blob: dff991e0681e85a5308ba097533e346e621bc954 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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...
David Rientjesa63d83f2010-08-09 17:19:46 -07007 * Copyright (C) 2010 Google, Inc.
8 * Rewritten by David Rientjes
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
10 * The routines in this file are used to kill a process when
Paul Jacksona49335c2005-09-06 15:18:09 -070011 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
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
Alexey Dobriyan8ac773b2006-10-19 23:28:32 -070020#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/mm.h>
Alexey Dobriyan4e950f62007-07-30 02:36:13 +040022#include <linux/err.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090023#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/sched.h>
25#include <linux/swap.h>
26#include <linux/timex.h>
27#include <linux/jiffies.h>
Paul Jacksonef08e3b2005-09-06 15:18:13 -070028#include <linux/cpuset.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040029#include <linux/export.h>
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -070030#include <linux/notifier.h>
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -080031#include <linux/memcontrol.h>
David Rientjes6f48d0eb2010-08-09 17:18:52 -070032#include <linux/mempolicy.h>
David Howells5cd9c582008-08-14 11:37:28 +010033#include <linux/security.h>
David Rientjesedd45542011-03-22 16:30:12 -070034#include <linux/ptrace.h>
David Rientjesf660daa2011-10-31 17:07:07 -070035#include <linux/freezer.h>
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080036#include <linux/ftrace.h>
David Rientjesdc3f21e2012-03-21 16:33:47 -070037#include <linux/ratelimit.h>
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080038
39#define CREATE_TRACE_POINTS
40#include <trace/events/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -070042int sysctl_panic_on_oom;
David Rientjesfe071d72007-10-16 23:25:56 -070043int sysctl_oom_kill_allocating_task;
David Rientjesad915c42010-08-09 17:18:53 -070044int sysctl_oom_dump_tasks = 1;
Johannes Weinerdc564012015-06-24 16:57:19 -070045
46DEFINE_MUTEX(oom_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
David Rientjes6f48d0eb2010-08-09 17:18:52 -070048#ifdef CONFIG_NUMA
49/**
50 * has_intersects_mems_allowed() - check task eligiblity for kill
Oleg Nesterovad962442014-01-21 15:50:00 -080051 * @start: task struct of which task to consider
David Rientjes6f48d0eb2010-08-09 17:18:52 -070052 * @mask: nodemask passed to page allocator for mempolicy ooms
53 *
54 * Task eligibility is determined by whether or not a candidate task, @tsk,
55 * shares the same mempolicy nodes as current if it is bound by such a policy
56 * and whether or not it has the same set of allowed cpuset nodes.
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070057 */
Oleg Nesterovad962442014-01-21 15:50:00 -080058static bool has_intersects_mems_allowed(struct task_struct *start,
David Rientjes6f48d0eb2010-08-09 17:18:52 -070059 const nodemask_t *mask)
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070060{
Oleg Nesterovad962442014-01-21 15:50:00 -080061 struct task_struct *tsk;
62 bool ret = false;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070063
Oleg Nesterovad962442014-01-21 15:50:00 -080064 rcu_read_lock();
Oleg Nesterov1da4db02014-01-21 15:49:58 -080065 for_each_thread(start, tsk) {
David Rientjes6f48d0eb2010-08-09 17:18:52 -070066 if (mask) {
67 /*
68 * If this is a mempolicy constrained oom, tsk's
69 * cpuset is irrelevant. Only return true if its
70 * mempolicy intersects current, otherwise it may be
71 * needlessly killed.
72 */
Oleg Nesterovad962442014-01-21 15:50:00 -080073 ret = mempolicy_nodemask_intersects(tsk, mask);
David Rientjes6f48d0eb2010-08-09 17:18:52 -070074 } else {
75 /*
76 * This is not a mempolicy constrained oom, so only
77 * check the mems of tsk's cpuset.
78 */
Oleg Nesterovad962442014-01-21 15:50:00 -080079 ret = cpuset_mems_allowed_intersects(current, tsk);
David Rientjes6f48d0eb2010-08-09 17:18:52 -070080 }
Oleg Nesterovad962442014-01-21 15:50:00 -080081 if (ret)
82 break;
Oleg Nesterov1da4db02014-01-21 15:49:58 -080083 }
Oleg Nesterovad962442014-01-21 15:50:00 -080084 rcu_read_unlock();
KOSAKI Motohirodf1090a2010-08-09 17:19:39 -070085
Oleg Nesterovad962442014-01-21 15:50:00 -080086 return ret;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070087}
David Rientjes6f48d0eb2010-08-09 17:18:52 -070088#else
89static bool has_intersects_mems_allowed(struct task_struct *tsk,
90 const nodemask_t *mask)
91{
92 return true;
93}
94#endif /* CONFIG_NUMA */
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070095
David Rientjes6f48d0eb2010-08-09 17:18:52 -070096/*
97 * The process p may have detached its own ->mm while exiting or through
98 * use_mm(), but one or more of its subthreads may still have a valid
99 * pointer. Return p, or any of its subthreads with a valid ->mm, with
100 * task_lock() held.
