]>
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 | |
c3b78b11 RF |
403 | pr_info("Tasks state (memory values in pages):\n"); |
404 | pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name\n"); | |
6b0c81b3 | 405 | rcu_read_lock(); |
c55db957 | 406 | for_each_process(p) { |
72835c86 | 407 | if (oom_unkillable_task(p, memcg, nodemask)) |
b4416d2b | 408 | continue; |
fef1bdd6 | 409 | |
c55db957 KM |
410 | task = find_lock_task_mm(p); |
411 | if (!task) { | |
6d2661ed | 412 | /* |
74ab7f1d DR |
413 | * This is a kthread or all of p's threads have already |
414 | * detached their mm's. There's no need to report | |
c55db957 | 415 | * them; they can't be oom killed anyway. |
6d2661ed | 416 | */ |
6d2661ed DR |
417 | continue; |
418 | } | |
c55db957 | 419 | |
c3b78b11 | 420 | pr_info("[%7d] %5d %5d %8lu %8lu %8ld %8lu %5hd %s\n", |
078de5f7 EB |
421 | task->pid, from_kuid(&init_user_ns, task_uid(task)), |
422 | task->tgid, task->mm->total_vm, get_mm_rss(task->mm), | |
af5b0f6a | 423 | mm_pgtables_bytes(task->mm), |
de34d965 | 424 | get_mm_counter(task->mm, MM_SWAPENTS), |
a63d83f4 | 425 | task->signal->oom_score_adj, task->comm); |
c55db957 KM |
426 | task_unlock(task); |
427 | } | |
6b0c81b3 | 428 | rcu_read_unlock(); |
fef1bdd6 DR |
429 | } |
430 | ||
2a966b77 | 431 | static void dump_header(struct oom_control *oc, struct task_struct *p) |
1b604d75 | 432 | { |
0205f755 MH |
433 | pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), nodemask=%*pbl, order=%d, oom_score_adj=%hd\n", |
434 | current->comm, oc->gfp_mask, &oc->gfp_mask, | |
435 | nodemask_pr_args(oc->nodemask), oc->order, | |
436 | current->signal->oom_score_adj); | |
9254990f MH |
437 | if (!IS_ENABLED(CONFIG_COMPACTION) && oc->order) |
438 | pr_warn("COMPACTION is disabled!!!\n"); | |
a0795cd4 | 439 | |
da39da3a | 440 | cpuset_print_current_mems_allowed(); |
1b604d75 | 441 | dump_stack(); |
852d8be0 | 442 | if (is_memcg_oom(oc)) |
2a966b77 | 443 | mem_cgroup_print_oom_info(oc->memcg, p); |
852d8be0 | 444 | else { |
299c517a | 445 | show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask); |
852d8be0 YS |
446 | if (is_dump_unreclaim_slabs()) |
447 | dump_unreclaimable_slab(); | |
448 | } | |
1b604d75 | 449 | if (sysctl_oom_dump_tasks) |
2a966b77 | 450 | dump_tasks(oc->memcg, oc->nodemask); |
1b604d75 DR |
451 | } |
452 | ||
5695be14 | 453 | /* |
c32b3cbe | 454 | * Number of OOM victims in flight |
5695be14 | 455 | */ |
c32b3cbe MH |
456 | static atomic_t oom_victims = ATOMIC_INIT(0); |
457 | static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait); | |
5695be14 | 458 | |
7c5f64f8 | 459 | static bool oom_killer_disabled __read_mostly; |
5695be14 | 460 | |
bc448e89 MH |
461 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
462 | ||
3ef22dff MH |
463 | /* |
464 | * task->mm can be NULL if the task is the exited group leader. So to | |
465 | * determine whether the task is using a particular mm, we examine all the | |
466 | * task's threads: if one of those is using this mm then this task was also | |
467 | * using it. | |
468 | */ | |
44a70ade | 469 | bool process_shares_mm(struct task_struct *p, struct mm_struct *mm) |
3ef22dff MH |
470 | { |
471 | struct task_struct *t; | |
472 | ||
473 | for_each_thread(p, t) { | |
474 | struct mm_struct *t_mm = READ_ONCE(t->mm); | |
475 | if (t_mm) | |
476 | return t_mm == mm; | |
477 | } | |
478 | return false; | |
479 | } | |
480 | ||
aac45363 MH |
481 | #ifdef CONFIG_MMU |
482 | /* | |
483 | * OOM Reaper kernel thread which tries to reap the memory used by the OOM | |
484 | * victim (if that is possible) to help the OOM killer to move on. | |
485 | */ | |
486 | static struct task_struct *oom_reaper_th; | |
aac45363 | 487 | static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait); |
