]>
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, |
6f48d0eb DR |
287 | struct mem_cgroup *mem, enum oom_constraint constraint, |
288 | const nodemask_t *mask) | |
1da177e4 | 289 | { |
495789a5 | 290 | struct task_struct *p; |
1da177e4 LT |
291 | struct task_struct *chosen = NULL; |
292 | struct timespec uptime; | |
9827b781 | 293 | *ppoints = 0; |
1da177e4 LT |
294 | |
295 | do_posix_clock_monotonic_gettime(&uptime); | |
495789a5 | 296 | for_each_process(p) { |
a49335cc | 297 | unsigned long points; |
a49335cc | 298 | |
455c0e5f ON |
299 | /* skip the init task and kthreads */ |
300 | if (is_global_init(p) || (p->flags & PF_KTHREAD)) | |
a49335cc | 301 | continue; |
4c4a2214 DR |
302 | if (mem && !task_in_mem_cgroup(p, mem)) |
303 | continue; | |
6f48d0eb DR |
304 | if (!has_intersects_mems_allowed(p, |
305 | constraint == CONSTRAINT_MEMORY_POLICY ? mask : | |
306 | NULL)) | |
6cf86ac6 | 307 | continue; |
ef08e3b4 | 308 | |
b78483a4 NP |
309 | /* |
310 | * This task already has access to memory reserves and is | |
311 | * being killed. Don't allow any other task access to the | |
312 | * memory reserve. | |
313 | * | |
314 | * Note: this may have a chance of deadlock if it gets | |
315 | * blocked waiting for another task which itself is waiting | |
316 | * for memory. Is there a better alternative? | |
317 | */ | |
318 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) | |
319 | return ERR_PTR(-1UL); | |
320 | ||
a49335cc | 321 | /* |
6937a25c | 322 | * This is in the process of releasing memory so wait for it |
a49335cc | 323 | * to finish before killing some other task by mistake. |
50ec3bbf NP |
324 | * |
325 | * However, if p is the current task, we allow the 'kill' to | |
326 | * go ahead if it is exiting: this will simply set TIF_MEMDIE, | |
327 | * which will allow it to gain access to memory reserves in | |
328 | * the process of exiting and releasing its resources. | |
b78483a4 | 329 | * Otherwise we could get an easy OOM deadlock. |
a49335cc | 330 | */ |
b5227940 | 331 | if ((p->flags & PF_EXITING) && p->mm) { |
b78483a4 NP |
332 | if (p != current) |
333 | return ERR_PTR(-1UL); | |
334 | ||
972c4ea5 ON |
335 | chosen = p; |
336 | *ppoints = ULONG_MAX; | |
50ec3bbf | 337 | } |
972c4ea5 | 338 | |
28b83c51 | 339 | if (p->signal->oom_adj == OOM_DISABLE) |
0753ba01 KM |
340 | continue; |
341 | ||
97d87c97 | 342 | points = badness(p, uptime.tv_sec); |
0753ba01 | 343 | if (points > *ppoints || !chosen) { |
a49335cc | 344 | chosen = p; |
9827b781 | 345 | *ppoints = points; |
1da177e4 | 346 | } |
495789a5 | 347 | } |
972c4ea5 | 348 | |
1da177e4 LT |
349 | return chosen; |
350 | } | |
351 | ||
fef1bdd6 | 352 | /** |
1b578df0 | 353 | * dump_tasks - dump current memory state of all system tasks |
74ab7f1d | 354 | * @mem: current's memory controller, if constrained |
1b578df0 | 355 | * |
fef1bdd6 DR |
356 | * Dumps the current memory state of all system tasks, excluding kernel threads. |
357 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj | |
358 | * score, and name. | |
359 | * | |
360 | * If the actual is non-NULL, only tasks that are a member of the mem_cgroup are | |
361 | * shown. | |
362 | * | |
363 | * Call with tasklist_lock read-locked. | |
364 | */ | |
365 | static void dump_tasks(const struct mem_cgroup *mem) | |
366 | { | |
c55db957 KM |
367 | struct task_struct *p; |
368 | struct task_struct *task; | |
fef1bdd6 DR |
369 | |
370 | printk(KERN_INFO "[ pid ] uid tgid total_vm rss cpu oom_adj " | |
371 | "name\n"); | |
c55db957 | 372 | for_each_process(p) { |
c55db957 | 373 | if (p->flags & PF_KTHREAD) |
