blob: ddf74487f8487f0d109416f93a433ba58117426e [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>
25#include <linux/swap.h>
26#include <linux/timex.h>
27#include <linux/jiffies.h>
Paul Jacksonef08e3b2005-09-06 15:18:13 -070028#include <linux/cpuset.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040029#include <linux/export.h>
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -070030#include <linux/notifier.h>
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -080031#include <linux/memcontrol.h>
David Rientjes6f48d0eb2010-08-09 17:18:52 -070032#include <linux/mempolicy.h>
David Howells5cd9c582008-08-14 11:37:28 +010033#include <linux/security.h>
David Rientjesedd45542011-03-22 16:30:12 -070034#include <linux/ptrace.h>
David Rientjesf660daa2011-10-31 17:07:07 -070035#include <linux/freezer.h>
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080036#include <linux/ftrace.h>
David Rientjesdc3f21e2012-03-21 16:33:47 -070037#include <linux/ratelimit.h>
Michal Hockoaac45362016-03-25 14:20:24 -070038#include <linux/kthread.h>
39#include <linux/init.h>
40
41#include <asm/tlb.h>
42#include "internal.h"
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080043
44#define CREATE_TRACE_POINTS
45#include <trace/events/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -070047int sysctl_panic_on_oom;
David Rientjesfe071d72007-10-16 23:25:56 -070048int sysctl_oom_kill_allocating_task;
David Rientjesad915c42010-08-09 17:18:53 -070049int sysctl_oom_dump_tasks = 1;
Johannes Weinerdc564012015-06-24 16:57:19 -070050
51DEFINE_MUTEX(oom_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
David Rientjes6f48d0eb2010-08-09 17:18:52 -070053#ifdef CONFIG_NUMA
54/**
55 * has_intersects_mems_allowed() - check task eligiblity for kill
Oleg Nesterovad962442014-01-21 15:50:00 -080056 * @start: task struct of which task to consider
David Rientjes6f48d0eb2010-08-09 17:18:52 -070057 * @mask: nodemask passed to page allocator for mempolicy ooms
58 *
59 * Task eligibility is determined by whether or not a candidate task, @tsk,
60 * shares the same mempolicy nodes as current if it is bound by such a policy
61 * and whether or not it has the same set of allowed cpuset nodes.
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070062 */
Oleg Nesterovad962442014-01-21 15:50:00 -080063static bool has_intersects_mems_allowed(struct task_struct *start,
David Rientjes6f48d0eb2010-08-09 17:18:52 -070064 const nodemask_t *mask)
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070065{
Oleg Nesterovad962442014-01-21 15:50:00 -080066 struct task_struct *tsk;
67 bool ret = false;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070068
Oleg Nesterovad962442014-01-21 15:50:00 -080069 rcu_read_lock();
Oleg Nesterov1da4db02014-01-21 15:49:58 -080070 for_each_thread(start, tsk) {
David Rientjes6f48d0eb2010-08-09 17:18:52 -070071 if (mask) {
72 /*
73 * If this is a mempolicy constrained oom, tsk's
74 * cpuset is irrelevant. Only return true if its
75 * mempolicy intersects current, otherwise it may be
76 * needlessly killed.
77 */
Oleg Nesterovad962442014-01-21 15:50:00 -080078 ret = mempolicy_nodemask_intersects(tsk, mask);
David Rientjes6f48d0eb2010-08-09 17:18:52 -070079 } else {
80 /*
81 * This is not a mempolicy constrained oom, so only
82 * check the mems of tsk's cpuset.
83 */
Oleg Nesterovad962442014-01-21 15:50:00 -080084 ret = cpuset_mems_allowed_intersects(current, tsk);
David Rientjes6f48d0eb2010-08-09 17:18:52 -070085 }
Oleg Nesterovad962442014-01-21 15:50:00 -080086 if (ret)
87 break;
Oleg Nesterov1da4db02014-01-21 15:49:58 -080088 }
Oleg Nesterovad962442014-01-21 15:50:00 -080089 rcu_read_unlock();
KOSAKI Motohirodf1090a2010-08-09 17:19:39 -070090
Oleg Nesterovad962442014-01-21 15:50:00 -080091 return ret;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070092}
David Rientjes6f48d0eb2010-08-09 17:18:52 -070093#else
94static bool has_intersects_mems_allowed(struct task_struct *tsk,
95 const nodemask_t *mask)
96{
97 return true;
98}
99#endif /* CONFIG_NUMA */
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700100
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700101/*
102 * The process p may have detached its own ->mm while exiting or through
103 * use_mm(), but one or more of its subthreads may still have a valid
104 * pointer. Return p, or any of its subthreads with a valid ->mm, with
105 * task_lock() held.
106 */
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -0700107struct task_struct *find_lock_task_mm(struct task_struct *p)
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700108{
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800109 struct task_struct *t;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700110
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800111 rcu_read_lock();
112
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800113 for_each_thread(p, t) {
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700114 task_lock(t);
115 if (likely(t->mm))
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800116 goto found;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700117 task_unlock(t);
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800118 }
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800119 t = NULL;
120found:
121 rcu_read_unlock();
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700122
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800123 return t;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700124}
125
Yaowei Baidb2a0dd2015-11-06 16:28:06 -0800126/*
127 * order == -1 means the oom kill is required by sysrq, otherwise only
128 * for display purposes.
129 */
130static inline bool is_sysrq_oom(struct oom_control *oc)
131{
132 return oc->order == -1;
133}
134
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700135/* return true if the task is not adequate as candidate victim task. */
David Rientjese85bfd32010-09-22 13:05:10 -0700136static bool oom_unkillable_task(struct task_struct *p,
Johannes Weiner2314b422014-12-10 15:44:33 -0800137 struct mem_cgroup *memcg, const nodemask_t *nodemask)
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700138{
139 if (is_global_init(p))
140 return true;
141 if (p->flags & PF_KTHREAD)
142 return true;
143
144 /* When mem_cgroup_out_of_memory() and p is not member of the group */
Johannes Weiner72835c82012-01-12 17:18:32 -0800145 if (memcg && !task_in_mem_cgroup(p, memcg))
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700146 return true;
147
148 /* p may not have freeable memory in nodemask */
149 if (!has_intersects_mems_allowed(p, nodemask))
150 return true;
151
152 return false;
153}
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155/**
David Rientjesa63d83f2010-08-09 17:19:46 -0700156 * oom_badness - heuristic function to determine which candidate task to kill
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 * @p: task struct of which task we should calculate
David Rientjesa63d83f2010-08-09 17:19:46 -0700158 * @totalpages: total present RAM allowed for page allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 *
David Rientjesa63d83f2010-08-09 17:19:46 -0700160 * The heuristic for determining which task to kill is made to be as simple and
161 * predictable as possible. The goal is to return the highest value for the
162 * task consuming the most memory to avoid subsequent oom failures.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163 */
David Rientjesa7f638f2012-05-29 15:06:47 -0700164unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
165 const nodemask_t *nodemask, unsigned long totalpages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166{
David Rientjes1e11ad82012-06-08 13:21:26 -0700167 long points;
David Rientjes61eafb02012-06-20 12:52:58 -0700168 long adj;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700169
Johannes Weiner72835c82012-01-12 17:18:32 -0800170 if (oom_unkillable_task(p, memcg, nodemask))
KOSAKI Motohiro26ebc982010-08-09 17:19:37 -0700171 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700173 p = find_lock_task_mm(p);
174 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 return 0;
176
Michal Hockobb8a4b72016-05-20 16:57:18 -0700177 /*
178 * Do not even consider tasks which are explicitly marked oom
179 * unkillable or have been already oom reaped.
