]>
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... | |
7 | * | |
8 | * The routines in this file are used to kill a process when | |
a49335cc PJ |
9 | * we're seriously out of memory. This gets called from __alloc_pages() |
10 | * in mm/page_alloc.c when we really run out of memory. | |
1da177e4 LT |
11 | * |
12 | * Since we won't call these routines often (on a well-configured | |
13 | * machine) this file will double as a 'coding guide' and a signpost | |
14 | * for newbie kernel hackers. It features several pointers to major | |
15 | * kernel subsystems and hints as to where to find out what things do. | |
16 | */ | |
17 | ||
8ac773b4 | 18 | #include <linux/oom.h> |
1da177e4 | 19 | #include <linux/mm.h> |
4e950f6f | 20 | #include <linux/err.h> |
5a0e3ad6 | 21 | #include <linux/gfp.h> |
1da177e4 LT |
22 | #include <linux/sched.h> |
23 | #include <linux/swap.h> | |
24 | #include <linux/timex.h> | |
25 | #include <linux/jiffies.h> | |
ef08e3b4 | 26 | #include <linux/cpuset.h> |
8bc719d3 MS |
27 | #include <linux/module.h> |
28 | #include <linux/notifier.h> | |
c7ba5c9e | 29 | #include <linux/memcontrol.h> |
6f48d0eb | 30 | #include <linux/mempolicy.h> |
5cd9c58f | 31 | #include <linux/security.h> |
1da177e4 | 32 | |
fadd8fbd | 33 | int sysctl_panic_on_oom; |
fe071d7e | 34 | int sysctl_oom_kill_allocating_task; |
ad915c43 | 35 | int sysctl_oom_dump_tasks = 1; |
c7d4caeb | 36 | static DEFINE_SPINLOCK(zone_scan_lock); |
1da177e4 LT |
37 | /* #define DEBUG */ |
38 | ||
6f48d0eb DR |
39 | #ifdef CONFIG_NUMA |
40 | /** | |
41 | * has_intersects_mems_allowed() - check task eligiblity for kill | |
42 | * @tsk: task struct of which task to consider | |
43 | * @mask: nodemask passed to page allocator for mempolicy ooms | |
44 | * | |
45 | * Task eligibility is determined by whether or not a candidate task, @tsk, | |
46 | * shares the same mempolicy nodes as current if it is bound by such a policy | |
47 | * and whether or not it has the same set of allowed cpuset nodes. | |
495789a5 | 48 | */ |
6f48d0eb DR |
49 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
50 | const nodemask_t *mask) | |
495789a5 | 51 | { |
6f48d0eb | 52 | struct task_struct *start = tsk; |
495789a5 | 53 | |
495789a5 | 54 | do { |
6f48d0eb DR |
55 | if (mask) { |
56 | /* | |
57 | * If this is a mempolicy constrained oom, tsk's | |
58 | * cpuset is irrelevant. Only return true if its | |
59 | * mempolicy intersects current, otherwise it may be | |
60 | * needlessly killed. | |
61 | */ | |
62 | if (mempolicy_nodemask_intersects(tsk, mask)) | |
63 | return true; | |
64 | } else { | |
65 | /* | |
66 | * This is not a mempolicy constrained oom, so only | |
67 | * check the mems of tsk's cpuset. | |
68 | */ | |
69 | if (cpuset_mems_allowed_intersects(current, tsk)) | |
70 | return true; | |
71 | } | |
72 | tsk = next_thread(tsk); | |
73 | } while (tsk != start); | |
74 | return false; | |
75 | } | |
76 | #else | |
77 | static bool has_intersects_mems_allowed(struct task_struct *tsk, | |
78 | const nodemask_t *mask) | |
79 | { | |
80 | return true; | |
495789a5 | 81 | } |
6f48d0eb | 82 | #endif /* CONFIG_NUMA */ |
495789a5 | 83 | |
6f48d0eb DR |
84 | /* |
85 | * The process p may have detached its own ->mm while exiting or through | |
86 | * use_mm(), but one or more of its subthreads may still have a valid | |
87 | * pointer. Return p, or any of its subthreads with a valid ->mm, with | |
88 | * task_lock() held. | |
89 | */ | |
dd8e8f40 ON |
90 | static struct task_struct *find_lock_task_mm(struct task_struct *p) |
91 | { | |
