]>
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> |
5cd9c58f | 30 | #include <linux/security.h> |
1da177e4 | 31 | |
fadd8fbd | 32 | int sysctl_panic_on_oom; |
fe071d7e | 33 | int sysctl_oom_kill_allocating_task; |
fef1bdd6 | 34 | int sysctl_oom_dump_tasks; |
c7d4caeb | 35 | static DEFINE_SPINLOCK(zone_scan_lock); |
1da177e4 LT |
36 | /* #define DEBUG */ |
37 | ||
495789a5 KM |
38 | /* |
39 | * Is all threads of the target process nodes overlap ours? | |
40 | */ | |
41 | static int has_intersects_mems_allowed(struct task_struct *tsk) | |
42 | { | |
43 | struct task_struct *t; | |
44 | ||
45 | t = tsk; | |
46 | do { | |
47 | if (cpuset_mems_allowed_intersects(current, t)) | |
48 | return 1; | |
49 | t = next_thread(t); | |
50 | } while (t != tsk); | |
51 | ||
52 | return 0; | |
53 | } | |
54 | ||
dd8e8f40 ON |
55 | static struct task_struct *find_lock_task_mm(struct task_struct *p) |
56 | { | |
57 | struct task_struct *t = p; | |
58 | ||
59 | do { | |
60 | task_lock(t); | |
61 | if (likely(t->mm)) | |
62 | return t; | |
63 | task_unlock(t); | |
64 | } while_each_thread(p, t); | |
65 | ||
66 | return NULL; | |
67 | } | |
68 | ||
1da177e4 | 69 | /** |
6937a25c | 70 | * badness - calculate a numeric value for how bad this task has been |
1da177e4 | 71 | * @p: task struct of which task we should calculate |
a49335cc | 72 | * @uptime: current uptime in seconds |
1da177e4 LT |
73 | * |
74 | * The formula used is relatively simple and documented inline in the | |
75 | * function. The main rationale is that we want to select a good task | |
76 | * to kill when we run out of memory. | |
77 | * | |
78 | * Good in this context means that: | |
79 | * 1) we lose the minimum amount of work done | |
80 | * 2) we recover a large amount of memory | |
81 | * 3) we don't kill anything innocent of eating tons of memory | |
82 | * 4) we want to kill the minimum amount of processes (one) | |
83 | * 5) we try to kill the process the user expects us to kill, this | |
84 | * algorithm has been meticulously tuned to meet the principle | |
85 | * of least surprise ... (be careful when you change it) | |
86 | */ | |
87 | ||
97d87c97 | 88 | unsigned long badness(struct task_struct *p, unsigned long uptime) |
1da177e4 | 89 | { |
a12888f7 | 90 | unsigned long points, cpu_time, run_time; |
97c2c9b8 | 91 | struct task_struct *child; |
dd8e8f40 | 92 | struct task_struct *c, *t; |
28b83c51 | 93 | int oom_adj = p->signal->oom_adj; |
495789a5 KM |
94 | struct task_cputime task_time; |
95 | unsigned long utime; | |
96 | unsigned long stime; | |
28b83c51 KM |
97 | |
98 | if (oom_adj == OOM_DISABLE) | |
99 | return 0; | |
1da177e4 | 100 | |
dd8e8f40 ON |
101 | p = find_lock_task_mm(p); |
102 | if (!p) | |
1da177e4 LT |
103 | return 0; |
104 | ||
105 | /* | |
106 | * The memory size of the process is the basis for the badness. | |
107 | */ | |
dd8e8f40 | 108 | points = p->mm->total_vm; |
97c2c9b8 AM |
109 | |
110 | /* | |
111 | * After this unlock we can no longer dereference local variable `mm' | |
112 | */ | |
113 | task_unlock(p); | |
1da177e4 | 114 | |
7ba34859 HD |
115 | /* |
116 | * swapoff can easily use up all memory, so kill those first. | |
117 | */ | |
35451bee | 118 | if (p->flags & PF_OOM_ORIGIN) |
7ba34859 HD |
119 | return ULONG_MAX; |
120 | ||
1da177e4 LT |
121 | /* |
122 | * Processes which fork a lot of child processes are likely | |
9827b781 | 123 | * a good choice. We add half the vmsize of the children if they |
1da177e4 | 124 | * have an own mm. This prevents forking servers to flood the |
9827b781 KG |
125 | * machine with an endless amount of children. In case a single |
