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 | */ | |
a9c58b90 | 56 | void compare_swap_oom_score_adj(short old_val, short new_val) |
43362a49 DR |
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 | */ | |
a9c58b90 | 75 | short test_set_oom_score_adj(short new_val) |
72788c38 DR |
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 | ||
a9c58b90 | 196 | adj = (long)p->signal->oom_score_adj; |
61eafb00 | 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 | |
9cbb78bb DR |
291 | enum oom_scan_t oom_scan_process_thread(struct task_struct *task, |
292 | unsigned long totalpages, const nodemask_t *nodemask, | |
293 | bool force_kill) | |
462607ec DR |
294 | { |
295 | if (task->exit_state) | |
296 | return OOM_SCAN_CONTINUE; | |
9cbb78bb | 297 | if (oom_unkillable_task(task, NULL, nodemask)) |
462607ec DR |
298 | return OOM_SCAN_CONTINUE; |
299 | ||
300 | /* | |
301 | * This task already has access to memory reserves and is being killed. | |
302 | * Don't allow any other task to have access to the reserves. | |
303 | */ | |
304 | if (test_tsk_thread_flag(task, TIF_MEMDIE)) { | |
305 | if (unlikely(frozen(task))) | |
306 | __thaw_task(task); | |
307 | if (!force_kill) | |
308 | return OOM_SCAN_ABORT; | |
309 | } | |
310 | if (!task->mm) | |
311 | return OOM_SCAN_CONTINUE; | |
312 | ||
9ff4868e | 313 | if (task->flags & PF_EXITING && !force_kill) { |
462607ec | 314 | /* |
9ff4868e DR |
315 | * If this task is not being ptraced on exit, then wait for it |
316 | * to finish before killing some other task unnecessarily. | |
462607ec | 317 | */ |
9ff4868e DR |
318 | if (!(task->group_leader->ptrace & PT_TRACE_EXIT)) |
319 | return OOM_SCAN_ABORT; | |
462607ec DR |
320 | } |
321 | return OOM_SCAN_OK; | |
322 | } | |
323 | ||
1da177e4 LT |
324 | /* |
325 | * Simple selection loop. We chose the process with the highest | |
6b0c81b3 | 326 | * number of 'points'. |
1da177e4 LT |
327 | * |
328 | * (not docbooked, we don't want this one cluttering up the manual) | |
329 | */ | |
a63d83f4 | 330 | static struct task_struct *select_bad_process(unsigned int *ppoints, |
9cbb78bb DR |
331 | unsigned long totalpages, const nodemask_t *nodemask, |
332 | bool force_kill) | |
1da177e4 | 333 | { |
3a5dda7a | 334 | struct task_struct *g, *p; |
1da177e4 | 335 | struct task_struct *chosen = NULL; |
a7f638f9 | 336 | unsigned long chosen_points = 0; |
1da177e4 | 337 | |
6b0c81b3 | 338 | rcu_read_lock(); |
3a5dda7a | 339 | do_each_thread(g, p) { |
a63d83f4 | 340 | unsigned int points; |
a49335cc | 341 | |
9cbb78bb | 342 | switch (oom_scan_process_thread(p, totalpages, nodemask, |
462607ec DR |
343 | force_kill)) { |
344 | case OOM_SCAN_SELECT: | |
345 | chosen = p; | |
346 | chosen_points = ULONG_MAX; | |
347 | /* fall through */ | |
348 | case OOM_SCAN_CONTINUE: | |
c027a474 | 349 | continue; |
462607ec | 350 | case OOM_SCAN_ABORT: |
6b0c81b3 | 351 | rcu_read_unlock(); |
462607ec DR |
352 | return ERR_PTR(-1UL); |
353 | case OOM_SCAN_OK: | |
354 | break; | |
355 | }; | |
9cbb78bb | 356 | points = oom_badness(p, NULL, nodemask, totalpages); |
a7f638f9 | 357 | if (points > chosen_points) { |
a49335cc | 358 | chosen = p; |
a7f638f9 | 359 | chosen_points = points; |
1da177e4 | 360 | } |
3a5dda7a | 361 | } while_each_thread(g, p); |
6b0c81b3 DR |
362 | if (chosen) |
363 | get_task_struct(chosen); | |
364 | rcu_read_unlock(); | |
972c4ea5 | 365 | |
a7f638f9 | 366 | *ppoints = chosen_points * 1000 / totalpages; |
1da177e4 LT |
367 | return chosen; |
368 | } | |
369 | ||
fef1bdd6 | 370 | /** |
1b578df0 | 371 | * dump_tasks - dump current memory state of all system tasks |
dad7557e | 372 | * @memcg: current's memory controller, if constrained |
e85bfd3a | 373 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
1b578df0 | 374 | * |
e85bfd3a DR |
375 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
376 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes | |
377 | * are not shown. | |
