Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/mm/oom_kill.c |
| 3 | * |
| 4 | * Copyright (C) 1998,2000 Rik van Riel |
| 5 | * Thanks go out to Claus Fischer for some serious inspiration and |
| 6 | * for goading me into coding this file... |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 7 | * Copyright (C) 2010 Google, Inc. |
| 8 | * Rewritten by David Rientjes |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 9 | * |
| 10 | * The routines in this file are used to kill a process when |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 11 | * we're seriously out of memory. This gets called from __alloc_pages() |
| 12 | * in mm/page_alloc.c when we really run out of memory. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | * |
| 14 | * Since we won't call these routines often (on a well-configured |
| 15 | * machine) this file will double as a 'coding guide' and a signpost |
| 16 | * for newbie kernel hackers. It features several pointers to major |
| 17 | * kernel subsystems and hints as to where to find out what things do. |
| 18 | */ |
| 19 | |
Alexey Dobriyan | 8ac773b | 2006-10-19 23:28:32 -0700 | [diff] [blame] | 20 | #include <linux/oom.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 21 | #include <linux/mm.h> |
Alexey Dobriyan | 4e950f6 | 2007-07-30 02:36:13 +0400 | [diff] [blame] | 22 | #include <linux/err.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 23 | #include <linux/gfp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include <linux/sched.h> |
Ingo Molnar | 6e84f31 | 2017-02-08 18:51:29 +0100 | [diff] [blame] | 25 | #include <linux/sched/mm.h> |
Ingo Molnar | f7ccbae | 2017-02-08 18:51:30 +0100 | [diff] [blame] | 26 | #include <linux/sched/coredump.h> |
Ingo Molnar | 2993002 | 2017-02-08 18:51:36 +0100 | [diff] [blame] | 27 | #include <linux/sched/task.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | #include <linux/swap.h> |
| 29 | #include <linux/timex.h> |
| 30 | #include <linux/jiffies.h> |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 31 | #include <linux/cpuset.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 32 | #include <linux/export.h> |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 33 | #include <linux/notifier.h> |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 34 | #include <linux/memcontrol.h> |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 35 | #include <linux/mempolicy.h> |
David Howells | 5cd9c58 | 2008-08-14 11:37:28 +0100 | [diff] [blame] | 36 | #include <linux/security.h> |
David Rientjes | edd4554 | 2011-03-22 16:30:12 -0700 | [diff] [blame] | 37 | #include <linux/ptrace.h> |
David Rientjes | f660daa | 2011-10-31 17:07:07 -0700 | [diff] [blame] | 38 | #include <linux/freezer.h> |
KAMEZAWA Hiroyuki | 43d2b11 | 2012-01-10 15:08:09 -0800 | [diff] [blame] | 39 | #include <linux/ftrace.h> |
David Rientjes | dc3f21e | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 40 | #include <linux/ratelimit.h> |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 41 | #include <linux/kthread.h> |
| 42 | #include <linux/init.h> |
Michal Hocko | 4d4bbd8 | 2017-10-03 16:14:50 -0700 | [diff] [blame] | 43 | #include <linux/mmu_notifier.h> |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 44 | |
| 45 | #include <asm/tlb.h> |
| 46 | #include "internal.h" |
Yang Shi | 852d8be | 2017-11-15 17:32:07 -0800 | [diff] [blame] | 47 | #include "slab.h" |
KAMEZAWA Hiroyuki | 43d2b11 | 2012-01-10 15:08:09 -0800 | [diff] [blame] | 48 | |
| 49 | #define CREATE_TRACE_POINTS |
| 50 | #include <trace/events/oom.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 51 | |
KAMEZAWA Hiroyuki | fadd8fb | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 52 | int sysctl_panic_on_oom; |
David Rientjes | fe071d7 | 2007-10-16 23:25:56 -0700 | [diff] [blame] | 53 | int sysctl_oom_kill_allocating_task; |
David Rientjes | ad915c4 | 2010-08-09 17:18:53 -0700 | [diff] [blame] | 54 | int sysctl_oom_dump_tasks = 1; |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 55 | |
| 56 | DEFINE_MUTEX(oom_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 57 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 58 | #ifdef CONFIG_NUMA |
| 59 | /** |
| 60 | * has_intersects_mems_allowed() - check task eligiblity for kill |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 61 | * @start: task struct of which task to consider |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 62 | * @mask: nodemask passed to page allocator for mempolicy ooms |
| 63 | * |
| 64 | * Task eligibility is determined by whether or not a candidate task, @tsk, |
| 65 | * shares the same mempolicy nodes as current if it is bound by such a policy |
| 66 | * and whether or not it has the same set of allowed cpuset nodes. |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 67 | */ |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 68 | static bool has_intersects_mems_allowed(struct task_struct *start, |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 69 | const nodemask_t *mask) |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 70 | { |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 71 | struct task_struct *tsk; |
| 72 | bool ret = false; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 73 | |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 74 | rcu_read_lock(); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 75 | for_each_thread(start, tsk) { |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 76 | if (mask) { |
| 77 | /* |
| 78 | * If this is a mempolicy constrained oom, tsk's |
| 79 | * cpuset is irrelevant. Only return true if its |
| 80 | * mempolicy intersects current, otherwise it may be |
| 81 | * needlessly killed. |
| 82 | */ |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 83 | ret = mempolicy_nodemask_intersects(tsk, mask); |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 84 | } else { |
| 85 | /* |
| 86 | * This is not a mempolicy constrained oom, so only |
| 87 | * check the mems of tsk's cpuset. |
| 88 | */ |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 89 | ret = cpuset_mems_allowed_intersects(current, tsk); |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 90 | } |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 91 | if (ret) |
| 92 | break; |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 93 | } |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 94 | rcu_read_unlock(); |
KOSAKI Motohiro | df1090a | 2010-08-09 17:19:39 -0700 | [diff] [blame] | 95 | |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 96 | return ret; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 97 | } |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 98 | #else |
| 99 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
| 100 | const nodemask_t *mask) |
| 101 | { |
| 102 | return true; |
| 103 | } |
| 104 | #endif /* CONFIG_NUMA */ |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 105 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 106 | /* |
| 107 | * The process p may have detached its own ->mm while exiting or through |
| 108 | * use_mm(), but one or more of its subthreads may still have a valid |
| 109 | * pointer. Return p, or any of its subthreads with a valid ->mm, with |
| 110 | * task_lock() held. |
| 111 | */ |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 112 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 113 | { |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 114 | struct task_struct *t; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 115 | |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 116 | rcu_read_lock(); |
| 117 | |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 118 | for_each_thread(p, t) { |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 119 | task_lock(t); |
| 120 | if (likely(t->mm)) |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 121 | goto found; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 122 | task_unlock(t); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 123 | } |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 124 | t = NULL; |
| 125 | found: |
| 126 | rcu_read_unlock(); |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 127 | |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 128 | return t; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 129 | } |
| 130 | |
Yaowei Bai | db2a0dd | 2015-11-06 16:28:06 -0800 | [diff] [blame] | 131 | /* |
| 132 | * order == -1 means the oom kill is required by sysrq, otherwise only |
| 133 | * for display purposes. |
| 134 | */ |
| 135 | static inline bool is_sysrq_oom(struct oom_control *oc) |
| 136 | { |
| 137 | return oc->order == -1; |
| 138 | } |
| 139 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 140 | static inline bool is_memcg_oom(struct oom_control *oc) |
| 141 | { |
| 142 | return oc->memcg != NULL; |
| 143 | } |