180 */
David Rientjesa9c58b902012-12-11 16:02:54 -0800181 adj = (long)p->signal->oom_score_adj;
Michal Hockobb8a4b72016-05-20 16:57:18 -0700182 if (adj == OOM_SCORE_ADJ_MIN ||
183 test_bit(MMF_OOM_REAPED, &p->mm->flags)) {
Michal Hocko5aecc852011-11-15 14:36:07 -0800184 task_unlock(p);
185 return 0;
186 }
187
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700189 * The baseline for the badness score is the proportion of RAM that each
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700190 * task's rss, pagetable and swap space use.
David Rientjesa63d83f2010-08-09 17:19:46 -0700191 */
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800192 points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
193 atomic_long_read(&p->mm->nr_ptes) + mm_nr_pmds(p->mm);
Andrew Morton97c2c9b82006-04-18 22:20:38 -0700194 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700195
196 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700197 * Root processes get 3% bonus, just like the __vm_enough_memory()
198 * implementation used by LSMs.
Hugh Dickins7ba34852007-01-05 16:37:03 -0800199 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700200 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
David Rientjes778c14a2014-01-30 15:46:11 -0800201 points -= (points * 3) / 100;
Hugh Dickins7ba34852007-01-05 16:37:03 -0800202
David Rientjes61eafb02012-06-20 12:52:58 -0700203 /* Normalize to oom_score_adj units */
204 adj *= totalpages / 1000;
205 points += adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700206
David Rientjesf19e8aa2010-09-22 13:04:52 -0700207 /*
David Rientjesa7f638f2012-05-29 15:06:47 -0700208 * Never return 0 for an eligible task regardless of the root bonus and
209 * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here).
David Rientjesf19e8aa2010-09-22 13:04:52 -0700210 */
David Rientjes1e11ad82012-06-08 13:21:26 -0700211 return points > 0 ? points : 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212}
213
214/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800215 * Determine the type of allocation constraint.
216 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800217#ifdef CONFIG_NUMA
David Rientjes6e0fc462015-09-08 15:00:36 -0700218static enum oom_constraint constrained_alloc(struct oom_control *oc,
219 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800220{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700221 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700222 struct zoneref *z;
David Rientjes6e0fc462015-09-08 15:00:36 -0700223 enum zone_type high_zoneidx = gfp_zone(oc->gfp_mask);
David Rientjesa63d83f2010-08-09 17:19:46 -0700224 bool cpuset_limited = false;
225 int nid;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800226
David Rientjesa63d83f2010-08-09 17:19:46 -0700227 /* Default to all available memory */
228 *totalpages = totalram_pages + total_swap_pages;
229
David Rientjes6e0fc462015-09-08 15:00:36 -0700230 if (!oc->zonelist)
David Rientjesa63d83f2010-08-09 17:19:46 -0700231 return CONSTRAINT_NONE;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800232 /*
233 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
234 * to kill current.We have to random task kill in this case.
235 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
236 */
David Rientjes6e0fc462015-09-08 15:00:36 -0700237 if (oc->gfp_mask & __GFP_THISNODE)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800238 return CONSTRAINT_NONE;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800239
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800240 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700241 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
242 * the page allocator means a mempolicy is in effect. Cpuset policy
243 * is enforced in get_page_from_freelist().
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800244 */
David Rientjes6e0fc462015-09-08 15:00:36 -0700245 if (oc->nodemask &&
246 !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700247 *totalpages = total_swap_pages;
David Rientjes6e0fc462015-09-08 15:00:36 -0700248 for_each_node_mask(nid, *oc->nodemask)
David Rientjesa63d83f2010-08-09 17:19:46 -0700249 *totalpages += node_spanned_pages(nid);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800250 return CONSTRAINT_MEMORY_POLICY;
David Rientjesa63d83f2010-08-09 17:19:46 -0700251 }
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800252
253 /* Check this allocation failure is caused by cpuset's wall function */
David Rientjes6e0fc462015-09-08 15:00:36 -0700254 for_each_zone_zonelist_nodemask(zone, z, oc->zonelist,
255 high_zoneidx, oc->nodemask)
256 if (!cpuset_zone_allowed(zone, oc->gfp_mask))
David Rientjesa63d83f2010-08-09 17:19:46 -0700257 cpuset_limited = true;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800258
David Rientjesa63d83f2010-08-09 17:19:46 -0700259 if (cpuset_limited) {
260 *totalpages = total_swap_pages;
261 for_each_node_mask(nid, cpuset_current_mems_allowed)
262 *totalpages += node_spanned_pages(nid);
263 return CONSTRAINT_CPUSET;
264 }
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800265 return CONSTRAINT_NONE;
266}
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800267#else
David Rientjes6e0fc462015-09-08 15:00:36 -0700268static enum oom_constraint constrained_alloc(struct oom_control *oc,
269 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800270{
David Rientjesa63d83f2010-08-09 17:19:46 -0700271 *totalpages = totalram_pages + total_swap_pages;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800272 return CONSTRAINT_NONE;
273}
274#endif
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800275
David Rientjes6e0fc462015-09-08 15:00:36 -0700276enum oom_scan_t oom_scan_process_thread(struct oom_control *oc,
277 struct task_struct *task, unsigned long totalpages)
David Rientjes462607e2012-07-31 16:43:40 -0700278{
David Rientjes6e0fc462015-09-08 15:00:36 -0700279 if (oom_unkillable_task(task, NULL, oc->nodemask))
David Rientjes462607e2012-07-31 16:43:40 -0700280 return OOM_SCAN_CONTINUE;
281
282 /*
283 * This task already has access to memory reserves and is being killed.
284 * Don't allow any other task to have access to the reserves.
285 */
Tetsuo Handaf44666b2016-05-20 16:57:27 -0700286 if (!is_sysrq_oom(oc) && atomic_read(&task->signal->oom_victims))
287 return OOM_SCAN_ABORT;
David Rientjes462607e2012-07-31 16:43:40 -0700288
David Rientjese1e12d22012-12-11 16:02:56 -0800289 /*
290 * If task is allocating a lot of memory and has been marked to be
291 * killed first if it triggers an oom, then select it.
