Thomas Gleixner | 457c899 | 2019-05-19 13:08:55 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2 | /* |
| 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 Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 8 | * Copyright (C) 2010 Google, Inc. |
| 9 | * Rewritten by David Rientjes |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 10 | * |
| 11 | * The routines in this file are used to kill a process when |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 12 | * 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | * |
| 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 Dobriyan | 8ac773b | 2006-10-19 23:28:32 -0700 | [diff] [blame] | 21 | #include <linux/oom.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | #include <linux/mm.h> |
Alexey Dobriyan | 4e950f6 | 2007-07-30 02:36:13 +0400 | [diff] [blame] | 23 | #include <linux/err.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 24 | #include <linux/gfp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 25 | #include <linux/sched.h> |
Ingo Molnar | 6e84f31 | 2017-02-08 18:51:29 +0100 | [diff] [blame] | 26 | #include <linux/sched/mm.h> |
Ingo Molnar | f7ccbae | 2017-02-08 18:51:30 +0100 | [diff] [blame] | 27 | #include <linux/sched/coredump.h> |
Ingo Molnar | 2993002 | 2017-02-08 18:51:36 +0100 | [diff] [blame] | 28 | #include <linux/sched/task.h> |
David Rientjes | 8a7ff02 | 2020-01-30 22:14:26 -0800 | [diff] [blame] | 29 | #include <linux/sched/debug.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | #include <linux/swap.h> |
Suren Baghdasaryan | 884a7e5 | 2021-09-02 15:00:29 -0700 | [diff] [blame] | 31 | #include <linux/syscalls.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | #include <linux/timex.h> |
| 33 | #include <linux/jiffies.h> |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 34 | #include <linux/cpuset.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 35 | #include <linux/export.h> |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 36 | #include <linux/notifier.h> |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 37 | #include <linux/memcontrol.h> |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 38 | #include <linux/mempolicy.h> |
David Howells | 5cd9c58 | 2008-08-14 11:37:28 +0100 | [diff] [blame] | 39 | #include <linux/security.h> |
David Rientjes | edd4554 | 2011-03-22 16:30:12 -0700 | [diff] [blame] | 40 | #include <linux/ptrace.h> |
David Rientjes | f660daa | 2011-10-31 17:07:07 -0700 | [diff] [blame] | 41 | #include <linux/freezer.h> |
KAMEZAWA Hiroyuki | 43d2b11 | 2012-01-10 15:08:09 -0800 | [diff] [blame] | 42 | #include <linux/ftrace.h> |
David Rientjes | dc3f21e | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 43 | #include <linux/ratelimit.h> |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 44 | #include <linux/kthread.h> |
| 45 | #include <linux/init.h> |
Michal Hocko | 4d4bbd8 | 2017-10-03 16:14:50 -0700 | [diff] [blame] | 46 | #include <linux/mmu_notifier.h> |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 47 | |
| 48 | #include <asm/tlb.h> |
| 49 | #include "internal.h" |
Yang Shi | 852d8be | 2017-11-15 17:32:07 -0800 | [diff] [blame] | 50 | #include "slab.h" |
KAMEZAWA Hiroyuki | 43d2b11 | 2012-01-10 15:08:09 -0800 | [diff] [blame] | 51 | |
| 52 | #define CREATE_TRACE_POINTS |
| 53 | #include <trace/events/oom.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | |
sujiaxun | 43fe219 | 2022-02-17 18:51:48 -0800 | [diff] [blame] | 55 | static int sysctl_panic_on_oom; |
| 56 | static int sysctl_oom_kill_allocating_task; |
| 57 | static int sysctl_oom_dump_tasks = 1; |
| 58 | |
Michal Hocko | a195d3f | 2018-08-17 15:49:10 -0700 | [diff] [blame] | 59 | /* |
| 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 Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 67 | DEFINE_MUTEX(oom_lock); |
Suren Baghdasaryan | 67197a4 | 2020-10-13 16:58:35 -0700 | [diff] [blame] | 68 | /* Serializes oom_score_adj and oom_score_adj_min updates */ |
| 69 | DEFINE_MUTEX(oom_adj_mutex); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | |
Shakeel Butt | ac311a1 | 2019-07-11 21:00:31 -0700 | [diff] [blame] | 71 | static inline bool is_memcg_oom(struct oom_control *oc) |
| 72 | { |
| 73 | return oc->memcg != NULL; |
| 74 | } |
| 75 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 76 | #ifdef CONFIG_NUMA |
| 77 | /** |
Ingo Molnar | f0953a1 | 2021-05-06 18:06:47 -0700 | [diff] [blame] | 78 | * oom_cpuset_eligible() - check task eligibility for kill |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 79 | * @start: task struct of which task to consider |
Yi Wang | f364f06 | 2019-09-23 15:37:14 -0700 | [diff] [blame] | 80 | * @oc: pointer to struct oom_control |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 81 | * |
| 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 Butt | ac311a1 | 2019-07-11 21:00:31 -0700 | [diff] [blame] | 85 | * |
| 86 | * This function is assuming oom-killer context and 'current' has triggered |
| 87 | * the oom-killer. |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 88 | */ |
Shakeel Butt | ac311a1 | 2019-07-11 21:00:31 -0700 | [diff] [blame] | 89 | static bool oom_cpuset_eligible(struct task_struct *start, |
| 90 | struct oom_control *oc) |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 91 | { |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 92 | struct task_struct *tsk; |
| 93 | bool ret = false; |
Shakeel Butt | ac311a1 | 2019-07-11 21:00:31 -0700 | [diff] [blame] | 94 | const nodemask_t *mask = oc->nodemask; |
| 95 | |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 96 | rcu_read_lock(); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 97 | for_each_thread(start, tsk) { |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 98 | 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 Tang | b26e517 | 2021-06-30 18:50:56 -0700 | [diff] [blame] | 105 | ret = mempolicy_in_oom_domain(tsk, mask); |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 106 | } else { |
| 107 | /* |
| 108 | * This is not a mempolicy constrained oom, so only |
| 109 | * check the mems of tsk's cpuset. |
| 110 | */ |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 111 | ret = cpuset_mems_allowed_intersects(current, tsk); |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 112 | } |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 113 | if (ret) |
| 114 | break; |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 115 | } |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 116 | rcu_read_unlock(); |
KOSAKI Motohiro | df1090a | 2010-08-09 17:19:39 -0700 | [diff] [blame] | 117 | |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 118 | return ret; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 119 | } |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 120 | #else |
Shakeel Butt | ac311a1 | 2019-07-11 21:00:31 -0700 | [diff] [blame] | 121 | static bool oom_cpuset_eligible(struct task_struct *tsk, struct oom_control *oc) |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 122 | { |
| 123 | return true; |
| 124 | } |
| 125 | #endif /* CONFIG_NUMA */ |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 126 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 127 | /* |
| 128 | * The process p may have detached its own ->mm while exiting or through |
Christoph Hellwig | f5678e7 | 2020-06-10 18:42:06 -0700 | [diff] [blame] | 129 | * kthread_use_mm(), but one or more of its subthreads may still have a valid |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 130 | * pointer. Return p, or any of its subthreads with a valid ->mm, with |
| 131 | * task_lock() held. |
| 132 | */ |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 133 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 134 | { |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 135 | struct task_struct *t; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 136 | |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 137 | rcu_read_lock(); |
| 138 | |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 139 | for_each_thread(p, t) { |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 140 | task_lock(t); |
| 141 | if (likely(t->mm)) |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 142 | goto found; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 143 | task_unlock(t); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 144 | } |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 145 | t = NULL; |
