]>
Commit | Line | Data |
---|---|---|
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 | ||
37 | int sysctl_panic_on_oom; | |
38 | int sysctl_oom_kill_allocating_task; | |
39 | int sysctl_oom_dump_tasks = 1; | |
40 | static DEFINE_SPINLOCK(zone_scan_lock); | |
41 | ||
42 | /* | |
43 | * compare_swap_oom_score_adj() - compare and swap current's oom_score_adj | |
44 | * @old_val: old oom_score_adj for compare | |
45 | * @new_val: new oom_score_adj for swap | |
46 | * | |
47 | * Sets the oom_score_adj value for current to @new_val iff its present value is | |
48 | * @old_val. Usually used to reinstate a previous value to prevent racing with | |
49 | * userspacing tuning the value in the interim. | |
50 | */ | |
51 | void compare_swap_oom_score_adj(int old_val, int new_val) | |
52 | { | |
53 | struct sighand_struct *sighand = current->sighand; | |
54 | ||
55 | spin_lock_irq(&sighand->siglock); | |
56 | if (current->signal->oom_score_adj == old_val) | |
57 | current->signal->oom_score_adj = new_val; | |
58 | spin_unlock_irq(&sighand->siglock); | |
59 | } | |
60 | ||
61 | /** | |
62 | * test_set_oom_score_adj() - set current's oom_score_adj and return old value | |
63 | * @new_val: new oom_score_adj value | |
64 | * | |
65 | * Sets the oom_score_adj value for current to @new_val with proper | |
66 | * synchronization and returns the old value. Usually used to temporarily | |
67 | * set a value, save the old value in the caller, and then reinstate it later. | |
68 | */ | |
69 | int test_set_oom_score_adj(int new_val) | |
70 | { | |
71 | struct sighand_struct *sighand = current->sighand; | |
72 | int old_val; | |
73 | ||
74 | spin_lock_irq(&sighand->siglock); | |
75 | old_val = current->signal->oom_score_adj; | |
76 | current->signal->oom_score_adj = new_val; | |
77 | spin_unlock_irq(&sighand->siglock); | |
78 | ||
79 | return old_val; | |
80 | } | |
81 | ||
82 | #ifdef CONFIG_NUMA | |
83 | /** | |
84 | * has_intersects_mems_allowed() - check task eligiblity for kill | |
85 | * @tsk: task struct of which task to consider | |
86 | * @mask: nodemask passed to page allocator for mempolicy ooms | |
87 | * | |
88 | * Task eligibility is determined by whether or not a candidate task, @tsk, | |
89 | * shares the same mempolicy nodes as current if it is bound by such a policy | |
90 | * and whether or not it has the same set of allowed cpuset nodes. | |
91 | */ | |
92 | static bool has_intersects_mems_allowed(struct task_struct *tsk, | |
93 | const nodemask_t *mask) | |
94 | { | |
95 | struct task_struct *start = tsk; | |
96 | ||
97 | do { | |
98 | if (mask) { | |
99 | /* | |
100 | * If this is a mempolicy constrained oom, tsk's | |
101 | * cpuset is irrelevant. Only return true if its | |
102 | * mempolicy intersects current, otherwise it may be | |
103 | * needlessly killed. | |
104 | */ | |
105 | if (mempolicy_nodemask_intersects(tsk, mask)) | |
106 | return true; | |
107 | } else { | |
108 | /* | |
109 | * This is not a mempolicy constrained oom, so only | |
110 | * check the mems of tsk's cpuset. | |
111 | */ | |
112 | if (cpuset_mems_allowed_intersects(current, tsk)) | |
113 | return true; | |
114 | } | |
115 | } while_each_thread(start, tsk); | |
116 | ||
117 | return false; | |
118 | } | |
119 | #else | |
120 | static bool has_intersects_mems_allowed(struct task_struct *tsk, | |
121 | const nodemask_t *mask) | |
122 | { | |
123 | return true; | |
124 | } | |
125 | #endif /* CONFIG_NUMA */ | |
126 | ||
127 | /* | |
128 | * The process p may have detached its own ->mm while exiting or through | |
129 | * use_mm(), but one or more of its subthreads may still have a valid | |
130 | * pointer. Return p, or any of its subthreads with a valid ->mm, with | |
131 | * task_lock() held. | |
132 | */ | |
133 | struct task_struct *find_lock_task_mm(struct task_struct *p) | |
134 | { | |
135 | struct task_struct *t = p; | |