292 */
293 if (oom_task_origin(task))
294 return OOM_SCAN_SELECT;
295
David Rientjes462607e2012-07-31 16:43:40 -0700296 return OOM_SCAN_OK;
297}
298
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800299/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 * Simple selection loop. We chose the process with the highest
Rusty Russell6b4f2b52013-07-15 11:24:08 +0930301 * number of 'points'. Returns -1 on scan abort.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302 */
David Rientjes6e0fc462015-09-08 15:00:36 -0700303static struct task_struct *select_bad_process(struct oom_control *oc,
304 unsigned int *ppoints, unsigned long totalpages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
Tetsuo Handaf44666b2016-05-20 16:57:27 -0700306 struct task_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 struct task_struct *chosen = NULL;
David Rientjesa7f638f2012-05-29 15:06:47 -0700308 unsigned long chosen_points = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309
David Rientjes6b0c81b2012-07-31 16:43:45 -0700310 rcu_read_lock();
Tetsuo Handaf44666b2016-05-20 16:57:27 -0700311 for_each_process(p) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700312 unsigned int points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700313
David Rientjes6e0fc462015-09-08 15:00:36 -0700314 switch (oom_scan_process_thread(oc, p, totalpages)) {
David Rientjes462607e2012-07-31 16:43:40 -0700315 case OOM_SCAN_SELECT:
316 chosen = p;
317 chosen_points = ULONG_MAX;
318 /* fall through */
319 case OOM_SCAN_CONTINUE:
Andrey Vagin30e2b412011-03-22 16:30:11 -0700320 continue;
David Rientjes462607e2012-07-31 16:43:40 -0700321 case OOM_SCAN_ABORT:
David Rientjes6b0c81b2012-07-31 16:43:45 -0700322 rcu_read_unlock();
Rusty Russell6b4f2b52013-07-15 11:24:08 +0930323 return (struct task_struct *)(-1UL);
David Rientjes462607e2012-07-31 16:43:40 -0700324 case OOM_SCAN_OK:
325 break;
326 };
David Rientjes6e0fc462015-09-08 15:00:36 -0700327 points = oom_badness(p, NULL, oc->nodemask, totalpages);
David Rientjesd49ad932014-01-23 15:53:34 -0800328 if (!points || points < chosen_points)
329 continue;
David Rientjesd49ad932014-01-23 15:53:34 -0800330
331 chosen = p;
332 chosen_points = points;
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800333 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700334 if (chosen)
335 get_task_struct(chosen);
336 rcu_read_unlock();
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700337
David Rientjesa7f638f2012-05-29 15:06:47 -0700338 *ppoints = chosen_points * 1000 / totalpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 return chosen;
340}
341
342/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700343 * dump_tasks - dump current memory state of all system tasks
Wanpeng Lidad75572012-06-20 12:53:01 -0700344 * @memcg: current's memory controller, if constrained
David Rientjese85bfd32010-09-22 13:05:10 -0700345 * @nodemask: nodemask passed to page allocator for mempolicy ooms
Randy Dunlap1b578df2008-03-19 17:00:42 -0700346 *
David Rientjese85bfd32010-09-22 13:05:10 -0700347 * Dumps the current memory state of all eligible tasks. Tasks not in the same
348 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
349 * are not shown.
David Rientjesde34d962012-07-31 16:42:56 -0700350 * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes,
351 * swapents, oom_score_adj value, and name.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800352 */
Johannes Weiner2314b422014-12-10 15:44:33 -0800353static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask)
David Rientjesfef1bdd2008-02-07 00:14:07 -0800354{
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700355 struct task_struct *p;
356 struct task_struct *task;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800357
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800358 pr_info("[ pid ] uid tgid total_vm rss nr_ptes nr_pmds swapents oom_score_adj name\n");
David Rientjes6b0c81b2012-07-31 16:43:45 -0700359 rcu_read_lock();
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700360 for_each_process(p) {
Johannes Weiner72835c82012-01-12 17:18:32 -0800361 if (oom_unkillable_task(p, memcg, nodemask))
David Rientjesfef1bdd2008-02-07 00:14:07 -0800362 continue;
363
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700364 task = find_lock_task_mm(p);
365 if (!task) {
David Rientjes6d2661e2009-05-28 14:34:19 -0700366 /*
David Rientjes74ab7f12010-08-09 17:18:46 -0700367 * This is a kthread or all of p's threads have already
368 * detached their mm's. There's no need to report
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700369 * them; they can't be oom killed anyway.
David Rientjes6d2661e2009-05-28 14:34:19 -0700370 */
David Rientjes6d2661e2009-05-28 14:34:19 -0700371 continue;
372 }
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700373
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800374 pr_info("[%5d] %5d %5d %8lu %8lu %7ld %7ld %8lu %5hd %s\n",
Eric W. Biederman078de5f2012-02-08 07:00:08 -0800375 task->pid, from_kuid(&init_user_ns, task_uid(task)),
376 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
Kirill A. Shutemove1f56c82013-11-14 14:30:48 -0800377 atomic_long_read(&task->mm->nr_ptes),
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800378 mm_nr_pmds(task->mm),
David Rientjesde34d962012-07-31 16:42:56 -0700379 get_mm_counter(task->mm, MM_SWAPENTS),
David Rientjesa63d83f2010-08-09 17:19:46 -0700380 task->signal->oom_score_adj, task->comm);
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700381 task_unlock(task);
382 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700383 rcu_read_unlock();
David Rientjesfef1bdd2008-02-07 00:14:07 -0800384}
385
David Rientjes6e0fc462015-09-08 15:00:36 -0700386static void dump_header(struct oom_control *oc, struct task_struct *p,
387 struct mem_cgroup *memcg)
David Rientjes1b604d72009-12-14 17:57:47 -0800388{
Joe Perches756a025f02016-03-17 14:19:47 -0700389 pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), order=%d, oom_score_adj=%hd\n",
Vlastimil Babkaa0795cd2016-03-15 14:56:05 -0700390 current->comm, oc->gfp_mask, &oc->gfp_mask, oc->order,
David Rientjesa63d83f2010-08-09 17:19:46 -0700391 current->signal->oom_score_adj);
Vlastimil Babkaa0795cd2016-03-15 14:56:05 -0700392
David Rientjesda39da32015-11-05 18:48:05 -0800393 cpuset_print_current_mems_allowed();
David Rientjes1b604d72009-12-14 17:57:47 -0800394 dump_stack();
Sha Zhengju58cf1882013-02-22 16:32:05 -0800395 if (memcg)
396 mem_cgroup_print_oom_info(memcg, p);
397 else
398 show_mem(SHOW_MEM_FILTER_NODES);
David Rientjes1b604d72009-12-14 17:57:47 -0800399 if (sysctl_oom_dump_tasks)
David Rientjes6e0fc462015-09-08 15:00:36 -0700400 dump_tasks(memcg, oc->nodemask);
David Rientjes1b604d72009-12-14 17:57:47 -0800401}
402
Michal Hocko5695be12014-10-20 18:12:32 +0200403/*
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800404 * Number of OOM victims in flight
Michal Hocko5695be12014-10-20 18:12:32 +0200405 */
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800406static atomic_t oom_victims = ATOMIC_INIT(0);
407static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
Michal Hocko5695be12014-10-20 18:12:32 +0200408
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800409bool oom_killer_disabled __read_mostly;
Michal Hocko5695be12014-10-20 18:12:32 +0200410
Michal Hockobc448e82016-03-25 14:20:30 -0700411#define K(x) ((x) << (PAGE_SHIFT-10))
412
Michal Hocko3ef22df2016-05-19 17:13:12 -0700413/*
414 * task->mm can be NULL if the task is the exited group leader. So to
415 * determine whether the task is using a particular mm, we examine all the
416 * task's threads: if one of those is using this mm then this task was also
417 * using it.
