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 | 43362a4 | 2011-10-31 17:07:18 -0700 | [diff] [blame] | 47 | /* |
| 48 | * compare_swap_oom_score_adj() - compare and swap current's oom_score_adj |
| 49 | * @old_val: old oom_score_adj for compare |
| 50 | * @new_val: new oom_score_adj for swap |
| 51 | * |
| 52 | * Sets the oom_score_adj value for current to @new_val iff its present value is |
| 53 | * @old_val. Usually used to reinstate a previous value to prevent racing with |
| 54 | * userspacing tuning the value in the interim. |
| 55 | */ |
| 56 | void compare_swap_oom_score_adj(int old_val, int new_val) |
| 57 | { |
| 58 | struct sighand_struct *sighand = current->sighand; |
| 59 | |
| 60 | spin_lock_irq(&sighand->siglock); |
| 61 | if (current->signal->oom_score_adj == old_val) |
| 62 | current->signal->oom_score_adj = new_val; |
KAMEZAWA Hiroyuki | 43d2b11 | 2012-01-10 15:08:09 -0800 | [diff] [blame] | 63 | trace_oom_score_adj_update(current); |
David Rientjes | 43362a4 | 2011-10-31 17:07:18 -0700 | [diff] [blame] | 64 | spin_unlock_irq(&sighand->siglock); |
| 65 | } |
| 66 | |
David Rientjes | 72788c3 | 2011-05-24 17:11:40 -0700 | [diff] [blame] | 67 | /** |
| 68 | * test_set_oom_score_adj() - set current's oom_score_adj and return old value |
| 69 | * @new_val: new oom_score_adj value |
| 70 | * |
| 71 | * Sets the oom_score_adj value for current to @new_val with proper |
| 72 | * synchronization and returns the old value. Usually used to temporarily |
| 73 | * set a value, save the old value in the caller, and then reinstate it later. |
| 74 | */ |
| 75 | int test_set_oom_score_adj(int new_val) |
| 76 | { |
| 77 | struct sighand_struct *sighand = current->sighand; |
| 78 | int old_val; |
| 79 | |
| 80 | spin_lock_irq(&sighand->siglock); |
| 81 | old_val = current->signal->oom_score_adj; |
David Rientjes | c9f0124 | 2011-10-31 17:07:15 -0700 | [diff] [blame] | 82 | current->signal->oom_score_adj = new_val; |
KAMEZAWA Hiroyuki | 43d2b11 | 2012-01-10 15:08:09 -0800 | [diff] [blame] | 83 | trace_oom_score_adj_update(current); |
David Rientjes | 72788c3 | 2011-05-24 17:11:40 -0700 | [diff] [blame] | 84 | spin_unlock_irq(&sighand->siglock); |
| 85 | |
| 86 | return old_val; |
| 87 | } |
| 88 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 89 | #ifdef CONFIG_NUMA |
| 90 | /** |
| 91 | * has_intersects_mems_allowed() - check task eligiblity for kill |
| 92 | * @tsk: task struct of which task to consider |
| 93 | * @mask: nodemask passed to page allocator for mempolicy ooms |
| 94 | * |
| 95 | * Task eligibility is determined by whether or not a candidate task, @tsk, |
| 96 | * shares the same mempolicy nodes as current if it is bound by such a policy |
| 97 | * 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] | 98 | */ |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 99 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
| 100 | const nodemask_t *mask) |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 101 | { |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 102 | struct task_struct *start = tsk; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 103 | |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 104 | do { |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 105 | if (mask) { |
| 106 | /* |
| 107 | * If this is a mempolicy constrained oom, tsk's |
| 108 | * cpuset is irrelevant. Only return true if its |
| 109 | * mempolicy intersects current, otherwise it may be |
| 110 | * needlessly killed. |
| 111 | */ |
| 112 | if (mempolicy_nodemask_intersects(tsk, mask)) |
| 113 | return true; |
| 114 | } else { |
| 115 | /* |
| 116 | * This is not a mempolicy constrained oom, so only |
| 117 | * check the mems of tsk's cpuset. |
| 118 | */ |
| 119 | if (cpuset_mems_allowed_intersects(current, tsk)) |
| 120 | return true; |
| 121 | } |
KOSAKI Motohiro | df1090a | 2010-08-09 17:19:39 -0700 | [diff] [blame] | 122 | } while_each_thread(start, tsk); |
| 123 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 124 | return false; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 125 | } |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 126 | #else |
| 127 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
| 128 | const nodemask_t *mask) |
| 129 | { |
| 130 | return true; |
| 131 | } |
| 132 | #endif /* CONFIG_NUMA */ |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 133 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 134 | /* |
| 135 | * The process p may have detached its own ->mm while exiting or through |
| 136 | * use_mm(), but one or more of its subthreads may still have a valid |
| 137 | * pointer. Return p, or any of its subthreads with a valid ->mm, with |
| 138 | * task_lock() held. |
| 139 | */ |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 140 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 141 | { |
| 142 | struct task_struct *t = p; |
| 143 | |
| 144 | do { |
| 145 | task_lock(t); |
| 146 | if (likely(t->mm)) |
| 147 | return t; |
| 148 | task_unlock(t); |
| 149 | } while_each_thread(p, t); |
| 150 | |
| 151 | return NULL; |
| 152 | } |
| 153 | |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 154 | /* return true if the task is not adequate as candidate victim task. */ |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 155 | static bool oom_unkillable_task(struct task_struct *p, |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 156 | const struct mem_cgroup *memcg, const nodemask_t *nodemask) |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 157 | { |