| 146 | found: |
| 147 | rcu_read_unlock(); |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 148 | |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 149 | return t; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 150 | } |
| 151 | |
Yaowei Bai | db2a0dd | 2015-11-06 16:28:06 -0800 | [diff] [blame] | 152 | /* |
| 153 | * order == -1 means the oom kill is required by sysrq, otherwise only |
| 154 | * for display purposes. |
| 155 | */ |
| 156 | static inline bool is_sysrq_oom(struct oom_control *oc) |
| 157 | { |
| 158 | return oc->order == -1; |
| 159 | } |
| 160 | |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 161 | /* return true if the task is not adequate as candidate victim task. */ |
Shakeel Butt | ac311a1 | 2019-07-11 21:00:31 -0700 | [diff] [blame] | 162 | static bool oom_unkillable_task(struct task_struct *p) |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 163 | { |
| 164 | if (is_global_init(p)) |
| 165 | return true; |
| 166 | if (p->flags & PF_KTHREAD) |
| 167 | return true; |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 168 | return false; |
| 169 | } |
| 170 | |
Randy Dunlap | 845be1c | 2021-04-16 15:45:54 -0700 | [diff] [blame] | 171 | /* |
Hui Su | 259b363 | 2020-12-14 19:12:49 -0800 | [diff] [blame] | 172 | * 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 | */ |
| 177 | static bool should_dump_unreclaim_slab(void) |
Yang Shi | 852d8be | 2017-11-15 17:32:07 -0800 | [diff] [blame] | 178 | { |
| 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 Gushchin | d42f324 | 2020-08-06 23:20:39 -0700 | [diff] [blame] | 189 | return (global_node_page_state_pages(NR_SLAB_UNRECLAIMABLE_B) > nr_lru); |
Yang Shi | 852d8be | 2017-11-15 17:32:07 -0800 | [diff] [blame] | 190 | } |
| 191 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | /** |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 193 | * oom_badness - heuristic function to determine which candidate task to kill |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | * @p: task struct of which task we should calculate |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 195 | * @totalpages: total present RAM allowed for page allocation |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 196 | * |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 197 | * 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 200 | */ |
Yafang Shao | 9066e5c | 2020-08-11 18:31:22 -0700 | [diff] [blame] | 201 | long oom_badness(struct task_struct *p, unsigned long totalpages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | { |
David Rientjes | 1e11ad8 | 2012-06-08 13:21:26 -0700 | [diff] [blame] | 203 | long points; |
David Rientjes | 61eafb0 | 2012-06-20 12:52:58 -0700 | [diff] [blame] | 204 | long adj; |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 205 | |
Shakeel Butt | ac311a1 | 2019-07-11 21:00:31 -0700 | [diff] [blame] | 206 | if (oom_unkillable_task(p)) |
Yafang Shao | 9066e5c | 2020-08-11 18:31:22 -0700 | [diff] [blame] | 207 | return LONG_MIN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 208 | |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 209 | p = find_lock_task_mm(p); |
| 210 | if (!p) |
Yafang Shao | 9066e5c | 2020-08-11 18:31:22 -0700 | [diff] [blame] | 211 | return LONG_MIN; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 212 | |
Michal Hocko | bb8a4b7 | 2016-05-20 16:57:18 -0700 | [diff] [blame] | 213 | /* |
| 214 | * Do not even consider tasks which are explicitly marked oom |
Michal Hocko | b18dc5f | 2016-07-28 15:44:46 -0700 | [diff] [blame] | 215 | * unkillable or have been already oom reaped or the are in |
| 216 | * the middle of vfork |
Michal Hocko | bb8a4b7 | 2016-05-20 16:57:18 -0700 | [diff] [blame] | 217 | */ |
David Rientjes | a9c58b90 | 2012-12-11 16:02:54 -0800 | [diff] [blame] | 218 | adj = (long)p->signal->oom_score_adj; |
Michal Hocko | bb8a4b7 | 2016-05-20 16:57:18 -0700 | [diff] [blame] | 219 | if (adj == OOM_SCORE_ADJ_MIN || |
Michal Hocko | 862e307 | 2016-10-07 16:58:57 -0700 | [diff] [blame] | 220 | test_bit(MMF_OOM_SKIP, &p->mm->flags) || |
Michal Hocko | b18dc5f | 2016-07-28 15:44:46 -0700 | [diff] [blame] | 221 | in_vfork(p)) { |
Michal Hocko | 5aecc85 | 2011-11-15 14:36:07 -0800 | [diff] [blame] | 222 | task_unlock(p); |
Yafang Shao | 9066e5c | 2020-08-11 18:31:22 -0700 | [diff] [blame] | 223 | return LONG_MIN; |
Michal Hocko | 5aecc85 | 2011-11-15 14:36:07 -0800 | [diff] [blame] | 224 | } |
| 225 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 226 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 227 | * The baseline for the badness score is the proportion of RAM that each |
KOSAKI Motohiro | f755a04 | 2011-04-27 15:26:50 -0700 | [diff] [blame] | 228 | * task's rss, pagetable and swap space use. |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 229 | */ |
Kirill A. Shutemov | dc6c9a3 | 2015-02-11 15:26:50 -0800 | [diff] [blame] | 230 | points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) + |
Kirill A. Shutemov | af5b0f6 | 2017-11-15 17:35:40 -0800 | [diff] [blame] | 231 | mm_pgtables_bytes(p->mm) / PAGE_SIZE; |
Andrew Morton | 97c2c9b8 | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 232 | task_unlock(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 233 | |
David Rientjes | 61eafb0 | 2012-06-20 12:52:58 -0700 | [diff] [blame] | 234 | /* Normalize to oom_score_adj units */ |
| 235 | adj *= totalpages / 1000; |
| 236 | points += adj; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 237 | |
Yafang Shao | 9066e5c | 2020-08-11 18:31:22 -0700 | [diff] [blame] | 238 | return points; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | } |
| 240 | |
yuzhoujian | ef8444e | 2018-12-28 00:36:07 -0800 | [diff] [blame] | 241 | static 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 Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 246 | }; |
| 247 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 248 | /* |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 249 | * Determine the type of allocation constraint. |
| 250 | */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 251 | static enum oom_constraint constrained_alloc(struct oom_control *oc) |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 252 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 253 | struct zone *zone; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 254 | struct zoneref *z; |
Joonsoo Kim | 97a225e | 2020-06-03 15:59:01 -0700 | [diff] [blame] | 255 | enum zone_type highest_zoneidx = gfp_zone(oc->gfp_mask); |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 256 | bool cpuset_limited = false; |
| 257 | int nid; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 258 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 259 | if (is_memcg_oom(oc)) { |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 260 | oc->totalpages = mem_cgroup_get_max(oc->memcg) ?: 1; |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 261 | return CONSTRAINT_MEMCG; |
| 262 | } |
| 263 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 264 | /* Default to all available memory */ |
Arun KS | ca79b0c | 2018-12-28 00:34:29 -0800 | [diff] [blame] | 265 | oc->totalpages = totalram_pages() + total_swap_pages; |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 266 | |
| 267 | if (!IS_ENABLED(CONFIG_NUMA)) |
| 268 | return CONSTRAINT_NONE; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 269 | |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 270 | if (!oc->zonelist) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 271 | return CONSTRAINT_NONE; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 272 | /* |
| 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 Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 277 | if (oc->gfp_mask & __GFP_THISNODE) |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 278 | return CONSTRAINT_NONE; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 279 | |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 280 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 281 | * 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 Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 284 | */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 285 | if (oc->nodemask && |
