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> |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 29 | #include <linux/module.h> |
| 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> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | |
KAMEZAWA Hiroyuki | fadd8fb | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 36 | int sysctl_panic_on_oom; |
David Rientjes | fe071d7 | 2007-10-16 23:25:56 -0700 | [diff] [blame] | 37 | int sysctl_oom_kill_allocating_task; |
David Rientjes | ad915c4 | 2010-08-09 17:18:53 -0700 | [diff] [blame] | 38 | int sysctl_oom_dump_tasks = 1; |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 39 | static DEFINE_SPINLOCK(zone_scan_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 40 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 41 | #ifdef CONFIG_NUMA |
| 42 | /** |
| 43 | * has_intersects_mems_allowed() - check task eligiblity for kill |
| 44 | * @tsk: task struct of which task to consider |
| 45 | * @mask: nodemask passed to page allocator for mempolicy ooms |
| 46 | * |
| 47 | * Task eligibility is determined by whether or not a candidate task, @tsk, |
| 48 | * shares the same mempolicy nodes as current if it is bound by such a policy |
| 49 | * 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] | 50 | */ |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 51 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
| 52 | const nodemask_t *mask) |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 53 | { |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 54 | struct task_struct *start = tsk; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 55 | |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 56 | do { |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 57 | if (mask) { |
| 58 | /* |
| 59 | * If this is a mempolicy constrained oom, tsk's |
| 60 | * cpuset is irrelevant. Only return true if its |
| 61 | * mempolicy intersects current, otherwise it may be |
| 62 | * needlessly killed. |
| 63 | */ |
| 64 | if (mempolicy_nodemask_intersects(tsk, mask)) |
| 65 | return true; |
| 66 | } else { |
| 67 | /* |
| 68 | * This is not a mempolicy constrained oom, so only |
| 69 | * check the mems of tsk's cpuset. |
| 70 | */ |
| 71 | if (cpuset_mems_allowed_intersects(current, tsk)) |
| 72 | return true; |
| 73 | } |
KOSAKI Motohiro | df1090a | 2010-08-09 17:19:39 -0700 | [diff] [blame] | 74 | } while_each_thread(start, tsk); |
| 75 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 76 | return false; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 77 | } |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 78 | #else |
| 79 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
| 80 | const nodemask_t *mask) |
| 81 | { |
| 82 | return true; |
| 83 | } |
| 84 | #endif /* CONFIG_NUMA */ |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 85 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 86 | /* |
Luis Claudio R. Goncalves | 93b43fa | 2010-08-09 17:19:41 -0700 | [diff] [blame] | 87 | * If this is a system OOM (not a memcg OOM) and the task selected to be |
| 88 | * killed is not already running at high (RT) priorities, speed up the |
| 89 | * recovery by boosting the dying task to the lowest FIFO priority. |
| 90 | * That helps with the recovery and avoids interfering with RT tasks. |
| 91 | */ |
| 92 | static void boost_dying_task_prio(struct task_struct *p, |
| 93 | struct mem_cgroup *mem) |
| 94 | { |
| 95 | struct sched_param param = { .sched_priority = 1 }; |
| 96 | |
| 97 | if (mem) |
| 98 | return; |
| 99 | |
| 100 | if (!rt_task(p)) |
| 101 | sched_setscheduler_nocheck(p, SCHED_FIFO, ¶m); |
| 102 | } |
| 103 | |
| 104 | /* |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 105 | * The process p may have detached its own ->mm while exiting or through |
| 106 | * use_mm(), but one or more of its subthreads may still have a valid |
| 107 | * pointer. Return p, or any of its subthreads with a valid ->mm, with |
| 108 | * task_lock() held. |
| 109 | */ |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 110 | struct task_struct *find_lock_task_mm(struct task_struct *p) |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 111 | { |
| 112 | struct task_struct *t = p; |
| 113 | |
| 114 | do { |
| 115 | task_lock(t); |
| 116 | if (likely(t->mm)) |
| 117 | return t; |
| 118 | task_unlock(t); |
| 119 | } while_each_thread(p, t); |
| 120 | |
| 121 | return NULL; |
| 122 | } |
| 123 | |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 124 | /* return true if the task is not adequate as candidate victim task. */ |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 125 | static bool oom_unkillable_task(struct task_struct *p, |
| 126 | const struct mem_cgroup *mem, const nodemask_t *nodemask) |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 127 | { |
| 128 | if (is_global_init(p)) |
| 129 | return true; |
| 130 | if (p->flags & PF_KTHREAD) |
| 131 | return true; |
| 132 | |
| 133 | /* When mem_cgroup_out_of_memory() and p is not member of the group */ |
| 134 | if (mem && !task_in_mem_cgroup(p, mem)) |
| 135 | return true; |
| 136 | |
| 137 | /* p may not have freeable memory in nodemask */ |
