blob: cd6520f7553d80a157e0cf15123b312891bfa929 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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...
David Rientjesa63d83f2010-08-09 17:19:46 -07007 * Copyright (C) 2010 Google, Inc.
8 * Rewritten by David Rientjes
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
10 * The routines in this file are used to kill a process when
Paul Jacksona49335c2005-09-06 15:18:09 -070011 * we're seriously out of memory. This gets called from __alloc_pages()
12 * in mm/page_alloc.c when we really run out of memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
14 * Since we won't call these routines often (on a well-configured
15 * machine) this file will double as a 'coding guide' and a signpost
16 * for newbie kernel hackers. It features several pointers to major
17 * kernel subsystems and hints as to where to find out what things do.
18 */
19
Alexey Dobriyan8ac773b2006-10-19 23:28:32 -070020#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/mm.h>
Alexey Dobriyan4e950f62007-07-30 02:36:13 +040022#include <linux/err.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090023#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/sched.h>
Ingo Molnar6e84f312017-02-08 18:51:29 +010025#include <linux/sched/mm.h>
Ingo Molnarf7ccbae2017-02-08 18:51:30 +010026#include <linux/sched/coredump.h>
Ingo Molnar29930022017-02-08 18:51:36 +010027#include <linux/sched/task.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/swap.h>
29#include <linux/timex.h>
30#include <linux/jiffies.h>
Paul Jacksonef08e3b2005-09-06 15:18:13 -070031#include <linux/cpuset.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040032#include <linux/export.h>
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -070033#include <linux/notifier.h>
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -080034#include <linux/memcontrol.h>
David Rientjes6f48d0eb2010-08-09 17:18:52 -070035#include <linux/mempolicy.h>
David Howells5cd9c582008-08-14 11:37:28 +010036#include <linux/security.h>
David Rientjesedd45542011-03-22 16:30:12 -070037#include <linux/ptrace.h>
David Rientjesf660daa2011-10-31 17:07:07 -070038#include <linux/freezer.h>
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080039#include <linux/ftrace.h>
David Rientjesdc3f21e2012-03-21 16:33:47 -070040#include <linux/ratelimit.h>
Michal Hockoaac45362016-03-25 14:20:24 -070041#include <linux/kthread.h>
42#include <linux/init.h>
Michal Hocko4d4bbd82017-10-03 16:14:50 -070043#include <linux/mmu_notifier.h>
Michal Hockoaac45362016-03-25 14:20:24 -070044
45#include <asm/tlb.h>
46#include "internal.h"
Yang Shi852d8be2017-11-15 17:32:07 -080047#include "slab.h"
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080048
49#define CREATE_TRACE_POINTS
50#include <trace/events/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -070052int sysctl_panic_on_oom;
David Rientjesfe071d72007-10-16 23:25:56 -070053int sysctl_oom_kill_allocating_task;
David Rientjesad915c42010-08-09 17:18:53 -070054int sysctl_oom_dump_tasks = 1;
Johannes Weinerdc564012015-06-24 16:57:19 -070055
56DEFINE_MUTEX(oom_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
David Rientjes6f48d0eb2010-08-09 17:18:52 -070058#ifdef CONFIG_NUMA
59/**
60 * has_intersects_mems_allowed() - check task eligiblity for kill
Oleg Nesterovad962442014-01-21 15:50:00 -080061 * @start: task struct of which task to consider
David Rientjes6f48d0eb2010-08-09 17:18:52 -070062 * @mask: nodemask passed to page allocator for mempolicy ooms
63 *
64 * Task eligibility is determined by whether or not a candidate task, @tsk,
65 * shares the same mempolicy nodes as current if it is bound by such a policy
66 * and whether or not it has the same set of allowed cpuset nodes.
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070067 */
Oleg Nesterovad962442014-01-21 15:50:00 -080068static bool has_intersects_mems_allowed(struct task_struct *start,
David Rientjes6f48d0eb2010-08-09 17:18:52 -070069 const nodemask_t *mask)
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070070{
Oleg Nesterovad962442014-01-21 15:50:00 -080071 struct task_struct *tsk;
72 bool ret = false;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070073
Oleg Nesterovad962442014-01-21 15:50:00 -080074 rcu_read_lock();
Oleg Nesterov1da4db02014-01-21 15:49:58 -080075 for_each_thread(start, tsk) {
David Rientjes6f48d0eb2010-08-09 17:18:52 -070076 if (mask) {
77 /*
78 * If this is a mempolicy constrained oom, tsk's
79 * cpuset is irrelevant. Only return true if its
80 * mempolicy intersects current, otherwise it may be
81 * needlessly killed.
82 */
Oleg Nesterovad962442014-01-21 15:50:00 -080083 ret = mempolicy_nodemask_intersects(tsk, mask);
David Rientjes6f48d0eb2010-08-09 17:18:52 -070084 } else {
85 /*
86 * This is not a mempolicy constrained oom, so only
87 * check the mems of tsk's cpuset.
88 */
Oleg Nesterovad962442014-01-21 15:50:00 -080089 ret = cpuset_mems_allowed_intersects(current, tsk);
David Rientjes6f48d0eb2010-08-09 17:18:52 -070090 }
Oleg Nesterovad962442014-01-21 15:50:00 -080091 if (ret)
92 break;
Oleg Nesterov1da4db02014-01-21 15:49:58 -080093 }
Oleg Nesterovad962442014-01-21 15:50:00 -080094 rcu_read_unlock();
KOSAKI Motohirodf1090a2010-08-09 17:19:39 -070095
Oleg Nesterovad962442014-01-21 15:50:00 -080096 return ret;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070097}
David Rientjes6f48d0eb2010-08-09 17:18:52 -070098#else
99static bool has_intersects_mems_allowed(struct task_struct *tsk,
100 const nodemask_t *mask)
101{
102 return true;
103}
104#endif /* CONFIG_NUMA */
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700105
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700106/*
107 * The process p may have detached its own ->mm while exiting or through
108 * use_mm(), but one or more of its subthreads may still have a valid
109 * pointer. Return p, or any of its subthreads with a valid ->mm, with
110 * task_lock() held.
111 */
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -0700112struct task_struct *find_lock_task_mm(struct task_struct *p)
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700113{
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800114 struct task_struct *t;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700115
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800116 rcu_read_lock();
117
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800118 for_each_thread(p, t) {
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700119 task_lock(t);
120 if (likely(t->mm))
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800121 goto found;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700122 task_unlock(t);
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800123 }
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800124 t = NULL;
125found:
126 rcu_read_unlock();
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700127
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800128 return t;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700129}
130
Yaowei Baidb2a0dd2015-11-06 16:28:06 -0800131/*
132 * order == -1 means the oom kill is required by sysrq, otherwise only
133 * for display purposes.