136 | ||
137 | do { | |
138 | task_lock(t); | |
139 | if (likely(t->mm)) | |
140 | return t; | |
141 | task_unlock(t); | |
142 | } while_each_thread(p, t); | |
143 | ||
144 | return NULL; | |
145 | } | |
146 | ||
147 | /* return true if the task is not adequate as candidate victim task. */ | |
148 | static bool oom_unkillable_task(struct task_struct *p, | |
149 | const struct mem_cgroup *mem, const nodemask_t *nodemask) | |
150 | { | |
151 | if (is_global_init(p)) | |
152 | return true; | |
153 | if (p->flags & PF_KTHREAD) | |
154 | return true; | |
155 | ||
156 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ | |
157 | if (mem && !task_in_mem_cgroup(p, mem)) | |
158 | return true; | |
159 | ||
160 | /* p may not have freeable memory in nodemask */ | |
161 | if (!has_intersects_mems_allowed(p, nodemask)) | |
162 | return true; | |
163 | ||
164 | return false; | |
165 | } | |
166 | ||
167 | /** | |
168 | * oom_badness - heuristic function to determine which candidate task to kill | |
169 | * @p: task struct of which task we should calculate | |
170 | * @totalpages: total present RAM allowed for page allocation | |
171 | * | |
172 | * The heuristic for determining which task to kill is made to be as simple and | |
173 | * predictable as possible. The goal is to return the highest value for the | |
174 | * task consuming the most memory to avoid subsequent oom failures. | |
175 | */ | |
176 | unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem, | |
177 | const nodemask_t *nodemask, unsigned long totalpages) | |
178 | { | |
179 | int points; | |
180 | ||
181 | if (oom_unkillable_task(p, mem, nodemask)) | |
182 | return 0; | |
183 | ||
184 | p = find_lock_task_mm(p); | |
185 | if (!p) | |
186 | return 0; | |
187 | ||
188 | /* | |
189 | * The memory controller may have a limit of 0 bytes, so avoid a divide | |
190 | * by zero, if necessary. | |
191 | */ | |
192 | if (!totalpages) | |
193 | totalpages = 1; | |
194 | ||
195 | /* | |
196 | * The baseline for the badness score is the proportion of RAM that each | |
197 | * task's rss, pagetable and swap space use. | |
198 | */ | |
199 | points = get_mm_rss(p->mm) + p->mm->nr_ptes; | |
200 | points += get_mm_counter(p->mm, MM_SWAPENTS); | |
201 | ||
202 | points *= 1000; | |
203 | points /= totalpages; | |
204 | task_unlock(p); | |
205 | ||
206 | /* | |
207 | * Root processes get 3% bonus, just like the __vm_enough_memory() | |
208 | * implementation used by LSMs. | |
209 | */ | |
210 | if (has_capability_noaudit(p, CAP_SYS_ADMIN)) | |
211 | points -= 30; | |
212 | ||
213 | /* | |
214 | * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may | |
215 | * either completely disable oom killing or always prefer a certain | |
216 | * task. | |
217 | */ | |
218 | points += p->signal->oom_score_adj; | |
219 | ||
220 | /* | |
221 | * Never return 0 for an eligible task that may be killed since it's | |
222 | * possible that no single user task uses more than 0.1% of memory and | |
223 | * no single admin tasks uses more than 3.0%. | |
224 | */ | |
225 | if (points <= 0) | |
226 | return 1; | |
227 | return (points < 1000) ? points : 1000; | |
228 | } | |
229 | ||
230 | /* | |
231 | * Determine the type of allocation constraint. | |
232 | */ | |
233 | #ifdef CONFIG_NUMA | |
234 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, | |
235 | gfp_t gfp_mask, nodemask_t *nodemask, | |
236 | unsigned long *totalpages) | |
237 | { | |
238 | struct zone *zone; | |
239 | struct zoneref *z; | |
240 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); | |
241 | bool cpuset_limited = false; | |
242 | int nid; | |
243 | ||
244 | /* Default to all available memory */ | |
245 | *totalpages = totalram_pages + total_swap_pages; | |
246 | ||
247 | if (!zonelist) | |
248 | return CONSTRAINT_NONE; | |
249 | /* | |
250 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid | |
251 | * to kill current.We have to random task kill in this case. | |
252 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. | |
253 | */ | |
254 | if (gfp_mask & __GFP_THISNODE) | |
255 | return CONSTRAINT_NONE; | |
256 | ||
257 | /* | |
258 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in | |
259 | * the page allocator means a mempolicy is in effect. Cpuset policy | |
260 | * is enforced in get_page_from_freelist(). | |
261 | */ | |
262 | if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) { | |
263 | *totalpages = total_swap_pages; | |
264 | for_each_node_mask(nid, *nodemask) | |
265 | *totalpages += node_spanned_pages(nid); | |
266 | return CONSTRAINT_MEMORY_POLICY; | |
267 | } | |
268 | ||
269 | /* Check this allocation failure is caused by cpuset's wall function */ | |
270 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
271 | high_zoneidx, nodemask) | |
272 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) | |
273 | cpuset_limited = true; | |
274 | ||
275 | if (cpuset_limited) { | |
276 | *totalpages = total_swap_pages; | |
277 | for_each_node_mask(nid, cpuset_current_mems_allowed) | |
278 | *totalpages += node_spanned_pages(nid); | |
279 | return CONSTRAINT_CPUSET; | |
280 | } | |
281 | return CONSTRAINT_NONE; | |
282 | } | |
283 | #else | |
284 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, | |
285 | gfp_t gfp_mask, nodemask_t *nodemask, | |
286 | unsigned long *totalpages) | |
287 | { | |
288 | *totalpages = totalram_pages + total_swap_pages; | |
289 | return CONSTRAINT_NONE; | |
290 | } | |
291 | #endif | |
292 | ||
293 | /* | |
294 | * Simple selection loop. We chose the process with the highest | |
295 | * number of 'points'. We expect the caller will lock the tasklist. | |
296 | * | |
297 | * (not docbooked, we don't want this one cluttering up the manual) | |
298 | */ | |
299 | static struct task_struct *select_bad_process(unsigned int *ppoints, | |
300 | unsigned long totalpages, struct mem_cgroup *mem, | |
301 | const nodemask_t *nodemask) | |
302 | { | |
303 | struct task_struct *g, *p; | |
304 | struct task_struct *chosen = NULL; | |
305 | *ppoints = 0; | |
306 | ||
307 | do_each_thread(g, p) { | |
308 | unsigned int points; | |
309 | ||
310 | if (p->exit_state) | |
311 | continue; | |
312 | if (oom_unkillable_task(p, mem, nodemask)) | |
313 | continue; | |
314 | ||
315 | /* | |
316 | * This task already has access to memory reserves and is | |
317 | * being killed. Don't allow any other task access to the | |
318 | * memory reserve. | |
319 | * | |
320 | * Note: this may have a chance of deadlock if it gets | |
321 | * blocked waiting for another task which itself is waiting | |
322 | * for memory. Is there a better alternative? | |
323 | */ | |
324 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) { | |
325 | if (unlikely(frozen(p))) | |
326 | thaw_process(p); | |
327 | return ERR_PTR(-1UL); | |
328 | } | |
329 | if (!p->mm) | |
330 | continue; | |
331 | ||
332 | if (p->flags & PF_EXITING) { | |
333 | /* | |
334 | * If p is the current task and is in the process of | |
335 | * releasing memory, we allow the "kill" to set | |
336 | * TIF_MEMDIE, which will allow it to gain access to | |
337 | * memory reserves. Otherwise, it may stall forever. | |
338 | * | |
339 | * The loop isn't broken here, however, in case other | |
340 | * threads are found to have already been oom killed. | |
341 | */ | |
342 | if (p == current) { | |
343 | chosen = p; | |
344 | *ppoints = 1000; | |
345 | } else { | |
346 | /* | |
347 | * If this task is not being ptraced on exit, | |
348 | * then wait for it to finish before killing | |
349 | * some other task unnecessarily. | |
350 | */ | |
351 | if (!(p->group_leader->ptrace & PT_TRACE_EXIT)) | |
352 | return ERR_PTR(-1UL); | |
353 | } | |
354 | } | |
355 | ||
356 | points = oom_badness(p, mem, nodemask, totalpages); | |
