blob: 40ba05061a4fbc8a43d3eb046a58fe756748cd06 [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{
58 unsigned long points, cpu_time, run_time, s;
Andrew Morton97c2c9b82006-04-18 22:20:38 -070059 struct mm_struct *mm;
60 struct task_struct *child;
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Andrew Morton97c2c9b82006-04-18 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 Morton97c2c9b82006-04-18 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 Morton97c2c9b82006-04-18 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 Morton97c2c9b82006-04-18 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
113 s = int_sqrt(cpu_time);
114 if (s)
115 points /= s;
116 s = int_sqrt(int_sqrt(run_time));
117 if (s)
118 points /= s;
119
120 /*
121 * Niced processes are most likely less important, so double
122 * their badness points.
123 */
124 if (task_nice(p) > 0)
125 points *= 2;
126
127 /*
128 * Superuser processes are usually more important, so we make it
129 * less likely that we kill those.
130 */
Eric Parisa2f29452008-11-11 22:02:54 +1100131 if (has_capability_noaudit(p, CAP_SYS_ADMIN) ||
132 has_capability_noaudit(p, CAP_SYS_RESOURCE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133 points /= 4;
134
135 /*
136 * We don't want to kill a process with direct hardware access.
137 * Not only could that mess up the hardware, but usually users
138 * tend to only have this flag set on applications they think
139 * of as important.
140 */
Eric Parisa2f29452008-11-11 22:02:54 +1100141 if (has_capability_noaudit(p, CAP_SYS_RAWIO))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 points /= 4;
143
144 /*
Nick Piggin7887a3d2006-09-25 23:31:29 -0700145 * If p's nodes don't overlap ours, it may still help to kill p
146 * because p may have allocated or otherwise mapped memory on
147 * this node before. However it will be less likely.
148 */
David Rientjesbbe373f2007-10-16 23:25:58 -0700149 if (!cpuset_mems_allowed_intersects(current, p))
Nick Piggin7887a3d2006-09-25 23:31:29 -0700150 points /= 8;
151
152 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 * Adjust the score by oomkilladj.
154 */
155 if (p->oomkilladj) {
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700156 if (p->oomkilladj > 0) {
157 if (!points)
158 points = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 points <<= p->oomkilladj;
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700160 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 points >>= -(p->oomkilladj);
162 }
163
164#ifdef DEBUG
David Rientjesa5e58a62007-07-31 00:37:32 -0700165 printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 p->pid, p->comm, points);
167#endif
168 return points;
169}
170
171/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800172 * Determine the type of allocation constraint.
173 */
David Rientjes70e24bd2007-10-16 23:25:53 -0700174static inline enum oom_constraint constrained_alloc(struct zonelist *zonelist,
175 gfp_t gfp_mask)
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800176{
177#ifdef CONFIG_NUMA
Mel Gorman54a6eb52008-04-28 02:12:16 -0700178 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700179 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700180 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Christoph Lameteree31af52007-10-16 01:25:30 -0700181 nodemask_t nodes = node_states[N_HIGH_MEMORY];
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800182
Mel Gorman54a6eb52008-04-28 02:12:16 -0700183 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
184 if (cpuset_zone_allowed_softwall(zone, gfp_mask))
185 node_clear(zone_to_nid(zone), nodes);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800186 else
187 return CONSTRAINT_CPUSET;
188
189 if (!nodes_empty(nodes))
190 return CONSTRAINT_MEMORY_POLICY;
191#endif
192
193 return CONSTRAINT_NONE;
194}
195
196/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197 * Simple selection loop. We chose the process with the highest
198 * number of 'points'. We expect the caller will lock the tasklist.
199 *
200 * (not docbooked, we don't want this one cluttering up the manual)
201 */
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800202static struct task_struct *select_bad_process(unsigned long *ppoints,
203 struct mem_cgroup *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 struct task_struct *g, *p;
206 struct task_struct *chosen = NULL;
207 struct timespec uptime;
Kurt Garloff9827b782006-02-20 18:27:51 -0800208 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209
210 do_posix_clock_monotonic_gettime(&uptime);
Paul Jacksona49335c2005-09-06 15:18:09 -0700211 do_each_thread(g, p) {
212 unsigned long points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700213
Oleg Nesterov28324d12006-09-29 02:01:12 -0700214 /*
215 * skip kernel threads and tasks which have already released
216 * their mm.
