blob: 92bcf1db16b24ea2c9b4d9f6c93043118c021615 [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
37/**
Dave Peterson6937a252006-06-23 02:03:13 -070038 * badness - calculate a numeric value for how bad this task has been
Linus Torvalds1da177e2005-04-16 15:20:36 -070039 * @p: task struct of which task we should calculate
Paul Jacksona49335c2005-09-06 15:18:09 -070040 * @uptime: current uptime in seconds
Linus Torvalds1da177e2005-04-16 15:20:36 -070041 *
42 * The formula used is relatively simple and documented inline in the
43 * function. The main rationale is that we want to select a good task
44 * to kill when we run out of memory.
45 *
46 * Good in this context means that:
47 * 1) we lose the minimum amount of work done
48 * 2) we recover a large amount of memory
49 * 3) we don't kill anything innocent of eating tons of memory
50 * 4) we want to kill the minimum amount of processes (one)
51 * 5) we try to kill the process the user expects us to kill, this
52 * algorithm has been meticulously tuned to meet the principle
53 * of least surprise ... (be careful when you change it)
54 */
55
Li Zefan97d87c92008-04-28 02:13:35 -070056unsigned long badness(struct task_struct *p, unsigned long uptime)
Linus Torvalds1da177e2005-04-16 15:20:36 -070057{
Cyrill Gorcunova12888f2009-03-31 15:19:27 -070058 unsigned long points, cpu_time, run_time;
Andrew Morton97c2c9b2006-04-18 22:20:38 -070059 struct mm_struct *mm;
60 struct task_struct *child;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Andrew Morton97c2c9b2006-04-18 22:20:38 -070062 task_lock(p);
63 mm = p->mm;
64 if (!mm) {
65 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070066 return 0;
Andrew Morton97c2c9b2006-04-18 22:20:38 -070067 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070068
69 /*
70 * The memory size of the process is the basis for the badness.
71 */
Andrew Morton97c2c9b2006-04-18 22:20:38 -070072 points = mm->total_vm;
73
74 /*
75 * After this unlock we can no longer dereference local variable `mm'
76 */
77 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078
79 /*
Hugh Dickins7ba34852007-01-05 16:37:03 -080080 * swapoff can easily use up all memory, so kill those first.
81 */
82 if (p->flags & PF_SWAPOFF)
83 return ULONG_MAX;
84
85 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 * Processes which fork a lot of child processes are likely
Kurt Garloff9827b782006-02-20 18:27:51 -080087 * a good choice. We add half the vmsize of the children if they
Linus Torvalds1da177e2005-04-16 15:20:36 -070088 * have an own mm. This prevents forking servers to flood the
Kurt Garloff9827b782006-02-20 18:27:51 -080089 * machine with an endless amount of children. In case a single
90 * child is eating the vast majority of memory, adding only half
91 * to the parents will make the child our kill candidate of choice.
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 */
Andrew Morton97c2c9b2006-04-18 22:20:38 -070093 list_for_each_entry(child, &p->children, sibling) {
94 task_lock(child);
95 if (child->mm != mm && child->mm)
96 points += child->mm->total_vm/2 + 1;
97 task_unlock(child);
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 }
99
100 /*
101 * CPU time is in tens of seconds and run time is in thousands
102 * of seconds. There is no particular reason for this other than
103 * that it turned out to work very well in practice.
104 */
105 cpu_time = (cputime_to_jiffies(p->utime) + cputime_to_jiffies(p->stime))
106 >> (SHIFT_HZ + 3);
107
108 if (uptime >= p->start_time.tv_sec)
109 run_time = (uptime - p->start_time.tv_sec) >> 10;
110 else
111 run_time = 0;
112
Cyrill Gorcunova12888f2009-03-31 15:19:27 -0700113 if (cpu_time)
114 points /= int_sqrt(cpu_time);
115 if (run_time)
116 points /= int_sqrt(int_sqrt(run_time));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
118 /*
119 * Niced processes are most likely less important, so double
120 * their badness points.
121 */
122 if (task_nice(p) > 0)
123 points *= 2;
124
125 /*
126 * Superuser processes are usually more important, so we make it
127 * less likely that we kill those.
