blob: 044e1eed720eedab7374fcf2f37de34b80ea6980 [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/mm/oom_kill.c
4 *
5 * Copyright (C) 1998,2000 Rik van Riel
6 * Thanks go out to Claus Fischer for some serious inspiration and
7 * for goading me into coding this file...
David Rientjesa63d83f2010-08-09 17:19:46 -07008 * Copyright (C) 2010 Google, Inc.
9 * Rewritten by David Rientjes
Linus Torvalds1da177e2005-04-16 15:20:36 -070010 *
11 * The routines in this file are used to kill a process when
Paul Jacksona49335c2005-09-06 15:18:09 -070012 * we're seriously out of memory. This gets called from __alloc_pages()
13 * in mm/page_alloc.c when we really run out of memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -070014 *
15 * Since we won't call these routines often (on a well-configured
16 * machine) this file will double as a 'coding guide' and a signpost
17 * for newbie kernel hackers. It features several pointers to major
18 * kernel subsystems and hints as to where to find out what things do.
19 */
20
Alexey Dobriyan8ac773b2006-10-19 23:28:32 -070021#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/mm.h>
Alexey Dobriyan4e950f62007-07-30 02:36:13 +040023#include <linux/err.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090024#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/sched.h>
Ingo Molnar6e84f312017-02-08 18:51:29 +010026#include <linux/sched/mm.h>
Ingo Molnarf7ccbae2017-02-08 18:51:30 +010027#include <linux/sched/coredump.h>
Ingo Molnar29930022017-02-08 18:51:36 +010028#include <linux/sched/task.h>
David Rientjes8a7ff022020-01-30 22:14:26 -080029#include <linux/sched/debug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/swap.h>
Suren Baghdasaryan884a7e52021-09-02 15:00:29 -070031#include <linux/syscalls.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/timex.h>
33#include <linux/jiffies.h>
Paul Jacksonef08e3b2005-09-06 15:18:13 -070034#include <linux/cpuset.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040035#include <linux/export.h>
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -070036#include <linux/notifier.h>
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -080037#include <linux/memcontrol.h>
David Rientjes6f48d0eb2010-08-09 17:18:52 -070038#include <linux/mempolicy.h>
David Howells5cd9c582008-08-14 11:37:28 +010039#include <linux/security.h>
David Rientjesedd45542011-03-22 16:30:12 -070040#include <linux/ptrace.h>
David Rientjesf660daa2011-10-31 17:07:07 -070041#include <linux/freezer.h>
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080042#include <linux/ftrace.h>
David Rientjesdc3f21e2012-03-21 16:33:47 -070043#include <linux/ratelimit.h>
Michal Hockoaac45362016-03-25 14:20:24 -070044#include <linux/kthread.h>
45#include <linux/init.h>
Michal Hocko4d4bbd82017-10-03 16:14:50 -070046#include <linux/mmu_notifier.h>
Michal Hockoaac45362016-03-25 14:20:24 -070047
48#include <asm/tlb.h>
49#include "internal.h"
Yang Shi852d8be2017-11-15 17:32:07 -080050#include "slab.h"
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080051
52#define CREATE_TRACE_POINTS
53#include <trace/events/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054
sujiaxun43fe2192022-02-17 18:51:48 -080055static int sysctl_panic_on_oom;
56static int sysctl_oom_kill_allocating_task;
57static int sysctl_oom_dump_tasks = 1;
58
Michal Hockoa195d3f2018-08-17 15:49:10 -070059/*
60 * Serializes oom killer invocations (out_of_memory()) from all contexts to
61 * prevent from over eager oom killing (e.g. when the oom killer is invoked
62 * from different domains).
63 *
64 * oom_killer_disable() relies on this lock to stabilize oom_killer_disabled
65 * and mark_oom_victim
66 */
Johannes Weinerdc564012015-06-24 16:57:19 -070067DEFINE_MUTEX(oom_lock);
Suren Baghdasaryan67197a42020-10-13 16:58:35 -070068/* Serializes oom_score_adj and oom_score_adj_min updates */
69DEFINE_MUTEX(oom_adj_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070070
Shakeel Buttac311a12019-07-11 21:00:31 -070071static inline bool is_memcg_oom(struct oom_control *oc)
72{
73 return oc->memcg != NULL;
74}
75
David Rientjes6f48d0eb2010-08-09 17:18:52 -070076#ifdef CONFIG_NUMA
77/**
Ingo Molnarf0953a12021-05-06 18:06:47 -070078 * oom_cpuset_eligible() - check task eligibility for kill
Oleg Nesterovad962442014-01-21 15:50:00 -080079 * @start: task struct of which task to consider
Yi Wangf364f062019-09-23 15:37:14 -070080 * @oc: pointer to struct oom_control
David Rientjes6f48d0eb2010-08-09 17:18:52 -070081 *
82 * Task eligibility is determined by whether or not a candidate task, @tsk,
83 * shares the same mempolicy nodes as current if it is bound by such a policy
84 * and whether or not it has the same set of allowed cpuset nodes.
Shakeel Buttac311a12019-07-11 21:00:31 -070085 *
86 * This function is assuming oom-killer context and 'current' has triggered
87 * the oom-killer.
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070088 */
Shakeel Buttac311a12019-07-11 21:00:31 -070089static bool oom_cpuset_eligible(struct task_struct *start,
90 struct oom_control *oc)
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070091{
Oleg Nesterovad962442014-01-21 15:50:00 -080092 struct task_struct *tsk;
93 bool ret = false;
Shakeel Buttac311a12019-07-11 21:00:31 -070094 const nodemask_t *mask = oc->nodemask;
95
Oleg Nesterovad962442014-01-21 15:50:00 -080096 rcu_read_lock();
Oleg Nesterov1da4db02014-01-21 15:49:58 -080097 for_each_thread(start, tsk) {
David Rientjes6f48d0eb2010-08-09 17:18:52 -070098 if (mask) {
99 /*
100 * If this is a mempolicy constrained oom, tsk's
101 * cpuset is irrelevant. Only return true if its
102 * mempolicy intersects current, otherwise it may be
103 * needlessly killed.
104 */
Feng Tangb26e5172021-06-30 18:50:56 -0700105 ret = mempolicy_in_oom_domain(tsk, mask);
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700106 } else {
107 /*
108 * This is not a mempolicy constrained oom, so only
109 * check the mems of tsk's cpuset.
110 */
Oleg Nesterovad962442014-01-21 15:50:00 -0800111 ret = cpuset_mems_allowed_intersects(current, tsk);
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700112 }
Oleg Nesterovad962442014-01-21 15:50:00 -0800113 if (ret)
114 break;
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800115 }
Oleg Nesterovad962442014-01-21 15:50:00 -0800116 rcu_read_unlock();
KOSAKI Motohirodf1090a2010-08-09 17:19:39 -0700117
Oleg Nesterovad962442014-01-21 15:50:00 -0800118 return ret;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700119}
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700120#else
Shakeel Buttac311a12019-07-11 21:00:31 -0700121static bool oom_cpuset_eligible(struct task_struct *tsk, struct oom_control *oc)
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700122{
123 return true;
124}
125#endif /* CONFIG_NUMA */
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700126
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700127/*
128 * The process p may have detached its own ->mm while exiting or through
Christoph Hellwigf5678e72020-06-10 18:42:06 -0700129 * kthread_use_mm(), but one or more of its subthreads may still have a valid
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700130 * pointer. Return p, or any of its subthreads with a valid ->mm, with
131 * task_lock() held.
132 */
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -0700133struct task_struct *find_lock_task_mm(struct task_struct *p)
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700134{
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800135 struct task_struct *t;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700136
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800137 rcu_read_lock();
138
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800139 for_each_thread(p, t) {
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700140 task_lock(t);
141 if (likely(t->mm))
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800142 goto found;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700143 task_unlock(t);
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800144 }
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800145 t = NULL;
146found:
147 rcu_read_unlock();
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700148
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800149 return t;
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700150}
151
Yaowei Baidb2a0dd2015-11-06 16:28:06 -0800152/*
153 * order == -1 means the oom kill is required by sysrq, otherwise only
154 * for display purposes.
155 */
156static inline bool is_sysrq_oom(struct oom_control *oc)
157{
158 return oc->order == -1;
159}
160
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700161/* return true if the task is not adequate as candidate victim task. */
Shakeel Buttac311a12019-07-11 21:00:31 -0700162static bool oom_unkillable_task(struct task_struct *p)
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700163{
164 if (is_global_init(p))
165 return true;
166 if (p->flags & PF_KTHREAD)
167 return true;
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700168 return false;
169}
170
Randy Dunlap845be1c2021-04-16 15:45:54 -0700171/*
Hui Su259b3632020-12-14 19:12:49 -0800172 * Check whether unreclaimable slab amount is greater than
173 * all user memory(LRU pages).
174 * dump_unreclaimable_slab() could help in the case that
175 * oom due to too much unreclaimable slab used by kernel.
