blob: 175a67a78a99e7b0368dfba11edf9888bc049674 [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;
David Rientjes2ff05b22009-06-16 15:32:56 -070061 int oom_adj;
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 }
David Rientjes2ff05b22009-06-16 15:32:56 -070069 oom_adj = mm->oom_adj;
David Rientjes4d8b9132009-06-16 15:32:57 -070070 if (oom_adj == OOM_DISABLE) {
71 task_unlock(p);
72 return 0;
73 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070074
75 /*
76 * The memory size of the process is the basis for the badness.
77 */
Andrew Morton97c2c9b2006-04-18 22:20:38 -070078 points = mm->total_vm;
79
80 /*
81 * After this unlock we can no longer dereference local variable `mm'
82 */
83 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
85 /*
Hugh Dickins7ba34852007-01-05 16:37:03 -080086 * swapoff can easily use up all memory, so kill those first.
87 */
88 if (p->flags & PF_SWAPOFF)
89 return ULONG_MAX;
90
91 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 * Processes which fork a lot of child processes are likely
Kurt Garloff9827b782006-02-20 18:27:51 -080093 * a good choice. We add half the vmsize of the children if they
Linus Torvalds1da177e2005-04-16 15:20:36 -070094 * have an own mm. This prevents forking servers to flood the
Kurt Garloff9827b782006-02-20 18:27:51 -080095 * machine with an endless amount of children. In case a single
96 * child is eating the vast majority of memory, adding only half
97 * to the parents will make the child our kill candidate of choice.
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 */
Andrew Morton97c2c9b2006-04-18 22:20:38 -070099 list_for_each_entry(child, &p->children, sibling) {
100 task_lock(child);
101 if (child->mm != mm && child->mm)
102 points += child->mm->total_vm/2 + 1;
103 task_unlock(child);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700104 }
105
106 /*
107 * CPU time is in tens of seconds and run time is in thousands
108 * of seconds. There is no particular reason for this other than
109 * that it turned out to work very well in practice.
110 */
111 cpu_time = (cputime_to_jiffies(p->utime) + cputime_to_jiffies(p->stime))
112 >> (SHIFT_HZ + 3);
113
114 if (uptime >= p->start_time.tv_sec)
115 run_time = (uptime - p->start_time.tv_sec) >> 10;
116 else
117 run_time = 0;
118
Cyrill Gorcunova12888f2009-03-31 15:19:27 -0700119 if (cpu_time)
120 points /= int_sqrt(cpu_time);
121 if (run_time)
122 points /= int_sqrt(int_sqrt(run_time));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
124 /*
125 * Niced processes are most likely less important, so double
126 * their badness points.
127 */
128 if (task_nice(p) > 0)
129 points *= 2;
130
131 /*
132 * Superuser processes are usually more important, so we make it
133 * less likely that we kill those.
134 */
Eric Parisa2f29452008-11-11 22:02:54 +1100135 if (has_capability_noaudit(p, CAP_SYS_ADMIN) ||
136 has_capability_noaudit(p, CAP_SYS_RESOURCE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137 points /= 4;
138
139 /*
140 * We don't want to kill a process with direct hardware access.
141 * Not only could that mess up the hardware, but usually users
142 * tend to only have this flag set on applications they think
143 * of as important.
144 */
Eric Parisa2f29452008-11-11 22:02:54 +1100145 if (has_capability_noaudit(p, CAP_SYS_RAWIO))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 points /= 4;
147
148 /*
Nick Piggin7887a3d2006-09-25 23:31:29 -0700149 * If p's nodes don't overlap ours, it may still help to kill p
150 * because p may have allocated or otherwise mapped memory on
151 * this node before. However it will be less likely.
