Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | |
a49335cc PJ |
9 | * 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. | |
1da177e4 LT |
11 | * |
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 | ||
8ac773b4 | 18 | #include <linux/oom.h> |
1da177e4 | 19 | #include <linux/mm.h> |
4e950f6f | 20 | #include <linux/err.h> |
5a0e3ad6 | 21 | #include <linux/gfp.h> |
1da177e4 LT |
22 | #include <linux/sched.h> |
23 | #include <linux/swap.h> | |
24 | #include <linux/timex.h> | |
25 | #include <linux/jiffies.h> | |
ef08e3b4 | 26 | #include <linux/cpuset.h> |
8bc719d3 MS |
27 | #include <linux/module.h> |
28 | #include <linux/notifier.h> | |
c7ba5c9e | 29 | #include <linux/memcontrol.h> |
5cd9c58f | 30 | #include <linux/security.h> |
1da177e4 | 31 | |
fadd8fbd | 32 | int sysctl_panic_on_oom; |
fe071d7e | 33 | int sysctl_oom_kill_allocating_task; |
fef1bdd6 | 34 | int sysctl_oom_dump_tasks; |
c7d4caeb | 35 | static DEFINE_SPINLOCK(zone_scan_lock); |
1da177e4 LT |
36 | /* #define DEBUG */ |
37 | ||
495789a5 KM |
38 | /* |
39 | * Is all threads of the target process nodes overlap ours? | |
40 | */ | |
41 | static int has_intersects_mems_allowed(struct task_struct *tsk) | |
42 | { | |
43 | struct task_struct *t; | |
44 | ||
45 | t = tsk; | |
46 | do { | |
47 | if (cpuset_mems_allowed_intersects(current, t)) | |
48 | return 1; | |
49 | t = next_thread(t); | |
50 | } while (t != tsk); | |
51 | ||
52 | return 0; | |
53 | } | |
54 | ||
dd8e8f40 ON |
55 | static struct task_struct *find_lock_task_mm(struct task_struct *p) |
56 | { | |
57 | struct task_struct *t = p; | |
58 | ||
59 | do { | |
60 | task_lock(t); | |
61 | if (likely(t->mm)) | |
62 | return t; | |
63 | task_unlock(t); | |
64 | } while_each_thread(p, t); | |
65 | ||
66 | return NULL; | |
67 | } | |
68 | ||
1da177e4 | 69 | /** |
6937a25c | 70 | * badness - calculate a numeric value for how bad this task has been |
1da177e4 | 71 | * @p: task struct of which task we should calculate |
a49335cc | 72 | * @uptime: current uptime in seconds |
1da177e4 LT |
73 | * |
74 | * The formula used is relatively simple and documented inline in the | |
75 | * function. The main rationale is that we want to select a good task | |
76 | * to kill when we run out of memory. | |
77 | * | |
78 | * Good in this context means that: | |
79 | * 1) we lose the minimum amount of work done | |
80 | * 2) we recover a large amount of memory | |
81 | * 3) we don't kill anything innocent of eating tons of memory | |
82 | * 4) we want to kill the minimum amount of processes (one) | |
83 | * 5) we try to kill the process the user expects us to kill, this | |
84 | * algorithm has been meticulously tuned to meet the principle | |
85 | * of least surprise ... (be careful when you change it) | |
86 | */ | |
87 | ||
97d87c97 | 88 | unsigned long badness(struct task_struct *p, unsigned long uptime) |
1da177e4 | 89 | { |
a12888f7 | 90 | unsigned long points, cpu_time, run_time; |
97c2c9b8 | 91 | struct task_struct *child; |
dd8e8f40 | 92 | struct task_struct *c, *t; |
28b83c51 | 93 | int oom_adj = p->signal->oom_adj; |
495789a5 KM |
94 | struct task_cputime task_time; |
95 | unsigned long utime; | |
96 | unsigned long stime; | |
28b83c51 KM |
97 | |
98 | if (oom_adj == OOM_DISABLE) | |
99 | return 0; | |
1da177e4 | 100 | |
dd8e8f40 ON |
101 | p = find_lock_task_mm(p); |
102 | if (!p) | |
1da177e4 LT |
103 | return 0; |
104 | ||
105 | /* | |
106 | * The memory size of the process is the basis for the badness. | |
107 | */ | |
dd8e8f40 | 108 | points = p->mm->total_vm; |
97c2c9b8 AM |
109 | |
110 | /* | |
111 | * After this unlock we can no longer dereference local variable `mm' | |
112 | */ | |
113 | task_unlock(p); | |
1da177e4 | 114 | |
