blob: beb592fe9389ffa9fe34c8c92ca82ada55abd960 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -070030int sysctl_panic_on_oom;
David Rientjesfe071d72007-10-16 23:25:56 -070031int sysctl_oom_kill_allocating_task;
David Rientjesfef1bdd2008-02-07 00:14:07 -080032int sysctl_oom_dump_tasks;
David Rientjesae741382007-10-16 23:26:00 -070033static DEFINE_SPINLOCK(zone_scan_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070034/* #define DEBUG */
35
36/**
Dave Peterson6937a252006-06-23 02:03:13 -070037 * badness - calculate a numeric value for how bad this task has been
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 * @p: task struct of which task we should calculate
Paul Jacksona49335c2005-09-06 15:18:09 -070039 * @uptime: current uptime in seconds
Randy Dunlap1b578df2008-03-19 17:00:42 -070040 * @mem: target memory controller
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
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -080056unsigned long badness(struct task_struct *p, unsigned long uptime,
57 struct mem_cgroup *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -070058{
59 unsigned long points, cpu_time, run_time, s;
Andrew Morton97c2c9b82006-04-18 22:20:38 -070060 struct mm_struct *mm;
61 struct task_struct *child;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Andrew Morton97c2c9b82006-04-18 22:20:38 -070063 task_lock(p);
64 mm = p->mm;
65 if (!mm) {
66 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070067 return 0;
Andrew Morton97c2c9b82006-04-18 22:20:38 -070068 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070069
70 /*
71 * The memory size of the process is the basis for the badness.
72 */
Andrew Morton97c2c9b82006-04-18 22:20:38 -070073 points = mm->total_vm;
74
75 /*
76 * After this unlock we can no longer dereference local variable `mm'
77 */
78 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070079
80 /*
Hugh Dickins7ba34852007-01-05 16:37:03 -080081 * swapoff can easily use up all memory, so kill those first.
82 */
83 if (p->flags & PF_SWAPOFF)
84 return ULONG_MAX;
85
86 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 * Processes which fork a lot of child processes are likely
Kurt Garloff9827b782006-02-20 18:27:51 -080088 * a good choice. We add half the vmsize of the children if they
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 * have an own mm. This prevents forking servers to flood the
Kurt Garloff9827b782006-02-20 18:27:51 -080090 * machine with an endless amount of children. In case a single
91 * child is eating the vast majority of memory, adding only half
92 * to the parents will make the child our kill candidate of choice.
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 */
Andrew Morton97c2c9b82006-04-18 22:20:38 -070094 list_for_each_entry(child, &p->children, sibling) {
95 task_lock(child);
96 if (child->mm != mm && child->mm)
97 points += child->mm->total_vm/2 + 1;
98 task_unlock(child);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 }
100
101 /*
102 * CPU time is in tens of seconds and run time is in thousands
103 * of seconds. There is no particular reason for this other than
104 * that it turned out to work very well in practice.
105 */
106 cpu_time = (cputime_to_jiffies(p->utime) + cputime_to_jiffies(p->stime))
107 >> (SHIFT_HZ + 3);
108
109 if (uptime >= p->start_time.tv_sec)
110 run_time = (uptime - p->start_time.tv_sec) >> 10;
111 else
112 run_time = 0;
113
114 s = int_sqrt(cpu_time);
115 if (s)
116 points /= s;
117 s = int_sqrt(int_sqrt(run_time));
118 if (s)
119 points /= s;
120
121 /*
122 * Niced processes are most likely less important, so double
123 * their badness points.
124 */
125 if (task_nice(p) > 0)
126 points *= 2;
127
128 /*
129 * Superuser processes are usually more important, so we make it
130 * less likely that we kill those.