128 */
Eric Parisa2f29452008-11-11 22:02:54 +1100129 if (has_capability_noaudit(p, CAP_SYS_ADMIN) ||
130 has_capability_noaudit(p, CAP_SYS_RESOURCE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131 points /= 4;
132
133 /*
134 * We don't want to kill a process with direct hardware access.
135 * Not only could that mess up the hardware, but usually users
136 * tend to only have this flag set on applications they think
137 * of as important.
138 */
Eric Parisa2f29452008-11-11 22:02:54 +1100139 if (has_capability_noaudit(p, CAP_SYS_RAWIO))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140 points /= 4;
141
142 /*
Nick Piggin7887a3d2006-09-25 23:31:29 -0700143 * If p's nodes don't overlap ours, it may still help to kill p
144 * because p may have allocated or otherwise mapped memory on
145 * this node before. However it will be less likely.
146 */
David Rientjesbbe373f2007-10-16 23:25:58 -0700147 if (!cpuset_mems_allowed_intersects(current, p))
Nick Piggin7887a3d2006-09-25 23:31:29 -0700148 points /= 8;
149
150 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 * Adjust the score by oomkilladj.
152 */
153 if (p->oomkilladj) {
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700154 if (p->oomkilladj > 0) {
155 if (!points)
156 points = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 points <<= p->oomkilladj;
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700158 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 points >>= -(p->oomkilladj);
160 }
161
162#ifdef DEBUG
David Rientjesa5e58a62007-07-31 00:37:32 -0700163 printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 p->pid, p->comm, points);
165#endif
166 return points;
167}
168
169/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800170 * Determine the type of allocation constraint.
171 */
David Rientjes70e24bd2007-10-16 23:25:53 -0700172static inline enum oom_constraint constrained_alloc(struct zonelist *zonelist,
173 gfp_t gfp_mask)
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800174{
175#ifdef CONFIG_NUMA
Mel Gorman54a6eb52008-04-28 02:12:16 -0700176 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700177 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700178 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Christoph Lameteree31af52007-10-16 01:25:30 -0700179 nodemask_t nodes = node_states[N_HIGH_MEMORY];
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800180
Mel Gorman54a6eb52008-04-28 02:12:16 -0700181 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
182 if (cpuset_zone_allowed_softwall(zone, gfp_mask))
183 node_clear(zone_to_nid(zone), nodes);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800184 else
185 return CONSTRAINT_CPUSET;
186
187 if (!nodes_empty(nodes))
188 return CONSTRAINT_MEMORY_POLICY;
189#endif
190
191 return CONSTRAINT_NONE;
192}
193
194/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195 * Simple selection loop. We chose the process with the highest
196 * number of 'points'. We expect the caller will lock the tasklist.
197 *
198 * (not docbooked, we don't want this one cluttering up the manual)
199 */
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800200static struct task_struct *select_bad_process(unsigned long *ppoints,
201 struct mem_cgroup *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 struct task_struct *g, *p;
204 struct task_struct *chosen = NULL;
205 struct timespec uptime;
Kurt Garloff9827b782006-02-20 18:27:51 -0800206 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
208 do_posix_clock_monotonic_gettime(&uptime);
Paul Jacksona49335c2005-09-06 15:18:09 -0700209 do_each_thread(g, p) {
210 unsigned long points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700211
Oleg Nesterov28324d12006-09-29 02:01:12 -0700212 /*
213 * skip kernel threads and tasks which have already released
214 * their mm.
215 */
Nick Piggin5081dde2006-09-25 23:31:32 -0700216 if (!p->mm)
217 continue;
Oleg Nesterov28324d12006-09-29 02:01:12 -0700218 /* skip the init task */
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700219 if (is_global_init(p))
Paul Jacksona49335c2005-09-06 15:18:09 -0700220 continue;
David Rientjes4c4a2212008-02-07 00:14:06 -0800221 if (mem && !task_in_mem_cgroup(p, mem))
222 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700223
Paul Jacksona49335c2005-09-06 15:18:09 -0700224 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700225 * This task already has access to memory reserves and is
226 * being killed. Don't allow any other task access to the
227 * memory reserve.
228 *
229 * Note: this may have a chance of deadlock if it gets
230 * blocked waiting for another task which itself is waiting
231 * for memory. Is there a better alternative?
