blob: 642f38cb175aa8321ab25d643d449d6cdd94f977 [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;
David Rientjesc7d4cae2009-01-06 14:39:00 -080045static DEFINE_SPINLOCK(zone_scan_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
David Rientjes6f48d0eb2010-08-09 17:18:52 -070047#ifdef CONFIG_NUMA
48/**
49 * has_intersects_mems_allowed() - check task eligiblity for kill
Oleg Nesterovad962442014-01-21 15:50:00 -080050 * @start: task struct of which task to consider
David Rientjes6f48d0eb2010-08-09 17:18:52 -070051 * @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.
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070056 */
Oleg Nesterovad962442014-01-21 15:50:00 -080057static bool has_intersects_mems_allowed(struct task_struct *start,
David Rientjes6f48d0eb2010-08-09 17:18:52 -070058 const nodemask_t *mask)
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070059{
Oleg Nesterovad962442014-01-21 15:50:00 -080060 struct task_struct *tsk;
61 bool ret = false;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070062
Oleg Nesterovad962442014-01-21 15:50:00 -080063 rcu_read_lock();
Oleg Nesterov1da4db02014-01-21 15:49:58 -080064 for_each_thread(start, tsk) {
David Rientjes6f48d0eb2010-08-09 17:18:52 -070065 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 */
Oleg Nesterovad962442014-01-21 15:50:00 -080072 ret = mempolicy_nodemask_intersects(tsk, mask);
David Rientjes6f48d0eb2010-08-09 17:18:52 -070073 } else {
74 /*
75 * This is not a mempolicy constrained oom, so only
76 * check the mems of tsk's cpuset.
77 */
Oleg Nesterovad962442014-01-21 15:50:00 -080078 ret = cpuset_mems_allowed_intersects(current, tsk);
David Rientjes6f48d0eb2010-08-09 17:18:52 -070079 }
Oleg Nesterovad962442014-01-21 15:50:00 -080080 if (ret)
81 break;
Oleg Nesterov1da4db02014-01-21 15:49:58 -080082 }
Oleg Nesterovad962442014-01-21 15:50:00 -080083 rcu_read_unlock();
KOSAKI Motohirodf1090a2010-08-09 17:19:39 -070084
Oleg Nesterovad962442014-01-21 15:50:00 -080085 return ret;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070086}
David Rientjes6f48d0eb2010-08-09 17:18:52 -070087#else
88static bool has_intersects_mems_allowed(struct task_struct *tsk,
89 const nodemask_t *mask)
90{
91 return true;
92}
93#endif /* CONFIG_NUMA */
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070094
David Rientjes6f48d0eb2010-08-09 17:18:52 -070095/*
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 */
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -0700101struct task_struct *find_lock_task_mm(struct task_struct *p)
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700102{
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800103 struct task_struct *t;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700104
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800105 rcu_read_lock();
106
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800107 for_each_thread(p, t) {
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700108 task_lock(t);
109 if (likely(t->mm))
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800110 goto found;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700111 task_unlock(t);
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800112 }
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800113 t = NULL;
114found:
115 rcu_read_unlock();
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700116
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800117 return t;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700118}
119
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700120/* return true if the task is not adequate as candidate victim task. */
David Rientjese85bfd32010-09-22 13:05:10 -0700121static bool oom_unkillable_task(struct task_struct *p,
Johannes Weiner2314b422014-12-10 15:44:33 -0800122 struct mem_cgroup *memcg, const nodemask_t *nodemask)
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700123{
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 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800130 if (memcg && !task_in_mem_cgroup(p, memcg))
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700131 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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140/**
David Rientjesa63d83f2010-08-09 17:19:46 -0700141 * oom_badness - heuristic function to determine which candidate task to kill
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 * @p: task struct of which task we should calculate
David Rientjesa63d83f2010-08-09 17:19:46 -0700143 * @totalpages: total present RAM allowed for page allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144 *
David Rientjesa63d83f2010-08-09 17:19:46 -0700145 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 */
David Rientjesa7f638f2012-05-29 15:06:47 -0700149unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
150 const nodemask_t *nodemask, unsigned long totalpages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151{
David Rientjes1e11ad82012-06-08 13:21:26 -0700152 long points;
David Rientjes61eafb02012-06-20 12:52:58 -0700153 long adj;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700154
Johannes Weiner72835c82012-01-12 17:18:32 -0800155 if (oom_unkillable_task(p, memcg, nodemask))
KOSAKI Motohiro26ebc982010-08-09 17:19:37 -0700156 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700158 p = find_lock_task_mm(p);
159 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 return 0;
161
David Rientjesa9c58b902012-12-11 16:02:54 -0800162 adj = (long)p->signal->oom_score_adj;
David Rientjes61eafb02012-06-20 12:52:58 -0700163 if (adj == OOM_SCORE_ADJ_MIN) {
Michal Hocko5aecc852011-11-15 14:36:07 -0800164 task_unlock(p);
165 return 0;
166 }
167
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700169 * The baseline for the badness score is the proportion of RAM that each
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700170 * task's rss, pagetable and swap space use.
