blob: 79e451a78c9ee342fea259d58c5ca1c0034ada2a [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
50 * @tsk: task struct of which task to consider
51 * @mask: nodemask passed to page allocator for mempolicy ooms
52 *
53 * Task eligibility is determined by whether or not a candidate task, @tsk,
54 * shares the same mempolicy nodes as current if it is bound by such a policy
55 * and whether or not it has the same set of allowed cpuset nodes.
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070056 */
David Rientjes6f48d0eb2010-08-09 17:18:52 -070057static bool has_intersects_mems_allowed(struct task_struct *tsk,
58 const nodemask_t *mask)
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070059{
David Rientjes6f48d0eb2010-08-09 17:18:52 -070060 struct task_struct *start = tsk;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070061
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070062 do {
David Rientjes6f48d0eb2010-08-09 17:18:52 -070063 if (mask) {
64 /*
65 * If this is a mempolicy constrained oom, tsk's
66 * cpuset is irrelevant. Only return true if its
67 * mempolicy intersects current, otherwise it may be
68 * needlessly killed.
69 */
70 if (mempolicy_nodemask_intersects(tsk, mask))
71 return true;
72 } else {
73 /*
74 * This is not a mempolicy constrained oom, so only
75 * check the mems of tsk's cpuset.
76 */
77 if (cpuset_mems_allowed_intersects(current, tsk))
78 return true;
79 }
KOSAKI Motohirodf1090a2010-08-09 17:19:39 -070080 } while_each_thread(start, tsk);
81
David Rientjes6f48d0eb2010-08-09 17:18:52 -070082 return false;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070083}
David Rientjes6f48d0eb2010-08-09 17:18:52 -070084#else
85static bool has_intersects_mems_allowed(struct task_struct *tsk,
86 const nodemask_t *mask)
87{
88 return true;
89}
90#endif /* CONFIG_NUMA */
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070091
David Rientjes6f48d0eb2010-08-09 17:18:52 -070092/*
93 * The process p may have detached its own ->mm while exiting or through
94 * use_mm(), but one or more of its subthreads may still have a valid
95 * pointer. Return p, or any of its subthreads with a valid ->mm, with
96 * task_lock() held.
97 */
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -070098struct task_struct *find_lock_task_mm(struct task_struct *p)
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -070099{
100 struct task_struct *t = p;
101
102 do {
103 task_lock(t);
104 if (likely(t->mm))
105 return t;
106 task_unlock(t);
107 } while_each_thread(p, t);
108
109 return NULL;
110}
111
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700112/* return true if the task is not adequate as candidate victim task. */
David Rientjese85bfd32010-09-22 13:05:10 -0700113static bool oom_unkillable_task(struct task_struct *p,
Johannes Weiner72835c82012-01-12 17:18:32 -0800114 const struct mem_cgroup *memcg, const nodemask_t *nodemask)
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700115{
116 if (is_global_init(p))
117 return true;
118 if (p->flags & PF_KTHREAD)
119 return true;
120
121 /* When mem_cgroup_out_of_memory() and p is not member of the group */
Johannes Weiner72835c82012-01-12 17:18:32 -0800122 if (memcg && !task_in_mem_cgroup(p, memcg))
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700123 return true;
124
125 /* p may not have freeable memory in nodemask */
126 if (!has_intersects_mems_allowed(p, nodemask))
127 return true;
128
129 return false;
130}
131
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132/**
David Rientjesa63d83f2010-08-09 17:19:46 -0700133 * oom_badness - heuristic function to determine which candidate task to kill
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 * @p: task struct of which task we should calculate
David Rientjesa63d83f2010-08-09 17:19:46 -0700135 * @totalpages: total present RAM allowed for page allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136 *
David Rientjesa63d83f2010-08-09 17:19:46 -0700137 * The heuristic for determining which task to kill is made to be as simple and
138 * predictable as possible. The goal is to return the highest value for the
139 * task consuming the most memory to avoid subsequent oom failures.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 */
David Rientjesa7f638f2012-05-29 15:06:47 -0700141unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
142 const nodemask_t *nodemask, unsigned long totalpages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143{
David Rientjes1e11ad82012-06-08 13:21:26 -0700144 long points;
David Rientjes61eafb02012-06-20 12:52:58 -0700145 long adj;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700146
Johannes Weiner72835c82012-01-12 17:18:32 -0800147 if (oom_unkillable_task(p, memcg, nodemask))
KOSAKI Motohiro26ebc982010-08-09 17:19:37 -0700148 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700150 p = find_lock_task_mm(p);
151 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 return 0;
153
David Rientjesa9c58b902012-12-11 16:02:54 -0800154 adj = (long)p->signal->oom_score_adj;
David Rientjes61eafb02012-06-20 12:52:58 -0700155 if (adj == OOM_SCORE_ADJ_MIN) {
Michal Hocko5aecc852011-11-15 14:36:07 -0800156 task_unlock(p);
157 return 0;
158 }
159
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700161 * The baseline for the badness score is the proportion of RAM that each
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700162 * task's rss, pagetable and swap space use.