232 */
233 if (test_tsk_thread_flag(p, TIF_MEMDIE))
234 return ERR_PTR(-1UL);
235
236 /*
Dave Peterson6937a252006-06-23 02:03:13 -0700237 * This is in the process of releasing memory so wait for it
Paul Jacksona49335c2005-09-06 15:18:09 -0700238 * to finish before killing some other task by mistake.
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700239 *
240 * However, if p is the current task, we allow the 'kill' to
241 * go ahead if it is exiting: this will simply set TIF_MEMDIE,
242 * which will allow it to gain access to memory reserves in
243 * the process of exiting and releasing its resources.
Nick Pigginb78483a2006-09-29 02:01:14 -0700244 * Otherwise we could get an easy OOM deadlock.
Paul Jacksona49335c2005-09-06 15:18:09 -0700245 */
Nick Pigginb78483a2006-09-29 02:01:14 -0700246 if (p->flags & PF_EXITING) {
247 if (p != current)
248 return ERR_PTR(-1UL);
249
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700250 chosen = p;
251 *ppoints = ULONG_MAX;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700252 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700253
Nick Piggin4a3ede12006-09-25 23:31:30 -0700254 if (p->oomkilladj == OOM_DISABLE)
255 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
Li Zefan97d87c92008-04-28 02:13:35 -0700257 points = badness(p, uptime.tv_sec);
Kurt Garloff9827b782006-02-20 18:27:51 -0800258 if (points > *ppoints || !chosen) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700259 chosen = p;
Kurt Garloff9827b782006-02-20 18:27:51 -0800260 *ppoints = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 }
Paul Jacksona49335c2005-09-06 15:18:09 -0700262 } while_each_thread(g, p);
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700263
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 return chosen;
265}
266
267/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700268 * dump_tasks - dump current memory state of all system tasks
269 * @mem: target memory controller
270 *
David Rientjesfef1bdd2008-02-07 00:14:07 -0800271 * Dumps the current memory state of all system tasks, excluding kernel threads.
272 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
273 * score, and name.
274 *
275 * If the actual is non-NULL, only tasks that are a member of the mem_cgroup are
276 * shown.
277 *
278 * Call with tasklist_lock read-locked.
279 */
280static void dump_tasks(const struct mem_cgroup *mem)
281{
282 struct task_struct *g, *p;
283
284 printk(KERN_INFO "[ pid ] uid tgid total_vm rss cpu oom_adj "
285 "name\n");
286 do_each_thread(g, p) {
287 /*
288 * total_vm and rss sizes do not exist for tasks with a
289 * detached mm so there's no need to report them.
290 */
291 if (!p->mm)
292 continue;
293 if (mem && !task_in_mem_cgroup(p, mem))
294 continue;
David Rientjesb4416d22008-11-06 12:53:29 -0800295 if (!thread_group_leader(p))
296 continue;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800297
298 task_lock(p);
299 printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3d %3d %s\n",
David Howellsc69e8d92008-11-14 10:39:19 +1100300 p->pid, __task_cred(p)->uid, p->tgid,
301 p->mm->total_vm, get_mm_rss(p->mm), (int)task_cpu(p),
302 p->oomkilladj, p->comm);
David Rientjesfef1bdd2008-02-07 00:14:07 -0800303 task_unlock(p);
304 } while_each_thread(g, p);
305}
306
Randy Dunlap1b578df2008-03-19 17:00:42 -0700307/*
Ram Gupta5a291b982006-09-25 23:31:54 -0700308 * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO
309 * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO
310 * set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311 */
Nick Pigginf3af38d2006-12-06 20:31:51 -0800312static void __oom_kill_task(struct task_struct *p, int verbose)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313{
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700314 if (is_global_init(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 WARN_ON(1);
316 printk(KERN_WARNING "tried to kill init!\n");
317 return;
318 }
319
Oleg Nesterov01017a22006-09-29 02:01:13 -0700320 if (!p->mm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 WARN_ON(1);
322 printk(KERN_WARNING "tried to kill an mm-less task!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 return;
324 }
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700325
Nick Pigginf3af38d2006-12-06 20:31:51 -0800326 if (verbose)
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700327 printk(KERN_ERR "Killed process %d (%s)\n",
328 task_pid_nr(p), p->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329
330 /*
331 * We give our sacrificial lamb high priority and access to
332 * all the memory it needs. That way it should be able to
333 * exit() and clear out its resources quickly...