152 */
David Rientjesbbe373f2007-10-16 23:25:58 -0700153 if (!cpuset_mems_allowed_intersects(current, p))
Nick Piggin7887a3d2006-09-25 23:31:29 -0700154 points /= 8;
155
156 /*
David Rientjes2ff05b22009-06-16 15:32:56 -0700157 * Adjust the score by oom_adj.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 */
David Rientjes2ff05b22009-06-16 15:32:56 -0700159 if (oom_adj) {
160 if (oom_adj > 0) {
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700161 if (!points)
162 points = 1;
David Rientjes2ff05b22009-06-16 15:32:56 -0700163 points <<= oom_adj;
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700164 } else
David Rientjes2ff05b22009-06-16 15:32:56 -0700165 points >>= -(oom_adj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 }
167
168#ifdef DEBUG
David Rientjesa5e58a62007-07-31 00:37:32 -0700169 printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 p->pid, p->comm, points);
171#endif
172 return points;
173}
174
175/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800176 * Determine the type of allocation constraint.
177 */
David Rientjes70e24bd2007-10-16 23:25:53 -0700178static inline enum oom_constraint constrained_alloc(struct zonelist *zonelist,
179 gfp_t gfp_mask)
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800180{
181#ifdef CONFIG_NUMA
Mel Gorman54a6eb52008-04-28 02:12:16 -0700182 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700183 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700184 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
Christoph Lameteree31af52007-10-16 01:25:30 -0700185 nodemask_t nodes = node_states[N_HIGH_MEMORY];
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800186
Mel Gorman54a6eb52008-04-28 02:12:16 -0700187 for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
188 if (cpuset_zone_allowed_softwall(zone, gfp_mask))
189 node_clear(zone_to_nid(zone), nodes);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800190 else
191 return CONSTRAINT_CPUSET;
192
193 if (!nodes_empty(nodes))
194 return CONSTRAINT_MEMORY_POLICY;
195#endif
196
197 return CONSTRAINT_NONE;
198}
199
200/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 * Simple selection loop. We chose the process with the highest
202 * number of 'points'. We expect the caller will lock the tasklist.
203 *
204 * (not docbooked, we don't want this one cluttering up the manual)
205 */
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800206static struct task_struct *select_bad_process(unsigned long *ppoints,
207 struct mem_cgroup *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 struct task_struct *g, *p;
210 struct task_struct *chosen = NULL;
211 struct timespec uptime;
Kurt Garloff9827b782006-02-20 18:27:51 -0800212 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
214 do_posix_clock_monotonic_gettime(&uptime);
Paul Jacksona49335c2005-09-06 15:18:09 -0700215 do_each_thread(g, p) {
216 unsigned long points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700217
Oleg Nesterov28324d12006-09-29 02:01:12 -0700218 /*
219 * skip kernel threads and tasks which have already released
220 * their mm.
221 */
Nick Piggin5081dde2006-09-25 23:31:32 -0700222 if (!p->mm)
223 continue;
Oleg Nesterov28324d12006-09-29 02:01:12 -0700224 /* skip the init task */
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700225 if (is_global_init(p))
Paul Jacksona49335c2005-09-06 15:18:09 -0700226 continue;
David Rientjes4c4a2212008-02-07 00:14:06 -0800227 if (mem && !task_in_mem_cgroup(p, mem))
228 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700229
Paul Jacksona49335c2005-09-06 15:18:09 -0700230 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700231 * This task already has access to memory reserves and is
232 * being killed. Don't allow any other task access to the
233 * memory reserve.
234 *
235 * Note: this may have a chance of deadlock if it gets
236 * blocked waiting for another task which itself is waiting
237 * for memory. Is there a better alternative?
238 */
239 if (test_tsk_thread_flag(p, TIF_MEMDIE))
240 return ERR_PTR(-1UL);
241
242 /*
Dave Peterson6937a252006-06-23 02:03:13 -0700243 * This is in the process of releasing memory so wait for it
Paul Jacksona49335c2005-09-06 15:18:09 -0700244 * to finish before killing some other task by mistake.
