oom: remove oom_disable_count
[linux-2.6-block.git] / mm / oom_kill.c
CommitLineData
1da177e4
LT
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...
a63d83f4
DR
7 * Copyright (C) 2010 Google, Inc.
8 * Rewritten by David Rientjes
1da177e4
LT
9 *
10 * The routines in this file are used to kill a process when
a49335cc
PJ
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.
1da177e4
LT
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
8ac773b4 20#include <linux/oom.h>
1da177e4 21#include <linux/mm.h>
4e950f6f 22#include <linux/err.h>
5a0e3ad6 23#include <linux/gfp.h>
1da177e4
LT
24#include <linux/sched.h>
25#include <linux/swap.h>
26#include <linux/timex.h>
27#include <linux/jiffies.h>
ef08e3b4 28#include <linux/cpuset.h>
8bc719d3
MS
29#include <linux/module.h>
30#include <linux/notifier.h>
c7ba5c9e 31#include <linux/memcontrol.h>
6f48d0eb 32#include <linux/mempolicy.h>
5cd9c58f 33#include <linux/security.h>
edd45544 34#include <linux/ptrace.h>
f660daac 35#include <linux/freezer.h>
1da177e4 36
fadd8fbd 37int sysctl_panic_on_oom;
fe071d7e 38int sysctl_oom_kill_allocating_task;
ad915c43 39int sysctl_oom_dump_tasks = 1;
c7d4caeb 40static DEFINE_SPINLOCK(zone_scan_lock);
1da177e4 41
72788c38
DR
42/**
43 * test_set_oom_score_adj() - set current's oom_score_adj and return old value
44 * @new_val: new oom_score_adj value
45 *
46 * Sets the oom_score_adj value for current to @new_val with proper
47 * synchronization and returns the old value. Usually used to temporarily
48 * set a value, save the old value in the caller, and then reinstate it later.
49 */
50int test_set_oom_score_adj(int new_val)
51{
52 struct sighand_struct *sighand = current->sighand;
53 int old_val;
54
55 spin_lock_irq(&sighand->siglock);
56 old_val = current->signal->oom_score_adj;
c9f01245 57 current->signal->oom_score_adj = new_val;
72788c38
DR
58 spin_unlock_irq(&sighand->siglock);
59
60 return old_val;
61}
62
6f48d0eb
DR
63#ifdef CONFIG_NUMA
64/**
65 * has_intersects_mems_allowed() - check task eligiblity for kill
66 * @tsk: task struct of which task to consider
67 * @mask: nodemask passed to page allocator for mempolicy ooms
68 *
69 * Task eligibility is determined by whether or not a candidate task, @tsk,
70 * shares the same mempolicy nodes as current if it is bound by such a policy
71 * and whether or not it has the same set of allowed cpuset nodes.
495789a5 72 */
6f48d0eb
DR
73static bool has_intersects_mems_allowed(struct task_struct *tsk,
74 const nodemask_t *mask)
495789a5 75{
6f48d0eb 76 struct task_struct *start = tsk;
495789a5 77
495789a5 78 do {
6f48d0eb
DR
79 if (mask) {
80 /*
81 * If this is a mempolicy constrained oom, tsk's
82 * cpuset is irrelevant. Only return true if its
83 * mempolicy intersects current, otherwise it may be
84 * needlessly killed.
85 */
86 if (mempolicy_nodemask_intersects(tsk, mask))
87 return true;
88 } else {
89 /*
90 * This is not a mempolicy constrained oom, so only
91 * check the mems of tsk's cpuset.
92 */
93 if (cpuset_mems_allowed_intersects(current, tsk))
94 return true;
95 }
df1090a8
KM
96 } while_each_thread(start, tsk);
97
6f48d0eb
DR
98 return false;
99}
100#else
101static bool has_intersects_mems_allowed(struct task_struct *tsk,
102 const nodemask_t *mask)
103{
104 return true;
495789a5 105}
6f48d0eb 106#endif /* CONFIG_NUMA */
495789a5 107
6f48d0eb
DR
108/*
109 * The process p may have detached its own ->mm while exiting or through
110 * use_mm(), but one or more of its subthreads may still have a valid
111 * pointer. Return p, or any of its subthreads with a valid ->mm, with
112 * task_lock() held.
