]>
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> |
43d2b113 | 36 | #include <linux/ftrace.h> |
dc3f21ea | 37 | #include <linux/ratelimit.h> |
43d2b113 KH |
38 | |
39 | #define CREATE_TRACE_POINTS | |
40 | #include <trace/events/oom.h> | |
1da177e4 | 41 | |
fadd8fbd | 42 | int sysctl_panic_on_oom; |
fe071d7e | 43 | int sysctl_oom_kill_allocating_task; |
ad915c43 | 44 | int sysctl_oom_dump_tasks = 1; |
c7d4caeb | 45 | static DEFINE_SPINLOCK(zone_scan_lock); |
1da177e4 | 46 | |
43362a49 DR |
47 | /* |
48 | * compare_swap_oom_score_adj() - compare and swap current's oom_score_adj | |
49 | * @old_val: old oom_score_adj for compare | |
50 | * @new_val: new oom_score_adj for swap | |
51 | * | |
52 | * Sets the oom_score_adj value for current to @new_val iff its present value is | |
53 | * @old_val. Usually used to reinstate a previous value to prevent racing with | |
54 | * userspacing tuning the value in the interim. | |
55 | */ | |
56 | void compare_swap_oom_score_adj(int old_val, int new_val) | |
57 | { | |
58 | struct sighand_struct *sighand = current->sighand; | |
59 | ||
60 | spin_lock_irq(&sighand->siglock); | |
61 | if (current->signal->oom_score_adj == old_val) | |
62 | current->signal->oom_score_adj = new_val; | |
43d2b113 | 63 | trace_oom_score_adj_update(current); |
43362a49 DR |
64 | spin_unlock_irq(&sighand->siglock); |
65 | } | |
66 | ||
72788c38 DR |
67 | /** |
68 | * test_set_oom_score_adj() - set current's oom_score_adj and return old value | |
69 | * @new_val: new oom_score_adj value | |
70 | * | |
71 | * Sets the oom_score_adj value for current to @new_val with proper | |
72 | * synchronization and returns the old value. Usually used to temporarily | |
73 | * set a value, save the old value in the caller, and then reinstate it later. | |
74 | */ | |
75 | int test_set_oom_score_adj(int new_val) | |
76 | { | |
77 | struct sighand_struct *sighand = current->sighand; | |
78 | int old_val; | |
79 | ||
80 | spin_lock_irq(&sighand->siglock); | |
81 | old_val = current->signal->oom_score_adj; | |
c9f01245 | 82 | current->signal->oom_score_adj = new_val; |
43d2b113 | 83 | trace_oom_score_adj_update(current); |
72788c38 DR |
84 | spin_unlock_irq(&sighand->siglock); |
85 | ||
86 | return old_val; | |
87 | } | |
88 | ||
6f48d0eb DR |
89 | #ifdef CONFIG_NUMA |
90 | /** | |
91 | * has_intersects_mems_allowed() - check task eligiblity for kill | |
92 | * @tsk: task struct of which task to consider | |
93 | * @mask: nodemask passed to page allocator for mempolicy ooms | |
94 | * | |
95 | * Task eligibility is determined by whether or not a candidate task, @tsk, | |
96 | * shares the same mempolicy nodes as current if it is bound by such a policy | |
97 | * and whether or not it has the same set of allowed cpuset nodes. | |
495789a5 | 98 | */ |
6f48d0eb DR |
99 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
100 | const nodemask_t *mask) | |
495789a5 | 101 | { |
6f48d0eb | 102 | struct task_struct *start = tsk; |
495789a5 | 103 | |
495789a5 | 104 | do { |
6f48d0eb DR |
105 | if (mask) { |
106 | /* | |
107 | * If this is a mempolicy constrained oom, tsk's | |
108 | * cpuset is irrelevant. Only return true if its | |
109 | * mempolicy intersects current, otherwise it may be | |
110 | * needlessly killed. | |
111 | */ | |
112 | if (mempolicy_nodemask_intersects(tsk, mask)) | |
113 | return true; | |
114 | } else { | |
115 | /* | |
116 | * This is not a mempolicy constrained oom, so only | |
117 | * check the mems of tsk's cpuset. | |
118 | */ | |
119 | if (cpuset_mems_allowed_intersects(current, tsk)) | |
120 | return true; | |
121 | } | |
df1090a8 KM |
122 | } while_each_thread(start, tsk); |
123 | ||
6f48d0eb DR |
124 | return false; |
125 | } | |
126 | #else | |