7ba34859 HD |
115 | /* |
116 | * swapoff can easily use up all memory, so kill those first. | |
117 | */ | |
35451bee | 118 | if (p->flags & PF_OOM_ORIGIN) |
7ba34859 HD |
119 | return ULONG_MAX; |
120 | ||
1da177e4 LT |
121 | /* |
122 | * Processes which fork a lot of child processes are likely | |
9827b781 | 123 | * a good choice. We add half the vmsize of the children if they |
1da177e4 | 124 | * have an own mm. This prevents forking servers to flood the |
9827b781 KG |
125 | * machine with an endless amount of children. In case a single |
126 | * child is eating the vast majority of memory, adding only half | |
127 | * to the parents will make the child our kill candidate of choice. | |
1da177e4 | 128 | */ |
dd8e8f40 ON |
129 | t = p; |
130 | do { | |
131 | list_for_each_entry(c, &t->children, sibling) { | |
132 | child = find_lock_task_mm(c); | |
133 | if (child) { | |
134 | if (child->mm != p->mm) | |
135 | points += child->mm->total_vm/2 + 1; | |
136 | task_unlock(child); | |
137 | } | |
138 | } | |
139 | } while_each_thread(p, t); | |
1da177e4 LT |
140 | |
141 | /* | |
142 | * CPU time is in tens of seconds and run time is in thousands | |
143 | * of seconds. There is no particular reason for this other than | |
144 | * that it turned out to work very well in practice. | |
145 | */ | |
495789a5 KM |
146 | thread_group_cputime(p, &task_time); |
147 | utime = cputime_to_jiffies(task_time.utime); | |
148 | stime = cputime_to_jiffies(task_time.stime); | |
149 | cpu_time = (utime + stime) >> (SHIFT_HZ + 3); | |
150 | ||
1da177e4 LT |
151 | |
152 | if (uptime >= p->start_time.tv_sec) | |
153 | run_time = (uptime - p->start_time.tv_sec) >> 10; | |
154 | else | |
155 | run_time = 0; | |
156 | ||
a12888f7 CG |
157 | if (cpu_time) |
158 | points /= int_sqrt(cpu_time); | |
159 | if (run_time) | |
160 | points /= int_sqrt(int_sqrt(run_time)); | |
1da177e4 LT |
161 | |
162 | /* | |
163 | * Niced processes are most likely less important, so double | |
164 | * their badness points. | |
165 | */ | |
166 | if (task_nice(p) > 0) | |
167 | points *= 2; | |
168 | ||
169 | /* | |
170 | * Superuser processes are usually more important, so we make it | |
171 | * less likely that we kill those. | |
172 | */ | |
a2f2945a EP |
173 | if (has_capability_noaudit(p, CAP_SYS_ADMIN) || |
174 | has_capability_noaudit(p, CAP_SYS_RESOURCE)) | |
1da177e4 LT |
175 | points /= 4; |
176 | ||
177 | /* | |
178 | * We don't want to kill a process with direct hardware access. | |
179 | * Not only could that mess up the hardware, but usually users | |
180 | * tend to only have this flag set on applications they think | |
181 | * of as important. | |
182 | */ | |
a2f2945a | 183 | if (has_capability_noaudit(p, CAP_SYS_RAWIO)) |
1da177e4 LT |
184 | points /= 4; |
185 | ||
7887a3da NP |
186 | /* |
187 | * If p's nodes don't overlap ours, it may still help to kill p | |
188 | * because p may have allocated or otherwise mapped memory on | |
189 | * this node before. However it will be less likely. | |
190 | */ | |
495789a5 | 191 | if (!has_intersects_mems_allowed(p)) |
7887a3da NP |
192 | points /= 8; |
193 | ||
1da177e4 | 194 | /* |
28b83c51 | 195 | * Adjust the score by oom_adj. |
1da177e4 | 196 | */ |
28b83c51 KM |
197 | if (oom_adj) { |
198 | if (oom_adj > 0) { | |
9a82782f JP |
199 | if (!points) |
200 | points = 1; | |
28b83c51 | 201 | points <<= oom_adj; |
9a82782f | 202 | } else |
28b83c51 | 203 | points >>= -(oom_adj); |
1da177e4 LT |
204 | } |
205 | ||
206 | #ifdef DEBUG | |
a5e58a61 | 207 | printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n", |
1da177e4 LT |
208 | p->pid, p->comm, points); |
209 | #endif | |
210 | return points; | |
211 | } | |
212 | ||
9b0f8b04 CL |
213 | /* |
