]>
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> |
b95f1b31 | 29 | #include <linux/export.h> |
8bc719d3 | 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 | |
43362a49 DR |
42 | /* |
43 | * compare_swap_oom_score_adj() - compare and swap current's oom_score_adj | |
44 | * @old_val: old oom_score_adj for compare | |
45 | * @new_val: new oom_score_adj for swap | |
46 | * | |
47 | * Sets the oom_score_adj value for current to @new_val iff its present value is | |
48 | * @old_val. Usually used to reinstate a previous value to prevent racing with | |
49 | * userspacing tuning the value in the interim. | |
50 | */ | |
51 | void compare_swap_oom_score_adj(int old_val, int new_val) | |
52 | { | |
53 | struct sighand_struct *sighand = current->sighand; | |
54 | ||
55 | spin_lock_irq(&sighand->siglock); | |
56 | if (current->signal->oom_score_adj == old_val) | |
57 | current->signal->oom_score_adj = new_val; | |
58 | spin_unlock_irq(&sighand->siglock); | |
59 | } | |
60 | ||
72788c38 DR |
61 | /** |
62 | * test_set_oom_score_adj() - set current's oom_score_adj and return old value | |
63 | * @new_val: new oom_score_adj value | |
64 | * | |
65 | * Sets the oom_score_adj value for current to @new_val with proper | |
66 | * synchronization and returns the old value. Usually used to temporarily | |
67 | * set a value, save the old value in the caller, and then reinstate it later. | |
68 | */ | |
69 | int test_set_oom_score_adj(int new_val) | |
70 | { | |
71 | struct sighand_struct *sighand = current->sighand; | |
72 | int old_val; | |
73 | ||
74 | spin_lock_irq(&sighand->siglock); | |
75 | old_val = current->signal->oom_score_adj; | |
c9f01245 | 76 | current->signal->oom_score_adj = new_val; |
72788c38 DR |
77 | spin_unlock_irq(&sighand->siglock); |
78 | ||
79 | return old_val; | |
80 | } | |
81 | ||
6f48d0eb DR |
82 | #ifdef CONFIG_NUMA |
83 | /** | |
84 | * has_intersects_mems_allowed() - check task eligiblity for kill | |
85 | * @tsk: task struct of which task to consider | |
86 | * @mask: nodemask passed to page allocator for mempolicy ooms | |
87 | * | |
88 | * Task eligibility is determined by whether or not a candidate task, @tsk, | |
89 | * shares the same mempolicy nodes as current if it is bound by such a policy | |
90 | * and whether or not it has the same set of allowed cpuset nodes. | |
495789a5 | 91 | */ |
6f48d0eb DR |
92 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
93 | const nodemask_t *mask) | |
495789a5 | 94 | { |
6f48d0eb | 95 | struct task_struct *start = tsk; |
495789a5 | 96 | |
495789a5 | 97 | do { |
6f48d0eb DR |
98 | if (mask) { |
99 | /* | |
100 | * If this is a mempolicy constrained oom, tsk's | |
101 | * cpuset is irrelevant. Only return true if its | |
102 | * mempolicy intersects current, otherwise it may be | |
103 | * needlessly killed. | |
104 | */ | |
105 | if (mempolicy_nodemask_intersects(tsk, mask)) | |
106 | return true; | |
107 | } else { | |
108 | /* | |
109 | * This is not a mempolicy constrained oom, so only | |
110 | * check the mems of tsk's cpuset. | |
111 | */ | |
112 | if (cpuset_mems_allowed_intersects(current, tsk)) | |
113 | return true; | |
114 | } | |
df1090a8 KM |
115 | } while_each_thread(start, tsk); |
116 | ||
6f48d0eb DR |
117 | return false; |
118 | } | |
119 | #else | |
120 | static bool has_intersects_mems_allowed(struct task_struct *tsk, | |
121 | const nodemask_t *mask) | |
122 | { | |
123 | return true; | |
495789a5 | 124 | } |
6f48d0eb | 125 | #endif /* CONFIG_NUMA */ |
495789a5 | 126 | |
6f48d0eb DR |
127 | /* |
128 | * The process p may have detached its own ->mm while exiting or through | |
129 | * use_mm(), but one or more of its subthreads may still have a valid | |
130 | * pointer. Return p, or any of its subthreads with a valid ->mm, with | |