127 | static bool has_intersects_mems_allowed(struct task_struct *tsk, | |
128 | const nodemask_t *mask) | |
129 | { | |
130 | return true; | |
495789a5 | 131 | } |
6f48d0eb | 132 | #endif /* CONFIG_NUMA */ |
495789a5 | 133 | |
6f48d0eb DR |
134 | /* |
135 | * The process p may have detached its own ->mm while exiting or through | |
136 | * use_mm(), but one or more of its subthreads may still have a valid | |
137 | * pointer. Return p, or any of its subthreads with a valid ->mm, with | |
138 | * task_lock() held. | |
139 | */ | |
158e0a2d | 140 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
dd8e8f40 ON |
141 | { |
142 | struct task_struct *t = p; | |
143 | ||
144 | do { | |
145 | task_lock(t); | |
146 | if (likely(t->mm)) | |
147 | return t; | |
148 | task_unlock(t); | |
149 | } while_each_thread(p, t); | |
150 | ||
151 | return NULL; | |
152 | } | |
153 | ||
ab290adb | 154 | /* return true if the task is not adequate as candidate victim task. */ |
e85bfd3a | 155 | static bool oom_unkillable_task(struct task_struct *p, |
72835c86 | 156 | const struct mem_cgroup *memcg, const nodemask_t *nodemask) |
ab290adb KM |
157 | { |
158 | if (is_global_init(p)) | |
159 | return true; | |
160 | if (p->flags & PF_KTHREAD) | |
161 | return true; | |
162 | ||
163 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ | |
72835c86 | 164 | if (memcg && !task_in_mem_cgroup(p, memcg)) |
ab290adb KM |
165 | return true; |
166 | ||
167 | /* p may not have freeable memory in nodemask */ | |
168 | if (!has_intersects_mems_allowed(p, nodemask)) | |
169 | return true; | |
170 | ||
171 | return false; | |
172 | } | |
173 | ||
1da177e4 | 174 | /** |
a63d83f4 | 175 | * oom_badness - heuristic function to determine which candidate task to kill |
1da177e4 | 176 | * @p: task struct of which task we should calculate |
a63d83f4 | 177 | * @totalpages: total present RAM allowed for page allocation |
1da177e4 | 178 | * |
a63d83f4 DR |
179 | * The heuristic for determining which task to kill is made to be as simple and |
180 | * predictable as possible. The goal is to return the highest value for the | |
181 | * task consuming the most memory to avoid subsequent oom failures. | |
1da177e4 | 182 | */ |
a7f638f9 DR |
183 | unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg, |
184 | const nodemask_t *nodemask, unsigned long totalpages) | |
1da177e4 | 185 | { |
1e11ad8d | 186 | long points; |
61eafb00 | 187 | long adj; |
28b83c51 | 188 | |
72835c86 | 189 | if (oom_unkillable_task(p, memcg, nodemask)) |
26ebc984 | 190 | return 0; |
1da177e4 | 191 | |
dd8e8f40 ON |
192 | p = find_lock_task_mm(p); |
193 | if (!p) | |
1da177e4 LT |
194 | return 0; |
195 | ||
61eafb00 DR |
196 | adj = p->signal->oom_score_adj; |
197 | if (adj == OOM_SCORE_ADJ_MIN) { | |
5aecc85a MH |
198 | task_unlock(p); |
199 | return 0; | |
200 | } | |
201 | ||
1da177e4 | 202 | /* |
a63d83f4 | 203 | * The baseline for the badness score is the proportion of RAM that each |
f755a042 | 204 | * task's rss, pagetable and swap space use. |
1da177e4 | 205 | */ |
a7f638f9 DR |
206 | points = get_mm_rss(p->mm) + p->mm->nr_ptes + |
207 | get_mm_counter(p->mm, MM_SWAPENTS); | |
a63d83f4 | 208 | task_unlock(p); |
1da177e4 LT |
209 | |
210 | /* | |
a63d83f4 DR |
211 | * Root processes get 3% bonus, just like the __vm_enough_memory() |
212 | * implementation used by LSMs. | |
1da177e4 | 213 | */ |
a63d83f4 | 214 | if (has_capability_noaudit(p, CAP_SYS_ADMIN)) |
61eafb00 | 215 | adj -= 30; |
1da177e4 | 216 | |
61eafb00 DR |
217 | /* Normalize to oom_score_adj units */ |
218 | adj *= totalpages / 1000; | |
219 | points += adj; | |
1da177e4 | 220 | |
f19e8aa1 | 221 | /* |
a7f638f9 DR |
222 | * Never return 0 for an eligible task regardless of the root bonus and |
223 | * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here). | |