113 */
158e0a2d 114struct task_struct *find_lock_task_mm(struct task_struct *p)
dd8e8f40
ON
115{
116 struct task_struct *t = p;
117
118 do {
119 task_lock(t);
120 if (likely(t->mm))
121 return t;
122 task_unlock(t);
123 } while_each_thread(p, t);
124
125 return NULL;
126}
127
ab290adb 128/* return true if the task is not adequate as candidate victim task. */
e85bfd3a
DR
129static bool oom_unkillable_task(struct task_struct *p,
130 const struct mem_cgroup *mem, const nodemask_t *nodemask)
ab290adb
KM
131{
132 if (is_global_init(p))
133 return true;
134 if (p->flags & PF_KTHREAD)
135 return true;
136
137 /* When mem_cgroup_out_of_memory() and p is not member of the group */
138 if (mem && !task_in_mem_cgroup(p, mem))
139 return true;
140
141 /* p may not have freeable memory in nodemask */
142 if (!has_intersects_mems_allowed(p, nodemask))
143 return true;
144
145 return false;
146}
147
1da177e4 148/**
a63d83f4 149 * oom_badness - heuristic function to determine which candidate task to kill
1da177e4 150 * @p: task struct of which task we should calculate
a63d83f4 151 * @totalpages: total present RAM allowed for page allocation
1da177e4 152 *
a63d83f4
DR
153 * The heuristic for determining which task to kill is made to be as simple and
154 * predictable as possible. The goal is to return the highest value for the
155 * task consuming the most memory to avoid subsequent oom failures.
1da177e4 156 */
a63d83f4
DR
157unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem,
158 const nodemask_t *nodemask, unsigned long totalpages)
1da177e4 159{
a63d83f4 160 int points;
28b83c51 161
26ebc984
KM
162 if (oom_unkillable_task(p, mem, nodemask))
163 return 0;
1da177e4 164
dd8e8f40
ON
165 p = find_lock_task_mm(p);
166 if (!p)
1da177e4
LT
167 return 0;
168
1da177e4 169 /*
a63d83f4
DR
170 * The memory controller may have a limit of 0 bytes, so avoid a divide
171 * by zero, if necessary.
1da177e4 172 */
a63d83f4
DR
173 if (!totalpages)
174 totalpages = 1;
1da177e4
LT
175
176 /*
a63d83f4 177 * The baseline for the badness score is the proportion of RAM that each
f755a042 178 * task's rss, pagetable and swap space use.
1da177e4 179 */
f755a042
KM
180 points = get_mm_rss(p->mm) + p->mm->nr_ptes;
181 points += get_mm_counter(p->mm, MM_SWAPENTS);
182
183 points *= 1000;
184 points /= totalpages;
a63d83f4 185 task_unlock(p);
1da177e4
LT
186
187 /*
a63d83f4
DR
188 * Root processes get 3% bonus, just like the __vm_enough_memory()
189 * implementation used by LSMs.
1da177e4 190 */
a63d83f4
DR
191 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
192 points -= 30;
1da177e4
LT
193
194 /*
a63d83f4
DR
195 * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may
196 * either completely disable oom killing or always prefer a certain
197 * task.
1da177e4 198 */
a63d83f4 199 points += p->signal->oom_score_adj;
1da177e4 200
f19e8aa1
DR
201 /*
202 * Never return 0 for an eligible task that may be killed since it's
203 * possible that no single user task uses more than 0.1% of memory and
204 * no single admin tasks uses more than 3.0%.
205 */
206 if (points <= 0)
207 return 1;
a63d83f4 208 return (points < 1000) ? points : 1000;
1da177e4
LT
209}
210
9b0f8b04
CL
211/*
212 * Determine the type of allocation constraint.
213 */
9b0f8b04 214#ifdef CONFIG_NUMA
4365a567 215static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
a63d83f4
DR
216 gfp_t gfp_mask, nodemask_t *nodemask,
217 unsigned long *totalpages)
4365a567 218{
54a6eb5c 219 struct zone *zone;
dd1a239f 220 struct zoneref *z;
54a6eb5c 221 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
a63d83f4
DR
222 bool cpuset_limited = false;
223 int nid;
9b0f8b04 224
a63d83f4
DR
225 /* Default to all available memory */
226 *totalpages = totalram_pages + total_swap_pages;
227
228 if (!zonelist)
229 return CONSTRAINT_NONE;
4365a567
KH
230 /*
231 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
232 * to kill current.We have to random task kill in this case.
