]>
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; |
dc56401f JW |
45 | |
46 | DEFINE_MUTEX(oom_lock); | |
1da177e4 | 47 | |
6f48d0eb DR |
48 | #ifdef CONFIG_NUMA |
49 | /** | |
50 | * has_intersects_mems_allowed() - check task eligiblity for kill | |
ad962441 | 51 | * @start: task struct of which task to consider |
6f48d0eb DR |
52 | * @mask: nodemask passed to page allocator for mempolicy ooms |
53 | * | |
54 | * Task eligibility is determined by whether or not a candidate task, @tsk, | |
55 | * shares the same mempolicy nodes as current if it is bound by such a policy | |
56 | * and whether or not it has the same set of allowed cpuset nodes. | |
495789a5 | 57 | */ |
ad962441 | 58 | static bool has_intersects_mems_allowed(struct task_struct *start, |
6f48d0eb | 59 | const nodemask_t *mask) |
495789a5 | 60 | { |
ad962441 ON |
61 | struct task_struct *tsk; |
62 | bool ret = false; | |
495789a5 | 63 | |
ad962441 | 64 | rcu_read_lock(); |
1da4db0c | 65 | for_each_thread(start, tsk) { |
6f48d0eb DR |
66 | if (mask) { |
67 | /* | |
68 | * If this is a mempolicy constrained oom, tsk's | |
69 | * cpuset is irrelevant. Only return true if its | |
70 | * mempolicy intersects current, otherwise it may be | |
71 | * needlessly killed. | |
72 | */ | |
ad962441 | 73 | ret = mempolicy_nodemask_intersects(tsk, mask); |
6f48d0eb DR |
74 | } else { |
75 | /* | |
76 | * This is not a mempolicy constrained oom, so only | |
77 | * check the mems of tsk's cpuset. | |
78 | */ | |
ad962441 | 79 | ret = cpuset_mems_allowed_intersects(current, tsk); |
6f48d0eb | 80 | } |
ad962441 ON |
81 | if (ret) |
82 | break; | |
1da4db0c | 83 | } |
ad962441 | 84 | rcu_read_unlock(); |
df1090a8 | 85 | |
ad962441 | 86 | return ret; |
6f48d0eb DR |
87 | } |
88 | #else | |
89 | static bool has_intersects_mems_allowed(struct task_struct *tsk, | |
90 | const nodemask_t *mask) | |
91 | { | |
92 | return true; | |
495789a5 | 93 | } |
6f48d0eb | 94 | #endif /* CONFIG_NUMA */ |
495789a5 | 95 | |
6f48d0eb DR |
96 | /* |
97 | * The process p may have detached its own ->mm while exiting or through | |
98 | * use_mm(), but one or more of its subthreads may still have a valid | |
99 | * pointer. Return p, or any of its subthreads with a valid ->mm, with | |
100 | * task_lock() held. | |
101 | */ | |
158e0a2d | 102 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
dd8e8f40 | 103 | { |
1da4db0c | 104 | struct task_struct *t; |
dd8e8f40 | 105 | |
4d4048be ON |
106 | rcu_read_lock(); |
107 | ||
1da4db0c | 108 | for_each_thread(p, t) { |
dd8e8f40 ON |
109 | task_lock(t); |
110 | if (likely(t->mm)) | |
4d4048be | 111 | goto found; |
dd8e8f40 | 112 | task_unlock(t); |
1da4db0c | 113 | } |
4d4048be ON |
114 | t = NULL; |
115 | found: | |
116 | rcu_read_unlock(); | |
dd8e8f40 | 117 | |
4d4048be | 118 | return t; |
dd8e8f40 ON |
119 | } |
120 | ||
ab290adb | 121 | /* return true if the task is not adequate as candidate victim task. */ |
e85bfd3a | 122 | static bool oom_unkillable_task(struct task_struct *p, |
2314b42d | 123 | struct mem_cgroup *memcg, const nodemask_t *nodemask) |
ab290adb KM |
124 | { |
125 | if (is_global_init(p)) | |
126 | return true; | |
127 | if (p->flags & PF_KTHREAD) | |
128 | return true; | |
129 | ||
130 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ | |
72835c86 | 131 | if (memcg && !task_in_mem_cgroup(p, memcg)) |
ab290adb KM |
132 | return true; |
133 | ||
134 | /* p may not have freeable memory in nodemask */ | |
135 | if (!has_intersects_mems_allowed(p, nodemask)) | |
136 | return true; | |
137 | ||
138 | return false; | |
139 | } | |
140 | ||
1da177e4 | 141 | /** |
