blob: 64e5b4bcd964cd912b5a5f6b22dc392621269d84 [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 Rientjesae741382007-10-16 23:26:00 -070034static DEFINE_SPINLOCK(zone_scan_mutex);
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
Randy Dunlap1b578df2008-03-19 17:00:42 -070041 * @mem: target memory controller
Linus Torvalds1da177e2005-04-16 15:20:36 -070042 *
43 * The formula used is relatively simple and documented inline in the
44 * function. The main rationale is that we want to select a good task
45 * to kill when we run out of memory.
46 *
47 * Good in this context means that:
48 * 1) we lose the minimum amount of work done
49 * 2) we recover a large amount of memory
50 * 3) we don't kill anything innocent of eating tons of memory
51 * 4) we want to kill the minimum amount of processes (one)
52 * 5) we try to kill the process the user expects us to kill, this
53 * algorithm has been meticulously tuned to meet the principle
54 * of least surprise ... (be careful when you change it)
55 */
56
Li Zefan97d87c92008-04-28 02:13:35 -070057unsigned long badness(struct task_struct *p, unsigned long uptime)
Linus Torvalds1da177e2005-04-16 15:20:36 -070058{
59 unsigned long points, cpu_time, run_time, s;
Andrew Morton97c2c9b2006-04-18 22:20:38 -070060 struct mm_struct *mm;
61 struct task_struct *child;
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Andrew Morton97c2c9b2006-04-18 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 Morton97c2c9b2006-04-18 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 Morton97c2c9b2006-04-18 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 Morton97c2c9b2006-04-18 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 */
David Howells5cd9c582008-08-14 11:37:28 +0100132 if (has_capability(p, CAP_SYS_ADMIN) ||
133 has_capability(p, CAP_SYS_RESOURCE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134 points /= 4;
135
136 /*
137 * We don't want to kill a process with direct hardware access.
138 * Not only could that mess up the hardware, but usually users
139 * tend to only have this flag set on applications they think
140 * of as important.
141 */
David Howells5cd9c582008-08-14 11:37:28 +0100142 if (has_capability(p, CAP_SYS_RAWIO))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143 points /= 4;
144
145 /*
Nick Piggin7887a3d2006-09-25 23:31:29 -0700146 * If p's nodes don't overlap ours, it may still help to kill p
147 * because p may have allocated or otherwise mapped memory on
148 * this node before. However it will be less likely.
149 */
David Rientjesbbe373f2007-10-16 23:25:58 -0700150 if (!cpuset_mems_allowed_intersects(current, p))
Nick Piggin7887a3d2006-09-25 23:31:29 -0700151 points /= 8;
152
153 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 * Adjust the score by oomkilladj.
155 */
156 if (p->oomkilladj) {
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700157 if (p->oomkilladj > 0) {
158 if (!points)
159 points = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 points <<= p->oomkilladj;
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700161 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 points >>= -(p->oomkilladj);
163 }
164
165#ifdef DEBUG
David Rientjesa5e58a62007-07-31 00:37:32 -0700166 printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 p->pid, p->comm, points);
168#endif
169 return points;
170}
171
172/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800173 * Determine the type of allocation constraint.
174 */
David Rientjes70e24bd2007-10-16 23:25:53 -0700175static inline enum oom_constraint constrained_alloc(struct zonelist *zonelist,
176 gfp_t gfp_mask)
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800177{
178#ifdef CONFIG_NUMA
Mel Gorman54a6eb52008-04-28 02:12:16 -0700179 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700180 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700181 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Christoph Lameteree31af52007-10-16 01:25:30 -0700182 nodemask_t nodes = node_states[N_HIGH_MEMORY];
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800183
Mel Gorman54a6eb52008-04-28 02:12:16 -0700184 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
185 if (cpuset_zone_allowed_softwall(zone, gfp_mask))
186 node_clear(zone_to_nid(zone), nodes);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800187 else
188 return CONSTRAINT_CPUSET;
189
190 if (!nodes_empty(nodes))
191 return CONSTRAINT_MEMORY_POLICY;
192#endif
193
194 return CONSTRAINT_NONE;
195}
196
197/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 * Simple selection loop. We chose the process with the highest
199 * number of 'points'. We expect the caller will lock the tasklist.
