quite low.
325
	 */
326 327
	if (!is_sysrq_oom(oc) && tsk_is_oom_victim(task)) {
		if (test_bit(MMF_OOM_SKIP, &task->signal->oom_mm->flags))
328 329
			goto next;
		goto abort;
330
	}
331

332 333 334 335
	/*
	 * If task is allocating a lot of memory and has been marked to be
	 * killed first if it triggers an oom, then select it.
	 */
336 337 338 339
	if (oom_task_origin(task)) {
		points = ULONG_MAX;
		goto select;
	}
340

341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360
	points = oom_badness(task, NULL, oc->nodemask, oc->totalpages);
	if (!points || points < oc->chosen_points)
		goto next;

	/* Prefer thread group leaders for display purposes */
	if (points == oc->chosen_points && thread_group_leader(oc->chosen))
		goto next;
select:
	if (oc->chosen)
		put_task_struct(oc->chosen);
	get_task_struct(task);
	oc->chosen = task;
	oc->chosen_points = points;
next:
	return 0;
abort:
	if (oc->chosen)
		put_task_struct(oc->chosen);
	oc->chosen = (void *)-1UL;
	return 1;
361 362
}

Linus Torvalds's avatar
Linus Torvalds committed
363
/*
364 365
 * Simple selection loop. We choose the process with the highest number of
 * 'points'. In case scan was aborted, oc->chosen is set to -1.
Linus Torvalds's avatar
Linus Torvalds committed
366
 */
367
static void select_bad_process(struct oom_control *oc)
Linus Torvalds's avatar
Linus Torvalds committed
368
{
369 370 371 372
	if (is_memcg_oom(oc))
		mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc);
	else {
		struct task_struct *p;
373

374 375 376 377 378
		rcu_read_lock();
		for_each_process(p)
			if (oom_evaluate_task(p, oc))
				break;
		rcu_read_unlock();
379
	}
380

381
	oc->chosen_points = oc->chosen_points * 1000 / oc->totalpages;
Linus Torvalds's avatar
Linus Torvalds committed
382 383
}

384
/**
Randy Dunlap's avatar
Randy Dunlap committed
385
 * dump_tasks - dump current memory state of all system tasks
Wanpeng Li's avatar
Wanpeng Li committed
386
 * @memcg: current's memory controller, if constrained
387
 * @nodemask: nodemask passed to page allocator for mempolicy ooms
Randy Dunlap's avatar
Randy Dunlap committed
388
 *
389 390 391
 * 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.
392 393
 * State information includes task's pid, uid, tgid, vm size, rss,
 * pgtables_bytes, swapents, oom_score_adj value, and name.
394
 */
395
static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask)
396
{
397 398
	struct task_struct *p;
	struct task_struct *task;
399

400
	pr_info("[ pid ]   uid  tgid total_vm      rss pgtables_bytes swapents oom_score_adj name\n");
401
	rcu_read_lock();
402
	for_each_process(p) {
403
		if (oom_unkillable_task(p, memcg, nodemask))
404
			continue;
405

406 407
		task = find_lock_task_mm(p);
		if (!task) {
408
			/*
409 410
			 * This is a kthread or all of p's threads have already
			 * detached their mm's.  There's no need to report
411
			 * them; they can't be oom killed anyway.
412 413 414
			 */
			continue;
		}
415

416
		pr_info("[%5d] %5d %5d %8lu %8lu %8ld %8lu         %5hd %s\n",
417 418
			task->pid, from_kuid(&init_user_ns, task_uid(task)),
			task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
419
			mm_pgtables_bytes(task->mm),
420
			get_mm_counter(task->mm, MM_SWAPENTS),
421
			task->signal->oom_score_adj, task->comm);
422 423
		task_unlock(task);
	}
424
	rcu_read_unlock();
425 426
}

427
static void dump_header(struct oom_control *oc, struct task_struct *p)
428
{
Michal Hocko's avatar
Michal Hocko committed
429 430 431 432
	pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), nodemask=%*pbl, order=%d, oom_score_adj=%hd\n",
		current->comm, oc->gfp_mask, &oc->gfp_mask,
		nodemask_pr_args(oc->nodemask), oc->order,
			current->signal->oom_score_adj);
433 434
	if (!IS_ENABLED(CONFIG_COMPACTION) && oc->order)
		pr_warn("COMPACTION is disabled!!!\n");
435

