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> |
f660daac | 35 | #include <linux/freezer.h> |
1da177e4 | 36 | |
fadd8fbd | 37 | int sysctl_panic_on_oom; |
fe071d7e | 38 | int sysctl_oom_kill_allocating_task; |
ad915c43 | 39 | int sysctl_oom_dump_tasks = 1; |
c7d4caeb | 40 | static DEFINE_SPINLOCK(zone_scan_lock); |
1da177e4 | 41 | |
72788c38 DR |
42 | /** |
43 | * test_set_oom_score_adj() - set current's oom_score_adj and return old value | |
44 | * @new_val: new oom_score_adj value | |
45 | * | |
46 | * Sets the oom_score_adj value for current to @new_val with proper | |
47 | * synchronization and returns the old value. Usually used to temporarily | |
48 | * set a value, save the old value in the caller, and then reinstate it later. | |
49 | */ | |
50 | int test_set_oom_score_adj(int new_val) | |
51 | { | |
52 | struct sighand_struct *sighand = current->sighand; | |
53 | int old_val; | |
54 | ||
55 | spin_lock_irq(&sighand->siglock); | |
56 | old_val = current->signal->oom_score_adj; | |
57 | if (new_val != old_val) { | |
58 | if (new_val == OOM_SCORE_ADJ_MIN) | |
59 | atomic_inc(¤t->mm->oom_disable_count); | |
60 | else if (old_val == OOM_SCORE_ADJ_MIN) | |
61 | atomic_dec(¤t->mm->oom_disable_count); | |
62 | current->signal->oom_score_adj = new_val; | |
63 | } | |
64 | spin_unlock_irq(&sighand->siglock); | |
65 | ||
66 | return old_val; | |
67 | } | |
68 | ||
6f48d0eb DR |
69 | #ifdef CONFIG_NUMA |
70 | /** | |
71 | * has_intersects_mems_allowed() - check task eligiblity for kill | |
72 | * @tsk: task struct of which task to consider | |
73 | * @mask: nodemask passed to page allocator for mempolicy ooms | |
74 | * | |
75 | * Task eligibility is determined by whether or not a candidate task, @tsk, | |
76 | * shares the same mempolicy nodes as current if it is bound by such a policy | |
77 | * and whether or not it has the same set of allowed cpuset nodes. | |
495789a5 | 78 | */ |
6f48d0eb DR |
79 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
80 | const nodemask_t *mask) | |
495789a5 | 81 | { |
6f48d0eb | 82 | struct task_struct *start = tsk; |
495789a5 | 83 | |
495789a5 | 84 | do { |
6f48d0eb DR |
85 | if (mask) { |
86 | /* | |
87 | * If this is a mempolicy constrained oom, tsk's | |
88 | * cpuset is irrelevant. Only return true if its | |
89 | * mempolicy intersects current, otherwise it may be | |
90 | * needlessly killed. | |
91 | */ | |
92 | if (mempolicy_nodemask_intersects(tsk, mask)) | |
93 | return true; | |
94 | } else { | |
95 | /* | |
96 | * This is not a mempolicy constrained oom, so only | |
97 | * check the mems of tsk's cpuset. | |
98 | */ | |
99 | if (cpuset_mems_allowed_intersects(current, tsk)) | |
100 | return true; | |
101 | } | |
df1090a8 KM |
102 | } while_each_thread(start, tsk); |
103 | ||
6f48d0eb DR |
104 | return false; |
105 | } | |
106 | #else | |
107 | static bool has_intersects_mems_allowed(struct task_struct *tsk, | |
108 | const nodemask_t *mask) | |
109 | { | |
110 | return true; | |
495789a5 | 111 | } |
6f48d0eb | 112 | #endif /* CONFIG_NUMA */ |
495789a5 | 113 | |
6f48d0eb DR |
114 | /* |
115 | * The process p may have detached its own ->mm while exiting or through | |
116 | * use_mm(), but one or more of its subthreads may still have a valid | |
117 | * pointer. Return p, or any of its subthreads with a valid ->mm, with | |
118 | * task_lock() held. | |
119 | */ | |
158e0a2d | 120 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
