Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/mm/oom_kill.c | |
4 | * | |
5 | * Copyright (C) 1998,2000 Rik van Riel | |
6 | * Thanks go out to Claus Fischer for some serious inspiration and | |
7 | * for goading me into coding this file... | |
a63d83f4 DR |
8 | * Copyright (C) 2010 Google, Inc. |
9 | * Rewritten by David Rientjes | |
1da177e4 LT |
10 | * |
11 | * The routines in this file are used to kill a process when | |
a49335cc PJ |
12 | * we're seriously out of memory. This gets called from __alloc_pages() |
13 | * in mm/page_alloc.c when we really run out of memory. | |
1da177e4 LT |
14 | * |
15 | * Since we won't call these routines often (on a well-configured | |
16 | * machine) this file will double as a 'coding guide' and a signpost | |
17 | * for newbie kernel hackers. It features several pointers to major | |
18 | * kernel subsystems and hints as to where to find out what things do. | |
19 | */ | |
20 | ||
8ac773b4 | 21 | #include <linux/oom.h> |
1da177e4 | 22 | #include <linux/mm.h> |
4e950f6f | 23 | #include <linux/err.h> |
5a0e3ad6 | 24 | #include <linux/gfp.h> |
1da177e4 | 25 | #include <linux/sched.h> |
6e84f315 | 26 | #include <linux/sched/mm.h> |
f7ccbae4 | 27 | #include <linux/sched/coredump.h> |
29930025 | 28 | #include <linux/sched/task.h> |
1da177e4 LT |
29 | #include <linux/swap.h> |
30 | #include <linux/timex.h> | |
31 | #include <linux/jiffies.h> | |
ef08e3b4 | 32 | #include <linux/cpuset.h> |
b95f1b31 | 33 | #include <linux/export.h> |
8bc719d3 | 34 | #include <linux/notifier.h> |
c7ba5c9e | 35 | #include <linux/memcontrol.h> |
6f48d0eb | 36 | #include <linux/mempolicy.h> |
5cd9c58f | 37 | #include <linux/security.h> |
edd45544 | 38 | #include <linux/ptrace.h> |
f660daac | 39 | #include <linux/freezer.h> |
43d2b113 | 40 | #include <linux/ftrace.h> |
dc3f21ea | 41 | #include <linux/ratelimit.h> |
aac45363 MH |
42 | #include <linux/kthread.h> |
43 | #include <linux/init.h> | |
4d4bbd85 | 44 | #include <linux/mmu_notifier.h> |
aac45363 MH |
45 | |
46 | #include <asm/tlb.h> | |
47 | #include "internal.h" | |
852d8be0 | 48 | #include "slab.h" |
43d2b113 KH |
49 | |
50 | #define CREATE_TRACE_POINTS | |
51 | #include <trace/events/oom.h> | |
1da177e4 | 52 | |
fadd8fbd | 53 | int sysctl_panic_on_oom; |
fe071d7e | 54 | int sysctl_oom_kill_allocating_task; |
ad915c43 | 55 | int sysctl_oom_dump_tasks = 1; |
dc56401f | 56 | |
a195d3f5 MH |
57 | /* |
58 | * Serializes oom killer invocations (out_of_memory()) from all contexts to | |
59 | * prevent from over eager oom killing (e.g. when the oom killer is invoked | |
60 | * from different domains). | |
61 | * | |
62 | * oom_killer_disable() relies on this lock to stabilize oom_killer_disabled | |
63 | * and mark_oom_victim | |
64 | */ | |
dc56401f | 65 | DEFINE_MUTEX(oom_lock); |
1da177e4 | 66 | |
6f48d0eb DR |
67 | #ifdef CONFIG_NUMA |
68 | /** | |
69 | * has_intersects_mems_allowed() - check task eligiblity for kill | |
ad962441 | 70 | * @start: task struct of which task to consider |
6f48d0eb DR |
71 | * @mask: nodemask passed to page allocator for mempolicy ooms |
72 | * | |
73 | * Task eligibility is determined by whether or not a candidate task, @tsk, | |
74 | * shares the same mempolicy nodes as current if it is bound by such a policy | |
75 | * and whether or not it has the same set of allowed cpuset nodes. | |
495789a5 | 76 | */ |
ad962441 | 77 | static bool has_intersects_mems_allowed(struct task_struct *start, |
6f48d0eb | 78 | const nodemask_t *mask) |
495789a5 | 79 | { |
ad962441 ON |
80 | struct task_struct *tsk; |
81 | bool ret = false; | |
495789a5 | 82 | |
ad962441 | 83 | rcu_read_lock(); |
1da4db0c | 84 | for_each_thread(start, tsk) { |
6f48d0eb DR |
85 | if (mask) { |
86 | /* | |
87 | * If this is a mempolicy constrained oom, tsk's | |
88 | * cpuset is irrelevant. Only return true if its | |
89 | * mempolicy intersects current, otherwise it may be | |
90 | * needlessly killed. | |
91 | */ | |
ad962441 | 92 | ret = mempolicy_nodemask_intersects(tsk, mask); |
6f48d0eb DR |
93 | } else { |
94 | /* | |
95 | * This is not a mempolicy constrained oom, so only | |
96 | * check the mems of tsk's cpuset. | |
97 | */ | |
ad962441 | 98 | ret = cpuset_mems_allowed_intersects(current, tsk); |
6f48d0eb | 99 | } |
ad962441 ON |
100 | if (ret) |
101 | break; | |
1da4db0c | 102 | } |
ad962441 | 103 | rcu_read_unlock(); |
df1090a8 | 104 | |
ad962441 | 105 | return ret; |
6f48d0eb DR |
106 | } |
107 | #else | |
108 | static bool has_intersects_mems_allowed(struct task_struct *tsk, | |
109 | const nodemask_t *mask) | |
110 | { | |
111 | return true; | |
495789a5 | 112 | } |
6f48d0eb | 113 | #endif /* CONFIG_NUMA */ |
495789a5 | 114 | |
6f48d0eb DR |
115 | /* |
116 | * The process p may have detached its own ->mm while exiting or through | |
117 | * use_mm(), but one or more of its subthreads may still have a valid | |
118 | * pointer. Return p, or any of its subthreads with a valid ->mm, with | |
119 | * task_lock() held. | |
120 | */ | |
158e0a2d | 121 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
dd8e8f40 | 122 | { |
1da4db0c | 123 | struct task_struct *t; |
dd8e8f40 | 124 | |
4d4048be ON |
125 | rcu_read_lock(); |
126 | ||
1da4db0c | 127 | for_each_thread(p, t) { |
dd8e8f40 ON |
128 | task_lock(t); |
129 | if (likely(t->mm)) | |
4d4048be | 130 | goto found; |
dd8e8f40 | 131 | task_unlock(t); |
1da4db0c | 132 | } |
4d4048be ON |
133 | t = NULL; |
134 | found: | |
135 | rcu_read_unlock(); | |
dd8e8f40 | 136 | |
4d4048be | 137 | return t; |
dd8e8f40 ON |
138 | } |
139 | ||
db2a0dd7 YB |
140 | /* |
141 | * order == -1 means the oom kill is required by sysrq, otherwise only | |
142 | * for display purposes. | |
143 | */ | |
144 | static inline bool is_sysrq_oom(struct oom_control *oc) | |