101 */
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -0700102struct task_struct *find_lock_task_mm(struct task_struct *p)
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700103{
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800104 struct task_struct *t;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700105
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800106 rcu_read_lock();
107
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800108 for_each_thread(p, t) {
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700109 task_lock(t);
110 if (likely(t->mm))
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800111 goto found;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700112 task_unlock(t);
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800113 }
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800114 t = NULL;
115found:
116 rcu_read_unlock();
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700117
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800118 return t;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700119}
120
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700121/* return true if the task is not adequate as candidate victim task. */
David Rientjese85bfd32010-09-22 13:05:10 -0700122static bool oom_unkillable_task(struct task_struct *p,
Johannes Weiner2314b422014-12-10 15:44:33 -0800123 struct mem_cgroup *memcg, const nodemask_t *nodemask)
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700124{
125 if (is_global_init(p))
126 return true;
127 if (p->flags & PF_KTHREAD)
128 return true;
129
130 /* When mem_cgroup_out_of_memory() and p is not member of the group */
Johannes Weiner72835c82012-01-12 17:18:32 -0800131 if (memcg && !task_in_mem_cgroup(p, memcg))
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700132 return true;
133
134 /* p may not have freeable memory in nodemask */
135 if (!has_intersects_mems_allowed(p, nodemask))
136 return true;
137
138 return false;
139}
140
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141/**
David Rientjesa63d83f2010-08-09 17:19:46 -0700142 * oom_badness - heuristic function to determine which candidate task to kill
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 * @p: task struct of which task we should calculate
David Rientjesa63d83f2010-08-09 17:19:46 -0700144 * @totalpages: total present RAM allowed for page allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 *
David Rientjesa63d83f2010-08-09 17:19:46 -0700146 * The heuristic for determining which task to kill is made to be as simple and
147 * predictable as possible. The goal is to return the highest value for the
148 * task consuming the most memory to avoid subsequent oom failures.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 */
David Rientjesa7f638f2012-05-29 15:06:47 -0700150unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
151 const nodemask_t *nodemask, unsigned long totalpages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152{
David Rientjes1e11ad82012-06-08 13:21:26 -0700153 long points;
David Rientjes61eafb02012-06-20 12:52:58 -0700154 long adj;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700155
Johannes Weiner72835c82012-01-12 17:18:32 -0800156 if (oom_unkillable_task(p, memcg, nodemask))
KOSAKI Motohiro26ebc982010-08-09 17:19:37 -0700157 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700159 p = find_lock_task_mm(p);
160 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 return 0;
162
David Rientjesa9c58b902012-12-11 16:02:54 -0800163 adj = (long)p->signal->oom_score_adj;
David Rientjes61eafb02012-06-20 12:52:58 -0700164 if (adj == OOM_SCORE_ADJ_MIN) {
Michal Hocko5aecc852011-11-15 14:36:07 -0800165 task_unlock(p);
166 return 0;
167 }
168
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700170 * The baseline for the badness score is the proportion of RAM that each
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700171 * task's rss, pagetable and swap space use.
David Rientjesa63d83f2010-08-09 17:19:46 -0700172 */
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800173 points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
174 atomic_long_read(&p->mm->nr_ptes) + mm_nr_pmds(p->mm);
Andrew Morton97c2c9b82006-04-18 22:20:38 -0700175 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
177 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700178 * Root processes get 3% bonus, just like the __vm_enough_memory()
179 * implementation used by LSMs.
Hugh Dickins7ba34852007-01-05 16:37:03 -0800180 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700181 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
David Rientjes778c14a2014-01-30 15:46:11 -0800182 points -= (points * 3) / 100;
Hugh Dickins7ba34852007-01-05 16:37:03 -0800183
David Rientjes61eafb02012-06-20 12:52:58 -0700184 /* Normalize to oom_score_adj units */
185 adj *= totalpages / 1000;
186 points += adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
David Rientjesf19e8aa2010-09-22 13:04:52 -0700188 /*
David Rientjesa7f638f2012-05-29 15:06:47 -0700189 * Never return 0 for an eligible task regardless of the root bonus and
190 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
David Rientjesf19e8aa2010-09-22 13:04:52 -0700191 */
David Rientjes1e11ad82012-06-08 13:21:26 -0700192 return points > 0 ? points : 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193}
194
195/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800196 * Determine the type of allocation constraint.