| 158 | if (is_global_init(p)) |
| 159 | return true; |
| 160 | if (p->flags & PF_KTHREAD) |
| 161 | return true; |
| 162 | |
| 163 | /* 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] | 164 | if (memcg && !task_in_mem_cgroup(p, memcg)) |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 165 | return true; |
| 166 | |
| 167 | /* p may not have freeable memory in nodemask */ |
| 168 | if (!has_intersects_mems_allowed(p, nodemask)) |
| 169 | return true; |
| 170 | |
| 171 | return false; |
| 172 | } |
| 173 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 174 | /** |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 175 | * oom_badness - heuristic function to determine which candidate task to kill |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 176 | * @p: task struct of which task we should calculate |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 177 | * @totalpages: total present RAM allowed for page allocation |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 178 | * |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 179 | * The heuristic for determining which task to kill is made to be as simple and |
| 180 | * predictable as possible. The goal is to return the highest value for the |
| 181 | * task consuming the most memory to avoid subsequent oom failures. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | */ |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 183 | unsigned long oom_badness(struct task_struct *p, struct mem_cgroup *memcg, |
| 184 | const nodemask_t *nodemask, unsigned long totalpages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 185 | { |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 186 | unsigned long points; |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 187 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 188 | if (oom_unkillable_task(p, memcg, nodemask)) |
KOSAKI Motohiro | 26ebc98 | 2010-08-09 17:19:37 -0700 | [diff] [blame] | 189 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 191 | p = find_lock_task_mm(p); |
| 192 | if (!p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 193 | return 0; |
| 194 | |
Michal Hocko | 5aecc85 | 2011-11-15 14:36:07 -0800 | [diff] [blame] | 195 | if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) { |
| 196 | task_unlock(p); |
| 197 | return 0; |
| 198 | } |
| 199 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 200 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 201 | * 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] | 202 | * task's rss, pagetable and swap space use. |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 203 | */ |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 204 | points = get_mm_rss(p->mm) + p->mm->nr_ptes + |
| 205 | get_mm_counter(p->mm, MM_SWAPENTS); |
Andrew Morton | 97c2c9b8 | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 206 | task_unlock(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 207 | |
| 208 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 209 | * Root processes get 3% bonus, just like the __vm_enough_memory() |
| 210 | * implementation used by LSMs. |
Hugh Dickins | 7ba3485 | 2007-01-05 16:37:03 -0800 | [diff] [blame] | 211 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 212 | if (has_capability_noaudit(p, CAP_SYS_ADMIN)) |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 213 | points -= 30 * totalpages / 1000; |
Hugh Dickins | 7ba3485 | 2007-01-05 16:37:03 -0800 | [diff] [blame] | 214 | |
| 215 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 216 | * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may |
| 217 | * either completely disable oom killing or always prefer a certain |
| 218 | * task. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 219 | */ |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 220 | points += p->signal->oom_score_adj * totalpages / 1000; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 221 | |
David Rientjes | f19e8aa | 2010-09-22 13:04:52 -0700 | [diff] [blame] | 222 | /* |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 223 | * Never return 0 for an eligible task regardless of the root bonus and |
| 224 | * 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] | 225 | */ |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 226 | return points ? points : 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | } |
| 228 | |
| 229 | /* |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 230 | * Determine the type of allocation constraint. |
| 231 | */ |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 232 | #ifdef CONFIG_NUMA |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 233 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 234 | gfp_t gfp_mask, nodemask_t *nodemask, |
| 235 | unsigned long *totalpages) |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 236 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 237 | struct zone *zone; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 238 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 239 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 240 | bool cpuset_limited = false; |