126 | * child is eating the vast majority of memory, adding only half | |
127 | * to the parents will make the child our kill candidate of choice. | |
1da177e4 | 128 | */ |
dd8e8f40 ON |
129 | t = p; |
130 | do { | |
131 | list_for_each_entry(c, &t->children, sibling) { | |
132 | child = find_lock_task_mm(c); | |
133 | if (child) { | |
134 | if (child->mm != p->mm) | |
135 | points += child->mm->total_vm/2 + 1; | |
136 | task_unlock(child); | |
137 | } | |
138 | } | |
139 | } while_each_thread(p, t); | |
1da177e4 LT |
140 | |
141 | /* | |
142 | * CPU time is in tens of seconds and run time is in thousands | |
143 | * of seconds. There is no particular reason for this other than | |
144 | * that it turned out to work very well in practice. | |
145 | */ | |
495789a5 KM |
146 | thread_group_cputime(p, &task_time); |
147 | utime = cputime_to_jiffies(task_time.utime); | |
148 | stime = cputime_to_jiffies(task_time.stime); | |
149 | cpu_time = (utime + stime) >> (SHIFT_HZ + 3); | |
150 | ||
1da177e4 LT |
151 | |
152 | if (uptime >= p->start_time.tv_sec) | |
153 | run_time = (uptime - p->start_time.tv_sec) >> 10; | |
154 | else | |
155 | run_time = 0; | |
156 | ||
a12888f7 CG |
157 | if (cpu_time) |
158 | points /= int_sqrt(cpu_time); | |
159 | if (run_time) | |
160 | points /= int_sqrt(int_sqrt(run_time)); | |
1da177e4 LT |
161 | |
162 | /* | |
163 | * Niced processes are most likely less important, so double | |
164 | * their badness points. | |
165 | */ | |
166 | if (task_nice(p) > 0) | |
167 | points *= 2; | |
168 | ||
169 | /* | |
170 | * Superuser processes are usually more important, so we make it | |
171 | * less likely that we kill those. | |
172 | */ | |
a2f2945a EP |
173 | if (has_capability_noaudit(p, CAP_SYS_ADMIN) || |
174 | has_capability_noaudit(p, CAP_SYS_RESOURCE)) | |
1da177e4 LT |
175 | points /= 4; |
176 | ||
177 | /* | |
178 | * We don't want to kill a process with direct hardware access. | |
179 | * Not only could that mess up the hardware, but usually users | |
180 | * tend to only have this flag set on applications they think | |
181 | * of as important. | |
182 | */ | |
a2f2945a | 183 | if (has_capability_noaudit(p, CAP_SYS_RAWIO)) |
1da177e4 LT |
184 | points /= 4; |
185 | ||
186 | /* | |
28b83c51 | 187 | * Adjust the score by oom_adj. |
1da177e4 | 188 | */ |
28b83c51 KM |
189 | if (oom_adj) { |
190 | if (oom_adj > 0) { | |
9a82782f JP |
191 | if (!points) |
192 | points = 1; | |
28b83c51 | 193 | points <<= oom_adj; |
9a82782f | 194 | } else |
28b83c51 | 195 | points >>= -(oom_adj); |
1da177e4 LT |
196 | } |
197 | ||
198 | #ifdef DEBUG | |
a5e58a61 | 199 | printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n", |
1da177e4 LT |
200 | p->pid, p->comm, points); |
201 | #endif | |
202 | return points; | |
203 | } | |
204 | ||
9b0f8b04 CL |
205 | /* |
206 | * Determine the type of allocation constraint. | |
207 | */ | |
9b0f8b04 | 208 | #ifdef CONFIG_NUMA |
4365a567 KH |
209 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
210 | gfp_t gfp_mask, nodemask_t *nodemask) | |
211 | { | |
54a6eb5c | 212 | struct zone *zone; |
dd1a239f | 213 | struct zoneref *z; |
54a6eb5c | 214 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
9b0f8b04 | 215 | |
4365a567 KH |
216 | /* |
217 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid | |
218 | * to kill current.We have to random task kill in this case. | |
219 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. | |
220 | */ | |
221 | if (gfp_mask & __GFP_THISNODE) | |
222 | return CONSTRAINT_NONE; | |
9b0f8b04 | 223 | |
4365a567 KH |
224 | /* |
225 | * The nodemask here is a nodemask passed to alloc_pages(). Now, | |