131 | * task_lock() held. | |
132 | */ | |
158e0a2d | 133 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
dd8e8f40 ON |
134 | { |
135 | struct task_struct *t = p; | |
136 | ||
137 | do { | |
138 | task_lock(t); | |
139 | if (likely(t->mm)) | |
140 | return t; | |
141 | task_unlock(t); | |
142 | } while_each_thread(p, t); | |
143 | ||
144 | return NULL; | |
145 | } | |
146 | ||
ab290adb | 147 | /* return true if the task is not adequate as candidate victim task. */ |
e85bfd3a DR |
148 | static bool oom_unkillable_task(struct task_struct *p, |
149 | const struct mem_cgroup *mem, const nodemask_t *nodemask) | |
ab290adb KM |
150 | { |
151 | if (is_global_init(p)) | |
152 | return true; | |
153 | if (p->flags & PF_KTHREAD) | |
154 | return true; | |
155 | ||
156 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ | |
157 | if (mem && !task_in_mem_cgroup(p, mem)) | |
158 | return true; | |
159 | ||
160 | /* p may not have freeable memory in nodemask */ | |
161 | if (!has_intersects_mems_allowed(p, nodemask)) | |
162 | return true; | |
163 | ||
164 | return false; | |
165 | } | |
166 | ||
1da177e4 | 167 | /** |
a63d83f4 | 168 | * oom_badness - heuristic function to determine which candidate task to kill |
1da177e4 | 169 | * @p: task struct of which task we should calculate |
a63d83f4 | 170 | * @totalpages: total present RAM allowed for page allocation |
1da177e4 | 171 | * |
a63d83f4 DR |
172 | * The heuristic for determining which task to kill is made to be as simple and |
173 | * predictable as possible. The goal is to return the highest value for the | |
174 | * task consuming the most memory to avoid subsequent oom failures. | |
1da177e4 | 175 | */ |
a63d83f4 DR |
176 | unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem, |
177 | const nodemask_t *nodemask, unsigned long totalpages) | |
1da177e4 | 178 | { |
ff05b6f7 | 179 | long points; |
28b83c51 | 180 | |
26ebc984 KM |
181 | if (oom_unkillable_task(p, mem, nodemask)) |
182 | return 0; | |
1da177e4 | 183 | |
dd8e8f40 ON |
184 | p = find_lock_task_mm(p); |
185 | if (!p) | |
1da177e4 LT |
186 | return 0; |
187 | ||
5aecc85a MH |
188 | if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) { |
189 | task_unlock(p); | |
190 | return 0; | |
191 | } | |
192 | ||
1da177e4 | 193 | /* |
a63d83f4 DR |
194 | * The memory controller may have a limit of 0 bytes, so avoid a divide |
195 | * by zero, if necessary. | |
1da177e4 | 196 | */ |
a63d83f4 DR |
197 | if (!totalpages) |
198 | totalpages = 1; | |
1da177e4 LT |
199 | |
200 | /* | |
a63d83f4 | 201 | * The baseline for the badness score is the proportion of RAM that each |
f755a042 | 202 | * task's rss, pagetable and swap space use. |
1da177e4 | 203 | */ |
f755a042 KM |
204 | points = get_mm_rss(p->mm) + p->mm->nr_ptes; |
205 | points += get_mm_counter(p->mm, MM_SWAPENTS); | |
206 | ||
207 | points *= 1000; | |
208 | points /= totalpages; | |
a63d83f4 | 209 | task_unlock(p); |
1da177e4 LT |
210 | |
211 | /* | |
a63d83f4 DR |
212 | * Root processes get 3% bonus, just like the __vm_enough_memory() |
213 | * implementation used by LSMs. | |
1da177e4 | 214 | */ |
a63d83f4 DR |
215 | if (has_capability_noaudit(p, CAP_SYS_ADMIN)) |
216 | points -= 30; | |
1da177e4 LT |
217 | |
218 | /* | |
a63d83f4 DR |
219 | * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may |
220 | * either completely disable oom killing or always prefer a certain | |
221 | * task. | |
1da177e4 | 222 | */ |
a63d83f4 | 223 | points += p->signal->oom_score_adj; |
1da177e4 | 224 | |
f19e8aa1 DR |
225 | /* |
226 | * Never return 0 for an eligible task that may be killed since it's | |
227 | * possible that no single user task uses more than 0.1% of memory and | |
228 | * no single admin tasks uses more than 3.0%. | |
229 | */ | |
230 | if (points <= 0) | |