233 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
234 */
235 if (gfp_mask & __GFP_THISNODE)
236 return CONSTRAINT_NONE;
9b0f8b04 237
4365a567 238 /*
a63d83f4
DR
239 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
240 * the page allocator means a mempolicy is in effect. Cpuset policy
241 * is enforced in get_page_from_freelist().
4365a567 242 */
a63d83f4
DR
243 if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) {
244 *totalpages = total_swap_pages;
245 for_each_node_mask(nid, *nodemask)
246 *totalpages += node_spanned_pages(nid);
9b0f8b04 247 return CONSTRAINT_MEMORY_POLICY;
a63d83f4 248 }
4365a567
KH
249
250 /* Check this allocation failure is caused by cpuset's wall function */
251 for_each_zone_zonelist_nodemask(zone, z, zonelist,
252 high_zoneidx, nodemask)
253 if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
a63d83f4 254 cpuset_limited = true;
9b0f8b04 255
a63d83f4
DR
256 if (cpuset_limited) {
257 *totalpages = total_swap_pages;
258 for_each_node_mask(nid, cpuset_current_mems_allowed)
259 *totalpages += node_spanned_pages(nid);
260 return CONSTRAINT_CPUSET;
261 }
9b0f8b04
CL
262 return CONSTRAINT_NONE;
263}
4365a567
KH
264#else
265static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
a63d83f4
DR
266 gfp_t gfp_mask, nodemask_t *nodemask,
267 unsigned long *totalpages)
4365a567 268{
a63d83f4 269 *totalpages = totalram_pages + total_swap_pages;
4365a567
KH
270 return CONSTRAINT_NONE;
271}
272#endif
9b0f8b04 273
1da177e4
LT
274/*
275 * Simple selection loop. We chose the process with the highest
276 * number of 'points'. We expect the caller will lock the tasklist.
277 *
278 * (not docbooked, we don't want this one cluttering up the manual)
279 */
a63d83f4
DR
280static struct task_struct *select_bad_process(unsigned int *ppoints,
281 unsigned long totalpages, struct mem_cgroup *mem,
282 const nodemask_t *nodemask)
1da177e4 283{
3a5dda7a 284 struct task_struct *g, *p;
1da177e4 285 struct task_struct *chosen = NULL;
9827b781 286 *ppoints = 0;
1da177e4 287
3a5dda7a 288 do_each_thread(g, p) {
a63d83f4 289 unsigned int points;
a49335cc 290
c027a474 291 if (p->exit_state)
30e2b41f 292 continue;
ab290adb 293 if (oom_unkillable_task(p, mem, nodemask))
6cf86ac6 294 continue;
ef08e3b4 295
b78483a4
NP
296 /*
297 * This task already has access to memory reserves and is
298 * being killed. Don't allow any other task access to the
299 * memory reserve.
300 *
301 * Note: this may have a chance of deadlock if it gets
302 * blocked waiting for another task which itself is waiting
303 * for memory. Is there a better alternative?
304 */
f660daac
DR
305 if (test_tsk_thread_flag(p, TIF_MEMDIE)) {
306 if (unlikely(frozen(p)))
307 thaw_process(p);
b78483a4 308 return ERR_PTR(-1UL);
f660daac 309 }
c027a474
ON
310 if (!p->mm)
311 continue;
b78483a4 312
30e2b41f 313 if (p->flags & PF_EXITING) {
edd45544
DR
314 /*
315 * If p is the current task and is in the process of
316 * releasing memory, we allow the "kill" to set
317 * TIF_MEMDIE, which will allow it to gain access to
318 * memory reserves. Otherwise, it may stall forever.
319 *
320 * The loop isn't broken here, however, in case other
321 * threads are found to have already been oom killed.
322 */
323 if (p == current) {
324 chosen = p;
325 *ppoints = 1000;
326 } else {
327 /*
328 * If this task is not being ptraced on exit,
329 * then wait for it to finish before killing
330 * some other task unnecessarily.