226 | * cpuset doesn't use this nodemask for its hardwall/softwall/hierarchy | |
227 | * feature. mempolicy is an only user of nodemask here. | |
228 | * check mempolicy's nodemask contains all N_HIGH_MEMORY | |
229 | */ | |
230 | if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) | |
9b0f8b04 | 231 | return CONSTRAINT_MEMORY_POLICY; |
4365a567 KH |
232 | |
233 | /* Check this allocation failure is caused by cpuset's wall function */ | |
234 | for_each_zone_zonelist_nodemask(zone, z, zonelist, | |
235 | high_zoneidx, nodemask) | |
236 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) | |
237 | return CONSTRAINT_CPUSET; | |
9b0f8b04 CL |
238 | |
239 | return CONSTRAINT_NONE; | |
240 | } | |
4365a567 KH |
241 | #else |
242 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, | |
243 | gfp_t gfp_mask, nodemask_t *nodemask) | |
244 | { | |
245 | return CONSTRAINT_NONE; | |
246 | } | |
247 | #endif | |
9b0f8b04 | 248 | |
1da177e4 LT |
249 | /* |
250 | * Simple selection loop. We chose the process with the highest | |
251 | * number of 'points'. We expect the caller will lock the tasklist. | |
252 | * | |
253 | * (not docbooked, we don't want this one cluttering up the manual) | |
254 | */ | |
c7ba5c9e PE |
255 | static struct task_struct *select_bad_process(unsigned long *ppoints, |
256 | struct mem_cgroup *mem) | |
1da177e4 | 257 | { |
495789a5 | 258 | struct task_struct *p; |
1da177e4 LT |
259 | struct task_struct *chosen = NULL; |
260 | struct timespec uptime; | |
9827b781 | 261 | *ppoints = 0; |
1da177e4 LT |
262 | |
263 | do_posix_clock_monotonic_gettime(&uptime); | |
495789a5 | 264 | for_each_process(p) { |
a49335cc | 265 | unsigned long points; |
a49335cc | 266 | |
455c0e5f ON |
267 | /* skip the init task and kthreads */ |
268 | if (is_global_init(p) || (p->flags & PF_KTHREAD)) | |
a49335cc | 269 | continue; |
4c4a2214 DR |
270 | if (mem && !task_in_mem_cgroup(p, mem)) |
271 | continue; | |
6cf86ac6 DR |
272 | if (!has_intersects_mems_allowed(p)) |
273 | continue; | |
ef08e3b4 | 274 | |
b78483a4 NP |
275 | /* |
276 | * This task already has access to memory reserves and is | |
277 | * being killed. Don't allow any other task access to the | |
278 | * memory reserve. | |
279 | * | |
280 | * Note: this may have a chance of deadlock if it gets | |
281 | * blocked waiting for another task which itself is waiting | |
282 | * for memory. Is there a better alternative? | |
283 | */ | |
284 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) | |
285 | return ERR_PTR(-1UL); | |
286 | ||
a49335cc | 287 | /* |
6937a25c | 288 | * This is in the process of releasing memory so wait for it |
a49335cc | 289 | * to finish before killing some other task by mistake. |
50ec3bbf NP |
290 | * |
291 | * However, if p is the current task, we allow the 'kill' to | |
292 | * go ahead if it is exiting: this will simply set TIF_MEMDIE, | |
293 | * which will allow it to gain access to memory reserves in | |
294 | * the process of exiting and releasing its resources. | |
b78483a4 | 295 | * Otherwise we could get an easy OOM deadlock. |
a49335cc | 296 | */ |
b5227940 | 297 | if ((p->flags & PF_EXITING) && p->mm) { |
b78483a4 NP |
298 | if (p != current) |
299 | return ERR_PTR(-1UL); | |
300 | ||
972c4ea5 ON |
301 | chosen = p; |
302 | *ppoints = ULONG_MAX; | |
50ec3bbf | 303 | } |
972c4ea5 | 304 | |
28b83c51 | 305 | if (p->signal->oom_adj == OOM_DISABLE) |
0753ba01 KM |
306 | continue; |
307 | ||
97d87c97 | 308 | points = badness(p, uptime.tv_sec); |
0753ba01 | 309 | if (points > *ppoints || !chosen) { |
a49335cc | 310 | chosen = p; |
