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... | |
a63d83f4 DR |
7 | * Copyright (C) 2010 Google, Inc. |
8 | * Rewritten by David Rientjes | |
1da177e4 LT |
9 | * |
10 | * The routines in this file are used to kill a process when | |
a49335cc PJ |
11 | * we're seriously out of memory. This gets called from __alloc_pages() |
12 | * in mm/page_alloc.c when we really run out of memory. | |
1da177e4 LT |
13 | * |
14 | * Since we won't call these routines often (on a well-configured | |
15 | * machine) this file will double as a 'coding guide' and a signpost | |
16 | * for newbie kernel hackers. It features several pointers to major | |
17 | * kernel subsystems and hints as to where to find out what things do. | |
18 | */ | |
19 | ||
8ac773b4 | 20 | #include <linux/oom.h> |
1da177e4 | 21 | #include <linux/mm.h> |
4e950f6f | 22 | #include <linux/err.h> |
5a0e3ad6 | 23 | #include <linux/gfp.h> |
1da177e4 LT |
24 | #include <linux/sched.h> |
25 | #include <linux/swap.h> | |
26 | #include <linux/timex.h> | |
27 | #include <linux/jiffies.h> | |
ef08e3b4 | 28 | #include <linux/cpuset.h> |
8bc719d3 MS |
29 | #include <linux/module.h> |
30 | #include <linux/notifier.h> | |
c7ba5c9e | 31 | #include <linux/memcontrol.h> |
6f48d0eb | 32 | #include <linux/mempolicy.h> |
5cd9c58f | 33 | #include <linux/security.h> |
edd45544 | 34 | #include <linux/ptrace.h> |
1da177e4 | 35 | |
fadd8fbd | 36 | int sysctl_panic_on_oom; |
fe071d7e | 37 | int sysctl_oom_kill_allocating_task; |
ad915c43 | 38 | int sysctl_oom_dump_tasks = 1; |
c7d4caeb | 39 | static DEFINE_SPINLOCK(zone_scan_lock); |
1da177e4 | 40 | |
6f48d0eb DR |
41 | #ifdef CONFIG_NUMA |
42 | /** | |
43 | * has_intersects_mems_allowed() - check task eligiblity for kill | |
44 | * @tsk: task struct of which task to consider | |
45 | * @mask: nodemask passed to page allocator for mempolicy ooms | |
46 | * | |
47 | * Task eligibility is determined by whether or not a candidate task, @tsk, | |
48 | * shares the same mempolicy nodes as current if it is bound by such a policy | |
49 | * and whether or not it has the same set of allowed cpuset nodes. | |
495789a5 | 50 | */ |
6f48d0eb DR |
51 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
52 | const nodemask_t *mask) | |
495789a5 | 53 | { |
6f48d0eb | 54 | struct task_struct *start = tsk; |
495789a5 | 55 | |
495789a5 | 56 | do { |
6f48d0eb DR |
57 | if (mask) { |
58 | /* | |
59 | * If this is a mempolicy constrained oom, tsk's | |
60 | * cpuset is irrelevant. Only return true if its | |
61 | * mempolicy intersects current, otherwise it may be | |
62 | * needlessly killed. | |
63 | */ | |
64 | if (mempolicy_nodemask_intersects(tsk, mask)) | |
65 | return true; | |
66 | } else { | |
67 | /* | |
68 | * This is not a mempolicy constrained oom, so only | |
69 | * check the mems of tsk's cpuset. | |
70 | */ | |
71 | if (cpuset_mems_allowed_intersects(current, tsk)) | |
72 | return true; | |
73 | } | |
df1090a8 KM |
74 | } while_each_thread(start, tsk); |
75 | ||
6f48d0eb DR |
76 | return false; |
77 | } | |
78 | #else | |
79 | static bool has_intersects_mems_allowed(struct task_struct *tsk, | |
80 | const nodemask_t *mask) | |
81 | { | |
82 | return true; | |
495789a5 | 83 | } |
6f48d0eb | 84 | #endif /* CONFIG_NUMA */ |
495789a5 | 85 | |
93b43fa5 LCG |
86 | /* |
87 | * If this is a system OOM (not a memcg OOM) and the task selected to be | |
88 | * killed is not already running at high (RT) priorities, speed up the | |
89 | * recovery by boosting the dying task to the lowest FIFO priority. | |
90 | * That helps with the recovery and avoids interfering with RT tasks. | |
91 | */ | |
92 | static void boost_dying_task_prio(struct task_struct *p, | |
93 | struct mem_cgroup *mem) | |
94 | { | |
95 | struct sched_param param = { .sched_priority = 1 }; | |
96 | ||
97 | if (mem) | |
98 | return; | |
99 | ||
100 | if (!rt_task(p)) | |
101 | sched_setscheduler_nocheck(p, SCHED_FIFO, ¶m); | |