418 */
419static bool process_shares_mm(struct task_struct *p, struct mm_struct *mm)
420{
421 struct task_struct *t;
422
423 for_each_thread(p, t) {
424 struct mm_struct *t_mm = READ_ONCE(t->mm);
425 if (t_mm)
426 return t_mm == mm;
427 }
428 return false;
429}
430
431
Michal Hockoaac45362016-03-25 14:20:24 -0700432#ifdef CONFIG_MMU
433/*
434 * OOM Reaper kernel thread which tries to reap the memory used by the OOM
435 * victim (if that is possible) to help the OOM killer to move on.
436 */
437static struct task_struct *oom_reaper_th;
Michal Hockoaac45362016-03-25 14:20:24 -0700438static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait);
Vladimir Davydov29c696e2016-03-25 14:20:39 -0700439static struct task_struct *oom_reaper_list;
Michal Hocko03049262016-03-25 14:20:33 -0700440static DEFINE_SPINLOCK(oom_reaper_lock);
441
Michal Hocko36324a92016-03-25 14:20:27 -0700442static bool __oom_reap_task(struct task_struct *tsk)
Michal Hockoaac45362016-03-25 14:20:24 -0700443{
444 struct mmu_gather tlb;
445 struct vm_area_struct *vma;
Michal Hockoe2fe1452016-05-27 14:27:35 -0700446 struct mm_struct *mm = NULL;
Michal Hocko36324a92016-03-25 14:20:27 -0700447 struct task_struct *p;
Michal Hockoaac45362016-03-25 14:20:24 -0700448 struct zap_details details = {.check_swap_entries = true,
449 .ignore_dirty = true};
450 bool ret = true;
451
Michal Hocko36324a92016-03-25 14:20:27 -0700452 /*
Michal Hockoe2fe1452016-05-27 14:27:35 -0700453 * We have to make sure to not race with the victim exit path
454 * and cause premature new oom victim selection:
455 * __oom_reap_task exit_mm
456 * atomic_inc_not_zero
457 * mmput
458 * atomic_dec_and_test
459 * exit_oom_victim
460 * [...]
461 * out_of_memory
462 * select_bad_process
463 * # no TIF_MEMDIE task selects new victim
464 * unmap_page_range # frees some memory
465 */
466 mutex_lock(&oom_lock);
467
468 /*
Michal Hocko36324a92016-03-25 14:20:27 -0700469 * Make sure we find the associated mm_struct even when the particular
470 * thread has already terminated and cleared its mm.
471 * We might have race with exit path so consider our work done if there
472 * is no mm.
473 */
474 p = find_lock_task_mm(tsk);
475 if (!p)
Michal Hockoe2fe1452016-05-27 14:27:35 -0700476 goto unlock_oom;
Michal Hocko36324a92016-03-25 14:20:27 -0700477 mm = p->mm;
Tetsuo Handa9df10fb2016-06-24 14:48:38 -0700478 atomic_inc(&mm->mm_users);
Michal Hocko36324a92016-03-25 14:20:27 -0700479 task_unlock(p);
480
Michal Hockoaac45362016-03-25 14:20:24 -0700481 if (!down_read_trylock(&mm->mmap_sem)) {
482 ret = false;
Michal Hockoe2fe1452016-05-27 14:27:35 -0700483 goto unlock_oom;
Michal Hockoaac45362016-03-25 14:20:24 -0700484 }
485
486 tlb_gather_mmu(&tlb, mm, 0, -1);
487 for (vma = mm->mmap ; vma; vma = vma->vm_next) {
488 if (is_vm_hugetlb_page(vma))
489 continue;
490
491 /*
492 * mlocked VMAs require explicit munlocking before unmap.
493 * Let's keep it simple here and skip such VMAs.
494 */
495 if (vma->vm_flags & VM_LOCKED)
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 unmap_page_range(&tlb, vma, vma->vm_start, vma->vm_end,
510 &details);
511 }
512 tlb_finish_mmu(&tlb, 0, -1);
Michal Hockobc448e82016-03-25 14:20:30 -0700513 pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
514 task_pid_nr(tsk), tsk->comm,
515 K(get_mm_counter(mm, MM_ANONPAGES)),
516 K(get_mm_counter(mm, MM_FILEPAGES)),
517 K(get_mm_counter(mm, MM_SHMEMPAGES)));
Michal Hockoaac45362016-03-25 14:20:24 -0700518 up_read(&mm->mmap_sem);
Michal Hocko36324a92016-03-25 14:20:27 -0700519
520 /*
Michal Hocko449d7772016-05-19 17:13:15 -0700521 * This task can be safely ignored because we cannot do much more
522 * to release its memory.
Michal Hocko36324a92016-03-25 14:20:27 -0700523 */
Michal Hockobb8a4b72016-05-20 16:57:18 -0700524 set_bit(MMF_OOM_REAPED, &mm->flags);
Michal Hockoe2fe1452016-05-27 14:27:35 -0700525unlock_oom:
526 mutex_unlock(&oom_lock);
Michal Hockoec8d7c12016-05-20 16:57:21 -0700527 /*
528 * Drop our reference but make sure the mmput slow path is called from a
529 * different context because we shouldn't risk we get stuck there and
530 * put the oom_reaper out of the way.
