blob: e4b0991ca3516b3fc3590f40da1ede9f8a4ba637 [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>
David Rientjesedd45542011-03-22 16:30:12 -070034#include <linux/ptrace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -070036int sysctl_panic_on_oom;
David Rientjesfe071d72007-10-16 23:25:56 -070037int sysctl_oom_kill_allocating_task;
David Rientjesad915c42010-08-09 17:18:53 -070038int sysctl_oom_dump_tasks = 1;
David Rientjesc7d4cae2009-01-06 14:39:00 -080039static DEFINE_SPINLOCK(zone_scan_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
David Rientjes72788c32011-05-24 17:11:40 -070041/**
42 * test_set_oom_score_adj() - set current's oom_score_adj and return old value
43 * @new_val: new oom_score_adj value
44 *
45 * Sets the oom_score_adj value for current to @new_val with proper
46 * synchronization and returns the old value. Usually used to temporarily
47 * set a value, save the old value in the caller, and then reinstate it later.
48 */
49int test_set_oom_score_adj(int new_val)
50{
51 struct sighand_struct *sighand = current->sighand;
52 int old_val;
53
54 spin_lock_irq(&sighand->siglock);
55 old_val = current->signal->oom_score_adj;
56 if (new_val != old_val) {
57 if (new_val == OOM_SCORE_ADJ_MIN)
58 atomic_inc(&current->mm->oom_disable_count);
59 else if (old_val == OOM_SCORE_ADJ_MIN)
60 atomic_dec(&current->mm->oom_disable_count);
61 current->signal->oom_score_adj = new_val;
62 }
63 spin_unlock_irq(&sighand->siglock);
64
65 return old_val;
66}
67
David Rientjes6f48d0eb2010-08-09 17:18:52 -070068#ifdef CONFIG_NUMA
69/**
70 * has_intersects_mems_allowed() - check task eligiblity for kill
71 * @tsk: task struct of which task to consider
72 * @mask: nodemask passed to page allocator for mempolicy ooms
73 *
74 * Task eligibility is determined by whether or not a candidate task, @tsk,
75 * shares the same mempolicy nodes as current if it is bound by such a policy
76 * and whether or not it has the same set of allowed cpuset nodes.
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070077 */
David Rientjes6f48d0eb2010-08-09 17:18:52 -070078static bool has_intersects_mems_allowed(struct task_struct *tsk,
79 const nodemask_t *mask)
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070080{
David Rientjes6f48d0eb2010-08-09 17:18:52 -070081 struct task_struct *start = tsk;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070082
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070083 do {
David Rientjes6f48d0eb2010-08-09 17:18:52 -070084 if (mask) {
85 /*
86 * If this is a mempolicy constrained oom, tsk's
87 * cpuset is irrelevant. Only return true if its
88 * mempolicy intersects current, otherwise it may be
89 * needlessly killed.
90 */
91 if (mempolicy_nodemask_intersects(tsk, mask))
92 return true;
93 } else {
94 /*
95 * This is not a mempolicy constrained oom, so only
96 * check the mems of tsk's cpuset.
97 */
98 if (cpuset_mems_allowed_intersects(current, tsk))
99 return true;
100 }
KOSAKI Motohirodf1090a2010-08-09 17:19:39 -0700101 } while_each_thread(start, tsk);
102
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700103 return false;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700104}
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700105#else
106static bool has_intersects_mems_allowed(struct task_struct *tsk,
107 const nodemask_t *mask)
108{
109 return true;
110}
111#endif /* CONFIG_NUMA */
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700112
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700113/*
114 * The process p may have detached its own ->mm while exiting or through
115 * use_mm(), but one or more of its subthreads may still have a valid
116 * pointer. Return p, or any of its subthreads with a valid ->mm, with
117 * task_lock() held.