29c696e1 | 488 | static struct task_struct *oom_reaper_list; |
03049269 MH |
489 | static DEFINE_SPINLOCK(oom_reaper_lock); |
490 | ||
93065ac7 | 491 | bool __oom_reap_task_mm(struct mm_struct *mm) |
aac45363 | 492 | { |
aac45363 | 493 | struct vm_area_struct *vma; |
93065ac7 | 494 | bool ret = true; |
27ae357f DR |
495 | |
496 | /* | |
497 | * Tell all users of get_user/copy_from_user etc... that the content | |
498 | * is no longer stable. No barriers really needed because unmapping | |
499 | * should imply barriers already and the reader would hit a page fault | |
500 | * if it stumbled over a reaped memory. | |
501 | */ | |
502 | set_bit(MMF_UNSTABLE, &mm->flags); | |
503 | ||
504 | for (vma = mm->mmap ; vma; vma = vma->vm_next) { | |
505 | if (!can_madv_dontneed_vma(vma)) | |
506 | continue; | |
507 | ||
508 | /* | |
509 | * Only anonymous pages have a good chance to be dropped | |
510 | * without additional steps which we cannot afford as we | |
511 | * are OOM already. | |
512 | * | |
513 | * We do not even care about fs backed pages because all | |
514 | * which are reclaimable have already been reclaimed and | |
515 | * we do not want to block exit_mmap by keeping mm ref | |
516 | * count elevated without a good reason. | |
517 | */ | |
518 | if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) { | |
519 | const unsigned long start = vma->vm_start; | |
520 | const unsigned long end = vma->vm_end; | |
521 | struct mmu_gather tlb; | |
522 | ||
523 | tlb_gather_mmu(&tlb, mm, start, end); | |
93065ac7 | 524 | if (mmu_notifier_invalidate_range_start_nonblock(mm, start, end)) { |
79cc8105 | 525 | tlb_finish_mmu(&tlb, start, end); |
93065ac7 MH |
526 | ret = false; |
527 | continue; | |
528 | } | |
27ae357f DR |
529 | unmap_page_range(&tlb, vma, start, end, NULL); |
530 | mmu_notifier_invalidate_range_end(mm, start, end); | |
531 | tlb_finish_mmu(&tlb, start, end); | |
532 | } | |
533 | } | |
93065ac7 MH |
534 | |
535 | return ret; | |
27ae357f DR |
536 | } |
537 | ||
431f42fd MH |
538 | /* |
539 | * Reaps the address space of the give task. | |
540 | * | |
541 | * Returns true on success and false if none or part of the address space | |
542 | * has been reclaimed and the caller should retry later. | |
543 | */ | |
27ae357f DR |
544 | static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm) |
545 | { | |
aac45363 MH |
546 | bool ret = true; |
547 | ||
aac45363 | 548 | if (!down_read_trylock(&mm->mmap_sem)) { |
422580c3 | 549 | trace_skip_task_reaping(tsk->pid); |
af5679fb | 550 | return false; |
4d4bbd85 MH |
551 | } |
552 | ||
e5e3f4c4 | 553 | /* |
21292580 AA |
554 | * MMF_OOM_SKIP is set by exit_mmap when the OOM reaper can't |
555 | * work on the mm anymore. The check for MMF_OOM_SKIP must run | |
556 | * under mmap_sem for reading because it serializes against the | |
557 | * down_write();up_write() cycle in exit_mmap(). | |
e5e3f4c4 | 558 | */ |
21292580 | 559 | if (test_bit(MMF_OOM_SKIP, &mm->flags)) { |
422580c3 | 560 | trace_skip_task_reaping(tsk->pid); |
431f42fd | 561 | goto out_unlock; |
aac45363 MH |
562 | } |
563 | ||
422580c3 RG |
564 | trace_start_task_reaping(tsk->pid); |
565 | ||
93065ac7 | 566 | /* failed to reap part of the address space. Try again later */ |
431f42fd MH |
567 | ret = __oom_reap_task_mm(mm); |
568 | if (!ret) | |
569 | goto out_finish; | |
aac45363 | 570 | |
bc448e89 MH |
571 | pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n", |
572 | task_pid_nr(tsk), tsk->comm, | |
573 | K(get_mm_counter(mm, MM_ANONPAGES)), | |
574 | K(get_mm_counter(mm, MM_FILEPAGES)), | |
575 | K(get_mm_counter(mm, MM_SHMEMPAGES))); | |
431f42fd MH |
576 | out_finish: |
577 | trace_finish_task_reaping(tsk->pid); | |
578 | out_unlock: | |