de34d965 DR |
378 | * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes, |
379 | * swapents, oom_score_adj value, and name. | |
fef1bdd6 | 380 | */ |
72835c86 | 381 | static void dump_tasks(const struct mem_cgroup *memcg, const nodemask_t *nodemask) |
fef1bdd6 | 382 | { |
c55db957 KM |
383 | struct task_struct *p; |
384 | struct task_struct *task; | |
fef1bdd6 | 385 | |
de34d965 | 386 | pr_info("[ pid ] uid tgid total_vm rss nr_ptes swapents oom_score_adj name\n"); |
6b0c81b3 | 387 | rcu_read_lock(); |
c55db957 | 388 | for_each_process(p) { |
72835c86 | 389 | if (oom_unkillable_task(p, memcg, nodemask)) |
b4416d2b | 390 | continue; |
fef1bdd6 | 391 | |
c55db957 KM |
392 | task = find_lock_task_mm(p); |
393 | if (!task) { | |
6d2661ed | 394 | /* |
74ab7f1d DR |
395 | * This is a kthread or all of p's threads have already |
396 | * detached their mm's. There's no need to report | |
c55db957 | 397 | * them; they can't be oom killed anyway. |
6d2661ed | 398 | */ |
6d2661ed DR |
399 | continue; |
400 | } | |
c55db957 | 401 | |
a9c58b90 | 402 | pr_info("[%5d] %5d %5d %8lu %8lu %7lu %8lu %5hd %s\n", |
078de5f7 EB |
403 | task->pid, from_kuid(&init_user_ns, task_uid(task)), |
404 | task->tgid, task->mm->total_vm, get_mm_rss(task->mm), | |
de34d965 DR |
405 | task->mm->nr_ptes, |
406 | get_mm_counter(task->mm, MM_SWAPENTS), | |
a63d83f4 | 407 | task->signal->oom_score_adj, task->comm); |
c55db957 KM |
408 | task_unlock(task); |
409 | } | |
6b0c81b3 | 410 | rcu_read_unlock(); |
fef1bdd6 DR |
411 | } |
412 | ||
d31f56db | 413 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
72835c86 | 414 | struct mem_cgroup *memcg, 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, " |
a9c58b90 | 418 | "oom_score_adj=%hd\n", |
01dc52eb | 419 | current->comm, gfp_mask, order, |
a63d83f4 | 420 | current->signal->oom_score_adj); |
1b604d75 DR |
421 | cpuset_print_task_mems_allowed(current); |
422 | task_unlock(current); | |
423 | dump_stack(); | |
72835c86 | 424 | mem_cgroup_print_oom_info(memcg, p); |
b2b755b5 | 425 | show_mem(SHOW_MEM_FILTER_NODES); |
1b604d75 | 426 | if (sysctl_oom_dump_tasks) |
72835c86 | 427 | dump_tasks(memcg, nodemask); |
1b604d75 DR |
428 | } |
429 | ||
3b4798cb | 430 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
6b0c81b3 DR |
431 | /* |
432 | * Must be called while holding a reference to p, which will be released upon | |
433 | * returning. | |
434 | */ | |
9cbb78bb DR |
435 | void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
436 | unsigned int points, unsigned long totalpages, | |
437 | struct mem_cgroup *memcg, nodemask_t *nodemask, | |
438 | const char *message) | |
1da177e4 | 439 | { |
52d3c036 | 440 | struct task_struct *victim = p; |
5e9d834a | 441 | struct task_struct *child; |
52d3c036 | 442 | struct task_struct *t = p; |
647f2bdf | 443 | struct mm_struct *mm; |
52d3c036 | 444 | unsigned int victim_points = 0; |
dc3f21ea DR |
445 | static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
446 | DEFAULT_RATELIMIT_BURST); | |
1da177e4 | 447 | |
50ec3bbf NP |
448 | /* |
449 | * If the task is already exiting, don't alarm the sysadmin or kill | |
450 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
451 | */ | |
0753ba01 | 452 | if (p->flags & PF_EXITING) { |
4358997a | 453 | set_tsk_thread_flag(p, TIF_MEMDIE); |
6b0c81b3 | 454 | put_task_struct(p); |
2a1c9b1f | 455 | return; |
50ec3bbf NP |
456 | } |
457 | ||
dc3f21ea | 458 | if (__ratelimit(&oom_rs)) |
8447d950 DR |
459 | dump_header(p, gfp_mask, order, memcg, nodemask); |
460 | ||
5e9d834a | 461 | task_lock(p); |
a63d83f4 | 462 | pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n", |
5e9d834a DR |
463 | message, task_pid_nr(p), p->comm, points); |
464 | task_unlock(p); | |
f3af38d3 | 465 | |
5e9d834a DR |
466 | /* |
467 | * If any of p's children has a different mm and is eligible for kill, | |
11239836 | 468 | * the one with the highest oom_badness() score is sacrificed for its |