200 *
201 * (not docbooked, we don't want this one cluttering up the manual)
202 */
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800203static struct task_struct *select_bad_process(unsigned long *ppoints,
204 struct mem_cgroup *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 struct task_struct *g, *p;
207 struct task_struct *chosen = NULL;
208 struct timespec uptime;
Kurt Garloff9827b782006-02-20 18:27:51 -0800209 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
211 do_posix_clock_monotonic_gettime(&uptime);
Paul Jacksona49335c2005-09-06 15:18:09 -0700212 do_each_thread(g, p) {
213 unsigned long points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700214
Oleg Nesterov28324d12006-09-29 02:01:12 -0700215 /*
216 * skip kernel threads and tasks which have already released
217 * their mm.
218 */
Nick Piggin5081dde2006-09-25 23:31:32 -0700219 if (!p->mm)
220 continue;
Oleg Nesterov28324d12006-09-29 02:01:12 -0700221 /* skip the init task */
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700222 if (is_global_init(p))
Paul Jacksona49335c2005-09-06 15:18:09 -0700223 continue;
David Rientjes4c4a2212008-02-07 00:14:06 -0800224 if (mem && !task_in_mem_cgroup(p, mem))
225 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700226
Paul Jacksona49335c2005-09-06 15:18:09 -0700227 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700228 * This task already has access to memory reserves and is
229 * being killed. Don't allow any other task access to the
230 * memory reserve.
231 *
232 * Note: this may have a chance of deadlock if it gets
233 * blocked waiting for another task which itself is waiting
234 * for memory. Is there a better alternative?
235 */
236 if (test_tsk_thread_flag(p, TIF_MEMDIE))
237 return ERR_PTR(-1UL);
238
239 /*
Dave Peterson6937a252006-06-23 02:03:13 -0700240 * This is in the process of releasing memory so wait for it
Paul Jacksona49335c2005-09-06 15:18:09 -0700241 * to finish before killing some other task by mistake.
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700242 *
243 * However, if p is the current task, we allow the 'kill' to
244 * go ahead if it is exiting: this will simply set TIF_MEMDIE,
245 * which will allow it to gain access to memory reserves in
246 * the process of exiting and releasing its resources.
Nick Pigginb78483a2006-09-29 02:01:14 -0700247 * Otherwise we could get an easy OOM deadlock.
Paul Jacksona49335c2005-09-06 15:18:09 -0700248 */
Nick Pigginb78483a2006-09-29 02:01:14 -0700249 if (p->flags & PF_EXITING) {
250 if (p != current)
251 return ERR_PTR(-1UL);
252
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700253 chosen = p;
254 *ppoints = ULONG_MAX;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700255 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700256
Nick Piggin4a3ede12006-09-25 23:31:30 -0700257 if (p->oomkilladj == OOM_DISABLE)
258 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259
Li Zefan97d87c92008-04-28 02:13:35 -0700260 points = badness(p, uptime.tv_sec);
Kurt Garloff9827b782006-02-20 18:27:51 -0800261 if (points > *ppoints || !chosen) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700262 chosen = p;
Kurt Garloff9827b782006-02-20 18:27:51 -0800263 *ppoints = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 }
Paul Jacksona49335c2005-09-06 15:18:09 -0700265 } while_each_thread(g, p);
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700266
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 return chosen;
268}
269
270/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700271 * dump_tasks - dump current memory state of all system tasks
272 * @mem: target memory controller
273 *
David Rientjesfef1bdd2008-02-07 00:14:07 -0800274 * Dumps the current memory state of all system tasks, excluding kernel threads.
275 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
276 * score, and name.
277 *
278 * If the actual is non-NULL, only tasks that are a member of the mem_cgroup are
279 * shown.
280 *
281 * Call with tasklist_lock read-locked.
282 */
283static void dump_tasks(const struct mem_cgroup *mem)
284{
285 struct task_struct *g, *p;
286
287 printk(KERN_INFO "[ pid ] uid tgid total_vm rss cpu oom_adj "
288 "name\n");
289 do_each_thread(g, p) {
290 /*
291 * total_vm and rss sizes do not exist for tasks with a
292 * detached mm so there's no need to report them.