David Rientjesa63d83f2010-08-09 17:19:46 -0700171 */
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800172 points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
173 atomic_long_read(&p->mm->nr_ptes) + mm_nr_pmds(p->mm);
Andrew Morton97c2c9b82006-04-18 22:20:38 -0700174 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
176 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700177 * Root processes get 3% bonus, just like the __vm_enough_memory()
178 * implementation used by LSMs.
Hugh Dickins7ba34852007-01-05 16:37:03 -0800179 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700180 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
David Rientjes778c14a2014-01-30 15:46:11 -0800181 points -= (points * 3) / 100;
Hugh Dickins7ba34852007-01-05 16:37:03 -0800182
David Rientjes61eafb02012-06-20 12:52:58 -0700183 /* Normalize to oom_score_adj units */
184 adj *= totalpages / 1000;
185 points += adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
David Rientjesf19e8aa2010-09-22 13:04:52 -0700187 /*
David Rientjesa7f638f2012-05-29 15:06:47 -0700188 * 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).
David Rientjesf19e8aa2010-09-22 13:04:52 -0700190 */
David Rientjes1e11ad82012-06-08 13:21:26 -0700191 return points > 0 ? points : 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192}
193
194/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800195 * Determine the type of allocation constraint.
196 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800197#ifdef CONFIG_NUMA
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800198static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700199 gfp_t gfp_mask, nodemask_t *nodemask,
200 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800201{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700202 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700203 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700204 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
David Rientjesa63d83f2010-08-09 17:19:46 -0700205 bool cpuset_limited = false;
206 int nid;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800207
David Rientjesa63d83f2010-08-09 17:19:46 -0700208 /* Default to all available memory */
209 *totalpages = totalram_pages + total_swap_pages;
210
211 if (!zonelist)
212 return CONSTRAINT_NONE;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800213 /*
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;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800220
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800221 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700222 * 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().
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800225 */
Lai Jiangshanbd3a66c2012-12-12 13:51:28 -0800226 if (nodemask && !nodes_subset(node_states[N_MEMORY], *nodemask)) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700227 *totalpages = total_swap_pages;
228 for_each_node_mask(nid, *nodemask)
229 *totalpages += node_spanned_pages(nid);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800230 return CONSTRAINT_MEMORY_POLICY;
David Rientjesa63d83f2010-08-09 17:19:46 -0700231 }
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800232
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)
Vladimir Davydov344736f2014-10-20 15:50:30 +0400236 if (!cpuset_zone_allowed(zone, gfp_mask))
David Rientjesa63d83f2010-08-09 17:19:46 -0700237 cpuset_limited = true;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800238
David Rientjesa63d83f2010-08-09 17:19:46 -0700239 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 }
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800245 return CONSTRAINT_NONE;
246}
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800247#else
248static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700249 gfp_t gfp_mask, nodemask_t *nodemask,
250 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800251{
David Rientjesa63d83f2010-08-09 17:19:46 -0700252 *totalpages = totalram_pages + total_swap_pages;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800253 return CONSTRAINT_NONE;
254}
255#endif
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800256
David Rientjes9cbb78b2012-07-31 16:43:44 -0700257enum oom_scan_t oom_scan_process_thread(struct task_struct *task,
258 unsigned long totalpages, const nodemask_t *nodemask,
259 bool force_kill)
David Rientjes462607e2012-07-31 16:43:40 -0700260{
David Rientjes9cbb78b2012-07-31 16:43:44 -0700261 if (oom_unkillable_task(task, NULL, nodemask))
David Rientjes462607e2012-07-31 16:43:40 -0700262 return OOM_SCAN_CONTINUE;
263
264 /*
265 * This task already has access to memory reserves and is being killed.
266 * Don't allow any other task to have access to the reserves.
267 */
268 if (test_tsk_thread_flag(task, TIF_MEMDIE)) {
David Rientjes462607e2012-07-31 16:43:40 -0700269 if (!force_kill)
270 return OOM_SCAN_ABORT;
271 }
272 if (!task->mm)
273 return OOM_SCAN_CONTINUE;
274
David Rientjese1e12d22012-12-11 16:02:56 -0800275 /*
276 * If task is allocating a lot of memory and has been marked to be
277 * killed first if it triggers an oom, then select it.