145 | { | |
146 | return oc->order == -1; | |
147 | } | |
148 | ||
7c5f64f8 VD |
149 | static inline bool is_memcg_oom(struct oom_control *oc) |
150 | { | |
151 | return oc->memcg != NULL; | |
152 | } | |
153 | ||
ab290adb | 154 | /* return true if the task is not adequate as candidate victim task. */ |
e85bfd3a | 155 | static bool oom_unkillable_task(struct task_struct *p, |
2314b42d | 156 | struct mem_cgroup *memcg, const nodemask_t *nodemask) |
ab290adb KM |
157 | { |
158 | if (is_global_init(p)) | |
159 | return true; | |
160 | if (p->flags & PF_KTHREAD) | |
161 | return true; | |
162 | ||
163 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ | |
72835c86 | 164 | if (memcg && !task_in_mem_cgroup(p, memcg)) |
ab290adb KM |
165 | return true; |
166 | ||
167 | /* p may not have freeable memory in nodemask */ | |
168 | if (!has_intersects_mems_allowed(p, nodemask)) | |
169 | return true; | |
170 | ||
171 | return false; | |
172 | } | |
173 | ||
852d8be0 YS |
174 | /* |
175 | * Print out unreclaimble slabs info when unreclaimable slabs amount is greater | |
176 | * than all user memory (LRU pages) | |
177 | */ | |
178 | static bool is_dump_unreclaim_slabs(void) | |
179 | { | |
180 | unsigned long nr_lru; | |
181 | ||
182 | nr_lru = global_node_page_state(NR_ACTIVE_ANON) + | |
183 | global_node_page_state(NR_INACTIVE_ANON) + | |
184 | global_node_page_state(NR_ACTIVE_FILE) + | |
185 | global_node_page_state(NR_INACTIVE_FILE) + | |
186 | global_node_page_state(NR_ISOLATED_ANON) + | |
187 | global_node_page_state(NR_ISOLATED_FILE) + | |
188 | global_node_page_state(NR_UNEVICTABLE); | |
189 | ||
190 | return (global_node_page_state(NR_SLAB_UNRECLAIMABLE) > nr_lru); | |
191 | } | |
192 | ||
1da177e4 | 193 | /** |
a63d83f4 | 194 | * oom_badness - heuristic function to determine which candidate task to kill |
1da177e4 | 195 | * @p: task struct of which task we should calculate |
a63d83f4 | 196 | * @totalpages: total present RAM allowed for page allocation |
e8b098fc MR |
197 | * @memcg: task's memory controller, if constrained |
198 | * @nodemask: nodemask passed to page allocator for mempolicy ooms | |
1da177e4 | 199 | * |
a63d83f4 DR |
200 | * The heuristic for determining which task to kill is made to be as simple and |
201 | * predictable as possible. The goal is to return the highest value for the | |
202 | * task consuming the most memory to avoid subsequent oom failures. | |
1da177e4 | 203 | */ |
a7f638f9 DR |
204 | unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg, |
205 | const nodemask_t *nodemask, unsigned long totalpages) | |
1da177e4 | 206 | { |
1e11ad8d | 207 | long points; |
61eafb00 | 208 | long adj; |
28b83c51 | 209 | |
72835c86 | 210 | if (oom_unkillable_task(p, memcg, nodemask)) |
26ebc984 | 211 | return 0; |
1da177e4 | 212 | |
dd8e8f40 ON |
213 | p = find_lock_task_mm(p); |
214 | if (!p) | |
1da177e4 LT |
215 | return 0; |
216 | ||
bb8a4b7f MH |
217 | /* |
218 | * Do not even consider tasks which are explicitly marked oom | |
b18dc5f2 MH |
219 | * unkillable or have been already oom reaped or the are in |
220 | * the middle of vfork | |
bb8a4b7f | 221 | */ |
a9c58b90 | 222 | adj = (long)p->signal->oom_score_adj; |
bb8a4b7f | 223 | if (adj == OOM_SCORE_ADJ_MIN || |
862e3073 | 224 | test_bit(MMF_OOM_SKIP, &p->mm->flags) || |
b18dc5f2 | 225 | in_vfork(p)) { |
5aecc85a MH |
226 | task_unlock(p); |
227 | return 0; | |
228 | } | |
229 | ||
1da177e4 | 230 | /* |
a63d83f4 | 231 | * The baseline for the badness score is the proportion of RAM that each |
f755a042 | 232 | * task's rss, pagetable and swap space use. |
1da177e4 | 233 | */ |
dc6c9a35 | 234 | points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) + |
af5b0f6a | 235 | mm_pgtables_bytes(p->mm) / PAGE_SIZE; |
a63d83f4 | 236 | task_unlock(p); |
1da177e4 | 237 | |
61eafb00 DR |
238 | /* Normalize to oom_score_adj units */ |
239 | adj *= totalpages / 1000; | |
240 | points += adj; | |
1da177e4 | 241 | |
f19e8aa1 | 242 | /* |
a7f638f9 DR |
243 | * Never return 0 for an eligible task regardless of the root bonus and |
244 | * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here). | |
f19e8aa1 | 245 | */ |
1e11ad8d | 246 | return points > 0 ? points : 1; |
1da177e4 LT |
247 | } |
248 | ||
ef8444ea | 249 | static const char * const oom_constraint_text[] = { |
250 | [CONSTRAINT_NONE] = "CONSTRAINT_NONE", | |
251 | [CONSTRAINT_CPUSET] = "CONSTRAINT_CPUSET", | |
252 | [CONSTRAINT_MEMORY_POLICY] = "CONSTRAINT_MEMORY_POLICY", | |
253 | [CONSTRAINT_MEMCG] = "CONSTRAINT_MEMCG", | |
7c5f64f8 VD |
254 | }; |
255 | ||
9b0f8b04 CL |
256 | /* |
257 | * Determine the type of allocation constraint. | |
258 | */ | |
7c5f64f8 | 259 | static enum oom_constraint constrained_alloc(struct oom_control *oc) |
4365a567 | 260 | { |
54a6eb5c | 261 | struct zone *zone; |
dd1a239f | 262 | struct zoneref *z; |
6e0fc46d | 263 | enum zone_type high_zoneidx = gfp_zone(oc->gfp_mask); |
a63d83f4 DR |
264 | bool cpuset_limited = false; |
265 | int nid; | |
9b0f8b04 | 266 | |
7c5f64f8 | 267 | if (is_memcg_oom(oc)) { |
bbec2e15 | 268 | oc->totalpages = mem_cgroup_get_max(oc->memcg) ?: 1; |
7c5f64f8 VD |
269 | return CONSTRAINT_MEMCG; |
270 | } | |
271 | ||
a63d83f4 | 272 | /* Default to all available memory */ |
ca79b0c2 | 273 | oc->totalpages = totalram_pages() + total_swap_pages; |
7c5f64f8 VD |
274 | |
275 | if (!IS_ENABLED(CONFIG_NUMA)) | |
276 | return CONSTRAINT_NONE; | |
a63d83f4 | 277 | |
6e0fc46d | 278 | if (!oc->zonelist) |
a63d83f4 | 279 | return CONSTRAINT_NONE; |
4365a567 KH |
280 | /* |
281 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid | |
282 | * to kill current.We have to random task kill in this case. | |
283 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. | |
284 | */ | |