102 | } | |
103 | ||
6f48d0eb DR |
104 | /* |
105 | * The process p may have detached its own ->mm while exiting or through | |
106 | * use_mm(), but one or more of its subthreads may still have a valid | |
107 | * pointer. Return p, or any of its subthreads with a valid ->mm, with | |
108 | * task_lock() held. | |
109 | */ | |
158e0a2d | 110 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
dd8e8f40 ON |
111 | { |
112 | struct task_struct *t = p; | |
113 | ||
114 | do { | |
115 | task_lock(t); | |
116 | if (likely(t->mm)) | |
117 | return t; | |
118 | task_unlock(t); | |
119 | } while_each_thread(p, t); | |
120 | ||
121 | return NULL; | |
122 | } | |
123 | ||
ab290adb | 124 | /* return true if the task is not adequate as candidate victim task. */ |
e85bfd3a DR |
125 | static bool oom_unkillable_task(struct task_struct *p, |
126 | const struct mem_cgroup *mem, const nodemask_t *nodemask) | |
ab290adb KM |
127 | { |
128 | if (is_global_init(p)) | |
129 | return true; | |
130 | if (p->flags & PF_KTHREAD) | |
131 | return true; | |
132 | ||
133 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ | |
134 | if (mem && !task_in_mem_cgroup(p, mem)) | |
135 | return true; | |
136 | ||
137 | /* p may not have freeable memory in nodemask */ | |
138 | if (!has_intersects_mems_allowed(p, nodemask)) | |
139 | return true; | |
140 | ||
141 | return false; | |
142 | } | |
143 | ||
1da177e4 | 144 | /** |
a63d83f4 | 145 | * oom_badness - heuristic function to determine which candidate task to kill |
1da177e4 | 146 | * @p: task struct of which task we should calculate |
a63d83f4 | 147 | * @totalpages: total present RAM allowed for page allocation |
1da177e4 | 148 | * |
a63d83f4 DR |
149 | * The heuristic for determining which task to kill is made to be as simple and |
150 | * predictable as possible. The goal is to return the highest value for the | |
151 | * task consuming the most memory to avoid subsequent oom failures. | |
1da177e4 | 152 | */ |
a63d83f4 DR |
153 | unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem, |
154 | const nodemask_t *nodemask, unsigned long totalpages) | |
1da177e4 | 155 | { |
a63d83f4 | 156 | int points; |
28b83c51 | 157 | |
26ebc984 KM |
158 | if (oom_unkillable_task(p, mem, nodemask)) |
159 | return 0; | |
1da177e4 | 160 | |
dd8e8f40 ON |
161 | p = find_lock_task_mm(p); |
162 | if (!p) | |
1da177e4 LT |
163 | return 0; |
164 | ||
165 | /* | |
e18641e1 DR |
166 | * Shortcut check for a thread sharing p->mm that is OOM_SCORE_ADJ_MIN |
167 | * so the entire heuristic doesn't need to be executed for something | |
168 | * that cannot be killed. | |
1da177e4 | 169 | */ |
e18641e1 | 170 | if (atomic_read(&p->mm->oom_disable_count)) { |
a63d83f4 DR |
171 | task_unlock(p); |
172 | return 0; | |
173 | } | |
1da177e4 LT |
174 | |
175 | /* | |
a63d83f4 DR |
176 | * When the PF_OOM_ORIGIN bit is set, it indicates the task should have |
177 | * priority for oom killing. | |
1da177e4 | 178 | */ |
a63d83f4 DR |
179 | if (p->flags & PF_OOM_ORIGIN) { |
180 | task_unlock(p); | |
181 | return 1000; | |
182 | } | |
1da177e4 LT |
183 | |
184 | /* | |
a63d83f4 DR |
185 | * The memory controller may have a limit of 0 bytes, so avoid a divide |
186 | * by zero, if necessary. | |
1da177e4 | 187 | */ |
a63d83f4 DR |
188 | if (!totalpages) |
189 | totalpages = 1; | |
1da177e4 LT |
190 | |
191 | /* | |
a63d83f4 DR |
192 | * The baseline for the badness score is the proportion of RAM that each |
193 | * task's rss and swap space use. | |
1da177e4 | 194 | */ |
a63d83f4 DR |
195 | points = (get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS)) * 1000 / |
196 | totalpages; | |
197 | task_unlock(p); | |
1da177e4 LT |
198 | |
199 | /* | |
a63d83f4 DR |
200 | * Root processes get 3% bonus, just like the __vm_enough_memory() |
201 | * implementation used by LSMs. | |
1da177e4 | 202 | */ |
a63d83f4 DR |