118 */
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -0700119struct task_struct *find_lock_task_mm(struct task_struct *p)
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700120{
121 struct task_struct *t = p;
122
123 do {
124 task_lock(t);
125 if (likely(t->mm))
126 return t;
127 task_unlock(t);
128 } while_each_thread(p, t);
129
130 return NULL;
131}
132
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700133/* return true if the task is not adequate as candidate victim task. */
David Rientjese85bfd32010-09-22 13:05:10 -0700134static bool oom_unkillable_task(struct task_struct *p,
135 const struct mem_cgroup *mem, const nodemask_t *nodemask)
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700136{
137 if (is_global_init(p))
138 return true;
139 if (p->flags & PF_KTHREAD)
140 return true;
141
142 /* When mem_cgroup_out_of_memory() and p is not member of the group */
143 if (mem && !task_in_mem_cgroup(p, mem))
144 return true;
145
146 /* p may not have freeable memory in nodemask */
147 if (!has_intersects_mems_allowed(p, nodemask))
148 return true;
149
150 return false;
151}
152
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153/**
David Rientjesa63d83f2010-08-09 17:19:46 -0700154 * oom_badness - heuristic function to determine which candidate task to kill
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 * @p: task struct of which task we should calculate
David Rientjesa63d83f2010-08-09 17:19:46 -0700156 * @totalpages: total present RAM allowed for page allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 *
David Rientjesa63d83f2010-08-09 17:19:46 -0700158 * The heuristic for determining which task to kill is made to be as simple and
159 * predictable as possible. The goal is to return the highest value for the
160 * task consuming the most memory to avoid subsequent oom failures.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700162unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem,
163 const nodemask_t *nodemask, unsigned long totalpages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164{
David Rientjesa63d83f2010-08-09 17:19:46 -0700165 int points;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700166
KOSAKI Motohiro26ebc982010-08-09 17:19:37 -0700167 if (oom_unkillable_task(p, mem, nodemask))
168 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700170 p = find_lock_task_mm(p);
171 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 return 0;
173
174 /*
David Rientjese18641e2010-10-26 14:21:23 -0700175 * Shortcut check for a thread sharing p->mm that is OOM_SCORE_ADJ_MIN
176 * so the entire heuristic doesn't need to be executed for something
177 * that cannot be killed.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 */
David Rientjese18641e2010-10-26 14:21:23 -0700179 if (atomic_read(&p->mm->oom_disable_count)) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700180 task_unlock(p);
181 return 0;
182 }
183
184 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700185 * The memory controller may have a limit of 0 bytes, so avoid a divide
186 * by zero, if necessary.
187 */
188 if (!totalpages)
189 totalpages = 1;
190
191 /*
192 * The baseline for the badness score is the proportion of RAM that each
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700193 * task's rss, pagetable and swap space use.
David Rientjesa63d83f2010-08-09 17:19:46 -0700194 */
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700195 points = get_mm_rss(p->mm) + p->mm->nr_ptes;
196 points += get_mm_counter(p->mm, MM_SWAPENTS);
197
198 points *= 1000;
199 points /= totalpages;
Andrew Morton97c2c9b82006-04-18 22:20:38 -0700200 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
202 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700203 * Root processes get 3% bonus, just like the __vm_enough_memory()
204 * implementation used by LSMs.
Hugh Dickins7ba34852007-01-05 16:37:03 -0800205 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700206 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
207 points -= 30;
Hugh Dickins7ba34852007-01-05 16:37:03 -0800208
209 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700210 * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may
211 * either completely disable oom killing or always prefer a certain
212 * task.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700214 points += p->signal->oom_score_adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215
David Rientjesf19e8aa2010-09-22 13:04:52 -0700216 /*
217 * Never return 0 for an eligible task that may be killed since it's
218 * possible that no single user task uses more than 0.1% of memory and
219 * no single admin tasks uses more than 3.0%.
220 */
221 if (points <= 0)
222 return 1;
David Rientjesa63d83f2010-08-09 17:19:46 -0700223 return (points < 1000) ? points : 1000;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224}
225
226/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800227 * Determine the type of allocation constraint.