David Rientjesa63d83f2010-08-09 17:19:46 -0700163 */
David Rientjesa7f638f2012-05-29 15:06:47 -0700164 points = get_mm_rss(p->mm) + p->mm->nr_ptes +
165 get_mm_counter(p->mm, MM_SWAPENTS);
Andrew Morton97c2c9b82006-04-18 22:20:38 -0700166 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167
168 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700169 * Root processes get 3% bonus, just like the __vm_enough_memory()
170 * implementation used by LSMs.
Hugh Dickins7ba34852007-01-05 16:37:03 -0800171 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700172 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
David Rientjes61eafb02012-06-20 12:52:58 -0700173 adj -= 30;
Hugh Dickins7ba34852007-01-05 16:37:03 -0800174
David Rientjes61eafb02012-06-20 12:52:58 -0700175 /* Normalize to oom_score_adj units */
176 adj *= totalpages / 1000;
177 points += adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
David Rientjesf19e8aa2010-09-22 13:04:52 -0700179 /*
David Rientjesa7f638f2012-05-29 15:06:47 -0700180 * Never return 0 for an eligible task regardless of the root bonus and
181 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
David Rientjesf19e8aa2010-09-22 13:04:52 -0700182 */
David Rientjes1e11ad82012-06-08 13:21:26 -0700183 return points > 0 ? points : 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184}
185
186/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800187 * Determine the type of allocation constraint.
188 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800189#ifdef CONFIG_NUMA
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800190static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700191 gfp_t gfp_mask, nodemask_t *nodemask,
192 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800193{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700194 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700195 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700196 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
David Rientjesa63d83f2010-08-09 17:19:46 -0700197 bool cpuset_limited = false;
198 int nid;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800199
David Rientjesa63d83f2010-08-09 17:19:46 -0700200 /* Default to all available memory */
201 *totalpages = totalram_pages + total_swap_pages;
202
203 if (!zonelist)
204 return CONSTRAINT_NONE;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800205 /*
206 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
207 * to kill current.We have to random task kill in this case.
208 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
209 */
210 if (gfp_mask & __GFP_THISNODE)
211 return CONSTRAINT_NONE;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800212
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800213 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700214 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
215 * the page allocator means a mempolicy is in effect. Cpuset policy
216 * is enforced in get_page_from_freelist().