a63d83f4 | 142 | * oom_badness - heuristic function to determine which candidate task to kill |
1da177e4 | 143 | * @p: task struct of which task we should calculate |
a63d83f4 | 144 | * @totalpages: total present RAM allowed for page allocation |
1da177e4 | 145 | * |
a63d83f4 DR |
146 | * The heuristic for determining which task to kill is made to be as simple and |
147 | * predictable as possible. The goal is to return the highest value for the | |
148 | * task consuming the most memory to avoid subsequent oom failures. | |
1da177e4 | 149 | */ |
a7f638f9 DR |
150 | unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg, |
151 | const nodemask_t *nodemask, unsigned long totalpages) | |
1da177e4 | 152 | { |
1e11ad8d | 153 | long points; |
61eafb00 | 154 | long adj; |
28b83c51 | 155 | |
72835c86 | 156 | if (oom_unkillable_task(p, memcg, nodemask)) |
26ebc984 | 157 | return 0; |
1da177e4 | 158 | |
dd8e8f40 ON |
159 | p = find_lock_task_mm(p); |
160 | if (!p) | |
1da177e4 LT |
161 | return 0; |
162 | ||
a9c58b90 | 163 | adj = (long)p->signal->oom_score_adj; |
61eafb00 | 164 | if (adj == OOM_SCORE_ADJ_MIN) { |
5aecc85a MH |
165 | task_unlock(p); |
166 | return 0; | |
167 | } | |
168 | ||
1da177e4 | 169 | /* |
a63d83f4 | 170 | * The baseline for the badness score is the proportion of RAM that each |
f755a042 | 171 | * task's rss, pagetable and swap space use. |
1da177e4 | 172 | */ |
dc6c9a35 KS |
173 | points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) + |
174 | atomic_long_read(&p->mm->nr_ptes) + mm_nr_pmds(p->mm); | |
a63d83f4 | 175 | task_unlock(p); |
1da177e4 LT |
176 | |
177 | /* | |
a63d83f4 DR |
178 | * Root processes get 3% bonus, just like the __vm_enough_memory() |
179 | * implementation used by LSMs. | |
1da177e4 | 180 | */ |
a63d83f4 | 181 | if (has_capability_noaudit(p, CAP_SYS_ADMIN)) |
778c14af | 182 | points -= (points * 3) / 100; |
1da177e4 | 183 | |
61eafb00 DR |
184 | /* Normalize to oom_score_adj units */ |
185 | adj *= totalpages / 1000; | |
186 | points += adj; | |
1da177e4 | 187 | |
f19e8aa1 | 188 | /* |
a7f638f9 DR |
189 | * Never return 0 for an eligible task regardless of the root bonus and |
190 | * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here). | |
f19e8aa1 | 191 | */ |
1e11ad8d | 192 | return points > 0 ? points : 1; |
1da177e4 LT |
193 | } |
194 | ||
9b0f8b04 CL |
195 | /* |
196 | * Determine the type of allocation constraint. | |
197 | */ | |
9b0f8b04 | 198 | #ifdef CONFIG_NUMA |
4365a567 | 199 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
a63d83f4 DR |
200 | gfp_t gfp_mask, nodemask_t *nodemask, |
201 | unsigned long *totalpages) | |
4365a567 | 202 | { |
54a6eb5c | 203 | struct zone *zone; |
dd1a239f | 204 | struct zoneref *z; |
54a6eb5c | 205 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
a63d83f4 DR |
206 | bool cpuset_limited = false; |
207 | int nid; | |
9b0f8b04 | 208 | |
a63d83f4 DR |
209 | /* Default to all available memory */ |
210 | *totalpages = totalram_pages + total_swap_pages; | |
211 | ||
212 | if (!zonelist) | |
213 | return CONSTRAINT_NONE; | |
4365a567 KH |
214 | /* |
215 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid | |
216 | * to kill current.We have to random task kill in this case. | |
217 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. | |
218 | */ | |
219 | if (gfp_mask & __GFP_THISNODE) | |
220 | return CONSTRAINT_NONE; | |
9b0f8b04 | 221 | |
4365a567 | 222 | /* |
a63d83f4 DR |
223 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
224 | * the page allocator means a mempolicy is in effect. Cpuset policy | |
225 | * is enforced in get_page_from_freelist(). | |