| 286 | !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) { |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 287 | oc->totalpages = total_swap_pages; |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 288 | for_each_node_mask(nid, *oc->nodemask) |
Michal Hocko | 1eb41bb0 | 2019-09-23 15:37:16 -0700 | [diff] [blame] | 289 | oc->totalpages += node_present_pages(nid); |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 290 | return CONSTRAINT_MEMORY_POLICY; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 291 | } |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 292 | |
| 293 | /* Check this allocation failure is caused by cpuset's wall function */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 294 | for_each_zone_zonelist_nodemask(zone, z, oc->zonelist, |
Joonsoo Kim | 97a225e | 2020-06-03 15:59:01 -0700 | [diff] [blame] | 295 | highest_zoneidx, oc->nodemask) |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 296 | if (!cpuset_zone_allowed(zone, oc->gfp_mask)) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 297 | cpuset_limited = true; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 298 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 299 | if (cpuset_limited) { |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 300 | oc->totalpages = total_swap_pages; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 301 | for_each_node_mask(nid, cpuset_current_mems_allowed) |
Michal Hocko | 1eb41bb0 | 2019-09-23 15:37:16 -0700 | [diff] [blame] | 302 | oc->totalpages += node_present_pages(nid); |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 303 | return CONSTRAINT_CPUSET; |
| 304 | } |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 305 | return CONSTRAINT_NONE; |
| 306 | } |
| 307 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 308 | static int oom_evaluate_task(struct task_struct *task, void *arg) |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 309 | { |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 310 | struct oom_control *oc = arg; |
Yafang Shao | 9066e5c | 2020-08-11 18:31:22 -0700 | [diff] [blame] | 311 | long points; |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 312 | |
Shakeel Butt | ac311a1 | 2019-07-11 21:00:31 -0700 | [diff] [blame] | 313 | 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 Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 318 | goto next; |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 319 | |
| 320 | /* |
| 321 | * This task already has access to memory reserves and is being killed. |
Michal Hocko | a373966 | 2016-07-28 15:45:01 -0700 | [diff] [blame] | 322 | * Don't allow any other task to have access to the reserves unless |
Michal Hocko | 862e307 | 2016-10-07 16:58:57 -0700 | [diff] [blame] | 323 | * the task has MMF_OOM_SKIP because chances that it would release |
Michal Hocko | a373966 | 2016-07-28 15:45:01 -0700 | [diff] [blame] | 324 | * any memory is quite low. |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 325 | */ |
Michal Hocko | 862e307 | 2016-10-07 16:58:57 -0700 | [diff] [blame] | 326 | if (!is_sysrq_oom(oc) && tsk_is_oom_victim(task)) { |
| 327 | if (test_bit(MMF_OOM_SKIP, &task->signal->oom_mm->flags)) |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 328 | goto next; |
| 329 | goto abort; |
Michal Hocko | a373966 | 2016-07-28 15:45:01 -0700 | [diff] [blame] | 330 | } |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 331 | |
David Rientjes | e1e12d2 | 2012-12-11 16:02:56 -0800 | [diff] [blame] | 332 | /* |
| 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 Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 336 | if (oom_task_origin(task)) { |
Yafang Shao | 9066e5c | 2020-08-11 18:31:22 -0700 | [diff] [blame] | 337 | points = LONG_MAX; |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 338 | goto select; |
| 339 | } |
David Rientjes | e1e12d2 | 2012-12-11 16:02:56 -0800 | [diff] [blame] | 340 | |
Shakeel Butt | ac311a1 | 2019-07-11 21:00:31 -0700 | [diff] [blame] | 341 | points = oom_badness(task, oc->totalpages); |
Yafang Shao | 9066e5c | 2020-08-11 18:31:22 -0700 | [diff] [blame] | 342 | if (points == LONG_MIN || points < oc->chosen_points) |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 343 | goto next; |
| 344 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 345 | select: |
| 346 | if (oc->chosen) |
| 347 | put_task_struct(oc->chosen); |
| 348 | get_task_struct(task); |
| 349 | oc->chosen = task; |
| 350 | oc->chosen_points = points; |
| 351 | next: |
| 352 | return 0; |
| 353 | abort: |
| 354 | if (oc->chosen) |
| 355 | put_task_struct(oc->chosen); |
| 356 | oc->chosen = (void *)-1UL; |
| 357 | return 1; |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 358 | } |
| 359 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 360 | /* |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 361 | * 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 363 | */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 364 | static void select_bad_process(struct oom_control *oc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 365 | { |
Yafang Shao | 9066e5c | 2020-08-11 18:31:22 -0700 | [diff] [blame] | 366 | oc->chosen_points = LONG_MIN; |
| 367 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 368 | if (is_memcg_oom(oc)) |
| 369 | mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc); |
| 370 | else { |
| 371 | struct task_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 372 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 373 | rcu_read_lock(); |
| 374 | for_each_process(p) |
| 375 | if (oom_evaluate_task(p, oc)) |
| 376 | break; |
| 377 | rcu_read_unlock(); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 378 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | } |
| 380 | |
Shakeel Butt | 5eee7e1 | 2019-07-11 21:00:23 -0700 | [diff] [blame] | 381 | static int dump_task(struct task_struct *p, void *arg) |
| 382 | { |
| 383 | struct oom_control *oc = arg; |
| 384 | struct task_struct *task; |
| 385 | |
Shakeel Butt | ac311a1 | 2019-07-11 21:00:31 -0700 | [diff] [blame] | 386 | 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 Butt | 5eee7e1 | 2019-07-11 21:00:23 -0700 | [diff] [blame] | 391 | return 0; |
| 392 | |
| 393 | task = find_lock_task_mm(p); |
| 394 | if (!task) { |
| 395 | /* |
Tang Yizhou | f8159c1 | 2021-02-24 12:09:50 -0800 | [diff] [blame] | 396 | * 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 Butt | 5eee7e1 | 2019-07-11 21:00:23 -0700 | [diff] [blame] | 398 | */ |
| 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 413 | /** |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 414 | * dump_tasks - dump current memory state of all system tasks |
Shakeel Butt | 5eee7e1 | 2019-07-11 21:00:23 -0700 | [diff] [blame] | 415 | * @oc: pointer to struct oom_control |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 416 | * |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 417 | * 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. Shutemov | af5b0f6 | 2017-11-15 17:35:40 -0800 | [diff] [blame] | 420 | * State information includes task's pid, uid, tgid, vm size, rss, |
| 421 | * pgtables_bytes, swapents, oom_score_adj value, and name. |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 422 | */ |
Shakeel Butt | 5eee7e1 | 2019-07-11 21:00:23 -0700 | [diff] [blame] | 423 | static void dump_tasks(struct oom_control *oc) |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 424 | { |
Rodrigo Freire | c3b78b1 | 2018-08-21 21:52:41 -0700 | [diff] [blame] | 425 | 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 Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 427 | |
Shakeel Butt | 5eee7e1 | 2019-07-11 21:00:23 -0700 | [diff] [blame] | 428 | if (is_memcg_oom(oc)) |
| 429 | mem_cgroup_scan_tasks(oc->memcg, dump_task, oc); |
| 430 | else { |
| 431 | struct task_struct *p; |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 432 | |
Shakeel Butt | 5eee7e1 | 2019-07-11 21:00:23 -0700 | [diff] [blame] | 433 | rcu_read_lock(); |
| 434 | for_each_process(p) |
| 435 | dump_task(p, oc); |
| 436 | rcu_read_unlock(); |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 437 | } |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 438 | } |
| 439 | |
yuzhoujian | ef8444e | 2018-12-28 00:36:07 -0800 | [diff] [blame] | 440 | static 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(); |
yuzhoujian | f0c867d | 2018-12-28 00:36:10 -0800 | [diff] [blame] | 447 | mem_cgroup_print_oom_context(oc->memcg, victim); |