278 */
279 if (oom_task_origin(task))
280 return OOM_SCAN_SELECT;
281
Oleg Nesterov6a2d5672014-12-12 16:56:27 -0800282 if (task_will_free_mem(task) && !force_kill)
283 return OOM_SCAN_ABORT;
284
David Rientjes462607e2012-07-31 16:43:40 -0700285 return OOM_SCAN_OK;
286}
287
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800288/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 * Simple selection loop. We chose the process with the highest
Rusty Russell6b4f2b52013-07-15 11:24:08 +0930290 * number of 'points'. Returns -1 on scan abort.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 *
292 * (not docbooked, we don't want this one cluttering up the manual)
293 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700294static struct task_struct *select_bad_process(unsigned int *ppoints,
David Rientjes9cbb78b2012-07-31 16:43:44 -0700295 unsigned long totalpages, const nodemask_t *nodemask,
296 bool force_kill)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297{
David Rientjes3a5dda72011-03-22 16:30:09 -0700298 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 struct task_struct *chosen = NULL;
David Rientjesa7f638f2012-05-29 15:06:47 -0700300 unsigned long chosen_points = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700301
David Rientjes6b0c81b2012-07-31 16:43:45 -0700302 rcu_read_lock();
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800303 for_each_process_thread(g, p) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700304 unsigned int points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700305
David Rientjes9cbb78b2012-07-31 16:43:44 -0700306 switch (oom_scan_process_thread(p, totalpages, nodemask,
David Rientjes462607e2012-07-31 16:43:40 -0700307 force_kill)) {
308 case OOM_SCAN_SELECT:
309 chosen = p;
310 chosen_points = ULONG_MAX;
311 /* fall through */
312 case OOM_SCAN_CONTINUE:
Andrey Vagin30e2b412011-03-22 16:30:11 -0700313 continue;
David Rientjes462607e2012-07-31 16:43:40 -0700314 case OOM_SCAN_ABORT:
David Rientjes6b0c81b2012-07-31 16:43:45 -0700315 rcu_read_unlock();
Rusty Russell6b4f2b52013-07-15 11:24:08 +0930316 return (struct task_struct *)(-1UL);
David Rientjes462607e2012-07-31 16:43:40 -0700317 case OOM_SCAN_OK:
318 break;
319 };
David Rientjes9cbb78b2012-07-31 16:43:44 -0700320 points = oom_badness(p, NULL, nodemask, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -0800321 if (!points || points < chosen_points)
322 continue;
323 /* Prefer thread group leaders for display purposes */
324 if (points == chosen_points && thread_group_leader(chosen))
325 continue;
326
327 chosen = p;
328 chosen_points = points;
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800329 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700330 if (chosen)
331 get_task_struct(chosen);
332 rcu_read_unlock();
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700333
David Rientjesa7f638f2012-05-29 15:06:47 -0700334 *ppoints = chosen_points * 1000 / totalpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335 return chosen;
336}
337
338/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700339 * dump_tasks - dump current memory state of all system tasks
Wanpeng Lidad75572012-06-20 12:53:01 -0700340 * @memcg: current's memory controller, if constrained
David Rientjese85bfd32010-09-22 13:05:10 -0700341 * @nodemask: nodemask passed to page allocator for mempolicy ooms
Randy Dunlap1b578df2008-03-19 17:00:42 -0700342 *
David Rientjese85bfd32010-09-22 13:05:10 -0700343 * Dumps the current memory state of all eligible tasks. Tasks not in the same
344 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
345 * are not shown.
David Rientjesde34d962012-07-31 16:42:56 -0700346 * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
347 * swapents, oom_score_adj value, and name.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800348 */
Johannes Weiner2314b422014-12-10 15:44:33 -0800349static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask)
David Rientjesfef1bdd2008-02-07 00:14:07 -0800350{
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700351 struct task_struct *p;
352 struct task_struct *task;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800353
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800354 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 -0700355 rcu_read_lock();
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700356 for_each_process(p) {
Johannes Weiner72835c82012-01-12 17:18:32 -0800357 if (oom_unkillable_task(p, memcg, nodemask))
David Rientjesfef1bdd2008-02-07 00:14:07 -0800358 continue;
359
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700360 task = find_lock_task_mm(p);
361 if (!task) {
David Rientjes6d2661e2009-05-28 14:34:19 -0700362 /*
David Rientjes74ab7f12010-08-09 17:18:46 -0700363 * This is a kthread or all of p's threads have already
364 * detached their mm's. There's no need to report
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700365 * them; they can't be oom killed anyway.