217 */
Nick Piggin5081dde2006-09-25 23:31:32 -0700218 if (!p->mm)
219 continue;
Oleg Nesterov28324d12006-09-29 02:01:12 -0700220 /* skip the init task */
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700221 if (is_global_init(p))
Paul Jacksona49335c2005-09-06 15:18:09 -0700222 continue;
David Rientjes4c4a2212008-02-07 00:14:06 -0800223 if (mem && !task_in_mem_cgroup(p, mem))
224 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700225
Paul Jacksona49335c2005-09-06 15:18:09 -0700226 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700227 * This task already has access to memory reserves and is
228 * being killed. Don't allow any other task access to the
229 * memory reserve.
230 *
231 * Note: this may have a chance of deadlock if it gets
232 * blocked waiting for another task which itself is waiting
233 * for memory. Is there a better alternative?
234 */
235 if (test_tsk_thread_flag(p, TIF_MEMDIE))
236 return ERR_PTR(-1UL);
237
238 /*
Dave Peterson6937a252006-06-23 02:03:13 -0700239 * This is in the process of releasing memory so wait for it
Paul Jacksona49335c2005-09-06 15:18:09 -0700240 * to finish before killing some other task by mistake.
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700241 *
242 * However, if p is the current task, we allow the 'kill' to
243 * go ahead if it is exiting: this will simply set TIF_MEMDIE,
244 * which will allow it to gain access to memory reserves in
245 * the process of exiting and releasing its resources.
Nick Pigginb78483a2006-09-29 02:01:14 -0700246 * Otherwise we could get an easy OOM deadlock.
Paul Jacksona49335c2005-09-06 15:18:09 -0700247 */
Nick Pigginb78483a2006-09-29 02:01:14 -0700248 if (p->flags & PF_EXITING) {
249 if (p != current)
250 return ERR_PTR(-1UL);
251
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700252 chosen = p;
253 *ppoints = ULONG_MAX;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700254 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700255
Nick Piggin4a3ede12006-09-25 23:31:30 -0700256 if (p->oomkilladj == OOM_DISABLE)
257 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
Li Zefan97d87c92008-04-28 02:13:35 -0700259 points = badness(p, uptime.tv_sec);
Kurt Garloff9827b782006-02-20 18:27:51 -0800260 if (points > *ppoints || !chosen) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700261 chosen = p;
Kurt Garloff9827b782006-02-20 18:27:51 -0800262 *ppoints = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 }
Paul Jacksona49335c2005-09-06 15:18:09 -0700264 } while_each_thread(g, p);
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700265
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 return chosen;
267}
268
269/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700270 * dump_tasks - dump current memory state of all system tasks
271 * @mem: target memory controller
272 *
David Rientjesfef1bdd2008-02-07 00:14:07 -0800273 * Dumps the current memory state of all system tasks, excluding kernel threads.
274 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
275 * score, and name.
276 *
277 * If the actual is non-NULL, only tasks that are a member of the mem_cgroup are
278 * shown.
279 *
280 * Call with tasklist_lock read-locked.
281 */
282static void dump_tasks(const struct mem_cgroup *mem)
283{
284 struct task_struct *g, *p;
285
286 printk(KERN_INFO "[ pid ] uid tgid total_vm rss cpu oom_adj "
287 "name\n");
288 do_each_thread(g, p) {
289 /*
290 * total_vm and rss sizes do not exist for tasks with a
291 * detached mm so there's no need to report them.