231 | return 1; | |
a63d83f4 | 232 | return (points < 1000) ? points : 1000; |
1da177e4 LT |
233 | } |
234 | ||
9b0f8b04 CL |
235 | /* |
236 | * Determine the type of allocation constraint. | |
237 | */ | |
9b0f8b04 | 238 | #ifdef CONFIG_NUMA |
4365a567 | 239 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
a63d83f4 DR |
240 | gfp_t gfp_mask, nodemask_t *nodemask, |
241 | unsigned long *totalpages) | |
4365a567 | 242 | { |
54a6eb5c | 243 | struct zone *zone; |
dd1a239f | 244 | struct zoneref *z; |
54a6eb5c | 245 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
a63d83f4 DR |
246 | bool cpuset_limited = false; |
247 | int nid; | |
9b0f8b04 | 248 | |
a63d83f4 DR |
249 | /* Default to all available memory */ |
250 | *totalpages = totalram_pages + total_swap_pages; | |
251 | ||
252 | if (!zonelist) | |
253 | return CONSTRAINT_NONE; | |
4365a567 KH |
254 | /* |
255 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid | |
256 | * to kill current.We have to random task kill in this case. | |
257 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. | |
258 | */ | |
259 | if (gfp_mask & __GFP_THISNODE) | |
260 | return CONSTRAINT_NONE; | |
9b0f8b04 | 261 | |
4365a567 | 262 | /* |
a63d83f4 DR |
263 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
264 | * the page allocator means a mempolicy is in effect. Cpuset policy | |
265 | * is enforced in get_page_from_freelist(). | |
4365a567 | 266 | */ |
a63d83f4 DR |
267 | if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) { |
268 | *totalpages = total_swap_pages; | |
269 | for_each_node_mask(nid, *nodemask) | |
270 | *totalpages += node_spanned_pages(nid); | |
9b0f8b04 | 271 | return CONSTRAINT_MEMORY_POLICY; |
a63d83f4 | 272 | } |
4365a567 KH |
273 | |
274 | /* Check this allocation failure is caused by cpuset's wall function */ | |
275 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
276 | high_zoneidx, nodemask) | |
277 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) | |
a63d83f4 | 278 | cpuset_limited = true; |
9b0f8b04 | 279 | |
a63d83f4 DR |
280 | if (cpuset_limited) { |
281 | *totalpages = total_swap_pages; | |
282 | for_each_node_mask(nid, cpuset_current_mems_allowed) | |
283 | *totalpages += node_spanned_pages(nid); | |
284 | return CONSTRAINT_CPUSET; | |
285 | } | |
9b0f8b04 CL |
286 | return CONSTRAINT_NONE; |
287 | } | |
4365a567 KH |
288 | #else |
289 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, | |
a63d83f4 DR |
290 | gfp_t gfp_mask, nodemask_t *nodemask, |
291 | unsigned long *totalpages) | |
4365a567 | 292 | { |
a63d83f4 | 293 | *totalpages = totalram_pages + total_swap_pages; |
4365a567 KH |
294 | return CONSTRAINT_NONE; |
295 | } | |
296 | #endif | |
9b0f8b04 | 297 | |
1da177e4 LT |
298 | /* |
299 | * Simple selection loop. We chose the process with the highest | |
300 | * number of 'points'. We expect the caller will lock the tasklist. | |
301 | * | |
302 | * (not docbooked, we don't want this one cluttering up the manual) | |
303 | */ | |
a63d83f4 DR |
304 | static struct task_struct *select_bad_process(unsigned int *ppoints, |
305 | unsigned long totalpages, struct mem_cgroup *mem, | |
306 | const nodemask_t *nodemask) | |
1da177e4 | 307 | { |
3a5dda7a | 308 | struct task_struct *g, *p; |
1da177e4 | 309 | struct task_struct *chosen = NULL; |
9827b781 | 310 | *ppoints = 0; |
1da177e4 | 311 | |
3a5dda7a | 312 | do_each_thread(g, p) { |
a63d83f4 | 313 | unsigned int points; |
a49335cc | 314 | |
c027a474 | 315 | if (p->exit_state) |
30e2b41f | 316 | continue; |
ab290adb | 317 | if (oom_unkillable_task(p, mem, nodemask)) |
6cf86ac6 | 318 | continue; |
ef08e3b4 | 319 | |
b78483a4 NP |
320 | /* |
321 | * This task already has access to memory reserves and is | |
322 | * being killed. Don't allow any other task access to the | |