176*/
177static bool should_dump_unreclaim_slab(void)
Yang Shi852d8be2017-11-15 17:32:07 -0800178{
179 unsigned long nr_lru;
180
181 nr_lru = global_node_page_state(NR_ACTIVE_ANON) +
182 global_node_page_state(NR_INACTIVE_ANON) +
183 global_node_page_state(NR_ACTIVE_FILE) +
184 global_node_page_state(NR_INACTIVE_FILE) +
185 global_node_page_state(NR_ISOLATED_ANON) +
186 global_node_page_state(NR_ISOLATED_FILE) +
187 global_node_page_state(NR_UNEVICTABLE);
188
Roman Gushchind42f3242020-08-06 23:20:39 -0700189 return (global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B) > nr_lru);
Yang Shi852d8be2017-11-15 17:32:07 -0800190}
191
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192/**
David Rientjesa63d83f2010-08-09 17:19:46 -0700193 * oom_badness - heuristic function to determine which candidate task to kill
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 * @p: task struct of which task we should calculate
David Rientjesa63d83f2010-08-09 17:19:46 -0700195 * @totalpages: total present RAM allowed for page allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 *
David Rientjesa63d83f2010-08-09 17:19:46 -0700197 * The heuristic for determining which task to kill is made to be as simple and
198 * predictable as possible. The goal is to return the highest value for the
199 * task consuming the most memory to avoid subsequent oom failures.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 */
Yafang Shao9066e5c2020-08-11 18:31:22 -0700201long oom_badness(struct task_struct *p, unsigned long totalpages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
David Rientjes1e11ad82012-06-08 13:21:26 -0700203 long points;
David Rientjes61eafb02012-06-20 12:52:58 -0700204 long adj;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700205
Shakeel Buttac311a12019-07-11 21:00:31 -0700206 if (oom_unkillable_task(p))
Yafang Shao9066e5c2020-08-11 18:31:22 -0700207 return LONG_MIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700209 p = find_lock_task_mm(p);
210 if (!p)
Yafang Shao9066e5c2020-08-11 18:31:22 -0700211 return LONG_MIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
Michal Hockobb8a4b72016-05-20 16:57:18 -0700213 /*
214 * Do not even consider tasks which are explicitly marked oom
Michal Hockob18dc5f2016-07-28 15:44:46 -0700215 * unkillable or have been already oom reaped or the are in
216 * the middle of vfork
Michal Hockobb8a4b72016-05-20 16:57:18 -0700217 */
David Rientjesa9c58b902012-12-11 16:02:54 -0800218 adj = (long)p->signal->oom_score_adj;
Michal Hockobb8a4b72016-05-20 16:57:18 -0700219 if (adj == OOM_SCORE_ADJ_MIN ||
Michal Hocko862e3072016-10-07 16:58:57 -0700220 test_bit(MMF_OOM_SKIP, &p->mm->flags) ||
Michal Hockob18dc5f2016-07-28 15:44:46 -0700221 in_vfork(p)) {
Michal Hocko5aecc852011-11-15 14:36:07 -0800222 task_unlock(p);
Yafang Shao9066e5c2020-08-11 18:31:22 -0700223 return LONG_MIN;
Michal Hocko5aecc852011-11-15 14:36:07 -0800224 }
225
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700227 * The baseline for the badness score is the proportion of RAM that each
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700228 * task's rss, pagetable and swap space use.
David Rientjesa63d83f2010-08-09 17:19:46 -0700229 */
Kirill A. Shutemovdc6c9a32015-02-11 15:26:50 -0800230 points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) +
Kirill A. Shutemovaf5b0f62017-11-15 17:35:40 -0800231 mm_pgtables_bytes(p->mm) / PAGE_SIZE;
Andrew Morton97c2c9b82006-04-18 22:20:38 -0700232 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
David Rientjes61eafb02012-06-20 12:52:58 -0700234 /* Normalize to oom_score_adj units */
235 adj *= totalpages / 1000;
236 points += adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237
Yafang Shao9066e5c2020-08-11 18:31:22 -0700238 return points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239}
240
yuzhoujianef8444e2018-12-28 00:36:07 -0800241static const char * const oom_constraint_text[] = {
242 [CONSTRAINT_NONE] = "CONSTRAINT_NONE",
243 [CONSTRAINT_CPUSET] = "CONSTRAINT_CPUSET",
244 [CONSTRAINT_MEMORY_POLICY] = "CONSTRAINT_MEMORY_POLICY",
245 [CONSTRAINT_MEMCG] = "CONSTRAINT_MEMCG",
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700246};
247
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800249 * Determine the type of allocation constraint.
250 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700251static enum oom_constraint constrained_alloc(struct oom_control *oc)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800252{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700253 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700254 struct zoneref *z;
Joonsoo Kim97a225e2020-06-03 15:59:01 -0700255 enum zone_type highest_zoneidx = gfp_zone(oc->gfp_mask);
David Rientjesa63d83f2010-08-09 17:19:46 -0700256 bool cpuset_limited = false;
257 int nid;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800258
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700259 if (is_memcg_oom(oc)) {
Roman Gushchinbbec2e12018-06-07 17:06:18 -0700260 oc->totalpages = mem_cgroup_get_max(oc->memcg) ?: 1;
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700261 return CONSTRAINT_MEMCG;
262 }
263
David Rientjesa63d83f2010-08-09 17:19:46 -0700264 /* Default to all available memory */
Arun KSca79b0c2018-12-28 00:34:29 -0800265 oc->totalpages = totalram_pages() + total_swap_pages;
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700266
267 if (!IS_ENABLED(CONFIG_NUMA))
268 return CONSTRAINT_NONE;
David Rientjesa63d83f2010-08-09 17:19:46 -0700269
David Rientjes6e0fc462015-09-08 15:00:36 -0700270 if (!oc->zonelist)
David Rientjesa63d83f2010-08-09 17:19:46 -0700271 return CONSTRAINT_NONE;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800272 /*
273 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
274 * to kill current.We have to random task kill in this case.
275 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
276 */
David Rientjes6e0fc462015-09-08 15:00:36 -0700277 if (oc->gfp_mask & __GFP_THISNODE)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800278 return CONSTRAINT_NONE;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800279
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800280 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700281 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
282 * the page allocator means a mempolicy is in effect. Cpuset policy
283 * is enforced in get_page_from_freelist().
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800284 */
David Rientjes6e0fc462015-09-08 15:00:36 -0700285 if (oc->nodemask &&
286 !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) {
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700287 oc->totalpages = total_swap_pages;
David Rientjes6e0fc462015-09-08 15:00:36 -0700288 for_each_node_mask(nid, *oc->nodemask)
Michal Hocko1eb41bb02019-09-23 15:37:16 -0700289 oc->totalpages += node_present_pages(nid);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800290 return CONSTRAINT_MEMORY_POLICY;
David Rientjesa63d83f2010-08-09 17:19:46 -0700291 }
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800292
293 /* Check this allocation failure is caused by cpuset's wall function */
David Rientjes6e0fc462015-09-08 15:00:36 -0700294 for_each_zone_zonelist_nodemask(zone, z, oc->zonelist,
Joonsoo Kim97a225e2020-06-03 15:59:01 -0700295 highest_zoneidx, oc->nodemask)
David Rientjes6e0fc462015-09-08 15:00:36 -0700296 if (!cpuset_zone_allowed(zone, oc->gfp_mask))
David Rientjesa63d83f2010-08-09 17:19:46 -0700297 cpuset_limited = true;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800298
David Rientjesa63d83f2010-08-09 17:19:46 -0700299 if (cpuset_limited) {
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700300 oc->totalpages = total_swap_pages;
David Rientjesa63d83f2010-08-09 17:19:46 -0700301 for_each_node_mask(nid, cpuset_current_mems_allowed)
Michal Hocko1eb41bb02019-09-23 15:37:16 -0700302 oc->totalpages += node_present_pages(nid);
David Rientjesa63d83f2010-08-09 17:19:46 -0700303 return CONSTRAINT_CPUSET;
304 }
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800305 return CONSTRAINT_NONE;
306}
307
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700308static int oom_evaluate_task(struct task_struct *task, void *arg)
David Rientjes462607e2012-07-31 16:43:40 -0700309{
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700310 struct oom_control *oc = arg;
Yafang Shao9066e5c2020-08-11 18:31:22 -0700311 long points;
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700312
Shakeel Buttac311a12019-07-11 21:00:31 -0700313 if (oom_unkillable_task(task))
314 goto next;
315
316 /* p may not have freeable memory in nodemask */
317 if (!is_memcg_oom(oc) && !oom_cpuset_eligible(task, oc))
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700318 goto next;
David Rientjes462607e2012-07-31 16:43:40 -0700319
320 /*
321 * This task already has access to memory reserves and is being killed.
Michal Hockoa3739662016-07-28 15:45:01 -0700322 * Don't allow any other task to have access to the reserves unless
Michal Hocko862e3072016-10-07 16:58:57 -0700323 * the task has MMF_OOM_SKIP because chances that it would release
Michal Hockoa3739662016-07-28 15:45:01 -0700324 * any memory is quite low.