197 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800198#ifdef CONFIG_NUMA
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800199static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700200 gfp_t gfp_mask, nodemask_t *nodemask,
201 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800202{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700203 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700204 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700205 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
David Rientjesa63d83f2010-08-09 17:19:46 -0700206 bool cpuset_limited = false;
207 int nid;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800208
David Rientjesa63d83f2010-08-09 17:19:46 -0700209 /* Default to all available memory */
210 *totalpages = totalram_pages + total_swap_pages;
211
212 if (!zonelist)
213 return CONSTRAINT_NONE;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800214 /*
215 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
216 * to kill current.We have to random task kill in this case.
217 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
218 */
219 if (gfp_mask & __GFP_THISNODE)
220 return CONSTRAINT_NONE;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800221
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800222 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700223 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
224 * the page allocator means a mempolicy is in effect. Cpuset policy
225 * is enforced in get_page_from_freelist().
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800226 */
Lai Jiangshanbd3a66c2012-12-12 13:51:28 -0800227 if (nodemask && !nodes_subset(node_states[N_MEMORY], *nodemask)) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700228 *totalpages = total_swap_pages;
229 for_each_node_mask(nid, *nodemask)
230 *totalpages += node_spanned_pages(nid);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800231 return CONSTRAINT_MEMORY_POLICY;
David Rientjesa63d83f2010-08-09 17:19:46 -0700232 }
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800233
234 /* Check this allocation failure is caused by cpuset's wall function */
235 for_each_zone_zonelist_nodemask(zone, z, zonelist,
236 high_zoneidx, nodemask)
Vladimir Davydov344736f2014-10-20 15:50:30 +0400237 if (!cpuset_zone_allowed(zone, gfp_mask))
David Rientjesa63d83f2010-08-09 17:19:46 -0700238 cpuset_limited = true;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800239
David Rientjesa63d83f2010-08-09 17:19:46 -0700240 if (cpuset_limited) {
241 *totalpages = total_swap_pages;
242 for_each_node_mask(nid, cpuset_current_mems_allowed)
243 *totalpages += node_spanned_pages(nid);
244 return CONSTRAINT_CPUSET;
245 }
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800246 return CONSTRAINT_NONE;
247}
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800248#else
249static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700250 gfp_t gfp_mask, nodemask_t *nodemask,
251 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800252{
David Rientjesa63d83f2010-08-09 17:19:46 -0700253 *totalpages = totalram_pages + total_swap_pages;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800254 return CONSTRAINT_NONE;
255}
256#endif
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800257
David Rientjes9cbb78b2012-07-31 16:43:44 -0700258enum oom_scan_t oom_scan_process_thread(struct task_struct *task,
259 unsigned long totalpages, const nodemask_t *nodemask,
260 bool force_kill)
David Rientjes462607e2012-07-31 16:43:40 -0700261{
David Rientjes9cbb78b2012-07-31 16:43:44 -0700262 if (oom_unkillable_task(task, NULL, nodemask))
David Rientjes462607e2012-07-31 16:43:40 -0700263 return OOM_SCAN_CONTINUE;
264
265 /*
266 * This task already has access to memory reserves and is being killed.
267 * Don't allow any other task to have access to the reserves.
268 */
269 if (test_tsk_thread_flag(task, TIF_MEMDIE)) {
David Rientjes462607e2012-07-31 16:43:40 -0700270 if (!force_kill)
271 return OOM_SCAN_ABORT;
272 }
273 if (!task->mm)
274 return OOM_SCAN_CONTINUE;
275
David Rientjese1e12d22012-12-11 16:02:56 -0800276 /*
277 * If task is allocating a lot of memory and has been marked to be
278 * killed first if it triggers an oom, then select it.
279 */
280 if (oom_task_origin(task))
281 return OOM_SCAN_SELECT;
282
Oleg Nesterov6a2d5672014-12-12 16:56:27 -0800283 if (task_will_free_mem(task) && !force_kill)
284 return OOM_SCAN_ABORT;
285
David Rientjes462607e2012-07-31 16:43:40 -0700286 return OOM_SCAN_OK;
287}
288
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800289/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 * Simple selection loop. We chose the process with the highest
Rusty Russell6b4f2b52013-07-15 11:24:08 +0930291 * number of 'points'. Returns -1 on scan abort.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 *
293 * (not docbooked, we don't want this one cluttering up the manual)
294 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700295static struct task_struct *select_bad_process(unsigned int *ppoints,
David Rientjes9cbb78b2012-07-31 16:43:44 -0700296 unsigned long totalpages, const nodemask_t *nodemask,
297 bool force_kill)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
