blob: 96473b482099793384af969b8158e1306e2d4f58 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -070029int sysctl_panic_on_oom;
David Rientjesfe071d72007-10-16 23:25:56 -070030int sysctl_oom_kill_allocating_task;
David Rientjesae741382007-10-16 23:26:00 -070031static DEFINE_SPINLOCK(zone_scan_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070032/* #define DEBUG */
33
34/**
Dave Peterson6937a252006-06-23 02:03:13 -070035 * badness - calculate a numeric value for how bad this task has been
Linus Torvalds1da177e2005-04-16 15:20:36 -070036 * @p: task struct of which task we should calculate
Paul Jacksona49335c2005-09-06 15:18:09 -070037 * @uptime: current uptime in seconds
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 *
39 * The formula used is relatively simple and documented inline in the
40 * function. The main rationale is that we want to select a good task
41 * to kill when we run out of memory.
42 *
43 * Good in this context means that:
44 * 1) we lose the minimum amount of work done
45 * 2) we recover a large amount of memory
46 * 3) we don't kill anything innocent of eating tons of memory
47 * 4) we want to kill the minimum amount of processes (one)
48 * 5) we try to kill the process the user expects us to kill, this
49 * algorithm has been meticulously tuned to meet the principle
50 * of least surprise ... (be careful when you change it)
51 */
52
53unsigned long badness(struct task_struct *p, unsigned long uptime)
54{
55 unsigned long points, cpu_time, run_time, s;
Andrew Morton97c2c9b82006-04-18 22:20:38 -070056 struct mm_struct *mm;
57 struct task_struct *child;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Andrew Morton97c2c9b82006-04-18 22:20:38 -070059 task_lock(p);
60 mm = p->mm;
61 if (!mm) {
62 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 return 0;
Andrew Morton97c2c9b82006-04-18 22:20:38 -070064 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
66 /*
67 * The memory size of the process is the basis for the badness.
68 */
Andrew Morton97c2c9b82006-04-18 22:20:38 -070069 points = mm->total_vm;
70
71 /*
72 * After this unlock we can no longer dereference local variable `mm'
73 */
74 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76 /*
Hugh Dickins7ba34852007-01-05 16:37:03 -080077 * swapoff can easily use up all memory, so kill those first.
78 */
79 if (p->flags & PF_SWAPOFF)
80 return ULONG_MAX;
81
82 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 * Processes which fork a lot of child processes are likely
Kurt Garloff9827b782006-02-20 18:27:51 -080084 * a good choice. We add half the vmsize of the children if they
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 * have an own mm. This prevents forking servers to flood the
Kurt Garloff9827b782006-02-20 18:27:51 -080086 * machine with an endless amount of children. In case a single
87 * child is eating the vast majority of memory, adding only half
88 * to the parents will make the child our kill candidate of choice.
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 */
Andrew Morton97c2c9b82006-04-18 22:20:38 -070090 list_for_each_entry(child, &p->children, sibling) {
91 task_lock(child);
92 if (child->mm != mm && child->mm)
93 points += child->mm->total_vm/2 + 1;
94 task_unlock(child);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 }
96
97 /*
98 * CPU time is in tens of seconds and run time is in thousands
99 * of seconds. There is no particular reason for this other than
100 * that it turned out to work very well in practice.
101 */
102 cpu_time = (cputime_to_jiffies(p->utime) + cputime_to_jiffies(p->stime))
103 >> (SHIFT_HZ + 3);
104
105 if (uptime >= p->start_time.tv_sec)
106 run_time = (uptime - p->start_time.tv_sec) >> 10;
107 else
108 run_time = 0;
109
110 s = int_sqrt(cpu_time);
111 if (s)
112 points /= s;
113 s = int_sqrt(int_sqrt(run_time));
114 if (s)
115 points /= s;
116
117 /*
118 * Niced processes are most likely less important, so double
119 * their badness points.
120 */
121 if (task_nice(p) > 0)
122 points *= 2;
123
124 /*
125 * Superuser processes are usually more important, so we make it
126 * less likely that we kill those.
