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> |
KAMEZAWA Hiroyuki | 43d2b11 | 2012-01-10 15:08:09 -0800 | [diff] [blame] | 38 | |
| 39 | #define CREATE_TRACE_POINTS |
| 40 | #include <trace/events/oom.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 41 | |
KAMEZAWA Hiroyuki | fadd8fb | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 42 | int sysctl_panic_on_oom; |
David Rientjes | fe071d7 | 2007-10-16 23:25:56 -0700 | [diff] [blame] | 43 | int sysctl_oom_kill_allocating_task; |
David Rientjes | ad915c4 | 2010-08-09 17:18:53 -0700 | [diff] [blame] | 44 | int sysctl_oom_dump_tasks = 1; |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 45 | static DEFINE_SPINLOCK(zone_scan_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 46 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 47 | #ifdef CONFIG_NUMA |
| 48 | /** |
| 49 | * has_intersects_mems_allowed() - check task eligiblity for kill |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 50 | * @start: task struct of which task to consider |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 51 | * @mask: nodemask passed to page allocator for mempolicy ooms |
| 52 | * |
| 53 | * Task eligibility is determined by whether or not a candidate task, @tsk, |
| 54 | * shares the same mempolicy nodes as current if it is bound by such a policy |
| 55 | * 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] | 56 | */ |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 57 | static bool has_intersects_mems_allowed(struct task_struct *start, |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 58 | const nodemask_t *mask) |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 59 | { |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 60 | struct task_struct *tsk; |
| 61 | bool ret = false; |
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 | rcu_read_lock(); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 64 | for_each_thread(start, tsk) { |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 65 | if (mask) { |
| 66 | /* |
| 67 | * If this is a mempolicy constrained oom, tsk's |
| 68 | * cpuset is irrelevant. Only return true if its |
| 69 | * mempolicy intersects current, otherwise it may be |
| 70 | * needlessly killed. |
| 71 | */ |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 72 | ret = mempolicy_nodemask_intersects(tsk, mask); |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 73 | } else { |
| 74 | /* |
| 75 | * This is not a mempolicy constrained oom, so only |
| 76 | * check the mems of tsk's cpuset. |
| 77 | */ |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 78 | ret = cpuset_mems_allowed_intersects(current, tsk); |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 79 | } |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 80 | if (ret) |
| 81 | break; |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 82 | } |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 83 | rcu_read_unlock(); |
KOSAKI Motohiro | df1090a | 2010-08-09 17:19:39 -0700 | [diff] [blame] | 84 | |
Oleg Nesterov | ad96244 | 2014-01-21 15:50:00 -0800 | [diff] [blame] | 85 | return ret; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 86 | } |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 87 | #else |
| 88 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
| 89 | const nodemask_t *mask) |
| 90 | { |
| 91 | return true; |
| 92 | } |
| 93 | #endif /* CONFIG_NUMA */ |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 94 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 95 | /* |
| 96 | * The process p may have detached its own ->mm while exiting or through |
| 97 | * use_mm(), but one or more of its subthreads may still have a valid |
| 98 | * pointer. Return p, or any of its subthreads with a valid ->mm, with |
| 99 | * task_lock() held. |
| 100 | */ |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 101 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 102 | { |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 103 | struct task_struct *t; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 104 | |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 105 | rcu_read_lock(); |
| 106 | |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 107 | for_each_thread(p, t) { |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 108 | task_lock(t); |
| 109 | if (likely(t->mm)) |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 110 | goto found; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 111 | task_unlock(t); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 112 | } |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 113 | t = NULL; |
| 114 | found: |
| 115 | rcu_read_unlock(); |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 116 | |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 117 | return t; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 118 | } |
| 119 | |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 120 | /* return true if the task is not adequate as candidate victim task. */ |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 121 | static bool oom_unkillable_task(struct task_struct *p, |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 122 | const struct mem_cgroup *memcg, const nodemask_t *nodemask) |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 123 | { |