4365a567 | 226 | */ |
bd3a66c1 | 227 | if (nodemask && !nodes_subset(node_states[N_MEMORY], *nodemask)) { |
a63d83f4 DR |
228 | *totalpages = total_swap_pages; |
229 | for_each_node_mask(nid, *nodemask) | |
230 | *totalpages += node_spanned_pages(nid); | |
9b0f8b04 | 231 | return CONSTRAINT_MEMORY_POLICY; |
a63d83f4 | 232 | } |
4365a567 KH |
233 | |
234 | /* Check this allocation failure is caused by cpuset's wall function */ | |
235 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
236 | high_zoneidx, nodemask) | |
344736f2 | 237 | if (!cpuset_zone_allowed(zone, gfp_mask)) |
a63d83f4 | 238 | cpuset_limited = true; |
9b0f8b04 | 239 | |
a63d83f4 DR |
240 | if (cpuset_limited) { |
241 | *totalpages = total_swap_pages; | |
242 | for_each_node_mask(nid, cpuset_current_mems_allowed) | |
243 | *totalpages += node_spanned_pages(nid); | |
244 | return CONSTRAINT_CPUSET; | |
245 | } | |
9b0f8b04 CL |
246 | return CONSTRAINT_NONE; |
247 | } | |
4365a567 KH |
248 | #else |
249 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, | |
a63d83f4 DR |
250 | gfp_t gfp_mask, nodemask_t *nodemask, |
251 | unsigned long *totalpages) | |
4365a567 | 252 | { |
a63d83f4 | 253 | *totalpages = totalram_pages + total_swap_pages; |
4365a567 KH |
254 | return CONSTRAINT_NONE; |
255 | } | |
256 | #endif | |
9b0f8b04 | 257 | |
9cbb78bb DR |
258 | enum oom_scan_t oom_scan_process_thread(struct task_struct *task, |
259 | unsigned long totalpages, const nodemask_t *nodemask, | |
260 | bool force_kill) | |
462607ec | 261 | { |
9cbb78bb | 262 | if (oom_unkillable_task(task, NULL, nodemask)) |
462607ec DR |
263 | return OOM_SCAN_CONTINUE; |
264 | ||
265 | /* | |
266 | * This task already has access to memory reserves and is being killed. | |
267 | * Don't allow any other task to have access to the reserves. | |
268 | */ | |
269 | if (test_tsk_thread_flag(task, TIF_MEMDIE)) { | |
462607ec DR |
270 | if (!force_kill) |
271 | return OOM_SCAN_ABORT; | |
272 | } | |
273 | if (!task->mm) | |
274 | return OOM_SCAN_CONTINUE; | |
275 | ||
e1e12d2f DR |
276 | /* |
277 | * If task is allocating a lot of memory and has been marked to be | |
278 | * killed first if it triggers an oom, then select it. | |
279 | */ | |
280 | if (oom_task_origin(task)) | |
281 | return OOM_SCAN_SELECT; | |
282 | ||
6a2d5679 ON |
283 | if (task_will_free_mem(task) && !force_kill) |
284 | return OOM_SCAN_ABORT; | |
285 | ||
462607ec DR |
286 | return OOM_SCAN_OK; |
287 | } | |
288 | ||
1da177e4 LT |
289 | /* |
290 | * Simple selection loop. We chose the process with the highest | |
6b4f2b56 | 291 | * number of 'points'. Returns -1 on scan abort. |
1da177e4 LT |
292 | * |
293 | * (not docbooked, we don't want this one cluttering up the manual) | |
294 | */ | |
a63d83f4 | 295 | static struct task_struct *select_bad_process(unsigned int *ppoints, |
9cbb78bb DR |
296 | unsigned long totalpages, const nodemask_t *nodemask, |
297 | bool force_kill) | |
1da177e4 | 298 | { |
3a5dda7a | 299 | struct task_struct *g, *p; |
1da177e4 | 300 | struct task_struct *chosen = NULL; |
a7f638f9 | 301 | unsigned long chosen_points = 0; |
1da177e4 | 302 | |
6b0c81b3 | 303 | rcu_read_lock(); |
1da4db0c | 304 | for_each_process_thread(g, p) { |
a63d83f4 | 305 | unsigned int points; |
a49335cc | 306 | |
9cbb78bb | 307 | switch (oom_scan_process_thread(p, totalpages, nodemask, |
462607ec DR |
308 | force_kill)) { |
309 | case OOM_SCAN_SELECT: | |
310 | chosen = p; | |
311 | chosen_points = ULONG_MAX; | |
312 | /* fall through */ | |
313 | case OOM_SCAN_CONTINUE: | |
c027a474 | 314 | continue; |
462607ec | 315 | case OOM_SCAN_ABORT: |
6b0c81b3 | 316 | rcu_read_unlock(); |