9827b781 | 311 | *ppoints = points; |
1da177e4 | 312 | } |
495789a5 | 313 | } |
972c4ea5 | 314 | |
1da177e4 LT |
315 | return chosen; |
316 | } | |
317 | ||
fef1bdd6 | 318 | /** |
1b578df0 | 319 | * dump_tasks - dump current memory state of all system tasks |
74ab7f1d | 320 | * @mem: current's memory controller, if constrained |
1b578df0 | 321 | * |
fef1bdd6 DR |
322 | * Dumps the current memory state of all system tasks, excluding kernel threads. |
323 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj | |
324 | * score, and name. | |
325 | * | |
326 | * If the actual is non-NULL, only tasks that are a member of the mem_cgroup are | |
327 | * shown. | |
328 | * | |
329 | * Call with tasklist_lock read-locked. | |
330 | */ | |
331 | static void dump_tasks(const struct mem_cgroup *mem) | |
332 | { | |
c55db957 KM |
333 | struct task_struct *p; |
334 | struct task_struct *task; | |
fef1bdd6 DR |
335 | |
336 | printk(KERN_INFO "[ pid ] uid tgid total_vm rss cpu oom_adj " | |
337 | "name\n"); | |
c55db957 | 338 | for_each_process(p) { |
c55db957 | 339 | if (p->flags & PF_KTHREAD) |
fef1bdd6 | 340 | continue; |
c55db957 | 341 | if (mem && !task_in_mem_cgroup(p, mem)) |
b4416d2b | 342 | continue; |
fef1bdd6 | 343 | |
c55db957 KM |
344 | task = find_lock_task_mm(p); |
345 | if (!task) { | |
6d2661ed | 346 | /* |
74ab7f1d DR |
347 | * This is a kthread or all of p's threads have already |
348 | * detached their mm's. There's no need to report | |
c55db957 | 349 | * them; they can't be oom killed anyway. |
6d2661ed | 350 | */ |
6d2661ed DR |
351 | continue; |
352 | } | |
c55db957 | 353 | |
c81fac5c | 354 | printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3u %3d %s\n", |
c55db957 KM |
355 | task->pid, __task_cred(task)->uid, task->tgid, |
356 | task->mm->total_vm, get_mm_rss(task->mm), | |
c81fac5c | 357 | task_cpu(task), task->signal->oom_adj, task->comm); |
c55db957 KM |
358 | task_unlock(task); |
359 | } | |
fef1bdd6 DR |
360 | } |
361 | ||
d31f56db DN |
362 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
363 | struct mem_cgroup *mem) | |
1b604d75 DR |
364 | { |
365 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " | |
366 | "oom_adj=%d\n", | |
367 | current->comm, gfp_mask, order, current->signal->oom_adj); | |
368 | task_lock(current); | |
369 | cpuset_print_task_mems_allowed(current); | |
370 | task_unlock(current); | |
371 | dump_stack(); | |
d31f56db | 372 | mem_cgroup_print_oom_info(mem, p); |
1b604d75 DR |
373 | show_mem(); |
374 | if (sysctl_oom_dump_tasks) | |
375 | dump_tasks(mem); | |
376 | } | |
377 | ||
3b4798cb KM |
378 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
379 | ||
1b578df0 | 380 | /* |
5a291b98 RG |
381 | * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO |
382 | * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO | |
383 | * set. | |
1da177e4 | 384 | */ |
f3af38d3 | 385 | static void __oom_kill_task(struct task_struct *p, int verbose) |
1da177e4 | 386 | { |
b460cbc5 | 387 | if (is_global_init(p)) { |
1da177e4 LT |
388 | WARN_ON(1); |
389 | printk(KERN_WARNING "tried to kill init!\n"); | |
390 | return; | |
391 | } | |
392 | ||
dd8e8f40 ON |
393 | p = find_lock_task_mm(p); |
394 | if (!p) | |
1da177e4 | 395 | return; |
50ec3bbf | 396 | |
f3af38d3 | 397 | if (verbose) |
3b4798cb KM |
398 | printk(KERN_ERR "Killed process %d (%s) " |
399 | "vsz:%lukB, anon-rss:%lukB, file-rss:%lukB\n", | |
400 | task_pid_nr(p), p->comm, | |
401 | K(p->mm->total_vm), | |
d559db08 KH |
402 | K(get_mm_counter(p->mm, MM_ANONPAGES)), |
403 | K(get_mm_counter(p->mm, MM_FILEPAGES))); | |
3b4798cb | 404 | task_unlock(p); |