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700245 *
246 * However, if p is the current task, we allow the 'kill' to
247 * go ahead if it is exiting: this will simply set TIF_MEMDIE,
248 * which will allow it to gain access to memory reserves in
249 * the process of exiting and releasing its resources.
Nick Pigginb78483a2006-09-29 02:01:14 -0700250 * Otherwise we could get an easy OOM deadlock.
Paul Jacksona49335c2005-09-06 15:18:09 -0700251 */
Nick Pigginb78483a2006-09-29 02:01:14 -0700252 if (p->flags & PF_EXITING) {
253 if (p != current)
254 return ERR_PTR(-1UL);
255
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700256 chosen = p;
257 *ppoints = ULONG_MAX;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700258 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700259
Li Zefan97d87c92008-04-28 02:13:35 -0700260 points = badness(p, uptime.tv_sec);
David Rientjes4d8b9132009-06-16 15:32:57 -0700261 if (points > *ppoints) {
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) {
David Rientjes6d2661e2009-05-28 14:34:19 -0700290 struct mm_struct *mm;
291
David Rientjesfef1bdd2008-02-07 00:14:07 -0800292 if (mem && !task_in_mem_cgroup(p, mem))
293 continue;
David Rientjesb4416d22008-11-06 12:53:29 -0800294 if (!thread_group_leader(p))
295 continue;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800296
297 task_lock(p);
David Rientjes6d2661e2009-05-28 14:34:19 -0700298 mm = p->mm;
299 if (!mm) {
300 /*
301 * total_vm and rss sizes do not exist for tasks with no
302 * mm so there's no need to report them; they can't be
303 * oom killed anyway.
304 */
305 task_unlock(p);
306 continue;
307 }
David Rientjesfef1bdd2008-02-07 00:14:07 -0800308 printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3d %3d %s\n",
David Rientjes6d2661e2009-05-28 14:34:19 -0700309 p->pid, __task_cred(p)->uid, p->tgid, mm->total_vm,
David Rientjes2ff05b22009-06-16 15:32:56 -0700310 get_mm_rss(mm), (int)task_cpu(p), mm->oom_adj, p->comm);
David Rientjesfef1bdd2008-02-07 00:14:07 -0800311 task_unlock(p);
312 } while_each_thread(g, p);
313}
314
Randy Dunlap1b578df2008-03-19 17:00:42 -0700315/*
Ram Gupta5a291b982006-09-25 23:31:54 -0700316 * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO
317 * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO
318 * set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319 */
Nick Pigginf3af38d2006-12-06 20:31:51 -0800320static void __oom_kill_task(struct task_struct *p, int verbose)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321{
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700322 if (is_global_init(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 WARN_ON(1);
324 printk(KERN_WARNING "tried to kill init!\n");
325 return;
326 }
327
David Rientjes81236812009-06-16 15:33:16 -0700328 if (!p->mm)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329 return;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700330
Nick Pigginf3af38d2006-12-06 20:31:51 -0800331 if (verbose)
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700332 printk(KERN_ERR "Killed process %d (%s)\n",
333 task_pid_nr(p), p->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334
335 /*
336 * We give our sacrificial lamb high priority and access to
337 * all the memory it needs. That way it should be able to
338 * exit() and clear out its resources quickly...