David Rientjes462607e2012-07-31 16:43:40 -0700325 */
Michal Hocko862e3072016-10-07 16:58:57 -0700326 if (!is_sysrq_oom(oc) && tsk_is_oom_victim(task)) {
327 if (test_bit(MMF_OOM_SKIP, &task->signal->oom_mm->flags))
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700328 goto next;
329 goto abort;
Michal Hockoa3739662016-07-28 15:45:01 -0700330 }
David Rientjes462607e2012-07-31 16:43:40 -0700331
David Rientjese1e12d22012-12-11 16:02:56 -0800332 /*
333 * If task is allocating a lot of memory and has been marked to be
334 * killed first if it triggers an oom, then select it.
335 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700336 if (oom_task_origin(task)) {
Yafang Shao9066e5c2020-08-11 18:31:22 -0700337 points = LONG_MAX;
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700338 goto select;
339 }
David Rientjese1e12d22012-12-11 16:02:56 -0800340
Shakeel Buttac311a12019-07-11 21:00:31 -0700341 points = oom_badness(task, oc->totalpages);
Yafang Shao9066e5c2020-08-11 18:31:22 -0700342 if (points == LONG_MIN || points < oc->chosen_points)
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700343 goto next;
344
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700345select:
346 if (oc->chosen)
347 put_task_struct(oc->chosen);
348 get_task_struct(task);
349 oc->chosen = task;
350 oc->chosen_points = points;
351next:
352 return 0;
353abort:
354 if (oc->chosen)
355 put_task_struct(oc->chosen);
356 oc->chosen = (void *)-1UL;
357 return 1;
David Rientjes462607e2012-07-31 16:43:40 -0700358}
359
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800360/*
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700361 * Simple selection loop. We choose the process with the highest number of
362 * 'points'. In case scan was aborted, oc->chosen is set to -1.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700364static void select_bad_process(struct oom_control *oc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365{
Yafang Shao9066e5c2020-08-11 18:31:22 -0700366 oc->chosen_points = LONG_MIN;
367
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700368 if (is_memcg_oom(oc))
369 mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc);
370 else {
371 struct task_struct *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700373 rcu_read_lock();
374 for_each_process(p)
375 if (oom_evaluate_task(p, oc))
376 break;
377 rcu_read_unlock();
Oleg Nesterov1da4db02014-01-21 15:49:58 -0800378 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700379}
380
Shakeel Butt5eee7e12019-07-11 21:00:23 -0700381static int dump_task(struct task_struct *p, void *arg)
382{
383 struct oom_control *oc = arg;
384 struct task_struct *task;
385
Shakeel Buttac311a12019-07-11 21:00:31 -0700386 if (oom_unkillable_task(p))
387 return 0;
388
389 /* p may not have freeable memory in nodemask */
390 if (!is_memcg_oom(oc) && !oom_cpuset_eligible(p, oc))
Shakeel Butt5eee7e12019-07-11 21:00:23 -0700391 return 0;
392
393 task = find_lock_task_mm(p);
394 if (!task) {
395 /*
Tang Yizhouf8159c12021-02-24 12:09:50 -0800396 * All of p's threads have already detached their mm's. There's
397 * no need to report them; they can't be oom killed anyway.
Shakeel Butt5eee7e12019-07-11 21:00:23 -0700398 */
399 return 0;
400 }
401
402 pr_info("[%7d] %5d %5d %8lu %8lu %8ld %8lu %5hd %s\n",
403 task->pid, from_kuid(&init_user_ns, task_uid(task)),
404 task->tgid, task->mm->total_vm, get_mm_rss(task->mm),
405 mm_pgtables_bytes(task->mm),
406 get_mm_counter(task->mm, MM_SWAPENTS),
407 task->signal->oom_score_adj, task->comm);
408 task_unlock(task);
409
410 return 0;
411}
412
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700414 * dump_tasks - dump current memory state of all system tasks
Shakeel Butt5eee7e12019-07-11 21:00:23 -0700415 * @oc: pointer to struct oom_control
Randy Dunlap1b578df2008-03-19 17:00:42 -0700416 *
David Rientjese85bfd32010-09-22 13:05:10 -0700417 * Dumps the current memory state of all eligible tasks. Tasks not in the same
418 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
419 * are not shown.
Kirill A. Shutemovaf5b0f62017-11-15 17:35:40 -0800420 * State information includes task's pid, uid, tgid, vm size, rss,
421 * pgtables_bytes, swapents, oom_score_adj value, and name.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800422 */
Shakeel Butt5eee7e12019-07-11 21:00:23 -0700423static void dump_tasks(struct oom_control *oc)
David Rientjesfef1bdd2008-02-07 00:14:07 -0800424{
Rodrigo Freirec3b78b12018-08-21 21:52:41 -0700425 pr_info("Tasks state (memory values in pages):\n");
426 pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name\n");
David Rientjesfef1bdd2008-02-07 00:14:07 -0800427
Shakeel Butt5eee7e12019-07-11 21:00:23 -0700428 if (is_memcg_oom(oc))
429 mem_cgroup_scan_tasks(oc->memcg, dump_task, oc);
430 else {
431 struct task_struct *p;
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700432
Shakeel Butt5eee7e12019-07-11 21:00:23 -0700433 rcu_read_lock();
434 for_each_process(p)
435 dump_task(p, oc);
436 rcu_read_unlock();
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700437 }
David Rientjesfef1bdd2008-02-07 00:14:07 -0800438}
439
yuzhoujianef8444e2018-12-28 00:36:07 -0800440static void dump_oom_summary(struct oom_control *oc, struct task_struct *victim)
441{
442 /* one line summary of the oom killer context. */
443 pr_info("oom-kill:constraint=%s,nodemask=%*pbl",
444 oom_constraint_text[oc->constraint],
445 nodemask_pr_args(oc->nodemask));
446 cpuset_print_current_mems_allowed();
yuzhoujianf0c867d2018-12-28 00:36:10 -0800447 mem_cgroup_print_oom_context(oc->memcg, victim);
yuzhoujianef8444e2018-12-28 00:36:07 -0800448 pr_cont(",task=%s,pid=%d,uid=%d\n", victim->comm, victim->pid,
449 from_kuid(&init_user_ns, task_uid(victim)));
450}
451
Vladimir Davydov2a966b72016-07-26 15:22:33 -0700452static void dump_header(struct oom_control *oc, struct task_struct *p)
David Rientjes1b604d72009-12-14 17:57:47 -0800453{
yuzhoujianef8444e2018-12-28 00:36:07 -0800454 pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), order=%d, oom_score_adj=%hd\n",
455 current->comm, oc->gfp_mask, &oc->gfp_mask, oc->order,
Michal Hocko0205f752017-11-15 17:39:14 -0800456 current->signal->oom_score_adj);
Michal Hocko92549902016-10-07 16:59:33 -0700457 if (!IS_ENABLED(CONFIG_COMPACTION) && oc->order)
458 pr_warn("COMPACTION is disabled!!!\n");
Vlastimil Babkaa0795cd2016-03-15 14:56:05 -0700459
David Rientjes1b604d72009-12-14 17:57:47 -0800460 dump_stack();
Yang Shi852d8be2017-11-15 17:32:07 -0800461 if (is_memcg_oom(oc))
yuzhoujianf0c867d2018-12-28 00:36:10 -0800462 mem_cgroup_print_oom_meminfo(oc->memcg);
Yang Shi852d8be2017-11-15 17:32:07 -0800463 else {
Michal Hocko974f4362022-08-23 11:22:30 +0200464 __show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask, gfp_zone(oc->gfp_mask));
Hui Su259b3632020-12-14 19:12:49 -0800465 if (should_dump_unreclaim_slab())
Yang Shi852d8be2017-11-15 17:32:07 -0800466 dump_unreclaimable_slab();
467 }
David Rientjes1b604d72009-12-14 17:57:47 -0800468 if (sysctl_oom_dump_tasks)
Shakeel Butt5eee7e12019-07-11 21:00:23 -0700469 dump_tasks(oc);
yuzhoujianef8444e2018-12-28 00:36:07 -0800470 if (p)
471 dump_oom_summary(oc, p);
David Rientjes1b604d72009-12-14 17:57:47 -0800472}
473
Michal Hocko5695be12014-10-20 18:12:32 +0200474/*
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800475 * Number of OOM victims in flight
Michal Hocko5695be12014-10-20 18:12:32 +0200476 */
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800477static atomic_t oom_victims = ATOMIC_INIT(0);
478static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait);
Michal Hocko5695be12014-10-20 18:12:32 +0200479
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700480static bool oom_killer_disabled __read_mostly;
Michal Hocko5695be12014-10-20 18:12:32 +0200481
Michal Hockobc448e82016-03-25 14:20:30 -0700482#define K(x) ((x) << (PAGE_SHIFT-10))
483
Michal Hocko3ef22df2016-05-19 17:13:12 -0700484/*
485 * task->mm can be NULL if the task is the exited group leader. So to
486 * determine whether the task is using a particular mm, we examine all the
487 * task's threads: if one of those is using this mm then this task was also
488 * using it.
