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> |
| 25 | #include <linux/swap.h> |
| 26 | #include <linux/timex.h> |
| 27 | #include <linux/jiffies.h> |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 28 | #include <linux/cpuset.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 29 | #include <linux/export.h> |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 30 | #include <linux/notifier.h> |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 31 | #include <linux/memcontrol.h> |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 32 | #include <linux/mempolicy.h> |
David Howells | 5cd9c58 | 2008-08-14 11:37:28 +0100 | [diff] [blame] | 33 | #include <linux/security.h> |
David Rientjes | edd4554 | 2011-03-22 16:30:12 -0700 | [diff] [blame] | 34 | #include <linux/ptrace.h> |
David Rientjes | f660daa | 2011-10-31 17:07:07 -0700 | [diff] [blame] | 35 | #include <linux/freezer.h> |
KAMEZAWA Hiroyuki | 43d2b11 | 2012-01-10 15:08:09 -0800 | [diff] [blame] | 36 | #include <linux/ftrace.h> |
David Rientjes | dc3f21e | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 37 | #include <linux/ratelimit.h> |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 38 | #include <linux/kthread.h> |
| 39 | #include <linux/init.h> |
| 40 | |
| 41 | #include <asm/tlb.h> |
| 42 | #include "internal.h" |
KAMEZAWA Hiroyuki | 43d2b11 | 2012-01-10 15:08:09 -0800 | [diff] [blame] | 43 | |
| 44 | #define CREATE_TRACE_POINTS |
| 45 | #include <trace/events/oom.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 46 | |
KAMEZAWA Hiroyuki | fadd8fb | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 47 | int sysctl_panic_on_oom; |
David Rientjes | fe071d7 | 2007-10-16 23:25:56 -0700 | [diff] [blame] | 48 | int sysctl_oom_kill_allocating_task; |
David Rientjes | ad915c4 | 2010-08-09 17:18:53 -0700 | [diff] [blame] | 49 | int sysctl_oom_dump_tasks = 1; |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 50 | |
| 51 | DEFINE_MUTEX(oom_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 52 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 53 | #ifdef CONFIG_NUMA |
| 54 | /** |
| 55 | * has_intersects_mems_allowed() - check task eligiblity for kill |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 56 | * @start: task struct of which task to consider |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 57 | * @mask: nodemask passed to page allocator for mempolicy ooms |
| 58 | * |
| 59 | * Task eligibility is determined by whether or not a candidate task, @tsk, |
| 60 | * shares the same mempolicy nodes as current if it is bound by such a policy |
| 61 | * and whether or not it has the same set of allowed cpuset nodes. |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 62 | */ |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 63 | static bool has_intersects_mems_allowed(struct task_struct *start, |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 64 | const nodemask_t *mask) |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 65 | { |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 66 | struct task_struct *tsk; |
| 67 | bool ret = false; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 68 | |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 69 | rcu_read_lock(); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 70 | for_each_thread(start, tsk) { |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 71 | if (mask) { |
| 72 | /* |
| 73 | * If this is a mempolicy constrained oom, tsk's |
| 74 | * cpuset is irrelevant. Only return true if its |
| 75 | * mempolicy intersects current, otherwise it may be |
| 76 | * needlessly killed. |
| 77 | */ |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 78 | ret = mempolicy_nodemask_intersects(tsk, mask); |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 79 | } else { |
| 80 | /* |
| 81 | * This is not a mempolicy constrained oom, so only |
| 82 | * check the mems of tsk's cpuset. |
| 83 | */ |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 84 | ret = cpuset_mems_allowed_intersects(current, tsk); |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 85 | } |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 86 | if (ret) |
| 87 | break; |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 88 | } |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 89 | rcu_read_unlock(); |
KOSAKI Motohiro | df1090a | 2010-08-09 17:19:39 -0700 | [diff] [blame] | 90 | |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 91 | return ret; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 92 | } |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 93 | #else |
| 94 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
| 95 | const nodemask_t *mask) |
| 96 | { |
| 97 | return true; |
| 98 | } |
| 99 | #endif /* CONFIG_NUMA */ |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 100 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 101 | /* |
| 102 | * The process p may have detached its own ->mm while exiting or through |
| 103 | * use_mm(), but one or more of its subthreads may still have a valid |
| 104 | * pointer. Return p, or any of its subthreads with a valid ->mm, with |
| 105 | * task_lock() held. |
| 106 | */ |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 107 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 108 | { |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 109 | struct task_struct *t; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 110 | |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 111 | rcu_read_lock(); |
| 112 | |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 113 | for_each_thread(p, t) { |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 114 | task_lock(t); |
| 115 | if (likely(t->mm)) |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 116 | goto found; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 117 | task_unlock(t); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 118 | } |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 119 | t = NULL; |
| 120 | found: |
| 121 | rcu_read_unlock(); |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 122 | |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 123 | return t; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 124 | } |
| 125 | |
Yaowei Bai | db2a0dd | 2015-11-06 16:28:06 -0800 | [diff] [blame] | 126 | /* |
| 127 | * order == -1 means the oom kill is required by sysrq, otherwise only |
| 128 | * for display purposes. |
| 129 | */ |
| 130 | static inline bool is_sysrq_oom(struct oom_control *oc) |
| 131 | { |
| 132 | return oc->order == -1; |
| 133 | } |
| 134 | |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 135 | /* return true if the task is not adequate as candidate victim task. */ |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 136 | static bool oom_unkillable_task(struct task_struct *p, |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 137 | struct mem_cgroup *memcg, const nodemask_t *nodemask) |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 138 | { |
| 139 | if (is_global_init(p)) |
| 140 | return true; |
| 141 | if (p->flags & PF_KTHREAD) |
| 142 | return true; |
