blob: 4029583a10241aaa84e3937ee216740e0a88a363 [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>
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -070029#include <linux/module.h>
30#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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -070035int sysctl_panic_on_oom;
David Rientjesfe071d72007-10-16 23:25:56 -070036int sysctl_oom_kill_allocating_task;
David Rientjesad915c42010-08-09 17:18:53 -070037int sysctl_oom_dump_tasks = 1;
David Rientjesc7d4cae2009-01-06 14:39:00 -080038static DEFINE_SPINLOCK(zone_scan_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
David Rientjes6f48d0eb2010-08-09 17:18:52 -070040#ifdef CONFIG_NUMA
41/**
42 * has_intersects_mems_allowed() - check task eligiblity for kill
43 * @tsk: task struct of which task to consider
44 * @mask: nodemask passed to page allocator for mempolicy ooms
45 *
46 * Task eligibility is determined by whether or not a candidate task, @tsk,
47 * shares the same mempolicy nodes as current if it is bound by such a policy
48 * and whether or not it has the same set of allowed cpuset nodes.
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070049 */
David Rientjes6f48d0eb2010-08-09 17:18:52 -070050static bool has_intersects_mems_allowed(struct task_struct *tsk,
51 const nodemask_t *mask)
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070052{
David Rientjes6f48d0eb2010-08-09 17:18:52 -070053 struct task_struct *start = tsk;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070054
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070055 do {
David Rientjes6f48d0eb2010-08-09 17:18:52 -070056 if (mask) {
57 /*
58 * If this is a mempolicy constrained oom, tsk's
59 * cpuset is irrelevant. Only return true if its
60 * mempolicy intersects current, otherwise it may be
61 * needlessly killed.
62 */
63 if (mempolicy_nodemask_intersects(tsk, mask))
64 return true;
65 } else {
66 /*
67 * This is not a mempolicy constrained oom, so only
68 * check the mems of tsk's cpuset.
69 */
70 if (cpuset_mems_allowed_intersects(current, tsk))
71 return true;
72 }
KOSAKI Motohirodf1090a2010-08-09 17:19:39 -070073 } while_each_thread(start, tsk);
74
David Rientjes6f48d0eb2010-08-09 17:18:52 -070075 return false;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070076}
David Rientjes6f48d0eb2010-08-09 17:18:52 -070077#else
78static bool has_intersects_mems_allowed(struct task_struct *tsk,
79 const nodemask_t *mask)
80{
81 return true;
82}
83#endif /* CONFIG_NUMA */
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070084
David Rientjes6f48d0eb2010-08-09 17:18:52 -070085/*
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -070086 * If this is a system OOM (not a memcg OOM) and the task selected to be
87 * killed is not already running at high (RT) priorities, speed up the
88 * recovery by boosting the dying task to the lowest FIFO priority.
89 * That helps with the recovery and avoids interfering with RT tasks.
90 */
91static void boost_dying_task_prio(struct task_struct *p,
92 struct mem_cgroup *mem)
93{
94 struct sched_param param = { .sched_priority = 1 };
95
96 if (mem)
97 return;
98
99 if (!rt_task(p))
100 sched_setscheduler_nocheck(p, SCHED_FIFO, &param);
101}
102
103/*
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700104 * The process p may have detached its own ->mm while exiting or through
105 * use_mm(), but one or more of its subthreads may still have a valid
106 * pointer. Return p, or any of its subthreads with a valid ->mm, with
107 * task_lock() held.