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800217 */
Lai Jiangshanbd3a66c2012-12-12 13:51:28 -0800218 if (nodemask && !nodes_subset(node_states[N_MEMORY], *nodemask)) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700219 *totalpages = total_swap_pages;
220 for_each_node_mask(nid, *nodemask)
221 *totalpages += node_spanned_pages(nid);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800222 return CONSTRAINT_MEMORY_POLICY;
David Rientjesa63d83f2010-08-09 17:19:46 -0700223 }
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800224
225 /* Check this allocation failure is caused by cpuset's wall function */
226 for_each_zone_zonelist_nodemask(zone, z, zonelist,
227 high_zoneidx, nodemask)
228 if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
David Rientjesa63d83f2010-08-09 17:19:46 -0700229 cpuset_limited = true;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800230
David Rientjesa63d83f2010-08-09 17:19:46 -0700231 if (cpuset_limited) {
232 *totalpages = total_swap_pages;
233 for_each_node_mask(nid, cpuset_current_mems_allowed)
234 *totalpages += node_spanned_pages(nid);
235 return CONSTRAINT_CPUSET;
236 }
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800237 return CONSTRAINT_NONE;
238}
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800239#else
240static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700241 gfp_t gfp_mask, nodemask_t *nodemask,
242 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800243{
David Rientjesa63d83f2010-08-09 17:19:46 -0700244 *totalpages = totalram_pages + total_swap_pages;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800245 return CONSTRAINT_NONE;
246}
247#endif
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800248
David Rientjes9cbb78b2012-07-31 16:43:44 -0700249enum oom_scan_t oom_scan_process_thread(struct task_struct *task,
250 unsigned long totalpages, const nodemask_t *nodemask,
251 bool force_kill)
David Rientjes462607e2012-07-31 16:43:40 -0700252{
253 if (task->exit_state)
254 return OOM_SCAN_CONTINUE;
David Rientjes9cbb78b2012-07-31 16:43:44 -0700255 if (oom_unkillable_task(task, NULL, nodemask))
David Rientjes462607e2012-07-31 16:43:40 -0700256 return OOM_SCAN_CONTINUE;
257
258 /*
259 * This task already has access to memory reserves and is being killed.
260 * Don't allow any other task to have access to the reserves.
261 */
262 if (test_tsk_thread_flag(task, TIF_MEMDIE)) {
263 if (unlikely(frozen(task)))
264 __thaw_task(task);
265 if (!force_kill)
266 return OOM_SCAN_ABORT;
267 }
268 if (!task->mm)
269 return OOM_SCAN_CONTINUE;
270
David Rientjese1e12d22012-12-11 16:02:56 -0800271 /*
272 * If task is allocating a lot of memory and has been marked to be
273 * killed first if it triggers an oom, then select it.
274 */
275 if (oom_task_origin(task))
276 return OOM_SCAN_SELECT;
277
David Rientjes9ff48682012-12-11 16:01:30 -0800278 if (task->flags & PF_EXITING && !force_kill) {
David Rientjes462607e2012-07-31 16:43:40 -0700279 /*
David Rientjes9ff48682012-12-11 16:01:30 -0800280 * If this task is not being ptraced on exit, then wait for it
281 * to finish before killing some other task unnecessarily.
David Rientjes462607e2012-07-31 16:43:40 -0700282 */
David Rientjes9ff48682012-12-11 16:01:30 -0800283 if (!(task->group_leader->ptrace & PT_TRACE_EXIT))
284 return OOM_SCAN_ABORT;
David Rientjes462607e2012-07-31 16:43:40 -0700285 }
286 return OOM_SCAN_OK;
287}
288
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800289/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 * Simple selection loop. We chose the process with the highest
David Rientjes6b0c81b2012-07-31 16:43:45 -0700291 * number of 'points'.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292 *
293 * (not docbooked, we don't want this one cluttering up the manual)
294 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700295static struct task_struct *select_bad_process(unsigned int *ppoints,
David Rientjes9cbb78b2012-07-31 16:43:44 -0700296 unsigned long totalpages, const nodemask_t *nodemask,