f19e8aa1 | 224 | */ |
1e11ad8d | 225 | return points > 0 ? points : 1; |
1da177e4 LT |
226 | } |
227 | ||
9b0f8b04 CL |
228 | /* |
229 | * Determine the type of allocation constraint. | |
230 | */ | |
9b0f8b04 | 231 | #ifdef CONFIG_NUMA |
4365a567 | 232 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
a63d83f4 DR |
233 | gfp_t gfp_mask, nodemask_t *nodemask, |
234 | unsigned long *totalpages) | |
4365a567 | 235 | { |
54a6eb5c | 236 | struct zone *zone; |
dd1a239f | 237 | struct zoneref *z; |
54a6eb5c | 238 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
a63d83f4 DR |
239 | bool cpuset_limited = false; |
240 | int nid; | |
9b0f8b04 | 241 | |
a63d83f4 DR |
242 | /* Default to all available memory */ |
243 | *totalpages = totalram_pages + total_swap_pages; | |
244 | ||
245 | if (!zonelist) | |
246 | return CONSTRAINT_NONE; | |
4365a567 KH |
247 | /* |
248 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid | |
249 | * to kill current.We have to random task kill in this case. | |
250 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. | |
251 | */ | |
252 | if (gfp_mask & __GFP_THISNODE) | |
253 | return CONSTRAINT_NONE; | |
9b0f8b04 | 254 | |
4365a567 | 255 | /* |
a63d83f4 DR |
256 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
257 | * the page allocator means a mempolicy is in effect. Cpuset policy | |
258 | * is enforced in get_page_from_freelist(). | |
4365a567 | 259 | */ |
a63d83f4 DR |
260 | if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) { |
261 | *totalpages = total_swap_pages; | |
262 | for_each_node_mask(nid, *nodemask) | |
263 | *totalpages += node_spanned_pages(nid); | |
9b0f8b04 | 264 | return CONSTRAINT_MEMORY_POLICY; |
a63d83f4 | 265 | } |
4365a567 KH |
266 | |
267 | /* Check this allocation failure is caused by cpuset's wall function */ | |
268 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
269 | high_zoneidx, nodemask) | |
270 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) | |
a63d83f4 | 271 | cpuset_limited = true; |
9b0f8b04 | 272 | |
a63d83f4 DR |
273 | if (cpuset_limited) { |
274 | *totalpages = total_swap_pages; | |
275 | for_each_node_mask(nid, cpuset_current_mems_allowed) | |
276 | *totalpages += node_spanned_pages(nid); | |
277 | return CONSTRAINT_CPUSET; | |
278 | } | |
9b0f8b04 CL |
279 | return CONSTRAINT_NONE; |
280 | } | |
4365a567 KH |
281 | #else |
282 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, | |
a63d83f4 DR |
283 | gfp_t gfp_mask, nodemask_t *nodemask, |
284 | unsigned long *totalpages) | |
4365a567 | 285 | { |
a63d83f4 | 286 | *totalpages = totalram_pages + total_swap_pages; |
4365a567 KH |
287 | return CONSTRAINT_NONE; |
288 | } | |
289 | #endif | |
9b0f8b04 | 290 | |
1da177e4 LT |
291 | /* |
292 | * Simple selection loop. We chose the process with the highest | |
293 | * number of 'points'. We expect the caller will lock the tasklist. | |
294 | * | |
295 | * (not docbooked, we don't want this one cluttering up the manual) | |
296 | */ | |
a63d83f4 | 297 | static struct task_struct *select_bad_process(unsigned int *ppoints, |
72835c86 | 298 | unsigned long totalpages, struct mem_cgroup *memcg, |
08ab9b10 | 299 | const nodemask_t *nodemask, bool force_kill) |
1da177e4 | 300 | { |
3a5dda7a | 301 | struct task_struct *g, *p; |
1da177e4 | 302 | struct task_struct *chosen = NULL; |
a7f638f9 | 303 | unsigned long chosen_points = 0; |
1da177e4 | 304 | |
3a5dda7a | 305 | do_each_thread(g, p) { |
a63d83f4 | 306 | unsigned int points; |
a49335cc | 307 | |
c027a474 | 308 | if (p->exit_state) |
30e2b41f | 309 | continue; |
72835c86 | 310 | if (oom_unkillable_task(p, memcg, nodemask)) |
6cf86ac6 | 311 | continue; |
ef08e3b4 | 312 | |
b78483a4 NP |
313 | /* |
314 | * This task already has access to memory reserves and is | |