108 */
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -0700109struct task_struct *find_lock_task_mm(struct task_struct *p)
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700110{
111 struct task_struct *t = p;
112
113 do {
114 task_lock(t);
115 if (likely(t->mm))
116 return t;
117 task_unlock(t);
118 } while_each_thread(p, t);
119
120 return NULL;
121}
122
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700123/* return true if the task is not adequate as candidate victim task. */
David Rientjese85bfd32010-09-22 13:05:10 -0700124static bool oom_unkillable_task(struct task_struct *p,
125 const struct mem_cgroup *mem, const nodemask_t *nodemask)
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700126{
127 if (is_global_init(p))
128 return true;
129 if (p->flags & PF_KTHREAD)
130 return true;
131
132 /* When mem_cgroup_out_of_memory() and p is not member of the group */
133 if (mem && !task_in_mem_cgroup(p, mem))
134 return true;
135
136 /* p may not have freeable memory in nodemask */
137 if (!has_intersects_mems_allowed(p, nodemask))
138 return true;
139
140 return false;
141}
142
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143/**
David Rientjesa63d83f2010-08-09 17:19:46 -0700144 * oom_badness - heuristic function to determine which candidate task to kill
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 * @p: task struct of which task we should calculate
David Rientjesa63d83f2010-08-09 17:19:46 -0700146 * @totalpages: total present RAM allowed for page allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 *
David Rientjesa63d83f2010-08-09 17:19:46 -0700148 * The heuristic for determining which task to kill is made to be as simple and
149 * predictable as possible. The goal is to return the highest value for the
150 * task consuming the most memory to avoid subsequent oom failures.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700152unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem,
153 const nodemask_t *nodemask, unsigned long totalpages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154{
David Rientjesa63d83f2010-08-09 17:19:46 -0700155 int points;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700156
KOSAKI Motohiro26ebc982010-08-09 17:19:37 -0700157 if (oom_unkillable_task(p, mem, nodemask))
158 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700160 p = find_lock_task_mm(p);
161 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 return 0;
163
164 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700165 * Shortcut check for OOM_SCORE_ADJ_MIN so the entire heuristic doesn't
166 * need to be executed for something that cannot be killed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700168 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
169 task_unlock(p);
170 return 0;
171 }
172
173 /*
174 * When the PF_OOM_ORIGIN bit is set, it indicates the task should have
175 * priority for oom killing.
176 */
177 if (p->flags & PF_OOM_ORIGIN) {
178 task_unlock(p);
179 return 1000;
180 }
181
182 /*
183 * The memory controller may have a limit of 0 bytes, so avoid a divide
184 * by zero, if necessary.
185 */
186 if (!totalpages)
187 totalpages = 1;
188
189 /*
190 * The baseline for the badness score is the proportion of RAM that each
191 * task's rss and swap space use.
192 */
193 points = (get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS)) * 1000 /
194 totalpages;
Andrew Morton97c2c9b82006-04-18 22:20:38 -0700195 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196
197 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700198 * Root processes get 3% bonus, just like the __vm_enough_memory()
199 * implementation used by LSMs.
Hugh Dickins7ba34852007-01-05 16:37:03 -0800200 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700201 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
202 points -= 30;
Hugh Dickins7ba34852007-01-05 16:37:03 -0800203
204 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700205 * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may
206 * either completely disable oom killing or always prefer a certain
207 * task.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700209 points += p->signal->oom_score_adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
David Rientjesf19e8aa2010-09-22 13:04:52 -0700211 /*
212 * Never return 0 for an eligible task that may be killed since it's
213 * possible that no single user task uses more than 0.1% of memory and
214 * no single admin tasks uses more than 3.0%.
215 */
216 if (points <= 0)
217 return 1;
David Rientjesa63d83f2010-08-09 17:19:46 -0700218 return (points < 1000) ? points : 1000;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219}
220
221/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800222 * Determine the type of allocation constraint.
223 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800224#ifdef CONFIG_NUMA
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800225static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700226 gfp_t gfp_mask, nodemask_t *nodemask,
227 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800228{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700229 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700230 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700231 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
David Rientjesa63d83f2010-08-09 17:19:46 -0700232 bool cpuset_limited = false;
233 int nid;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800234
David Rientjesa63d83f2010-08-09 17:19:46 -0700235 /* Default to all available memory */
236 *totalpages = totalram_pages + total_swap_pages;
237
238 if (!zonelist)
239 return CONSTRAINT_NONE;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800240 /*
241 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
242 * to kill current.We have to random task kill in this case.
243 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
244 */
245 if (gfp_mask & __GFP_THISNODE)
246 return CONSTRAINT_NONE;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800247
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800248 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700249 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
250 * the page allocator means a mempolicy is in effect. Cpuset policy
251 * is enforced in get_page_from_freelist().