| 124 | if (is_global_init(p)) |
| 125 | return true; |
| 126 | if (p->flags & PF_KTHREAD) |
| 127 | return true; |
| 128 | |
| 129 | /* 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] | 130 | if (memcg && !task_in_mem_cgroup(p, memcg)) |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 131 | return true; |
| 132 | |
| 133 | /* p may not have freeable memory in nodemask */ |
| 134 | if (!has_intersects_mems_allowed(p, nodemask)) |
| 135 | return true; |
| 136 | |
| 137 | return false; |
| 138 | } |
| 139 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 140 | /** |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 141 | * oom_badness - heuristic function to determine which candidate task to kill |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | * @p: task struct of which task we should calculate |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 143 | * @totalpages: total present RAM allowed for page allocation |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | * |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 145 | * The heuristic for determining which task to kill is made to be as simple and |
| 146 | * predictable as possible. The goal is to return the highest value for the |
| 147 | * task consuming the most memory to avoid subsequent oom failures. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 148 | */ |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 149 | unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg, |
| 150 | const nodemask_t *nodemask, unsigned long totalpages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 151 | { |
David Rientjes | 1e11ad8 | 2012-06-08 13:21:26 -0700 | [diff] [blame] | 152 | long points; |
David Rientjes | 61eafb0 | 2012-06-20 12:52:58 -0700 | [diff] [blame] | 153 | long adj; |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 154 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 155 | if (oom_unkillable_task(p, memcg, nodemask)) |
KOSAKI Motohiro | 26ebc98 | 2010-08-09 17:19:37 -0700 | [diff] [blame] | 156 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 157 | |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 158 | p = find_lock_task_mm(p); |
| 159 | if (!p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 160 | return 0; |
| 161 | |
David Rientjes | a9c58b90 | 2012-12-11 16:02:54 -0800 | [diff] [blame] | 162 | adj = (long)p->signal->oom_score_adj; |
David Rientjes | 61eafb0 | 2012-06-20 12:52:58 -0700 | [diff] [blame] | 163 | if (adj == OOM_SCORE_ADJ_MIN) { |
Michal Hocko | 5aecc85 | 2011-11-15 14:36:07 -0800 | [diff] [blame] | 164 | task_unlock(p); |
| 165 | return 0; |
| 166 | } |
| 167 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 169 | * 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] | 170 | * task's rss, pagetable and swap space use. |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 171 | */ |
Kirill A. Shutemov | e1f56c8 | 2013-11-14 14:30:48 -0800 | [diff] [blame] | 172 | points = get_mm_rss(p->mm) + atomic_long_read(&p->mm->nr_ptes) + |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 173 | get_mm_counter(p->mm, MM_SWAPENTS); |
Andrew Morton | 97c2c9b8 | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 174 | task_unlock(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 175 | |
| 176 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 177 | * Root processes get 3% bonus, just like the __vm_enough_memory() |
| 178 | * implementation used by LSMs. |
Hugh Dickins | 7ba3485 | 2007-01-05 16:37:03 -0800 | [diff] [blame] | 179 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 180 | if (has_capability_noaudit(p, CAP_SYS_ADMIN)) |
David Rientjes | 778c14a | 2014-01-30 15:46:11 -0800 | [diff] [blame] | 181 | points -= (points * 3) / 100; |
Hugh Dickins | 7ba3485 | 2007-01-05 16:37:03 -0800 | [diff] [blame] | 182 | |
David Rientjes | 61eafb0 | 2012-06-20 12:52:58 -0700 | [diff] [blame] | 183 | /* Normalize to oom_score_adj units */ |
| 184 | adj *= totalpages / 1000; |
| 185 | points += adj; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | |
David Rientjes | f19e8aa | 2010-09-22 13:04:52 -0700 | [diff] [blame] | 187 | /* |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 188 | * Never return 0 for an eligible task regardless of the root bonus and |
| 189 | * 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] | 190 | */ |
David Rientjes | 1e11ad8 | 2012-06-08 13:21:26 -0700 | [diff] [blame] | 191 | return points > 0 ? points : 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | } |
| 193 | |
| 194 | /* |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 195 | * Determine the type of allocation constraint. |
| 196 | */ |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 197 | #ifdef CONFIG_NUMA |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 198 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 199 | gfp_t gfp_mask, nodemask_t *nodemask, |
| 200 | unsigned long *totalpages) |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 201 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 202 | struct zone *zone; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 203 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 204 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 205 | bool cpuset_limited = false; |