357 | if (points > *ppoints) { | |
358 | chosen = p; | |
359 | *ppoints = points; | |
360 | } | |
361 | } while_each_thread(g, p); | |
362 | ||
363 | return chosen; | |
364 | } | |
365 | ||
366 | /** | |
367 | * dump_tasks - dump current memory state of all system tasks | |
368 | * @mem: current's memory controller, if constrained | |
369 | * @nodemask: nodemask passed to page allocator for mempolicy ooms | |
370 | * | |
371 | * Dumps the current memory state of all eligible tasks. Tasks not in the same | |
372 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes | |
373 | * are not shown. | |
374 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj | |
375 | * value, oom_score_adj value, and name. | |
376 | * | |
377 | * Call with tasklist_lock read-locked. | |
378 | */ | |
379 | static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask) | |
380 | { | |
381 | struct task_struct *p; | |
382 | struct task_struct *task; | |
383 | ||
384 | pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n"); | |
385 | for_each_process(p) { | |
386 | if (oom_unkillable_task(p, mem, nodemask)) | |
387 | continue; | |
388 | ||
389 | task = find_lock_task_mm(p); | |
390 | if (!task) { | |
391 | /* | |
392 | * This is a kthread or all of p's threads have already | |
393 | * detached their mm's. There's no need to report | |
394 | * them; they can't be oom killed anyway. | |
395 | */ | |
396 | continue; | |
397 | } | |
398 | ||
399 | pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n", | |
400 | task->pid, task_uid(task), task->tgid, | |
401 | task->mm->total_vm, get_mm_rss(task->mm), | |
402 | task_cpu(task), task->signal->oom_adj, | |
403 | task->signal->oom_score_adj, task->comm); | |
404 | task_unlock(task); | |
405 | } | |
406 | } | |
407 | ||
408 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, | |
409 | struct mem_cgroup *mem, const nodemask_t *nodemask) | |
410 | { | |
411 | task_lock(current); | |
412 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " | |
413 | "oom_adj=%d, oom_score_adj=%d\n", | |
414 | current->comm, gfp_mask, order, current->signal->oom_adj, | |
415 | current->signal->oom_score_adj); | |
416 | cpuset_print_task_mems_allowed(current); | |
417 | task_unlock(current); | |
418 | dump_stack(); | |
419 | mem_cgroup_print_oom_info(mem, p); | |
420 | show_mem(SHOW_MEM_FILTER_NODES); | |
421 | if (sysctl_oom_dump_tasks) | |
422 | dump_tasks(mem, nodemask); | |
423 | } | |
424 | ||
425 | #define K(x) ((x) << (PAGE_SHIFT-10)) | |
426 | static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem) | |
427 | { | |
428 | struct task_struct *q; | |
429 | struct mm_struct *mm; | |
430 | ||
431 | p = find_lock_task_mm(p); | |
432 | if (!p) | |
433 | return 1; | |
434 | ||
435 | /* mm cannot be safely dereferenced after task_unlock(p) */ | |
436 | mm = p->mm; | |
437 | ||
438 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", | |
439 | task_pid_nr(p), p->comm, K(p->mm->total_vm), | |
440 | K(get_mm_counter(p->mm, MM_ANONPAGES)), | |
441 | K(get_mm_counter(p->mm, MM_FILEPAGES))); | |
442 | task_unlock(p); | |
443 | ||
444 | /* | |
445 | * Kill all user processes sharing p->mm in other thread groups, if any. | |
446 | * They don't get access to memory reserves or a higher scheduler | |
447 | * priority, though, to avoid depletion of all memory or task | |
448 | * starvation. This prevents mm->mmap_sem livelock when an oom killed | |
449 | * task cannot exit because it requires the semaphore and its contended | |
450 | * by another thread trying to allocate memory itself. That thread will | |
451 | * now get access to memory reserves since it has a pending fatal | |
452 | * signal. | |
453 | */ | |
454 | for_each_process(q) | |
455 | if (q->mm == mm && !same_thread_group(q, p) && | |
456 | !(q->flags & PF_KTHREAD)) { | |
457 | if (q->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) | |