David Rientjes3a5dda72011-03-22 16:30:09 -0700299 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 struct task_struct *chosen = NULL;
David Rientjesa7f638f2012-05-29 15:06:47 -0700301 unsigned long chosen_points = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
David Rientjes6b0c81b2012-07-31 16:43:45 -0700303 rcu_read_lock();
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800304 for_each_process_thread(g, p) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700305 unsigned int points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700306
David Rientjes9cbb78b2012-07-31 16:43:44 -0700307 switch (oom_scan_process_thread(p, totalpages, nodemask,
David Rientjes462607e2012-07-31 16:43:40 -0700308 force_kill)) {
309 case OOM_SCAN_SELECT:
310 chosen = p;
311 chosen_points = ULONG_MAX;
312 /* fall through */
313 case OOM_SCAN_CONTINUE:
Andrey Vagin30e2b412011-03-22 16:30:11 -0700314 continue;
David Rientjes462607e2012-07-31 16:43:40 -0700315 case OOM_SCAN_ABORT:
David Rientjes6b0c81b2012-07-31 16:43:45 -0700316 rcu_read_unlock();
Rusty Russell6b4f2b52013-07-15 11:24:08 +0930317 return (struct task_struct *)(-1UL);
David Rientjes462607e2012-07-31 16:43:40 -0700318 case OOM_SCAN_OK:
319 break;
320 };
David Rientjes9cbb78b2012-07-31 16:43:44 -0700321 points = oom_badness(p, NULL, nodemask, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -0800322 if (!points || points < chosen_points)
323 continue;
324 /* Prefer thread group leaders for display purposes */
325 if (points == chosen_points && thread_group_leader(chosen))
326 continue;
327
328 chosen = p;
329 chosen_points = points;
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800330 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700331 if (chosen)
332 get_task_struct(chosen);
333 rcu_read_unlock();
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700334
David Rientjesa7f638f2012-05-29 15:06:47 -0700335 *ppoints = chosen_points * 1000 / totalpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 return chosen;
337}
338
339/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700340 * dump_tasks - dump current memory state of all system tasks
Wanpeng Lidad75572012-06-20 12:53:01 -0700341 * @memcg: current's memory controller, if constrained
David Rientjese85bfd32010-09-22 13:05:10 -0700342 * @nodemask: nodemask passed to page allocator for mempolicy ooms
Randy Dunlap1b578df2008-03-19 17:00:42 -0700343 *
David Rientjese85bfd32010-09-22 13:05:10 -0700344 * Dumps the current memory state of all eligible tasks. Tasks not in the same
345 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
346 * are not shown.
David Rientjesde34d962012-07-31 16:42:56 -0700347 * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
348 * swapents, oom_score_adj value, and name.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800349 */
Johannes Weiner2314b422014-12-10 15:44:33 -0800350static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask)
David Rientjesfef1bdd2008-02-07 00:14:07 -0800351{
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700352 struct task_struct *p;
353 struct task_struct *task;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800354
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800355 pr_info("[ pid ] uid tgid total_vm rss nr_ptes nr_pmds swapents oom_score_adj name\n");
David Rientjes6b0c81b2012-07-31 16:43:45 -0700356 rcu_read_lock();
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700357 for_each_process(p) {
Johannes Weiner72835c82012-01-12 17:18:32 -0800358 if (oom_unkillable_task(p, memcg, nodemask))
David Rientjesfef1bdd2008-02-07 00:14:07 -0800359 continue;
360
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700361 task = find_lock_task_mm(p);
362 if (!task) {
David Rientjes6d2661e2009-05-28 14:34:19 -0700363 /*
David Rientjes74ab7f12010-08-09 17:18:46 -0700364 * This is a kthread or all of p's threads have already
365 * detached their mm's. There's no need to report
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700366 * them; they can't be oom killed anyway.
David Rientjes6d2661e2009-05-28 14:34:19 -0700367 */
David Rientjes6d2661e2009-05-28 14:34:19 -0700368 continue;
369 }
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700370
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800371 pr_info("[%5d] %5d %5d %8lu %8lu %7ld %7ld %8lu %5hd %s\n",
Eric W. Biederman078de5f2012-02-08 07:00:08 -0800372 task->pid, from_kuid(&init_user_ns, task_uid(task)),
373 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -0800374 atomic_long_read(&task->mm->nr_ptes),
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800375 mm_nr_pmds(task->mm),
David Rientjesde34d962012-07-31 16:42:56 -0700376 get_mm_counter(task->mm, MM_SWAPENTS),
David Rientjesa63d83f2010-08-09 17:19:46 -0700377 task->signal->oom_score_adj, task->comm);
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700378 task_unlock(task);
379 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700380 rcu_read_unlock();
David Rientjesfef1bdd2008-02-07 00:14:07 -0800381}
382
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -0800383static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
Johannes Weiner72835c82012-01-12 17:18:32 -0800384 struct mem_cgroup *memcg, const nodemask_t *nodemask)