| 144 | |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 145 | /* return true if the task is not adequate as candidate victim task. */ |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 146 | static bool oom_unkillable_task(struct task_struct *p, |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 147 | struct mem_cgroup *memcg, const nodemask_t *nodemask) |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 148 | { |
| 149 | if (is_global_init(p)) |
| 150 | return true; |
| 151 | if (p->flags & PF_KTHREAD) |
| 152 | return true; |
| 153 | |
| 154 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 155 | if (memcg && !task_in_mem_cgroup(p, memcg)) |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 156 | return true; |
| 157 | |
| 158 | /* p may not have freeable memory in nodemask */ |
| 159 | if (!has_intersects_mems_allowed(p, nodemask)) |
| 160 | return true; |
| 161 | |
| 162 | return false; |
| 163 | } |
| 164 | |
Yang Shi | 852d8be | 2017-11-15 17:32:07 -0800 | [diff] [blame] | 165 | /* |
| 166 | * Print out unreclaimble slabs info when unreclaimable slabs amount is greater |
| 167 | * than all user memory (LRU pages) |
| 168 | */ |
| 169 | static bool is_dump_unreclaim_slabs(void) |
| 170 | { |
| 171 | unsigned long nr_lru; |
| 172 | |
| 173 | nr_lru = global_node_page_state(NR_ACTIVE_ANON) + |
| 174 | global_node_page_state(NR_INACTIVE_ANON) + |
| 175 | global_node_page_state(NR_ACTIVE_FILE) + |
| 176 | global_node_page_state(NR_INACTIVE_FILE) + |
| 177 | global_node_page_state(NR_ISOLATED_ANON) + |
| 178 | global_node_page_state(NR_ISOLATED_FILE) + |
| 179 | global_node_page_state(NR_UNEVICTABLE); |
| 180 | |
| 181 | return (global_node_page_state(NR_SLAB_UNRECLAIMABLE) > nr_lru); |
| 182 | } |
| 183 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | /** |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 185 | * oom_badness - heuristic function to determine which candidate task to kill |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | * @p: task struct of which task we should calculate |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 187 | * @totalpages: total present RAM allowed for page allocation |
Mike Rapoport | e8b098f | 2018-04-05 16:24:57 -0700 | [diff] [blame] | 188 | * @memcg: task's memory controller, if constrained |
| 189 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | * |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 191 | * The heuristic for determining which task to kill is made to be as simple and |
| 192 | * predictable as possible. The goal is to return the highest value for the |
| 193 | * task consuming the most memory to avoid subsequent oom failures. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | */ |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 195 | unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg, |
| 196 | const nodemask_t *nodemask, unsigned long totalpages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 197 | { |
David Rientjes | 1e11ad8 | 2012-06-08 13:21:26 -0700 | [diff] [blame] | 198 | long points; |
David Rientjes | 61eafb0 | 2012-06-20 12:52:58 -0700 | [diff] [blame] | 199 | long adj; |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 200 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 201 | if (oom_unkillable_task(p, memcg, nodemask)) |
KOSAKI Motohiro | 26ebc98 | 2010-08-09 17:19:37 -0700 | [diff] [blame] | 202 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 203 | |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 204 | p = find_lock_task_mm(p); |
| 205 | if (!p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 206 | return 0; |
| 207 | |
Michal Hocko | bb8a4b7 | 2016-05-20 16:57:18 -0700 | [diff] [blame] | 208 | /* |
| 209 | * Do not even consider tasks which are explicitly marked oom |
Michal Hocko | b18dc5f | 2016-07-28 15:44:46 -0700 | [diff] [blame] | 210 | * unkillable or have been already oom reaped or the are in |
| 211 | * the middle of vfork |
Michal Hocko | bb8a4b7 | 2016-05-20 16:57:18 -0700 | [diff] [blame] | 212 | */ |
David Rientjes | a9c58b90 | 2012-12-11 16:02:54 -0800 | [diff] [blame] | 213 | adj = (long)p->signal->oom_score_adj; |
Michal Hocko | bb8a4b7 | 2016-05-20 16:57:18 -0700 | [diff] [blame] | 214 | if (adj == OOM_SCORE_ADJ_MIN || |
Michal Hocko | 862e307 | 2016-10-07 16:58:57 -0700 | [diff] [blame] | 215 | test_bit(MMF_OOM_SKIP, &p->mm->flags) || |
Michal Hocko | b18dc5f | 2016-07-28 15:44:46 -0700 | [diff] [blame] | 216 | in_vfork(p)) { |
Michal Hocko | 5aecc85 | 2011-11-15 14:36:07 -0800 | [diff] [blame] | 217 | task_unlock(p); |
| 218 | return 0; |
| 219 | } |
| 220 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 221 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 222 | * 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] | 223 | * task's rss, pagetable and swap space use. |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 224 | */ |
Kirill A. Shutemov | dc6c9a3 | 2015-02-11 15:26:50 -0800 | [diff] [blame] | 225 | 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] | 226 | mm_pgtables_bytes(p->mm) / PAGE_SIZE; |
Andrew Morton | 97c2c9b8 | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 227 | task_unlock(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 228 | |
David Rientjes | 61eafb0 | 2012-06-20 12:52:58 -0700 | [diff] [blame] | 229 | /* Normalize to oom_score_adj units */ |
| 230 | adj *= totalpages / 1000; |
| 231 | points += adj; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 232 | |
David Rientjes | f19e8aa | 2010-09-22 13:04:52 -0700 | [diff] [blame] | 233 | /* |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 234 | * Never return 0 for an eligible task regardless of the root bonus and |
| 235 | * oom_score_adj (oom_score_adj can't be OOM_SCORE_ADJ_MIN here). |
David Rientjes | f19e8aa | 2010-09-22 13:04:52 -0700 | [diff] [blame] | 236 | */ |
David Rientjes | 1e11ad8 | 2012-06-08 13:21:26 -0700 | [diff] [blame] | 237 | return points > 0 ? points : 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 238 | } |
| 239 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 240 | enum oom_constraint { |
| 241 | CONSTRAINT_NONE, |
| 242 | CONSTRAINT_CPUSET, |
| 243 | CONSTRAINT_MEMORY_POLICY, |
| 244 | CONSTRAINT_MEMCG, |
| 245 | }; |
| 246 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 247 | /* |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 248 | * Determine the type of allocation constraint. |
| 249 | */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 250 | static enum oom_constraint constrained_alloc(struct oom_control *oc) |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 251 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 252 | struct zone *zone; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 253 | struct zoneref *z; |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 254 | enum zone_type high_zoneidx = gfp_zone(oc->gfp_mask); |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 255 | bool cpuset_limited = false; |
| 256 | int nid; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 257 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 258 | if (is_memcg_oom(oc)) { |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 259 | oc->totalpages = mem_cgroup_get_max(oc->memcg) ?: 1; |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 260 | return CONSTRAINT_MEMCG; |
| 261 | } |
| 262 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 263 | /* Default to all available memory */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 264 | oc->totalpages = totalram_pages + total_swap_pages; |
| 265 | |
| 266 | if (!IS_ENABLED(CONFIG_NUMA)) |
| 267 | return CONSTRAINT_NONE; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 268 | |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 269 | if (!oc->zonelist) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 270 | return CONSTRAINT_NONE; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 271 | /* |
| 272 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid |
| 273 | * to kill current.We have to random task kill in this case. |
| 274 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. |
| 275 | */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 276 | if (oc->gfp_mask & __GFP_THISNODE) |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 277 | return CONSTRAINT_NONE; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 278 | |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 279 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 280 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
| 281 | * the page allocator means a mempolicy is in effect. Cpuset policy |
| 282 | * is enforced in get_page_from_freelist(). |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 283 | */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 284 | if (oc->nodemask && |
