]>
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> |
6f48d0eb | 30 | #include <linux/mempolicy.h> |
5cd9c58f | 31 | #include <linux/security.h> |
1da177e4 | 32 | |
fadd8fbd | 33 | int sysctl_panic_on_oom; |
fe071d7e | 34 | int sysctl_oom_kill_allocating_task; |
ad915c43 | 35 | int sysctl_oom_dump_tasks = 1; |
c7d4caeb | 36 | static DEFINE_SPINLOCK(zone_scan_lock); |
1da177e4 LT |
37 | /* #define DEBUG */ |
38 | ||
6f48d0eb DR |
39 | #ifdef CONFIG_NUMA |
40 | /** | |
41 | * has_intersects_mems_allowed() - check task eligiblity for kill | |
42 | * @tsk: task struct of which task to consider | |
43 | * @mask: nodemask passed to page allocator for mempolicy ooms | |
44 | * | |
45 | * Task eligibility is determined by whether or not a candidate task, @tsk, | |
46 | * shares the same mempolicy nodes as current if it is bound by such a policy | |
47 | * and whether or not it has the same set of allowed cpuset nodes. | |
495789a5 | 48 | */ |
6f48d0eb DR |
49 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
50 | const nodemask_t *mask) | |
495789a5 | 51 | { |
6f48d0eb | 52 | struct task_struct *start = tsk; |
495789a5 | 53 | |
495789a5 | 54 | do { |
6f48d0eb DR |
55 | if (mask) { |
56 | /* | |
57 | * If this is a mempolicy constrained oom, tsk's | |
58 | * cpuset is irrelevant. Only return true if its | |
59 | * mempolicy intersects current, otherwise it may be | |
60 | * needlessly killed. | |
61 | */ | |
62 | if (mempolicy_nodemask_intersects(tsk, mask)) | |
63 | return true; | |
64 | } else { | |
65 | /* | |
66 | * This is not a mempolicy constrained oom, so only | |
67 | * check the mems of tsk's cpuset. | |
68 | */ | |
69 | if (cpuset_mems_allowed_intersects(current, tsk)) | |
70 | return true; | |
71 | } | |
df1090a8 KM |
72 | } while_each_thread(start, tsk); |
73 | ||
6f48d0eb DR |
74 | return false; |
75 | } | |
76 | #else | |
77 | static bool has_intersects_mems_allowed(struct task_struct *tsk, | |
78 | const nodemask_t *mask) | |
79 | { | |
80 | return true; | |
495789a5 | 81 | } |
6f48d0eb | 82 | #endif /* CONFIG_NUMA */ |
495789a5 | 83 | |
93b43fa5 LCG |
84 | /* |
85 | * If this is a system OOM (not a memcg OOM) and the task selected to be | |
86 | * killed is not already running at high (RT) priorities, speed up the | |
87 | * recovery by boosting the dying task to the lowest FIFO priority. | |
88 | * That helps with the recovery and avoids interfering with RT tasks. | |
89 | */ | |
90 | static void boost_dying_task_prio(struct task_struct *p, | |
91 | struct mem_cgroup *mem) | |
92 | { | |
93 | struct sched_param param = { .sched_priority = 1 }; | |
94 | ||
95 | if (mem) | |
96 | return; | |
97 | ||
98 | if (!rt_task(p)) | |
99 | sched_setscheduler_nocheck(p, SCHED_FIFO, ¶m); | |
100 | } | |
101 | ||
6f48d0eb DR |
102 | /* |
103 | * The process p may have detached its own ->mm while exiting or through | |
104 | * use_mm(), but one or more of its subthreads may still have a valid | |
105 | * pointer. Return p, or any of its subthreads with a valid ->mm, with | |
106 | * task_lock() held. | |
107 | */ | |
dd8e8f40 ON |
108 | static struct task_struct *find_lock_task_mm(struct task_struct *p) |
109 | { | |
110 | struct task_struct *t = p; | |
111 | ||
112 | do { | |
113 | task_lock(t); | |
114 | if (likely(t->mm)) | |
115 | return t; | |
116 | task_unlock(t); | |
117 | } while_each_thread(p, t); | |
118 | ||
119 | return NULL; | |
120 | } | |
121 | ||
ab290adb KM |
122 | /* return true if the task is not adequate as candidate victim task. */ |
123 | static bool oom_unkillable_task(struct task_struct *p, struct mem_cgroup *mem, | |