fef1bdd6 | 374 | continue; |
c55db957 | 375 | if (mem && !task_in_mem_cgroup(p, mem)) |
b4416d2b | 376 | continue; |
fef1bdd6 | 377 | |
c55db957 KM |
378 | task = find_lock_task_mm(p); |
379 | if (!task) { | |
6d2661ed | 380 | /* |
74ab7f1d DR |
381 | * This is a kthread or all of p's threads have already |
382 | * detached their mm's. There's no need to report | |
c55db957 | 383 | * them; they can't be oom killed anyway. |
6d2661ed | 384 | */ |
6d2661ed DR |
385 | continue; |
386 | } | |
c55db957 | 387 | |
c81fac5c | 388 | printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3u %3d %s\n", |
c55db957 KM |
389 | task->pid, __task_cred(task)->uid, task->tgid, |
390 | task->mm->total_vm, get_mm_rss(task->mm), | |
c81fac5c | 391 | task_cpu(task), task->signal->oom_adj, task->comm); |
c55db957 KM |
392 | task_unlock(task); |
393 | } | |
fef1bdd6 DR |
394 | } |
395 | ||
d31f56db DN |
396 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
397 | struct mem_cgroup *mem) | |
1b604d75 | 398 | { |
5e9d834a | 399 | task_lock(current); |
1b604d75 DR |
400 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " |
401 | "oom_adj=%d\n", | |
402 | current->comm, gfp_mask, order, current->signal->oom_adj); | |
1b604d75 DR |
403 | cpuset_print_task_mems_allowed(current); |
404 | task_unlock(current); | |
405 | dump_stack(); | |
d31f56db | 406 | mem_cgroup_print_oom_info(mem, p); |
1b604d75 DR |
407 | show_mem(); |
408 | if (sysctl_oom_dump_tasks) | |
409 | dump_tasks(mem); | |
410 | } | |
411 | ||
3b4798cb KM |
412 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
413 | ||
1b578df0 | 414 | /* |
5a291b98 RG |
415 | * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO |
416 | * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO | |
417 | * set. | |
1da177e4 | 418 | */ |
f3af38d3 | 419 | static void __oom_kill_task(struct task_struct *p, int verbose) |
1da177e4 | 420 | { |
b460cbc5 | 421 | if (is_global_init(p)) { |
1da177e4 LT |
422 | WARN_ON(1); |
423 | printk(KERN_WARNING "tried to kill init!\n"); | |
424 | return; | |
425 | } | |
426 | ||
dd8e8f40 ON |
427 | p = find_lock_task_mm(p); |
428 | if (!p) | |
1da177e4 | 429 | return; |
50ec3bbf | 430 | |
f3af38d3 | 431 | if (verbose) |
3b4798cb KM |
432 | printk(KERN_ERR "Killed process %d (%s) " |
433 | "vsz:%lukB, anon-rss:%lukB, file-rss:%lukB\n", | |
434 | task_pid_nr(p), p->comm, | |
435 | K(p->mm->total_vm), | |
d559db08 KH |
436 | K(get_mm_counter(p->mm, MM_ANONPAGES)), |
437 | K(get_mm_counter(p->mm, MM_FILEPAGES))); | |
3b4798cb | 438 | task_unlock(p); |
1da177e4 LT |
439 | |
440 | /* | |
441 | * We give our sacrificial lamb high priority and access to | |
442 | * all the memory it needs. That way it should be able to | |
443 | * exit() and clear out its resources quickly... | |
444 | */ | |
fa717060 | 445 | p->rt.time_slice = HZ; |
1da177e4 LT |
446 | set_tsk_thread_flag(p, TIF_MEMDIE); |
447 | ||
448 | force_sig(SIGKILL, p); | |
449 | } | |
450 | ||
f3af38d3 | 451 | static int oom_kill_task(struct task_struct *p) |
1da177e4 | 452 | { |
0753ba01 KM |
453 | /* WARNING: mm may not be dereferenced since we did not obtain its |
454 | * value from get_task_mm(p). This is OK since all we need to do is | |
455 | * compare mm to q->mm below. | |
456 | * | |
457 | * Furthermore, even if mm contains a non-NULL value, p->mm may | |
458 | * change to NULL at any time since we do not hold task_lock(p). | |
459 | * However, this is of no concern to us. | |
460 | */ | |
8c5cd6f3 | 461 | if (!p->mm || p->signal->oom_adj == OOM_DISABLE) |
01315922 | 462 | return 1; |
0753ba01 | 463 | |