David Rientjes6d2661e2009-05-28 14:34:19 -0700366 */
David Rientjes6d2661e2009-05-28 14:34:19 -0700367 continue;
368 }
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700369
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800370 pr_info("[%5d] %5d %5d %8lu %8lu %7ld %7ld %8lu %5hd %s\n",
Eric W. Biederman078de5f2012-02-08 07:00:08 -0800371 task->pid, from_kuid(&init_user_ns, task_uid(task)),
372 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -0800373 atomic_long_read(&task->mm->nr_ptes),
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800374 mm_nr_pmds(task->mm),
David Rientjesde34d962012-07-31 16:42:56 -0700375 get_mm_counter(task->mm, MM_SWAPENTS),
David Rientjesa63d83f2010-08-09 17:19:46 -0700376 task->signal->oom_score_adj, task->comm);
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700377 task_unlock(task);
378 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700379 rcu_read_unlock();
David Rientjesfef1bdd2008-02-07 00:14:07 -0800380}
381
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -0800382static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
Johannes Weiner72835c82012-01-12 17:18:32 -0800383 struct mem_cgroup *memcg, const nodemask_t *nodemask)
David Rientjes1b604d72009-12-14 17:57:47 -0800384{
David Rientjes5e9d8342010-08-09 17:18:51 -0700385 task_lock(current);
David Rientjes1b604d72009-12-14 17:57:47 -0800386 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
David Rientjesa9c58b902012-12-11 16:02:54 -0800387 "oom_score_adj=%hd\n",
Davidlohr Bueso01dc52e2012-10-08 16:29:30 -0700388 current->comm, gfp_mask, order,
David Rientjesa63d83f2010-08-09 17:19:46 -0700389 current->signal->oom_score_adj);
David Rientjes1b604d72009-12-14 17:57:47 -0800390 cpuset_print_task_mems_allowed(current);
391 task_unlock(current);
392 dump_stack();
Sha Zhengju58cf1882013-02-22 16:32:05 -0800393 if (memcg)
394 mem_cgroup_print_oom_info(memcg, p);
395 else
396 show_mem(SHOW_MEM_FILTER_NODES);
David Rientjes1b604d72009-12-14 17:57:47 -0800397 if (sysctl_oom_dump_tasks)
Johannes Weiner72835c82012-01-12 17:18:32 -0800398 dump_tasks(memcg, nodemask);
David Rientjes1b604d72009-12-14 17:57:47 -0800399}
400
Michal Hocko5695be12014-10-20 18:12:32 +0200401/*
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800402 * Number of OOM victims in flight
Michal Hocko5695be12014-10-20 18:12:32 +0200403 */
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800404static atomic_t oom_victims = ATOMIC_INIT(0);
405static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
Michal Hocko5695be12014-10-20 18:12:32 +0200406
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800407bool oom_killer_disabled __read_mostly;
408static DECLARE_RWSEM(oom_sem);
Michal Hocko5695be12014-10-20 18:12:32 +0200409
Michal Hocko49550b62015-02-11 15:26:12 -0800410/**
411 * mark_tsk_oom_victim - marks the given taks as OOM victim.
412 * @tsk: task to mark
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800413 *
414 * Has to be called with oom_sem taken for read and never after
415 * oom has been disabled already.
Michal Hocko49550b62015-02-11 15:26:12 -0800416 */
417void mark_tsk_oom_victim(struct task_struct *tsk)
418{
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/**
434 * unmark_oom_victim - unmarks the current task as OOM victim.
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800435 *
436 * Wakes up all waiters in oom_killer_disable()
Michal Hocko49550b62015-02-11 15:26:12 -0800437 */
438void unmark_oom_victim(void)
439{
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800440 if (!test_and_clear_thread_flag(TIF_MEMDIE))
441 return;
442
443 down_read(&oom_sem);
444 /*
445 * There is no need to signal the lasst oom_victim if there
446 * is nobody who cares.
447 */
448 if (!atomic_dec_return(&oom_victims) && oom_killer_disabled)
449 wake_up_all(&oom_victims_wait);
450 up_read(&oom_sem);
451}
452
453/**
454 * oom_killer_disable - disable OOM killer
455 *
456 * Forces all page allocations to fail rather than trigger OOM killer.
457 * Will block and wait until all OOM victims are killed.
458 *
459 * The function cannot be called when there are runnable user tasks because
460 * the userspace would see unexpected allocation failures as a result. Any
461 * new usage of this function should be consulted with MM people.
462 *
463 * Returns true if successful and false if the OOM killer cannot be
464 * disabled.
465 */
466bool oom_killer_disable(void)
467{
468 /*
469 * Make sure to not race with an ongoing OOM killer
470 * and that the current is not the victim.