124 | const nodemask_t *nodemask) | |
125 | { | |
126 | if (is_global_init(p)) | |
127 | return true; | |
128 | if (p->flags & PF_KTHREAD) | |
129 | return true; | |
130 | ||
131 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ | |
132 | if (mem && !task_in_mem_cgroup(p, mem)) | |
133 | return true; | |
134 | ||
135 | /* p may not have freeable memory in nodemask */ | |
136 | if (!has_intersects_mems_allowed(p, nodemask)) | |
137 | return true; | |
138 | ||
139 | return false; | |
140 | } | |
141 | ||
1da177e4 | 142 | /** |
6937a25c | 143 | * badness - calculate a numeric value for how bad this task has been |
1da177e4 | 144 | * @p: task struct of which task we should calculate |
a49335cc | 145 | * @uptime: current uptime in seconds |
1da177e4 LT |
146 | * |
147 | * The formula used is relatively simple and documented inline in the | |
148 | * function. The main rationale is that we want to select a good task | |
149 | * to kill when we run out of memory. | |
150 | * | |
151 | * Good in this context means that: | |
152 | * 1) we lose the minimum amount of work done | |
153 | * 2) we recover a large amount of memory | |
154 | * 3) we don't kill anything innocent of eating tons of memory | |
155 | * 4) we want to kill the minimum amount of processes (one) | |
156 | * 5) we try to kill the process the user expects us to kill, this | |
157 | * algorithm has been meticulously tuned to meet the principle | |
158 | * of least surprise ... (be careful when you change it) | |
159 | */ | |
26ebc984 KM |
160 | unsigned long badness(struct task_struct *p, struct mem_cgroup *mem, |
161 | const nodemask_t *nodemask, unsigned long uptime) | |
1da177e4 | 162 | { |
a12888f7 | 163 | unsigned long points, cpu_time, run_time; |
97c2c9b8 | 164 | struct task_struct *child; |
dd8e8f40 | 165 | struct task_struct *c, *t; |
28b83c51 | 166 | int oom_adj = p->signal->oom_adj; |
495789a5 KM |
167 | struct task_cputime task_time; |
168 | unsigned long utime; | |
169 | unsigned long stime; | |
28b83c51 | 170 | |
26ebc984 KM |
171 | if (oom_unkillable_task(p, mem, nodemask)) |
172 | return 0; | |
28b83c51 KM |
173 | if (oom_adj == OOM_DISABLE) |
174 | return 0; | |
1da177e4 | 175 | |
dd8e8f40 ON |
176 | p = find_lock_task_mm(p); |
177 | if (!p) | |
1da177e4 LT |
178 | return 0; |
179 | ||
180 | /* | |
181 | * The memory size of the process is the basis for the badness. | |
182 | */ | |
dd8e8f40 | 183 | points = p->mm->total_vm; |
97c2c9b8 | 184 | task_unlock(p); |
1da177e4 | 185 | |
7ba34859 HD |
186 | /* |
187 | * swapoff can easily use up all memory, so kill those first. | |
188 | */ | |
35451bee | 189 | if (p->flags & PF_OOM_ORIGIN) |
7ba34859 HD |
190 | return ULONG_MAX; |
191 | ||
1da177e4 LT |
192 | /* |
193 | * Processes which fork a lot of child processes are likely | |
9827b781 | 194 | * a good choice. We add half the vmsize of the children if they |
1da177e4 | 195 | * have an own mm. This prevents forking servers to flood the |
9827b781 KG |
196 | * machine with an endless amount of children. In case a single |
197 | * child is eating the vast majority of memory, adding only half | |
198 | * to the parents will make the child our kill candidate of choice. | |
1da177e4 | 199 | */ |
dd8e8f40 ON |
200 | t = p; |
201 | do { | |
202 | list_for_each_entry(c, &t->children, sibling) { | |
203 | child = find_lock_task_mm(c); | |
204 | if (child) { | |
205 | if (child->mm != p->mm) | |
206 | points += child->mm->total_vm/2 + 1; | |
207 | task_unlock(child); | |
208 | } | |
209 | } | |
210 | } while_each_thread(p, t); | |
1da177e4 LT |
211 | |
212 | /* | |
213 | * CPU time is in tens of seconds and run time is in thousands | |