| 138 | if (!has_intersects_mems_allowed(p, nodemask)) |
| 139 | return true; |
| 140 | |
| 141 | return false; |
| 142 | } |
| 143 | |
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 | * oom_badness - heuristic function to determine which candidate task to kill |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 146 | * @p: task struct of which task we should calculate |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 147 | * @totalpages: total present RAM allowed for page allocation |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 148 | * |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 149 | * The heuristic for determining which task to kill is made to be as simple and |
| 150 | * predictable as possible. The goal is to return the highest value for the |
| 151 | * task consuming the most memory to avoid subsequent oom failures. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 152 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 153 | unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem, |
| 154 | const nodemask_t *nodemask, unsigned long totalpages) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 155 | { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 156 | int points; |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 157 | |
KOSAKI Motohiro | 26ebc98 | 2010-08-09 17:19:37 -0700 | [diff] [blame] | 158 | if (oom_unkillable_task(p, mem, nodemask)) |
| 159 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 160 | |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 161 | p = find_lock_task_mm(p); |
| 162 | if (!p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 163 | return 0; |
| 164 | |
| 165 | /* |
David Rientjes | e18641e | 2010-10-26 14:21:23 -0700 | [diff] [blame] | 166 | * Shortcut check for a thread sharing p->mm that is OOM_SCORE_ADJ_MIN |
| 167 | * so the entire heuristic doesn't need to be executed for something |
| 168 | * that cannot be killed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 169 | */ |
David Rientjes | e18641e | 2010-10-26 14:21:23 -0700 | [diff] [blame] | 170 | if (atomic_read(&p->mm->oom_disable_count)) { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 171 | task_unlock(p); |
| 172 | return 0; |
| 173 | } |
| 174 | |
| 175 | /* |
| 176 | * When the PF_OOM_ORIGIN bit is set, it indicates the task should have |
| 177 | * priority for oom killing. |
| 178 | */ |
| 179 | if (p->flags & PF_OOM_ORIGIN) { |
| 180 | task_unlock(p); |
| 181 | return 1000; |
| 182 | } |
| 183 | |
| 184 | /* |
| 185 | * The memory controller may have a limit of 0 bytes, so avoid a divide |
| 186 | * by zero, if necessary. |
| 187 | */ |
| 188 | if (!totalpages) |
| 189 | totalpages = 1; |
| 190 | |
| 191 | /* |
| 192 | * The baseline for the badness score is the proportion of RAM that each |
| 193 | * task's rss and swap space use. |
| 194 | */ |
| 195 | points = (get_mm_rss(p->mm) + get_mm_counter(p->mm, MM_SWAPENTS)) * 1000 / |
| 196 | totalpages; |
Andrew Morton | 97c2c9b8 | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 197 | task_unlock(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 198 | |
| 199 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 200 | * Root processes get 3% bonus, just like the __vm_enough_memory() |
| 201 | * implementation used by LSMs. |
Hugh Dickins | 7ba3485 | 2007-01-05 16:37:03 -0800 | [diff] [blame] | 202 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 203 | if (has_capability_noaudit(p, CAP_SYS_ADMIN)) |
| 204 | points -= 30; |
Hugh Dickins | 7ba3485 | 2007-01-05 16:37:03 -0800 | [diff] [blame] | 205 | |
| 206 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 207 | * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may |
| 208 | * either completely disable oom killing or always prefer a certain |
| 209 | * task. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 211 | points += p->signal->oom_score_adj; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 212 | |
David Rientjes | f19e8aa | 2010-09-22 13:04:52 -0700 | [diff] [blame] | 213 | /* |
| 214 | * Never return 0 for an eligible task that may be killed since it's |
| 215 | * possible that no single user task uses more than 0.1% of memory and |
| 216 | * no single admin tasks uses more than 3.0%. |
| 217 | */ |
| 218 | if (points <= 0) |
| 219 | return 1; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 220 | return (points < 1000) ? points : 1000; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 221 | } |
| 222 | |
| 223 | /* |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 224 | * Determine the type of allocation constraint. |
| 225 | */ |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 226 | #ifdef CONFIG_NUMA |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 227 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 228 | gfp_t gfp_mask, nodemask_t *nodemask, |
| 229 | unsigned long *totalpages) |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 230 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 231 | struct zone *zone; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 232 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 233 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 234 | bool cpuset_limited = false; |