5e9d834a DR |
469 | * parent. This attempts to lose the minimal amount of work done while |
470 | * still freeing memory. | |
471 | */ | |
6b0c81b3 | 472 | read_lock(&tasklist_lock); |
dd8e8f40 | 473 | do { |
5e9d834a | 474 | list_for_each_entry(child, &t->children, sibling) { |
a63d83f4 | 475 | unsigned int child_points; |
5e9d834a | 476 | |
edd45544 DR |
477 | if (child->mm == p->mm) |
478 | continue; | |
a63d83f4 DR |
479 | /* |
480 | * oom_badness() returns 0 if the thread is unkillable | |
481 | */ | |
72835c86 | 482 | child_points = oom_badness(child, memcg, nodemask, |
a63d83f4 | 483 | totalpages); |
5e9d834a | 484 | if (child_points > victim_points) { |
6b0c81b3 | 485 | put_task_struct(victim); |
5e9d834a DR |
486 | victim = child; |
487 | victim_points = child_points; | |
6b0c81b3 | 488 | get_task_struct(victim); |
5e9d834a | 489 | } |
dd8e8f40 ON |
490 | } |
491 | } while_each_thread(p, t); | |
6b0c81b3 | 492 | read_unlock(&tasklist_lock); |
dd8e8f40 | 493 | |
6b0c81b3 DR |
494 | rcu_read_lock(); |
495 | p = find_lock_task_mm(victim); | |
496 | if (!p) { | |
497 | rcu_read_unlock(); | |
498 | put_task_struct(victim); | |
647f2bdf | 499 | return; |
6b0c81b3 DR |
500 | } else if (victim != p) { |
501 | get_task_struct(p); | |
502 | put_task_struct(victim); | |
503 | victim = p; | |
504 | } | |
647f2bdf DR |
505 | |
506 | /* mm cannot safely be dereferenced after task_unlock(victim) */ | |
507 | mm = victim->mm; | |
508 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", | |
509 | task_pid_nr(victim), victim->comm, K(victim->mm->total_vm), | |
510 | K(get_mm_counter(victim->mm, MM_ANONPAGES)), | |
511 | K(get_mm_counter(victim->mm, MM_FILEPAGES))); | |
512 | task_unlock(victim); | |
513 | ||
514 | /* | |
515 | * Kill all user processes sharing victim->mm in other thread groups, if | |
516 | * any. They don't get access to memory reserves, though, to avoid | |
517 | * depletion of all memory. This prevents mm->mmap_sem livelock when an | |
518 | * oom killed thread cannot exit because it requires the semaphore and | |
519 | * its contended by another thread trying to allocate memory itself. | |
520 | * That thread will now get access to memory reserves since it has a | |
521 | * pending fatal signal. | |
522 | */ | |
523 | for_each_process(p) | |
524 | if (p->mm == mm && !same_thread_group(p, victim) && | |
525 | !(p->flags & PF_KTHREAD)) { | |
526 | if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) | |
527 | continue; | |
528 | ||
529 | task_lock(p); /* Protect ->comm from prctl() */ | |
530 | pr_err("Kill process %d (%s) sharing same memory\n", | |
531 | task_pid_nr(p), p->comm); | |
532 | task_unlock(p); | |
d2d39309 | 533 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true); |
647f2bdf | 534 | } |
6b0c81b3 | 535 | rcu_read_unlock(); |
647f2bdf DR |
536 | |
537 | set_tsk_thread_flag(victim, TIF_MEMDIE); | |
d2d39309 | 538 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true); |
6b0c81b3 | 539 | put_task_struct(victim); |
1da177e4 | 540 | } |
647f2bdf | 541 | #undef K |
1da177e4 | 542 | |
309ed882 DR |
543 | /* |
544 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. | |
545 | */ | |
876aafbf DR |
546 | void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask, |
547 | int order, const nodemask_t *nodemask) | |
309ed882 DR |
548 | { |
549 | if (likely(!sysctl_panic_on_oom)) | |
550 | return; | |
551 | if (sysctl_panic_on_oom != 2) { | |
552 | /* | |
553 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel | |
554 | * does not panic for cpuset, mempolicy, or memcg allocation | |
555 | * failures. | |
556 | */ | |
557 | if (constraint != CONSTRAINT_NONE) | |
558 | return; | |
559 | } | |
e85bfd3a | 560 | dump_header(NULL, gfp_mask, order, NULL, nodemask); |
309ed882 DR |
561 | panic("Out of memory: %s panic_on_oom is enabled\n", |
562 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); | |