| 206 | int nid; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 207 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 208 | /* Default to all available memory */ |
| 209 | *totalpages = totalram_pages + total_swap_pages; |
| 210 | |
| 211 | if (!zonelist) |
| 212 | return CONSTRAINT_NONE; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 213 | /* |
| 214 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid |
| 215 | * to kill current.We have to random task kill in this case. |
| 216 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. |
| 217 | */ |
| 218 | if (gfp_mask & __GFP_THISNODE) |
| 219 | return CONSTRAINT_NONE; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 220 | |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 221 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 222 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
| 223 | * the page allocator means a mempolicy is in effect. Cpuset policy |
| 224 | * is enforced in get_page_from_freelist(). |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 225 | */ |
Lai Jiangshan | bd3a66c | 2012-12-12 13:51:28 -0800 | [diff] [blame] | 226 | if (nodemask && !nodes_subset(node_states[N_MEMORY], *nodemask)) { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 227 | *totalpages = total_swap_pages; |
| 228 | for_each_node_mask(nid, *nodemask) |
| 229 | *totalpages += node_spanned_pages(nid); |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 230 | return CONSTRAINT_MEMORY_POLICY; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 231 | } |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 232 | |
| 233 | /* Check this allocation failure is caused by cpuset's wall function */ |
| 234 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 235 | high_zoneidx, nodemask) |
| 236 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 237 | cpuset_limited = true; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 238 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 239 | if (cpuset_limited) { |
| 240 | *totalpages = total_swap_pages; |
| 241 | for_each_node_mask(nid, cpuset_current_mems_allowed) |
| 242 | *totalpages += node_spanned_pages(nid); |
| 243 | return CONSTRAINT_CPUSET; |
| 244 | } |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 245 | return CONSTRAINT_NONE; |
| 246 | } |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 247 | #else |
| 248 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 249 | gfp_t gfp_mask, nodemask_t *nodemask, |
| 250 | unsigned long *totalpages) |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 251 | { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 252 | *totalpages = totalram_pages + total_swap_pages; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 253 | return CONSTRAINT_NONE; |
| 254 | } |
| 255 | #endif |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 256 | |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 257 | enum oom_scan_t oom_scan_process_thread(struct task_struct *task, |
| 258 | unsigned long totalpages, const nodemask_t *nodemask, |
| 259 | bool force_kill) |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 260 | { |
| 261 | if (task->exit_state) |
| 262 | return OOM_SCAN_CONTINUE; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 263 | if (oom_unkillable_task(task, NULL, nodemask)) |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 264 | return OOM_SCAN_CONTINUE; |
| 265 | |
| 266 | /* |
| 267 | * This task already has access to memory reserves and is being killed. |
| 268 | * Don't allow any other task to have access to the reserves. |
| 269 | */ |
| 270 | if (test_tsk_thread_flag(task, TIF_MEMDIE)) { |
| 271 | if (unlikely(frozen(task))) |
| 272 | __thaw_task(task); |
| 273 | if (!force_kill) |
| 274 | return OOM_SCAN_ABORT; |
| 275 | } |
| 276 | if (!task->mm) |
| 277 | return OOM_SCAN_CONTINUE; |
| 278 | |
David Rientjes | e1e12d2 | 2012-12-11 16:02:56 -0800 | [diff] [blame] | 279 | /* |
| 280 | * If task is allocating a lot of memory and has been marked to be |
| 281 | * killed first if it triggers an oom, then select it. |
| 282 | */ |
| 283 | if (oom_task_origin(task)) |
| 284 | return OOM_SCAN_SELECT; |
| 285 | |
David Rientjes | 9ff4868 | 2012-12-11 16:01:30 -0800 | [diff] [blame] | 286 | if (task->flags & PF_EXITING && !force_kill) { |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 287 | /* |
David Rientjes | 9ff4868 | 2012-12-11 16:01:30 -0800 | [diff] [blame] | 288 | * If this task is not being ptraced on exit, then wait for it |
| 289 | * to finish before killing some other task unnecessarily. |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 290 | */ |
David Rientjes | 9ff4868 | 2012-12-11 16:01:30 -0800 | [diff] [blame] | 291 | if (!(task->group_leader->ptrace & PT_TRACE_EXIT)) |
| 292 | return OOM_SCAN_ABORT; |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 293 | } |
| 294 | return OOM_SCAN_OK; |
| 295 | } |
| 296 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 297 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 298 | * Simple selection loop. We chose the process with the highest |
Rusty Russell | 6b4f2b5 | 2013-07-15 11:24:08 +0930 | [diff] [blame] | 299 | * number of 'points'. Returns -1 on scan abort. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 300 | * |