297 bool force_kill)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298{
David Rientjes3a5dda72011-03-22 16:30:09 -0700299 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 struct task_struct *chosen = NULL;
David Rientjesa7f638f2012-05-29 15:06:47 -0700301 unsigned long chosen_points = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
David Rientjes6b0c81b2012-07-31 16:43:45 -0700303 rcu_read_lock();
David Rientjes3a5dda72011-03-22 16:30:09 -0700304 do_each_thread(g, p) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700305 unsigned int points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700306
David Rientjes9cbb78b2012-07-31 16:43:44 -0700307 switch (oom_scan_process_thread(p, totalpages, nodemask,
David Rientjes462607e2012-07-31 16:43:40 -0700308 force_kill)) {
309 case OOM_SCAN_SELECT:
310 chosen = p;
311 chosen_points = ULONG_MAX;
312 /* fall through */
313 case OOM_SCAN_CONTINUE:
Andrey Vagin30e2b412011-03-22 16:30:11 -0700314 continue;
David Rientjes462607e2012-07-31 16:43:40 -0700315 case OOM_SCAN_ABORT:
David Rientjes6b0c81b2012-07-31 16:43:45 -0700316 rcu_read_unlock();
David Rientjes462607e2012-07-31 16:43:40 -0700317 return ERR_PTR(-1UL);
318 case OOM_SCAN_OK:
319 break;
320 };
David Rientjes9cbb78b2012-07-31 16:43:44 -0700321 points = oom_badness(p, NULL, nodemask, totalpages);
David Rientjesa7f638f2012-05-29 15:06:47 -0700322 if (points > chosen_points) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700323 chosen = p;
David Rientjesa7f638f2012-05-29 15:06:47 -0700324 chosen_points = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 }
David Rientjes3a5dda72011-03-22 16:30:09 -0700326 } while_each_thread(g, p);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700327 if (chosen)
328 get_task_struct(chosen);
329 rcu_read_unlock();
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700330
David Rientjesa7f638f2012-05-29 15:06:47 -0700331 *ppoints = chosen_points * 1000 / totalpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 return chosen;
333}
334
335/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700336 * dump_tasks - dump current memory state of all system tasks
Wanpeng Lidad75572012-06-20 12:53:01 -0700337 * @memcg: current's memory controller, if constrained
David Rientjese85bfd32010-09-22 13:05:10 -0700338 * @nodemask: nodemask passed to page allocator for mempolicy ooms
Randy Dunlap1b578df2008-03-19 17:00:42 -0700339 *
David Rientjese85bfd32010-09-22 13:05:10 -0700340 * Dumps the current memory state of all eligible tasks. Tasks not in the same
341 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
342 * are not shown.
David Rientjesde34d962012-07-31 16:42:56 -0700343 * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
344 * swapents, oom_score_adj value, and name.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800345 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800346static void dump_tasks(const struct mem_cgroup *memcg, const nodemask_t *nodemask)
David Rientjesfef1bdd2008-02-07 00:14:07 -0800347{
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700348 struct task_struct *p;
349 struct task_struct *task;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800350
David Rientjesde34d962012-07-31 16:42:56 -0700351 pr_info("[ pid ] uid tgid total_vm rss nr_ptes swapents oom_score_adj name\n");
David Rientjes6b0c81b2012-07-31 16:43:45 -0700352 rcu_read_lock();
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700353 for_each_process(p) {
Johannes Weiner72835c82012-01-12 17:18:32 -0800354 if (oom_unkillable_task(p, memcg, nodemask))
David Rientjesfef1bdd2008-02-07 00:14:07 -0800355 continue;
356
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700357 task = find_lock_task_mm(p);
358 if (!task) {
David Rientjes6d2661e2009-05-28 14:34:19 -0700359 /*
David Rientjes74ab7f12010-08-09 17:18:46 -0700360 * This is a kthread or all of p's threads have already
361 * detached their mm's. There's no need to report
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700362 * them; they can't be oom killed anyway.