214 | * Determine the type of allocation constraint. | |
215 | */ | |
9b0f8b04 | 216 | #ifdef CONFIG_NUMA |
4365a567 KH |
217 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
218 | gfp_t gfp_mask, nodemask_t *nodemask) | |
219 | { | |
54a6eb5c | 220 | struct zone *zone; |
dd1a239f | 221 | struct zoneref *z; |
54a6eb5c | 222 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
9b0f8b04 | 223 | |
4365a567 KH |
224 | /* |
225 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid | |
226 | * to kill current.We have to random task kill in this case. | |
227 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. | |
228 | */ | |
229 | if (gfp_mask & __GFP_THISNODE) | |
230 | return CONSTRAINT_NONE; | |
9b0f8b04 | 231 | |
4365a567 KH |
232 | /* |
233 | * The nodemask here is a nodemask passed to alloc_pages(). Now, | |
234 | * cpuset doesn't use this nodemask for its hardwall/softwall/hierarchy | |
235 | * feature. mempolicy is an only user of nodemask here. | |
236 | * check mempolicy's nodemask contains all N_HIGH_MEMORY | |
237 | */ | |
238 | if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) | |
9b0f8b04 | 239 | return CONSTRAINT_MEMORY_POLICY; |
4365a567 KH |
240 | |
241 | /* Check this allocation failure is caused by cpuset's wall function */ | |
242 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
243 | high_zoneidx, nodemask) | |
244 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) | |
245 | return CONSTRAINT_CPUSET; | |
9b0f8b04 CL |
246 | |
247 | return CONSTRAINT_NONE; | |
248 | } | |
4365a567 KH |
249 | #else |
250 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, | |
251 | gfp_t gfp_mask, nodemask_t *nodemask) | |
252 | { | |
253 | return CONSTRAINT_NONE; | |
254 | } | |
255 | #endif | |
9b0f8b04 | 256 | |
1da177e4 LT |
257 | /* |
258 | * Simple selection loop. We chose the process with the highest | |
259 | * number of 'points'. We expect the caller will lock the tasklist. | |
260 | * | |
261 | * (not docbooked, we don't want this one cluttering up the manual) | |
262 | */ | |
c7ba5c9e PE |
263 | static struct task_struct *select_bad_process(unsigned long *ppoints, |
264 | struct mem_cgroup *mem) | |
1da177e4 | 265 | { |
495789a5 | 266 | struct task_struct *p; |
1da177e4 LT |
267 | struct task_struct *chosen = NULL; |
268 | struct timespec uptime; | |
9827b781 | 269 | *ppoints = 0; |
1da177e4 LT |
270 | |
271 | do_posix_clock_monotonic_gettime(&uptime); | |
495789a5 | 272 | for_each_process(p) { |
a49335cc | 273 | unsigned long points; |
a49335cc | 274 | |
455c0e5f ON |
275 | /* skip the init task and kthreads */ |
276 | if (is_global_init(p) || (p->flags & PF_KTHREAD)) | |
a49335cc | 277 | continue; |
4c4a2214 DR |
278 | if (mem && !task_in_mem_cgroup(p, mem)) |
279 | continue; | |
ef08e3b4 | 280 | |
b78483a4 NP |
281 | /* |
282 | * This task already has access to memory reserves and is | |
283 | * being killed. Don't allow any other task access to the | |
284 | * memory reserve. | |
285 | * | |
286 | * Note: this may have a chance of deadlock if it gets | |
287 | * blocked waiting for another task which itself is waiting | |
288 | * for memory. Is there a better alternative? | |
289 | */ | |
290 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) | |
291 | return ERR_PTR(-1UL); | |
292 | ||
a49335cc | 293 | /* |
6937a25c | 294 | * This is in the process of releasing memory so wait for it |
a49335cc | 295 | * to finish before killing some other task by mistake. |
50ec3bbf NP |
296 | * |
297 | * However, if p is the current task, we allow the 'kill' to | |
298 | * go ahead if it is exiting: this will simply set TIF_MEMDIE, | |
299 | * which will allow it to gain access to memory reserves in | |
300 | * the process of exiting and releasing its resources. | |