131 */
Serge E. Hallyn97829952008-02-04 22:29:47 -0800132 if (__capable(p, CAP_SYS_ADMIN) || __capable(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 */
Andrew Morgane338d262008-02-04 22:29:42 -0800141 if (__capable(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
178 struct zone **z;
Christoph Lameteree31af52007-10-16 01:25:30 -0700179 nodemask_t nodes = node_states[N_HIGH_MEMORY];
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800180
181 for (z = zonelist->zones; *z; z++)
Paul Jackson02a0e532006-12-13 00:34:25 -0800182 if (cpuset_zone_allowed_softwall(*z, gfp_mask))
Christoph Lameter89fa3022006-09-25 23:31:55 -0700183 node_clear(zone_to_nid(*z), 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
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800257 points = badness(p, uptime.tv_sec, mem);
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;
295
296 task_lock(p);
297 printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3d %3d %s\n",
298 p->pid, p->uid, p->tgid, p->mm->total_vm,
299 get_mm_rss(p->mm), (int)task_cpu(p), p->oomkilladj,
300 p->comm);
301 task_unlock(p);
302 } while_each_thread(g, p);
303}
304
Randy Dunlap1b578df2008-03-19 17:00:42 -0700305/*
Ram Gupta5a291b982006-09-25 23:31:54 -0700306 * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO
307 * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO
308 * set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 */
Nick Pigginf3af38d2006-12-06 20:31:51 -0800310static void __oom_kill_task(struct task_struct *p, int verbose)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311{
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700312 if (is_global_init(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 WARN_ON(1);
314 printk(KERN_WARNING "tried to kill init!\n");
315 return;
316 }
317
Oleg Nesterov01017a22006-09-29 02:01:13 -0700318 if (!p->mm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 WARN_ON(1);
320 printk(KERN_WARNING "tried to kill an mm-less task!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321 return;
322 }
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700323
Nick Pigginf3af38d2006-12-06 20:31:51 -0800324 if (verbose)
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700325 printk(KERN_ERR "Killed process %d (%s)\n",
326 task_pid_nr(p), p->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327
328 /*
329 * We give our sacrificial lamb high priority and access to
330 * all the memory it needs. That way it should be able to
331 * exit() and clear out its resources quickly...
332 */
Peter Zijlstrafa717062008-01-25 21:08:27 +0100333 p->rt.time_slice = HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 set_tsk_thread_flag(p, TIF_MEMDIE);
335
336 force_sig(SIGKILL, p);
337}
338
Nick Pigginf3af38d2006-12-06 20:31:51 -0800339static int oom_kill_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340{
Dave Peterson01315922006-04-18 22:20:44 -0700341 struct mm_struct *mm;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700342 struct task_struct *g, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343
Dave Peterson01315922006-04-18 22:20:44 -0700344 mm = p->mm;
345
346 /* WARNING: mm may not be dereferenced since we did not obtain its
347 * value from get_task_mm(p). This is OK since all we need to do is
348 * compare mm to q->mm below.
349 *
350 * Furthermore, even if mm contains a non-NULL value, p->mm may
351 * change to NULL at any time since we do not hold task_lock(p).
352 * However, this is of no concern to us.
353 */
354
Oleg Nesterov01017a22006-09-29 02:01:13 -0700355 if (mm == NULL)
Dave Peterson01315922006-04-18 22:20:44 -0700356 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800358 /*
359 * Don't kill the process if any threads are set to OOM_DISABLE
360 */
361 do_each_thread(g, q) {
Ankita Garg35ae8342007-03-16 13:38:33 -0800362 if (q->mm == mm && q->oomkilladj == OOM_DISABLE)
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800363 return 1;
364 } while_each_thread(g, q);
365
Nick Pigginf3af38d2006-12-06 20:31:51 -0800366 __oom_kill_task(p, 1);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 /*
369 * kill all processes that share the ->mm (i.e. all threads),
Nick Pigginf2a2a712006-12-06 20:31:52 -0800370 * but are in a different thread group. Don't let them have access
371 * to memory reserves though, otherwise we might deplete all memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 */
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800373 do_each_thread(g, q) {