yuzhoujian | ef8444e | 2018-12-28 00:36:07 -0800 | [diff] [blame] | 448 | 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 Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 452 | static void dump_header(struct oom_control *oc, struct task_struct *p) |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 453 | { |
yuzhoujian | ef8444e | 2018-12-28 00:36:07 -0800 | [diff] [blame] | 454 | 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 Hocko | 0205f75 | 2017-11-15 17:39:14 -0800 | [diff] [blame] | 456 | current->signal->oom_score_adj); |
Michal Hocko | 9254990 | 2016-10-07 16:59:33 -0700 | [diff] [blame] | 457 | if (!IS_ENABLED(CONFIG_COMPACTION) && oc->order) |
| 458 | pr_warn("COMPACTION is disabled!!!\n"); |
Vlastimil Babka | a0795cd | 2016-03-15 14:56:05 -0700 | [diff] [blame] | 459 | |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 460 | dump_stack(); |
Yang Shi | 852d8be | 2017-11-15 17:32:07 -0800 | [diff] [blame] | 461 | if (is_memcg_oom(oc)) |
yuzhoujian | f0c867d | 2018-12-28 00:36:10 -0800 | [diff] [blame] | 462 | mem_cgroup_print_oom_meminfo(oc->memcg); |
Yang Shi | 852d8be | 2017-11-15 17:32:07 -0800 | [diff] [blame] | 463 | else { |
Michal Hocko | 974f436 | 2022-08-23 11:22:30 +0200 | [diff] [blame] | 464 | __show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask, gfp_zone(oc->gfp_mask)); |
Hui Su | 259b363 | 2020-12-14 19:12:49 -0800 | [diff] [blame] | 465 | if (should_dump_unreclaim_slab()) |
Yang Shi | 852d8be | 2017-11-15 17:32:07 -0800 | [diff] [blame] | 466 | dump_unreclaimable_slab(); |
| 467 | } |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 468 | if (sysctl_oom_dump_tasks) |
Shakeel Butt | 5eee7e1 | 2019-07-11 21:00:23 -0700 | [diff] [blame] | 469 | dump_tasks(oc); |
yuzhoujian | ef8444e | 2018-12-28 00:36:07 -0800 | [diff] [blame] | 470 | if (p) |
| 471 | dump_oom_summary(oc, p); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 472 | } |
| 473 | |
Michal Hocko | 5695be1 | 2014-10-20 18:12:32 +0200 | [diff] [blame] | 474 | /* |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 475 | * Number of OOM victims in flight |
Michal Hocko | 5695be1 | 2014-10-20 18:12:32 +0200 | [diff] [blame] | 476 | */ |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 477 | static atomic_t oom_victims = ATOMIC_INIT(0); |
| 478 | static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait); |
Michal Hocko | 5695be1 | 2014-10-20 18:12:32 +0200 | [diff] [blame] | 479 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 480 | static bool oom_killer_disabled __read_mostly; |
Michal Hocko | 5695be1 | 2014-10-20 18:12:32 +0200 | [diff] [blame] | 481 | |
Michal Hocko | bc448e8 | 2016-03-25 14:20:30 -0700 | [diff] [blame] | 482 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
| 483 | |
Michal Hocko | 3ef22df | 2016-05-19 17:13:12 -0700 | [diff] [blame] | 484 | /* |
| 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 Hocko | 44a70ade | 2016-07-28 15:44:43 -0700 | [diff] [blame] | 490 | bool process_shares_mm(struct task_struct *p, struct mm_struct *mm) |
Michal Hocko | 3ef22df | 2016-05-19 17:13:12 -0700 | [diff] [blame] | 491 | { |
| 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 Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 502 | #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 | */ |
| 507 | static struct task_struct *oom_reaper_th; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 508 | static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait); |
Vladimir Davydov | 29c696e | 2016-03-25 14:20:39 -0700 | [diff] [blame] | 509 | static struct task_struct *oom_reaper_list; |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 510 | static DEFINE_SPINLOCK(oom_reaper_lock); |
| 511 | |
Suren Baghdasaryan | bf3980c | 2022-05-31 15:30:59 -0700 | [diff] [blame] | 512 | static bool __oom_reap_task_mm(struct mm_struct *mm) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 513 | { |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 514 | struct vm_area_struct *vma; |
Michal Hocko | 93065ac | 2018-08-21 21:52:33 -0700 | [diff] [blame] | 515 | bool ret = true; |
Liam R. Howlett | e1c2c77 | 2022-09-06 19:49:03 +0000 | [diff] [blame] | 516 | VMA_ITERATOR(vmi, mm, 0); |
David Rientjes | 27ae357 | 2018-05-11 16:02:04 -0700 | [diff] [blame] | 517 | |
| 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. Howlett | e1c2c77 | 2022-09-06 19:49:03 +0000 | [diff] [blame] | 526 | for_each_vma(vmi, vma) { |
Hugh Dickins | a213e5c | 2022-02-14 18:23:29 -0800 | [diff] [blame] | 527 | if (vma->vm_flags & (VM_HUGETLB|VM_PFNMAP)) |
David Rientjes | 27ae357 | 2018-05-11 16:02:04 -0700 | [diff] [blame] | 528 | 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 Glisse | ac46d4f | 2018-12-28 00:38:09 -0800 | [diff] [blame] | 541 | struct mmu_notifier_range range; |
David Rientjes | 27ae357 | 2018-05-11 16:02:04 -0700 | [diff] [blame] | 542 | struct mmu_gather tlb; |
| 543 | |
Jérôme Glisse | 6f4f13e | 2019-05-13 17:20:49 -0700 | [diff] [blame] | 544 | mmu_notifier_range_init(&range, MMU_NOTIFY_UNMAP, 0, |
Alistair Popple | 7d4a8be | 2023-01-10 13:57:22 +1100 | [diff] [blame] | 545 | mm, vma->vm_start, |
Jérôme Glisse | ac46d4f | 2018-12-28 00:38:09 -0800 | [diff] [blame] | 546 | vma->vm_end); |
Will Deacon | a72afd8 | 2021-01-27 23:53:45 +0000 | [diff] [blame] | 547 | tlb_gather_mmu(&tlb, mm); |
Jérôme Glisse | ac46d4f | 2018-12-28 00:38:09 -0800 | [diff] [blame] | 548 | if (mmu_notifier_invalidate_range_start_nonblock(&range)) { |
Will Deacon | ae8eba8 | 2021-01-27 23:53:43 +0000 | [diff] [blame] | 549 | tlb_finish_mmu(&tlb); |
Michal Hocko | 93065ac | 2018-08-21 21:52:33 -0700 | [diff] [blame] | 550 | ret = false; |
| 551 | continue; |
| 552 | } |
Jérôme Glisse | ac46d4f | 2018-12-28 00:38:09 -0800 | [diff] [blame] | 553 | unmap_page_range(&tlb, vma, range.start, range.end, NULL); |
| 554 | mmu_notifier_invalidate_range_end(&range); |
Will Deacon | ae8eba8 | 2021-01-27 23:53:43 +0000 | [diff] [blame] | 555 | tlb_finish_mmu(&tlb); |
David Rientjes | 27ae357 | 2018-05-11 16:02:04 -0700 | [diff] [blame] | 556 | } |
| 557 | } |
Michal Hocko | 93065ac | 2018-08-21 21:52:33 -0700 | [diff] [blame] | 558 | |
| 559 | return ret; |
David Rientjes | 27ae357 | 2018-05-11 16:02:04 -0700 | [diff] [blame] | 560 | } |
| 561 | |
Michal Hocko | 431f42fd | 2018-08-21 21:52:45 -0700 | [diff] [blame] | 562 | /* |
| 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 Rientjes | 27ae357 | 2018-05-11 16:02:04 -0700 | [diff] [blame] | 568 | static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm) |
| 569 | { |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 570 | bool ret = true; |
| 571 | |
Michel Lespinasse | d8ed45c | 2020-06-08 21:33:25 -0700 | [diff] [blame] | 572 | if (!mmap_read_trylock(mm)) { |
Roman Gushchin | 422580c | 2017-07-10 15:49:05 -0700 | [diff] [blame] | 573 | trace_skip_task_reaping(tsk->pid); |
Michal Hocko | af5679f | 2018-08-21 21:52:37 -0700 | [diff] [blame] | 574 | return false; |
Michal Hocko | 4d4bbd8 | 2017-10-03 16:14:50 -0700 | [diff] [blame] | 575 | } |
| 576 | |
| 577 | /* |
Andrea Arcangeli | 2129258 | 2017-09-06 16:25:00 -0700 | [diff] [blame] | 578 | * 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 Lespinasse | 3e4e28c | 2020-06-08 21:33:51 -0700 | [diff] [blame] | 580 | * under mmap_lock for reading because it serializes against the |
| 581 | * mmap_write_lock();mmap_write_unlock() cycle in exit_mmap(). |
Michal Hocko | e5e3f4c | 2016-07-26 15:24:50 -0700 | [diff] [blame] | 582 | */ |
Andrea Arcangeli | 2129258 | 2017-09-06 16:25:00 -0700 | [diff] [blame] | 583 | if (test_bit(MMF_OOM_SKIP, &mm->flags)) { |
Roman Gushchin | 422580c | 2017-07-10 15:49:05 -0700 | [diff] [blame] | 584 | trace_skip_task_reaping(tsk->pid); |
Michal Hocko | 431f42fd | 2018-08-21 21:52:45 -0700 | [diff] [blame] | 585 | goto out_unlock; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 586 | } |
| 587 | |
Roman Gushchin | 422580c | 2017-07-10 15:49:05 -0700 | [diff] [blame] | 588 | trace_start_task_reaping(tsk->pid); |
| 589 | |
Michal Hocko | 93065ac | 2018-08-21 21:52:33 -0700 | [diff] [blame] | 590 | /* failed to reap part of the address space. Try again later */ |
Michal Hocko | 431f42fd | 2018-08-21 21:52:45 -0700 | [diff] [blame] | 591 | ret = __oom_reap_task_mm(mm); |
| 592 | if (!ret) |
| 593 | goto out_finish; |
Michal Hocko | 3f70dc3 | 2016-10-07 16:59:06 -0700 | [diff] [blame] | 594 | |