334 */
Peter Zijlstrafa717062008-01-25 21:08:27 +0100335 p->rt.time_slice = HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 set_tsk_thread_flag(p, TIF_MEMDIE);
337
338 force_sig(SIGKILL, p);
339}
340
Nick Pigginf3af38d2006-12-06 20:31:51 -0800341static int oom_kill_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
Dave Peterson01315922006-04-18 22:20:44 -0700343 struct mm_struct *mm;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700344 struct task_struct *g, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
Dave Peterson01315922006-04-18 22:20:44 -0700346 mm = p->mm;
347
348 /* WARNING: mm may not be dereferenced since we did not obtain its
349 * value from get_task_mm(p). This is OK since all we need to do is
350 * compare mm to q->mm below.
351 *
352 * Furthermore, even if mm contains a non-NULL value, p->mm may
353 * change to NULL at any time since we do not hold task_lock(p).
354 * However, this is of no concern to us.
355 */
356
Oleg Nesterov01017a22006-09-29 02:01:13 -0700357 if (mm == NULL)
Dave Peterson01315922006-04-18 22:20:44 -0700358 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800360 /*
361 * Don't kill the process if any threads are set to OOM_DISABLE
362 */
363 do_each_thread(g, q) {
Ankita Garg35ae8342007-03-16 13:38:33 -0800364 if (q->mm == mm && q->oomkilladj == OOM_DISABLE)
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800365 return 1;
366 } while_each_thread(g, q);
367
Nick Pigginf3af38d2006-12-06 20:31:51 -0800368 __oom_kill_task(p, 1);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800369
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370 /*
371 * kill all processes that share the ->mm (i.e. all threads),
Nick Pigginf2a2a712006-12-06 20:31:52 -0800372 * but are in a different thread group. Don't let them have access
373 * to memory reserves though, otherwise we might deplete all memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 */
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800375 do_each_thread(g, q) {
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700376 if (q->mm == mm && !same_thread_group(q, p))
David Rientjes650a7c92007-04-23 21:36:13 -0700377 force_sig(SIGKILL, q);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800378 } while_each_thread(g, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379
Dave Peterson01315922006-04-18 22:20:44 -0700380 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381}
382
David Rientjes7213f502007-10-16 23:25:57 -0700383static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
David Rientjesfef1bdd2008-02-07 00:14:07 -0800384 unsigned long points, struct mem_cgroup *mem,
385 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 struct task_struct *c;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388
David Rientjes7213f502007-10-16 23:25:57 -0700389 if (printk_ratelimit()) {
390 printk(KERN_WARNING "%s invoked oom-killer: "
391 "gfp_mask=0x%x, order=%d, oomkilladj=%d\n",
392 current->comm, gfp_mask, order, current->oomkilladj);
David Rientjes75aa1992009-01-06 14:39:01 -0800393 task_lock(current);
394 cpuset_print_task_mems_allowed(current);
395 task_unlock(current);
David Rientjes7213f502007-10-16 23:25:57 -0700396 dump_stack();
Balbir Singhe2224322009-04-02 16:57:39 -0700397 mem_cgroup_print_oom_info(mem, current);
David Rientjes7213f502007-10-16 23:25:57 -0700398 show_mem();
David Rientjesfef1bdd2008-02-07 00:14:07 -0800399 if (sysctl_oom_dump_tasks)
400 dump_tasks(mem);
David Rientjes7213f502007-10-16 23:25:57 -0700401 }
402
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700403 /*
404 * If the task is already exiting, don't alarm the sysadmin or kill
405 * its children or threads, just set TIF_MEMDIE so it can die quickly
406 */
407 if (p->flags & PF_EXITING) {
Nick Pigginf3af38d2006-12-06 20:31:51 -0800408 __oom_kill_task(p, 0);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700409 return 0;
410 }
411
Nick Pigginf3af38d2006-12-06 20:31:51 -0800412 printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700413 message, task_pid_nr(p), p->comm, points);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800414
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 /* Try to kill a child first */
Matthias Kaehlcke7b1915a2007-10-18 23:39:58 -0700416 list_for_each_entry(c, &p->children, sibling) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 if (c->mm == p->mm)
418 continue;
Nick Pigginf3af38d2006-12-06 20:31:51 -0800419 if (!oom_kill_task(c))