1da177e4 LT |
405 | |
406 | /* | |
407 | * We give our sacrificial lamb high priority and access to | |
408 | * all the memory it needs. That way it should be able to | |
409 | * exit() and clear out its resources quickly... | |
410 | */ | |
fa717060 | 411 | p->rt.time_slice = HZ; |
1da177e4 LT |
412 | set_tsk_thread_flag(p, TIF_MEMDIE); |
413 | ||
414 | force_sig(SIGKILL, p); | |
415 | } | |
416 | ||
f3af38d3 | 417 | static int oom_kill_task(struct task_struct *p) |
1da177e4 | 418 | { |
0753ba01 KM |
419 | /* WARNING: mm may not be dereferenced since we did not obtain its |
420 | * value from get_task_mm(p). This is OK since all we need to do is | |
421 | * compare mm to q->mm below. | |
422 | * | |
423 | * Furthermore, even if mm contains a non-NULL value, p->mm may | |
424 | * change to NULL at any time since we do not hold task_lock(p). | |
425 | * However, this is of no concern to us. | |
426 | */ | |
8c5cd6f3 | 427 | if (!p->mm || p->signal->oom_adj == OOM_DISABLE) |
01315922 | 428 | return 1; |
0753ba01 | 429 | |
f3af38d3 | 430 | __oom_kill_task(p, 1); |
c33e0fca | 431 | |
01315922 | 432 | return 0; |
1da177e4 LT |
433 | } |
434 | ||
7213f506 | 435 | static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
fef1bdd6 DR |
436 | unsigned long points, struct mem_cgroup *mem, |
437 | const char *message) | |
1da177e4 | 438 | { |
1da177e4 | 439 | struct task_struct *c; |
dd8e8f40 | 440 | struct task_struct *t = p; |
1da177e4 | 441 | |
1b604d75 | 442 | if (printk_ratelimit()) |
d31f56db | 443 | dump_header(p, gfp_mask, order, mem); |
7213f506 | 444 | |
50ec3bbf NP |
445 | /* |
446 | * If the task is already exiting, don't alarm the sysadmin or kill | |
447 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
448 | */ | |
0753ba01 | 449 | if (p->flags & PF_EXITING) { |
4358997a | 450 | set_tsk_thread_flag(p, TIF_MEMDIE); |
50ec3bbf NP |
451 | return 0; |
452 | } | |
453 | ||
f3af38d3 | 454 | printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n", |
ba25f9dc | 455 | message, task_pid_nr(p), p->comm, points); |
f3af38d3 | 456 | |
1da177e4 | 457 | /* Try to kill a child first */ |
dd8e8f40 ON |
458 | do { |
459 | list_for_each_entry(c, &t->children, sibling) { | |
460 | if (c->mm == p->mm) | |
461 | continue; | |
462 | if (mem && !task_in_mem_cgroup(c, mem)) | |
463 | continue; | |
464 | if (!oom_kill_task(c)) | |
465 | return 0; | |
466 | } | |
467 | } while_each_thread(p, t); | |
468 | ||
f3af38d3 | 469 | return oom_kill_task(p); |
1da177e4 LT |
470 | } |
471 | ||
00f0b825 | 472 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
c7ba5c9e PE |
473 | void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) |
474 | { | |
475 | unsigned long points = 0; | |
476 | struct task_struct *p; | |
477 | ||
daaf1e68 KH |
478 | if (sysctl_panic_on_oom == 2) |
479 | panic("out of memory(memcg). panic_on_oom is selected.\n"); | |
e115f2d8 | 480 | read_lock(&tasklist_lock); |
c7ba5c9e PE |
481 | retry: |
482 | p = select_bad_process(&points, mem); | |
df64f81b | 483 | if (!p || PTR_ERR(p) == -1UL) |
c7ba5c9e PE |
484 | goto out; |
485 | ||
fef1bdd6 | 486 | if (oom_kill_process(p, gfp_mask, 0, points, mem, |
c7ba5c9e PE |
487 | "Memory cgroup out of memory")) |
488 | goto retry; | |
489 | out: | |
e115f2d8 | 490 | read_unlock(&tasklist_lock); |
c7ba5c9e PE |
491 | } |
492 | #endif | |
493 | ||
8bc719d3 MS |
494 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
495 | ||
496 | int register_oom_notifier(struct notifier_block *nb) | |
497 | { | |
498 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
499 | } | |
500 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