6b4f2b56 | 317 | return (struct task_struct *)(-1UL); |
462607ec DR |
318 | case OOM_SCAN_OK: |
319 | break; | |
320 | }; | |
9cbb78bb | 321 | points = oom_badness(p, NULL, nodemask, totalpages); |
d49ad935 DR |
322 | if (!points || points < chosen_points) |
323 | continue; | |
324 | /* Prefer thread group leaders for display purposes */ | |
325 | if (points == chosen_points && thread_group_leader(chosen)) | |
326 | continue; | |
327 | ||
328 | chosen = p; | |
329 | chosen_points = points; | |
1da4db0c | 330 | } |
6b0c81b3 DR |
331 | if (chosen) |
332 | get_task_struct(chosen); | |
333 | rcu_read_unlock(); | |
972c4ea5 | 334 | |
a7f638f9 | 335 | *ppoints = chosen_points * 1000 / totalpages; |
1da177e4 LT |
336 | return chosen; |
337 | } | |
338 | ||
fef1bdd6 | 339 | /** |
1b578df0 | 340 | * dump_tasks - dump current memory state of all system tasks |
dad7557e | 341 | * @memcg: current's memory controller, if constrained |
e85bfd3a | 342 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
1b578df0 | 343 | * |
e85bfd3a DR |
344 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
345 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes | |
346 | * are not shown. | |
de34d965 DR |
347 | * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes, |
348 | * swapents, oom_score_adj value, and name. | |
fef1bdd6 | 349 | */ |
2314b42d | 350 | static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask) |
fef1bdd6 | 351 | { |
c55db957 KM |
352 | struct task_struct *p; |
353 | struct task_struct *task; | |
fef1bdd6 | 354 | |
dc6c9a35 | 355 | pr_info("[ pid ] uid tgid total_vm rss nr_ptes nr_pmds swapents oom_score_adj name\n"); |
6b0c81b3 | 356 | rcu_read_lock(); |
c55db957 | 357 | for_each_process(p) { |
72835c86 | 358 | if (oom_unkillable_task(p, memcg, nodemask)) |
b4416d2b | 359 | continue; |
fef1bdd6 | 360 | |
c55db957 KM |
361 | task = find_lock_task_mm(p); |
362 | if (!task) { | |
6d2661ed | 363 | /* |
74ab7f1d DR |
364 | * This is a kthread or all of p's threads have already |
365 | * detached their mm's. There's no need to report | |
c55db957 | 366 | * them; they can't be oom killed anyway. |
6d2661ed | 367 | */ |
6d2661ed DR |
368 | continue; |
369 | } | |
c55db957 | 370 | |
dc6c9a35 | 371 | pr_info("[%5d] %5d %5d %8lu %8lu %7ld %7ld %8lu %5hd %s\n", |
078de5f7 EB |
372 | task->pid, from_kuid(&init_user_ns, task_uid(task)), |
373 | task->tgid, task->mm->total_vm, get_mm_rss(task->mm), | |
e1f56c89 | 374 | atomic_long_read(&task->mm->nr_ptes), |
dc6c9a35 | 375 | mm_nr_pmds(task->mm), |
de34d965 | 376 | get_mm_counter(task->mm, MM_SWAPENTS), |
a63d83f4 | 377 | task->signal->oom_score_adj, task->comm); |
c55db957 KM |
378 | task_unlock(task); |
379 | } | |
6b0c81b3 | 380 | rcu_read_unlock(); |
fef1bdd6 DR |
381 | } |
382 | ||
d31f56db | 383 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
72835c86 | 384 | struct mem_cgroup *memcg, const nodemask_t *nodemask) |
1b604d75 | 385 | { |
5e9d834a | 386 | task_lock(current); |
1b604d75 | 387 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " |
a9c58b90 | 388 | "oom_score_adj=%hd\n", |
01dc52eb | 389 | current->comm, gfp_mask, order, |
a63d83f4 | 390 | current->signal->oom_score_adj); |
1b604d75 DR |
391 | cpuset_print_task_mems_allowed(current); |
392 | task_unlock(current); | |
393 | dump_stack(); | |
58cf188e SZ |
394 | if (memcg) |
395 | mem_cgroup_print_oom_info(memcg, p); | |
396 | else | |
397 | show_mem(SHOW_MEM_FILTER_NODES); | |
1b604d75 | 398 | if (sysctl_oom_dump_tasks) |
72835c86 | 399 | dump_tasks(memcg, nodemask); |
1b604d75 DR |
400 | } |
401 | ||
5695be14 | 402 | /* |
