mm/oom_kill.c

   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
   9  *  we're seriously out of memory. This gets called from kswapd()
  10  *  in linux/mm/vmscan.c when we really run out of memory.
  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/swapctl.h>
  22 #include <linux/timex.h>
  23
  24 /* #define DEBUG */
  25
  26 /**
  27  * int_sqrt - oom_kill.c internal function, rough approximation to sqrt
  28  * @x: integer of which to calculate the sqrt
  29  *
  30  * A very rough approximation to the sqrt() function.
  31  */
  32 static unsigned int int_sqrt(unsigned int x)
  33 {
  34         unsigned int out = x;
  35         while (x & ~(unsigned int)1) x >>=2, out >>=1;
  36         if (x) out -= out >> 2;
  37         return (out ? out : 1);
  38 }
  39
  40 /**
  41  * oom_badness - calculate a numeric value for how bad this task has been
  42  * @p: task struct of which task we should calculate
  43  *
  44  * The formula used is relatively simple and documented inline in the
  45  * function. The main rationale is that we want to select a good task
  46  * to kill when we run out of memory.
  47  *
  48  * Good in this context means that:
  49  * 1) we lose the minimum amount of work done
  50  * 2) we recover a large amount of memory
  51  * 3) we don't kill anything innocent of eating tons of memory
  52  * 4) we want to kill the minimum amount of processes (one)
  53  * 5) we try to kill the process the user expects us to kill, this
  54  *    algorithm has been meticulously tuned to meet the priniciple
  55  *    of least surprise ... (be careful when you change it)
  56  */
  57
  58 static int badness(struct task_struct *p)
  59 {
  60         int points, cpu_time, run_time;
  61
  62         if (!p->mm)
  63                 return 0;
  64         /*
  65          * The memory size of the process is the basis for the badness.
  66          */
  67         points = p->mm->total_vm;
  68
  69         /*
  70          * CPU time is in seconds and run time is in minutes. There is no
  71          * particular reason for this other than that it turned out to work
  72          * very well in practice. This is not safe against jiffie wraps
  73          * but we don't care _that_ much...
  74          */
  75         cpu_time = (p->times.tms_utime + p->times.tms_stime) >> (SHIFT_HZ + 3);
  76         run_time = (jiffies - p->start_time) >> (SHIFT_HZ + 10);
  77
  78         points /= int_sqrt(cpu_time);
  79         points /= int_sqrt(int_sqrt(run_time));
  80
  81         /*
  82          * Niced processes are most likely less important, so double
  83          * their badness points.
  84          */
  85         if (p->nice > 0)
  86                 points *= 2;
  87
  88         /*
  89          * Superuser processes are usually more important, so we make it
  90          * less likely that we kill those.
  91          */
  92         if (cap_t(p->cap_effective) & CAP_TO_MASK(CAP_SYS_ADMIN) ||
  93                                 p->uid == 0 || p->euid == 0)
  94                 points /= 4;
  95
  96         /*
  97          * We don't want to kill a process with direct hardware access.
  98          * Not only could that mess up the hardware, but usually users
  99          * tend to only have this flag set on applications they think
 100          * of as important.
 101          */
 102         if (cap_t(p->cap_effective) & CAP_TO_MASK(CAP_SYS_RAWIO))
 103                 points /= 4;
 104 #ifdef DEBUG
 105         printk(KERN_DEBUG "OOMkill: task %d (%s) got %d points\n",
 106         p->pid, p->comm, points);
 107 #endif
 108         return points;
 109 }
 110
 111 /*
 112  * Simple selection loop. We chose the process with the highest
 113  * number of 'points'. We need the locks to make sure that the
 114  * list of task structs doesn't change while we look the other way.
 115  *
 116  * (not docbooked, we don't want this one cluttering up the manual)
 117  */
 118 static struct task_struct * select_bad_process(void)
 119 {
 120         int maxpoints = 0;
 121         struct task_struct *p = NULL;
 122         struct task_struct *chosen = NULL;
 123
 124         read_lock(&tasklist_lock);
 125         for_each_task(p) {
 126                 if (p->pid) {
 127                         int points = badness(p);
 128                         if (points > maxpoints) {
 129                                 chosen = p;
 130                                 maxpoints = points;
 131                         }
 132                 }
 133         }
 134         read_unlock(&tasklist_lock);
 135         return chosen;
 136 }
 137
 138 /**
 139  * oom_kill - kill the "best" process when we run out of memory
 140  *
 141  * If we run out of memory, we have the choice between either
 142  * killing a random task (bad), letting the system crash (worse)
 143  * OR try to be smart about which process to kill. Note that we
 144  * don't have to be perfect here, we just have to be good.
 145  *
 146  * We must be careful though to never send SIGKILL a process with
 147  * CAP_SYS_RAW_IO set, send SIGTERM instead (but it's unlikely that
 148  * we select a process with CAP_SYS_RAW_IO set).
 149  */
 150 void oom_kill(void)
 151 {
 152
 153         struct task_struct *p = select_bad_process();
 154
 155         /* Found nothing?!?! Either we hang forever, or we panic. */
 156         if (p == NULL)
 157                 panic("Out of memory and no killable processes...\n");
 158
 159         printk(KERN_ERR "Out of Memory: Killed process %d (%s).\n", p->pid, p->comm);
 160
 161         /*
 162          * We give our sacrificial lamb high priority and access to
 163          * all the memory it needs. That way it should be able to
 164          * exit() and clear out its resources quickly...
 165          */
 166         p->counter = 5 * HZ;
 167         p->flags |= PF_MEMALLOC;
 168
 169         /* This process has hardware access, be more careful. */
 170         if (cap_t(p->cap_effective) & CAP_TO_MASK(CAP_SYS_RAWIO)) {
 171                 force_sig(SIGTERM, p);
 172         } else {
 173                 force_sig(SIGKILL, p);
 174         }
 175
 176         /*
 177          * Make kswapd go out of the way, so "p" has a good chance of
 178          * killing itself before someone else gets the chance to ask
 179          * for more memory.
 180          */
 181         current->policy |= SCHED_YIELD;
 182         schedule();
 183         return;
 184 }
 185
 186 /**
 187  * out_of_memory - is the system out of memory?
 188  *
 189  * Returns 0 if there is still enough memory left,
 190  * 1 when we are out of memory (otherwise).
 191  */
 192 int out_of_memory(void)
 193 {
 194         struct sysinfo swp_info;
 195
 196         /* Enough free memory?  Not OOM. */
 197         if (nr_free_pages() > freepages.min)
 198                 return 0;
 199
 200         if (nr_free_pages() + nr_inactive_clean_pages() > freepages.low)
 201                 return 0;
 202
 203         /* Enough swap space left?  Not OOM. */
 204         si_swapinfo(&swp_info);
 205         if (swp_info.freeswap > 0)
 206                 return 0;
 207
 208         /* Else... */
 209         return 1;
 210 }