| 285 | !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) { |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 286 | oc->totalpages = total_swap_pages; |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 287 | for_each_node_mask(nid, *oc->nodemask) |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 288 | oc->totalpages += node_spanned_pages(nid); |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 289 | return CONSTRAINT_MEMORY_POLICY; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 290 | } |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 291 | |
| 292 | /* Check this allocation failure is caused by cpuset's wall function */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 293 | for_each_zone_zonelist_nodemask(zone, z, oc->zonelist, |
| 294 | high_zoneidx, oc->nodemask) |
| 295 | if (!cpuset_zone_allowed(zone, oc->gfp_mask)) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 296 | cpuset_limited = true; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 297 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 298 | if (cpuset_limited) { |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 299 | oc->totalpages = total_swap_pages; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 300 | for_each_node_mask(nid, cpuset_current_mems_allowed) |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 301 | oc->totalpages += node_spanned_pages(nid); |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 302 | return CONSTRAINT_CPUSET; |
| 303 | } |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 304 | return CONSTRAINT_NONE; |
| 305 | } |
| 306 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 307 | static int oom_evaluate_task(struct task_struct *task, void *arg) |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 308 | { |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 309 | struct oom_control *oc = arg; |
| 310 | unsigned long points; |
| 311 | |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 312 | if (oom_unkillable_task(task, NULL, oc->nodemask)) |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 313 | goto next; |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 314 | |
| 315 | /* |
| 316 | * This task already has access to memory reserves and is being killed. |
Michal Hocko | a373966 | 2016-07-28 15:45:01 -0700 | [diff] [blame] | 317 | * 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] | 318 | * the task has MMF_OOM_SKIP because chances that it would release |
Michal Hocko | a373966 | 2016-07-28 15:45:01 -0700 | [diff] [blame] | 319 | * any memory is quite low. |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 320 | */ |
Michal Hocko | 862e307 | 2016-10-07 16:58:57 -0700 | [diff] [blame] | 321 | if (!is_sysrq_oom(oc) && tsk_is_oom_victim(task)) { |
| 322 | if (test_bit(MMF_OOM_SKIP, &task->signal->oom_mm->flags)) |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 323 | goto next; |
| 324 | goto abort; |
Michal Hocko | a373966 | 2016-07-28 15:45:01 -0700 | [diff] [blame] | 325 | } |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 326 | |
David Rientjes | e1e12d2 | 2012-12-11 16:02:56 -0800 | [diff] [blame] | 327 | /* |
| 328 | * If task is allocating a lot of memory and has been marked to be |
| 329 | * killed first if it triggers an oom, then select it. |
| 330 | */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 331 | if (oom_task_origin(task)) { |
| 332 | points = ULONG_MAX; |
| 333 | goto select; |
| 334 | } |
David Rientjes | e1e12d2 | 2012-12-11 16:02:56 -0800 | [diff] [blame] | 335 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 336 | points = oom_badness(task, NULL, oc->nodemask, oc->totalpages); |
| 337 | if (!points || points < oc->chosen_points) |
| 338 | goto next; |
| 339 | |
| 340 | /* Prefer thread group leaders for display purposes */ |
| 341 | if (points == oc->chosen_points && thread_group_leader(oc->chosen)) |
| 342 | goto next; |
| 343 | select: |
| 344 | if (oc->chosen) |
| 345 | put_task_struct(oc->chosen); |
| 346 | get_task_struct(task); |
| 347 | oc->chosen = task; |
| 348 | oc->chosen_points = points; |
| 349 | next: |
| 350 | return 0; |
| 351 | abort: |
| 352 | if (oc->chosen) |
| 353 | put_task_struct(oc->chosen); |
| 354 | oc->chosen = (void *)-1UL; |
| 355 | return 1; |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 356 | } |
| 357 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 358 | /* |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 359 | * Simple selection loop. We choose the process with the highest number of |
| 360 | * 'points'. In case scan was aborted, oc->chosen is set to -1. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 361 | */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 362 | static void select_bad_process(struct oom_control *oc) |
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 | if (is_memcg_oom(oc)) |
| 365 | mem_cgroup_scan_tasks(oc->memcg, oom_evaluate_task, oc); |
| 366 | else { |
| 367 | struct task_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 369 | rcu_read_lock(); |
| 370 | for_each_process(p) |
| 371 | if (oom_evaluate_task(p, oc)) |
| 372 | break; |
| 373 | rcu_read_unlock(); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 374 | } |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 375 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 376 | oc->chosen_points = oc->chosen_points * 1000 / oc->totalpages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | } |
| 378 | |
| 379 | /** |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 380 | * dump_tasks - dump current memory state of all system tasks |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 381 | * @memcg: current's memory controller, if constrained |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 382 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 383 | * |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 384 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
| 385 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes |
| 386 | * are not shown. |
Kirill A. Shutemov | af5b0f6 | 2017-11-15 17:35:40 -0800 | [diff] [blame] | 387 | * State information includes task's pid, uid, tgid, vm size, rss, |
| 388 | * pgtables_bytes, swapents, oom_score_adj value, and name. |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 389 | */ |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 390 | static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask) |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 391 | { |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 392 | struct task_struct *p; |
| 393 | struct task_struct *task; |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 394 | |
Kirill A. Shutemov | af5b0f6 | 2017-11-15 17:35:40 -0800 | [diff] [blame] | 395 | pr_info("[ pid ] uid tgid total_vm rss pgtables_bytes swapents oom_score_adj name\n"); |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 396 | rcu_read_lock(); |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 397 | for_each_process(p) { |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 398 | if (oom_unkillable_task(p, memcg, nodemask)) |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 399 | continue; |
| 400 | |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 401 | task = find_lock_task_mm(p); |
| 402 | if (!task) { |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 403 | /* |
David Rientjes | 74ab7f1 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 404 | * This is a kthread or all of p's threads have already |
| 405 | * detached their mm's. There's no need to report |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 406 | * them; they can't be oom killed anyway. |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 407 | */ |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 408 | continue; |
| 409 | } |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 410 | |
Kirill A. Shutemov | af5b0f6 | 2017-11-15 17:35:40 -0800 | [diff] [blame] | 411 | pr_info("[%5d] %5d %5d %8lu %8lu %8ld %8lu %5hd %s\n", |
Eric W. Biederman | 078de5f | 2012-02-08 07:00:08 -0800 | [diff] [blame] | 412 | task->pid, from_kuid(&init_user_ns, task_uid(task)), |
| 413 | task->tgid, task->mm->total_vm, get_mm_rss(task->mm), |
Kirill A. Shutemov | af5b0f6 | 2017-11-15 17:35:40 -0800 | [diff] [blame] | 414 | mm_pgtables_bytes(task->mm), |
David Rientjes | de34d96 | 2012-07-31 16:42:56 -0700 | [diff] [blame] | 415 | get_mm_counter(task->mm, MM_SWAPENTS), |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 416 | task->signal->oom_score_adj, task->comm); |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 417 | task_unlock(task); |
| 418 | } |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 419 | rcu_read_unlock(); |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 420 | } |
| 421 | |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 422 | static void dump_header(struct oom_control *oc, struct task_struct *p) |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 423 | { |