dd8e8f40 ON |
121 | { |
122 | struct task_struct *t = p; | |
123 | ||
124 | do { | |
125 | task_lock(t); | |
126 | if (likely(t->mm)) | |
127 | return t; | |
128 | task_unlock(t); | |
129 | } while_each_thread(p, t); | |
130 | ||
131 | return NULL; | |
132 | } | |
133 | ||
ab290adb | 134 | /* return true if the task is not adequate as candidate victim task. */ |
e85bfd3a DR |
135 | static bool oom_unkillable_task(struct task_struct *p, |
136 | const struct mem_cgroup *mem, const nodemask_t *nodemask) | |
ab290adb KM |
137 | { |
138 | if (is_global_init(p)) | |
139 | return true; | |
140 | if (p->flags & PF_KTHREAD) | |
141 | return true; | |
142 | ||
143 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ | |
144 | if (mem && !task_in_mem_cgroup(p, mem)) | |
145 | return true; | |
146 | ||
147 | /* p may not have freeable memory in nodemask */ | |
148 | if (!has_intersects_mems_allowed(p, nodemask)) | |
149 | return true; | |
150 | ||
151 | return false; | |
152 | } | |
153 | ||
1da177e4 | 154 | /** |
a63d83f4 | 155 | * oom_badness - heuristic function to determine which candidate task to kill |
1da177e4 | 156 | * @p: task struct of which task we should calculate |
a63d83f4 | 157 | * @totalpages: total present RAM allowed for page allocation |
1da177e4 | 158 | * |
a63d83f4 DR |
159 | * The heuristic for determining which task to kill is made to be as simple and |
160 | * predictable as possible. The goal is to return the highest value for the | |
161 | * task consuming the most memory to avoid subsequent oom failures. | |
1da177e4 | 162 | */ |
a63d83f4 DR |
163 | unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem, |
164 | const nodemask_t *nodemask, unsigned long totalpages) | |
1da177e4 | 165 | { |
a63d83f4 | 166 | int points; |
28b83c51 | 167 | |
26ebc984 KM |
168 | if (oom_unkillable_task(p, mem, nodemask)) |
169 | return 0; | |
1da177e4 | 170 | |
dd8e8f40 ON |
171 | p = find_lock_task_mm(p); |
172 | if (!p) | |
1da177e4 LT |
173 | return 0; |
174 | ||
175 | /* | |
e18641e1 DR |
176 | * Shortcut check for a thread sharing p->mm that is OOM_SCORE_ADJ_MIN |
177 | * so the entire heuristic doesn't need to be executed for something | |
178 | * that cannot be killed. | |
1da177e4 | 179 | */ |
e18641e1 | 180 | if (atomic_read(&p->mm->oom_disable_count)) { |
a63d83f4 DR |
181 | task_unlock(p); |
182 | return 0; | |
183 | } | |
1da177e4 | 184 | |
1da177e4 | 185 | /* |
a63d83f4 DR |
186 | * The memory controller may have a limit of 0 bytes, so avoid a divide |
187 | * by zero, if necessary. | |
1da177e4 | 188 | */ |
a63d83f4 DR |
189 | if (!totalpages) |
190 | totalpages = 1; | |
1da177e4 LT |
191 | |
192 | /* | |
a63d83f4 | 193 | * The baseline for the badness score is the proportion of RAM that each |
f755a042 | 194 | * task's rss, pagetable and swap space use. |
1da177e4 | 195 | */ |
f755a042 KM |
196 | points = get_mm_rss(p->mm) + p->mm->nr_ptes; |
197 | points += get_mm_counter(p->mm, MM_SWAPENTS); | |
198 | ||
199 | points *= 1000; | |
200 | points /= totalpages; | |
a63d83f4 | 201 | task_unlock(p); |
1da177e4 LT |
202 | |
203 | /* | |
a63d83f4 DR |
204 | * Root processes get 3% bonus, just like the __vm_enough_memory() |
205 | * implementation used by LSMs. | |
1da177e4 | 206 | */ |
a63d83f4 DR |
207 | if (has_capability_noaudit(p, CAP_SYS_ADMIN)) |
208 | points -= 30; | |
1da177e4 LT |
209 | |
210 | /* | |
a63d83f4 DR |