f3af38d3 | 464 | __oom_kill_task(p, 1); |
c33e0fca | 465 | |
01315922 | 466 | return 0; |
1da177e4 LT |
467 | } |
468 | ||
7213f506 | 469 | static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
fef1bdd6 DR |
470 | unsigned long points, struct mem_cgroup *mem, |
471 | const char *message) | |
1da177e4 | 472 | { |
5e9d834a DR |
473 | struct task_struct *victim = p; |
474 | struct task_struct *child; | |
dd8e8f40 | 475 | struct task_struct *t = p; |
5e9d834a DR |
476 | unsigned long victim_points = 0; |
477 | struct timespec uptime; | |
1da177e4 | 478 | |
1b604d75 | 479 | if (printk_ratelimit()) |
d31f56db | 480 | dump_header(p, gfp_mask, order, mem); |
7213f506 | 481 | |
50ec3bbf NP |
482 | /* |
483 | * If the task is already exiting, don't alarm the sysadmin or kill | |
484 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
485 | */ | |
0753ba01 | 486 | if (p->flags & PF_EXITING) { |
4358997a | 487 | set_tsk_thread_flag(p, TIF_MEMDIE); |
50ec3bbf NP |
488 | return 0; |
489 | } | |
490 | ||
5e9d834a DR |
491 | task_lock(p); |
492 | pr_err("%s: Kill process %d (%s) score %lu or sacrifice child\n", | |
493 | message, task_pid_nr(p), p->comm, points); | |
494 | task_unlock(p); | |
f3af38d3 | 495 | |
5e9d834a DR |
496 | /* |
497 | * If any of p's children has a different mm and is eligible for kill, | |
498 | * the one with the highest badness() score is sacrificed for its | |
499 | * parent. This attempts to lose the minimal amount of work done while | |
500 | * still freeing memory. | |
501 | */ | |
502 | do_posix_clock_monotonic_gettime(&uptime); | |
dd8e8f40 | 503 | do { |
5e9d834a DR |
504 | list_for_each_entry(child, &t->children, sibling) { |
505 | unsigned long child_points; | |
506 | ||
507 | if (child->mm == p->mm) | |
dd8e8f40 | 508 | continue; |
5e9d834a | 509 | if (mem && !task_in_mem_cgroup(child, mem)) |
dd8e8f40 | 510 | continue; |
5e9d834a DR |
511 | |
512 | /* badness() returns 0 if the thread is unkillable */ | |
513 | child_points = badness(child, uptime.tv_sec); | |
514 | if (child_points > victim_points) { | |
515 | victim = child; | |
516 | victim_points = child_points; | |
517 | } | |
dd8e8f40 ON |
518 | } |
519 | } while_each_thread(p, t); | |
520 | ||
5e9d834a | 521 | return oom_kill_task(victim); |
1da177e4 LT |
522 | } |
523 | ||
00f0b825 | 524 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
c7ba5c9e PE |
525 | void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) |
526 | { | |
527 | unsigned long points = 0; | |
528 | struct task_struct *p; | |
529 | ||
daaf1e68 KH |
530 | if (sysctl_panic_on_oom == 2) |
531 | panic("out of memory(memcg). panic_on_oom is selected.\n"); | |
e115f2d8 | 532 | read_lock(&tasklist_lock); |
c7ba5c9e | 533 | retry: |
6f48d0eb | 534 | p = select_bad_process(&points, mem, CONSTRAINT_NONE, NULL); |
df64f81b | 535 | if (!p || PTR_ERR(p) == -1UL) |
c7ba5c9e PE |
536 | goto out; |
537 | ||
fef1bdd6 | 538 | if (oom_kill_process(p, gfp_mask, 0, points, mem, |
c7ba5c9e PE |
539 | "Memory cgroup out of memory")) |
540 | goto retry; | |
541 | out: | |
e115f2d8 | 542 | read_unlock(&tasklist_lock); |
c7ba5c9e PE |
543 | } |
544 | #endif | |
545 | ||
8bc719d3 MS |
546 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
547 | ||
548 | int register_oom_notifier(struct notifier_block *nb) | |
549 | { | |
550 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
551 | } | |
552 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
553 | ||
554 | int unregister_oom_notifier(struct notifier_block *nb) | |
555 | { | |
556 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
557 | } | |