323 | * memory reserve. | |
324 | * | |
325 | * Note: this may have a chance of deadlock if it gets | |
326 | * blocked waiting for another task which itself is waiting | |
327 | * for memory. Is there a better alternative? | |
328 | */ | |
f660daac DR |
329 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) { |
330 | if (unlikely(frozen(p))) | |
a5be2d0d | 331 | __thaw_task(p); |
b78483a4 | 332 | return ERR_PTR(-1UL); |
f660daac | 333 | } |
c027a474 ON |
334 | if (!p->mm) |
335 | continue; | |
b78483a4 | 336 | |
30e2b41f | 337 | if (p->flags & PF_EXITING) { |
edd45544 DR |
338 | /* |
339 | * If p is the current task and is in the process of | |
340 | * releasing memory, we allow the "kill" to set | |
341 | * TIF_MEMDIE, which will allow it to gain access to | |
342 | * memory reserves. Otherwise, it may stall forever. | |
343 | * | |
344 | * The loop isn't broken here, however, in case other | |
345 | * threads are found to have already been oom killed. | |
346 | */ | |
347 | if (p == current) { | |
348 | chosen = p; | |
349 | *ppoints = 1000; | |
350 | } else { | |
351 | /* | |
352 | * If this task is not being ptraced on exit, | |
353 | * then wait for it to finish before killing | |
354 | * some other task unnecessarily. | |
355 | */ | |
d21142ec | 356 | if (!(p->group_leader->ptrace & PT_TRACE_EXIT)) |
edd45544 DR |
357 | return ERR_PTR(-1UL); |
358 | } | |
50ec3bbf | 359 | } |
972c4ea5 | 360 | |
a63d83f4 DR |
361 | points = oom_badness(p, mem, nodemask, totalpages); |
362 | if (points > *ppoints) { | |
a49335cc | 363 | chosen = p; |
9827b781 | 364 | *ppoints = points; |
1da177e4 | 365 | } |
3a5dda7a | 366 | } while_each_thread(g, p); |
972c4ea5 | 367 | |
1da177e4 LT |
368 | return chosen; |
369 | } | |
370 | ||
fef1bdd6 | 371 | /** |
1b578df0 | 372 | * dump_tasks - dump current memory state of all system tasks |
74ab7f1d | 373 | * @mem: current's memory controller, if constrained |
e85bfd3a | 374 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
1b578df0 | 375 | * |
e85bfd3a DR |
376 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
377 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes | |
378 | * are not shown. | |
fef1bdd6 | 379 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj |
a63d83f4 | 380 | * value, oom_score_adj value, and name. |
fef1bdd6 | 381 | * |
fef1bdd6 DR |
382 | * Call with tasklist_lock read-locked. |
383 | */ | |
e85bfd3a | 384 | static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask) |
fef1bdd6 | 385 | { |
c55db957 KM |
386 | struct task_struct *p; |
387 | struct task_struct *task; | |
fef1bdd6 | 388 | |
a63d83f4 | 389 | pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n"); |
c55db957 | 390 | for_each_process(p) { |
e85bfd3a | 391 | if (oom_unkillable_task(p, mem, nodemask)) |
b4416d2b | 392 | continue; |
fef1bdd6 | 393 | |
c55db957 KM |
394 | task = find_lock_task_mm(p); |
395 | if (!task) { | |
6d2661ed | 396 | /* |
74ab7f1d DR |
397 | * This is a kthread or all of p's threads have already |
398 | * detached their mm's. There's no need to report | |
c55db957 | 399 | * them; they can't be oom killed anyway. |
6d2661ed | 400 | */ |
6d2661ed DR |
401 | continue; |
402 | } | |
c55db957 | 403 | |
a63d83f4 | 404 | pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n", |
8d6c83f0 | 405 | task->pid, task_uid(task), task->tgid, |
a63d83f4 DR |
406 | task->mm->total_vm, get_mm_rss(task->mm), |
407 | task_cpu(task), task->signal->oom_adj, | |
408 | task->signal->oom_score_adj, task->comm); | |
c55db957 KM |
409 | task_unlock(task); |
410 | } | |
fef1bdd6 DR |
411 | } |
412 | ||
d31f56db | 413 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
e85bfd3a | 414 | struct mem_cgroup *mem, const nodemask_t *nodemask) |
1b604d75 | 415 | { |
5e9d834a | 416 | task_lock(current); |