292 */
293 if (!p->mm)
294 continue;
295 if (mem && !task_in_mem_cgroup(p, mem))
296 continue;
David Rientjesb4416d22008-11-06 12:53:29 -0800297 if (!thread_group_leader(p))
298 continue;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800299
300 task_lock(p);
301 printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3d %3d %s\n",
David Howellsc69e8d92008-11-14 10:39:19 +1100302 p->pid, __task_cred(p)->uid, p->tgid,
303 p->mm->total_vm, get_mm_rss(p->mm), (int)task_cpu(p),
304 p->oomkilladj, p->comm);
David Rientjesfef1bdd2008-02-07 00:14:07 -0800305 task_unlock(p);
306 } while_each_thread(g, p);
307}
308
Randy Dunlap1b578df2008-03-19 17:00:42 -0700309/*
Ram Gupta5a291b982006-09-25 23:31:54 -0700310 * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO
311 * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO
312 * set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 */
Nick Pigginf3af38d2006-12-06 20:31:51 -0800314static void __oom_kill_task(struct task_struct *p, int verbose)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315{
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700316 if (is_global_init(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317 WARN_ON(1);
318 printk(KERN_WARNING "tried to kill init!\n");
319 return;
320 }
321
Oleg Nesterov01017a22006-09-29 02:01:13 -0700322 if (!p->mm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 WARN_ON(1);
324 printk(KERN_WARNING "tried to kill an mm-less task!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700325 return;
326 }
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700327
Nick Pigginf3af38d2006-12-06 20:31:51 -0800328 if (verbose)
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700329 printk(KERN_ERR "Killed process %d (%s)\n",
330 task_pid_nr(p), p->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331
332 /*
333 * We give our sacrificial lamb high priority and access to
334 * all the memory it needs. That way it should be able to
335 * exit() and clear out its resources quickly...
336 */
Peter Zijlstrafa717062008-01-25 21:08:27 +0100337 p->rt.time_slice = HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 set_tsk_thread_flag(p, TIF_MEMDIE);
339
340 force_sig(SIGKILL, p);
341}
342
Nick Pigginf3af38d2006-12-06 20:31:51 -0800343static int oom_kill_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344{
Dave Peterson01315922006-04-18 22:20:44 -0700345 struct mm_struct *mm;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700346 struct task_struct *g, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347
Dave Peterson01315922006-04-18 22:20:44 -0700348 mm = p->mm;
349
350 /* WARNING: mm may not be dereferenced since we did not obtain its
351 * value from get_task_mm(p). This is OK since all we need to do is
352 * compare mm to q->mm below.
353 *
354 * Furthermore, even if mm contains a non-NULL value, p->mm may
355 * change to NULL at any time since we do not hold task_lock(p).
356 * However, this is of no concern to us.
357 */
358
Oleg Nesterov01017a22006-09-29 02:01:13 -0700359 if (mm == NULL)
Dave Peterson01315922006-04-18 22:20:44 -0700360 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800362 /*
363 * Don't kill the process if any threads are set to OOM_DISABLE
364 */
365 do_each_thread(g, q) {
Ankita Garg35ae8342007-03-16 13:38:33 -0800366 if (q->mm == mm && q->oomkilladj == OOM_DISABLE)
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800367 return 1;
368 } while_each_thread(g, q);
369
Nick Pigginf3af38d2006-12-06 20:31:51 -0800370 __oom_kill_task(p, 1);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800371
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 /*
373 * kill all processes that share the ->mm (i.e. all threads),
Nick Pigginf2a2a712006-12-06 20:31:52 -0800374 * but are in a different thread group. Don't let them have access
375 * to memory reserves though, otherwise we might deplete all memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 */
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800377 do_each_thread(g, q) {
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700378 if (q->mm == mm && !same_thread_group(q, p))
David Rientjes650a7c92007-04-23 21:36:13 -0700379 force_sig(SIGKILL, q);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800380 } while_each_thread(g, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381
Dave Peterson01315922006-04-18 22:20:44 -0700382 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383}
384
David Rientjes7213f502007-10-16 23:25:57 -0700385static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
David Rientjesfef1bdd2008-02-07 00:14:07 -0800386 unsigned long points, struct mem_cgroup *mem,
387 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389 struct task_struct *c;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700390
David Rientjes7213f502007-10-16 23:25:57 -0700391 if (printk_ratelimit()) {
392 printk(KERN_WARNING "%s invoked oom-killer: "
393 "gfp_mask=0x%x, order=%d, oomkilladj=%d\n",
394 current->comm, gfp_mask, order, current->oomkilladj);
David Rientjes75aa1992009-01-06 14:39:01 -0800395 task_lock(current);
396 cpuset_print_task_mems_allowed(current);