134 */
135static inline bool is_sysrq_oom(struct oom_control *oc)
136{
137 return oc->order == -1;
138}
139
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700140static inline bool is_memcg_oom(struct oom_control *oc)
141{
142 return oc->memcg != NULL;
143}
144
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700145/* return true if the task is not adequate as candidate victim task. */
David Rientjese85bfd32010-09-22 13:05:10 -0700146static bool oom_unkillable_task(struct task_struct *p,
Johannes Weiner2314b422014-12-10 15:44:33 -0800147 struct mem_cgroup *memcg, const nodemask_t *nodemask)
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700148{
149 if (is_global_init(p))
150 return true;
151 if (p->flags & PF_KTHREAD)
152 return true;
153
154 /* When mem_cgroup_out_of_memory() and p is not member of the group */
Johannes Weiner72835c82012-01-12 17:18:32 -0800155 if (memcg && !task_in_mem_cgroup(p, memcg))
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700156 return true;
157
158 /* p may not have freeable memory in nodemask */
159 if (!has_intersects_mems_allowed(p, nodemask))
160 return true;
161
162 return false;
163}
164
Yang Shi852d8be2017-11-15 17:32:07 -0800165/*
166 * Print out unreclaimble slabs info when unreclaimable slabs amount is greater
167 * than all user memory (LRU pages)
168 */
169static bool is_dump_unreclaim_slabs(void)
170{
171 unsigned long nr_lru;
172
173 nr_lru = global_node_page_state(NR_ACTIVE_ANON) +
174 global_node_page_state(NR_INACTIVE_ANON) +
175 global_node_page_state(NR_ACTIVE_FILE) +
176 global_node_page_state(NR_INACTIVE_FILE) +
177 global_node_page_state(NR_ISOLATED_ANON) +
178 global_node_page_state(NR_ISOLATED_FILE) +
179 global_node_page_state(NR_UNEVICTABLE);
180
181 return (global_node_page_state(NR_SLAB_UNRECLAIMABLE) > nr_lru);
182}
183
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184/**
David Rientjesa63d83f2010-08-09 17:19:46 -0700185 * oom_badness - heuristic function to determine which candidate task to kill
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 * @p: task struct of which task we should calculate
David Rientjesa63d83f2010-08-09 17:19:46 -0700187 * @totalpages: total present RAM allowed for page allocation
Mike Rapoporte8b098f2018-04-05 16:24:57 -0700188 * @memcg: task's memory controller, if constrained
189 * @nodemask: nodemask passed to page allocator for mempolicy ooms
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 *
David Rientjesa63d83f2010-08-09 17:19:46 -0700191 * The heuristic for determining which task to kill is made to be as simple and
192 * predictable as possible. The goal is to return the highest value for the
193 * task consuming the most memory to avoid subsequent oom failures.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 */
David Rientjesa7f638f2012-05-29 15:06:47 -0700195unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
196 const nodemask_t *nodemask, unsigned long totalpages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197{
David Rientjes1e11ad82012-06-08 13:21:26 -0700198 long points;
David Rientjes61eafb02012-06-20 12:52:58 -0700199 long adj;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700200
Johannes Weiner72835c82012-01-12 17:18:32 -0800201 if (oom_unkillable_task(p, memcg, nodemask))
KOSAKI Motohiro26ebc982010-08-09 17:19:37 -0700202 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700204 p = find_lock_task_mm(p);
205 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206 return 0;
207
Michal Hockobb8a4b72016-05-20 16:57:18 -0700208 /*
209 * Do not even consider tasks which are explicitly marked oom
Michal Hockob18dc5f2016-07-28 15:44:46 -0700210 * unkillable or have been already oom reaped or the are in
211 * the middle of vfork
Michal Hockobb8a4b72016-05-20 16:57:18 -0700212 */
David Rientjesa9c58b902012-12-11 16:02:54 -0800213 adj = (long)p->signal->oom_score_adj;
Michal Hockobb8a4b72016-05-20 16:57:18 -0700214 if (adj == OOM_SCORE_ADJ_MIN ||
Michal Hocko862e3072016-10-07 16:58:57 -0700215 test_bit(MMF_OOM_SKIP, &p->mm->flags) ||
Michal Hockob18dc5f2016-07-28 15:44:46 -0700216 in_vfork(p)) {
Michal Hocko5aecc852011-11-15 14:36:07 -0800217 task_unlock(p);
218 return 0;
219 }
220
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700222 * The baseline for the badness score is the proportion of RAM that each
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700223 * task's rss, pagetable and swap space use.
David Rientjesa63d83f2010-08-09 17:19:46 -0700224 */
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800225 points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
Kirill A. Shutemovaf5b0f62017-11-15 17:35:40 -0800226 mm_pgtables_bytes(p->mm) / PAGE_SIZE;
Andrew Morton97c2c9b82006-04-18 22:20:38 -0700227 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228
David Rientjes61eafb02012-06-20 12:52:58 -0700229 /* Normalize to oom_score_adj units */
230 adj *= totalpages / 1000;
231 points += adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232
David Rientjesf19e8aa2010-09-22 13:04:52 -0700233 /*
David Rientjesa7f638f2012-05-29 15:06:47 -0700234 * Never return 0 for an eligible task regardless of the root bonus and
235 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
David Rientjesf19e8aa2010-09-22 13:04:52 -0700236 */
David Rientjes1e11ad82012-06-08 13:21:26 -0700237 return points > 0 ? points : 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238}
239
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700240enum oom_constraint {
241 CONSTRAINT_NONE,
242 CONSTRAINT_CPUSET,
243 CONSTRAINT_MEMORY_POLICY,
244 CONSTRAINT_MEMCG,
245};
246
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800248 * Determine the type of allocation constraint.
249 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700250static enum oom_constraint constrained_alloc(struct oom_control *oc)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800251{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700252 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700253 struct zoneref *z;
David Rientjes6e0fc462015-09-08 15:00:36 -0700254 enum zone_type high_zoneidx = gfp_zone(oc->gfp_mask);
David Rientjesa63d83f2010-08-09 17:19:46 -0700255 bool cpuset_limited = false;
256 int nid;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800257
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700258 if (is_memcg_oom(oc)) {
Roman Gushchinbbec2e12018-06-07 17:06:18 -0700259 oc->totalpages = mem_cgroup_get_max(oc->memcg) ?: 1;
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700260 return CONSTRAINT_MEMCG;
261 }
262
David Rientjesa63d83f2010-08-09 17:19:46 -0700263 /* Default to all available memory */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700264 oc->totalpages = totalram_pages + total_swap_pages;
265
266 if (!IS_ENABLED(CONFIG_NUMA))
267 return CONSTRAINT_NONE;
David Rientjesa63d83f2010-08-09 17:19:46 -0700268
David Rientjes6e0fc462015-09-08 15:00:36 -0700269 if (!oc->zonelist)
David Rientjesa63d83f2010-08-09 17:19:46 -0700270 return CONSTRAINT_NONE;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800271 /*
272 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
273 * to kill current.We have to random task kill in this case.
274 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
275 */
David Rientjes6e0fc462015-09-08 15:00:36 -0700276 if (oc->gfp_mask & __GFP_THISNODE)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800277 return CONSTRAINT_NONE;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800278
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800279 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700280 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
281 * the page allocator means a mempolicy is in effect. Cpuset policy
282 * is enforced in get_page_from_freelist().
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800283 */
David Rientjes6e0fc462015-09-08 15:00:36 -0700284 if (oc->nodemask &&
285 !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) {
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700286 oc->totalpages = total_swap_pages;
David Rientjes6e0fc462015-09-08 15:00:36 -0700287 for_each_node_mask(nid, *oc->nodemask)
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700288 oc->totalpages += node_spanned_pages(nid);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800289 return CONSTRAINT_MEMORY_POLICY;
David Rientjesa63d83f2010-08-09 17:19:46 -0700290 }
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800291
292 /* Check this allocation failure is caused by cpuset's wall function */
David Rientjes6e0fc462015-09-08 15:00:36 -0700293 for_each_zone_zonelist_nodemask(zone, z, oc->zonelist,
294 high_zoneidx, oc->nodemask)
295 if (!cpuset_zone_allowed(zone, oc->gfp_mask))
David Rientjesa63d83f2010-08-09 17:19:46 -0700296 cpuset_limited = true;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800297
David Rientjesa63d83f2010-08-09 17:19:46 -0700298 if (cpuset_limited) {
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700299 oc->totalpages = total_swap_pages;
David Rientjesa63d83f2010-08-09 17:19:46 -0700300 for_each_node_mask(nid, cpuset_current_mems_allowed)
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700301 oc->totalpages += node_spanned_pages(nid);
David Rientjesa63d83f2010-08-09 17:19:46 -0700302 return CONSTRAINT_CPUSET;
303 }
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800304 return CONSTRAINT_NONE;
305}
306
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700307static int oom_evaluate_task(struct task_struct *task, void *arg)
David Rientjes462607e2012-07-31 16:43:40 -0700308{
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700309 struct oom_control *oc = arg;
310 unsigned long points;
311
David Rientjes6e0fc462015-09-08 15:00:36 -0700312 if (oom_unkillable_task(task, NULL, oc->nodemask))
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700313 goto next;
David Rientjes462607e2012-07-31 16:43:40 -0700314
315 /*
316 * This task already has access to memory reserves and is being killed.
