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 | |
6f48d0eb DR |
86 | /* |
87 | * The process p may have detached its own ->mm while exiting or through | |
88 | * use_mm(), but one or more of its subthreads may still have a valid | |
89 | * pointer. Return p, or any of its subthreads with a valid ->mm, with | |
90 | * task_lock() held. | |
91 | */ | |
158e0a2d | 92 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
dd8e8f40 ON |
93 | { |
94 | struct task_struct *t = p; | |
95 | ||
96 | do { | |
97 | task_lock(t); | |
98 | if (likely(t->mm)) | |
99 | return t; | |
100 | task_unlock(t); | |
101 | } while_each_thread(p, t); | |
102 | ||
103 | return NULL; | |
104 | } | |
105 | ||
ab290adb | 106 | /* return true if the task is not adequate as candidate victim task. */ |
e85bfd3a DR |
107 | static bool oom_unkillable_task(struct task_struct *p, |
108 | const struct mem_cgroup *mem, const nodemask_t *nodemask) | |
ab290adb KM |
109 | { |
110 | if (is_global_init(p)) | |
111 | return true; | |
112 | if (p->flags & PF_KTHREAD) | |
113 | return true; | |
114 | ||
115 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ | |
116 | if (mem && !task_in_mem_cgroup(p, mem)) | |
117 | return true; | |
118 | ||
119 | /* p may not have freeable memory in nodemask */ | |
120 | if (!has_intersects_mems_allowed(p, nodemask)) | |
121 | return true; | |
122 | ||
123 | return false; | |
124 | } | |
125 | ||
1da177e4 | 126 | /** |
a63d83f4 | 127 | * oom_badness - heuristic function to determine which candidate task to kill |
1da177e4 | 128 | * @p: task struct of which task we should calculate |
a63d83f4 | 129 | * @totalpages: total present RAM allowed for page allocation |
1da177e4 | 130 | * |
a63d83f4 DR |
131 | * The heuristic for determining which task to kill is made to be as simple and |
132 | * predictable as possible. The goal is to return the highest value for the | |
133 | * task consuming the most memory to avoid subsequent oom failures. | |
1da177e4 | 134 | */ |
a63d83f4 DR |
135 | unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem, |
136 | const nodemask_t *nodemask, unsigned long totalpages) | |
1da177e4 | 137 | { |
a63d83f4 | 138 | int points; |
28b83c51 | 139 | |
26ebc984 KM |
140 | if (oom_unkillable_task(p, mem, nodemask)) |
141 | return 0; | |
1da177e4 | 142 | |
dd8e8f40 ON |
143 | p = find_lock_task_mm(p); |
144 | if (!p) | |
1da177e4 LT |
145 | return 0; |
146 | ||
147 | /* | |
e18641e1 DR |
148 | * Shortcut check for a thread sharing p->mm that is OOM_SCORE_ADJ_MIN |
149 | * so the entire heuristic doesn't need to be executed for something | |
150 | * that cannot be killed. | |
1da177e4 | 151 | */ |
e18641e1 | 152 | if (atomic_read(&p->mm->oom_disable_count)) { |
a63d83f4 DR |
153 | task_unlock(p); |
154 | return 0; | |
155 | } | |
1da177e4 LT |
156 | |
157 | /* | |
a63d83f4 DR |
158 | * When the PF_OOM_ORIGIN bit is set, it indicates the task should have |
159 | * priority for oom killing. | |
1da177e4 | 160 | */ |
a63d83f4 DR |
161 | if (p->flags & PF_OOM_ORIGIN) { |
162 | task_unlock(p); | |
163 | return 1000; | |
164 | } | |
1da177e4 LT |
165 | |
166 | /* | |
a63d83f4 DR |
167 | * The memory controller may have a limit of 0 bytes, so avoid a divide |
168 | * by zero, if necessary. | |
1da177e4 | 169 | */ |
a63d83f4 DR |
170 | if (!totalpages) |
171 | totalpages = 1; | |
1da177e4 LT |
172 | |
173 | /* | |
a63d83f4 | 174 | * The baseline for the badness score is the proportion of RAM that each |
f755a042 | 175 | * task's rss, pagetable and swap space use. |
1da177e4 | 176 | */ |
f755a042 KM |
177 | points = get_mm_rss(p->mm) + p->mm->nr_ptes; |
178 | points += get_mm_counter(p->mm, MM_SWAPENTS); | |
179 | ||
180 | points *= 1000; | |
181 | points /= totalpages; | |
a63d83f4 | 182 | task_unlock(p); |
1da177e4 LT |
183 | |
184 | /* | |
a63d83f4 DR |