293 */
294 if (!p->mm)
295 continue;
296 if (mem && !task_in_mem_cgroup(p, mem))
297 continue;
298
299 task_lock(p);
300 printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3d %3d %s\n",
301 p->pid, p->uid, p->tgid, p->mm->total_vm,
302 get_mm_rss(p->mm), (int)task_cpu(p), p->oomkilladj,
303 p->comm);
304 task_unlock(p);
305 } while_each_thread(g, p);
306}
307
Randy Dunlap1b578df2008-03-19 17:00:42 -0700308/*
Ram Gupta5a291b982006-09-25 23:31:54 -0700309 * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO
310 * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO
311 * set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 */
Nick Pigginf3af38d2006-12-06 20:31:51 -0800313static void __oom_kill_task(struct task_struct *p, int verbose)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314{
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700315 if (is_global_init(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 WARN_ON(1);
317 printk(KERN_WARNING "tried to kill init!\n");
318 return;
319 }
320
Oleg Nesterov01017a22006-09-29 02:01:13 -0700321 if (!p->mm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 WARN_ON(1);
323 printk(KERN_WARNING "tried to kill an mm-less task!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 return;
325 }
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700326
Nick Pigginf3af38d2006-12-06 20:31:51 -0800327 if (verbose)
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700328 printk(KERN_ERR "Killed process %d (%s)\n",
329 task_pid_nr(p), p->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700330
331 /*
332 * We give our sacrificial lamb high priority and access to
333 * all the memory it needs. That way it should be able to
334 * exit() and clear out its resources quickly...
335 */
Peter Zijlstrafa717062008-01-25 21:08:27 +0100336 p->rt.time_slice = HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 set_tsk_thread_flag(p, TIF_MEMDIE);
338
339 force_sig(SIGKILL, p);
340}
341
Nick Pigginf3af38d2006-12-06 20:31:51 -0800342static int oom_kill_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343{
Dave Peterson01315922006-04-18 22:20:44 -0700344 struct mm_struct *mm;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700345 struct task_struct *g, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346
Dave Peterson01315922006-04-18 22:20:44 -0700347 mm = p->mm;
348
349 /* WARNING: mm may not be dereferenced since we did not obtain its
350 * value from get_task_mm(p). This is OK since all we need to do is
351 * compare mm to q->mm below.
352 *
353 * Furthermore, even if mm contains a non-NULL value, p->mm may
354 * change to NULL at any time since we do not hold task_lock(p).
355 * However, this is of no concern to us.
356 */
357
Oleg Nesterov01017a22006-09-29 02:01:13 -0700358 if (mm == NULL)
Dave Peterson01315922006-04-18 22:20:44 -0700359 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800361 /*
362 * Don't kill the process if any threads are set to OOM_DISABLE
363 */
364 do_each_thread(g, q) {
Ankita Garg35ae8342007-03-16 13:38:33 -0800365 if (q->mm == mm && q->oomkilladj == OOM_DISABLE)
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800366 return 1;
367 } while_each_thread(g, q);
368
Nick Pigginf3af38d2006-12-06 20:31:51 -0800369 __oom_kill_task(p, 1);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800370
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 /*
372 * kill all processes that share the ->mm (i.e. all threads),
Nick Pigginf2a2a712006-12-06 20:31:52 -0800373 * but are in a different thread group. Don't let them have access
374 * to memory reserves though, otherwise we might deplete all memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 */
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800376 do_each_thread(g, q) {
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700377 if (q->mm == mm && !same_thread_group(q, p))
David Rientjes650a7c92007-04-23 21:36:13 -0700378 force_sig(SIGKILL, q);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800379 } while_each_thread(g, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380
Dave Peterson01315922006-04-18 22:20:44 -0700381 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382}
383
David Rientjes7213f502007-10-16 23:25:57 -0700384static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
