~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/mm/oom_kill.c

Version: ~ [ linux-5.4-rc3 ] ~ [ linux-5.3.6 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.79 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.149 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.196 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.196 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.75 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 #include <linux/ftrace.h>
 37 #include <linux/ratelimit.h>
 38 
 39 #define CREATE_TRACE_POINTS
 40 #include <trace/events/oom.h>
 41 
 42 int sysctl_panic_on_oom;
 43 int sysctl_oom_kill_allocating_task;
 44 int sysctl_oom_dump_tasks = 1;
 45 static DEFINE_SPINLOCK(zone_scan_lock);
 46 
 47 #ifdef CONFIG_NUMA
 48 /**
 49  * has_intersects_mems_allowed() - check task eligiblity for kill
 50  * @start: task struct of which task to consider
 51  * @mask: nodemask passed to page allocator for mempolicy ooms
 52  *
 53  * Task eligibility is determined by whether or not a candidate task, @tsk,
 54  * shares the same mempolicy nodes as current if it is bound by such a policy
 55  * and whether or not it has the same set of allowed cpuset nodes.
 56  */
 57 static bool has_intersects_mems_allowed(struct task_struct *start,
 58                                         const nodemask_t *mask)
 59 {
 60         struct task_struct *tsk;
 61         bool ret = false;
 62 
 63         rcu_read_lock();
 64         for_each_thread(start, tsk) {
 65                 if (mask) {
 66                         /*
 67                          * If this is a mempolicy constrained oom, tsk's
 68                          * cpuset is irrelevant.  Only return true if its
 69                          * mempolicy intersects current, otherwise it may be
 70                          * needlessly killed.
 71                          */
 72                         ret = mempolicy_nodemask_intersects(tsk, mask);
 73                 } else {
 74                         /*
 75                          * This is not a mempolicy constrained oom, so only
 76                          * check the mems of tsk's cpuset.
 77                          */
 78                         ret = cpuset_mems_allowed_intersects(current, tsk);
 79                 }
 80                 if (ret)
 81                         break;
 82         }
 83         rcu_read_unlock();
 84 
 85         return ret;
 86 }
 87 #else
 88 static bool has_intersects_mems_allowed(struct task_struct *tsk,
 89                                         const nodemask_t *mask)
 90 {
 91         return true;
 92 }
 93 #endif /* CONFIG_NUMA */
 94 
 95 /*
 96  * The process p may have detached its own ->mm while exiting or through
 97  * use_mm(), but one or more of its subthreads may still have a valid
 98  * pointer.  Return p, or any of its subthreads with a valid ->mm, with
 99  * task_lock() held.
100  */
101 struct task_struct *find_lock_task_mm(struct task_struct *p)
102 {
103         struct task_struct *t;
104 
105         rcu_read_lock();
106 
107         for_each_thread(p, t) {
108                 task_lock(t);
109                 if (likely(t->mm))
110                         goto found;
111                 task_unlock(t);
112         }
113         t = NULL;
114 found:
115         rcu_read_unlock();
116 
117         return t;
118 }
119 
120 /* return true if the task is not adequate as candidate victim task. */
121 static bool oom_unkillable_task(struct task_struct *p,
122                 const struct mem_cgroup *memcg, const nodemask_t *nodemask)
123 {
124         if (is_global_init(p))
125                 return true;
126         if (p->flags & PF_KTHREAD)
127                 return true;
128 
129         /* When mem_cgroup_out_of_memory() and p is not member of the group */
130         if (memcg && !task_in_mem_cgroup(p, memcg))
131                 return true;
132 
133         /* p may not have freeable memory in nodemask */
134         if (!has_intersects_mems_allowed(p, nodemask))
135                 return true;
136 
137         return false;
138 }
139 
140 /**
141  * oom_badness - heuristic function to determine which candidate task to kill
142  * @p: task struct of which task we should calculate
143  * @totalpages: total present RAM allowed for page allocation
144  *
145  * The heuristic for determining which task to kill is made to be as simple and
146  * predictable as possible.  The goal is to return the highest value for the
147  * task consuming the most memory to avoid subsequent oom failures.
148  */
149 unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
150                           const nodemask_t *nodemask, unsigned long totalpages)
151 {
152         long points;
153         long adj;
154 
155         if (oom_unkillable_task(p, memcg, nodemask))
156                 return 0;
157 
158         p = find_lock_task_mm(p);
159         if (!p)
160                 return 0;
161 
162         adj = (long)p->signal->oom_score_adj;
163         if (adj == OOM_SCORE_ADJ_MIN) {
164                 task_unlock(p);
165                 return 0;
166         }
167 
168         /*
169          * The baseline for the badness score is the proportion of RAM that each
170          * task's rss, pagetable and swap space use.