331 */
d21142ec 332 if (!(p->group_leader->ptrace & PT_TRACE_EXIT))
edd45544
DR
333 return ERR_PTR(-1UL);
334 }
50ec3bbf 335 }
972c4ea5 336
a63d83f4
DR
337 points = oom_badness(p, mem, nodemask, totalpages);
338 if (points > *ppoints) {
a49335cc 339 chosen = p;
9827b781 340 *ppoints = points;
1da177e4 341 }
3a5dda7a 342 } while_each_thread(g, p);
972c4ea5 343
1da177e4
LT
344 return chosen;
345}
346
fef1bdd6 347/**
1b578df0 348 * dump_tasks - dump current memory state of all system tasks
74ab7f1d 349 * @mem: current's memory controller, if constrained
e85bfd3a 350 * @nodemask: nodemask passed to page allocator for mempolicy ooms
1b578df0 351 *
e85bfd3a
DR
352 * Dumps the current memory state of all eligible tasks. Tasks not in the same
353 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
354 * are not shown.
fef1bdd6 355 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
a63d83f4 356 * value, oom_score_adj value, and name.
fef1bdd6 357 *
fef1bdd6
DR
358 * Call with tasklist_lock read-locked.
359 */
e85bfd3a 360static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask)
fef1bdd6 361{
c55db957
KM
362 struct task_struct *p;
363 struct task_struct *task;
fef1bdd6 364
a63d83f4 365 pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n");
c55db957 366 for_each_process(p) {
e85bfd3a 367 if (oom_unkillable_task(p, mem, nodemask))
b4416d2b 368 continue;
fef1bdd6 369
c55db957
KM
370 task = find_lock_task_mm(p);
371 if (!task) {
6d2661ed 372 /*
74ab7f1d
DR
373 * This is a kthread or all of p's threads have already
374 * detached their mm's. There's no need to report
c55db957 375 * them; they can't be oom killed anyway.
6d2661ed 376 */
6d2661ed
DR
377 continue;
378 }
c55db957 379
a63d83f4 380 pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n",
8d6c83f0 381 task->pid, task_uid(task), task->tgid,
a63d83f4
DR
382 task->mm->total_vm, get_mm_rss(task->mm),
383 task_cpu(task), task->signal->oom_adj,
384 task->signal->oom_score_adj, task->comm);
c55db957
KM
385 task_unlock(task);
386 }
fef1bdd6
DR
387}
388
d31f56db 389static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
e85bfd3a 390 struct mem_cgroup *mem, const nodemask_t *nodemask)
1b604d75 391{
5e9d834a 392 task_lock(current);
1b604d75 393 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
a63d83f4
DR
394 "oom_adj=%d, oom_score_adj=%d\n",
395 current->comm, gfp_mask, order, current->signal->oom_adj,
396 current->signal->oom_score_adj);
1b604d75
DR
397 cpuset_print_task_mems_allowed(current);
398 task_unlock(current);
399 dump_stack();
d31f56db 400 mem_cgroup_print_oom_info(mem, p);
b2b755b5 401 show_mem(SHOW_MEM_FILTER_NODES);
1b604d75 402 if (sysctl_oom_dump_tasks)
e85bfd3a 403 dump_tasks(mem, nodemask);
1b604d75
DR
404}
405
3b4798cb 406#define K(x) ((x) << (PAGE_SHIFT-10))
93b43fa5 407static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem)
1da177e4 408{
1e99bad0
DR
409 struct task_struct *q;
410 struct mm_struct *mm;
411
dd8e8f40 412 p = find_lock_task_mm(p);
be71cf22 413 if (!p)
b940fd70 414 return 1;
be71cf22 415
1e99bad0
DR
416 /* mm cannot be safely dereferenced after task_unlock(p) */
417 mm = p->mm;
418
b940fd70
DR
419 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
420 task_pid_nr(p), p->comm, K(p->mm->total_vm),
421 K(get_mm_counter(p->mm, MM_ANONPAGES)),
422 K(get_mm_counter(p->mm, MM_FILEPAGES)));
3b4798cb 423 task_unlock(p);
1da177e4 424
1e99bad0 425 /*
7b0d44fa 426 * Kill all user processes sharing p->mm in other thread groups, if any.