Michal Hocko | 0205f75 | 2017-11-15 17:39:14 -0800 | [diff] [blame] | 424 | pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), nodemask=%*pbl, order=%d, oom_score_adj=%hd\n", |
| 425 | current->comm, oc->gfp_mask, &oc->gfp_mask, |
| 426 | nodemask_pr_args(oc->nodemask), oc->order, |
| 427 | current->signal->oom_score_adj); |
Michal Hocko | 9254990 | 2016-10-07 16:59:33 -0700 | [diff] [blame] | 428 | if (!IS_ENABLED(CONFIG_COMPACTION) && oc->order) |
| 429 | pr_warn("COMPACTION is disabled!!!\n"); |
Vlastimil Babka | a0795cd | 2016-03-15 14:56:05 -0700 | [diff] [blame] | 430 | |
David Rientjes | da39da3 | 2015-11-05 18:48:05 -0800 | [diff] [blame] | 431 | cpuset_print_current_mems_allowed(); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 432 | dump_stack(); |
Yang Shi | 852d8be | 2017-11-15 17:32:07 -0800 | [diff] [blame] | 433 | if (is_memcg_oom(oc)) |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 434 | mem_cgroup_print_oom_info(oc->memcg, p); |
Yang Shi | 852d8be | 2017-11-15 17:32:07 -0800 | [diff] [blame] | 435 | else { |
David Rientjes | 299c517 | 2017-02-24 14:55:42 -0800 | [diff] [blame] | 436 | show_mem(SHOW_MEM_FILTER_NODES, oc->nodemask); |
Yang Shi | 852d8be | 2017-11-15 17:32:07 -0800 | [diff] [blame] | 437 | if (is_dump_unreclaim_slabs()) |
| 438 | dump_unreclaimable_slab(); |
| 439 | } |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 440 | if (sysctl_oom_dump_tasks) |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 441 | dump_tasks(oc->memcg, oc->nodemask); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 442 | } |
| 443 | |
Michal Hocko | 5695be1 | 2014-10-20 18:12:32 +0200 | [diff] [blame] | 444 | /* |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 445 | * Number of OOM victims in flight |
Michal Hocko | 5695be1 | 2014-10-20 18:12:32 +0200 | [diff] [blame] | 446 | */ |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 447 | static atomic_t oom_victims = ATOMIC_INIT(0); |
| 448 | static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait); |
Michal Hocko | 5695be1 | 2014-10-20 18:12:32 +0200 | [diff] [blame] | 449 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 450 | static bool oom_killer_disabled __read_mostly; |
Michal Hocko | 5695be1 | 2014-10-20 18:12:32 +0200 | [diff] [blame] | 451 | |
Michal Hocko | bc448e8 | 2016-03-25 14:20:30 -0700 | [diff] [blame] | 452 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
| 453 | |
Michal Hocko | 3ef22df | 2016-05-19 17:13:12 -0700 | [diff] [blame] | 454 | /* |
| 455 | * task->mm can be NULL if the task is the exited group leader. So to |
| 456 | * determine whether the task is using a particular mm, we examine all the |
| 457 | * task's threads: if one of those is using this mm then this task was also |
| 458 | * using it. |
| 459 | */ |
Michal Hocko | 44a70ade | 2016-07-28 15:44:43 -0700 | [diff] [blame] | 460 | bool process_shares_mm(struct task_struct *p, struct mm_struct *mm) |
Michal Hocko | 3ef22df | 2016-05-19 17:13:12 -0700 | [diff] [blame] | 461 | { |
| 462 | struct task_struct *t; |
| 463 | |
| 464 | for_each_thread(p, t) { |
| 465 | struct mm_struct *t_mm = READ_ONCE(t->mm); |
| 466 | if (t_mm) |
| 467 | return t_mm == mm; |
| 468 | } |
| 469 | return false; |
| 470 | } |
| 471 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 472 | #ifdef CONFIG_MMU |
| 473 | /* |
| 474 | * OOM Reaper kernel thread which tries to reap the memory used by the OOM |
| 475 | * victim (if that is possible) to help the OOM killer to move on. |
| 476 | */ |
| 477 | static struct task_struct *oom_reaper_th; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 478 | static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait); |
Vladimir Davydov | 29c696e | 2016-03-25 14:20:39 -0700 | [diff] [blame] | 479 | static struct task_struct *oom_reaper_list; |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 480 | static DEFINE_SPINLOCK(oom_reaper_lock); |
| 481 | |
David Rientjes | 27ae357 | 2018-05-11 16:02:04 -0700 | [diff] [blame] | 482 | void __oom_reap_task_mm(struct mm_struct *mm) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 483 | { |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 484 | struct vm_area_struct *vma; |
David Rientjes | 27ae357 | 2018-05-11 16:02:04 -0700 | [diff] [blame] | 485 | |
| 486 | /* |
| 487 | * Tell all users of get_user/copy_from_user etc... that the content |
| 488 | * is no longer stable. No barriers really needed because unmapping |
| 489 | * should imply barriers already and the reader would hit a page fault |
| 490 | * if it stumbled over a reaped memory. |
| 491 | */ |
| 492 | set_bit(MMF_UNSTABLE, &mm->flags); |
| 493 | |
| 494 | for (vma = mm->mmap ; vma; vma = vma->vm_next) { |
| 495 | if (!can_madv_dontneed_vma(vma)) |
| 496 | continue; |
| 497 | |
| 498 | /* |
| 499 | * Only anonymous pages have a good chance to be dropped |
| 500 | * without additional steps which we cannot afford as we |
| 501 | * are OOM already. |
| 502 | * |
| 503 | * We do not even care about fs backed pages because all |
| 504 | * which are reclaimable have already been reclaimed and |
| 505 | * we do not want to block exit_mmap by keeping mm ref |
| 506 | * count elevated without a good reason. |
| 507 | */ |
| 508 | if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) { |
| 509 | const unsigned long start = vma->vm_start; |
| 510 | const unsigned long end = vma->vm_end; |
| 511 | struct mmu_gather tlb; |
| 512 | |
| 513 | tlb_gather_mmu(&tlb, mm, start, end); |
| 514 | mmu_notifier_invalidate_range_start(mm, start, end); |
| 515 | unmap_page_range(&tlb, vma, start, end, NULL); |
| 516 | mmu_notifier_invalidate_range_end(mm, start, end); |
| 517 | tlb_finish_mmu(&tlb, start, end); |
| 518 | } |
| 519 | } |
| 520 | } |
| 521 | |
| 522 | static bool oom_reap_task_mm(struct task_struct *tsk, struct mm_struct *mm) |
| 523 | { |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 524 | bool ret = true; |
| 525 | |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 526 | /* |
Michal Hocko | e2fe145 | 2016-05-27 14:27:35 -0700 | [diff] [blame] | 527 | * We have to make sure to not race with the victim exit path |
| 528 | * and cause premature new oom victim selection: |
David Rientjes | 27ae357 | 2018-05-11 16:02:04 -0700 | [diff] [blame] | 529 | * oom_reap_task_mm exit_mm |
Michal Hocko | e5e3f4c | 2016-07-26 15:24:50 -0700 | [diff] [blame] | 530 | * mmget_not_zero |
Michal Hocko | e2fe145 | 2016-05-27 14:27:35 -0700 | [diff] [blame] | 531 | * mmput |
| 532 | * atomic_dec_and_test |
| 533 | * exit_oom_victim |
| 534 | * [...] |
| 535 | * out_of_memory |
| 536 | * select_bad_process |
| 537 | * # no TIF_MEMDIE task selects new victim |
| 538 | * unmap_page_range # frees some memory |
| 539 | */ |
| 540 | mutex_lock(&oom_lock); |
| 541 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 542 | if (!down_read_trylock(&mm->mmap_sem)) { |
| 543 | ret = false; |
Roman Gushchin | 422580c | 2017-07-10 15:49:05 -0700 | [diff] [blame] | 544 | trace_skip_task_reaping(tsk->pid); |
Tetsuo Handa | 7ebffa4 | 2016-10-07 16:58:45 -0700 | [diff] [blame] | 545 | goto unlock_oom; |
Michal Hocko | e5e3f4c | 2016-07-26 15:24:50 -0700 | [diff] [blame] | 546 | } |
| 547 | |
| 548 | /* |
David Rientjes | f340ff8 | 2018-01-31 16:18:36 -0800 | [diff] [blame] | 549 | * If the mm has invalidate_{start,end}() notifiers that could block, |
Michal Hocko | 4d4bbd8 | 2017-10-03 16:14:50 -0700 | [diff] [blame] | 550 | * sleep to give the oom victim some more time. |
| 551 | * TODO: we really want to get rid of this ugly hack and make sure that |
David Rientjes | f340ff8 | 2018-01-31 16:18:36 -0800 | [diff] [blame] | 552 | * notifiers cannot block for unbounded amount of time |
Michal Hocko | 4d4bbd8 | 2017-10-03 16:14:50 -0700 | [diff] [blame] | 553 | */ |
David Rientjes | f340ff8 | 2018-01-31 16:18:36 -0800 | [diff] [blame] | 554 | if (mm_has_blockable_invalidate_notifiers(mm)) { |
Michal Hocko | 4d4bbd8 | 2017-10-03 16:14:50 -0700 | [diff] [blame] | 555 | up_read(&mm->mmap_sem); |
| 556 | schedule_timeout_idle(HZ); |
| 557 | goto unlock_oom; |
| 558 | } |
| 559 | |
| 560 | /* |
Andrea Arcangeli | 2129258 | 2017-09-06 16:25:00 -0700 | [diff] [blame] | 561 | * MMF_OOM_SKIP is set by exit_mmap when the OOM reaper can't |
| 562 | * work on the mm anymore. The check for MMF_OOM_SKIP must run |
| 563 | * under mmap_sem for reading because it serializes against the |
| 564 | * down_write();up_write() cycle in exit_mmap(). |
Michal Hocko | e5e3f4c | 2016-07-26 15:24:50 -0700 | [diff] [blame] | 565 | */ |
Andrea Arcangeli | 2129258 | 2017-09-06 16:25:00 -0700 | [diff] [blame] | 566 | if (test_bit(MMF_OOM_SKIP, &mm->flags)) { |
Michal Hocko | e5e3f4c | 2016-07-26 15:24:50 -0700 | [diff] [blame] | 567 | up_read(&mm->mmap_sem); |