228 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800229#ifdef CONFIG_NUMA
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800230static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700231 gfp_t gfp_mask, nodemask_t *nodemask,
232 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800233{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700234 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700235 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700236 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
David Rientjesa63d83f2010-08-09 17:19:46 -0700237 bool cpuset_limited = false;
238 int nid;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800239
David Rientjesa63d83f2010-08-09 17:19:46 -0700240 /* Default to all available memory */
241 *totalpages = totalram_pages + total_swap_pages;
242
243 if (!zonelist)
244 return CONSTRAINT_NONE;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800245 /*
246 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
247 * to kill current.We have to random task kill in this case.
248 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
249 */
250 if (gfp_mask & __GFP_THISNODE)
251 return CONSTRAINT_NONE;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800252
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800253 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700254 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
255 * the page allocator means a mempolicy is in effect. Cpuset policy
256 * is enforced in get_page_from_freelist().
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800257 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700258 if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) {
259 *totalpages = total_swap_pages;
260 for_each_node_mask(nid, *nodemask)
261 *totalpages += node_spanned_pages(nid);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800262 return CONSTRAINT_MEMORY_POLICY;
David Rientjesa63d83f2010-08-09 17:19:46 -0700263 }
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800264
265 /* Check this allocation failure is caused by cpuset's wall function */
266 for_each_zone_zonelist_nodemask(zone, z, zonelist,
267 high_zoneidx, nodemask)
268 if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
David Rientjesa63d83f2010-08-09 17:19:46 -0700269 cpuset_limited = true;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800270
David Rientjesa63d83f2010-08-09 17:19:46 -0700271 if (cpuset_limited) {
272 *totalpages = total_swap_pages;
273 for_each_node_mask(nid, cpuset_current_mems_allowed)
274 *totalpages += node_spanned_pages(nid);
275 return CONSTRAINT_CPUSET;
276 }
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800277 return CONSTRAINT_NONE;
278}
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800279#else
280static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700281 gfp_t gfp_mask, nodemask_t *nodemask,
282 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800283{
David Rientjesa63d83f2010-08-09 17:19:46 -0700284 *totalpages = totalram_pages + total_swap_pages;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800285 return CONSTRAINT_NONE;
286}
287#endif
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800288
289/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 * Simple selection loop. We chose the process with the highest
291 * number of 'points'. We expect the caller will lock the tasklist.
292 *
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,
296 unsigned long totalpages, struct mem_cgroup *mem,
297 const nodemask_t *nodemask)
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;
Kurt Garloff9827b782006-02-20 18:27:51 -0800301 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
David Rientjes3a5dda72011-03-22 16:30:09 -0700303 do_each_thread(g, p) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700304 unsigned int points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700305
Andrey Vagin30e2b412011-03-22 16:30:11 -0700306 if (!p->mm)
307 continue;
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700308 if (oom_unkillable_task(p, mem, nodemask))
David Rientjes6cf86ac2010-08-09 17:18:50 -0700309 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700310
Paul Jacksona49335c2005-09-06 15:18:09 -0700311 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700312 * This task already has access to memory reserves and is
313 * being killed. Don't allow any other task access to the
314 * memory reserve.
315 *
316 * Note: this may have a chance of deadlock if it gets
317 * blocked waiting for another task which itself is waiting
318 * for memory. Is there a better alternative?
319 */
320 if (test_tsk_thread_flag(p, TIF_MEMDIE))
321 return ERR_PTR(-1UL);
322
Andrey Vagin30e2b412011-03-22 16:30:11 -0700323 if (p->flags & PF_EXITING) {
David Rientjesedd45542011-03-22 16:30:12 -0700324 /*
325 * If p is the current task and is in the process of
326 * releasing memory, we allow the "kill" to set
327 * TIF_MEMDIE, which will allow it to gain access to
328 * memory reserves. Otherwise, it may stall forever.
329 *
330 * The loop isn't broken here, however, in case other
331 * threads are found to have already been oom killed.
332 */
333 if (p == current) {
334 chosen = p;
335 *ppoints = 1000;
336 } else {
337 /*
338 * If this task is not being ptraced on exit,
339 * then wait for it to finish before killing
340 * some other task unnecessarily.