6e0fc46d | 285 | if (oc->gfp_mask & __GFP_THISNODE) |
4365a567 | 286 | return CONSTRAINT_NONE; |
9b0f8b04 | 287 | |
4365a567 | 288 | /* |
a63d83f4 DR |
289 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
290 | * the page allocator means a mempolicy is in effect. Cpuset policy | |
291 | * is enforced in get_page_from_freelist(). | |
4365a567 | 292 | */ |
6e0fc46d DR |
293 | if (oc->nodemask && |
294 | !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) { | |
7c5f64f8 | 295 | oc->totalpages = total_swap_pages; |
6e0fc46d | 296 | for_each_node_mask(nid, *oc->nodemask) |
7c5f64f8 | 297 | oc->totalpages += node_spanned_pages(nid); |
9b0f8b04 | 298 | return CONSTRAINT_MEMORY_POLICY; |
a63d83f4 | 299 | } |
4365a567 KH |
300 | |
301 | /* Check this allocation failure is caused by cpuset's wall function */ | |
6e0fc46d DR |
302 | for_each_zone_zonelist_nodemask(zone, z, oc->zonelist, |
303 | high_zoneidx, oc->nodemask) | |
304 | if (!cpuset_zone_allowed(zone, oc->gfp_mask)) | |
a63d83f4 | 305 | cpuset_limited = true; |
9b0f8b04 | 306 | |
a63d83f4 | 307 | if (cpuset_limited) { |
7c5f64f8 | 308 | oc->totalpages = total_swap_pages; |
a63d83f4 | 309 | for_each_node_mask(nid, cpuset_current_mems_allowed) |
7c5f64f8 | 310 | oc->totalpages += node_spanned_pages(nid); |
a63d83f4 DR |
311 | return CONSTRAINT_CPUSET; |
312 | } | |
9b0f8b04 CL |
313 | return CONSTRAINT_NONE; |
314 | } | |
315 | ||
7c5f64f8 | 316 | static int oom_evaluate_task(struct task_struct *task, void *arg) |
462607ec | 317 | { |
7c5f64f8 VD |
318 | struct oom_control *oc = arg; |
319 | unsigned long points; | |
320 | ||
6e0fc46d | 321 | if (oom_unkillable_task(task, NULL, oc->nodemask)) |
7c5f64f8 | 322 | goto next; |
462607ec DR |
323 | |
324 | /* | |
325 | * This task already has access to memory reserves and is being killed. | |
a373966d | 326 | * Don't allow any other task to have access to the reserves unless |
862e3073 | 327 | * the task has MMF_OOM_SKIP because chances that it would release |
a373966d | 328 | * any memory is quite low. |
462607ec | 329 | */ |
862e3073 MH |
330 | if (!is_sysrq_oom(oc) && tsk_is_oom_victim(task)) { |
331 | if (test_bit(MMF_OOM_SKIP, &task->signal->oom_mm->flags)) | |
7c5f64f8 VD |
332 | goto next; |
333 | goto abort; | |
a373966d | 334 | } |
462607ec | 335 | |
e1e12d2f DR |
336 | /* |
337 | * If task is allocating a lot of memory and has been marked to be | |
338 | * killed first if it triggers an oom, then select it. | |
339 | */ | |
7c5f64f8 VD |
340 | if (oom_task_origin(task)) { |
341 | points = ULONG_MAX; | |
342 | goto select; | |
343 | } | |
e1e12d2f | 344 | |
7c5f64f8 VD |
345 | points = oom_badness(task, NULL, oc->nodemask, oc->totalpages); |
346 | if (!points || points < oc->chosen_points) | |
347 | goto next; | |
348 | ||
349 | /* Prefer thread group leaders for display purposes */ | |
350 | if (points == oc->chosen_points && thread_group_leader(oc->chosen)) | |
351 | goto next; | |
352 | select: | |
353 | if (oc->chosen) | |
354 | put_task_struct(oc->chosen); | |
355 | get_task_struct(task); | |
356 | oc->chosen = task; | |
357 | oc->chosen_points = points; | |
358 | next: | |
359 | return 0; | |
360 | abort: | |
361 | if (oc->chosen) | |
362 | put_task_struct(oc->chosen); | |
363 | oc->chosen = (void *)-1UL; | |
364 | return 1; | |
462607ec DR |
365 | } |
366 | ||
1da177e4 | 367 | /* |
7c5f64f8 VD |
368 | * Simple selection loop. We choose the process with the highest number of |
369 | * 'points'. In case scan was aborted, oc->chosen is set to -1. | |
1da177e4 | 370 | */ |
7c5f64f8 | 371 | static void select_bad_process(struct oom_control *oc) |
1da177e4 | 372 | { |
7c5f64f8 VD |
373 | if (is_memcg_oom(oc)) |
374 | mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc); | |
375 | else { | |
376 | struct task_struct *p; | |
d49ad935 | 377 | |
7c5f64f8 VD |
378 | rcu_read_lock(); |
379 | for_each_process(p) | |
380 | if (oom_evaluate_task(p, oc)) | |
381 | break; | |
382 | rcu_read_unlock(); | |
1da4db0c | 383 | } |
972c4ea5 | 384 | |
7c5f64f8 | 385 | oc->chosen_points = oc->chosen_points * 1000 / oc->totalpages; |
1da177e4 LT |
386 | } |
387 | ||
fef1bdd6 | 388 | /** |
1b578df0 | 389 | * dump_tasks - dump current memory state of all system tasks |
dad7557e | 390 | * @memcg: current's memory controller, if constrained |
e85bfd3a | 391 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
1b578df0 | 392 | * |
e85bfd3a DR |
393 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
394 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes | |
395 | * are not shown. | |
af5b0f6a KS |
396 | * State information includes task's pid, uid, tgid, vm size, rss, |
397 | * pgtables_bytes, swapents, oom_score_adj value, and name. | |
fef1bdd6 | 398 | */ |
2314b42d | 399 | static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask) |
fef1bdd6 | 400 | { |
c55db957 KM |
401 | struct task_struct *p; |
402 | struct task_struct *task; | |
fef1bdd6 | 403 | |
c3b78b11 RF |
404 | pr_info("Tasks state (memory values in pages):\n"); |
405 | pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name\n"); | |
6b0c81b3 | 406 | rcu_read_lock(); |
c55db957 | 407 | for_each_process(p) { |
72835c86 | 408 | if (oom_unkillable_task(p, memcg, nodemask)) |
b4416d2b | 409 | continue; |
fef1bdd6 | 410 | |
c55db957 KM |
411 | task = find_lock_task_mm(p); |
412 | if (!task) { | |
6d2661ed | 413 | /* |
74ab7f1d DR |
414 | * This is a kthread or all of p's threads have already |
415 | * detached their mm's. There's no need to report | |
c55db957 | 416 | * them; they can't be oom killed anyway. |
6d2661ed | 417 | */ |
6d2661ed DR |
418 | continue; |
419 | } | |
c55db957 | 420 | |
c3b78b11 | 421 | pr_info("[%7d] %5d %5d %8lu %8lu %8ld %8lu %5hd %s\n", |
078de5f7 EB |