1e99bad0
DR
427 * They don't get access to memory reserves or a higher scheduler
428 * priority, though, to avoid depletion of all memory or task
429 * starvation. This prevents mm->mmap_sem livelock when an oom killed
430 * task cannot exit because it requires the semaphore and its contended
431 * by another thread trying to allocate memory itself. That thread will
432 * now get access to memory reserves since it has a pending fatal
433 * signal.
434 */
435 for_each_process(q)
7b0d44fa
DR
436 if (q->mm == mm && !same_thread_group(q, p) &&
437 !(q->flags & PF_KTHREAD)) {
c9f01245
DR
438 if (q->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
439 continue;
440
1e99bad0
DR
441 task_lock(q); /* Protect ->comm from prctl() */
442 pr_err("Kill process %d (%s) sharing same memory\n",
443 task_pid_nr(q), q->comm);
444 task_unlock(q);
445 force_sig(SIGKILL, q);
446 }
93b43fa5 447
1da177e4 448 set_tsk_thread_flag(p, TIF_MEMDIE);
1da177e4 449 force_sig(SIGKILL, p);
93b43fa5 450
01315922 451 return 0;
1da177e4 452}
b940fd70 453#undef K
1da177e4 454
7213f506 455static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
a63d83f4
DR
456 unsigned int points, unsigned long totalpages,
457 struct mem_cgroup *mem, nodemask_t *nodemask,
458 const char *message)
1da177e4 459{
52d3c036 460 struct task_struct *victim = p;
5e9d834a 461 struct task_struct *child;
52d3c036
LT
462 struct task_struct *t = p;
463 unsigned int victim_points = 0;
1da177e4 464
1b604d75 465 if (printk_ratelimit())
e85bfd3a 466 dump_header(p, gfp_mask, order, mem, nodemask);
7213f506 467
50ec3bbf
NP
468 /*
469 * If the task is already exiting, don't alarm the sysadmin or kill
470 * its children or threads, just set TIF_MEMDIE so it can die quickly
471 */
0753ba01 472 if (p->flags & PF_EXITING) {
4358997a 473 set_tsk_thread_flag(p, TIF_MEMDIE);
50ec3bbf
NP
474 return 0;
475 }
476
5e9d834a 477 task_lock(p);
a63d83f4 478 pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
5e9d834a
DR
479 message, task_pid_nr(p), p->comm, points);
480 task_unlock(p);
f3af38d3 481
5e9d834a
DR
482 /*
483 * If any of p's children has a different mm and is eligible for kill,
11239836 484 * the one with the highest oom_badness() score is sacrificed for its
5e9d834a
DR
485 * parent. This attempts to lose the minimal amount of work done while
486 * still freeing memory.
487 */
dd8e8f40 488 do {
5e9d834a 489 list_for_each_entry(child, &t->children, sibling) {
a63d83f4 490 unsigned int child_points;
5e9d834a 491
edd45544
DR
492 if (child->mm == p->mm)
493 continue;
a63d83f4
DR
494 /*
495 * oom_badness() returns 0 if the thread is unkillable
496 */
497 child_points = oom_badness(child, mem, nodemask,
498 totalpages);
5e9d834a
DR
499 if (child_points > victim_points) {
500 victim = child;
501 victim_points = child_points;
502 }
dd8e8f40
ON
503 }
504 } while_each_thread(p, t);
505
93b43fa5 506 return oom_kill_task(victim, mem);
1da177e4
LT
507}
508
309ed882
DR
509/*
510 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
511 */
512static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
e85bfd3a 513 int order, const nodemask_t *nodemask)
309ed882
DR
514{
515 if (likely(!sysctl_panic_on_oom))
516 return;
517 if (sysctl_panic_on_oom != 2) {
518 /*
519 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
520 * does not panic for cpuset, mempolicy, or memcg allocation
521 * failures.
522 */
523 if (constraint != CONSTRAINT_NONE)
524 return;
525 }
526 read_lock(&tasklist_lock);
e85bfd3a 527 dump_header(NULL, gfp_mask, order, NULL, nodemask);
309ed882
DR
528 read_unlock(&tasklist_lock);
529 panic("Out of memory: %s panic_on_oom is enabled\n",
530 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
531}
532
00f0b825 533#ifdef CONFIG_CGROUP_MEM_RES_CTLR
c7ba5c9e
PE
534void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
535{
a63d83f4
DR
536 unsigned long limit;
537 unsigned int points = 0;
c7ba5c9e
PE
538 struct task_struct *p;
539
f9434ad1
DR
540 /*
541 * If current has a pending SIGKILL, then automatically select it. The
542 * goal is to allow it to allocate so that it may quickly exit and free
543 * its memory.