315 | * being killed. Don't allow any other task access to the | |
316 | * memory reserve. | |
317 | * | |
318 | * Note: this may have a chance of deadlock if it gets | |
319 | * blocked waiting for another task which itself is waiting | |
320 | * for memory. Is there a better alternative? | |
321 | */ | |
f660daac DR |
322 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) { |
323 | if (unlikely(frozen(p))) | |
a5be2d0d | 324 | __thaw_task(p); |
08ab9b10 DR |
325 | if (!force_kill) |
326 | return ERR_PTR(-1UL); | |
f660daac | 327 | } |
c027a474 ON |
328 | if (!p->mm) |
329 | continue; | |
b78483a4 | 330 | |
30e2b41f | 331 | if (p->flags & PF_EXITING) { |
edd45544 DR |
332 | /* |
333 | * If p is the current task and is in the process of | |
334 | * releasing memory, we allow the "kill" to set | |
335 | * TIF_MEMDIE, which will allow it to gain access to | |
336 | * memory reserves. Otherwise, it may stall forever. | |
337 | * | |
338 | * The loop isn't broken here, however, in case other | |
339 | * threads are found to have already been oom killed. | |
340 | */ | |
341 | if (p == current) { | |
342 | chosen = p; | |
a7f638f9 | 343 | chosen_points = ULONG_MAX; |
08ab9b10 | 344 | } else if (!force_kill) { |
edd45544 DR |
345 | /* |
346 | * If this task is not being ptraced on exit, | |
347 | * then wait for it to finish before killing | |
348 | * some other task unnecessarily. | |
349 | */ | |
d21142ec | 350 | if (!(p->group_leader->ptrace & PT_TRACE_EXIT)) |
edd45544 DR |
351 | return ERR_PTR(-1UL); |
352 | } | |
50ec3bbf | 353 | } |
972c4ea5 | 354 | |
72835c86 | 355 | points = oom_badness(p, memcg, nodemask, totalpages); |
a7f638f9 | 356 | if (points > chosen_points) { |
a49335cc | 357 | chosen = p; |
a7f638f9 | 358 | chosen_points = points; |
1da177e4 | 359 | } |
3a5dda7a | 360 | } while_each_thread(g, p); |
972c4ea5 | 361 | |
a7f638f9 | 362 | *ppoints = chosen_points * 1000 / totalpages; |
1da177e4 LT |
363 | return chosen; |
364 | } | |
365 | ||
fef1bdd6 | 366 | /** |
1b578df0 | 367 | * dump_tasks - dump current memory state of all system tasks |
74ab7f1d | 368 | * @mem: current's memory controller, if constrained |
e85bfd3a | 369 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
1b578df0 | 370 | * |
e85bfd3a DR |
371 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
372 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes | |
373 | * are not shown. | |
fef1bdd6 | 374 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj |
a63d83f4 | 375 | * value, oom_score_adj value, and name. |
fef1bdd6 | 376 | * |
fef1bdd6 DR |
377 | * Call with tasklist_lock read-locked. |
378 | */ | |
72835c86 | 379 | static void dump_tasks(const struct mem_cgroup *memcg, const nodemask_t *nodemask) |
fef1bdd6 | 380 | { |
c55db957 KM |
381 | struct task_struct *p; |
382 | struct task_struct *task; | |
fef1bdd6 | 383 | |
a63d83f4 | 384 | pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n"); |
c55db957 | 385 | for_each_process(p) { |
72835c86 | 386 | if (oom_unkillable_task(p, memcg, nodemask)) |
b4416d2b | 387 | continue; |
fef1bdd6 | 388 | |
c55db957 KM |
389 | task = find_lock_task_mm(p); |
390 | if (!task) { | |
6d2661ed | 391 | /* |
74ab7f1d DR |
392 | * This is a kthread or all of p's threads have already |
393 | * detached their mm's. There's no need to report | |
c55db957 | 394 | * them; they can't be oom killed anyway. |
6d2661ed | 395 | */ |
6d2661ed DR |
396 | continue; |
397 | } | |
c55db957 | 398 | |
a63d83f4 | 399 | pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n", |
078de5f7 EB |
400 | task->pid, from_kuid(&init_user_ns, task_uid(task)), |
401 | task->tgid, task->mm->total_vm, get_mm_rss(task->mm), | |
a63d83f4 DR |
402 | task_cpu(task), task->signal->oom_adj, |