Dave Peterson01315922006-04-18 22:20:44 -0700420 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421 }
Nick Pigginf3af38d2006-12-06 20:31:51 -0800422 return oom_kill_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423}
424
Balbir Singh00f0b822008-03-04 14:28:39 -0800425#ifdef CONFIG_CGROUP_MEM_RES_CTLR
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800426void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
427{
428 unsigned long points = 0;
429 struct task_struct *p;
430
Li Zefane115f2d2008-04-15 14:34:37 -0700431 read_lock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800432retry:
433 p = select_bad_process(&points, mem);
434 if (PTR_ERR(p) == -1UL)
435 goto out;
436
437 if (!p)
438 p = current;
439
David Rientjesfef1bdd2008-02-07 00:14:07 -0800440 if (oom_kill_process(p, gfp_mask, 0, points, mem,
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800441 "Memory cgroup out of memory"))
442 goto retry;
443out:
Li Zefane115f2d2008-04-15 14:34:37 -0700444 read_unlock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800445}
446#endif
447
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700448static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
449
450int register_oom_notifier(struct notifier_block *nb)
451{
452 return blocking_notifier_chain_register(&oom_notify_list, nb);
453}
454EXPORT_SYMBOL_GPL(register_oom_notifier);
455
456int unregister_oom_notifier(struct notifier_block *nb)
457{
458 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
459}
460EXPORT_SYMBOL_GPL(unregister_oom_notifier);
461
David Rientjes098d7f12007-10-16 23:25:55 -0700462/*
463 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
464 * if a parallel OOM killing is already taking place that includes a zone in
465 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
466 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700467int try_set_zone_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700468{
Mel Gormandd1a2392008-04-28 02:12:17 -0700469 struct zoneref *z;
470 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700471 int ret = 1;
472
David Rientjesc7d4cae2009-01-06 14:39:00 -0800473 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700474 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
475 if (zone_is_oom_locked(zone)) {
David Rientjes098d7f12007-10-16 23:25:55 -0700476 ret = 0;
477 goto out;
478 }
Mel Gormandd1a2392008-04-28 02:12:17 -0700479 }
David Rientjes098d7f12007-10-16 23:25:55 -0700480
Mel Gormandd1a2392008-04-28 02:12:17 -0700481 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
482 /*
David Rientjesc7d4cae2009-01-06 14:39:00 -0800483 * Lock each zone in the zonelist under zone_scan_lock so a
Mel Gormandd1a2392008-04-28 02:12:17 -0700484 * parallel invocation of try_set_zone_oom() doesn't succeed
485 * when it shouldn't.
486 */
487 zone_set_flag(zone, ZONE_OOM_LOCKED);
488 }
489
David Rientjes098d7f12007-10-16 23:25:55 -0700490out:
David Rientjesc7d4cae2009-01-06 14:39:00 -0800491 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700492 return ret;
493}
494
495/*
496 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
497 * allocation attempts with zonelists containing them may now recall the OOM
498 * killer, if necessary.
499 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700500void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700501{
Mel Gormandd1a2392008-04-28 02:12:17 -0700502 struct zoneref *z;
503 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700504
David Rientjesc7d4cae2009-01-06 14:39:00 -0800505 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700506 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
507 zone_clear_flag(zone, ZONE_OOM_LOCKED);
508 }
David Rientjesc7d4cae2009-01-06 14:39:00 -0800509 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700510}
511
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800512/*
513 * Must be called with tasklist_lock held for read.
514 */
515static void __out_of_memory(gfp_t gfp_mask, int order)
516{
David Rientjes184101b2009-05-06 16:02:55 -0700517 struct task_struct *p;
518 unsigned long points;
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800519
David Rientjes184101b2009-05-06 16:02:55 -0700520 if (sysctl_oom_kill_allocating_task)
521 if (!oom_kill_process(current, gfp_mask, order, 0, NULL,
522 "Out of memory (oom_kill_allocating_task)"))
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800523 return;
David Rientjes184101b2009-05-06 16:02:55 -0700524retry:
525 /*
526 * Rambo mode: Shoot down a process and hope it solves whatever
527 * issues we may have.