| 301 | * (not docbooked, we don't want this one cluttering up the manual) |
| 302 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 303 | static struct task_struct *select_bad_process(unsigned int *ppoints, |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 304 | unsigned long totalpages, const nodemask_t *nodemask, |
| 305 | bool force_kill) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 306 | { |
David Rientjes | 3a5dda7 | 2011-03-22 16:30:09 -0700 | [diff] [blame] | 307 | struct task_struct *g, *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 308 | struct task_struct *chosen = NULL; |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 309 | unsigned long chosen_points = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 310 | |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 311 | rcu_read_lock(); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 312 | for_each_process_thread(g, p) { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 313 | unsigned int points; |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 314 | |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 315 | switch (oom_scan_process_thread(p, totalpages, nodemask, |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 316 | force_kill)) { |
| 317 | case OOM_SCAN_SELECT: |
| 318 | chosen = p; |
| 319 | chosen_points = ULONG_MAX; |
| 320 | /* fall through */ |
| 321 | case OOM_SCAN_CONTINUE: |
Andrey Vagin | 30e2b41 | 2011-03-22 16:30:11 -0700 | [diff] [blame] | 322 | continue; |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 323 | case OOM_SCAN_ABORT: |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 324 | rcu_read_unlock(); |
Rusty Russell | 6b4f2b5 | 2013-07-15 11:24:08 +0930 | [diff] [blame] | 325 | return (struct task_struct *)(-1UL); |
David Rientjes | 462607e | 2012-07-31 16:43:40 -0700 | [diff] [blame] | 326 | case OOM_SCAN_OK: |
| 327 | break; |
| 328 | }; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 329 | points = oom_badness(p, NULL, nodemask, totalpages); |
David Rientjes | d49ad93 | 2014-01-23 15:53:34 -0800 | [diff] [blame] | 330 | if (!points || points < chosen_points) |
| 331 | continue; |
| 332 | /* Prefer thread group leaders for display purposes */ |
| 333 | if (points == chosen_points && thread_group_leader(chosen)) |
| 334 | continue; |
| 335 | |
| 336 | chosen = p; |
| 337 | chosen_points = points; |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 338 | } |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 339 | if (chosen) |
| 340 | get_task_struct(chosen); |
| 341 | rcu_read_unlock(); |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 342 | |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 343 | *ppoints = chosen_points * 1000 / totalpages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 344 | return chosen; |
| 345 | } |
| 346 | |
| 347 | /** |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 348 | * dump_tasks - dump current memory state of all system tasks |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 349 | * @memcg: current's memory controller, if constrained |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 350 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 351 | * |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 352 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
| 353 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes |
| 354 | * are not shown. |
David Rientjes | de34d96 | 2012-07-31 16:42:56 -0700 | [diff] [blame] | 355 | * State information includes task's pid, uid, tgid, vm size, rss, nr_ptes, |
| 356 | * swapents, oom_score_adj value, and name. |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 357 | */ |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 358 | static void dump_tasks(const struct mem_cgroup *memcg, const nodemask_t *nodemask) |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 359 | { |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 360 | struct task_struct *p; |
| 361 | struct task_struct *task; |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 362 | |
David Rientjes | de34d96 | 2012-07-31 16:42:56 -0700 | [diff] [blame] | 363 | pr_info("[ pid ] uid tgid total_vm rss nr_ptes swapents oom_score_adj name\n"); |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 364 | rcu_read_lock(); |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 365 | for_each_process(p) { |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 366 | if (oom_unkillable_task(p, memcg, nodemask)) |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 367 | continue; |
| 368 | |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 369 | task = find_lock_task_mm(p); |
| 370 | if (!task) { |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 371 | /* |
David Rientjes | 74ab7f1 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 372 | * This is a kthread or all of p's threads have already |
| 373 | * detached their mm's. There's no need to report |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 374 | * them; they can't be oom killed anyway. |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 375 | */ |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 376 | continue; |
| 377 | } |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 378 | |