| 235 | int nid; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 236 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 237 | /* Default to all available memory */ |
| 238 | *totalpages = totalram_pages + total_swap_pages; |
| 239 | |
| 240 | if (!zonelist) |
| 241 | return CONSTRAINT_NONE; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 242 | /* |
| 243 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid |
| 244 | * to kill current.We have to random task kill in this case. |
| 245 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. |
| 246 | */ |
| 247 | if (gfp_mask & __GFP_THISNODE) |
| 248 | return CONSTRAINT_NONE; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 249 | |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 250 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 251 | * This is not a __GFP_THISNODE allocation, so a truncated nodemask in |
| 252 | * the page allocator means a mempolicy is in effect. Cpuset policy |
| 253 | * is enforced in get_page_from_freelist(). |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 254 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 255 | if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) { |
| 256 | *totalpages = total_swap_pages; |
| 257 | for_each_node_mask(nid, *nodemask) |
| 258 | *totalpages += node_spanned_pages(nid); |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 259 | return CONSTRAINT_MEMORY_POLICY; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 260 | } |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 261 | |
| 262 | /* Check this allocation failure is caused by cpuset's wall function */ |
| 263 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 264 | high_zoneidx, nodemask) |
| 265 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 266 | cpuset_limited = true; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 267 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 268 | if (cpuset_limited) { |
| 269 | *totalpages = total_swap_pages; |
| 270 | for_each_node_mask(nid, cpuset_current_mems_allowed) |
| 271 | *totalpages += node_spanned_pages(nid); |
| 272 | return CONSTRAINT_CPUSET; |
| 273 | } |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 274 | return CONSTRAINT_NONE; |
| 275 | } |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 276 | #else |
| 277 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 278 | gfp_t gfp_mask, nodemask_t *nodemask, |
| 279 | unsigned long *totalpages) |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 280 | { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 281 | *totalpages = totalram_pages + total_swap_pages; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 282 | return CONSTRAINT_NONE; |
| 283 | } |
| 284 | #endif |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 285 | |
| 286 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 287 | * Simple selection loop. We chose the process with the highest |
| 288 | * number of 'points'. We expect the caller will lock the tasklist. |
| 289 | * |
| 290 | * (not docbooked, we don't want this one cluttering up the manual) |
| 291 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 292 | static struct task_struct *select_bad_process(unsigned int *ppoints, |
| 293 | unsigned long totalpages, struct mem_cgroup *mem, |
| 294 | const nodemask_t *nodemask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 295 | { |
David Rientjes | 3a5dda7 | 2011-03-22 16:30:09 -0700 | [diff] [blame] | 296 | struct task_struct *g, *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 297 | struct task_struct *chosen = NULL; |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 298 | *ppoints = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 299 | |
David Rientjes | 3a5dda7 | 2011-03-22 16:30:09 -0700 | [diff] [blame] | 300 | do_each_thread(g, p) { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 301 | unsigned int points; |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 302 | |
Andrey Vagin | 30e2b41 | 2011-03-22 16:30:11 -0700 | [diff] [blame] | 303 | if (!p->mm) |
| 304 | continue; |
KOSAKI Motohiro | ab290ad | 2010-08-09 17:19:35 -0700 | [diff] [blame] | 305 | if (oom_unkillable_task(p, mem, nodemask)) |
David Rientjes | 6cf86ac | 2010-08-09 17:18:50 -0700 | [diff] [blame] | 306 | continue; |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 307 | |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 308 | /* |
Nick Piggin | b78483a | 2006-09-29 02:01:14 -0700 | [diff] [blame] | 309 | * This task already has access to memory reserves and is |
| 310 | * being killed. Don't allow any other task access to the |
| 311 | * memory reserve. |
| 312 | * |
| 313 | * Note: this may have a chance of deadlock if it gets |
| 314 | * blocked waiting for another task which itself is waiting |
| 315 | * for memory. Is there a better alternative? |
| 316 | */ |
| 317 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) |
| 318 | return ERR_PTR(-1UL); |
| 319 | |
Andrey Vagin | 30e2b41 | 2011-03-22 16:30:11 -0700 | [diff] [blame] | 320 | if (p->flags & PF_EXITING) { |