| 143 | |
| 144 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 145 | if (memcg && !task_in_mem_cgroup(p, memcg)) |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 146 | return true; |
| 147 | |
| 148 | /* p may not have freeable memory in nodemask */ |
| 149 | if (!has_intersects_mems_allowed(p, nodemask)) |
| 150 | return true; |
| 151 | |
| 152 | return false; |
| 153 | } |
| 154 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | /** |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 156 | * oom_badness - heuristic function to determine which candidate task to kill |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | * @p: task struct of which task we should calculate |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 158 | * @totalpages: total present RAM allowed for page allocation |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 159 | * |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 160 | * The heuristic for determining which task to kill is made to be as simple and |
| 161 | * predictable as possible. The goal is to return the highest value for the |
| 162 | * task consuming the most memory to avoid subsequent oom failures. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 163 | */ |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 164 | unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg, |
| 165 | const nodemask_t *nodemask, unsigned long totalpages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 166 | { |
David Rientjes | 1e11ad8 | 2012-06-08 13:21:26 -0700 | [diff] [blame] | 167 | long points; |
David Rientjes | 61eafb0 | 2012-06-20 12:52:58 -0700 | [diff] [blame] | 168 | long adj; |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 169 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 170 | if (oom_unkillable_task(p, memcg, nodemask)) |
KOSAKI Motohiro | 26ebc98 | 2010-08-09 17:19:37 -0700 | [diff] [blame] | 171 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 173 | p = find_lock_task_mm(p); |
| 174 | if (!p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | return 0; |
| 176 | |
David Rientjes | a9c58b90 | 2012-12-11 16:02:54 -0800 | [diff] [blame] | 177 | adj = (long)p->signal->oom_score_adj; |
David Rientjes | 61eafb0 | 2012-06-20 12:52:58 -0700 | [diff] [blame] | 178 | if (adj == OOM_SCORE_ADJ_MIN) { |
Michal Hocko | 5aecc85 | 2011-11-15 14:36:07 -0800 | [diff] [blame] | 179 | task_unlock(p); |
| 180 | return 0; |
| 181 | } |
| 182 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 183 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 184 | * 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] | 185 | * task's rss, pagetable and swap space use. |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 186 | */ |
Kirill A. Shutemov | dc6c9a3 | 2015-02-11 15:26:50 -0800 | [diff] [blame] | 187 | points = get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS) + |
| 188 | atomic_long_read(&p->mm->nr_ptes) + mm_nr_pmds(p->mm); |
Andrew Morton | 97c2c9b8 | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 189 | task_unlock(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | |
| 191 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 192 | * Root processes get 3% bonus, just like the __vm_enough_memory() |
| 193 | * implementation used by LSMs. |
Hugh Dickins | 7ba3485 | 2007-01-05 16:37:03 -0800 | [diff] [blame] | 194 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 195 | if (has_capability_noaudit(p, CAP_SYS_ADMIN)) |
David Rientjes | 778c14a | 2014-01-30 15:46:11 -0800 | [diff] [blame] | 196 | points -= (points * 3) / 100; |
Hugh Dickins | 7ba3485 | 2007-01-05 16:37:03 -0800 | [diff] [blame] | 197 | |
David Rientjes | 61eafb0 | 2012-06-20 12:52:58 -0700 | [diff] [blame] | 198 | /* Normalize to oom_score_adj units */ |
| 199 | adj *= totalpages / 1000; |
| 200 | points += adj; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 201 | |
David Rientjes | f19e8aa | 2010-09-22 13:04:52 -0700 | [diff] [blame] | 202 | /* |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 203 | * Never return 0 for an eligible task regardless of the root bonus and |
| 204 | * 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] | 205 | */ |
David Rientjes | 1e11ad8 | 2012-06-08 13:21:26 -0700 | [diff] [blame] | 206 | return points > 0 ? points : 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 207 | } |
| 208 | |
| 209 | /* |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 210 | * Determine the type of allocation constraint. |
| 211 | */ |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 212 | #ifdef CONFIG_NUMA |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 213 | static enum oom_constraint constrained_alloc(struct oom_control *oc, |
| 214 | unsigned long *totalpages) |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 215 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 216 | struct zone *zone; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 217 | struct zoneref *z; |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 218 | enum zone_type high_zoneidx = gfp_zone(oc->gfp_mask); |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 219 | bool cpuset_limited = false; |
| 220 | int nid; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 221 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 222 | /* Default to all available memory */ |
| 223 | *totalpages = totalram_pages + total_swap_pages; |
| 224 | |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 225 | if (!oc->zonelist) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 226 | return CONSTRAINT_NONE; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 227 | /* |
| 228 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid |
| 229 | * to kill current.We have to random task kill in this case. |
| 230 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. |
| 231 | */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 232 | if (oc->gfp_mask & __GFP_THISNODE) |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 233 | return CONSTRAINT_NONE; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 234 | |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 235 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 236 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
| 237 | * the page allocator means a mempolicy is in effect. Cpuset policy |
| 238 | * is enforced in get_page_from_freelist(). |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 239 | */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 240 | if (oc->nodemask && |
| 241 | !nodes_subset(node_states[N_MEMORY], *oc->nodemask)) { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 242 | *totalpages = total_swap_pages; |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 243 | for_each_node_mask(nid, *oc->nodemask) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 244 | *totalpages += node_spanned_pages(nid); |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 245 | return CONSTRAINT_MEMORY_POLICY; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 246 | } |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 247 | |