Roman Gushchin | 422580c | 2017-07-10 15:49:05 -0700 | [diff] [blame] | 568 | trace_skip_task_reaping(tsk->pid); |
Tetsuo Handa | 7ebffa4 | 2016-10-07 16:58:45 -0700 | [diff] [blame] | 569 | goto unlock_oom; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 570 | } |
| 571 | |
Roman Gushchin | 422580c | 2017-07-10 15:49:05 -0700 | [diff] [blame] | 572 | trace_start_task_reaping(tsk->pid); |
| 573 | |
David Rientjes | 27ae357 | 2018-05-11 16:02:04 -0700 | [diff] [blame] | 574 | __oom_reap_task_mm(mm); |
Michal Hocko | 3f70dc3 | 2016-10-07 16:59:06 -0700 | [diff] [blame] | 575 | |
Michal Hocko | bc448e8 | 2016-03-25 14:20:30 -0700 | [diff] [blame] | 576 | pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n", |
| 577 | task_pid_nr(tsk), tsk->comm, |
| 578 | K(get_mm_counter(mm, MM_ANONPAGES)), |
| 579 | K(get_mm_counter(mm, MM_FILEPAGES)), |
| 580 | K(get_mm_counter(mm, MM_SHMEMPAGES))); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 581 | up_read(&mm->mmap_sem); |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 582 | |
Roman Gushchin | 422580c | 2017-07-10 15:49:05 -0700 | [diff] [blame] | 583 | trace_finish_task_reaping(tsk->pid); |
Michal Hocko | e5e3f4c | 2016-07-26 15:24:50 -0700 | [diff] [blame] | 584 | unlock_oom: |
| 585 | mutex_unlock(&oom_lock); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 586 | return ret; |
| 587 | } |
| 588 | |
Michal Hocko | bc448e8 | 2016-03-25 14:20:30 -0700 | [diff] [blame] | 589 | #define MAX_OOM_REAP_RETRIES 10 |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 590 | static void oom_reap_task(struct task_struct *tsk) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 591 | { |
| 592 | int attempts = 0; |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 593 | struct mm_struct *mm = tsk->signal->oom_mm; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 594 | |
| 595 | /* Retry the down_read_trylock(mmap_sem) a few times */ |
David Rientjes | 27ae357 | 2018-05-11 16:02:04 -0700 | [diff] [blame] | 596 | while (attempts++ < MAX_OOM_REAP_RETRIES && !oom_reap_task_mm(tsk, mm)) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 597 | schedule_timeout_idle(HZ/10); |
| 598 | |
Tetsuo Handa | 97b1255 | 2018-04-05 16:25:45 -0700 | [diff] [blame] | 599 | if (attempts <= MAX_OOM_REAP_RETRIES || |
| 600 | test_bit(MMF_OOM_SKIP, &mm->flags)) |
Tetsuo Handa | 7ebffa4 | 2016-10-07 16:58:45 -0700 | [diff] [blame] | 601 | goto done; |
Michal Hocko | 11a410d | 2016-07-28 15:44:58 -0700 | [diff] [blame] | 602 | |
Tetsuo Handa | 7ebffa4 | 2016-10-07 16:58:45 -0700 | [diff] [blame] | 603 | pr_info("oom_reaper: unable to reap pid:%d (%s)\n", |
| 604 | task_pid_nr(tsk), tsk->comm); |
Tetsuo Handa | 7ebffa4 | 2016-10-07 16:58:45 -0700 | [diff] [blame] | 605 | debug_show_all_locks(); |
Michal Hocko | bc448e8 | 2016-03-25 14:20:30 -0700 | [diff] [blame] | 606 | |
Tetsuo Handa | 7ebffa4 | 2016-10-07 16:58:45 -0700 | [diff] [blame] | 607 | done: |
Michal Hocko | 449d777 | 2016-05-19 17:13:15 -0700 | [diff] [blame] | 608 | tsk->oom_reaper_list = NULL; |
Michal Hocko | 449d777 | 2016-05-19 17:13:15 -0700 | [diff] [blame] | 609 | |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 610 | /* |
| 611 | * Hide this mm from OOM killer because it has been either reaped or |
| 612 | * somebody can't call up_write(mmap_sem). |
| 613 | */ |
Michal Hocko | 862e307 | 2016-10-07 16:58:57 -0700 | [diff] [blame] | 614 | set_bit(MMF_OOM_SKIP, &mm->flags); |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 615 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 616 | /* Drop a reference taken by wake_oom_reaper */ |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 617 | put_task_struct(tsk); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 618 | } |
| 619 | |
| 620 | static int oom_reaper(void *unused) |
| 621 | { |
| 622 | while (true) { |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 623 | struct task_struct *tsk = NULL; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 624 | |
Vladimir Davydov | 29c696e | 2016-03-25 14:20:39 -0700 | [diff] [blame] | 625 | wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL); |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 626 | spin_lock(&oom_reaper_lock); |
Vladimir Davydov | 29c696e | 2016-03-25 14:20:39 -0700 | [diff] [blame] | 627 | if (oom_reaper_list != NULL) { |
| 628 | tsk = oom_reaper_list; |
| 629 | oom_reaper_list = tsk->oom_reaper_list; |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 630 | } |
| 631 | spin_unlock(&oom_reaper_lock); |
| 632 | |
| 633 | if (tsk) |
| 634 | oom_reap_task(tsk); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 635 | } |
| 636 | |
| 637 | return 0; |
| 638 | } |
| 639 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 640 | static void wake_oom_reaper(struct task_struct *tsk) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 641 | { |
Michal Hocko | af8e15c | 2016-04-01 14:31:34 -0700 | [diff] [blame] | 642 | /* tsk is already queued? */ |
| 643 | if (tsk == oom_reaper_list || tsk->oom_reaper_list) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 644 | return; |
| 645 | |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 646 | get_task_struct(tsk); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 647 | |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 648 | spin_lock(&oom_reaper_lock); |
Vladimir Davydov | 29c696e | 2016-03-25 14:20:39 -0700 | [diff] [blame] | 649 | tsk->oom_reaper_list = oom_reaper_list; |
| 650 | oom_reaper_list = tsk; |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 651 | spin_unlock(&oom_reaper_lock); |
Roman Gushchin | 422580c | 2017-07-10 15:49:05 -0700 | [diff] [blame] | 652 | trace_wake_reaper(tsk->pid); |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 653 | wake_up(&oom_reaper_wait); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 654 | } |
| 655 | |
| 656 | static int __init oom_init(void) |
| 657 | { |
| 658 | oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper"); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 659 | return 0; |
| 660 | } |
| 661 | subsys_initcall(oom_init) |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 662 | #else |
| 663 | static inline void wake_oom_reaper(struct task_struct *tsk) |
| 664 | { |
| 665 | } |
| 666 | #endif /* CONFIG_MMU */ |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 667 | |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 668 | /** |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 669 | * mark_oom_victim - mark the given task as OOM victim |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 670 | * @tsk: task to mark |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 671 | * |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 672 | * Has to be called with oom_lock held and never after |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 673 | * oom has been disabled already. |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 674 | * |
| 675 | * tsk->mm has to be non NULL and caller has to guarantee it is stable (either |
| 676 | * under task_lock or operate on the current). |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 677 | */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 678 | static void mark_oom_victim(struct task_struct *tsk) |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 679 | { |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 680 | struct mm_struct *mm = tsk->mm; |
| 681 | |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 682 | WARN_ON(oom_killer_disabled); |
| 683 | /* OOM killer might race with memcg OOM */ |
| 684 | if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE)) |
| 685 | return; |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 686 | |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 687 | /* oom_mm is bound to the signal struct life time. */ |
Michal Hocko | 4837fe3 | 2017-12-14 15:33:15 -0800 | [diff] [blame] | 688 | if (!cmpxchg(&tsk->signal->oom_mm, NULL, mm)) { |
Vegard Nossum | f1f1007 | 2017-02-27 14:30:07 -0800 | [diff] [blame] | 689 | mmgrab(tsk->signal->oom_mm); |
Michal Hocko | 4837fe3 | 2017-12-14 15:33:15 -0800 | [diff] [blame] | 690 | set_bit(MMF_OOM_VICTIM, &mm->flags); |
| 691 | } |
Michal Hocko | 26db62f | 2016-10-07 16:58:51 -0700 | [diff] [blame] | 692 | |
Michal Hocko | 63a8ca9 | 2015-02-11 15:26:15 -0800 | [diff] [blame] | 693 | /* |
| 694 | * Make sure that the task is woken up from uninterruptible sleep |
| 695 | * if it is frozen because OOM killer wouldn't be able to free |
| 696 | * any memory and livelock. freezing_slow_path will tell the freezer |
| 697 | * that TIF_MEMDIE tasks should be ignored. |
| 698 | */ |
| 699 | __thaw_task(tsk); |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 700 | atomic_inc(&oom_victims); |
Roman Gushchin | 422580c | 2017-07-10 15:49:05 -0700 | [diff] [blame] | 701 | trace_mark_victim(tsk->pid); |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 702 | } |