436
	cpuset_print_current_mems_allowed();
437
	dump_stack();
438
	if (is_memcg_oom(oc))
439
		mem_cgroup_print_oom_info(oc->memcg, p);
440
	else {
441
		show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask);
442 443 444
		if (is_dump_unreclaim_slabs())
			dump_unreclaimable_slab();
	}
445
	if (sysctl_oom_dump_tasks)
446
		dump_tasks(oc->memcg, oc->nodemask);
447 448
}

449
/*
450
 * Number of OOM victims in flight
451
 */
452 453
static atomic_t oom_victims = ATOMIC_INIT(0);
static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
454

455
static bool oom_killer_disabled __read_mostly;
456

457 458
#define K(x) ((x) << (PAGE_SHIFT-10))

459 460 461 462 463 464
/*
 * task->mm can be NULL if the task is the exited group leader.  So to
 * determine whether the task is using a particular mm, we examine all the
 * task's threads: if one of those is using this mm then this task was also
 * using it.
 */
465
bool process_shares_mm(struct task_struct *p, struct mm_struct *mm)
466 467 468 469 470 471 472 473 474 475 476 477
{
	struct task_struct *t;

	for_each_thread(p, t) {
		struct mm_struct *t_mm = READ_ONCE(t->mm);
		if (t_mm)
			return t_mm == mm;
	}
	return false;
}


Michal Hocko's avatar
Michal Hocko committed
478 479 480 481 482 483 484
#ifdef CONFIG_MMU
/*
 * OOM Reaper kernel thread which tries to reap the memory used by the OOM
 * victim (if that is possible) to help the OOM killer to move on.
 */
static struct task_struct *oom_reaper_th;
static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait);
485
static struct task_struct *oom_reaper_list;
486 487
static DEFINE_SPINLOCK(oom_reaper_lock);

488
static bool __oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
Michal Hocko's avatar
Michal Hocko committed
489 490 491 492 493
{
	struct mmu_gather tlb;
	struct vm_area_struct *vma;
	bool ret = true;

494 495 496
	/*
	 * We have to make sure to not race with the victim exit path
	 * and cause premature new oom victim selection:
497
	 * __oom_reap_task_mm		exit_mm
498
	 *   mmget_not_zero
499 500 501 502 503 504 505 506 507 508 509
	 *				  mmput
	 *				    atomic_dec_and_test
	 *				  exit_oom_victim
	 *				[...]
	 *				out_of_memory
	 *				  select_bad_process
	 *				    # no TIF_MEMDIE task selects new victim
	 *  unmap_page_range # frees some memory
	 */
	mutex_lock(&oom_lock);

Michal Hocko's avatar
Michal Hocko committed
510 511
	if (!down_read_trylock(&mm->mmap_sem)) {
		ret = false;
512
		trace_skip_task_reaping(tsk->pid);
513
		goto unlock_oom;
514 515
	}

516 517 518 519 520 521 522 523 524 525 526 527 528 529 530
	/*
	 * If the mm has notifiers then we would need to invalidate them around
	 * unmap_page_range and that is risky because notifiers can sleep and
	 * what they do is basically undeterministic.  So let's have a short
	 * sleep to give the oom victim some more time.
	 * TODO: we really want to get rid of this ugly hack and make sure that
	 * notifiers cannot block for unbounded amount of time and add
	 * mmu_notifier_invalidate_range_{start,end} around unmap_page_range
	 */
	if (mm_has_notifiers(mm)) {
		up_read(&mm->mmap_sem);
		schedule_timeout_idle(HZ);
		goto unlock_oom;
	}

531
	/*
532 533 534 535
	 * MMF_OOM_SKIP is set by exit_mmap when the OOM reaper can't
	 * work on the mm anymore. The check for MMF_OOM_SKIP must run
	 * under mmap_sem for reading because it serializes against the
	 * down_write();up_write() cycle in exit_mmap().
536
	 */
537
	if (test_bit(MMF_OOM_SKIP, &mm->flags)) {
538
		up_read(&mm->mmap_sem);
539
		trace_skip_task_reaping(tsk->pid);
540
		goto unlock_oom;
Michal Hocko's avatar
Michal Hocko committed
541 542
	}

