oom_kill.c 21 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759
  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. #include <linux/oom.h>
  20. #include <linux/mm.h>
  21. #include <linux/err.h>
  22. #include <linux/gfp.h>
  23. #include <linux/sched.h>
  24. #include <linux/swap.h>
  25. #include <linux/timex.h>
  26. #include <linux/jiffies.h>
  27. #include <linux/cpuset.h>
  28. #include <linux/module.h>
  29. #include <linux/notifier.h>
  30. #include <linux/memcontrol.h>
  31. #include <linux/mempolicy.h>
  32. #include <linux/security.h>
  33. #include <linux/ptrace.h>
  34. int sysctl_panic_on_oom;
  35. int sysctl_oom_kill_allocating_task;
  36. int sysctl_oom_dump_tasks = 1;
  37. static DEFINE_SPINLOCK(zone_scan_lock);
  38. #ifdef CONFIG_NUMA
  39. /**
  40. * has_intersects_mems_allowed() - check task eligiblity for kill
  41. * @tsk: task struct of which task to consider
  42. * @mask: nodemask passed to page allocator for mempolicy ooms
  43. *
  44. * Task eligibility is determined by whether or not a candidate task, @tsk,
  45. * shares the same mempolicy nodes as current if it is bound by such a policy
  46. * and whether or not it has the same set of allowed cpuset nodes.
  47. */
  48. static bool has_intersects_mems_allowed(struct task_struct *tsk,
  49. const nodemask_t *mask)
  50. {
  51. struct task_struct *start = tsk;
  52. do {
  53. if (mask) {
  54. /*
  55. * If this is a mempolicy constrained oom, tsk's
  56. * cpuset is irrelevant. Only return true if its
  57. * mempolicy intersects current, otherwise it may be
  58. * needlessly killed.
  59. */
  60. if (mempolicy_nodemask_intersects(tsk, mask))
  61. return true;
  62. } else {
  63. /*
  64. * This is not a mempolicy constrained oom, so only
  65. * check the mems of tsk's cpuset.
  66. */
  67. if (cpuset_mems_allowed_intersects(current, tsk))
  68. return true;
  69. }
  70. } while_each_thread(start, tsk);
  71. return false;
  72. }
  73. #else
  74. static bool has_intersects_mems_allowed(struct task_struct *tsk,
  75. const nodemask_t *mask)
  76. {
  77. return true;
  78. }
  79. #endif /* CONFIG_NUMA */
  80. /*
  81. * The process p may have detached its own ->mm while exiting or through
  82. * use_mm(), but one or more of its subthreads may still have a valid
  83. * pointer. Return p, or any of its subthreads with a valid ->mm, with
  84. * task_lock() held.
  85. */
  86. struct task_struct *find_lock_task_mm(struct task_struct *p)
  87. {
  88. struct task_struct *t = p;
  89. do {
  90. task_lock(t);
  91. if (likely(t->mm))
  92. return t;
  93. task_unlock(t);
  94. } while_each_thread(p, t);
  95. return NULL;
  96. }
  97. /* return true if the task is not adequate as candidate victim task. */
  98. static bool oom_unkillable_task(struct task_struct *p,
  99. const struct mem_cgroup *mem, const nodemask_t *nodemask)
  100. {
  101. if (is_global_init(p))
  102. return true;
  103. if (p->flags & PF_KTHREAD)
  104. return true;
  105. /* When mem_cgroup_out_of_memory() and p is not member of the group */
  106. if (mem && !task_in_mem_cgroup(p, mem))
  107. return true;
  108. /* p may not have freeable memory in nodemask */
  109. if (!has_intersects_mems_allowed(p, nodemask))
  110. return true;
  111. return false;
  112. }
  113. /**
  114. * oom_badness - heuristic function to determine which candidate task to kill
  115. * @p: task struct of which task we should calculate
  116. * @totalpages: total present RAM allowed for page allocation
  117. *
  118. * The heuristic for determining which task to kill is made to be as simple and
  119. * predictable as possible. The goal is to return the highest value for the
  120. * task consuming the most memory to avoid subsequent oom failures.