| 248 | /* Check this allocation failure is caused by cpuset's wall function */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 249 | for_each_zone_zonelist_nodemask(zone, z, oc->zonelist, |
| 250 | high_zoneidx, oc->nodemask) |
| 251 | if (!cpuset_zone_allowed(zone, oc->gfp_mask)) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 252 | cpuset_limited = true; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 253 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 254 | if (cpuset_limited) { |
| 255 | *totalpages = total_swap_pages; |
| 256 | for_each_node_mask(nid, cpuset_current_mems_allowed) |
| 257 | *totalpages += node_spanned_pages(nid); |
| 258 | return CONSTRAINT_CPUSET; |
| 259 | } |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 260 | return CONSTRAINT_NONE; |
| 261 | } |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 262 | #else |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 263 | static enum oom_constraint constrained_alloc(struct oom_control *oc, |
| 264 | unsigned long *totalpages) |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 265 | { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 266 | *totalpages = totalram_pages + total_swap_pages; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 267 | return CONSTRAINT_NONE; |
| 268 | } |
| 269 | #endif |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 270 | |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 271 | enum oom_scan_t oom_scan_process_thread(struct oom_control *oc, |
| 272 | struct task_struct *task, unsigned long totalpages) |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 273 | { |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 274 | if (oom_unkillable_task(task, NULL, oc->nodemask)) |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 275 | return OOM_SCAN_CONTINUE; |
| 276 | |
| 277 | /* |
| 278 | * This task already has access to memory reserves and is being killed. |
| 279 | * Don't allow any other task to have access to the reserves. |
| 280 | */ |
| 281 | if (test_tsk_thread_flag(task, TIF_MEMDIE)) { |
Yaowei Bai | db2a0dd | 2015-11-06 16:28:06 -0800 | [diff] [blame] | 282 | if (!is_sysrq_oom(oc)) |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 283 | return OOM_SCAN_ABORT; |
| 284 | } |
| 285 | if (!task->mm) |
| 286 | return OOM_SCAN_CONTINUE; |
| 287 | |
David Rientjes | e1e12d2 | 2012-12-11 16:02:56 -0800 | [diff] [blame] | 288 | /* |
| 289 | * If task is allocating a lot of memory and has been marked to be |
| 290 | * killed first if it triggers an oom, then select it. |
| 291 | */ |
| 292 | if (oom_task_origin(task)) |
| 293 | return OOM_SCAN_SELECT; |
| 294 | |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 295 | return OOM_SCAN_OK; |
| 296 | } |
| 297 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 298 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | * Simple selection loop. We chose the process with the highest |
Rusty Russell | 6b4f2b5 | 2013-07-15 11:24:08 +0930 | [diff] [blame] | 300 | * number of 'points'. Returns -1 on scan abort. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 301 | */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 302 | static struct task_struct *select_bad_process(struct oom_control *oc, |
| 303 | unsigned int *ppoints, unsigned long totalpages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 304 | { |
David Rientjes | 3a5dda7 | 2011-03-22 16:30:09 -0700 | [diff] [blame] | 305 | struct task_struct *g, *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 306 | struct task_struct *chosen = NULL; |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 307 | unsigned long chosen_points = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 308 | |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 309 | rcu_read_lock(); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 310 | for_each_process_thread(g, p) { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 311 | unsigned int points; |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 312 | |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 313 | switch (oom_scan_process_thread(oc, p, totalpages)) { |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 314 | case OOM_SCAN_SELECT: |
| 315 | chosen = p; |
| 316 | chosen_points = ULONG_MAX; |
| 317 | /* fall through */ |
| 318 | case OOM_SCAN_CONTINUE: |
Andrey Vagin | 30e2b41 | 2011-03-22 16:30:11 -0700 | [diff] [blame] | 319 | continue; |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 320 | case OOM_SCAN_ABORT: |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 321 | rcu_read_unlock(); |
Rusty Russell | 6b4f2b5 | 2013-07-15 11:24:08 +0930 | [diff] [blame] | 322 | return (struct task_struct *)(-1UL); |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 323 | case OOM_SCAN_OK: |
| 324 | break; |
| 325 | }; |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 326 | points = oom_badness(p, NULL, oc->nodemask, totalpages); |
David Rientjes | d49ad93 | 2014-01-23 15:53:34 -0800 | [diff] [blame] | 327 | if (!points || points < chosen_points) |
| 328 | continue; |
| 329 | /* Prefer thread group leaders for display purposes */ |
| 330 | if (points == chosen_points && thread_group_leader(chosen)) |
| 331 | continue; |
| 332 | |
| 333 | chosen = p; |
| 334 | chosen_points = points; |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 335 | } |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 336 | if (chosen) |
| 337 | get_task_struct(chosen); |
| 338 | rcu_read_unlock(); |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 339 | |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 340 | *ppoints = chosen_points * 1000 / totalpages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 341 | return chosen; |
| 342 | } |
| 343 | |
| 344 | /** |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 345 | * dump_tasks - dump current memory state of all system tasks |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 346 | * @memcg: current's memory controller, if constrained |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 347 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 348 | * |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 349 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
| 350 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes |
| 351 | * are not shown. |
David Rientjes | de34d96 | 2012-07-31 16:42:56 -0700 | [diff] [blame] | 352 | * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes, |
| 353 | * swapents, oom_score_adj value, and name. |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 354 | */ |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 355 | static void dump_tasks(struct mem_cgroup *memcg, const nodemask_t *nodemask) |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 356 | { |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 357 | struct task_struct *p; |
| 358 | struct task_struct *task; |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 359 | |