214 | * of seconds. There is no particular reason for this other than | |
215 | * that it turned out to work very well in practice. | |
216 | */ | |
495789a5 KM |
217 | thread_group_cputime(p, &task_time); |
218 | utime = cputime_to_jiffies(task_time.utime); | |
219 | stime = cputime_to_jiffies(task_time.stime); | |
220 | cpu_time = (utime + stime) >> (SHIFT_HZ + 3); | |
221 | ||
1da177e4 LT |
222 | |
223 | if (uptime >= p->start_time.tv_sec) | |
224 | run_time = (uptime - p->start_time.tv_sec) >> 10; | |
225 | else | |
226 | run_time = 0; | |
227 | ||
a12888f7 CG |
228 | if (cpu_time) |
229 | points /= int_sqrt(cpu_time); | |
230 | if (run_time) | |
231 | points /= int_sqrt(int_sqrt(run_time)); | |
1da177e4 LT |
232 | |
233 | /* | |
234 | * Niced processes are most likely less important, so double | |
235 | * their badness points. | |
236 | */ | |
237 | if (task_nice(p) > 0) | |
238 | points *= 2; | |
239 | ||
240 | /* | |
241 | * Superuser processes are usually more important, so we make it | |
242 | * less likely that we kill those. | |
243 | */ | |
a2f2945a EP |
244 | if (has_capability_noaudit(p, CAP_SYS_ADMIN) || |
245 | has_capability_noaudit(p, CAP_SYS_RESOURCE)) | |
1da177e4 LT |
246 | points /= 4; |
247 | ||
248 | /* | |
249 | * We don't want to kill a process with direct hardware access. | |
250 | * Not only could that mess up the hardware, but usually users | |
251 | * tend to only have this flag set on applications they think | |
252 | * of as important. | |
253 | */ | |
a2f2945a | 254 | if (has_capability_noaudit(p, CAP_SYS_RAWIO)) |
1da177e4 LT |
255 | points /= 4; |
256 | ||
257 | /* | |
28b83c51 | 258 | * Adjust the score by oom_adj. |
1da177e4 | 259 | */ |
28b83c51 KM |
260 | if (oom_adj) { |
261 | if (oom_adj > 0) { | |
9a82782f JP |
262 | if (!points) |
263 | points = 1; | |
28b83c51 | 264 | points <<= oom_adj; |
9a82782f | 265 | } else |
28b83c51 | 266 | points >>= -(oom_adj); |
1da177e4 LT |
267 | } |
268 | ||
269 | #ifdef DEBUG | |
a5e58a61 | 270 | printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n", |
1da177e4 LT |
271 | p->pid, p->comm, points); |
272 | #endif | |
273 | return points; | |
274 | } | |
275 | ||
9b0f8b04 CL |
276 | /* |
277 | * Determine the type of allocation constraint. | |
278 | */ | |
9b0f8b04 | 279 | #ifdef CONFIG_NUMA |
4365a567 KH |
280 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
281 | gfp_t gfp_mask, nodemask_t *nodemask) | |
282 | { | |
54a6eb5c | 283 | struct zone *zone; |
dd1a239f | 284 | struct zoneref *z; |
54a6eb5c | 285 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
9b0f8b04 | 286 | |
4365a567 KH |
287 | /* |
288 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid | |
289 | * to kill current.We have to random task kill in this case. | |
290 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. | |
291 | */ | |
292 | if (gfp_mask & __GFP_THISNODE) | |
293 | return CONSTRAINT_NONE; | |
9b0f8b04 | 294 | |
4365a567 KH |
295 | /* |
296 | * The nodemask here is a nodemask passed to alloc_pages(). Now, | |
297 | * cpuset doesn't use this nodemask for its hardwall/softwall/hierarchy | |
298 | * feature. mempolicy is an only user of nodemask here. | |
299 | * check mempolicy's nodemask contains all N_HIGH_MEMORY | |
300 | */ | |
301 | if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) | |
9b0f8b04 | 302 | return CONSTRAINT_MEMORY_POLICY; |
4365a567 KH |
303 | |
304 | /* Check this allocation failure is caused by cpuset's wall function */ | |
305 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
306 | high_zoneidx, nodemask) | |