458 | continue; | |
459 | ||
460 | task_lock(q); /* Protect ->comm from prctl() */ | |
461 | pr_err("Kill process %d (%s) sharing same memory\n", | |
462 | task_pid_nr(q), q->comm); | |
463 | task_unlock(q); | |
464 | force_sig(SIGKILL, q); | |
465 | } | |
466 | ||
467 | set_tsk_thread_flag(p, TIF_MEMDIE); | |
468 | force_sig(SIGKILL, p); | |
469 | ||
470 | return 0; | |
471 | } | |
472 | #undef K | |
473 | ||
474 | static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, | |
475 | unsigned int points, unsigned long totalpages, | |
476 | struct mem_cgroup *mem, nodemask_t *nodemask, | |
477 | const char *message) | |
478 | { | |
479 | struct task_struct *victim = p; | |
480 | struct task_struct *child; | |
481 | struct task_struct *t = p; | |
482 | unsigned int victim_points = 0; | |
483 | ||
484 | if (printk_ratelimit()) | |
485 | dump_header(p, gfp_mask, order, mem, nodemask); | |
486 | ||
487 | /* | |
488 | * If the task is already exiting, don't alarm the sysadmin or kill | |
489 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
490 | */ | |
491 | if (p->flags & PF_EXITING) { | |
492 | set_tsk_thread_flag(p, TIF_MEMDIE); | |
493 | return 0; | |
494 | } | |
495 | ||
496 | task_lock(p); | |
497 | pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n", | |
498 | message, task_pid_nr(p), p->comm, points); | |
499 | task_unlock(p); | |
500 | ||
501 | /* | |
502 | * If any of p's children has a different mm and is eligible for kill, | |
503 | * the one with the highest oom_badness() score is sacrificed for its | |
504 | * parent. This attempts to lose the minimal amount of work done while | |
505 | * still freeing memory. | |
506 | */ | |
507 | do { | |
508 | list_for_each_entry(child, &t->children, sibling) { | |
509 | unsigned int child_points; | |
510 | ||
511 | if (child->mm == p->mm) | |
512 | continue; | |
513 | /* | |
514 | * oom_badness() returns 0 if the thread is unkillable | |
515 | */ | |
516 | child_points = oom_badness(child, mem, nodemask, | |
517 | totalpages); | |
518 | if (child_points > victim_points) { | |
519 | victim = child; | |
520 | victim_points = child_points; | |
521 | } | |
522 | } | |
523 | } while_each_thread(p, t); | |
524 | ||
525 | return oom_kill_task(victim, mem); | |
526 | } | |
527 | ||
528 | /* | |
529 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. | |
530 | */ | |
531 | static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask, | |
532 | int order, const nodemask_t *nodemask) | |
533 | { | |
534 | if (likely(!sysctl_panic_on_oom)) | |
535 | return; | |
536 | if (sysctl_panic_on_oom != 2) { | |
537 | /* | |
538 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel | |
539 | * does not panic for cpuset, mempolicy, or memcg allocation | |
540 | * failures. | |
541 | */ | |
542 | if (constraint != CONSTRAINT_NONE) | |
543 | return; | |
544 | } | |
545 | read_lock(&tasklist_lock); | |
546 | dump_header(NULL, gfp_mask, order, NULL, nodemask); | |
547 | read_unlock(&tasklist_lock); | |
548 | panic("Out of memory: %s panic_on_oom is enabled\n", | |
549 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); | |
550 | } | |
551 | ||
552 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR | |
553 | void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) | |
554 | { | |
555 | unsigned long limit; | |
556 | unsigned int points = 0; | |
557 | struct task_struct *p; | |
558 | ||
559 | /* | |
560 | * If current has a pending SIGKILL, then automatically select it. The | |
561 | * goal is to allow it to allocate so that it may quickly exit and free | |
562 | * its memory. | |
563 | */ | |
564 | if (fatal_signal_pending(current)) { | |
565 | set_thread_flag(TIF_MEMDIE); | |
566 | return; | |
567 | } | |
568 | ||
569 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL); | |
570 | limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT; | |
