]>
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 | 24 | #include <linux/sched.h> |
6e84f315 | 25 | #include <linux/sched/mm.h> |
f7ccbae4 | 26 | #include <linux/sched/coredump.h> |
29930025 | 27 | #include <linux/sched/task.h> |
1da177e4 LT |
28 | #include <linux/swap.h> |
29 | #include <linux/timex.h> | |
30 | #include <linux/jiffies.h> | |
ef08e3b4 | 31 | #include <linux/cpuset.h> |
b95f1b31 | 32 | #include <linux/export.h> |
8bc719d3 | 33 | #include <linux/notifier.h> |
c7ba5c9e | 34 | #include <linux/memcontrol.h> |
6f48d0eb | 35 | #include <linux/mempolicy.h> |
5cd9c58f | 36 | #include <linux/security.h> |
edd45544 | 37 | #include <linux/ptrace.h> |
f660daac | 38 | #include <linux/freezer.h> |
43d2b113 | 39 | #include <linux/ftrace.h> |
dc3f21ea | 40 | #include <linux/ratelimit.h> |
aac45363 MH |
41 | #include <linux/kthread.h> |
42 | #include <linux/init.h> | |
4d4bbd85 | 43 | #include <linux/mmu_notifier.h> |
aac45363 MH |
44 | |
45 | #include <asm/tlb.h> | |
46 | #include "internal.h" | |
852d8be0 | 47 | #include "slab.h" |
43d2b113 KH |
48 | |
49 | #define CREATE_TRACE_POINTS | |
50 | #include <trace/events/oom.h> | |
1da177e4 | 51 | |
fadd8fbd | 52 | int sysctl_panic_on_oom; |
fe071d7e | 53 | int sysctl_oom_kill_allocating_task; |
ad915c43 | 54 | int sysctl_oom_dump_tasks = 1; |
dc56401f | 55 | |
a195d3f5 MH |
56 | /* |
57 | * Serializes oom killer invocations (out_of_memory()) from all contexts to | |
58 | * prevent from over eager oom killing (e.g. when the oom killer is invoked | |
59 | * from different domains). | |
60 | * | |
61 | * oom_killer_disable() relies on this lock to stabilize oom_killer_disabled | |
62 | * and mark_oom_victim | |
63 | */ | |
dc56401f | 64 | DEFINE_MUTEX(oom_lock); |
1da177e4 | 65 | |
6f48d0eb DR |
66 | #ifdef CONFIG_NUMA |
67 | /** | |
68 | * has_intersects_mems_allowed() - check task eligiblity for kill | |
ad962441 | 69 | * @start: task struct of which task to consider |
6f48d0eb DR |
70 | * @mask: nodemask passed to page allocator for mempolicy ooms |
71 | * | |
72 | * Task eligibility is determined by whether or not a candidate task, @tsk, | |
73 | * shares the same mempolicy nodes as current if it is bound by such a policy | |
74 | * and whether or not it has the same set of allowed cpuset nodes. | |
495789a5 | 75 | */ |
ad962441 | 76 | static bool has_intersects_mems_allowed(struct task_struct *start, |
6f48d0eb | 77 | const nodemask_t *mask) |
495789a5 | 78 | { |
ad962441 ON |
79 | struct task_struct *tsk; |
80 | bool ret = false; | |
495789a5 | 81 | |
ad962441 | 82 | rcu_read_lock(); |
1da4db0c | 83 | for_each_thread(start, tsk) { |
6f48d0eb DR |
84 | if (mask) { |
85 | /* | |
86 | * If this is a mempolicy constrained oom, tsk's | |
87 | * cpuset is irrelevant. Only return true if its | |
88 | * mempolicy intersects current, otherwise it may be | |
89 | * needlessly killed. | |
90 | */ | |
ad962441 | 91 | ret = mempolicy_nodemask_intersects(tsk, mask); |
6f48d0eb DR |
92 | } else { |
93 | /* | |
94 | * This is not a mempolicy constrained oom, so only | |
95 | * check the mems of tsk's cpuset. | |
96 | */ | |
ad962441 | 97 | ret = cpuset_mems_allowed_intersects(current, tsk); |
6f48d0eb | 98 | } |
ad962441 ON |
99 | if (ret) |
100 | break; | |
1da4db0c | 101 | } |
ad962441 | 102 | rcu_read_unlock(); |
df1090a8 | 103 | |
ad962441 | 104 | return ret; |
6f48d0eb DR |
105 | } |
106 | #else | |
107 | static bool has_intersects_mems_allowed(struct task_struct *tsk, | |
108 | const nodemask_t *mask) | |
109 | { | |
110 | return true; | |
495789a5 | 111 | } |
6f48d0eb | 112 | #endif /* CONFIG_NUMA */ |
495789a5 | 113 | |
6f48d0eb DR |
114 | /* |
115 | * The process p may have detached its own ->mm while exiting or through | |
116 | * use_mm(), but one or more of its subthreads may still have a valid | |
117 | * pointer. Return p, or any of its subthreads with a valid ->mm, with | |
118 | * task_lock() held. | |
119 | */ | |
158e0a2d | 120 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
dd8e8f40 | 121 | { |
1da4db0c | 122 | struct task_struct *t; |
dd8e8f40 | 123 | |
4d4048be ON |
124 | rcu_read_lock(); |
125 | ||
1da4db0c | 126 | for_each_thread(p, t) { |
dd8e8f40 ON |
127 | task_lock(t); |
128 | if (likely(t->mm)) | |
4d4048be | 129 | goto found; |
dd8e8f40 | 130 | task_unlock(t); |
1da4db0c | 131 | } |
4d4048be ON |
132 | t = NULL; |
133 | found: | |
134 | rcu_read_unlock(); | |
dd8e8f40 | 135 | |
4d4048be | 136 | return t; |
dd8e8f40 ON |
137 | } |
138 | ||
db2a0dd7 YB |
139 | /* |
140 | * order == -1 means the oom kill is required by sysrq, otherwise only | |
141 | * for display purposes. | |
142 | */ | |
143 | static inline bool is_sysrq_oom(struct oom_control *oc) | |
144 | { | |
145 | return oc->order == -1; | |
146 | } | |
147 | ||
7c5f64f8 VD |
148 | static inline bool is_memcg_oom(struct oom_control *oc) |
149 | { | |
150 | return oc->memcg != NULL; | |
151 | } | |
152 | ||
ab290adb | 153 | /* return true if the task is not adequate as candidate victim task. */ |
e85bfd3a | 154 | static bool oom_unkillable_task(struct task_struct *p, |
2314b42d | 155 | struct mem_cgroup *memcg, const nodemask_t *nodemask) |
ab290adb KM |
156 | { |
157 | if (is_global_init(p)) | |
158 | return true; | |
159 | if (p->flags & PF_KTHREAD) | |
160 | return true; | |
161 | ||
162 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ | |
72835c86 | 163 | if (memcg && !task_in_mem_cgroup(p, memcg)) |
ab290adb KM |
164 | return true; |
165 | ||
166 | /* p may not have freeable memory in nodemask */ | |