Michal Hocko | bc448e8 | 2016-03-25 14:20:30 -0700 | [diff] [blame] | 595 | 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 Hocko | 431f42fd | 2018-08-21 21:52:45 -0700 | [diff] [blame] | 600 | out_finish: |
| 601 | trace_finish_task_reaping(tsk->pid); |
| 602 | out_unlock: |
Michel Lespinasse | d8ed45c | 2020-06-08 21:33:25 -0700 | [diff] [blame] | 603 | mmap_read_unlock(mm); |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 604 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 605 | return ret; |
| 606 | } |
| 607 | |
Michal Hocko | bc448e8 | 2016-03-25 14:20:30 -0700 | [diff] [blame] | 608 | #define MAX_OOM_REAP_RETRIES 10 |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 609 | static void oom_reap_task(struct task_struct *tsk) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 610 | { |
| 611 | int attempts = 0; |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 612 | struct mm_struct *mm = tsk->signal->oom_mm; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 613 | |
Michel Lespinasse | 3e4e28c | 2020-06-08 21:33:51 -0700 | [diff] [blame] | 614 | /* Retry the mmap_read_trylock(mm) a few times */ |
David Rientjes | 27ae357 | 2018-05-11 16:02:04 -0700 | [diff] [blame] | 615 | while (attempts++ < MAX_OOM_REAP_RETRIES && !oom_reap_task_mm(tsk, mm)) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 616 | schedule_timeout_idle(HZ/10); |
| 617 | |
Tetsuo Handa | 97b1255 | 2018-04-05 16:25:45 -0700 | [diff] [blame] | 618 | if (attempts <= MAX_OOM_REAP_RETRIES || |
| 619 | test_bit(MMF_OOM_SKIP, &mm->flags)) |
Tetsuo Handa | 7ebffa4 | 2016-10-07 16:58:45 -0700 | [diff] [blame] | 620 | goto done; |
Michal Hocko | 11a410d | 2016-07-28 15:44:58 -0700 | [diff] [blame] | 621 | |
Tetsuo Handa | 7ebffa4 | 2016-10-07 16:58:45 -0700 | [diff] [blame] | 622 | pr_info("oom_reaper: unable to reap pid:%d (%s)\n", |
| 623 | task_pid_nr(tsk), tsk->comm); |
David Rientjes | 8a7ff02 | 2020-01-30 22:14:26 -0800 | [diff] [blame] | 624 | sched_show_task(tsk); |
Tetsuo Handa | 7ebffa4 | 2016-10-07 16:58:45 -0700 | [diff] [blame] | 625 | debug_show_all_locks(); |
Michal Hocko | bc448e8 | 2016-03-25 14:20:30 -0700 | [diff] [blame] | 626 | |
Tetsuo Handa | 7ebffa4 | 2016-10-07 16:58:45 -0700 | [diff] [blame] | 627 | done: |
Michal Hocko | 449d777 | 2016-05-19 17:13:15 -0700 | [diff] [blame] | 628 | tsk->oom_reaper_list = NULL; |
Michal Hocko | 449d777 | 2016-05-19 17:13:15 -0700 | [diff] [blame] | 629 | |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 630 | /* |
| 631 | * Hide this mm from OOM killer because it has been either reaped or |
Michel Lespinasse | 3e4e28c | 2020-06-08 21:33:51 -0700 | [diff] [blame] | 632 | * somebody can't call mmap_write_unlock(mm). |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 633 | */ |
Michal Hocko | 862e307 | 2016-10-07 16:58:57 -0700 | [diff] [blame] | 634 | set_bit(MMF_OOM_SKIP, &mm->flags); |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 635 | |
Nico Pache | e4a3840 | 2022-04-21 16:36:01 -0700 | [diff] [blame] | 636 | /* Drop a reference taken by queue_oom_reaper */ |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 637 | put_task_struct(tsk); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 638 | } |
| 639 | |
| 640 | static int oom_reaper(void *unused) |
| 641 | { |
Sultan Alsawaf | 3723929 | 2021-11-05 13:43:25 -0700 | [diff] [blame] | 642 | set_freezable(); |
| 643 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 644 | while (true) { |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 645 | struct task_struct *tsk = NULL; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 646 | |
Vladimir Davydov | 29c696e | 2016-03-25 14:20:39 -0700 | [diff] [blame] | 647 | wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL); |
Nico Pache | e4a3840 | 2022-04-21 16:36:01 -0700 | [diff] [blame] | 648 | spin_lock_irq(&oom_reaper_lock); |
Vladimir Davydov | 29c696e | 2016-03-25 14:20:39 -0700 | [diff] [blame] | 649 | if (oom_reaper_list != NULL) { |
| 650 | tsk = oom_reaper_list; |
| 651 | oom_reaper_list = tsk->oom_reaper_list; |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 652 | } |
Nico Pache | e4a3840 | 2022-04-21 16:36:01 -0700 | [diff] [blame] | 653 | spin_unlock_irq(&oom_reaper_lock); |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 654 | |
| 655 | if (tsk) |
| 656 | oom_reap_task(tsk); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 657 | } |
| 658 | |
| 659 | return 0; |
| 660 | } |
| 661 | |
Nico Pache | e4a3840 | 2022-04-21 16:36:01 -0700 | [diff] [blame] | 662 | static 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) |
| 692 | static void queue_oom_reaper(struct task_struct *tsk) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 693 | { |
Tetsuo Handa | 9bcdeb5 | 2019-02-01 14:20:31 -0800 | [diff] [blame] | 694 | /* mm is already queued? */ |
| 695 | if (test_and_set_bit(MMF_OOM_REAP_QUEUED, &tsk->signal->oom_mm->flags)) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 696 | return; |
| 697 | |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 698 | get_task_struct(tsk); |
Nico Pache | e4a3840 | 2022-04-21 16:36:01 -0700 | [diff] [blame] | 699 | 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 Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 702 | } |
| 703 | |
Andrew Morton | a19cad0 | 2022-06-01 15:57:16 -0700 | [diff] [blame] | 704 | #ifdef CONFIG_SYSCTL |
| 705 | static 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 Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 733 | static int __init oom_init(void) |
| 734 | { |
| 735 | oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper"); |
sujiaxun | 43fe219 | 2022-02-17 18:51:48 -0800 | [diff] [blame] | 736 | #ifdef CONFIG_SYSCTL |
| 737 | register_sysctl_init("vm", vm_oom_kill_table); |
| 738 | #endif |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 739 | return 0; |
| 740 | } |
| 741 | subsys_initcall(oom_init) |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 742 | #else |
Nico Pache | e4a3840 | 2022-04-21 16:36:01 -0700 | [diff] [blame] | 743 | static inline void queue_oom_reaper(struct task_struct *tsk) |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 744 | { |
| 745 | } |
| 746 | #endif /* CONFIG_MMU */ |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 747 | |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 748 | /** |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 749 | * mark_oom_victim - mark the given task as OOM victim |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 750 | * @tsk: task to mark |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 751 | * |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 752 | * Has to be called with oom_lock held and never after |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 753 | * oom has been disabled already. |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 754 | * |
| 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 Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 757 | */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 758 | static void mark_oom_victim(struct task_struct *tsk) |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 759 | { |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 760 | struct mm_struct *mm = tsk->mm; |
| 761 | |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 762 | 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 Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 766 | |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 767 | /* oom_mm is bound to the signal struct life time. */ |
Suren Baghdasaryan | b3541d9 | 2022-05-31 15:31:00 -0700 | [diff] [blame] | 768 | if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm)) |
Vegard Nossum | f1f1007 | 2017-02-27 14:30:07 -0800 | [diff] [blame] | 769 | mmgrab(tsk->signal->oom_mm); |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 770 | |
Michal Hocko | 63a8ca9 | 2015-02-11 15:26:15 -0800 | [diff] [blame] | 771 | /* |
| 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 Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 778 | atomic_inc(&oom_victims); |
Roman Gushchin | 422580c | 2017-07-10 15:49:05 -0700 | [diff] [blame] | 779 | trace_mark_victim(tsk->pid); |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 780 | } |
| 781 | |
| 782 | /** |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 783 | * exit_oom_victim - note the exit of an OOM victim |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 784 | */ |