403 | task->signal->oom_score_adj, task->comm); | |
c55db957 KM |
404 | task_unlock(task); |
405 | } | |
fef1bdd6 DR |
406 | } |
407 | ||
d31f56db | 408 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
72835c86 | 409 | struct mem_cgroup *memcg, const nodemask_t *nodemask) |
1b604d75 | 410 | { |
5e9d834a | 411 | task_lock(current); |
1b604d75 | 412 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " |
a63d83f4 DR |
413 | "oom_adj=%d, oom_score_adj=%d\n", |
414 | current->comm, gfp_mask, order, current->signal->oom_adj, | |
415 | current->signal->oom_score_adj); | |
1b604d75 DR |
416 | cpuset_print_task_mems_allowed(current); |
417 | task_unlock(current); | |
418 | dump_stack(); | |
72835c86 | 419 | mem_cgroup_print_oom_info(memcg, p); |
b2b755b5 | 420 | show_mem(SHOW_MEM_FILTER_NODES); |
1b604d75 | 421 | if (sysctl_oom_dump_tasks) |
72835c86 | 422 | dump_tasks(memcg, nodemask); |
1b604d75 DR |
423 | } |
424 | ||
3b4798cb | 425 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
2a1c9b1f DR |
426 | static void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
427 | unsigned int points, unsigned long totalpages, | |
428 | struct mem_cgroup *memcg, nodemask_t *nodemask, | |
429 | const char *message) | |
1da177e4 | 430 | { |
52d3c036 | 431 | struct task_struct *victim = p; |
5e9d834a | 432 | struct task_struct *child; |
52d3c036 | 433 | struct task_struct *t = p; |
647f2bdf | 434 | struct mm_struct *mm; |
52d3c036 | 435 | unsigned int victim_points = 0; |
dc3f21ea DR |
436 | static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
437 | DEFAULT_RATELIMIT_BURST); | |
1da177e4 | 438 | |
50ec3bbf NP |
439 | /* |
440 | * If the task is already exiting, don't alarm the sysadmin or kill | |
441 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
442 | */ | |
0753ba01 | 443 | if (p->flags & PF_EXITING) { |
4358997a | 444 | set_tsk_thread_flag(p, TIF_MEMDIE); |
2a1c9b1f | 445 | return; |
50ec3bbf NP |
446 | } |
447 | ||
dc3f21ea | 448 | if (__ratelimit(&oom_rs)) |
8447d950 DR |
449 | dump_header(p, gfp_mask, order, memcg, nodemask); |
450 | ||
5e9d834a | 451 | task_lock(p); |
a63d83f4 | 452 | pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n", |
5e9d834a DR |
453 | message, task_pid_nr(p), p->comm, points); |
454 | task_unlock(p); | |
f3af38d3 | 455 | |
5e9d834a DR |
456 | /* |
457 | * If any of p's children has a different mm and is eligible for kill, | |
11239836 | 458 | * the one with the highest oom_badness() score is sacrificed for its |
5e9d834a DR |
459 | * parent. This attempts to lose the minimal amount of work done while |
460 | * still freeing memory. | |
461 | */ | |
dd8e8f40 | 462 | do { |
5e9d834a | 463 | list_for_each_entry(child, &t->children, sibling) { |
a63d83f4 | 464 | unsigned int child_points; |
5e9d834a | 465 | |
edd45544 DR |
466 | if (child->mm == p->mm) |
467 | continue; | |
a63d83f4 DR |
468 | /* |
469 | * oom_badness() returns 0 if the thread is unkillable | |
470 | */ | |
72835c86 | 471 | child_points = oom_badness(child, memcg, nodemask, |
a63d83f4 | 472 | totalpages); |
5e9d834a DR |
473 | if (child_points > victim_points) { |
474 | victim = child; | |
475 | victim_points = child_points; | |
476 | } | |
dd8e8f40 ON |
477 | } |
478 | } while_each_thread(p, t); | |
479 | ||
647f2bdf DR |
480 | victim = find_lock_task_mm(victim); |
481 | if (!victim) | |
482 | return; | |
483 | ||
484 | /* mm cannot safely be dereferenced after task_unlock(victim) */ | |
485 | mm = victim->mm; | |
486 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", | |
487 | task_pid_nr(victim), victim->comm, K(victim->mm->total_vm), | |
488 | K(get_mm_counter(victim->mm, MM_ANONPAGES)), | |
489 | K(get_mm_counter(victim->mm, MM_FILEPAGES))); | |
490 | task_unlock(victim); | |