185 | * Root processes get 3% bonus, just like the __vm_enough_memory() |
186 | * implementation used by LSMs. | |
1da177e4 | 187 | */ |
a63d83f4 DR |
188 | if (has_capability_noaudit(p, CAP_SYS_ADMIN)) |
189 | points -= 30; | |
1da177e4 LT |
190 | |
191 | /* | |
a63d83f4 DR |
192 | * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may |
193 | * either completely disable oom killing or always prefer a certain | |
194 | * task. | |
1da177e4 | 195 | */ |
a63d83f4 | 196 | points += p->signal->oom_score_adj; |
1da177e4 | 197 | |
f19e8aa1 DR |
198 | /* |
199 | * Never return 0 for an eligible task that may be killed since it's | |
200 | * possible that no single user task uses more than 0.1% of memory and | |
201 | * no single admin tasks uses more than 3.0%. | |
202 | */ | |
203 | if (points <= 0) | |
204 | return 1; | |
a63d83f4 | 205 | return (points < 1000) ? points : 1000; |
1da177e4 LT |
206 | } |
207 | ||
9b0f8b04 CL |
208 | /* |
209 | * Determine the type of allocation constraint. | |
210 | */ | |
9b0f8b04 | 211 | #ifdef CONFIG_NUMA |
4365a567 | 212 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
a63d83f4 DR |
213 | gfp_t gfp_mask, nodemask_t *nodemask, |
214 | unsigned long *totalpages) | |
4365a567 | 215 | { |
54a6eb5c | 216 | struct zone *zone; |
dd1a239f | 217 | struct zoneref *z; |
54a6eb5c | 218 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
a63d83f4 DR |
219 | bool cpuset_limited = false; |
220 | int nid; | |
9b0f8b04 | 221 | |
a63d83f4 DR |
222 | /* Default to all available memory */ |
223 | *totalpages = totalram_pages + total_swap_pages; | |
224 | ||
225 | if (!zonelist) | |
226 | return CONSTRAINT_NONE; | |
4365a567 KH |
227 | /* |
228 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid | |
229 | * to kill current.We have to random task kill in this case. | |
230 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. | |
231 | */ | |
232 | if (gfp_mask & __GFP_THISNODE) | |
233 | return CONSTRAINT_NONE; | |
9b0f8b04 | 234 | |
4365a567 | 235 | /* |
a63d83f4 DR |
236 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
237 | * the page allocator means a mempolicy is in effect. Cpuset policy | |
238 | * is enforced in get_page_from_freelist(). | |
4365a567 | 239 | */ |
a63d83f4 DR |
240 | if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) { |
241 | *totalpages = total_swap_pages; | |
242 | for_each_node_mask(nid, *nodemask) | |
243 | *totalpages += node_spanned_pages(nid); | |
9b0f8b04 | 244 | return CONSTRAINT_MEMORY_POLICY; |
a63d83f4 | 245 | } |
4365a567 KH |
246 | |
247 | /* Check this allocation failure is caused by cpuset's wall function */ | |
248 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
249 | high_zoneidx, nodemask) | |
250 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) | |
a63d83f4 | 251 | cpuset_limited = true; |
9b0f8b04 | 252 | |
a63d83f4 DR |
253 | if (cpuset_limited) { |
254 | *totalpages = total_swap_pages; | |
255 | for_each_node_mask(nid, cpuset_current_mems_allowed) | |
256 | *totalpages += node_spanned_pages(nid); | |
257 | return CONSTRAINT_CPUSET; | |
258 | } | |
9b0f8b04 CL |
259 | return CONSTRAINT_NONE; |
260 | } | |
4365a567 KH |
261 | #else |
262 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, | |
a63d83f4 DR |
263 | gfp_t gfp_mask, nodemask_t *nodemask, |
264 | unsigned long *totalpages) | |
4365a567 | 265 | { |
a63d83f4 | 266 | *totalpages = totalram_pages + total_swap_pages; |
4365a567 KH |
267 | return CONSTRAINT_NONE; |
268 | } | |
269 | #endif | |
9b0f8b04 | 270 | |
1da177e4 LT |
271 | /* |
272 | * Simple selection loop. We chose the process with the highest | |
273 | * number of 'points'. We expect the caller will lock the tasklist. | |
274 | * | |
275 | * (not docbooked, we don't want this one cluttering up the manual) | |
276 | */ | |
a63d83f4 DR |
277 | static struct task_struct *select_bad_process(unsigned int *ppoints, |