c32b3cbe | 403 | * Number of OOM victims in flight |
5695be14 | 404 | */ |
c32b3cbe MH |
405 | static atomic_t oom_victims = ATOMIC_INIT(0); |
406 | static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait); | |
5695be14 | 407 | |
c32b3cbe | 408 | bool oom_killer_disabled __read_mostly; |
5695be14 | 409 | |
49550b60 | 410 | /** |
16e95196 | 411 | * mark_oom_victim - mark the given task as OOM victim |
49550b60 | 412 | * @tsk: task to mark |
c32b3cbe | 413 | * |
dc56401f | 414 | * Has to be called with oom_lock held and never after |
c32b3cbe | 415 | * oom has been disabled already. |
49550b60 | 416 | */ |
16e95196 | 417 | void mark_oom_victim(struct task_struct *tsk) |
49550b60 | 418 | { |
c32b3cbe MH |
419 | WARN_ON(oom_killer_disabled); |
420 | /* OOM killer might race with memcg OOM */ | |
421 | if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE)) | |
422 | return; | |
63a8ca9b MH |
423 | /* |
424 | * Make sure that the task is woken up from uninterruptible sleep | |
425 | * if it is frozen because OOM killer wouldn't be able to free | |
426 | * any memory and livelock. freezing_slow_path will tell the freezer | |
427 | * that TIF_MEMDIE tasks should be ignored. | |
428 | */ | |
429 | __thaw_task(tsk); | |
c32b3cbe | 430 | atomic_inc(&oom_victims); |
49550b60 MH |
431 | } |
432 | ||
433 | /** | |
16e95196 | 434 | * exit_oom_victim - note the exit of an OOM victim |
49550b60 | 435 | */ |
16e95196 | 436 | void exit_oom_victim(void) |
49550b60 | 437 | { |
46402778 | 438 | clear_thread_flag(TIF_MEMDIE); |
c32b3cbe | 439 | |
c38f1025 | 440 | if (!atomic_dec_return(&oom_victims)) |
c32b3cbe | 441 | wake_up_all(&oom_victims_wait); |
c32b3cbe MH |
442 | } |
443 | ||
444 | /** | |
445 | * oom_killer_disable - disable OOM killer | |
446 | * | |
447 | * Forces all page allocations to fail rather than trigger OOM killer. | |
448 | * Will block and wait until all OOM victims are killed. | |
449 | * | |
450 | * The function cannot be called when there are runnable user tasks because | |
451 | * the userspace would see unexpected allocation failures as a result. Any | |
452 | * new usage of this function should be consulted with MM people. | |
453 | * | |
454 | * Returns true if successful and false if the OOM killer cannot be | |
455 | * disabled. | |
456 | */ | |
457 | bool oom_killer_disable(void) | |
458 | { | |
459 | /* | |
460 | * Make sure to not race with an ongoing OOM killer | |
461 | * and that the current is not the victim. | |
462 | */ | |
dc56401f | 463 | mutex_lock(&oom_lock); |
c32b3cbe | 464 | if (test_thread_flag(TIF_MEMDIE)) { |
dc56401f | 465 | mutex_unlock(&oom_lock); |
c32b3cbe MH |
466 | return false; |
467 | } | |
468 | ||
469 | oom_killer_disabled = true; | |
dc56401f | 470 | mutex_unlock(&oom_lock); |
c32b3cbe MH |
471 | |
472 | wait_event(oom_victims_wait, !atomic_read(&oom_victims)); | |
473 | ||
474 | return true; | |
475 | } | |
476 | ||
477 | /** | |
478 | * oom_killer_enable - enable OOM killer | |
479 | */ | |
480 | void oom_killer_enable(void) | |
481 | { | |
c32b3cbe | 482 | oom_killer_disabled = false; |
49550b60 MH |
483 | } |
484 | ||
3b4798cb | 485 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
6b0c81b3 DR |
486 | /* |
487 | * Must be called while holding a reference to p, which will be released upon | |
488 | * returning. | |
489 | */ | |
9cbb78bb DR |
490 | void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
491 | unsigned int points, unsigned long totalpages, | |
492 | struct mem_cgroup *memcg, nodemask_t *nodemask, | |
493 | const char *message) | |
1da177e4 | 494 | { |
52d3c036 | 495 | struct task_struct *victim = p; |