211 | * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may |
212 | * either completely disable oom killing or always prefer a certain | |
213 | * task. | |
1da177e4 | 214 | */ |
a63d83f4 | 215 | points += p->signal->oom_score_adj; |
1da177e4 | 216 | |
f19e8aa1 DR |
217 | /* |
218 | * Never return 0 for an eligible task that may be killed since it's | |
219 | * possible that no single user task uses more than 0.1% of memory and | |
220 | * no single admin tasks uses more than 3.0%. | |
221 | */ | |
222 | if (points <= 0) | |
223 | return 1; | |
a63d83f4 | 224 | return (points < 1000) ? points : 1000; |
1da177e4 LT |
225 | } |
226 | ||
9b0f8b04 CL |
227 | /* |
228 | * Determine the type of allocation constraint. | |
229 | */ | |
9b0f8b04 | 230 | #ifdef CONFIG_NUMA |
4365a567 | 231 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
a63d83f4 DR |
232 | gfp_t gfp_mask, nodemask_t *nodemask, |
233 | unsigned long *totalpages) | |
4365a567 | 234 | { |
54a6eb5c | 235 | struct zone *zone; |
dd1a239f | 236 | struct zoneref *z; |
54a6eb5c | 237 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
a63d83f4 DR |
238 | bool cpuset_limited = false; |
239 | int nid; | |
9b0f8b04 | 240 | |
a63d83f4 DR |
241 | /* Default to all available memory */ |
242 | *totalpages = totalram_pages + total_swap_pages; | |
243 | ||
244 | if (!zonelist) | |
245 | return CONSTRAINT_NONE; | |
4365a567 KH |
246 | /* |
247 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid | |
248 | * to kill current.We have to random task kill in this case. | |
249 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. | |
250 | */ | |
251 | if (gfp_mask & __GFP_THISNODE) | |
252 | return CONSTRAINT_NONE; | |
9b0f8b04 | 253 | |
4365a567 | 254 | /* |
a63d83f4 DR |
255 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
256 | * the page allocator means a mempolicy is in effect. Cpuset policy | |
257 | * is enforced in get_page_from_freelist(). | |
4365a567 | 258 | */ |
a63d83f4 DR |
259 | if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) { |
260 | *totalpages = total_swap_pages; | |
261 | for_each_node_mask(nid, *nodemask) | |
262 | *totalpages += node_spanned_pages(nid); | |
9b0f8b04 | 263 | return CONSTRAINT_MEMORY_POLICY; |
a63d83f4 | 264 | } |
4365a567 KH |
265 | |
266 | /* Check this allocation failure is caused by cpuset's wall function */ | |
267 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
268 | high_zoneidx, nodemask) | |
269 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) | |
a63d83f4 | 270 | cpuset_limited = true; |
9b0f8b04 | 271 | |
a63d83f4 DR |
272 | if (cpuset_limited) { |
273 | *totalpages = total_swap_pages; | |
274 | for_each_node_mask(nid, cpuset_current_mems_allowed) | |
275 | *totalpages += node_spanned_pages(nid); | |
276 | return CONSTRAINT_CPUSET; | |
277 | } | |
9b0f8b04 CL |
278 | return CONSTRAINT_NONE; |
279 | } | |
4365a567 KH |
280 | #else |
281 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, | |
a63d83f4 DR |
282 | gfp_t gfp_mask, nodemask_t *nodemask, |
283 | unsigned long *totalpages) | |
4365a567 | 284 | { |
a63d83f4 | 285 | *totalpages = totalram_pages + total_swap_pages; |
4365a567 KH |
286 | return CONSTRAINT_NONE; |
287 | } | |
288 | #endif | |
9b0f8b04 | 289 | |
1da177e4 LT |
290 | /* |
291 | * Simple selection loop. We chose the process with the highest | |
292 | * number of 'points'. We expect the caller will lock the tasklist. | |