278 | unsigned long totalpages, struct mem_cgroup *mem, | |
279 | const nodemask_t *nodemask) | |
1da177e4 | 280 | { |
3a5dda7a | 281 | struct task_struct *g, *p; |
1da177e4 | 282 | struct task_struct *chosen = NULL; |
9827b781 | 283 | *ppoints = 0; |
1da177e4 | 284 | |
3a5dda7a | 285 | do_each_thread(g, p) { |
a63d83f4 | 286 | unsigned int points; |
a49335cc | 287 | |
30e2b41f AV |
288 | if (!p->mm) |
289 | continue; | |
ab290adb | 290 | if (oom_unkillable_task(p, mem, nodemask)) |
6cf86ac6 | 291 | continue; |
ef08e3b4 | 292 | |
b78483a4 NP |
293 | /* |
294 | * This task already has access to memory reserves and is | |
295 | * being killed. Don't allow any other task access to the | |
296 | * memory reserve. | |
297 | * | |
298 | * Note: this may have a chance of deadlock if it gets | |
299 | * blocked waiting for another task which itself is waiting | |
300 | * for memory. Is there a better alternative? | |
301 | */ | |
302 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) | |
303 | return ERR_PTR(-1UL); | |
304 | ||
30e2b41f | 305 | if (p->flags & PF_EXITING) { |
edd45544 DR |
306 | /* |
307 | * If p is the current task and is in the process of | |
308 | * releasing memory, we allow the "kill" to set | |
309 | * TIF_MEMDIE, which will allow it to gain access to | |
310 | * memory reserves. Otherwise, it may stall forever. | |
311 | * | |
312 | * The loop isn't broken here, however, in case other | |
313 | * threads are found to have already been oom killed. | |
314 | */ | |
315 | if (p == current) { | |
316 | chosen = p; | |
317 | *ppoints = 1000; | |
318 | } else { | |
319 | /* | |
320 | * If this task is not being ptraced on exit, | |
321 | * then wait for it to finish before killing | |
322 | * some other task unnecessarily. | |
323 | */ | |
324 | if (!(task_ptrace(p->group_leader) & | |
325 | PT_TRACE_EXIT)) | |
326 | return ERR_PTR(-1UL); | |
327 | } | |
50ec3bbf | 328 | } |
972c4ea5 | 329 | |
a63d83f4 DR |
330 | points = oom_badness(p, mem, nodemask, totalpages); |
331 | if (points > *ppoints) { | |
a49335cc | 332 | chosen = p; |
9827b781 | 333 | *ppoints = points; |
1da177e4 | 334 | } |
3a5dda7a | 335 | } while_each_thread(g, p); |
972c4ea5 | 336 | |
1da177e4 LT |
337 | return chosen; |
338 | } | |
339 | ||
fef1bdd6 | 340 | /** |
1b578df0 | 341 | * dump_tasks - dump current memory state of all system tasks |
74ab7f1d | 342 | * @mem: current's memory controller, if constrained |
e85bfd3a | 343 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
1b578df0 | 344 | * |
e85bfd3a DR |
345 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
346 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes | |
347 | * are not shown. | |
fef1bdd6 | 348 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj |
a63d83f4 | 349 | * value, oom_score_adj value, and name. |
fef1bdd6 | 350 | * |
fef1bdd6 DR |
351 | * Call with tasklist_lock read-locked. |
352 | */ | |
e85bfd3a | 353 | static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask) |
fef1bdd6 | 354 | { |
c55db957 KM |
355 | struct task_struct *p; |
356 | struct task_struct *task; | |
fef1bdd6 | 357 | |
a63d83f4 | 358 | pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n"); |
c55db957 | 359 | for_each_process(p) { |
e85bfd3a | 360 | if (oom_unkillable_task(p, mem, nodemask)) |
b4416d2b | 361 | continue; |
fef1bdd6 | 362 | |
c55db957 KM |
363 | task = find_lock_task_mm(p); |
364 | if (!task) { | |
6d2661ed | 365 | /* |
74ab7f1d DR |
366 | * This is a kthread or all of p's threads have already |
367 | * detached their mm's. There's no need to report | |
c55db957 | 368 | * them; they can't be oom killed anyway. |
6d2661ed | 369 | */ |
6d2661ed DR |
370 | continue; |
371 | } | |
c55db957 | 372 | |
a63d83f4 | 373 | pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n", |
8d6c83f0 | 374 | task->pid, task_uid(task), task->tgid, |