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700374 if (q->mm == mm && !same_thread_group(q, p))
David Rientjes650a7c92007-04-23 21:36:13 -0700375 force_sig(SIGKILL, q);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800376 } while_each_thread(g, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377
Dave Peterson01315922006-04-18 22:20:44 -0700378 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379}
380
David Rientjes7213f502007-10-16 23:25:57 -0700381static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
David Rientjesfef1bdd2008-02-07 00:14:07 -0800382 unsigned long points, struct mem_cgroup *mem,
383 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 struct task_struct *c;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
David Rientjes7213f502007-10-16 23:25:57 -0700387 if (printk_ratelimit()) {
388 printk(KERN_WARNING "%s invoked oom-killer: "
389 "gfp_mask=0x%x, order=%d, oomkilladj=%d\n",
390 current->comm, gfp_mask, order, current->oomkilladj);
391 dump_stack();
392 show_mem();
David Rientjesfef1bdd2008-02-07 00:14:07 -0800393 if (sysctl_oom_dump_tasks)
394 dump_tasks(mem);
David Rientjes7213f502007-10-16 23:25:57 -0700395 }
396
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700397 /*
398 * If the task is already exiting, don't alarm the sysadmin or kill
399 * its children or threads, just set TIF_MEMDIE so it can die quickly
400 */
401 if (p->flags & PF_EXITING) {
Nick Pigginf3af38d2006-12-06 20:31:51 -0800402 __oom_kill_task(p, 0);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700403 return 0;
404 }
405
Nick Pigginf3af38d2006-12-06 20:31:51 -0800406 printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700407 message, task_pid_nr(p), p->comm, points);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800408
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 /* Try to kill a child first */
Matthias Kaehlcke7b1915a2007-10-18 23:39:58 -0700410 list_for_each_entry(c, &p->children, sibling) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 if (c->mm == p->mm)
412 continue;
Nick Pigginf3af38d2006-12-06 20:31:51 -0800413 if (!oom_kill_task(c))
Dave Peterson01315922006-04-18 22:20:44 -0700414 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 }
Nick Pigginf3af38d2006-12-06 20:31:51 -0800416 return oom_kill_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417}
418
Balbir Singh00f0b822008-03-04 14:28:39 -0800419#ifdef CONFIG_CGROUP_MEM_RES_CTLR
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800420void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
421{
422 unsigned long points = 0;
423 struct task_struct *p;
424
425 cgroup_lock();
Li Zefane115f2d2008-04-15 14:34:37 -0700426 read_lock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800427retry:
428 p = select_bad_process(&points, mem);
429 if (PTR_ERR(p) == -1UL)
430 goto out;
431
432 if (!p)
433 p = current;
434
David Rientjesfef1bdd2008-02-07 00:14:07 -0800435 if (oom_kill_process(p, gfp_mask, 0, points, mem,
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800436 "Memory cgroup out of memory"))
437 goto retry;
438out:
Li Zefane115f2d2008-04-15 14:34:37 -0700439 read_unlock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800440 cgroup_unlock();
441}
442#endif
443
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700444static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
445
446int register_oom_notifier(struct notifier_block *nb)
447{
448 return blocking_notifier_chain_register(&oom_notify_list, nb);
449}
450EXPORT_SYMBOL_GPL(register_oom_notifier);
451
452int unregister_oom_notifier(struct notifier_block *nb)
453{
454 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
455}
456EXPORT_SYMBOL_GPL(unregister_oom_notifier);
457
David Rientjes098d7f12007-10-16 23:25:55 -0700458/*
459 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
460 * if a parallel OOM killing is already taking place that includes a zone in
461 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
462 */
463int try_set_zone_oom(struct zonelist *zonelist)
464{
465 struct zone **z;
466 int ret = 1;
467
468 z = zonelist->zones;
469
David Rientjesae741382007-10-16 23:26:00 -0700470 spin_lock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700471 do {
472 if (zone_is_oom_locked(*z)) {
473 ret = 0;
474 goto out;
475 }
476 } while (*(++z) != NULL);
477
478 /*
479 * Lock each zone in the zonelist under zone_scan_mutex so a parallel
480 * invocation of try_set_zone_oom() doesn't succeed when it shouldn't.