Tetsuo Handa | 3853120 | 2016-10-07 16:59:03 -0700 | [diff] [blame] | 785 | void exit_oom_victim(void) |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 786 | { |
Tetsuo Handa | 3853120 | 2016-10-07 16:59:03 -0700 | [diff] [blame] | 787 | clear_thread_flag(TIF_MEMDIE); |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 788 | |
Johannes Weiner | c38f102 | 2015-06-24 16:57:13 -0700 | [diff] [blame] | 789 | if (!atomic_dec_return(&oom_victims)) |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 790 | wake_up_all(&oom_victims_wait); |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 791 | } |
| 792 | |
| 793 | /** |
Michal Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 794 | * oom_killer_enable - enable OOM killer |
| 795 | */ |
| 796 | void oom_killer_enable(void) |
| 797 | { |
| 798 | oom_killer_disabled = false; |
Michal Hocko | d75da00 | 2017-05-03 14:54:57 -0700 | [diff] [blame] | 799 | pr_info("OOM killer enabled.\n"); |
Michal Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 800 | } |
| 801 | |
| 802 | /** |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 803 | * oom_killer_disable - disable OOM killer |
Michal Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 804 | * @timeout: maximum timeout to wait for oom victims in jiffies |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 805 | * |
| 806 | * Forces all page allocations to fail rather than trigger OOM killer. |
Michal Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 807 | * Will block and wait until all OOM victims are killed or the given |
| 808 | * timeout expires. |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 809 | * |
| 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 Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 817 | bool oom_killer_disable(signed long timeout) |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 818 | { |
Michal Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 819 | signed long ret; |
| 820 | |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 821 | /* |
Tetsuo Handa | 6afcf28 | 2016-03-17 14:20:45 -0700 | [diff] [blame] | 822 | * 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 Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 824 | */ |
Tetsuo Handa | 6afcf28 | 2016-03-17 14:20:45 -0700 | [diff] [blame] | 825 | if (mutex_lock_killable(&oom_lock)) |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 826 | return false; |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 827 | oom_killer_disabled = true; |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 828 | mutex_unlock(&oom_lock); |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 829 | |
Michal Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 830 | 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 Hocko | d75da00 | 2017-05-03 14:54:57 -0700 | [diff] [blame] | 836 | pr_info("OOM killer disabled.\n"); |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 837 | |
| 838 | return true; |
| 839 | } |
| 840 | |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 841 | static inline bool __task_will_free_mem(struct task_struct *task) |
| 842 | { |
| 843 | struct signal_struct *sig = task->signal; |
| 844 | |
| 845 | /* |
Eric W. Biederman | d67e03e | 2021-09-01 11:23:38 -0500 | [diff] [blame] | 846 | * A coredumping process may sleep for an extended period in |
Eric W. Biederman | 9230738 | 2021-09-01 11:33:50 -0500 | [diff] [blame] | 847 | * coredump_task_exit(), so the oom killer cannot assume that |
Eric W. Biederman | d67e03e | 2021-09-01 11:23:38 -0500 | [diff] [blame] | 848 | * the process will promptly exit and release memory. |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 849 | */ |
Eric W. Biederman | 98b24b1 | 2021-11-19 11:29:48 -0600 | [diff] [blame] | 850 | if (sig->core_state) |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 851 | 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 Hocko | 091f362 | 2016-07-28 15:45:04 -0700 | [diff] [blame] | 866 | * Caller has to make sure that task->mm is stable (hold task_lock or |
| 867 | * it operates on the current). |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 868 | */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 869 | static bool task_will_free_mem(struct task_struct *task) |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 870 | { |
Michal Hocko | 091f362 | 2016-07-28 15:45:04 -0700 | [diff] [blame] | 871 | struct mm_struct *mm = task->mm; |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 872 | struct task_struct *p; |
Geert Uytterhoeven | f33e6f0 | 2016-08-11 15:33:09 -0700 | [diff] [blame] | 873 | bool ret = true; |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 874 | |
Michal Hocko | 091f362 | 2016-07-28 15:45:04 -0700 | [diff] [blame] | 875 | /* |
| 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 Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 883 | if (!__task_will_free_mem(task)) |
| 884 | return false; |
| 885 | |
| 886 | /* |
Michal Hocko | 696453e | 2016-07-28 15:44:55 -0700 | [diff] [blame] | 887 | * This task has already been drained by the oom reaper so there are |
| 888 | * only small chances it will free some more |
| 889 | */ |
Michal Hocko | 862e307 | 2016-10-07 16:58:57 -0700 | [diff] [blame] | 890 | if (test_bit(MMF_OOM_SKIP, &mm->flags)) |
Michal Hocko | 696453e | 2016-07-28 15:44:55 -0700 | [diff] [blame] | 891 | return false; |
Michal Hocko | 696453e | 2016-07-28 15:44:55 -0700 | [diff] [blame] | 892 | |
Michal Hocko | 091f362 | 2016-07-28 15:45:04 -0700 | [diff] [blame] | 893 | if (atomic_read(&mm->mm_users) <= 1) |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 894 | return true; |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 895 | |
| 896 | /* |
Michal Hocko | 5870c2e | 2016-10-07 16:57:32 -0700 | [diff] [blame] | 897 | * 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 Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 900 | */ |
| 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 Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 912 | |
| 913 | return ret; |
| 914 | } |
| 915 | |
Shakeel Butt | bbbe480 | 2019-03-05 15:46:12 -0800 | [diff] [blame] | 916 | static void __oom_kill_process(struct task_struct *victim, const char *message) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 917 | { |
Roman Gushchin | 5989ad7 | 2018-08-21 21:53:50 -0700 | [diff] [blame] | 918 | struct task_struct *p; |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 919 | struct mm_struct *mm; |
Tetsuo Handa | bb29902 | 2016-03-25 14:20:44 -0700 | [diff] [blame] | 920 | bool can_oom_reap = true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 921 | |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 922 | p = find_lock_task_mm(victim); |
| 923 | if (!p) { |
Yafang Shao | 619b5b4 | 2020-08-11 18:31:32 -0700 | [diff] [blame] | 924 | pr_info("%s: OOM victim %d (%s) is already exiting. Skip killing the task\n", |
| 925 | message, task_pid_nr(victim), victim->comm); |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 926 | put_task_struct(victim); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 927 | return; |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 928 | } else if (victim != p) { |
| 929 | get_task_struct(p); |
| 930 | put_task_struct(victim); |
| 931 | victim = p; |
| 932 | } |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 933 | |
Tetsuo Handa | 880b7689 | 2015-11-05 18:47:51 -0800 | [diff] [blame] | 934 | /* Get a reference to safely compare mm after task_unlock(victim) */ |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 935 | mm = victim->mm; |
Vegard Nossum | f1f1007 | 2017-02-27 14:30:07 -0800 | [diff] [blame] | 936 | mmgrab(mm); |
Konstantin Khlebnikov | 8e675f7 | 2017-07-06 15:40:28 -0700 | [diff] [blame] | 937 | |
| 938 | /* Raise event before sending signal: task reaper must see this */ |
| 939 | count_vm_event(OOM_KILL); |
Roman Gushchin | fe6bdfc | 2018-06-14 15:28:05 -0700 | [diff] [blame] | 940 | memcg_memory_event_mm(mm, MEMCG_OOM_KILL); |
Konstantin Khlebnikov | 8e675f7 | 2017-07-06 15:40:28 -0700 | [diff] [blame] | 941 | |
Tetsuo Handa | 426fb5e | 2015-11-05 18:47:44 -0800 | [diff] [blame] | 942 | /* |
Michal Hocko | cd04ae1 | 2017-09-06 16:24:50 -0700 | [diff] [blame] | 943 | * 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 Handa | 426fb5e | 2015-11-05 18:47:44 -0800 | [diff] [blame] | 946 | */ |