491 | ||
492 | /* | |
493 | * Kill all user processes sharing victim->mm in other thread groups, if | |
494 | * any. They don't get access to memory reserves, though, to avoid | |
495 | * depletion of all memory. This prevents mm->mmap_sem livelock when an | |
496 | * oom killed thread cannot exit because it requires the semaphore and | |
497 | * its contended by another thread trying to allocate memory itself. | |
498 | * That thread will now get access to memory reserves since it has a | |
499 | * pending fatal signal. | |
500 | */ | |
501 | for_each_process(p) | |
502 | if (p->mm == mm && !same_thread_group(p, victim) && | |
503 | !(p->flags & PF_KTHREAD)) { | |
504 | if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) | |
505 | continue; | |
506 | ||
507 | task_lock(p); /* Protect ->comm from prctl() */ | |
508 | pr_err("Kill process %d (%s) sharing same memory\n", | |
509 | task_pid_nr(p), p->comm); | |
510 | task_unlock(p); | |
d2d39309 | 511 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true); |
647f2bdf DR |
512 | } |
513 | ||
514 | set_tsk_thread_flag(victim, TIF_MEMDIE); | |
d2d39309 | 515 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true); |
1da177e4 | 516 | } |
647f2bdf | 517 | #undef K |
1da177e4 | 518 | |
309ed882 DR |
519 | /* |
520 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. | |
521 | */ | |
522 | static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask, | |
e85bfd3a | 523 | int order, const nodemask_t *nodemask) |
309ed882 DR |
524 | { |
525 | if (likely(!sysctl_panic_on_oom)) | |
526 | return; | |
527 | if (sysctl_panic_on_oom != 2) { | |
528 | /* | |
529 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel | |
530 | * does not panic for cpuset, mempolicy, or memcg allocation | |
531 | * failures. | |
532 | */ | |
533 | if (constraint != CONSTRAINT_NONE) | |
534 | return; | |
535 | } | |
536 | read_lock(&tasklist_lock); | |
e85bfd3a | 537 | dump_header(NULL, gfp_mask, order, NULL, nodemask); |
309ed882 DR |
538 | read_unlock(&tasklist_lock); |
539 | panic("Out of memory: %s panic_on_oom is enabled\n", | |
540 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); | |
541 | } | |
542 | ||
00f0b825 | 543 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
e845e199 DR |
544 | void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
545 | int order) | |
c7ba5c9e | 546 | { |
a63d83f4 DR |
547 | unsigned long limit; |
548 | unsigned int points = 0; | |
c7ba5c9e PE |
549 | struct task_struct *p; |
550 | ||
f9434ad1 DR |
551 | /* |
552 | * If current has a pending SIGKILL, then automatically select it. The | |
553 | * goal is to allow it to allocate so that it may quickly exit and free | |
554 | * its memory. | |
555 | */ | |
556 | if (fatal_signal_pending(current)) { | |
557 | set_thread_flag(TIF_MEMDIE); | |
f9434ad1 DR |
558 | return; |
559 | } | |
560 | ||
e845e199 | 561 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); |
a7f638f9 | 562 | limit = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
e115f2d8 | 563 | read_lock(&tasklist_lock); |
08ab9b10 | 564 | p = select_bad_process(&points, limit, memcg, NULL, false); |
2a1c9b1f | 565 | if (p && PTR_ERR(p) != -1UL) |
e845e199 | 566 | oom_kill_process(p, gfp_mask, order, points, limit, memcg, NULL, |
2a1c9b1f | 567 | "Memory cgroup out of memory"); |
e115f2d8 | 568 | read_unlock(&tasklist_lock); |
c7ba5c9e PE |
569 | } |
570 | #endif | |
571 | ||
8bc719d3 MS |
572 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
573 | ||
574 | int register_oom_notifier(struct notifier_block *nb) | |
575 | { | |
576 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
577 | } | |
578 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
579 | ||
580 | int unregister_oom_notifier(struct notifier_block *nb) | |
