blob: ea2147dabba60febfb381b8fb788c9a0cbf9ed30 [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...
7 *
8 * The routines in this file are used to kill a process when
Paul Jacksona49335c2005-09-06 15:18:09 -07009 * we're seriously out of memory. This gets called from __alloc_pages()
10 * in mm/page_alloc.c when we really run out of memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 *
12 * Since we won't call these routines often (on a well-configured
13 * machine) this file will double as a 'coding guide' and a signpost
14 * for newbie kernel hackers. It features several pointers to major
15 * kernel subsystems and hints as to where to find out what things do.
16 */
17
Alexey Dobriyan8ac773b2006-10-19 23:28:32 -070018#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/mm.h>
Alexey Dobriyan4e950f62007-07-30 02:36:13 +040020#include <linux/err.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/sched.h>
22#include <linux/swap.h>
23#include <linux/timex.h>
24#include <linux/jiffies.h>
Paul Jacksonef08e3b2005-09-06 15:18:13 -070025#include <linux/cpuset.h>
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -070026#include <linux/module.h>
27#include <linux/notifier.h>
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -080028#include <linux/memcontrol.h>
David Howells5cd9c582008-08-14 11:37:28 +010029#include <linux/security.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -070031int sysctl_panic_on_oom;
David Rientjesfe071d72007-10-16 23:25:56 -070032int sysctl_oom_kill_allocating_task;
David Rientjesfef1bdd2008-02-07 00:14:07 -080033int sysctl_oom_dump_tasks;
David Rientjesc7d4cae2009-01-06 14:39:00 -080034static DEFINE_SPINLOCK(zone_scan_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070035/* #define DEBUG */
36
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070037/*
38 * Is all threads of the target process nodes overlap ours?
39 */
40static int has_intersects_mems_allowed(struct task_struct *tsk)
41{
42 struct task_struct *t;
43
44 t = tsk;
45 do {
46 if (cpuset_mems_allowed_intersects(current, t))
47 return 1;
48 t = next_thread(t);
49 } while (t != tsk);
50
51 return 0;
52}
53
Linus Torvalds1da177e2005-04-16 15:20:36 -070054/**
Dave Peterson6937a252006-06-23 02:03:13 -070055 * badness - calculate a numeric value for how bad this task has been
Linus Torvalds1da177e2005-04-16 15:20:36 -070056 * @p: task struct of which task we should calculate
Paul Jacksona49335c2005-09-06 15:18:09 -070057 * @uptime: current uptime in seconds
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 *
59 * The formula used is relatively simple and documented inline in the
60 * function. The main rationale is that we want to select a good task
61 * to kill when we run out of memory.
62 *
63 * Good in this context means that:
64 * 1) we lose the minimum amount of work done
65 * 2) we recover a large amount of memory
66 * 3) we don't kill anything innocent of eating tons of memory
67 * 4) we want to kill the minimum amount of processes (one)
68 * 5) we try to kill the process the user expects us to kill, this
69 * algorithm has been meticulously tuned to meet the principle
70 * of least surprise ... (be careful when you change it)
71 */
72
Li Zefan97d87c92008-04-28 02:13:35 -070073unsigned long badness(struct task_struct *p, unsigned long uptime)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
Cyrill Gorcunova12888f2009-03-31 15:19:27 -070075 unsigned long points, cpu_time, run_time;
Andrew Morton97c2c9b82006-04-18 22:20:38 -070076 struct mm_struct *mm;
77 struct task_struct *child;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -070078 int oom_adj = p->signal->oom_adj;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070079 struct task_cputime task_time;
80 unsigned long utime;
81 unsigned long stime;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -070082
83 if (oom_adj == OOM_DISABLE)
84 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -070085
Andrew Morton97c2c9b82006-04-18 22:20:38 -070086 task_lock(p);
87 mm = p->mm;
88 if (!mm) {
89 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090 return 0;
Andrew Morton97c2c9b82006-04-18 22:20:38 -070091 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
93 /*
94 * The memory size of the process is the basis for the badness.
95 */
Andrew Morton97c2c9b82006-04-18 22:20:38 -070096 points = mm->total_vm;
97
98 /*
99 * After this unlock we can no longer dereference local variable `mm'
100 */
101 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102
103 /*
Hugh Dickins7ba34852007-01-05 16:37:03 -0800104 * swapoff can easily use up all memory, so kill those first.