397 task_unlock(current);
David Rientjes7213f502007-10-16 23:25:57 -0700398 dump_stack();
399 show_mem();
David Rientjesfef1bdd2008-02-07 00:14:07 -0800400 if (sysctl_oom_dump_tasks)
401 dump_tasks(mem);
David Rientjes7213f502007-10-16 23:25:57 -0700402 }
403
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700404 /*
405 * If the task is already exiting, don't alarm the sysadmin or kill
406 * its children or threads, just set TIF_MEMDIE so it can die quickly
407 */
408 if (p->flags & PF_EXITING) {
Nick Pigginf3af38d2006-12-06 20:31:51 -0800409 __oom_kill_task(p, 0);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700410 return 0;
411 }
412
Nick Pigginf3af38d2006-12-06 20:31:51 -0800413 printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700414 message, task_pid_nr(p), p->comm, points);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800415
Linus Torvalds1da177e2005-04-16 15:20:36 -0700416 /* Try to kill a child first */
Matthias Kaehlcke7b1915a2007-10-18 23:39:58 -0700417 list_for_each_entry(c, &p->children, sibling) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 if (c->mm == p->mm)
419 continue;
Nick Pigginf3af38d2006-12-06 20:31:51 -0800420 if (!oom_kill_task(c))
Dave Peterson01315922006-04-18 22:20:44 -0700421 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 }
Nick Pigginf3af38d2006-12-06 20:31:51 -0800423 return oom_kill_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
425
Balbir Singh00f0b822008-03-04 14:28:39 -0800426#ifdef CONFIG_CGROUP_MEM_RES_CTLR
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800427void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
428{
429 unsigned long points = 0;
430 struct task_struct *p;
431
Li Zefane115f2d2008-04-15 14:34:37 -0700432 read_lock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800433retry:
434 p = select_bad_process(&points, mem);
435 if (PTR_ERR(p) == -1UL)
436 goto out;
437
438 if (!p)
439 p = current;
440
David Rientjesfef1bdd2008-02-07 00:14:07 -0800441 if (oom_kill_process(p, gfp_mask, 0, points, mem,
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800442 "Memory cgroup out of memory"))
443 goto retry;
444out:
Li Zefane115f2d2008-04-15 14:34:37 -0700445 read_unlock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800446}
447#endif
448
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700449static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
450
451int register_oom_notifier(struct notifier_block *nb)
452{
453 return blocking_notifier_chain_register(&oom_notify_list, nb);
454}
455EXPORT_SYMBOL_GPL(register_oom_notifier);
456
457int unregister_oom_notifier(struct notifier_block *nb)
458{
459 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
460}
461EXPORT_SYMBOL_GPL(unregister_oom_notifier);
462
David Rientjes098d7f12007-10-16 23:25:55 -0700463/*
464 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
465 * if a parallel OOM killing is already taking place that includes a zone in
466 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
467 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700468int try_set_zone_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700469{
Mel Gormandd1a2392008-04-28 02:12:17 -0700470 struct zoneref *z;
471 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700472 int ret = 1;
473
David Rientjesc7d4cae2009-01-06 14:39:00 -0800474 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700475 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
476 if (zone_is_oom_locked(zone)) {
David Rientjes098d7f12007-10-16 23:25:55 -0700477 ret = 0;
478 goto out;
479 }
Mel Gormandd1a2392008-04-28 02:12:17 -0700480 }
David Rientjes098d7f12007-10-16 23:25:55 -0700481
Mel Gormandd1a2392008-04-28 02:12:17 -0700482 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
483 /*
David Rientjesc7d4cae2009-01-06 14:39:00 -0800484 * Lock each zone in the zonelist under zone_scan_lock so a
Mel Gormandd1a2392008-04-28 02:12:17 -0700485 * parallel invocation of try_set_zone_oom() doesn't succeed
486 * when it shouldn't.
487 */
488 zone_set_flag(zone, ZONE_OOM_LOCKED);
489 }
490
David Rientjes098d7f12007-10-16 23:25:55 -0700491out:
David Rientjesc7d4cae2009-01-06 14:39:00 -0800492 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700493 return ret;
494}
495
496/*
497 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
498 * allocation attempts with zonelists containing them may now recall the OOM
499 * killer, if necessary.
500 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700501void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700502{
Mel Gormandd1a2392008-04-28 02:12:17 -0700503 struct zoneref *z;
504 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700505
David Rientjesc7d4cae2009-01-06 14:39:00 -0800506 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700507 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
508 zone_clear_flag(zone, ZONE_OOM_LOCKED);
509 }
David Rientjesc7d4cae2009-01-06 14:39:00 -0800510 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700511}
512
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800513/*
514 * Must be called with tasklist_lock held for read.