558 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
559 | ||
098d7f12 DR |
560 | /* |
561 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero | |
562 | * if a parallel OOM killing is already taking place that includes a zone in | |
563 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. | |
564 | */ | |
dd1a239f | 565 | int try_set_zone_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 566 | { |
dd1a239f MG |
567 | struct zoneref *z; |
568 | struct zone *zone; | |
098d7f12 DR |
569 | int ret = 1; |
570 | ||
c7d4caeb | 571 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
572 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
573 | if (zone_is_oom_locked(zone)) { | |
098d7f12 DR |
574 | ret = 0; |
575 | goto out; | |
576 | } | |
dd1a239f MG |
577 | } |
578 | ||
579 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
580 | /* | |
c7d4caeb | 581 | * Lock each zone in the zonelist under zone_scan_lock so a |
dd1a239f MG |
582 | * parallel invocation of try_set_zone_oom() doesn't succeed |
583 | * when it shouldn't. | |
584 | */ | |
585 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
586 | } | |
098d7f12 | 587 | |
098d7f12 | 588 | out: |
c7d4caeb | 589 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
590 | return ret; |
591 | } | |
592 | ||
593 | /* | |
594 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed | |
595 | * allocation attempts with zonelists containing them may now recall the OOM | |
596 | * killer, if necessary. | |
597 | */ | |
dd1a239f | 598 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 599 | { |
dd1a239f MG |
600 | struct zoneref *z; |
601 | struct zone *zone; | |
098d7f12 | 602 | |
c7d4caeb | 603 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
604 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
605 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
606 | } | |
c7d4caeb | 607 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
608 | } |
609 | ||
1c0fe6e3 NP |
610 | /* |
611 | * Must be called with tasklist_lock held for read. | |
612 | */ | |
6f48d0eb DR |
613 | static void __out_of_memory(gfp_t gfp_mask, int order, |
614 | enum oom_constraint constraint, const nodemask_t *mask) | |
1c0fe6e3 | 615 | { |
184101bf DR |
616 | struct task_struct *p; |
617 | unsigned long points; | |
1c0fe6e3 | 618 | |
184101bf DR |
619 | if (sysctl_oom_kill_allocating_task) |
620 | if (!oom_kill_process(current, gfp_mask, order, 0, NULL, | |
621 | "Out of memory (oom_kill_allocating_task)")) | |
1c0fe6e3 | 622 | return; |
184101bf DR |
623 | retry: |
624 | /* | |
625 | * Rambo mode: Shoot down a process and hope it solves whatever | |
626 | * issues we may have. | |
627 | */ | |
6f48d0eb | 628 | p = select_bad_process(&points, NULL, constraint, mask); |
1c0fe6e3 | 629 | |
184101bf DR |
630 | if (PTR_ERR(p) == -1UL) |
631 | return; | |
1c0fe6e3 | 632 | |
184101bf DR |
633 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
634 | if (!p) { | |
635 | read_unlock(&tasklist_lock); | |
d31f56db | 636 | dump_header(NULL, gfp_mask, order, NULL); |
184101bf | 637 | panic("Out of memory and no killable processes...\n"); |
1c0fe6e3 | 638 | } |
184101bf DR |
639 | |
640 | if (oom_kill_process(p, gfp_mask, order, points, NULL, | |
641 | "Out of memory")) | |
642 | goto retry; | |
1c0fe6e3 NP |
643 | } |
644 | ||
645 | /* | |
646 | * pagefault handler calls into here because it is out of memory but | |
647 | * doesn't know exactly how or why. | |
648 | */ | |
649 | void pagefault_out_of_memory(void) | |
650 | { | |
651 | unsigned long freed = 0; | |
652 | ||
653 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