Michal Hockoa3739662016-07-28 15:45:01 -0700317 * Don't allow any other task to have access to the reserves unless
Michal Hocko862e3072016-10-07 16:58:57 -0700318 * the task has MMF_OOM_SKIP because chances that it would release
Michal Hockoa3739662016-07-28 15:45:01 -0700319 * any memory is quite low.
David Rientjes462607e2012-07-31 16:43:40 -0700320 */
Michal Hocko862e3072016-10-07 16:58:57 -0700321 if (!is_sysrq_oom(oc) && tsk_is_oom_victim(task)) {
322 if (test_bit(MMF_OOM_SKIP, &task->signal->oom_mm->flags))
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700323 goto next;
324 goto abort;
Michal Hockoa3739662016-07-28 15:45:01 -0700325 }
David Rientjes462607e2012-07-31 16:43:40 -0700326
David Rientjese1e12d22012-12-11 16:02:56 -0800327 /*
328 * If task is allocating a lot of memory and has been marked to be
329 * killed first if it triggers an oom, then select it.
330 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700331 if (oom_task_origin(task)) {
332 points = ULONG_MAX;
333 goto select;
334 }
David Rientjese1e12d22012-12-11 16:02:56 -0800335
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700336 points = oom_badness(task, NULL, oc->nodemask, oc->totalpages);
337 if (!points || points < oc->chosen_points)
338 goto next;
339
340 /* Prefer thread group leaders for display purposes */
341 if (points == oc->chosen_points && thread_group_leader(oc->chosen))
342 goto next;
343select:
344 if (oc->chosen)
345 put_task_struct(oc->chosen);
346 get_task_struct(task);
347 oc->chosen = task;
348 oc->chosen_points = points;
349next:
350 return 0;
351abort:
352 if (oc->chosen)
353 put_task_struct(oc->chosen);
354 oc->chosen = (void *)-1UL;
355 return 1;
David Rientjes462607e2012-07-31 16:43:40 -0700356}
357
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800358/*
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700359 * Simple selection loop. We choose the process with the highest number of
360 * 'points'. In case scan was aborted, oc->chosen is set to -1.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700362static void select_bad_process(struct oom_control *oc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363{
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700364 if (is_memcg_oom(oc))
365 mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc);
366 else {
367 struct task_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700369 rcu_read_lock();
370 for_each_process(p)
371 if (oom_evaluate_task(p, oc))
372 break;
373 rcu_read_unlock();
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800374 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700375
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700376 oc->chosen_points = oc->chosen_points * 1000 / oc->totalpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377}
378
379/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700380 * dump_tasks - dump current memory state of all system tasks
Wanpeng Lidad75572012-06-20 12:53:01 -0700381 * @memcg: current's memory controller, if constrained
David Rientjese85bfd32010-09-22 13:05:10 -0700382 * @nodemask: nodemask passed to page allocator for mempolicy ooms
Randy Dunlap1b578df2008-03-19 17:00:42 -0700383 *
David Rientjese85bfd32010-09-22 13:05:10 -0700384 * Dumps the current memory state of all eligible tasks. Tasks not in the same
385 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
386 * are not shown.
Kirill A. Shutemovaf5b0f62017-11-15 17:35:40 -0800387 * State information includes task's pid, uid, tgid, vm size, rss,
388 * pgtables_bytes, swapents, oom_score_adj value, and name.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800389 */
Johannes Weiner2314b422014-12-10 15:44:33 -0800390static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask)
David Rientjesfef1bdd2008-02-07 00:14:07 -0800391{
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700392 struct task_struct *p;
393 struct task_struct *task;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800394
Kirill A. Shutemovaf5b0f62017-11-15 17:35:40 -0800395 pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name\n");
David Rientjes6b0c81b2012-07-31 16:43:45 -0700396 rcu_read_lock();
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700397 for_each_process(p) {
Johannes Weiner72835c82012-01-12 17:18:32 -0800398 if (oom_unkillable_task(p, memcg, nodemask))
David Rientjesfef1bdd2008-02-07 00:14:07 -0800399 continue;
400
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700401 task = find_lock_task_mm(p);
402 if (!task) {
David Rientjes6d2661e2009-05-28 14:34:19 -0700403 /*
David Rientjes74ab7f12010-08-09 17:18:46 -0700404 * This is a kthread or all of p's threads have already
405 * detached their mm's. There's no need to report
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700406 * them; they can't be oom killed anyway.
David Rientjes6d2661e2009-05-28 14:34:19 -0700407 */
David Rientjes6d2661e2009-05-28 14:34:19 -0700408 continue;
409 }
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700410
Kirill A. Shutemovaf5b0f62017-11-15 17:35:40 -0800411 pr_info("[%5d] %5d %5d %8lu %8lu %8ld %8lu %5hd %s\n",
Eric W. Biederman078de5f2012-02-08 07:00:08 -0800412 task->pid, from_kuid(&init_user_ns, task_uid(task)),
413 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
Kirill A. Shutemovaf5b0f62017-11-15 17:35:40 -0800414 mm_pgtables_bytes(task->mm),
David Rientjesde34d962012-07-31 16:42:56 -0700415 get_mm_counter(task->mm, MM_SWAPENTS),
David Rientjesa63d83f2010-08-09 17:19:46 -0700416 task->signal->oom_score_adj, task->comm);
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700417 task_unlock(task);
418 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700419 rcu_read_unlock();
David Rientjesfef1bdd2008-02-07 00:14:07 -0800420}
421
Vladimir Davydov2a966b72016-07-26 15:22:33 -0700422static void dump_header(struct oom_control *oc, struct task_struct *p)
David Rientjes1b604d72009-12-14 17:57:47 -0800423{
Michal Hocko0205f752017-11-15 17:39:14 -0800424 pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), nodemask=%*pbl, order=%d, oom_score_adj=%hd\n",
425 current->comm, oc->gfp_mask, &oc->gfp_mask,
426 nodemask_pr_args(oc->nodemask), oc->order,
427 current->signal->oom_score_adj);
Michal Hocko92549902016-10-07 16:59:33 -0700428 if (!IS_ENABLED(CONFIG_COMPACTION) && oc->order)
429 pr_warn("COMPACTION is disabled!!!\n");
Vlastimil Babkaa0795cd2016-03-15 14:56:05 -0700430
David Rientjesda39da32015-11-05 18:48:05 -0800431 cpuset_print_current_mems_allowed();
David Rientjes1b604d72009-12-14 17:57:47 -0800432 dump_stack();
Yang Shi852d8be2017-11-15 17:32:07 -0800433 if (is_memcg_oom(oc))
Vladimir Davydov2a966b72016-07-26 15:22:33 -0700434 mem_cgroup_print_oom_info(oc->memcg, p);
Yang Shi852d8be2017-11-15 17:32:07 -0800435 else {
David Rientjes299c5172017-02-24 14:55:42 -0800436 show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask);
Yang Shi852d8be2017-11-15 17:32:07 -0800437 if (is_dump_unreclaim_slabs())
438 dump_unreclaimable_slab();
439 }
David Rientjes1b604d72009-12-14 17:57:47 -0800440 if (sysctl_oom_dump_tasks)
Vladimir Davydov2a966b72016-07-26 15:22:33 -0700441 dump_tasks(oc->memcg, oc->nodemask);
David Rientjes1b604d72009-12-14 17:57:47 -0800442}
443
Michal Hocko5695be12014-10-20 18:12:32 +0200444/*
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800445 * Number of OOM victims in flight
Michal Hocko5695be12014-10-20 18:12:32 +0200446 */
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800447static atomic_t oom_victims = ATOMIC_INIT(0);
448static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
Michal Hocko5695be12014-10-20 18:12:32 +0200449
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700450static bool oom_killer_disabled __read_mostly;
Michal Hocko5695be12014-10-20 18:12:32 +0200451
Michal Hockobc448e82016-03-25 14:20:30 -0700452#define K(x) ((x) << (PAGE_SHIFT-10))
453
Michal Hocko3ef22df2016-05-19 17:13:12 -0700454/*
455 * task->mm can be NULL if the task is the exited group leader. So to
456 * determine whether the task is using a particular mm, we examine all the
457 * task's threads: if one of those is using this mm then this task was also
458 * using it.