  121. */
  122. unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem,
  123. const nodemask_t *nodemask, unsigned long totalpages)
  124. {
  125. int points;
  126. if (oom_unkillable_task(p, mem, nodemask))
  127. return 0;
  128. p = find_lock_task_mm(p);
  129. if (!p)
  130. return 0;
  131. /*
  132. * Shortcut check for a thread sharing p->mm that is OOM_SCORE_ADJ_MIN
  133. * so the entire heuristic doesn't need to be executed for something
  134. * that cannot be killed.
  135. */
  136. if (atomic_read(&p->mm->oom_disable_count)) {
  137. task_unlock(p);
  138. return 0;
  139. }
  140. /*
  141. * When the PF_OOM_ORIGIN bit is set, it indicates the task should have
  142. * priority for oom killing.
  143. */
  144. if (p->flags & PF_OOM_ORIGIN) {
  145. task_unlock(p);
  146. return 1000;
  147. }
  148. /*
  149. * The memory controller may have a limit of 0 bytes, so avoid a divide
  150. * by zero, if necessary.
  151. */
  152. if (!totalpages)
  153. totalpages = 1;
  154. /*
  155. * The baseline for the badness score is the proportion of RAM that each
  156. * task's rss, pagetable and swap space use.
  157. */
  158. points = get_mm_rss(p->mm) + p->mm->nr_ptes;
  159. points += get_mm_counter(p->mm, MM_SWAPENTS);
  160. points *= 1000;
  161. points /= totalpages;
  162. task_unlock(p);
  163. /*
  164. * Root processes get 3% bonus, just like the __vm_enough_memory()
  165. * implementation used by LSMs.
  166. */
  167. if (has_capability_noaudit(p, CAP_SYS_ADMIN))
  168. points -= 30;
  169. /*
  170. * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may
  171. * either completely disable oom killing or always prefer a certain
  172. * task.
  173. */
  174. points += p->signal->oom_score_adj;
  175. /*
  176. * Never return 0 for an eligible task that may be killed since it's
  177. * possible that no single user task uses more than 0.1% of memory and
  178. * no single admin tasks uses more than 3.0%.
  179. */
  180. if (points <= 0)
  181. return 1;
  182. return (points < 1000) ? points : 1000;
  183. }
  184. /*
  185. * Determine the type of allocation constraint.
  186. */
  187. #ifdef CONFIG_NUMA
  188. static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
  189. gfp_t gfp_mask, nodemask_t *nodemask,
  190. unsigned long *totalpages)
  191. {
  192. struct zone *zone;
  193. struct zoneref *z;
  194. enum zone_type high_zoneidx = gfp_zone(gfp_mask);
  195. bool cpuset_limited = false;
  196. int nid;
  197. /* Default to all available memory */
  198. *totalpages = totalram_pages + total_swap_pages;
  199. if (!zonelist)
  200. return CONSTRAINT_NONE;
  201. /*
  202. * Reach here only when __GFP_NOFAIL is used. So, we should avoid
  203. * to kill current.We have to random task kill in this case.
  204. * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
  205. */
  206. if (gfp_mask & __GFP_THISNODE)
  207. return CONSTRAINT_NONE;
  208. /*
  209. * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
  210. * the page allocator means a mempolicy is in effect. Cpuset policy
  211. * is enforced in get_page_from_freelist().
  212. */
  213. if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) {
  214. *totalpages = total_swap_pages;
  215. for_each_node_mask(nid, *nodemask)
  216. *totalpages += node_spanned_pages(nid);
  217. return CONSTRAINT_MEMORY_POLICY;
  218. }
  219. /* Check this allocation failure is caused by cpuset's wall function */
  220. for_each_zone_zonelist_nodemask(zone, z, zonelist,
  221. high_zoneidx, nodemask)
  222. if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
  223. cpuset_limited = true;
  224. if (cpuset_limited) {
  225. *totalpages = total_swap_pages;
  226. for_each_node_mask(nid, cpuset_current_mems_allowed)
  227. *totalpages += node_spanned_pages(nid);
  228. return CONSTRAINT_CPUSET;
  229. }
  230. return CONSTRAINT_NONE;
  231. }
  232. #else
  233. static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
  234. gfp_t gfp_mask, nodemask_t *nodemask,
  235. unsigned long *totalpages)
  236. {
  237. *totalpages = totalram_pages + total_swap_pages;
  238. return CONSTRAINT_NONE;
  239. }
  240. #endif
  241. /*
  242. * Simple selection loop. We chose the process with the highest
  243. * number of 'points'. We expect the caller will lock the tasklist.