528 */
529 p = select_bad_process(&points, NULL);
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800530
David Rientjes184101b2009-05-06 16:02:55 -0700531 if (PTR_ERR(p) == -1UL)
532 return;
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800533
David Rientjes184101b2009-05-06 16:02:55 -0700534 /* Found nothing?!?! Either we hang forever, or we panic. */
535 if (!p) {
536 read_unlock(&tasklist_lock);
537 panic("Out of memory and no killable processes...\n");
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800538 }
David Rientjes184101b2009-05-06 16:02:55 -0700539
540 if (oom_kill_process(p, gfp_mask, order, points, NULL,
541 "Out of memory"))
542 goto retry;
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800543}
544
545/*
546 * pagefault handler calls into here because it is out of memory but
547 * doesn't know exactly how or why.
548 */
549void pagefault_out_of_memory(void)
550{
551 unsigned long freed = 0;
552
553 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
554 if (freed > 0)
555 /* Got some memory back in the last second. */
556 return;
557
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800558 /*
559 * If this is from memcg, oom-killer is already invoked.
560 * and not worth to go system-wide-oom.
561 */
562 if (mem_cgroup_oom_called(current))
563 goto rest_and_return;
564
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800565 if (sysctl_panic_on_oom)
566 panic("out of memory from page fault. panic_on_oom is selected.\n");
567
568 read_lock(&tasklist_lock);
569 __out_of_memory(0, 0); /* unknown gfp_mask and order */
570 read_unlock(&tasklist_lock);
571
572 /*
573 * Give "p" a good chance of killing itself before we
574 * retry to allocate memory.
575 */
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800576rest_and_return:
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800577 if (!test_thread_flag(TIF_MEMDIE))
578 schedule_timeout_uninterruptible(1);
579}
580
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581/**
Dave Peterson6937a252006-06-23 02:03:13 -0700582 * out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700583 * @zonelist: zonelist pointer
584 * @gfp_mask: memory allocation flags
585 * @order: amount of memory being requested as a power of 2
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 *
587 * If we run out of memory, we have the choice between either
588 * killing a random task (bad), letting the system crash (worse)
589 * OR try to be smart about which process to kill. Note that we
590 * don't have to be perfect here, we just have to be good.
591 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800592void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593{
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700594 unsigned long freed = 0;
David Rientjes70e24bd2007-10-16 23:25:53 -0700595 enum oom_constraint constraint;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700596
597 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
598 if (freed > 0)
599 /* Got some memory back in the last second. */
600 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601
Yasunori Goto2b744c02007-05-06 14:49:59 -0700602 if (sysctl_panic_on_oom == 2)
603 panic("out of memory. Compulsory panic_on_oom is selected.\n");
604
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800605 /*
606 * Check if there were limitations on the allocation (only relevant for
607 * NUMA) that may require different handling.
608 */
David Rientjes2b45ab32007-05-06 14:50:00 -0700609 constraint = constrained_alloc(zonelist, gfp_mask);
David Rientjes2b45ab32007-05-06 14:50:00 -0700610 read_lock(&tasklist_lock);
611
612 switch (constraint) {
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800613 case CONSTRAINT_MEMORY_POLICY:
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800614 oom_kill_process(current, gfp_mask, order, 0, NULL,
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800615 "No available memory (MPOL_BIND)");
616 break;
617
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800618 case CONSTRAINT_NONE:
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -0700619 if (sysctl_panic_on_oom)
620 panic("out of memory. panic_on_oom is selected\n");
David Rientjesfe071d72007-10-16 23:25:56 -0700621 /* Fall-through */
622 case CONSTRAINT_CPUSET:
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800623 __out_of_memory(gfp_mask, order);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800624 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 }
626
Andrew Morton140ffce2006-03-02 02:54:28 -0800627 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628
629 /*
630 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800631 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 */
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800633 if (!test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800634 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635}