fs/ioprio.c

   1 /*
   2  * fs/ioprio.c
   3  *
   4  * Copyright (C) 2004 Jens Axboe <axboe@kernel.dk>
   5  *
   6  * Helper functions for setting/querying io priorities of processes. The
   7  * system calls closely mimmick getpriority/setpriority, see the man page for
   8  * those. The prio argument is a composite of prio class and prio data, where
   9  * the data argument has meaning within that class. The standard scheduling
  10  * classes have 8 distinct prio levels, with 0 being the highest prio and 7
  11  * being the lowest.
  12  *
  13  * IOW, setting BE scheduling class with prio 2 is done ala:
  14  *
  15  * unsigned int prio = (IOPRIO_CLASS_BE << IOPRIO_CLASS_SHIFT) | 2;
  16  *
  17  * ioprio_set(PRIO_PROCESS, pid, prio);
  18  *
  19  * See also Documentation/block/ioprio.txt
  20  *
  21  */
  22 #include <linux/kernel.h>
  23 #include <linux/ioprio.h>
  24 #include <linux/blkdev.h>
  25 #include <linux/capability.h>
  26 #include <linux/syscalls.h>
  27 #include <linux/security.h>
  28 #include <linux/pid_namespace.h>
  29
  30 static int set_task_ioprio(struct task_struct *task, int ioprio)
  31 {
  32         int err;
  33         struct io_context *ioc;
  34
  35         if (task->uid != current->euid &&
  36             task->uid != current->uid && !capable(CAP_SYS_NICE))
  37                 return -EPERM;
  38
  39         err = security_task_setioprio(task, ioprio);
  40         if (err)
  41                 return err;
  42
  43         task_lock(task);
  44
  45         task->ioprio = ioprio;
  46
  47         ioc = task->io_context;
  48         /* see wmb() in current_io_context() */
  49         smp_read_barrier_depends();
  50
  51         if (ioc)
  52                 ioc->ioprio_changed = 1;
  53
  54         task_unlock(task);
  55         return 0;
  56 }
  57
  58 asmlinkage long sys_ioprio_set(int which, int who, int ioprio)
  59 {
  60         int class = IOPRIO_PRIO_CLASS(ioprio);
  61         int data = IOPRIO_PRIO_DATA(ioprio);
  62         struct task_struct *p, *g;
  63         struct user_struct *user;
  64         struct pid *pgrp;
  65         int ret;
  66
  67         switch (class) {
  68                 case IOPRIO_CLASS_RT:
  69                         if (!capable(CAP_SYS_ADMIN))
  70                                 return -EPERM;
  71                         /* fall through, rt has prio field too */
  72                 case IOPRIO_CLASS_BE:
  73                         if (data >= IOPRIO_BE_NR || data < 0)
  74                                 return -EINVAL;
  75
  76                         break;
  77                 case IOPRIO_CLASS_IDLE:
  78                         if (!capable(CAP_SYS_ADMIN))
  79                                 return -EPERM;
  80                         break;
  81                 default:
  82                         return -EINVAL;
  83         }
  84
  85         ret = -ESRCH;
  86         /*
  87          * We want IOPRIO_WHO_PGRP/IOPRIO_WHO_USER to be "atomic",
  88          * so we can't use rcu_read_lock(). See re-copy of ->ioprio
  89          * in copy_process().