David Rientjesfef1bdd2008-02-07 00:14:07 -0800385 unsigned long points, struct mem_cgroup *mem,
386 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 struct task_struct *c;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
David Rientjes7213f502007-10-16 23:25:57 -0700390 if (printk_ratelimit()) {
391 printk(KERN_WARNING "%s invoked oom-killer: "
392 "gfp_mask=0x%x, order=%d, oomkilladj=%d\n",
393 current->comm, gfp_mask, order, current->oomkilladj);
394 dump_stack();
395 show_mem();
David Rientjesfef1bdd2008-02-07 00:14:07 -0800396 if (sysctl_oom_dump_tasks)
397 dump_tasks(mem);
David Rientjes7213f502007-10-16 23:25:57 -0700398 }
399
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700400 /*
401 * If the task is already exiting, don't alarm the sysadmin or kill
402 * its children or threads, just set TIF_MEMDIE so it can die quickly
403 */
404 if (p->flags & PF_EXITING) {
Nick Pigginf3af38d2006-12-06 20:31:51 -0800405 __oom_kill_task(p, 0);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700406 return 0;
407 }
408
Nick Pigginf3af38d2006-12-06 20:31:51 -0800409 printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700410 message, task_pid_nr(p), p->comm, points);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800411
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 /* Try to kill a child first */
Matthias Kaehlcke7b1915a2007-10-18 23:39:58 -0700413 list_for_each_entry(c, &p->children, sibling) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 if (c->mm == p->mm)
415 continue;
Nick Pigginf3af38d2006-12-06 20:31:51 -0800416 if (!oom_kill_task(c))
Dave Peterson01315922006-04-18 22:20:44 -0700417 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 }
Nick Pigginf3af38d2006-12-06 20:31:51 -0800419 return oom_kill_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420}
421
Balbir Singh00f0b822008-03-04 14:28:39 -0800422#ifdef CONFIG_CGROUP_MEM_RES_CTLR
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800423void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
424{
425 unsigned long points = 0;
426 struct task_struct *p;
427
428 cgroup_lock();
Li Zefane115f2d2008-04-15 14:34:37 -0700429 read_lock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800430retry:
431 p = select_bad_process(&points, mem);
432 if (PTR_ERR(p) == -1UL)
433 goto out;
434
435 if (!p)
436 p = current;
437
David Rientjesfef1bdd2008-02-07 00:14:07 -0800438 if (oom_kill_process(p, gfp_mask, 0, points, mem,
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800439 "Memory cgroup out of memory"))
440 goto retry;
441out:
Li Zefane115f2d2008-04-15 14:34:37 -0700442 read_unlock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800443 cgroup_unlock();
444}
445#endif
446
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700447static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
448
449int register_oom_notifier(struct notifier_block *nb)
450{
451 return blocking_notifier_chain_register(&oom_notify_list, nb);
452}
453EXPORT_SYMBOL_GPL(register_oom_notifier);
454
455int unregister_oom_notifier(struct notifier_block *nb)
456{
457 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
458}
459EXPORT_SYMBOL_GPL(unregister_oom_notifier);
460
David Rientjes098d7f12007-10-16 23:25:55 -0700461/*
462 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
463 * if a parallel OOM killing is already taking place that includes a zone in
464 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
465 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700466int try_set_zone_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700467{
Mel Gormandd1a2392008-04-28 02:12:17 -0700468 struct zoneref *z;
469 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700470 int ret = 1;
471
David Rientjesae741382007-10-16 23:26:00 -0700472 spin_lock(&zone_scan_mutex);
Mel Gormandd1a2392008-04-28 02:12:17 -0700473 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
474 if (zone_is_oom_locked(zone)) {
David Rientjes098d7f12007-10-16 23:25:55 -0700475 ret = 0;
476 goto out;
477 }
Mel Gormandd1a2392008-04-28 02:12:17 -0700478 }
David Rientjes098d7f12007-10-16 23:25:55 -0700479
Mel Gormandd1a2392008-04-28 02:12:17 -0700480 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
481 /*
482 * Lock each zone in the zonelist under zone_scan_mutex so a
483 * parallel invocation of try_set_zone_oom() doesn't succeed
484 * when it shouldn't.