aac45363 | 579 | up_read(&mm->mmap_sem); |
36324a99 | 580 | |
aac45363 MH |
581 | return ret; |
582 | } | |
583 | ||
bc448e89 | 584 | #define MAX_OOM_REAP_RETRIES 10 |
36324a99 | 585 | static void oom_reap_task(struct task_struct *tsk) |
aac45363 MH |
586 | { |
587 | int attempts = 0; | |
26db62f1 | 588 | struct mm_struct *mm = tsk->signal->oom_mm; |
aac45363 MH |
589 | |
590 | /* Retry the down_read_trylock(mmap_sem) a few times */ | |
27ae357f | 591 | while (attempts++ < MAX_OOM_REAP_RETRIES && !oom_reap_task_mm(tsk, mm)) |
aac45363 MH |
592 | schedule_timeout_idle(HZ/10); |
593 | ||
97b1255c TH |
594 | if (attempts <= MAX_OOM_REAP_RETRIES || |
595 | test_bit(MMF_OOM_SKIP, &mm->flags)) | |
7ebffa45 | 596 | goto done; |
11a410d5 | 597 | |
7ebffa45 TH |
598 | pr_info("oom_reaper: unable to reap pid:%d (%s)\n", |
599 | task_pid_nr(tsk), tsk->comm); | |
7ebffa45 | 600 | debug_show_all_locks(); |
bc448e89 | 601 | |
7ebffa45 | 602 | done: |
449d777d | 603 | tsk->oom_reaper_list = NULL; |
449d777d | 604 | |
26db62f1 MH |
605 | /* |
606 | * Hide this mm from OOM killer because it has been either reaped or | |
607 | * somebody can't call up_write(mmap_sem). | |
608 | */ | |
862e3073 | 609 | set_bit(MMF_OOM_SKIP, &mm->flags); |
26db62f1 | 610 | |
aac45363 | 611 | /* Drop a reference taken by wake_oom_reaper */ |
36324a99 | 612 | put_task_struct(tsk); |
aac45363 MH |
613 | } |
614 | ||
615 | static int oom_reaper(void *unused) | |
616 | { | |
617 | while (true) { | |
03049269 | 618 | struct task_struct *tsk = NULL; |
aac45363 | 619 | |
29c696e1 | 620 | wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL); |
03049269 | 621 | spin_lock(&oom_reaper_lock); |
29c696e1 VD |
622 | if (oom_reaper_list != NULL) { |
623 | tsk = oom_reaper_list; | |
624 | oom_reaper_list = tsk->oom_reaper_list; | |
03049269 MH |
625 | } |
626 | spin_unlock(&oom_reaper_lock); | |
627 | ||
628 | if (tsk) | |
629 | oom_reap_task(tsk); | |
aac45363 MH |
630 | } |
631 | ||
632 | return 0; | |
633 | } | |
634 | ||
7c5f64f8 | 635 | static void wake_oom_reaper(struct task_struct *tsk) |
aac45363 | 636 | { |
af8e15cc MH |
637 | /* tsk is already queued? */ |
638 | if (tsk == oom_reaper_list || tsk->oom_reaper_list) | |
aac45363 MH |
639 | return; |
640 | ||
36324a99 | 641 | get_task_struct(tsk); |
aac45363 | 642 | |
03049269 | 643 | spin_lock(&oom_reaper_lock); |
29c696e1 VD |
644 | tsk->oom_reaper_list = oom_reaper_list; |
645 | oom_reaper_list = tsk; | |
03049269 | 646 | spin_unlock(&oom_reaper_lock); |
422580c3 | 647 | trace_wake_reaper(tsk->pid); |
03049269 | 648 | wake_up(&oom_reaper_wait); |
aac45363 MH |
649 | } |
650 | ||
651 | static int __init oom_init(void) | |
652 | { | |
653 | oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper"); | |
aac45363 MH |
654 | return 0; |
655 | } | |
656 | subsys_initcall(oom_init) | |
7c5f64f8 VD |
657 | #else |
658 | static inline void wake_oom_reaper(struct task_struct *tsk) | |
659 | { | |
660 | } | |
661 | #endif /* CONFIG_MMU */ | |
aac45363 | 662 | |
49550b60 | 663 | /** |
16e95196 | 664 | * mark_oom_victim - mark the given task as OOM victim |
49550b60 | 665 | * @tsk: task to mark |
c32b3cbe | 666 | * |
dc56401f | 667 | * Has to be called with oom_lock held and never after |
c32b3cbe | 668 | * oom has been disabled already. |
26db62f1 MH |
669 | * |
670 | * tsk->mm has to be non NULL and caller has to guarantee it is stable (either | |
671 | * under task_lock or operate on the current). | |
49550b60 | 672 | */ |
7c5f64f8 | 673 | static void mark_oom_victim(struct task_struct *tsk) |
49550b60 | 674 | { |
26db62f1 MH |
675 | struct mm_struct *mm = tsk->mm; |
676 | ||
c32b3cbe MH |
677 | WARN_ON(oom_killer_disabled); |
678 | /* OOM killer might race with memcg OOM */ | |