1b604d75 | 417 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " |
a63d83f4 DR |
418 | "oom_adj=%d, oom_score_adj=%d\n", |
419 | current->comm, gfp_mask, order, current->signal->oom_adj, | |
420 | current->signal->oom_score_adj); | |
1b604d75 DR |
421 | cpuset_print_task_mems_allowed(current); |
422 | task_unlock(current); | |
423 | dump_stack(); | |
d31f56db | 424 | mem_cgroup_print_oom_info(mem, p); |
b2b755b5 | 425 | show_mem(SHOW_MEM_FILTER_NODES); |
1b604d75 | 426 | if (sysctl_oom_dump_tasks) |
e85bfd3a | 427 | dump_tasks(mem, nodemask); |
1b604d75 DR |
428 | } |
429 | ||
3b4798cb | 430 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
93b43fa5 | 431 | static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem) |
1da177e4 | 432 | { |
1e99bad0 DR |
433 | struct task_struct *q; |
434 | struct mm_struct *mm; | |
435 | ||
dd8e8f40 | 436 | p = find_lock_task_mm(p); |
be71cf22 | 437 | if (!p) |
b940fd70 | 438 | return 1; |
be71cf22 | 439 | |
1e99bad0 DR |
440 | /* mm cannot be safely dereferenced after task_unlock(p) */ |
441 | mm = p->mm; | |
442 | ||
b940fd70 DR |
443 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", |
444 | task_pid_nr(p), p->comm, K(p->mm->total_vm), | |
445 | K(get_mm_counter(p->mm, MM_ANONPAGES)), | |
446 | K(get_mm_counter(p->mm, MM_FILEPAGES))); | |
3b4798cb | 447 | task_unlock(p); |
1da177e4 | 448 | |
1e99bad0 | 449 | /* |
7b0d44fa | 450 | * Kill all user processes sharing p->mm in other thread groups, if any. |
1e99bad0 DR |
451 | * They don't get access to memory reserves or a higher scheduler |
452 | * priority, though, to avoid depletion of all memory or task | |
453 | * starvation. This prevents mm->mmap_sem livelock when an oom killed | |
454 | * task cannot exit because it requires the semaphore and its contended | |
455 | * by another thread trying to allocate memory itself. That thread will | |
456 | * now get access to memory reserves since it has a pending fatal | |
457 | * signal. | |
458 | */ | |
459 | for_each_process(q) | |
7b0d44fa DR |
460 | if (q->mm == mm && !same_thread_group(q, p) && |
461 | !(q->flags & PF_KTHREAD)) { | |
c9f01245 DR |
462 | if (q->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) |
463 | continue; | |
464 | ||
1e99bad0 DR |
465 | task_lock(q); /* Protect ->comm from prctl() */ |
466 | pr_err("Kill process %d (%s) sharing same memory\n", | |
467 | task_pid_nr(q), q->comm); | |
468 | task_unlock(q); | |
469 | force_sig(SIGKILL, q); | |
470 | } | |
93b43fa5 | 471 | |
1da177e4 | 472 | set_tsk_thread_flag(p, TIF_MEMDIE); |
1da177e4 | 473 | force_sig(SIGKILL, p); |
93b43fa5 | 474 | |
01315922 | 475 | return 0; |
1da177e4 | 476 | } |
b940fd70 | 477 | #undef K |
1da177e4 | 478 | |
7213f506 | 479 | static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
a63d83f4 DR |
480 | unsigned int points, unsigned long totalpages, |
481 | struct mem_cgroup *mem, nodemask_t *nodemask, | |
482 | const char *message) | |
1da177e4 | 483 | { |
52d3c036 | 484 | struct task_struct *victim = p; |
5e9d834a | 485 | struct task_struct *child; |
52d3c036 LT |
486 | struct task_struct *t = p; |
487 | unsigned int victim_points = 0; | |
1da177e4 | 488 | |
1b604d75 | 489 | if (printk_ratelimit()) |
e85bfd3a | 490 | dump_header(p, gfp_mask, order, mem, nodemask); |
7213f506 | 491 | |
50ec3bbf NP |
492 | /* |
493 | * If the task is already exiting, don't alarm the sysadmin or kill | |
494 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
495 | */ | |
0753ba01 | 496 | if (p->flags & PF_EXITING) { |
4358997a | 497 | set_tsk_thread_flag(p, TIF_MEMDIE); |
50ec3bbf NP |
498 | return 0; |
499 | } | |
500 | ||
5e9d834a | 501 | task_lock(p); |
a63d83f4 | 502 | pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n", |
5e9d834a DR |
503 | message, task_pid_nr(p), p->comm, points); |