471 */
472 down_write(&oom_sem);
473 if (test_thread_flag(TIF_MEMDIE)) {
474 up_write(&oom_sem);
475 return false;
476 }
477
478 oom_killer_disabled = true;
479 up_write(&oom_sem);
480
481 wait_event(oom_victims_wait, !atomic_read(&oom_victims));
482
483 return true;
484}
485
486/**
487 * oom_killer_enable - enable OOM killer
488 */
489void oom_killer_enable(void)
490{
491 down_write(&oom_sem);
492 oom_killer_disabled = false;
493 up_write(&oom_sem);
Michal Hocko49550b62015-02-11 15:26:12 -0800494}
495
KOSAKI Motohiro3b4798c2009-12-15 16:45:32 -0800496#define K(x) ((x) << (PAGE_SHIFT-10))
David Rientjes6b0c81b2012-07-31 16:43:45 -0700497/*
498 * Must be called while holding a reference to p, which will be released upon
499 * returning.
500 */
David Rientjes9cbb78b2012-07-31 16:43:44 -0700501void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
502 unsigned int points, unsigned long totalpages,
503 struct mem_cgroup *memcg, nodemask_t *nodemask,
504 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505{
Linus Torvalds52d3c032011-03-14 15:17:07 -0700506 struct task_struct *victim = p;
David Rientjes5e9d8342010-08-09 17:18:51 -0700507 struct task_struct *child;
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800508 struct task_struct *t;
David Rientjes647f2bd2012-03-21 16:33:46 -0700509 struct mm_struct *mm;
Linus Torvalds52d3c032011-03-14 15:17:07 -0700510 unsigned int victim_points = 0;
David Rientjesdc3f21e2012-03-21 16:33:47 -0700511 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
512 DEFAULT_RATELIMIT_BURST);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700514 /*
515 * If the task is already exiting, don't alarm the sysadmin or kill
516 * its children or threads, just set TIF_MEMDIE so it can die quickly
517 */
Michal Hocko83363b92015-02-11 15:24:56 -0800518 task_lock(p);
519 if (p->mm && task_will_free_mem(p)) {
Michal Hocko49550b62015-02-11 15:26:12 -0800520 mark_tsk_oom_victim(p);
Michal Hocko83363b92015-02-11 15:24:56 -0800521 task_unlock(p);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700522 put_task_struct(p);
David Rientjes2a1c9b12012-03-21 16:33:46 -0700523 return;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700524 }
Michal Hocko83363b92015-02-11 15:24:56 -0800525 task_unlock(p);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700526
David Rientjesdc3f21e2012-03-21 16:33:47 -0700527 if (__ratelimit(&oom_rs))
David Rientjes8447d952012-03-21 16:33:47 -0700528 dump_header(p, gfp_mask, order, memcg, nodemask);
529
David Rientjes5e9d8342010-08-09 17:18:51 -0700530 task_lock(p);
David Rientjesa63d83f2010-08-09 17:19:46 -0700531 pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
David Rientjes5e9d8342010-08-09 17:18:51 -0700532 message, task_pid_nr(p), p->comm, points);
533 task_unlock(p);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800534
David Rientjes5e9d8342010-08-09 17:18:51 -0700535 /*
536 * If any of p's children has a different mm and is eligible for kill,
David Rientjes11239832011-07-25 17:12:17 -0700537 * the one with the highest oom_badness() score is sacrificed for its
David Rientjes5e9d8342010-08-09 17:18:51 -0700538 * parent. This attempts to lose the minimal amount of work done while
539 * still freeing memory.