Eric W. Biederman | 079b22d | 2018-09-03 10:32:52 +0200 | [diff] [blame] | 947 | do_send_sig_info(SIGKILL, SEND_SIG_PRIV, victim, PIDTYPE_TGID); |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 948 | mark_oom_victim(victim); |
Edward Chron | 70cb6d2 | 2019-09-23 15:37:11 -0700 | [diff] [blame] | 949 | 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 Dryomov | 941f762 | 2020-01-04 13:00:09 -0800 | [diff] [blame] | 955 | mm_pgtables_bytes(mm) >> 10, victim->signal->oom_score_adj); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 956 | 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 Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 961 | * depletion of all memory. This prevents mm->mmap_lock livelock when an |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 962 | * 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 Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 967 | rcu_read_lock(); |
Oleg Nesterov | c319025 | 2015-11-05 18:48:23 -0800 | [diff] [blame] | 968 | for_each_process(p) { |
Oleg Nesterov | 4d7b339 | 2015-11-05 18:48:26 -0800 | [diff] [blame] | 969 | if (!process_shares_mm(p, mm)) |
Oleg Nesterov | c319025 | 2015-11-05 18:48:23 -0800 | [diff] [blame] | 970 | continue; |
| 971 | if (same_thread_group(p, victim)) |
| 972 | continue; |
Michal Hocko | 1b51e65 | 2016-10-07 16:59:09 -0700 | [diff] [blame] | 973 | if (is_global_init(p)) { |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 974 | can_oom_reap = false; |
Michal Hocko | 862e307 | 2016-10-07 16:58:57 -0700 | [diff] [blame] | 975 | set_bit(MMF_OOM_SKIP, &mm->flags); |
Michal Hocko | a373966 | 2016-07-28 15:45:01 -0700 | [diff] [blame] | 976 | 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 Nesterov | c319025 | 2015-11-05 18:48:23 -0800 | [diff] [blame] | 979 | continue; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 980 | } |
Michal Hocko | 1b51e65 | 2016-10-07 16:59:09 -0700 | [diff] [blame] | 981 | /* |
Rolf Eike Beer | 4c9c380 | 2021-03-17 10:24:51 +0100 | [diff] [blame] | 982 | * No kthread_use_mm() user needs to read from the userspace so |
Christoph Hellwig | f5678e7 | 2020-06-10 18:42:06 -0700 | [diff] [blame] | 983 | * we are ok to reap it. |
Michal Hocko | 1b51e65 | 2016-10-07 16:59:09 -0700 | [diff] [blame] | 984 | */ |
| 985 | if (unlikely(p->flags & PF_KTHREAD)) |
| 986 | continue; |
Eric W. Biederman | 079b22d | 2018-09-03 10:32:52 +0200 | [diff] [blame] | 987 | do_send_sig_info(SIGKILL, SEND_SIG_PRIV, p, PIDTYPE_TGID); |
Oleg Nesterov | c319025 | 2015-11-05 18:48:23 -0800 | [diff] [blame] | 988 | } |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 989 | rcu_read_unlock(); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 990 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 991 | if (can_oom_reap) |
Nico Pache | e4a3840 | 2022-04-21 16:36:01 -0700 | [diff] [blame] | 992 | queue_oom_reaper(victim); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 993 | |
Tetsuo Handa | 880b7689 | 2015-11-05 18:47:51 -0800 | [diff] [blame] | 994 | mmdrop(mm); |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 995 | put_task_struct(victim); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 996 | } |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 997 | #undef K |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 998 | |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 999 | /* |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 1000 | * Kill provided task unless it's secured by setting |
| 1001 | * oom_score_adj to OOM_SCORE_ADJ_MIN. |
| 1002 | */ |
Shakeel Butt | bbbe480 | 2019-03-05 15:46:12 -0800 | [diff] [blame] | 1003 | static int oom_kill_memcg_member(struct task_struct *task, void *message) |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 1004 | { |
Tetsuo Handa | d342a0b | 2019-03-05 15:48:22 -0800 | [diff] [blame] | 1005 | if (task->signal->oom_score_adj != OOM_SCORE_ADJ_MIN && |
| 1006 | !is_global_init(task)) { |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 1007 | get_task_struct(task); |
Shakeel Butt | bbbe480 | 2019-03-05 15:46:12 -0800 | [diff] [blame] | 1008 | __oom_kill_process(task, message); |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 1009 | } |
| 1010 | return 0; |
| 1011 | } |
| 1012 | |
Roman Gushchin | 5989ad7 | 2018-08-21 21:53:50 -0700 | [diff] [blame] | 1013 | static void oom_kill_process(struct oom_control *oc, const char *message) |
| 1014 | { |
Shakeel Butt | bbbe480 | 2019-03-05 15:46:12 -0800 | [diff] [blame] | 1015 | struct task_struct *victim = oc->chosen; |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 1016 | struct mem_cgroup *oom_group; |
Roman Gushchin | 5989ad7 | 2018-08-21 21:53:50 -0700 | [diff] [blame] | 1017 | 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 Butt | bbbe480 | 2019-03-05 15:46:12 -0800 | [diff] [blame] | 1025 | task_lock(victim); |
| 1026 | if (task_will_free_mem(victim)) { |
| 1027 | mark_oom_victim(victim); |
Nico Pache | e4a3840 | 2022-04-21 16:36:01 -0700 | [diff] [blame] | 1028 | queue_oom_reaper(victim); |
Shakeel Butt | bbbe480 | 2019-03-05 15:46:12 -0800 | [diff] [blame] | 1029 | task_unlock(victim); |
| 1030 | put_task_struct(victim); |
Roman Gushchin | 5989ad7 | 2018-08-21 21:53:50 -0700 | [diff] [blame] | 1031 | return; |
| 1032 | } |
Shakeel Butt | bbbe480 | 2019-03-05 15:46:12 -0800 | [diff] [blame] | 1033 | task_unlock(victim); |
Roman Gushchin | 5989ad7 | 2018-08-21 21:53:50 -0700 | [diff] [blame] | 1034 | |
| 1035 | if (__ratelimit(&oom_rs)) |
Shakeel Butt | bbbe480 | 2019-03-05 15:46:12 -0800 | [diff] [blame] | 1036 | dump_header(oc, victim); |
Roman Gushchin | 5989ad7 | 2018-08-21 21:53:50 -0700 | [diff] [blame] | 1037 | |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 1038 | /* |
| 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 Butt | bbbe480 | 2019-03-05 15:46:12 -0800 | [diff] [blame] | 1045 | __oom_kill_process(victim, message); |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 1046 | |
| 1047 | /* |
| 1048 | * If necessary, kill all tasks in the selected memory cgroup. |
| 1049 | */ |
| 1050 | if (oom_group) { |
Dan Schatzberg | b6bf9ab | 2022-01-14 14:05:35 -0800 | [diff] [blame] | 1051 | memcg_memory_event(oom_group, MEMCG_OOM_GROUP_KILL); |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 1052 | mem_cgroup_print_oom_group(oom_group); |
Shakeel Butt | bbbe480 | 2019-03-05 15:46:12 -0800 | [diff] [blame] | 1053 | mem_cgroup_scan_tasks(oom_group, oom_kill_memcg_member, |
Zhiyuan Dai | 68d68ff | 2021-05-04 18:40:12 -0700 | [diff] [blame] | 1054 | (void *)message); |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 1055 | mem_cgroup_put(oom_group); |
| 1056 | } |
Roman Gushchin | 5989ad7 | 2018-08-21 21:53:50 -0700 | [diff] [blame] | 1057 | } |
| 1058 | |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 1059 | /* |
| 1060 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. |
| 1061 | */ |
Yafang Shao | 432b1de | 2019-06-28 12:06:59 -0700 | [diff] [blame] | 1062 | static void check_panic_on_oom(struct oom_control *oc) |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 1063 | { |
| 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 Shao | 432b1de | 2019-06-28 12:06:59 -0700 | [diff] [blame] | 1072 | if (oc->constraint != CONSTRAINT_NONE) |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 1073 | return; |
| 1074 | } |
David Rientjes | 071a4be | 2015-09-08 15:00:42 -0700 | [diff] [blame] | 1075 | /* Do not panic for oom kills triggered by sysrq */ |
Yaowei Bai | db2a0dd | 2015-11-06 16:28:06 -0800 | [diff] [blame] | 1076 | if (is_sysrq_oom(oc)) |
David Rientjes | 071a4be | 2015-09-08 15:00:42 -0700 | [diff] [blame] | 1077 | return; |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 1078 | dump_header(oc, NULL); |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 1079 | panic("Out of memory: %s panic_on_oom is enabled\n", |
| 1080 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); |
| 1081 | } |
| 1082 | |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 1083 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
| 1084 | |
| 1085 | int register_oom_notifier(struct notifier_block *nb) |
| 1086 | { |
| 1087 | return blocking_notifier_chain_register(&oom_notify_list, nb); |
| 1088 | } |
| 1089 | EXPORT_SYMBOL_GPL(register_oom_notifier); |
| 1090 | |
| 1091 | int unregister_oom_notifier(struct notifier_block *nb) |