422 | task->pid, from_kuid(&init_user_ns, task_uid(task)), |
423 | task->tgid, task->mm->total_vm, get_mm_rss(task->mm), | |
af5b0f6a | 424 | mm_pgtables_bytes(task->mm), |
de34d965 | 425 | get_mm_counter(task->mm, MM_SWAPENTS), |
a63d83f4 | 426 | task->signal->oom_score_adj, task->comm); |
c55db957 KM |
427 | task_unlock(task); |
428 | } | |
6b0c81b3 | 429 | rcu_read_unlock(); |
fef1bdd6 DR |
430 | } |
431 | ||
ef8444ea | 432 | static void dump_oom_summary(struct oom_control *oc, struct task_struct *victim) |
433 | { | |
434 | /* one line summary of the oom killer context. */ | |
435 | pr_info("oom-kill:constraint=%s,nodemask=%*pbl", | |
436 | oom_constraint_text[oc->constraint], | |
437 | nodemask_pr_args(oc->nodemask)); | |
438 | cpuset_print_current_mems_allowed(); | |
f0c867d9 | 439 | mem_cgroup_print_oom_context(oc->memcg, victim); |
ef8444ea | 440 | pr_cont(",task=%s,pid=%d,uid=%d\n", victim->comm, victim->pid, |
441 | from_kuid(&init_user_ns, task_uid(victim))); | |
442 | } | |
443 | ||
2a966b77 | 444 | static void dump_header(struct oom_control *oc, struct task_struct *p) |
1b604d75 | 445 | { |
ef8444ea | 446 | pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), order=%d, oom_score_adj=%hd\n", |
447 | current->comm, oc->gfp_mask, &oc->gfp_mask, oc->order, | |
0205f755 | 448 | current->signal->oom_score_adj); |
9254990f MH |
449 | if (!IS_ENABLED(CONFIG_COMPACTION) && oc->order) |
450 | pr_warn("COMPACTION is disabled!!!\n"); | |
a0795cd4 | 451 | |
1b604d75 | 452 | dump_stack(); |
852d8be0 | 453 | if (is_memcg_oom(oc)) |
f0c867d9 | 454 | mem_cgroup_print_oom_meminfo(oc->memcg); |
852d8be0 | 455 | else { |
299c517a | 456 | show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask); |
852d8be0 YS |
457 | if (is_dump_unreclaim_slabs()) |
458 | dump_unreclaimable_slab(); | |
459 | } | |
1b604d75 | 460 | if (sysctl_oom_dump_tasks) |
2a966b77 | 461 | dump_tasks(oc->memcg, oc->nodemask); |
ef8444ea | 462 | if (p) |
463 | dump_oom_summary(oc, p); | |
1b604d75 DR |
464 | } |
465 | ||
5695be14 | 466 | /* |
c32b3cbe | 467 | * Number of OOM victims in flight |
5695be14 | 468 | */ |
c32b3cbe MH |
469 | static atomic_t oom_victims = ATOMIC_INIT(0); |
470 | static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait); | |
5695be14 | 471 | |
7c5f64f8 | 472 | static bool oom_killer_disabled __read_mostly; |
5695be14 | 473 | |
bc448e89 MH |
474 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
475 | ||
3ef22dff MH |
476 | /* |
477 | * task->mm can be NULL if the task is the exited group leader. So to | |
478 | * determine whether the task is using a particular mm, we examine all the | |
479 | * task's threads: if one of those is using this mm then this task was also | |
480 | * using it. | |
481 | */ | |
44a70ade | 482 | bool process_shares_mm(struct task_struct *p, struct mm_struct *mm) |
3ef22dff MH |
483 | { |
484 | struct task_struct *t; | |
485 | ||
486 | for_each_thread(p, t) { | |
487 | struct mm_struct *t_mm = READ_ONCE(t->mm); | |
488 | if (t_mm) | |
489 | return t_mm == mm; | |
490 | } | |
491 | return false; | |
492 | } | |
493 | ||
aac45363 MH |
494 | #ifdef CONFIG_MMU |
495 | /* | |
496 | * OOM Reaper kernel thread which tries to reap the memory used by the OOM | |
497 | * victim (if that is possible) to help the OOM killer to move on. | |
498 | */ | |
499 | static struct task_struct *oom_reaper_th; | |
aac45363 | 500 | static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait); |
29c696e1 | 501 | static struct task_struct *oom_reaper_list; |
03049269 MH |
502 | static DEFINE_SPINLOCK(oom_reaper_lock); |
503 | ||
93065ac7 | 504 | bool __oom_reap_task_mm(struct mm_struct *mm) |
aac45363 | 505 | { |
aac45363 | 506 | struct vm_area_struct *vma; |
93065ac7 | 507 | bool ret = true; |
27ae357f DR |
508 | |
509 | /* | |
510 | * Tell all users of get_user/copy_from_user etc... that the content | |
511 | * is no longer stable. No barriers really needed because unmapping | |
512 | * should imply barriers already and the reader would hit a page fault | |
513 | * if it stumbled over a reaped memory. | |
514 | */ | |
515 | set_bit(MMF_UNSTABLE, &mm->flags); | |
516 | ||
517 | for (vma = mm->mmap ; vma; vma = vma->vm_next) { | |
518 | if (!can_madv_dontneed_vma(vma)) | |
519 | continue; | |
520 | ||
521 | /* | |
522 | * Only anonymous pages have a good chance to be dropped | |
523 | * without additional steps which we cannot afford as we | |
524 | * are OOM already. | |
525 | * | |
526 | * We do not even care about fs backed pages because all | |
527 | * which are reclaimable have already been reclaimed and | |
528 | * we do not want to block exit_mmap by keeping mm ref | |
529 | * count elevated without a good reason. | |
530 | */ | |
531 | if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) { | |
ac46d4f3 | 532 | struct mmu_notifier_range range; |
27ae357f DR |
533 | struct mmu_gather tlb; |
534 | ||
6f4f13e8 JG |
535 | mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, |
536 | vma, mm, vma->vm_start, | |
ac46d4f3 JG |
537 | vma->vm_end); |
538 | tlb_gather_mmu(&tlb, mm, range.start, range.end); | |
539 | if (mmu_notifier_invalidate_range_start_nonblock(&range)) { | |
540 | tlb_finish_mmu(&tlb, range.start, range.end); | |
93065ac7 MH |
541 | ret = false; |
542 | continue; | |
543 | } | |
ac46d4f3 JG |
544 | unmap_page_range(&tlb, vma, range.start, range.end, NULL); |
545 | mmu_notifier_invalidate_range_end(&range); | |
546 | tlb_finish_mmu(&tlb, range.start, range.end); | |
27ae357f DR |
547 | } |
548 | } | |
93065ac7 MH |
549 | |
550 | return ret; | |
27ae357f DR |
551 | } |
552 | ||
431f42fd MH |
553 | /* |
554 | * Reaps the address space of the give task. | |
555 | * | |
556 | * Returns true on success and false if none or part of the address space | |