David Rientjes1b604d72009-12-14 17:57:47 -0800385{
David Rientjes5e9d8342010-08-09 17:18:51 -0700386 task_lock(current);
David Rientjes1b604d72009-12-14 17:57:47 -0800387 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
David Rientjesa9c58b902012-12-11 16:02:54 -0800388 "oom_score_adj=%hd\n",
Davidlohr Bueso01dc52e2012-10-08 16:29:30 -0700389 current->comm, gfp_mask, order,
David Rientjesa63d83f2010-08-09 17:19:46 -0700390 current->signal->oom_score_adj);
David Rientjes1b604d72009-12-14 17:57:47 -0800391 cpuset_print_task_mems_allowed(current);
392 task_unlock(current);
393 dump_stack();
Sha Zhengju58cf1882013-02-22 16:32:05 -0800394 if (memcg)
395 mem_cgroup_print_oom_info(memcg, p);
396 else
397 show_mem(SHOW_MEM_FILTER_NODES);
David Rientjes1b604d72009-12-14 17:57:47 -0800398 if (sysctl_oom_dump_tasks)
Johannes Weiner72835c82012-01-12 17:18:32 -0800399 dump_tasks(memcg, nodemask);
David Rientjes1b604d72009-12-14 17:57:47 -0800400}
401
Michal Hocko5695be12014-10-20 18:12:32 +0200402/*
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800403 * Number of OOM victims in flight
Michal Hocko5695be12014-10-20 18:12:32 +0200404 */
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800405static atomic_t oom_victims = ATOMIC_INIT(0);
406static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
Michal Hocko5695be12014-10-20 18:12:32 +0200407
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800408bool oom_killer_disabled __read_mostly;
Michal Hocko5695be12014-10-20 18:12:32 +0200409
Michal Hocko49550b62015-02-11 15:26:12 -0800410/**
Johannes Weiner16e95192015-06-24 16:57:07 -0700411 * mark_oom_victim - mark the given task as OOM victim
Michal Hocko49550b62015-02-11 15:26:12 -0800412 * @tsk: task to mark
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800413 *
Johannes Weinerdc564012015-06-24 16:57:19 -0700414 * Has to be called with oom_lock held and never after
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800415 * oom has been disabled already.
Michal Hocko49550b62015-02-11 15:26:12 -0800416 */
Johannes Weiner16e95192015-06-24 16:57:07 -0700417void mark_oom_victim(struct task_struct *tsk)
Michal Hocko49550b62015-02-11 15:26:12 -0800418{
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800419 WARN_ON(oom_killer_disabled);
420 /* OOM killer might race with memcg OOM */
421 if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
422 return;
Michal Hocko63a8ca92015-02-11 15:26:15 -0800423 /*
424 * Make sure that the task is woken up from uninterruptible sleep
425 * if it is frozen because OOM killer wouldn't be able to free
426 * any memory and livelock. freezing_slow_path will tell the freezer
427 * that TIF_MEMDIE tasks should be ignored.
428 */
429 __thaw_task(tsk);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800430 atomic_inc(&oom_victims);
Michal Hocko49550b62015-02-11 15:26:12 -0800431}
432
433/**
Johannes Weiner16e95192015-06-24 16:57:07 -0700434 * exit_oom_victim - note the exit of an OOM victim
Michal Hocko49550b62015-02-11 15:26:12 -0800435 */
Johannes Weiner16e95192015-06-24 16:57:07 -0700436void exit_oom_victim(void)
Michal Hocko49550b62015-02-11 15:26:12 -0800437{
Johannes Weiner46402772015-06-24 16:57:10 -0700438 clear_thread_flag(TIF_MEMDIE);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800439
Johannes Weinerc38f1022015-06-24 16:57:13 -0700440 if (!atomic_dec_return(&oom_victims))
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800441 wake_up_all(&oom_victims_wait);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800442}
443
444/**
445 * oom_killer_disable - disable OOM killer
446 *
447 * Forces all page allocations to fail rather than trigger OOM killer.
448 * Will block and wait until all OOM victims are killed.
449 *
450 * The function cannot be called when there are runnable user tasks because
451 * the userspace would see unexpected allocation failures as a result. Any
452 * new usage of this function should be consulted with MM people.
453 *
454 * Returns true if successful and false if the OOM killer cannot be
455 * disabled.
456 */
457bool oom_killer_disable(void)
458{
459 /*
460 * Make sure to not race with an ongoing OOM killer
461 * and that the current is not the victim.
462 */
Johannes Weinerdc564012015-06-24 16:57:19 -0700463 mutex_lock(&oom_lock);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800464 if (test_thread_flag(TIF_MEMDIE)) {
Johannes Weinerdc564012015-06-24 16:57:19 -0700465 mutex_unlock(&oom_lock);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800466 return false;
467 }
468
469 oom_killer_disabled = true;
Johannes Weinerdc564012015-06-24 16:57:19 -0700470 mutex_unlock(&oom_lock);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800471
472 wait_event(oom_victims_wait, !atomic_read(&oom_victims));
473
474 return true;
475}
476
477/**
478 * oom_killer_enable - enable OOM killer
479 */
480void oom_killer_enable(void)
481{
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800482 oom_killer_disabled = false;
Michal Hocko49550b62015-02-11 15:26:12 -0800483}
484
KOSAKI Motohiro3b4798c2009-12-15 16:45:32 -0800485#define K(x) ((x) << (PAGE_SHIFT-10))
David Rientjes6b0c81b2012-07-31 16:43:45 -0700486/*
487 * Must be called while holding a reference to p, which will be released upon
488 * returning.