167 | if (!has_intersects_mems_allowed(p, nodemask)) | |
168 | return true; | |
169 | ||
170 | return false; | |
171 | } | |
172 | ||
852d8be0 YS |
173 | /* |
174 | * Print out unreclaimble slabs info when unreclaimable slabs amount is greater | |
175 | * than all user memory (LRU pages) | |
176 | */ | |
177 | static bool is_dump_unreclaim_slabs(void) | |
178 | { | |
179 | unsigned long nr_lru; | |
180 | ||
181 | nr_lru = global_node_page_state(NR_ACTIVE_ANON) + | |
182 | global_node_page_state(NR_INACTIVE_ANON) + | |
183 | global_node_page_state(NR_ACTIVE_FILE) + | |
184 | global_node_page_state(NR_INACTIVE_FILE) + | |
185 | global_node_page_state(NR_ISOLATED_ANON) + | |
186 | global_node_page_state(NR_ISOLATED_FILE) + | |
187 | global_node_page_state(NR_UNEVICTABLE); | |
188 | ||
189 | return (global_node_page_state(NR_SLAB_UNRECLAIMABLE) > nr_lru); | |
190 | } | |
191 | ||
1da177e4 | 192 | /** |
a63d83f4 | 193 | * oom_badness - heuristic function to determine which candidate task to kill |
1da177e4 | 194 | * @p: task struct of which task we should calculate |
a63d83f4 | 195 | * @totalpages: total present RAM allowed for page allocation |
e8b098fc MR |
196 | * @memcg: task's memory controller, if constrained |
197 | * @nodemask: nodemask passed to page allocator for mempolicy ooms | |
1da177e4 | 198 | * |
a63d83f4 DR |
199 | * The heuristic for determining which task to kill is made to be as simple and |
200 | * predictable as possible. The goal is to return the highest value for the | |
201 | * task consuming the most memory to avoid subsequent oom failures. | |
1da177e4 | 202 | */ |
a7f638f9 DR |
203 | unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg, |
204 | const nodemask_t *nodemask, unsigned long totalpages) | |
1da177e4 | 205 | { |
1e11ad8d | 206 | long points; |
61eafb00 | 207 | long adj; |
28b83c51 | 208 | |
72835c86 | 209 | if (oom_unkillable_task(p, memcg, nodemask)) |
26ebc984 | 210 | return 0; |
1da177e4 | 211 | |
dd8e8f40 ON |
212 | p = find_lock_task_mm(p); |
213 | if (!p) | |
1da177e4 LT |
214 | return 0; |
215 | ||
bb8a4b7f MH |
216 | /* |
217 | * Do not even consider tasks which are explicitly marked oom | |
b18dc5f2 MH |
218 | * unkillable or have been already oom reaped or the are in |
219 | * the middle of vfork | |
bb8a4b7f | 220 | */ |
a9c58b90 | 221 | adj = (long)p->signal->oom_score_adj; |
bb8a4b7f | 222 | if (adj == OOM_SCORE_ADJ_MIN || |
862e3073 | 223 | test_bit(MMF_OOM_SKIP, &p->mm->flags) || |
b18dc5f2 | 224 | in_vfork(p)) { |
5aecc85a MH |
225 | task_unlock(p); |
226 | return 0; | |
227 | } | |
228 | ||
1da177e4 | 229 | /* |
a63d83f4 | 230 | * The baseline for the badness score is the proportion of RAM that each |
f755a042 | 231 | * task's rss, pagetable and swap space use. |
1da177e4 | 232 | */ |
dc6c9a35 | 233 | points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) + |
af5b0f6a | 234 | mm_pgtables_bytes(p->mm) / PAGE_SIZE; |
a63d83f4 | 235 | task_unlock(p); |
1da177e4 | 236 | |
61eafb00 DR |
237 | /* Normalize to oom_score_adj units */ |
238 | adj *= totalpages / 1000; | |
239 | points += adj; | |
1da177e4 | 240 | |
f19e8aa1 | 241 | /* |
a7f638f9 DR |
242 | * Never return 0 for an eligible task regardless of the root bonus and |
243 | * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here). | |
f19e8aa1 | 244 | */ |
1e11ad8d | 245 | return points > 0 ? points : 1; |
1da177e4 LT |
246 | } |
247 | ||
7c5f64f8 VD |
248 | enum oom_constraint { |
249 | CONSTRAINT_NONE, | |
250 | CONSTRAINT_CPUSET, | |
251 | CONSTRAINT_MEMORY_POLICY, | |
252 | CONSTRAINT_MEMCG, | |
253 | }; | |
254 | ||
9b0f8b04 CL |
255 | /* |
256 | * Determine the type of allocation constraint. | |
257 | */ | |
7c5f64f8 | 258 | static enum oom_constraint constrained_alloc(struct oom_control *oc) |
4365a567 | 259 | { |
54a6eb5c | 260 | struct zone *zone; |
dd1a239f | 261 | struct zoneref *z; |
6e0fc46d | 262 | enum zone_type high_zoneidx = gfp_zone(oc->gfp_mask); |
a63d83f4 DR |
263 | bool cpuset_limited = false; |
264 | int nid; | |
9b0f8b04 | 265 | |
7c5f64f8 | 266 | if (is_memcg_oom(oc)) { |
bbec2e15 | 267 | oc->totalpages = mem_cgroup_get_max(oc->memcg) ?: 1; |
7c5f64f8 VD |
268 | return CONSTRAINT_MEMCG; |
269 | } | |
270 | ||
a63d83f4 | 271 | /* Default to all available memory */ |
7c5f64f8 VD |
272 | oc->totalpages = totalram_pages + total_swap_pages; |
273 | ||
274 | if (!IS_ENABLED(CONFIG_NUMA)) | |
275 | return CONSTRAINT_NONE; | |
a63d83f4 | 276 | |
6e0fc46d | 277 | if (!oc->zonelist) |
a63d83f4 | 278 | return CONSTRAINT_NONE; |
4365a567 KH |
279 | /* |
280 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid | |
281 | * to kill current.We have to random task kill in this case. | |
282 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. | |
283 | */ | |
6e0fc46d | 284 | if (oc->gfp_mask & __GFP_THISNODE) |
4365a567 | 285 | return CONSTRAINT_NONE; |
9b0f8b04 | 286 | |
4365a567 | 287 | /* |
a63d83f4 DR |
288 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
289 | * the page allocator means a mempolicy is in effect. Cpuset policy | |
290 | * is enforced in get_page_from_freelist(). | |
4365a567 | 291 | */ |
6e0fc46d DR |
292 | if (oc->nodemask && |
293 | !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) { | |
7c5f64f8 | 294 | oc->totalpages = total_swap_pages; |
6e0fc46d | 295 | for_each_node_mask(nid, *oc->nodemask) |
7c5f64f8 | 296 | oc->totalpages += node_spanned_pages(nid); |
9b0f8b04 | 297 | return CONSTRAINT_MEMORY_POLICY; |
a63d83f4 | 298 | } |
4365a567 KH |
299 | |