105 */
Hugh Dickins35451be2009-09-21 17:02:27 -0700106 if (p->flags & PF_OOM_ORIGIN)
Hugh Dickins7ba34852007-01-05 16:37:03 -0800107 return ULONG_MAX;
108
109 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700110 * Processes which fork a lot of child processes are likely
Kurt Garloff9827b782006-02-20 18:27:51 -0800111 * a good choice. We add half the vmsize of the children if they
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 * have an own mm. This prevents forking servers to flood the
Kurt Garloff9827b782006-02-20 18:27:51 -0800113 * machine with an endless amount of children. In case a single
114 * child is eating the vast majority of memory, adding only half
115 * to the parents will make the child our kill candidate of choice.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 */
Andrew Morton97c2c9b82006-04-18 22:20:38 -0700117 list_for_each_entry(child, &p->children, sibling) {
118 task_lock(child);
119 if (child->mm != mm && child->mm)
120 points += child->mm->total_vm/2 + 1;
121 task_unlock(child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 }
123
124 /*
125 * CPU time is in tens of seconds and run time is in thousands
126 * of seconds. There is no particular reason for this other than
127 * that it turned out to work very well in practice.
128 */
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700129 thread_group_cputime(p, &task_time);
130 utime = cputime_to_jiffies(task_time.utime);
131 stime = cputime_to_jiffies(task_time.stime);
132 cpu_time = (utime + stime) >> (SHIFT_HZ + 3);
133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
135 if (uptime >= p->start_time.tv_sec)
136 run_time = (uptime - p->start_time.tv_sec) >> 10;
137 else
138 run_time = 0;
139
Cyrill Gorcunova12888f2009-03-31 15:19:27 -0700140 if (cpu_time)
141 points /= int_sqrt(cpu_time);
142 if (run_time)
143 points /= int_sqrt(int_sqrt(run_time));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144
145 /*
146 * Niced processes are most likely less important, so double
147 * their badness points.
148 */
149 if (task_nice(p) > 0)
150 points *= 2;
151
152 /*
153 * Superuser processes are usually more important, so we make it
154 * less likely that we kill those.
155 */
Eric Parisa2f29452008-11-11 22:02:54 +1100156 if (has_capability_noaudit(p, CAP_SYS_ADMIN) ||
157 has_capability_noaudit(p, CAP_SYS_RESOURCE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 points /= 4;
159
160 /*
161 * We don't want to kill a process with direct hardware access.
162 * Not only could that mess up the hardware, but usually users
163 * tend to only have this flag set on applications they think
164 * of as important.
165 */
Eric Parisa2f29452008-11-11 22:02:54 +1100166 if (has_capability_noaudit(p, CAP_SYS_RAWIO))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 points /= 4;
168
169 /*
Nick Piggin7887a3d2006-09-25 23:31:29 -0700170 * If p's nodes don't overlap ours, it may still help to kill p
171 * because p may have allocated or otherwise mapped memory on
172 * this node before. However it will be less likely.
173 */
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700174 if (!has_intersects_mems_allowed(p))
Nick Piggin7887a3d2006-09-25 23:31:29 -0700175 points /= 8;
176
177 /*
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700178 * Adjust the score by oom_adj.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 */
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700180 if (oom_adj) {
181 if (oom_adj > 0) {
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700182 if (!points)
183 points = 1;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700184 points <<= oom_adj;
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700185 } else
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700186 points >>= -(oom_adj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 }
188
189#ifdef DEBUG
David Rientjesa5e58a62007-07-31 00:37:32 -0700190 printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 p->pid, p->comm, points);
192#endif
193 return points;
194}
195
196/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800197 * Determine the type of allocation constraint.
198 */
David Rientjes70e24bd2007-10-16 23:25:53 -0700199static inline enum oom_constraint constrained_alloc(struct zonelist *zonelist,
200 gfp_t gfp_mask)
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800201{
202#ifdef CONFIG_NUMA
Mel Gorman54a6eb52008-04-28 02:12:16 -0700203 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700204 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700205 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Christoph Lameteree31af52007-10-16 01:25:30 -0700206 nodemask_t nodes = node_states[N_HIGH_MEMORY];
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800207
Mel Gorman54a6eb52008-04-28 02:12:16 -0700208 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
209 if (cpuset_zone_allowed_softwall(zone, gfp_mask))
210 node_clear(zone_to_nid(zone), nodes);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800211 else
212 return CONSTRAINT_CPUSET;
213
214 if (!nodes_empty(nodes))
215 return CONSTRAINT_MEMORY_POLICY;
216#endif
217
218 return CONSTRAINT_NONE;
219}
220
221/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 * Simple selection loop. We chose the process with the highest
223 * number of 'points'. We expect the caller will lock the tasklist.