307 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) | |
308 | return CONSTRAINT_CPUSET; | |
9b0f8b04 CL |
309 | |
310 | return CONSTRAINT_NONE; | |
311 | } | |
4365a567 KH |
312 | #else |
313 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, | |
314 | gfp_t gfp_mask, nodemask_t *nodemask) | |
315 | { | |
316 | return CONSTRAINT_NONE; | |
317 | } | |
318 | #endif | |
9b0f8b04 | 319 | |
1da177e4 LT |
320 | /* |
321 | * Simple selection loop. We chose the process with the highest | |
322 | * number of 'points'. We expect the caller will lock the tasklist. | |
323 | * | |
324 | * (not docbooked, we don't want this one cluttering up the manual) | |
325 | */ | |
c7ba5c9e | 326 | static struct task_struct *select_bad_process(unsigned long *ppoints, |
f4420032 | 327 | struct mem_cgroup *mem, const nodemask_t *nodemask) |
1da177e4 | 328 | { |
495789a5 | 329 | struct task_struct *p; |
1da177e4 LT |
330 | struct task_struct *chosen = NULL; |
331 | struct timespec uptime; | |
9827b781 | 332 | *ppoints = 0; |
1da177e4 LT |
333 | |
334 | do_posix_clock_monotonic_gettime(&uptime); | |
495789a5 | 335 | for_each_process(p) { |
a49335cc | 336 | unsigned long points; |
a49335cc | 337 | |
ab290adb | 338 | if (oom_unkillable_task(p, mem, nodemask)) |
6cf86ac6 | 339 | continue; |
ef08e3b4 | 340 | |
b78483a4 NP |
341 | /* |
342 | * This task already has access to memory reserves and is | |
343 | * being killed. Don't allow any other task access to the | |
344 | * memory reserve. | |
345 | * | |
346 | * Note: this may have a chance of deadlock if it gets | |
347 | * blocked waiting for another task which itself is waiting | |
348 | * for memory. Is there a better alternative? | |
349 | */ | |
350 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) | |
351 | return ERR_PTR(-1UL); | |
352 | ||
a49335cc | 353 | /* |
6937a25c | 354 | * This is in the process of releasing memory so wait for it |
a49335cc | 355 | * to finish before killing some other task by mistake. |
50ec3bbf NP |
356 | * |
357 | * However, if p is the current task, we allow the 'kill' to | |
358 | * go ahead if it is exiting: this will simply set TIF_MEMDIE, | |
359 | * which will allow it to gain access to memory reserves in | |
360 | * the process of exiting and releasing its resources. | |
b78483a4 | 361 | * Otherwise we could get an easy OOM deadlock. |
a49335cc | 362 | */ |
b5227940 | 363 | if ((p->flags & PF_EXITING) && p->mm) { |
b78483a4 NP |
364 | if (p != current) |
365 | return ERR_PTR(-1UL); | |
366 | ||
972c4ea5 ON |
367 | chosen = p; |
368 | *ppoints = ULONG_MAX; | |
50ec3bbf | 369 | } |
972c4ea5 | 370 | |
26ebc984 | 371 | points = badness(p, mem, nodemask, uptime.tv_sec); |
0753ba01 | 372 | if (points > *ppoints || !chosen) { |
a49335cc | 373 | chosen = p; |
9827b781 | 374 | *ppoints = points; |
1da177e4 | 375 | } |
495789a5 | 376 | } |
972c4ea5 | 377 | |
1da177e4 LT |
378 | return chosen; |
379 | } | |
380 | ||
fef1bdd6 | 381 | /** |
1b578df0 | 382 | * dump_tasks - dump current memory state of all system tasks |
74ab7f1d | 383 | * @mem: current's memory controller, if constrained |
1b578df0 | 384 | * |
fef1bdd6 DR |
385 | * Dumps the current memory state of all system tasks, excluding kernel threads. |
386 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj | |
387 | * score, and name. | |
388 | * | |
389 | * If the actual is non-NULL, only tasks that are a member of the mem_cgroup are | |
390 | * shown. | |
391 | * | |
392 | * Call with tasklist_lock read-locked. | |
393 | */ | |
394 | static void dump_tasks(const struct mem_cgroup *mem) | |
395 | { | |
c55db957 KM |
396 | struct task_struct *p; |