| 241 | int nid; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 242 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 243 | /* Default to all available memory */ |
| 244 | *totalpages = totalram_pages + total_swap_pages; |
| 245 | |
| 246 | if (!zonelist) |
| 247 | return CONSTRAINT_NONE; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 248 | /* |
| 249 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid |
| 250 | * to kill current.We have to random task kill in this case. |
| 251 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. |
| 252 | */ |
| 253 | if (gfp_mask & __GFP_THISNODE) |
| 254 | return CONSTRAINT_NONE; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 255 | |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 256 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 257 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
| 258 | * the page allocator means a mempolicy is in effect. Cpuset policy |
| 259 | * is enforced in get_page_from_freelist(). |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 260 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 261 | if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) { |
| 262 | *totalpages = total_swap_pages; |
| 263 | for_each_node_mask(nid, *nodemask) |
| 264 | *totalpages += node_spanned_pages(nid); |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 265 | return CONSTRAINT_MEMORY_POLICY; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 266 | } |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 267 | |
| 268 | /* Check this allocation failure is caused by cpuset's wall function */ |
| 269 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 270 | high_zoneidx, nodemask) |
| 271 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 272 | cpuset_limited = true; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 273 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 274 | if (cpuset_limited) { |
| 275 | *totalpages = total_swap_pages; |
| 276 | for_each_node_mask(nid, cpuset_current_mems_allowed) |
| 277 | *totalpages += node_spanned_pages(nid); |
| 278 | return CONSTRAINT_CPUSET; |
| 279 | } |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 280 | return CONSTRAINT_NONE; |
| 281 | } |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 282 | #else |
| 283 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 284 | gfp_t gfp_mask, nodemask_t *nodemask, |
| 285 | unsigned long *totalpages) |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 286 | { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 287 | *totalpages = totalram_pages + total_swap_pages; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 288 | return CONSTRAINT_NONE; |
| 289 | } |
| 290 | #endif |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 291 | |
| 292 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 293 | * Simple selection loop. We chose the process with the highest |
| 294 | * number of 'points'. We expect the caller will lock the tasklist. |
| 295 | * |
| 296 | * (not docbooked, we don't want this one cluttering up the manual) |
| 297 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 298 | static struct task_struct *select_bad_process(unsigned int *ppoints, |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 299 | unsigned long totalpages, struct mem_cgroup *memcg, |
David Rientjes | 08ab9b1 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 300 | const nodemask_t *nodemask, bool force_kill) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 301 | { |
David Rientjes | 3a5dda7 | 2011-03-22 16:30:09 -0700 | [diff] [blame] | 302 | struct task_struct *g, *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 303 | struct task_struct *chosen = NULL; |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 304 | unsigned long chosen_points = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | |
David Rientjes | 3a5dda7 | 2011-03-22 16:30:09 -0700 | [diff] [blame] | 306 | do_each_thread(g, p) { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 307 | unsigned int points; |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 308 | |
Oleg Nesterov | c027a47 | 2011-07-30 16:35:02 +0200 | [diff] [blame] | 309 | if (p->exit_state) |
Andrey Vagin | 30e2b41 | 2011-03-22 16:30:11 -0700 | [diff] [blame] | 310 | continue; |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 311 | if (oom_unkillable_task(p, memcg, nodemask)) |
David Rientjes | 6cf86ac | 2010-08-09 17:18:50 -0700 | [diff] [blame] | 312 | continue; |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 313 | |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 314 | /* |
Nick Piggin | b78483a | 2006-09-29 02:01:14 -0700 | [diff] [blame] | 315 | * This task already has access to memory reserves and is |
| 316 | * being killed. Don't allow any other task access to the |
| 317 | * memory reserve. |
| 318 | * |
| 319 | * Note: this may have a chance of deadlock if it gets |
| 320 | * blocked waiting for another task which itself is waiting |
| 321 | * for memory. Is there a better alternative? |
| 322 | */ |
David Rientjes | f660daa | 2011-10-31 17:07:07 -0700 | [diff] [blame] | 323 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) { |
| 324 | if (unlikely(frozen(p))) |
Tejun Heo | a5be2d0 | 2011-11-21 12:32:23 -0800 | [diff] [blame] | 325 | __thaw_task(p); |