489 */
Michal Hocko44a70ade2016-07-28 15:44:43 -0700490bool process_shares_mm(struct task_struct *p, struct mm_struct *mm)
Michal Hocko3ef22df2016-05-19 17:13:12 -0700491{
492 struct task_struct *t;
493
494 for_each_thread(p, t) {
495 struct mm_struct *t_mm = READ_ONCE(t->mm);
496 if (t_mm)
497 return t_mm == mm;
498 }
499 return false;
500}
501
Michal Hockoaac45362016-03-25 14:20:24 -0700502#ifdef CONFIG_MMU
503/*
504 * OOM Reaper kernel thread which tries to reap the memory used by the OOM
505 * victim (if that is possible) to help the OOM killer to move on.
506 */
507static struct task_struct *oom_reaper_th;
Michal Hockoaac45362016-03-25 14:20:24 -0700508static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait);
Vladimir Davydov29c696e2016-03-25 14:20:39 -0700509static struct task_struct *oom_reaper_list;
Michal Hocko03049262016-03-25 14:20:33 -0700510static DEFINE_SPINLOCK(oom_reaper_lock);
511
Suren Baghdasaryanbf3980c2022-05-31 15:30:59 -0700512static bool __oom_reap_task_mm(struct mm_struct *mm)
Michal Hockoaac45362016-03-25 14:20:24 -0700513{
Michal Hockoaac45362016-03-25 14:20:24 -0700514 struct vm_area_struct *vma;
Michal Hocko93065ac2018-08-21 21:52:33 -0700515 bool ret = true;
Liam R. Howlette1c2c772022-09-06 19:49:03 +0000516 VMA_ITERATOR(vmi, mm, 0);
David Rientjes27ae3572018-05-11 16:02:04 -0700517
518 /*
519 * Tell all users of get_user/copy_from_user etc... that the content
520 * is no longer stable. No barriers really needed because unmapping
521 * should imply barriers already and the reader would hit a page fault
522 * if it stumbled over a reaped memory.
523 */
524 set_bit(MMF_UNSTABLE, &mm->flags);
525
Liam R. Howlette1c2c772022-09-06 19:49:03 +0000526 for_each_vma(vmi, vma) {
Hugh Dickinsa213e5c2022-02-14 18:23:29 -0800527 if (vma->vm_flags & (VM_HUGETLB|VM_PFNMAP))
David Rientjes27ae3572018-05-11 16:02:04 -0700528 continue;
529
530 /*
531 * Only anonymous pages have a good chance to be dropped
532 * without additional steps which we cannot afford as we
533 * are OOM already.
534 *
535 * We do not even care about fs backed pages because all
536 * which are reclaimable have already been reclaimed and
537 * we do not want to block exit_mmap by keeping mm ref
538 * count elevated without a good reason.
539 */
540 if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) {
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800541 struct mmu_notifier_range range;
David Rientjes27ae3572018-05-11 16:02:04 -0700542 struct mmu_gather tlb;
543
Jérôme Glisse6f4f13e2019-05-13 17:20:49 -0700544 mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0,
Alistair Popple7d4a8be2023-01-10 13:57:22 +1100545 mm, vma->vm_start,
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800546 vma->vm_end);
Will Deacona72afd82021-01-27 23:53:45 +0000547 tlb_gather_mmu(&tlb, mm);
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800548 if (mmu_notifier_invalidate_range_start_nonblock(&range)) {
Will Deaconae8eba82021-01-27 23:53:43 +0000549 tlb_finish_mmu(&tlb);
Michal Hocko93065ac2018-08-21 21:52:33 -0700550 ret = false;
551 continue;
552 }
Jérôme Glisseac46d4f2018-12-28 00:38:09 -0800553 unmap_page_range(&tlb, vma, range.start, range.end, NULL);
554 mmu_notifier_invalidate_range_end(&range);
Will Deaconae8eba82021-01-27 23:53:43 +0000555 tlb_finish_mmu(&tlb);
David Rientjes27ae3572018-05-11 16:02:04 -0700556 }
557 }
Michal Hocko93065ac2018-08-21 21:52:33 -0700558
559 return ret;
David Rientjes27ae3572018-05-11 16:02:04 -0700560}
561
Michal Hocko431f42fd2018-08-21 21:52:45 -0700562/*
563 * Reaps the address space of the give task.
564 *
565 * Returns true on success and false if none or part of the address space
566 * has been reclaimed and the caller should retry later.
567 */
David Rientjes27ae3572018-05-11 16:02:04 -0700568static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm)
569{
Michal Hockoaac45362016-03-25 14:20:24 -0700570 bool ret = true;
571
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700572 if (!mmap_read_trylock(mm)) {
Roman Gushchin422580c2017-07-10 15:49:05 -0700573 trace_skip_task_reaping(tsk->pid);
Michal Hockoaf5679f2018-08-21 21:52:37 -0700574 return false;
Michal Hocko4d4bbd82017-10-03 16:14:50 -0700575 }
576
577 /*
Andrea Arcangeli21292582017-09-06 16:25:00 -0700578 * MMF_OOM_SKIP is set by exit_mmap when the OOM reaper can't
579 * work on the mm anymore. The check for MMF_OOM_SKIP must run
Michel Lespinasse3e4e28c2020-06-08 21:33:51 -0700580 * under mmap_lock for reading because it serializes against the
581 * mmap_write_lock();mmap_write_unlock() cycle in exit_mmap().
Michal Hockoe5e3f4c2016-07-26 15:24:50 -0700582 */
Andrea Arcangeli21292582017-09-06 16:25:00 -0700583 if (test_bit(MMF_OOM_SKIP, &mm->flags)) {
Roman Gushchin422580c2017-07-10 15:49:05 -0700584 trace_skip_task_reaping(tsk->pid);
Michal Hocko431f42fd2018-08-21 21:52:45 -0700585 goto out_unlock;
Michal Hockoaac45362016-03-25 14:20:24 -0700586 }
587
Roman Gushchin422580c2017-07-10 15:49:05 -0700588 trace_start_task_reaping(tsk->pid);
589
Michal Hocko93065ac2018-08-21 21:52:33 -0700590 /* failed to reap part of the address space. Try again later */
Michal Hocko431f42fd2018-08-21 21:52:45 -0700591 ret = __oom_reap_task_mm(mm);
592 if (!ret)
593 goto out_finish;
Michal Hocko3f70dc32016-10-07 16:59:06 -0700594
Michal Hockobc448e82016-03-25 14:20:30 -0700595 pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n",
596 task_pid_nr(tsk), tsk->comm,
597 K(get_mm_counter(mm, MM_ANONPAGES)),
598 K(get_mm_counter(mm, MM_FILEPAGES)),
599 K(get_mm_counter(mm, MM_SHMEMPAGES)));
Michal Hocko431f42fd2018-08-21 21:52:45 -0700600out_finish:
601 trace_finish_task_reaping(tsk->pid);
602out_unlock:
Michel Lespinassed8ed45c2020-06-08 21:33:25 -0700603 mmap_read_unlock(mm);
Michal Hocko36324a92016-03-25 14:20:27 -0700604
Michal Hockoaac45362016-03-25 14:20:24 -0700605 return ret;
606}
607
Michal Hockobc448e82016-03-25 14:20:30 -0700608#define MAX_OOM_REAP_RETRIES 10
Michal Hocko36324a92016-03-25 14:20:27 -0700609static void oom_reap_task(struct task_struct *tsk)
Michal Hockoaac45362016-03-25 14:20:24 -0700610{
611 int attempts = 0;
Michal Hocko26db62f2016-10-07 16:58:51 -0700612 struct mm_struct *mm = tsk->signal->oom_mm;
Michal Hockoaac45362016-03-25 14:20:24 -0700613
Michel Lespinasse3e4e28c2020-06-08 21:33:51 -0700614 /* Retry the mmap_read_trylock(mm) a few times */
David Rientjes27ae3572018-05-11 16:02:04 -0700615 while (attempts++ < MAX_OOM_REAP_RETRIES && !oom_reap_task_mm(tsk, mm))
Michal Hockoaac45362016-03-25 14:20:24 -0700616 schedule_timeout_idle(HZ/10);
617
Tetsuo Handa97b12552018-04-05 16:25:45 -0700618 if (attempts <= MAX_OOM_REAP_RETRIES ||
619 test_bit(MMF_OOM_SKIP, &mm->flags))
Tetsuo Handa7ebffa42016-10-07 16:58:45 -0700620 goto done;
Michal Hocko11a410d2016-07-28 15:44:58 -0700621
Tetsuo Handa7ebffa42016-10-07 16:58:45 -0700622 pr_info("oom_reaper: unable to reap pid:%d (%s)\n",
623 task_pid_nr(tsk), tsk->comm);
David Rientjes8a7ff022020-01-30 22:14:26 -0800624 sched_show_task(tsk);
Tetsuo Handa7ebffa42016-10-07 16:58:45 -0700625 debug_show_all_locks();
Michal Hockobc448e82016-03-25 14:20:30 -0700626
Tetsuo Handa7ebffa42016-10-07 16:58:45 -0700627done:
Michal Hocko449d7772016-05-19 17:13:15 -0700628 tsk->oom_reaper_list = NULL;
Michal Hocko449d7772016-05-19 17:13:15 -0700629
Michal Hocko26db62f2016-10-07 16:58:51 -0700630 /*
631 * Hide this mm from OOM killer because it has been either reaped or
Michel Lespinasse3e4e28c2020-06-08 21:33:51 -0700632 * somebody can't call mmap_write_unlock(mm).