5e9d834a | 496 | struct task_struct *child; |
1da4db0c | 497 | struct task_struct *t; |
647f2bdf | 498 | struct mm_struct *mm; |
52d3c036 | 499 | unsigned int victim_points = 0; |
dc3f21ea DR |
500 | static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
501 | DEFAULT_RATELIMIT_BURST); | |
1da177e4 | 502 | |
50ec3bbf NP |
503 | /* |
504 | * If the task is already exiting, don't alarm the sysadmin or kill | |
505 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
506 | */ | |
83363b91 MH |
507 | task_lock(p); |
508 | if (p->mm && task_will_free_mem(p)) { | |
16e95196 | 509 | mark_oom_victim(p); |
83363b91 | 510 | task_unlock(p); |
6b0c81b3 | 511 | put_task_struct(p); |
2a1c9b1f | 512 | return; |
50ec3bbf | 513 | } |
83363b91 | 514 | task_unlock(p); |
50ec3bbf | 515 | |
dc3f21ea | 516 | if (__ratelimit(&oom_rs)) |
8447d950 DR |
517 | dump_header(p, gfp_mask, order, memcg, nodemask); |
518 | ||
5e9d834a | 519 | task_lock(p); |
a63d83f4 | 520 | pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n", |
5e9d834a DR |
521 | message, task_pid_nr(p), p->comm, points); |
522 | task_unlock(p); | |
f3af38d3 | 523 | |
5e9d834a DR |
524 | /* |
525 | * If any of p's children has a different mm and is eligible for kill, | |
11239836 | 526 | * the one with the highest oom_badness() score is sacrificed for its |
5e9d834a DR |
527 | * parent. This attempts to lose the minimal amount of work done while |
528 | * still freeing memory. | |
529 | */ | |
6b0c81b3 | 530 | read_lock(&tasklist_lock); |
1da4db0c | 531 | for_each_thread(p, t) { |
5e9d834a | 532 | list_for_each_entry(child, &t->children, sibling) { |
a63d83f4 | 533 | unsigned int child_points; |
5e9d834a | 534 | |
edd45544 DR |
535 | if (child->mm == p->mm) |
536 | continue; | |
a63d83f4 DR |
537 | /* |
538 | * oom_badness() returns 0 if the thread is unkillable | |
539 | */ | |
72835c86 | 540 | child_points = oom_badness(child, memcg, nodemask, |
a63d83f4 | 541 | totalpages); |
5e9d834a | 542 | if (child_points > victim_points) { |
6b0c81b3 | 543 | put_task_struct(victim); |
5e9d834a DR |
544 | victim = child; |
545 | victim_points = child_points; | |
6b0c81b3 | 546 | get_task_struct(victim); |
5e9d834a | 547 | } |
dd8e8f40 | 548 | } |
1da4db0c | 549 | } |
6b0c81b3 | 550 | read_unlock(&tasklist_lock); |
dd8e8f40 | 551 | |
6b0c81b3 DR |
552 | p = find_lock_task_mm(victim); |
553 | if (!p) { | |
6b0c81b3 | 554 | put_task_struct(victim); |
647f2bdf | 555 | return; |
6b0c81b3 DR |
556 | } else if (victim != p) { |
557 | get_task_struct(p); | |
558 | put_task_struct(victim); | |
559 | victim = p; | |
560 | } | |
647f2bdf DR |
561 | |
562 | /* mm cannot safely be dereferenced after task_unlock(victim) */ | |
563 | mm = victim->mm; | |
16e95196 | 564 | mark_oom_victim(victim); |
647f2bdf DR |
565 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", |
566 | task_pid_nr(victim), victim->comm, K(victim->mm->total_vm), | |
567 | K(get_mm_counter(victim->mm, MM_ANONPAGES)), | |
568 | K(get_mm_counter(victim->mm, MM_FILEPAGES))); | |
569 | task_unlock(victim); | |
570 | ||
571 | /* | |
572 | * Kill all user processes sharing victim->mm in other thread groups, if | |
573 | * any. They don't get access to memory reserves, though, to avoid | |
574 | * depletion of all memory. This prevents mm->mmap_sem livelock when an | |
575 | * oom killed thread cannot exit because it requires the semaphore and | |
576 | * its contended by another thread trying to allocate memory itself. | |
577 | * That thread will now get access to memory reserves since it has a | |
578 | * pending fatal signal. | |
579 | */ | |