Kirill A. Shutemov | e1f56c8 | 2013-11-14 14:30:48 -0800 | [diff] [blame] | 379 | pr_info("[%5d] %5d %5d %8lu %8lu %7ld %8lu %5hd %s\n", |
Eric W. Biederman | 078de5f | 2012-02-08 07:00:08 -0800 | [diff] [blame] | 380 | task->pid, from_kuid(&init_user_ns, task_uid(task)), |
| 381 | task->tgid, task->mm->total_vm, get_mm_rss(task->mm), |
Kirill A. Shutemov | e1f56c8 | 2013-11-14 14:30:48 -0800 | [diff] [blame] | 382 | atomic_long_read(&task->mm->nr_ptes), |
David Rientjes | de34d96 | 2012-07-31 16:42:56 -0700 | [diff] [blame] | 383 | get_mm_counter(task->mm, MM_SWAPENTS), |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 384 | task->signal->oom_score_adj, task->comm); |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 385 | task_unlock(task); |
| 386 | } |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 387 | rcu_read_unlock(); |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 388 | } |
| 389 | |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 390 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 391 | struct mem_cgroup *memcg, const nodemask_t *nodemask) |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 392 | { |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 393 | task_lock(current); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 394 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " |
David Rientjes | a9c58b90 | 2012-12-11 16:02:54 -0800 | [diff] [blame] | 395 | "oom_score_adj=%hd\n", |
Davidlohr Bueso | 01dc52e | 2012-10-08 16:29:30 -0700 | [diff] [blame] | 396 | current->comm, gfp_mask, order, |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 397 | current->signal->oom_score_adj); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 398 | cpuset_print_task_mems_allowed(current); |
| 399 | task_unlock(current); |
| 400 | dump_stack(); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 401 | if (memcg) |
| 402 | mem_cgroup_print_oom_info(memcg, p); |
| 403 | else |
| 404 | show_mem(SHOW_MEM_FILTER_NODES); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 405 | if (sysctl_oom_dump_tasks) |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 406 | dump_tasks(memcg, nodemask); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 407 | } |
| 408 | |
KOSAKI Motohiro | 3b4798c | 2009-12-15 16:45:32 -0800 | [diff] [blame] | 409 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 410 | /* |
| 411 | * Must be called while holding a reference to p, which will be released upon |
| 412 | * returning. |
| 413 | */ |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 414 | void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
| 415 | unsigned int points, unsigned long totalpages, |
| 416 | struct mem_cgroup *memcg, nodemask_t *nodemask, |
| 417 | const char *message) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 418 | { |
Linus Torvalds | 52d3c03 | 2011-03-14 15:17:07 -0700 | [diff] [blame] | 419 | struct task_struct *victim = p; |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 420 | struct task_struct *child; |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 421 | struct task_struct *t; |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 422 | struct mm_struct *mm; |
Linus Torvalds | 52d3c03 | 2011-03-14 15:17:07 -0700 | [diff] [blame] | 423 | unsigned int victim_points = 0; |
David Rientjes | dc3f21e | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 424 | static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
| 425 | DEFAULT_RATELIMIT_BURST); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 426 | |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 427 | /* |
| 428 | * If the task is already exiting, don't alarm the sysadmin or kill |
| 429 | * its children or threads, just set TIF_MEMDIE so it can die quickly |
| 430 | */ |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 431 | if (p->flags & PF_EXITING) { |
David Rientjes | 4358997 | 2010-08-09 17:18:49 -0700 | [diff] [blame] | 432 | set_tsk_thread_flag(p, TIF_MEMDIE); |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 433 | put_task_struct(p); |
David Rientjes | 2a1c9b1 | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 434 | return; |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 435 | } |
| 436 | |
David Rientjes | dc3f21e | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 437 | if (__ratelimit(&oom_rs)) |
David Rientjes | 8447d95 | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 438 | dump_header(p, gfp_mask, order, memcg, nodemask); |
| 439 | |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 440 | task_lock(p); |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 441 | pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n", |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 442 | message, task_pid_nr(p), p->comm, points); |
| 443 | task_unlock(p); |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 444 | |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 445 | /* |
| 446 | * 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] | 447 | * the one with the highest oom_badness() score is sacrificed for its |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 448 | * parent. This attempts to lose the minimal amount of work done while |
| 449 | * still freeing memory. |