171          */
172         points = get_mm_rss(p->mm) + p->mm->nr_ptes +
173                  get_mm_counter(p->mm, MM_SWAPENTS);
174         task_unlock(p);
175 
176         /*
177          * Root processes get 3% bonus, just like the __vm_enough_memory()
178          * implementation used by LSMs.
179          */
180         if (has_capability_noaudit(p, CAP_SYS_ADMIN))
181                 points -= (points * 3) / 100;
182 
183         /* Normalize to oom_score_adj units */
184         adj *= totalpages / 1000;
185         points += adj;
186 
187         /*
188          * Never return 0 for an eligible task regardless of the root bonus and
189          * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
190          */
191         return points > 0 ? points : 1;
192 }
193 
194 /*
195  * Determine the type of allocation constraint.
196  */
197 #ifdef CONFIG_NUMA
198 static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
199                                 gfp_t gfp_mask, nodemask_t *nodemask,
200                                 unsigned long *totalpages)
201 {
202         struct zone *zone;
203         struct zoneref *z;
204         enum zone_type high_zoneidx = gfp_zone(gfp_mask);
205         bool cpuset_limited = false;
206         int nid;
207 
208         /* Default to all available memory */
209         *totalpages = totalram_pages + total_swap_pages;
210 
211         if (!zonelist)
212                 return CONSTRAINT_NONE;
213         /*
214          * Reach here only when __GFP_NOFAIL is used. So, we should avoid
215          * to kill current.We have to random task kill in this case.
216          * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
217          */
218         if (gfp_mask & __GFP_THISNODE)
219                 return CONSTRAINT_NONE;
220 
221         /*
222          * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
223          * the page allocator means a mempolicy is in effect.  Cpuset policy
224          * is enforced in get_page_from_freelist().
225          */
226         if (nodemask && !nodes_subset(node_states[N_MEMORY], *nodemask)) {
227                 *totalpages = total_swap_pages;
228                 for_each_node_mask(nid, *nodemask)
229                         *totalpages += node_spanned_pages(nid);
230                 return CONSTRAINT_MEMORY_POLICY;
231         }
232 
233         /* Check this allocation failure is caused by cpuset's wall function */
234         for_each_zone_zonelist_nodemask(zone, z, zonelist,
235                         high_zoneidx, nodemask)
236                 if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
237                         cpuset_limited = true;
238 
239         if (cpuset_limited) {
240                 *totalpages = total_swap_pages;
241                 for_each_node_mask(nid, cpuset_current_mems_allowed)
242                         *totalpages += node_spanned_pages(nid);
243                 return CONSTRAINT_CPUSET;
244         }
245         return CONSTRAINT_NONE;
246 }
247 #else
248 static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
249                                 gfp_t gfp_mask, nodemask_t *nodemask,
250                                 unsigned long *totalpages)
251 {
252         *totalpages = totalram_pages + total_swap_pages;
253         return CONSTRAINT_NONE;
254 }
255 #endif
256 
257 enum oom_scan_t oom_scan_process_thread(struct task_struct *task,
258                 unsigned long totalpages, const nodemask_t *nodemask,
259                 bool force_kill)
260 {
261         if (task->exit_state)
262                 return OOM_SCAN_CONTINUE;
263         if (oom_unkillable_task(task, NULL, nodemask))
264                 return OOM_SCAN_CONTINUE;
265 
266         /*
267          * This task already has access to memory reserves and is being killed.
268          * Don't allow any other task to have access to the reserves.
269          */
270         if (test_tsk_thread_flag(task, TIF_MEMDIE)) {
271                 if (unlikely(frozen(task)))
272                         __thaw_task(task);
273                 if (!force_kill)
274                         return OOM_SCAN_ABORT;
275         }
276         if (!task->mm)
277                 return OOM_SCAN_CONTINUE;
278 
279         /*
280          * If task is allocating a lot of memory and has been marked to be
281          * killed first if it triggers an oom, then select it.
282          */
283         if (oom_task_origin(task))
284                 return OOM_SCAN_SELECT;
285 
286         if (task->flags & PF_EXITING && !force_kill) {
287                 /*
288                  * If this task is not being ptraced on exit, then wait for it
289                  * to finish before killing some other task unnecessarily.
290                  */
291                 if (!(task->group_leader->ptrace & PT_TRACE_EXIT))
292                         return OOM_SCAN_ABORT;
293         }
294         return OOM_SCAN_OK;
295 }
296 
297 /*
298  * Simple selection loop. We chose the process with the highest
299  * number of 'points'.  Returns -1 on scan abort.