David Rientjes6d2661e2009-05-28 14:34:19 -0700363 */
David Rientjes6d2661e2009-05-28 14:34:19 -0700364 continue;
365 }
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700366
David Rientjesa9c58b902012-12-11 16:02:54 -0800367 pr_info("[%5d] %5d %5d %8lu %8lu %7lu %8lu %5hd %s\n",
Eric W. Biederman078de5f2012-02-08 07:00:08 -0800368 task->pid, from_kuid(&init_user_ns, task_uid(task)),
369 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
David Rientjesde34d962012-07-31 16:42:56 -0700370 task->mm->nr_ptes,
371 get_mm_counter(task->mm, MM_SWAPENTS),
David Rientjesa63d83f2010-08-09 17:19:46 -0700372 task->signal->oom_score_adj, task->comm);
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700373 task_unlock(task);
374 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700375 rcu_read_unlock();
David Rientjesfef1bdd2008-02-07 00:14:07 -0800376}
377
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -0800378static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
Johannes Weiner72835c82012-01-12 17:18:32 -0800379 struct mem_cgroup *memcg, const nodemask_t *nodemask)
David Rientjes1b604d72009-12-14 17:57:47 -0800380{
David Rientjes5e9d8342010-08-09 17:18:51 -0700381 task_lock(current);
David Rientjes1b604d72009-12-14 17:57:47 -0800382 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
David Rientjesa9c58b902012-12-11 16:02:54 -0800383 "oom_score_adj=%hd\n",
Davidlohr Bueso01dc52e2012-10-08 16:29:30 -0700384 current->comm, gfp_mask, order,
David Rientjesa63d83f2010-08-09 17:19:46 -0700385 current->signal->oom_score_adj);
David Rientjes1b604d72009-12-14 17:57:47 -0800386 cpuset_print_task_mems_allowed(current);
387 task_unlock(current);
388 dump_stack();
Sha Zhengju58cf1882013-02-22 16:32:05 -0800389 if (memcg)
390 mem_cgroup_print_oom_info(memcg, p);
391 else
392 show_mem(SHOW_MEM_FILTER_NODES);
David Rientjes1b604d72009-12-14 17:57:47 -0800393 if (sysctl_oom_dump_tasks)
Johannes Weiner72835c82012-01-12 17:18:32 -0800394 dump_tasks(memcg, nodemask);
David Rientjes1b604d72009-12-14 17:57:47 -0800395}
396
KOSAKI Motohiro3b4798c2009-12-15 16:45:32 -0800397#define K(x) ((x) << (PAGE_SHIFT-10))
David Rientjes6b0c81b2012-07-31 16:43:45 -0700398/*
399 * Must be called while holding a reference to p, which will be released upon
400 * returning.
401 */
David Rientjes9cbb78b2012-07-31 16:43:44 -0700402void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
403 unsigned int points, unsigned long totalpages,
404 struct mem_cgroup *memcg, nodemask_t *nodemask,
405 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406{
Linus Torvalds52d3c032011-03-14 15:17:07 -0700407 struct task_struct *victim = p;
David Rientjes5e9d8342010-08-09 17:18:51 -0700408 struct task_struct *child;
Linus Torvalds52d3c032011-03-14 15:17:07 -0700409 struct task_struct *t = p;
David Rientjes647f2bd2012-03-21 16:33:46 -0700410 struct mm_struct *mm;
Linus Torvalds52d3c032011-03-14 15:17:07 -0700411 unsigned int victim_points = 0;
David Rientjesdc3f21e2012-03-21 16:33:47 -0700412 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
413 DEFAULT_RATELIMIT_BURST);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700415 /*
416 * If the task is already exiting, don't alarm the sysadmin or kill
417 * its children or threads, just set TIF_MEMDIE so it can die quickly
418 */
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700419 if (p->flags & PF_EXITING) {
David Rientjes43589972010-08-09 17:18:49 -0700420 set_tsk_thread_flag(p, TIF_MEMDIE);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700421 put_task_struct(p);
David Rientjes2a1c9b12012-03-21 16:33:46 -0700422 return;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700423 }
424
David Rientjesdc3f21e2012-03-21 16:33:47 -0700425 if (__ratelimit(&oom_rs))
David Rientjes8447d952012-03-21 16:33:47 -0700426 dump_header(p, gfp_mask, order, memcg, nodemask);
427
David Rientjes5e9d8342010-08-09 17:18:51 -0700428 task_lock(p);
David Rientjesa63d83f2010-08-09 17:19:46 -0700429 pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
David Rientjes5e9d8342010-08-09 17:18:51 -0700430 message, task_pid_nr(p), p->comm, points);
431 task_unlock(p);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800432
David Rientjes5e9d8342010-08-09 17:18:51 -0700433 /*
434 * If any of p's children has a different mm and is eligible for kill,
David Rientjes11239832011-07-25 17:12:17 -0700435 * the one with the highest oom_badness() score is sacrificed for its
David Rientjes5e9d8342010-08-09 17:18:51 -0700436 * parent. This attempts to lose the minimal amount of work done while
437 * still freeing memory.