293 | * | |
294 | * (not docbooked, we don't want this one cluttering up the manual) | |
295 | */ | |
a63d83f4 DR |
296 | static struct task_struct *select_bad_process(unsigned int *ppoints, |
297 | unsigned long totalpages, struct mem_cgroup *mem, | |
298 | const nodemask_t *nodemask) | |
1da177e4 | 299 | { |
3a5dda7a | 300 | struct task_struct *g, *p; |
1da177e4 | 301 | struct task_struct *chosen = NULL; |
9827b781 | 302 | *ppoints = 0; |
1da177e4 | 303 | |
3a5dda7a | 304 | do_each_thread(g, p) { |
a63d83f4 | 305 | unsigned int points; |
a49335cc | 306 | |
c027a474 | 307 | if (p->exit_state) |
30e2b41f | 308 | continue; |
ab290adb | 309 | if (oom_unkillable_task(p, mem, nodemask)) |
6cf86ac6 | 310 | continue; |
ef08e3b4 | 311 | |
b78483a4 NP |
312 | /* |
313 | * This task already has access to memory reserves and is | |
314 | * being killed. Don't allow any other task access to the | |
315 | * memory reserve. | |
316 | * | |
317 | * Note: this may have a chance of deadlock if it gets | |
318 | * blocked waiting for another task which itself is waiting | |
319 | * for memory. Is there a better alternative? | |
320 | */ | |
f660daac DR |
321 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) { |
322 | if (unlikely(frozen(p))) | |
323 | thaw_process(p); | |
b78483a4 | 324 | return ERR_PTR(-1UL); |
f660daac | 325 | } |
c027a474 ON |
326 | if (!p->mm) |
327 | continue; | |
b78483a4 | 328 | |
30e2b41f | 329 | if (p->flags & PF_EXITING) { |
edd45544 DR |
330 | /* |
331 | * If p is the current task and is in the process of | |
332 | * releasing memory, we allow the "kill" to set | |
333 | * TIF_MEMDIE, which will allow it to gain access to | |
334 | * memory reserves. Otherwise, it may stall forever. | |
335 | * | |
336 | * The loop isn't broken here, however, in case other | |
337 | * threads are found to have already been oom killed. | |
338 | */ | |
339 | if (p == current) { | |
340 | chosen = p; | |
341 | *ppoints = 1000; | |
342 | } else { | |
343 | /* | |
344 | * If this task is not being ptraced on exit, | |
345 | * then wait for it to finish before killing | |
346 | * some other task unnecessarily. | |
347 | */ | |
d21142ec | 348 | if (!(p->group_leader->ptrace & PT_TRACE_EXIT)) |
edd45544 DR |
349 | return ERR_PTR(-1UL); |
350 | } | |
50ec3bbf | 351 | } |
972c4ea5 | 352 | |
a63d83f4 DR |
353 | points = oom_badness(p, mem, nodemask, totalpages); |
354 | if (points > *ppoints) { | |
a49335cc | 355 | chosen = p; |
9827b781 | 356 | *ppoints = points; |
1da177e4 | 357 | } |
3a5dda7a | 358 | } while_each_thread(g, p); |
972c4ea5 | 359 | |
1da177e4 LT |
360 | return chosen; |
361 | } | |
362 | ||
fef1bdd6 | 363 | /** |
1b578df0 | 364 | * dump_tasks - dump current memory state of all system tasks |
74ab7f1d | 365 | * @mem: current's memory controller, if constrained |
e85bfd3a | 366 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
1b578df0 | 367 | * |
e85bfd3a DR |
368 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
369 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes | |
370 | * are not shown. | |
fef1bdd6 | 371 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj |
a63d83f4 | 372 | * value, oom_score_adj value, and name. |
fef1bdd6 | 373 | * |
fef1bdd6 DR |
374 | * Call with tasklist_lock read-locked. |
375 | */ | |
e85bfd3a | 376 | static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask) |
fef1bdd6 | 377 | { |
c55db957 KM |