543 544
	trace_start_task_reaping(tsk->pid);

545 546 547 548 549 550 551 552
	/*
	 * Tell all users of get_user/copy_from_user etc... that the content
	 * is no longer stable. No barriers really needed because unmapping
	 * should imply barriers already and the reader would hit a page fault
	 * if it stumbled over a reaped memory.
	 */
	set_bit(MMF_UNSTABLE, &mm->flags);

Michal Hocko's avatar
Michal Hocko committed
553
	for (vma = mm->mmap ; vma; vma = vma->vm_next) {
554
		if (!can_madv_dontneed_vma(vma))
Michal Hocko's avatar
Michal Hocko committed
555 556 557 558 559 560 561 562 563 564 565 566
			continue;

		/*
		 * Only anonymous pages have a good chance to be dropped
		 * without additional steps which we cannot afford as we
		 * are OOM already.
		 *
		 * We do not even care about fs backed pages because all
		 * which are reclaimable have already been reclaimed and
		 * we do not want to block exit_mmap by keeping mm ref
		 * count elevated without a good reason.
		 */
567 568
		if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) {
			tlb_gather_mmu(&tlb, mm, vma->vm_start, vma->vm_end);
Michal Hocko's avatar
Michal Hocko committed
569
			unmap_page_range(&tlb, vma, vma->vm_start, vma->vm_end,
570
					 NULL);
571 572
			tlb_finish_mmu(&tlb, vma->vm_start, vma->vm_end);
		}
Michal Hocko's avatar
Michal Hocko committed
573
	}
574 575 576 577 578
	pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
			task_pid_nr(tsk), tsk->comm,
			K(get_mm_counter(mm, MM_ANONPAGES)),
			K(get_mm_counter(mm, MM_FILEPAGES)),
			K(get_mm_counter(mm, MM_SHMEMPAGES)));
Michal Hocko's avatar
Michal Hocko committed
579
	up_read(&mm->mmap_sem);
580

581
	trace_finish_task_reaping(tsk->pid);
582 583
unlock_oom:
	mutex_unlock(&oom_lock);
Michal Hocko's avatar
Michal Hocko committed
584 585 586
	return ret;
}

587
#define MAX_OOM_REAP_RETRIES 10
588
static void oom_reap_task(struct task_struct *tsk)
Michal Hocko's avatar
Michal Hocko committed
589 590
{
	int attempts = 0;
591
	struct mm_struct *mm = tsk->signal->oom_mm;
Michal Hocko's avatar
Michal Hocko committed
592 593

	/* Retry the down_read_trylock(mmap_sem) a few times */
594
	while (attempts++ < MAX_OOM_REAP_RETRIES && !__oom_reap_task_mm(tsk, mm))
Michal Hocko's avatar
Michal Hocko committed
595 596
		schedule_timeout_idle(HZ/10);

597 598
	if (attempts <= MAX_OOM_REAP_RETRIES)
		goto done;
599

600

601 602 603
	pr_info("oom_reaper: unable to reap pid:%d (%s)\n",
		task_pid_nr(tsk), tsk->comm);
	debug_show_all_locks();
604

605
done:
606 607
	tsk->oom_reaper_list = NULL;

608 609 610 611
	/*
	 * Hide this mm from OOM killer because it has been either reaped or
	 * somebody can't call up_write(mmap_sem).
	 */
612
	set_bit(MMF_OOM_SKIP, &mm->flags);
613

Michal Hocko's avatar
Michal Hocko committed
614
	/* Drop a reference taken by wake_oom_reaper */
615
	put_task_struct(tsk);
Michal Hocko's avatar
Michal Hocko committed
616 617 618 619 620
}

static int oom_reaper(void *unused)
{
	while (true) {
621
		struct task_struct *tsk = NULL;
Michal Hocko's avatar
Michal Hocko committed
622