485 */
486 zone_set_flag(zone, ZONE_OOM_LOCKED);
487 }
488
David Rientjes098d7f12007-10-16 23:25:55 -0700489out:
David Rientjesae741382007-10-16 23:26:00 -0700490 spin_unlock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700491 return ret;
492}
493
494/*
495 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
496 * allocation attempts with zonelists containing them may now recall the OOM
497 * killer, if necessary.
498 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700499void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700500{
Mel Gormandd1a2392008-04-28 02:12:17 -0700501 struct zoneref *z;
502 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700503
David Rientjesae741382007-10-16 23:26:00 -0700504 spin_lock(&zone_scan_mutex);
Mel Gormandd1a2392008-04-28 02:12:17 -0700505 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
506 zone_clear_flag(zone, ZONE_OOM_LOCKED);
507 }
David Rientjesae741382007-10-16 23:26:00 -0700508 spin_unlock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700509}
510
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511/**
Dave Peterson6937a252006-06-23 02:03:13 -0700512 * out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700513 * @zonelist: zonelist pointer
514 * @gfp_mask: memory allocation flags
515 * @order: amount of memory being requested as a power of 2
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516 *
517 * If we run out of memory, we have the choice between either
518 * killing a random task (bad), letting the system crash (worse)
519 * OR try to be smart about which process to kill. Note that we
520 * don't have to be perfect here, we just have to be good.
521 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800522void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523{
Ingo Molnar36c8b582006-07-03 00:25:41 -0700524 struct task_struct *p;
Andrew Mortond6713e02006-02-28 16:59:19 -0800525 unsigned long points = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700526 unsigned long freed = 0;
David Rientjes70e24bd2007-10-16 23:25:53 -0700527 enum oom_constraint constraint;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700528
529 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
530 if (freed > 0)
531 /* Got some memory back in the last second. */
532 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
Yasunori Goto2b744c02007-05-06 14:49:59 -0700534 if (sysctl_panic_on_oom == 2)
535 panic("out of memory. Compulsory panic_on_oom is selected.\n");
536
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800537 /*
538 * Check if there were limitations on the allocation (only relevant for
539 * NUMA) that may require different handling.
540 */
David Rientjes2b45ab32007-05-06 14:50:00 -0700541 constraint = constrained_alloc(zonelist, gfp_mask);
David Rientjes2b45ab32007-05-06 14:50:00 -0700542 read_lock(&tasklist_lock);
543
544 switch (constraint) {
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800545 case CONSTRAINT_MEMORY_POLICY:
David Rientjesfef1bdd2008-02-07 00:14:07 -0800546 oom_kill_process(current, gfp_mask, order, points, NULL,
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800547 "No available memory (MPOL_BIND)");
548 break;
549
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800550 case CONSTRAINT_NONE:
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -0700551 if (sysctl_panic_on_oom)
552 panic("out of memory. panic_on_oom is selected\n");
David Rientjesfe071d72007-10-16 23:25:56 -0700553 /* Fall-through */
554 case CONSTRAINT_CPUSET:
555 if (sysctl_oom_kill_allocating_task) {
David Rientjesfef1bdd2008-02-07 00:14:07 -0800556 oom_kill_process(current, gfp_mask, order, points, NULL,
David Rientjesfe071d72007-10-16 23:25:56 -0700557 "Out of memory (oom_kill_allocating_task)");
558 break;
559 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560retry:
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800561 /*
562 * Rambo mode: Shoot down a process and hope it solves whatever
563 * issues we may have.
564 */
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800565 p = select_bad_process(&points, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800567 if (PTR_ERR(p) == -1UL)
568 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800570 /* Found nothing?!?! Either we hang forever, or we panic. */
571 if (!p) {
572 read_unlock(&tasklist_lock);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800573 panic("Out of memory and no killable processes...\n");
574 }
575
David Rientjesfef1bdd2008-02-07 00:14:07 -0800576 if (oom_kill_process(p, gfp_mask, order, points, NULL,
David Rientjes7213f502007-10-16 23:25:57 -0700577 "Out of memory"))
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800578 goto retry;
579
580 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 }
582
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800583out:
Andrew Morton140ffce2006-03-02 02:54:28 -0800584 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
586 /*
587 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800588 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 */
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800590 if (!test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800591 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592}