378 | struct task_struct *p; |
379 | struct task_struct *task; | |
fef1bdd6 | 380 | |
a63d83f4 | 381 | pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n"); |
c55db957 | 382 | for_each_process(p) { |
e85bfd3a | 383 | if (oom_unkillable_task(p, mem, nodemask)) |
b4416d2b | 384 | continue; |
fef1bdd6 | 385 | |
c55db957 KM |
386 | task = find_lock_task_mm(p); |
387 | if (!task) { | |
6d2661ed | 388 | /* |
74ab7f1d DR |
389 | * This is a kthread or all of p's threads have already |
390 | * detached their mm's. There's no need to report | |
c55db957 | 391 | * them; they can't be oom killed anyway. |
6d2661ed | 392 | */ |
6d2661ed DR |
393 | continue; |
394 | } | |
c55db957 | 395 | |
a63d83f4 | 396 | pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n", |
8d6c83f0 | 397 | task->pid, task_uid(task), task->tgid, |
a63d83f4 DR |
398 | task->mm->total_vm, get_mm_rss(task->mm), |
399 | task_cpu(task), task->signal->oom_adj, | |
400 | task->signal->oom_score_adj, task->comm); | |
c55db957 KM |
401 | task_unlock(task); |
402 | } | |
fef1bdd6 DR |
403 | } |
404 | ||
d31f56db | 405 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
e85bfd3a | 406 | struct mem_cgroup *mem, const nodemask_t *nodemask) |
1b604d75 | 407 | { |
5e9d834a | 408 | task_lock(current); |
1b604d75 | 409 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " |
a63d83f4 DR |
410 | "oom_adj=%d, oom_score_adj=%d\n", |
411 | current->comm, gfp_mask, order, current->signal->oom_adj, | |
412 | current->signal->oom_score_adj); | |
1b604d75 DR |
413 | cpuset_print_task_mems_allowed(current); |
414 | task_unlock(current); | |
415 | dump_stack(); | |
d31f56db | 416 | mem_cgroup_print_oom_info(mem, p); |
b2b755b5 | 417 | show_mem(SHOW_MEM_FILTER_NODES); |
1b604d75 | 418 | if (sysctl_oom_dump_tasks) |
e85bfd3a | 419 | dump_tasks(mem, nodemask); |
1b604d75 DR |
420 | } |
421 | ||
3b4798cb | 422 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
93b43fa5 | 423 | static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem) |
1da177e4 | 424 | { |
1e99bad0 DR |
425 | struct task_struct *q; |
426 | struct mm_struct *mm; | |
427 | ||
dd8e8f40 | 428 | p = find_lock_task_mm(p); |
be71cf22 | 429 | if (!p) |
b940fd70 | 430 | return 1; |
be71cf22 | 431 | |
1e99bad0 DR |
432 | /* mm cannot be safely dereferenced after task_unlock(p) */ |
433 | mm = p->mm; | |
434 | ||
b940fd70 DR |
435 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", |
436 | task_pid_nr(p), p->comm, K(p->mm->total_vm), | |
437 | K(get_mm_counter(p->mm, MM_ANONPAGES)), | |
438 | K(get_mm_counter(p->mm, MM_FILEPAGES))); | |
3b4798cb | 439 | task_unlock(p); |
1da177e4 | 440 | |
1e99bad0 DR |
441 | /* |
442 | * Kill all processes sharing p->mm in other thread groups, if any. | |
443 | * They don't get access to memory reserves or a higher scheduler | |
444 | * priority, though, to avoid depletion of all memory or task | |
445 | * starvation. This prevents mm->mmap_sem livelock when an oom killed | |
446 | * task cannot exit because it requires the semaphore and its contended | |
447 | * by another thread trying to allocate memory itself. That thread will | |
448 | * now get access to memory reserves since it has a pending fatal | |
449 | * signal. | |
450 | */ | |
451 | for_each_process(q) | |
452 | if (q->mm == mm && !same_thread_group(q, p)) { | |
453 | task_lock(q); /* Protect ->comm from prctl() */ | |
454 | pr_err("Kill process %d (%s) sharing same memory\n", | |