David Rientjes | edd4554 | 2011-03-22 16:30:12 -0700 | [diff] [blame] | 321 | /* |
| 322 | * If p is the current task and is in the process of |
| 323 | * releasing memory, we allow the "kill" to set |
| 324 | * TIF_MEMDIE, which will allow it to gain access to |
| 325 | * memory reserves. Otherwise, it may stall forever. |
| 326 | * |
| 327 | * The loop isn't broken here, however, in case other |
| 328 | * threads are found to have already been oom killed. |
| 329 | */ |
| 330 | if (p == current) { |
| 331 | chosen = p; |
| 332 | *ppoints = 1000; |
| 333 | } else { |
| 334 | /* |
| 335 | * If this task is not being ptraced on exit, |
| 336 | * then wait for it to finish before killing |
| 337 | * some other task unnecessarily. |
| 338 | */ |
| 339 | if (!(task_ptrace(p->group_leader) & |
| 340 | PT_TRACE_EXIT)) |
| 341 | return ERR_PTR(-1UL); |
| 342 | } |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 343 | } |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 344 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 345 | points = oom_badness(p, mem, nodemask, totalpages); |
| 346 | if (points > *ppoints) { |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 347 | chosen = p; |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 348 | *ppoints = points; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 349 | } |
David Rientjes | 3a5dda7 | 2011-03-22 16:30:09 -0700 | [diff] [blame] | 350 | } while_each_thread(g, p); |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 351 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 352 | return chosen; |
| 353 | } |
| 354 | |
| 355 | /** |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 356 | * dump_tasks - dump current memory state of all system tasks |
David Rientjes | 74ab7f1 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 357 | * @mem: current's memory controller, if constrained |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 358 | * @nodemask: nodemask passed to page allocator for mempolicy ooms |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 359 | * |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 360 | * Dumps the current memory state of all eligible tasks. Tasks not in the same |
| 361 | * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes |
| 362 | * are not shown. |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 363 | * 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] | 364 | * value, oom_score_adj value, and name. |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 365 | * |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 366 | * Call with tasklist_lock read-locked. |
| 367 | */ |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 368 | static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask) |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 369 | { |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 370 | struct task_struct *p; |
| 371 | struct task_struct *task; |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 372 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 373 | 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] | 374 | for_each_process(p) { |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 375 | if (oom_unkillable_task(p, mem, nodemask)) |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 376 | continue; |
| 377 | |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 378 | task = find_lock_task_mm(p); |
| 379 | if (!task) { |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 380 | /* |
David Rientjes | 74ab7f1 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 381 | * This is a kthread or all of p's threads have already |
| 382 | * detached their mm's. There's no need to report |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 383 | * them; they can't be oom killed anyway. |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 384 | */ |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 385 | continue; |
| 386 | } |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 387 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 388 | pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n", |
KOSAKI Motohiro | 8d6c83f | 2010-08-19 14:13:39 -0700 | [diff] [blame] | 389 | task->pid, task_uid(task), task->tgid, |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 390 | task->mm->total_vm, get_mm_rss(task->mm), |
| 391 | task_cpu(task), task->signal->oom_adj, |
| 392 | task->signal->oom_score_adj, task->comm); |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 393 | task_unlock(task); |
| 394 | } |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 395 | } |
| 396 | |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 397 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 398 | struct mem_cgroup *mem, const nodemask_t *nodemask) |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 399 | { |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 400 | task_lock(current); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 401 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 402 | "oom_adj=%d, oom_score_adj=%d\n", |