571 | read_lock(&tasklist_lock); | |
572 | retry: | |
573 | p = select_bad_process(&points, limit, mem, NULL); | |
574 | if (!p || PTR_ERR(p) == -1UL) | |
575 | goto out; | |
576 | ||
577 | if (oom_kill_process(p, gfp_mask, 0, points, limit, mem, NULL, | |
578 | "Memory cgroup out of memory")) | |
579 | goto retry; | |
580 | out: | |
581 | read_unlock(&tasklist_lock); | |
582 | } | |
583 | #endif | |
584 | ||
585 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); | |
586 | ||
587 | int register_oom_notifier(struct notifier_block *nb) | |
588 | { | |
589 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
590 | } | |
591 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
592 | ||
593 | int unregister_oom_notifier(struct notifier_block *nb) | |
594 | { | |
595 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
596 | } | |
597 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
598 | ||
599 | /* | |
600 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero | |
601 | * if a parallel OOM killing is already taking place that includes a zone in | |
602 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. | |
603 | */ | |
604 | int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) | |
605 | { | |
606 | struct zoneref *z; | |
607 | struct zone *zone; | |
608 | int ret = 1; | |
609 | ||
610 | spin_lock(&zone_scan_lock); | |
611 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
612 | if (zone_is_oom_locked(zone)) { | |
613 | ret = 0; | |
614 | goto out; | |
615 | } | |
616 | } | |
617 | ||
618 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
619 | /* | |
620 | * Lock each zone in the zonelist under zone_scan_lock so a | |
621 | * parallel invocation of try_set_zonelist_oom() doesn't succeed | |
622 | * when it shouldn't. | |
623 | */ | |
624 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
625 | } | |
626 | ||
627 | out: | |
628 | spin_unlock(&zone_scan_lock); | |
629 | return ret; | |
630 | } | |
631 | ||
632 | /* | |
633 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed | |
634 | * allocation attempts with zonelists containing them may now recall the OOM | |
635 | * killer, if necessary. | |
636 | */ | |
637 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) | |
638 | { | |
639 | struct zoneref *z; | |
640 | struct zone *zone; | |
641 | ||
642 | spin_lock(&zone_scan_lock); | |
643 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
644 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
645 | } | |
646 | spin_unlock(&zone_scan_lock); | |
647 | } | |
648 | ||
649 | /* | |
650 | * Try to acquire the oom killer lock for all system zones. Returns zero if a | |
651 | * parallel oom killing is taking place, otherwise locks all zones and returns | |
652 | * non-zero. | |
653 | */ | |
654 | static int try_set_system_oom(void) | |
655 | { | |
656 | struct zone *zone; | |
657 | int ret = 1; | |
658 | ||
659 | spin_lock(&zone_scan_lock); | |
660 | for_each_populated_zone(zone) | |
661 | if (zone_is_oom_locked(zone)) { | |
662 | ret = 0; | |
663 | goto out; | |
664 | } | |
665 | for_each_populated_zone(zone) | |
666 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
667 | out: | |
668 | spin_unlock(&zone_scan_lock); | |
669 | return ret; | |
670 | } | |
671 | ||
672 | /* | |
673 | * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation | |
674 | * attempts or page faults may now recall the oom killer, if necessary. | |
675 | */ | |
676 | static void clear_system_oom(void) | |
677 | { | |
678 | struct zone *zone; | |
679 | ||
680 | spin_lock(&zone_scan_lock); | |
681 | for_each_populated_zone(zone) | |
682 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
683 | spin_unlock(&zone_scan_lock); | |
684 | } | |
685 | ||
686 | /** | |
687 | * out_of_memory - kill the "best" process when we run out of memory | |
688 | * @zonelist: zonelist pointer | |