515 */
516static void __out_of_memory(gfp_t gfp_mask, int order)
517{
518 if (sysctl_oom_kill_allocating_task) {
519 oom_kill_process(current, gfp_mask, order, 0, NULL,
520 "Out of memory (oom_kill_allocating_task)");
521
522 } else {
523 unsigned long points;
524 struct task_struct *p;
525
526retry:
527 /*
528 * Rambo mode: Shoot down a process and hope it solves whatever
529 * issues we may have.
530 */
531 p = select_bad_process(&points, NULL);
532
533 if (PTR_ERR(p) == -1UL)
534 return;
535
536 /* Found nothing?!?! Either we hang forever, or we panic. */
537 if (!p) {
538 read_unlock(&tasklist_lock);
539 panic("Out of memory and no killable processes...\n");
540 }
541
542 if (oom_kill_process(p, gfp_mask, order, points, NULL,
543 "Out of memory"))
544 goto retry;
545 }
546}
547
548/*
549 * pagefault handler calls into here because it is out of memory but
550 * doesn't know exactly how or why.
551 */
552void pagefault_out_of_memory(void)
553{
554 unsigned long freed = 0;
555
556 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
557 if (freed > 0)
558 /* Got some memory back in the last second. */
559 return;
560
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800561 /*
562 * If this is from memcg, oom-killer is already invoked.
563 * and not worth to go system-wide-oom.
564 */
565 if (mem_cgroup_oom_called(current))
566 goto rest_and_return;
567
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800568 if (sysctl_panic_on_oom)
569 panic("out of memory from page fault. panic_on_oom is selected.\n");
570
571 read_lock(&tasklist_lock);
572 __out_of_memory(0, 0); /* unknown gfp_mask and order */
573 read_unlock(&tasklist_lock);
574
575 /*
576 * Give "p" a good chance of killing itself before we
577 * retry to allocate memory.
578 */
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800579rest_and_return:
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800580 if (!test_thread_flag(TIF_MEMDIE))
581 schedule_timeout_uninterruptible(1);
582}
583
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584/**
Dave Peterson6937a252006-06-23 02:03:13 -0700585 * out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700586 * @zonelist: zonelist pointer
587 * @gfp_mask: memory allocation flags
588 * @order: amount of memory being requested as a power of 2
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 *
590 * If we run out of memory, we have the choice between either
591 * killing a random task (bad), letting the system crash (worse)
592 * OR try to be smart about which process to kill. Note that we
593 * don't have to be perfect here, we just have to be good.
594 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800595void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596{
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700597 unsigned long freed = 0;
David Rientjes70e24bd2007-10-16 23:25:53 -0700598 enum oom_constraint constraint;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700599
600 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
601 if (freed > 0)
602 /* Got some memory back in the last second. */
603 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604
Yasunori Goto2b744c02007-05-06 14:49:59 -0700605 if (sysctl_panic_on_oom == 2)
606 panic("out of memory. Compulsory panic_on_oom is selected.\n");
607
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800608 /*
609 * Check if there were limitations on the allocation (only relevant for
610 * NUMA) that may require different handling.
611 */
David Rientjes2b45ab32007-05-06 14:50:00 -0700612 constraint = constrained_alloc(zonelist, gfp_mask);
David Rientjes2b45ab32007-05-06 14:50:00 -0700613 read_lock(&tasklist_lock);
614
615 switch (constraint) {
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800616 case CONSTRAINT_MEMORY_POLICY:
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800617 oom_kill_process(current, gfp_mask, order, 0, NULL,
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800618 "No available memory (MPOL_BIND)");
619 break;
620
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800621 case CONSTRAINT_NONE:
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -0700622 if (sysctl_panic_on_oom)
623 panic("out of memory. panic_on_oom is selected\n");
David Rientjesfe071d72007-10-16 23:25:56 -0700624 /* Fall-through */
625 case CONSTRAINT_CPUSET:
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800626 __out_of_memory(gfp_mask, order);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800627 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 }
629
Andrew Morton140ffce2006-03-02 02:54:28 -0800630 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
632 /*
633 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800634 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 */
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800636 if (!test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800637 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638}