531 */
Michal Hockoe2fe1452016-05-27 14:27:35 -0700532 if (mm)
533 mmput_async(mm);
Michal Hockoaac45362016-03-25 14:20:24 -0700534 return ret;
535}
536
Michal Hockobc448e82016-03-25 14:20:30 -0700537#define MAX_OOM_REAP_RETRIES 10
Michal Hocko36324a92016-03-25 14:20:27 -0700538static void oom_reap_task(struct task_struct *tsk)
Michal Hockoaac45362016-03-25 14:20:24 -0700539{
540 int attempts = 0;
541
542 /* Retry the down_read_trylock(mmap_sem) a few times */
Michal Hockobc448e82016-03-25 14:20:30 -0700543 while (attempts++ < MAX_OOM_REAP_RETRIES && !__oom_reap_task(tsk))
Michal Hockoaac45362016-03-25 14:20:24 -0700544 schedule_timeout_idle(HZ/10);
545
Michal Hockobc448e82016-03-25 14:20:30 -0700546 if (attempts > MAX_OOM_REAP_RETRIES) {
547 pr_info("oom_reaper: unable to reap pid:%d (%s)\n",
548 task_pid_nr(tsk), tsk->comm);
549 debug_show_all_locks();
550 }
551
Michal Hocko449d7772016-05-19 17:13:15 -0700552 /*
553 * Clear TIF_MEMDIE because the task shouldn't be sitting on a
554 * reasonably reclaimable memory anymore or it is not a good candidate
555 * for the oom victim right now because it cannot release its memory
556 * itself nor by the oom reaper.
557 */
558 tsk->oom_reaper_list = NULL;
559 exit_oom_victim(tsk);
560
Michal Hockoaac45362016-03-25 14:20:24 -0700561 /* Drop a reference taken by wake_oom_reaper */
Michal Hocko36324a92016-03-25 14:20:27 -0700562 put_task_struct(tsk);
Michal Hockoaac45362016-03-25 14:20:24 -0700563}
564
565static int oom_reaper(void *unused)
566{
Michal Hockoe2679602016-03-25 14:20:41 -0700567 set_freezable();
568
Michal Hockoaac45362016-03-25 14:20:24 -0700569 while (true) {
Michal Hocko03049262016-03-25 14:20:33 -0700570 struct task_struct *tsk = NULL;
Michal Hockoaac45362016-03-25 14:20:24 -0700571
Vladimir Davydov29c696e2016-03-25 14:20:39 -0700572 wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL);
Michal Hocko03049262016-03-25 14:20:33 -0700573 spin_lock(&oom_reaper_lock);
Vladimir Davydov29c696e2016-03-25 14:20:39 -0700574 if (oom_reaper_list != NULL) {
575 tsk = oom_reaper_list;
576 oom_reaper_list = tsk->oom_reaper_list;
Michal Hocko03049262016-03-25 14:20:33 -0700577 }
578 spin_unlock(&oom_reaper_lock);
579
580 if (tsk)
581 oom_reap_task(tsk);
Michal Hockoaac45362016-03-25 14:20:24 -0700582 }
583
584 return 0;
585}
586
Michal Hocko36324a92016-03-25 14:20:27 -0700587static void wake_oom_reaper(struct task_struct *tsk)
Michal Hockoaac45362016-03-25 14:20:24 -0700588{
Michal Hockoaf8e15c2016-04-01 14:31:34 -0700589 if (!oom_reaper_th)
590 return;
591
592 /* tsk is already queued? */
593 if (tsk == oom_reaper_list || tsk->oom_reaper_list)
Michal Hockoaac45362016-03-25 14:20:24 -0700594 return;
595
Michal Hocko36324a92016-03-25 14:20:27 -0700596 get_task_struct(tsk);
Michal Hockoaac45362016-03-25 14:20:24 -0700597
Michal Hocko03049262016-03-25 14:20:33 -0700598 spin_lock(&oom_reaper_lock);
Vladimir Davydov29c696e2016-03-25 14:20:39 -0700599 tsk->oom_reaper_list = oom_reaper_list;
600 oom_reaper_list = tsk;
Michal Hocko03049262016-03-25 14:20:33 -0700601 spin_unlock(&oom_reaper_lock);
602 wake_up(&oom_reaper_wait);
Michal Hockoaac45362016-03-25 14:20:24 -0700603}
604
Michal Hocko3ef22df2016-05-19 17:13:12 -0700605/* Check if we can reap the given task. This has to be called with stable
606 * tsk->mm
607 */
608void try_oom_reaper(struct task_struct *tsk)
609{
610 struct mm_struct *mm = tsk->mm;
611 struct task_struct *p;
612
613 if (!mm)
614 return;
615
616 /*
617 * There might be other threads/processes which are either not
618 * dying or even not killable.
619 */
620 if (atomic_read(&mm->mm_users) > 1) {
621 rcu_read_lock();
622 for_each_process(p) {
Michal Hocko3ef22df2016-05-19 17:13:12 -0700623 if (!process_shares_mm(p, mm))
624 continue;
Michal Hocko3ef22df2016-05-19 17:13:12 -0700625 if (fatal_signal_pending(p))
626 continue;
627
628 /*
629 * If the task is exiting make sure the whole thread group
630 * is exiting and cannot acces mm anymore.
631 */
Michal Hockocbdcf7f2016-06-03 14:55:55 -0700632 if (signal_group_exit(p->signal))
Michal Hocko3ef22df2016-05-19 17:13:12 -0700633 continue;
634
635 /* Give up */
636 rcu_read_unlock();
637 return;
638 }
639 rcu_read_unlock();
640 }
641
642 wake_oom_reaper(tsk);
643}
644
Michal Hockoaac45362016-03-25 14:20:24 -0700645static int __init oom_init(void)
646{
647 oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper");
648 if (IS_ERR(oom_reaper_th)) {
649 pr_err("Unable to start OOM reaper %ld. Continuing regardless\n",
650 PTR_ERR(oom_reaper_th));
651 oom_reaper_th = NULL;
652 }
653 return 0;
654}
655subsys_initcall(oom_init)
656#else
Michal Hocko36324a92016-03-25 14:20:27 -0700657static void wake_oom_reaper(struct task_struct *tsk)
Michal Hockoaac45362016-03-25 14:20:24 -0700658{
659}
660#endif
661
Michal Hocko49550b62015-02-11 15:26:12 -0800662/**
Johannes Weiner16e95192015-06-24 16:57:07 -0700663 * mark_oom_victim - mark the given task as OOM victim
Michal Hocko49550b62015-02-11 15:26:12 -0800664 * @tsk: task to mark
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800665 *
Johannes Weinerdc564012015-06-24 16:57:19 -0700666 * Has to be called with oom_lock held and never after
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800667 * oom has been disabled already.
Michal Hocko49550b62015-02-11 15:26:12 -0800668 */
Johannes Weiner16e95192015-06-24 16:57:07 -0700669void mark_oom_victim(struct task_struct *tsk)
Michal Hocko49550b62015-02-11 15:26:12 -0800670{
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800671 WARN_ON(oom_killer_disabled);
672 /* OOM killer might race with memcg OOM */
673 if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
674 return;
Tetsuo Handaf44666b2016-05-20 16:57:27 -0700675 atomic_inc(&tsk->signal->oom_victims);
Michal Hocko63a8ca92015-02-11 15:26:15 -0800676 /*
677 * Make sure that the task is woken up from uninterruptible sleep
678 * if it is frozen because OOM killer wouldn't be able to free
679 * any memory and livelock. freezing_slow_path will tell the freezer
680 * that TIF_MEMDIE tasks should be ignored.