438 */
David Rientjes6b0c81b2012-07-31 16:43:45 -0700439 read_lock(&tasklist_lock);
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700440 do {
David Rientjes5e9d8342010-08-09 17:18:51 -0700441 list_for_each_entry(child, &t->children, sibling) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700442 unsigned int child_points;
David Rientjes5e9d8342010-08-09 17:18:51 -0700443
David Rientjesedd45542011-03-22 16:30:12 -0700444 if (child->mm == p->mm)
445 continue;
David Rientjesa63d83f2010-08-09 17:19:46 -0700446 /*
447 * oom_badness() returns 0 if the thread is unkillable
448 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800449 child_points = oom_badness(child, memcg, nodemask,
David Rientjesa63d83f2010-08-09 17:19:46 -0700450 totalpages);
David Rientjes5e9d8342010-08-09 17:18:51 -0700451 if (child_points > victim_points) {
David Rientjes6b0c81b2012-07-31 16:43:45 -0700452 put_task_struct(victim);
David Rientjes5e9d8342010-08-09 17:18:51 -0700453 victim = child;
454 victim_points = child_points;
David Rientjes6b0c81b2012-07-31 16:43:45 -0700455 get_task_struct(victim);
David Rientjes5e9d8342010-08-09 17:18:51 -0700456 }
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700457 }
458 } while_each_thread(p, t);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700459 read_unlock(&tasklist_lock);
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700460
David Rientjes6b0c81b2012-07-31 16:43:45 -0700461 rcu_read_lock();
462 p = find_lock_task_mm(victim);
463 if (!p) {
464 rcu_read_unlock();
465 put_task_struct(victim);
David Rientjes647f2bd2012-03-21 16:33:46 -0700466 return;
David Rientjes6b0c81b2012-07-31 16:43:45 -0700467 } else if (victim != p) {
468 get_task_struct(p);
469 put_task_struct(victim);
470 victim = p;
471 }
David Rientjes647f2bd2012-03-21 16:33:46 -0700472
473 /* mm cannot safely be dereferenced after task_unlock(victim) */
474 mm = victim->mm;
475 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
476 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
477 K(get_mm_counter(victim->mm, MM_ANONPAGES)),
478 K(get_mm_counter(victim->mm, MM_FILEPAGES)));
479 task_unlock(victim);
480
481 /*
482 * Kill all user processes sharing victim->mm in other thread groups, if
483 * any. They don't get access to memory reserves, though, to avoid
484 * depletion of all memory. This prevents mm->mmap_sem livelock when an
485 * oom killed thread cannot exit because it requires the semaphore and
486 * its contended by another thread trying to allocate memory itself.
487 * That thread will now get access to memory reserves since it has a
488 * pending fatal signal.
489 */
490 for_each_process(p)
491 if (p->mm == mm && !same_thread_group(p, victim) &&
492 !(p->flags & PF_KTHREAD)) {
493 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
494 continue;
495
496 task_lock(p); /* Protect ->comm from prctl() */
497 pr_err("Kill process %d (%s) sharing same memory\n",
498 task_pid_nr(p), p->comm);
499 task_unlock(p);
Oleg Nesterovd2d39302012-03-23 15:02:45 -0700500 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
David Rientjes647f2bd2012-03-21 16:33:46 -0700501 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700502 rcu_read_unlock();
David Rientjes647f2bd2012-03-21 16:33:46 -0700503
504 set_tsk_thread_flag(victim, TIF_MEMDIE);
Oleg Nesterovd2d39302012-03-23 15:02:45 -0700505 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700506 put_task_struct(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507}
David Rientjes647f2bd2012-03-21 16:33:46 -0700508#undef K
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509
David Rientjes309ed882010-08-09 17:18:54 -0700510/*
511 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
512 */
David Rientjes876aafb2012-07-31 16:43:48 -0700513void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
514 int order, const nodemask_t *nodemask)
David Rientjes309ed882010-08-09 17:18:54 -0700515{
516 if (likely(!sysctl_panic_on_oom))
517 return;
518 if (sysctl_panic_on_oom != 2) {
519 /*
520 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
521 * does not panic for cpuset, mempolicy, or memcg allocation
522 * failures.