557 | * has been reclaimed and the caller should retry later. | |
558 | */ | |
27ae357f DR |
559 | static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm) |
560 | { | |
aac45363 MH |
561 | bool ret = true; |
562 | ||
aac45363 | 563 | if (!down_read_trylock(&mm->mmap_sem)) { |
422580c3 | 564 | trace_skip_task_reaping(tsk->pid); |
af5679fb | 565 | return false; |
4d4bbd85 MH |
566 | } |
567 | ||
e5e3f4c4 | 568 | /* |
21292580 AA |
569 | * MMF_OOM_SKIP is set by exit_mmap when the OOM reaper can't |
570 | * work on the mm anymore. The check for MMF_OOM_SKIP must run | |
571 | * under mmap_sem for reading because it serializes against the | |
572 | * down_write();up_write() cycle in exit_mmap(). | |
e5e3f4c4 | 573 | */ |
21292580 | 574 | if (test_bit(MMF_OOM_SKIP, &mm->flags)) { |
422580c3 | 575 | trace_skip_task_reaping(tsk->pid); |
431f42fd | 576 | goto out_unlock; |
aac45363 MH |
577 | } |
578 | ||
422580c3 RG |
579 | trace_start_task_reaping(tsk->pid); |
580 | ||
93065ac7 | 581 | /* failed to reap part of the address space. Try again later */ |
431f42fd MH |
582 | ret = __oom_reap_task_mm(mm); |
583 | if (!ret) | |
584 | goto out_finish; | |
aac45363 | 585 | |
bc448e89 MH |
586 | pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n", |
587 | task_pid_nr(tsk), tsk->comm, | |
588 | K(get_mm_counter(mm, MM_ANONPAGES)), | |
589 | K(get_mm_counter(mm, MM_FILEPAGES)), | |
590 | K(get_mm_counter(mm, MM_SHMEMPAGES))); | |
431f42fd MH |
591 | out_finish: |
592 | trace_finish_task_reaping(tsk->pid); | |
593 | out_unlock: | |
aac45363 | 594 | up_read(&mm->mmap_sem); |
36324a99 | 595 | |
aac45363 MH |
596 | return ret; |
597 | } | |
598 | ||
bc448e89 | 599 | #define MAX_OOM_REAP_RETRIES 10 |
36324a99 | 600 | static void oom_reap_task(struct task_struct *tsk) |
aac45363 MH |
601 | { |
602 | int attempts = 0; | |
26db62f1 | 603 | struct mm_struct *mm = tsk->signal->oom_mm; |
aac45363 MH |
604 | |
605 | /* Retry the down_read_trylock(mmap_sem) a few times */ | |
27ae357f | 606 | while (attempts++ < MAX_OOM_REAP_RETRIES && !oom_reap_task_mm(tsk, mm)) |
aac45363 MH |
607 | schedule_timeout_idle(HZ/10); |
608 | ||
97b1255c TH |
609 | if (attempts <= MAX_OOM_REAP_RETRIES || |
610 | test_bit(MMF_OOM_SKIP, &mm->flags)) | |
7ebffa45 | 611 | goto done; |
11a410d5 | 612 | |
7ebffa45 TH |
613 | pr_info("oom_reaper: unable to reap pid:%d (%s)\n", |
614 | task_pid_nr(tsk), tsk->comm); | |
7ebffa45 | 615 | debug_show_all_locks(); |
bc448e89 | 616 | |
7ebffa45 | 617 | done: |
449d777d | 618 | tsk->oom_reaper_list = NULL; |
449d777d | 619 | |
26db62f1 MH |
620 | /* |
621 | * Hide this mm from OOM killer because it has been either reaped or | |
622 | * somebody can't call up_write(mmap_sem). | |
623 | */ | |
862e3073 | 624 | set_bit(MMF_OOM_SKIP, &mm->flags); |
26db62f1 | 625 | |
aac45363 | 626 | /* Drop a reference taken by wake_oom_reaper */ |
36324a99 | 627 | put_task_struct(tsk); |
aac45363 MH |
628 | } |
629 | ||
630 | static int oom_reaper(void *unused) | |
631 | { | |
632 | while (true) { | |
03049269 | 633 | struct task_struct *tsk = NULL; |
aac45363 | 634 | |
29c696e1 | 635 | wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL); |
03049269 | 636 | spin_lock(&oom_reaper_lock); |
29c696e1 VD |
637 | if (oom_reaper_list != NULL) { |
638 | tsk = oom_reaper_list; | |
639 | oom_reaper_list = tsk->oom_reaper_list; | |
03049269 MH |
640 | } |
641 | spin_unlock(&oom_reaper_lock); | |
642 | ||
643 | if (tsk) | |
644 | oom_reap_task(tsk); | |
aac45363 MH |
645 | } |
646 | ||
647 | return 0; | |
648 | } | |
649 | ||
7c5f64f8 | 650 | static void wake_oom_reaper(struct task_struct *tsk) |
aac45363 | 651 | { |
9bcdeb51 TH |
652 | /* mm is already queued? */ |
653 | if (test_and_set_bit(MMF_OOM_REAP_QUEUED, &tsk->signal->oom_mm->flags)) | |
aac45363 MH |
654 | return; |
655 | ||
36324a99 | 656 | get_task_struct(tsk); |
aac45363 | 657 | |
03049269 | 658 | spin_lock(&oom_reaper_lock); |
29c696e1 VD |
659 | tsk->oom_reaper_list = oom_reaper_list; |
660 | oom_reaper_list = tsk; | |
03049269 | 661 | spin_unlock(&oom_reaper_lock); |
422580c3 | 662 | trace_wake_reaper(tsk->pid); |
03049269 | 663 | wake_up(&oom_reaper_wait); |
aac45363 MH |
664 | } |
665 | ||
666 | static int __init oom_init(void) | |
667 | { | |
668 | oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper"); | |
aac45363 MH |
669 | return 0; |
670 | } | |
671 | subsys_initcall(oom_init) | |
7c5f64f8 VD |
672 | #else |
673 | static inline void wake_oom_reaper(struct task_struct *tsk) | |
674 | { | |
675 | } | |
676 | #endif /* CONFIG_MMU */ | |
aac45363 | 677 | |
49550b60 | 678 | /** |
16e95196 | 679 | * mark_oom_victim - mark the given task as OOM victim |
49550b60 | 680 | * @tsk: task to mark |
c32b3cbe | 681 | * |
dc56401f | 682 | * Has to be called with oom_lock held and never after |
c32b3cbe | 683 | * oom has been disabled already. |
26db62f1 MH |
684 | * |
685 | * tsk->mm has to be non NULL and caller has to guarantee it is stable (either | |
686 | * under task_lock or operate on the current). | |
49550b60 | 687 | */ |
7c5f64f8 | 688 | static void mark_oom_victim(struct task_struct *tsk) |
49550b60 | 689 | { |
26db62f1 MH |
690 | struct mm_struct *mm = tsk->mm; |
691 | ||
c32b3cbe MH |
692 | WARN_ON(oom_killer_disabled); |
693 | /* OOM killer might race with memcg OOM */ | |
694 | if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE)) | |
695 | return; | |
26db62f1 | 696 | |
26db62f1 | 697 | /* oom_mm is bound to the signal struct life time. */ |
4837fe37 | 698 | if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm)) { |
f1f10076 | 699 | mmgrab(tsk->signal->oom_mm); |
4837fe37 MH |
700 | set_bit(MMF_OOM_VICTIM, &mm->flags); |
701 | } | |
26db62f1 | 702 | |
63a8ca9b MH |
703 | /* |