563 | } | |
564 | ||
8bc719d3 MS |
565 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
566 | ||
567 | int register_oom_notifier(struct notifier_block *nb) | |
568 | { | |
569 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
570 | } | |
571 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
572 | ||
573 | int unregister_oom_notifier(struct notifier_block *nb) | |
574 | { | |
575 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
576 | } | |
577 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
578 | ||
098d7f12 DR |
579 | /* |
580 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero | |
581 | * if a parallel OOM killing is already taking place that includes a zone in | |
582 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. | |
583 | */ | |
ff321fea | 584 | int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 585 | { |
dd1a239f MG |
586 | struct zoneref *z; |
587 | struct zone *zone; | |
098d7f12 DR |
588 | int ret = 1; |
589 | ||
c7d4caeb | 590 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
591 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
592 | if (zone_is_oom_locked(zone)) { | |
098d7f12 DR |
593 | ret = 0; |
594 | goto out; | |
595 | } | |
dd1a239f MG |
596 | } |
597 | ||
598 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
599 | /* | |
c7d4caeb | 600 | * Lock each zone in the zonelist under zone_scan_lock so a |
ff321fea | 601 | * parallel invocation of try_set_zonelist_oom() doesn't succeed |
dd1a239f MG |
602 | * when it shouldn't. |
603 | */ | |
604 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
605 | } | |
098d7f12 | 606 | |
098d7f12 | 607 | out: |
c7d4caeb | 608 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
609 | return ret; |
610 | } | |
611 | ||
612 | /* | |
613 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed | |
614 | * allocation attempts with zonelists containing them may now recall the OOM | |
615 | * killer, if necessary. | |
616 | */ | |
dd1a239f | 617 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 618 | { |
dd1a239f MG |
619 | struct zoneref *z; |
620 | struct zone *zone; | |
098d7f12 | 621 | |
c7d4caeb | 622 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
623 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
624 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
625 | } | |
c7d4caeb | 626 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
627 | } |
628 | ||
e3658932 DR |
629 | /* |
630 | * Try to acquire the oom killer lock for all system zones. Returns zero if a | |
631 | * parallel oom killing is taking place, otherwise locks all zones and returns | |
632 | * non-zero. | |
633 | */ | |
634 | static int try_set_system_oom(void) | |
635 | { | |
636 | struct zone *zone; | |
637 | int ret = 1; | |
638 | ||
639 | spin_lock(&zone_scan_lock); | |
640 | for_each_populated_zone(zone) | |
641 | if (zone_is_oom_locked(zone)) { | |
642 | ret = 0; | |
643 | goto out; | |
644 | } | |
645 | for_each_populated_zone(zone) | |
646 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
647 | out: | |
648 | spin_unlock(&zone_scan_lock); | |
649 | return ret; | |
650 | } | |
651 | ||
652 | /* | |
653 | * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation | |
654 | * attempts or page faults may now recall the oom killer, if necessary. | |
655 | */ | |
656 | static void clear_system_oom(void) | |
657 | { | |
658 | struct zone *zone; | |
659 | ||
660 | spin_lock(&zone_scan_lock); | |
661 | for_each_populated_zone(zone) | |
662 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
663 | spin_unlock(&zone_scan_lock); | |
664 | } | |
665 | ||
1da177e4 | 666 | /** |
6937a25c | 667 | * out_of_memory - kill the "best" process when we run out of memory |
1b578df0 RD |
668 | * @zonelist: zonelist pointer |
669 | * @gfp_mask: memory allocation flags | |
670 | * @order: amount of memory being requested as a power of 2 | |
6f48d0eb | 671 | * @nodemask: nodemask passed to page allocator |