92 | struct task_struct *t = p; | |
93 | ||
94 | do { | |
95 | task_lock(t); | |
96 | if (likely(t->mm)) | |
97 | return t; | |
98 | task_unlock(t); | |
99 | } while_each_thread(p, t); | |
100 | ||
101 | return NULL; | |
102 | } | |
103 | ||
1da177e4 | 104 | /** |
6937a25c | 105 | * badness - calculate a numeric value for how bad this task has been |
1da177e4 | 106 | * @p: task struct of which task we should calculate |
a49335cc | 107 | * @uptime: current uptime in seconds |
1da177e4 LT |
108 | * |
109 | * The formula used is relatively simple and documented inline in the | |
110 | * function. The main rationale is that we want to select a good task | |
111 | * to kill when we run out of memory. | |
112 | * | |
113 | * Good in this context means that: | |
114 | * 1) we lose the minimum amount of work done | |
115 | * 2) we recover a large amount of memory | |
116 | * 3) we don't kill anything innocent of eating tons of memory | |
117 | * 4) we want to kill the minimum amount of processes (one) | |
118 | * 5) we try to kill the process the user expects us to kill, this | |
119 | * algorithm has been meticulously tuned to meet the principle | |
120 | * of least surprise ... (be careful when you change it) | |
121 | */ | |
122 | ||
97d87c97 | 123 | unsigned long badness(struct task_struct *p, unsigned long uptime) |
1da177e4 | 124 | { |
a12888f7 | 125 | unsigned long points, cpu_time, run_time; |
97c2c9b8 | 126 | struct task_struct *child; |
dd8e8f40 | 127 | struct task_struct *c, *t; |
28b83c51 | 128 | int oom_adj = p->signal->oom_adj; |
495789a5 KM |
129 | struct task_cputime task_time; |
130 | unsigned long utime; | |
131 | unsigned long stime; | |
28b83c51 KM |
132 | |
133 | if (oom_adj == OOM_DISABLE) | |
134 | return 0; | |
1da177e4 | 135 | |
dd8e8f40 ON |
136 | p = find_lock_task_mm(p); |
137 | if (!p) | |
1da177e4 LT |
138 | return 0; |
139 | ||
140 | /* | |
141 | * The memory size of the process is the basis for the badness. | |
142 | */ | |
dd8e8f40 | 143 | points = p->mm->total_vm; |
97c2c9b8 | 144 | task_unlock(p); |
1da177e4 | 145 | |
7ba34859 HD |
146 | /* |
147 | * swapoff can easily use up all memory, so kill those first. | |
148 | */ | |
35451bee | 149 | if (p->flags & PF_OOM_ORIGIN) |
7ba34859 HD |
150 | return ULONG_MAX; |
151 | ||
1da177e4 LT |
152 | /* |
153 | * Processes which fork a lot of child processes are likely | |
9827b781 | 154 | * a good choice. We add half the vmsize of the children if they |
1da177e4 | 155 | * have an own mm. This prevents forking servers to flood the |
9827b781 KG |
156 | * machine with an endless amount of children. In case a single |
157 | * child is eating the vast majority of memory, adding only half | |
158 | * to the parents will make the child our kill candidate of choice. | |
1da177e4 | 159 | */ |
dd8e8f40 ON |
160 | t = p; |
161 | do { | |
162 | list_for_each_entry(c, &t->children, sibling) { | |
163 | child = find_lock_task_mm(c); | |
164 | if (child) { | |
165 | if (child->mm != p->mm) | |
166 | points += child->mm->total_vm/2 + 1; | |
167 | task_unlock(child); | |
168 | } | |
169 | } | |
170 | } while_each_thread(p, t); | |
1da177e4 LT |
171 | |
172 | /* | |
173 | * CPU time is in tens of seconds and run time is in thousands | |
174 | * of seconds. There is no particular reason for this other than | |
175 | * that it turned out to work very well in practice. | |
176 | */ | |
495789a5 KM |
177 | thread_group_cputime(p, &task_time); |
178 | utime = cputime_to_jiffies(task_time.utime); | |
179 | stime = cputime_to_jiffies(task_time.stime); | |