300  *
301  * (not docbooked, we don't want this one cluttering up the manual)
302  */
303 static struct task_struct *select_bad_process(unsigned int *ppoints,
304                 unsigned long totalpages, const nodemask_t *nodemask,
305                 bool force_kill)
306 {
307         struct task_struct *g, *p;
308         struct task_struct *chosen = NULL;
309         unsigned long chosen_points = 0;
310 
311         rcu_read_lock();
312         for_each_process_thread(g, p) {
313                 unsigned int points;
314 
315                 switch (oom_scan_process_thread(p, totalpages, nodemask,
316                                                 force_kill)) {
317                 case OOM_SCAN_SELECT:
318                         chosen = p;
319                         chosen_points = ULONG_MAX;
320                         /* fall through */
321                 case OOM_SCAN_CONTINUE:
322                         continue;
323                 case OOM_SCAN_ABORT:
324                         rcu_read_unlock();
325                         return (struct task_struct *)(-1UL);
326                 case OOM_SCAN_OK:
327                         break;
328                 };
329                 points = oom_badness(p, NULL, nodemask, totalpages);
330                 if (!points || points < chosen_points)
331                         continue;
332                 /* Prefer thread group leaders for display purposes */
333                 if (points == chosen_points && thread_group_leader(chosen))
334                         continue;
335 
336                 chosen = p;
337                 chosen_points = points;
338         }
339         if (chosen)
340                 get_task_struct(chosen);
341         rcu_read_unlock();
342 
343         *ppoints = chosen_points * 1000 / totalpages;
344         return chosen;
345 }
346 
347 /**
348  * dump_tasks - dump current memory state of all system tasks
349  * @memcg: current's memory controller, if constrained
350  * @nodemask: nodemask passed to page allocator for mempolicy ooms
351  *
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.
355  * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
356  * swapents, oom_score_adj value, and name.
357  */
358 static void dump_tasks(const struct mem_cgroup *memcg, const nodemask_t *nodemask)
359 {
360         struct task_struct *p;
361         struct task_struct *task;
362 
363         pr_info("[ pid ]   uid  tgid total_vm      rss nr_ptes swapents oom_score_adj name\n");
364         rcu_read_lock();
365         for_each_process(p) {
366                 if (oom_unkillable_task(p, memcg, nodemask))
367                         continue;
368 
369                 task = find_lock_task_mm(p);
370                 if (!task) {
371                         /*
372                          * This is a kthread or all of p's threads have already
373                          * detached their mm's.  There's no need to report
374                          * them; they can't be oom killed anyway.
375                          */
376                         continue;
377                 }
378 
379                 pr_info("[%5d] %5d %5d %8lu %8lu %7lu %8lu         %5hd %s\n",
380                         task->pid, from_kuid(&init_user_ns, task_uid(task)),
381                         task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
382                         task->mm->nr_ptes,
383                         get_mm_counter(task->mm, MM_SWAPENTS),
384                         task->signal->oom_score_adj, task->comm);
385                 task_unlock(task);
386         }
387         rcu_read_unlock();
388 }
389 
390 static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
391                         struct mem_cgroup *memcg, const nodemask_t *nodemask)
392 {
393         task_lock(current);
394         pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
395                 "oom_score_adj=%hd\n",
396                 current->comm, gfp_mask, order,
397                 current->signal->oom_score_adj);
398         cpuset_print_task_mems_allowed(current);
399         task_unlock(current);
400         dump_stack();
401         if (memcg)
402                 mem_cgroup_print_oom_info(memcg, p);
403         else
404                 show_mem(SHOW_MEM_FILTER_NODES);
405         if (sysctl_oom_dump_tasks)
406                 dump_tasks(memcg, nodemask);
407 }
408 
409 /*
410  * Number of OOM killer invocations (including memcg OOM killer).
411  * Primarily used by PM freezer to check for potential races with
412  * OOM killed frozen task.
413  */
414 static atomic_t oom_kills = ATOMIC_INIT(0);
415 
416 int oom_kills_count(void)
417 {
418         return atomic_read(&oom_kills);
419 }
420 
421 void note_oom_kill(void)
422 {
423         atomic_inc(&oom_kills);
424 }
425 
426 #define K(x) ((x) << (PAGE_SHIFT-10))
427 /*
428  * Must be called while holding a reference to p, which will be released upon
429  * returning.