459 */
Michal Hocko44a70ade2016-07-28 15:44:43 -0700460bool process_shares_mm(struct task_struct *p, struct mm_struct *mm)
Michal Hocko3ef22df2016-05-19 17:13:12 -0700461{
462 struct task_struct *t;
463
464 for_each_thread(p, t) {
465 struct mm_struct *t_mm = READ_ONCE(t->mm);
466 if (t_mm)
467 return t_mm == mm;
468 }
469 return false;
470}
471
Michal Hockoaac45362016-03-25 14:20:24 -0700472#ifdef CONFIG_MMU
473/*
474 * OOM Reaper kernel thread which tries to reap the memory used by the OOM
475 * victim (if that is possible) to help the OOM killer to move on.
476 */
477static struct task_struct *oom_reaper_th;
Michal Hockoaac45362016-03-25 14:20:24 -0700478static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait);
Vladimir Davydov29c696e2016-03-25 14:20:39 -0700479static struct task_struct *oom_reaper_list;
Michal Hocko03049262016-03-25 14:20:33 -0700480static DEFINE_SPINLOCK(oom_reaper_lock);
481
David Rientjes27ae3572018-05-11 16:02:04 -0700482void __oom_reap_task_mm(struct mm_struct *mm)
Michal Hockoaac45362016-03-25 14:20:24 -0700483{
Michal Hockoaac45362016-03-25 14:20:24 -0700484 struct vm_area_struct *vma;
David Rientjes27ae3572018-05-11 16:02:04 -0700485
486 /*
487 * Tell all users of get_user/copy_from_user etc... that the content
488 * is no longer stable. No barriers really needed because unmapping
489 * should imply barriers already and the reader would hit a page fault
490 * if it stumbled over a reaped memory.
491 */
492 set_bit(MMF_UNSTABLE, &mm->flags);
493
494 for (vma = mm->mmap ; vma; vma = vma->vm_next) {
495 if (!can_madv_dontneed_vma(vma))
496 continue;
497
498 /*
499 * Only anonymous pages have a good chance to be dropped
500 * without additional steps which we cannot afford as we
501 * are OOM already.
502 *
503 * We do not even care about fs backed pages because all
504 * which are reclaimable have already been reclaimed and
505 * we do not want to block exit_mmap by keeping mm ref
506 * count elevated without a good reason.
507 */
508 if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) {
509 const unsigned long start = vma->vm_start;
510 const unsigned long end = vma->vm_end;
511 struct mmu_gather tlb;
512
513 tlb_gather_mmu(&tlb, mm, start, end);
514 mmu_notifier_invalidate_range_start(mm, start, end);
515 unmap_page_range(&tlb, vma, start, end, NULL);
516 mmu_notifier_invalidate_range_end(mm, start, end);
517 tlb_finish_mmu(&tlb, start, end);
518 }
519 }
520}
521
522static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
523{
Michal Hockoaac45362016-03-25 14:20:24 -0700524 bool ret = true;
525
Michal Hocko36324a92016-03-25 14:20:27 -0700526 /*
Michal Hockoe2fe1452016-05-27 14:27:35 -0700527 * We have to make sure to not race with the victim exit path
528 * and cause premature new oom victim selection:
David Rientjes27ae3572018-05-11 16:02:04 -0700529 * oom_reap_task_mm exit_mm
Michal Hockoe5e3f4c2016-07-26 15:24:50 -0700530 * mmget_not_zero
Michal Hockoe2fe1452016-05-27 14:27:35 -0700531 * mmput
532 * atomic_dec_and_test
533 * exit_oom_victim
534 * [...]
535 * out_of_memory
536 * select_bad_process
537 * # no TIF_MEMDIE task selects new victim
538 * unmap_page_range # frees some memory
539 */
540 mutex_lock(&oom_lock);
541
Michal Hockoaac45362016-03-25 14:20:24 -0700542 if (!down_read_trylock(&mm->mmap_sem)) {
543 ret = false;
Roman Gushchin422580c2017-07-10 15:49:05 -0700544 trace_skip_task_reaping(tsk->pid);
Tetsuo Handa7ebffa42016-10-07 16:58:45 -0700545 goto unlock_oom;
Michal Hockoe5e3f4c2016-07-26 15:24:50 -0700546 }
547
548 /*
David Rientjesf340ff82018-01-31 16:18:36 -0800549 * If the mm has invalidate_{start,end}() notifiers that could block,
Michal Hocko4d4bbd82017-10-03 16:14:50 -0700550 * sleep to give the oom victim some more time.
551 * TODO: we really want to get rid of this ugly hack and make sure that
David Rientjesf340ff82018-01-31 16:18:36 -0800552 * notifiers cannot block for unbounded amount of time
Michal Hocko4d4bbd82017-10-03 16:14:50 -0700553 */
David Rientjesf340ff82018-01-31 16:18:36 -0800554 if (mm_has_blockable_invalidate_notifiers(mm)) {
Michal Hocko4d4bbd82017-10-03 16:14:50 -0700555 up_read(&mm->mmap_sem);
556 schedule_timeout_idle(HZ);
557 goto unlock_oom;
558 }
559
560 /*
Andrea Arcangeli21292582017-09-06 16:25:00 -0700561 * MMF_OOM_SKIP is set by exit_mmap when the OOM reaper can't
562 * work on the mm anymore. The check for MMF_OOM_SKIP must run
563 * under mmap_sem for reading because it serializes against the
564 * down_write();up_write() cycle in exit_mmap().