  90          */
  91         read_lock(&tasklist_lock);
  92         switch (which) {
  93                 case IOPRIO_WHO_PROCESS:
  94                         if (!who)
  95                                 p = current;
  96                         else
  97                                 p = find_task_by_vpid(who);
  98                         if (p)
  99                                 ret = set_task_ioprio(p, ioprio);
 100                         break;
 101                 case IOPRIO_WHO_PGRP:
 102                         if (!who)
 103                                 pgrp = task_pgrp(current);
 104                         else
 105                                 pgrp = find_vpid(who);
 106                         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 107                                 ret = set_task_ioprio(p, ioprio);
 108                                 if (ret)
 109                                         break;
 110                         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 111                         break;
 112                 case IOPRIO_WHO_USER:
 113                         if (!who)
 114                                 user = current->user;
 115                         else
 116                                 user = find_user(who);
 117
 118                         if (!user)
 119                                 break;
 120
 121                         do_each_thread(g, p) {
 122                                 if (p->uid != who)
 123                                         continue;
 124                                 ret = set_task_ioprio(p, ioprio);
 125                                 if (ret)
 126                                         goto free_uid;
 127                         } while_each_thread(g, p);
 128 free_uid:
 129                         if (who)
 130                                 free_uid(user);
 131                         break;
 132                 default:
 133                         ret = -EINVAL;
 134         }
 135
 136         read_unlock(&tasklist_lock);
 137         return ret;
 138 }
 139
 140 static int get_task_ioprio(struct task_struct *p)
 141 {
 142         int ret;
 143
 144         ret = security_task_getioprio(p);
 145         if (ret)
 146                 goto out;
 147         ret = p->ioprio;
 148 out:
 149         return ret;
 150 }
 151
 152 int ioprio_best(unsigned short aprio, unsigned short bprio)
 153 {
 154         unsigned short aclass = IOPRIO_PRIO_CLASS(aprio);
 155         unsigned short bclass = IOPRIO_PRIO_CLASS(bprio);
 156
 157         if (aclass == IOPRIO_CLASS_NONE)
 158                 aclass = IOPRIO_CLASS_BE;
 159         if (bclass == IOPRIO_CLASS_NONE)
 160                 bclass = IOPRIO_CLASS_BE;
 161
 162         if (aclass == bclass)
 163                 return min(aprio, bprio);
 164         if (aclass > bclass)
 165                 return bprio;
 166         else
 167                 return aprio;
 168 }
 169
 170 asmlinkage long sys_ioprio_get(int which, int who)
 171 {
 172         struct task_struct *g, *p;
 173         struct user_struct *user;
 174         struct pid *pgrp;
 175         int ret = -ESRCH;
 176         int tmpio;
 177
 178         read_lock(&tasklist_lock);
 179         switch (which) {
 180                 case IOPRIO_WHO_PROCESS:
 181                         if (!who)
 182                                 p = current;
 183                         else
 184                                 p = find_task_by_vpid(who);
 185                         if (p)
 186                                 ret = get_task_ioprio(p);
 187                         break;
 188                 case IOPRIO_WHO_PGRP:
 189                         if (!who)
 190                                 pgrp = task_pgrp(current);
 191                         else
 192                                 pgrp = find_vpid(who);
 193                         do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
 194                                 tmpio = get_task_ioprio(p);
 195                                 if (tmpio < 0)
 196                                         continue;
 197                                 if (ret == -ESRCH)
 198                                         ret = tmpio;
 199                                 else
 200                                         ret = ioprio_best(ret, tmpio);
 201                         } while_each_pid_task(pgrp, PIDTYPE_PGID, p);
 202                         break;
 203                 case IOPRIO_WHO_USER:
 204                         if (!who)
 205                                 user = current->user;
 206                         else
 207                                 user = find_user(who);
 208
 209                         if (!user)
 210                                 break;
 211
 212                         do_each_thread(g, p) {
 213                                 if (p->uid != user->uid)
 214                                         continue;
 215                                 tmpio = get_task_ioprio(p);
 216                                 if (tmpio < 0)
 217                                         continue;
 218                                 if (ret == -ESRCH)
 219                                         ret = tmpio;
 220                                 else
 221                                         ret = ioprio_best(ret, tmpio);
 222                         } while_each_thread(g, p);
 223
 224                         if (who)
 225                                 free_uid(user);
 226                         break;
 227                 default:
 228                         ret = -EINVAL;
 229         }
 230
 231         read_unlock(&tasklist_lock);
 232         return ret;
 233 }
 234