| 1092 | { |
| 1093 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); |
| 1094 | } |
| 1095 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); |
| 1096 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1097 | /** |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1098 | * out_of_memory - kill the "best" process when we run out of memory |
| 1099 | * @oc: pointer to struct oom_control |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1100 | * |
| 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 Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1106 | bool out_of_memory(struct oom_control *oc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1107 | { |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 1108 | unsigned long freed = 0; |
| 1109 | |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1110 | if (oom_killer_disabled) |
| 1111 | return false; |
| 1112 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1113 | if (!is_memcg_oom(oc)) { |
| 1114 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); |
Jann Horn | f530243 | 2022-01-14 14:08:27 -0800 | [diff] [blame] | 1115 | if (freed > 0 && !is_sysrq_oom(oc)) |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1116 | /* Got some memory back in the last second. */ |
| 1117 | return true; |
| 1118 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1119 | |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 1120 | /* |
David Rientjes | 9ff4868 | 2012-12-11 16:01:30 -0800 | [diff] [blame] | 1121 | * 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 Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 1124 | */ |
Michal Hocko | 091f362 | 2016-07-28 15:45:04 -0700 | [diff] [blame] | 1125 | if (task_will_free_mem(current)) { |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 1126 | mark_oom_victim(current); |
Nico Pache | e4a3840 | 2022-04-21 16:36:01 -0700 | [diff] [blame] | 1127 | queue_oom_reaper(current); |
David Rientjes | 75e8f8b | 2015-09-08 15:00:47 -0700 | [diff] [blame] | 1128 | return true; |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 1129 | } |
| 1130 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 1131 | /* |
Michal Hocko | 3da88fb3 | 2016-05-19 17:13:09 -0700 | [diff] [blame] | 1132 | * 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 Handa | f9c6456 | 2019-09-23 15:37:08 -0700 | [diff] [blame] | 1135 | * ___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 Hocko | 3da88fb3 | 2016-05-19 17:13:09 -0700 | [diff] [blame] | 1137 | */ |
Tetsuo Handa | f9c6456 | 2019-09-23 15:37:08 -0700 | [diff] [blame] | 1138 | if (oc->gfp_mask && !(oc->gfp_mask & __GFP_FS) && !is_memcg_oom(oc)) |
Michal Hocko | 3da88fb3 | 2016-05-19 17:13:09 -0700 | [diff] [blame] | 1139 | return true; |
| 1140 | |
| 1141 | /* |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 1142 | * Check if there were limitations on the allocation (only relevant for |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1143 | * NUMA and memcg) that may require different handling. |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 1144 | */ |
Yafang Shao | 432b1de | 2019-06-28 12:06:59 -0700 | [diff] [blame] | 1145 | oc->constraint = constrained_alloc(oc); |
| 1146 | if (oc->constraint != CONSTRAINT_MEMORY_POLICY) |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1147 | oc->nodemask = NULL; |
Yafang Shao | 432b1de | 2019-06-28 12:06:59 -0700 | [diff] [blame] | 1148 | check_panic_on_oom(oc); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 1149 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1150 | if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task && |
Shakeel Butt | ac311a1 | 2019-07-11 21:00:31 -0700 | [diff] [blame] | 1151 | current->mm && !oom_unkillable_task(current) && |
| 1152 | oom_cpuset_eligible(current, oc) && |
David Rientjes | 121d1ba | 2012-07-31 16:42:55 -0700 | [diff] [blame] | 1153 | current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) { |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 1154 | get_task_struct(current); |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1155 | oc->chosen = current; |
| 1156 | oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)"); |
David Rientjes | 75e8f8b | 2015-09-08 15:00:47 -0700 | [diff] [blame] | 1157 | return true; |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 1158 | } |
| 1159 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1160 | select_bad_process(oc); |
Johannes Weiner | 3100dab | 2018-09-04 15:45:34 -0700 | [diff] [blame] | 1161 | /* Found nothing?!?! */ |
| 1162 | if (!oc->chosen) { |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 1163 | dump_header(oc, NULL); |
Johannes Weiner | 3100dab | 2018-09-04 15:45:34 -0700 | [diff] [blame] | 1164 | 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 Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 1172 | } |
Michal Hocko | 9bfe5de | 2018-08-17 15:49:04 -0700 | [diff] [blame] | 1173 | if (oc->chosen && oc->chosen != (void *)-1UL) |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1174 | oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" : |
| 1175 | "Memory cgroup out of memory"); |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1176 | return !!oc->chosen; |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 1177 | } |
| 1178 | |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 1179 | /* |
Michal Hocko | 60e2793 | 2021-11-05 13:38:06 -0700 | [diff] [blame] | 1180 | * 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 Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 1184 | */ |
| 1185 | void pagefault_out_of_memory(void) |
| 1186 | { |
Michal Hocko | 60e2793 | 2021-11-05 13:38:06 -0700 | [diff] [blame] | 1187 | static DEFINE_RATELIMIT_STATE(pfoom_rs, DEFAULT_RATELIMIT_INTERVAL, |
| 1188 | DEFAULT_RATELIMIT_BURST); |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1189 | |
Johannes Weiner | 49426420 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1190 | if (mem_cgroup_oom_synchronize(true)) |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1191 | return; |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1192 | |
Vasily Averin | 0b28179 | 2021-11-05 13:38:02 -0700 | [diff] [blame] | 1193 | if (fatal_signal_pending(current)) |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1194 | return; |
Vasily Averin | 0b28179 | 2021-11-05 13:38:02 -0700 | [diff] [blame] | 1195 | |
Michal Hocko | 60e2793 | 2021-11-05 13:38:06 -0700 | [diff] [blame] | 1196 | if (__ratelimit(&pfoom_rs)) |
| 1197 | pr_warn("Huh VM_FAULT_OOM leaked out to the #PF handler. Retrying PF\n"); |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 1198 | } |
Suren Baghdasaryan | 884a7e5 | 2021-09-02 15:00:29 -0700 | [diff] [blame] | 1199 | |
| 1200 | SYSCALL_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 Baghdasaryan | 337546e | 2021-10-28 14:36:14 -0700 | [diff] [blame] | 1207 | bool reap = false; |
Suren Baghdasaryan | 884a7e5 | 2021-09-02 15:00:29 -0700 | [diff] [blame] | 1208 | long ret = 0; |
| 1209 | |
| 1210 | if (flags) |
| 1211 | return -EINVAL; |
| 1212 | |
Christian Brauner | ee9955d | 2021-10-11 15:32:45 +0200 | [diff] [blame] | 1213 | task = pidfd_get_task(pidfd, &f_flags); |
| 1214 | if (IS_ERR(task)) |
| 1215 | return PTR_ERR(task); |
Suren Baghdasaryan | 884a7e5 | 2021-09-02 15:00:29 -0700 | [diff] [blame] | 1216 | |
| 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 Baghdasaryan | ba535c1 | 2022-01-14 14:06:22 -0800 | [diff] [blame] | 1227 | 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 Baghdasaryan | 884a7e5 | 2021-09-02 15:00:29 -0700 | [diff] [blame] | 1236 | } |
| 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 Baghdasaryan | ba535c1 | 2022-01-14 14:06:22 -0800 | [diff] [blame] | 1246 | /* |
| 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 Baghdasaryan | 884a7e5 | 2021-09-02 15:00:29 -0700 | [diff] [blame] | 1251 | ret = -EAGAIN; |
| 1252 | mmap_read_unlock(mm); |
| 1253 | |
| 1254 | drop_mm: |
Suren Baghdasaryan | ba535c1 | 2022-01-14 14:06:22 -0800 | [diff] [blame] | 1255 | mmdrop(mm); |
Suren Baghdasaryan | 884a7e5 | 2021-09-02 15:00:29 -0700 | [diff] [blame] | 1256 | put_task: |
| 1257 | put_task_struct(task); |
Suren Baghdasaryan | 884a7e5 | 2021-09-02 15:00:29 -0700 | [diff] [blame] | 1258 | return ret; |
| 1259 | #else |
| 1260 | return -ENOSYS; |
| 1261 | #endif /* CONFIG_MMU */ |
| 1262 | } |