455 | task_pid_nr(q), q->comm); | |
456 | task_unlock(q); | |
457 | force_sig(SIGKILL, q); | |
458 | } | |
93b43fa5 | 459 | |
1da177e4 | 460 | set_tsk_thread_flag(p, TIF_MEMDIE); |
1da177e4 | 461 | force_sig(SIGKILL, p); |
93b43fa5 | 462 | |
01315922 | 463 | return 0; |
1da177e4 | 464 | } |
b940fd70 | 465 | #undef K |
1da177e4 | 466 | |
7213f506 | 467 | static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
a63d83f4 DR |
468 | unsigned int points, unsigned long totalpages, |
469 | struct mem_cgroup *mem, nodemask_t *nodemask, | |
470 | const char *message) | |
1da177e4 | 471 | { |
52d3c036 | 472 | struct task_struct *victim = p; |
5e9d834a | 473 | struct task_struct *child; |
52d3c036 LT |
474 | struct task_struct *t = p; |
475 | unsigned int victim_points = 0; | |
1da177e4 | 476 | |
1b604d75 | 477 | if (printk_ratelimit()) |
e85bfd3a | 478 | dump_header(p, gfp_mask, order, mem, nodemask); |
7213f506 | 479 | |
50ec3bbf NP |
480 | /* |
481 | * If the task is already exiting, don't alarm the sysadmin or kill | |
482 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
483 | */ | |
0753ba01 | 484 | if (p->flags & PF_EXITING) { |
4358997a | 485 | set_tsk_thread_flag(p, TIF_MEMDIE); |
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, | |
11239836 | 496 | * the one with the highest oom_badness() score is sacrificed for its |
5e9d834a DR |
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); | |
f9434ad1 DR |
559 | return; |
560 | } | |
561 | ||
e85bfd3a | 562 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL); |
a63d83f4 | 563 | limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT; |
e115f2d8 | 564 | read_lock(&tasklist_lock); |
c7ba5c9e | 565 | retry: |
a63d83f4 | 566 | p = select_bad_process(&points, limit, mem, NULL); |
df64f81b | 567 | if (!p || PTR_ERR(p) == -1UL) |
c7ba5c9e PE |
568 | goto out; |
569 | ||
a63d83f4 | 570 | if (oom_kill_process(p, gfp_mask, 0, points, limit, mem, NULL, |
c7ba5c9e PE |
571 | "Memory cgroup out of memory")) |
572 | goto retry; | |
573 | out: | |
e115f2d8 | 574 | read_unlock(&tasklist_lock); |
c7ba5c9e PE |
575 | } |
576 | #endif | |
577 | ||
8bc719d3 MS |
578 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
579 | ||
580 | int register_oom_notifier(struct notifier_block *nb) | |
581 | { | |
582 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
583 | } | |
584 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
585 | ||
586 | int unregister_oom_notifier(struct notifier_block *nb) | |
587 | { | |
588 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
589 | } | |
590 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
591 | ||
098d7f12 DR |
592 | /* |
593 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero | |
594 | * if a parallel OOM killing is already taking place that includes a zone in | |
595 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. | |
596 | */ | |
ff321fea | 597 | int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 598 | { |
dd1a239f MG |
599 | struct zoneref *z; |
600 | struct zone *zone; | |
098d7f12 DR |
601 | int ret = 1; |
602 | ||
c7d4caeb | 603 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
604 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
605 | if (zone_is_oom_locked(zone)) { | |
098d7f12 DR |
606 | ret = 0; |
607 | goto out; | |
608 | } | |
dd1a239f MG |
609 | } |
610 | ||
611 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