| 403 | current->comm, gfp_mask, order, current->signal->oom_adj, |
| 404 | current->signal->oom_score_adj); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 405 | cpuset_print_task_mems_allowed(current); |
| 406 | task_unlock(current); |
| 407 | dump_stack(); |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 408 | mem_cgroup_print_oom_info(mem, p); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 409 | show_mem(); |
| 410 | if (sysctl_oom_dump_tasks) |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 411 | dump_tasks(mem, nodemask); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 412 | } |
| 413 | |
KOSAKI Motohiro | 3b4798c | 2009-12-15 16:45:32 -0800 | [diff] [blame] | 414 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
Luis Claudio R. Goncalves | 93b43fa | 2010-08-09 17:19:41 -0700 | [diff] [blame] | 415 | static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | { |
David Rientjes | 1e99bad | 2010-10-26 14:21:24 -0700 | [diff] [blame] | 417 | struct task_struct *q; |
| 418 | struct mm_struct *mm; |
| 419 | |
David Rientjes | b940fd7 | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 420 | p = find_lock_task_mm(p); |
KOSAKI Motohiro | be71cf2 | 2010-08-19 14:13:38 -0700 | [diff] [blame] | 421 | if (!p) |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 422 | return 1; |
KOSAKI Motohiro | be71cf2 | 2010-08-19 14:13:38 -0700 | [diff] [blame] | 423 | |
David Rientjes | 1e99bad | 2010-10-26 14:21:24 -0700 | [diff] [blame] | 424 | /* mm cannot be safely dereferenced after task_unlock(p) */ |
| 425 | mm = p->mm; |
| 426 | |
David Rientjes | b940fd7 | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 427 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", |
| 428 | task_pid_nr(p), p->comm, K(p->mm->total_vm), |
| 429 | K(get_mm_counter(p->mm, MM_ANONPAGES)), |
| 430 | K(get_mm_counter(p->mm, MM_FILEPAGES))); |
| 431 | task_unlock(p); |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 432 | |
David Rientjes | 1e99bad | 2010-10-26 14:21:24 -0700 | [diff] [blame] | 433 | /* |
| 434 | * Kill all processes sharing p->mm in other thread groups, if any. |
| 435 | * They don't get access to memory reserves or a higher scheduler |
| 436 | * priority, though, to avoid depletion of all memory or task |
| 437 | * starvation. This prevents mm->mmap_sem livelock when an oom killed |
| 438 | * task cannot exit because it requires the semaphore and its contended |
| 439 | * by another thread trying to allocate memory itself. That thread will |
| 440 | * now get access to memory reserves since it has a pending fatal |
| 441 | * signal. |
| 442 | */ |
| 443 | for_each_process(q) |
| 444 | if (q->mm == mm && !same_thread_group(q, p)) { |
| 445 | task_lock(q); /* Protect ->comm from prctl() */ |
| 446 | pr_err("Kill process %d (%s) sharing same memory\n", |
| 447 | task_pid_nr(q), q->comm); |
| 448 | task_unlock(q); |
| 449 | force_sig(SIGKILL, q); |
| 450 | } |
Luis Claudio R. Goncalves | 93b43fa | 2010-08-09 17:19:41 -0700 | [diff] [blame] | 451 | |
David Rientjes | b940fd7 | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 452 | set_tsk_thread_flag(p, TIF_MEMDIE); |
| 453 | force_sig(SIGKILL, p); |
Luis Claudio R. Goncalves | 93b43fa | 2010-08-09 17:19:41 -0700 | [diff] [blame] | 454 | |
| 455 | /* |
| 456 | * We give our sacrificial lamb high priority and access to |
| 457 | * all the memory it needs. That way it should be able to |
| 458 | * exit() and clear out its resources quickly... |
| 459 | */ |
| 460 | boost_dying_task_prio(p, mem); |
| 461 | |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 462 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 463 | } |
David Rientjes | b940fd7 | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 464 | #undef K |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 465 | |
David Rientjes | 7213f50 | 2007-10-16 23:25:57 -0700 | [diff] [blame] | 466 | static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 467 | unsigned int points, unsigned long totalpages, |
| 468 | struct mem_cgroup *mem, nodemask_t *nodemask, |
| 469 | const char *message) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | { |
Linus Torvalds | 52d3c03 | 2011-03-14 15:17:07 -0700 | [diff] [blame] | 471 | struct task_struct *victim = p; |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 472 | struct task_struct *child; |
Linus Torvalds | 52d3c03 | 2011-03-14 15:17:07 -0700 | [diff] [blame] | 473 | struct task_struct *t = p; |
| 474 | unsigned int victim_points = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 475 | |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 476 | if (printk_ratelimit()) |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 477 | dump_header(p, gfp_mask, order, mem, nodemask); |
David Rientjes | 7213f50 | 2007-10-16 23:25:57 -0700 | [diff] [blame] | 478 | |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 479 | /* |
| 480 | * If the task is already exiting, don't alarm the sysadmin or kill |
| 481 | * its children or threads, just set TIF_MEMDIE so it can die quickly |
| 482 | */ |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 483 | if (p->flags & PF_EXITING) { |