180 | cpu_time = (utime + stime) >> (SHIFT_HZ + 3); | |
181 | ||
1da177e4 LT |
182 | |
183 | if (uptime >= p->start_time.tv_sec) | |
184 | run_time = (uptime - p->start_time.tv_sec) >> 10; | |
185 | else | |
186 | run_time = 0; | |
187 | ||
a12888f7 CG |
188 | if (cpu_time) |
189 | points /= int_sqrt(cpu_time); | |
190 | if (run_time) | |
191 | points /= int_sqrt(int_sqrt(run_time)); | |
1da177e4 LT |
192 | |
193 | /* | |
194 | * Niced processes are most likely less important, so double | |
195 | * their badness points. | |
196 | */ | |
197 | if (task_nice(p) > 0) | |
198 | points *= 2; | |
199 | ||
200 | /* | |
201 | * Superuser processes are usually more important, so we make it | |
202 | * less likely that we kill those. | |
203 | */ | |
a2f2945a EP |
204 | if (has_capability_noaudit(p, CAP_SYS_ADMIN) || |
205 | has_capability_noaudit(p, CAP_SYS_RESOURCE)) | |
1da177e4 LT |
206 | points /= 4; |
207 | ||
208 | /* | |
209 | * We don't want to kill a process with direct hardware access. | |
210 | * Not only could that mess up the hardware, but usually users | |
211 | * tend to only have this flag set on applications they think | |
212 | * of as important. | |
213 | */ | |
a2f2945a | 214 | if (has_capability_noaudit(p, CAP_SYS_RAWIO)) |
1da177e4 LT |
215 | points /= 4; |
216 | ||
217 | /* | |
28b83c51 | 218 | * Adjust the score by oom_adj. |
1da177e4 | 219 | */ |
28b83c51 KM |
220 | if (oom_adj) { |
221 | if (oom_adj > 0) { | |
9a82782f JP |
222 | if (!points) |
223 | points = 1; | |
28b83c51 | 224 | points <<= oom_adj; |
9a82782f | 225 | } else |
28b83c51 | 226 | points >>= -(oom_adj); |
1da177e4 LT |
227 | } |
228 | ||
229 | #ifdef DEBUG | |
a5e58a61 | 230 | printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n", |
1da177e4 LT |
231 | p->pid, p->comm, points); |
232 | #endif | |
233 | return points; | |
234 | } | |
235 | ||
9b0f8b04 CL |
236 | /* |
237 | * Determine the type of allocation constraint. | |
238 | */ | |
9b0f8b04 | 239 | #ifdef CONFIG_NUMA |
4365a567 KH |
240 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
241 | gfp_t gfp_mask, nodemask_t *nodemask) | |
242 | { | |
54a6eb5c | 243 | struct zone *zone; |
dd1a239f | 244 | struct zoneref *z; |
54a6eb5c | 245 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
9b0f8b04 | 246 | |
4365a567 KH |
247 | /* |
248 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid | |
249 | * to kill current.We have to random task kill in this case. | |
250 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. | |
251 | */ | |
252 | if (gfp_mask & __GFP_THISNODE) | |
253 | return CONSTRAINT_NONE; | |
9b0f8b04 | 254 | |
4365a567 KH |
255 | /* |
256 | * The nodemask here is a nodemask passed to alloc_pages(). Now, | |
257 | * cpuset doesn't use this nodemask for its hardwall/softwall/hierarchy | |
258 | * feature. mempolicy is an only user of nodemask here. | |
259 | * check mempolicy's nodemask contains all N_HIGH_MEMORY | |
260 | */ | |
261 | if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) | |
9b0f8b04 | 262 | return CONSTRAINT_MEMORY_POLICY; |
4365a567 KH |
263 | |
264 | /* Check this allocation failure is caused by cpuset's wall function */ | |
265 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
266 | high_zoneidx, nodemask) | |
267 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) | |
268 | return CONSTRAINT_CPUSET; | |
9b0f8b04 CL |
269 | |
270 | return CONSTRAINT_NONE; | |
271 | } | |
4365a567 KH |
272 | #else |
273 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, | |
274 | gfp_t gfp_mask, nodemask_t *nodemask) | |
275 | { | |
276 | return CONSTRAINT_NONE; | |