Kirill A. Shutemov | dc6c9a3 | 2015-02-11 15:26:50 -0800 | [diff] [blame] | 360 | pr_info("[ pid ] uid tgid total_vm rss nr_ptes nr_pmds swapents oom_score_adj name\n"); |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 361 | rcu_read_lock(); |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 362 | for_each_process(p) { |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 363 | if (oom_unkillable_task(p, memcg, nodemask)) |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 364 | continue; |
| 365 | |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 366 | task = find_lock_task_mm(p); |
| 367 | if (!task) { |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 368 | /* |
David Rientjes | 74ab7f1 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 369 | * This is a kthread or all of p's threads have already |
| 370 | * detached their mm's. There's no need to report |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 371 | * them; they can't be oom killed anyway. |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 372 | */ |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 373 | continue; |
| 374 | } |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 375 | |
Kirill A. Shutemov | dc6c9a3 | 2015-02-11 15:26:50 -0800 | [diff] [blame] | 376 | pr_info("[%5d] %5d %5d %8lu %8lu %7ld %7ld %8lu %5hd %s\n", |
Eric W. Biederman | 078de5f | 2012-02-08 07:00:08 -0800 | [diff] [blame] | 377 | task->pid, from_kuid(&init_user_ns, task_uid(task)), |
| 378 | task->tgid, task->mm->total_vm, get_mm_rss(task->mm), |
Kirill A. Shutemov | e1f56c8 | 2013-11-14 14:30:48 -0800 | [diff] [blame] | 379 | atomic_long_read(&task->mm->nr_ptes), |
Kirill A. Shutemov | dc6c9a3 | 2015-02-11 15:26:50 -0800 | [diff] [blame] | 380 | mm_nr_pmds(task->mm), |
David Rientjes | de34d96 | 2012-07-31 16:42:56 -0700 | [diff] [blame] | 381 | get_mm_counter(task->mm, MM_SWAPENTS), |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 382 | task->signal->oom_score_adj, task->comm); |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 383 | task_unlock(task); |
| 384 | } |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 385 | rcu_read_unlock(); |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 386 | } |
| 387 | |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 388 | static void dump_header(struct oom_control *oc, struct task_struct *p, |
| 389 | struct mem_cgroup *memcg) |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 390 | { |
Joe Perches | 756a025 | 2016-03-17 14:19:47 -0700 | [diff] [blame] | 391 | pr_warn("%s invoked oom-killer: gfp_mask=%#x(%pGg), order=%d, oom_score_adj=%hd\n", |
Vlastimil Babka | a0795cd | 2016-03-15 14:56:05 -0700 | [diff] [blame] | 392 | current->comm, oc->gfp_mask, &oc->gfp_mask, oc->order, |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 393 | current->signal->oom_score_adj); |
Vlastimil Babka | a0795cd | 2016-03-15 14:56:05 -0700 | [diff] [blame] | 394 | |
David Rientjes | da39da3 | 2015-11-05 18:48:05 -0800 | [diff] [blame] | 395 | cpuset_print_current_mems_allowed(); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 396 | dump_stack(); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 397 | if (memcg) |
| 398 | mem_cgroup_print_oom_info(memcg, p); |
| 399 | else |
| 400 | show_mem(SHOW_MEM_FILTER_NODES); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 401 | if (sysctl_oom_dump_tasks) |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 402 | dump_tasks(memcg, oc->nodemask); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 403 | } |
| 404 | |
Michal Hocko | 5695be1 | 2014-10-20 18:12:32 +0200 | [diff] [blame] | 405 | /* |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 406 | * Number of OOM victims in flight |
Michal Hocko | 5695be1 | 2014-10-20 18:12:32 +0200 | [diff] [blame] | 407 | */ |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 408 | static atomic_t oom_victims = ATOMIC_INIT(0); |
| 409 | static DECLARE_WAIT_QUEUE_HEAD(oom_victims_wait); |
Michal Hocko | 5695be1 | 2014-10-20 18:12:32 +0200 | [diff] [blame] | 410 | |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 411 | bool oom_killer_disabled __read_mostly; |
Michal Hocko | 5695be1 | 2014-10-20 18:12:32 +0200 | [diff] [blame] | 412 | |
Michal Hocko | bc448e8 | 2016-03-25 14:20:30 -0700 | [diff] [blame] | 413 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
| 414 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 415 | #ifdef CONFIG_MMU |
| 416 | /* |
| 417 | * OOM Reaper kernel thread which tries to reap the memory used by the OOM |
| 418 | * victim (if that is possible) to help the OOM killer to move on. |
| 419 | */ |
| 420 | static struct task_struct *oom_reaper_th; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 421 | static DECLARE_WAIT_QUEUE_HEAD(oom_reaper_wait); |
Vladimir Davydov | 29c696e | 2016-03-25 14:20:39 -0700 | [diff] [blame] | 422 | static struct task_struct *oom_reaper_list; |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 423 | static DEFINE_SPINLOCK(oom_reaper_lock); |
| 424 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 425 | |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 426 | static bool __oom_reap_task(struct task_struct *tsk) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 427 | { |
| 428 | struct mmu_gather tlb; |
| 429 | struct vm_area_struct *vma; |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 430 | struct mm_struct *mm; |
| 431 | struct task_struct *p; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 432 | struct zap_details details = {.check_swap_entries = true, |
| 433 | .ignore_dirty = true}; |
| 434 | bool ret = true; |
| 435 | |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 436 | /* |
| 437 | * Make sure we find the associated mm_struct even when the particular |
| 438 | * thread has already terminated and cleared its mm. |
| 439 | * We might have race with exit path so consider our work done if there |
| 440 | * is no mm. |
| 441 | */ |
| 442 | p = find_lock_task_mm(tsk); |
| 443 | if (!p) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 444 | return true; |
| 445 | |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 446 | mm = p->mm; |
| 447 | if (!atomic_inc_not_zero(&mm->mm_users)) { |
| 448 | task_unlock(p); |
| 449 | return true; |
| 450 | } |
| 451 | |
| 452 | task_unlock(p); |
| 453 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 454 | if (!down_read_trylock(&mm->mmap_sem)) { |
| 455 | ret = false; |
| 456 | goto out; |
| 457 | } |
| 458 | |
| 459 | tlb_gather_mmu(&tlb, mm, 0, -1); |
| 460 | for (vma = mm->mmap ; vma; vma = vma->vm_next) { |
| 461 | if (is_vm_hugetlb_page(vma)) |
| 462 | continue; |
| 463 | |
| 464 | /* |
| 465 | * mlocked VMAs require explicit munlocking before unmap. |
| 466 | * Let's keep it simple here and skip such VMAs. |
| 467 | */ |
| 468 | if (vma->vm_flags & VM_LOCKED) |
| 469 | continue; |
| 470 | |
| 471 | /* |
| 472 | * Only anonymous pages have a good chance to be dropped |
| 473 | * without additional steps which we cannot afford as we |
| 474 | * are OOM already. |
| 475 | * |
| 476 | * We do not even care about fs backed pages because all |
| 477 | * which are reclaimable have already been reclaimed and |
| 478 | * we do not want to block exit_mmap by keeping mm ref |
| 479 | * count elevated without a good reason. |
| 480 | */ |