300 | /* Check this allocation failure is caused by cpuset's wall function */ | |
6e0fc46d DR |
301 | for_each_zone_zonelist_nodemask(zone, z, oc->zonelist, |
302 | high_zoneidx, oc->nodemask) | |
303 | if (!cpuset_zone_allowed(zone, oc->gfp_mask)) | |
a63d83f4 | 304 | cpuset_limited = true; |
9b0f8b04 | 305 | |
a63d83f4 | 306 | if (cpuset_limited) { |
7c5f64f8 | 307 | oc->totalpages = total_swap_pages; |
a63d83f4 | 308 | for_each_node_mask(nid, cpuset_current_mems_allowed) |
7c5f64f8 | 309 | oc->totalpages += node_spanned_pages(nid); |
a63d83f4 DR |
310 | return CONSTRAINT_CPUSET; |
311 | } | |
9b0f8b04 CL |
312 | return CONSTRAINT_NONE; |
313 | } | |
314 | ||
7c5f64f8 | 315 | static int oom_evaluate_task(struct task_struct *task, void *arg) |
462607ec | 316 | { |
7c5f64f8 VD |
317 | struct oom_control *oc = arg; |
318 | unsigned long points; | |
319 | ||
6e0fc46d | 320 | if (oom_unkillable_task(task, NULL, oc->nodemask)) |
7c5f64f8 | 321 | goto next; |
462607ec DR |
322 | |
323 | /* | |
324 | * This task already has access to memory reserves and is being killed. | |
a373966d | 325 | * Don't allow any other task to have access to the reserves unless |
862e3073 | 326 | * the task has MMF_OOM_SKIP because chances that it would release |
a373966d | 327 | * any memory is quite low. |
462607ec | 328 | */ |
862e3073 MH |
329 | if (!is_sysrq_oom(oc) && tsk_is_oom_victim(task)) { |
330 | if (test_bit(MMF_OOM_SKIP, &task->signal->oom_mm->flags)) | |
7c5f64f8 VD |
331 | goto next; |
332 | goto abort; | |
a373966d | 333 | } |
462607ec | 334 | |
e1e12d2f DR |
335 | /* |
336 | * If task is allocating a lot of memory and has been marked to be | |
337 | * killed first if it triggers an oom, then select it. | |
338 | */ | |
7c5f64f8 VD |
339 | if (oom_task_origin(task)) { |
340 | points = ULONG_MAX; | |
341 | goto select; | |
342 | } | |
e1e12d2f | 343 | |
7c5f64f8 VD |
344 | points = oom_badness(task, NULL, oc->nodemask, oc->totalpages); |
345 | if (!points || points < oc->chosen_points) | |
346 | goto next; | |
347 | ||
348 | /* Prefer thread group leaders for display purposes */ | |
349 | if (points == oc->chosen_points && thread_group_leader(oc->chosen)) | |
350 | goto next; | |
351 | select: | |
352 | if (oc->chosen) | |
353 | put_task_struct(oc->chosen); | |
354 | get_task_struct(task); | |
355 | oc->chosen = task; | |
356 | oc->chosen_points = points; | |
357 | next: | |
358 | return 0; | |
359 | abort: | |
360 | if (oc->chosen) | |
361 | put_task_struct(oc->chosen); | |
362 | oc->chosen = (void *)-1UL; | |
363 | return 1; | |
462607ec DR |
364 | } |
365 | ||
1da177e4 | 366 | /* |
7c5f64f8 VD |
367 | * Simple selection loop. We choose the process with the highest number of |
368 | * 'points'. In case scan was aborted, oc->chosen is set to -1. | |
1da177e4 | 369 | */ |
7c5f64f8 | 370 | static void select_bad_process(struct oom_control *oc) |
1da177e4 | 371 | { |
7c5f64f8 VD |
372 | if (is_memcg_oom(oc)) |
373 | mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc); | |
374 | else { | |
375 | struct task_struct *p; | |
d49ad935 | 376 | |
7c5f64f8 VD |
377 | rcu_read_lock(); |
378 | for_each_process(p) | |
379 | if (oom_evaluate_task(p, oc)) | |
380 | break; | |
381 | rcu_read_unlock(); | |
1da4db0c | 382 | } |
972c4ea5 | 383 | |
7c5f64f8 | 384 | oc->chosen_points = oc->chosen_points * 1000 / oc->totalpages; |
1da177e4 LT |
385 | } |
386 | ||
fef1bdd6 | 387 | /** |
1b578df0 | 388 | * dump_tasks - dump current memory state of all system tasks |
dad7557e | 389 | * @memcg: current's memory controller, if constrained |
e85bfd3a | 390 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
1b578df0 | 391 | * |
e85bfd3a DR |
392 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
393 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes | |
394 | * are not shown. | |
af5b0f6a KS |
395 | * State information includes task's pid, uid, tgid, vm size, rss, |
396 | * pgtables_bytes, swapents, oom_score_adj value, and name. | |
fef1bdd6 | 397 | */ |
2314b42d | 398 | static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask) |
fef1bdd6 | 399 | { |
c55db957 KM |
400 | struct task_struct *p; |
401 | struct task_struct *task; | |
fef1bdd6 | 402 | |
af5b0f6a | 403 | pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name\n"); |
6b0c81b3 | 404 | rcu_read_lock(); |
c55db957 | 405 | for_each_process(p) { |
72835c86 | 406 | if (oom_unkillable_task(p, memcg, nodemask)) |
b4416d2b | 407 | continue; |
fef1bdd6 | 408 | |
c55db957 KM |
409 | task = find_lock_task_mm(p); |
410 | if (!task) { | |
6d2661ed | 411 | /* |
74ab7f1d DR |
412 | * This is a kthread or all of p's threads have already |
413 | * detached their mm's. There's no need to report | |
c55db957 | 414 | * them; they can't be oom killed anyway. |
6d2661ed | 415 | */ |
6d2661ed DR |
416 | continue; |
417 | } | |
c55db957 | 418 | |
af5b0f6a | 419 | pr_info("[%5d] %5d %5d %8lu %8lu %8ld %8lu %5hd %s\n", |
078de5f7 EB |
420 | task->pid, from_kuid(&init_user_ns, task_uid(task)), |
421 | task->tgid, task->mm->total_vm, get_mm_rss(task->mm), | |
af5b0f6a | 422 | mm_pgtables_bytes(task->mm), |
de34d965 | 423 | get_mm_counter(task->mm, MM_SWAPENTS), |
a63d83f4 | 424 | task->signal->oom_score_adj, task->comm); |
c55db957 KM |
425 | task_unlock(task); |
426 | } | |
6b0c81b3 | 427 | rcu_read_unlock(); |
fef1bdd6 DR |
428 | } |
429 | ||
2a966b77 | 430 | static void dump_header(struct oom_control *oc, struct task_struct *p) |
1b604d75 | 431 | { |
0205f755 MH |
432 | pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), nodemask=%*pbl, order=%d, oom_score_adj=%hd\n", |