277 | } | |
278 | #endif | |
9b0f8b04 | 279 | |
1da177e4 LT |
280 | /* |
281 | * Simple selection loop. We chose the process with the highest | |
282 | * number of 'points'. We expect the caller will lock the tasklist. | |
283 | * | |
284 | * (not docbooked, we don't want this one cluttering up the manual) | |
285 | */ | |
c7ba5c9e | 286 | static struct task_struct *select_bad_process(unsigned long *ppoints, |
f4420032 | 287 | struct mem_cgroup *mem, const nodemask_t *nodemask) |
1da177e4 | 288 | { |
495789a5 | 289 | struct task_struct *p; |
1da177e4 LT |
290 | struct task_struct *chosen = NULL; |
291 | struct timespec uptime; | |
9827b781 | 292 | *ppoints = 0; |
1da177e4 LT |
293 | |
294 | do_posix_clock_monotonic_gettime(&uptime); | |
495789a5 | 295 | for_each_process(p) { |
a49335cc | 296 | unsigned long points; |
a49335cc | 297 | |
455c0e5f ON |
298 | /* skip the init task and kthreads */ |
299 | if (is_global_init(p) || (p->flags & PF_KTHREAD)) | |
a49335cc | 300 | continue; |
4c4a2214 DR |
301 | if (mem && !task_in_mem_cgroup(p, mem)) |
302 | continue; | |
f4420032 | 303 | if (!has_intersects_mems_allowed(p, nodemask)) |
6cf86ac6 | 304 | continue; |
ef08e3b4 | 305 | |
b78483a4 NP |
306 | /* |
307 | * This task already has access to memory reserves and is | |
308 | * being killed. Don't allow any other task access to the | |
309 | * memory reserve. | |
310 | * | |
311 | * Note: this may have a chance of deadlock if it gets | |
312 | * blocked waiting for another task which itself is waiting | |
313 | * for memory. Is there a better alternative? | |
314 | */ | |
315 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) | |
316 | return ERR_PTR(-1UL); | |
317 | ||
a49335cc | 318 | /* |
6937a25c | 319 | * This is in the process of releasing memory so wait for it |
a49335cc | 320 | * to finish before killing some other task by mistake. |
50ec3bbf NP |
321 | * |
322 | * However, if p is the current task, we allow the 'kill' to | |
323 | * go ahead if it is exiting: this will simply set TIF_MEMDIE, | |
324 | * which will allow it to gain access to memory reserves in | |
325 | * the process of exiting and releasing its resources. | |
b78483a4 | 326 | * Otherwise we could get an easy OOM deadlock. |
a49335cc | 327 | */ |
b5227940 | 328 | if ((p->flags & PF_EXITING) && p->mm) { |
b78483a4 NP |
329 | if (p != current) |
330 | return ERR_PTR(-1UL); | |
331 | ||
972c4ea5 ON |
332 | chosen = p; |
333 | *ppoints = ULONG_MAX; | |
50ec3bbf | 334 | } |
972c4ea5 | 335 | |
28b83c51 | 336 | if (p->signal->oom_adj == OOM_DISABLE) |
0753ba01 KM |
337 | continue; |
338 | ||
97d87c97 | 339 | points = badness(p, uptime.tv_sec); |
0753ba01 | 340 | if (points > *ppoints || !chosen) { |
a49335cc | 341 | chosen = p; |
9827b781 | 342 | *ppoints = points; |
1da177e4 | 343 | } |
495789a5 | 344 | } |
972c4ea5 | 345 | |
1da177e4 LT |
346 | return chosen; |
347 | } | |
348 | ||
fef1bdd6 | 349 | /** |
1b578df0 | 350 | * dump_tasks - dump current memory state of all system tasks |
74ab7f1d | 351 | * @mem: current's memory controller, if constrained |
1b578df0 | 352 | * |
fef1bdd6 DR |
353 | * Dumps the current memory state of all system tasks, excluding kernel threads. |
354 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj | |
355 | * score, and name. | |
356 | * | |
357 | * If the actual is non-NULL, only tasks that are a member of the mem_cgroup are | |
358 | * shown. | |
359 | * | |
360 | * Call with tasklist_lock read-locked. | |
361 | */ | |
362 | static void dump_tasks(const struct mem_cgroup *mem) | |
363 | { | |