David Rientjes | 08ab9b1 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 326 | if (!force_kill) |
| 327 | return ERR_PTR(-1UL); |
David Rientjes | f660daa | 2011-10-31 17:07:07 -0700 | [diff] [blame] | 328 | } |
Oleg Nesterov | c027a47 | 2011-07-30 16:35:02 +0200 | [diff] [blame] | 329 | if (!p->mm) |
| 330 | continue; |
Nick Piggin | b78483a | 2006-09-29 02:01:14 -0700 | [diff] [blame] | 331 | |
Andrey Vagin | 30e2b41 | 2011-03-22 16:30:11 -0700 | [diff] [blame] | 332 | if (p->flags & PF_EXITING) { |
David Rientjes | edd4554 | 2011-03-22 16:30:12 -0700 | [diff] [blame] | 333 | /* |
| 334 | * If p is the current task and is in the process of |
| 335 | * releasing memory, we allow the "kill" to set |
| 336 | * TIF_MEMDIE, which will allow it to gain access to |
| 337 | * memory reserves. Otherwise, it may stall forever. |
| 338 | * |
| 339 | * The loop isn't broken here, however, in case other |
| 340 | * threads are found to have already been oom killed. |
| 341 | */ |
| 342 | if (p == current) { |
| 343 | chosen = p; |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 344 | chosen_points = ULONG_MAX; |
David Rientjes | 08ab9b1 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 345 | } else if (!force_kill) { |
David Rientjes | edd4554 | 2011-03-22 16:30:12 -0700 | [diff] [blame] | 346 | /* |
| 347 | * If this task is not being ptraced on exit, |
| 348 | * then wait for it to finish before killing |
| 349 | * some other task unnecessarily. |
| 350 | */ |
Tejun Heo | d21142e | 2011-06-17 16:50:34 +0200 | [diff] [blame] | 351 | if (!(p->group_leader->ptrace & PT_TRACE_EXIT)) |
David Rientjes | edd4554 | 2011-03-22 16:30:12 -0700 | [diff] [blame] | 352 | return ERR_PTR(-1UL); |
| 353 | } |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 354 | } |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 355 | |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 356 | points = oom_badness(p, memcg, nodemask, totalpages); |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 357 | if (points > chosen_points) { |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 358 | chosen = p; |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 359 | chosen_points = points; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 360 | } |
David Rientjes | 3a5dda7 | 2011-03-22 16:30:09 -0700 | [diff] [blame] | 361 | } while_each_thread(g, p); |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 362 | |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 363 | *ppoints = chosen_points * 1000 / totalpages; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 364 | return chosen; |
| 365 | } |
| 366 | |
| 367 | /** |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 368 | * dump_tasks - dump current memory state of all system tasks |
David Rientjes | 74ab7f1 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 369 | * @mem: current's memory controller, if constrained |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 370 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 371 | * |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 372 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
| 373 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes |
| 374 | * are not shown. |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 375 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 376 | * value, oom_score_adj value, and name. |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 377 | * |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 378 | * Call with tasklist_lock read-locked. |
| 379 | */ |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 380 | 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] | 381 | { |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 382 | struct task_struct *p; |
| 383 | struct task_struct *task; |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 384 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 385 | pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n"); |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 386 | for_each_process(p) { |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 387 | if (oom_unkillable_task(p, memcg, nodemask)) |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 388 | continue; |
| 389 | |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 390 | task = find_lock_task_mm(p); |
| 391 | if (!task) { |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 392 | /* |
David Rientjes | 74ab7f1 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 393 | * This is a kthread or all of p's threads have already |
| 394 | * detached their mm's. There's no need to report |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 395 | * them; they can't be oom killed anyway. |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 396 | */ |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 397 | continue; |
| 398 | } |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 399 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 400 | pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n", |
Eric W. Biederman | 078de5f | 2012-02-08 07:00:08 -0800 | [diff] [blame] | 401 | task->pid, from_kuid(&init_user_ns, task_uid(task)), |
| 402 | task->tgid, task->mm->total_vm, get_mm_rss(task->mm), |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 403 | task_cpu(task), task->signal->oom_adj, |