433 | current->comm, oc->gfp_mask, &oc->gfp_mask, | |
434 | nodemask_pr_args(oc->nodemask), oc->order, | |
435 | current->signal->oom_score_adj); | |
9254990f MH |
436 | if (!IS_ENABLED(CONFIG_COMPACTION) && oc->order) |
437 | pr_warn("COMPACTION is disabled!!!\n"); | |
a0795cd4 | 438 | |
da39da3a | 439 | cpuset_print_current_mems_allowed(); |
1b604d75 | 440 | dump_stack(); |
852d8be0 | 441 | if (is_memcg_oom(oc)) |
2a966b77 | 442 | mem_cgroup_print_oom_info(oc->memcg, p); |
852d8be0 | 443 | else { |
299c517a | 444 | show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask); |
852d8be0 YS |
445 | if (is_dump_unreclaim_slabs()) |
446 | dump_unreclaimable_slab(); | |
447 | } | |
1b604d75 | 448 | if (sysctl_oom_dump_tasks) |
2a966b77 | 449 | dump_tasks(oc->memcg, oc->nodemask); |
1b604d75 DR |
450 | } |
451 | ||
5695be14 | 452 | /* |
c32b3cbe | 453 | * Number of OOM victims in flight |
5695be14 | 454 | */ |
c32b3cbe MH |
455 | static atomic_t oom_victims = ATOMIC_INIT(0); |
456 | static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait); | |
5695be14 | 457 | |
7c5f64f8 | 458 | static bool oom_killer_disabled __read_mostly; |
5695be14 | 459 | |
bc448e89 MH |
460 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
461 | ||
3ef22dff MH |
462 | /* |
463 | * task->mm can be NULL if the task is the exited group leader. So to | |
464 | * determine whether the task is using a particular mm, we examine all the | |
465 | * task's threads: if one of those is using this mm then this task was also | |
466 | * using it. | |
467 | */ | |
44a70ade | 468 | bool process_shares_mm(struct task_struct *p, struct mm_struct *mm) |
3ef22dff MH |
469 | { |
470 | struct task_struct *t; | |
471 | ||
472 | for_each_thread(p, t) { | |
473 | struct mm_struct *t_mm = READ_ONCE(t->mm); | |
474 | if (t_mm) | |
475 | return t_mm == mm; | |
476 | } | |
477 | return false; | |
478 | } | |
479 | ||
aac45363 MH |
480 | #ifdef CONFIG_MMU |
481 | /* | |
482 | * OOM Reaper kernel thread which tries to reap the memory used by the OOM | |
483 | * victim (if that is possible) to help the OOM killer to move on. | |
484 | */ | |
485 | static struct task_struct *oom_reaper_th; | |
aac45363 | 486 | static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait); |
29c696e1 | 487 | static struct task_struct *oom_reaper_list; |
03049269 MH |
488 | static DEFINE_SPINLOCK(oom_reaper_lock); |
489 | ||
93065ac7 | 490 | bool __oom_reap_task_mm(struct mm_struct *mm) |
aac45363 | 491 | { |
aac45363 | 492 | struct vm_area_struct *vma; |
93065ac7 | 493 | bool ret = true; |
27ae357f DR |
494 | |
495 | /* | |
496 | * Tell all users of get_user/copy_from_user etc... that the content | |
497 | * is no longer stable. No barriers really needed because unmapping | |
498 | * should imply barriers already and the reader would hit a page fault | |
499 | * if it stumbled over a reaped memory. | |
500 | */ | |
501 | set_bit(MMF_UNSTABLE, &mm->flags); | |
502 | ||
503 | for (vma = mm->mmap ; vma; vma = vma->vm_next) { | |
504 | if (!can_madv_dontneed_vma(vma)) | |
505 | continue; | |
506 | ||
507 | /* | |
508 | * Only anonymous pages have a good chance to be dropped | |
509 | * without additional steps which we cannot afford as we | |
510 | * are OOM already. | |
511 | * | |
512 | * We do not even care about fs backed pages because all | |
513 | * which are reclaimable have already been reclaimed and | |
514 | * we do not want to block exit_mmap by keeping mm ref | |
515 | * count elevated without a good reason. | |
516 | */ | |
517 | if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) { | |
518 | const unsigned long start = vma->vm_start; | |
519 | const unsigned long end = vma->vm_end; | |
520 | struct mmu_gather tlb; | |
521 | ||
522 | tlb_gather_mmu(&tlb, mm, start, end); | |
93065ac7 MH |
523 | if (mmu_notifier_invalidate_range_start_nonblock(mm, start, end)) { |
524 | ret = false; | |
525 | continue; | |
526 | } | |
27ae357f DR |
527 | unmap_page_range(&tlb, vma, start, end, NULL); |
528 | mmu_notifier_invalidate_range_end(mm, start, end); | |
529 | tlb_finish_mmu(&tlb, start, end); | |
530 | } | |
531 | } | |
93065ac7 MH |
532 | |
533 | return ret; | |
27ae357f DR |
534 | } |
535 | ||
536 | static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm) | |
537 | { | |
aac45363 MH |
538 | bool ret = true; |
539 | ||
e2fe1456 MH |
540 | /* |
541 | * We have to make sure to not race with the victim exit path | |
542 | * and cause premature new oom victim selection: | |
27ae357f | 543 | * oom_reap_task_mm exit_mm |
e5e3f4c4 | 544 | * mmget_not_zero |
e2fe1456 MH |
545 | * mmput |
546 | * atomic_dec_and_test | |
547 | * exit_oom_victim | |
548 | * [...] | |
549 | * out_of_memory | |
550 | * select_bad_process | |
551 | * # no TIF_MEMDIE task selects new victim | |
552 | * unmap_page_range # frees some memory | |
553 | */ | |
554 | mutex_lock(&oom_lock); | |
555 | ||
aac45363 MH |
556 | if (!down_read_trylock(&mm->mmap_sem)) { |
557 | ret = false; | |
422580c3 | 558 | trace_skip_task_reaping(tsk->pid); |
7ebffa45 | 559 | goto unlock_oom; |
e5e3f4c4 MH |
560 | } |
561 | ||
562 | /* | |
21292580 AA |
563 | * MMF_OOM_SKIP is set by exit_mmap when the OOM reaper can't |
564 | * work on the mm anymore. The check for MMF_OOM_SKIP must run | |
565 | * under mmap_sem for reading because it serializes against the | |
566 | * down_write();up_write() cycle in exit_mmap(). | |
e5e3f4c4 | 567 | */ |
21292580 | 568 | if (test_bit(MMF_OOM_SKIP, &mm->flags)) { |
e5e3f4c4 | 569 | up_read(&mm->mmap_sem); |