544 */
545 if (fatal_signal_pending(current)) {
546 set_thread_flag(TIF_MEMDIE);
f9434ad1
DR
547 return;
548 }
549
e85bfd3a 550 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL);
a63d83f4 551 limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT;
e115f2d8 552 read_lock(&tasklist_lock);
c7ba5c9e 553retry:
a63d83f4 554 p = select_bad_process(&points, limit, mem, NULL);
df64f81b 555 if (!p || PTR_ERR(p) == -1UL)
c7ba5c9e
PE
556 goto out;
557
a63d83f4 558 if (oom_kill_process(p, gfp_mask, 0, points, limit, mem, NULL,
c7ba5c9e
PE
559 "Memory cgroup out of memory"))
560 goto retry;
561out:
e115f2d8 562 read_unlock(&tasklist_lock);
c7ba5c9e
PE
563}
564#endif
565
8bc719d3
MS
566static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
567
568int register_oom_notifier(struct notifier_block *nb)
569{
570 return blocking_notifier_chain_register(&oom_notify_list, nb);
571}
572EXPORT_SYMBOL_GPL(register_oom_notifier);
573
574int unregister_oom_notifier(struct notifier_block *nb)
575{
576 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
577}
578EXPORT_SYMBOL_GPL(unregister_oom_notifier);
579
098d7f12
DR
580/*
581 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
582 * if a parallel OOM killing is already taking place that includes a zone in
583 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
584 */
ff321fea 585int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
098d7f12 586{
dd1a239f
MG
587 struct zoneref *z;
588 struct zone *zone;
098d7f12
DR
589 int ret = 1;
590
c7d4caeb 591 spin_lock(&zone_scan_lock);
dd1a239f
MG
592 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
593 if (zone_is_oom_locked(zone)) {
098d7f12
DR
594 ret = 0;
595 goto out;
596 }
dd1a239f
MG
597 }
598
599 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
600 /*
c7d4caeb 601 * Lock each zone in the zonelist under zone_scan_lock so a
ff321fea 602 * parallel invocation of try_set_zonelist_oom() doesn't succeed
dd1a239f
MG
603 * when it shouldn't.
604 */
605 zone_set_flag(zone, ZONE_OOM_LOCKED);
606 }
098d7f12 607
098d7f12 608out:
c7d4caeb 609 spin_unlock(&zone_scan_lock);
098d7f12
DR
610 return ret;
611}
612
613/*
614 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
615 * allocation attempts with zonelists containing them may now recall the OOM
616 * killer, if necessary.
617 */
dd1a239f 618void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
098d7f12 619{
dd1a239f
MG
620 struct zoneref *z;
621 struct zone *zone;
098d7f12 622
c7d4caeb 623 spin_lock(&zone_scan_lock);
dd1a239f
MG
624 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
625 zone_clear_flag(zone, ZONE_OOM_LOCKED);
626 }
c7d4caeb 627 spin_unlock(&zone_scan_lock);
098d7f12
DR
628}
629
e3658932
DR
630/*
631 * Try to acquire the oom killer lock for all system zones. Returns zero if a
632 * parallel oom killing is taking place, otherwise locks all zones and returns
633 * non-zero.
634 */
635static int try_set_system_oom(void)
636{
637 struct zone *zone;
638 int ret = 1;
639
640 spin_lock(&zone_scan_lock);
641 for_each_populated_zone(zone)
642 if (zone_is_oom_locked(zone)) {
643 ret = 0;
644 goto out;
645 }
646 for_each_populated_zone(zone)
647 zone_set_flag(zone, ZONE_OOM_LOCKED);
648out:
649 spin_unlock(&zone_scan_lock);
650 return ret;
651}
652
653/*
654 * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation
655 * attempts or page faults may now recall the oom killer, if necessary.