681 */
682 __thaw_task(tsk);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800683 atomic_inc(&oom_victims);
Michal Hocko49550b62015-02-11 15:26:12 -0800684}
685
686/**
Johannes Weiner16e95192015-06-24 16:57:07 -0700687 * exit_oom_victim - note the exit of an OOM victim
Michal Hocko49550b62015-02-11 15:26:12 -0800688 */
Michal Hocko36324a92016-03-25 14:20:27 -0700689void exit_oom_victim(struct task_struct *tsk)
Michal Hocko49550b62015-02-11 15:26:12 -0800690{
Michal Hocko36324a92016-03-25 14:20:27 -0700691 if (!test_and_clear_tsk_thread_flag(tsk, TIF_MEMDIE))
692 return;
Tetsuo Handaf44666b2016-05-20 16:57:27 -0700693 atomic_dec(&tsk->signal->oom_victims);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800694
Johannes Weinerc38f1022015-06-24 16:57:13 -0700695 if (!atomic_dec_return(&oom_victims))
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800696 wake_up_all(&oom_victims_wait);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800697}
698
699/**
700 * oom_killer_disable - disable OOM killer
701 *
702 * Forces all page allocations to fail rather than trigger OOM killer.
703 * Will block and wait until all OOM victims are killed.
704 *
705 * The function cannot be called when there are runnable user tasks because
706 * the userspace would see unexpected allocation failures as a result. Any
707 * new usage of this function should be consulted with MM people.
708 *
709 * Returns true if successful and false if the OOM killer cannot be
710 * disabled.
711 */
712bool oom_killer_disable(void)
713{
714 /*
Tetsuo Handa6afcf282016-03-17 14:20:45 -0700715 * Make sure to not race with an ongoing OOM killer. Check that the
716 * current is not killed (possibly due to sharing the victim's memory).
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800717 */
Tetsuo Handa6afcf282016-03-17 14:20:45 -0700718 if (mutex_lock_killable(&oom_lock))
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800719 return false;
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800720 oom_killer_disabled = true;
Johannes Weinerdc564012015-06-24 16:57:19 -0700721 mutex_unlock(&oom_lock);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800722
723 wait_event(oom_victims_wait, !atomic_read(&oom_victims));
724
725 return true;
726}
727
728/**
729 * oom_killer_enable - enable OOM killer
730 */
731void oom_killer_enable(void)
732{
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800733 oom_killer_disabled = false;
Michal Hocko49550b62015-02-11 15:26:12 -0800734}
735
Oleg Nesterov4d7b3392015-11-05 18:48:26 -0800736/*
David Rientjes6b0c81b2012-07-31 16:43:45 -0700737 * Must be called while holding a reference to p, which will be released upon
738 * returning.
739 */
David Rientjes6e0fc462015-09-08 15:00:36 -0700740void oom_kill_process(struct oom_control *oc, struct task_struct *p,
David Rientjes9cbb78b2012-07-31 16:43:44 -0700741 unsigned int points, unsigned long totalpages,
David Rientjes6e0fc462015-09-08 15:00:36 -0700742 struct mem_cgroup *memcg, const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743{
Linus Torvalds52d3c032011-03-14 15:17:07 -0700744 struct task_struct *victim = p;
David Rientjes5e9d8342010-08-09 17:18:51 -0700745 struct task_struct *child;
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800746 struct task_struct *t;
David Rientjes647f2bd2012-03-21 16:33:46 -0700747 struct mm_struct *mm;
Linus Torvalds52d3c032011-03-14 15:17:07 -0700748 unsigned int victim_points = 0;
David Rientjesdc3f21e2012-03-21 16:33:47 -0700749 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
750 DEFAULT_RATELIMIT_BURST);
Tetsuo Handabb299022016-03-25 14:20:44 -0700751 bool can_oom_reap = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700753 /*
754 * If the task is already exiting, don't alarm the sysadmin or kill
755 * its children or threads, just set TIF_MEMDIE so it can die quickly
756 */
Michal Hocko83363b92015-02-11 15:24:56 -0800757 task_lock(p);
758 if (p->mm && task_will_free_mem(p)) {
Johannes Weiner16e95192015-06-24 16:57:07 -0700759 mark_oom_victim(p);
Michal Hocko3ef22df2016-05-19 17:13:12 -0700760 try_oom_reaper(p);
Michal Hocko83363b92015-02-11 15:24:56 -0800761 task_unlock(p);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700762 put_task_struct(p);
David Rientjes2a1c9b12012-03-21 16:33:46 -0700763 return;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700764 }
Michal Hocko83363b92015-02-11 15:24:56 -0800765 task_unlock(p);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700766
David Rientjesdc3f21e2012-03-21 16:33:47 -0700767 if (__ratelimit(&oom_rs))
David Rientjes6e0fc462015-09-08 15:00:36 -0700768 dump_header(oc, p, memcg);
David Rientjes8447d952012-03-21 16:33:47 -0700769
Wang Longf0d66472015-06-24 16:58:01 -0700770 pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n",
David Rientjes5e9d8342010-08-09 17:18:51 -0700771 message, task_pid_nr(p), p->comm, points);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800772
David Rientjes5e9d8342010-08-09 17:18:51 -0700773 /*
774 * If any of p's children has a different mm and is eligible for kill,
David Rientjes11239832011-07-25 17:12:17 -0700775 * the one with the highest oom_badness() score is sacrificed for its
David Rientjes5e9d8342010-08-09 17:18:51 -0700776 * parent. This attempts to lose the minimal amount of work done while
777 * still freeing memory.