224 *
225 * (not docbooked, we don't want this one cluttering up the manual)
226 */
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800227static struct task_struct *select_bad_process(unsigned long *ppoints,
228 struct mem_cgroup *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229{
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700230 struct task_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 struct task_struct *chosen = NULL;
232 struct timespec uptime;
Kurt Garloff9827b782006-02-20 18:27:51 -0800233 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234
235 do_posix_clock_monotonic_gettime(&uptime);
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700236 for_each_process(p) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700237 unsigned long points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700238
Oleg Nesterov28324d12006-09-29 02:01:12 -0700239 /*
240 * skip kernel threads and tasks which have already released
241 * their mm.
242 */
Nick Piggin5081dde2006-09-25 23:31:32 -0700243 if (!p->mm)
244 continue;
Oleg Nesterov28324d12006-09-29 02:01:12 -0700245 /* skip the init task */
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700246 if (is_global_init(p))
Paul Jacksona49335c2005-09-06 15:18:09 -0700247 continue;
David Rientjes4c4a2212008-02-07 00:14:06 -0800248 if (mem && !task_in_mem_cgroup(p, mem))
249 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700250
Paul Jacksona49335c2005-09-06 15:18:09 -0700251 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700252 * This task already has access to memory reserves and is
253 * being killed. Don't allow any other task access to the
254 * memory reserve.
255 *
256 * Note: this may have a chance of deadlock if it gets
257 * blocked waiting for another task which itself is waiting
258 * for memory. Is there a better alternative?
259 */
260 if (test_tsk_thread_flag(p, TIF_MEMDIE))
261 return ERR_PTR(-1UL);
262
263 /*
Dave Peterson6937a252006-06-23 02:03:13 -0700264 * This is in the process of releasing memory so wait for it
Paul Jacksona49335c2005-09-06 15:18:09 -0700265 * to finish before killing some other task by mistake.
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700266 *
267 * However, if p is the current task, we allow the 'kill' to
268 * go ahead if it is exiting: this will simply set TIF_MEMDIE,
269 * which will allow it to gain access to memory reserves in
270 * the process of exiting and releasing its resources.
Nick Pigginb78483a2006-09-29 02:01:14 -0700271 * Otherwise we could get an easy OOM deadlock.
Paul Jacksona49335c2005-09-06 15:18:09 -0700272 */
Nick Pigginb78483a2006-09-29 02:01:14 -0700273 if (p->flags & PF_EXITING) {
274 if (p != current)
275 return ERR_PTR(-1UL);
276
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700277 chosen = p;
278 *ppoints = ULONG_MAX;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700279 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700280
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700281 if (p->signal->oom_adj == OOM_DISABLE)
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700282 continue;
283
Li Zefan97d87c92008-04-28 02:13:35 -0700284 points = badness(p, uptime.tv_sec);
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700285 if (points > *ppoints || !chosen) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700286 chosen = p;
Kurt Garloff9827b782006-02-20 18:27:51 -0800287 *ppoints = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288 }
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700289 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700290
Linus Torvalds1da177e2005-04-16 15:20:36 -0700291 return chosen;
292}
293
294/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700295 * dump_tasks - dump current memory state of all system tasks
296 * @mem: target memory controller
297 *
David Rientjesfef1bdd2008-02-07 00:14:07 -0800298 * Dumps the current memory state of all system tasks, excluding kernel threads.
299 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
300 * score, and name.
301 *
302 * If the actual is non-NULL, only tasks that are a member of the mem_cgroup are
303 * shown.
304 *
305 * Call with tasklist_lock read-locked.
306 */
307static void dump_tasks(const struct mem_cgroup *mem)
308{
309 struct task_struct *g, *p;
310
311 printk(KERN_INFO "[ pid ] uid tgid total_vm rss cpu oom_adj "
312 "name\n");
313 do_each_thread(g, p) {
David Rientjes6d2661e2009-05-28 14:34:19 -0700314 struct mm_struct *mm;
315
David Rientjesfef1bdd2008-02-07 00:14:07 -0800316 if (mem && !task_in_mem_cgroup(p, mem))
317 continue;
David Rientjesb4416d22008-11-06 12:53:29 -0800318 if (!thread_group_leader(p))
319 continue;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800320
321 task_lock(p);
David Rientjes6d2661e2009-05-28 14:34:19 -0700322 mm = p->mm;
323 if (!mm) {
324 /*
325 * total_vm and rss sizes do not exist for tasks with no
326 * mm so there's no need to report them; they can't be
327 * oom killed anyway.