  244. *
  245. * (not docbooked, we don't want this one cluttering up the manual)
  246. */
  247. static struct task_struct *select_bad_process(unsigned int *ppoints,
  248. unsigned long totalpages, struct mem_cgroup *mem,
  249. const nodemask_t *nodemask)
  250. {
  251. struct task_struct *g, *p;
  252. struct task_struct *chosen = NULL;
  253. *ppoints = 0;
  254. do_each_thread(g, p) {
  255. unsigned int points;
  256. if (!p->mm)
  257. continue;
  258. if (oom_unkillable_task(p, mem, nodemask))
  259. continue;
  260. /*
  261. * This task already has access to memory reserves and is
  262. * being killed. Don't allow any other task access to the
  263. * memory reserve.
  264. *
  265. * Note: this may have a chance of deadlock if it gets
  266. * blocked waiting for another task which itself is waiting
  267. * for memory. Is there a better alternative?
  268. */
  269. if (test_tsk_thread_flag(p, TIF_MEMDIE))
  270. return ERR_PTR(-1UL);
  271. if (p->flags & PF_EXITING) {
  272. /*
  273. * If p is the current task and is in the process of
  274. * releasing memory, we allow the "kill" to set
  275. * TIF_MEMDIE, which will allow it to gain access to
  276. * memory reserves. Otherwise, it may stall forever.
  277. *
  278. * The loop isn't broken here, however, in case other
  279. * threads are found to have already been oom killed.
  280. */
  281. if (p == current) {
  282. chosen = p;
  283. *ppoints = 1000;
  284. } else {
  285. /*
  286. * If this task is not being ptraced on exit,
  287. * then wait for it to finish before killing
  288. * some other task unnecessarily.
  289. */
  290. if (!(task_ptrace(p->group_leader) &
  291. PT_TRACE_EXIT))
  292. return ERR_PTR(-1UL);
  293. }
  294. }
  295. points = oom_badness(p, mem, nodemask, totalpages);
  296. if (points > *ppoints) {
  297. chosen = p;
  298. *ppoints = points;
  299. }
  300. } while_each_thread(g, p);
  301. return chosen;
  302. }
  303. /**
  304. * dump_tasks - dump current memory state of all system tasks
  305. * @mem: current's memory controller, if constrained
  306. * @nodemask: nodemask passed to page allocator for mempolicy ooms
  307. *
  308. * Dumps the current memory state of all eligible tasks. Tasks not in the same
  309. * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
  310. * are not shown.
  311. * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
  312. * value, oom_score_adj value, and name.
  313. *
  314. * Call with tasklist_lock read-locked.
  315. */
  316. static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask)
  317. {
  318. struct task_struct *p;
  319. struct task_struct *task;
  320. pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n");
  321. for_each_process(p) {
  322. if (oom_unkillable_task(p, mem, nodemask))
  323. continue;
  324. task = find_lock_task_mm(p);
  325. if (!task) {
  326. /*
  327. * This is a kthread or all of p's threads have already
  328. * detached their mm's. There's no need to report
  329. * them; they can't be oom killed anyway.