489 */
David Rientjes9cbb78b2012-07-31 16:43:44 -0700490void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
491 unsigned int points, unsigned long totalpages,
492 struct mem_cgroup *memcg, nodemask_t *nodemask,
493 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494{
Linus Torvalds52d3c032011-03-14 15:17:07 -0700495 struct task_struct *victim = p;
David Rientjes5e9d8342010-08-09 17:18:51 -0700496 struct task_struct *child;
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800497 struct task_struct *t;
David Rientjes647f2bd2012-03-21 16:33:46 -0700498 struct mm_struct *mm;
Linus Torvalds52d3c032011-03-14 15:17:07 -0700499 unsigned int victim_points = 0;
David Rientjesdc3f21e2012-03-21 16:33:47 -0700500 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
501 DEFAULT_RATELIMIT_BURST);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700503 /*
504 * If the task is already exiting, don't alarm the sysadmin or kill
505 * its children or threads, just set TIF_MEMDIE so it can die quickly
506 */
Michal Hocko83363b92015-02-11 15:24:56 -0800507 task_lock(p);
508 if (p->mm && task_will_free_mem(p)) {
Johannes Weiner16e95192015-06-24 16:57:07 -0700509 mark_oom_victim(p);
Michal Hocko83363b92015-02-11 15:24:56 -0800510 task_unlock(p);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700511 put_task_struct(p);
David Rientjes2a1c9b12012-03-21 16:33:46 -0700512 return;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700513 }
Michal Hocko83363b92015-02-11 15:24:56 -0800514 task_unlock(p);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700515
David Rientjesdc3f21e2012-03-21 16:33:47 -0700516 if (__ratelimit(&oom_rs))
David Rientjes8447d952012-03-21 16:33:47 -0700517 dump_header(p, gfp_mask, order, memcg, nodemask);
518
David Rientjes5e9d8342010-08-09 17:18:51 -0700519 task_lock(p);
Wang Longf0d66472015-06-24 16:58:01 -0700520 pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n",
David Rientjes5e9d8342010-08-09 17:18:51 -0700521 message, task_pid_nr(p), p->comm, points);
522 task_unlock(p);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800523
David Rientjes5e9d8342010-08-09 17:18:51 -0700524 /*
525 * If any of p's children has a different mm and is eligible for kill,
David Rientjes11239832011-07-25 17:12:17 -0700526 * the one with the highest oom_badness() score is sacrificed for its
David Rientjes5e9d8342010-08-09 17:18:51 -0700527 * parent. This attempts to lose the minimal amount of work done while
528 * still freeing memory.
529 */
David Rientjes6b0c81b2012-07-31 16:43:45 -0700530 read_lock(&tasklist_lock);
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800531 for_each_thread(p, t) {
David Rientjes5e9d8342010-08-09 17:18:51 -0700532 list_for_each_entry(child, &t->children, sibling) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700533 unsigned int child_points;
David Rientjes5e9d8342010-08-09 17:18:51 -0700534
David Rientjesedd45542011-03-22 16:30:12 -0700535 if (child->mm == p->mm)
536 continue;
David Rientjesa63d83f2010-08-09 17:19:46 -0700537 /*
538 * oom_badness() returns 0 if the thread is unkillable
539 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800540 child_points = oom_badness(child, memcg, nodemask,
David Rientjesa63d83f2010-08-09 17:19:46 -0700541 totalpages);
David Rientjes5e9d8342010-08-09 17:18:51 -0700542 if (child_points > victim_points) {
David Rientjes6b0c81b2012-07-31 16:43:45 -0700543 put_task_struct(victim);
David Rientjes5e9d8342010-08-09 17:18:51 -0700544 victim = child;
545 victim_points = child_points;
David Rientjes6b0c81b2012-07-31 16:43:45 -0700546 get_task_struct(victim);
David Rientjes5e9d8342010-08-09 17:18:51 -0700547 }
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700548 }
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800549 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700550 read_unlock(&tasklist_lock);
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700551
David Rientjes6b0c81b2012-07-31 16:43:45 -0700552 p = find_lock_task_mm(victim);
553 if (!p) {
David Rientjes6b0c81b2012-07-31 16:43:45 -0700554 put_task_struct(victim);
David Rientjes647f2bd2012-03-21 16:33:46 -0700555 return;
David Rientjes6b0c81b2012-07-31 16:43:45 -0700556 } else if (victim != p) {
557 get_task_struct(p);
558 put_task_struct(victim);
559 victim = p;
560 }
David Rientjes647f2bd2012-03-21 16:33:46 -0700561
562 /* mm cannot safely be dereferenced after task_unlock(victim) */
563 mm = victim->mm;
Johannes Weiner16e95192015-06-24 16:57:07 -0700564 mark_oom_victim(victim);
David Rientjes647f2bd2012-03-21 16:33:46 -0700565 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
566 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
567 K(get_mm_counter(victim->mm, MM_ANONPAGES)),
568 K(get_mm_counter(victim->mm, MM_FILEPAGES)));
569 task_unlock(victim);
570
571 /*
572 * Kill all user processes sharing victim->mm in other thread groups, if
573 * any. They don't get access to memory reserves, though, to avoid
574 * depletion of all memory. This prevents mm->mmap_sem livelock when an
575 * oom killed thread cannot exit because it requires the semaphore and
576 * its contended by another thread trying to allocate memory itself.