341 */
342 if (!(task_ptrace(p->group_leader) &
343 PT_TRACE_EXIT))
344 return ERR_PTR(-1UL);
345 }
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700346 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700347
David Rientjesa63d83f2010-08-09 17:19:46 -0700348 points = oom_badness(p, mem, nodemask, totalpages);
349 if (points > *ppoints) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700350 chosen = p;
Kurt Garloff9827b782006-02-20 18:27:51 -0800351 *ppoints = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700352 }
David Rientjes3a5dda72011-03-22 16:30:09 -0700353 } while_each_thread(g, p);
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700354
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 return chosen;
356}
357
358/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700359 * dump_tasks - dump current memory state of all system tasks
David Rientjes74ab7f12010-08-09 17:18:46 -0700360 * @mem: current's memory controller, if constrained
David Rientjese85bfd32010-09-22 13:05:10 -0700361 * @nodemask: nodemask passed to page allocator for mempolicy ooms
Randy Dunlap1b578df2008-03-19 17:00:42 -0700362 *
David Rientjese85bfd32010-09-22 13:05:10 -0700363 * Dumps the current memory state of all eligible tasks. Tasks not in the same
364 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
365 * are not shown.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800366 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
David Rientjesa63d83f2010-08-09 17:19:46 -0700367 * value, oom_score_adj value, and name.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800368 *
David Rientjesfef1bdd2008-02-07 00:14:07 -0800369 * Call with tasklist_lock read-locked.
370 */
David Rientjese85bfd32010-09-22 13:05:10 -0700371static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask)
David Rientjesfef1bdd2008-02-07 00:14:07 -0800372{
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700373 struct task_struct *p;
374 struct task_struct *task;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800375
David Rientjesa63d83f2010-08-09 17:19:46 -0700376 pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n");
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700377 for_each_process(p) {
David Rientjese85bfd32010-09-22 13:05:10 -0700378 if (oom_unkillable_task(p, mem, nodemask))
David Rientjesfef1bdd2008-02-07 00:14:07 -0800379 continue;
380
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700381 task = find_lock_task_mm(p);
382 if (!task) {
David Rientjes6d2661e2009-05-28 14:34:19 -0700383 /*
David Rientjes74ab7f12010-08-09 17:18:46 -0700384 * This is a kthread or all of p's threads have already
385 * detached their mm's. There's no need to report
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700386 * them; they can't be oom killed anyway.
David Rientjes6d2661e2009-05-28 14:34:19 -0700387 */
David Rientjes6d2661e2009-05-28 14:34:19 -0700388 continue;
389 }
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700390
David Rientjesa63d83f2010-08-09 17:19:46 -0700391 pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n",
KOSAKI Motohiro8d6c83f2010-08-19 14:13:39 -0700392 task->pid, task_uid(task), task->tgid,
David Rientjesa63d83f2010-08-09 17:19:46 -0700393 task->mm->total_vm, get_mm_rss(task->mm),
394 task_cpu(task), task->signal->oom_adj,
395 task->signal->oom_score_adj, task->comm);
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700396 task_unlock(task);
397 }
David Rientjesfef1bdd2008-02-07 00:14:07 -0800398}
399
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -0800400static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
David Rientjese85bfd32010-09-22 13:05:10 -0700401 struct mem_cgroup *mem, const nodemask_t *nodemask)
David Rientjes1b604d72009-12-14 17:57:47 -0800402{
David Rientjes5e9d8342010-08-09 17:18:51 -0700403 task_lock(current);
David Rientjes1b604d72009-12-14 17:57:47 -0800404 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
David Rientjesa63d83f2010-08-09 17:19:46 -0700405 "oom_adj=%d, oom_score_adj=%d\n",
406 current->comm, gfp_mask, order, current->signal->oom_adj,
407 current->signal->oom_score_adj);
David Rientjes1b604d72009-12-14 17:57:47 -0800408 cpuset_print_task_mems_allowed(current);
409 task_unlock(current);
410 dump_stack();
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -0800411 mem_cgroup_print_oom_info(mem, p);
David Rientjesb2b755b2011-03-24 15:18:15 -0700412 show_mem(SHOW_MEM_FILTER_NODES);