  330. */
  331. continue;
  332. }
  333. pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n",
  334. task->pid, task_uid(task), task->tgid,
  335. task->mm->total_vm, get_mm_rss(task->mm),
  336. task_cpu(task), task->signal->oom_adj,
  337. task->signal->oom_score_adj, task->comm);
  338. task_unlock(task);
  339. }
  340. }
  341. static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
  342. struct mem_cgroup *mem, const nodemask_t *nodemask)
  343. {
  344. task_lock(current);
  345. pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
  346. "oom_adj=%d, oom_score_adj=%d\n",
  347. current->comm, gfp_mask, order, current->signal->oom_adj,
  348. current->signal->oom_score_adj);
  349. cpuset_print_task_mems_allowed(current);
  350. task_unlock(current);
  351. dump_stack();
  352. mem_cgroup_print_oom_info(mem, p);
  353. show_mem(SHOW_MEM_FILTER_NODES);
  354. if (sysctl_oom_dump_tasks)
  355. dump_tasks(mem, nodemask);
  356. }
  357. #define K(x) ((x) << (PAGE_SHIFT-10))
  358. static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem)
  359. {
  360. struct task_struct *q;
  361. struct mm_struct *mm;
  362. p = find_lock_task_mm(p);
  363. if (!p)
  364. return 1;
  365. /* mm cannot be safely dereferenced after task_unlock(p) */
  366. mm = p->mm;
  367. pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
  368. task_pid_nr(p), p->comm, K(p->mm->total_vm),
  369. K(get_mm_counter(p->mm, MM_ANONPAGES)),
  370. K(get_mm_counter(p->mm, MM_FILEPAGES)));
  371. task_unlock(p);
  372. /*
  373. * Kill all processes sharing p->mm in other thread groups, if any.
  374. * They don't get access to memory reserves or a higher scheduler
  375. * priority, though, to avoid depletion of all memory or task
  376. * starvation. This prevents mm->mmap_sem livelock when an oom killed
  377. * task cannot exit because it requires the semaphore and its contended
  378. * by another thread trying to allocate memory itself. That thread will
  379. * now get access to memory reserves since it has a pending fatal
  380. * signal.
  381. */
  382. for_each_process(q)
  383. if (q->mm == mm && !same_thread_group(q, p)) {
  384. task_lock(q); /* Protect ->comm from prctl() */
  385. pr_err("Kill process %d (%s) sharing same memory\n",
  386. task_pid_nr(q), q->comm);
  387. task_unlock(q);
  388. force_sig(SIGKILL, q);
  389. }
  390. set_tsk_thread_flag(p, TIF_MEMDIE);
  391. force_sig(SIGKILL, p);
  392. return 0;
  393. }
  394. #undef K
  395. static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
  396. unsigned int points, unsigned long totalpages,
  397. struct mem_cgroup *mem, nodemask_t *nodemask,
  398. const char *message)
  399. {
  400. struct task_struct *victim = p;
  401. struct task_struct *child;
  402. struct task_struct *t = p;
  403. unsigned int victim_points = 0;
  404. if (printk_ratelimit())
  405. dump_header(p, gfp_mask, order, mem, nodemask);
  406. /*
  407. * If the task is already exiting, don't alarm the sysadmin or kill
  408. * its children or threads, just set TIF_MEMDIE so it can die quickly
  409. */
  410. if (p->flags & PF_EXITING) {
  411. set_tsk_thread_flag(p, TIF_MEMDIE);
  412. return 0;
  413. }
  414. task_lock(p);
  415. pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
  416. message, task_pid_nr(p), p->comm, points);
  417. task_unlock(p);
  418. /*
  419. * If any of p's children has a different mm and is eligible for kill,
  420. * the one with the highest badness() score is sacrificed for its
  421. * parent. This attempts to lose the minimal amount of work done while
  422. * still freeing memory.
  423. */
  424. do {
  425. list_for_each_entry(child, &t->children, sibling) {
  426. unsigned int child_points;
  427. if (child->mm == p->mm)
  428. continue;
  429. /*
  430. * oom_badness() returns 0 if the thread is unkillable
  431. */
  432. child_points = oom_badness(child, mem, nodemask,
  433. totalpages);
  434. if (child_points > victim_points) {
  435. victim = child;
  436. victim_points = child_points;
  437. }
  438. }
  439. } while_each_thread(p, t);
  440. return oom_kill_task(victim, mem);
  441. }
  442. /*
  443. * Determines whether the kernel must panic because of the panic_on_oom sysctl.