422580c3 | 570 | trace_skip_task_reaping(tsk->pid); |
7ebffa45 | 571 | goto unlock_oom; |
aac45363 MH |
572 | } |
573 | ||
422580c3 RG |
574 | trace_start_task_reaping(tsk->pid); |
575 | ||
93065ac7 MH |
576 | /* failed to reap part of the address space. Try again later */ |
577 | if (!__oom_reap_task_mm(mm)) { | |
578 | up_read(&mm->mmap_sem); | |
579 | ret = false; | |
580 | goto unlock_oom; | |
581 | } | |
aac45363 | 582 | |
bc448e89 MH |
583 | pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n", |
584 | task_pid_nr(tsk), tsk->comm, | |
585 | K(get_mm_counter(mm, MM_ANONPAGES)), | |
586 | K(get_mm_counter(mm, MM_FILEPAGES)), | |
587 | K(get_mm_counter(mm, MM_SHMEMPAGES))); | |
aac45363 | 588 | up_read(&mm->mmap_sem); |
36324a99 | 589 | |
422580c3 | 590 | trace_finish_task_reaping(tsk->pid); |
e5e3f4c4 MH |
591 | unlock_oom: |
592 | mutex_unlock(&oom_lock); | |
aac45363 MH |
593 | return ret; |
594 | } | |
595 | ||
bc448e89 | 596 | #define MAX_OOM_REAP_RETRIES 10 |
36324a99 | 597 | static void oom_reap_task(struct task_struct *tsk) |
aac45363 MH |
598 | { |
599 | int attempts = 0; | |
26db62f1 | 600 | struct mm_struct *mm = tsk->signal->oom_mm; |
aac45363 MH |
601 | |
602 | /* Retry the down_read_trylock(mmap_sem) a few times */ | |
27ae357f | 603 | while (attempts++ < MAX_OOM_REAP_RETRIES && !oom_reap_task_mm(tsk, mm)) |
aac45363 MH |
604 | schedule_timeout_idle(HZ/10); |
605 | ||
97b1255c TH |
606 | if (attempts <= MAX_OOM_REAP_RETRIES || |
607 | test_bit(MMF_OOM_SKIP, &mm->flags)) | |
7ebffa45 | 608 | goto done; |
11a410d5 | 609 | |
7ebffa45 TH |
610 | pr_info("oom_reaper: unable to reap pid:%d (%s)\n", |
611 | task_pid_nr(tsk), tsk->comm); | |
7ebffa45 | 612 | debug_show_all_locks(); |
bc448e89 | 613 | |
7ebffa45 | 614 | done: |
449d777d | 615 | tsk->oom_reaper_list = NULL; |
449d777d | 616 | |
26db62f1 MH |
617 | /* |
618 | * Hide this mm from OOM killer because it has been either reaped or | |
619 | * somebody can't call up_write(mmap_sem). | |
620 | */ | |
862e3073 | 621 | set_bit(MMF_OOM_SKIP, &mm->flags); |
26db62f1 | 622 | |
aac45363 | 623 | /* Drop a reference taken by wake_oom_reaper */ |
36324a99 | 624 | put_task_struct(tsk); |
aac45363 MH |
625 | } |
626 | ||
627 | static int oom_reaper(void *unused) | |
628 | { | |
629 | while (true) { | |
03049269 | 630 | struct task_struct *tsk = NULL; |
aac45363 | 631 | |
29c696e1 | 632 | wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL); |
03049269 | 633 | spin_lock(&oom_reaper_lock); |
29c696e1 VD |
634 | if (oom_reaper_list != NULL) { |
635 | tsk = oom_reaper_list; | |
636 | oom_reaper_list = tsk->oom_reaper_list; | |
03049269 MH |
637 | } |
638 | spin_unlock(&oom_reaper_lock); | |
639 | ||
640 | if (tsk) | |
641 | oom_reap_task(tsk); | |
aac45363 MH |
642 | } |
643 | ||
644 | return 0; | |
645 | } | |
646 | ||
7c5f64f8 | 647 | static void wake_oom_reaper(struct task_struct *tsk) |
aac45363 | 648 | { |
af8e15cc MH |
649 | /* tsk is already queued? */ |
650 | if (tsk == oom_reaper_list || tsk->oom_reaper_list) | |
aac45363 MH |
651 | return; |
652 | ||
36324a99 | 653 | get_task_struct(tsk); |
aac45363 | 654 | |
03049269 | 655 | spin_lock(&oom_reaper_lock); |
29c696e1 VD |
656 | tsk->oom_reaper_list = oom_reaper_list; |
657 | oom_reaper_list = tsk; | |
03049269 | 658 | spin_unlock(&oom_reaper_lock); |
422580c3 | 659 | trace_wake_reaper(tsk->pid); |
03049269 | 660 | wake_up(&oom_reaper_wait); |
aac45363 MH |
661 | } |
662 | ||
663 | static int __init oom_init(void) | |
664 | { | |
665 | oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper"); | |
aac45363 MH |
666 | return 0; |
667 | } | |
668 | subsys_initcall(oom_init) | |
7c5f64f8 VD |
669 | #else |
670 | static inline void wake_oom_reaper(struct task_struct *tsk) | |
671 | { | |
672 | } | |
673 | #endif /* CONFIG_MMU */ | |
aac45363 | 674 | |
49550b60 | 675 | /** |
16e95196 | 676 | * mark_oom_victim - mark the given task as OOM victim |
49550b60 | 677 | * @tsk: task to mark |
c32b3cbe | 678 | * |
dc56401f | 679 | * Has to be called with oom_lock held and never after |
c32b3cbe | 680 | * oom has been disabled already. |
26db62f1 MH |
681 | * |
682 | * tsk->mm has to be non NULL and caller has to guarantee it is stable (either | |
683 | * under task_lock or operate on the current). | |
49550b60 | 684 | */ |
7c5f64f8 | 685 | static void mark_oom_victim(struct task_struct *tsk) |
49550b60 | 686 | { |
26db62f1 MH |
687 | struct mm_struct *mm = tsk->mm; |
688 | ||
c32b3cbe MH |
689 | WARN_ON(oom_killer_disabled); |
690 | /* OOM killer might race with memcg OOM */ | |
691 | if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE)) | |
692 | return; | |
26db62f1 | 693 | |
26db62f1 | 694 | /* oom_mm is bound to the signal struct life time. */ |
4837fe37 | 695 | if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm)) { |
f1f10076 | 696 | mmgrab(tsk->signal->oom_mm); |
4837fe37 MH |
697 | set_bit(MMF_OOM_VICTIM, &mm->flags); |
698 | } | |
26db62f1 | 699 | |
63a8ca9b MH |
700 | /* |
701 | * Make sure that the task is woken up from uninterruptible sleep | |
702 | * if it is frozen because OOM killer wouldn't be able to free | |
703 | * any memory and livelock. freezing_slow_path will tell the freezer | |
704 | * that TIF_MEMDIE tasks should be ignored. | |
705 | */ | |
706 | __thaw_task(tsk); | |
c32b3cbe | 707 | atomic_inc(&oom_victims); |
422580c3 | 708 | trace_mark_victim(tsk->pid); |
49550b60 MH |
709 | } |
710 | ||
711 | /** | |
16e95196 | 712 | * exit_oom_victim - note the exit of an OOM victim |
49550b60 | 713 | */ |