| 481 | if (vma_is_anonymous(vma) || !(vma->vm_flags & VM_SHARED)) |
| 482 | unmap_page_range(&tlb, vma, vma->vm_start, vma->vm_end, |
| 483 | &details); |
| 484 | } |
| 485 | tlb_finish_mmu(&tlb, 0, -1); |
Michal Hocko | bc448e8 | 2016-03-25 14:20:30 -0700 | [diff] [blame] | 486 | pr_info("oom_reaper: reaped process %d (%s), now anon-rss:%lukB, file-rss:%lukB, shmem-rss:%lukB\n", |
| 487 | task_pid_nr(tsk), tsk->comm, |
| 488 | K(get_mm_counter(mm, MM_ANONPAGES)), |
| 489 | K(get_mm_counter(mm, MM_FILEPAGES)), |
| 490 | K(get_mm_counter(mm, MM_SHMEMPAGES))); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 491 | up_read(&mm->mmap_sem); |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 492 | |
| 493 | /* |
| 494 | * Clear TIF_MEMDIE because the task shouldn't be sitting on a |
| 495 | * reasonably reclaimable memory anymore. OOM killer can continue |
| 496 | * by selecting other victim if unmapping hasn't led to any |
| 497 | * improvements. This also means that selecting this task doesn't |
| 498 | * make any sense. |
| 499 | */ |
| 500 | tsk->signal->oom_score_adj = OOM_SCORE_ADJ_MIN; |
| 501 | exit_oom_victim(tsk); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 502 | out: |
| 503 | mmput(mm); |
| 504 | return ret; |
| 505 | } |
| 506 | |
Michal Hocko | bc448e8 | 2016-03-25 14:20:30 -0700 | [diff] [blame] | 507 | #define MAX_OOM_REAP_RETRIES 10 |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 508 | static void oom_reap_task(struct task_struct *tsk) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 509 | { |
| 510 | int attempts = 0; |
| 511 | |
| 512 | /* Retry the down_read_trylock(mmap_sem) a few times */ |
Michal Hocko | bc448e8 | 2016-03-25 14:20:30 -0700 | [diff] [blame] | 513 | while (attempts++ < MAX_OOM_REAP_RETRIES && !__oom_reap_task(tsk)) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 514 | schedule_timeout_idle(HZ/10); |
| 515 | |
Michal Hocko | bc448e8 | 2016-03-25 14:20:30 -0700 | [diff] [blame] | 516 | if (attempts > MAX_OOM_REAP_RETRIES) { |
| 517 | pr_info("oom_reaper: unable to reap pid:%d (%s)\n", |
| 518 | task_pid_nr(tsk), tsk->comm); |
| 519 | debug_show_all_locks(); |
| 520 | } |
| 521 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 522 | /* Drop a reference taken by wake_oom_reaper */ |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 523 | put_task_struct(tsk); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 524 | } |
| 525 | |
| 526 | static int oom_reaper(void *unused) |
| 527 | { |
Michal Hocko | e267960 | 2016-03-25 14:20:41 -0700 | [diff] [blame] | 528 | set_freezable(); |
| 529 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 530 | while (true) { |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 531 | struct task_struct *tsk = NULL; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 532 | |
Vladimir Davydov | 29c696e | 2016-03-25 14:20:39 -0700 | [diff] [blame] | 533 | wait_event_freezable(oom_reaper_wait, oom_reaper_list != NULL); |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 534 | spin_lock(&oom_reaper_lock); |
Vladimir Davydov | 29c696e | 2016-03-25 14:20:39 -0700 | [diff] [blame] | 535 | if (oom_reaper_list != NULL) { |
| 536 | tsk = oom_reaper_list; |
| 537 | oom_reaper_list = tsk->oom_reaper_list; |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 538 | } |
| 539 | spin_unlock(&oom_reaper_lock); |
| 540 | |
| 541 | if (tsk) |
| 542 | oom_reap_task(tsk); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 543 | } |
| 544 | |
| 545 | return 0; |
| 546 | } |
| 547 | |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 548 | static void wake_oom_reaper(struct task_struct *tsk) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 549 | { |
Tetsuo Handa | bb29902 | 2016-03-25 14:20:44 -0700 | [diff] [blame^] | 550 | if (!oom_reaper_th || tsk->oom_reaper_list) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 551 | return; |
| 552 | |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 553 | get_task_struct(tsk); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 554 | |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 555 | spin_lock(&oom_reaper_lock); |
Vladimir Davydov | 29c696e | 2016-03-25 14:20:39 -0700 | [diff] [blame] | 556 | tsk->oom_reaper_list = oom_reaper_list; |
| 557 | oom_reaper_list = tsk; |
Michal Hocko | 0304926 | 2016-03-25 14:20:33 -0700 | [diff] [blame] | 558 | spin_unlock(&oom_reaper_lock); |
| 559 | wake_up(&oom_reaper_wait); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 560 | } |
| 561 | |
| 562 | static int __init oom_init(void) |
| 563 | { |
| 564 | oom_reaper_th = kthread_run(oom_reaper, NULL, "oom_reaper"); |
| 565 | if (IS_ERR(oom_reaper_th)) { |
| 566 | pr_err("Unable to start OOM reaper %ld. Continuing regardless\n", |
| 567 | PTR_ERR(oom_reaper_th)); |
| 568 | oom_reaper_th = NULL; |
| 569 | } |
| 570 | return 0; |
| 571 | } |
| 572 | subsys_initcall(oom_init) |
| 573 | #else |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 574 | static void wake_oom_reaper(struct task_struct *tsk) |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 575 | { |
| 576 | } |
| 577 | #endif |
| 578 | |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 579 | /** |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 580 | * mark_oom_victim - mark the given task as OOM victim |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 581 | * @tsk: task to mark |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 582 | * |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 583 | * Has to be called with oom_lock held and never after |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 584 | * oom has been disabled already. |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 585 | */ |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 586 | void mark_oom_victim(struct task_struct *tsk) |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 587 | { |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 588 | WARN_ON(oom_killer_disabled); |
| 589 | /* OOM killer might race with memcg OOM */ |
| 590 | if (test_and_set_tsk_thread_flag(tsk, TIF_MEMDIE)) |
| 591 | return; |
Michal Hocko | 63a8ca9 | 2015-02-11 15:26:15 -0800 | [diff] [blame] | 592 | /* |
| 593 | * Make sure that the task is woken up from uninterruptible sleep |
| 594 | * if it is frozen because OOM killer wouldn't be able to free |
| 595 | * any memory and livelock. freezing_slow_path will tell the freezer |
| 596 | * that TIF_MEMDIE tasks should be ignored. |
| 597 | */ |
| 598 | __thaw_task(tsk); |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 599 | atomic_inc(&oom_victims); |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 600 | } |
| 601 | |
| 602 | /** |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 603 | * exit_oom_victim - note the exit of an OOM victim |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 604 | */ |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 605 | void exit_oom_victim(struct task_struct *tsk) |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 606 | { |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 607 | if (!test_and_clear_tsk_thread_flag(tsk, TIF_MEMDIE)) |