203 | if (has_capability_noaudit(p, CAP_SYS_ADMIN)) |
204 | points -= 30; | |
1da177e4 LT |
205 | |
206 | /* | |
a63d83f4 DR |
207 | * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may |
208 | * either completely disable oom killing or always prefer a certain | |
209 | * task. | |
1da177e4 | 210 | */ |
a63d83f4 | 211 | points += p->signal->oom_score_adj; |
1da177e4 | 212 | |
f19e8aa1 DR |
213 | /* |
214 | * Never return 0 for an eligible task that may be killed since it's | |
215 | * possible that no single user task uses more than 0.1% of memory and | |
216 | * no single admin tasks uses more than 3.0%. | |
217 | */ | |
218 | if (points <= 0) | |
219 | return 1; | |
a63d83f4 | 220 | return (points < 1000) ? points : 1000; |
1da177e4 LT |
221 | } |
222 | ||
9b0f8b04 CL |
223 | /* |
224 | * Determine the type of allocation constraint. | |
225 | */ | |
9b0f8b04 | 226 | #ifdef CONFIG_NUMA |
4365a567 | 227 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
a63d83f4 DR |
228 | gfp_t gfp_mask, nodemask_t *nodemask, |
229 | unsigned long *totalpages) | |
4365a567 | 230 | { |
54a6eb5c | 231 | struct zone *zone; |
dd1a239f | 232 | struct zoneref *z; |
54a6eb5c | 233 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
a63d83f4 DR |
234 | bool cpuset_limited = false; |
235 | int nid; | |
9b0f8b04 | 236 | |
a63d83f4 DR |
237 | /* Default to all available memory */ |
238 | *totalpages = totalram_pages + total_swap_pages; | |
239 | ||
240 | if (!zonelist) | |
241 | return CONSTRAINT_NONE; | |
4365a567 KH |
242 | /* |
243 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid | |
244 | * to kill current.We have to random task kill in this case. | |
245 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. | |
246 | */ | |
247 | if (gfp_mask & __GFP_THISNODE) | |
248 | return CONSTRAINT_NONE; | |
9b0f8b04 | 249 | |
4365a567 | 250 | /* |
a63d83f4 DR |
251 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
252 | * the page allocator means a mempolicy is in effect. Cpuset policy | |
253 | * is enforced in get_page_from_freelist(). | |
4365a567 | 254 | */ |
a63d83f4 DR |
255 | if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) { |
256 | *totalpages = total_swap_pages; | |
257 | for_each_node_mask(nid, *nodemask) | |
258 | *totalpages += node_spanned_pages(nid); | |
9b0f8b04 | 259 | return CONSTRAINT_MEMORY_POLICY; |
a63d83f4 | 260 | } |
4365a567 KH |
261 | |
262 | /* Check this allocation failure is caused by cpuset's wall function */ | |
263 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
264 | high_zoneidx, nodemask) | |
265 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) | |
a63d83f4 | 266 | cpuset_limited = true; |
9b0f8b04 | 267 | |
a63d83f4 DR |
268 | if (cpuset_limited) { |
269 | *totalpages = total_swap_pages; | |
270 | for_each_node_mask(nid, cpuset_current_mems_allowed) | |
271 | *totalpages += node_spanned_pages(nid); | |
272 | return CONSTRAINT_CPUSET; | |
273 | } | |
9b0f8b04 CL |
274 | return CONSTRAINT_NONE; |
275 | } | |
4365a567 KH |
276 | #else |
277 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, | |
a63d83f4 DR |
278 | gfp_t gfp_mask, nodemask_t *nodemask, |
279 | unsigned long *totalpages) | |
4365a567 | 280 | { |
a63d83f4 | 281 | *totalpages = totalram_pages + total_swap_pages; |
4365a567 KH |
282 | return CONSTRAINT_NONE; |
283 | } | |
284 | #endif | |
9b0f8b04 | 285 | |
1da177e4 LT |
286 | /* |
287 | * Simple selection loop. We chose the process with the highest | |
288 | * number of 'points'. We expect the caller will lock the tasklist. | |
289 | * | |
290 | * (not docbooked, we don't want this one cluttering up the manual) | |
291 | */ | |
a63d83f4 DR |
292 | static struct task_struct *select_bad_process(unsigned int *ppoints, |
293 | unsigned long totalpages, struct mem_cgroup *mem, | |
294 | const nodemask_t *nodemask) | |
1da177e4 | 295 | { |
3a5dda7a | 296 | struct task_struct *g, *p; |