679 | if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE)) | |
680 | return; | |
26db62f1 | 681 | |
26db62f1 | 682 | /* oom_mm is bound to the signal struct life time. */ |
4837fe37 | 683 | if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm)) { |
f1f10076 | 684 | mmgrab(tsk->signal->oom_mm); |
4837fe37 MH |
685 | set_bit(MMF_OOM_VICTIM, &mm->flags); |
686 | } | |
26db62f1 | 687 | |
63a8ca9b MH |
688 | /* |
689 | * Make sure that the task is woken up from uninterruptible sleep | |
690 | * if it is frozen because OOM killer wouldn't be able to free | |
691 | * any memory and livelock. freezing_slow_path will tell the freezer | |
692 | * that TIF_MEMDIE tasks should be ignored. | |
693 | */ | |
694 | __thaw_task(tsk); | |
c32b3cbe | 695 | atomic_inc(&oom_victims); |
422580c3 | 696 | trace_mark_victim(tsk->pid); |
49550b60 MH |
697 | } |
698 | ||
699 | /** | |
16e95196 | 700 | * exit_oom_victim - note the exit of an OOM victim |
49550b60 | 701 | */ |
38531201 | 702 | void exit_oom_victim(void) |
49550b60 | 703 | { |
38531201 | 704 | clear_thread_flag(TIF_MEMDIE); |
c32b3cbe | 705 | |
c38f1025 | 706 | if (!atomic_dec_return(&oom_victims)) |
c32b3cbe | 707 | wake_up_all(&oom_victims_wait); |
c32b3cbe MH |
708 | } |
709 | ||
7d2e7a22 MH |
710 | /** |
711 | * oom_killer_enable - enable OOM killer | |
712 | */ | |
713 | void oom_killer_enable(void) | |
714 | { | |
715 | oom_killer_disabled = false; | |
d75da004 | 716 | pr_info("OOM killer enabled.\n"); |
7d2e7a22 MH |
717 | } |
718 | ||
c32b3cbe MH |
719 | /** |
720 | * oom_killer_disable - disable OOM killer | |
7d2e7a22 | 721 | * @timeout: maximum timeout to wait for oom victims in jiffies |
c32b3cbe MH |
722 | * |
723 | * Forces all page allocations to fail rather than trigger OOM killer. | |
7d2e7a22 MH |
724 | * Will block and wait until all OOM victims are killed or the given |
725 | * timeout expires. | |
c32b3cbe MH |
726 | * |
727 | * The function cannot be called when there are runnable user tasks because | |
728 | * the userspace would see unexpected allocation failures as a result. Any | |
729 | * new usage of this function should be consulted with MM people. | |
730 | * | |
731 | * Returns true if successful and false if the OOM killer cannot be | |
732 | * disabled. | |
733 | */ | |
7d2e7a22 | 734 | bool oom_killer_disable(signed long timeout) |
c32b3cbe | 735 | { |
7d2e7a22 MH |
736 | signed long ret; |
737 | ||
c32b3cbe | 738 | /* |
6afcf289 TH |
739 | * Make sure to not race with an ongoing OOM killer. Check that the |
740 | * current is not killed (possibly due to sharing the victim's memory). | |
c32b3cbe | 741 | */ |
6afcf289 | 742 | if (mutex_lock_killable(&oom_lock)) |
c32b3cbe | 743 | return false; |
c32b3cbe | 744 | oom_killer_disabled = true; |
dc56401f | 745 | mutex_unlock(&oom_lock); |
c32b3cbe | 746 | |
7d2e7a22 MH |
747 | ret = wait_event_interruptible_timeout(oom_victims_wait, |
748 | !atomic_read(&oom_victims), timeout); | |
749 | if (ret <= 0) { | |
750 | oom_killer_enable(); | |
751 | return false; | |
752 | } | |
d75da004 | 753 | pr_info("OOM killer disabled.\n"); |
c32b3cbe MH |
754 | |
755 | return true; | |
756 | } | |
757 | ||
1af8bb43 MH |
758 | static inline bool __task_will_free_mem(struct task_struct *task) |
759 | { | |
760 | struct signal_struct *sig = task->signal; | |
761 | ||
762 | /* | |
763 | * A coredumping process may sleep for an extended period in exit_mm(), | |
764 | * so the oom killer cannot assume that the process will promptly exit | |
765 | * and release memory. | |
766 | */ | |
767 | if (sig->flags & SIGNAL_GROUP_COREDUMP) | |
768 | return false; | |
769 | ||
770 | if (sig->flags & SIGNAL_GROUP_EXIT) | |
771 | return true; | |
772 | ||
773 | if (thread_group_empty(task) && (task->flags & PF_EXITING)) | |