612 | /* | |
c7d4caeb | 613 | * Lock each zone in the zonelist under zone_scan_lock so a |
ff321fea | 614 | * parallel invocation of try_set_zonelist_oom() doesn't succeed |
dd1a239f MG |
615 | * when it shouldn't. |
616 | */ | |
617 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
618 | } | |
098d7f12 | 619 | |
098d7f12 | 620 | out: |
c7d4caeb | 621 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
622 | return ret; |
623 | } | |
624 | ||
625 | /* | |
626 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed | |
627 | * allocation attempts with zonelists containing them may now recall the OOM | |
628 | * killer, if necessary. | |
629 | */ | |
dd1a239f | 630 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 631 | { |
dd1a239f MG |
632 | struct zoneref *z; |
633 | struct zone *zone; | |
098d7f12 | 634 | |
c7d4caeb | 635 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
636 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
637 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
638 | } | |
c7d4caeb | 639 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
640 | } |
641 | ||
e3658932 DR |
642 | /* |
643 | * Try to acquire the oom killer lock for all system zones. Returns zero if a | |
644 | * parallel oom killing is taking place, otherwise locks all zones and returns | |
645 | * non-zero. | |
646 | */ | |
647 | static int try_set_system_oom(void) | |
648 | { | |
649 | struct zone *zone; | |
650 | int ret = 1; | |
651 | ||
652 | spin_lock(&zone_scan_lock); | |
653 | for_each_populated_zone(zone) | |
654 | if (zone_is_oom_locked(zone)) { | |
655 | ret = 0; | |
656 | goto out; | |
657 | } | |
658 | for_each_populated_zone(zone) | |
659 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
660 | out: | |
661 | spin_unlock(&zone_scan_lock); | |
662 | return ret; | |
663 | } | |
664 | ||
665 | /* | |
666 | * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation | |
667 | * attempts or page faults may now recall the oom killer, if necessary. | |
668 | */ | |
669 | static void clear_system_oom(void) | |
670 | { | |
671 | struct zone *zone; | |
672 | ||
673 | spin_lock(&zone_scan_lock); | |
674 | for_each_populated_zone(zone) | |
675 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
676 | spin_unlock(&zone_scan_lock); | |
677 | } | |
678 | ||
1da177e4 | 679 | /** |
6937a25c | 680 | * out_of_memory - kill the "best" process when we run out of memory |
1b578df0 RD |
681 | * @zonelist: zonelist pointer |
682 | * @gfp_mask: memory allocation flags | |
683 | * @order: amount of memory being requested as a power of 2 | |
6f48d0eb | 684 | * @nodemask: nodemask passed to page allocator |
1da177e4 LT |
685 | * |
686 | * If we run out of memory, we have the choice between either | |
687 | * killing a random task (bad), letting the system crash (worse) | |
688 | * OR try to be smart about which process to kill. Note that we | |
689 | * don't have to be perfect here, we just have to be good. | |
690 | */ | |
4365a567 KH |
691 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
692 | int order, nodemask_t *nodemask) | |
1da177e4 | 693 | { |
e85bfd3a | 694 | const nodemask_t *mpol_mask; |
0aad4b31 | 695 | struct task_struct *p; |
a63d83f4 | 696 | unsigned long totalpages; |
8bc719d3 | 697 | unsigned long freed = 0; |
a63d83f4 | 698 | unsigned int points; |
e3658932 | 699 | enum oom_constraint constraint = CONSTRAINT_NONE; |