704 | * Make sure that the task is woken up from uninterruptible sleep | |
705 | * if it is frozen because OOM killer wouldn't be able to free | |
706 | * any memory and livelock. freezing_slow_path will tell the freezer | |
707 | * that TIF_MEMDIE tasks should be ignored. | |
708 | */ | |
709 | __thaw_task(tsk); | |
c32b3cbe | 710 | atomic_inc(&oom_victims); |
422580c3 | 711 | trace_mark_victim(tsk->pid); |
49550b60 MH |
712 | } |
713 | ||
714 | /** | |
16e95196 | 715 | * exit_oom_victim - note the exit of an OOM victim |
49550b60 | 716 | */ |
38531201 | 717 | void exit_oom_victim(void) |
49550b60 | 718 | { |
38531201 | 719 | clear_thread_flag(TIF_MEMDIE); |
c32b3cbe | 720 | |
c38f1025 | 721 | if (!atomic_dec_return(&oom_victims)) |
c32b3cbe | 722 | wake_up_all(&oom_victims_wait); |
c32b3cbe MH |
723 | } |
724 | ||
7d2e7a22 MH |
725 | /** |
726 | * oom_killer_enable - enable OOM killer | |
727 | */ | |
728 | void oom_killer_enable(void) | |
729 | { | |
730 | oom_killer_disabled = false; | |
d75da004 | 731 | pr_info("OOM killer enabled.\n"); |
7d2e7a22 MH |
732 | } |
733 | ||
c32b3cbe MH |
734 | /** |
735 | * oom_killer_disable - disable OOM killer | |
7d2e7a22 | 736 | * @timeout: maximum timeout to wait for oom victims in jiffies |
c32b3cbe MH |
737 | * |
738 | * Forces all page allocations to fail rather than trigger OOM killer. | |
7d2e7a22 MH |
739 | * Will block and wait until all OOM victims are killed or the given |
740 | * timeout expires. | |
c32b3cbe MH |
741 | * |
742 | * The function cannot be called when there are runnable user tasks because | |
743 | * the userspace would see unexpected allocation failures as a result. Any | |
744 | * new usage of this function should be consulted with MM people. | |
745 | * | |
746 | * Returns true if successful and false if the OOM killer cannot be | |
747 | * disabled. | |
748 | */ | |
7d2e7a22 | 749 | bool oom_killer_disable(signed long timeout) |
c32b3cbe | 750 | { |
7d2e7a22 MH |
751 | signed long ret; |
752 | ||
c32b3cbe | 753 | /* |
6afcf289 TH |
754 | * Make sure to not race with an ongoing OOM killer. Check that the |
755 | * current is not killed (possibly due to sharing the victim's memory). | |
c32b3cbe | 756 | */ |
6afcf289 | 757 | if (mutex_lock_killable(&oom_lock)) |
c32b3cbe | 758 | return false; |
c32b3cbe | 759 | oom_killer_disabled = true; |
dc56401f | 760 | mutex_unlock(&oom_lock); |
c32b3cbe | 761 | |
7d2e7a22 MH |
762 | ret = wait_event_interruptible_timeout(oom_victims_wait, |
763 | !atomic_read(&oom_victims), timeout); | |
764 | if (ret <= 0) { | |
765 | oom_killer_enable(); | |
766 | return false; | |
767 | } | |
d75da004 | 768 | pr_info("OOM killer disabled.\n"); |
c32b3cbe MH |
769 | |
770 | return true; | |
771 | } | |
772 | ||
1af8bb43 MH |
773 | static inline bool __task_will_free_mem(struct task_struct *task) |
774 | { | |
775 | struct signal_struct *sig = task->signal; | |
776 | ||
777 | /* | |
778 | * A coredumping process may sleep for an extended period in exit_mm(), | |
779 | * so the oom killer cannot assume that the process will promptly exit | |
780 | * and release memory. | |
781 | */ | |
782 | if (sig->flags & SIGNAL_GROUP_COREDUMP) | |
783 | return false; | |
784 | ||
785 | if (sig->flags & SIGNAL_GROUP_EXIT) | |
786 | return true; | |
787 | ||
788 | if (thread_group_empty(task) && (task->flags & PF_EXITING)) | |
789 | return true; | |
790 | ||
791 | return false; | |
792 | } | |
793 | ||
794 | /* | |
795 | * Checks whether the given task is dying or exiting and likely to | |
796 | * release its address space. This means that all threads and processes | |
797 | * sharing the same mm have to be killed or exiting. | |
091f362c MH |
798 | * Caller has to make sure that task->mm is stable (hold task_lock or |
799 | * it operates on the current). | |
1af8bb43 | 800 | */ |
7c5f64f8 | 801 | static bool task_will_free_mem(struct task_struct *task) |
1af8bb43 | 802 | { |
091f362c | 803 | struct mm_struct *mm = task->mm; |
1af8bb43 | 804 | struct task_struct *p; |
f33e6f06 | 805 | bool ret = true; |
1af8bb43 | 806 | |
1af8bb43 | 807 | /* |
091f362c MH |
808 | * Skip tasks without mm because it might have passed its exit_mm and |
809 | * exit_oom_victim. oom_reaper could have rescued that but do not rely | |
810 | * on that for now. We can consider find_lock_task_mm in future. | |
1af8bb43 | 811 | */ |
091f362c | 812 | if (!mm) |
1af8bb43 MH |
813 | return false; |
814 | ||
091f362c MH |
815 | if (!__task_will_free_mem(task)) |
816 | return false; | |
696453e6 MH |
817 | |
818 | /* | |
819 | * This task has already been drained by the oom reaper so there are | |
820 | * only small chances it will free some more | |
821 | */ | |
862e3073 | 822 | if (test_bit(MMF_OOM_SKIP, &mm->flags)) |
696453e6 | 823 | return false; |
696453e6 | 824 | |
091f362c | 825 | if (atomic_read(&mm->mm_users) <= 1) |
1af8bb43 | 826 | return true; |
1af8bb43 MH |
827 | |
828 | /* | |
5870c2e1 MH |
829 | * Make sure that all tasks which share the mm with the given tasks |
830 | * are dying as well to make sure that a) nobody pins its mm and | |
831 | * b) the task is also reapable by the oom reaper. | |
1af8bb43 MH |
832 | */ |
833 | rcu_read_lock(); | |
834 | for_each_process(p) { | |
835 | if (!process_shares_mm(p, mm)) | |
836 | continue; | |
837 | if (same_thread_group(task, p)) | |
838 | continue; | |
839 | ret = __task_will_free_mem(p); | |
840 | if (!ret) | |
841 | break; | |
842 | } | |
843 | rcu_read_unlock(); | |
1af8bb43 MH |
844 | |
845 | return ret; | |
846 | } | |
847 | ||
bbbe4802 | 848 | static void __oom_kill_process(struct task_struct *victim, const char *message) |
1da177e4 | 849 | { |
5989ad7b | 850 | struct task_struct *p; |
647f2bdf | 851 | struct mm_struct *mm; |
bb29902a | 852 | bool can_oom_reap = true; |