504 | task_unlock(p); | |
f3af38d3 | 505 | |
5e9d834a DR |
506 | /* |
507 | * If any of p's children has a different mm and is eligible for kill, | |
11239836 | 508 | * the one with the highest oom_badness() score is sacrificed for its |
5e9d834a DR |
509 | * parent. This attempts to lose the minimal amount of work done while |
510 | * still freeing memory. | |
511 | */ | |
dd8e8f40 | 512 | do { |
5e9d834a | 513 | list_for_each_entry(child, &t->children, sibling) { |
a63d83f4 | 514 | unsigned int child_points; |
5e9d834a | 515 | |
edd45544 DR |
516 | if (child->mm == p->mm) |
517 | continue; | |
a63d83f4 DR |
518 | /* |
519 | * oom_badness() returns 0 if the thread is unkillable | |
520 | */ | |
521 | child_points = oom_badness(child, mem, nodemask, | |
522 | totalpages); | |
5e9d834a DR |
523 | if (child_points > victim_points) { |
524 | victim = child; | |
525 | victim_points = child_points; | |
526 | } | |
dd8e8f40 ON |
527 | } |
528 | } while_each_thread(p, t); | |
529 | ||
93b43fa5 | 530 | return oom_kill_task(victim, mem); |
1da177e4 LT |
531 | } |
532 | ||
309ed882 DR |
533 | /* |
534 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. | |
535 | */ | |
536 | static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask, | |
e85bfd3a | 537 | int order, const nodemask_t *nodemask) |
309ed882 DR |
538 | { |
539 | if (likely(!sysctl_panic_on_oom)) | |
540 | return; | |
541 | if (sysctl_panic_on_oom != 2) { | |
542 | /* | |
543 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel | |
544 | * does not panic for cpuset, mempolicy, or memcg allocation | |
545 | * failures. | |
546 | */ | |
547 | if (constraint != CONSTRAINT_NONE) | |
548 | return; | |
549 | } | |
550 | read_lock(&tasklist_lock); | |
e85bfd3a | 551 | dump_header(NULL, gfp_mask, order, NULL, nodemask); |
309ed882 DR |
552 | read_unlock(&tasklist_lock); |
553 | panic("Out of memory: %s panic_on_oom is enabled\n", | |
554 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); | |
555 | } | |
556 | ||
00f0b825 | 557 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
c7ba5c9e PE |
558 | void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) |
559 | { | |
a63d83f4 DR |
560 | unsigned long limit; |
561 | unsigned int points = 0; | |
c7ba5c9e PE |
562 | struct task_struct *p; |
563 | ||
f9434ad1 DR |
564 | /* |
565 | * If current has a pending SIGKILL, then automatically select it. The | |
566 | * goal is to allow it to allocate so that it may quickly exit and free | |
567 | * its memory. | |
568 | */ | |
569 | if (fatal_signal_pending(current)) { | |
570 | set_thread_flag(TIF_MEMDIE); | |
f9434ad1 DR |
571 | return; |
572 | } | |
573 | ||
e85bfd3a | 574 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL); |
a63d83f4 | 575 | limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT; |
e115f2d8 | 576 | read_lock(&tasklist_lock); |
c7ba5c9e | 577 | retry: |
a63d83f4 | 578 | p = select_bad_process(&points, limit, mem, NULL); |
df64f81b | 579 | if (!p || PTR_ERR(p) == -1UL) |
c7ba5c9e PE |
580 | goto out; |
581 | ||
a63d83f4 | 582 | if (oom_kill_process(p, gfp_mask, 0, points, limit, mem, NULL, |
c7ba5c9e PE |
583 | "Memory cgroup out of memory")) |
584 | goto retry; | |
585 | out: | |
e115f2d8 | 586 | read_unlock(&tasklist_lock); |
c7ba5c9e PE |
587 | } |
588 | #endif | |
589 | ||
8bc719d3 MS |
590 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
591 | ||
592 | int register_oom_notifier(struct notifier_block *nb) | |
593 | { | |
594 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
595 | } | |
596 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
597 | ||
598 | int unregister_oom_notifier(struct notifier_block *nb) | |
599 | { | |
600 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
601 | } | |
602 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