540 */
David Rientjes6b0c81b2012-07-31 16:43:45 -0700541 read_lock(&tasklist_lock);
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800542 for_each_thread(p, t) {
David Rientjes5e9d8342010-08-09 17:18:51 -0700543 list_for_each_entry(child, &t->children, sibling) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700544 unsigned int child_points;
David Rientjes5e9d8342010-08-09 17:18:51 -0700545
David Rientjesedd45542011-03-22 16:30:12 -0700546 if (child->mm == p->mm)
547 continue;
David Rientjesa63d83f2010-08-09 17:19:46 -0700548 /*
549 * oom_badness() returns 0 if the thread is unkillable
550 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800551 child_points = oom_badness(child, memcg, nodemask,
David Rientjesa63d83f2010-08-09 17:19:46 -0700552 totalpages);
David Rientjes5e9d8342010-08-09 17:18:51 -0700553 if (child_points > victim_points) {
David Rientjes6b0c81b2012-07-31 16:43:45 -0700554 put_task_struct(victim);
David Rientjes5e9d8342010-08-09 17:18:51 -0700555 victim = child;
556 victim_points = child_points;
David Rientjes6b0c81b2012-07-31 16:43:45 -0700557 get_task_struct(victim);
David Rientjes5e9d8342010-08-09 17:18:51 -0700558 }
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700559 }
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800560 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700561 read_unlock(&tasklist_lock);
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700562
David Rientjes6b0c81b2012-07-31 16:43:45 -0700563 p = find_lock_task_mm(victim);
564 if (!p) {
David Rientjes6b0c81b2012-07-31 16:43:45 -0700565 put_task_struct(victim);
David Rientjes647f2bd2012-03-21 16:33:46 -0700566 return;
David Rientjes6b0c81b2012-07-31 16:43:45 -0700567 } else if (victim != p) {
568 get_task_struct(p);
569 put_task_struct(victim);
570 victim = p;
571 }
David Rientjes647f2bd2012-03-21 16:33:46 -0700572
573 /* mm cannot safely be dereferenced after task_unlock(victim) */
574 mm = victim->mm;
Michal Hocko49550b62015-02-11 15:26:12 -0800575 mark_tsk_oom_victim(victim);
David Rientjes647f2bd2012-03-21 16:33:46 -0700576 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
577 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
578 K(get_mm_counter(victim->mm, MM_ANONPAGES)),
579 K(get_mm_counter(victim->mm, MM_FILEPAGES)));
580 task_unlock(victim);
581
582 /*
583 * Kill all user processes sharing victim->mm in other thread groups, if
584 * any. They don't get access to memory reserves, though, to avoid
585 * depletion of all memory. This prevents mm->mmap_sem livelock when an
586 * oom killed thread cannot exit because it requires the semaphore and
587 * its contended by another thread trying to allocate memory itself.
588 * That thread will now get access to memory reserves since it has a
589 * pending fatal signal.
590 */
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800591 rcu_read_lock();
David Rientjes647f2bd2012-03-21 16:33:46 -0700592 for_each_process(p)
593 if (p->mm == mm && !same_thread_group(p, victim) &&
594 !(p->flags & PF_KTHREAD)) {
595 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
596 continue;
597
598 task_lock(p); /* Protect ->comm from prctl() */
599 pr_err("Kill process %d (%s) sharing same memory\n",
600 task_pid_nr(p), p->comm);
601 task_unlock(p);
Oleg Nesterovd2d39302012-03-23 15:02:45 -0700602 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
David Rientjes647f2bd2012-03-21 16:33:46 -0700603 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700604 rcu_read_unlock();
David Rientjes647f2bd2012-03-21 16:33:46 -0700605
Oleg Nesterovd2d39302012-03-23 15:02:45 -0700606 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700607 put_task_struct(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608}
David Rientjes647f2bd2012-03-21 16:33:46 -0700609#undef K
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610
David Rientjes309ed882010-08-09 17:18:54 -0700611/*
612 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
613 */
David Rientjes876aafb2012-07-31 16:43:48 -0700614void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
615 int order, const nodemask_t *nodemask)
David Rientjes309ed882010-08-09 17:18:54 -0700616{
617 if (likely(!sysctl_panic_on_oom))
618 return;
619 if (sysctl_panic_on_oom != 2) {
620 /*
621 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
622 * does not panic for cpuset, mempolicy, or memcg allocation
623 * failures.
624 */
625 if (constraint != CONSTRAINT_NONE)
626 return;
627 }
David Rientjese85bfd32010-09-22 13:05:10 -0700628 dump_header(NULL, gfp_mask, order, NULL, nodemask);
David Rientjes309ed882010-08-09 17:18:54 -0700629 panic("Out of memory: %s panic_on_oom is enabled\n",
630 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
631}
632
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700633static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
634
635int register_oom_notifier(struct notifier_block *nb)
636{
637 return blocking_notifier_chain_register(&oom_notify_list, nb);
638}
639EXPORT_SYMBOL_GPL(register_oom_notifier);
640
641int unregister_oom_notifier(struct notifier_block *nb)
642{
643 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
644}
645EXPORT_SYMBOL_GPL(unregister_oom_notifier);
646
David Rientjes098d7f12007-10-16 23:25:55 -0700647/*
648 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
649 * if a parallel OOM killing is already taking place that includes a zone in
650 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
651 */
David Rientjese972a072014-08-06 16:07:52 -0700652bool oom_zonelist_trylock(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700653{
Mel Gormandd1a2392008-04-28 02:12:17 -0700654 struct zoneref *z;
655 struct zone *zone;
David Rientjese972a072014-08-06 16:07:52 -0700656 bool ret = true;
David Rientjes098d7f12007-10-16 23:25:55 -0700657
David Rientjesc7d4cae2009-01-06 14:39:00 -0800658 spin_lock(&zone_scan_lock);
David Rientjese972a072014-08-06 16:07:52 -0700659 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask))
Johannes Weiner57054652014-10-09 15:28:17 -0700660 if (test_bit(ZONE_OOM_LOCKED, &zone->flags)) {
David Rientjese972a072014-08-06 16:07:52 -0700661 ret = false;
David Rientjes098d7f12007-10-16 23:25:55 -0700662 goto out;
663 }
David Rientjes098d7f12007-10-16 23:25:55 -0700664
David Rientjese972a072014-08-06 16:07:52 -0700665 /*
666 * Lock each zone in the zonelist under zone_scan_lock so a parallel
667 * call to oom_zonelist_trylock() doesn't succeed when it shouldn't.