397 | struct task_struct *task; | |
fef1bdd6 DR |
398 | |
399 | printk(KERN_INFO "[ pid ] uid tgid total_vm rss cpu oom_adj " | |
400 | "name\n"); | |
c55db957 | 401 | for_each_process(p) { |
c55db957 | 402 | if (p->flags & PF_KTHREAD) |
fef1bdd6 | 403 | continue; |
c55db957 | 404 | if (mem && !task_in_mem_cgroup(p, mem)) |
b4416d2b | 405 | continue; |
fef1bdd6 | 406 | |
c55db957 KM |
407 | task = find_lock_task_mm(p); |
408 | if (!task) { | |
6d2661ed | 409 | /* |
74ab7f1d DR |
410 | * This is a kthread or all of p's threads have already |
411 | * detached their mm's. There's no need to report | |
c55db957 | 412 | * them; they can't be oom killed anyway. |
6d2661ed | 413 | */ |
6d2661ed DR |
414 | continue; |
415 | } | |
c55db957 | 416 | |
c81fac5c | 417 | printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3u %3d %s\n", |
c55db957 KM |
418 | task->pid, __task_cred(task)->uid, task->tgid, |
419 | task->mm->total_vm, get_mm_rss(task->mm), | |
c81fac5c | 420 | task_cpu(task), task->signal->oom_adj, task->comm); |
c55db957 KM |
421 | task_unlock(task); |
422 | } | |
fef1bdd6 DR |
423 | } |
424 | ||
d31f56db DN |
425 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
426 | struct mem_cgroup *mem) | |
1b604d75 | 427 | { |
5e9d834a | 428 | task_lock(current); |
1b604d75 DR |
429 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " |
430 | "oom_adj=%d\n", | |
431 | current->comm, gfp_mask, order, current->signal->oom_adj); | |
1b604d75 DR |
432 | cpuset_print_task_mems_allowed(current); |
433 | task_unlock(current); | |
434 | dump_stack(); | |
d31f56db | 435 | mem_cgroup_print_oom_info(mem, p); |
1b604d75 DR |
436 | show_mem(); |
437 | if (sysctl_oom_dump_tasks) | |
438 | dump_tasks(mem); | |
439 | } | |
440 | ||
3b4798cb | 441 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
93b43fa5 | 442 | static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem) |
1da177e4 | 443 | { |
dd8e8f40 | 444 | p = find_lock_task_mm(p); |
a96cfd6e | 445 | if (!p) { |
b940fd70 DR |
446 | task_unlock(p); |
447 | return 1; | |
448 | } | |
449 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", | |
450 | task_pid_nr(p), p->comm, K(p->mm->total_vm), | |
451 | K(get_mm_counter(p->mm, MM_ANONPAGES)), | |
452 | K(get_mm_counter(p->mm, MM_FILEPAGES))); | |
3b4798cb | 453 | task_unlock(p); |
1da177e4 | 454 | |
93b43fa5 | 455 | |
1da177e4 | 456 | set_tsk_thread_flag(p, TIF_MEMDIE); |
1da177e4 | 457 | force_sig(SIGKILL, p); |
93b43fa5 LCG |
458 | |
459 | /* | |
460 | * We give our sacrificial lamb high priority and access to | |
461 | * all the memory it needs. That way it should be able to | |
462 | * exit() and clear out its resources quickly... | |
463 | */ | |
464 | boost_dying_task_prio(p, mem); | |
465 | ||
01315922 | 466 | return 0; |
1da177e4 | 467 | } |
b940fd70 | 468 | #undef K |
1da177e4 | 469 | |
7213f506 | 470 | static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
fef1bdd6 | 471 | unsigned long points, struct mem_cgroup *mem, |
7c59aec8 | 472 | nodemask_t *nodemask, const char *message) |
1da177e4 | 473 | { |
5e9d834a DR |
474 | struct task_struct *victim = p; |
475 | struct task_struct *child; | |
dd8e8f40 | 476 | struct task_struct *t = p; |
5e9d834a DR |
477 | unsigned long victim_points = 0; |
478 | struct timespec uptime; | |
1da177e4 | 479 | |
1b604d75 | 480 | if (printk_ratelimit()) |
d31f56db | 481 | dump_header(p, gfp_mask, order, mem); |
7213f506 | 482 | |
50ec3bbf NP |
483 | /* |
484 | * If the task is already exiting, don't alarm the sysadmin or kill | |
485 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
486 | */ | |