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800252 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700253 if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) {
254 *totalpages = total_swap_pages;
255 for_each_node_mask(nid, *nodemask)
256 *totalpages += node_spanned_pages(nid);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800257 return CONSTRAINT_MEMORY_POLICY;
David Rientjesa63d83f2010-08-09 17:19:46 -0700258 }
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800259
260 /* Check this allocation failure is caused by cpuset's wall function */
261 for_each_zone_zonelist_nodemask(zone, z, zonelist,
262 high_zoneidx, nodemask)
263 if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
David Rientjesa63d83f2010-08-09 17:19:46 -0700264 cpuset_limited = true;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800265
David Rientjesa63d83f2010-08-09 17:19:46 -0700266 if (cpuset_limited) {
267 *totalpages = total_swap_pages;
268 for_each_node_mask(nid, cpuset_current_mems_allowed)
269 *totalpages += node_spanned_pages(nid);
270 return CONSTRAINT_CPUSET;
271 }
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800272 return CONSTRAINT_NONE;
273}
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800274#else
275static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700276 gfp_t gfp_mask, nodemask_t *nodemask,
277 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800278{
David Rientjesa63d83f2010-08-09 17:19:46 -0700279 *totalpages = totalram_pages + total_swap_pages;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800280 return CONSTRAINT_NONE;
281}
282#endif
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800283
284/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 * Simple selection loop. We chose the process with the highest
286 * number of 'points'. We expect the caller will lock the tasklist.
287 *
288 * (not docbooked, we don't want this one cluttering up the manual)
289 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700290static struct task_struct *select_bad_process(unsigned int *ppoints,
291 unsigned long totalpages, struct mem_cgroup *mem,
292 const nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700293{
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700294 struct task_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 struct task_struct *chosen = NULL;
Kurt Garloff9827b782006-02-20 18:27:51 -0800296 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700298 for_each_process(p) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700299 unsigned int points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700300
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700301 if (oom_unkillable_task(p, mem, nodemask))
David Rientjes6cf86ac2010-08-09 17:18:50 -0700302 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700303
Paul Jacksona49335c2005-09-06 15:18:09 -0700304 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700305 * This task already has access to memory reserves and is
306 * being killed. Don't allow any other task access to the
307 * memory reserve.
308 *
309 * Note: this may have a chance of deadlock if it gets
310 * blocked waiting for another task which itself is waiting
311 * for memory. Is there a better alternative?
312 */
313 if (test_tsk_thread_flag(p, TIF_MEMDIE))
314 return ERR_PTR(-1UL);
315
316 /*
Dave Peterson6937a252006-06-23 02:03:13 -0700317 * This is in the process of releasing memory so wait for it
Paul Jacksona49335c2005-09-06 15:18:09 -0700318 * to finish before killing some other task by mistake.
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700319 *
320 * However, if p is the current task, we allow the 'kill' to
321 * go ahead if it is exiting: this will simply set TIF_MEMDIE,
322 * which will allow it to gain access to memory reserves in
323 * the process of exiting and releasing its resources.
Nick Pigginb78483a2006-09-29 02:01:14 -0700324 * Otherwise we could get an easy OOM deadlock.
Paul Jacksona49335c2005-09-06 15:18:09 -0700325 */
KOSAKI Motohirocef1d352010-08-09 17:19:42 -0700326 if (thread_group_empty(p) && (p->flags & PF_EXITING) && p->mm) {
Nick Pigginb78483a2006-09-29 02:01:14 -0700327 if (p != current)
328 return ERR_PTR(-1UL);
329
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700330 chosen = p;
David Rientjesa63d83f2010-08-09 17:19:46 -0700331 *ppoints = 1000;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700332 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700333
David Rientjesa63d83f2010-08-09 17:19:46 -0700334 points = oom_badness(p, mem, nodemask, totalpages);
335 if (points > *ppoints) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700336 chosen = p;
Kurt Garloff9827b782006-02-20 18:27:51 -0800337 *ppoints = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 }
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700339 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700340
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 return chosen;
342}
343
344/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700345 * dump_tasks - dump current memory state of all system tasks
David Rientjes74ab7f12010-08-09 17:18:46 -0700346 * @mem: current's memory controller, if constrained
David Rientjese85bfd32010-09-22 13:05:10 -0700347 * @nodemask: nodemask passed to page allocator for mempolicy ooms
Randy Dunlap1b578df2008-03-19 17:00:42 -0700348 *
David Rientjese85bfd32010-09-22 13:05:10 -0700349 * Dumps the current memory state of all eligible tasks. Tasks not in the same
350 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
351 * are not shown.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800352 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
David Rientjesa63d83f2010-08-09 17:19:46 -0700353 * value, oom_score_adj value, and name.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800354 *
David Rientjesfef1bdd2008-02-07 00:14:07 -0800355 * Call with tasklist_lock read-locked.