668 */
669 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask))
Johannes Weiner57054652014-10-09 15:28:17 -0700670 set_bit(ZONE_OOM_LOCKED, &zone->flags);
Mel Gormandd1a2392008-04-28 02:12:17 -0700671
David Rientjes098d7f12007-10-16 23:25:55 -0700672out:
David Rientjesc7d4cae2009-01-06 14:39:00 -0800673 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700674 return ret;
675}
676
677/*
678 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
679 * allocation attempts with zonelists containing them may now recall the OOM
680 * killer, if necessary.
681 */
David Rientjese972a072014-08-06 16:07:52 -0700682void oom_zonelist_unlock(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700683{
Mel Gormandd1a2392008-04-28 02:12:17 -0700684 struct zoneref *z;
685 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700686
David Rientjesc7d4cae2009-01-06 14:39:00 -0800687 spin_lock(&zone_scan_lock);
David Rientjese972a072014-08-06 16:07:52 -0700688 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask))
Johannes Weiner57054652014-10-09 15:28:17 -0700689 clear_bit(ZONE_OOM_LOCKED, &zone->flags);
David Rientjesc7d4cae2009-01-06 14:39:00 -0800690 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700691}
692
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693/**
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800694 * __out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700695 * @zonelist: zonelist pointer
696 * @gfp_mask: memory allocation flags
697 * @order: amount of memory being requested as a power of 2
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700698 * @nodemask: nodemask passed to page allocator
David Rientjes08ab9b12012-03-21 16:34:04 -0700699 * @force_kill: true if a task must be killed, even if others are exiting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 *
701 * If we run out of memory, we have the choice between either
702 * killing a random task (bad), letting the system crash (worse)
703 * OR try to be smart about which process to kill. Note that we
704 * don't have to be perfect here, we just have to be good.
705 */
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800706static void __out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
David Rientjes08ab9b12012-03-21 16:34:04 -0700707 int order, nodemask_t *nodemask, bool force_kill)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700708{
David Rientjese85bfd32010-09-22 13:05:10 -0700709 const nodemask_t *mpol_mask;
David Rientjes0aad4b32010-08-09 17:18:59 -0700710 struct task_struct *p;
David Rientjesa63d83f2010-08-09 17:19:46 -0700711 unsigned long totalpages;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700712 unsigned long freed = 0;
David Rientjes9cbb78b2012-07-31 16:43:44 -0700713 unsigned int uninitialized_var(points);
David Rientjese3658932010-08-09 17:18:55 -0700714 enum oom_constraint constraint = CONSTRAINT_NONE;
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700715 int killed = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700716
717 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
718 if (freed > 0)
719 /* Got some memory back in the last second. */
720 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
David Rientjes7b98c2e2010-08-09 17:18:48 -0700722 /*
David Rientjes9ff48682012-12-11 16:01:30 -0800723 * If current has a pending SIGKILL or is exiting, then automatically
724 * select it. The goal is to allow it to allocate so that it may
725 * quickly exit and free its memory.
Tetsuo Handad7a94e72015-02-11 15:24:54 -0800726 *
727 * But don't select if current has already released its mm and cleared
728 * TIF_MEMDIE flag at exit_mm(), otherwise an OOM livelock may occur.
David Rientjes7b98c2e2010-08-09 17:18:48 -0700729 */
Tetsuo Handad7a94e72015-02-11 15:24:54 -0800730 if (current->mm &&
731 (fatal_signal_pending(current) || task_will_free_mem(current))) {
Michal Hocko49550b62015-02-11 15:26:12 -0800732 mark_tsk_oom_victim(current);
David Rientjes7b98c2e2010-08-09 17:18:48 -0700733 return;
734 }
735
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800736 /*
737 * Check if there were limitations on the allocation (only relevant for
738 * NUMA) that may require different handling.