David Rientjes | 4358997 | 2010-08-09 17:18:49 -0700 | [diff] [blame] | 484 | set_tsk_thread_flag(p, TIF_MEMDIE); |
Luis Claudio R. Goncalves | 93b43fa | 2010-08-09 17:19:41 -0700 | [diff] [blame] | 485 | boost_dying_task_prio(p, mem); |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 486 | return 0; |
| 487 | } |
| 488 | |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 489 | task_lock(p); |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 490 | 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] | 491 | message, task_pid_nr(p), p->comm, points); |
| 492 | task_unlock(p); |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 493 | |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 494 | /* |
| 495 | * If any of p's children has a different mm and is eligible for kill, |
| 496 | * the one with the highest badness() score is sacrificed for its |
| 497 | * parent. This attempts to lose the minimal amount of work done while |
| 498 | * still freeing memory. |
| 499 | */ |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 500 | do { |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 501 | list_for_each_entry(child, &t->children, sibling) { |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 502 | unsigned int child_points; |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 503 | |
David Rientjes | edd4554 | 2011-03-22 16:30:12 -0700 | [diff] [blame] | 504 | if (child->mm == p->mm) |
| 505 | continue; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 506 | /* |
| 507 | * oom_badness() returns 0 if the thread is unkillable |
| 508 | */ |
| 509 | child_points = oom_badness(child, mem, nodemask, |
| 510 | totalpages); |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 511 | if (child_points > victim_points) { |
| 512 | victim = child; |
| 513 | victim_points = child_points; |
| 514 | } |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 515 | } |
| 516 | } while_each_thread(p, t); |
| 517 | |
Luis Claudio R. Goncalves | 93b43fa | 2010-08-09 17:19:41 -0700 | [diff] [blame] | 518 | return oom_kill_task(victim, mem); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 519 | } |
| 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 | */ |
| 524 | 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] | 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 | } |
| 538 | read_lock(&tasklist_lock); |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 539 | dump_header(NULL, gfp_mask, order, NULL, nodemask); |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 540 | read_unlock(&tasklist_lock); |
| 541 | panic("Out of memory: %s panic_on_oom is enabled\n", |
| 542 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); |
| 543 | } |
| 544 | |
Balbir Singh | 00f0b82 | 2008-03-04 14:28:39 -0800 | [diff] [blame] | 545 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 546 | void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) |
| 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 | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 552 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL); |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 553 | limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT; |
Li Zefan | e115f2d | 2008-04-15 14:34:37 -0700 | [diff] [blame] | 554 | read_lock(&tasklist_lock); |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 555 | retry: |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 556 | p = select_bad_process(&points, limit, mem, NULL); |
David Rientjes | df64f81 | 2010-05-26 14:42:41 -0700 | [diff] [blame] | 557 | if (!p || PTR_ERR(p) == -1UL) |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 558 | goto out; |
| 559 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 560 | if (oom_kill_process(p, gfp_mask, 0, points, limit, mem, NULL, |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 561 | "Memory cgroup out of memory")) |
| 562 | goto retry; |
| 563 | out: |
Li Zefan | e115f2d | 2008-04-15 14:34:37 -0700 | [diff] [blame] | 564 | read_unlock(&tasklist_lock); |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 565 | } |
| 566 | #endif |
| 567 | |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 568 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
| 569 | |
| 570 | int register_oom_notifier(struct notifier_block *nb) |
| 571 | { |
| 572 | return blocking_notifier_chain_register(&oom_notify_list, nb); |
| 573 | } |
| 574 | EXPORT_SYMBOL_GPL(register_oom_notifier); |
| 575 | |
| 576 | int unregister_oom_notifier(struct notifier_block *nb) |
| 577 | { |
| 578 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); |
| 579 | } |
| 580 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); |
| 581 | |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 582 | /* |
| 583 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero |
| 584 | * if a parallel OOM killing is already taking place that includes a zone in |
| 585 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. |
| 586 | */ |
Minchan Kim | ff321fe | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 587 | int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 588 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 589 | struct zoneref *z; |