c55db957 KM |
364 | struct task_struct *p; |
365 | struct task_struct *task; | |
fef1bdd6 DR |
366 | |
367 | printk(KERN_INFO "[ pid ] uid tgid total_vm rss cpu oom_adj " | |
368 | "name\n"); | |
c55db957 | 369 | for_each_process(p) { |
c55db957 | 370 | if (p->flags & PF_KTHREAD) |
fef1bdd6 | 371 | continue; |
c55db957 | 372 | if (mem && !task_in_mem_cgroup(p, mem)) |
b4416d2b | 373 | continue; |
fef1bdd6 | 374 | |
c55db957 KM |
375 | task = find_lock_task_mm(p); |
376 | if (!task) { | |
6d2661ed | 377 | /* |
74ab7f1d DR |
378 | * This is a kthread or all of p's threads have already |
379 | * detached their mm's. There's no need to report | |
c55db957 | 380 | * them; they can't be oom killed anyway. |
6d2661ed | 381 | */ |
6d2661ed DR |
382 | continue; |
383 | } | |
c55db957 | 384 | |
c81fac5c | 385 | printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3u %3d %s\n", |
c55db957 KM |
386 | task->pid, __task_cred(task)->uid, task->tgid, |
387 | task->mm->total_vm, get_mm_rss(task->mm), | |
c81fac5c | 388 | task_cpu(task), task->signal->oom_adj, task->comm); |
c55db957 KM |
389 | task_unlock(task); |
390 | } | |
fef1bdd6 DR |
391 | } |
392 | ||
d31f56db DN |
393 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
394 | struct mem_cgroup *mem) | |
1b604d75 | 395 | { |
5e9d834a | 396 | task_lock(current); |
1b604d75 DR |
397 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " |
398 | "oom_adj=%d\n", | |
399 | current->comm, gfp_mask, order, current->signal->oom_adj); | |
1b604d75 DR |
400 | cpuset_print_task_mems_allowed(current); |
401 | task_unlock(current); | |
402 | dump_stack(); | |
d31f56db | 403 | mem_cgroup_print_oom_info(mem, p); |
1b604d75 DR |
404 | show_mem(); |
405 | if (sysctl_oom_dump_tasks) | |
406 | dump_tasks(mem); | |
407 | } | |
408 | ||
3b4798cb | 409 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
b940fd70 | 410 | static int oom_kill_task(struct task_struct *p) |
1da177e4 | 411 | { |
dd8e8f40 | 412 | p = find_lock_task_mm(p); |
b940fd70 DR |
413 | if (!p || p->signal->oom_adj == OOM_DISABLE) { |
414 | task_unlock(p); | |
415 | return 1; | |
416 | } | |
417 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", | |
418 | task_pid_nr(p), p->comm, K(p->mm->total_vm), | |
419 | K(get_mm_counter(p->mm, MM_ANONPAGES)), | |
420 | K(get_mm_counter(p->mm, MM_FILEPAGES))); | |
3b4798cb | 421 | task_unlock(p); |
1da177e4 | 422 | |
fa717060 | 423 | p->rt.time_slice = HZ; |
1da177e4 | 424 | set_tsk_thread_flag(p, TIF_MEMDIE); |
1da177e4 | 425 | force_sig(SIGKILL, p); |
01315922 | 426 | return 0; |
1da177e4 | 427 | } |
b940fd70 | 428 | #undef K |
1da177e4 | 429 | |
7213f506 | 430 | static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
fef1bdd6 | 431 | unsigned long points, struct mem_cgroup *mem, |
7c59aec8 | 432 | nodemask_t *nodemask, const char *message) |
1da177e4 | 433 | { |
5e9d834a DR |
434 | struct task_struct *victim = p; |
435 | struct task_struct *child; | |
dd8e8f40 | 436 | struct task_struct *t = p; |
5e9d834a DR |
437 | unsigned long victim_points = 0; |
438 | struct timespec uptime; | |
1da177e4 | 439 | |
1b604d75 | 440 | if (printk_ratelimit()) |
d31f56db | 441 | dump_header(p, gfp_mask, order, mem); |
7213f506 | 442 | |
50ec3bbf NP |
443 | /* |
444 | * If the task is already exiting, don't alarm the sysadmin or kill | |
445 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
446 | */ | |
0753ba01 | 447 | if (p->flags & PF_EXITING) { |
4358997a | 448 | set_tsk_thread_flag(p, TIF_MEMDIE); |
50ec3bbf NP |
449 | return 0; |
450 | } | |
451 | ||
5e9d834a DR |
452 | task_lock(p); |