577 * That thread will now get access to memory reserves since it has a
578 * pending fatal signal.
579 */
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800580 rcu_read_lock();
David Rientjes647f2bd2012-03-21 16:33:46 -0700581 for_each_process(p)
582 if (p->mm == mm && !same_thread_group(p, victim) &&
583 !(p->flags & PF_KTHREAD)) {
584 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
585 continue;
586
587 task_lock(p); /* Protect ->comm from prctl() */
588 pr_err("Kill process %d (%s) sharing same memory\n",
589 task_pid_nr(p), p->comm);
590 task_unlock(p);
Oleg Nesterovd2d39302012-03-23 15:02:45 -0700591 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
David Rientjes647f2bd2012-03-21 16:33:46 -0700592 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700593 rcu_read_unlock();
David Rientjes647f2bd2012-03-21 16:33:46 -0700594
Oleg Nesterovd2d39302012-03-23 15:02:45 -0700595 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700596 put_task_struct(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597}
David Rientjes647f2bd2012-03-21 16:33:46 -0700598#undef K
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
David Rientjes309ed882010-08-09 17:18:54 -0700600/*
601 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
602 */
David Rientjes876aafb2012-07-31 16:43:48 -0700603void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -0700604 int order, const nodemask_t *nodemask,
605 struct mem_cgroup *memcg)
David Rientjes309ed882010-08-09 17:18:54 -0700606{
607 if (likely(!sysctl_panic_on_oom))
608 return;
609 if (sysctl_panic_on_oom != 2) {
610 /*
611 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
612 * does not panic for cpuset, mempolicy, or memcg allocation
613 * failures.
614 */
615 if (constraint != CONSTRAINT_NONE)
616 return;
617 }
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -0700618 dump_header(NULL, gfp_mask, order, memcg, nodemask);
David Rientjes309ed882010-08-09 17:18:54 -0700619 panic("Out of memory: %s panic_on_oom is enabled\n",
620 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
621}
622
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700623static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
624
625int register_oom_notifier(struct notifier_block *nb)
626{
627 return blocking_notifier_chain_register(&oom_notify_list, nb);
628}
629EXPORT_SYMBOL_GPL(register_oom_notifier);
630
631int unregister_oom_notifier(struct notifier_block *nb)
632{
633 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
634}
635EXPORT_SYMBOL_GPL(unregister_oom_notifier);
636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637/**
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800638 * __out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700639 * @zonelist: zonelist pointer
640 * @gfp_mask: memory allocation flags
641 * @order: amount of memory being requested as a power of 2
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700642 * @nodemask: nodemask passed to page allocator
David Rientjes08ab9b12012-03-21 16:34:04 -0700643 * @force_kill: true if a task must be killed, even if others are exiting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 *
645 * If we run out of memory, we have the choice between either
646 * killing a random task (bad), letting the system crash (worse)
647 * OR try to be smart about which process to kill. Note that we
648 * don't have to be perfect here, we just have to be good.
649 */
Johannes Weinerdc564012015-06-24 16:57:19 -0700650bool out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
651 int order, nodemask_t *nodemask, bool force_kill)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652{
David Rientjese85bfd32010-09-22 13:05:10 -0700653 const nodemask_t *mpol_mask;
David Rientjes0aad4b32010-08-09 17:18:59 -0700654 struct task_struct *p;
David Rientjesa63d83f2010-08-09 17:19:46 -0700655 unsigned long totalpages;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700656 unsigned long freed = 0;
David Rientjes9cbb78b2012-07-31 16:43:44 -0700657 unsigned int uninitialized_var(points);
David Rientjese3658932010-08-09 17:18:55 -0700658 enum oom_constraint constraint = CONSTRAINT_NONE;
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700659 int killed = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700660
Johannes Weinerdc564012015-06-24 16:57:19 -0700661 if (oom_killer_disabled)
662 return false;
663
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700664 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
665 if (freed > 0)
666 /* Got some memory back in the last second. */
Johannes Weinerdc564012015-06-24 16:57:19 -0700667 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
David Rientjes7b98c2e2010-08-09 17:18:48 -0700669 /*
David Rientjes9ff48682012-12-11 16:01:30 -0800670 * If current has a pending SIGKILL or is exiting, then automatically
671 * select it. The goal is to allow it to allocate so that it may
672 * quickly exit and free its memory.