| 608 | return; |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 609 | |
Johannes Weiner | c38f102 | 2015-06-24 16:57:13 -0700 | [diff] [blame] | 610 | if (!atomic_dec_return(&oom_victims)) |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 611 | wake_up_all(&oom_victims_wait); |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 612 | } |
| 613 | |
| 614 | /** |
| 615 | * oom_killer_disable - disable OOM killer |
| 616 | * |
| 617 | * Forces all page allocations to fail rather than trigger OOM killer. |
| 618 | * Will block and wait until all OOM victims are killed. |
| 619 | * |
| 620 | * The function cannot be called when there are runnable user tasks because |
| 621 | * the userspace would see unexpected allocation failures as a result. Any |
| 622 | * new usage of this function should be consulted with MM people. |
| 623 | * |
| 624 | * Returns true if successful and false if the OOM killer cannot be |
| 625 | * disabled. |
| 626 | */ |
| 627 | bool oom_killer_disable(void) |
| 628 | { |
| 629 | /* |
Tetsuo Handa | 6afcf28 | 2016-03-17 14:20:45 -0700 | [diff] [blame] | 630 | * Make sure to not race with an ongoing OOM killer. Check that the |
| 631 | * current is not killed (possibly due to sharing the victim's memory). |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 632 | */ |
Tetsuo Handa | 6afcf28 | 2016-03-17 14:20:45 -0700 | [diff] [blame] | 633 | if (mutex_lock_killable(&oom_lock)) |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 634 | return false; |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 635 | oom_killer_disabled = true; |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 636 | mutex_unlock(&oom_lock); |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 637 | |
| 638 | wait_event(oom_victims_wait, !atomic_read(&oom_victims)); |
| 639 | |
| 640 | return true; |
| 641 | } |
| 642 | |
| 643 | /** |
| 644 | * oom_killer_enable - enable OOM killer |
| 645 | */ |
| 646 | void oom_killer_enable(void) |
| 647 | { |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 648 | oom_killer_disabled = false; |
Michal Hocko | 49550b6 | 2015-02-11 15:26:12 -0800 | [diff] [blame] | 649 | } |
| 650 | |
Oleg Nesterov | 4d7b339 | 2015-11-05 18:48:26 -0800 | [diff] [blame] | 651 | /* |
| 652 | * task->mm can be NULL if the task is the exited group leader. So to |
| 653 | * determine whether the task is using a particular mm, we examine all the |
| 654 | * task's threads: if one of those is using this mm then this task was also |
| 655 | * using it. |
| 656 | */ |
| 657 | static bool process_shares_mm(struct task_struct *p, struct mm_struct *mm) |
| 658 | { |
| 659 | struct task_struct *t; |
| 660 | |
| 661 | for_each_thread(p, t) { |
| 662 | struct mm_struct *t_mm = READ_ONCE(t->mm); |
| 663 | if (t_mm) |
| 664 | return t_mm == mm; |
| 665 | } |
| 666 | return false; |
| 667 | } |
| 668 | |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 669 | /* |
| 670 | * Must be called while holding a reference to p, which will be released upon |
| 671 | * returning. |
| 672 | */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 673 | void oom_kill_process(struct oom_control *oc, struct task_struct *p, |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 674 | unsigned int points, unsigned long totalpages, |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 675 | struct mem_cgroup *memcg, const char *message) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 676 | { |
Linus Torvalds | 52d3c03 | 2011-03-14 15:17:07 -0700 | [diff] [blame] | 677 | struct task_struct *victim = p; |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 678 | struct task_struct *child; |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 679 | struct task_struct *t; |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 680 | struct mm_struct *mm; |
Linus Torvalds | 52d3c03 | 2011-03-14 15:17:07 -0700 | [diff] [blame] | 681 | unsigned int victim_points = 0; |
David Rientjes | dc3f21e | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 682 | static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
| 683 | DEFAULT_RATELIMIT_BURST); |
Tetsuo Handa | bb29902 | 2016-03-25 14:20:44 -0700 | [diff] [blame^] | 684 | bool can_oom_reap = true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 685 | |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 686 | /* |
| 687 | * If the task is already exiting, don't alarm the sysadmin or kill |
| 688 | * its children or threads, just set TIF_MEMDIE so it can die quickly |
| 689 | */ |
Michal Hocko | 83363b9 | 2015-02-11 15:24:56 -0800 | [diff] [blame] | 690 | task_lock(p); |
| 691 | if (p->mm && task_will_free_mem(p)) { |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 692 | mark_oom_victim(p); |
Michal Hocko | 83363b9 | 2015-02-11 15:24:56 -0800 | [diff] [blame] | 693 | task_unlock(p); |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 694 | put_task_struct(p); |
David Rientjes | 2a1c9b1 | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 695 | return; |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 696 | } |
Michal Hocko | 83363b9 | 2015-02-11 15:24:56 -0800 | [diff] [blame] | 697 | task_unlock(p); |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 698 | |
David Rientjes | dc3f21e | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 699 | if (__ratelimit(&oom_rs)) |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 700 | dump_header(oc, p, memcg); |
David Rientjes | 8447d95 | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 701 | |
Wang Long | f0d6647 | 2015-06-24 16:58:01 -0700 | [diff] [blame] | 702 | 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] | 703 | message, task_pid_nr(p), p->comm, points); |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 704 | |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 705 | /* |
| 706 | * 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] | 707 | * the one with the highest oom_badness() score is sacrificed for its |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 708 | * parent. This attempts to lose the minimal amount of work done while |
| 709 | * still freeing memory. |
| 710 | */ |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 711 | read_lock(&tasklist_lock); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 712 | for_each_thread(p, t) { |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 713 | list_for_each_entry(child, &t->children, sibling) { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 714 | unsigned int child_points; |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 715 | |
Oleg Nesterov | 4d7b339 | 2015-11-05 18:48:26 -0800 | [diff] [blame] | 716 | if (process_shares_mm(child, p->mm)) |
David Rientjes | edd4554 | 2011-03-22 16:30:12 -0700 | [diff] [blame] | 717 | continue; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 718 | /* |
| 719 | * oom_badness() returns 0 if the thread is unkillable |
| 720 | */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 721 | child_points = oom_badness(child, memcg, oc->nodemask, |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 722 | totalpages); |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 723 | if (child_points > victim_points) { |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 724 | put_task_struct(victim); |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 725 | victim = child; |