| 404 | task->signal->oom_score_adj, task->comm); |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 405 | task_unlock(task); |
| 406 | } |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 407 | } |
| 408 | |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 409 | 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] | 410 | struct mem_cgroup *memcg, const nodemask_t *nodemask) |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 411 | { |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 412 | task_lock(current); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 413 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 414 | "oom_adj=%d, oom_score_adj=%d\n", |
| 415 | current->comm, gfp_mask, order, current->signal->oom_adj, |
| 416 | current->signal->oom_score_adj); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 417 | cpuset_print_task_mems_allowed(current); |
| 418 | task_unlock(current); |
| 419 | dump_stack(); |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 420 | mem_cgroup_print_oom_info(memcg, p); |
David Rientjes | b2b755b | 2011-03-24 15:18:15 -0700 | [diff] [blame] | 421 | show_mem(SHOW_MEM_FILTER_NODES); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 422 | if (sysctl_oom_dump_tasks) |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 423 | dump_tasks(memcg, nodemask); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 424 | } |
| 425 | |
KOSAKI Motohiro | 3b4798c | 2009-12-15 16:45:32 -0800 | [diff] [blame] | 426 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
David Rientjes | 2a1c9b1 | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 427 | static void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
| 428 | unsigned int points, unsigned long totalpages, |
| 429 | struct mem_cgroup *memcg, nodemask_t *nodemask, |
| 430 | const char *message) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 431 | { |
Linus Torvalds | 52d3c03 | 2011-03-14 15:17:07 -0700 | [diff] [blame] | 432 | struct task_struct *victim = p; |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 433 | struct task_struct *child; |
Linus Torvalds | 52d3c03 | 2011-03-14 15:17:07 -0700 | [diff] [blame] | 434 | struct task_struct *t = p; |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 435 | struct mm_struct *mm; |
Linus Torvalds | 52d3c03 | 2011-03-14 15:17:07 -0700 | [diff] [blame] | 436 | unsigned int victim_points = 0; |
David Rientjes | dc3f21e | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 437 | static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL, |
| 438 | DEFAULT_RATELIMIT_BURST); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 439 | |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 440 | /* |
| 441 | * If the task is already exiting, don't alarm the sysadmin or kill |
| 442 | * its children or threads, just set TIF_MEMDIE so it can die quickly |
| 443 | */ |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 444 | if (p->flags & PF_EXITING) { |
David Rientjes | 4358997 | 2010-08-09 17:18:49 -0700 | [diff] [blame] | 445 | set_tsk_thread_flag(p, TIF_MEMDIE); |
David Rientjes | 2a1c9b1 | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 446 | return; |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 447 | } |
| 448 | |
David Rientjes | dc3f21e | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 449 | if (__ratelimit(&oom_rs)) |
David Rientjes | 8447d95 | 2012-03-21 16:33:47 -0700 | [diff] [blame] | 450 | dump_header(p, gfp_mask, order, memcg, nodemask); |
| 451 | |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 452 | task_lock(p); |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 453 | 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] | 454 | message, task_pid_nr(p), p->comm, points); |
| 455 | task_unlock(p); |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 456 | |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 457 | /* |
| 458 | * 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] | 459 | * the one with the highest oom_badness() score is sacrificed for its |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 460 | * parent. This attempts to lose the minimal amount of work done while |
| 461 | * still freeing memory. |
| 462 | */ |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 463 | do { |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 464 | list_for_each_entry(child, &t->children, sibling) { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 465 | unsigned int child_points; |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 466 | |
David Rientjes | edd4554 | 2011-03-22 16:30:12 -0700 | [diff] [blame] | 467 | if (child->mm == p->mm) |
| 468 | continue; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 469 | /* |
| 470 | * oom_badness() returns 0 if the thread is unkillable |
| 471 | */ |
Johannes Weiner | 72835c8 | 2012-01-12 17:18:32 -0800 | [diff] [blame] | 472 | child_points = oom_badness(child, memcg, nodemask, |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 473 | totalpages); |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 474 | if (child_points > victim_points) { |
| 475 | victim = child; |
| 476 | victim_points = child_points; |
| 477 | } |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 478 | } |
| 479 | } while_each_thread(p, t); |