339 */
Peter Zijlstrafa717062008-01-25 21:08:27 +0100340 p->rt.time_slice = HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 set_tsk_thread_flag(p, TIF_MEMDIE);
342
343 force_sig(SIGKILL, p);
344}
345
Nick Pigginf3af38d2006-12-06 20:31:51 -0800346static int oom_kill_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347{
Dave Peterson01315922006-04-18 22:20:44 -0700348 struct mm_struct *mm;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700349 struct task_struct *g, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700350
David Rientjes4d8b9132009-06-16 15:32:57 -0700351 task_lock(p);
Dave Peterson01315922006-04-18 22:20:44 -0700352 mm = p->mm;
David Rientjes4d8b9132009-06-16 15:32:57 -0700353 if (!mm || mm->oom_adj == OOM_DISABLE) {
354 task_unlock(p);
Dave Peterson01315922006-04-18 22:20:44 -0700355 return 1;
David Rientjes4d8b9132009-06-16 15:32:57 -0700356 }
357 task_unlock(p);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800358 __oom_kill_task(p, 1);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800359
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 /*
361 * kill all processes that share the ->mm (i.e. all threads),
Nick Pigginf2a2a712006-12-06 20:31:52 -0800362 * but are in a different thread group. Don't let them have access
363 * to memory reserves though, otherwise we might deplete all memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 */
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800365 do_each_thread(g, q) {
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700366 if (q->mm == mm && !same_thread_group(q, p))
David Rientjes650a7c92007-04-23 21:36:13 -0700367 force_sig(SIGKILL, q);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800368 } while_each_thread(g, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369
Dave Peterson01315922006-04-18 22:20:44 -0700370 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371}
372
David Rientjes7213f502007-10-16 23:25:57 -0700373static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
David Rientjesfef1bdd2008-02-07 00:14:07 -0800374 unsigned long points, struct mem_cgroup *mem,
375 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 struct task_struct *c;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378
David Rientjes7213f502007-10-16 23:25:57 -0700379 if (printk_ratelimit()) {
David Rientjes75aa1992009-01-06 14:39:01 -0800380 task_lock(current);
David Rientjes2ff05b22009-06-16 15:32:56 -0700381 printk(KERN_WARNING "%s invoked oom-killer: "
382 "gfp_mask=0x%x, order=%d, oom_adj=%d\n",
383 current->comm, gfp_mask, order,
384 current->mm ? current->mm->oom_adj : OOM_DISABLE);
David Rientjes75aa1992009-01-06 14:39:01 -0800385 cpuset_print_task_mems_allowed(current);
386 task_unlock(current);
David Rientjes7213f502007-10-16 23:25:57 -0700387 dump_stack();
Balbir Singhe2224322009-04-02 16:57:39 -0700388 mem_cgroup_print_oom_info(mem, current);
David Rientjes7213f502007-10-16 23:25:57 -0700389 show_mem();
David Rientjesfef1bdd2008-02-07 00:14:07 -0800390 if (sysctl_oom_dump_tasks)
391 dump_tasks(mem);
David Rientjes7213f502007-10-16 23:25:57 -0700392 }
393
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700394 /*
395 * If the task is already exiting, don't alarm the sysadmin or kill
396 * its children or threads, just set TIF_MEMDIE so it can die quickly
David Rientjes81236812009-06-16 15:33:16 -0700397 * if its mm is still attached.