| 726 | victim_points = child_points; |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 727 | get_task_struct(victim); |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 728 | } |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 729 | } |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 730 | } |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 731 | read_unlock(&tasklist_lock); |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 732 | |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 733 | p = find_lock_task_mm(victim); |
| 734 | if (!p) { |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 735 | put_task_struct(victim); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 736 | return; |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 737 | } else if (victim != p) { |
| 738 | get_task_struct(p); |
| 739 | put_task_struct(victim); |
| 740 | victim = p; |
| 741 | } |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 742 | |
Tetsuo Handa | 880b768 | 2015-11-05 18:47:51 -0800 | [diff] [blame] | 743 | /* Get a reference to safely compare mm after task_unlock(victim) */ |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 744 | mm = victim->mm; |
Tetsuo Handa | 880b768 | 2015-11-05 18:47:51 -0800 | [diff] [blame] | 745 | atomic_inc(&mm->mm_count); |
Tetsuo Handa | 426fb5e | 2015-11-05 18:47:44 -0800 | [diff] [blame] | 746 | /* |
| 747 | * We should send SIGKILL before setting TIF_MEMDIE in order to prevent |
| 748 | * the OOM victim from depleting the memory reserves from the user |
| 749 | * space under its control. |
| 750 | */ |
| 751 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true); |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 752 | mark_oom_victim(victim); |
Jerome Marchand | eca56ff | 2016-01-14 15:19:26 -0800 | [diff] [blame] | 753 | 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] | 754 | task_pid_nr(victim), victim->comm, K(victim->mm->total_vm), |
| 755 | K(get_mm_counter(victim->mm, MM_ANONPAGES)), |
Jerome Marchand | eca56ff | 2016-01-14 15:19:26 -0800 | [diff] [blame] | 756 | K(get_mm_counter(victim->mm, MM_FILEPAGES)), |
| 757 | K(get_mm_counter(victim->mm, MM_SHMEMPAGES))); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 758 | task_unlock(victim); |
| 759 | |
| 760 | /* |
| 761 | * Kill all user processes sharing victim->mm in other thread groups, if |
| 762 | * any. They don't get access to memory reserves, though, to avoid |
| 763 | * depletion of all memory. This prevents mm->mmap_sem livelock when an |
| 764 | * oom killed thread cannot exit because it requires the semaphore and |
| 765 | * its contended by another thread trying to allocate memory itself. |
| 766 | * That thread will now get access to memory reserves since it has a |
| 767 | * pending fatal signal. |
| 768 | */ |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 769 | rcu_read_lock(); |
Oleg Nesterov | c319025 | 2015-11-05 18:48:23 -0800 | [diff] [blame] | 770 | for_each_process(p) { |
Oleg Nesterov | 4d7b339 | 2015-11-05 18:48:26 -0800 | [diff] [blame] | 771 | if (!process_shares_mm(p, mm)) |
Oleg Nesterov | c319025 | 2015-11-05 18:48:23 -0800 | [diff] [blame] | 772 | continue; |
| 773 | if (same_thread_group(p, victim)) |
| 774 | continue; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 775 | if (unlikely(p->flags & PF_KTHREAD) || is_global_init(p) || |
| 776 | p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) { |
| 777 | /* |
| 778 | * We cannot use oom_reaper for the mm shared by this |
| 779 | * process because it wouldn't get killed and so the |
| 780 | * memory might be still used. |
| 781 | */ |
| 782 | can_oom_reap = false; |
Oleg Nesterov | c319025 | 2015-11-05 18:48:23 -0800 | [diff] [blame] | 783 | continue; |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 784 | } |
Oleg Nesterov | c319025 | 2015-11-05 18:48:23 -0800 | [diff] [blame] | 785 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true); |
| 786 | } |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 787 | rcu_read_unlock(); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 788 | |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 789 | if (can_oom_reap) |
Michal Hocko | 36324a9 | 2016-03-25 14:20:27 -0700 | [diff] [blame] | 790 | wake_oom_reaper(victim); |
Michal Hocko | aac4536 | 2016-03-25 14:20:24 -0700 | [diff] [blame] | 791 | |
Tetsuo Handa | 880b768 | 2015-11-05 18:47:51 -0800 | [diff] [blame] | 792 | mmdrop(mm); |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 793 | put_task_struct(victim); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 794 | } |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 795 | #undef K |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 796 | |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 797 | /* |
| 798 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. |
| 799 | */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 800 | void check_panic_on_oom(struct oom_control *oc, enum oom_constraint constraint, |
Balasubramani Vivekanandan | 2415b9f | 2015-04-14 15:48:18 -0700 | [diff] [blame] | 801 | struct mem_cgroup *memcg) |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 802 | { |
| 803 | if (likely(!sysctl_panic_on_oom)) |
| 804 | return; |
| 805 | if (sysctl_panic_on_oom != 2) { |
| 806 | /* |
| 807 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel |
| 808 | * does not panic for cpuset, mempolicy, or memcg allocation |
| 809 | * failures. |
| 810 | */ |
| 811 | if (constraint != CONSTRAINT_NONE) |
| 812 | return; |
| 813 | } |
David Rientjes | 071a4be | 2015-09-08 15:00:42 -0700 | [diff] [blame] | 814 | /* Do not panic for oom kills triggered by sysrq */ |
Yaowei Bai | db2a0dd | 2015-11-06 16:28:06 -0800 | [diff] [blame] | 815 | if (is_sysrq_oom(oc)) |
David Rientjes | 071a4be | 2015-09-08 15:00:42 -0700 | [diff] [blame] | 816 | return; |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 817 | dump_header(oc, NULL, memcg); |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 818 | panic("Out of memory: %s panic_on_oom is enabled\n", |
| 819 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); |
| 820 | } |
| 821 | |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 822 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
| 823 | |
| 824 | int register_oom_notifier(struct notifier_block *nb) |
| 825 | { |
| 826 | return blocking_notifier_chain_register(&oom_notify_list, nb); |
| 827 | } |
| 828 | EXPORT_SYMBOL_GPL(register_oom_notifier); |
| 829 | |
| 830 | int unregister_oom_notifier(struct notifier_block *nb) |
| 831 | { |
| 832 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); |
| 833 | } |
| 834 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); |
| 835 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 836 | /** |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 837 | * out_of_memory - kill the "best" process when we run out of memory |