0753ba01 | 487 | if (p->flags & PF_EXITING) { |
4358997a | 488 | set_tsk_thread_flag(p, TIF_MEMDIE); |
93b43fa5 | 489 | boost_dying_task_prio(p, mem); |
50ec3bbf NP |
490 | return 0; |
491 | } | |
492 | ||
5e9d834a DR |
493 | task_lock(p); |
494 | pr_err("%s: Kill process %d (%s) score %lu or sacrifice child\n", | |
495 | message, task_pid_nr(p), p->comm, points); | |
496 | task_unlock(p); | |
f3af38d3 | 497 | |
5e9d834a DR |
498 | /* |
499 | * If any of p's children has a different mm and is eligible for kill, | |
500 | * the one with the highest badness() score is sacrificed for its | |
501 | * parent. This attempts to lose the minimal amount of work done while | |
502 | * still freeing memory. | |
503 | */ | |
504 | do_posix_clock_monotonic_gettime(&uptime); | |
dd8e8f40 | 505 | do { |
5e9d834a DR |
506 | list_for_each_entry(child, &t->children, sibling) { |
507 | unsigned long child_points; | |
508 | ||
5e9d834a | 509 | /* badness() returns 0 if the thread is unkillable */ |
26ebc984 KM |
510 | child_points = badness(child, mem, nodemask, |
511 | uptime.tv_sec); | |
5e9d834a DR |
512 | if (child_points > victim_points) { |
513 | victim = child; | |
514 | victim_points = child_points; | |
515 | } | |
dd8e8f40 ON |
516 | } |
517 | } while_each_thread(p, t); | |
518 | ||
93b43fa5 | 519 | return oom_kill_task(victim, mem); |
1da177e4 LT |
520 | } |
521 | ||
309ed882 DR |
522 | /* |
523 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. | |
524 | */ | |
525 | static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask, | |
526 | int order) | |
527 | { | |
528 | if (likely(!sysctl_panic_on_oom)) | |
529 | return; | |
530 | if (sysctl_panic_on_oom != 2) { | |
531 | /* | |
532 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel | |
533 | * does not panic for cpuset, mempolicy, or memcg allocation | |
534 | * failures. | |
535 | */ | |
536 | if (constraint != CONSTRAINT_NONE) | |
537 | return; | |
538 | } | |
539 | read_lock(&tasklist_lock); | |
540 | dump_header(NULL, gfp_mask, order, NULL); | |
541 | read_unlock(&tasklist_lock); | |
542 | panic("Out of memory: %s panic_on_oom is enabled\n", | |
543 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); | |
544 | } | |
545 | ||
00f0b825 | 546 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
c7ba5c9e PE |
547 | void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) |
548 | { | |
549 | unsigned long points = 0; | |
550 | struct task_struct *p; | |
551 | ||
309ed882 | 552 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0); |
e115f2d8 | 553 | read_lock(&tasklist_lock); |
c7ba5c9e | 554 | retry: |
f4420032 | 555 | p = select_bad_process(&points, mem, NULL); |
df64f81b | 556 | if (!p || PTR_ERR(p) == -1UL) |
c7ba5c9e PE |
557 | goto out; |
558 | ||
7c59aec8 | 559 | if (oom_kill_process(p, gfp_mask, 0, points, mem, NULL, |
c7ba5c9e PE |
560 | "Memory cgroup out of memory")) |
561 | goto retry; | |
562 | out: | |
e115f2d8 | 563 | read_unlock(&tasklist_lock); |
c7ba5c9e PE |
564 | } |
565 | #endif | |
566 | ||
8bc719d3 MS |
567 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
568 | ||
569 | int register_oom_notifier(struct notifier_block *nb) | |
570 | { | |
571 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
572 | } | |
573 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
574 | ||
575 | int unregister_oom_notifier(struct notifier_block *nb) | |
576 | { | |
577 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
578 | } | |
579 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
580 | ||
098d7f12 DR |
581 | /* |
582 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero | |
583 | * if a parallel OOM killing is already taking place that includes a zone in | |
584 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. | |
585 | */ | |
ff321fea | 586 | int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 587 | { |
dd1a239f MG |
588 | struct zoneref *z; |
589 | struct zone *zone; | |
098d7f12 DR |
590 | int ret = 1; |
591 | ||
c7d4caeb | 592 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
593 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
594 | if (zone_is_oom_locked(zone)) { | |
098d7f12 DR |
595 | ret = 0; |
596 | goto out; | |
597 | } | |
dd1a239f MG |
598 | } |
599 | ||
600 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
601 | /* | |
c7d4caeb | 602 | * Lock each zone in the zonelist under zone_scan_lock so a |
ff321fea | 603 | * parallel invocation of try_set_zonelist_oom() doesn't succeed |
dd1a239f MG |
604 | * when it shouldn't. |
605 | */ | |
606 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
607 | } | |
098d7f12 | 608 | |
098d7f12 | 609 | out: |
c7d4caeb | 610 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
611 | return ret; |
612 | } | |
613 | ||
614 | /* | |
615 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed | |
616 | * allocation attempts with zonelists containing them may now recall the OOM | |
617 | * killer, if necessary. | |
618 | */ | |
dd1a239f | 619 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 620 | { |
dd1a239f MG |
621 | struct zoneref *z; |
622 | struct zone *zone; | |
098d7f12 | 623 | |
c7d4caeb | 624 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
625 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
626 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
627 | } | |
c7d4caeb | 628 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
629 | } |
630 | ||
e3658932 DR |
631 | /* |
632 | * Try to acquire the oom killer lock for all system zones. Returns zero if a | |
633 | * parallel oom killing is taking place, otherwise locks all zones and returns | |
634 | * non-zero. | |
635 | */ | |
636 | static int try_set_system_oom(void) | |
637 | { | |
638 | struct zone *zone; | |
639 | int ret = 1; | |
640 | ||
641 | spin_lock(&zone_scan_lock); | |
642 | for_each_populated_zone(zone) | |
643 | if (zone_is_oom_locked(zone)) { | |
644 | ret = 0; | |
645 | goto out; | |
646 | } | |
647 | for_each_populated_zone(zone) | |
648 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
649 | out: | |
650 | spin_unlock(&zone_scan_lock); | |
651 | return ret; | |
652 | } | |
653 | ||
654 | /* | |
655 | * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation | |
656 | * attempts or page faults may now recall the oom killer, if necessary. | |
657 | */ | |
658 | static void clear_system_oom(void) | |
659 | { | |
660 | struct zone *zone; | |
661 | ||
662 | spin_lock(&zone_scan_lock); | |
663 | for_each_populated_zone(zone) | |
664 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
665 | spin_unlock(&zone_scan_lock); | |
666 | } | |
667 | ||
1da177e4 | 668 | /** |
6937a25c | 669 | * out_of_memory - kill the "best" process when we run out of memory |
1b578df0 RD |
670 | * @zonelist: zonelist pointer |
671 | * @gfp_mask: memory allocation flags | |
672 | * @order: amount of memory being requested as a power of 2 | |
6f48d0eb | 673 | * @nodemask: nodemask passed to page allocator |
1da177e4 LT |
674 | * |
675 | * If we run out of memory, we have the choice between either | |
676 | * killing a random task (bad), letting the system crash (worse) | |
677 | * OR try to be smart about which process to kill. Note that we | |
678 | * don't have to be perfect here, we just have to be good. | |