| 480 | |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 481 | victim = find_lock_task_mm(victim); |
| 482 | if (!victim) |
| 483 | return; |
| 484 | |
| 485 | /* mm cannot safely be dereferenced after task_unlock(victim) */ |
| 486 | mm = victim->mm; |
| 487 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", |
| 488 | task_pid_nr(victim), victim->comm, K(victim->mm->total_vm), |
| 489 | K(get_mm_counter(victim->mm, MM_ANONPAGES)), |
| 490 | K(get_mm_counter(victim->mm, MM_FILEPAGES))); |
| 491 | task_unlock(victim); |
| 492 | |
| 493 | /* |
| 494 | * Kill all user processes sharing victim->mm in other thread groups, if |
| 495 | * any. They don't get access to memory reserves, though, to avoid |
| 496 | * depletion of all memory. This prevents mm->mmap_sem livelock when an |
| 497 | * oom killed thread cannot exit because it requires the semaphore and |
| 498 | * its contended by another thread trying to allocate memory itself. |
| 499 | * That thread will now get access to memory reserves since it has a |
| 500 | * pending fatal signal. |
| 501 | */ |
| 502 | for_each_process(p) |
| 503 | if (p->mm == mm && !same_thread_group(p, victim) && |
| 504 | !(p->flags & PF_KTHREAD)) { |
| 505 | if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) |
| 506 | continue; |
| 507 | |
| 508 | task_lock(p); /* Protect ->comm from prctl() */ |
| 509 | pr_err("Kill process %d (%s) sharing same memory\n", |
| 510 | task_pid_nr(p), p->comm); |
| 511 | task_unlock(p); |
Oleg Nesterov | d2d3930 | 2012-03-23 15:02:45 -0700 | [diff] [blame] | 512 | do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true); |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 513 | } |
| 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); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 517 | } |
David Rientjes | 647f2bd | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 518 | #undef K |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 519 | |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 520 | /* |
| 521 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. |
| 522 | */ |
| 523 | static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask, |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 524 | int order, const nodemask_t *nodemask) |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 525 | { |
| 526 | if (likely(!sysctl_panic_on_oom)) |
| 527 | return; |
| 528 | if (sysctl_panic_on_oom != 2) { |
| 529 | /* |
| 530 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel |
| 531 | * does not panic for cpuset, mempolicy, or memcg allocation |
| 532 | * failures. |
| 533 | */ |
| 534 | if (constraint != CONSTRAINT_NONE) |
| 535 | return; |
| 536 | } |
| 537 | read_lock(&tasklist_lock); |
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 | read_unlock(&tasklist_lock); |
| 540 | panic("Out of memory: %s panic_on_oom is enabled\n", |
| 541 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); |
| 542 | } |
| 543 | |
Balbir Singh | 00f0b82 | 2008-03-04 14:28:39 -0800 | [diff] [blame] | 544 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
David Rientjes | e845e19 | 2012-03-21 16:34:10 -0700 | [diff] [blame] | 545 | void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
| 546 | int order) |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 547 | { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 548 | unsigned long limit; |
| 549 | unsigned int points = 0; |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 550 | struct task_struct *p; |
| 551 | |
David Rientjes | f9434ad | 2011-03-23 16:42:44 -0700 | [diff] [blame] | 552 | /* |
| 553 | * If current has a pending SIGKILL, then automatically select it. The |
| 554 | * goal is to allow it to allocate so that it may quickly exit and free |
| 555 | * its memory. |
| 556 | */ |
| 557 | if (fatal_signal_pending(current)) { |
| 558 | set_thread_flag(TIF_MEMDIE); |
David Rientjes | f9434ad | 2011-03-23 16:42:44 -0700 | [diff] [blame] | 559 | return; |
| 560 | } |
| 561 | |
David Rientjes | e845e19 | 2012-03-21 16:34:10 -0700 | [diff] [blame] | 562 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL); |
David Rientjes | a7f638f | 2012-05-29 15:06:47 -0700 | [diff] [blame] | 563 | limit = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT ? : 1; |
Li Zefan | e115f2d | 2008-04-15 14:34:37 -0700 | [diff] [blame] | 564 | read_lock(&tasklist_lock); |
David Rientjes | 08ab9b1 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 565 | p = select_bad_process(&points, limit, memcg, NULL, false); |
David Rientjes | 2a1c9b1 | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 566 | if (p && PTR_ERR(p) != -1UL) |
David Rientjes | e845e19 | 2012-03-21 16:34:10 -0700 | [diff] [blame] | 567 | oom_kill_process(p, gfp_mask, order, points, limit, memcg, NULL, |
David Rientjes | 2a1c9b1 | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 568 | "Memory cgroup out of memory"); |
Li Zefan | e115f2d | 2008-04-15 14:34:37 -0700 | [diff] [blame] | 569 | read_unlock(&tasklist_lock); |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 570 | } |
| 571 | #endif |
| 572 | |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 573 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
| 574 | |
| 575 | int register_oom_notifier(struct notifier_block *nb) |
| 576 | { |
| 577 | return blocking_notifier_chain_register(&oom_notify_list, nb); |
| 578 | } |