1da177e4 | 297 | struct task_struct *chosen = NULL; |
9827b781 | 298 | *ppoints = 0; |
1da177e4 | 299 | |
3a5dda7a | 300 | do_each_thread(g, p) { |
a63d83f4 | 301 | unsigned int points; |
a49335cc | 302 | |
30e2b41f AV |
303 | if (!p->mm) |
304 | continue; | |
ab290adb | 305 | if (oom_unkillable_task(p, mem, nodemask)) |
6cf86ac6 | 306 | continue; |
ef08e3b4 | 307 | |
b78483a4 NP |
308 | /* |
309 | * This task already has access to memory reserves and is | |
310 | * being killed. Don't allow any other task access to the | |
311 | * memory reserve. | |
312 | * | |
313 | * Note: this may have a chance of deadlock if it gets | |
314 | * blocked waiting for another task which itself is waiting | |
315 | * for memory. Is there a better alternative? | |
316 | */ | |
317 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) | |
318 | return ERR_PTR(-1UL); | |
319 | ||
30e2b41f | 320 | if (p->flags & PF_EXITING) { |
edd45544 DR |
321 | /* |
322 | * If p is the current task and is in the process of | |
323 | * releasing memory, we allow the "kill" to set | |
324 | * TIF_MEMDIE, which will allow it to gain access to | |
325 | * memory reserves. Otherwise, it may stall forever. | |
326 | * | |
327 | * The loop isn't broken here, however, in case other | |
328 | * threads are found to have already been oom killed. | |
329 | */ | |
330 | if (p == current) { | |
331 | chosen = p; | |
332 | *ppoints = 1000; | |
333 | } else { | |
334 | /* | |
335 | * If this task is not being ptraced on exit, | |
336 | * then wait for it to finish before killing | |
337 | * some other task unnecessarily. | |
338 | */ | |
339 | if (!(task_ptrace(p->group_leader) & | |
340 | PT_TRACE_EXIT)) | |
341 | return ERR_PTR(-1UL); | |
342 | } | |
50ec3bbf | 343 | } |
972c4ea5 | 344 | |
a63d83f4 DR |
345 | points = oom_badness(p, mem, nodemask, totalpages); |
346 | if (points > *ppoints) { | |
a49335cc | 347 | chosen = p; |
9827b781 | 348 | *ppoints = points; |
1da177e4 | 349 | } |
3a5dda7a | 350 | } while_each_thread(g, p); |
972c4ea5 | 351 | |
1da177e4 LT |
352 | return chosen; |
353 | } | |
354 | ||
fef1bdd6 | 355 | /** |
1b578df0 | 356 | * dump_tasks - dump current memory state of all system tasks |
74ab7f1d | 357 | * @mem: current's memory controller, if constrained |
e85bfd3a | 358 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
1b578df0 | 359 | * |
e85bfd3a DR |
360 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
361 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes | |
362 | * are not shown. | |
fef1bdd6 | 363 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj |
a63d83f4 | 364 | * value, oom_score_adj value, and name. |
fef1bdd6 | 365 | * |
fef1bdd6 DR |
366 | * Call with tasklist_lock read-locked. |
367 | */ | |
e85bfd3a | 368 | static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask) |
fef1bdd6 | 369 | { |
c55db957 KM |
370 | struct task_struct *p; |
371 | struct task_struct *task; | |
fef1bdd6 | 372 | |
a63d83f4 | 373 | pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n"); |
c55db957 | 374 | for_each_process(p) { |
e85bfd3a | 375 | if (oom_unkillable_task(p, mem, nodemask)) |
b4416d2b | 376 | continue; |
fef1bdd6 | 377 | |
c55db957 KM |
378 | task = find_lock_task_mm(p); |
379 | if (!task) { | |
6d2661ed | 380 | /* |
74ab7f1d DR |
381 | * This is a kthread or all of p's threads have already |
382 | * detached their mm's. There's no need to report | |
c55db957 | 383 | * them; they can't be oom killed anyway. |
6d2661ed | 384 | */ |
6d2661ed DR |
385 | continue; |
386 | } | |
c55db957 | 387 | |
a63d83f4 | 388 | pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n", |
8d6c83f0 | 389 | task->pid, task_uid(task), task->tgid, |
a63d83f4 DR |
390 | task->mm->total_vm, get_mm_rss(task->mm), |
391 | task_cpu(task), task->signal->oom_adj, | |
392 | task->signal->oom_score_adj, task->comm); | |
c55db957 KM |
393 | task_unlock(task); |
394 | } | |
fef1bdd6 DR |