739 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700740 constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
741 &totalpages);
David Rientjese85bfd32010-09-22 13:05:10 -0700742 mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
743 check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700744
David Rientjes121d1ba2012-07-31 16:42:55 -0700745 if (sysctl_oom_kill_allocating_task && current->mm &&
KOSAKI Motohiroa96cfd62010-08-09 17:19:38 -0700746 !oom_unkillable_task(current, NULL, nodemask) &&
David Rientjes121d1ba2012-07-31 16:42:55 -0700747 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
David Rientjes6b0c81b2012-07-31 16:43:45 -0700748 get_task_struct(current);
David Rientjes2a1c9b12012-03-21 16:33:46 -0700749 oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL,
750 nodemask,
751 "Out of memory (oom_kill_allocating_task)");
752 goto out;
David Rientjes0aad4b32010-08-09 17:18:59 -0700753 }
754
David Rientjes9cbb78b2012-07-31 16:43:44 -0700755 p = select_bad_process(&points, totalpages, mpol_mask, force_kill);
David Rientjes0aad4b32010-08-09 17:18:59 -0700756 /* Found nothing?!?! Either we hang forever, or we panic. */
757 if (!p) {
David Rientjese85bfd32010-09-22 13:05:10 -0700758 dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700759 panic("Out of memory and no killable processes...\n");
760 }
Rusty Russell6b4f2b52013-07-15 11:24:08 +0930761 if (p != (void *)-1UL) {
David Rientjes2a1c9b12012-03-21 16:33:46 -0700762 oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
763 nodemask, "Out of memory");
764 killed = 1;
765 }
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700766out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 /*
David Rientjes4f774b92012-07-31 16:42:37 -0700768 * Give the killed threads a good chance of exiting before trying to
769 * allocate memory again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 */
David Rientjes4f774b92012-07-31 16:42:37 -0700771 if (killed)
772 schedule_timeout_killable(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773}
David Rientjese3658932010-08-09 17:18:55 -0700774
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800775/**
776 * out_of_memory - tries to invoke OOM killer.
777 * @zonelist: zonelist pointer
778 * @gfp_mask: memory allocation flags
779 * @order: amount of memory being requested as a power of 2
780 * @nodemask: nodemask passed to page allocator
781 * @force_kill: true if a task must be killed, even if others are exiting
782 *
783 * invokes __out_of_memory if the OOM is not disabled by oom_killer_disable()
784 * when it returns false. Otherwise returns true.
785 */
786bool out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
787 int order, nodemask_t *nodemask, bool force_kill)
788{
789 bool ret = false;
790
791 down_read(&oom_sem);
792 if (!oom_killer_disabled) {
793 __out_of_memory(zonelist, gfp_mask, order, nodemask, force_kill);
794 ret = true;
795 }
796 up_read(&oom_sem);
797
798 return ret;
799}
800
David Rientjese3658932010-08-09 17:18:55 -0700801/*
802 * The pagefault handler calls here because it is out of memory, so kill a
David Rientjesefacd022012-12-12 13:52:06 -0800803 * memory-hogging task. If any populated zone has ZONE_OOM_LOCKED set, a
804 * parallel oom killing is already in progress so do nothing.
David Rientjese3658932010-08-09 17:18:55 -0700805 */
806void pagefault_out_of_memory(void)
807{
Johannes Weiner3812c8c2013-09-12 15:13:44 -0700808 struct zonelist *zonelist;
David Rientjesefacd022012-12-12 13:52:06 -0800809
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800810 down_read(&oom_sem);
Johannes Weiner49426422013-10-16 13:46:59 -0700811 if (mem_cgroup_oom_synchronize(true))
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800812 goto unlock;
Johannes Weiner3812c8c2013-09-12 15:13:44 -0700813
David Rientjes8d060bf2014-08-06 16:07:50 -0700814 zonelist = node_zonelist(first_memory_node, GFP_KERNEL);
David Rientjese972a072014-08-06 16:07:52 -0700815 if (oom_zonelist_trylock(zonelist, GFP_KERNEL)) {
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800816 if (!oom_killer_disabled)
817 __out_of_memory(NULL, 0, 0, NULL, false);
818 else
819 /*
820 * There shouldn't be any user tasks runable while the
821 * OOM killer is disabled so the current task has to
822 * be a racing OOM victim for which oom_killer_disable()
823 * is waiting for.
824 */
825 WARN_ON(test_thread_flag(TIF_MEMDIE));
826
David Rientjese972a072014-08-06 16:07:52 -0700827 oom_zonelist_unlock(zonelist, GFP_KERNEL);
David Rientjese3658932010-08-09 17:18:55 -0700828 }
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800829unlock:
830 up_read(&oom_sem);
David Rientjese3658932010-08-09 17:18:55 -0700831}