  444. */
  445. static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
  446. int order, const nodemask_t *nodemask)
  447. {
  448. if (likely(!sysctl_panic_on_oom))
  449. return;
  450. if (sysctl_panic_on_oom != 2) {
  451. /*
  452. * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
  453. * does not panic for cpuset, mempolicy, or memcg allocation
  454. * failures.
  455. */
  456. if (constraint != CONSTRAINT_NONE)
  457. return;
  458. }
  459. read_lock(&tasklist_lock);
  460. dump_header(NULL, gfp_mask, order, NULL, nodemask);
  461. read_unlock(&tasklist_lock);
  462. panic("Out of memory: %s panic_on_oom is enabled\n",
  463. sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
  464. }
  465. #ifdef CONFIG_CGROUP_MEM_RES_CTLR
  466. void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
  467. {
  468. unsigned long limit;
  469. unsigned int points = 0;
  470. struct task_struct *p;
  471. /*
  472. * If current has a pending SIGKILL, then automatically select it. The
  473. * goal is to allow it to allocate so that it may quickly exit and free
  474. * its memory.
  475. */
  476. if (fatal_signal_pending(current)) {
  477. set_thread_flag(TIF_MEMDIE);
  478. return;
  479. }
  480. check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL);
  481. limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT;
  482. read_lock(&tasklist_lock);
  483. retry:
  484. p = select_bad_process(&points, limit, mem, NULL);
  485. if (!p || PTR_ERR(p) == -1UL)
  486. goto out;
  487. if (oom_kill_process(p, gfp_mask, 0, points, limit, mem, NULL,
  488. "Memory cgroup out of memory"))
  489. goto retry;
  490. out:
  491. read_unlock(&tasklist_lock);
  492. }
  493. #endif
  494. static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
  495. int register_oom_notifier(struct notifier_block *nb)
  496. {
  497. return blocking_notifier_chain_register(&oom_notify_list, nb);
  498. }
  499. EXPORT_SYMBOL_GPL(register_oom_notifier);
  500. int unregister_oom_notifier(struct notifier_block *nb)
  501. {
  502. return blocking_notifier_chain_unregister(&oom_notify_list, nb);
  503. }
  504. EXPORT_SYMBOL_GPL(unregister_oom_notifier);
  505. /*
  506. * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
  507. * if a parallel OOM killing is already taking place that includes a zone in
  508. * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
  509. */
  510. int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
  511. {
  512. struct zoneref *z;
  513. struct zone *zone;
  514. int ret = 1;
  515. spin_lock(&zone_scan_lock);
  516. for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
  517. if (zone_is_oom_locked(zone)) {
  518. ret = 0;
  519. goto out;
  520. }
  521. }
  522. for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
  523. /*
  524. * Lock each zone in the zonelist under zone_scan_lock so a
  525. * parallel invocation of try_set_zonelist_oom() doesn't succeed
  526. * when it shouldn't.
  527. */
  528. zone_set_flag(zone, ZONE_OOM_LOCKED);
  529. }
  530. out:
  531. spin_unlock(&zone_scan_lock);
  532. return ret;
  533. }
  534. /*
  535. * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
  536. * allocation attempts with zonelists containing them may now recall the OOM
  537. * killer, if necessary.
  538. */
  539. void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
  540. {
  541. struct zoneref *z;
  542. struct zone *zone;
  543. spin_lock(&zone_scan_lock);
  544. for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
  545. zone_clear_flag(zone, ZONE_OOM_LOCKED);
  546. }
  547. spin_unlock(&zone_scan_lock);
  548. }
  549. /*
  550. * Try to acquire the oom killer lock for all system zones. Returns zero if a
  551. * parallel oom killing is taking place, otherwise locks all zones and returns
  552. * non-zero.
  553. */
  554. static int try_set_system_oom(void)
  555. {
  556. struct zone *zone;
  557. int ret = 1;
  558. spin_lock(&zone_scan_lock);
  559. for_each_populated_zone(zone)
  560. if (zone_is_oom_locked(zone)) {
  561. ret = 0;
  562. goto out;
  563. }
  564. for_each_populated_zone(zone)
  565. zone_set_flag(zone, ZONE_OOM_LOCKED);
  566. out:
  567. spin_unlock(&zone_scan_lock);
  568. return ret;
  569. }
  570. /*
  571. * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation
  572. * attempts or page faults may now recall the oom killer, if necessary.