b78483a4 | 301 | * Otherwise we could get an easy OOM deadlock. |
a49335cc | 302 | */ |
b5227940 | 303 | if ((p->flags & PF_EXITING) && p->mm) { |
b78483a4 NP |
304 | if (p != current) |
305 | return ERR_PTR(-1UL); | |
306 | ||
972c4ea5 ON |
307 | chosen = p; |
308 | *ppoints = ULONG_MAX; | |
50ec3bbf | 309 | } |
972c4ea5 | 310 | |
28b83c51 | 311 | if (p->signal->oom_adj == OOM_DISABLE) |
0753ba01 KM |
312 | continue; |
313 | ||
97d87c97 | 314 | points = badness(p, uptime.tv_sec); |
0753ba01 | 315 | if (points > *ppoints || !chosen) { |
a49335cc | 316 | chosen = p; |
9827b781 | 317 | *ppoints = points; |
1da177e4 | 318 | } |
495789a5 | 319 | } |
972c4ea5 | 320 | |
1da177e4 LT |
321 | return chosen; |
322 | } | |
323 | ||
fef1bdd6 | 324 | /** |
1b578df0 RD |
325 | * dump_tasks - dump current memory state of all system tasks |
326 | * @mem: target memory controller | |
327 | * | |
fef1bdd6 DR |
328 | * Dumps the current memory state of all system tasks, excluding kernel threads. |
329 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj | |
330 | * score, and name. | |
331 | * | |
332 | * If the actual is non-NULL, only tasks that are a member of the mem_cgroup are | |
333 | * shown. | |
334 | * | |
335 | * Call with tasklist_lock read-locked. | |
336 | */ | |
337 | static void dump_tasks(const struct mem_cgroup *mem) | |
338 | { | |
c55db957 KM |
339 | struct task_struct *p; |
340 | struct task_struct *task; | |
fef1bdd6 DR |
341 | |
342 | printk(KERN_INFO "[ pid ] uid tgid total_vm rss cpu oom_adj " | |
343 | "name\n"); | |
c55db957 KM |
344 | for_each_process(p) { |
345 | /* | |
346 | * We don't have is_global_init() check here, because the old | |
347 | * code do that. printing init process is not big matter. But | |
348 | * we don't hope to make unnecessary compatibility breaking. | |
349 | */ | |
350 | if (p->flags & PF_KTHREAD) | |
fef1bdd6 | 351 | continue; |
c55db957 | 352 | if (mem && !task_in_mem_cgroup(p, mem)) |
b4416d2b | 353 | continue; |
fef1bdd6 | 354 | |
c55db957 KM |
355 | task = find_lock_task_mm(p); |
356 | if (!task) { | |
6d2661ed | 357 | /* |
c55db957 KM |
358 | * Probably oom vs task-exiting race was happen and ->mm |
359 | * have been detached. thus there's no need to report | |
360 | * them; they can't be oom killed anyway. | |
6d2661ed | 361 | */ |
6d2661ed DR |
362 | continue; |
363 | } | |
c55db957 | 364 | |
fef1bdd6 | 365 | printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3d %3d %s\n", |
c55db957 KM |
366 | task->pid, __task_cred(task)->uid, task->tgid, |
367 | task->mm->total_vm, get_mm_rss(task->mm), | |
368 | (int)task_cpu(task), task->signal->oom_adj, p->comm); | |
369 | task_unlock(task); | |
370 | } | |
fef1bdd6 DR |
371 | } |
372 | ||
d31f56db DN |
373 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
374 | struct mem_cgroup *mem) | |
1b604d75 DR |
375 | { |
376 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " | |
377 | "oom_adj=%d\n", | |
378 | current->comm, gfp_mask, order, current->signal->oom_adj); | |
379 | task_lock(current); | |
380 | cpuset_print_task_mems_allowed(current); | |
381 | task_unlock(current); | |
382 | dump_stack(); | |
d31f56db | 383 | mem_cgroup_print_oom_info(mem, p); |
1b604d75 DR |
384 | show_mem(); |
385 | if (sysctl_oom_dump_tasks) | |
386 | dump_tasks(mem); | |
387 | } | |
388 | ||
3b4798cb KM |
389 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
390 | ||
1b578df0 | 391 | /* |
5a291b98 RG |
392 | * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO |
393 | * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO | |
394 | * set. | |
1da177e4 | 395 | */ |
f3af38d3 | 396 | static void __oom_kill_task(struct task_struct *p, int verbose) |