| 838 | * @oc: pointer to struct oom_control |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 839 | * |
| 840 | * If we run out of memory, we have the choice between either |
| 841 | * killing a random task (bad), letting the system crash (worse) |
| 842 | * OR try to be smart about which process to kill. Note that we |
| 843 | * don't have to be perfect here, we just have to be good. |
| 844 | */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 845 | bool out_of_memory(struct oom_control *oc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 846 | { |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 847 | struct task_struct *p; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 848 | unsigned long totalpages; |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 849 | unsigned long freed = 0; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 850 | unsigned int uninitialized_var(points); |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 851 | enum oom_constraint constraint = CONSTRAINT_NONE; |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 852 | |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 853 | if (oom_killer_disabled) |
| 854 | return false; |
| 855 | |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 856 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); |
| 857 | if (freed > 0) |
| 858 | /* Got some memory back in the last second. */ |
David Rientjes | 75e8f8b | 2015-09-08 15:00:47 -0700 | [diff] [blame] | 859 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 860 | |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 861 | /* |
David Rientjes | 9ff4868 | 2012-12-11 16:01:30 -0800 | [diff] [blame] | 862 | * If current has a pending SIGKILL or is exiting, then automatically |
| 863 | * select it. The goal is to allow it to allocate so that it may |
| 864 | * quickly exit and free its memory. |
Tetsuo Handa | d7a94e7 | 2015-02-11 15:24:54 -0800 | [diff] [blame] | 865 | * |
| 866 | * But don't select if current has already released its mm and cleared |
| 867 | * TIF_MEMDIE flag at exit_mm(), otherwise an OOM livelock may occur. |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 868 | */ |
Tetsuo Handa | d7a94e7 | 2015-02-11 15:24:54 -0800 | [diff] [blame] | 869 | if (current->mm && |
| 870 | (fatal_signal_pending(current) || task_will_free_mem(current))) { |
Johannes Weiner | 16e9519 | 2015-06-24 16:57:07 -0700 | [diff] [blame] | 871 | mark_oom_victim(current); |
David Rientjes | 75e8f8b | 2015-09-08 15:00:47 -0700 | [diff] [blame] | 872 | return true; |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 873 | } |
| 874 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 875 | /* |
| 876 | * Check if there were limitations on the allocation (only relevant for |
| 877 | * NUMA) that may require different handling. |
| 878 | */ |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 879 | constraint = constrained_alloc(oc, &totalpages); |
| 880 | if (constraint != CONSTRAINT_MEMORY_POLICY) |
| 881 | oc->nodemask = NULL; |
| 882 | check_panic_on_oom(oc, constraint, NULL); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 883 | |
David Rientjes | 121d1ba | 2012-07-31 16:42:55 -0700 | [diff] [blame] | 884 | if (sysctl_oom_kill_allocating_task && current->mm && |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 885 | !oom_unkillable_task(current, NULL, oc->nodemask) && |
David Rientjes | 121d1ba | 2012-07-31 16:42:55 -0700 | [diff] [blame] | 886 | current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) { |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 887 | get_task_struct(current); |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 888 | oom_kill_process(oc, current, 0, totalpages, NULL, |
David Rientjes | 2a1c9b1 | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 889 | "Out of memory (oom_kill_allocating_task)"); |
David Rientjes | 75e8f8b | 2015-09-08 15:00:47 -0700 | [diff] [blame] | 890 | return true; |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 891 | } |
| 892 | |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 893 | p = select_bad_process(oc, &points, totalpages); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 894 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
Yaowei Bai | db2a0dd | 2015-11-06 16:28:06 -0800 | [diff] [blame] | 895 | if (!p && !is_sysrq_oom(oc)) { |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 896 | dump_header(oc, NULL, NULL); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 897 | panic("Out of memory and no killable processes...\n"); |
| 898 | } |
David Rientjes | 071a4be | 2015-09-08 15:00:42 -0700 | [diff] [blame] | 899 | if (p && p != (void *)-1UL) { |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 900 | oom_kill_process(oc, p, points, totalpages, NULL, |
| 901 | "Out of memory"); |
David Rientjes | 75e8f8b | 2015-09-08 15:00:47 -0700 | [diff] [blame] | 902 | /* |
| 903 | * Give the killed process a good chance to exit before trying |
| 904 | * to allocate memory again. |
| 905 | */ |
David Rientjes | 4f774b9 | 2012-07-31 16:42:37 -0700 | [diff] [blame] | 906 | schedule_timeout_killable(1); |
David Rientjes | 75e8f8b | 2015-09-08 15:00:47 -0700 | [diff] [blame] | 907 | } |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 908 | return true; |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 909 | } |
| 910 | |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 911 | /* |
| 912 | * The pagefault handler calls here because it is out of memory, so kill a |
David Rientjes | efacd02 | 2012-12-12 13:52:06 -0800 | [diff] [blame] | 913 | * memory-hogging task. If any populated zone has ZONE_OOM_LOCKED set, a |
| 914 | * parallel oom killing is already in progress so do nothing. |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 915 | */ |
| 916 | void pagefault_out_of_memory(void) |
| 917 | { |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 918 | struct oom_control oc = { |
| 919 | .zonelist = NULL, |
| 920 | .nodemask = NULL, |
| 921 | .gfp_mask = 0, |
| 922 | .order = 0, |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 923 | }; |
| 924 | |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 925 | if (mem_cgroup_oom_synchronize(true)) |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 926 | return; |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 927 | |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 928 | if (!mutex_trylock(&oom_lock)) |
| 929 | return; |
Michal Hocko | c32b3cb | 2015-02-11 15:26:24 -0800 | [diff] [blame] | 930 | |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 931 | if (!out_of_memory(&oc)) { |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 932 | /* |
| 933 | * There shouldn't be any user tasks runnable while the |
| 934 | * OOM killer is disabled, so the current task has to |
| 935 | * be a racing OOM victim for which oom_killer_disable() |
| 936 | * is waiting for. |
| 937 | */ |
| 938 | WARN_ON(test_thread_flag(TIF_MEMDIE)); |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 939 | } |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 940 | |
| 941 | mutex_unlock(&oom_lock); |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 942 | } |