Michal Hocko26db62f2016-10-07 16:58:51 -0700633 */
Michal Hocko862e3072016-10-07 16:58:57 -0700634 set_bit(MMF_OOM_SKIP, &mm->flags);
Michal Hocko26db62f2016-10-07 16:58:51 -0700635
Nico Pachee4a38402022-04-21 16:36:01 -0700636 /* Drop a reference taken by queue_oom_reaper */
Michal Hocko36324a92016-03-25 14:20:27 -0700637 put_task_struct(tsk);
Michal Hockoaac45362016-03-25 14:20:24 -0700638}
639
640static int oom_reaper(void *unused)
641{
Sultan Alsawaf37239292021-11-05 13:43:25 -0700642 set_freezable();
643
Michal Hockoaac45362016-03-25 14:20:24 -0700644 while (true) {
Michal Hocko03049262016-03-25 14:20:33 -0700645 struct task_struct *tsk = NULL;
Michal Hockoaac45362016-03-25 14:20:24 -0700646
Vladimir Davydov29c696e2016-03-25 14:20:39 -0700647 wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL);
Nico Pachee4a38402022-04-21 16:36:01 -0700648 spin_lock_irq(&oom_reaper_lock);
Vladimir Davydov29c696e2016-03-25 14:20:39 -0700649 if (oom_reaper_list != NULL) {
650 tsk = oom_reaper_list;
651 oom_reaper_list = tsk->oom_reaper_list;
Michal Hocko03049262016-03-25 14:20:33 -0700652 }
Nico Pachee4a38402022-04-21 16:36:01 -0700653 spin_unlock_irq(&oom_reaper_lock);
Michal Hocko03049262016-03-25 14:20:33 -0700654
655 if (tsk)
656 oom_reap_task(tsk);
Michal Hockoaac45362016-03-25 14:20:24 -0700657 }
658
659 return 0;
660}
661
Nico Pachee4a38402022-04-21 16:36:01 -0700662static void wake_oom_reaper(struct timer_list *timer)
663{
664 struct task_struct *tsk = container_of(timer, struct task_struct,
665 oom_reaper_timer);
666 struct mm_struct *mm = tsk->signal->oom_mm;
667 unsigned long flags;
668
669 /* The victim managed to terminate on its own - see exit_mmap */
670 if (test_bit(MMF_OOM_SKIP, &mm->flags)) {
671 put_task_struct(tsk);
672 return;
673 }
674
675 spin_lock_irqsave(&oom_reaper_lock, flags);
676 tsk->oom_reaper_list = oom_reaper_list;
677 oom_reaper_list = tsk;
678 spin_unlock_irqrestore(&oom_reaper_lock, flags);
679 trace_wake_reaper(tsk->pid);
680 wake_up(&oom_reaper_wait);
681}
682
683/*
684 * Give the OOM victim time to exit naturally before invoking the oom_reaping.
685 * The timers timeout is arbitrary... the longer it is, the longer the worst
686 * case scenario for the OOM can take. If it is too small, the oom_reaper can
687 * get in the way and release resources needed by the process exit path.
688 * e.g. The futex robust list can sit in Anon|Private memory that gets reaped
689 * before the exit path is able to wake the futex waiters.
690 */
691#define OOM_REAPER_DELAY (2*HZ)
692static void queue_oom_reaper(struct task_struct *tsk)
Michal Hockoaac45362016-03-25 14:20:24 -0700693{
Tetsuo Handa9bcdeb52019-02-01 14:20:31 -0800694 /* mm is already queued? */
695 if (test_and_set_bit(MMF_OOM_REAP_QUEUED, &tsk->signal->oom_mm->flags))
Michal Hockoaac45362016-03-25 14:20:24 -0700696 return;
697
Michal Hocko36324a92016-03-25 14:20:27 -0700698 get_task_struct(tsk);
Nico Pachee4a38402022-04-21 16:36:01 -0700699 timer_setup(&tsk->oom_reaper_timer, wake_oom_reaper, 0);
700 tsk->oom_reaper_timer.expires = jiffies + OOM_REAPER_DELAY;
701 add_timer(&tsk->oom_reaper_timer);
Michal Hockoaac45362016-03-25 14:20:24 -0700702}
703
Andrew Mortona19cad02022-06-01 15:57:16 -0700704#ifdef CONFIG_SYSCTL
705static struct ctl_table vm_oom_kill_table[] = {
706 {
707 .procname = "panic_on_oom",
708 .data = &sysctl_panic_on_oom,
709 .maxlen = sizeof(sysctl_panic_on_oom),
710 .mode = 0644,
711 .proc_handler = proc_dointvec_minmax,
712 .extra1 = SYSCTL_ZERO,
713 .extra2 = SYSCTL_TWO,
714 },
715 {
716 .procname = "oom_kill_allocating_task",
717 .data = &sysctl_oom_kill_allocating_task,
718 .maxlen = sizeof(sysctl_oom_kill_allocating_task),
719 .mode = 0644,
720 .proc_handler = proc_dointvec,
721 },
722 {
723 .procname = "oom_dump_tasks",
724 .data = &sysctl_oom_dump_tasks,
725 .maxlen = sizeof(sysctl_oom_dump_tasks),
726 .mode = 0644,
727 .proc_handler = proc_dointvec,
728 },
729 {}
730};
731#endif
732
Michal Hockoaac45362016-03-25 14:20:24 -0700733static int __init oom_init(void)
734{
735 oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper");
sujiaxun43fe2192022-02-17 18:51:48 -0800736#ifdef CONFIG_SYSCTL
737 register_sysctl_init("vm", vm_oom_kill_table);
738#endif
Michal Hockoaac45362016-03-25 14:20:24 -0700739 return 0;
740}
741subsys_initcall(oom_init)
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700742#else
Nico Pachee4a38402022-04-21 16:36:01 -0700743static inline void queue_oom_reaper(struct task_struct *tsk)
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700744{
745}
746#endif /* CONFIG_MMU */
Michal Hockoaac45362016-03-25 14:20:24 -0700747
Michal Hocko49550b62015-02-11 15:26:12 -0800748/**
Johannes Weiner16e95192015-06-24 16:57:07 -0700749 * mark_oom_victim - mark the given task as OOM victim
Michal Hocko49550b62015-02-11 15:26:12 -0800750 * @tsk: task to mark
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800751 *
Johannes Weinerdc564012015-06-24 16:57:19 -0700752 * Has to be called with oom_lock held and never after
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800753 * oom has been disabled already.
Michal Hocko26db62f2016-10-07 16:58:51 -0700754 *
755 * tsk->mm has to be non NULL and caller has to guarantee it is stable (either
756 * under task_lock or operate on the current).
Michal Hocko49550b62015-02-11 15:26:12 -0800757 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700758static void mark_oom_victim(struct task_struct *tsk)
Michal Hocko49550b62015-02-11 15:26:12 -0800759{
Michal Hocko26db62f2016-10-07 16:58:51 -0700760 struct mm_struct *mm = tsk->mm;
761
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800762 WARN_ON(oom_killer_disabled);
763 /* OOM killer might race with memcg OOM */
764 if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE))
765 return;
Michal Hocko26db62f2016-10-07 16:58:51 -0700766
Michal Hocko26db62f2016-10-07 16:58:51 -0700767 /* oom_mm is bound to the signal struct life time. */
Suren Baghdasaryanb3541d92022-05-31 15:31:00 -0700768 if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm))
Vegard Nossumf1f10072017-02-27 14:30:07 -0800769 mmgrab(tsk->signal->oom_mm);
Michal Hocko26db62f2016-10-07 16:58:51 -0700770
Michal Hocko63a8ca92015-02-11 15:26:15 -0800771 /*
772 * Make sure that the task is woken up from uninterruptible sleep
773 * if it is frozen because OOM killer wouldn't be able to free
774 * any memory and livelock. freezing_slow_path will tell the freezer
775 * that TIF_MEMDIE tasks should be ignored.
776 */
777 __thaw_task(tsk);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800778 atomic_inc(&oom_victims);
Roman Gushchin422580c2017-07-10 15:49:05 -0700779 trace_mark_victim(tsk->pid);
Michal Hocko49550b62015-02-11 15:26:12 -0800780}
781
782/**
Johannes Weiner16e95192015-06-24 16:57:07 -0700783 * exit_oom_victim - note the exit of an OOM victim
Michal Hocko49550b62015-02-11 15:26:12 -0800784 */
Tetsuo Handa38531202016-10-07 16:59:03 -0700785void exit_oom_victim(void)
Michal Hocko49550b62015-02-11 15:26:12 -0800786{
Tetsuo Handa38531202016-10-07 16:59:03 -0700787 clear_thread_flag(TIF_MEMDIE);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800788
Johannes Weinerc38f1022015-06-24 16:57:13 -0700789 if (!atomic_dec_return(&oom_victims))
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800790 wake_up_all(&oom_victims_wait);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800791}
792
793/**
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700794 * oom_killer_enable - enable OOM killer
795 */
796void oom_killer_enable(void)
797{
798 oom_killer_disabled = false;
Michal Hockod75da002017-05-03 14:54:57 -0700799 pr_info("OOM killer enabled.\n");
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700800}
801
802/**
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800803 * oom_killer_disable - disable OOM killer
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700804 * @timeout: maximum timeout to wait for oom victims in jiffies
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800805 *
806 * Forces all page allocations to fail rather than trigger OOM killer.