1da177e4 | 397 | { |
b460cbc5 | 398 | if (is_global_init(p)) { |
1da177e4 LT |
399 | WARN_ON(1); |
400 | printk(KERN_WARNING "tried to kill init!\n"); | |
401 | return; | |
402 | } | |
403 | ||
dd8e8f40 ON |
404 | p = find_lock_task_mm(p); |
405 | if (!p) | |
1da177e4 | 406 | return; |
50ec3bbf | 407 | |
f3af38d3 | 408 | if (verbose) |
3b4798cb KM |
409 | printk(KERN_ERR "Killed process %d (%s) " |
410 | "vsz:%lukB, anon-rss:%lukB, file-rss:%lukB\n", | |
411 | task_pid_nr(p), p->comm, | |
412 | K(p->mm->total_vm), | |
d559db08 KH |
413 | K(get_mm_counter(p->mm, MM_ANONPAGES)), |
414 | K(get_mm_counter(p->mm, MM_FILEPAGES))); | |
3b4798cb | 415 | task_unlock(p); |
1da177e4 LT |
416 | |
417 | /* | |
418 | * We give our sacrificial lamb high priority and access to | |
419 | * all the memory it needs. That way it should be able to | |
420 | * exit() and clear out its resources quickly... | |
421 | */ | |
fa717060 | 422 | p->rt.time_slice = HZ; |
1da177e4 LT |
423 | set_tsk_thread_flag(p, TIF_MEMDIE); |
424 | ||
425 | force_sig(SIGKILL, p); | |
426 | } | |
427 | ||
f3af38d3 | 428 | static int oom_kill_task(struct task_struct *p) |
1da177e4 | 429 | { |
0753ba01 KM |
430 | /* WARNING: mm may not be dereferenced since we did not obtain its |
431 | * value from get_task_mm(p). This is OK since all we need to do is | |
432 | * compare mm to q->mm below. | |
433 | * | |
434 | * Furthermore, even if mm contains a non-NULL value, p->mm may | |
435 | * change to NULL at any time since we do not hold task_lock(p). | |
436 | * However, this is of no concern to us. | |
437 | */ | |
8c5cd6f3 | 438 | if (!p->mm || p->signal->oom_adj == OOM_DISABLE) |
01315922 | 439 | return 1; |
0753ba01 | 440 | |
f3af38d3 | 441 | __oom_kill_task(p, 1); |
c33e0fca | 442 | |
01315922 | 443 | return 0; |
1da177e4 LT |
444 | } |
445 | ||
7213f506 | 446 | static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
fef1bdd6 DR |
447 | unsigned long points, struct mem_cgroup *mem, |
448 | const char *message) | |
1da177e4 | 449 | { |
1da177e4 | 450 | struct task_struct *c; |
dd8e8f40 | 451 | struct task_struct *t = p; |
1da177e4 | 452 | |
1b604d75 | 453 | if (printk_ratelimit()) |
d31f56db | 454 | dump_header(p, gfp_mask, order, mem); |
7213f506 | 455 | |
50ec3bbf NP |
456 | /* |
457 | * If the task is already exiting, don't alarm the sysadmin or kill | |
458 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
459 | */ | |
0753ba01 | 460 | if (p->flags & PF_EXITING) { |
f3af38d3 | 461 | __oom_kill_task(p, 0); |
50ec3bbf NP |
462 | return 0; |
463 | } | |
464 | ||
f3af38d3 | 465 | printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n", |
ba25f9dc | 466 | message, task_pid_nr(p), p->comm, points); |
f3af38d3 | 467 | |
1da177e4 | 468 | /* Try to kill a child first */ |
dd8e8f40 ON |
469 | do { |
470 | list_for_each_entry(c, &t->children, sibling) { | |
471 | if (c->mm == p->mm) | |
472 | continue; | |
473 | if (mem && !task_in_mem_cgroup(c, mem)) | |
474 | continue; | |
475 | if (!oom_kill_task(c)) | |
476 | return 0; | |
477 | } | |
478 | } while_each_thread(p, t); | |
479 | ||
f3af38d3 | 480 | return oom_kill_task(p); |
1da177e4 LT |
481 | } |
482 | ||
00f0b825 | 483 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
c7ba5c9e PE |
484 | void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) |
485 | { | |
486 | unsigned long points = 0; | |
487 | struct task_struct *p; | |
488 | ||
daaf1e68 KH |
489 | if (sysctl_panic_on_oom == 2) |
490 | panic("out of memory(memcg). panic_on_oom is selected.\n"); | |
e115f2d8 | 491 | read_lock(&tasklist_lock); |
c7ba5c9e PE |
492 | retry: |
493 | p = select_bad_process(&points, mem); | |
df64f81b | 494 | if (!p || PTR_ERR(p) == -1UL) |
c7ba5c9e PE |