356 */
David Rientjese85bfd32010-09-22 13:05:10 -0700357static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask)
David Rientjesfef1bdd2008-02-07 00:14:07 -0800358{
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700359 struct task_struct *p;
360 struct task_struct *task;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800361
David Rientjesa63d83f2010-08-09 17:19:46 -0700362 pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n");
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700363 for_each_process(p) {
David Rientjese85bfd32010-09-22 13:05:10 -0700364 if (oom_unkillable_task(p, mem, nodemask))
David Rientjesfef1bdd2008-02-07 00:14:07 -0800365 continue;
366
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700367 task = find_lock_task_mm(p);
368 if (!task) {
David Rientjes6d2661e2009-05-28 14:34:19 -0700369 /*
David Rientjes74ab7f12010-08-09 17:18:46 -0700370 * This is a kthread or all of p's threads have already
371 * detached their mm's. There's no need to report
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700372 * them; they can't be oom killed anyway.
David Rientjes6d2661e2009-05-28 14:34:19 -0700373 */
David Rientjes6d2661e2009-05-28 14:34:19 -0700374 continue;
375 }
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700376
David Rientjesa63d83f2010-08-09 17:19:46 -0700377 pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n",
KOSAKI Motohiro8d6c83f2010-08-19 14:13:39 -0700378 task->pid, task_uid(task), task->tgid,
David Rientjesa63d83f2010-08-09 17:19:46 -0700379 task->mm->total_vm, get_mm_rss(task->mm),
380 task_cpu(task), task->signal->oom_adj,
381 task->signal->oom_score_adj, task->comm);
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700382 task_unlock(task);
383 }
David Rientjesfef1bdd2008-02-07 00:14:07 -0800384}
385
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -0800386static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
David Rientjese85bfd32010-09-22 13:05:10 -0700387 struct mem_cgroup *mem, const nodemask_t *nodemask)
David Rientjes1b604d72009-12-14 17:57:47 -0800388{
David Rientjes5e9d8342010-08-09 17:18:51 -0700389 task_lock(current);
David Rientjes1b604d72009-12-14 17:57:47 -0800390 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
David Rientjesa63d83f2010-08-09 17:19:46 -0700391 "oom_adj=%d, oom_score_adj=%d\n",
392 current->comm, gfp_mask, order, current->signal->oom_adj,
393 current->signal->oom_score_adj);
David Rientjes1b604d72009-12-14 17:57:47 -0800394 cpuset_print_task_mems_allowed(current);
395 task_unlock(current);
396 dump_stack();
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -0800397 mem_cgroup_print_oom_info(mem, p);
David Rientjes1b604d72009-12-14 17:57:47 -0800398 show_mem();
399 if (sysctl_oom_dump_tasks)
David Rientjese85bfd32010-09-22 13:05:10 -0700400 dump_tasks(mem, nodemask);
David Rientjes1b604d72009-12-14 17:57:47 -0800401}
402
KOSAKI Motohiro3b4798c2009-12-15 16:45:32 -0800403#define K(x) ((x) << (PAGE_SHIFT-10))
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700404static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405{
David Rientjesb940fd72010-08-09 17:18:57 -0700406 p = find_lock_task_mm(p);
KOSAKI Motohirobe71cf22010-08-19 14:13:38 -0700407 if (!p)
Dave Peterson01315922006-04-18 22:20:44 -0700408 return 1;
KOSAKI Motohirobe71cf22010-08-19 14:13:38 -0700409
David Rientjesb940fd72010-08-09 17:18:57 -0700410 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
411 task_pid_nr(p), p->comm, K(p->mm->total_vm),
412 K(get_mm_counter(p->mm, MM_ANONPAGES)),
413 K(get_mm_counter(p->mm, MM_FILEPAGES)));
414 task_unlock(p);
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700415
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700416
David Rientjesb940fd72010-08-09 17:18:57 -0700417 set_tsk_thread_flag(p, TIF_MEMDIE);
418 force_sig(SIGKILL, p);
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700419
420 /*
421 * We give our sacrificial lamb high priority and access to
422 * all the memory it needs. That way it should be able to
423 * exit() and clear out its resources quickly...