523 */
524 if (constraint != CONSTRAINT_NONE)
525 return;
526 }
David Rientjese85bfd32010-09-22 13:05:10 -0700527 dump_header(NULL, gfp_mask, order, NULL, nodemask);
David Rientjes309ed882010-08-09 17:18:54 -0700528 panic("Out of memory: %s panic_on_oom is enabled\n",
529 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
530}
531
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700532static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
533
534int register_oom_notifier(struct notifier_block *nb)
535{
536 return blocking_notifier_chain_register(&oom_notify_list, nb);
537}
538EXPORT_SYMBOL_GPL(register_oom_notifier);
539
540int unregister_oom_notifier(struct notifier_block *nb)
541{
542 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
543}
544EXPORT_SYMBOL_GPL(unregister_oom_notifier);
545
David Rientjes098d7f12007-10-16 23:25:55 -0700546/*
547 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
548 * if a parallel OOM killing is already taking place that includes a zone in
549 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
550 */
Minchan Kimff321fe2010-08-09 17:18:57 -0700551int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700552{
Mel Gormandd1a2392008-04-28 02:12:17 -0700553 struct zoneref *z;
554 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700555 int ret = 1;
556
David Rientjesc7d4cae2009-01-06 14:39:00 -0800557 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700558 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
559 if (zone_is_oom_locked(zone)) {
David Rientjes098d7f12007-10-16 23:25:55 -0700560 ret = 0;
561 goto out;
562 }
Mel Gormandd1a2392008-04-28 02:12:17 -0700563 }
David Rientjes098d7f12007-10-16 23:25:55 -0700564
Mel Gormandd1a2392008-04-28 02:12:17 -0700565 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
566 /*
David Rientjesc7d4cae2009-01-06 14:39:00 -0800567 * Lock each zone in the zonelist under zone_scan_lock so a
Minchan Kimff321fe2010-08-09 17:18:57 -0700568 * parallel invocation of try_set_zonelist_oom() doesn't succeed
Mel Gormandd1a2392008-04-28 02:12:17 -0700569 * when it shouldn't.
570 */
571 zone_set_flag(zone, ZONE_OOM_LOCKED);
572 }
573
David Rientjes098d7f12007-10-16 23:25:55 -0700574out:
David Rientjesc7d4cae2009-01-06 14:39:00 -0800575 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700576 return ret;
577}
578
579/*
580 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
581 * allocation attempts with zonelists containing them may now recall the OOM
582 * killer, if necessary.
583 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700584void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700585{
Mel Gormandd1a2392008-04-28 02:12:17 -0700586 struct zoneref *z;
587 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700588
David Rientjesc7d4cae2009-01-06 14:39:00 -0800589 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700590 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
591 zone_clear_flag(zone, ZONE_OOM_LOCKED);
592 }
David Rientjesc7d4cae2009-01-06 14:39:00 -0800593 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700594}
595
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596/**
Dave Peterson6937a252006-06-23 02:03:13 -0700597 * out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700598 * @zonelist: zonelist pointer
599 * @gfp_mask: memory allocation flags
600 * @order: amount of memory being requested as a power of 2
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700601 * @nodemask: nodemask passed to page allocator
David Rientjes08ab9b12012-03-21 16:34:04 -0700602 * @force_kill: true if a task must be killed, even if others are exiting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 *
604 * If we run out of memory, we have the choice between either
605 * killing a random task (bad), letting the system crash (worse)
606 * OR try to be smart about which process to kill. Note that we
607 * don't have to be perfect here, we just have to be good.