| 579 | EXPORT_SYMBOL_GPL(register_oom_notifier); |
| 580 | |
| 581 | int unregister_oom_notifier(struct notifier_block *nb) |
| 582 | { |
| 583 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); |
| 584 | } |
| 585 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); |
| 586 | |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 587 | /* |
| 588 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero |
| 589 | * if a parallel OOM killing is already taking place that includes a zone in |
| 590 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. |
| 591 | */ |
Minchan Kim | ff321fe | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 592 | int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 593 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 594 | struct zoneref *z; |
| 595 | struct zone *zone; |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 596 | int ret = 1; |
| 597 | |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 598 | spin_lock(&zone_scan_lock); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 599 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 600 | if (zone_is_oom_locked(zone)) { |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 601 | ret = 0; |
| 602 | goto out; |
| 603 | } |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 604 | } |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 605 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 606 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 607 | /* |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 608 | * Lock each zone in the zonelist under zone_scan_lock so a |
Minchan Kim | ff321fe | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 609 | * parallel invocation of try_set_zonelist_oom() doesn't succeed |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 610 | * when it shouldn't. |
| 611 | */ |
| 612 | zone_set_flag(zone, ZONE_OOM_LOCKED); |
| 613 | } |
| 614 | |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 615 | out: |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 616 | spin_unlock(&zone_scan_lock); |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 617 | return ret; |
| 618 | } |
| 619 | |
| 620 | /* |
| 621 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed |
| 622 | * allocation attempts with zonelists containing them may now recall the OOM |
| 623 | * killer, if necessary. |
| 624 | */ |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 625 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 626 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 627 | struct zoneref *z; |
| 628 | struct zone *zone; |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 629 | |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 630 | spin_lock(&zone_scan_lock); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 631 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 632 | zone_clear_flag(zone, ZONE_OOM_LOCKED); |
| 633 | } |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 634 | spin_unlock(&zone_scan_lock); |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 635 | } |
| 636 | |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 637 | /* |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 638 | * Try to acquire the oom killer lock for all system zones. Returns zero if a |
| 639 | * parallel oom killing is taking place, otherwise locks all zones and returns |
| 640 | * non-zero. |
| 641 | */ |
| 642 | static int try_set_system_oom(void) |
| 643 | { |
| 644 | struct zone *zone; |
| 645 | int ret = 1; |
| 646 | |
| 647 | spin_lock(&zone_scan_lock); |
| 648 | for_each_populated_zone(zone) |
| 649 | if (zone_is_oom_locked(zone)) { |
| 650 | ret = 0; |
| 651 | goto out; |
| 652 | } |
| 653 | for_each_populated_zone(zone) |
| 654 | zone_set_flag(zone, ZONE_OOM_LOCKED); |
| 655 | out: |
| 656 | spin_unlock(&zone_scan_lock); |
| 657 | return ret; |
| 658 | } |
| 659 | |
| 660 | /* |
| 661 | * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation |
| 662 | * attempts or page faults may now recall the oom killer, if necessary. |
| 663 | */ |
| 664 | static void clear_system_oom(void) |
| 665 | { |
| 666 | struct zone *zone; |
| 667 | |
| 668 | spin_lock(&zone_scan_lock); |
| 669 | for_each_populated_zone(zone) |
| 670 | zone_clear_flag(zone, ZONE_OOM_LOCKED); |
| 671 | spin_unlock(&zone_scan_lock); |
| 672 | } |
| 673 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 674 | /** |
Dave Peterson | 6937a25 | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 675 | * 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] | 676 | * @zonelist: zonelist pointer |
| 677 | * @gfp_mask: memory allocation flags |
| 678 | * @order: amount of memory being requested as a power of 2 |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 679 | * @nodemask: nodemask passed to page allocator |
David Rientjes | 08ab9b1 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 680 | * @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] | 681 | * |
| 682 | * If we run out of memory, we have the choice between either |
| 683 | * killing a random task (bad), letting the system crash (worse) |
| 684 | * OR try to be smart about which process to kill. Note that we |
| 685 | * don't have to be perfect here, we just have to be good. |
| 686 | */ |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 687 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