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700398 */
David Rientjes81236812009-06-16 15:33:16 -0700399 if (p->mm && (p->flags & PF_EXITING)) {
Nick Pigginf3af38d2006-12-06 20:31:51 -0800400 __oom_kill_task(p, 0);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700401 return 0;
402 }
403
Nick Pigginf3af38d2006-12-06 20:31:51 -0800404 printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700405 message, task_pid_nr(p), p->comm, points);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800406
Linus Torvalds1da177e2005-04-16 15:20:36 -0700407 /* Try to kill a child first */
Matthias Kaehlcke7b1915a2007-10-18 23:39:58 -0700408 list_for_each_entry(c, &p->children, sibling) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409 if (c->mm == p->mm)
410 continue;
Nick Pigginf3af38d2006-12-06 20:31:51 -0800411 if (!oom_kill_task(c))
Dave Peterson01315922006-04-18 22:20:44 -0700412 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 }
Nick Pigginf3af38d2006-12-06 20:31:51 -0800414 return oom_kill_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415}
416
Balbir Singh00f0b822008-03-04 14:28:39 -0800417#ifdef CONFIG_CGROUP_MEM_RES_CTLR
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800418void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
419{
420 unsigned long points = 0;
421 struct task_struct *p;
422
Li Zefane115f2d2008-04-15 14:34:37 -0700423 read_lock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800424retry:
425 p = select_bad_process(&points, mem);
426 if (PTR_ERR(p) == -1UL)
427 goto out;
428
429 if (!p)
430 p = current;
431
David Rientjesfef1bdd2008-02-07 00:14:07 -0800432 if (oom_kill_process(p, gfp_mask, 0, points, mem,
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800433 "Memory cgroup out of memory"))
434 goto retry;
435out:
Li Zefane115f2d2008-04-15 14:34:37 -0700436 read_unlock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800437}
438#endif
439
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700440static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
441
442int register_oom_notifier(struct notifier_block *nb)
443{
444 return blocking_notifier_chain_register(&oom_notify_list, nb);
445}
446EXPORT_SYMBOL_GPL(register_oom_notifier);
447
448int unregister_oom_notifier(struct notifier_block *nb)
449{
450 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
451}
452EXPORT_SYMBOL_GPL(unregister_oom_notifier);
453
David Rientjes098d7f12007-10-16 23:25:55 -0700454/*
455 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
456 * if a parallel OOM killing is already taking place that includes a zone in
457 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
458 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700459int try_set_zone_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700460{
Mel Gormandd1a2392008-04-28 02:12:17 -0700461 struct zoneref *z;
462 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700463 int ret = 1;
464
David Rientjesc7d4cae2009-01-06 14:39:00 -0800465 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700466 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
467 if (zone_is_oom_locked(zone)) {
David Rientjes098d7f12007-10-16 23:25:55 -0700468 ret = 0;
469 goto out;
470 }
Mel Gormandd1a2392008-04-28 02:12:17 -0700471 }
David Rientjes098d7f12007-10-16 23:25:55 -0700472
Mel Gormandd1a2392008-04-28 02:12:17 -0700473 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
474 /*
David Rientjesc7d4cae2009-01-06 14:39:00 -0800475 * Lock each zone in the zonelist under zone_scan_lock so a
Mel Gormandd1a2392008-04-28 02:12:17 -0700476 * parallel invocation of try_set_zone_oom() doesn't succeed
477 * when it shouldn't.
478 */
479 zone_set_flag(zone, ZONE_OOM_LOCKED);
480 }
481
David Rientjes098d7f12007-10-16 23:25:55 -0700482out:
David Rientjesc7d4cae2009-01-06 14:39:00 -0800483 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700484 return ret;
485}
486
487/*
488 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
489 * allocation attempts with zonelists containing them may now recall the OOM
490 * killer, if necessary.
491 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700492void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700493{
Mel Gormandd1a2392008-04-28 02:12:17 -0700494 struct zoneref *z;
495 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700496
David Rientjesc7d4cae2009-01-06 14:39:00 -0800497 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700498 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
499 zone_clear_flag(zone, ZONE_OOM_LOCKED);
500 }
David Rientjesc7d4cae2009-01-06 14:39:00 -0800501 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700502}
503
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800504/*
505 * Must be called with tasklist_lock held for read.
506 */
507static void __out_of_memory(gfp_t gfp_mask, int order)
508{
David Rientjes184101b2009-05-06 16:02:55 -0700509 struct task_struct *p;
510 unsigned long points;
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800511
David Rientjes184101b2009-05-06 16:02:55 -0700512 if (sysctl_oom_kill_allocating_task)
513 if (!oom_kill_process(current, gfp_mask, order, 0, NULL,
514 "Out of memory (oom_kill_allocating_task)"))
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800515 return;
David Rientjes184101b2009-05-06 16:02:55 -0700516retry:
517 /*
518 * Rambo mode: Shoot down a process and hope it solves whatever
519 * issues we may have.