430  */
431 void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
432                       unsigned int points, unsigned long totalpages,
433                       struct mem_cgroup *memcg, nodemask_t *nodemask,
434                       const char *message)
435 {
436         struct task_struct *victim = p;
437         struct task_struct *child;
438         struct task_struct *t;
439         struct mm_struct *mm;
440         unsigned int victim_points = 0;
441         static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
442                                               DEFAULT_RATELIMIT_BURST);
443 
444         /*
445          * If the task is already exiting, don't alarm the sysadmin or kill
446          * its children or threads, just set TIF_MEMDIE so it can die quickly
447          */
448         if (p->flags & PF_EXITING) {
449                 set_tsk_thread_flag(p, TIF_MEMDIE);
450                 put_task_struct(p);
451                 return;
452         }
453 
454         if (__ratelimit(&oom_rs))
455                 dump_header(p, gfp_mask, order, memcg, nodemask);
456 
457         task_lock(p);
458         pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
459                 message, task_pid_nr(p), p->comm, points);
460         task_unlock(p);
461 
462         /*
463          * If any of p's children has a different mm and is eligible for kill,
464          * the one with the highest oom_badness() score is sacrificed for its
465          * parent.  This attempts to lose the minimal amount of work done while
466          * still freeing memory.
467          */
468         read_lock(&tasklist_lock);
469         for_each_thread(p, t) {
470                 list_for_each_entry(child, &t->children, sibling) {
471                         unsigned int child_points;
472 
473                         if (child->mm == p->mm)
474                                 continue;
475                         /*
476                          * oom_badness() returns 0 if the thread is unkillable
477                          */
478                         child_points = oom_badness(child, memcg, nodemask,
479                                                                 totalpages);
480                         if (child_points > victim_points) {
481                                 put_task_struct(victim);
482                                 victim = child;
483                                 victim_points = child_points;
484                                 get_task_struct(victim);
485                         }
486                 }
487         }
488         read_unlock(&tasklist_lock);
489 
490         p = find_lock_task_mm(victim);
491         if (!p) {
492                 put_task_struct(victim);
493                 return;
494         } else if (victim != p) {
495                 get_task_struct(p);
496                 put_task_struct(victim);
497                 victim = p;
498         }
499 
500         /* mm cannot safely be dereferenced after task_unlock(victim) */
501         mm = victim->mm;
502         pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
503                 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
504                 K(get_mm_counter(victim->mm, MM_ANONPAGES)),
505                 K(get_mm_counter(victim->mm, MM_FILEPAGES)));
506         task_unlock(victim);
507 
508         /*
509          * Kill all user processes sharing victim->mm in other thread groups, if
510          * any.  They don't get access to memory reserves, though, to avoid
511          * depletion of all memory.  This prevents mm->mmap_sem livelock when an
512          * oom killed thread cannot exit because it requires the semaphore and
513          * its contended by another thread trying to allocate memory itself.
514          * That thread will now get access to memory reserves since it has a
515          * pending fatal signal.
516          */
517         rcu_read_lock();
518         for_each_process(p)
519                 if (p->mm == mm && !same_thread_group(p, victim) &&
520                     !(p->flags & PF_KTHREAD)) {
521                         if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
522                                 continue;
523 
524                         task_lock(p);   /* Protect ->comm from prctl() */
525                         pr_err("Kill process %d (%s) sharing same memory\n",
526                                 task_pid_nr(p), p->comm);
527                         task_unlock(p);
528                         do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
529                 }
530         rcu_read_unlock();
531 
532         set_tsk_thread_flag(victim, TIF_MEMDIE);
533         do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
534         put_task_struct(victim);
535 }
536 #undef K
537 
538 /*
539  * Determines whether the kernel must panic because of the panic_on_oom sysctl.
540  */
541 void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
542                         int order, const nodemask_t *nodemask)
543 {
544         if (likely(!sysctl_panic_on_oom))
545                 return;
546         if (sysctl_panic_on_oom != 2) {
547                 /*
548                  * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
549                  * does not panic for cpuset, mempolicy, or memcg allocation
550                  * failures.
551                  */
552                 if (constraint != CONSTRAINT_NONE)
553                         return;
554         }
555         dump_header(NULL, gfp_mask, order, NULL, nodemask);
556         panic("Out of memory: %s panic_on_oom is enabled\n",
557                 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
558 }
559 
560 static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
561 
562 int register_oom_notifier(struct notifier_block *nb)
563 {
564         return blocking_notifier_chain_register(&oom_notify_list, nb);
565 }
566 EXPORT_SYMBOL_GPL(register_oom_notifier);
567 
568 int unregister_oom_notifier(struct notifier_block *nb)
569 {
570         return blocking_notifier_chain_unregister(&oom_notify_list, nb);
571 }
572 EXPORT_SYMBOL_GPL(unregister_oom_notifier);
573 
574 /*
575  * Try to acquire the OOM killer lock for the zones in zonelist.  Returns zero
576  * if a parallel OOM killing is already taking place that includes a zone in
577  * the zonelist.  Otherwise, locks all zones in the zonelist and returns 1.