395 | } |
396 | ||
d31f56db | 397 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
e85bfd3a | 398 | struct mem_cgroup *mem, const nodemask_t *nodemask) |
1b604d75 | 399 | { |
5e9d834a | 400 | task_lock(current); |
1b604d75 | 401 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " |
a63d83f4 DR |
402 | "oom_adj=%d, oom_score_adj=%d\n", |
403 | current->comm, gfp_mask, order, current->signal->oom_adj, | |
404 | current->signal->oom_score_adj); | |
1b604d75 DR |
405 | cpuset_print_task_mems_allowed(current); |
406 | task_unlock(current); | |
407 | dump_stack(); | |
d31f56db | 408 | mem_cgroup_print_oom_info(mem, p); |
b2b755b5 | 409 | show_mem(SHOW_MEM_FILTER_NODES); |
1b604d75 | 410 | if (sysctl_oom_dump_tasks) |
e85bfd3a | 411 | dump_tasks(mem, nodemask); |
1b604d75 DR |
412 | } |
413 | ||
3b4798cb | 414 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
93b43fa5 | 415 | static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem) |
1da177e4 | 416 | { |
1e99bad0 DR |
417 | struct task_struct *q; |
418 | struct mm_struct *mm; | |
419 | ||
dd8e8f40 | 420 | p = find_lock_task_mm(p); |
be71cf22 | 421 | if (!p) |
b940fd70 | 422 | return 1; |
be71cf22 | 423 | |
1e99bad0 DR |
424 | /* mm cannot be safely dereferenced after task_unlock(p) */ |
425 | mm = p->mm; | |
426 | ||
b940fd70 DR |
427 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", |
428 | task_pid_nr(p), p->comm, K(p->mm->total_vm), | |
429 | K(get_mm_counter(p->mm, MM_ANONPAGES)), | |
430 | K(get_mm_counter(p->mm, MM_FILEPAGES))); | |
3b4798cb | 431 | task_unlock(p); |
1da177e4 | 432 | |
1e99bad0 DR |
433 | /* |
434 | * Kill all processes sharing p->mm in other thread groups, if any. | |
435 | * They don't get access to memory reserves or a higher scheduler | |
436 | * priority, though, to avoid depletion of all memory or task | |
437 | * starvation. This prevents mm->mmap_sem livelock when an oom killed | |
438 | * task cannot exit because it requires the semaphore and its contended | |
439 | * by another thread trying to allocate memory itself. That thread will | |
440 | * now get access to memory reserves since it has a pending fatal | |
441 | * signal. | |
442 | */ | |
443 | for_each_process(q) | |
444 | if (q->mm == mm && !same_thread_group(q, p)) { | |
445 | task_lock(q); /* Protect ->comm from prctl() */ | |
446 | pr_err("Kill process %d (%s) sharing same memory\n", | |
447 | task_pid_nr(q), q->comm); | |
448 | task_unlock(q); | |
449 | force_sig(SIGKILL, q); | |
450 | } | |
93b43fa5 | 451 | |
1da177e4 | 452 | set_tsk_thread_flag(p, TIF_MEMDIE); |
1da177e4 | 453 | force_sig(SIGKILL, p); |
93b43fa5 LCG |
454 | |
455 | /* | |
456 | * We give our sacrificial lamb high priority and access to | |
457 | * all the memory it needs. That way it should be able to | |
458 | * exit() and clear out its resources quickly... | |
459 | */ | |
460 | boost_dying_task_prio(p, mem); | |
461 | ||
01315922 | 462 | return 0; |
1da177e4 | 463 | } |
b940fd70 | 464 | #undef K |
1da177e4 | 465 | |
7213f506 | 466 | static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
a63d83f4 DR |
467 | unsigned int points, unsigned long totalpages, |
468 | struct mem_cgroup *mem, nodemask_t *nodemask, | |
469 | const char *message) | |
1da177e4 | 470 | { |
52d3c036 | 471 | struct task_struct *victim = p; |
5e9d834a | 472 | struct task_struct *child; |
52d3c036 LT |
473 | struct task_struct *t = p; |
474 | unsigned int victim_points = 0; | |
1da177e4 | 475 | |
1b604d75 | 476 | if (printk_ratelimit()) |
e85bfd3a | 477 | dump_header(p, gfp_mask, order, mem, nodemask); |
7213f506 | 478 | |
50ec3bbf NP |
479 | /* |
480 | * If the task is already exiting, don't alarm the sysadmin or kill | |
481 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
482 | */ | |
0753ba01 | 483 | if (p->flags & PF_EXITING) { |
4358997a | 484 | set_tsk_thread_flag(p, TIF_MEMDIE); |
93b43fa5 | 485 | boost_dying_task_prio(p, mem); |