| 703 | |
| 704 | /** |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 705 | * exit_oom_victim - note the exit of an OOM victim |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 706 | */ |
Tetsuo Handa | 3853120 | 2016-10-07 16:59:03 -0700 | [diff] [blame] | 707 | void exit_oom_victim(void) |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 708 | { |
Tetsuo Handa | 3853120 | 2016-10-07 16:59:03 -0700 | [diff] [blame] | 709 | clear_thread_flag(TIF_MEMDIE); |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 710 | |
Johannes Weiner | c38f102 | 2015-06-24 16:57:13 -0700 | [diff] [blame] | 711 | if (!atomic_dec_return(&oom_victims)) |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 712 | wake_up_all(&oom_victims_wait); |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 713 | } |
| 714 | |
| 715 | /** |
Michal Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 716 | * oom_killer_enable - enable OOM killer |
| 717 | */ |
| 718 | void oom_killer_enable(void) |
| 719 | { |
| 720 | oom_killer_disabled = false; |
Michal Hocko | d75da00 | 2017-05-03 14:54:57 -0700 | [diff] [blame] | 721 | pr_info("OOM killer enabled.\n"); |
Michal Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 722 | } |
| 723 | |
| 724 | /** |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 725 | * oom_killer_disable - disable OOM killer |
Michal Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 726 | * @timeout: maximum timeout to wait for oom victims in jiffies |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 727 | * |
| 728 | * Forces all page allocations to fail rather than trigger OOM killer. |
Michal Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 729 | * Will block and wait until all OOM victims are killed or the given |
| 730 | * timeout expires. |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 731 | * |
| 732 | * The function cannot be called when there are runnable user tasks because |
| 733 | * the userspace would see unexpected allocation failures as a result. Any |
| 734 | * new usage of this function should be consulted with MM people. |
| 735 | * |
| 736 | * Returns true if successful and false if the OOM killer cannot be |
| 737 | * disabled. |
| 738 | */ |
Michal Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 739 | bool oom_killer_disable(signed long timeout) |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 740 | { |
Michal Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 741 | signed long ret; |
| 742 | |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 743 | /* |
Tetsuo Handa | 6afcf28 | 2016-03-17 14:20:45 -0700 | [diff] [blame] | 744 | * Make sure to not race with an ongoing OOM killer. Check that the |
| 745 | * current is not killed (possibly due to sharing the victim's memory). |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 746 | */ |
Tetsuo Handa | 6afcf28 | 2016-03-17 14:20:45 -0700 | [diff] [blame] | 747 | if (mutex_lock_killable(&oom_lock)) |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 748 | return false; |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 749 | oom_killer_disabled = true; |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 750 | mutex_unlock(&oom_lock); |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 751 | |
Michal Hocko | 7d2e7a2 | 2016-10-07 16:59:00 -0700 | [diff] [blame] | 752 | ret = wait_event_interruptible_timeout(oom_victims_wait, |
| 753 | !atomic_read(&oom_victims), timeout); |
| 754 | if (ret <= 0) { |
| 755 | oom_killer_enable(); |
| 756 | return false; |
| 757 | } |
Michal Hocko | d75da00 | 2017-05-03 14:54:57 -0700 | [diff] [blame] | 758 | pr_info("OOM killer disabled.\n"); |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 759 | |
| 760 | return true; |
| 761 | } |
| 762 | |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 763 | static inline bool __task_will_free_mem(struct task_struct *task) |
| 764 | { |
| 765 | struct signal_struct *sig = task->signal; |
| 766 | |
| 767 | /* |
| 768 | * A coredumping process may sleep for an extended period in exit_mm(), |
| 769 | * so the oom killer cannot assume that the process will promptly exit |
| 770 | * and release memory. |
| 771 | */ |
| 772 | if (sig->flags & SIGNAL_GROUP_COREDUMP) |
| 773 | return false; |
| 774 | |
| 775 | if (sig->flags & SIGNAL_GROUP_EXIT) |
| 776 | return true; |
| 777 | |
| 778 | if (thread_group_empty(task) && (task->flags & PF_EXITING)) |
| 779 | return true; |
| 780 | |
| 781 | return false; |
| 782 | } |
| 783 | |
| 784 | /* |
| 785 | * Checks whether the given task is dying or exiting and likely to |
| 786 | * release its address space. This means that all threads and processes |
| 787 | * sharing the same mm have to be killed or exiting. |
Michal Hocko | 091f362 | 2016-07-28 15:45:04 -0700 | [diff] [blame] | 788 | * Caller has to make sure that task->mm is stable (hold task_lock or |
| 789 | * it operates on the current). |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 790 | */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 791 | static bool task_will_free_mem(struct task_struct *task) |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 792 | { |
Michal Hocko | 091f362 | 2016-07-28 15:45:04 -0700 | [diff] [blame] | 793 | struct mm_struct *mm = task->mm; |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 794 | struct task_struct *p; |
Geert Uytterhoeven | f33e6f0 | 2016-08-11 15:33:09 -0700 | [diff] [blame] | 795 | bool ret = true; |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 796 | |
Michal Hocko | 091f362 | 2016-07-28 15:45:04 -0700 | [diff] [blame] | 797 | /* |
| 798 | * Skip tasks without mm because it might have passed its exit_mm and |
| 799 | * exit_oom_victim. oom_reaper could have rescued that but do not rely |
| 800 | * on that for now. We can consider find_lock_task_mm in future. |
| 801 | */ |
| 802 | if (!mm) |
| 803 | return false; |
| 804 | |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 805 | if (!__task_will_free_mem(task)) |
| 806 | return false; |
| 807 | |
| 808 | /* |
Michal Hocko | 696453e | 2016-07-28 15:44:55 -0700 | [diff] [blame] | 809 | * This task has already been drained by the oom reaper so there are |
| 810 | * only small chances it will free some more |
| 811 | */ |
Michal Hocko | 862e307 | 2016-10-07 16:58:57 -0700 | [diff] [blame] | 812 | if (test_bit(MMF_OOM_SKIP, &mm->flags)) |
Michal Hocko | 696453e | 2016-07-28 15:44:55 -0700 | [diff] [blame] | 813 | return false; |
Michal Hocko | 696453e | 2016-07-28 15:44:55 -0700 | [diff] [blame] | 814 | |
Michal Hocko | 091f362 | 2016-07-28 15:45:04 -0700 | [diff] [blame] | 815 | if (atomic_read(&mm->mm_users) <= 1) |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 816 | return true; |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 817 | |
| 818 | /* |
Michal Hocko | 5870c2e | 2016-10-07 16:57:32 -0700 | [diff] [blame] | 819 | * Make sure that all tasks which share the mm with the given tasks |
| 820 | * are dying as well to make sure that a) nobody pins its mm and |
| 821 | * b) the task is also reapable by the oom reaper. |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 822 | */ |
| 823 | rcu_read_lock(); |
| 824 | for_each_process(p) { |
| 825 | if (!process_shares_mm(p, mm)) |
| 826 | continue; |
| 827 | if (same_thread_group(task, p)) |
| 828 | continue; |
| 829 | ret = __task_will_free_mem(p); |
| 830 | if (!ret) |
| 831 | break; |
| 832 | } |
| 833 | rcu_read_unlock(); |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 834 | |
| 835 | return ret; |
| 836 | } |
| 837 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 838 | static void oom_kill_process(struct oom_control *oc, const char *message) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 839 | { |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 840 | struct task_struct *p = oc->chosen; |
| 841 | unsigned int points = oc->chosen_points; |
Linus Torvalds | 52d3c03 | 2011-03-14 15:17:07 -0700 | [diff] [blame] | 842 | struct task_struct *victim = p; |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 843 | struct task_struct *child; |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 844 | struct task_struct *t; |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 845 | struct mm_struct *mm; |
Linus Torvalds | 52d3c03 | 2011-03-14 15:17:07 -0700 | [diff] [blame] | 846 | unsigned int victim_points = 0; |
David Rientjes | dc3f21e | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 847 | static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
| 848 | DEFAULT_RATELIMIT_BURST); |
Tetsuo Handa | bb29902 | 2016-03-25 14:20:44 -0700 | [diff] [blame] | 849 | bool can_oom_reap = true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 850 | |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 851 | /* |
| 852 | * If the task is already exiting, don't alarm the sysadmin or kill |