| 450 | */ |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 451 | read_lock(&tasklist_lock); |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 452 | for_each_thread(p, t) { |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 453 | list_for_each_entry(child, &t->children, sibling) { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 454 | unsigned int child_points; |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 455 | |
David Rientjes | edd4554 | 2011-03-22 16:30:12 -0700 | [diff] [blame] | 456 | if (child->mm == p->mm) |
| 457 | continue; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 458 | /* |
| 459 | * oom_badness() returns 0 if the thread is unkillable |
| 460 | */ |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 461 | child_points = oom_badness(child, memcg, nodemask, |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 462 | totalpages); |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 463 | if (child_points > victim_points) { |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 464 | put_task_struct(victim); |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 465 | victim = child; |
| 466 | victim_points = child_points; |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 467 | get_task_struct(victim); |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 468 | } |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 469 | } |
Oleg Nesterov | 1da4db0 | 2014-01-21 15:49:58 -0800 | [diff] [blame] | 470 | } |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 471 | read_unlock(&tasklist_lock); |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 472 | |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 473 | p = find_lock_task_mm(victim); |
| 474 | if (!p) { |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 475 | put_task_struct(victim); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 476 | return; |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 477 | } else if (victim != p) { |
| 478 | get_task_struct(p); |
| 479 | put_task_struct(victim); |
| 480 | victim = p; |
| 481 | } |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 482 | |
| 483 | /* mm cannot safely be dereferenced after task_unlock(victim) */ |
| 484 | mm = victim->mm; |
| 485 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", |
| 486 | task_pid_nr(victim), victim->comm, K(victim->mm->total_vm), |
| 487 | K(get_mm_counter(victim->mm, MM_ANONPAGES)), |
| 488 | K(get_mm_counter(victim->mm, MM_FILEPAGES))); |
| 489 | task_unlock(victim); |
| 490 | |
| 491 | /* |
| 492 | * Kill all user processes sharing victim->mm in other thread groups, if |
| 493 | * any. They don't get access to memory reserves, though, to avoid |
| 494 | * depletion of all memory. This prevents mm->mmap_sem livelock when an |
| 495 | * oom killed thread cannot exit because it requires the semaphore and |
| 496 | * its contended by another thread trying to allocate memory itself. |
| 497 | * That thread will now get access to memory reserves since it has a |
| 498 | * pending fatal signal. |
| 499 | */ |
Oleg Nesterov | 4d4048b | 2014-01-21 15:50:01 -0800 | [diff] [blame] | 500 | rcu_read_lock(); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 501 | for_each_process(p) |
| 502 | if (p->mm == mm && !same_thread_group(p, victim) && |
| 503 | !(p->flags & PF_KTHREAD)) { |
| 504 | if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) |
| 505 | continue; |
| 506 | |
| 507 | task_lock(p); /* Protect ->comm from prctl() */ |
| 508 | pr_err("Kill process %d (%s) sharing same memory\n", |
| 509 | task_pid_nr(p), p->comm); |
| 510 | task_unlock(p); |
Oleg Nesterov | d2d3930 | 2012-03-23 15:02:45 -0700 | [diff] [blame] | 511 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 512 | } |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 513 | rcu_read_unlock(); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 514 | |
| 515 | set_tsk_thread_flag(victim, TIF_MEMDIE); |
Oleg Nesterov | d2d3930 | 2012-03-23 15:02:45 -0700 | [diff] [blame] | 516 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true); |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 517 | put_task_struct(victim); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 518 | } |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 519 | #undef K |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 520 | |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 521 | /* |
| 522 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. |
| 523 | */ |
David Rientjes | 876aafb | 2012-07-31 16:43:48 -0700 | [diff] [blame] | 524 | void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask, |
| 525 | int order, const nodemask_t *nodemask) |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 526 | { |
| 527 | if (likely(!sysctl_panic_on_oom)) |
| 528 | return; |
| 529 | if (sysctl_panic_on_oom != 2) { |
| 530 | /* |
| 531 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel |
| 532 | * does not panic for cpuset, mempolicy, or memcg allocation |
| 533 | * failures. |
| 534 | */ |
| 535 | if (constraint != CONSTRAINT_NONE) |
| 536 | return; |
| 537 | } |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 538 | dump_header(NULL, gfp_mask, order, NULL, nodemask); |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 539 | panic("Out of memory: %s panic_on_oom is enabled\n", |
| 540 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); |