127 */
128 if (cap_t(p->cap_effective) & CAP_TO_MASK(CAP_SYS_ADMIN) ||
129 p->uid == 0 || p->euid == 0)
130 points /= 4;
131
132 /*
133 * We don't want to kill a process with direct hardware access.
134 * Not only could that mess up the hardware, but usually users
135 * tend to only have this flag set on applications they think
136 * of as important.
137 */
138 if (cap_t(p->cap_effective) & CAP_TO_MASK(CAP_SYS_RAWIO))
139 points /= 4;
140
141 /*
Nick Piggin7887a3d2006-09-25 23:31:29 -0700142 * If p's nodes don't overlap ours, it may still help to kill p
143 * because p may have allocated or otherwise mapped memory on
144 * this node before. However it will be less likely.
145 */
David Rientjesbbe373f2007-10-16 23:25:58 -0700146 if (!cpuset_mems_allowed_intersects(current, p))
Nick Piggin7887a3d2006-09-25 23:31:29 -0700147 points /= 8;
148
149 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 * Adjust the score by oomkilladj.
151 */
152 if (p->oomkilladj) {
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700153 if (p->oomkilladj > 0) {
154 if (!points)
155 points = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 points <<= p->oomkilladj;
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700157 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 points >>= -(p->oomkilladj);
159 }
160
161#ifdef DEBUG
David Rientjesa5e58a62007-07-31 00:37:32 -0700162 printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 p->pid, p->comm, points);
164#endif
165 return points;
166}
167
168/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800169 * Determine the type of allocation constraint.
170 */
David Rientjes70e24bd2007-10-16 23:25:53 -0700171static inline enum oom_constraint constrained_alloc(struct zonelist *zonelist,
172 gfp_t gfp_mask)
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800173{
174#ifdef CONFIG_NUMA
175 struct zone **z;
Christoph Lameteree31af52007-10-16 01:25:30 -0700176 nodemask_t nodes = node_states[N_HIGH_MEMORY];
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800177
178 for (z = zonelist->zones; *z; z++)
Paul Jackson02a0e532006-12-13 00:34:25 -0800179 if (cpuset_zone_allowed_softwall(*z, gfp_mask))
Christoph Lameter89fa3022006-09-25 23:31:55 -0700180 node_clear(zone_to_nid(*z), nodes);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800181 else
182 return CONSTRAINT_CPUSET;
183
184 if (!nodes_empty(nodes))
185 return CONSTRAINT_MEMORY_POLICY;
186#endif
187
188 return CONSTRAINT_NONE;
189}
190
191/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 * Simple selection loop. We chose the process with the highest
193 * number of 'points'. We expect the caller will lock the tasklist.
194 *
195 * (not docbooked, we don't want this one cluttering up the manual)
196 */
Kurt Garloff9827b782006-02-20 18:27:51 -0800197static struct task_struct *select_bad_process(unsigned long *ppoints)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 struct task_struct *g, *p;
200 struct task_struct *chosen = NULL;
201 struct timespec uptime;
Kurt Garloff9827b782006-02-20 18:27:51 -0800202 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
204 do_posix_clock_monotonic_gettime(&uptime);
Paul Jacksona49335c2005-09-06 15:18:09 -0700205 do_each_thread(g, p) {
206 unsigned long points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700207
Oleg Nesterov28324d12006-09-29 02:01:12 -0700208 /*
209 * skip kernel threads and tasks which have already released
210 * their mm.
211 */
Nick Piggin5081dde2006-09-25 23:31:32 -0700212 if (!p->mm)
213 continue;
Oleg Nesterov28324d12006-09-29 02:01:12 -0700214 /* skip the init task */
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700215 if (is_global_init(p))
Paul Jacksona49335c2005-09-06 15:18:09 -0700216 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700217
Paul Jacksona49335c2005-09-06 15:18:09 -0700218 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700219 * This task already has access to memory reserves and is
220 * being killed. Don't allow any other task access to the
221 * memory reserve.