38531201 | 714 | void exit_oom_victim(void) |
49550b60 | 715 | { |
38531201 | 716 | clear_thread_flag(TIF_MEMDIE); |
c32b3cbe | 717 | |
c38f1025 | 718 | if (!atomic_dec_return(&oom_victims)) |
c32b3cbe | 719 | wake_up_all(&oom_victims_wait); |
c32b3cbe MH |
720 | } |
721 | ||
7d2e7a22 MH |
722 | /** |
723 | * oom_killer_enable - enable OOM killer | |
724 | */ | |
725 | void oom_killer_enable(void) | |
726 | { | |
727 | oom_killer_disabled = false; | |
d75da004 | 728 | pr_info("OOM killer enabled.\n"); |
7d2e7a22 MH |
729 | } |
730 | ||
c32b3cbe MH |
731 | /** |
732 | * oom_killer_disable - disable OOM killer | |
7d2e7a22 | 733 | * @timeout: maximum timeout to wait for oom victims in jiffies |
c32b3cbe MH |
734 | * |
735 | * Forces all page allocations to fail rather than trigger OOM killer. | |
7d2e7a22 MH |
736 | * Will block and wait until all OOM victims are killed or the given |
737 | * timeout expires. | |
c32b3cbe MH |
738 | * |
739 | * The function cannot be called when there are runnable user tasks because | |
740 | * the userspace would see unexpected allocation failures as a result. Any | |
741 | * new usage of this function should be consulted with MM people. | |
742 | * | |
743 | * Returns true if successful and false if the OOM killer cannot be | |
744 | * disabled. | |
745 | */ | |
7d2e7a22 | 746 | bool oom_killer_disable(signed long timeout) |
c32b3cbe | 747 | { |
7d2e7a22 MH |
748 | signed long ret; |
749 | ||
c32b3cbe | 750 | /* |
6afcf289 TH |
751 | * Make sure to not race with an ongoing OOM killer. Check that the |
752 | * current is not killed (possibly due to sharing the victim's memory). | |
c32b3cbe | 753 | */ |
6afcf289 | 754 | if (mutex_lock_killable(&oom_lock)) |
c32b3cbe | 755 | return false; |
c32b3cbe | 756 | oom_killer_disabled = true; |
dc56401f | 757 | mutex_unlock(&oom_lock); |
c32b3cbe | 758 | |
7d2e7a22 MH |
759 | ret = wait_event_interruptible_timeout(oom_victims_wait, |
760 | !atomic_read(&oom_victims), timeout); | |
761 | if (ret <= 0) { | |
762 | oom_killer_enable(); | |
763 | return false; | |
764 | } | |
d75da004 | 765 | pr_info("OOM killer disabled.\n"); |
c32b3cbe MH |
766 | |
767 | return true; | |
768 | } | |
769 | ||
1af8bb43 MH |
770 | static inline bool __task_will_free_mem(struct task_struct *task) |
771 | { | |
772 | struct signal_struct *sig = task->signal; | |
773 | ||
774 | /* | |
775 | * A coredumping process may sleep for an extended period in exit_mm(), | |
776 | * so the oom killer cannot assume that the process will promptly exit | |
777 | * and release memory. | |
778 | */ | |
779 | if (sig->flags & SIGNAL_GROUP_COREDUMP) | |
780 | return false; | |
781 | ||
782 | if (sig->flags & SIGNAL_GROUP_EXIT) | |
783 | return true; | |
784 | ||
785 | if (thread_group_empty(task) && (task->flags & PF_EXITING)) | |
786 | return true; | |
787 | ||
788 | return false; | |
789 | } | |
790 | ||
791 | /* | |
792 | * Checks whether the given task is dying or exiting and likely to | |
793 | * release its address space. This means that all threads and processes | |
794 | * sharing the same mm have to be killed or exiting. | |
091f362c MH |
795 | * Caller has to make sure that task->mm is stable (hold task_lock or |
796 | * it operates on the current). | |
1af8bb43 | 797 | */ |
7c5f64f8 | 798 | static bool task_will_free_mem(struct task_struct *task) |
1af8bb43 | 799 | { |
091f362c | 800 | struct mm_struct *mm = task->mm; |
1af8bb43 | 801 | struct task_struct *p; |
f33e6f06 | 802 | bool ret = true; |
1af8bb43 | 803 | |
1af8bb43 | 804 | /* |
091f362c MH |
805 | * Skip tasks without mm because it might have passed its exit_mm and |
806 | * exit_oom_victim. oom_reaper could have rescued that but do not rely | |
807 | * on that for now. We can consider find_lock_task_mm in future. | |
1af8bb43 | 808 | */ |
091f362c | 809 | if (!mm) |
1af8bb43 MH |
810 | return false; |
811 | ||
091f362c MH |
812 | if (!__task_will_free_mem(task)) |
813 | return false; | |
696453e6 MH |
814 | |
815 | /* | |
816 | * This task has already been drained by the oom reaper so there are | |
817 | * only small chances it will free some more | |
818 | */ | |
862e3073 | 819 | if (test_bit(MMF_OOM_SKIP, &mm->flags)) |
696453e6 | 820 | return false; |
696453e6 | 821 | |
091f362c | 822 | if (atomic_read(&mm->mm_users) <= 1) |
1af8bb43 | 823 | return true; |
1af8bb43 MH |
824 | |
825 | /* | |
5870c2e1 MH |
826 | * Make sure that all tasks which share the mm with the given tasks |
827 | * are dying as well to make sure that a) nobody pins its mm and | |
828 | * b) the task is also reapable by the oom reaper. | |
1af8bb43 MH |
829 | */ |
830 | rcu_read_lock(); | |
831 | for_each_process(p) { | |
832 | if (!process_shares_mm(p, mm)) | |
833 | continue; | |
834 | if (same_thread_group(task, p)) | |
835 | continue; | |
836 | ret = __task_will_free_mem(p); | |
837 | if (!ret) | |
838 | break; | |
839 | } | |
840 | rcu_read_unlock(); | |
1af8bb43 MH |
841 | |
842 | return ret; | |
843 | } | |
844 | ||
7c5f64f8 | 845 | static void oom_kill_process(struct oom_control *oc, const char *message) |
1da177e4 | 846 | { |
7c5f64f8 VD |
847 | struct task_struct *p = oc->chosen; |
848 | unsigned int points = oc->chosen_points; | |
52d3c036 | 849 | struct task_struct *victim = p; |
5e9d834a | 850 | struct task_struct *child; |
1da4db0c | 851 | struct task_struct *t; |
647f2bdf | 852 | struct mm_struct *mm; |
52d3c036 | 853 | unsigned int victim_points = 0; |
dc3f21ea DR |
854 | static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