774 | return true; | |
775 | ||
776 | return false; | |
777 | } | |
778 | ||
779 | /* | |
780 | * Checks whether the given task is dying or exiting and likely to | |
781 | * release its address space. This means that all threads and processes | |
782 | * sharing the same mm have to be killed or exiting. | |
091f362c MH |
783 | * Caller has to make sure that task->mm is stable (hold task_lock or |
784 | * it operates on the current). | |
1af8bb43 | 785 | */ |
7c5f64f8 | 786 | static bool task_will_free_mem(struct task_struct *task) |
1af8bb43 | 787 | { |
091f362c | 788 | struct mm_struct *mm = task->mm; |
1af8bb43 | 789 | struct task_struct *p; |
f33e6f06 | 790 | bool ret = true; |
1af8bb43 | 791 | |
1af8bb43 | 792 | /* |
091f362c MH |
793 | * Skip tasks without mm because it might have passed its exit_mm and |
794 | * exit_oom_victim. oom_reaper could have rescued that but do not rely | |
795 | * on that for now. We can consider find_lock_task_mm in future. | |
1af8bb43 | 796 | */ |
091f362c | 797 | if (!mm) |
1af8bb43 MH |
798 | return false; |
799 | ||
091f362c MH |
800 | if (!__task_will_free_mem(task)) |
801 | return false; | |
696453e6 MH |
802 | |
803 | /* | |
804 | * This task has already been drained by the oom reaper so there are | |
805 | * only small chances it will free some more | |
806 | */ | |
862e3073 | 807 | if (test_bit(MMF_OOM_SKIP, &mm->flags)) |
696453e6 | 808 | return false; |
696453e6 | 809 | |
091f362c | 810 | if (atomic_read(&mm->mm_users) <= 1) |
1af8bb43 | 811 | return true; |
1af8bb43 MH |
812 | |
813 | /* | |
5870c2e1 MH |
814 | * Make sure that all tasks which share the mm with the given tasks |
815 | * are dying as well to make sure that a) nobody pins its mm and | |
816 | * b) the task is also reapable by the oom reaper. | |
1af8bb43 MH |
817 | */ |
818 | rcu_read_lock(); | |
819 | for_each_process(p) { | |
820 | if (!process_shares_mm(p, mm)) | |
821 | continue; | |
822 | if (same_thread_group(task, p)) | |
823 | continue; | |
824 | ret = __task_will_free_mem(p); | |
825 | if (!ret) | |
826 | break; | |
827 | } | |
828 | rcu_read_unlock(); | |
1af8bb43 MH |
829 | |
830 | return ret; | |
831 | } | |
832 | ||
5989ad7b | 833 | static void __oom_kill_process(struct task_struct *victim) |
1da177e4 | 834 | { |
5989ad7b | 835 | struct task_struct *p; |
647f2bdf | 836 | struct mm_struct *mm; |
bb29902a | 837 | bool can_oom_reap = true; |
1da177e4 | 838 | |
6b0c81b3 DR |
839 | p = find_lock_task_mm(victim); |
840 | if (!p) { | |
6b0c81b3 | 841 | put_task_struct(victim); |
647f2bdf | 842 | return; |
6b0c81b3 DR |
843 | } else if (victim != p) { |
844 | get_task_struct(p); | |
845 | put_task_struct(victim); | |
846 | victim = p; | |
847 | } | |
647f2bdf | 848 | |
880b7689 | 849 | /* Get a reference to safely compare mm after task_unlock(victim) */ |
647f2bdf | 850 | mm = victim->mm; |
f1f10076 | 851 | mmgrab(mm); |
8e675f7a KK |
852 | |
853 | /* Raise event before sending signal: task reaper must see this */ | |
854 | count_vm_event(OOM_KILL); | |
fe6bdfc8 | 855 | memcg_memory_event_mm(mm, MEMCG_OOM_KILL); |
8e675f7a | 856 | |
426fb5e7 | 857 | /* |
cd04ae1e MH |
858 | * We should send SIGKILL before granting access to memory reserves |
859 | * in order to prevent the OOM victim from depleting the memory | |
860 | * reserves from the user space under its control. | |
426fb5e7 | 861 | */ |
079b22dc | 862 | do_send_sig_info(SIGKILL, SEND_SIG_PRIV, victim, PIDTYPE_TGID); |
16e95196 | 863 | mark_oom_victim(victim); |
eca56ff9 | 864 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n", |
647f2bdf DR |
865 | task_pid_nr(victim), victim->comm, K(victim->mm->total_vm), |
866 | K(get_mm_counter(victim->mm, MM_ANONPAGES)), | |
eca56ff9 JM |