| 590 | struct zone *zone; |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 591 | int ret = 1; |
| 592 | |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 593 | spin_lock(&zone_scan_lock); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 594 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 595 | if (zone_is_oom_locked(zone)) { |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 596 | ret = 0; |
| 597 | goto out; |
| 598 | } |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 599 | } |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 600 | |
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 | /* |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 603 | * Lock each zone in the zonelist under zone_scan_lock so a |
Minchan Kim | ff321fe | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 604 | * parallel invocation of try_set_zonelist_oom() doesn't succeed |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 605 | * when it shouldn't. |
| 606 | */ |
| 607 | zone_set_flag(zone, ZONE_OOM_LOCKED); |
| 608 | } |
| 609 | |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 610 | out: |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 611 | spin_unlock(&zone_scan_lock); |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 612 | return ret; |
| 613 | } |
| 614 | |
| 615 | /* |
| 616 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed |
| 617 | * allocation attempts with zonelists containing them may now recall the OOM |
| 618 | * killer, if necessary. |
| 619 | */ |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 620 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 621 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 622 | struct zoneref *z; |
| 623 | struct zone *zone; |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 624 | |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 625 | spin_lock(&zone_scan_lock); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 626 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 627 | zone_clear_flag(zone, ZONE_OOM_LOCKED); |
| 628 | } |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 629 | spin_unlock(&zone_scan_lock); |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 630 | } |
| 631 | |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 632 | /* |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 633 | * Try to acquire the oom killer lock for all system zones. Returns zero if a |
| 634 | * parallel oom killing is taking place, otherwise locks all zones and returns |
| 635 | * non-zero. |
| 636 | */ |
| 637 | static int try_set_system_oom(void) |
| 638 | { |
| 639 | struct zone *zone; |
| 640 | int ret = 1; |
| 641 | |
| 642 | spin_lock(&zone_scan_lock); |
| 643 | for_each_populated_zone(zone) |
| 644 | if (zone_is_oom_locked(zone)) { |
| 645 | ret = 0; |
| 646 | goto out; |
| 647 | } |
| 648 | for_each_populated_zone(zone) |
| 649 | zone_set_flag(zone, ZONE_OOM_LOCKED); |
| 650 | out: |
| 651 | spin_unlock(&zone_scan_lock); |
| 652 | return ret; |
| 653 | } |
| 654 | |
| 655 | /* |
| 656 | * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation |
| 657 | * attempts or page faults may now recall the oom killer, if necessary. |
| 658 | */ |
| 659 | static void clear_system_oom(void) |
| 660 | { |
| 661 | struct zone *zone; |
| 662 | |
| 663 | spin_lock(&zone_scan_lock); |
| 664 | for_each_populated_zone(zone) |
| 665 | zone_clear_flag(zone, ZONE_OOM_LOCKED); |
| 666 | spin_unlock(&zone_scan_lock); |
| 667 | } |
| 668 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 669 | /** |
Dave Peterson | 6937a25 | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 670 | * 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] | 671 | * @zonelist: zonelist pointer |
| 672 | * @gfp_mask: memory allocation flags |
| 673 | * @order: amount of memory being requested as a power of 2 |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 674 | * @nodemask: nodemask passed to page allocator |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 675 | * |
| 676 | * If we run out of memory, we have the choice between either |
| 677 | * killing a random task (bad), letting the system crash (worse) |
| 678 | * OR try to be smart about which process to kill. Note that we |
| 679 | * don't have to be perfect here, we just have to be good. |
| 680 | */ |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 681 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
| 682 | int order, nodemask_t *nodemask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 683 | { |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 684 | const nodemask_t *mpol_mask; |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 685 | struct task_struct *p; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 686 | unsigned long totalpages; |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 687 | unsigned long freed = 0; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 688 | unsigned int points; |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 689 | enum oom_constraint constraint = CONSTRAINT_NONE; |