328 */
329 task_unlock(p);
330 continue;
331 }
David Rientjesfef1bdd2008-02-07 00:14:07 -0800332 printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3d %3d %s\n",
David Rientjes6d2661e2009-05-28 14:34:19 -0700333 p->pid, __task_cred(p)->uid, p->tgid, mm->total_vm,
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700334 get_mm_rss(mm), (int)task_cpu(p), p->signal->oom_adj,
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700335 p->comm);
David Rientjesfef1bdd2008-02-07 00:14:07 -0800336 task_unlock(p);
337 } while_each_thread(g, p);
338}
339
Randy Dunlap1b578df2008-03-19 17:00:42 -0700340/*
Ram Gupta5a291b982006-09-25 23:31:54 -0700341 * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO
342 * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO
343 * set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 */
Nick Pigginf3af38d2006-12-06 20:31:51 -0800345static void __oom_kill_task(struct task_struct *p, int verbose)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346{
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700347 if (is_global_init(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 WARN_ON(1);
349 printk(KERN_WARNING "tried to kill init!\n");
350 return;
351 }
352
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700353 if (!p->mm) {
354 WARN_ON(1);
355 printk(KERN_WARNING "tried to kill an mm-less task!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 return;
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700357 }
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700358
Nick Pigginf3af38d2006-12-06 20:31:51 -0800359 if (verbose)
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700360 printk(KERN_ERR "Killed process %d (%s)\n",
361 task_pid_nr(p), p->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362
363 /*
364 * We give our sacrificial lamb high priority and access to
365 * all the memory it needs. That way it should be able to
366 * exit() and clear out its resources quickly...
367 */
Peter Zijlstrafa717062008-01-25 21:08:27 +0100368 p->rt.time_slice = HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369 set_tsk_thread_flag(p, TIF_MEMDIE);
370
371 force_sig(SIGKILL, p);
372}
373
Nick Pigginf3af38d2006-12-06 20:31:51 -0800374static int oom_kill_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375{
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700376 /* WARNING: mm may not be dereferenced since we did not obtain its
377 * value from get_task_mm(p). This is OK since all we need to do is
378 * compare mm to q->mm below.
379 *
380 * Furthermore, even if mm contains a non-NULL value, p->mm may
381 * change to NULL at any time since we do not hold task_lock(p).
382 * However, this is of no concern to us.
383 */
KOSAKI Motohiro8c5cd6f2009-09-21 17:03:15 -0700384 if (!p->mm || p->signal->oom_adj == OOM_DISABLE)
Dave Peterson01315922006-04-18 22:20:44 -0700385 return 1;
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700386
Nick Pigginf3af38d2006-12-06 20:31:51 -0800387 __oom_kill_task(p, 1);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800388
Dave Peterson01315922006-04-18 22:20:44 -0700389 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390}
391
David Rientjes7213f502007-10-16 23:25:57 -0700392static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
David Rientjesfef1bdd2008-02-07 00:14:07 -0800393 unsigned long points, struct mem_cgroup *mem,
394 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700396 struct task_struct *c;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700397
David Rientjes7213f502007-10-16 23:25:57 -0700398 if (printk_ratelimit()) {
David Rientjes2ff05b22009-06-16 15:32:56 -0700399 printk(KERN_WARNING "%s invoked oom-killer: "
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700400 "gfp_mask=0x%x, order=%d, oom_adj=%d\n",
401 current->comm, gfp_mask, order,
402 current->signal->oom_adj);
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700403 task_lock(current);
David Rientjes75aa1992009-01-06 14:39:01 -0800404 cpuset_print_task_mems_allowed(current);
405 task_unlock(current);
David Rientjes7213f502007-10-16 23:25:57 -0700406 dump_stack();
Balbir Singhe2224322009-04-02 16:57:39 -0700407 mem_cgroup_print_oom_info(mem, current);
David Rientjes7213f502007-10-16 23:25:57 -0700408 show_mem();
David Rientjesfef1bdd2008-02-07 00:14:07 -0800409 if (sysctl_oom_dump_tasks)
410 dump_tasks(mem);
David Rientjes7213f502007-10-16 23:25:57 -0700411 }
412
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700413 /*
414 * If the task is already exiting, don't alarm the sysadmin or kill
415 * its children or threads, just set TIF_MEMDIE so it can die quickly
416 */