501 | ||
502 | int unregister_oom_notifier(struct notifier_block *nb) | |
503 | { | |
504 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
505 | } | |
506 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
507 | ||
098d7f12 DR |
508 | /* |
509 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero | |
510 | * if a parallel OOM killing is already taking place that includes a zone in | |
511 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. | |
512 | */ | |
dd1a239f | 513 | int try_set_zone_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 514 | { |
dd1a239f MG |
515 | struct zoneref *z; |
516 | struct zone *zone; | |
098d7f12 DR |
517 | int ret = 1; |
518 | ||
c7d4caeb | 519 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
520 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
521 | if (zone_is_oom_locked(zone)) { | |
098d7f12 DR |
522 | ret = 0; |
523 | goto out; | |
524 | } | |
dd1a239f MG |
525 | } |
526 | ||
527 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
528 | /* | |
c7d4caeb | 529 | * Lock each zone in the zonelist under zone_scan_lock so a |
dd1a239f MG |
530 | * parallel invocation of try_set_zone_oom() doesn't succeed |
531 | * when it shouldn't. | |
532 | */ | |
533 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
534 | } | |
098d7f12 | 535 | |
098d7f12 | 536 | out: |
c7d4caeb | 537 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
538 | return ret; |
539 | } | |
540 | ||
541 | /* | |
542 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed | |
543 | * allocation attempts with zonelists containing them may now recall the OOM | |
544 | * killer, if necessary. | |
545 | */ | |
dd1a239f | 546 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 547 | { |
dd1a239f MG |
548 | struct zoneref *z; |
549 | struct zone *zone; | |
098d7f12 | 550 | |
c7d4caeb | 551 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
552 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
553 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
554 | } | |
c7d4caeb | 555 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
556 | } |
557 | ||
1c0fe6e3 NP |
558 | /* |
559 | * Must be called with tasklist_lock held for read. | |
560 | */ | |
561 | static void __out_of_memory(gfp_t gfp_mask, int order) | |
562 | { | |
184101bf DR |
563 | struct task_struct *p; |
564 | unsigned long points; | |
1c0fe6e3 | 565 | |
184101bf DR |
566 | if (sysctl_oom_kill_allocating_task) |
567 | if (!oom_kill_process(current, gfp_mask, order, 0, NULL, | |
568 | "Out of memory (oom_kill_allocating_task)")) | |
1c0fe6e3 | 569 | return; |
184101bf DR |
570 | retry: |
571 | /* | |
572 | * Rambo mode: Shoot down a process and hope it solves whatever | |
573 | * issues we may have. | |
574 | */ | |
575 | p = select_bad_process(&points, NULL); | |
1c0fe6e3 | 576 | |
184101bf DR |
577 | if (PTR_ERR(p) == -1UL) |
578 | return; | |
1c0fe6e3 | 579 | |
184101bf DR |
580 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
581 | if (!p) { | |
582 | read_unlock(&tasklist_lock); | |
d31f56db | 583 | dump_header(NULL, gfp_mask, order, NULL); |
184101bf | 584 | panic("Out of memory and no killable processes...\n"); |
1c0fe6e3 | 585 | } |
184101bf DR |
586 | |
587 | if (oom_kill_process(p, gfp_mask, order, points, NULL, | |
588 | "Out of memory")) | |
589 | goto retry; | |
1c0fe6e3 NP |
590 | } |
591 | ||
592 | /* | |
593 | * pagefault handler calls into here because it is out of memory but | |
594 | * doesn't know exactly how or why. | |
595 | */ | |
596 | void pagefault_out_of_memory(void) | |
597 | { | |
598 | unsigned long freed = 0; | |
599 | ||