867 | K(get_mm_counter(victim->mm, MM_FILEPAGES)), |
868 | K(get_mm_counter(victim->mm, MM_SHMEMPAGES))); | |
647f2bdf DR |
869 | task_unlock(victim); |
870 | ||
871 | /* | |
872 | * Kill all user processes sharing victim->mm in other thread groups, if | |
873 | * any. They don't get access to memory reserves, though, to avoid | |
874 | * depletion of all memory. This prevents mm->mmap_sem livelock when an | |
875 | * oom killed thread cannot exit because it requires the semaphore and | |
876 | * its contended by another thread trying to allocate memory itself. | |
877 | * That thread will now get access to memory reserves since it has a | |
878 | * pending fatal signal. | |
879 | */ | |
4d4048be | 880 | rcu_read_lock(); |
c319025a | 881 | for_each_process(p) { |
4d7b3394 | 882 | if (!process_shares_mm(p, mm)) |
c319025a ON |
883 | continue; |
884 | if (same_thread_group(p, victim)) | |
885 | continue; | |
1b51e65e | 886 | if (is_global_init(p)) { |
aac45363 | 887 | can_oom_reap = false; |
862e3073 | 888 | set_bit(MMF_OOM_SKIP, &mm->flags); |
a373966d MH |
889 | pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n", |
890 | task_pid_nr(victim), victim->comm, | |
891 | task_pid_nr(p), p->comm); | |
c319025a | 892 | continue; |
aac45363 | 893 | } |
1b51e65e MH |
894 | /* |
895 | * No use_mm() user needs to read from the userspace so we are | |
896 | * ok to reap it. | |
897 | */ | |
898 | if (unlikely(p->flags & PF_KTHREAD)) | |
899 | continue; | |
079b22dc | 900 | do_send_sig_info(SIGKILL, SEND_SIG_PRIV, p, PIDTYPE_TGID); |
c319025a | 901 | } |
6b0c81b3 | 902 | rcu_read_unlock(); |
647f2bdf | 903 | |
aac45363 | 904 | if (can_oom_reap) |
36324a99 | 905 | wake_oom_reaper(victim); |
aac45363 | 906 | |
880b7689 | 907 | mmdrop(mm); |
6b0c81b3 | 908 | put_task_struct(victim); |
1da177e4 | 909 | } |
647f2bdf | 910 | #undef K |
1da177e4 | 911 | |
3d8b38eb RG |
912 | /* |
913 | * Kill provided task unless it's secured by setting | |
914 | * oom_score_adj to OOM_SCORE_ADJ_MIN. | |
915 | */ | |
916 | static int oom_kill_memcg_member(struct task_struct *task, void *unused) | |
917 | { | |
918 | if (task->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) { | |
919 | get_task_struct(task); | |
920 | __oom_kill_process(task); | |
921 | } | |
922 | return 0; | |
923 | } | |
924 | ||
5989ad7b RG |
925 | static void oom_kill_process(struct oom_control *oc, const char *message) |
926 | { | |
927 | struct task_struct *p = oc->chosen; | |
928 | unsigned int points = oc->chosen_points; | |
929 | struct task_struct *victim = p; | |
930 | struct task_struct *child; | |
931 | struct task_struct *t; | |
3d8b38eb | 932 | struct mem_cgroup *oom_group; |
5989ad7b RG |
933 | unsigned int victim_points = 0; |
934 | static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL, | |
935 | DEFAULT_RATELIMIT_BURST); | |
936 | ||
937 | /* | |
938 | * If the task is already exiting, don't alarm the sysadmin or kill | |
939 | * its children or threads, just give it access to memory reserves | |
940 | * so it can die quickly | |
941 | */ | |
942 | task_lock(p); | |
943 | if (task_will_free_mem(p)) { | |
944 | mark_oom_victim(p); | |
945 | wake_oom_reaper(p); | |
946 | task_unlock(p); | |
947 | put_task_struct(p); | |
948 | return; | |
949 | } | |
950 | task_unlock(p); | |
951 | ||
952 | if (__ratelimit(&oom_rs)) | |
953 | dump_header(oc, p); | |
954 | ||
955 | pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n", | |
956 | message, task_pid_nr(p), p->comm, points); | |
957 | ||
958 | /* | |
959 | * If any of p's children has a different mm and is eligible for kill, | |
960 | * the one with the highest oom_badness() score is sacrificed for its | |
961 | * parent. This attempts to lose the minimal amount of work done while | |