David Rientjes1b604d72009-12-14 17:57:47 -0800413 if (sysctl_oom_dump_tasks)
David Rientjese85bfd32010-09-22 13:05:10 -0700414 dump_tasks(mem, nodemask);
David Rientjes1b604d72009-12-14 17:57:47 -0800415}
416
KOSAKI Motohiro3b4798c2009-12-15 16:45:32 -0800417#define K(x) ((x) << (PAGE_SHIFT-10))
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700418static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419{
David Rientjes1e99bad2010-10-26 14:21:24 -0700420 struct task_struct *q;
421 struct mm_struct *mm;
422
David Rientjesb940fd72010-08-09 17:18:57 -0700423 p = find_lock_task_mm(p);
KOSAKI Motohirobe71cf22010-08-19 14:13:38 -0700424 if (!p)
Dave Peterson01315922006-04-18 22:20:44 -0700425 return 1;
KOSAKI Motohirobe71cf22010-08-19 14:13:38 -0700426
David Rientjes1e99bad2010-10-26 14:21:24 -0700427 /* mm cannot be safely dereferenced after task_unlock(p) */
428 mm = p->mm;
429
David Rientjesb940fd72010-08-09 17:18:57 -0700430 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
431 task_pid_nr(p), p->comm, K(p->mm->total_vm),
432 K(get_mm_counter(p->mm, MM_ANONPAGES)),
433 K(get_mm_counter(p->mm, MM_FILEPAGES)));
434 task_unlock(p);
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700435
David Rientjes1e99bad2010-10-26 14:21:24 -0700436 /*
437 * Kill all processes sharing p->mm in other thread groups, if any.
438 * They don't get access to memory reserves or a higher scheduler
439 * priority, though, to avoid depletion of all memory or task
440 * starvation. This prevents mm->mmap_sem livelock when an oom killed
441 * task cannot exit because it requires the semaphore and its contended
442 * by another thread trying to allocate memory itself. That thread will
443 * now get access to memory reserves since it has a pending fatal
444 * signal.
445 */
446 for_each_process(q)
447 if (q->mm == mm && !same_thread_group(q, p)) {
448 task_lock(q); /* Protect ->comm from prctl() */
449 pr_err("Kill process %d (%s) sharing same memory\n",
450 task_pid_nr(q), q->comm);
451 task_unlock(q);
452 force_sig(SIGKILL, q);
453 }
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700454
David Rientjesb940fd72010-08-09 17:18:57 -0700455 set_tsk_thread_flag(p, TIF_MEMDIE);
456 force_sig(SIGKILL, p);
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700457
Dave Peterson01315922006-04-18 22:20:44 -0700458 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459}
David Rientjesb940fd72010-08-09 17:18:57 -0700460#undef K
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461
David Rientjes7213f502007-10-16 23:25:57 -0700462static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
David Rientjesa63d83f2010-08-09 17:19:46 -0700463 unsigned int points, unsigned long totalpages,
464 struct mem_cgroup *mem, nodemask_t *nodemask,
465 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466{
Linus Torvalds52d3c032011-03-14 15:17:07 -0700467 struct task_struct *victim = p;
David Rientjes5e9d8342010-08-09 17:18:51 -0700468 struct task_struct *child;
Linus Torvalds52d3c032011-03-14 15:17:07 -0700469 struct task_struct *t = p;
470 unsigned int victim_points = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471
David Rientjes1b604d72009-12-14 17:57:47 -0800472 if (printk_ratelimit())
David Rientjese85bfd32010-09-22 13:05:10 -0700473 dump_header(p, gfp_mask, order, mem, nodemask);
David Rientjes7213f502007-10-16 23:25:57 -0700474
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700475 /*
476 * If the task is already exiting, don't alarm the sysadmin or kill
477 * its children or threads, just set TIF_MEMDIE so it can die quickly
478 */
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700479 if (p->flags & PF_EXITING) {
David Rientjes43589972010-08-09 17:18:49 -0700480 set_tsk_thread_flag(p, TIF_MEMDIE);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700481 return 0;
482 }
483
David Rientjes5e9d8342010-08-09 17:18:51 -0700484 task_lock(p);
David Rientjesa63d83f2010-08-09 17:19:46 -0700485 pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
David Rientjes5e9d8342010-08-09 17:18:51 -0700486 message, task_pid_nr(p), p->comm, points);
487 task_unlock(p);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800488
David Rientjes5e9d8342010-08-09 17:18:51 -0700489 /*
490 * If any of p's children has a different mm and is eligible for kill,
491 * the one with the highest badness() score is sacrificed for its
492 * parent. This attempts to lose the minimal amount of work done while
493 * still freeing memory.