424 */
425 boost_dying_task_prio(p, mem);
426
Dave Peterson01315922006-04-18 22:20:44 -0700427 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700428}
David Rientjesb940fd72010-08-09 17:18:57 -0700429#undef K
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430
David Rientjes7213f502007-10-16 23:25:57 -0700431static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
David Rientjesa63d83f2010-08-09 17:19:46 -0700432 unsigned int points, unsigned long totalpages,
433 struct mem_cgroup *mem, nodemask_t *nodemask,
434 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435{
David Rientjes5e9d8342010-08-09 17:18:51 -0700436 struct task_struct *victim = p;
437 struct task_struct *child;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700438 struct task_struct *t = p;
David Rientjesa63d83f2010-08-09 17:19:46 -0700439 unsigned int victim_points = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440
David Rientjes1b604d72009-12-14 17:57:47 -0800441 if (printk_ratelimit())
David Rientjese85bfd32010-09-22 13:05:10 -0700442 dump_header(p, gfp_mask, order, mem, nodemask);
David Rientjes7213f502007-10-16 23:25:57 -0700443
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700444 /*
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 */
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700448 if (p->flags & PF_EXITING) {
David Rientjes43589972010-08-09 17:18:49 -0700449 set_tsk_thread_flag(p, TIF_MEMDIE);
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700450 boost_dying_task_prio(p, mem);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700451 return 0;
452 }
453
David Rientjes5e9d8342010-08-09 17:18:51 -0700454 task_lock(p);
David Rientjesa63d83f2010-08-09 17:19:46 -0700455 pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
David Rientjes5e9d8342010-08-09 17:18:51 -0700456 message, task_pid_nr(p), p->comm, points);
457 task_unlock(p);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800458
David Rientjes5e9d8342010-08-09 17:18:51 -0700459 /*
460 * If any of p's children has a different mm and is eligible for kill,
461 * the one with the highest badness() score is sacrificed for its
462 * parent. This attempts to lose the minimal amount of work done while
463 * still freeing memory.
464 */
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700465 do {
David Rientjes5e9d8342010-08-09 17:18:51 -0700466 list_for_each_entry(child, &t->children, sibling) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700467 unsigned int child_points;
David Rientjes5e9d8342010-08-09 17:18:51 -0700468
David Rientjesa63d83f2010-08-09 17:19:46 -0700469 /*
470 * oom_badness() returns 0 if the thread is unkillable
471 */
472 child_points = oom_badness(child, mem, nodemask,
473 totalpages);
David Rientjes5e9d8342010-08-09 17:18:51 -0700474 if (child_points > victim_points) {
475 victim = child;
476 victim_points = child_points;
477 }
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700478 }
479 } while_each_thread(p, t);
480
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700481 return oom_kill_task(victim, mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483
David Rientjes309ed882010-08-09 17:18:54 -0700484/*
485 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
486 */
487static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
David Rientjese85bfd32010-09-22 13:05:10 -0700488 int order, const nodemask_t *nodemask)
David Rientjes309ed882010-08-09 17:18:54 -0700489{
490 if (likely(!sysctl_panic_on_oom))
491 return;
492 if (sysctl_panic_on_oom != 2) {
493 /*
494 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
495 * does not panic for cpuset, mempolicy, or memcg allocation
496 * failures.