David Rientjes | 08ab9b1 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 688 | int order, nodemask_t *nodemask, bool force_kill) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 689 | { |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 690 | const nodemask_t *mpol_mask; |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 691 | struct task_struct *p; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 692 | unsigned long totalpages; |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 693 | unsigned long freed = 0; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 694 | unsigned int points; |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 695 | enum oom_constraint constraint = CONSTRAINT_NONE; |
KOSAKI Motohiro | b52723c | 2010-08-19 14:13:39 -0700 | [diff] [blame] | 696 | int killed = 0; |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 697 | |
| 698 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); |
| 699 | if (freed > 0) |
| 700 | /* Got some memory back in the last second. */ |
| 701 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 702 | |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 703 | /* |
| 704 | * If current has a pending SIGKILL, then automatically select it. The |
| 705 | * goal is to allow it to allocate so that it may quickly exit and free |
| 706 | * its memory. |
| 707 | */ |
| 708 | if (fatal_signal_pending(current)) { |
| 709 | set_thread_flag(TIF_MEMDIE); |
| 710 | return; |
| 711 | } |
| 712 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 713 | /* |
| 714 | * Check if there were limitations on the allocation (only relevant for |
| 715 | * NUMA) that may require different handling. |
| 716 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 717 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask, |
| 718 | &totalpages); |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 719 | mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL; |
| 720 | check_panic_on_oom(constraint, gfp_mask, order, mpol_mask); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 721 | |
David Rientjes | 2b45ab3 | 2007-05-06 14:50:00 -0700 | [diff] [blame] | 722 | read_lock(&tasklist_lock); |
KOSAKI Motohiro | f88ccad | 2010-08-09 17:19:36 -0700 | [diff] [blame] | 723 | if (sysctl_oom_kill_allocating_task && |
KOSAKI Motohiro | a96cfd6 | 2010-08-09 17:19:38 -0700 | [diff] [blame] | 724 | !oom_unkillable_task(current, NULL, nodemask) && |
David Rientjes | c9f0124 | 2011-10-31 17:07:15 -0700 | [diff] [blame] | 725 | current->mm) { |
David Rientjes | 2a1c9b1 | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 726 | oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL, |
| 727 | nodemask, |
| 728 | "Out of memory (oom_kill_allocating_task)"); |
| 729 | goto out; |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 730 | } |
| 731 | |
David Rientjes | 08ab9b1 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 732 | p = select_bad_process(&points, totalpages, NULL, mpol_mask, |
| 733 | force_kill); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 734 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
| 735 | if (!p) { |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 736 | dump_header(NULL, gfp_mask, order, NULL, mpol_mask); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 737 | read_unlock(&tasklist_lock); |
| 738 | panic("Out of memory and no killable processes...\n"); |
| 739 | } |
David Rientjes | 2a1c9b1 | 2012-03-21 16:33:46 -0700 | [diff] [blame] | 740 | if (PTR_ERR(p) != -1UL) { |
| 741 | oom_kill_process(p, gfp_mask, order, points, totalpages, NULL, |
| 742 | nodemask, "Out of memory"); |
| 743 | killed = 1; |
| 744 | } |
KOSAKI Motohiro | b52723c | 2010-08-19 14:13:39 -0700 | [diff] [blame] | 745 | out: |
Andrew Morton | 140ffce | 2006-03-02 02:54:28 -0800 | [diff] [blame] | 746 | read_unlock(&tasklist_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 747 | |
| 748 | /* |
| 749 | * Give "p" a good chance of killing itself before we |
Kirill Korotaev | 2f659f4 | 2006-01-08 01:01:05 -0800 | [diff] [blame] | 750 | * retry to allocate memory unless "p" is current |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 751 | */ |
KOSAKI Motohiro | b52723c | 2010-08-19 14:13:39 -0700 | [diff] [blame] | 752 | if (killed && !test_thread_flag(TIF_MEMDIE)) |
Andrew Morton | 140ffce | 2006-03-02 02:54:28 -0800 | [diff] [blame] | 753 | schedule_timeout_uninterruptible(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 754 | } |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 755 | |
| 756 | /* |
| 757 | * The pagefault handler calls here because it is out of memory, so kill a |
| 758 | * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel |
| 759 | * oom killing is already in progress so do nothing. If a task is found with |
| 760 | * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit. |
| 761 | */ |
| 762 | void pagefault_out_of_memory(void) |
| 763 | { |
| 764 | if (try_set_system_oom()) { |
David Rientjes | 08ab9b1 | 2012-03-21 16:34:04 -0700 | [diff] [blame] | 765 | out_of_memory(NULL, 0, 0, NULL, false); |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 766 | clear_system_oom(); |
| 767 | } |
| 768 | if (!test_thread_flag(TIF_MEMDIE)) |
| 769 | schedule_timeout_uninterruptible(1); |
| 770 | } |