578  */
579 int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
580 {
581         struct zoneref *z;
582         struct zone *zone;
583         int ret = 1;
584 
585         spin_lock(&zone_scan_lock);
586         for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
587                 if (zone_is_oom_locked(zone)) {
588                         ret = 0;
589                         goto out;
590                 }
591         }
592 
593         for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
594                 /*
595                  * Lock each zone in the zonelist under zone_scan_lock so a
596                  * parallel invocation of try_set_zonelist_oom() doesn't succeed
597                  * when it shouldn't.
598                  */
599                 zone_set_flag(zone, ZONE_OOM_LOCKED);
600         }
601 
602 out:
603         spin_unlock(&zone_scan_lock);
604         return ret;
605 }
606 
607 /*
608  * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
609  * allocation attempts with zonelists containing them may now recall the OOM
610  * killer, if necessary.
611  */
612 void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
613 {
614         struct zoneref *z;
615         struct zone *zone;
616 
617         spin_lock(&zone_scan_lock);
618         for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
619                 zone_clear_flag(zone, ZONE_OOM_LOCKED);
620         }
621         spin_unlock(&zone_scan_lock);
622 }
623 
624 /**
625  * out_of_memory - kill the "best" process when we run out of memory
626  * @zonelist: zonelist pointer
627  * @gfp_mask: memory allocation flags
628  * @order: amount of memory being requested as a power of 2
629  * @nodemask: nodemask passed to page allocator
630  * @force_kill: true if a task must be killed, even if others are exiting
631  *
632  * If we run out of memory, we have the choice between either
633  * killing a random task (bad), letting the system crash (worse)
634  * OR try to be smart about which process to kill. Note that we
635  * don't have to be perfect here, we just have to be good.
636  */
637 void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
638                 int order, nodemask_t *nodemask, bool force_kill)
639 {
640         const nodemask_t *mpol_mask;
641         struct task_struct *p;
642         unsigned long totalpages;
643         unsigned long freed = 0;
644         unsigned int uninitialized_var(points);
645         enum oom_constraint constraint = CONSTRAINT_NONE;
646         int killed = 0;
647 
648         blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
649         if (freed > 0)
650                 /* Got some memory back in the last second. */
651                 return;
652 
653         /*
654          * If current has a pending SIGKILL or is exiting, then automatically
655          * select it.  The goal is to allow it to allocate so that it may
656          * quickly exit and free its memory.
657          */
658         if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
659                 set_thread_flag(TIF_MEMDIE);
660                 return;
661         }
662 
663         /*
664          * Check if there were limitations on the allocation (only relevant for
665          * NUMA) that may require different handling.
666          */
667         constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
668                                                 &totalpages);
669         mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
670         check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
671 
672         if (sysctl_oom_kill_allocating_task && current->mm &&
673             !oom_unkillable_task(current, NULL, nodemask) &&
674             current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
675                 get_task_struct(current);
676                 oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL,
677                                  nodemask,
678                                  "Out of memory (oom_kill_allocating_task)");
679                 goto out;
680         }
681 
682         p = select_bad_process(&points, totalpages, mpol_mask, force_kill);
683         /* Found nothing?!?! Either we hang forever, or we panic. */
684         if (!p) {
685                 dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
686                 panic("Out of memory and no killable processes...\n");
687         }
688         if (p != (void *)-1UL) {
689                 oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
690                                  nodemask, "Out of memory");
691                 killed = 1;
692         }
693 out:
694         /*
695          * Give the killed threads a good chance of exiting before trying to
696          * allocate memory again.
697          */
698         if (killed)
699                 schedule_timeout_killable(1);
700 }
701 
702 /*
703  * The pagefault handler calls here because it is out of memory, so kill a
704  * memory-hogging task.  If any populated zone has ZONE_OOM_LOCKED set, a
705  * parallel oom killing is already in progress so do nothing.
706  */
707 void pagefault_out_of_memory(void)
708 {
709         struct zonelist *zonelist;
710 
711         if (mem_cgroup_oom_synchronize(true))
712                 return;
713 
714         zonelist = node_zonelist(first_online_node, GFP_KERNEL);
715         if (try_set_zonelist_oom(zonelist, GFP_KERNEL)) {
716                 out_of_memory(NULL, 0, 0, NULL, false);
717                 clear_zonelist_oom(zonelist, GFP_KERNEL);
718         }
719 }
720 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp