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... |
| 7 | * |
| 8 | * The routines in this file are used to kill a process when |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 9 | * we're seriously out of memory. This gets called from __alloc_pages() |
| 10 | * in mm/page_alloc.c when we really run out of memory. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 11 | * |
| 12 | * Since we won't call these routines often (on a well-configured |
| 13 | * machine) this file will double as a 'coding guide' and a signpost |
| 14 | * for newbie kernel hackers. It features several pointers to major |
| 15 | * kernel subsystems and hints as to where to find out what things do. |
| 16 | */ |
| 17 | |
| 18 | #include <linux/mm.h> |
| 19 | #include <linux/sched.h> |
| 20 | #include <linux/swap.h> |
| 21 | #include <linux/timex.h> |
| 22 | #include <linux/jiffies.h> |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 23 | #include <linux/cpuset.h> |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 24 | #include <linux/module.h> |
| 25 | #include <linux/notifier.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 26 | |
KAMEZAWA Hiroyuki | fadd8fb | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 27 | int sysctl_panic_on_oom; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 28 | /* #define DEBUG */ |
| 29 | |
| 30 | /** |
Dave Peterson | 6937a25 | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 31 | * badness - calculate a numeric value for how bad this task has been |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | * @p: task struct of which task we should calculate |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 33 | * @uptime: current uptime in seconds |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 34 | * |
| 35 | * The formula used is relatively simple and documented inline in the |
| 36 | * function. The main rationale is that we want to select a good task |
| 37 | * to kill when we run out of memory. |
| 38 | * |
| 39 | * Good in this context means that: |
| 40 | * 1) we lose the minimum amount of work done |
| 41 | * 2) we recover a large amount of memory |
| 42 | * 3) we don't kill anything innocent of eating tons of memory |
| 43 | * 4) we want to kill the minimum amount of processes (one) |
| 44 | * 5) we try to kill the process the user expects us to kill, this |
| 45 | * algorithm has been meticulously tuned to meet the principle |
| 46 | * of least surprise ... (be careful when you change it) |
| 47 | */ |
| 48 | |
| 49 | unsigned long badness(struct task_struct *p, unsigned long uptime) |
| 50 | { |
| 51 | unsigned long points, cpu_time, run_time, s; |
Andrew Morton | 97c2c9b8 | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 52 | struct mm_struct *mm; |
| 53 | struct task_struct *child; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | |
Andrew Morton | 97c2c9b8 | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 55 | task_lock(p); |
| 56 | mm = p->mm; |
| 57 | if (!mm) { |
| 58 | task_unlock(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | return 0; |
Andrew Morton | 97c2c9b8 | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 60 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 61 | |
| 62 | /* |
Nick Piggin | af5b912 | 2006-09-25 23:31:31 -0700 | [diff] [blame] | 63 | * swapoff can easily use up all memory, so kill those first. |
| 64 | */ |
| 65 | if (p->flags & PF_SWAPOFF) |
| 66 | return ULONG_MAX; |
| 67 | |
| 68 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 69 | * The memory size of the process is the basis for the badness. |
| 70 | */ |
Andrew Morton | 97c2c9b8 | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 71 | points = mm->total_vm; |
| 72 | |
| 73 | /* |
| 74 | * After this unlock we can no longer dereference local variable `mm' |
| 75 | */ |
| 76 | task_unlock(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 77 | |
| 78 | /* |
| 79 | * Processes which fork a lot of child processes are likely |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 80 | * a good choice. We add half the vmsize of the children if they |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 81 | * have an own mm. This prevents forking servers to flood the |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 82 | * machine with an endless amount of children. In case a single |
| 83 | * child is eating the vast majority of memory, adding only half |
| 84 | * to the parents will make the child our kill candidate of choice. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 85 | */ |
Andrew Morton | 97c2c9b8 | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 86 | list_for_each_entry(child, &p->children, sibling) { |
| 87 | task_lock(child); |
| 88 | if (child->mm != mm && child->mm) |
| 89 | points += child->mm->total_vm/2 + 1; |
| 90 | task_unlock(child); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 91 | } |
| 92 | |
| 93 | /* |
| 94 | * CPU time is in tens of seconds and run time is in thousands |
| 95 | * of seconds. There is no particular reason for this other than |
| 96 | * that it turned out to work very well in practice. |
| 97 | */ |
| 98 | cpu_time = (cputime_to_jiffies(p->utime) + cputime_to_jiffies(p->stime)) |
| 99 | >> (SHIFT_HZ + 3); |
| 100 | |
| 101 | if (uptime >= p->start_time.tv_sec) |
| 102 | run_time = (uptime - p->start_time.tv_sec) >> 10; |
| 103 | else |
| 104 | run_time = 0; |
| 105 | |
| 106 | s = int_sqrt(cpu_time); |
| 107 | if (s) |
| 108 | points /= s; |
| 109 | s = int_sqrt(int_sqrt(run_time)); |
| 110 | if (s) |
| 111 | points /= s; |
| 112 | |
| 113 | /* |
| 114 | * Niced processes are most likely less important, so double |
| 115 | * their badness points. |
| 116 | */ |
| 117 | if (task_nice(p) > 0) |
| 118 | points *= 2; |
| 119 | |
| 120 | /* |
| 121 | * Superuser processes are usually more important, so we make it |
| 122 | * less likely that we kill those. |
| 123 | */ |
| 124 | if (cap_t(p->cap_effective) & CAP_TO_MASK(CAP_SYS_ADMIN) || |
| 125 | p->uid == 0 || p->euid == 0) |
| 126 | points /= 4; |
| 127 | |
| 128 | /* |
| 129 | * We don't want to kill a process with direct hardware access. |
| 130 | * Not only could that mess up the hardware, but usually users |
| 131 | * tend to only have this flag set on applications they think |
| 132 | * of as important. |
| 133 | */ |
| 134 | if (cap_t(p->cap_effective) & CAP_TO_MASK(CAP_SYS_RAWIO)) |
| 135 | points /= 4; |
| 136 | |
| 137 | /* |
Nick Piggin | 7887a3d | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 138 | * If p's nodes don't overlap ours, it may still help to kill p |
| 139 | * because p may have allocated or otherwise mapped memory on |
| 140 | * this node before. However it will be less likely. |
| 141 | */ |
| 142 | if (!cpuset_excl_nodes_overlap(p)) |
| 143 | points /= 8; |
| 144 | |
| 145 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 146 | * Adjust the score by oomkilladj. |
| 147 | */ |
| 148 | if (p->oomkilladj) { |
| 149 | if (p->oomkilladj > 0) |
| 150 | points <<= p->oomkilladj; |
| 151 | else |
| 152 | points >>= -(p->oomkilladj); |
| 153 | } |
| 154 | |
| 155 | #ifdef DEBUG |
| 156 | printk(KERN_DEBUG "OOMkill: task %d (%s) got %d points\n", |
| 157 | p->pid, p->comm, points); |
| 158 | #endif |
| 159 | return points; |
| 160 | } |
| 161 | |
| 162 | /* |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 163 | * Types of limitations to the nodes from which allocations may occur |
| 164 | */ |
| 165 | #define CONSTRAINT_NONE 1 |
| 166 | #define CONSTRAINT_MEMORY_POLICY 2 |
| 167 | #define CONSTRAINT_CPUSET 3 |
| 168 | |
| 169 | /* |
| 170 | * Determine the type of allocation constraint. |
| 171 | */ |
| 172 | static inline int constrained_alloc(struct zonelist *zonelist, gfp_t gfp_mask) |
| 173 | { |
| 174 | #ifdef CONFIG_NUMA |
| 175 | struct zone **z; |
| 176 | nodemask_t nodes = node_online_map; |
| 177 | |
| 178 | for (z = zonelist->zones; *z; z++) |
| 179 | if (cpuset_zone_allowed(*z, gfp_mask)) |
Christoph Lameter | 89fa302 | 2006-09-25 23:31:55 -0700 | [diff] [blame] | 180 | node_clear(zone_to_nid(*z), nodes); |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 181 | else |
| 182 | return CONSTRAINT_CPUSET; |
| 183 | |
| 184 | if (!nodes_empty(nodes)) |
| 185 | return CONSTRAINT_MEMORY_POLICY; |
| 186 | #endif |
| 187 | |
| 188 | return CONSTRAINT_NONE; |
| 189 | } |
| 190 | |
| 191 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | * Simple selection loop. We chose the process with the highest |
| 193 | * number of 'points'. We expect the caller will lock the tasklist. |
| 194 | * |
| 195 | * (not docbooked, we don't want this one cluttering up the manual) |
| 196 | */ |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 197 | static struct task_struct *select_bad_process(unsigned long *ppoints) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 198 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | struct task_struct *g, *p; |
| 200 | struct task_struct *chosen = NULL; |
| 201 | struct timespec uptime; |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 202 | *ppoints = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 203 | |
| 204 | do_posix_clock_monotonic_gettime(&uptime); |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 205 | do_each_thread(g, p) { |
| 206 | unsigned long points; |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 207 | |
Oleg Nesterov | 28324d1 | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 208 | /* |
| 209 | * skip kernel threads and tasks which have already released |
| 210 | * their mm. |
| 211 | */ |
Nick Piggin | 5081dde | 2006-09-25 23:31:32 -0700 | [diff] [blame] | 212 | if (!p->mm) |
| 213 | continue; |
Oleg Nesterov | 28324d1 | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 214 | /* skip the init task */ |
| 215 | if (is_init(p)) |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 216 | continue; |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 217 | |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 218 | /* |
Dave Peterson | 6937a25 | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 219 | * This is in the process of releasing memory so wait for it |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 220 | * to finish before killing some other task by mistake. |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 221 | * |
| 222 | * However, if p is the current task, we allow the 'kill' to |
| 223 | * go ahead if it is exiting: this will simply set TIF_MEMDIE, |
| 224 | * which will allow it to gain access to memory reserves in |
| 225 | * the process of exiting and releasing its resources. |
| 226 | * Otherwise we could get an OOM deadlock. |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 227 | */ |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 228 | if ((p->flags & PF_EXITING) && p == current) { |
| 229 | chosen = p; |
| 230 | *ppoints = ULONG_MAX; |
| 231 | break; |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 232 | } |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 233 | if ((p->flags & PF_EXITING) || |
| 234 | test_tsk_thread_flag(p, TIF_MEMDIE)) |
| 235 | return ERR_PTR(-1UL); |
| 236 | |
Nick Piggin | 4a3ede1 | 2006-09-25 23:31:30 -0700 | [diff] [blame] | 237 | if (p->oomkilladj == OOM_DISABLE) |
| 238 | continue; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 240 | points = badness(p, uptime.tv_sec); |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 241 | if (points > *ppoints || !chosen) { |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 242 | chosen = p; |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 243 | *ppoints = points; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 244 | } |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 245 | } while_each_thread(g, p); |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 246 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 247 | return chosen; |
| 248 | } |
| 249 | |
| 250 | /** |
Ram Gupta | 5a291b98 | 2006-09-25 23:31:54 -0700 | [diff] [blame] | 251 | * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO |
| 252 | * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO |
| 253 | * set. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 254 | */ |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 255 | static void __oom_kill_task(struct task_struct *p, const char *message) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 256 | { |
Sukadev Bhattiprolu | f400e19 | 2006-09-29 02:00:07 -0700 | [diff] [blame] | 257 | if (is_init(p)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 258 | WARN_ON(1); |
| 259 | printk(KERN_WARNING "tried to kill init!\n"); |
| 260 | return; |
| 261 | } |
| 262 | |
Oleg Nesterov | 01017a2 | 2006-09-29 02:01:13 -0700 | [diff] [blame^] | 263 | if (!p->mm) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 264 | WARN_ON(1); |
| 265 | printk(KERN_WARNING "tried to kill an mm-less task!\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 266 | return; |
| 267 | } |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 268 | |
| 269 | if (message) { |
| 270 | printk(KERN_ERR "%s: Killed process %d (%s).\n", |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 271 | message, p->pid, p->comm); |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 272 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 273 | |
| 274 | /* |
| 275 | * We give our sacrificial lamb high priority and access to |
| 276 | * all the memory it needs. That way it should be able to |
| 277 | * exit() and clear out its resources quickly... |
| 278 | */ |
| 279 | p->time_slice = HZ; |
| 280 | set_tsk_thread_flag(p, TIF_MEMDIE); |
| 281 | |
| 282 | force_sig(SIGKILL, p); |
| 283 | } |
| 284 | |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 285 | static int oom_kill_task(struct task_struct *p, const char *message) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 286 | { |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 287 | struct mm_struct *mm; |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 288 | struct task_struct *g, *q; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 289 | |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 290 | mm = p->mm; |
| 291 | |
| 292 | /* WARNING: mm may not be dereferenced since we did not obtain its |
| 293 | * value from get_task_mm(p). This is OK since all we need to do is |
| 294 | * compare mm to q->mm below. |
| 295 | * |
| 296 | * Furthermore, even if mm contains a non-NULL value, p->mm may |
| 297 | * change to NULL at any time since we do not hold task_lock(p). |
| 298 | * However, this is of no concern to us. |
| 299 | */ |
| 300 | |
Oleg Nesterov | 01017a2 | 2006-09-29 02:01:13 -0700 | [diff] [blame^] | 301 | if (mm == NULL) |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 302 | return 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 303 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 304 | __oom_kill_task(p, message); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 305 | /* |
| 306 | * kill all processes that share the ->mm (i.e. all threads), |
| 307 | * but are in a different thread group |
| 308 | */ |
| 309 | do_each_thread(g, q) |
| 310 | if (q->mm == mm && q->tgid != p->tgid) |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 311 | __oom_kill_task(q, message); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 312 | while_each_thread(g, q); |
| 313 | |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 314 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 315 | } |
| 316 | |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 317 | static int oom_kill_process(struct task_struct *p, unsigned long points, |
| 318 | const char *message) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 319 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 320 | struct task_struct *c; |
| 321 | struct list_head *tsk; |
| 322 | |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 323 | /* |
| 324 | * If the task is already exiting, don't alarm the sysadmin or kill |
| 325 | * its children or threads, just set TIF_MEMDIE so it can die quickly |
| 326 | */ |
| 327 | if (p->flags & PF_EXITING) { |
| 328 | __oom_kill_task(p, NULL); |
| 329 | return 0; |
| 330 | } |
| 331 | |
| 332 | printk(KERN_ERR "Out of Memory: Kill process %d (%s) score %li" |
| 333 | " and children.\n", p->pid, p->comm, points); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 334 | /* Try to kill a child first */ |
| 335 | list_for_each(tsk, &p->children) { |
| 336 | c = list_entry(tsk, struct task_struct, sibling); |
| 337 | if (c->mm == p->mm) |
| 338 | continue; |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 339 | if (!oom_kill_task(c, message)) |
| 340 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 341 | } |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 342 | return oom_kill_task(p, message); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 343 | } |
| 344 | |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 345 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
| 346 | |
| 347 | int register_oom_notifier(struct notifier_block *nb) |
| 348 | { |
| 349 | return blocking_notifier_chain_register(&oom_notify_list, nb); |
| 350 | } |
| 351 | EXPORT_SYMBOL_GPL(register_oom_notifier); |
| 352 | |
| 353 | int unregister_oom_notifier(struct notifier_block *nb) |
| 354 | { |
| 355 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); |
| 356 | } |
| 357 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); |
| 358 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 359 | /** |
Dave Peterson | 6937a25 | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 360 | * out_of_memory - kill the "best" process when we run out of memory |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 361 | * |
| 362 | * If we run out of memory, we have the choice between either |
| 363 | * killing a random task (bad), letting the system crash (worse) |
| 364 | * OR try to be smart about which process to kill. Note that we |
| 365 | * don't have to be perfect here, we just have to be good. |
| 366 | */ |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 367 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 368 | { |
Ingo Molnar | 36c8b58 | 2006-07-03 00:25:41 -0700 | [diff] [blame] | 369 | struct task_struct *p; |
Andrew Morton | d6713e0 | 2006-02-28 16:59:19 -0800 | [diff] [blame] | 370 | unsigned long points = 0; |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 371 | unsigned long freed = 0; |
| 372 | |
| 373 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); |
| 374 | if (freed > 0) |
| 375 | /* Got some memory back in the last second. */ |
| 376 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | |
Anton Blanchard | 4263926 | 2005-07-07 17:56:06 -0700 | [diff] [blame] | 378 | if (printk_ratelimit()) { |
Nick Piggin | b72f160 | 2006-09-25 23:31:32 -0700 | [diff] [blame] | 379 | printk(KERN_WARNING "%s invoked oom-killer: " |
| 380 | "gfp_mask=0x%x, order=%d, oomkilladj=%d\n", |
| 381 | current->comm, gfp_mask, order, current->oomkilladj); |
Andrew Morton | b958f7d | 2006-02-01 03:05:51 -0800 | [diff] [blame] | 382 | dump_stack(); |
Anton Blanchard | 4263926 | 2005-07-07 17:56:06 -0700 | [diff] [blame] | 383 | show_mem(); |
| 384 | } |
Janet Morgan | 578c2fd | 2005-06-21 17:14:56 -0700 | [diff] [blame] | 385 | |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 386 | cpuset_lock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 387 | read_lock(&tasklist_lock); |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 388 | |
| 389 | /* |
| 390 | * Check if there were limitations on the allocation (only relevant for |
| 391 | * NUMA) that may require different handling. |
| 392 | */ |
| 393 | switch (constrained_alloc(zonelist, gfp_mask)) { |
| 394 | case CONSTRAINT_MEMORY_POLICY: |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 395 | oom_kill_process(current, points, |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 396 | "No available memory (MPOL_BIND)"); |
| 397 | break; |
| 398 | |
| 399 | case CONSTRAINT_CPUSET: |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 400 | oom_kill_process(current, points, |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 401 | "No available memory in cpuset"); |
| 402 | break; |
| 403 | |
| 404 | case CONSTRAINT_NONE: |
KAMEZAWA Hiroyuki | fadd8fb | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 405 | if (sysctl_panic_on_oom) |
| 406 | panic("out of memory. panic_on_oom is selected\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 407 | retry: |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 408 | /* |
| 409 | * Rambo mode: Shoot down a process and hope it solves whatever |
| 410 | * issues we may have. |
| 411 | */ |
| 412 | p = select_bad_process(&points); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 413 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 414 | if (PTR_ERR(p) == -1UL) |
| 415 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 417 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
| 418 | if (!p) { |
| 419 | read_unlock(&tasklist_lock); |
| 420 | cpuset_unlock(); |
| 421 | panic("Out of memory and no killable processes...\n"); |
| 422 | } |
| 423 | |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 424 | if (oom_kill_process(p, points, "Out of memory")) |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 425 | goto retry; |
| 426 | |
| 427 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 428 | } |
| 429 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 430 | out: |
Andrew Morton | 140ffce | 2006-03-02 02:54:28 -0800 | [diff] [blame] | 431 | read_unlock(&tasklist_lock); |
Paul Jackson | 505970b | 2006-01-14 13:21:06 -0800 | [diff] [blame] | 432 | cpuset_unlock(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | |
| 434 | /* |
| 435 | * Give "p" a good chance of killing itself before we |
Kirill Korotaev | 2f659f4 | 2006-01-08 01:01:05 -0800 | [diff] [blame] | 436 | * retry to allocate memory unless "p" is current |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 437 | */ |
Kirill Korotaev | 2f659f4 | 2006-01-08 01:01:05 -0800 | [diff] [blame] | 438 | if (!test_thread_flag(TIF_MEMDIE)) |
Andrew Morton | 140ffce | 2006-03-02 02:54:28 -0800 | [diff] [blame] | 439 | schedule_timeout_uninterruptible(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 440 | } |