453 | pr_err("%s: Kill process %d (%s) score %lu or sacrifice child\n", | |
454 | message, task_pid_nr(p), p->comm, points); | |
455 | task_unlock(p); | |
f3af38d3 | 456 | |
5e9d834a DR |
457 | /* |
458 | * If any of p's children has a different mm and is eligible for kill, | |
459 | * the one with the highest badness() score is sacrificed for its | |
460 | * parent. This attempts to lose the minimal amount of work done while | |
461 | * still freeing memory. | |
462 | */ | |
463 | do_posix_clock_monotonic_gettime(&uptime); | |
dd8e8f40 | 464 | do { |
5e9d834a DR |
465 | list_for_each_entry(child, &t->children, sibling) { |
466 | unsigned long child_points; | |
467 | ||
468 | if (child->mm == p->mm) | |
dd8e8f40 | 469 | continue; |
2c5ea53c KM |
470 | if (child->flags & PF_KTHREAD) |
471 | continue; | |
5e9d834a | 472 | if (mem && !task_in_mem_cgroup(child, mem)) |
dd8e8f40 | 473 | continue; |
7c59aec8 KM |
474 | if (!has_intersects_mems_allowed(child, nodemask)) |
475 | continue; | |
5e9d834a DR |
476 | |
477 | /* badness() returns 0 if the thread is unkillable */ | |
478 | child_points = badness(child, uptime.tv_sec); | |
479 | if (child_points > victim_points) { | |
480 | victim = child; | |
481 | victim_points = child_points; | |
482 | } | |
dd8e8f40 ON |
483 | } |
484 | } while_each_thread(p, t); | |
485 | ||
5e9d834a | 486 | return oom_kill_task(victim); |
1da177e4 LT |
487 | } |
488 | ||
309ed882 DR |
489 | /* |
490 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. | |
491 | */ | |
492 | static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask, | |
493 | int order) | |
494 | { | |
495 | if (likely(!sysctl_panic_on_oom)) | |
496 | return; | |
497 | if (sysctl_panic_on_oom != 2) { | |
498 | /* | |
499 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel | |
500 | * does not panic for cpuset, mempolicy, or memcg allocation | |
501 | * failures. | |
502 | */ | |
503 | if (constraint != CONSTRAINT_NONE) | |
504 | return; | |
505 | } | |
506 | read_lock(&tasklist_lock); | |
507 | dump_header(NULL, gfp_mask, order, NULL); | |
508 | read_unlock(&tasklist_lock); | |
509 | panic("Out of memory: %s panic_on_oom is enabled\n", | |
510 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); | |
511 | } | |
512 | ||
00f0b825 | 513 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
c7ba5c9e PE |
514 | void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) |
515 | { | |
516 | unsigned long points = 0; | |
517 | struct task_struct *p; | |
518 | ||
309ed882 | 519 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0); |
e115f2d8 | 520 | read_lock(&tasklist_lock); |
c7ba5c9e | 521 | retry: |
f4420032 | 522 | p = select_bad_process(&points, mem, NULL); |
df64f81b | 523 | if (!p || PTR_ERR(p) == -1UL) |
c7ba5c9e PE |
524 | goto out; |
525 | ||
7c59aec8 | 526 | if (oom_kill_process(p, gfp_mask, 0, points, mem, NULL, |
c7ba5c9e PE |
527 | "Memory cgroup out of memory")) |
528 | goto retry; | |
529 | out: | |
e115f2d8 | 530 | read_unlock(&tasklist_lock); |
c7ba5c9e PE |
531 | } |
532 | #endif | |
533 | ||
8bc719d3 MS |
534 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
535 | ||
536 | int register_oom_notifier(struct notifier_block *nb) | |
537 | { | |
538 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
539 | } | |
540 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
541 | ||
542 | int unregister_oom_notifier(struct notifier_block *nb) | |
543 | { | |
544 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
545 | } | |
546 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