4d4048be | 580 | rcu_read_lock(); |
647f2bdf DR |
581 | for_each_process(p) |
582 | if (p->mm == mm && !same_thread_group(p, victim) && | |
583 | !(p->flags & PF_KTHREAD)) { | |
584 | if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) | |
585 | continue; | |
586 | ||
587 | task_lock(p); /* Protect ->comm from prctl() */ | |
588 | pr_err("Kill process %d (%s) sharing same memory\n", | |
589 | task_pid_nr(p), p->comm); | |
590 | task_unlock(p); | |
d2d39309 | 591 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true); |
647f2bdf | 592 | } |
6b0c81b3 | 593 | rcu_read_unlock(); |
647f2bdf | 594 | |
d2d39309 | 595 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true); |
6b0c81b3 | 596 | put_task_struct(victim); |
1da177e4 | 597 | } |
647f2bdf | 598 | #undef K |
1da177e4 | 599 | |
309ed882 DR |
600 | /* |
601 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. | |
602 | */ | |
876aafbf | 603 | void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask, |
2415b9f5 BV |
604 | int order, const nodemask_t *nodemask, |
605 | struct mem_cgroup *memcg) | |
309ed882 DR |
606 | { |
607 | if (likely(!sysctl_panic_on_oom)) | |
608 | return; | |
609 | if (sysctl_panic_on_oom != 2) { | |
610 | /* | |
611 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel | |
612 | * does not panic for cpuset, mempolicy, or memcg allocation | |
613 | * failures. | |
614 | */ | |
615 | if (constraint != CONSTRAINT_NONE) | |
616 | return; | |
617 | } | |
2415b9f5 | 618 | dump_header(NULL, gfp_mask, order, memcg, nodemask); |
309ed882 DR |
619 | panic("Out of memory: %s panic_on_oom is enabled\n", |
620 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); | |
621 | } | |
622 | ||
8bc719d3 MS |
623 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
624 | ||
625 | int register_oom_notifier(struct notifier_block *nb) | |
626 | { | |
627 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
628 | } | |
629 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
630 | ||
631 | int unregister_oom_notifier(struct notifier_block *nb) | |
632 | { | |
633 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
634 | } | |
635 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
636 | ||
1da177e4 | 637 | /** |
c32b3cbe | 638 | * __out_of_memory - kill the "best" process when we run out of memory |
1b578df0 RD |
639 | * @zonelist: zonelist pointer |
640 | * @gfp_mask: memory allocation flags | |
641 | * @order: amount of memory being requested as a power of 2 | |
6f48d0eb | 642 | * @nodemask: nodemask passed to page allocator |
08ab9b10 | 643 | * @force_kill: true if a task must be killed, even if others are exiting |
1da177e4 LT |
644 | * |
645 | * If we run out of memory, we have the choice between either | |
646 | * killing a random task (bad), letting the system crash (worse) | |
647 | * OR try to be smart about which process to kill. Note that we | |
648 | * don't have to be perfect here, we just have to be good. | |
649 | */ | |
dc56401f JW |
650 | bool out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
651 | int order, nodemask_t *nodemask, bool force_kill) | |
1da177e4 | 652 | { |
e85bfd3a | 653 | const nodemask_t *mpol_mask; |
0aad4b31 | 654 | struct task_struct *p; |
a63d83f4 | 655 | unsigned long totalpages; |
8bc719d3 | 656 | unsigned long freed = 0; |
9cbb78bb | 657 | unsigned int uninitialized_var(points); |
e3658932 | 658 | enum oom_constraint constraint = CONSTRAINT_NONE; |
b52723c5 | 659 | int killed = 0; |
8bc719d3 | 660 | |
dc56401f JW |
661 | if (oom_killer_disabled) |
662 | return false; | |
663 | ||
8bc719d3 MS |
664 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); |