689 | * @gfp_mask: memory allocation flags | |
690 | * @order: amount of memory being requested as a power of 2 | |
691 | * @nodemask: nodemask passed to page allocator | |
692 | * | |
693 | * If we run out of memory, we have the choice between either | |
694 | * killing a random task (bad), letting the system crash (worse) | |
695 | * OR try to be smart about which process to kill. Note that we | |
696 | * don't have to be perfect here, we just have to be good. | |
697 | */ | |
698 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, | |
699 | int order, nodemask_t *nodemask) | |
700 | { | |
701 | const nodemask_t *mpol_mask; | |
702 | struct task_struct *p; | |
703 | unsigned long totalpages; | |
704 | unsigned long freed = 0; | |
705 | unsigned int points; | |
706 | enum oom_constraint constraint = CONSTRAINT_NONE; | |
707 | int killed = 0; | |
708 | ||
709 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
710 | if (freed > 0) | |
711 | /* Got some memory back in the last second. */ | |
712 | return; | |
713 | ||
714 | /* | |
715 | * If current has a pending SIGKILL, then automatically select it. The | |
716 | * goal is to allow it to allocate so that it may quickly exit and free | |
717 | * its memory. | |
718 | */ | |
719 | if (fatal_signal_pending(current)) { | |
720 | set_thread_flag(TIF_MEMDIE); | |
721 | return; | |
722 | } | |
723 | ||
724 | /* | |
725 | * Check if there were limitations on the allocation (only relevant for | |
726 | * NUMA) that may require different handling. | |
727 | */ | |
728 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask, | |
729 | &totalpages); | |
730 | mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL; | |
731 | check_panic_on_oom(constraint, gfp_mask, order, mpol_mask); | |
732 | ||
733 | read_lock(&tasklist_lock); | |
734 | if (sysctl_oom_kill_allocating_task && | |
735 | !oom_unkillable_task(current, NULL, nodemask) && | |
736 | current->mm) { | |
737 | /* | |
738 | * oom_kill_process() needs tasklist_lock held. If it returns | |
739 | * non-zero, current could not be killed so we must fallback to | |
740 | * the tasklist scan. | |
741 | */ | |
742 | if (!oom_kill_process(current, gfp_mask, order, 0, totalpages, | |
743 | NULL, nodemask, | |
744 | "Out of memory (oom_kill_allocating_task)")) | |
745 | goto out; | |
746 | } | |
747 | ||
748 | retry: | |
749 | p = select_bad_process(&points, totalpages, NULL, mpol_mask); | |
750 | if (PTR_ERR(p) == -1UL) | |
751 | goto out; | |
752 | ||
753 | /* Found nothing?!?! Either we hang forever, or we panic. */ | |
754 | if (!p) { | |
755 | dump_header(NULL, gfp_mask, order, NULL, mpol_mask); | |
756 | read_unlock(&tasklist_lock); | |
757 | panic("Out of memory and no killable processes...\n"); | |
758 | } | |
759 | ||
760 | if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL, | |
761 | nodemask, "Out of memory")) | |
762 | goto retry; | |
763 | killed = 1; | |
764 | out: | |
765 | read_unlock(&tasklist_lock); | |
766 | ||
767 | /* | |
768 | * Give "p" a good chance of killing itself before we | |
769 | * retry to allocate memory unless "p" is current | |
770 | */ | |
771 | if (killed && !test_thread_flag(TIF_MEMDIE)) | |
772 | schedule_timeout_uninterruptible(1); | |
773 | } | |
774 | ||
775 | /* | |
776 | * The pagefault handler calls here because it is out of memory, so kill a | |
777 | * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel | |
778 | * oom killing is already in progress so do nothing. If a task is found with | |
779 | * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit. | |
780 | */ | |
781 | void pagefault_out_of_memory(void) | |
782 | { | |
783 | if (try_set_system_oom()) { | |
784 | out_of_memory(NULL, 0, 0, NULL); | |
785 | clear_system_oom(); | |
786 | } | |
787 | if (!test_thread_flag(TIF_MEMDIE)) | |
788 | schedule_timeout_uninterruptible(1); | |
789 | } |