494 */
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700495 do {
David Rientjes5e9d8342010-08-09 17:18:51 -0700496 list_for_each_entry(child, &t->children, sibling) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700497 unsigned int child_points;
David Rientjes5e9d8342010-08-09 17:18:51 -0700498
David Rientjesedd45542011-03-22 16:30:12 -0700499 if (child->mm == p->mm)
500 continue;
David Rientjesa63d83f2010-08-09 17:19:46 -0700501 /*
502 * oom_badness() returns 0 if the thread is unkillable
503 */
504 child_points = oom_badness(child, mem, nodemask,
505 totalpages);
David Rientjes5e9d8342010-08-09 17:18:51 -0700506 if (child_points > victim_points) {
507 victim = child;
508 victim_points = child_points;
509 }
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700510 }
511 } while_each_thread(p, t);
512
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700513 return oom_kill_task(victim, mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514}
515
David Rientjes309ed882010-08-09 17:18:54 -0700516/*
517 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
518 */
519static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
David Rientjese85bfd32010-09-22 13:05:10 -0700520 int order, const nodemask_t *nodemask)
David Rientjes309ed882010-08-09 17:18:54 -0700521{
522 if (likely(!sysctl_panic_on_oom))
523 return;
524 if (sysctl_panic_on_oom != 2) {
525 /*
526 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
527 * does not panic for cpuset, mempolicy, or memcg allocation
528 * failures.
529 */
530 if (constraint != CONSTRAINT_NONE)
531 return;
532 }
533 read_lock(&tasklist_lock);
David Rientjese85bfd32010-09-22 13:05:10 -0700534 dump_header(NULL, gfp_mask, order, NULL, nodemask);
David Rientjes309ed882010-08-09 17:18:54 -0700535 read_unlock(&tasklist_lock);
536 panic("Out of memory: %s panic_on_oom is enabled\n",
537 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
538}
539
Balbir Singh00f0b822008-03-04 14:28:39 -0800540#ifdef CONFIG_CGROUP_MEM_RES_CTLR
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800541void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
542{
David Rientjesa63d83f2010-08-09 17:19:46 -0700543 unsigned long limit;
544 unsigned int points = 0;
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800545 struct task_struct *p;
546
David Rientjesf9434ad2011-03-23 16:42:44 -0700547 /*
548 * If current has a pending SIGKILL, then automatically select it. The
549 * goal is to allow it to allocate so that it may quickly exit and free
550 * its memory.