656 */
657static void clear_system_oom(void)
658{
659 struct zone *zone;
660
661 spin_lock(&zone_scan_lock);
662 for_each_populated_zone(zone)
663 zone_clear_flag(zone, ZONE_OOM_LOCKED);
664 spin_unlock(&zone_scan_lock);
665}
666
1da177e4 667/**
6937a25c 668 * out_of_memory - kill the "best" process when we run out of memory
1b578df0
RD
669 * @zonelist: zonelist pointer
670 * @gfp_mask: memory allocation flags
671 * @order: amount of memory being requested as a power of 2
6f48d0eb 672 * @nodemask: nodemask passed to page allocator
1da177e4
LT
673 *
674 * If we run out of memory, we have the choice between either
675 * killing a random task (bad), letting the system crash (worse)
676 * OR try to be smart about which process to kill. Note that we
677 * don't have to be perfect here, we just have to be good.
678 */
4365a567
KH
679void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
680 int order, nodemask_t *nodemask)
1da177e4 681{
e85bfd3a 682 const nodemask_t *mpol_mask;
0aad4b31 683 struct task_struct *p;
a63d83f4 684 unsigned long totalpages;
8bc719d3 685 unsigned long freed = 0;
a63d83f4 686 unsigned int points;
e3658932 687 enum oom_constraint constraint = CONSTRAINT_NONE;
b52723c5 688 int killed = 0;
8bc719d3
MS
689
690 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
691 if (freed > 0)
692 /* Got some memory back in the last second. */
693 return;
1da177e4 694
7b98c2e4
DR
695 /*
696 * If current has a pending SIGKILL, then automatically select it. The
697 * goal is to allow it to allocate so that it may quickly exit and free
698 * its memory.
699 */
700 if (fatal_signal_pending(current)) {
701 set_thread_flag(TIF_MEMDIE);
702 return;
703 }
704
9b0f8b04
CL
705 /*
706 * Check if there were limitations on the allocation (only relevant for
707 * NUMA) that may require different handling.
708 */
a63d83f4
DR
709 constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
710 &totalpages);
e85bfd3a
DR
711 mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
712 check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
0aad4b31 713
2b45ab33 714 read_lock(&tasklist_lock);
f88ccad5 715 if (sysctl_oom_kill_allocating_task &&
a96cfd6e 716 !oom_unkillable_task(current, NULL, nodemask) &&
c9f01245 717 current->mm) {
0aad4b31
DR
718 /*
719 * oom_kill_process() needs tasklist_lock held. If it returns
720 * non-zero, current could not be killed so we must fallback to
721 * the tasklist scan.
722 */
a63d83f4
DR
723 if (!oom_kill_process(current, gfp_mask, order, 0, totalpages,
724 NULL, nodemask,
0aad4b31 725 "Out of memory (oom_kill_allocating_task)"))
b52723c5 726 goto out;
0aad4b31
DR
727 }
728
729retry:
e85bfd3a 730 p = select_bad_process(&points, totalpages, NULL, mpol_mask);
0aad4b31 731 if (PTR_ERR(p) == -1UL)
b52723c5 732 goto out;
0aad4b31
DR
733
734 /* Found nothing?!?! Either we hang forever, or we panic. */
735 if (!p) {
e85bfd3a 736 dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
0aad4b31
DR
737 read_unlock(&tasklist_lock);
738 panic("Out of memory and no killable processes...\n");
739 }
740
a63d83f4
DR
741 if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
742 nodemask, "Out of memory"))
0aad4b31 743 goto retry;
b52723c5
KM
744 killed = 1;
745out:
140ffcec 746 read_unlock(&tasklist_lock);
1da177e4
LT
747
748 /*
749 * Give "p" a good chance of killing itself before we
2f659f46 750 * retry to allocate memory unless "p" is current
1da177e4 751 */
b52723c5 752 if (killed && !test_thread_flag(TIF_MEMDIE))
140ffcec 753 schedule_timeout_uninterruptible(1);
1da177e4 754}
e3658932
DR
755
756/*
757 * The pagefault handler calls here because it is out of memory, so kill a
758 * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel
759 * oom killing is already in progress so do nothing. If a task is found with
760 * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit.
761 */
762void pagefault_out_of_memory(void)
763{
764 if (try_set_system_oom()) {
765 out_of_memory(NULL, 0, 0, NULL);
766 clear_system_oom();
767 }
768 if (!test_thread_flag(TIF_MEMDIE))
769 schedule_timeout_uninterruptible(1);
770}