665 | if (freed > 0) | |
666 | /* Got some memory back in the last second. */ | |
dc56401f | 667 | goto out; |
1da177e4 | 668 | |
7b98c2e4 | 669 | /* |
9ff4868e DR |
670 | * If current has a pending SIGKILL or is exiting, then automatically |
671 | * select it. The goal is to allow it to allocate so that it may | |
672 | * quickly exit and free its memory. | |
d7a94e7e TH |
673 | * |
674 | * But don't select if current has already released its mm and cleared | |
675 | * TIF_MEMDIE flag at exit_mm(), otherwise an OOM livelock may occur. | |
7b98c2e4 | 676 | */ |
d7a94e7e TH |
677 | if (current->mm && |
678 | (fatal_signal_pending(current) || task_will_free_mem(current))) { | |
16e95196 | 679 | mark_oom_victim(current); |
dc56401f | 680 | goto out; |
7b98c2e4 DR |
681 | } |
682 | ||
9b0f8b04 CL |
683 | /* |
684 | * Check if there were limitations on the allocation (only relevant for | |
685 | * NUMA) that may require different handling. | |
686 | */ | |
a63d83f4 DR |
687 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask, |
688 | &totalpages); | |
e85bfd3a | 689 | mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL; |
2415b9f5 | 690 | check_panic_on_oom(constraint, gfp_mask, order, mpol_mask, NULL); |
0aad4b31 | 691 | |
121d1ba0 | 692 | if (sysctl_oom_kill_allocating_task && current->mm && |
a96cfd6e | 693 | !oom_unkillable_task(current, NULL, nodemask) && |
121d1ba0 | 694 | current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) { |
6b0c81b3 | 695 | get_task_struct(current); |
2a1c9b1f DR |
696 | oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL, |
697 | nodemask, | |
698 | "Out of memory (oom_kill_allocating_task)"); | |
699 | goto out; | |
0aad4b31 DR |
700 | } |
701 | ||
9cbb78bb | 702 | p = select_bad_process(&points, totalpages, mpol_mask, force_kill); |
0aad4b31 DR |
703 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
704 | if (!p) { | |
e85bfd3a | 705 | dump_header(NULL, gfp_mask, order, NULL, mpol_mask); |
0aad4b31 DR |
706 | panic("Out of memory and no killable processes...\n"); |
707 | } | |
6b4f2b56 | 708 | if (p != (void *)-1UL) { |
2a1c9b1f DR |
709 | oom_kill_process(p, gfp_mask, order, points, totalpages, NULL, |
710 | nodemask, "Out of memory"); | |
711 | killed = 1; | |
712 | } | |
b52723c5 | 713 | out: |
1da177e4 | 714 | /* |
4f774b91 DR |
715 | * Give the killed threads a good chance of exiting before trying to |
716 | * allocate memory again. | |
1da177e4 | 717 | */ |
4f774b91 DR |
718 | if (killed) |
719 | schedule_timeout_killable(1); | |
c32b3cbe | 720 | |
dc56401f | 721 | return true; |
c32b3cbe MH |
722 | } |
723 | ||
e3658932 DR |
724 | /* |
725 | * The pagefault handler calls here because it is out of memory, so kill a | |
efacd02e DR |
726 | * memory-hogging task. If any populated zone has ZONE_OOM_LOCKED set, a |
727 | * parallel oom killing is already in progress so do nothing. | |
e3658932 DR |
728 | */ |
729 | void pagefault_out_of_memory(void) | |
730 | { | |
49426420 | 731 | if (mem_cgroup_oom_synchronize(true)) |
dc56401f | 732 | return; |
3812c8c8 | 733 | |
dc56401f JW |
734 | if (!mutex_trylock(&oom_lock)) |
735 | return; | |
c32b3cbe | 736 | |
dc56401f JW |
737 | if (!out_of_memory(NULL, 0, 0, NULL, false)) { |
738 | /* | |
739 | * There shouldn't be any user tasks runnable while the | |
740 | * OOM killer is disabled, so the current task has to | |
741 | * be a racing OOM victim for which oom_killer_disable() | |
742 | * is waiting for. | |
743 | */ | |
744 | WARN_ON(test_thread_flag(TIF_MEMDIE)); | |
e3658932 | 745 | } |
dc56401f JW |
746 | |
747 | mutex_unlock(&oom_lock); | |
e3658932 | 748 | } |