497 */
498 if (constraint != CONSTRAINT_NONE)
499 return;
500 }
501 read_lock(&tasklist_lock);
David Rientjese85bfd32010-09-22 13:05:10 -0700502 dump_header(NULL, gfp_mask, order, NULL, nodemask);
David Rientjes309ed882010-08-09 17:18:54 -0700503 read_unlock(&tasklist_lock);
504 panic("Out of memory: %s panic_on_oom is enabled\n",
505 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
506}
507
Balbir Singh00f0b822008-03-04 14:28:39 -0800508#ifdef CONFIG_CGROUP_MEM_RES_CTLR
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800509void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
510{
David Rientjesa63d83f2010-08-09 17:19:46 -0700511 unsigned long limit;
512 unsigned int points = 0;
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800513 struct task_struct *p;
514
David Rientjese85bfd32010-09-22 13:05:10 -0700515 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL);
David Rientjesa63d83f2010-08-09 17:19:46 -0700516 limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT;
Li Zefane115f2d2008-04-15 14:34:37 -0700517 read_lock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800518retry:
David Rientjesa63d83f2010-08-09 17:19:46 -0700519 p = select_bad_process(&points, limit, mem, NULL);
David Rientjesdf64f812010-05-26 14:42:41 -0700520 if (!p || PTR_ERR(p) == -1UL)
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800521 goto out;
522
David Rientjesa63d83f2010-08-09 17:19:46 -0700523 if (oom_kill_process(p, gfp_mask, 0, points, limit, mem, NULL,
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800524 "Memory cgroup out of memory"))
525 goto retry;
526out:
Li Zefane115f2d2008-04-15 14:34:37 -0700527 read_unlock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800528}
529#endif
530
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700531static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
532
533int register_oom_notifier(struct notifier_block *nb)
534{
535 return blocking_notifier_chain_register(&oom_notify_list, nb);
536}
537EXPORT_SYMBOL_GPL(register_oom_notifier);
538
539int unregister_oom_notifier(struct notifier_block *nb)
540{
541 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
542}
543EXPORT_SYMBOL_GPL(unregister_oom_notifier);
544
David Rientjes098d7f12007-10-16 23:25:55 -0700545/*
546 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
547 * if a parallel OOM killing is already taking place that includes a zone in
548 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
549 */
Minchan Kimff321fe2010-08-09 17:18:57 -0700550int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700551{
Mel Gormandd1a2392008-04-28 02:12:17 -0700552 struct zoneref *z;
553 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700554 int ret = 1;
555
David Rientjesc7d4cae2009-01-06 14:39:00 -0800556 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700557 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
558 if (zone_is_oom_locked(zone)) {
David Rientjes098d7f12007-10-16 23:25:55 -0700559 ret = 0;
560 goto out;
561 }
Mel Gormandd1a2392008-04-28 02:12:17 -0700562 }
David Rientjes098d7f12007-10-16 23:25:55 -0700563
Mel Gormandd1a2392008-04-28 02:12:17 -0700564 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
565 /*
David Rientjesc7d4cae2009-01-06 14:39:00 -0800566 * Lock each zone in the zonelist under zone_scan_lock so a
Minchan Kimff321fe2010-08-09 17:18:57 -0700567 * parallel invocation of try_set_zonelist_oom() doesn't succeed
Mel Gormandd1a2392008-04-28 02:12:17 -0700568 * when it shouldn't.
569 */
570 zone_set_flag(zone, ZONE_OOM_LOCKED);
571 }
572
David Rientjes098d7f12007-10-16 23:25:55 -0700573out:
David Rientjesc7d4cae2009-01-06 14:39:00 -0800574 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700575 return ret;
576}
577
578/*
579 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
580 * allocation attempts with zonelists containing them may now recall the OOM
581 * killer, if necessary.
582 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700583void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700584{
Mel Gormandd1a2392008-04-28 02:12:17 -0700585 struct zoneref *z;
586 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700587
David Rientjesc7d4cae2009-01-06 14:39:00 -0800588 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700589 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
590 zone_clear_flag(zone, ZONE_OOM_LOCKED);
591 }
David Rientjesc7d4cae2009-01-06 14:39:00 -0800592 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700593}
594
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800595/*
David Rientjese3658932010-08-09 17:18:55 -0700596 * Try to acquire the oom killer lock for all system zones. Returns zero if a
597 * parallel oom killing is taking place, otherwise locks all zones and returns
598 * non-zero.
599 */
600static int try_set_system_oom(void)
601{
602 struct zone *zone;
603 int ret = 1;
604
605 spin_lock(&zone_scan_lock);
606 for_each_populated_zone(zone)
607 if (zone_is_oom_locked(zone)) {
608 ret = 0;
609 goto out;
610 }
611 for_each_populated_zone(zone)
612 zone_set_flag(zone, ZONE_OOM_LOCKED);
613out:
614 spin_unlock(&zone_scan_lock);
615 return ret;
616}
617
618/*
619 * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation
620 * attempts or page faults may now recall the oom killer, if necessary.
621 */
622static void clear_system_oom(void)
623{
624 struct zone *zone;
625
626 spin_lock(&zone_scan_lock);
627 for_each_populated_zone(zone)
628 zone_clear_flag(zone, ZONE_OOM_LOCKED);
629 spin_unlock(&zone_scan_lock);
630}
631
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632/**
Dave Peterson6937a252006-06-23 02:03:13 -0700633 * out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700634 * @zonelist: zonelist pointer
635 * @gfp_mask: memory allocation flags
636 * @order: amount of memory being requested as a power of 2
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700637 * @nodemask: nodemask passed to page allocator
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 *
639 * If we run out of memory, we have the choice between either
640 * killing a random task (bad), letting the system crash (worse)
641 * OR try to be smart about which process to kill. Note that we
642 * don't have to be perfect here, we just have to be good.