1da177e4 | 853 | |
6b0c81b3 DR |
854 | p = find_lock_task_mm(victim); |
855 | if (!p) { | |
6b0c81b3 | 856 | put_task_struct(victim); |
647f2bdf | 857 | return; |
6b0c81b3 DR |
858 | } else if (victim != p) { |
859 | get_task_struct(p); | |
860 | put_task_struct(victim); | |
861 | victim = p; | |
862 | } | |
647f2bdf | 863 | |
880b7689 | 864 | /* Get a reference to safely compare mm after task_unlock(victim) */ |
647f2bdf | 865 | mm = victim->mm; |
f1f10076 | 866 | mmgrab(mm); |
8e675f7a KK |
867 | |
868 | /* Raise event before sending signal: task reaper must see this */ | |
869 | count_vm_event(OOM_KILL); | |
fe6bdfc8 | 870 | memcg_memory_event_mm(mm, MEMCG_OOM_KILL); |
8e675f7a | 871 | |
426fb5e7 | 872 | /* |
cd04ae1e MH |
873 | * We should send SIGKILL before granting access to memory reserves |
874 | * in order to prevent the OOM victim from depleting the memory | |
875 | * reserves from the user space under its control. | |
426fb5e7 | 876 | */ |
079b22dc | 877 | do_send_sig_info(SIGKILL, SEND_SIG_PRIV, victim, PIDTYPE_TGID); |
16e95196 | 878 | mark_oom_victim(victim); |
bbbe4802 SB |
879 | pr_err("%s: Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n", |
880 | message, task_pid_nr(victim), victim->comm, | |
881 | K(victim->mm->total_vm), | |
647f2bdf | 882 | K(get_mm_counter(victim->mm, MM_ANONPAGES)), |
eca56ff9 JM |
883 | K(get_mm_counter(victim->mm, MM_FILEPAGES)), |
884 | K(get_mm_counter(victim->mm, MM_SHMEMPAGES))); | |
647f2bdf DR |
885 | task_unlock(victim); |
886 | ||
887 | /* | |
888 | * Kill all user processes sharing victim->mm in other thread groups, if | |
889 | * any. They don't get access to memory reserves, though, to avoid | |
890 | * depletion of all memory. This prevents mm->mmap_sem livelock when an | |
891 | * oom killed thread cannot exit because it requires the semaphore and | |
892 | * its contended by another thread trying to allocate memory itself. | |
893 | * That thread will now get access to memory reserves since it has a | |
894 | * pending fatal signal. | |
895 | */ | |
4d4048be | 896 | rcu_read_lock(); |
c319025a | 897 | for_each_process(p) { |
4d7b3394 | 898 | if (!process_shares_mm(p, mm)) |
c319025a ON |
899 | continue; |
900 | if (same_thread_group(p, victim)) | |
901 | continue; | |
1b51e65e | 902 | if (is_global_init(p)) { |
aac45363 | 903 | can_oom_reap = false; |
862e3073 | 904 | set_bit(MMF_OOM_SKIP, &mm->flags); |
a373966d MH |
905 | pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n", |
906 | task_pid_nr(victim), victim->comm, | |
907 | task_pid_nr(p), p->comm); | |
c319025a | 908 | continue; |
aac45363 | 909 | } |
1b51e65e MH |
910 | /* |
911 | * No use_mm() user needs to read from the userspace so we are | |
912 | * ok to reap it. | |
913 | */ | |
914 | if (unlikely(p->flags & PF_KTHREAD)) | |
915 | continue; | |
079b22dc | 916 | do_send_sig_info(SIGKILL, SEND_SIG_PRIV, p, PIDTYPE_TGID); |
c319025a | 917 | } |
6b0c81b3 | 918 | rcu_read_unlock(); |
647f2bdf | 919 | |
aac45363 | 920 | if (can_oom_reap) |
36324a99 | 921 | wake_oom_reaper(victim); |
aac45363 | 922 | |
880b7689 | 923 | mmdrop(mm); |
6b0c81b3 | 924 | put_task_struct(victim); |
1da177e4 | 925 | } |
647f2bdf | 926 | #undef K |
1da177e4 | 927 | |
3d8b38eb RG |
928 | /* |
929 | * Kill provided task unless it's secured by setting | |
930 | * oom_score_adj to OOM_SCORE_ADJ_MIN. | |
931 | */ | |
bbbe4802 | 932 | static int oom_kill_memcg_member(struct task_struct *task, void *message) |
3d8b38eb | 933 | { |
d342a0b3 TH |
934 | if (task->signal->oom_score_adj != OOM_SCORE_ADJ_MIN && |
935 | !is_global_init(task)) { | |
3d8b38eb | 936 | get_task_struct(task); |
bbbe4802 | 937 | __oom_kill_process(task, message); |
3d8b38eb RG |
938 | } |
939 | return 0; | |
940 | } | |
941 | ||
5989ad7b RG |
942 | static void oom_kill_process(struct oom_control *oc, const char *message) |
943 | { | |
bbbe4802 | 944 | struct task_struct *victim = oc->chosen; |
3d8b38eb | 945 | struct mem_cgroup *oom_group; |
5989ad7b RG |
946 | static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
947 | DEFAULT_RATELIMIT_BURST); | |
948 | ||
949 | /* | |
950 | * If the task is already exiting, don't alarm the sysadmin or kill | |
951 | * its children or threads, just give it access to memory reserves | |
952 | * so it can die quickly | |
953 | */ | |
bbbe4802 SB |
954 | task_lock(victim); |
955 | if (task_will_free_mem(victim)) { | |
956 | mark_oom_victim(victim); | |
957 | wake_oom_reaper(victim); | |
958 | task_unlock(victim); | |
959 | put_task_struct(victim); | |
5989ad7b RG |
960 | return; |
961 | } | |
bbbe4802 | 962 | task_unlock(victim); |
5989ad7b RG |
963 | |
964 | if (__ratelimit(&oom_rs)) | |
bbbe4802 | 965 | dump_header(oc, victim); |
5989ad7b | 966 | |
3d8b38eb RG |
967 | /* |
968 | * Do we need to kill the entire memory cgroup? | |
969 | * Or even one of the ancestor memory cgroups? | |
970 | * Check this out before killing the victim task. | |
971 | */ | |
972 | oom_group = mem_cgroup_get_oom_group(victim, oc->memcg); | |
973 | ||
bbbe4802 | 974 | __oom_kill_process(victim, message); |
3d8b38eb RG |
975 | |
976 | /* | |
977 | * If necessary, kill all tasks in the selected memory cgroup. | |
978 | */ | |
979 | if (oom_group) { | |
980 | mem_cgroup_print_oom_group(oom_group); | |
bbbe4802 SB |
981 | mem_cgroup_scan_tasks(oom_group, oom_kill_memcg_member, |
982 | (void*)message); | |
3d8b38eb RG |
983 | mem_cgroup_put(oom_group); |
984 | } | |
5989ad7b RG |
985 | } |
986 | ||
309ed882 DR |
987 | /* |
988 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. | |
989 | */ | |
7c5f64f8 VD |
990 | static void check_panic_on_oom(struct oom_control *oc, |
991 | enum oom_constraint constraint) | |
309ed882 DR |
992 | { |
993 | if (likely(!sysctl_panic_on_oom)) | |
994 | return; | |