547 | ||
098d7f12 DR |
548 | /* |
549 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero | |
550 | * if a parallel OOM killing is already taking place that includes a zone in | |
551 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. | |
552 | */ | |
ff321fea | 553 | int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 554 | { |
dd1a239f MG |
555 | struct zoneref *z; |
556 | struct zone *zone; | |
098d7f12 DR |
557 | int ret = 1; |
558 | ||
c7d4caeb | 559 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
560 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
561 | if (zone_is_oom_locked(zone)) { | |
098d7f12 DR |
562 | ret = 0; |
563 | goto out; | |
564 | } | |
dd1a239f MG |
565 | } |
566 | ||
567 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
568 | /* | |
c7d4caeb | 569 | * Lock each zone in the zonelist under zone_scan_lock so a |
ff321fea | 570 | * parallel invocation of try_set_zonelist_oom() doesn't succeed |
dd1a239f MG |
571 | * when it shouldn't. |
572 | */ | |
573 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
574 | } | |
098d7f12 | 575 | |
098d7f12 | 576 | out: |
c7d4caeb | 577 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
578 | return ret; |
579 | } | |
580 | ||
581 | /* | |
582 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed | |
583 | * allocation attempts with zonelists containing them may now recall the OOM | |
584 | * killer, if necessary. | |
585 | */ | |
dd1a239f | 586 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 587 | { |
dd1a239f MG |
588 | struct zoneref *z; |
589 | struct zone *zone; | |
098d7f12 | 590 | |
c7d4caeb | 591 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
592 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
593 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
594 | } | |
c7d4caeb | 595 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
596 | } |
597 | ||
e3658932 DR |
598 | /* |
599 | * Try to acquire the oom killer lock for all system zones. Returns zero if a | |
600 | * parallel oom killing is taking place, otherwise locks all zones and returns | |
601 | * non-zero. | |
602 | */ | |
603 | static int try_set_system_oom(void) | |
604 | { | |
605 | struct zone *zone; | |
606 | int ret = 1; | |
607 | ||
608 | spin_lock(&zone_scan_lock); | |
609 | for_each_populated_zone(zone) | |
610 | if (zone_is_oom_locked(zone)) { | |
611 | ret = 0; | |
612 | goto out; | |
613 | } | |
614 | for_each_populated_zone(zone) | |
615 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
616 | out: | |
617 | spin_unlock(&zone_scan_lock); | |
618 | return ret; | |
619 | } | |
620 | ||
621 | /* | |
622 | * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation | |
623 | * attempts or page faults may now recall the oom killer, if necessary. | |
624 | */ | |
625 | static void clear_system_oom(void) | |
626 | { | |
627 | struct zone *zone; | |
628 | ||
629 | spin_lock(&zone_scan_lock); | |
630 | for_each_populated_zone(zone) | |
631 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
632 | spin_unlock(&zone_scan_lock); | |
633 | } | |
634 | ||
1da177e4 | 635 | /** |
6937a25c | 636 | * out_of_memory - kill the "best" process when we run out of memory |
1b578df0 RD |
637 | * @zonelist: zonelist pointer |
638 | * @gfp_mask: memory allocation flags | |
639 | * @order: amount of memory being requested as a power of 2 | |
6f48d0eb | 640 | * @nodemask: nodemask passed to page allocator |
1da177e4 LT |
641 | * |
642 | * If we run out of memory, we have the choice between either | |
643 | * killing a random task (bad), letting the system crash (worse) | |