50ec3bbf NP |
486 | return 0; |
487 | } | |
488 | ||
5e9d834a | 489 | task_lock(p); |
a63d83f4 | 490 | pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n", |
5e9d834a DR |
491 | message, task_pid_nr(p), p->comm, points); |
492 | task_unlock(p); | |
f3af38d3 | 493 | |
5e9d834a DR |
494 | /* |
495 | * If any of p's children has a different mm and is eligible for kill, | |
496 | * the one with the highest badness() score is sacrificed for its | |
497 | * parent. This attempts to lose the minimal amount of work done while | |
498 | * still freeing memory. | |
499 | */ | |
dd8e8f40 | 500 | do { |
5e9d834a | 501 | list_for_each_entry(child, &t->children, sibling) { |
a63d83f4 | 502 | unsigned int child_points; |
5e9d834a | 503 | |
edd45544 DR |
504 | if (child->mm == p->mm) |
505 | continue; | |
a63d83f4 DR |
506 | /* |
507 | * oom_badness() returns 0 if the thread is unkillable | |
508 | */ | |
509 | child_points = oom_badness(child, mem, nodemask, | |
510 | totalpages); | |
5e9d834a DR |
511 | if (child_points > victim_points) { |
512 | victim = child; | |
513 | victim_points = child_points; | |
514 | } | |
dd8e8f40 ON |
515 | } |
516 | } while_each_thread(p, t); | |
517 | ||
93b43fa5 | 518 | return oom_kill_task(victim, mem); |
1da177e4 LT |
519 | } |
520 | ||
309ed882 DR |
521 | /* |
522 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. | |
523 | */ | |
524 | static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask, | |
e85bfd3a | 525 | int order, const nodemask_t *nodemask) |
309ed882 DR |
526 | { |
527 | if (likely(!sysctl_panic_on_oom)) | |
528 | return; | |
529 | if (sysctl_panic_on_oom != 2) { | |
530 | /* | |
531 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel | |
532 | * does not panic for cpuset, mempolicy, or memcg allocation | |
533 | * failures. | |
534 | */ | |
535 | if (constraint != CONSTRAINT_NONE) | |
536 | return; | |
537 | } | |
538 | read_lock(&tasklist_lock); | |
e85bfd3a | 539 | dump_header(NULL, gfp_mask, order, NULL, nodemask); |
309ed882 DR |
540 | read_unlock(&tasklist_lock); |
541 | panic("Out of memory: %s panic_on_oom is enabled\n", | |
542 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); | |
543 | } | |
544 | ||
00f0b825 | 545 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
c7ba5c9e PE |
546 | void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) |
547 | { | |
a63d83f4 DR |
548 | unsigned long limit; |
549 | unsigned int points = 0; | |
c7ba5c9e PE |
550 | struct task_struct *p; |
551 | ||
f9434ad1 DR |
552 | /* |
553 | * If current has a pending SIGKILL, then automatically select it. The | |
554 | * goal is to allow it to allocate so that it may quickly exit and free | |
555 | * its memory. | |
556 | */ | |
557 | if (fatal_signal_pending(current)) { | |
558 | set_thread_flag(TIF_MEMDIE); | |
559 | boost_dying_task_prio(current, NULL); | |
560 | return; | |
561 | } | |
562 | ||
e85bfd3a | 563 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL); |
a63d83f4 | 564 | limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT; |
e115f2d8 | 565 | read_lock(&tasklist_lock); |
c7ba5c9e | 566 | retry: |
a63d83f4 | 567 | p = select_bad_process(&points, limit, mem, NULL); |
df64f81b | 568 | if (!p || PTR_ERR(p) == -1UL) |
c7ba5c9e PE |
569 | goto out; |
570 | ||
a63d83f4 | 571 | if (oom_kill_process(p, gfp_mask, 0, points, limit, mem, NULL, |
c7ba5c9e PE |
572 | "Memory cgroup out of memory")) |
573 | goto retry; | |
574 | out: | |
e115f2d8 | 575 | read_unlock(&tasklist_lock); |
c7ba5c9e PE |
576 | } |
577 | #endif | |
578 | ||
8bc719d3 MS |
579 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
580 | ||
581 | int register_oom_notifier(struct notifier_block *nb) | |
582 | { | |
583 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
584 | } | |
585 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
586 | ||
587 | int unregister_oom_notifier(struct notifier_block *nb) | |