520 */
521 p = select_bad_process(&points, NULL);
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800522
David Rientjes184101b2009-05-06 16:02:55 -0700523 if (PTR_ERR(p) == -1UL)
524 return;
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800525
David Rientjes184101b2009-05-06 16:02:55 -0700526 /* Found nothing?!?! Either we hang forever, or we panic. */
527 if (!p) {
528 read_unlock(&tasklist_lock);
529 panic("Out of memory and no killable processes...\n");
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800530 }
David Rientjes184101b2009-05-06 16:02:55 -0700531
532 if (oom_kill_process(p, gfp_mask, order, points, NULL,
533 "Out of memory"))
534 goto retry;
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800535}
536
537/*
538 * pagefault handler calls into here because it is out of memory but
539 * doesn't know exactly how or why.
540 */
541void pagefault_out_of_memory(void)
542{
543 unsigned long freed = 0;
544
545 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
546 if (freed > 0)
547 /* Got some memory back in the last second. */
548 return;
549
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800550 /*
551 * If this is from memcg, oom-killer is already invoked.
552 * and not worth to go system-wide-oom.
553 */
554 if (mem_cgroup_oom_called(current))
555 goto rest_and_return;
556
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800557 if (sysctl_panic_on_oom)
558 panic("out of memory from page fault. panic_on_oom is selected.\n");
559
560 read_lock(&tasklist_lock);
561 __out_of_memory(0, 0); /* unknown gfp_mask and order */
562 read_unlock(&tasklist_lock);
563
564 /*
565 * Give "p" a good chance of killing itself before we
566 * retry to allocate memory.
567 */
KAMEZAWA Hiroyukia636b322009-01-07 18:08:08 -0800568rest_and_return:
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800569 if (!test_thread_flag(TIF_MEMDIE))
570 schedule_timeout_uninterruptible(1);
571}
572
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573/**
Dave Peterson6937a252006-06-23 02:03:13 -0700574 * out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700575 * @zonelist: zonelist pointer
576 * @gfp_mask: memory allocation flags
577 * @order: amount of memory being requested as a power of 2
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 *
579 * If we run out of memory, we have the choice between either
580 * killing a random task (bad), letting the system crash (worse)
581 * OR try to be smart about which process to kill. Note that we
582 * don't have to be perfect here, we just have to be good.
583 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800584void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585{
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700586 unsigned long freed = 0;
David Rientjes70e24bd2007-10-16 23:25:53 -0700587 enum oom_constraint constraint;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700588
589 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
590 if (freed > 0)
591 /* Got some memory back in the last second. */
592 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
Yasunori Goto2b744c02007-05-06 14:49:59 -0700594 if (sysctl_panic_on_oom == 2)
595 panic("out of memory. Compulsory panic_on_oom is selected.\n");
596
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800597 /*
598 * Check if there were limitations on the allocation (only relevant for
599 * NUMA) that may require different handling.
600 */
David Rientjes2b45ab32007-05-06 14:50:00 -0700601 constraint = constrained_alloc(zonelist, gfp_mask);
David Rientjes2b45ab32007-05-06 14:50:00 -0700602 read_lock(&tasklist_lock);
603
604 switch (constraint) {
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800605 case CONSTRAINT_MEMORY_POLICY:
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800606 oom_kill_process(current, gfp_mask, order, 0, NULL,
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800607 "No available memory (MPOL_BIND)");
608 break;
609
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800610 case CONSTRAINT_NONE:
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -0700611 if (sysctl_panic_on_oom)
612 panic("out of memory. panic_on_oom is selected\n");
David Rientjesfe071d72007-10-16 23:25:56 -0700613 /* Fall-through */
614 case CONSTRAINT_CPUSET:
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800615 __out_of_memory(gfp_mask, order);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800616 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 }
618
Andrew Morton140ffce2006-03-02 02:54:28 -0800619 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621 /*
622 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800623 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 */
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800625 if (!test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800626 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627}