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700807 * Will block and wait until all OOM victims are killed or the given
808 * timeout expires.
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800809 *
810 * The function cannot be called when there are runnable user tasks because
811 * the userspace would see unexpected allocation failures as a result. Any
812 * new usage of this function should be consulted with MM people.
813 *
814 * Returns true if successful and false if the OOM killer cannot be
815 * disabled.
816 */
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700817bool oom_killer_disable(signed long timeout)
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800818{
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700819 signed long ret;
820
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800821 /*
Tetsuo Handa6afcf282016-03-17 14:20:45 -0700822 * Make sure to not race with an ongoing OOM killer. Check that the
823 * current is not killed (possibly due to sharing the victim's memory).
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800824 */
Tetsuo Handa6afcf282016-03-17 14:20:45 -0700825 if (mutex_lock_killable(&oom_lock))
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800826 return false;
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800827 oom_killer_disabled = true;
Johannes Weinerdc564012015-06-24 16:57:19 -0700828 mutex_unlock(&oom_lock);
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800829
Michal Hocko7d2e7a22016-10-07 16:59:00 -0700830 ret = wait_event_interruptible_timeout(oom_victims_wait,
831 !atomic_read(&oom_victims), timeout);
832 if (ret <= 0) {
833 oom_killer_enable();
834 return false;
835 }
Michal Hockod75da002017-05-03 14:54:57 -0700836 pr_info("OOM killer disabled.\n");
Michal Hockoc32b3cb2015-02-11 15:26:24 -0800837
838 return true;
839}
840
Michal Hocko1af8bb42016-07-28 15:44:52 -0700841static inline bool __task_will_free_mem(struct task_struct *task)
842{
843 struct signal_struct *sig = task->signal;
844
845 /*
Eric W. Biedermand67e03e2021-09-01 11:23:38 -0500846 * A coredumping process may sleep for an extended period in
Eric W. Biederman92307382021-09-01 11:33:50 -0500847 * coredump_task_exit(), so the oom killer cannot assume that
Eric W. Biedermand67e03e2021-09-01 11:23:38 -0500848 * the process will promptly exit and release memory.
Michal Hocko1af8bb42016-07-28 15:44:52 -0700849 */
Eric W. Biederman98b24b12021-11-19 11:29:48 -0600850 if (sig->core_state)
Michal Hocko1af8bb42016-07-28 15:44:52 -0700851 return false;
852
853 if (sig->flags & SIGNAL_GROUP_EXIT)
854 return true;
855
856 if (thread_group_empty(task) && (task->flags & PF_EXITING))
857 return true;
858
859 return false;
860}
861
862/*
863 * Checks whether the given task is dying or exiting and likely to
864 * release its address space. This means that all threads and processes
865 * sharing the same mm have to be killed or exiting.
Michal Hocko091f3622016-07-28 15:45:04 -0700866 * Caller has to make sure that task->mm is stable (hold task_lock or
867 * it operates on the current).
Michal Hocko1af8bb42016-07-28 15:44:52 -0700868 */
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -0700869static bool task_will_free_mem(struct task_struct *task)
Michal Hocko1af8bb42016-07-28 15:44:52 -0700870{
Michal Hocko091f3622016-07-28 15:45:04 -0700871 struct mm_struct *mm = task->mm;
Michal Hocko1af8bb42016-07-28 15:44:52 -0700872 struct task_struct *p;
Geert Uytterhoevenf33e6f02016-08-11 15:33:09 -0700873 bool ret = true;
Michal Hocko1af8bb42016-07-28 15:44:52 -0700874
Michal Hocko091f3622016-07-28 15:45:04 -0700875 /*
876 * Skip tasks without mm because it might have passed its exit_mm and
877 * exit_oom_victim. oom_reaper could have rescued that but do not rely
878 * on that for now. We can consider find_lock_task_mm in future.
879 */
880 if (!mm)
881 return false;
882
Michal Hocko1af8bb42016-07-28 15:44:52 -0700883 if (!__task_will_free_mem(task))
884 return false;
885
886 /*
Michal Hocko696453e2016-07-28 15:44:55 -0700887 * This task has already been drained by the oom reaper so there are
888 * only small chances it will free some more
889 */
Michal Hocko862e3072016-10-07 16:58:57 -0700890 if (test_bit(MMF_OOM_SKIP, &mm->flags))
Michal Hocko696453e2016-07-28 15:44:55 -0700891 return false;
Michal Hocko696453e2016-07-28 15:44:55 -0700892
Michal Hocko091f3622016-07-28 15:45:04 -0700893 if (atomic_read(&mm->mm_users) <= 1)
Michal Hocko1af8bb42016-07-28 15:44:52 -0700894 return true;
Michal Hocko1af8bb42016-07-28 15:44:52 -0700895
896 /*
Michal Hocko5870c2e2016-10-07 16:57:32 -0700897 * Make sure that all tasks which share the mm with the given tasks
898 * are dying as well to make sure that a) nobody pins its mm and
899 * b) the task is also reapable by the oom reaper.
Michal Hocko1af8bb42016-07-28 15:44:52 -0700900 */
901 rcu_read_lock();
902 for_each_process(p) {
903 if (!process_shares_mm(p, mm))
904 continue;
905 if (same_thread_group(task, p))
906 continue;
907 ret = __task_will_free_mem(p);
908 if (!ret)
909 break;
910 }
911 rcu_read_unlock();
Michal Hocko1af8bb42016-07-28 15:44:52 -0700912
913 return ret;
914}
915
Shakeel Buttbbbe4802019-03-05 15:46:12 -0800916static void __oom_kill_process(struct task_struct *victim, const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700917{
Roman Gushchin5989ad72018-08-21 21:53:50 -0700918 struct task_struct *p;
David Rientjes647f2bd2012-03-21 16:33:46 -0700919 struct mm_struct *mm;
Tetsuo Handabb299022016-03-25 14:20:44 -0700920 bool can_oom_reap = true;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921
David Rientjes6b0c81b2012-07-31 16:43:45 -0700922 p = find_lock_task_mm(victim);
923 if (!p) {
Yafang Shao619b5b42020-08-11 18:31:32 -0700924 pr_info("%s: OOM victim %d (%s) is already exiting. Skip killing the task\n",
925 message, task_pid_nr(victim), victim->comm);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700926 put_task_struct(victim);
David Rientjes647f2bd2012-03-21 16:33:46 -0700927 return;
David Rientjes6b0c81b2012-07-31 16:43:45 -0700928 } else if (victim != p) {
929 get_task_struct(p);
930 put_task_struct(victim);
931 victim = p;
932 }
David Rientjes647f2bd2012-03-21 16:33:46 -0700933
Tetsuo Handa880b76892015-11-05 18:47:51 -0800934 /* Get a reference to safely compare mm after task_unlock(victim) */
David Rientjes647f2bd2012-03-21 16:33:46 -0700935 mm = victim->mm;
Vegard Nossumf1f10072017-02-27 14:30:07 -0800936 mmgrab(mm);
Konstantin Khlebnikov8e675f72017-07-06 15:40:28 -0700937
938 /* Raise event before sending signal: task reaper must see this */
939 count_vm_event(OOM_KILL);
Roman Gushchinfe6bdfc2018-06-14 15:28:05 -0700940 memcg_memory_event_mm(mm, MEMCG_OOM_KILL);
Konstantin Khlebnikov8e675f72017-07-06 15:40:28 -0700941
Tetsuo Handa426fb5e2015-11-05 18:47:44 -0800942 /*
Michal Hockocd04ae12017-09-06 16:24:50 -0700943 * We should send SIGKILL before granting access to memory reserves
944 * in order to prevent the OOM victim from depleting the memory
945 * reserves from the user space under its control.
Tetsuo Handa426fb5e2015-11-05 18:47:44 -0800946 */
Eric W. Biederman079b22d2018-09-03 10:32:52 +0200947 do_send_sig_info(SIGKILL, SEND_SIG_PRIV, victim, PIDTYPE_TGID);
Johannes Weiner16e95192015-06-24 16:57:07 -0700948 mark_oom_victim(victim);
Edward Chron70cb6d22019-09-23 15:37:11 -0700949 pr_err("%s: Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB, UID:%u pgtables:%lukB oom_score_adj:%hd\n",
950 message, task_pid_nr(victim), victim->comm, K(mm->total_vm),
951 K(get_mm_counter(mm, MM_ANONPAGES)),
952 K(get_mm_counter(mm, MM_FILEPAGES)),
953 K(get_mm_counter(mm, MM_SHMEMPAGES)),
954 from_kuid(&init_user_ns, task_uid(victim)),
Ilya Dryomov941f7622020-01-04 13:00:09 -0800955 mm_pgtables_bytes(mm) >> 10, victim->signal->oom_score_adj);
David Rientjes647f2bd2012-03-21 16:33:46 -0700956 task_unlock(victim);
957
958 /*
959 * Kill all user processes sharing victim->mm in other thread groups, if
960 * any. They don't get access to memory reserves, though, to avoid
Michel Lespinassec1e8d7c2020-06-08 21:33:54 -0700961 * depletion of all memory. This prevents mm->mmap_lock livelock when an
David Rientjes647f2bd2012-03-21 16:33:46 -0700962 * oom killed thread cannot exit because it requires the semaphore and
963 * its contended by another thread trying to allocate memory itself.