551 */
552 if (fatal_signal_pending(current)) {
553 set_thread_flag(TIF_MEMDIE);
David Rientjesf9434ad2011-03-23 16:42:44 -0700554 return;
555 }
556
David Rientjese85bfd32010-09-22 13:05:10 -0700557 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL);
David Rientjesa63d83f2010-08-09 17:19:46 -0700558 limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT;
Li Zefane115f2d2008-04-15 14:34:37 -0700559 read_lock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800560retry:
David Rientjesa63d83f2010-08-09 17:19:46 -0700561 p = select_bad_process(&points, limit, mem, NULL);
David Rientjesdf64f812010-05-26 14:42:41 -0700562 if (!p || PTR_ERR(p) == -1UL)
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800563 goto out;
564
David Rientjesa63d83f2010-08-09 17:19:46 -0700565 if (oom_kill_process(p, gfp_mask, 0, points, limit, mem, NULL,
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800566 "Memory cgroup out of memory"))
567 goto retry;
568out:
Li Zefane115f2d2008-04-15 14:34:37 -0700569 read_unlock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800570}
571#endif
572
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700573static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
574
575int register_oom_notifier(struct notifier_block *nb)
576{
577 return blocking_notifier_chain_register(&oom_notify_list, nb);
578}
579EXPORT_SYMBOL_GPL(register_oom_notifier);
580
581int unregister_oom_notifier(struct notifier_block *nb)
582{
583 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
584}
585EXPORT_SYMBOL_GPL(unregister_oom_notifier);
586
David Rientjes098d7f12007-10-16 23:25:55 -0700587/*
588 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
589 * if a parallel OOM killing is already taking place that includes a zone in
590 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
591 */
Minchan Kimff321fe2010-08-09 17:18:57 -0700592int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700593{
Mel Gormandd1a2392008-04-28 02:12:17 -0700594 struct zoneref *z;
595 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700596 int ret = 1;
597
David Rientjesc7d4cae2009-01-06 14:39:00 -0800598 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700599 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
600 if (zone_is_oom_locked(zone)) {
David Rientjes098d7f12007-10-16 23:25:55 -0700601 ret = 0;
602 goto out;
603 }
Mel Gormandd1a2392008-04-28 02:12:17 -0700604 }
David Rientjes098d7f12007-10-16 23:25:55 -0700605
Mel Gormandd1a2392008-04-28 02:12:17 -0700606 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
607 /*
David Rientjesc7d4cae2009-01-06 14:39:00 -0800608 * Lock each zone in the zonelist under zone_scan_lock so a
Minchan Kimff321fe2010-08-09 17:18:57 -0700609 * parallel invocation of try_set_zonelist_oom() doesn't succeed
Mel Gormandd1a2392008-04-28 02:12:17 -0700610 * when it shouldn't.
611 */
612 zone_set_flag(zone, ZONE_OOM_LOCKED);
613 }
614
David Rientjes098d7f12007-10-16 23:25:55 -0700615out:
David Rientjesc7d4cae2009-01-06 14:39:00 -0800616 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700617 return ret;
618}
619
620/*
621 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
622 * allocation attempts with zonelists containing them may now recall the OOM
623 * killer, if necessary.
624 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700625void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700626{
Mel Gormandd1a2392008-04-28 02:12:17 -0700627 struct zoneref *z;
628 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700629
David Rientjesc7d4cae2009-01-06 14:39:00 -0800630 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700631 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
632 zone_clear_flag(zone, ZONE_OOM_LOCKED);
633 }
David Rientjesc7d4cae2009-01-06 14:39:00 -0800634 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700635}
636
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800637/*
David Rientjese3658932010-08-09 17:18:55 -0700638 * Try to acquire the oom killer lock for all system zones. Returns zero if a
639 * parallel oom killing is taking place, otherwise locks all zones and returns
640 * non-zero.
641 */
642static int try_set_system_oom(void)
643{
644 struct zone *zone;
645 int ret = 1;
646
647 spin_lock(&zone_scan_lock);
648 for_each_populated_zone(zone)
649 if (zone_is_oom_locked(zone)) {
650 ret = 0;
651 goto out;
652 }
653 for_each_populated_zone(zone)
654 zone_set_flag(zone, ZONE_OOM_LOCKED);
655out:
656 spin_unlock(&zone_scan_lock);
657 return ret;
658}
659
660/*
661 * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation
662 * attempts or page faults may now recall the oom killer, if necessary.
663 */
664static void clear_system_oom(void)
665{
666 struct zone *zone;
667
668 spin_lock(&zone_scan_lock);
669 for_each_populated_zone(zone)
670 zone_clear_flag(zone, ZONE_OOM_LOCKED);
671 spin_unlock(&zone_scan_lock);
672}
673
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674/**
Dave Peterson6937a252006-06-23 02:03:13 -0700675 * out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700676 * @zonelist: zonelist pointer
677 * @gfp_mask: memory allocation flags
678 * @order: amount of memory being requested as a power of 2
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700679 * @nodemask: nodemask passed to page allocator
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680 *
681 * If we run out of memory, we have the choice between either
682 * killing a random task (bad), letting the system crash (worse)
683 * OR try to be smart about which process to kill. Note that we
684 * don't have to be perfect here, we just have to be good.