679 | */ | |
4365a567 KH |
680 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
681 | int order, nodemask_t *nodemask) | |
1da177e4 | 682 | { |
0aad4b31 | 683 | struct task_struct *p; |
8bc719d3 | 684 | unsigned long freed = 0; |
0aad4b31 | 685 | unsigned long points; |
e3658932 | 686 | enum oom_constraint constraint = CONSTRAINT_NONE; |
8bc719d3 MS |
687 | |
688 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
689 | if (freed > 0) | |
690 | /* Got some memory back in the last second. */ | |
691 | return; | |
1da177e4 | 692 | |
7b98c2e4 DR |
693 | /* |
694 | * If current has a pending SIGKILL, then automatically select it. The | |
695 | * goal is to allow it to allocate so that it may quickly exit and free | |
696 | * its memory. | |
697 | */ | |
698 | if (fatal_signal_pending(current)) { | |
699 | set_thread_flag(TIF_MEMDIE); | |
93b43fa5 | 700 | boost_dying_task_prio(current, NULL); |
7b98c2e4 DR |
701 | return; |
702 | } | |
703 | ||
9b0f8b04 CL |
704 | /* |
705 | * Check if there were limitations on the allocation (only relevant for | |
706 | * NUMA) that may require different handling. | |
707 | */ | |
e3658932 DR |
708 | if (zonelist) |
709 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask); | |
309ed882 | 710 | check_panic_on_oom(constraint, gfp_mask, order); |
0aad4b31 | 711 | |
2b45ab33 | 712 | read_lock(&tasklist_lock); |
f88ccad5 | 713 | if (sysctl_oom_kill_allocating_task && |
a96cfd6e KM |
714 | !oom_unkillable_task(current, NULL, nodemask) && |
715 | (current->signal->oom_adj != OOM_DISABLE)) { | |
0aad4b31 DR |
716 | /* |
717 | * oom_kill_process() needs tasklist_lock held. If it returns | |
718 | * non-zero, current could not be killed so we must fallback to | |
719 | * the tasklist scan. | |
720 | */ | |
721 | if (!oom_kill_process(current, gfp_mask, order, 0, NULL, | |
7c59aec8 | 722 | nodemask, |
0aad4b31 DR |
723 | "Out of memory (oom_kill_allocating_task)")) |
724 | return; | |
725 | } | |
726 | ||
727 | retry: | |
728 | p = select_bad_process(&points, NULL, | |
f4420032 DR |
729 | constraint == CONSTRAINT_MEMORY_POLICY ? nodemask : |
730 | NULL); | |
0aad4b31 DR |
731 | if (PTR_ERR(p) == -1UL) |
732 | return; | |
733 | ||
734 | /* Found nothing?!?! Either we hang forever, or we panic. */ | |
735 | if (!p) { | |
736 | dump_header(NULL, gfp_mask, order, NULL); | |
737 | read_unlock(&tasklist_lock); | |
738 | panic("Out of memory and no killable processes...\n"); | |
739 | } | |
740 | ||
7c59aec8 | 741 | if (oom_kill_process(p, gfp_mask, order, points, NULL, nodemask, |
0aad4b31 DR |
742 | "Out of memory")) |
743 | goto retry; | |
140ffcec | 744 | read_unlock(&tasklist_lock); |
1da177e4 LT |
745 | |
746 | /* | |
747 | * Give "p" a good chance of killing itself before we | |
2f659f46 | 748 | * retry to allocate memory unless "p" is current |
1da177e4 | 749 | */ |
2f659f46 | 750 | if (!test_thread_flag(TIF_MEMDIE)) |
140ffcec | 751 | schedule_timeout_uninterruptible(1); |
1da177e4 | 752 | } |
e3658932 DR |
753 | |
754 | /* | |
755 | * The pagefault handler calls here because it is out of memory, so kill a | |
756 | * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel | |
757 | * oom killing is already in progress so do nothing. If a task is found with | |
758 | * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit. | |
759 | */ | |
760 | void pagefault_out_of_memory(void) | |
761 | { | |
762 | if (try_set_system_oom()) { | |
763 | out_of_memory(NULL, 0, 0, NULL); | |
764 | clear_system_oom(); | |
765 | } | |
766 | if (!test_thread_flag(TIF_MEMDIE)) | |
767 | schedule_timeout_uninterruptible(1); | |
768 | } |