603 | ||
098d7f12 DR |
604 | /* |
605 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero | |
606 | * if a parallel OOM killing is already taking place that includes a zone in | |
607 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. | |
608 | */ | |
ff321fea | 609 | int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 610 | { |
dd1a239f MG |
611 | struct zoneref *z; |
612 | struct zone *zone; | |
098d7f12 DR |
613 | int ret = 1; |
614 | ||
c7d4caeb | 615 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
616 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
617 | if (zone_is_oom_locked(zone)) { | |
098d7f12 DR |
618 | ret = 0; |
619 | goto out; | |
620 | } | |
dd1a239f MG |
621 | } |
622 | ||
623 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
624 | /* | |
c7d4caeb | 625 | * Lock each zone in the zonelist under zone_scan_lock so a |
ff321fea | 626 | * parallel invocation of try_set_zonelist_oom() doesn't succeed |
dd1a239f MG |
627 | * when it shouldn't. |
628 | */ | |
629 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
630 | } | |
098d7f12 | 631 | |
098d7f12 | 632 | out: |
c7d4caeb | 633 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
634 | return ret; |
635 | } | |
636 | ||
637 | /* | |
638 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed | |
639 | * allocation attempts with zonelists containing them may now recall the OOM | |
640 | * killer, if necessary. | |
641 | */ | |
dd1a239f | 642 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 643 | { |
dd1a239f MG |
644 | struct zoneref *z; |
645 | struct zone *zone; | |
098d7f12 | 646 | |
c7d4caeb | 647 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
648 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
649 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
650 | } | |
c7d4caeb | 651 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
652 | } |
653 | ||
e3658932 DR |
654 | /* |
655 | * Try to acquire the oom killer lock for all system zones. Returns zero if a | |
656 | * parallel oom killing is taking place, otherwise locks all zones and returns | |
657 | * non-zero. | |
658 | */ | |
659 | static int try_set_system_oom(void) | |
660 | { | |
661 | struct zone *zone; | |
662 | int ret = 1; | |
663 | ||
664 | spin_lock(&zone_scan_lock); | |
665 | for_each_populated_zone(zone) | |
666 | if (zone_is_oom_locked(zone)) { | |
667 | ret = 0; | |
668 | goto out; | |
669 | } | |
670 | for_each_populated_zone(zone) | |
671 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
672 | out: | |
673 | spin_unlock(&zone_scan_lock); | |
674 | return ret; | |
675 | } | |
676 | ||
677 | /* | |
678 | * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation | |
679 | * attempts or page faults may now recall the oom killer, if necessary. | |
680 | */ | |
681 | static void clear_system_oom(void) | |
682 | { | |
683 | struct zone *zone; | |
684 | ||
685 | spin_lock(&zone_scan_lock); | |
686 | for_each_populated_zone(zone) | |
687 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
688 | spin_unlock(&zone_scan_lock); | |
689 | } | |
690 | ||
1da177e4 | 691 | /** |
6937a25c | 692 | * out_of_memory - kill the "best" process when we run out of memory |
1b578df0 RD |
693 | * @zonelist: zonelist pointer |
694 | * @gfp_mask: memory allocation flags | |
695 | * @order: amount of memory being requested as a power of 2 | |
6f48d0eb | 696 | * @nodemask: nodemask passed to page allocator |
1da177e4 LT |
697 | * |
698 | * If we run out of memory, we have the choice between either | |
699 | * killing a random task (bad), letting the system crash (worse) | |
700 | * OR try to be smart about which process to kill. Note that we | |
701 | * don't have to be perfect here, we just have to be good. | |
702 | */ | |
4365a567 KH |