Michal Hockoe5e3f4c2016-07-26 15:24:50 -0700565 */
Andrea Arcangeli21292582017-09-06 16:25:00 -0700566 if (test_bit(MMF_OOM_SKIP, &mm->flags)) {
Michal Hockoe5e3f4c2016-07-26 15:24:50 -0700567 up_read(&mm->mmap_sem);
Roman Gushchin422580c2017-07-10 15:49:05 -0700568 trace_skip_task_reaping(tsk->pid);
Tetsuo Handa7ebffa42016-10-07 16:58:45 -0700569 goto unlock_oom;
Michal Hockoaac45362016-03-25 14:20:24 -0700570 }
571
Roman Gushchin422580c2017-07-10 15:49:05 -0700572 trace_start_task_reaping(tsk->pid);
573
David Rientjes27ae3572018-05-11 16:02:04 -0700574 __oom_reap_task_mm(mm);
Michal Hocko3f70dc32016-10-07 16:59:06 -0700575
Michal Hockobc448e82016-03-25 14:20:30 -0700576 pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
577 task_pid_nr(tsk), tsk->comm,
578 K(get_mm_counter(mm, MM_ANONPAGES)),
579 K(get_mm_counter(mm, MM_FILEPAGES)),
580 K(get_mm_counter(mm, MM_SHMEMPAGES)));
Michal Hockoaac45362016-03-25 14:20:24 -0700581 up_read(&mm->mmap_sem);
Michal Hocko36324a92016-03-25 14:20:27 -0700582
Roman Gushchin422580c2017-07-10 15:49:05 -0700583 trace_finish_task_reaping(tsk->pid);
Michal Hockoe5e3f4c2016-07-26 15:24:50 -0700584unlock_oom:
585 mutex_unlock(&oom_lock);
Michal Hockoaac45362016-03-25 14:20:24 -0700586 return ret;
587}
588
Michal Hockobc448e82016-03-25 14:20:30 -0700589#define MAX_OOM_REAP_RETRIES 10
Michal Hocko36324a92016-03-25 14:20:27 -0700590static void oom_reap_task(struct task_struct *tsk)
Michal Hockoaac45362016-03-25 14:20:24 -0700591{
592 int attempts = 0;
Michal Hocko26db62f2016-10-07 16:58:51 -0700593 struct mm_struct *mm = tsk->signal->oom_mm;
Michal Hockoaac45362016-03-25 14:20:24 -0700594
595 /* Retry the down_read_trylock(mmap_sem) a few times */
David Rientjes27ae3572018-05-11 16:02:04 -0700596 while (attempts++ < MAX_OOM_REAP_RETRIES && !oom_reap_task_mm(tsk, mm))
Michal Hockoaac45362016-03-25 14:20:24 -0700597 schedule_timeout_idle(HZ/10);
598
Tetsuo Handa97b12552018-04-05 16:25:45 -0700599 if (attempts <= MAX_OOM_REAP_RETRIES ||
600 test_bit(MMF_OOM_SKIP, &mm->flags))
Tetsuo Handa7ebffa42016-10-07 16:58:45 -0700601 goto done;
Michal Hocko11a410d2016-07-28 15:44:58 -0700602
Tetsuo Handa7ebffa42016-10-07 16:58:45 -0700603 pr_info("oom_reaper: unable to reap pid:%d (%s)\n",
604 task_pid_nr(tsk), tsk->comm);
Tetsuo Handa7ebffa42016-10-07 16:58:45 -0700605 debug_show_all_locks();
Michal Hockobc448e82016-03-25 14:20:30 -0700606
Tetsuo Handa7ebffa42016-10-07 16:58:45 -0700607done:
Michal Hocko449d7772016-05-19 17:13:15 -0700608 tsk->oom_reaper_list = NULL;
Michal Hocko449d7772016-05-19 17:13:15 -0700609
Michal Hocko26db62f2016-10-07 16:58:51 -0700610 /*
611 * Hide this mm from OOM killer because it has been either reaped or
612 * somebody can't call up_write(mmap_sem).
613 */
Michal Hocko862e3072016-10-07 16:58:57 -0700614 set_bit(MMF_OOM_SKIP, &mm->flags);
Michal Hocko26db62f2016-10-07 16:58:51 -0700615
Michal Hockoaac45362016-03-25 14:20:24 -0700616 /* Drop a reference taken by wake_oom_reaper */
Michal Hocko36324a92016-03-25 14:20:27 -0700617 put_task_struct(tsk);
Michal Hockoaac45362016-03-25 14:20:24 -0700618}
619
620static int oom_reaper(void *unused)
621{
622 while (true) {
Michal Hocko03049262016-03-25 14:20:33 -0700623 struct task_struct *tsk = NULL;
Michal Hockoaac45362016-03-25 14:20:24 -0700624
Vladimir Davydov29c696e2016-03-25 14:20:39 -0700625 wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL);
Michal Hocko03049262016-03-25 14:20:33 -0700626 spin_lock(&oom_reaper_lock);
Vladimir Davydov29c696e2016-03-25 14:20:39 -0700627 if (oom_reaper_list != NULL) {
628 tsk = oom_reaper_list;
629 oom_reaper_list = tsk->oom_reaper_list;
Michal Hocko03049262016-03-25 14:20:33 -0700630 }
631 spin_unlock(&oom_reaper_lock);
632
633 if (tsk)
634 oom_reap_task(tsk);
Michal Hockoaac45362016-03-25 14:20:24 -0700635 }
636
637 return 0;
638}
639
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700640static void wake_oom_reaper(struct task_struct *tsk)
Michal Hockoaac45362016-03-25 14:20:24 -0700641{
Michal Hockoaf8e15c2016-04-01 14:31:34 -0700642 /* tsk is already queued? */
643 if (tsk == oom_reaper_list || tsk->oom_reaper_list)
Michal Hockoaac45362016-03-25 14:20:24 -0700644 return;
645
Michal Hocko36324a92016-03-25 14:20:27 -0700646 get_task_struct(tsk);
Michal Hockoaac45362016-03-25 14:20:24 -0700647
Michal Hocko03049262016-03-25 14:20:33 -0700648 spin_lock(&oom_reaper_lock);
Vladimir Davydov29c696e2016-03-25 14:20:39 -0700649 tsk->oom_reaper_list = oom_reaper_list;
650 oom_reaper_list = tsk;
Michal Hocko03049262016-03-25 14:20:33 -0700651 spin_unlock(&oom_reaper_lock);
Roman Gushchin422580c2017-07-10 15:49:05 -0700652 trace_wake_reaper(tsk->pid);
Michal Hocko03049262016-03-25 14:20:33 -0700653 wake_up(&oom_reaper_wait);
Michal Hockoaac45362016-03-25 14:20:24 -0700654}
655
656static int __init oom_init(void)
657{
658 oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper");
Michal Hockoaac45362016-03-25 14:20:24 -0700659 return 0;
660}
661subsys_initcall(oom_init)
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700662#else
663static inline void wake_oom_reaper(struct task_struct *tsk)
664{
665}
666#endif /* CONFIG_MMU */
Michal Hockoaac45362016-03-25 14:20:24 -0700667
Michal Hocko49550b62015-02-11 15:26:12 -0800668/**
Johannes Weiner16e95192015-06-24 16:57:07 -0700669 * mark_oom_victim - mark the given task as OOM victim
Michal Hocko49550b62015-02-11 15:26:12 -0800670 * @tsk: task to mark
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800671 *
Johannes Weinerdc564012015-06-24 16:57:19 -0700672 * Has to be called with oom_lock held and never after
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800673 * oom has been disabled already.
Michal Hocko26db62f2016-10-07 16:58:51 -0700674 *
675 * tsk->mm has to be non NULL and caller has to guarantee it is stable (either
676 * under task_lock or operate on the current).
Michal Hocko49550b62015-02-11 15:26:12 -0800677 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700678static void mark_oom_victim(struct task_struct *tsk)
Michal Hocko49550b62015-02-11 15:26:12 -0800679{
Michal Hocko26db62f2016-10-07 16:58:51 -0700680 struct mm_struct *mm = tsk->mm;
681
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800682 WARN_ON(oom_killer_disabled);
683 /* OOM killer might race with memcg OOM */
684 if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
685 return;
Michal Hocko26db62f2016-10-07 16:58:51 -0700686
Michal Hocko26db62f2016-10-07 16:58:51 -0700687 /* oom_mm is bound to the signal struct life time. */
Michal Hocko4837fe32017-12-14 15:33:15 -0800688 if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm)) {
Vegard Nossumf1f10072017-02-27 14:30:07 -0800689 mmgrab(tsk->signal->oom_mm);
Michal Hocko4837fe32017-12-14 15:33:15 -0800690 set_bit(MMF_OOM_VICTIM, &mm->flags);
691 }
Michal Hocko26db62f2016-10-07 16:58:51 -0700692
Michal Hocko63a8ca92015-02-11 15:26:15 -0800693 /*
694 * Make sure that the task is woken up from uninterruptible sleep
695 * if it is frozen because OOM killer wouldn't be able to free
696 * any memory and livelock. freezing_slow_path will tell the freezer
697 * that TIF_MEMDIE tasks should be ignored.