222 *
223 * Note: this may have a chance of deadlock if it gets
224 * blocked waiting for another task which itself is waiting
225 * for memory. Is there a better alternative?
226 */
227 if (test_tsk_thread_flag(p, TIF_MEMDIE))
228 return ERR_PTR(-1UL);
229
230 /*
Dave Peterson6937a252006-06-23 02:03:13 -0700231 * This is in the process of releasing memory so wait for it
Paul Jacksona49335c2005-09-06 15:18:09 -0700232 * to finish before killing some other task by mistake.
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700233 *
234 * However, if p is the current task, we allow the 'kill' to
235 * go ahead if it is exiting: this will simply set TIF_MEMDIE,
236 * which will allow it to gain access to memory reserves in
237 * the process of exiting and releasing its resources.
Nick Pigginb78483a2006-09-29 02:01:14 -0700238 * Otherwise we could get an easy OOM deadlock.
Paul Jacksona49335c2005-09-06 15:18:09 -0700239 */
Nick Pigginb78483a2006-09-29 02:01:14 -0700240 if (p->flags & PF_EXITING) {
241 if (p != current)
242 return ERR_PTR(-1UL);
243
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700244 chosen = p;
245 *ppoints = ULONG_MAX;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700246 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700247
Nick Piggin4a3ede12006-09-25 23:31:30 -0700248 if (p->oomkilladj == OOM_DISABLE)
249 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250
Paul Jacksona49335c2005-09-06 15:18:09 -0700251 points = badness(p, uptime.tv_sec);
Kurt Garloff9827b782006-02-20 18:27:51 -0800252 if (points > *ppoints || !chosen) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700253 chosen = p;
Kurt Garloff9827b782006-02-20 18:27:51 -0800254 *ppoints = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 }
Paul Jacksona49335c2005-09-06 15:18:09 -0700256 } while_each_thread(g, p);
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 return chosen;
259}
260
261/**
Ram Gupta5a291b982006-09-25 23:31:54 -0700262 * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO
263 * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO
264 * set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 */
Nick Pigginf3af38d2006-12-06 20:31:51 -0800266static void __oom_kill_task(struct task_struct *p, int verbose)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267{
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700268 if (is_global_init(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 WARN_ON(1);
270 printk(KERN_WARNING "tried to kill init!\n");
271 return;
272 }
273
Oleg Nesterov01017a22006-09-29 02:01:13 -0700274 if (!p->mm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 WARN_ON(1);
276 printk(KERN_WARNING "tried to kill an mm-less task!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277 return;
278 }
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700279
Nick Pigginf3af38d2006-12-06 20:31:51 -0800280 if (verbose)
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700281 printk(KERN_ERR "Killed process %d (%s)\n",
282 task_pid_nr(p), p->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
284 /*
285 * We give our sacrificial lamb high priority and access to
286 * all the memory it needs. That way it should be able to
287 * exit() and clear out its resources quickly...
288 */
Peter Zijlstrafa717062008-01-25 21:08:27 +0100289 p->rt.time_slice = HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290 set_tsk_thread_flag(p, TIF_MEMDIE);
291
292 force_sig(SIGKILL, p);
293}
294
Nick Pigginf3af38d2006-12-06 20:31:51 -0800295static int oom_kill_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296{
Dave Peterson01315922006-04-18 22:20:44 -0700297 struct mm_struct *mm;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700298 struct task_struct *g, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
Dave Peterson01315922006-04-18 22:20:44 -0700300 mm = p->mm;
301
302 /* WARNING: mm may not be dereferenced since we did not obtain its
303 * value from get_task_mm(p). This is OK since all we need to do is
304 * compare mm to q->mm below.
305 *
306 * Furthermore, even if mm contains a non-NULL value, p->mm may
307 * change to NULL at any time since we do not hold task_lock(p).
308 * However, this is of no concern to us.