643 */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800644void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
645 int order, nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646{
David Rientjese85bfd32010-09-22 13:05:10 -0700647 const nodemask_t *mpol_mask;
David Rientjes0aad4b32010-08-09 17:18:59 -0700648 struct task_struct *p;
David Rientjesa63d83f2010-08-09 17:19:46 -0700649 unsigned long totalpages;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700650 unsigned long freed = 0;
David Rientjesa63d83f2010-08-09 17:19:46 -0700651 unsigned int points;
David Rientjese3658932010-08-09 17:18:55 -0700652 enum oom_constraint constraint = CONSTRAINT_NONE;
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700653 int killed = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700654
655 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
656 if (freed > 0)
657 /* Got some memory back in the last second. */
658 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
David Rientjes7b98c2e2010-08-09 17:18:48 -0700660 /*
661 * If current has a pending SIGKILL, then automatically select it. The
662 * goal is to allow it to allocate so that it may quickly exit and free
663 * its memory.
664 */
665 if (fatal_signal_pending(current)) {
666 set_thread_flag(TIF_MEMDIE);
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700667 boost_dying_task_prio(current, NULL);
David Rientjes7b98c2e2010-08-09 17:18:48 -0700668 return;
669 }
670
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800671 /*
672 * Check if there were limitations on the allocation (only relevant for
673 * NUMA) that may require different handling.
674 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700675 constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
676 &totalpages);
David Rientjese85bfd32010-09-22 13:05:10 -0700677 mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
678 check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700679
David Rientjes2b45ab32007-05-06 14:50:00 -0700680 read_lock(&tasklist_lock);
KOSAKI Motohirof88ccad2010-08-09 17:19:36 -0700681 if (sysctl_oom_kill_allocating_task &&
KOSAKI Motohiroa96cfd62010-08-09 17:19:38 -0700682 !oom_unkillable_task(current, NULL, nodemask) &&
683 (current->signal->oom_adj != OOM_DISABLE)) {
David Rientjes0aad4b32010-08-09 17:18:59 -0700684 /*
685 * oom_kill_process() needs tasklist_lock held. If it returns
686 * non-zero, current could not be killed so we must fallback to
687 * the tasklist scan.
688 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700689 if (!oom_kill_process(current, gfp_mask, order, 0, totalpages,
690 NULL, nodemask,
David Rientjes0aad4b32010-08-09 17:18:59 -0700691 "Out of memory (oom_kill_allocating_task)"))
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700692 goto out;
David Rientjes0aad4b32010-08-09 17:18:59 -0700693 }
694
695retry:
David Rientjese85bfd32010-09-22 13:05:10 -0700696 p = select_bad_process(&points, totalpages, NULL, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700697 if (PTR_ERR(p) == -1UL)
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700698 goto out;
David Rientjes0aad4b32010-08-09 17:18:59 -0700699
700 /* Found nothing?!?! Either we hang forever, or we panic. */
701 if (!p) {
David Rientjese85bfd32010-09-22 13:05:10 -0700702 dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700703 read_unlock(&tasklist_lock);
704 panic("Out of memory and no killable processes...\n");
705 }
706
David Rientjesa63d83f2010-08-09 17:19:46 -0700707 if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
708 nodemask, "Out of memory"))
David Rientjes0aad4b32010-08-09 17:18:59 -0700709 goto retry;
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700710 killed = 1;
711out:
Andrew Morton140ffce2006-03-02 02:54:28 -0800712 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713
714 /*
715 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800716 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 */
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700718 if (killed && !test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800719 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720}
David Rientjese3658932010-08-09 17:18:55 -0700721
722/*
723 * The pagefault handler calls here because it is out of memory, so kill a
724 * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel
725 * oom killing is already in progress so do nothing. If a task is found with
726 * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit.
727 */
728void pagefault_out_of_memory(void)
729{
730 if (try_set_system_oom()) {
731 out_of_memory(NULL, 0, 0, NULL);
732 clear_system_oom();
733 }
734 if (!test_thread_flag(TIF_MEMDIE))
735 schedule_timeout_uninterruptible(1);
736}