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700417 if (p->flags & PF_EXITING) {
Nick Pigginf3af38d2006-12-06 20:31:51 -0800418 __oom_kill_task(p, 0);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700419 return 0;
420 }
421
Nick Pigginf3af38d2006-12-06 20:31:51 -0800422 printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700423 message, task_pid_nr(p), p->comm, points);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800424
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 /* Try to kill a child first */
Matthias Kaehlcke7b1915a2007-10-18 23:39:58 -0700426 list_for_each_entry(c, &p->children, sibling) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 if (c->mm == p->mm)
428 continue;
Nick Pigginf3af38d2006-12-06 20:31:51 -0800429 if (!oom_kill_task(c))
Dave Peterson01315922006-04-18 22:20:44 -0700430 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 }
Nick Pigginf3af38d2006-12-06 20:31:51 -0800432 return oom_kill_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433}
434
Balbir Singh00f0b822008-03-04 14:28:39 -0800435#ifdef CONFIG_CGROUP_MEM_RES_CTLR
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800436void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
437{
438 unsigned long points = 0;
439 struct task_struct *p;
440
Li Zefane115f2d2008-04-15 14:34:37 -0700441 read_lock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800442retry:
443 p = select_bad_process(&points, mem);
444 if (PTR_ERR(p) == -1UL)
445 goto out;
446
447 if (!p)
448 p = current;
449
David Rientjesfef1bdd2008-02-07 00:14:07 -0800450 if (oom_kill_process(p, gfp_mask, 0, points, mem,
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800451 "Memory cgroup out of memory"))
452 goto retry;
453out:
Li Zefane115f2d2008-04-15 14:34:37 -0700454 read_unlock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800455}
456#endif
457
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700458static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
459
460int register_oom_notifier(struct notifier_block *nb)
461{
462 return blocking_notifier_chain_register(&oom_notify_list, nb);
463}
464EXPORT_SYMBOL_GPL(register_oom_notifier);
465
466int unregister_oom_notifier(struct notifier_block *nb)
467{
468 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
469}
470EXPORT_SYMBOL_GPL(unregister_oom_notifier);
471
David Rientjes098d7f12007-10-16 23:25:55 -0700472/*
473 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
474 * if a parallel OOM killing is already taking place that includes a zone in
475 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
476 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700477int try_set_zone_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700478{
Mel Gormandd1a2392008-04-28 02:12:17 -0700479 struct zoneref *z;
480 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700481 int ret = 1;
482
David Rientjesc7d4cae2009-01-06 14:39:00 -0800483 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700484 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
485 if (zone_is_oom_locked(zone)) {
David Rientjes098d7f12007-10-16 23:25:55 -0700486 ret = 0;
487 goto out;
488 }
Mel Gormandd1a2392008-04-28 02:12:17 -0700489 }
David Rientjes098d7f12007-10-16 23:25:55 -0700490
Mel Gormandd1a2392008-04-28 02:12:17 -0700491 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
492 /*
David Rientjesc7d4cae2009-01-06 14:39:00 -0800493 * Lock each zone in the zonelist under zone_scan_lock so a
Mel Gormandd1a2392008-04-28 02:12:17 -0700494 * parallel invocation of try_set_zone_oom() doesn't succeed
495 * when it shouldn't.
496 */
497 zone_set_flag(zone, ZONE_OOM_LOCKED);
498 }
499
David Rientjes098d7f12007-10-16 23:25:55 -0700500out:
David Rientjesc7d4cae2009-01-06 14:39:00 -0800501 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700502 return ret;
503}
504
505/*
506 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
507 * allocation attempts with zonelists containing them may now recall the OOM
508 * killer, if necessary.
509 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700510void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700511{
Mel Gormandd1a2392008-04-28 02:12:17 -0700512 struct zoneref *z;
513 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700514
David Rientjesc7d4cae2009-01-06 14:39:00 -0800515 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700516 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
517 zone_clear_flag(zone, ZONE_OOM_LOCKED);
518 }
David Rientjesc7d4cae2009-01-06 14:39:00 -0800519 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700520}
521
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800522/*
523 * Must be called with tasklist_lock held for read.