  573. */
  574. static void clear_system_oom(void)
  575. {
  576. struct zone *zone;
  577. spin_lock(&zone_scan_lock);
  578. for_each_populated_zone(zone)
  579. zone_clear_flag(zone, ZONE_OOM_LOCKED);
  580. spin_unlock(&zone_scan_lock);
  581. }
  582. /**
  583. * out_of_memory - kill the "best" process when we run out of memory
  584. * @zonelist: zonelist pointer
  585. * @gfp_mask: memory allocation flags
  586. * @order: amount of memory being requested as a power of 2
  587. * @nodemask: nodemask passed to page allocator
  588. *
  589. * If we run out of memory, we have the choice between either
  590. * killing a random task (bad), letting the system crash (worse)
  591. * OR try to be smart about which process to kill. Note that we
  592. * don't have to be perfect here, we just have to be good.
  593. */
  594. void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
  595. int order, nodemask_t *nodemask)
  596. {
  597. const nodemask_t *mpol_mask;
  598. struct task_struct *p;
  599. unsigned long totalpages;
  600. unsigned long freed = 0;
  601. unsigned int points;
  602. enum oom_constraint constraint = CONSTRAINT_NONE;
  603. int killed = 0;
  604. blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
  605. if (freed > 0)
  606. /* Got some memory back in the last second. */
  607. return;
  608. /*
  609. * If current has a pending SIGKILL, then automatically select it. The
  610. * goal is to allow it to allocate so that it may quickly exit and free
  611. * its memory.
  612. */
  613. if (fatal_signal_pending(current)) {
  614. set_thread_flag(TIF_MEMDIE);
  615. return;
  616. }
  617. /*
  618. * Check if there were limitations on the allocation (only relevant for
  619. * NUMA) that may require different handling.
  620. */
  621. constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
  622. &totalpages);
  623. mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
  624. check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
  625. read_lock(&tasklist_lock);
  626. if (sysctl_oom_kill_allocating_task &&
  627. !oom_unkillable_task(current, NULL, nodemask) &&
  628. current->mm && !atomic_read(&current->mm->oom_disable_count)) {
  629. /*
  630. * oom_kill_process() needs tasklist_lock held. If it returns
  631. * non-zero, current could not be killed so we must fallback to
  632. * the tasklist scan.
  633. */
  634. if (!oom_kill_process(current, gfp_mask, order, 0, totalpages,
  635. NULL, nodemask,
  636. "Out of memory (oom_kill_allocating_task)"))
  637. goto out;
  638. }
  639. retry:
  640. p = select_bad_process(&points, totalpages, NULL, mpol_mask);
  641. if (PTR_ERR(p) == -1UL)
  642. goto out;
  643. /* Found nothing?!?! Either we hang forever, or we panic. */
  644. if (!p) {
  645. dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
  646. read_unlock(&tasklist_lock);
  647. panic("Out of memory and no killable processes...\n");
  648. }
  649. if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
  650. nodemask, "Out of memory"))
  651. goto retry;
  652. killed = 1;
  653. out:
  654. read_unlock(&tasklist_lock);
  655. /*
  656. * Give "p" a good chance of killing itself before we
  657. * retry to allocate memory unless "p" is current
  658. */
  659. if (killed && !test_thread_flag(TIF_MEMDIE))
  660. schedule_timeout_uninterruptible(1);
  661. }
  662. /*
  663. * The pagefault handler calls here because it is out of memory, so kill a
  664. * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel
  665. * oom killing is already in progress so do nothing. If a task is found with
  666. * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit.
  667. */
  668. void pagefault_out_of_memory(void)
  669. {
  670. if (try_set_system_oom()) {
  671. out_of_memory(NULL, 0, 0, NULL);
  672. clear_system_oom();
  673. }
  674. if (!test_thread_flag(TIF_MEMDIE))
  675. schedule_timeout_uninterruptible(1);
  676. }