698 */
699 __thaw_task(tsk);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800700 atomic_inc(&oom_victims);
Roman Gushchin422580c2017-07-10 15:49:05 -0700701 trace_mark_victim(tsk->pid);
Michal Hocko49550b62015-02-11 15:26:12 -0800702}
703
704/**
Johannes Weiner16e95192015-06-24 16:57:07 -0700705 * exit_oom_victim - note the exit of an OOM victim
Michal Hocko49550b62015-02-11 15:26:12 -0800706 */
Tetsuo Handa38531202016-10-07 16:59:03 -0700707void exit_oom_victim(void)
Michal Hocko49550b62015-02-11 15:26:12 -0800708{
Tetsuo Handa38531202016-10-07 16:59:03 -0700709 clear_thread_flag(TIF_MEMDIE);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800710
Johannes Weinerc38f1022015-06-24 16:57:13 -0700711 if (!atomic_dec_return(&oom_victims))
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800712 wake_up_all(&oom_victims_wait);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800713}
714
715/**
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700716 * oom_killer_enable - enable OOM killer
717 */
718void oom_killer_enable(void)
719{
720 oom_killer_disabled = false;
Michal Hockod75da002017-05-03 14:54:57 -0700721 pr_info("OOM killer enabled.\n");
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700722}
723
724/**
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800725 * oom_killer_disable - disable OOM killer
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700726 * @timeout: maximum timeout to wait for oom victims in jiffies
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800727 *
728 * Forces all page allocations to fail rather than trigger OOM killer.
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700729 * Will block and wait until all OOM victims are killed or the given
730 * timeout expires.
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800731 *
732 * The function cannot be called when there are runnable user tasks because
733 * the userspace would see unexpected allocation failures as a result. Any
734 * new usage of this function should be consulted with MM people.
735 *
736 * Returns true if successful and false if the OOM killer cannot be
737 * disabled.
738 */
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700739bool oom_killer_disable(signed long timeout)
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800740{
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700741 signed long ret;
742
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800743 /*
Tetsuo Handa6afcf282016-03-17 14:20:45 -0700744 * Make sure to not race with an ongoing OOM killer. Check that the
745 * current is not killed (possibly due to sharing the victim's memory).
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800746 */
Tetsuo Handa6afcf282016-03-17 14:20:45 -0700747 if (mutex_lock_killable(&oom_lock))
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800748 return false;
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800749 oom_killer_disabled = true;
Johannes Weinerdc564012015-06-24 16:57:19 -0700750 mutex_unlock(&oom_lock);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800751
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700752 ret = wait_event_interruptible_timeout(oom_victims_wait,
753 !atomic_read(&oom_victims), timeout);
754 if (ret <= 0) {
755 oom_killer_enable();
756 return false;
757 }
Michal Hockod75da002017-05-03 14:54:57 -0700758 pr_info("OOM killer disabled.\n");
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800759
760 return true;
761}
762
Michal Hocko1af8bb42016-07-28 15:44:52 -0700763static inline bool __task_will_free_mem(struct task_struct *task)
764{
765 struct signal_struct *sig = task->signal;
766
767 /*
768 * A coredumping process may sleep for an extended period in exit_mm(),
769 * so the oom killer cannot assume that the process will promptly exit
770 * and release memory.
771 */
772 if (sig->flags & SIGNAL_GROUP_COREDUMP)
773 return false;
774
775 if (sig->flags & SIGNAL_GROUP_EXIT)
776 return true;
777
778 if (thread_group_empty(task) && (task->flags & PF_EXITING))
779 return true;
780
781 return false;
782}
783
784/*
785 * Checks whether the given task is dying or exiting and likely to
786 * release its address space. This means that all threads and processes
787 * sharing the same mm have to be killed or exiting.
Michal Hocko091f3622016-07-28 15:45:04 -0700788 * Caller has to make sure that task->mm is stable (hold task_lock or
789 * it operates on the current).
Michal Hocko1af8bb42016-07-28 15:44:52 -0700790 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700791static bool task_will_free_mem(struct task_struct *task)
Michal Hocko1af8bb42016-07-28 15:44:52 -0700792{
Michal Hocko091f3622016-07-28 15:45:04 -0700793 struct mm_struct *mm = task->mm;
Michal Hocko1af8bb42016-07-28 15:44:52 -0700794 struct task_struct *p;
Geert Uytterhoevenf33e6f02016-08-11 15:33:09 -0700795 bool ret = true;
Michal Hocko1af8bb42016-07-28 15:44:52 -0700796
Michal Hocko091f3622016-07-28 15:45:04 -0700797 /*
798 * Skip tasks without mm because it might have passed its exit_mm and
799 * exit_oom_victim. oom_reaper could have rescued that but do not rely
800 * on that for now. We can consider find_lock_task_mm in future.
801 */
802 if (!mm)
803 return false;
804
Michal Hocko1af8bb42016-07-28 15:44:52 -0700805 if (!__task_will_free_mem(task))
806 return false;
807
808 /*
Michal Hocko696453e2016-07-28 15:44:55 -0700809 * This task has already been drained by the oom reaper so there are
810 * only small chances it will free some more
811 */
Michal Hocko862e3072016-10-07 16:58:57 -0700812 if (test_bit(MMF_OOM_SKIP, &mm->flags))
Michal Hocko696453e2016-07-28 15:44:55 -0700813 return false;
Michal Hocko696453e2016-07-28 15:44:55 -0700814
Michal Hocko091f3622016-07-28 15:45:04 -0700815 if (atomic_read(&mm->mm_users) <= 1)
Michal Hocko1af8bb42016-07-28 15:44:52 -0700816 return true;
Michal Hocko1af8bb42016-07-28 15:44:52 -0700817
818 /*
Michal Hocko5870c2e2016-10-07 16:57:32 -0700819 * Make sure that all tasks which share the mm with the given tasks
820 * are dying as well to make sure that a) nobody pins its mm and
821 * b) the task is also reapable by the oom reaper.