588 | { | |
589 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
590 | } | |
591 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
592 | ||
098d7f12 DR |
593 | /* |
594 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero | |
595 | * if a parallel OOM killing is already taking place that includes a zone in | |
596 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. | |
597 | */ | |
ff321fea | 598 | int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 599 | { |
dd1a239f MG |
600 | struct zoneref *z; |
601 | struct zone *zone; | |
098d7f12 DR |
602 | int ret = 1; |
603 | ||
c7d4caeb | 604 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
605 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
606 | if (zone_is_oom_locked(zone)) { | |
098d7f12 DR |
607 | ret = 0; |
608 | goto out; | |
609 | } | |
dd1a239f MG |
610 | } |
611 | ||
612 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
613 | /* | |
c7d4caeb | 614 | * Lock each zone in the zonelist under zone_scan_lock so a |
ff321fea | 615 | * parallel invocation of try_set_zonelist_oom() doesn't succeed |
dd1a239f MG |
616 | * when it shouldn't. |
617 | */ | |
618 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
619 | } | |
098d7f12 | 620 | |
098d7f12 | 621 | out: |
c7d4caeb | 622 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
623 | return ret; |
624 | } | |
625 | ||
626 | /* | |
627 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed | |
628 | * allocation attempts with zonelists containing them may now recall the OOM | |
629 | * killer, if necessary. | |
630 | */ | |
dd1a239f | 631 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 632 | { |
dd1a239f MG |
633 | struct zoneref *z; |
634 | struct zone *zone; | |
098d7f12 | 635 | |
c7d4caeb | 636 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
637 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
638 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
639 | } | |
c7d4caeb | 640 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
641 | } |
642 | ||
e3658932 DR |
643 | /* |
644 | * Try to acquire the oom killer lock for all system zones. Returns zero if a | |
645 | * parallel oom killing is taking place, otherwise locks all zones and returns | |
646 | * non-zero. | |
647 | */ | |
648 | static int try_set_system_oom(void) | |
649 | { | |
650 | struct zone *zone; | |
651 | int ret = 1; | |
652 | ||
653 | spin_lock(&zone_scan_lock); | |
654 | for_each_populated_zone(zone) | |
655 | if (zone_is_oom_locked(zone)) { | |
656 | ret = 0; | |
657 | goto out; | |
658 | } | |
659 | for_each_populated_zone(zone) | |
660 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
661 | out: | |
662 | spin_unlock(&zone_scan_lock); | |
663 | return ret; | |
664 | } | |
665 | ||
666 | /* | |
667 | * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation | |
668 | * attempts or page faults may now recall the oom killer, if necessary. | |
669 | */ | |
670 | static void clear_system_oom(void) | |
671 | { | |
672 | struct zone *zone; | |
673 | ||
674 | spin_lock(&zone_scan_lock); | |
675 | for_each_populated_zone(zone) | |
676 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
677 | spin_unlock(&zone_scan_lock); | |
678 | } | |
679 | ||
1da177e4 | 680 | /** |
6937a25c | 681 | * out_of_memory - kill the "best" process when we run out of memory |
1b578df0 RD |
682 | * @zonelist: zonelist pointer |
683 | * @gfp_mask: memory allocation flags | |
684 | * @order: amount of memory being requested as a power of 2 | |
6f48d0eb | 685 | * @nodemask: nodemask passed to page allocator |
1da177e4 LT |
686 | * |
687 | * If we run out of memory, we have the choice between either | |
688 | * killing a random task (bad), letting the system crash (worse) | |
689 | * OR try to be smart about which process to kill. Note that we | |
690 | * don't have to be perfect here, we just have to be good. | |
691 | */ | |
4365a567 KH |