a63d83f4 DR |
375 | task->mm->total_vm, get_mm_rss(task->mm), |
376 | task_cpu(task), task->signal->oom_adj, | |
377 | task->signal->oom_score_adj, task->comm); | |
c55db957 KM |
378 | task_unlock(task); |
379 | } | |
fef1bdd6 DR |
380 | } |
381 | ||
d31f56db | 382 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
e85bfd3a | 383 | struct mem_cgroup *mem, const nodemask_t *nodemask) |
1b604d75 | 384 | { |
5e9d834a | 385 | task_lock(current); |
1b604d75 | 386 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " |
a63d83f4 DR |
387 | "oom_adj=%d, oom_score_adj=%d\n", |
388 | current->comm, gfp_mask, order, current->signal->oom_adj, | |
389 | current->signal->oom_score_adj); | |
1b604d75 DR |
390 | cpuset_print_task_mems_allowed(current); |
391 | task_unlock(current); | |
392 | dump_stack(); | |
d31f56db | 393 | mem_cgroup_print_oom_info(mem, p); |
b2b755b5 | 394 | show_mem(SHOW_MEM_FILTER_NODES); |
1b604d75 | 395 | if (sysctl_oom_dump_tasks) |
e85bfd3a | 396 | dump_tasks(mem, nodemask); |
1b604d75 DR |
397 | } |
398 | ||
3b4798cb | 399 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
93b43fa5 | 400 | static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem) |
1da177e4 | 401 | { |
1e99bad0 DR |
402 | struct task_struct *q; |
403 | struct mm_struct *mm; | |
404 | ||
dd8e8f40 | 405 | p = find_lock_task_mm(p); |
be71cf22 | 406 | if (!p) |
b940fd70 | 407 | return 1; |
be71cf22 | 408 | |
1e99bad0 DR |
409 | /* mm cannot be safely dereferenced after task_unlock(p) */ |
410 | mm = p->mm; | |
411 | ||
b940fd70 DR |
412 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", |
413 | task_pid_nr(p), p->comm, K(p->mm->total_vm), | |
414 | K(get_mm_counter(p->mm, MM_ANONPAGES)), | |
415 | K(get_mm_counter(p->mm, MM_FILEPAGES))); | |
3b4798cb | 416 | task_unlock(p); |
1da177e4 | 417 | |
1e99bad0 DR |
418 | /* |
419 | * Kill all processes sharing p->mm in other thread groups, if any. | |
420 | * They don't get access to memory reserves or a higher scheduler | |
421 | * priority, though, to avoid depletion of all memory or task | |
422 | * starvation. This prevents mm->mmap_sem livelock when an oom killed | |
423 | * task cannot exit because it requires the semaphore and its contended | |
424 | * by another thread trying to allocate memory itself. That thread will | |
425 | * now get access to memory reserves since it has a pending fatal | |
426 | * signal. | |
427 | */ | |
428 | for_each_process(q) | |
429 | if (q->mm == mm && !same_thread_group(q, p)) { | |
430 | task_lock(q); /* Protect ->comm from prctl() */ | |
431 | pr_err("Kill process %d (%s) sharing same memory\n", | |
432 | task_pid_nr(q), q->comm); | |
433 | task_unlock(q); | |
434 | force_sig(SIGKILL, q); | |
435 | } | |
93b43fa5 | 436 | |
1da177e4 | 437 | set_tsk_thread_flag(p, TIF_MEMDIE); |
1da177e4 | 438 | force_sig(SIGKILL, p); |
93b43fa5 | 439 | |
01315922 | 440 | return 0; |
1da177e4 | 441 | } |
b940fd70 | 442 | #undef K |
1da177e4 | 443 | |
7213f506 | 444 | static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
a63d83f4 DR |
445 | unsigned int points, unsigned long totalpages, |
446 | struct mem_cgroup *mem, nodemask_t *nodemask, | |
447 | const char *message) | |
1da177e4 | 448 | { |
52d3c036 | 449 | struct task_struct *victim = p; |
5e9d834a | 450 | struct task_struct *child; |
52d3c036 LT |
451 | struct task_struct *t = p; |
452 | unsigned int victim_points = 0; | |
1da177e4 | 453 | |
1b604d75 | 454 | if (printk_ratelimit()) |
e85bfd3a | 455 | dump_header(p, gfp_mask, order, mem, nodemask); |
7213f506 | 456 | |
50ec3bbf NP |
457 | /* |
458 | * If the task is already exiting, don't alarm the sysadmin or kill | |
459 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
460 | */ | |
0753ba01 | 461 | if (p->flags & PF_EXITING) { |
4358997a | 462 | set_tsk_thread_flag(p, TIF_MEMDIE); |
50ec3bbf NP |
463 | return 0; |