Michal Hocko | cd04ae1 | 2017-09-06 16:24:50 -0700 | [diff] [blame] | 853 | * its children or threads, just give it access to memory reserves |
| 854 | * so it can die quickly |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 855 | */ |
Michal Hocko | 091f362 | 2016-07-28 15:45:04 -0700 | [diff] [blame] | 856 | task_lock(p); |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 857 | if (task_will_free_mem(p)) { |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 858 | mark_oom_victim(p); |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 859 | wake_oom_reaper(p); |
Michal Hocko | 091f362 | 2016-07-28 15:45:04 -0700 | [diff] [blame] | 860 | task_unlock(p); |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 861 | put_task_struct(p); |
David Rientjes | 2a1c9b1 | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 862 | return; |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 863 | } |
Michal Hocko | 091f362 | 2016-07-28 15:45:04 -0700 | [diff] [blame] | 864 | task_unlock(p); |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 865 | |
David Rientjes | dc3f21e | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 866 | if (__ratelimit(&oom_rs)) |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 867 | dump_header(oc, p); |
David Rientjes | 8447d95 | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 868 | |
Wang Long | f0d6647 | 2015-06-24 16:58:01 -0700 | [diff] [blame] | 869 | pr_err("%s: Kill process %d (%s) score %u or sacrifice child\n", |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 870 | message, task_pid_nr(p), p->comm, points); |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 871 | |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 872 | /* |
| 873 | * If any of p's children has a different mm and is eligible for kill, |
David Rientjes | 1123983 | 2011-07-25 17:12:17 -0700 | [diff] [blame] | 874 | * the one with the highest oom_badness() score is sacrificed for its |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 875 | * parent. This attempts to lose the minimal amount of work done while |
| 876 | * still freeing memory. |
| 877 | */ |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 878 | read_lock(&tasklist_lock); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 879 | for_each_thread(p, t) { |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 880 | list_for_each_entry(child, &t->children, sibling) { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 881 | unsigned int child_points; |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 882 | |
Oleg Nesterov | 4d7b339 | 2015-11-05 18:48:26 -0800 | [diff] [blame] | 883 | if (process_shares_mm(child, p->mm)) |
David Rientjes | edd4554 | 2011-03-22 16:30:12 -0700 | [diff] [blame] | 884 | continue; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 885 | /* |
| 886 | * oom_badness() returns 0 if the thread is unkillable |
| 887 | */ |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 888 | child_points = oom_badness(child, |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 889 | oc->memcg, oc->nodemask, oc->totalpages); |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 890 | if (child_points > victim_points) { |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 891 | put_task_struct(victim); |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 892 | victim = child; |
| 893 | victim_points = child_points; |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 894 | get_task_struct(victim); |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 895 | } |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 896 | } |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 897 | } |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 898 | read_unlock(&tasklist_lock); |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 899 | |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 900 | p = find_lock_task_mm(victim); |
| 901 | if (!p) { |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 902 | put_task_struct(victim); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 903 | return; |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 904 | } else if (victim != p) { |
| 905 | get_task_struct(p); |
| 906 | put_task_struct(victim); |
| 907 | victim = p; |
| 908 | } |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 909 | |
Tetsuo Handa | 880b768 | 2015-11-05 18:47:51 -0800 | [diff] [blame] | 910 | /* Get a reference to safely compare mm after task_unlock(victim) */ |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 911 | mm = victim->mm; |
Vegard Nossum | f1f1007 | 2017-02-27 14:30:07 -0800 | [diff] [blame] | 912 | mmgrab(mm); |
Konstantin Khlebnikov | 8e675f7 | 2017-07-06 15:40:28 -0700 | [diff] [blame] | 913 | |
| 914 | /* Raise event before sending signal: task reaper must see this */ |
| 915 | count_vm_event(OOM_KILL); |
Roman Gushchin | fe6bdfc | 2018-06-14 15:28:05 -0700 | [diff] [blame] | 916 | memcg_memory_event_mm(mm, MEMCG_OOM_KILL); |
Konstantin Khlebnikov | 8e675f7 | 2017-07-06 15:40:28 -0700 | [diff] [blame] | 917 | |
Tetsuo Handa | 426fb5e | 2015-11-05 18:47:44 -0800 | [diff] [blame] | 918 | /* |
Michal Hocko | cd04ae1 | 2017-09-06 16:24:50 -0700 | [diff] [blame] | 919 | * We should send SIGKILL before granting access to memory reserves |
| 920 | * in order to prevent the OOM victim from depleting the memory |
| 921 | * reserves from the user space under its control. |
Tetsuo Handa | 426fb5e | 2015-11-05 18:47:44 -0800 | [diff] [blame] | 922 | */ |
| 923 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true); |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 924 | mark_oom_victim(victim); |
Jerome Marchand | eca56ff | 2016-01-14 15:19:26 -0800 | [diff] [blame] | 925 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n", |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 926 | task_pid_nr(victim), victim->comm, K(victim->mm->total_vm), |
| 927 | K(get_mm_counter(victim->mm, MM_ANONPAGES)), |
Jerome Marchand | eca56ff | 2016-01-14 15:19:26 -0800 | [diff] [blame] | 928 | K(get_mm_counter(victim->mm, MM_FILEPAGES)), |
| 929 | K(get_mm_counter(victim->mm, MM_SHMEMPAGES))); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 930 | task_unlock(victim); |
| 931 | |
| 932 | /* |
| 933 | * Kill all user processes sharing victim->mm in other thread groups, if |
| 934 | * any. They don't get access to memory reserves, though, to avoid |
| 935 | * depletion of all memory. This prevents mm->mmap_sem livelock when an |
| 936 | * oom killed thread cannot exit because it requires the semaphore and |
| 937 | * its contended by another thread trying to allocate memory itself. |
| 938 | * That thread will now get access to memory reserves since it has a |
| 939 | * pending fatal signal. |
| 940 | */ |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 941 | rcu_read_lock(); |
Oleg Nesterov | c319025 | 2015-11-05 18:48:23 -0800 | [diff] [blame] | 942 | for_each_process(p) { |
Oleg Nesterov | 4d7b339 | 2015-11-05 18:48:26 -0800 | [diff] [blame] | 943 | if (!process_shares_mm(p, mm)) |
Oleg Nesterov | c319025 | 2015-11-05 18:48:23 -0800 | [diff] [blame] | 944 | continue; |
| 945 | if (same_thread_group(p, victim)) |
| 946 | continue; |
Michal Hocko | 1b51e65 | 2016-10-07 16:59:09 -0700 | [diff] [blame] | 947 | if (is_global_init(p)) { |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 948 | can_oom_reap = false; |
Michal Hocko | 862e307 | 2016-10-07 16:58:57 -0700 | [diff] [blame] | 949 | set_bit(MMF_OOM_SKIP, &mm->flags); |
Michal Hocko | a373966 | 2016-07-28 15:45:01 -0700 | [diff] [blame] | 950 | pr_info("oom killer %d (%s) has mm pinned by %d (%s)\n", |
| 951 | task_pid_nr(victim), victim->comm, |
| 952 | task_pid_nr(p), p->comm); |
Oleg Nesterov | c319025 | 2015-11-05 18:48:23 -0800 | [diff] [blame] | 953 | continue; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 954 | } |
Michal Hocko | 1b51e65 | 2016-10-07 16:59:09 -0700 | [diff] [blame] | 955 | /* |
| 956 | * No use_mm() user needs to read from the userspace so we are |
| 957 | * ok to reap it. |
| 958 | */ |
| 959 | if (unlikely(p->flags & PF_KTHREAD)) |
| 960 | continue; |
Oleg Nesterov | c319025 | 2015-11-05 18:48:23 -0800 | [diff] [blame] | 961 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true); |
| 962 | } |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 963 | rcu_read_unlock(); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 964 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 965 | if (can_oom_reap) |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 966 | wake_oom_reaper(victim); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 967 | |
Tetsuo Handa | 880b768 | 2015-11-05 18:47:51 -0800 | [diff] [blame] | 968 | mmdrop(mm); |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 969 | put_task_struct(victim); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 970 | } |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 971 | #undef K |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 972 | |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 973 | /* |