495 | goto out; |
496 | ||
fef1bdd6 | 497 | if (oom_kill_process(p, gfp_mask, 0, points, mem, |
c7ba5c9e PE |
498 | "Memory cgroup out of memory")) |
499 | goto retry; | |
500 | out: | |
e115f2d8 | 501 | read_unlock(&tasklist_lock); |
c7ba5c9e PE |
502 | } |
503 | #endif | |
504 | ||
8bc719d3 MS |
505 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
506 | ||
507 | int register_oom_notifier(struct notifier_block *nb) | |
508 | { | |
509 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
510 | } | |
511 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
512 | ||
513 | int unregister_oom_notifier(struct notifier_block *nb) | |
514 | { | |
515 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
516 | } | |
517 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
518 | ||
098d7f12 DR |
519 | /* |
520 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero | |
521 | * if a parallel OOM killing is already taking place that includes a zone in | |
522 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. | |
523 | */ | |
dd1a239f | 524 | int try_set_zone_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 525 | { |
dd1a239f MG |
526 | struct zoneref *z; |
527 | struct zone *zone; | |
098d7f12 DR |
528 | int ret = 1; |
529 | ||
c7d4caeb | 530 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
531 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
532 | if (zone_is_oom_locked(zone)) { | |
098d7f12 DR |
533 | ret = 0; |
534 | goto out; | |
535 | } | |
dd1a239f MG |
536 | } |
537 | ||
538 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
539 | /* | |
c7d4caeb | 540 | * Lock each zone in the zonelist under zone_scan_lock so a |
dd1a239f MG |
541 | * parallel invocation of try_set_zone_oom() doesn't succeed |
542 | * when it shouldn't. | |
543 | */ | |
544 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
545 | } | |
098d7f12 | 546 | |
098d7f12 | 547 | out: |
c7d4caeb | 548 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
549 | return ret; |
550 | } | |
551 | ||
552 | /* | |
553 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed | |
554 | * allocation attempts with zonelists containing them may now recall the OOM | |
555 | * killer, if necessary. | |
556 | */ | |
dd1a239f | 557 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 558 | { |
dd1a239f MG |
559 | struct zoneref *z; |
560 | struct zone *zone; | |
098d7f12 | 561 | |
c7d4caeb | 562 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
563 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
564 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
565 | } | |
c7d4caeb | 566 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
567 | } |
568 | ||
1c0fe6e3 NP |
569 | /* |
570 | * Must be called with tasklist_lock held for read. | |
571 | */ | |
572 | static void __out_of_memory(gfp_t gfp_mask, int order) | |
573 | { | |
184101bf DR |
574 | struct task_struct *p; |
575 | unsigned long points; | |
1c0fe6e3 | 576 | |
184101bf DR |
577 | if (sysctl_oom_kill_allocating_task) |
578 | if (!oom_kill_process(current, gfp_mask, order, 0, NULL, | |
579 | "Out of memory (oom_kill_allocating_task)")) | |
1c0fe6e3 | 580 | return; |
184101bf DR |
581 | retry: |
582 | /* | |
583 | * Rambo mode: Shoot down a process and hope it solves whatever | |
584 | * issues we may have. | |
585 | */ | |
586 | p = select_bad_process(&points, NULL); | |
1c0fe6e3 | 587 | |
184101bf DR |
588 | if (PTR_ERR(p) == -1UL) |
589 | return; | |
1c0fe6e3 | 590 | |
184101bf DR |
591 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
592 | if (!p) { | |
593 | read_unlock(&tasklist_lock); | |
d31f56db | 594 | dump_header(NULL, gfp_mask, order, NULL); |
184101bf | 595 | panic("Out of memory and no killable processes...\n"); |