Tetsuo Handad7a94e72015-02-11 15:24:54 -0800673 *
674 * But don't select if current has already released its mm and cleared
675 * TIF_MEMDIE flag at exit_mm(), otherwise an OOM livelock may occur.
David Rientjes7b98c2e2010-08-09 17:18:48 -0700676 */
Tetsuo Handad7a94e72015-02-11 15:24:54 -0800677 if (current->mm &&
678 (fatal_signal_pending(current) || task_will_free_mem(current))) {
Johannes Weiner16e95192015-06-24 16:57:07 -0700679 mark_oom_victim(current);
Johannes Weinerdc564012015-06-24 16:57:19 -0700680 goto out;
David Rientjes7b98c2e2010-08-09 17:18:48 -0700681 }
682
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800683 /*
684 * Check if there were limitations on the allocation (only relevant for
685 * NUMA) that may require different handling.
686 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700687 constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
688 &totalpages);
David Rientjese85bfd32010-09-22 13:05:10 -0700689 mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -0700690 check_panic_on_oom(constraint, gfp_mask, order, mpol_mask, NULL);
David Rientjes0aad4b32010-08-09 17:18:59 -0700691
David Rientjes121d1ba2012-07-31 16:42:55 -0700692 if (sysctl_oom_kill_allocating_task && current->mm &&
KOSAKI Motohiroa96cfd62010-08-09 17:19:38 -0700693 !oom_unkillable_task(current, NULL, nodemask) &&
David Rientjes121d1ba2012-07-31 16:42:55 -0700694 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
David Rientjes6b0c81b2012-07-31 16:43:45 -0700695 get_task_struct(current);
David Rientjes2a1c9b12012-03-21 16:33:46 -0700696 oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL,
697 nodemask,
698 "Out of memory (oom_kill_allocating_task)");
699 goto out;
David Rientjes0aad4b32010-08-09 17:18:59 -0700700 }
701
David Rientjes9cbb78b2012-07-31 16:43:44 -0700702 p = select_bad_process(&points, totalpages, mpol_mask, force_kill);
David Rientjes0aad4b32010-08-09 17:18:59 -0700703 /* Found nothing?!?! Either we hang forever, or we panic. */
704 if (!p) {
David Rientjese85bfd32010-09-22 13:05:10 -0700705 dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700706 panic("Out of memory and no killable processes...\n");
707 }
Rusty Russell6b4f2b52013-07-15 11:24:08 +0930708 if (p != (void *)-1UL) {
David Rientjes2a1c9b12012-03-21 16:33:46 -0700709 oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
710 nodemask, "Out of memory");
711 killed = 1;
712 }
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700713out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 /*
David Rientjes4f774b92012-07-31 16:42:37 -0700715 * Give the killed threads a good chance of exiting before trying to
716 * allocate memory again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 */
David Rientjes4f774b92012-07-31 16:42:37 -0700718 if (killed)
719 schedule_timeout_killable(1);
David Rientjese3658932010-08-09 17:18:55 -0700720
Johannes Weinerdc564012015-06-24 16:57:19 -0700721 return true;
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800722}
723
David Rientjese3658932010-08-09 17:18:55 -0700724/*
725 * The pagefault handler calls here because it is out of memory, so kill a
David Rientjesefacd022012-12-12 13:52:06 -0800726 * memory-hogging task. If any populated zone has ZONE_OOM_LOCKED set, a
727 * parallel oom killing is already in progress so do nothing.
David Rientjese3658932010-08-09 17:18:55 -0700728 */
729void pagefault_out_of_memory(void)
730{
Johannes Weiner49426422013-10-16 13:46:59 -0700731 if (mem_cgroup_oom_synchronize(true))
Johannes Weinerdc564012015-06-24 16:57:19 -0700732 return;
Johannes Weiner3812c8c2013-09-12 15:13:44 -0700733
Johannes Weinerdc564012015-06-24 16:57:19 -0700734 if (!mutex_trylock(&oom_lock))
735 return;
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800736
Johannes Weinerdc564012015-06-24 16:57:19 -0700737 if (!out_of_memory(NULL, 0, 0, NULL, false)) {
738 /*
739 * There shouldn't be any user tasks runnable while the
740 * OOM killer is disabled, so the current task has to
741 * be a racing OOM victim for which oom_killer_disable()
742 * is waiting for.
743 */
744 WARN_ON(test_thread_flag(TIF_MEMDIE));
David Rientjese3658932010-08-09 17:18:55 -0700745 }
Johannes Weinerdc564012015-06-24 16:57:19 -0700746
747 mutex_unlock(&oom_lock);
David Rientjese3658932010-08-09 17:18:55 -0700748}