855 | DEFAULT_RATELIMIT_BURST); | |
bb29902a | 856 | bool can_oom_reap = true; |
1da177e4 | 857 | |
50ec3bbf NP |
858 | /* |
859 | * If the task is already exiting, don't alarm the sysadmin or kill | |
cd04ae1e MH |
860 | * its children or threads, just give it access to memory reserves |
861 | * so it can die quickly | |
50ec3bbf | 862 | */ |
091f362c | 863 | task_lock(p); |
1af8bb43 | 864 | if (task_will_free_mem(p)) { |
16e95196 | 865 | mark_oom_victim(p); |
1af8bb43 | 866 | wake_oom_reaper(p); |
091f362c | 867 | task_unlock(p); |
6b0c81b3 | 868 | put_task_struct(p); |
2a1c9b1f | 869 | return; |
50ec3bbf | 870 | } |
091f362c | 871 | task_unlock(p); |
50ec3bbf | 872 | |
dc3f21ea | 873 | if (__ratelimit(&oom_rs)) |
2a966b77 | 874 | dump_header(oc, p); |
8447d950 | 875 | |
f0d6647e | 876 | pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n", |
5e9d834a | 877 | message, task_pid_nr(p), p->comm, points); |
f3af38d3 | 878 | |
5e9d834a DR |
879 | /* |
880 | * If any of p's children has a different mm and is eligible for kill, | |
11239836 | 881 | * the one with the highest oom_badness() score is sacrificed for its |
5e9d834a DR |
882 | * parent. This attempts to lose the minimal amount of work done while |
883 | * still freeing memory. | |
884 | */ | |
6b0c81b3 | 885 | read_lock(&tasklist_lock); |
1da4db0c | 886 | for_each_thread(p, t) { |
5e9d834a | 887 | list_for_each_entry(child, &t->children, sibling) { |
a63d83f4 | 888 | unsigned int child_points; |
5e9d834a | 889 | |
4d7b3394 | 890 | if (process_shares_mm(child, p->mm)) |
edd45544 | 891 | continue; |
a63d83f4 DR |
892 | /* |
893 | * oom_badness() returns 0 if the thread is unkillable | |
894 | */ | |
2a966b77 | 895 | child_points = oom_badness(child, |
7c5f64f8 | 896 | oc->memcg, oc->nodemask, oc->totalpages); |
5e9d834a | 897 | if (child_points > victim_points) { |
6b0c81b3 | 898 | put_task_struct(victim); |
5e9d834a DR |
899 | victim = child; |
900 | victim_points = child_points; | |
6b0c81b3 | 901 | get_task_struct(victim); |
5e9d834a | 902 | } |
dd8e8f40 | 903 | } |
1da4db0c | 904 | } |
6b0c81b3 | 905 | read_unlock(&tasklist_lock); |
dd8e8f40 | 906 | |
6b0c81b3 DR |
907 | p = find_lock_task_mm(victim); |
908 | if (!p) { | |
6b0c81b3 | 909 | put_task_struct(victim); |
647f2bdf | 910 | return; |
6b0c81b3 DR |
911 | } else if (victim != p) { |
912 | get_task_struct(p); | |
913 | put_task_struct(victim); | |
914 | victim = p; | |
915 | } | |
647f2bdf | 916 | |
880b7689 | 917 | /* Get a reference to safely compare mm after task_unlock(victim) */ |
647f2bdf | 918 | mm = victim->mm; |
f1f10076 | 919 | mmgrab(mm); |
8e675f7a KK |
920 | |
921 | /* Raise event before sending signal: task reaper must see this */ | |
922 | count_vm_event(OOM_KILL); | |
fe6bdfc8 | 923 | memcg_memory_event_mm(mm, MEMCG_OOM_KILL); |
8e675f7a | 924 | |
426fb5e7 | 925 | /* |
cd04ae1e MH |
926 | * We should send SIGKILL before granting access to memory reserves |
927 | * in order to prevent the OOM victim from depleting the memory | |
928 | * reserves from the user space under its control. | |
426fb5e7 TH |
929 | */ |
930 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true); | |
16e95196 | 931 | mark_oom_victim(victim); |
eca56ff9 | 932 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n", |
647f2bdf DR |
933 | task_pid_nr(victim), victim->comm, K(victim->mm->total_vm), |
934 | K(get_mm_counter(victim->mm, MM_ANONPAGES)), | |
eca56ff9 JM |
935 | K(get_mm_counter(victim->mm, MM_FILEPAGES)), |
936 | K(get_mm_counter(victim->mm, MM_SHMEMPAGES))); | |
647f2bdf DR |
937 | task_unlock(victim); |
938 | ||
939 | /* | |
940 | * Kill all user processes sharing victim->mm in other thread groups, if | |
941 | * any. They don't get access to memory reserves, though, to avoid | |
942 | * depletion of all memory. This prevents mm->mmap_sem livelock when an | |
943 | * oom killed thread cannot exit because it requires the semaphore and | |
944 | * its contended by another thread trying to allocate memory itself. | |
945 | * That thread will now get access to memory reserves since it has a | |
946 | * pending fatal signal. | |
947 | */ | |
4d4048be | 948 | rcu_read_lock(); |
c319025a | 949 | for_each_process(p) { |
4d7b3394 | 950 | if (!process_shares_mm(p, mm)) |
c319025a ON |
951 | continue; |
952 | if (same_thread_group(p, victim)) | |
953 | continue; | |
1b51e65e | 954 | if (is_global_init(p)) { |
aac45363 | 955 | can_oom_reap = false; |
862e3073 | 956 | set_bit(MMF_OOM_SKIP, &mm->flags); |
a373966d MH |
957 | pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n", |
958 | task_pid_nr(victim), victim->comm, | |
959 | task_pid_nr(p), p->comm); | |
c319025a | 960 | continue; |
aac45363 | 961 | } |
1b51e65e MH |
962 | /* |
963 | * No use_mm() user needs to read from the userspace so we are | |
964 | * ok to reap it. | |
965 | */ | |
966 | if (unlikely(p->flags & PF_KTHREAD)) | |
967 | continue; | |
c319025a ON |
968 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true); |
969 | } | |
6b0c81b3 | 970 | rcu_read_unlock(); |
647f2bdf | 971 | |
aac45363 | 972 | if (can_oom_reap) |
36324a99 | 973 | wake_oom_reaper(victim); |
aac45363 | 974 | |
880b7689 | 975 | mmdrop(mm); |
6b0c81b3 | 976 | put_task_struct(victim); |
1da177e4 | 977 | } |
647f2bdf | 978 | #undef K |
1da177e4 | 979 | |
309ed882 DR |
980 | /* |
981 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. | |
982 | */ | |
7c5f64f8 VD |
983 | static void check_panic_on_oom(struct oom_control *oc, |
984 | enum oom_constraint constraint) | |
309ed882 DR |
985 | { |