| 974 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. |
| 975 | */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 976 | static void check_panic_on_oom(struct oom_control *oc, |
| 977 | enum oom_constraint constraint) |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 978 | { |
| 979 | if (likely(!sysctl_panic_on_oom)) |
| 980 | return; |
| 981 | if (sysctl_panic_on_oom != 2) { |
| 982 | /* |
| 983 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel |
| 984 | * does not panic for cpuset, mempolicy, or memcg allocation |
| 985 | * failures. |
| 986 | */ |
| 987 | if (constraint != CONSTRAINT_NONE) |
| 988 | return; |
| 989 | } |
David Rientjes | 071a4be | 2015-09-08 15:00:42 -0700 | [diff] [blame] | 990 | /* Do not panic for oom kills triggered by sysrq */ |
Yaowei Bai | db2a0dd | 2015-11-06 16:28:06 -0800 | [diff] [blame] | 991 | if (is_sysrq_oom(oc)) |
David Rientjes | 071a4be | 2015-09-08 15:00:42 -0700 | [diff] [blame] | 992 | return; |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 993 | dump_header(oc, NULL); |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 994 | panic("Out of memory: %s panic_on_oom is enabled\n", |
| 995 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); |
| 996 | } |
| 997 | |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 998 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
| 999 | |
| 1000 | int register_oom_notifier(struct notifier_block *nb) |
| 1001 | { |
| 1002 | return blocking_notifier_chain_register(&oom_notify_list, nb); |
| 1003 | } |
| 1004 | EXPORT_SYMBOL_GPL(register_oom_notifier); |
| 1005 | |
| 1006 | int unregister_oom_notifier(struct notifier_block *nb) |
| 1007 | { |
| 1008 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); |
| 1009 | } |
| 1010 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); |
| 1011 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1012 | /** |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1013 | * out_of_memory - kill the "best" process when we run out of memory |
| 1014 | * @oc: pointer to struct oom_control |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1015 | * |
| 1016 | * If we run out of memory, we have the choice between either |
| 1017 | * killing a random task (bad), letting the system crash (worse) |
| 1018 | * OR try to be smart about which process to kill. Note that we |
| 1019 | * don't have to be perfect here, we just have to be good. |
| 1020 | */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1021 | bool out_of_memory(struct oom_control *oc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1022 | { |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 1023 | unsigned long freed = 0; |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 1024 | enum oom_constraint constraint = CONSTRAINT_NONE; |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 1025 | |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1026 | if (oom_killer_disabled) |
| 1027 | return false; |
| 1028 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1029 | if (!is_memcg_oom(oc)) { |
| 1030 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); |
| 1031 | if (freed > 0) |
| 1032 | /* Got some memory back in the last second. */ |
| 1033 | return true; |
| 1034 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1035 | |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 1036 | /* |
David Rientjes | 9ff4868 | 2012-12-11 16:01:30 -0800 | [diff] [blame] | 1037 | * If current has a pending SIGKILL or is exiting, then automatically |
| 1038 | * select it. The goal is to allow it to allocate so that it may |
| 1039 | * quickly exit and free its memory. |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 1040 | */ |
Michal Hocko | 091f362 | 2016-07-28 15:45:04 -0700 | [diff] [blame] | 1041 | if (task_will_free_mem(current)) { |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 1042 | mark_oom_victim(current); |
Michal Hocko | 1af8bb4 | 2016-07-28 15:44:52 -0700 | [diff] [blame] | 1043 | wake_oom_reaper(current); |
David Rientjes | 75e8f8b | 2015-09-08 15:00:47 -0700 | [diff] [blame] | 1044 | return true; |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 1045 | } |
| 1046 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 1047 | /* |
Michal Hocko | 3da88fb | 2016-05-19 17:13:09 -0700 | [diff] [blame] | 1048 | * The OOM killer does not compensate for IO-less reclaim. |
| 1049 | * pagefault_out_of_memory lost its gfp context so we have to |
| 1050 | * make sure exclude 0 mask - all other users should have at least |
| 1051 | * ___GFP_DIRECT_RECLAIM to get here. |
| 1052 | */ |
Michal Hocko | 06ad276 | 2017-02-22 15:46:22 -0800 | [diff] [blame] | 1053 | if (oc->gfp_mask && !(oc->gfp_mask & __GFP_FS)) |
Michal Hocko | 3da88fb | 2016-05-19 17:13:09 -0700 | [diff] [blame] | 1054 | return true; |
| 1055 | |
| 1056 | /* |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 1057 | * Check if there were limitations on the allocation (only relevant for |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1058 | * NUMA and memcg) that may require different handling. |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 1059 | */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1060 | constraint = constrained_alloc(oc); |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1061 | if (constraint != CONSTRAINT_MEMORY_POLICY) |
| 1062 | oc->nodemask = NULL; |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 1063 | check_panic_on_oom(oc, constraint); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 1064 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1065 | if (!is_memcg_oom(oc) && sysctl_oom_kill_allocating_task && |
| 1066 | current->mm && !oom_unkillable_task(current, NULL, oc->nodemask) && |
David Rientjes | 121d1ba | 2012-07-31 16:42:55 -0700 | [diff] [blame] | 1067 | current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) { |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 1068 | get_task_struct(current); |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1069 | oc->chosen = current; |
| 1070 | oom_kill_process(oc, "Out of memory (oom_kill_allocating_task)"); |
David Rientjes | 75e8f8b | 2015-09-08 15:00:47 -0700 | [diff] [blame] | 1071 | return true; |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 1072 | } |
| 1073 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1074 | select_bad_process(oc); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 1075 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1076 | if (!oc->chosen && !is_sysrq_oom(oc) && !is_memcg_oom(oc)) { |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 1077 | dump_header(oc, NULL); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 1078 | panic("Out of memory and no killable processes...\n"); |
| 1079 | } |
Michal Hocko | 9bfe5de | 2018-08-17 15:49:04 -0700 | [diff] [blame^] | 1080 | if (oc->chosen && oc->chosen != (void *)-1UL) |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1081 | oom_kill_process(oc, !is_memcg_oom(oc) ? "Out of memory" : |
| 1082 | "Memory cgroup out of memory"); |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1083 | return !!oc->chosen; |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 1084 | } |
| 1085 | |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 1086 | /* |
| 1087 | * The pagefault handler calls here because it is out of memory, so kill a |
Vladimir Davydov | 798fd75 | 2016-07-26 15:22:30 -0700 | [diff] [blame] | 1088 | * memory-hogging task. If oom_lock is held by somebody else, a parallel oom |
| 1089 | * killing is already in progress so do nothing. |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 1090 | */ |
| 1091 | void pagefault_out_of_memory(void) |
| 1092 | { |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1093 | struct oom_control oc = { |
| 1094 | .zonelist = NULL, |
| 1095 | .nodemask = NULL, |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 1096 | .memcg = NULL, |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1097 | .gfp_mask = 0, |
| 1098 | .order = 0, |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1099 | }; |
| 1100 | |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1101 | if (mem_cgroup_oom_synchronize(true)) |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1102 | return; |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1103 | |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1104 | if (!mutex_trylock(&oom_lock)) |
| 1105 | return; |
Tetsuo Handa | a104808 | 2016-10-07 17:00:49 -0700 | [diff] [blame] | 1106 | out_of_memory(&oc); |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1107 | mutex_unlock(&oom_lock); |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 1108 | } |