778 */
David Rientjes6b0c81b2012-07-31 16:43:45 -0700779 read_lock(&tasklist_lock);
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800780 for_each_thread(p, t) {
David Rientjes5e9d8342010-08-09 17:18:51 -0700781 list_for_each_entry(child, &t->children, sibling) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700782 unsigned int child_points;
David Rientjes5e9d8342010-08-09 17:18:51 -0700783
Oleg Nesterov4d7b3392015-11-05 18:48:26 -0800784 if (process_shares_mm(child, p->mm))
David Rientjesedd45542011-03-22 16:30:12 -0700785 continue;
David Rientjesa63d83f2010-08-09 17:19:46 -0700786 /*
787 * oom_badness() returns 0 if the thread is unkillable
788 */
David Rientjes6e0fc462015-09-08 15:00:36 -0700789 child_points = oom_badness(child, memcg, oc->nodemask,
David Rientjesa63d83f2010-08-09 17:19:46 -0700790 totalpages);
David Rientjes5e9d8342010-08-09 17:18:51 -0700791 if (child_points > victim_points) {
David Rientjes6b0c81b2012-07-31 16:43:45 -0700792 put_task_struct(victim);
David Rientjes5e9d8342010-08-09 17:18:51 -0700793 victim = child;
794 victim_points = child_points;
David Rientjes6b0c81b2012-07-31 16:43:45 -0700795 get_task_struct(victim);
David Rientjes5e9d8342010-08-09 17:18:51 -0700796 }
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700797 }
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800798 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700799 read_unlock(&tasklist_lock);
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700800
David Rientjes6b0c81b2012-07-31 16:43:45 -0700801 p = find_lock_task_mm(victim);
802 if (!p) {
David Rientjes6b0c81b2012-07-31 16:43:45 -0700803 put_task_struct(victim);
David Rientjes647f2bd2012-03-21 16:33:46 -0700804 return;
David Rientjes6b0c81b2012-07-31 16:43:45 -0700805 } else if (victim != p) {
806 get_task_struct(p);
807 put_task_struct(victim);
808 victim = p;
809 }
David Rientjes647f2bd2012-03-21 16:33:46 -0700810
Tetsuo Handa880b7682015-11-05 18:47:51 -0800811 /* Get a reference to safely compare mm after task_unlock(victim) */
David Rientjes647f2bd2012-03-21 16:33:46 -0700812 mm = victim->mm;
Tetsuo Handa880b7682015-11-05 18:47:51 -0800813 atomic_inc(&mm->mm_count);
Tetsuo Handa426fb5e2015-11-05 18:47:44 -0800814 /*
815 * We should send SIGKILL before setting TIF_MEMDIE in order to prevent
816 * the OOM victim from depleting the memory reserves from the user
817 * space under its control.
818 */
819 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
Johannes Weiner16e95192015-06-24 16:57:07 -0700820 mark_oom_victim(victim);
Jerome Marchandeca56ff2016-01-14 15:19:26 -0800821 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 -0700822 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
823 K(get_mm_counter(victim->mm, MM_ANONPAGES)),
Jerome Marchandeca56ff2016-01-14 15:19:26 -0800824 K(get_mm_counter(victim->mm, MM_FILEPAGES)),
825 K(get_mm_counter(victim->mm, MM_SHMEMPAGES)));
David Rientjes647f2bd2012-03-21 16:33:46 -0700826 task_unlock(victim);
827
828 /*
829 * Kill all user processes sharing victim->mm in other thread groups, if
830 * any. They don't get access to memory reserves, though, to avoid
831 * depletion of all memory. This prevents mm->mmap_sem livelock when an
832 * oom killed thread cannot exit because it requires the semaphore and
833 * its contended by another thread trying to allocate memory itself.
834 * That thread will now get access to memory reserves since it has a
835 * pending fatal signal.
836 */
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800837 rcu_read_lock();
Oleg Nesterovc3190252015-11-05 18:48:23 -0800838 for_each_process(p) {
Oleg Nesterov4d7b3392015-11-05 18:48:26 -0800839 if (!process_shares_mm(p, mm))
Oleg Nesterovc3190252015-11-05 18:48:23 -0800840 continue;
841 if (same_thread_group(p, victim))
842 continue;
Michal Hockoaac45362016-03-25 14:20:24 -0700843 if (unlikely(p->flags & PF_KTHREAD) || is_global_init(p) ||
844 p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
845 /*
846 * We cannot use oom_reaper for the mm shared by this
847 * process because it wouldn't get killed and so the
848 * memory might be still used.
849 */
850 can_oom_reap = false;
Oleg Nesterovc3190252015-11-05 18:48:23 -0800851 continue;
Michal Hockoaac45362016-03-25 14:20:24 -0700852 }
Oleg Nesterovc3190252015-11-05 18:48:23 -0800853 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
854 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700855 rcu_read_unlock();
David Rientjes647f2bd2012-03-21 16:33:46 -0700856
Michal Hockoaac45362016-03-25 14:20:24 -0700857 if (can_oom_reap)
Michal Hocko36324a92016-03-25 14:20:27 -0700858 wake_oom_reaper(victim);
Michal Hockoaac45362016-03-25 14:20:24 -0700859
Tetsuo Handa880b7682015-11-05 18:47:51 -0800860 mmdrop(mm);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700861 put_task_struct(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862}
David Rientjes647f2bd2012-03-21 16:33:46 -0700863#undef K
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864
David Rientjes309ed882010-08-09 17:18:54 -0700865/*
866 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
867 */
David Rientjes6e0fc462015-09-08 15:00:36 -0700868void check_panic_on_oom(struct oom_control *oc, enum oom_constraint constraint,
Balasubramani Vivekanandan2415b9f2015-04-14 15:48:18 -0700869 struct mem_cgroup *memcg)
David Rientjes309ed882010-08-09 17:18:54 -0700870{
871 if (likely(!sysctl_panic_on_oom))
872 return;
873 if (sysctl_panic_on_oom != 2) {
874 /*
875 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
876 * does not panic for cpuset, mempolicy, or memcg allocation
877 * failures.
878 */
879 if (constraint != CONSTRAINT_NONE)
880 return;
881 }
David Rientjes071a4be2015-09-08 15:00:42 -0700882 /* Do not panic for oom kills triggered by sysrq */
Yaowei Baidb2a0dd2015-11-06 16:28:06 -0800883 if (is_sysrq_oom(oc))
David Rientjes071a4be2015-09-08 15:00:42 -0700884 return;
David Rientjes6e0fc462015-09-08 15:00:36 -0700885 dump_header(oc, NULL, memcg);
David Rientjes309ed882010-08-09 17:18:54 -0700886 panic("Out of memory: %s panic_on_oom is enabled\n",
887 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
888}
889
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700890static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
891
892int register_oom_notifier(struct notifier_block *nb)
893{
894 return blocking_notifier_chain_register(&oom_notify_list, nb);
895}
896EXPORT_SYMBOL_GPL(register_oom_notifier);
897
898int unregister_oom_notifier(struct notifier_block *nb)
899{
900 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
901}
902EXPORT_SYMBOL_GPL(unregister_oom_notifier);
903
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904/**
David Rientjes6e0fc462015-09-08 15:00:36 -0700905 * out_of_memory - kill the "best" process when we run out of memory
906 * @oc: pointer to struct oom_control
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907 *
908 * If we run out of memory, we have the choice between either
909 * killing a random task (bad), letting the system crash (worse)
910 * OR try to be smart about which process to kill. Note that we
911 * don't have to be perfect here, we just have to be good.