Michal Hocko1af8bb42016-07-28 15:44:52 -0700822 */
823 rcu_read_lock();
824 for_each_process(p) {
825 if (!process_shares_mm(p, mm))
826 continue;
827 if (same_thread_group(task, p))
828 continue;
829 ret = __task_will_free_mem(p);
830 if (!ret)
831 break;
832 }
833 rcu_read_unlock();
Michal Hocko1af8bb42016-07-28 15:44:52 -0700834
835 return ret;
836}
837
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700838static void oom_kill_process(struct oom_control *oc, const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700839{
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700840 struct task_struct *p = oc->chosen;
841 unsigned int points = oc->chosen_points;
Linus Torvalds52d3c032011-03-14 15:17:07 -0700842 struct task_struct *victim = p;
David Rientjes5e9d8342010-08-09 17:18:51 -0700843 struct task_struct *child;
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800844 struct task_struct *t;
David Rientjes647f2bd2012-03-21 16:33:46 -0700845 struct mm_struct *mm;
Linus Torvalds52d3c032011-03-14 15:17:07 -0700846 unsigned int victim_points = 0;
David Rientjesdc3f21e2012-03-21 16:33:47 -0700847 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
848 DEFAULT_RATELIMIT_BURST);
Tetsuo Handabb299022016-03-25 14:20:44 -0700849 bool can_oom_reap = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700850
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700851 /*
852 * If the task is already exiting, don't alarm the sysadmin or kill
Michal Hockocd04ae12017-09-06 16:24:50 -0700853 * its children or threads, just give it access to memory reserves
854 * so it can die quickly
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700855 */
Michal Hocko091f3622016-07-28 15:45:04 -0700856 task_lock(p);
Michal Hocko1af8bb42016-07-28 15:44:52 -0700857 if (task_will_free_mem(p)) {
Johannes Weiner16e95192015-06-24 16:57:07 -0700858 mark_oom_victim(p);
Michal Hocko1af8bb42016-07-28 15:44:52 -0700859 wake_oom_reaper(p);
Michal Hocko091f3622016-07-28 15:45:04 -0700860 task_unlock(p);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700861 put_task_struct(p);
David Rientjes2a1c9b12012-03-21 16:33:46 -0700862 return;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700863 }
Michal Hocko091f3622016-07-28 15:45:04 -0700864 task_unlock(p);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700865
David Rientjesdc3f21e2012-03-21 16:33:47 -0700866 if (__ratelimit(&oom_rs))
Vladimir Davydov2a966b72016-07-26 15:22:33 -0700867 dump_header(oc, p);
David Rientjes8447d952012-03-21 16:33:47 -0700868
Wang Longf0d66472015-06-24 16:58:01 -0700869 pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n",
David Rientjes5e9d8342010-08-09 17:18:51 -0700870 message, task_pid_nr(p), p->comm, points);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800871
David Rientjes5e9d8342010-08-09 17:18:51 -0700872 /*
873 * If any of p's children has a different mm and is eligible for kill,
David Rientjes11239832011-07-25 17:12:17 -0700874 * the one with the highest oom_badness() score is sacrificed for its
David Rientjes5e9d8342010-08-09 17:18:51 -0700875 * parent. This attempts to lose the minimal amount of work done while
876 * still freeing memory.
877 */
David Rientjes6b0c81b2012-07-31 16:43:45 -0700878 read_lock(&tasklist_lock);
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800879 for_each_thread(p, t) {
David Rientjes5e9d8342010-08-09 17:18:51 -0700880 list_for_each_entry(child, &t->children, sibling) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700881 unsigned int child_points;
David Rientjes5e9d8342010-08-09 17:18:51 -0700882
Oleg Nesterov4d7b3392015-11-05 18:48:26 -0800883 if (process_shares_mm(child, p->mm))
David Rientjesedd45542011-03-22 16:30:12 -0700884 continue;
David Rientjesa63d83f2010-08-09 17:19:46 -0700885 /*
886 * oom_badness() returns 0 if the thread is unkillable
887 */
Vladimir Davydov2a966b72016-07-26 15:22:33 -0700888 child_points = oom_badness(child,
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700889 oc->memcg, oc->nodemask, oc->totalpages);
David Rientjes5e9d8342010-08-09 17:18:51 -0700890 if (child_points > victim_points) {
David Rientjes6b0c81b2012-07-31 16:43:45 -0700891 put_task_struct(victim);
David Rientjes5e9d8342010-08-09 17:18:51 -0700892 victim = child;
893 victim_points = child_points;
David Rientjes6b0c81b2012-07-31 16:43:45 -0700894 get_task_struct(victim);
David Rientjes5e9d8342010-08-09 17:18:51 -0700895 }
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700896 }
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800897 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700898 read_unlock(&tasklist_lock);
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700899
David Rientjes6b0c81b2012-07-31 16:43:45 -0700900 p = find_lock_task_mm(victim);
901 if (!p) {
David Rientjes6b0c81b2012-07-31 16:43:45 -0700902 put_task_struct(victim);
David Rientjes647f2bd2012-03-21 16:33:46 -0700903 return;
David Rientjes6b0c81b2012-07-31 16:43:45 -0700904 } else if (victim != p) {
905 get_task_struct(p);
906 put_task_struct(victim);
907 victim = p;
908 }
David Rientjes647f2bd2012-03-21 16:33:46 -0700909
Tetsuo Handa880b7682015-11-05 18:47:51 -0800910 /* Get a reference to safely compare mm after task_unlock(victim) */
David Rientjes647f2bd2012-03-21 16:33:46 -0700911 mm = victim->mm;
Vegard Nossumf1f10072017-02-27 14:30:07 -0800912 mmgrab(mm);
Konstantin Khlebnikov8e675f72017-07-06 15:40:28 -0700913
914 /* Raise event before sending signal: task reaper must see this */
915 count_vm_event(OOM_KILL);
Roman Gushchinfe6bdfc2018-06-14 15:28:05 -0700916 memcg_memory_event_mm(mm, MEMCG_OOM_KILL);
Konstantin Khlebnikov8e675f72017-07-06 15:40:28 -0700917
Tetsuo Handa426fb5e2015-11-05 18:47:44 -0800918 /*
Michal Hockocd04ae12017-09-06 16:24:50 -0700919 * We should send SIGKILL before granting access to memory reserves
920 * in order to prevent the OOM victim from depleting the memory
921 * reserves from the user space under its control.
Tetsuo Handa426fb5e2015-11-05 18:47:44 -0800922 */
923 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
Johannes Weiner16e95192015-06-24 16:57:07 -0700924 mark_oom_victim(victim);
Jerome Marchandeca56ff2016-01-14 15:19:26 -0800925 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
David Rientjes647f2bd2012-03-21 16:33:46 -0700926 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
927 K(get_mm_counter(victim->mm, MM_ANONPAGES)),
Jerome Marchandeca56ff2016-01-14 15:19:26 -0800928 K(get_mm_counter(victim->mm, MM_FILEPAGES)),
929 K(get_mm_counter(victim->mm, MM_SHMEMPAGES)));
David Rientjes647f2bd2012-03-21 16:33:46 -0700930 task_unlock(victim);
931
932 /*
933 * Kill all user processes sharing victim->mm in other thread groups, if
934 * any. They don't get access to memory reserves, though, to avoid
935 * depletion of all memory. This prevents mm->mmap_sem livelock when an
936 * oom killed thread cannot exit because it requires the semaphore and
937 * its contended by another thread trying to allocate memory itself.
938 * That thread will now get access to memory reserves since it has a
939 * pending fatal signal.
940 */
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800941 rcu_read_lock();
Oleg Nesterovc3190252015-11-05 18:48:23 -0800942 for_each_process(p) {
Oleg Nesterov4d7b3392015-11-05 18:48:26 -0800943 if (!process_shares_mm(p, mm))
Oleg Nesterovc3190252015-11-05 18:48:23 -0800944 continue;
945 if (same_thread_group(p, victim))
946 continue;
Michal Hocko1b51e652016-10-07 16:59:09 -0700947 if (is_global_init(p)) {
Michal Hockoaac45362016-03-25 14:20:24 -0700948 can_oom_reap = false;
Michal Hocko862e3072016-10-07 16:58:57 -0700949 set_bit(MMF_OOM_SKIP, &mm->flags);
Michal Hockoa3739662016-07-28 15:45:01 -0700950 pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n",
951 task_pid_nr(victim), victim->comm,
952 task_pid_nr(p), p->comm);
Oleg Nesterovc3190252015-11-05 18:48:23 -0800953 continue;
Michal Hockoaac45362016-03-25 14:20:24 -0700954 }
Michal Hocko1b51e652016-10-07 16:59:09 -0700955 /*
956 * No use_mm() user needs to read from the userspace so we are
957 * ok to reap it.