685 */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800686void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
687 int order, nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688{
David Rientjese85bfd32010-09-22 13:05:10 -0700689 const nodemask_t *mpol_mask;
David Rientjes0aad4b32010-08-09 17:18:59 -0700690 struct task_struct *p;
David Rientjesa63d83f2010-08-09 17:19:46 -0700691 unsigned long totalpages;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700692 unsigned long freed = 0;
David Rientjesa63d83f2010-08-09 17:19:46 -0700693 unsigned int points;
David Rientjese3658932010-08-09 17:18:55 -0700694 enum oom_constraint constraint = CONSTRAINT_NONE;
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700695 int killed = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700696
697 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
698 if (freed > 0)
699 /* Got some memory back in the last second. */
700 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
David Rientjes7b98c2e2010-08-09 17:18:48 -0700702 /*
703 * If current has a pending SIGKILL, then automatically select it. The
704 * goal is to allow it to allocate so that it may quickly exit and free
705 * its memory.
706 */
707 if (fatal_signal_pending(current)) {
708 set_thread_flag(TIF_MEMDIE);
709 return;
710 }
711
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800712 /*
713 * Check if there were limitations on the allocation (only relevant for
714 * NUMA) that may require different handling.
715 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700716 constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
717 &totalpages);
David Rientjese85bfd32010-09-22 13:05:10 -0700718 mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
719 check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700720
David Rientjes2b45ab32007-05-06 14:50:00 -0700721 read_lock(&tasklist_lock);
KOSAKI Motohirof88ccad2010-08-09 17:19:36 -0700722 if (sysctl_oom_kill_allocating_task &&
KOSAKI Motohiroa96cfd62010-08-09 17:19:38 -0700723 !oom_unkillable_task(current, NULL, nodemask) &&
David Rientjese18641e2010-10-26 14:21:23 -0700724 current->mm && !atomic_read(&current->mm->oom_disable_count)) {
David Rientjes0aad4b32010-08-09 17:18:59 -0700725 /*
726 * oom_kill_process() needs tasklist_lock held. If it returns
727 * non-zero, current could not be killed so we must fallback to
728 * the tasklist scan.
729 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700730 if (!oom_kill_process(current, gfp_mask, order, 0, totalpages,
731 NULL, nodemask,
David Rientjes0aad4b32010-08-09 17:18:59 -0700732 "Out of memory (oom_kill_allocating_task)"))
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700733 goto out;
David Rientjes0aad4b32010-08-09 17:18:59 -0700734 }
735
736retry:
David Rientjese85bfd32010-09-22 13:05:10 -0700737 p = select_bad_process(&points, totalpages, NULL, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700738 if (PTR_ERR(p) == -1UL)
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700739 goto out;
David Rientjes0aad4b32010-08-09 17:18:59 -0700740
741 /* Found nothing?!?! Either we hang forever, or we panic. */
742 if (!p) {
David Rientjese85bfd32010-09-22 13:05:10 -0700743 dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700744 read_unlock(&tasklist_lock);
745 panic("Out of memory and no killable processes...\n");
746 }
747
David Rientjesa63d83f2010-08-09 17:19:46 -0700748 if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
749 nodemask, "Out of memory"))
David Rientjes0aad4b32010-08-09 17:18:59 -0700750 goto retry;
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700751 killed = 1;
752out:
Andrew Morton140ffce2006-03-02 02:54:28 -0800753 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754
755 /*
756 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800757 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 */
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700759 if (killed && !test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800760 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761}
David Rientjese3658932010-08-09 17:18:55 -0700762
763/*
764 * The pagefault handler calls here because it is out of memory, so kill a
765 * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel
766 * oom killing is already in progress so do nothing. If a task is found with
767 * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit.
768 */
769void pagefault_out_of_memory(void)
770{
771 if (try_set_system_oom()) {
772 out_of_memory(NULL, 0, 0, NULL);
773 clear_system_oom();
774 }
775 if (!test_thread_flag(TIF_MEMDIE))
776 schedule_timeout_uninterruptible(1);
777}