309 */
310
Oleg Nesterov01017a22006-09-29 02:01:13 -0700311 if (mm == NULL)
Dave Peterson01315922006-04-18 22:20:44 -0700312 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800314 /*
315 * Don't kill the process if any threads are set to OOM_DISABLE
316 */
317 do_each_thread(g, q) {
Ankita Garg35ae8342007-03-16 13:38:33 -0800318 if (q->mm == mm && q->oomkilladj == OOM_DISABLE)
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800319 return 1;
320 } while_each_thread(g, q);
321
Nick Pigginf3af38d2006-12-06 20:31:51 -0800322 __oom_kill_task(p, 1);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800323
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 /*
325 * kill all processes that share the ->mm (i.e. all threads),
Nick Pigginf2a2a712006-12-06 20:31:52 -0800326 * but are in a different thread group. Don't let them have access
327 * to memory reserves though, otherwise we might deplete all memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328 */
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800329 do_each_thread(g, q) {
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700330 if (q->mm == mm && !same_thread_group(q, p))
David Rientjes650a7c92007-04-23 21:36:13 -0700331 force_sig(SIGKILL, q);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800332 } while_each_thread(g, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333
Dave Peterson01315922006-04-18 22:20:44 -0700334 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700335}
336
David Rientjes7213f502007-10-16 23:25:57 -0700337static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
338 unsigned long points, const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340 struct task_struct *c;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341
David Rientjes7213f502007-10-16 23:25:57 -0700342 if (printk_ratelimit()) {
343 printk(KERN_WARNING "%s invoked oom-killer: "
344 "gfp_mask=0x%x, order=%d, oomkilladj=%d\n",
345 current->comm, gfp_mask, order, current->oomkilladj);
346 dump_stack();
347 show_mem();
348 }
349
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700350 /*
351 * If the task is already exiting, don't alarm the sysadmin or kill
352 * its children or threads, just set TIF_MEMDIE so it can die quickly
353 */
354 if (p->flags & PF_EXITING) {
Nick Pigginf3af38d2006-12-06 20:31:51 -0800355 __oom_kill_task(p, 0);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700356 return 0;
357 }
358
Nick Pigginf3af38d2006-12-06 20:31:51 -0800359 printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700360 message, task_pid_nr(p), p->comm, points);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800361
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362 /* Try to kill a child first */
Matthias Kaehlcke7b1915a2007-10-18 23:39:58 -0700363 list_for_each_entry(c, &p->children, sibling) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 if (c->mm == p->mm)
365 continue;
Nick Pigginf3af38d2006-12-06 20:31:51 -0800366 if (!oom_kill_task(c))
Dave Peterson01315922006-04-18 22:20:44 -0700367 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 }
Nick Pigginf3af38d2006-12-06 20:31:51 -0800369 return oom_kill_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370}
371
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700372static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
373
374int register_oom_notifier(struct notifier_block *nb)
375{
376 return blocking_notifier_chain_register(&oom_notify_list, nb);
377}
378EXPORT_SYMBOL_GPL(register_oom_notifier);
379
380int unregister_oom_notifier(struct notifier_block *nb)
381{
382 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
383}
384EXPORT_SYMBOL_GPL(unregister_oom_notifier);
385
David Rientjes098d7f12007-10-16 23:25:55 -0700386/*
387 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
388 * if a parallel OOM killing is already taking place that includes a zone in
389 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
390 */
391int try_set_zone_oom(struct zonelist *zonelist)
392{
393 struct zone **z;
394 int ret = 1;
395
396 z = zonelist->zones;
397
David Rientjesae741382007-10-16 23:26:00 -0700398 spin_lock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700399 do {
400 if (zone_is_oom_locked(*z)) {
401 ret = 0;
402 goto out;
403 }
404 } while (*(++z) != NULL);
405
406 /*
407 * Lock each zone in the zonelist under zone_scan_mutex so a parallel
408 * invocation of try_set_zone_oom() doesn't succeed when it shouldn't.