608 */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800609void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
David Rientjes08ab9b12012-03-21 16:34:04 -0700610 int order, nodemask_t *nodemask, bool force_kill)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
David Rientjese85bfd32010-09-22 13:05:10 -0700612 const nodemask_t *mpol_mask;
David Rientjes0aad4b32010-08-09 17:18:59 -0700613 struct task_struct *p;
David Rientjesa63d83f2010-08-09 17:19:46 -0700614 unsigned long totalpages;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700615 unsigned long freed = 0;
David Rientjes9cbb78b2012-07-31 16:43:44 -0700616 unsigned int uninitialized_var(points);
David Rientjese3658932010-08-09 17:18:55 -0700617 enum oom_constraint constraint = CONSTRAINT_NONE;
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700618 int killed = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700619
620 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
621 if (freed > 0)
622 /* Got some memory back in the last second. */
623 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624
David Rientjes7b98c2e2010-08-09 17:18:48 -0700625 /*
David Rientjes9ff48682012-12-11 16:01:30 -0800626 * If current has a pending SIGKILL or is exiting, then automatically
627 * select it. The goal is to allow it to allocate so that it may
628 * quickly exit and free its memory.
David Rientjes7b98c2e2010-08-09 17:18:48 -0700629 */
David Rientjes9ff48682012-12-11 16:01:30 -0800630 if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
David Rientjes7b98c2e2010-08-09 17:18:48 -0700631 set_thread_flag(TIF_MEMDIE);
632 return;
633 }
634
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800635 /*
636 * Check if there were limitations on the allocation (only relevant for
637 * NUMA) that may require different handling.
638 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700639 constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
640 &totalpages);
David Rientjese85bfd32010-09-22 13:05:10 -0700641 mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
642 check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700643
David Rientjes121d1ba2012-07-31 16:42:55 -0700644 if (sysctl_oom_kill_allocating_task && current->mm &&
KOSAKI Motohiroa96cfd62010-08-09 17:19:38 -0700645 !oom_unkillable_task(current, NULL, nodemask) &&
David Rientjes121d1ba2012-07-31 16:42:55 -0700646 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
David Rientjes6b0c81b2012-07-31 16:43:45 -0700647 get_task_struct(current);
David Rientjes2a1c9b12012-03-21 16:33:46 -0700648 oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL,
649 nodemask,
650 "Out of memory (oom_kill_allocating_task)");
651 goto out;
David Rientjes0aad4b32010-08-09 17:18:59 -0700652 }
653
David Rientjes9cbb78b2012-07-31 16:43:44 -0700654 p = select_bad_process(&points, totalpages, mpol_mask, force_kill);
David Rientjes0aad4b32010-08-09 17:18:59 -0700655 /* Found nothing?!?! Either we hang forever, or we panic. */
656 if (!p) {
David Rientjese85bfd32010-09-22 13:05:10 -0700657 dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700658 panic("Out of memory and no killable processes...\n");
659 }
David Rientjes2a1c9b12012-03-21 16:33:46 -0700660 if (PTR_ERR(p) != -1UL) {
661 oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
662 nodemask, "Out of memory");
663 killed = 1;
664 }
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700665out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 /*
David Rientjes4f774b92012-07-31 16:42:37 -0700667 * Give the killed threads a good chance of exiting before trying to
668 * allocate memory again.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700669 */
David Rientjes4f774b92012-07-31 16:42:37 -0700670 if (killed)
671 schedule_timeout_killable(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700672}
David Rientjese3658932010-08-09 17:18:55 -0700673
674/*
675 * The pagefault handler calls here because it is out of memory, so kill a
David Rientjesefacd022012-12-12 13:52:06 -0800676 * memory-hogging task. If any populated zone has ZONE_OOM_LOCKED set, a
677 * parallel oom killing is already in progress so do nothing.
David Rientjese3658932010-08-09 17:18:55 -0700678 */
679void pagefault_out_of_memory(void)
680{
David Rientjesefacd022012-12-12 13:52:06 -0800681 struct zonelist *zonelist = node_zonelist(first_online_node,
682 GFP_KERNEL);
683
684 if (try_set_zonelist_oom(zonelist, GFP_KERNEL)) {
David Rientjes08ab9b12012-03-21 16:34:04 -0700685 out_of_memory(NULL, 0, 0, NULL, false);
David Rientjesefacd022012-12-12 13:52:06 -0800686 clear_zonelist_oom(zonelist, GFP_KERNEL);
David Rientjese3658932010-08-09 17:18:55 -0700687 }
David Rientjese3658932010-08-09 17:18:55 -0700688}