623
		wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL);
624
		spin_lock(&oom_reaper_lock);
625 626 627
		if (oom_reaper_list != NULL) {
			tsk = oom_reaper_list;
			oom_reaper_list = tsk->oom_reaper_list;
628 629 630 631 632
		}
		spin_unlock(&oom_reaper_lock);

		if (tsk)
			oom_reap_task(tsk);
Michal Hocko's avatar
Michal Hocko committed
633 634 635 636 637
	}

	return 0;
}

638
static void wake_oom_reaper(struct task_struct *tsk)
Michal Hocko's avatar
Michal Hocko committed
639
{
640 641
	/* tsk is already queued? */
	if (tsk == oom_reaper_list || tsk->oom_reaper_list)
Michal Hocko's avatar
Michal Hocko committed
642 643
		return;

644
	get_task_struct(tsk);
Michal Hocko's avatar
Michal Hocko committed
645

646
	spin_lock(&oom_reaper_lock);
647 648
	tsk->oom_reaper_list = oom_reaper_list;
	oom_reaper_list = tsk;
649
	spin_unlock(&oom_reaper_lock);
650
	trace_wake_reaper(tsk->pid);
651
	wake_up(&oom_reaper_wait);
Michal Hocko's avatar
Michal Hocko committed
652 653 654 655 656 657 658 659
}

static int __init oom_init(void)
{
	oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper");
	return 0;
}
subsys_initcall(oom_init)
660 661 662 663 664
#else
static inline void wake_oom_reaper(struct task_struct *tsk)
{
}
#endif /* CONFIG_MMU */
Michal Hocko's avatar
Michal Hocko committed
665

666
/**
667
 * mark_oom_victim - mark the given task as OOM victim
668
 * @tsk: task to mark
669
 *
670
 * Has to be called with oom_lock held and never after
671
 * oom has been disabled already.
672 673 674
 *
 * tsk->mm has to be non NULL and caller has to guarantee it is stable (either
 * under task_lock or operate on the current).
675
 */
676
static void mark_oom_victim(struct task_struct *tsk)
677
{
678 679
	struct mm_struct *mm = tsk->mm;

680 681 682 683
	WARN_ON(oom_killer_disabled);
	/* OOM killer might race with memcg OOM */
	if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
		return;
684 685 686

	/* oom_mm is bound to the signal struct life time. */
	if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm))
Vegard Nossum's avatar
Vegard Nossum committed
687
		mmgrab(tsk->signal->oom_mm);
688

689 690 691 692 693 694 695
	/*
	 * Make sure that the task is woken up from uninterruptible sleep
	 * if it is frozen because OOM killer wouldn't be able to free
	 * any memory and livelock. freezing_slow_path will tell the freezer
	 * that TIF_MEMDIE tasks should be ignored.
	 */
	__thaw_task(tsk);
696
	atomic_inc(&oom_victims);
697
	trace_mark_victim(tsk->pid);
698 699 700
}

/**
701
 * exit_oom_victim - note the exit of an OOM victim
702
 */
703
void exit_oom_victim(void)
704
{
705
	clear_thread_flag(TIF_MEMDIE);
706

707
	if (!atomic_dec_return(&oom_victims))
708 709 710
		wake_up_all(&oom_victims_wait);
}

711 712 713 714 715 716
/**
 * oom_killer_enable - enable OOM killer
 */
void oom_killer_enable(void)
{
	oom_killer_disabled = false;
717
	pr_info("OOM killer enabled.\n");
718 719
}

720 721
/**
 * oom_killer_disable - disable OOM killer
722
 * @timeout: maximum timeout to wait for oom victims in jiffies
723 724
 *
 * Forces all page allocations to fail rather than trigger OOM killer.
725 726
 * Will block and wait until all OOM victims are killed or the given
 * timeout expires.
727 728 729 730 731 732 733 734
 *
 * The function cannot be called when there are runnable user tasks because
 * the userspace would see unexpected allocation failures as a result. Any
 * new usage of this function should be consulted with MM people.
 *
 * Returns true if successful and false if the OOM killer cannot be
 * disabled.
 */
735
bool oom_killer_disable(signed long timeout)
736
{
737 738
	signed long ret;