958 */
959 if (unlikely(p->flags & PF_KTHREAD))
960 continue;
Oleg Nesterovc3190252015-11-05 18:48:23 -0800961 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
962 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700963 rcu_read_unlock();
David Rientjes647f2bd2012-03-21 16:33:46 -0700964
Michal Hockoaac45362016-03-25 14:20:24 -0700965 if (can_oom_reap)
Michal Hocko36324a92016-03-25 14:20:27 -0700966 wake_oom_reaper(victim);
Michal Hockoaac45362016-03-25 14:20:24 -0700967
Tetsuo Handa880b7682015-11-05 18:47:51 -0800968 mmdrop(mm);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700969 put_task_struct(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700970}
David Rientjes647f2bd2012-03-21 16:33:46 -0700971#undef K
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972
David Rientjes309ed882010-08-09 17:18:54 -0700973/*
974 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
975 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700976static void check_panic_on_oom(struct oom_control *oc,
977 enum oom_constraint constraint)
David Rientjes309ed882010-08-09 17:18:54 -0700978{
979 if (likely(!sysctl_panic_on_oom))
980 return;
981 if (sysctl_panic_on_oom != 2) {
982 /*
983 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
984 * does not panic for cpuset, mempolicy, or memcg allocation
985 * failures.
986 */
987 if (constraint != CONSTRAINT_NONE)
988 return;
989 }
David Rientjes071a4be2015-09-08 15:00:42 -0700990 /* Do not panic for oom kills triggered by sysrq */
Yaowei Baidb2a0dd2015-11-06 16:28:06 -0800991 if (is_sysrq_oom(oc))
David Rientjes071a4be2015-09-08 15:00:42 -0700992 return;
Vladimir Davydov2a966b72016-07-26 15:22:33 -0700993 dump_header(oc, NULL);
David Rientjes309ed882010-08-09 17:18:54 -0700994 panic("Out of memory: %s panic_on_oom is enabled\n",
995 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
996}
997
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700998static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
999
1000int register_oom_notifier(struct notifier_block *nb)
1001{
1002 return blocking_notifier_chain_register(&oom_notify_list, nb);
1003}
1004EXPORT_SYMBOL_GPL(register_oom_notifier);
1005
1006int unregister_oom_notifier(struct notifier_block *nb)
1007{
1008 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
1009}
1010EXPORT_SYMBOL_GPL(unregister_oom_notifier);
1011
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012/**
David Rientjes6e0fc462015-09-08 15:00:36 -07001013 * out_of_memory - kill the "best" process when we run out of memory
1014 * @oc: pointer to struct oom_control
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 *
1016 * If we run out of memory, we have the choice between either
1017 * killing a random task (bad), letting the system crash (worse)
1018 * OR try to be smart about which process to kill. Note that we
1019 * don't have to be perfect here, we just have to be good.
1020 */
David Rientjes6e0fc462015-09-08 15:00:36 -07001021bool out_of_memory(struct oom_control *oc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022{
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -07001023 unsigned long freed = 0;
David Rientjese3658932010-08-09 17:18:55 -07001024 enum oom_constraint constraint = CONSTRAINT_NONE;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -07001025
Johannes Weinerdc564012015-06-24 16:57:19 -07001026 if (oom_killer_disabled)
1027 return false;
1028
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001029 if (!is_memcg_oom(oc)) {
1030 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
1031 if (freed > 0)
1032 /* Got some memory back in the last second. */
1033 return true;
1034 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001035
David Rientjes7b98c2e2010-08-09 17:18:48 -07001036 /*
David Rientjes9ff48682012-12-11 16:01:30 -08001037 * If current has a pending SIGKILL or is exiting, then automatically
1038 * select it. The goal is to allow it to allocate so that it may
1039 * quickly exit and free its memory.
David Rientjes7b98c2e2010-08-09 17:18:48 -07001040 */
Michal Hocko091f3622016-07-28 15:45:04 -07001041 if (task_will_free_mem(current)) {
Johannes Weiner16e95192015-06-24 16:57:07 -07001042 mark_oom_victim(current);
Michal Hocko1af8bb42016-07-28 15:44:52 -07001043 wake_oom_reaper(current);
David Rientjes75e8f8b2015-09-08 15:00:47 -07001044 return true;
David Rientjes7b98c2e2010-08-09 17:18:48 -07001045 }
1046
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001047 /*
Michal Hocko3da88fb2016-05-19 17:13:09 -07001048 * The OOM killer does not compensate for IO-less reclaim.
1049 * pagefault_out_of_memory lost its gfp context so we have to
1050 * make sure exclude 0 mask - all other users should have at least
1051 * ___GFP_DIRECT_RECLAIM to get here.
1052 */
Michal Hocko06ad2762017-02-22 15:46:22 -08001053 if (oc->gfp_mask && !(oc->gfp_mask & __GFP_FS))
Michal Hocko3da88fb2016-05-19 17:13:09 -07001054 return true;
1055
1056 /*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001057 * Check if there were limitations on the allocation (only relevant for
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001058 * NUMA and memcg) that may require different handling.
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001059 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001060 constraint = constrained_alloc(oc);
David Rientjes6e0fc462015-09-08 15:00:36 -07001061 if (constraint != CONSTRAINT_MEMORY_POLICY)
1062 oc->nodemask = NULL;
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001063 check_panic_on_oom(oc, constraint);
David Rientjes0aad4b32010-08-09 17:18:59 -07001064
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001065 if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task &&
1066 current->mm && !oom_unkillable_task(current, NULL, oc->nodemask) &&
David Rientjes121d1ba2012-07-31 16:42:55 -07001067 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
David Rientjes6b0c81b2012-07-31 16:43:45 -07001068 get_task_struct(current);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001069 oc->chosen = current;
1070 oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)");
David Rientjes75e8f8b2015-09-08 15:00:47 -07001071 return true;
David Rientjes0aad4b32010-08-09 17:18:59 -07001072 }
1073
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001074 select_bad_process(oc);
David Rientjes0aad4b32010-08-09 17:18:59 -07001075 /* Found nothing?!?! Either we hang forever, or we panic. */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001076 if (!oc->chosen && !is_sysrq_oom(oc) && !is_memcg_oom(oc)) {
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001077 dump_header(oc, NULL);
David Rientjes0aad4b32010-08-09 17:18:59 -07001078 panic("Out of memory and no killable processes...\n");
1079 }
Michal Hocko9bfe5de2018-08-17 15:49:04 -07001080 if (oc->chosen && oc->chosen != (void *)-1UL)
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001081 oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" :
1082 "Memory cgroup out of memory");
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001083 return !!oc->chosen;
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001084}
1085
David Rientjese3658932010-08-09 17:18:55 -07001086/*
1087 * The pagefault handler calls here because it is out of memory, so kill a
Vladimir Davydov798fd752016-07-26 15:22:30 -07001088 * memory-hogging task. If oom_lock is held by somebody else, a parallel oom
1089 * killing is already in progress so do nothing.
David Rientjese3658932010-08-09 17:18:55 -07001090 */
1091void pagefault_out_of_memory(void)
1092{
David Rientjes6e0fc462015-09-08 15:00:36 -07001093 struct oom_control oc = {
1094 .zonelist = NULL,
1095 .nodemask = NULL,
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001096 .memcg = NULL,
David Rientjes6e0fc462015-09-08 15:00:36 -07001097 .gfp_mask = 0,
1098 .order = 0,
David Rientjes6e0fc462015-09-08 15:00:36 -07001099 };
1100
Johannes Weiner49426422013-10-16 13:46:59 -07001101 if (mem_cgroup_oom_synchronize(true))
Johannes Weinerdc564012015-06-24 16:57:19 -07001102 return;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001103
Johannes Weinerdc564012015-06-24 16:57:19 -07001104 if (!mutex_trylock(&oom_lock))
1105 return;
Tetsuo Handaa1048082016-10-07 17:00:49 -07001106 out_of_memory(&oc);
Johannes Weinerdc564012015-06-24 16:57:19 -07001107 mutex_unlock(&oom_lock);
David Rientjese3658932010-08-09 17:18:55 -07001108}