464 | } | |
465 | ||
5e9d834a | 466 | task_lock(p); |
a63d83f4 | 467 | pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n", |
5e9d834a DR |
468 | message, task_pid_nr(p), p->comm, points); |
469 | task_unlock(p); | |
f3af38d3 | 470 | |
5e9d834a DR |
471 | /* |
472 | * If any of p's children has a different mm and is eligible for kill, | |
473 | * the one with the highest badness() score is sacrificed for its | |
474 | * parent. This attempts to lose the minimal amount of work done while | |
475 | * still freeing memory. | |
476 | */ | |
dd8e8f40 | 477 | do { |
5e9d834a | 478 | list_for_each_entry(child, &t->children, sibling) { |
a63d83f4 | 479 | unsigned int child_points; |
5e9d834a | 480 | |
edd45544 DR |
481 | if (child->mm == p->mm) |
482 | continue; | |
a63d83f4 DR |
483 | /* |
484 | * oom_badness() returns 0 if the thread is unkillable | |
485 | */ | |
486 | child_points = oom_badness(child, mem, nodemask, | |
487 | totalpages); | |
5e9d834a DR |
488 | if (child_points > victim_points) { |
489 | victim = child; | |
490 | victim_points = child_points; | |
491 | } | |
dd8e8f40 ON |
492 | } |
493 | } while_each_thread(p, t); | |
494 | ||
93b43fa5 | 495 | return oom_kill_task(victim, mem); |
1da177e4 LT |
496 | } |
497 | ||
309ed882 DR |
498 | /* |
499 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. | |
500 | */ | |
501 | static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask, | |
e85bfd3a | 502 | int order, const nodemask_t *nodemask) |
309ed882 DR |
503 | { |
504 | if (likely(!sysctl_panic_on_oom)) | |
505 | return; | |
506 | if (sysctl_panic_on_oom != 2) { | |
507 | /* | |
508 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel | |
509 | * does not panic for cpuset, mempolicy, or memcg allocation | |
510 | * failures. | |
511 | */ | |
512 | if (constraint != CONSTRAINT_NONE) | |
513 | return; | |
514 | } | |
515 | read_lock(&tasklist_lock); | |
e85bfd3a | 516 | dump_header(NULL, gfp_mask, order, NULL, nodemask); |
309ed882 DR |
517 | read_unlock(&tasklist_lock); |
518 | panic("Out of memory: %s panic_on_oom is enabled\n", | |
519 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); | |
520 | } | |
521 | ||
00f0b825 | 522 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
c7ba5c9e PE |
523 | void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) |
524 | { | |
a63d83f4 DR |
525 | unsigned long limit; |
526 | unsigned int points = 0; | |
c7ba5c9e PE |
527 | struct task_struct *p; |
528 | ||
f9434ad1 DR |
529 | /* |
530 | * If current has a pending SIGKILL, then automatically select it. The | |
531 | * goal is to allow it to allocate so that it may quickly exit and free | |
532 | * its memory. | |
533 | */ | |
534 | if (fatal_signal_pending(current)) { | |
535 | set_thread_flag(TIF_MEMDIE); | |
f9434ad1 DR |
536 | return; |
537 | } | |
538 | ||
e85bfd3a | 539 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL); |
a63d83f4 | 540 | limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT; |
e115f2d8 | 541 | read_lock(&tasklist_lock); |
c7ba5c9e | 542 | retry: |
a63d83f4 | 543 | p = select_bad_process(&points, limit, mem, NULL); |
df64f81b | 544 | if (!p || PTR_ERR(p) == -1UL) |
c7ba5c9e PE |
545 | goto out; |
546 | ||
a63d83f4 | 547 | if (oom_kill_process(p, gfp_mask, 0, points, limit, mem, NULL, |
c7ba5c9e PE |
548 | "Memory cgroup out of memory")) |
549 | goto retry; | |
550 | out: | |
e115f2d8 | 551 | read_unlock(&tasklist_lock); |
c7ba5c9e PE |
552 | } |
553 | #endif | |
554 | ||
8bc719d3 MS |
555 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
556 | ||
557 | int register_oom_notifier(struct notifier_block *nb) | |
558 | { | |
559 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
560 | } | |
561 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
562 | ||
563 | int unregister_oom_notifier(struct notifier_block *nb) | |
564 | { | |