995 | if (sysctl_panic_on_oom != 2) { | |
996 | /* | |
997 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel | |
998 | * does not panic for cpuset, mempolicy, or memcg allocation | |
999 | * failures. | |
1000 | */ | |
1001 | if (constraint != CONSTRAINT_NONE) | |
1002 | return; | |
1003 | } | |
071a4bef | 1004 | /* Do not panic for oom kills triggered by sysrq */ |
db2a0dd7 | 1005 | if (is_sysrq_oom(oc)) |
071a4bef | 1006 | return; |
2a966b77 | 1007 | dump_header(oc, NULL); |
309ed882 DR |
1008 | panic("Out of memory: %s panic_on_oom is enabled\n", |
1009 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); | |
1010 | } | |
1011 | ||
8bc719d3 MS |
1012 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
1013 | ||
1014 | int register_oom_notifier(struct notifier_block *nb) | |
1015 | { | |
1016 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
1017 | } | |
1018 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
1019 | ||
1020 | int unregister_oom_notifier(struct notifier_block *nb) | |
1021 | { | |
1022 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
1023 | } | |
1024 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
1025 | ||
1da177e4 | 1026 | /** |
6e0fc46d DR |
1027 | * out_of_memory - kill the "best" process when we run out of memory |
1028 | * @oc: pointer to struct oom_control | |
1da177e4 LT |
1029 | * |
1030 | * If we run out of memory, we have the choice between either | |
1031 | * killing a random task (bad), letting the system crash (worse) | |
1032 | * OR try to be smart about which process to kill. Note that we | |
1033 | * don't have to be perfect here, we just have to be good. | |
1034 | */ | |
6e0fc46d | 1035 | bool out_of_memory(struct oom_control *oc) |
1da177e4 | 1036 | { |
8bc719d3 | 1037 | unsigned long freed = 0; |
e3658932 | 1038 | enum oom_constraint constraint = CONSTRAINT_NONE; |
8bc719d3 | 1039 | |
dc56401f JW |
1040 | if (oom_killer_disabled) |
1041 | return false; | |
1042 | ||
7c5f64f8 VD |
1043 | if (!is_memcg_oom(oc)) { |
1044 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
1045 | if (freed > 0) | |
1046 | /* Got some memory back in the last second. */ | |
1047 | return true; | |
1048 | } | |
1da177e4 | 1049 | |
7b98c2e4 | 1050 | /* |
9ff4868e DR |
1051 | * If current has a pending SIGKILL or is exiting, then automatically |
1052 | * select it. The goal is to allow it to allocate so that it may | |
1053 | * quickly exit and free its memory. | |
7b98c2e4 | 1054 | */ |
091f362c | 1055 | if (task_will_free_mem(current)) { |
16e95196 | 1056 | mark_oom_victim(current); |
1af8bb43 | 1057 | wake_oom_reaper(current); |
75e8f8b2 | 1058 | return true; |
7b98c2e4 DR |
1059 | } |
1060 | ||
3da88fb3 MH |
1061 | /* |
1062 | * The OOM killer does not compensate for IO-less reclaim. | |
1063 | * pagefault_out_of_memory lost its gfp context so we have to | |
1064 | * make sure exclude 0 mask - all other users should have at least | |
1065 | * ___GFP_DIRECT_RECLAIM to get here. | |
1066 | */ | |
06ad276a | 1067 | if (oc->gfp_mask && !(oc->gfp_mask & __GFP_FS)) |
3da88fb3 MH |
1068 | return true; |
1069 | ||
9b0f8b04 CL |
1070 | /* |
1071 | * Check if there were limitations on the allocation (only relevant for | |
7c5f64f8 | 1072 | * NUMA and memcg) that may require different handling. |
9b0f8b04 | 1073 | */ |
7c5f64f8 | 1074 | constraint = constrained_alloc(oc); |
6e0fc46d DR |
1075 | if (constraint != CONSTRAINT_MEMORY_POLICY) |
1076 | oc->nodemask = NULL; | |
2a966b77 | 1077 | check_panic_on_oom(oc, constraint); |
0aad4b31 | 1078 | |
7c5f64f8 VD |
1079 | if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task && |
1080 | current->mm && !oom_unkillable_task(current, NULL, oc->nodemask) && | |
121d1ba0 | 1081 | current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) { |
6b0c81b3 | 1082 | get_task_struct(current); |
7c5f64f8 VD |
1083 | oc->chosen = current; |
1084 | oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)"); | |
75e8f8b2 | 1085 | return true; |
0aad4b31 DR |
1086 | } |
1087 | ||
7c5f64f8 | 1088 | select_bad_process(oc); |
3100dab2 JW |
1089 | /* Found nothing?!?! */ |
1090 | if (!oc->chosen) { | |
2a966b77 | 1091 | dump_header(oc, NULL); |
3100dab2 JW |
1092 | pr_warn("Out of memory and no killable processes...\n"); |
1093 | /* | |
1094 | * If we got here due to an actual allocation at the | |
1095 | * system level, we cannot survive this and will enter | |
1096 | * an endless loop in the allocator. Bail out now. | |
1097 | */ | |
1098 | if (!is_sysrq_oom(oc) && !is_memcg_oom(oc)) | |
1099 | panic("System is deadlocked on memory\n"); | |
0aad4b31 | 1100 | } |
9bfe5ded | 1101 | if (oc->chosen && oc->chosen != (void *)-1UL) |
7c5f64f8 VD |
1102 | oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" : |
1103 | "Memory cgroup out of memory"); | |
7c5f64f8 | 1104 | return !!oc->chosen; |
c32b3cbe MH |
1105 | } |
1106 | ||
e3658932 DR |
1107 | /* |
1108 | * The pagefault handler calls here because it is out of memory, so kill a | |
798fd756 VD |
1109 | * memory-hogging task. If oom_lock is held by somebody else, a parallel oom |
1110 | * killing is already in progress so do nothing. | |
e3658932 DR |
1111 | */ |
1112 | void pagefault_out_of_memory(void) | |
1113 | { | |
6e0fc46d DR |
1114 | struct oom_control oc = { |
1115 | .zonelist = NULL, | |
1116 | .nodemask = NULL, | |
2a966b77 | 1117 | .memcg = NULL, |
6e0fc46d DR |
1118 | .gfp_mask = 0, |
1119 | .order = 0, | |
6e0fc46d DR |
1120 | }; |
1121 | ||
49426420 | 1122 | if (mem_cgroup_oom_synchronize(true)) |
dc56401f | 1123 | return; |
3812c8c8 | 1124 | |
dc56401f JW |
1125 | if (!mutex_trylock(&oom_lock)) |
1126 | return; | |
a104808e | 1127 | out_of_memory(&oc); |
dc56401f | 1128 | mutex_unlock(&oom_lock); |
e3658932 | 1129 | } |