692 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
693 | int order, nodemask_t *nodemask) | |
1da177e4 | 694 | { |
e85bfd3a | 695 | const nodemask_t *mpol_mask; |
0aad4b31 | 696 | struct task_struct *p; |
a63d83f4 | 697 | unsigned long totalpages; |
8bc719d3 | 698 | unsigned long freed = 0; |
a63d83f4 | 699 | unsigned int points; |
e3658932 | 700 | enum oom_constraint constraint = CONSTRAINT_NONE; |
b52723c5 | 701 | int killed = 0; |
8bc719d3 MS |
702 | |
703 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
704 | if (freed > 0) | |
705 | /* Got some memory back in the last second. */ | |
706 | return; | |
1da177e4 | 707 | |
7b98c2e4 DR |
708 | /* |
709 | * If current has a pending SIGKILL, then automatically select it. The | |
710 | * goal is to allow it to allocate so that it may quickly exit and free | |
711 | * its memory. | |
712 | */ | |
713 | if (fatal_signal_pending(current)) { | |
714 | set_thread_flag(TIF_MEMDIE); | |
93b43fa5 | 715 | boost_dying_task_prio(current, NULL); |
7b98c2e4 DR |
716 | return; |
717 | } | |
718 | ||
9b0f8b04 CL |
719 | /* |
720 | * Check if there were limitations on the allocation (only relevant for | |
721 | * NUMA) that may require different handling. | |
722 | */ | |
a63d83f4 DR |
723 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask, |
724 | &totalpages); | |
e85bfd3a DR |
725 | mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL; |
726 | check_panic_on_oom(constraint, gfp_mask, order, mpol_mask); | |
0aad4b31 | 727 | |
2b45ab33 | 728 | read_lock(&tasklist_lock); |
f88ccad5 | 729 | if (sysctl_oom_kill_allocating_task && |
a96cfd6e | 730 | !oom_unkillable_task(current, NULL, nodemask) && |
e18641e1 | 731 | current->mm && !atomic_read(¤t->mm->oom_disable_count)) { |
0aad4b31 DR |
732 | /* |
733 | * oom_kill_process() needs tasklist_lock held. If it returns | |
734 | * non-zero, current could not be killed so we must fallback to | |
735 | * the tasklist scan. | |
736 | */ | |
a63d83f4 DR |
737 | if (!oom_kill_process(current, gfp_mask, order, 0, totalpages, |
738 | NULL, nodemask, | |
0aad4b31 | 739 | "Out of memory (oom_kill_allocating_task)")) |
b52723c5 | 740 | goto out; |
0aad4b31 DR |
741 | } |
742 | ||
743 | retry: | |
e85bfd3a | 744 | p = select_bad_process(&points, totalpages, NULL, mpol_mask); |
0aad4b31 | 745 | if (PTR_ERR(p) == -1UL) |
b52723c5 | 746 | goto out; |
0aad4b31 DR |
747 | |
748 | /* Found nothing?!?! Either we hang forever, or we panic. */ | |
749 | if (!p) { | |
e85bfd3a | 750 | dump_header(NULL, gfp_mask, order, NULL, mpol_mask); |
0aad4b31 DR |
751 | read_unlock(&tasklist_lock); |
752 | panic("Out of memory and no killable processes...\n"); | |
753 | } | |
754 | ||
a63d83f4 DR |
755 | if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL, |
756 | nodemask, "Out of memory")) | |
0aad4b31 | 757 | goto retry; |
b52723c5 KM |
758 | killed = 1; |
759 | out: | |
140ffcec | 760 | read_unlock(&tasklist_lock); |
1da177e4 LT |
761 | |
762 | /* | |
763 | * Give "p" a good chance of killing itself before we | |
2f659f46 | 764 | * retry to allocate memory unless "p" is current |
1da177e4 | 765 | */ |
b52723c5 | 766 | if (killed && !test_thread_flag(TIF_MEMDIE)) |
140ffcec | 767 | schedule_timeout_uninterruptible(1); |
1da177e4 | 768 | } |
e3658932 DR |
769 | |
770 | /* | |
771 | * The pagefault handler calls here because it is out of memory, so kill a | |
772 | * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel | |
773 | * oom killing is already in progress so do nothing. If a task is found with | |
774 | * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit. | |
775 | */ | |
776 | void pagefault_out_of_memory(void) | |
777 | { | |
778 | if (try_set_system_oom()) { | |
779 | out_of_memory(NULL, 0, 0, NULL); | |
780 | clear_system_oom(); | |
781 | } | |
782 | if (!test_thread_flag(TIF_MEMDIE)) | |
783 | schedule_timeout_uninterruptible(1); | |
784 | } |