600 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
601 | if (freed > 0) | |
602 | /* Got some memory back in the last second. */ | |
603 | return; | |
604 | ||
605 | if (sysctl_panic_on_oom) | |
606 | panic("out of memory from page fault. panic_on_oom is selected.\n"); | |
607 | ||
608 | read_lock(&tasklist_lock); | |
609 | __out_of_memory(0, 0); /* unknown gfp_mask and order */ | |
610 | read_unlock(&tasklist_lock); | |
611 | ||
612 | /* | |
613 | * Give "p" a good chance of killing itself before we | |
614 | * retry to allocate memory. | |
615 | */ | |
616 | if (!test_thread_flag(TIF_MEMDIE)) | |
617 | schedule_timeout_uninterruptible(1); | |
618 | } | |
619 | ||
1da177e4 | 620 | /** |
6937a25c | 621 | * out_of_memory - kill the "best" process when we run out of memory |
1b578df0 RD |
622 | * @zonelist: zonelist pointer |
623 | * @gfp_mask: memory allocation flags | |
624 | * @order: amount of memory being requested as a power of 2 | |
1da177e4 LT |
625 | * |
626 | * If we run out of memory, we have the choice between either | |
627 | * killing a random task (bad), letting the system crash (worse) | |
628 | * OR try to be smart about which process to kill. Note that we | |
629 | * don't have to be perfect here, we just have to be good. | |
630 | */ | |
4365a567 KH |
631 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
632 | int order, nodemask_t *nodemask) | |
1da177e4 | 633 | { |
8bc719d3 | 634 | unsigned long freed = 0; |
70e24bdf | 635 | enum oom_constraint constraint; |
8bc719d3 MS |
636 | |
637 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
638 | if (freed > 0) | |
639 | /* Got some memory back in the last second. */ | |
640 | return; | |
1da177e4 | 641 | |
7b98c2e4 DR |
642 | /* |
643 | * If current has a pending SIGKILL, then automatically select it. The | |
644 | * goal is to allow it to allocate so that it may quickly exit and free | |
645 | * its memory. | |
646 | */ | |
647 | if (fatal_signal_pending(current)) { | |
648 | set_thread_flag(TIF_MEMDIE); | |
649 | return; | |
650 | } | |
651 | ||
1b604d75 | 652 | if (sysctl_panic_on_oom == 2) { |
d31f56db | 653 | dump_header(NULL, gfp_mask, order, NULL); |
2b744c01 | 654 | panic("out of memory. Compulsory panic_on_oom is selected.\n"); |
1b604d75 | 655 | } |
2b744c01 | 656 | |
9b0f8b04 CL |
657 | /* |
658 | * Check if there were limitations on the allocation (only relevant for | |
659 | * NUMA) that may require different handling. | |
660 | */ | |
4365a567 | 661 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask); |
2b45ab33 DR |
662 | read_lock(&tasklist_lock); |
663 | ||
664 | switch (constraint) { | |
9b0f8b04 | 665 | case CONSTRAINT_MEMORY_POLICY: |
1c0fe6e3 | 666 | oom_kill_process(current, gfp_mask, order, 0, NULL, |
9b0f8b04 CL |
667 | "No available memory (MPOL_BIND)"); |
668 | break; | |
669 | ||
9b0f8b04 | 670 | case CONSTRAINT_NONE: |
1b604d75 | 671 | if (sysctl_panic_on_oom) { |
d31f56db | 672 | dump_header(NULL, gfp_mask, order, NULL); |
fadd8fbd | 673 | panic("out of memory. panic_on_oom is selected\n"); |
1b604d75 | 674 | } |
fe071d7e DR |
675 | /* Fall-through */ |
676 | case CONSTRAINT_CPUSET: | |
1c0fe6e3 | 677 | __out_of_memory(gfp_mask, order); |
9b0f8b04 CL |
678 | break; |
679 | } | |
1da177e4 | 680 | |
140ffcec | 681 | read_unlock(&tasklist_lock); |
1da177e4 LT |
682 | |
683 | /* | |
684 | * Give "p" a good chance of killing itself before we | |
2f659f46 | 685 | * retry to allocate memory unless "p" is current |
1da177e4 | 686 | */ |
2f659f46 | 687 | if (!test_thread_flag(TIF_MEMDIE)) |
140ffcec | 688 | schedule_timeout_uninterruptible(1); |
1da177e4 | 689 | } |