912 */
David Rientjes6e0fc462015-09-08 15:00:36 -0700913bool out_of_memory(struct oom_control *oc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914{
David Rientjes0aad4b32010-08-09 17:18:59 -0700915 struct task_struct *p;
David Rientjesa63d83f2010-08-09 17:19:46 -0700916 unsigned long totalpages;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700917 unsigned long freed = 0;
David Rientjes9cbb78b2012-07-31 16:43:44 -0700918 unsigned int uninitialized_var(points);
David Rientjese3658932010-08-09 17:18:55 -0700919 enum oom_constraint constraint = CONSTRAINT_NONE;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700920
Johannes Weinerdc564012015-06-24 16:57:19 -0700921 if (oom_killer_disabled)
922 return false;
923
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700924 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
925 if (freed > 0)
926 /* Got some memory back in the last second. */
David Rientjes75e8f8b2015-09-08 15:00:47 -0700927 return true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928
David Rientjes7b98c2e2010-08-09 17:18:48 -0700929 /*
David Rientjes9ff48682012-12-11 16:01:30 -0800930 * If current has a pending SIGKILL or is exiting, then automatically
931 * select it. The goal is to allow it to allocate so that it may
932 * quickly exit and free its memory.
Tetsuo Handad7a94e72015-02-11 15:24:54 -0800933 *
934 * But don't select if current has already released its mm and cleared
935 * TIF_MEMDIE flag at exit_mm(), otherwise an OOM livelock may occur.
David Rientjes7b98c2e2010-08-09 17:18:48 -0700936 */
Tetsuo Handad7a94e72015-02-11 15:24:54 -0800937 if (current->mm &&
938 (fatal_signal_pending(current) || task_will_free_mem(current))) {
Johannes Weiner16e95192015-06-24 16:57:07 -0700939 mark_oom_victim(current);
Michal Hocko3ef22df2016-05-19 17:13:12 -0700940 try_oom_reaper(current);
David Rientjes75e8f8b2015-09-08 15:00:47 -0700941 return true;
David Rientjes7b98c2e2010-08-09 17:18:48 -0700942 }
943
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800944 /*
Michal Hocko3da88fb2016-05-19 17:13:09 -0700945 * The OOM killer does not compensate for IO-less reclaim.
946 * pagefault_out_of_memory lost its gfp context so we have to
947 * make sure exclude 0 mask - all other users should have at least
948 * ___GFP_DIRECT_RECLAIM to get here.
949 */
950 if (oc->gfp_mask && !(oc->gfp_mask & (__GFP_FS|__GFP_NOFAIL)))
951 return true;
952
953 /*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800954 * Check if there were limitations on the allocation (only relevant for
955 * NUMA) that may require different handling.
956 */
David Rientjes6e0fc462015-09-08 15:00:36 -0700957 constraint = constrained_alloc(oc, &totalpages);
958 if (constraint != CONSTRAINT_MEMORY_POLICY)
959 oc->nodemask = NULL;
960 check_panic_on_oom(oc, constraint, NULL);
David Rientjes0aad4b32010-08-09 17:18:59 -0700961
David Rientjes121d1ba2012-07-31 16:42:55 -0700962 if (sysctl_oom_kill_allocating_task && current->mm &&
David Rientjes6e0fc462015-09-08 15:00:36 -0700963 !oom_unkillable_task(current, NULL, oc->nodemask) &&
David Rientjes121d1ba2012-07-31 16:42:55 -0700964 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
David Rientjes6b0c81b2012-07-31 16:43:45 -0700965 get_task_struct(current);
David Rientjes6e0fc462015-09-08 15:00:36 -0700966 oom_kill_process(oc, current, 0, totalpages, NULL,
David Rientjes2a1c9b12012-03-21 16:33:46 -0700967 "Out of memory (oom_kill_allocating_task)");
David Rientjes75e8f8b2015-09-08 15:00:47 -0700968 return true;
David Rientjes0aad4b32010-08-09 17:18:59 -0700969 }
970
David Rientjes6e0fc462015-09-08 15:00:36 -0700971 p = select_bad_process(oc, &points, totalpages);
David Rientjes0aad4b32010-08-09 17:18:59 -0700972 /* Found nothing?!?! Either we hang forever, or we panic. */
Yaowei Baidb2a0dd2015-11-06 16:28:06 -0800973 if (!p && !is_sysrq_oom(oc)) {
David Rientjes6e0fc462015-09-08 15:00:36 -0700974 dump_header(oc, NULL, NULL);
David Rientjes0aad4b32010-08-09 17:18:59 -0700975 panic("Out of memory and no killable processes...\n");
976 }
David Rientjes071a4be2015-09-08 15:00:42 -0700977 if (p && p != (void *)-1UL) {
David Rientjes6e0fc462015-09-08 15:00:36 -0700978 oom_kill_process(oc, p, points, totalpages, NULL,
979 "Out of memory");
David Rientjes75e8f8b2015-09-08 15:00:47 -0700980 /*
981 * Give the killed process a good chance to exit before trying
982 * to allocate memory again.
983 */
David Rientjes4f774b92012-07-31 16:42:37 -0700984 schedule_timeout_killable(1);
David Rientjes75e8f8b2015-09-08 15:00:47 -0700985 }
Johannes Weinerdc564012015-06-24 16:57:19 -0700986 return true;
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800987}
988
David Rientjese3658932010-08-09 17:18:55 -0700989/*
990 * The pagefault handler calls here because it is out of memory, so kill a
David Rientjesefacd022012-12-12 13:52:06 -0800991 * memory-hogging task. If any populated zone has ZONE_OOM_LOCKED set, a
992 * parallel oom killing is already in progress so do nothing.
David Rientjese3658932010-08-09 17:18:55 -0700993 */
994void pagefault_out_of_memory(void)
995{
David Rientjes6e0fc462015-09-08 15:00:36 -0700996 struct oom_control oc = {
997 .zonelist = NULL,
998 .nodemask = NULL,
999 .gfp_mask = 0,
1000 .order = 0,
David Rientjes6e0fc462015-09-08 15:00:36 -07001001 };
1002
Johannes Weiner49426422013-10-16 13:46:59 -07001003 if (mem_cgroup_oom_synchronize(true))
Johannes Weinerdc564012015-06-24 16:57:19 -07001004 return;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001005
Johannes Weinerdc564012015-06-24 16:57:19 -07001006 if (!mutex_trylock(&oom_lock))
1007 return;
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001008
David Rientjes6e0fc462015-09-08 15:00:36 -07001009 if (!out_of_memory(&oc)) {
Johannes Weinerdc564012015-06-24 16:57:19 -07001010 /*
1011 * There shouldn't be any user tasks runnable while the
1012 * OOM killer is disabled, so the current task has to
1013 * be a racing OOM victim for which oom_killer_disable()
1014 * is waiting for.
1015 */
1016 WARN_ON(test_thread_flag(TIF_MEMDIE));
David Rientjese3658932010-08-09 17:18:55 -07001017 }
Johannes Weinerdc564012015-06-24 16:57:19 -07001018
1019 mutex_unlock(&oom_lock);
David Rientjese3658932010-08-09 17:18:55 -07001020}