986 | if (likely(!sysctl_panic_on_oom)) | |
987 | return; | |
988 | if (sysctl_panic_on_oom != 2) { | |
989 | /* | |
990 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel | |
991 | * does not panic for cpuset, mempolicy, or memcg allocation | |
992 | * failures. | |
993 | */ | |
994 | if (constraint != CONSTRAINT_NONE) | |
995 | return; | |
996 | } | |
071a4bef | 997 | /* Do not panic for oom kills triggered by sysrq */ |
db2a0dd7 | 998 | if (is_sysrq_oom(oc)) |
071a4bef | 999 | return; |
2a966b77 | 1000 | dump_header(oc, NULL); |
309ed882 DR |
1001 | panic("Out of memory: %s panic_on_oom is enabled\n", |
1002 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); | |
1003 | } | |
1004 | ||
8bc719d3 MS |
1005 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
1006 | ||
1007 | int register_oom_notifier(struct notifier_block *nb) | |
1008 | { | |
1009 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
1010 | } | |
1011 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
1012 | ||
1013 | int unregister_oom_notifier(struct notifier_block *nb) | |
1014 | { | |
1015 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
1016 | } | |
1017 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
1018 | ||
1da177e4 | 1019 | /** |
6e0fc46d DR |
1020 | * out_of_memory - kill the "best" process when we run out of memory |
1021 | * @oc: pointer to struct oom_control | |
1da177e4 LT |
1022 | * |
1023 | * If we run out of memory, we have the choice between either | |
1024 | * killing a random task (bad), letting the system crash (worse) | |
1025 | * OR try to be smart about which process to kill. Note that we | |
1026 | * don't have to be perfect here, we just have to be good. | |
1027 | */ | |
6e0fc46d | 1028 | bool out_of_memory(struct oom_control *oc) |
1da177e4 | 1029 | { |
8bc719d3 | 1030 | unsigned long freed = 0; |
e3658932 | 1031 | enum oom_constraint constraint = CONSTRAINT_NONE; |
8bc719d3 | 1032 | |
dc56401f JW |
1033 | if (oom_killer_disabled) |
1034 | return false; | |
1035 | ||
7c5f64f8 VD |
1036 | if (!is_memcg_oom(oc)) { |
1037 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
1038 | if (freed > 0) | |
1039 | /* Got some memory back in the last second. */ | |
1040 | return true; | |
1041 | } | |
1da177e4 | 1042 | |
7b98c2e4 | 1043 | /* |
9ff4868e DR |
1044 | * If current has a pending SIGKILL or is exiting, then automatically |
1045 | * select it. The goal is to allow it to allocate so that it may | |
1046 | * quickly exit and free its memory. | |
7b98c2e4 | 1047 | */ |
091f362c | 1048 | if (task_will_free_mem(current)) { |
16e95196 | 1049 | mark_oom_victim(current); |
1af8bb43 | 1050 | wake_oom_reaper(current); |
75e8f8b2 | 1051 | return true; |
7b98c2e4 DR |
1052 | } |
1053 | ||
3da88fb3 MH |
1054 | /* |
1055 | * The OOM killer does not compensate for IO-less reclaim. | |
1056 | * pagefault_out_of_memory lost its gfp context so we have to | |
1057 | * make sure exclude 0 mask - all other users should have at least | |
1058 | * ___GFP_DIRECT_RECLAIM to get here. | |
1059 | */ | |
06ad276a | 1060 | if (oc->gfp_mask && !(oc->gfp_mask & __GFP_FS)) |
3da88fb3 MH |
1061 | return true; |
1062 | ||
9b0f8b04 CL |
1063 | /* |
1064 | * Check if there were limitations on the allocation (only relevant for | |
7c5f64f8 | 1065 | * NUMA and memcg) that may require different handling. |
9b0f8b04 | 1066 | */ |
7c5f64f8 | 1067 | constraint = constrained_alloc(oc); |
6e0fc46d DR |
1068 | if (constraint != CONSTRAINT_MEMORY_POLICY) |
1069 | oc->nodemask = NULL; | |
2a966b77 | 1070 | check_panic_on_oom(oc, constraint); |
0aad4b31 | 1071 | |
7c5f64f8 VD |
1072 | if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task && |
1073 | current->mm && !oom_unkillable_task(current, NULL, oc->nodemask) && | |
121d1ba0 | 1074 | current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) { |
6b0c81b3 | 1075 | get_task_struct(current); |
7c5f64f8 VD |
1076 | oc->chosen = current; |
1077 | oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)"); | |
75e8f8b2 | 1078 | return true; |
0aad4b31 DR |
1079 | } |
1080 | ||
7c5f64f8 | 1081 | select_bad_process(oc); |
0aad4b31 | 1082 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
7c5f64f8 | 1083 | if (!oc->chosen && !is_sysrq_oom(oc) && !is_memcg_oom(oc)) { |
2a966b77 | 1084 | dump_header(oc, NULL); |
0aad4b31 DR |
1085 | panic("Out of memory and no killable processes...\n"); |
1086 | } | |
9bfe5ded | 1087 | if (oc->chosen && oc->chosen != (void *)-1UL) |
7c5f64f8 VD |
1088 | oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" : |
1089 | "Memory cgroup out of memory"); | |
7c5f64f8 | 1090 | return !!oc->chosen; |
c32b3cbe MH |
1091 | } |
1092 | ||
e3658932 DR |
1093 | /* |
1094 | * The pagefault handler calls here because it is out of memory, so kill a | |
798fd756 VD |
1095 | * memory-hogging task. If oom_lock is held by somebody else, a parallel oom |
1096 | * killing is already in progress so do nothing. | |
e3658932 DR |
1097 | */ |
1098 | void pagefault_out_of_memory(void) | |
1099 | { | |
6e0fc46d DR |
1100 | struct oom_control oc = { |
1101 | .zonelist = NULL, | |
1102 | .nodemask = NULL, | |
2a966b77 | 1103 | .memcg = NULL, |
6e0fc46d DR |
1104 | .gfp_mask = 0, |
1105 | .order = 0, | |
6e0fc46d DR |
1106 | }; |
1107 | ||
49426420 | 1108 | if (mem_cgroup_oom_synchronize(true)) |
dc56401f | 1109 | return; |
3812c8c8 | 1110 | |
dc56401f JW |
1111 | if (!mutex_trylock(&oom_lock)) |
1112 | return; | |
a104808e | 1113 | out_of_memory(&oc); |
dc56401f | 1114 | mutex_unlock(&oom_lock); |
e3658932 | 1115 | } |