481 */
482 z = zonelist->zones;
483 do {
484 zone_set_flag(*z, ZONE_OOM_LOCKED);
485 } while (*(++z) != NULL);
486out:
David Rientjesae741382007-10-16 23:26:00 -0700487 spin_unlock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700488 return ret;
489}
490
491/*
492 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
493 * allocation attempts with zonelists containing them may now recall the OOM
494 * killer, if necessary.
495 */
496void clear_zonelist_oom(struct zonelist *zonelist)
497{
498 struct zone **z;
499
500 z = zonelist->zones;
501
David Rientjesae741382007-10-16 23:26:00 -0700502 spin_lock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700503 do {
504 zone_clear_flag(*z, ZONE_OOM_LOCKED);
505 } while (*(++z) != NULL);
David Rientjesae741382007-10-16 23:26:00 -0700506 spin_unlock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700507}
508
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509/**
Dave Peterson6937a252006-06-23 02:03:13 -0700510 * out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700511 * @zonelist: zonelist pointer
512 * @gfp_mask: memory allocation flags
513 * @order: amount of memory being requested as a power of 2
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514 *
515 * If we run out of memory, we have the choice between either
516 * killing a random task (bad), letting the system crash (worse)
517 * OR try to be smart about which process to kill. Note that we
518 * don't have to be perfect here, we just have to be good.
519 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800520void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700521{
Ingo Molnar36c8b582006-07-03 00:25:41 -0700522 struct task_struct *p;
Andrew Mortond6713e02006-02-28 16:59:19 -0800523 unsigned long points = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700524 unsigned long freed = 0;
David Rientjes70e24bd2007-10-16 23:25:53 -0700525 enum oom_constraint constraint;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700526
527 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
528 if (freed > 0)
529 /* Got some memory back in the last second. */
530 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
Yasunori Goto2b744c02007-05-06 14:49:59 -0700532 if (sysctl_panic_on_oom == 2)
533 panic("out of memory. Compulsory panic_on_oom is selected.\n");
534
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800535 /*
536 * Check if there were limitations on the allocation (only relevant for
537 * NUMA) that may require different handling.
538 */
David Rientjes2b45ab32007-05-06 14:50:00 -0700539 constraint = constrained_alloc(zonelist, gfp_mask);
David Rientjes2b45ab32007-05-06 14:50:00 -0700540 read_lock(&tasklist_lock);
541
542 switch (constraint) {
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800543 case CONSTRAINT_MEMORY_POLICY:
David Rientjesfef1bdd2008-02-07 00:14:07 -0800544 oom_kill_process(current, gfp_mask, order, points, NULL,
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800545 "No available memory (MPOL_BIND)");
546 break;
547
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800548 case CONSTRAINT_NONE:
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -0700549 if (sysctl_panic_on_oom)
550 panic("out of memory. panic_on_oom is selected\n");
David Rientjesfe071d72007-10-16 23:25:56 -0700551 /* Fall-through */
552 case CONSTRAINT_CPUSET:
553 if (sysctl_oom_kill_allocating_task) {
David Rientjesfef1bdd2008-02-07 00:14:07 -0800554 oom_kill_process(current, gfp_mask, order, points, NULL,
David Rientjesfe071d72007-10-16 23:25:56 -0700555 "Out of memory (oom_kill_allocating_task)");
556 break;
557 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558retry:
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800559 /*
560 * Rambo mode: Shoot down a process and hope it solves whatever
561 * issues we may have.
562 */
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800563 p = select_bad_process(&points, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800565 if (PTR_ERR(p) == -1UL)
566 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800568 /* Found nothing?!?! Either we hang forever, or we panic. */
569 if (!p) {
570 read_unlock(&tasklist_lock);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800571 panic("Out of memory and no killable processes...\n");
572 }
573
David Rientjesfef1bdd2008-02-07 00:14:07 -0800574 if (oom_kill_process(p, gfp_mask, order, points, NULL,
David Rientjes7213f502007-10-16 23:25:57 -0700575 "Out of memory"))
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800576 goto retry;
577
578 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 }
580
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800581out:
Andrew Morton140ffce2006-03-02 02:54:28 -0800582 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583
584 /*
585 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800586 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 */
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800588 if (!test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800589 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590}