644 | * OR try to be smart about which process to kill. Note that we | |
645 | * don't have to be perfect here, we just have to be good. | |
646 | */ | |
4365a567 KH |
647 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
648 | int order, nodemask_t *nodemask) | |
1da177e4 | 649 | { |
0aad4b31 | 650 | struct task_struct *p; |
8bc719d3 | 651 | unsigned long freed = 0; |
0aad4b31 | 652 | unsigned long points; |
e3658932 | 653 | enum oom_constraint constraint = CONSTRAINT_NONE; |
8bc719d3 MS |
654 | |
655 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
656 | if (freed > 0) | |
657 | /* Got some memory back in the last second. */ | |
658 | return; | |
1da177e4 | 659 | |
7b98c2e4 DR |
660 | /* |
661 | * If current has a pending SIGKILL, then automatically select it. The | |
662 | * goal is to allow it to allocate so that it may quickly exit and free | |
663 | * its memory. | |
664 | */ | |
665 | if (fatal_signal_pending(current)) { | |
666 | set_thread_flag(TIF_MEMDIE); | |
667 | return; | |
668 | } | |
669 | ||
9b0f8b04 CL |
670 | /* |
671 | * Check if there were limitations on the allocation (only relevant for | |
672 | * NUMA) that may require different handling. | |
673 | */ | |
e3658932 DR |
674 | if (zonelist) |
675 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask); | |
309ed882 | 676 | check_panic_on_oom(constraint, gfp_mask, order); |
0aad4b31 | 677 | |
2b45ab33 | 678 | read_lock(&tasklist_lock); |
0aad4b31 DR |
679 | if (sysctl_oom_kill_allocating_task) { |
680 | /* | |
681 | * oom_kill_process() needs tasklist_lock held. If it returns | |
682 | * non-zero, current could not be killed so we must fallback to | |
683 | * the tasklist scan. | |
684 | */ | |
685 | if (!oom_kill_process(current, gfp_mask, order, 0, NULL, | |
7c59aec8 | 686 | nodemask, |
0aad4b31 DR |
687 | "Out of memory (oom_kill_allocating_task)")) |
688 | return; | |
689 | } | |
690 | ||
691 | retry: | |
692 | p = select_bad_process(&points, NULL, | |
f4420032 DR |
693 | constraint == CONSTRAINT_MEMORY_POLICY ? nodemask : |
694 | NULL); | |
0aad4b31 DR |
695 | if (PTR_ERR(p) == -1UL) |
696 | return; | |
697 | ||
698 | /* Found nothing?!?! Either we hang forever, or we panic. */ | |
699 | if (!p) { | |
700 | dump_header(NULL, gfp_mask, order, NULL); | |
701 | read_unlock(&tasklist_lock); | |
702 | panic("Out of memory and no killable processes...\n"); | |
703 | } | |
704 | ||
7c59aec8 | 705 | if (oom_kill_process(p, gfp_mask, order, points, NULL, nodemask, |
0aad4b31 DR |
706 | "Out of memory")) |
707 | goto retry; | |
140ffcec | 708 | read_unlock(&tasklist_lock); |
1da177e4 LT |
709 | |
710 | /* | |
711 | * Give "p" a good chance of killing itself before we | |
2f659f46 | 712 | * retry to allocate memory unless "p" is current |
1da177e4 | 713 | */ |
2f659f46 | 714 | if (!test_thread_flag(TIF_MEMDIE)) |
140ffcec | 715 | schedule_timeout_uninterruptible(1); |
1da177e4 | 716 | } |
e3658932 DR |
717 | |
718 | /* | |
719 | * The pagefault handler calls here because it is out of memory, so kill a | |
720 | * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel | |
721 | * oom killing is already in progress so do nothing. If a task is found with | |
722 | * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit. | |
723 | */ | |
724 | void pagefault_out_of_memory(void) | |
725 | { | |
726 | if (try_set_system_oom()) { | |
727 | out_of_memory(NULL, 0, 0, NULL); | |
728 | clear_system_oom(); | |
729 | } | |
730 | if (!test_thread_flag(TIF_MEMDIE)) | |
731 | schedule_timeout_uninterruptible(1); | |
732 | } |