654 | if (freed > 0) | |
655 | /* Got some memory back in the last second. */ | |
656 | return; | |
657 | ||
658 | if (sysctl_panic_on_oom) | |
659 | panic("out of memory from page fault. panic_on_oom is selected.\n"); | |
660 | ||
661 | read_lock(&tasklist_lock); | |
6f48d0eb DR |
662 | /* unknown gfp_mask and order */ |
663 | __out_of_memory(0, 0, CONSTRAINT_NONE, NULL); | |
1c0fe6e3 NP |
664 | read_unlock(&tasklist_lock); |
665 | ||
666 | /* | |
667 | * Give "p" a good chance of killing itself before we | |
668 | * retry to allocate memory. | |
669 | */ | |
670 | if (!test_thread_flag(TIF_MEMDIE)) | |
671 | schedule_timeout_uninterruptible(1); | |
672 | } | |
673 | ||
1da177e4 | 674 | /** |
6937a25c | 675 | * out_of_memory - kill the "best" process when we run out of memory |
1b578df0 RD |
676 | * @zonelist: zonelist pointer |
677 | * @gfp_mask: memory allocation flags | |
678 | * @order: amount of memory being requested as a power of 2 | |
6f48d0eb | 679 | * @nodemask: nodemask passed to page allocator |
1da177e4 LT |
680 | * |
681 | * If we run out of memory, we have the choice between either | |
682 | * killing a random task (bad), letting the system crash (worse) | |
683 | * OR try to be smart about which process to kill. Note that we | |
684 | * don't have to be perfect here, we just have to be good. | |
685 | */ | |
4365a567 KH |
686 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
687 | int order, nodemask_t *nodemask) | |
1da177e4 | 688 | { |
8bc719d3 | 689 | unsigned long freed = 0; |
70e24bdf | 690 | enum oom_constraint constraint; |
8bc719d3 MS |
691 | |
692 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
693 | if (freed > 0) | |
694 | /* Got some memory back in the last second. */ | |
695 | return; | |
1da177e4 | 696 | |
7b98c2e4 DR |
697 | /* |
698 | * If current has a pending SIGKILL, then automatically select it. The | |
699 | * goal is to allow it to allocate so that it may quickly exit and free | |
700 | * its memory. | |
701 | */ | |
702 | if (fatal_signal_pending(current)) { | |
703 | set_thread_flag(TIF_MEMDIE); | |
704 | return; | |
705 | } | |
706 | ||
1b604d75 | 707 | if (sysctl_panic_on_oom == 2) { |
d31f56db | 708 | dump_header(NULL, gfp_mask, order, NULL); |
2b744c01 | 709 | panic("out of memory. Compulsory panic_on_oom is selected.\n"); |
1b604d75 | 710 | } |
2b744c01 | 711 | |
9b0f8b04 CL |
712 | /* |
713 | * Check if there were limitations on the allocation (only relevant for | |
714 | * NUMA) that may require different handling. | |
715 | */ | |
4365a567 | 716 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask); |
2b45ab33 | 717 | read_lock(&tasklist_lock); |
6f48d0eb DR |
718 | if (unlikely(sysctl_panic_on_oom)) { |
719 | /* | |
720 | * panic_on_oom only affects CONSTRAINT_NONE, the kernel | |
721 | * should not panic for cpuset or mempolicy induced memory | |
722 | * failures. | |
723 | */ | |
724 | if (constraint == CONSTRAINT_NONE) { | |
d31f56db | 725 | dump_header(NULL, gfp_mask, order, NULL); |
6f48d0eb DR |
726 | read_unlock(&tasklist_lock); |
727 | panic("Out of memory: panic_on_oom is enabled\n"); | |
1b604d75 | 728 | } |
9b0f8b04 | 729 | } |
6f48d0eb | 730 | __out_of_memory(gfp_mask, order, constraint, nodemask); |
140ffcec | 731 | read_unlock(&tasklist_lock); |
1da177e4 LT |
732 | |
733 | /* | |
734 | * Give "p" a good chance of killing itself before we | |
2f659f46 | 735 | * retry to allocate memory unless "p" is current |
1da177e4 | 736 | */ |
2f659f46 | 737 | if (!test_thread_flag(TIF_MEMDIE)) |
140ffcec | 738 | schedule_timeout_uninterruptible(1); |
1da177e4 | 739 | } |