KOSAKI Motohiro | b52723c | 2010-08-19 14:13:39 -0700 | [diff] [blame] | 690 | int killed = 0; |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 691 | |
| 692 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); |
| 693 | if (freed > 0) |
| 694 | /* Got some memory back in the last second. */ |
| 695 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 696 | |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 697 | /* |
| 698 | * If current has a pending SIGKILL, then automatically select it. The |
| 699 | * goal is to allow it to allocate so that it may quickly exit and free |
| 700 | * its memory. |
| 701 | */ |
| 702 | if (fatal_signal_pending(current)) { |
| 703 | set_thread_flag(TIF_MEMDIE); |
Luis Claudio R. Goncalves | 93b43fa | 2010-08-09 17:19:41 -0700 | [diff] [blame] | 704 | boost_dying_task_prio(current, NULL); |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 705 | return; |
| 706 | } |
| 707 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 708 | /* |
| 709 | * Check if there were limitations on the allocation (only relevant for |
| 710 | * NUMA) that may require different handling. |
| 711 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 712 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask, |
| 713 | &totalpages); |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 714 | mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL; |
| 715 | check_panic_on_oom(constraint, gfp_mask, order, mpol_mask); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 716 | |
David Rientjes | 2b45ab3 | 2007-05-06 14:50:00 -0700 | [diff] [blame] | 717 | read_lock(&tasklist_lock); |
KOSAKI Motohiro | f88ccad | 2010-08-09 17:19:36 -0700 | [diff] [blame] | 718 | if (sysctl_oom_kill_allocating_task && |
KOSAKI Motohiro | a96cfd6 | 2010-08-09 17:19:38 -0700 | [diff] [blame] | 719 | !oom_unkillable_task(current, NULL, nodemask) && |
David Rientjes | e18641e | 2010-10-26 14:21:23 -0700 | [diff] [blame] | 720 | current->mm && !atomic_read(¤t->mm->oom_disable_count)) { |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 721 | /* |
| 722 | * oom_kill_process() needs tasklist_lock held. If it returns |
| 723 | * non-zero, current could not be killed so we must fallback to |
| 724 | * the tasklist scan. |
| 725 | */ |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 726 | if (!oom_kill_process(current, gfp_mask, order, 0, totalpages, |
| 727 | NULL, nodemask, |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 728 | "Out of memory (oom_kill_allocating_task)")) |
KOSAKI Motohiro | b52723c | 2010-08-19 14:13:39 -0700 | [diff] [blame] | 729 | goto out; |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 730 | } |
| 731 | |
| 732 | retry: |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 733 | p = select_bad_process(&points, totalpages, NULL, mpol_mask); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 734 | if (PTR_ERR(p) == -1UL) |
KOSAKI Motohiro | b52723c | 2010-08-19 14:13:39 -0700 | [diff] [blame] | 735 | goto out; |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 736 | |
| 737 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
| 738 | if (!p) { |
David Rientjes | e85bfd3 | 2010-09-22 13:05:10 -0700 | [diff] [blame] | 739 | dump_header(NULL, gfp_mask, order, NULL, mpol_mask); |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 740 | read_unlock(&tasklist_lock); |
| 741 | panic("Out of memory and no killable processes...\n"); |
| 742 | } |
| 743 | |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 744 | if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL, |
| 745 | nodemask, "Out of memory")) |
David Rientjes | 0aad4b3 | 2010-08-09 17:18:59 -0700 | [diff] [blame] | 746 | goto retry; |
KOSAKI Motohiro | b52723c | 2010-08-19 14:13:39 -0700 | [diff] [blame] | 747 | killed = 1; |
| 748 | out: |
Andrew Morton | 140ffce | 2006-03-02 02:54:28 -0800 | [diff] [blame] | 749 | read_unlock(&tasklist_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 750 | |
| 751 | /* |
| 752 | * Give "p" a good chance of killing itself before we |
Kirill Korotaev | 2f659f4 | 2006-01-08 01:01:05 -0800 | [diff] [blame] | 753 | * retry to allocate memory unless "p" is current |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 754 | */ |
KOSAKI Motohiro | b52723c | 2010-08-19 14:13:39 -0700 | [diff] [blame] | 755 | if (killed && !test_thread_flag(TIF_MEMDIE)) |
Andrew Morton | 140ffce | 2006-03-02 02:54:28 -0800 | [diff] [blame] | 756 | schedule_timeout_uninterruptible(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 757 | } |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 758 | |
| 759 | /* |
| 760 | * The pagefault handler calls here because it is out of memory, so kill a |
| 761 | * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel |
| 762 | * oom killing is already in progress so do nothing. If a task is found with |
| 763 | * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit. |
| 764 | */ |
| 765 | void pagefault_out_of_memory(void) |
| 766 | { |
| 767 | if (try_set_system_oom()) { |
| 768 | out_of_memory(NULL, 0, 0, NULL); |
| 769 | clear_system_oom(); |
| 770 | } |
| 771 | if (!test_thread_flag(TIF_MEMDIE)) |
| 772 | schedule_timeout_uninterruptible(1); |
| 773 | } |