1c0fe6e3 | 596 | } |
184101bf DR |
597 | |
598 | if (oom_kill_process(p, gfp_mask, order, points, NULL, | |
599 | "Out of memory")) | |
600 | goto retry; | |
1c0fe6e3 NP |
601 | } |
602 | ||
603 | /* | |
604 | * pagefault handler calls into here because it is out of memory but | |
605 | * doesn't know exactly how or why. | |
606 | */ | |
607 | void pagefault_out_of_memory(void) | |
608 | { | |
609 | unsigned long freed = 0; | |
610 | ||
611 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
612 | if (freed > 0) | |
613 | /* Got some memory back in the last second. */ | |
614 | return; | |
615 | ||
616 | if (sysctl_panic_on_oom) | |
617 | panic("out of memory from page fault. panic_on_oom is selected.\n"); | |
618 | ||
619 | read_lock(&tasklist_lock); | |
620 | __out_of_memory(0, 0); /* unknown gfp_mask and order */ | |
621 | read_unlock(&tasklist_lock); | |
622 | ||
623 | /* | |
624 | * Give "p" a good chance of killing itself before we | |
625 | * retry to allocate memory. | |
626 | */ | |
627 | if (!test_thread_flag(TIF_MEMDIE)) | |
628 | schedule_timeout_uninterruptible(1); | |
629 | } | |
630 | ||
1da177e4 | 631 | /** |
6937a25c | 632 | * out_of_memory - kill the "best" process when we run out of memory |
1b578df0 RD |
633 | * @zonelist: zonelist pointer |
634 | * @gfp_mask: memory allocation flags | |
635 | * @order: amount of memory being requested as a power of 2 | |
1da177e4 LT |
636 | * |
637 | * If we run out of memory, we have the choice between either | |
638 | * killing a random task (bad), letting the system crash (worse) | |
639 | * OR try to be smart about which process to kill. Note that we | |
640 | * don't have to be perfect here, we just have to be good. | |
641 | */ | |
4365a567 KH |
642 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
643 | int order, nodemask_t *nodemask) | |
1da177e4 | 644 | { |
8bc719d3 | 645 | unsigned long freed = 0; |
70e24bdf | 646 | enum oom_constraint constraint; |
8bc719d3 MS |
647 | |
648 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
649 | if (freed > 0) | |
650 | /* Got some memory back in the last second. */ | |
651 | return; | |
1da177e4 | 652 | |
1b604d75 | 653 | if (sysctl_panic_on_oom == 2) { |
d31f56db | 654 | dump_header(NULL, gfp_mask, order, NULL); |
2b744c01 | 655 | panic("out of memory. Compulsory panic_on_oom is selected.\n"); |
1b604d75 | 656 | } |
2b744c01 | 657 | |
9b0f8b04 CL |
658 | /* |
659 | * Check if there were limitations on the allocation (only relevant for | |
660 | * NUMA) that may require different handling. | |
661 | */ | |
4365a567 | 662 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask); |
2b45ab33 DR |
663 | read_lock(&tasklist_lock); |
664 | ||
665 | switch (constraint) { | |
9b0f8b04 | 666 | case CONSTRAINT_MEMORY_POLICY: |
1c0fe6e3 | 667 | oom_kill_process(current, gfp_mask, order, 0, NULL, |
9b0f8b04 CL |
668 | "No available memory (MPOL_BIND)"); |
669 | break; | |
670 | ||
9b0f8b04 | 671 | case CONSTRAINT_NONE: |
1b604d75 | 672 | if (sysctl_panic_on_oom) { |
d31f56db | 673 | dump_header(NULL, gfp_mask, order, NULL); |
fadd8fbd | 674 | panic("out of memory. panic_on_oom is selected\n"); |
1b604d75 | 675 | } |
fe071d7e DR |
676 | /* Fall-through */ |
677 | case CONSTRAINT_CPUSET: | |
1c0fe6e3 | 678 | __out_of_memory(gfp_mask, order); |
9b0f8b04 CL |
679 | break; |
680 | } | |
1da177e4 | 681 | |
140ffcec | 682 | read_unlock(&tasklist_lock); |
1da177e4 LT |
683 | |
684 | /* | |
685 | * Give "p" a good chance of killing itself before we | |
2f659f46 | 686 | * retry to allocate memory unless "p" is current |
1da177e4 | 687 | */ |
2f659f46 | 688 | if (!test_thread_flag(TIF_MEMDIE)) |
140ffcec | 689 | schedule_timeout_uninterruptible(1); |
1da177e4 | 690 | } |