964 * That thread will now get access to memory reserves since it has a
965 * pending fatal signal.
966 */
Oleg Nesterov4d4048b2014-01-21 15:50:01 -0800967 rcu_read_lock();
Oleg Nesterovc3190252015-11-05 18:48:23 -0800968 for_each_process(p) {
Oleg Nesterov4d7b3392015-11-05 18:48:26 -0800969 if (!process_shares_mm(p, mm))
Oleg Nesterovc3190252015-11-05 18:48:23 -0800970 continue;
971 if (same_thread_group(p, victim))
972 continue;
Michal Hocko1b51e652016-10-07 16:59:09 -0700973 if (is_global_init(p)) {
Michal Hockoaac45362016-03-25 14:20:24 -0700974 can_oom_reap = false;
Michal Hocko862e3072016-10-07 16:58:57 -0700975 set_bit(MMF_OOM_SKIP, &mm->flags);
Michal Hockoa3739662016-07-28 15:45:01 -0700976 pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n",
977 task_pid_nr(victim), victim->comm,
978 task_pid_nr(p), p->comm);
Oleg Nesterovc3190252015-11-05 18:48:23 -0800979 continue;
Michal Hockoaac45362016-03-25 14:20:24 -0700980 }
Michal Hocko1b51e652016-10-07 16:59:09 -0700981 /*
Rolf Eike Beer4c9c3802021-03-17 10:24:51 +0100982 * No kthread_use_mm() user needs to read from the userspace so
Christoph Hellwigf5678e72020-06-10 18:42:06 -0700983 * we are ok to reap it.
Michal Hocko1b51e652016-10-07 16:59:09 -0700984 */
985 if (unlikely(p->flags & PF_KTHREAD))
986 continue;
Eric W. Biederman079b22d2018-09-03 10:32:52 +0200987 do_send_sig_info(SIGKILL, SEND_SIG_PRIV, p, PIDTYPE_TGID);
Oleg Nesterovc3190252015-11-05 18:48:23 -0800988 }
David Rientjes6b0c81b2012-07-31 16:43:45 -0700989 rcu_read_unlock();
David Rientjes647f2bd2012-03-21 16:33:46 -0700990
Michal Hockoaac45362016-03-25 14:20:24 -0700991 if (can_oom_reap)
Nico Pachee4a38402022-04-21 16:36:01 -0700992 queue_oom_reaper(victim);
Michal Hockoaac45362016-03-25 14:20:24 -0700993
Tetsuo Handa880b76892015-11-05 18:47:51 -0800994 mmdrop(mm);
David Rientjes6b0c81b2012-07-31 16:43:45 -0700995 put_task_struct(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996}
David Rientjes647f2bd2012-03-21 16:33:46 -0700997#undef K
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
David Rientjes309ed882010-08-09 17:18:54 -0700999/*
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07001000 * Kill provided task unless it's secured by setting
1001 * oom_score_adj to OOM_SCORE_ADJ_MIN.
1002 */
Shakeel Buttbbbe4802019-03-05 15:46:12 -08001003static int oom_kill_memcg_member(struct task_struct *task, void *message)
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07001004{
Tetsuo Handad342a0b2019-03-05 15:48:22 -08001005 if (task->signal->oom_score_adj != OOM_SCORE_ADJ_MIN &&
1006 !is_global_init(task)) {
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07001007 get_task_struct(task);
Shakeel Buttbbbe4802019-03-05 15:46:12 -08001008 __oom_kill_process(task, message);
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07001009 }
1010 return 0;
1011}
1012
Roman Gushchin5989ad72018-08-21 21:53:50 -07001013static void oom_kill_process(struct oom_control *oc, const char *message)
1014{
Shakeel Buttbbbe4802019-03-05 15:46:12 -08001015 struct task_struct *victim = oc->chosen;
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07001016 struct mem_cgroup *oom_group;
Roman Gushchin5989ad72018-08-21 21:53:50 -07001017 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
1018 DEFAULT_RATELIMIT_BURST);
1019
1020 /*
1021 * If the task is already exiting, don't alarm the sysadmin or kill
1022 * its children or threads, just give it access to memory reserves
1023 * so it can die quickly
1024 */
Shakeel Buttbbbe4802019-03-05 15:46:12 -08001025 task_lock(victim);
1026 if (task_will_free_mem(victim)) {
1027 mark_oom_victim(victim);
Nico Pachee4a38402022-04-21 16:36:01 -07001028 queue_oom_reaper(victim);
Shakeel Buttbbbe4802019-03-05 15:46:12 -08001029 task_unlock(victim);
1030 put_task_struct(victim);
Roman Gushchin5989ad72018-08-21 21:53:50 -07001031 return;
1032 }
Shakeel Buttbbbe4802019-03-05 15:46:12 -08001033 task_unlock(victim);
Roman Gushchin5989ad72018-08-21 21:53:50 -07001034
1035 if (__ratelimit(&oom_rs))
Shakeel Buttbbbe4802019-03-05 15:46:12 -08001036 dump_header(oc, victim);
Roman Gushchin5989ad72018-08-21 21:53:50 -07001037
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07001038 /*
1039 * Do we need to kill the entire memory cgroup?
1040 * Or even one of the ancestor memory cgroups?
1041 * Check this out before killing the victim task.
1042 */
1043 oom_group = mem_cgroup_get_oom_group(victim, oc->memcg);
1044
Shakeel Buttbbbe4802019-03-05 15:46:12 -08001045 __oom_kill_process(victim, message);
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07001046
1047 /*
1048 * If necessary, kill all tasks in the selected memory cgroup.
1049 */
1050 if (oom_group) {
Dan Schatzbergb6bf9ab2022-01-14 14:05:35 -08001051 memcg_memory_event(oom_group, MEMCG_OOM_GROUP_KILL);
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07001052 mem_cgroup_print_oom_group(oom_group);
Shakeel Buttbbbe4802019-03-05 15:46:12 -08001053 mem_cgroup_scan_tasks(oom_group, oom_kill_memcg_member,
Zhiyuan Dai68d68ff2021-05-04 18:40:12 -07001054 (void *)message);
Roman Gushchin3d8b38e2018-08-21 21:53:54 -07001055 mem_cgroup_put(oom_group);
1056 }
Roman Gushchin5989ad72018-08-21 21:53:50 -07001057}
1058
David Rientjes309ed882010-08-09 17:18:54 -07001059/*
1060 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
1061 */
Yafang Shao432b1de2019-06-28 12:06:59 -07001062static void check_panic_on_oom(struct oom_control *oc)
David Rientjes309ed882010-08-09 17:18:54 -07001063{
1064 if (likely(!sysctl_panic_on_oom))
1065 return;
1066 if (sysctl_panic_on_oom != 2) {
1067 /*
1068 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
1069 * does not panic for cpuset, mempolicy, or memcg allocation
1070 * failures.
1071 */
Yafang Shao432b1de2019-06-28 12:06:59 -07001072 if (oc->constraint != CONSTRAINT_NONE)
David Rientjes309ed882010-08-09 17:18:54 -07001073 return;
1074 }
David Rientjes071a4be2015-09-08 15:00:42 -07001075 /* Do not panic for oom kills triggered by sysrq */
Yaowei Baidb2a0dd2015-11-06 16:28:06 -08001076 if (is_sysrq_oom(oc))
David Rientjes071a4be2015-09-08 15:00:42 -07001077 return;
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001078 dump_header(oc, NULL);
David Rientjes309ed882010-08-09 17:18:54 -07001079 panic("Out of memory: %s panic_on_oom is enabled\n",
1080 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
1081}
1082
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -07001083static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
1084
1085int register_oom_notifier(struct notifier_block *nb)
1086{
1087 return blocking_notifier_chain_register(&oom_notify_list, nb);
1088}
1089EXPORT_SYMBOL_GPL(register_oom_notifier);
1090
1091int unregister_oom_notifier(struct notifier_block *nb)
1092{
1093 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
1094}
1095EXPORT_SYMBOL_GPL(unregister_oom_notifier);
1096
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097/**
David Rientjes6e0fc462015-09-08 15:00:36 -07001098 * out_of_memory - kill the "best" process when we run out of memory
1099 * @oc: pointer to struct oom_control
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 *
1101 * If we run out of memory, we have the choice between either
1102 * killing a random task (bad), letting the system crash (worse)
1103 * OR try to be smart about which process to kill. Note that we
1104 * don't have to be perfect here, we just have to be good.