581 | { | |
582 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
583 | } | |
584 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
585 | ||
098d7f12 DR |
586 | /* |
587 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero | |
588 | * if a parallel OOM killing is already taking place that includes a zone in | |
589 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. | |
590 | */ | |
ff321fea | 591 | int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 592 | { |
dd1a239f MG |
593 | struct zoneref *z; |
594 | struct zone *zone; | |
098d7f12 DR |
595 | int ret = 1; |
596 | ||
c7d4caeb | 597 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
598 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
599 | if (zone_is_oom_locked(zone)) { | |
098d7f12 DR |
600 | ret = 0; |
601 | goto out; | |
602 | } | |
dd1a239f MG |
603 | } |
604 | ||
605 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
606 | /* | |
c7d4caeb | 607 | * Lock each zone in the zonelist under zone_scan_lock so a |
ff321fea | 608 | * parallel invocation of try_set_zonelist_oom() doesn't succeed |
dd1a239f MG |
609 | * when it shouldn't. |
610 | */ | |
611 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
612 | } | |
098d7f12 | 613 | |
098d7f12 | 614 | out: |
c7d4caeb | 615 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
616 | return ret; |
617 | } | |
618 | ||
619 | /* | |
620 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed | |
621 | * allocation attempts with zonelists containing them may now recall the OOM | |
622 | * killer, if necessary. | |
623 | */ | |
dd1a239f | 624 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 625 | { |
dd1a239f MG |
626 | struct zoneref *z; |
627 | struct zone *zone; | |
098d7f12 | 628 | |
c7d4caeb | 629 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
630 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
631 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
632 | } | |
c7d4caeb | 633 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
634 | } |
635 | ||
e3658932 DR |
636 | /* |
637 | * Try to acquire the oom killer lock for all system zones. Returns zero if a | |
638 | * parallel oom killing is taking place, otherwise locks all zones and returns | |
639 | * non-zero. | |
640 | */ | |
641 | static int try_set_system_oom(void) | |
642 | { | |
643 | struct zone *zone; | |
644 | int ret = 1; | |
645 | ||
646 | spin_lock(&zone_scan_lock); | |
647 | for_each_populated_zone(zone) | |
648 | if (zone_is_oom_locked(zone)) { | |
649 | ret = 0; | |
650 | goto out; | |
651 | } | |
652 | for_each_populated_zone(zone) | |
653 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
654 | out: | |
655 | spin_unlock(&zone_scan_lock); | |
656 | return ret; | |
657 | } | |
658 | ||
659 | /* | |
660 | * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation | |
661 | * attempts or page faults may now recall the oom killer, if necessary. | |
662 | */ | |
663 | static void clear_system_oom(void) | |
664 | { | |
665 | struct zone *zone; | |
666 | ||
667 | spin_lock(&zone_scan_lock); | |
668 | for_each_populated_zone(zone) | |
669 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
670 | spin_unlock(&zone_scan_lock); | |
671 | } | |
672 | ||
1da177e4 | 673 | /** |
6937a25c | 674 | * out_of_memory - kill the "best" process when we run out of memory |
1b578df0 RD |
675 | * @zonelist: zonelist pointer |
676 | * @gfp_mask: memory allocation flags | |
677 | * @order: amount of memory being requested as a power of 2 | |
6f48d0eb | 678 | * @nodemask: nodemask passed to page allocator |
08ab9b10 | 679 | * @force_kill: true if a task must be killed, even if others are exiting |
1da177e4 LT |
680 | * |
681 | * If we run out of memory, we have the choice between either | |