565 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
566 | } | |
567 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
568 | ||
098d7f12 DR |
569 | /* |
570 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero | |
571 | * if a parallel OOM killing is already taking place that includes a zone in | |
572 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. | |
573 | */ | |
ff321fea | 574 | int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 575 | { |
dd1a239f MG |
576 | struct zoneref *z; |
577 | struct zone *zone; | |
098d7f12 DR |
578 | int ret = 1; |
579 | ||
c7d4caeb | 580 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
581 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
582 | if (zone_is_oom_locked(zone)) { | |
098d7f12 DR |
583 | ret = 0; |
584 | goto out; | |
585 | } | |
dd1a239f MG |
586 | } |
587 | ||
588 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
589 | /* | |
c7d4caeb | 590 | * Lock each zone in the zonelist under zone_scan_lock so a |
ff321fea | 591 | * parallel invocation of try_set_zonelist_oom() doesn't succeed |
dd1a239f MG |
592 | * when it shouldn't. |
593 | */ | |
594 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
595 | } | |
098d7f12 | 596 | |
098d7f12 | 597 | out: |
c7d4caeb | 598 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
599 | return ret; |
600 | } | |
601 | ||
602 | /* | |
603 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed | |
604 | * allocation attempts with zonelists containing them may now recall the OOM | |
605 | * killer, if necessary. | |
606 | */ | |
dd1a239f | 607 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 608 | { |
dd1a239f MG |
609 | struct zoneref *z; |
610 | struct zone *zone; | |
098d7f12 | 611 | |
c7d4caeb | 612 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
613 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
614 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
615 | } | |
c7d4caeb | 616 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
617 | } |
618 | ||
e3658932 DR |
619 | /* |
620 | * Try to acquire the oom killer lock for all system zones. Returns zero if a | |
621 | * parallel oom killing is taking place, otherwise locks all zones and returns | |
622 | * non-zero. | |
623 | */ | |
624 | static int try_set_system_oom(void) | |
625 | { | |
626 | struct zone *zone; | |
627 | int ret = 1; | |
628 | ||
629 | spin_lock(&zone_scan_lock); | |
630 | for_each_populated_zone(zone) | |
631 | if (zone_is_oom_locked(zone)) { | |
632 | ret = 0; | |
633 | goto out; | |
634 | } | |
635 | for_each_populated_zone(zone) | |
636 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
637 | out: | |
638 | spin_unlock(&zone_scan_lock); | |
639 | return ret; | |
640 | } | |
641 | ||
642 | /* | |
643 | * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation | |
644 | * attempts or page faults may now recall the oom killer, if necessary. | |
645 | */ | |
646 | static void clear_system_oom(void) | |
647 | { | |
648 | struct zone *zone; | |
649 | ||
650 | spin_lock(&zone_scan_lock); | |
651 | for_each_populated_zone(zone) | |
652 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
653 | spin_unlock(&zone_scan_lock); | |
654 | } | |
655 | ||
1da177e4 | 656 | /** |
6937a25c | 657 | * out_of_memory - kill the "best" process when we run out of memory |
1b578df0 RD |
658 | * @zonelist: zonelist pointer |
659 | * @gfp_mask: memory allocation flags | |
660 | * @order: amount of memory being requested as a power of 2 | |
6f48d0eb | 661 | * @nodemask: nodemask passed to page allocator |
1da177e4 LT |
662 | * |
663 | * If we run out of memory, we have the choice between either | |
664 | * killing a random task (bad), letting the system crash (worse) | |
665 | * OR try to be smart about which process to kill. Note that we | |
666 | * don't have to be perfect here, we just have to be good. | |