1105 */
David Rientjes6e0fc462015-09-08 15:00:36 -07001106bool out_of_memory(struct oom_control *oc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107{
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -07001108 unsigned long freed = 0;
1109
Johannes Weinerdc564012015-06-24 16:57:19 -07001110 if (oom_killer_disabled)
1111 return false;
1112
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001113 if (!is_memcg_oom(oc)) {
1114 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
Jann Hornf5302432022-01-14 14:08:27 -08001115 if (freed > 0 && !is_sysrq_oom(oc))
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001116 /* Got some memory back in the last second. */
1117 return true;
1118 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119
David Rientjes7b98c2e2010-08-09 17:18:48 -07001120 /*
David Rientjes9ff48682012-12-11 16:01:30 -08001121 * If current has a pending SIGKILL or is exiting, then automatically
1122 * select it. The goal is to allow it to allocate so that it may
1123 * quickly exit and free its memory.
David Rientjes7b98c2e2010-08-09 17:18:48 -07001124 */
Michal Hocko091f3622016-07-28 15:45:04 -07001125 if (task_will_free_mem(current)) {
Johannes Weiner16e95192015-06-24 16:57:07 -07001126 mark_oom_victim(current);
Nico Pachee4a38402022-04-21 16:36:01 -07001127 queue_oom_reaper(current);
David Rientjes75e8f8b2015-09-08 15:00:47 -07001128 return true;
David Rientjes7b98c2e2010-08-09 17:18:48 -07001129 }
1130
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001131 /*
Michal Hocko3da88fb32016-05-19 17:13:09 -07001132 * The OOM killer does not compensate for IO-less reclaim.
1133 * pagefault_out_of_memory lost its gfp context so we have to
1134 * make sure exclude 0 mask - all other users should have at least
Tetsuo Handaf9c64562019-09-23 15:37:08 -07001135 * ___GFP_DIRECT_RECLAIM to get here. But mem_cgroup_oom() has to
1136 * invoke the OOM killer even if it is a GFP_NOFS allocation.
Michal Hocko3da88fb32016-05-19 17:13:09 -07001137 */
Tetsuo Handaf9c64562019-09-23 15:37:08 -07001138 if (oc->gfp_mask && !(oc->gfp_mask & __GFP_FS) && !is_memcg_oom(oc))
Michal Hocko3da88fb32016-05-19 17:13:09 -07001139 return true;
1140
1141 /*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001142 * Check if there were limitations on the allocation (only relevant for
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001143 * NUMA and memcg) that may require different handling.
Christoph Lameter9b0f8b02006-02-20 18:27:52 -08001144 */
Yafang Shao432b1de2019-06-28 12:06:59 -07001145 oc->constraint = constrained_alloc(oc);
1146 if (oc->constraint != CONSTRAINT_MEMORY_POLICY)
David Rientjes6e0fc462015-09-08 15:00:36 -07001147 oc->nodemask = NULL;
Yafang Shao432b1de2019-06-28 12:06:59 -07001148 check_panic_on_oom(oc);
David Rientjes0aad4b32010-08-09 17:18:59 -07001149
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001150 if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task &&
Shakeel Buttac311a12019-07-11 21:00:31 -07001151 current->mm && !oom_unkillable_task(current) &&
1152 oom_cpuset_eligible(current, oc) &&
David Rientjes121d1ba2012-07-31 16:42:55 -07001153 current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) {
David Rientjes6b0c81b2012-07-31 16:43:45 -07001154 get_task_struct(current);
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001155 oc->chosen = current;
1156 oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)");
David Rientjes75e8f8b2015-09-08 15:00:47 -07001157 return true;
David Rientjes0aad4b32010-08-09 17:18:59 -07001158 }
1159
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001160 select_bad_process(oc);
Johannes Weiner3100dab2018-09-04 15:45:34 -07001161 /* Found nothing?!?! */
1162 if (!oc->chosen) {
Vladimir Davydov2a966b72016-07-26 15:22:33 -07001163 dump_header(oc, NULL);
Johannes Weiner3100dab2018-09-04 15:45:34 -07001164 pr_warn("Out of memory and no killable processes...\n");
1165 /*
1166 * If we got here due to an actual allocation at the
1167 * system level, we cannot survive this and will enter
1168 * an endless loop in the allocator. Bail out now.
1169 */
1170 if (!is_sysrq_oom(oc) && !is_memcg_oom(oc))
1171 panic("System is deadlocked on memory\n");
David Rientjes0aad4b32010-08-09 17:18:59 -07001172 }
Michal Hocko9bfe5de2018-08-17 15:49:04 -07001173 if (oc->chosen && oc->chosen != (void *)-1UL)
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001174 oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" :
1175 "Memory cgroup out of memory");
Vladimir Davydov7c5f64f2016-10-07 16:57:23 -07001176 return !!oc->chosen;
Michal Hockoc32b3cb2015-02-11 15:26:24 -08001177}
1178
David Rientjese3658932010-08-09 17:18:55 -07001179/*
Michal Hocko60e27932021-11-05 13:38:06 -07001180 * The pagefault handler calls here because some allocation has failed. We have
1181 * to take care of the memcg OOM here because this is the only safe context without
1182 * any locks held but let the oom killer triggered from the allocation context care
1183 * about the global OOM.
David Rientjese3658932010-08-09 17:18:55 -07001184 */
1185void pagefault_out_of_memory(void)
1186{
Michal Hocko60e27932021-11-05 13:38:06 -07001187 static DEFINE_RATELIMIT_STATE(pfoom_rs, DEFAULT_RATELIMIT_INTERVAL,
1188 DEFAULT_RATELIMIT_BURST);
David Rientjes6e0fc462015-09-08 15:00:36 -07001189
Johannes Weiner494264202013-10-16 13:46:59 -07001190 if (mem_cgroup_oom_synchronize(true))
Johannes Weinerdc564012015-06-24 16:57:19 -07001191 return;
Johannes Weiner3812c8c2013-09-12 15:13:44 -07001192
Vasily Averin0b281792021-11-05 13:38:02 -07001193 if (fatal_signal_pending(current))
Johannes Weinerdc564012015-06-24 16:57:19 -07001194 return;
Vasily Averin0b281792021-11-05 13:38:02 -07001195
Michal Hocko60e27932021-11-05 13:38:06 -07001196 if (__ratelimit(&pfoom_rs))
1197 pr_warn("Huh VM_FAULT_OOM leaked out to the #PF handler. Retrying PF\n");
David Rientjese3658932010-08-09 17:18:55 -07001198}
Suren Baghdasaryan884a7e52021-09-02 15:00:29 -07001199
1200SYSCALL_DEFINE2(process_mrelease, int, pidfd, unsigned int, flags)
1201{
1202#ifdef CONFIG_MMU
1203 struct mm_struct *mm = NULL;
1204 struct task_struct *task;
1205 struct task_struct *p;
1206 unsigned int f_flags;
Suren Baghdasaryan337546e2021-10-28 14:36:14 -07001207 bool reap = false;
Suren Baghdasaryan884a7e52021-09-02 15:00:29 -07001208 long ret = 0;
1209
1210 if (flags)
1211 return -EINVAL;
1212
Christian Brauneree9955d2021-10-11 15:32:45 +02001213 task = pidfd_get_task(pidfd, &f_flags);
1214 if (IS_ERR(task))
1215 return PTR_ERR(task);
Suren Baghdasaryan884a7e52021-09-02 15:00:29 -07001216
1217 /*
1218 * Make sure to choose a thread which still has a reference to mm
1219 * during the group exit
1220 */
1221 p = find_lock_task_mm(task);
1222 if (!p) {
1223 ret = -ESRCH;
1224 goto put_task;
1225 }
1226
Suren Baghdasaryanba535c12022-01-14 14:06:22 -08001227 mm = p->mm;
1228 mmgrab(mm);
1229
1230 if (task_will_free_mem(p))
1231 reap = true;
1232 else {
1233 /* Error only if the work has not been done already */
1234 if (!test_bit(MMF_OOM_SKIP, &mm->flags))
1235 ret = -EINVAL;
Suren Baghdasaryan884a7e52021-09-02 15:00:29 -07001236 }
1237 task_unlock(p);
1238
1239 if (!reap)
1240 goto drop_mm;
1241
1242 if (mmap_read_lock_killable(mm)) {
1243 ret = -EINTR;
1244 goto drop_mm;
1245 }
Suren Baghdasaryanba535c12022-01-14 14:06:22 -08001246 /*
1247 * Check MMF_OOM_SKIP again under mmap_read_lock protection to ensure
1248 * possible change in exit_mmap is seen
1249 */
1250 if (!test_bit(MMF_OOM_SKIP, &mm->flags) && !__oom_reap_task_mm(mm))
Suren Baghdasaryan884a7e52021-09-02 15:00:29 -07001251 ret = -EAGAIN;
1252 mmap_read_unlock(mm);
1253
1254drop_mm:
Suren Baghdasaryanba535c12022-01-14 14:06:22 -08001255 mmdrop(mm);
Suren Baghdasaryan884a7e52021-09-02 15:00:29 -07001256put_task:
1257 put_task_struct(task);
Suren Baghdasaryan884a7e52021-09-02 15:00:29 -07001258 return ret;
1259#else
1260 return -ENOSYS;
1261#endif /* CONFIG_MMU */
1262}