682 | * killing a random task (bad), letting the system crash (worse) | |
683 | * OR try to be smart about which process to kill. Note that we | |
684 | * don't have to be perfect here, we just have to be good. | |
685 | */ | |
4365a567 | 686 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
08ab9b10 | 687 | int order, nodemask_t *nodemask, bool force_kill) |
1da177e4 | 688 | { |
e85bfd3a | 689 | const nodemask_t *mpol_mask; |
0aad4b31 | 690 | struct task_struct *p; |
a63d83f4 | 691 | unsigned long totalpages; |
8bc719d3 | 692 | unsigned long freed = 0; |
a63d83f4 | 693 | unsigned int points; |
e3658932 | 694 | enum oom_constraint constraint = CONSTRAINT_NONE; |
b52723c5 | 695 | int killed = 0; |
8bc719d3 MS |
696 | |
697 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
698 | if (freed > 0) | |
699 | /* Got some memory back in the last second. */ | |
700 | return; | |
1da177e4 | 701 | |
7b98c2e4 DR |
702 | /* |
703 | * If current has a pending SIGKILL, then automatically select it. The | |
704 | * goal is to allow it to allocate so that it may quickly exit and free | |
705 | * its memory. | |
706 | */ | |
707 | if (fatal_signal_pending(current)) { | |
708 | set_thread_flag(TIF_MEMDIE); | |
709 | return; | |
710 | } | |
711 | ||
9b0f8b04 CL |
712 | /* |
713 | * Check if there were limitations on the allocation (only relevant for | |
714 | * NUMA) that may require different handling. | |
715 | */ | |
a63d83f4 DR |
716 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask, |
717 | &totalpages); | |
e85bfd3a DR |
718 | mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL; |
719 | check_panic_on_oom(constraint, gfp_mask, order, mpol_mask); | |
0aad4b31 | 720 | |
2b45ab33 | 721 | read_lock(&tasklist_lock); |
f88ccad5 | 722 | if (sysctl_oom_kill_allocating_task && |
a96cfd6e | 723 | !oom_unkillable_task(current, NULL, nodemask) && |
c9f01245 | 724 | current->mm) { |
2a1c9b1f DR |
725 | oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL, |
726 | nodemask, | |
727 | "Out of memory (oom_kill_allocating_task)"); | |
728 | goto out; | |
0aad4b31 DR |
729 | } |
730 | ||
08ab9b10 DR |
731 | p = select_bad_process(&points, totalpages, NULL, mpol_mask, |
732 | force_kill); | |
0aad4b31 DR |
733 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
734 | if (!p) { | |
e85bfd3a | 735 | dump_header(NULL, gfp_mask, order, NULL, mpol_mask); |
0aad4b31 DR |
736 | read_unlock(&tasklist_lock); |
737 | panic("Out of memory and no killable processes...\n"); | |
738 | } | |
2a1c9b1f DR |
739 | if (PTR_ERR(p) != -1UL) { |
740 | oom_kill_process(p, gfp_mask, order, points, totalpages, NULL, | |
741 | nodemask, "Out of memory"); | |
742 | killed = 1; | |
743 | } | |
b52723c5 | 744 | out: |
140ffcec | 745 | read_unlock(&tasklist_lock); |
1da177e4 LT |
746 | |
747 | /* | |
748 | * Give "p" a good chance of killing itself before we | |
2f659f46 | 749 | * retry to allocate memory unless "p" is current |
1da177e4 | 750 | */ |
b52723c5 | 751 | if (killed && !test_thread_flag(TIF_MEMDIE)) |
140ffcec | 752 | schedule_timeout_uninterruptible(1); |
1da177e4 | 753 | } |
e3658932 DR |
754 | |
755 | /* | |
756 | * The pagefault handler calls here because it is out of memory, so kill a | |
757 | * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel | |
758 | * oom killing is already in progress so do nothing. If a task is found with | |
759 | * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit. | |
760 | */ | |
761 | void pagefault_out_of_memory(void) | |
762 | { | |
763 | if (try_set_system_oom()) { | |
08ab9b10 | 764 | out_of_memory(NULL, 0, 0, NULL, false); |
e3658932 DR |
765 | clear_system_oom(); |
766 | } | |
767 | if (!test_thread_flag(TIF_MEMDIE)) | |
768 | schedule_timeout_uninterruptible(1); | |
769 | } |