962 | * still freeing memory. | |
963 | */ | |
964 | read_lock(&tasklist_lock); | |
965 | for_each_thread(p, t) { | |
966 | list_for_each_entry(child, &t->children, sibling) { | |
967 | unsigned int child_points; | |
968 | ||
969 | if (process_shares_mm(child, p->mm)) | |
970 | continue; | |
971 | /* | |
972 | * oom_badness() returns 0 if the thread is unkillable | |
973 | */ | |
974 | child_points = oom_badness(child, | |
975 | oc->memcg, oc->nodemask, oc->totalpages); | |
976 | if (child_points > victim_points) { | |
977 | put_task_struct(victim); | |
978 | victim = child; | |
979 | victim_points = child_points; | |
980 | get_task_struct(victim); | |
981 | } | |
982 | } | |
983 | } | |
984 | read_unlock(&tasklist_lock); | |
985 | ||
3d8b38eb RG |
986 | /* |
987 | * Do we need to kill the entire memory cgroup? | |
988 | * Or even one of the ancestor memory cgroups? | |
989 | * Check this out before killing the victim task. | |
990 | */ | |
991 | oom_group = mem_cgroup_get_oom_group(victim, oc->memcg); | |
992 | ||
5989ad7b | 993 | __oom_kill_process(victim); |
3d8b38eb RG |
994 | |
995 | /* | |
996 | * If necessary, kill all tasks in the selected memory cgroup. | |
997 | */ | |
998 | if (oom_group) { | |
999 | mem_cgroup_print_oom_group(oom_group); | |
1000 | mem_cgroup_scan_tasks(oom_group, oom_kill_memcg_member, NULL); | |
1001 | mem_cgroup_put(oom_group); | |
1002 | } | |
5989ad7b RG |
1003 | } |
1004 | ||
309ed882 DR |
1005 | /* |
1006 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. | |
1007 | */ | |
7c5f64f8 VD |
1008 | static void check_panic_on_oom(struct oom_control *oc, |
1009 | enum oom_constraint constraint) | |
309ed882 DR |
1010 | { |
1011 | if (likely(!sysctl_panic_on_oom)) | |
1012 | return; | |
1013 | if (sysctl_panic_on_oom != 2) { | |
1014 | /* | |
1015 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel | |
1016 | * does not panic for cpuset, mempolicy, or memcg allocation | |
1017 | * failures. | |
1018 | */ | |
1019 | if (constraint != CONSTRAINT_NONE) | |
1020 | return; | |
1021 | } | |
071a4bef | 1022 | /* Do not panic for oom kills triggered by sysrq */ |
db2a0dd7 | 1023 | if (is_sysrq_oom(oc)) |
071a4bef | 1024 | return; |
2a966b77 | 1025 | dump_header(oc, NULL); |
309ed882 DR |
1026 | panic("Out of memory: %s panic_on_oom is enabled\n", |
1027 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); | |
1028 | } | |
1029 | ||
8bc719d3 MS |
1030 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
1031 | ||
1032 | int register_oom_notifier(struct notifier_block *nb) | |
1033 | { | |
1034 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
1035 | } | |
1036 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
1037 | ||
1038 | int unregister_oom_notifier(struct notifier_block *nb) | |
1039 | { | |
1040 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
1041 | } | |
1042 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
1043 | ||
1da177e4 | 1044 | /** |
6e0fc46d DR |
1045 | * out_of_memory - kill the "best" process when we run out of memory |
1046 | * @oc: pointer to struct oom_control | |
1da177e4 LT |
1047 | * |
1048 | * If we run out of memory, we have the choice between either | |
1049 | * killing a random task (bad), letting the system crash (worse) | |
1050 | * OR try to be smart about which process to kill. Note that we | |
1051 | * don't have to be perfect here, we just have to be good. | |
1052 | */ | |
6e0fc46d | 1053 | bool out_of_memory(struct oom_control *oc) |
1da177e4 | 1054 | { |
8bc719d3 | 1055 | unsigned long freed = 0; |
e3658932 | 1056 | enum oom_constraint constraint = CONSTRAINT_NONE; |
8bc719d3 | 1057 | |
dc56401f JW |
1058 | if (oom_killer_disabled) |
1059 | return false; | |
1060 | ||
7c5f64f8 VD |
1061 | if (!is_memcg_oom(oc)) { |