08ab9b10 | 672 | * @force_kill: true if a task must be killed, even if others are exiting |
1da177e4 LT |
673 | * |
674 | * If we run out of memory, we have the choice between either | |
675 | * killing a random task (bad), letting the system crash (worse) | |
676 | * OR try to be smart about which process to kill. Note that we | |
677 | * don't have to be perfect here, we just have to be good. | |
678 | */ | |
4365a567 | 679 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
08ab9b10 | 680 | int order, nodemask_t *nodemask, bool force_kill) |
1da177e4 | 681 | { |
e85bfd3a | 682 | const nodemask_t *mpol_mask; |
0aad4b31 | 683 | struct task_struct *p; |
a63d83f4 | 684 | unsigned long totalpages; |
8bc719d3 | 685 | unsigned long freed = 0; |
9cbb78bb | 686 | unsigned int uninitialized_var(points); |
e3658932 | 687 | enum oom_constraint constraint = CONSTRAINT_NONE; |
b52723c5 | 688 | int killed = 0; |
8bc719d3 MS |
689 | |
690 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
691 | if (freed > 0) | |
692 | /* Got some memory back in the last second. */ | |
693 | return; | |
1da177e4 | 694 | |
7b98c2e4 | 695 | /* |
9ff4868e DR |
696 | * If current has a pending SIGKILL or is exiting, then automatically |
697 | * select it. The goal is to allow it to allocate so that it may | |
698 | * quickly exit and free its memory. | |
7b98c2e4 | 699 | */ |
9ff4868e | 700 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
7b98c2e4 DR |
701 | set_thread_flag(TIF_MEMDIE); |
702 | return; | |
703 | } | |
704 | ||
9b0f8b04 CL |
705 | /* |
706 | * Check if there were limitations on the allocation (only relevant for | |
707 | * NUMA) that may require different handling. | |
708 | */ | |
a63d83f4 DR |
709 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask, |
710 | &totalpages); | |
e85bfd3a DR |
711 | mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL; |
712 | check_panic_on_oom(constraint, gfp_mask, order, mpol_mask); | |
0aad4b31 | 713 | |
121d1ba0 | 714 | if (sysctl_oom_kill_allocating_task && current->mm && |
a96cfd6e | 715 | !oom_unkillable_task(current, NULL, nodemask) && |
121d1ba0 | 716 | current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) { |
6b0c81b3 | 717 | get_task_struct(current); |
2a1c9b1f DR |
718 | oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL, |
719 | nodemask, | |
720 | "Out of memory (oom_kill_allocating_task)"); | |
721 | goto out; | |
0aad4b31 DR |
722 | } |
723 | ||
9cbb78bb | 724 | p = select_bad_process(&points, totalpages, mpol_mask, force_kill); |
0aad4b31 DR |
725 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
726 | if (!p) { | |
e85bfd3a | 727 | dump_header(NULL, gfp_mask, order, NULL, mpol_mask); |
0aad4b31 DR |
728 | panic("Out of memory and no killable processes...\n"); |
729 | } | |
2a1c9b1f DR |
730 | if (PTR_ERR(p) != -1UL) { |
731 | oom_kill_process(p, gfp_mask, order, points, totalpages, NULL, | |
732 | nodemask, "Out of memory"); | |
733 | killed = 1; | |
734 | } | |
b52723c5 | 735 | out: |
1da177e4 | 736 | /* |
4f774b91 DR |
737 | * Give the killed threads a good chance of exiting before trying to |
738 | * allocate memory again. | |
1da177e4 | 739 | */ |
4f774b91 DR |
740 | if (killed) |
741 | schedule_timeout_killable(1); | |
1da177e4 | 742 | } |
e3658932 DR |
743 | |
744 | /* | |
745 | * The pagefault handler calls here because it is out of memory, so kill a | |
746 | * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel | |
747 | * oom killing is already in progress so do nothing. If a task is found with | |
748 | * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit. | |
749 | */ | |
750 | void pagefault_out_of_memory(void) | |
751 | { | |
752 | if (try_set_system_oom()) { | |
08ab9b10 | 753 | out_of_memory(NULL, 0, 0, NULL, false); |
e3658932 DR |
754 | clear_system_oom(); |
755 | } | |
4f774b91 | 756 | schedule_timeout_killable(1); |
e3658932 | 757 | } |