667 | */ | |
4365a567 KH |
668 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
669 | int order, nodemask_t *nodemask) | |
1da177e4 | 670 | { |
e85bfd3a | 671 | const nodemask_t *mpol_mask; |
0aad4b31 | 672 | struct task_struct *p; |
a63d83f4 | 673 | unsigned long totalpages; |
8bc719d3 | 674 | unsigned long freed = 0; |
a63d83f4 | 675 | unsigned int points; |
e3658932 | 676 | enum oom_constraint constraint = CONSTRAINT_NONE; |
b52723c5 | 677 | int killed = 0; |
8bc719d3 MS |
678 | |
679 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
680 | if (freed > 0) | |
681 | /* Got some memory back in the last second. */ | |
682 | return; | |
1da177e4 | 683 | |
7b98c2e4 DR |
684 | /* |
685 | * If current has a pending SIGKILL, then automatically select it. The | |
686 | * goal is to allow it to allocate so that it may quickly exit and free | |
687 | * its memory. | |
688 | */ | |
689 | if (fatal_signal_pending(current)) { | |
690 | set_thread_flag(TIF_MEMDIE); | |
691 | return; | |
692 | } | |
693 | ||
9b0f8b04 CL |
694 | /* |
695 | * Check if there were limitations on the allocation (only relevant for | |
696 | * NUMA) that may require different handling. | |
697 | */ | |
a63d83f4 DR |
698 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask, |
699 | &totalpages); | |
e85bfd3a DR |
700 | mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL; |
701 | check_panic_on_oom(constraint, gfp_mask, order, mpol_mask); | |
0aad4b31 | 702 | |
2b45ab33 | 703 | read_lock(&tasklist_lock); |
f88ccad5 | 704 | if (sysctl_oom_kill_allocating_task && |
a96cfd6e | 705 | !oom_unkillable_task(current, NULL, nodemask) && |
e18641e1 | 706 | current->mm && !atomic_read(¤t->mm->oom_disable_count)) { |
0aad4b31 DR |
707 | /* |
708 | * oom_kill_process() needs tasklist_lock held. If it returns | |
709 | * non-zero, current could not be killed so we must fallback to | |
710 | * the tasklist scan. | |
711 | */ | |
a63d83f4 DR |
712 | if (!oom_kill_process(current, gfp_mask, order, 0, totalpages, |
713 | NULL, nodemask, | |
0aad4b31 | 714 | "Out of memory (oom_kill_allocating_task)")) |
b52723c5 | 715 | goto out; |
0aad4b31 DR |
716 | } |
717 | ||
718 | retry: | |
e85bfd3a | 719 | p = select_bad_process(&points, totalpages, NULL, mpol_mask); |
0aad4b31 | 720 | if (PTR_ERR(p) == -1UL) |
b52723c5 | 721 | goto out; |
0aad4b31 DR |
722 | |
723 | /* Found nothing?!?! Either we hang forever, or we panic. */ | |
724 | if (!p) { | |
e85bfd3a | 725 | dump_header(NULL, gfp_mask, order, NULL, mpol_mask); |
0aad4b31 DR |
726 | read_unlock(&tasklist_lock); |
727 | panic("Out of memory and no killable processes...\n"); | |
728 | } | |
729 | ||
a63d83f4 DR |
730 | if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL, |
731 | nodemask, "Out of memory")) | |
0aad4b31 | 732 | goto retry; |
b52723c5 KM |
733 | killed = 1; |
734 | out: | |
140ffcec | 735 | read_unlock(&tasklist_lock); |
1da177e4 LT |
736 | |
737 | /* | |
738 | * Give "p" a good chance of killing itself before we | |
2f659f46 | 739 | * retry to allocate memory unless "p" is current |
1da177e4 | 740 | */ |
b52723c5 | 741 | if (killed && !test_thread_flag(TIF_MEMDIE)) |
140ffcec | 742 | schedule_timeout_uninterruptible(1); |
1da177e4 | 743 | } |
e3658932 DR |
744 | |
745 | /* | |
746 | * The pagefault handler calls here because it is out of memory, so kill a | |
747 | * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel | |
748 | * oom killing is already in progress so do nothing. If a task is found with | |
749 | * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit. | |
750 | */ | |
751 | void pagefault_out_of_memory(void) | |
752 | { | |
753 | if (try_set_system_oom()) { | |
754 | out_of_memory(NULL, 0, 0, NULL); | |
755 | clear_system_oom(); | |
756 | } | |
757 | if (!test_thread_flag(TIF_MEMDIE)) | |
758 | schedule_timeout_uninterruptible(1); | |
759 | } |