703 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
704 | int order, nodemask_t *nodemask) | |
1da177e4 | 705 | { |
e85bfd3a | 706 | const nodemask_t *mpol_mask; |
0aad4b31 | 707 | struct task_struct *p; |
a63d83f4 | 708 | unsigned long totalpages; |
8bc719d3 | 709 | unsigned long freed = 0; |
a63d83f4 | 710 | unsigned int points; |
e3658932 | 711 | enum oom_constraint constraint = CONSTRAINT_NONE; |
b52723c5 | 712 | int killed = 0; |
8bc719d3 MS |
713 | |
714 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
715 | if (freed > 0) | |
716 | /* Got some memory back in the last second. */ | |
717 | return; | |
1da177e4 | 718 | |
7b98c2e4 DR |
719 | /* |
720 | * If current has a pending SIGKILL, then automatically select it. The | |
721 | * goal is to allow it to allocate so that it may quickly exit and free | |
722 | * its memory. | |
723 | */ | |
724 | if (fatal_signal_pending(current)) { | |
725 | set_thread_flag(TIF_MEMDIE); | |
726 | return; | |
727 | } | |
728 | ||
9b0f8b04 CL |
729 | /* |
730 | * Check if there were limitations on the allocation (only relevant for | |
731 | * NUMA) that may require different handling. | |
732 | */ | |
a63d83f4 DR |
733 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask, |
734 | &totalpages); | |
e85bfd3a DR |
735 | mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL; |
736 | check_panic_on_oom(constraint, gfp_mask, order, mpol_mask); | |
0aad4b31 | 737 | |
2b45ab33 | 738 | read_lock(&tasklist_lock); |
f88ccad5 | 739 | if (sysctl_oom_kill_allocating_task && |
a96cfd6e | 740 | !oom_unkillable_task(current, NULL, nodemask) && |
c9f01245 | 741 | current->mm) { |
0aad4b31 DR |
742 | /* |
743 | * oom_kill_process() needs tasklist_lock held. If it returns | |
744 | * non-zero, current could not be killed so we must fallback to | |
745 | * the tasklist scan. | |
746 | */ | |
a63d83f4 DR |
747 | if (!oom_kill_process(current, gfp_mask, order, 0, totalpages, |
748 | NULL, nodemask, | |
0aad4b31 | 749 | "Out of memory (oom_kill_allocating_task)")) |
b52723c5 | 750 | goto out; |
0aad4b31 DR |
751 | } |
752 | ||
753 | retry: | |
e85bfd3a | 754 | p = select_bad_process(&points, totalpages, NULL, mpol_mask); |
0aad4b31 | 755 | if (PTR_ERR(p) == -1UL) |
b52723c5 | 756 | goto out; |
0aad4b31 DR |
757 | |
758 | /* Found nothing?!?! Either we hang forever, or we panic. */ | |
759 | if (!p) { | |
e85bfd3a | 760 | dump_header(NULL, gfp_mask, order, NULL, mpol_mask); |
0aad4b31 DR |
761 | read_unlock(&tasklist_lock); |
762 | panic("Out of memory and no killable processes...\n"); | |
763 | } | |
764 | ||
a63d83f4 DR |
765 | if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL, |
766 | nodemask, "Out of memory")) | |
0aad4b31 | 767 | goto retry; |
b52723c5 KM |
768 | killed = 1; |
769 | out: | |
140ffcec | 770 | read_unlock(&tasklist_lock); |
1da177e4 LT |
771 | |
772 | /* | |
773 | * Give "p" a good chance of killing itself before we | |
2f659f46 | 774 | * retry to allocate memory unless "p" is current |
1da177e4 | 775 | */ |
b52723c5 | 776 | if (killed && !test_thread_flag(TIF_MEMDIE)) |
140ffcec | 777 | schedule_timeout_uninterruptible(1); |
1da177e4 | 778 | } |
e3658932 DR |
779 | |
780 | /* | |
781 | * The pagefault handler calls here because it is out of memory, so kill a | |
782 | * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel | |
783 | * oom killing is already in progress so do nothing. If a task is found with | |
784 | * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit. | |
785 | */ | |
786 | void pagefault_out_of_memory(void) | |
787 | { | |
788 | if (try_set_system_oom()) { | |
789 | out_of_memory(NULL, 0, 0, NULL); | |
790 | clear_system_oom(); | |
791 | } | |
792 | if (!test_thread_flag(TIF_MEMDIE)) | |
793 | schedule_timeout_uninterruptible(1); | |
794 | } |