| 541 | } |
| 542 | |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 543 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
| 544 | |
| 545 | int register_oom_notifier(struct notifier_block *nb) |
| 546 | { |
| 547 | return blocking_notifier_chain_register(&oom_notify_list, nb); |
| 548 | } |
| 549 | EXPORT_SYMBOL_GPL(register_oom_notifier); |
| 550 | |
| 551 | int unregister_oom_notifier(struct notifier_block *nb) |
| 552 | { |
| 553 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); |
| 554 | } |
| 555 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); |
| 556 | |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 557 | /* |
| 558 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero |
| 559 | * if a parallel OOM killing is already taking place that includes a zone in |
| 560 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. |
| 561 | */ |
Minchan Kim | ff321fe | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 562 | int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 563 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 564 | struct zoneref *z; |
| 565 | struct zone *zone; |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 566 | int ret = 1; |
| 567 | |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 568 | spin_lock(&zone_scan_lock); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 569 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 570 | if (zone_is_oom_locked(zone)) { |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 571 | ret = 0; |
| 572 | goto out; |
| 573 | } |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 574 | } |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 575 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 576 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 577 | /* |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 578 | * Lock each zone in the zonelist under zone_scan_lock so a |
Minchan Kim | ff321fe | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 579 | * parallel invocation of try_set_zonelist_oom() doesn't succeed |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 580 | * when it shouldn't. |
| 581 | */ |
| 582 | zone_set_flag(zone, ZONE_OOM_LOCKED); |
| 583 | } |
| 584 | |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 585 | out: |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 586 | spin_unlock(&zone_scan_lock); |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 587 | return ret; |
| 588 | } |
| 589 | |
| 590 | /* |
| 591 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed |
| 592 | * allocation attempts with zonelists containing them may now recall the OOM |
| 593 | * killer, if necessary. |
| 594 | */ |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 595 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 596 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 597 | struct zoneref *z; |
| 598 | struct zone *zone; |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 599 | |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 600 | spin_lock(&zone_scan_lock); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 601 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 602 | zone_clear_flag(zone, ZONE_OOM_LOCKED); |
| 603 | } |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 604 | spin_unlock(&zone_scan_lock); |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 605 | } |
| 606 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 607 | /** |
Dave Peterson | 6937a25 | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 608 | * out_of_memory - kill the "best" process when we run out of memory |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 609 | * @zonelist: zonelist pointer |
| 610 | * @gfp_mask: memory allocation flags |
| 611 | * @order: amount of memory being requested as a power of 2 |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 612 | * @nodemask: nodemask passed to page allocator |
David Rientjes | 08ab9b1 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 613 | * @force_kill: true if a task must be killed, even if others are exiting |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 614 | * |
| 615 | * If we run out of memory, we have the choice between either |
| 616 | * killing a random task (bad), letting the system crash (worse) |
| 617 | * OR try to be smart about which process to kill. Note that we |
| 618 | * don't have to be perfect here, we just have to be good. |
| 619 | */ |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 620 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
David Rientjes | 08ab9b1 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 621 | int order, nodemask_t *nodemask, bool force_kill) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 622 | { |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 623 | const nodemask_t *mpol_mask; |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 624 | struct task_struct *p; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 625 | unsigned long totalpages; |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 626 | unsigned long freed = 0; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 627 | unsigned int uninitialized_var(points); |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 628 | enum oom_constraint constraint = CONSTRAINT_NONE; |