524 */
525static void __out_of_memory(gfp_t gfp_mask, int order)
526{
David Rientjes184101b2009-05-06 16:02:55 -0700527 struct task_struct *p;
528 unsigned long points;
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800529
David Rientjes184101b2009-05-06 16:02:55 -0700530 if (sysctl_oom_kill_allocating_task)
531 if (!oom_kill_process(current, gfp_mask, order, 0, NULL,
532 "Out of memory (oom_kill_allocating_task)"))
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800533 return;
David Rientjes184101b2009-05-06 16:02:55 -0700534retry:
535 /*
536 * Rambo mode: Shoot down a process and hope it solves whatever
537 * issues we may have.
538 */
539 p = select_bad_process(&points, NULL);
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800540
David Rientjes184101b2009-05-06 16:02:55 -0700541 if (PTR_ERR(p) == -1UL)
542 return;
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800543
David Rientjes184101b2009-05-06 16:02:55 -0700544 /* Found nothing?!?! Either we hang forever, or we panic. */
545 if (!p) {
546 read_unlock(&tasklist_lock);
547 panic("Out of memory and no killable processes...\n");
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800548 }
David Rientjes184101b2009-05-06 16:02:55 -0700549
550 if (oom_kill_process(p, gfp_mask, order, points, NULL,
551 "Out of memory"))
552 goto retry;
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800553}
554
555/*
556 * pagefault handler calls into here because it is out of memory but
557 * doesn't know exactly how or why.
558 */
559void pagefault_out_of_memory(void)
560{
561 unsigned long freed = 0;
562
563 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
564 if (freed > 0)
565 /* Got some memory back in the last second. */
566 return;
567
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800568 /*
569 * If this is from memcg, oom-killer is already invoked.
570 * and not worth to go system-wide-oom.
571 */
572 if (mem_cgroup_oom_called(current))
573 goto rest_and_return;
574
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800575 if (sysctl_panic_on_oom)
576 panic("out of memory from page fault. panic_on_oom is selected.\n");
577
578 read_lock(&tasklist_lock);
579 __out_of_memory(0, 0); /* unknown gfp_mask and order */
580 read_unlock(&tasklist_lock);
581
582 /*
583 * Give "p" a good chance of killing itself before we
584 * retry to allocate memory.
585 */
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800586rest_and_return:
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800587 if (!test_thread_flag(TIF_MEMDIE))
588 schedule_timeout_uninterruptible(1);
589}
590
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591/**
Dave Peterson6937a252006-06-23 02:03:13 -0700592 * out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700593 * @zonelist: zonelist pointer
594 * @gfp_mask: memory allocation flags
595 * @order: amount of memory being requested as a power of 2
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 *
597 * If we run out of memory, we have the choice between either
598 * killing a random task (bad), letting the system crash (worse)
599 * OR try to be smart about which process to kill. Note that we
600 * don't have to be perfect here, we just have to be good.
601 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800602void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603{
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700604 unsigned long freed = 0;
David Rientjes70e24bd2007-10-16 23:25:53 -0700605 enum oom_constraint constraint;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700606
607 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
608 if (freed > 0)
609 /* Got some memory back in the last second. */
610 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
Yasunori Goto2b744c02007-05-06 14:49:59 -0700612 if (sysctl_panic_on_oom == 2)
613 panic("out of memory. Compulsory panic_on_oom is selected.\n");
614
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800615 /*
616 * Check if there were limitations on the allocation (only relevant for
617 * NUMA) that may require different handling.
618 */
David Rientjes2b45ab32007-05-06 14:50:00 -0700619 constraint = constrained_alloc(zonelist, gfp_mask);
David Rientjes2b45ab32007-05-06 14:50:00 -0700620 read_lock(&tasklist_lock);
621
622 switch (constraint) {
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800623 case CONSTRAINT_MEMORY_POLICY:
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800624 oom_kill_process(current, gfp_mask, order, 0, NULL,
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800625 "No available memory (MPOL_BIND)");
626 break;
627
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800628 case CONSTRAINT_NONE:
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -0700629 if (sysctl_panic_on_oom)
630 panic("out of memory. panic_on_oom is selected\n");
David Rientjesfe071d72007-10-16 23:25:56 -0700631 /* Fall-through */
632 case CONSTRAINT_CPUSET:
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800633 __out_of_memory(gfp_mask, order);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800634 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 }
636
Andrew Morton140ffce2006-03-02 02:54:28 -0800637 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638
639 /*
640 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800641 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 */
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800643 if (!test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800644 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645}