739
	/*
740 741
	 * Make sure to not race with an ongoing OOM killer. Check that the
	 * current is not killed (possibly due to sharing the victim's memory).
742
	 */
743
	if (mutex_lock_killable(&oom_lock))
744 745
		return false;
	oom_killer_disabled = true;
746
	mutex_unlock(&oom_lock);
747

748 749 750 751 752 753
	ret = wait_event_interruptible_timeout(oom_victims_wait,
			!atomic_read(&oom_victims), timeout);
	if (ret <= 0) {
		oom_killer_enable();
		return false;
	}
754
	pr_info("OOM killer disabled.\n");
755 756 757 758

	return true;
}

759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783
static inline bool __task_will_free_mem(struct task_struct *task)
{
	struct signal_struct *sig = task->signal;

	/*
	 * A coredumping process may sleep for an extended period in exit_mm(),
	 * so the oom killer cannot assume that the process will promptly exit
	 * and release memory.
	 */
	if (sig->flags & SIGNAL_GROUP_COREDUMP)
		return false;

	if (sig->flags & SIGNAL_GROUP_EXIT)
		return true;

	if (thread_group_empty(task) && (task->flags & PF_EXITING))
		return true;

	return false;
}

/*
 * Checks whether the given task is dying or exiting and likely to
 * release its address space. This means that all threads and processes
 * sharing the same mm have to be killed or exiting.
784 785
 * Caller has to make sure that task->mm is stable (hold task_lock or
 * it operates on the current).
786
 */
787
static bool task_will_free_mem(struct task_struct *task)
788
{
789
	struct mm_struct *mm = task->mm;
790
	struct task_struct *p;
791
	bool ret = true;
792 793

	/*
794 795 796
	 * Skip tasks without mm because it might have passed its exit_mm and
	 * exit_oom_victim. oom_reaper could have rescued that but do not rely
	 * on that for now. We can consider find_lock_task_mm in future.
797
	 */
798
	if (!mm)
799 800
		return false;

801 802
	if (!__task_will_free_mem(task))
		return false;
803 804 805 806 807

	/*
	 * This task has already been drained by the oom reaper so there are
	 * only small chances it will free some more
	 */
808
	if (test_bit(MMF_OOM_SKIP, &mm->flags))
809 810
		return false;

811
	if (atomic_read(&mm->mm_users) <= 1)
812 813 814
		return true;

	/*
815 816 817
	 * Make sure that all tasks which share the mm with the given tasks
	 * are dying as well to make sure that a) nobody pins its mm and
	 * b) the task is also reapable by the oom reaper.
818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833
	 */
	rcu_read_lock();
	for_each_process(p) {
		if (!process_shares_mm(p, mm))
			continue;
		if (same_thread_group(task, p))
			continue;
		ret = __task_will_free_mem(p);
		if (!ret)
			break;
	}
	rcu_read_unlock();

	return ret;
}

834
static void oom_kill_process(struct oom_control *oc, const char *message)
Linus Torvalds's avatar
Linus Torvalds committed
835
{
836 837
	struct task_struct *p = oc->chosen;
	unsigned int points = oc->chosen_points;
838
	struct task_struct *victim = p;
839
	struct task_struct *child;
840
	struct task_struct *t;
841
	struct mm_struct *mm;
842
	unsigned int victim_points = 0;
843 844
	static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
					      DEFAULT_RATELIMIT_BURST);
845
	bool can_oom_reap = true;
Linus Torvalds's avatar
Linus Torvalds committed
846

847 848
	/*
	 * If the task is already exiting, don't alarm the sysadmin or kill
849 850
	 * its children or threads, just give it access to memory reserves
	 * so it can die quickly
851
	 */
852
	task_lock(p);
853
	if (task_will_free_mem(p)) {
854
		mark_oom_victim(p);
855
		wake_oom_reaper(p);
856
		task_unlock(p);
857
		put_task_struct(p);
858
		return;
859
	}
860
	task_unlock(p);
861

862
	if (__ratelimit(&oom_rs))
863
		dump_header(oc, p);
864

865
	pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n",
866
		message, task_pid_nr(p), p->comm, points);
Nick Piggin's avatar
Nick Piggin committed
867