b52723c5 | 700 | int killed = 0; |
8bc719d3 MS |
701 | |
702 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
703 | if (freed > 0) | |
704 | /* Got some memory back in the last second. */ | |
705 | return; | |
1da177e4 | 706 | |
7b98c2e4 DR |
707 | /* |
708 | * If current has a pending SIGKILL, then automatically select it. The | |
709 | * goal is to allow it to allocate so that it may quickly exit and free | |
710 | * its memory. | |
711 | */ | |
712 | if (fatal_signal_pending(current)) { | |
713 | set_thread_flag(TIF_MEMDIE); | |
714 | return; | |
715 | } | |
716 | ||
9b0f8b04 CL |
717 | /* |
718 | * Check if there were limitations on the allocation (only relevant for | |
719 | * NUMA) that may require different handling. | |
720 | */ | |
a63d83f4 DR |
721 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask, |
722 | &totalpages); | |
e85bfd3a DR |
723 | mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL; |
724 | check_panic_on_oom(constraint, gfp_mask, order, mpol_mask); | |
0aad4b31 | 725 | |
2b45ab33 | 726 | read_lock(&tasklist_lock); |
f88ccad5 | 727 | if (sysctl_oom_kill_allocating_task && |
a96cfd6e | 728 | !oom_unkillable_task(current, NULL, nodemask) && |
e18641e1 | 729 | current->mm && !atomic_read(¤t->mm->oom_disable_count)) { |
0aad4b31 DR |
730 | /* |
731 | * oom_kill_process() needs tasklist_lock held. If it returns | |
732 | * non-zero, current could not be killed so we must fallback to | |
733 | * the tasklist scan. | |
734 | */ | |
a63d83f4 DR |
735 | if (!oom_kill_process(current, gfp_mask, order, 0, totalpages, |
736 | NULL, nodemask, | |
0aad4b31 | 737 | "Out of memory (oom_kill_allocating_task)")) |
b52723c5 | 738 | goto out; |
0aad4b31 DR |
739 | } |
740 | ||
741 | retry: | |
e85bfd3a | 742 | p = select_bad_process(&points, totalpages, NULL, mpol_mask); |
0aad4b31 | 743 | if (PTR_ERR(p) == -1UL) |
b52723c5 | 744 | goto out; |
0aad4b31 DR |
745 | |
746 | /* Found nothing?!?! Either we hang forever, or we panic. */ | |
747 | if (!p) { | |
e85bfd3a | 748 | dump_header(NULL, gfp_mask, order, NULL, mpol_mask); |
0aad4b31 DR |
749 | read_unlock(&tasklist_lock); |
750 | panic("Out of memory and no killable processes...\n"); | |
751 | } | |
752 | ||
a63d83f4 DR |
753 | if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL, |
754 | nodemask, "Out of memory")) | |
0aad4b31 | 755 | goto retry; |
b52723c5 KM |
756 | killed = 1; |
757 | out: | |
140ffcec | 758 | read_unlock(&tasklist_lock); |
1da177e4 LT |
759 | |
760 | /* | |
761 | * Give "p" a good chance of killing itself before we | |
2f659f46 | 762 | * retry to allocate memory unless "p" is current |
1da177e4 | 763 | */ |
b52723c5 | 764 | if (killed && !test_thread_flag(TIF_MEMDIE)) |
140ffcec | 765 | schedule_timeout_uninterruptible(1); |
1da177e4 | 766 | } |
e3658932 DR |
767 | |
768 | /* | |
769 | * The pagefault handler calls here because it is out of memory, so kill a | |
770 | * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel | |
771 | * oom killing is already in progress so do nothing. If a task is found with | |
772 | * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit. | |
773 | */ | |
774 | void pagefault_out_of_memory(void) | |
775 | { | |
776 | if (try_set_system_oom()) { | |
777 | out_of_memory(NULL, 0, 0, NULL); | |
778 | clear_system_oom(); | |
779 | } | |
780 | if (!test_thread_flag(TIF_MEMDIE)) | |
781 | schedule_timeout_uninterruptible(1); | |
782 | } |