KOSAKI Motohiro | b52723c | 2010-08-19 14:13:39 -0700 | [diff] [blame] | 629 | int killed = 0; |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 630 | |
| 631 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); |
| 632 | if (freed > 0) |
| 633 | /* Got some memory back in the last second. */ |
| 634 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 635 | |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 636 | /* |
David Rientjes | 9ff4868 | 2012-12-11 16:01:30 -0800 | [diff] [blame] | 637 | * If current has a pending SIGKILL or is exiting, then automatically |
| 638 | * select it. The goal is to allow it to allocate so that it may |
| 639 | * quickly exit and free its memory. |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 640 | */ |
David Rientjes | 9ff4868 | 2012-12-11 16:01:30 -0800 | [diff] [blame] | 641 | if (fatal_signal_pending(current) || current->flags & PF_EXITING) { |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 642 | set_thread_flag(TIF_MEMDIE); |
| 643 | return; |
| 644 | } |
| 645 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 646 | /* |
| 647 | * Check if there were limitations on the allocation (only relevant for |
| 648 | * NUMA) that may require different handling. |
| 649 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 650 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask, |
| 651 | &totalpages); |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 652 | mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL; |
| 653 | check_panic_on_oom(constraint, gfp_mask, order, mpol_mask); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 654 | |
David Rientjes | 121d1ba | 2012-07-31 16:42:55 -0700 | [diff] [blame] | 655 | if (sysctl_oom_kill_allocating_task && current->mm && |
KOSAKI Motohiro | a96cfd6 | 2010-08-09 17:19:38 -0700 | [diff] [blame] | 656 | !oom_unkillable_task(current, NULL, nodemask) && |
David Rientjes | 121d1ba | 2012-07-31 16:42:55 -0700 | [diff] [blame] | 657 | current->signal->oom_score_adj != OOM_SCORE_ADJ_MIN) { |
David Rientjes | 6b0c81b | 2012-07-31 16:43:45 -0700 | [diff] [blame] | 658 | get_task_struct(current); |
David Rientjes | 2a1c9b1 | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 659 | oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL, |
| 660 | nodemask, |
| 661 | "Out of memory (oom_kill_allocating_task)"); |
| 662 | goto out; |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 663 | } |
| 664 | |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 665 | p = select_bad_process(&points, totalpages, mpol_mask, force_kill); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 666 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
| 667 | if (!p) { |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 668 | dump_header(NULL, gfp_mask, order, NULL, mpol_mask); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 669 | panic("Out of memory and no killable processes...\n"); |
| 670 | } |
Rusty Russell | 6b4f2b5 | 2013-07-15 11:24:08 +0930 | [diff] [blame] | 671 | if (p != (void *)-1UL) { |
David Rientjes | 2a1c9b1 | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 672 | oom_kill_process(p, gfp_mask, order, points, totalpages, NULL, |
| 673 | nodemask, "Out of memory"); |
| 674 | killed = 1; |
| 675 | } |
KOSAKI Motohiro | b52723c | 2010-08-19 14:13:39 -0700 | [diff] [blame] | 676 | out: |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 677 | /* |
David Rientjes | 4f774b9 | 2012-07-31 16:42:37 -0700 | [diff] [blame] | 678 | * Give the killed threads a good chance of exiting before trying to |
| 679 | * allocate memory again. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 680 | */ |
David Rientjes | 4f774b9 | 2012-07-31 16:42:37 -0700 | [diff] [blame] | 681 | if (killed) |
| 682 | schedule_timeout_killable(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 683 | } |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 684 | |
| 685 | /* |
| 686 | * 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] | 687 | * memory-hogging task. If any populated zone has ZONE_OOM_LOCKED set, a |
| 688 | * parallel oom killing is already in progress so do nothing. |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 689 | */ |
| 690 | void pagefault_out_of_memory(void) |
| 691 | { |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 692 | struct zonelist *zonelist; |
David Rientjes | efacd02 | 2012-12-12 13:52:06 -0800 | [diff] [blame] | 693 | |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 694 | if (mem_cgroup_oom_synchronize(true)) |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 695 | return; |
| 696 | |
David Rientjes | 8d060bf | 2014-08-06 16:07:50 -0700 | [diff] [blame^] | 697 | zonelist = node_zonelist(first_memory_node, GFP_KERNEL); |
David Rientjes | efacd02 | 2012-12-12 13:52:06 -0800 | [diff] [blame] | 698 | if (try_set_zonelist_oom(zonelist, GFP_KERNEL)) { |
David Rientjes | 08ab9b1 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 699 | out_of_memory(NULL, 0, 0, NULL, false); |
David Rientjes | efacd02 | 2012-12-12 13:52:06 -0800 | [diff] [blame] | 700 | clear_zonelist_oom(zonelist, GFP_KERNEL); |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 701 | } |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 702 | } |