868 869
	/*
	 * If any of p's children has a different mm and is eligible for kill,
870
	 * the one with the highest oom_badness() score is sacrificed for its
871 872 873
	 * parent.  This attempts to lose the minimal amount of work done while
	 * still freeing memory.
	 */
874
	read_lock(&tasklist_lock);
875
	for_each_thread(p, t) {
876
		list_for_each_entry(child, &t->children, sibling) {
877
			unsigned int child_points;
878

879
			if (process_shares_mm(child, p->mm))
880
				continue;
881 882 883
			/*
			 * oom_badness() returns 0 if the thread is unkillable
			 */
884
			child_points = oom_badness(child,
885
				oc->memcg, oc->nodemask, oc->totalpages);
886
			if (child_points > victim_points) {
887
				put_task_struct(victim);
888 889
				victim = child;
				victim_points = child_points;
890
				get_task_struct(victim);
891
			}
892
		}
893
	}
894
	read_unlock(&tasklist_lock);
895

896 897 898
	p = find_lock_task_mm(victim);
	if (!p) {
		put_task_struct(victim);
899
		return;
900 901 902 903 904
	} else if (victim != p) {
		get_task_struct(p);
		put_task_struct(victim);
		victim = p;
	}
905

906
	/* Get a reference to safely compare mm after task_unlock(victim) */
907
	mm = victim->mm;
Vegard Nossum's avatar
Vegard Nossum committed
908
	mmgrab(mm);
909 910 911 912 913

	/* Raise event before sending signal: task reaper must see this */
	count_vm_event(OOM_KILL);
	count_memcg_event_mm(mm, OOM_KILL);

914
	/*
915 916 917
	 * We should send SIGKILL before granting access to memory reserves
	 * in order to prevent the OOM victim from depleting the memory
	 * reserves from the user space under its control.
918 919
	 */
	do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
920
	mark_oom_victim(victim);
921
	pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
922 923
		task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
		K(get_mm_counter(victim->mm, MM_ANONPAGES)),
924 925
		K(get_mm_counter(victim->mm, MM_FILEPAGES)),
		K(get_mm_counter(victim->mm, MM_SHMEMPAGES)));
926 927 928 929 930 931 932 933 934 935 936
	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.
	 */
937
	rcu_read_lock();
938
	for_each_process(p) {
939
		if (!process_shares_mm(p, mm))
940 941 942
			continue;
		if (same_thread_group(p, victim))
			continue;
943
		if (is_global_init(p)) {
Michal Hocko's avatar
Michal Hocko committed
944
			can_oom_reap = false;
945
			set_bit(MMF_OOM_SKIP, &mm->flags);
946 947 948
			pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n",
					task_pid_nr(victim), victim->comm,
					task_pid_nr(p), p->comm);
949
			continue;
Michal Hocko's avatar
Michal Hocko committed
950
		}
951 952 953 954 955 956
		/*
		 * No use_mm() user needs to read from the userspace so we are
		 * ok to reap it.
		 */
		if (unlikely(p->flags & PF_KTHREAD))
			continue;
957 958
		do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
	}
959
	rcu_read_unlock();
960

Michal Hocko's avatar
Michal Hocko committed
961
	if (can_oom_reap)
962
		wake_oom_reaper(victim);
Michal Hocko's avatar
Michal Hocko committed
963

964
	mmdrop(mm);
965
	put_task_struct(victim);
Linus Torvalds's avatar
Linus Torvalds committed
966
}
967
#undef K
Linus Torvalds's avatar
Linus Torvalds committed
968

969 970 971
/*
 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
 */
972 973
static void check_panic_on_oom(struct oom_control *oc,
			       enum oom_constraint constraint)
974 975 976 977 978 979 980 981 982 983 984 985
{
	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;
	}
986
	/* Do not panic for oom kills triggered by sysrq */
987
	if (is_sysrq_oom(oc))
988
		return;
989
	dump_header(oc, NULL);
990 991 992 993
	panic("Out of memory: %s panic_on_oom is enabled\n",
		sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
}

994 995 996 997 998 999 1000 1001 1002 1003 1004 1005