409 */
410 z = zonelist->zones;
411 do {
412 zone_set_flag(*z, ZONE_OOM_LOCKED);
413 } while (*(++z) != NULL);
414out:
David Rientjesae741382007-10-16 23:26:00 -0700415 spin_unlock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700416 return ret;
417}
418
419/*
420 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
421 * allocation attempts with zonelists containing them may now recall the OOM
422 * killer, if necessary.
423 */
424void clear_zonelist_oom(struct zonelist *zonelist)
425{
426 struct zone **z;
427
428 z = zonelist->zones;
429
David Rientjesae741382007-10-16 23:26:00 -0700430 spin_lock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700431 do {
432 zone_clear_flag(*z, ZONE_OOM_LOCKED);
433 } while (*(++z) != NULL);
David Rientjesae741382007-10-16 23:26:00 -0700434 spin_unlock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700435}
436
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437/**
Dave Peterson6937a252006-06-23 02:03:13 -0700438 * out_of_memory - kill the "best" process when we run out of memory
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 *
440 * If we run out of memory, we have the choice between either
441 * killing a random task (bad), letting the system crash (worse)
442 * OR try to be smart about which process to kill. Note that we
443 * don't have to be perfect here, we just have to be good.
444 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800445void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700446{
Ingo Molnar36c8b582006-07-03 00:25:41 -0700447 struct task_struct *p;
Andrew Mortond6713e02006-02-28 16:59:19 -0800448 unsigned long points = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700449 unsigned long freed = 0;
David Rientjes70e24bd2007-10-16 23:25:53 -0700450 enum oom_constraint constraint;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700451
452 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
453 if (freed > 0)
454 /* Got some memory back in the last second. */
455 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456
Yasunori Goto2b744c02007-05-06 14:49:59 -0700457 if (sysctl_panic_on_oom == 2)
458 panic("out of memory. Compulsory panic_on_oom is selected.\n");
459
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800460 /*
461 * Check if there were limitations on the allocation (only relevant for
462 * NUMA) that may require different handling.
463 */
David Rientjes2b45ab32007-05-06 14:50:00 -0700464 constraint = constrained_alloc(zonelist, gfp_mask);
David Rientjes2b45ab32007-05-06 14:50:00 -0700465 read_lock(&tasklist_lock);
466
467 switch (constraint) {
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800468 case CONSTRAINT_MEMORY_POLICY:
David Rientjes7213f502007-10-16 23:25:57 -0700469 oom_kill_process(current, gfp_mask, order, points,
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800470 "No available memory (MPOL_BIND)");
471 break;
472
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800473 case CONSTRAINT_NONE:
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -0700474 if (sysctl_panic_on_oom)
475 panic("out of memory. panic_on_oom is selected\n");
David Rientjesfe071d72007-10-16 23:25:56 -0700476 /* Fall-through */
477 case CONSTRAINT_CPUSET:
478 if (sysctl_oom_kill_allocating_task) {
David Rientjes7213f502007-10-16 23:25:57 -0700479 oom_kill_process(current, gfp_mask, order, points,
David Rientjesfe071d72007-10-16 23:25:56 -0700480 "Out of memory (oom_kill_allocating_task)");
481 break;
482 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483retry:
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800484 /*
485 * Rambo mode: Shoot down a process and hope it solves whatever
486 * issues we may have.
487 */
488 p = select_bad_process(&points);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800490 if (PTR_ERR(p) == -1UL)
491 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700492
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800493 /* Found nothing?!?! Either we hang forever, or we panic. */
494 if (!p) {
495 read_unlock(&tasklist_lock);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800496 panic("Out of memory and no killable processes...\n");
497 }
498
Al Viroe91a8102007-10-20 21:45:33 +0100499 if (oom_kill_process(p, gfp_mask, order, points,
David Rientjes7213f502007-10-16 23:25:57 -0700500 "Out of memory"))
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800501 goto retry;
502
503 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 }
505
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800506out:
Andrew Morton140ffce2006-03-02 02:54:28 -0800507 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508
509 /*
510 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800511 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700512 */
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800513 if (!test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800514 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700515}