1062 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
1063 | if (freed > 0) | |
1064 | /* Got some memory back in the last second. */ | |
1065 | return true; | |
1066 | } | |
1da177e4 | 1067 | |
7b98c2e4 | 1068 | /* |
9ff4868e DR |
1069 | * If current has a pending SIGKILL or is exiting, then automatically |
1070 | * select it. The goal is to allow it to allocate so that it may | |
1071 | * quickly exit and free its memory. | |
7b98c2e4 | 1072 | */ |
091f362c | 1073 | if (task_will_free_mem(current)) { |
16e95196 | 1074 | mark_oom_victim(current); |
1af8bb43 | 1075 | wake_oom_reaper(current); |
75e8f8b2 | 1076 | return true; |
7b98c2e4 DR |
1077 | } |
1078 | ||
3da88fb3 MH |
1079 | /* |
1080 | * The OOM killer does not compensate for IO-less reclaim. | |
1081 | * pagefault_out_of_memory lost its gfp context so we have to | |
1082 | * make sure exclude 0 mask - all other users should have at least | |
1083 | * ___GFP_DIRECT_RECLAIM to get here. | |
1084 | */ | |
06ad276a | 1085 | if (oc->gfp_mask && !(oc->gfp_mask & __GFP_FS)) |
3da88fb3 MH |
1086 | return true; |
1087 | ||
9b0f8b04 CL |
1088 | /* |
1089 | * Check if there were limitations on the allocation (only relevant for | |
7c5f64f8 | 1090 | * NUMA and memcg) that may require different handling. |
9b0f8b04 | 1091 | */ |
7c5f64f8 | 1092 | constraint = constrained_alloc(oc); |
6e0fc46d DR |
1093 | if (constraint != CONSTRAINT_MEMORY_POLICY) |
1094 | oc->nodemask = NULL; | |
2a966b77 | 1095 | check_panic_on_oom(oc, constraint); |
0aad4b31 | 1096 | |
7c5f64f8 VD |
1097 | if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task && |
1098 | current->mm && !oom_unkillable_task(current, NULL, oc->nodemask) && | |
121d1ba0 | 1099 | current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) { |
6b0c81b3 | 1100 | get_task_struct(current); |
7c5f64f8 VD |
1101 | oc->chosen = current; |
1102 | oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)"); | |
75e8f8b2 | 1103 | return true; |
0aad4b31 DR |
1104 | } |
1105 | ||
7c5f64f8 | 1106 | select_bad_process(oc); |
3100dab2 JW |
1107 | /* Found nothing?!?! */ |
1108 | if (!oc->chosen) { | |
2a966b77 | 1109 | dump_header(oc, NULL); |
3100dab2 JW |
1110 | pr_warn("Out of memory and no killable processes...\n"); |
1111 | /* | |
1112 | * If we got here due to an actual allocation at the | |
1113 | * system level, we cannot survive this and will enter | |
1114 | * an endless loop in the allocator. Bail out now. | |
1115 | */ | |
1116 | if (!is_sysrq_oom(oc) && !is_memcg_oom(oc)) | |
1117 | panic("System is deadlocked on memory\n"); | |
0aad4b31 | 1118 | } |
9bfe5ded | 1119 | if (oc->chosen && oc->chosen != (void *)-1UL) |
7c5f64f8 VD |
1120 | oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" : |
1121 | "Memory cgroup out of memory"); | |
7c5f64f8 | 1122 | return !!oc->chosen; |
c32b3cbe MH |
1123 | } |
1124 | ||
e3658932 DR |
1125 | /* |
1126 | * The pagefault handler calls here because it is out of memory, so kill a | |
798fd756 VD |
1127 | * memory-hogging task. If oom_lock is held by somebody else, a parallel oom |
1128 | * killing is already in progress so do nothing. | |
e3658932 DR |
1129 | */ |
1130 | void pagefault_out_of_memory(void) | |
1131 | { | |
6e0fc46d DR |
1132 | struct oom_control oc = { |
1133 | .zonelist = NULL, | |
1134 | .nodemask = NULL, | |
2a966b77 | 1135 | .memcg = NULL, |
6e0fc46d DR |
1136 | .gfp_mask = 0, |
1137 | .order = 0, | |
6e0fc46d DR |
1138 | }; |
1139 | ||
49426420 | 1140 | if (mem_cgroup_oom_synchronize(true)) |
dc56401f | 1141 | return; |
3812c8c8 | 1142 | |
dc56401f JW |
1143 | if (!mutex_trylock(&oom_lock)) |
1144 | return; | |
a104808e | 1145 | out_of_memory(&oc); |
dc56401f | 1146 | mutex_unlock(&oom_lock); |
e3658932 | 1147 | } |