block/ioprio.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * fs/ioprio.c
   4  *
   5  * Copyright (C) 2004 Jens Axboe <axboe@kernel.dk>
   6  *
   7  * Helper functions for setting/querying io priorities of processes. The
   8  * system calls closely mimmick getpriority/setpriority, see the man page for
   9  * those. The prio argument is a composite of prio class and prio data, where
  10  * the data argument has meaning within that class. The standard scheduling
  11  * classes have 8 distinct prio levels, with 0 being the highest prio and 7
  12  * being the lowest.
  13  *
  14  * IOW, setting BE scheduling class with prio 2 is done ala:
  15  *
  16  * unsigned int prio = (IOPRIO_CLASS_BE << IOPRIO_CLASS_SHIFT) | 2;
  17  *
  18  * ioprio_set(PRIO_PROCESS, pid, prio);
  19  *
  20  * See also Documentation/block/ioprio.rst
  21  *
  22  */
  23 #include <linux/gfp.h>
  24 #include <linux/kernel.h>
  25 #include <linux/ioprio.h>
  26 #include <linux/cred.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/capability.h>
  29 #include <linux/syscalls.h>
  30 #include <linux/security.h>
  31 #include <linux/pid_namespace.h>
  32
  33 int ioprio_check_cap(int ioprio)
  34 {
  35         int class = IOPRIO_PRIO_CLASS(ioprio);
  36         int level = IOPRIO_PRIO_LEVEL(ioprio);
  37
  38         switch (class) {
  39                 case IOPRIO_CLASS_RT:
  40                         /*
  41                          * Originally this only checked for CAP_SYS_ADMIN,
  42                          * which was implicitly allowed for pid 0 by security
  43                          * modules such as SELinux. Make sure we check
  44                          * CAP_SYS_ADMIN first to avoid a denial/avc for
  45                          * possibly missing CAP_SYS_NICE permission.
  46                          */
  47                         if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE))
  48                                 return -EPERM;
  49                         fallthrough;
  50                         /* rt has prio field too */
  51                 case IOPRIO_CLASS_BE:
  52                         if (level >= IOPRIO_NR_LEVELS)
  53                                 return -EINVAL;
  54                         break;
  55                 case IOPRIO_CLASS_IDLE:
  56                         break;
  57                 case IOPRIO_CLASS_NONE:
  58                         if (level)
  59                                 return -EINVAL;
  60                         break;
  61                 case IOPRIO_CLASS_INVALID:
  62                 default:
  63                         return -EINVAL;
  64         }
  65
  66         return 0;
  67 }
  68
  69 SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio)
  70 {
  71         struct task_struct *p, *g;
  72         struct user_struct *user;
  73         struct pid *pgrp;
  74         kuid_t uid;
  75         int ret;
  76
  77         ret = ioprio_check_cap(ioprio);
  78         if (ret)
  79                 return ret;
  80
  81         ret = -ESRCH;
  82         rcu_read_lock();
  83         switch (which) {
  84                 case IOPRIO_WHO_PROCESS:
  85                         if (!who)
  86                                 p = current;
  87                         else
  88                                 p = find_task_by_vpid(who);
  89                         if (p)
  90                                 ret = set_task_ioprio(p, ioprio);
  91                         break;
  92                 case IOPRIO_WHO_PGRP:
  93                         if (!who)
  94                                 pgrp = task_pgrp(current);
  95                         else
  96                                 pgrp = find_vpid(who);
  97
  98                         read_lock(&tasklist_lock);
  99                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 100                                 ret = set_task_ioprio(p, ioprio);
 101                                 if (ret) {
 102                                         read_unlock(&tasklist_lock);
 103                                         goto out;
 104                                 }
 105                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 106                         read_unlock(&tasklist_lock);
 107
 108                         break;
 109                 case IOPRIO_WHO_USER:
 110                         uid = make_kuid(current_user_ns(), who);
 111                         if (!uid_valid(uid))
 112                                 break;
 113                         if (!who)
 114                                 user = current_user();
 115                         else
 116                                 user = find_user(uid);
 117
 118                         if (!user)
 119                                 break;
 120
 121                         for_each_process_thread(g, p) {
 122                                 if (!uid_eq(task_uid(p), uid) ||
 123                                     !task_pid_vnr(p))
 124                                         continue;
 125                                 ret = set_task_ioprio(p, ioprio);
 126                                 if (ret)
 127                                         goto free_uid;
 128                         }
 129 free_uid:
 130                         if (who)
 131                                 free_uid(user);
 132                         break;
 133                 default:
 134                         ret = -EINVAL;
 135         }
 136
 137 out:
 138         rcu_read_unlock();
 139         return ret;
 140 }
 141
 142 static int get_task_ioprio(struct task_struct *p)
 143 {
 144         int ret;
 145
 146         ret = security_task_getioprio(p);
 147         if (ret)
 148                 goto out;
 149         task_lock(p);
 150         ret = __get_task_ioprio(p);
 151         task_unlock(p);
 152 out:
 153         return ret;
 154 }
 155
 156 /*
 157  * Return raw IO priority value as set by userspace. We use this for
 158  * ioprio_get(pid, IOPRIO_WHO_PROCESS) so that we keep historical behavior and
 159  * also so that userspace can distinguish unset IO priority (which just gets
 160  * overriden based on task's nice value) from IO priority set to some value.
 161  */
 162 static int get_task_raw_ioprio(struct task_struct *p)
 163 {
 164         int ret;
 165
 166         ret = security_task_getioprio(p);
 167         if (ret)
 168                 goto out;
 169         task_lock(p);
 170         if (p->io_context)
 171                 ret = p->io_context->ioprio;
 172         else
 173                 ret = IOPRIO_DEFAULT;
 174         task_unlock(p);
 175 out:
 176         return ret;
 177 }
 178
 179 static int ioprio_best(unsigned short aprio, unsigned short bprio)
 180 {
 181         return min(aprio, bprio);
 182 }
 183
 184 SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
 185 {
 186         struct task_struct *g, *p;
 187         struct user_struct *user;
 188         struct pid *pgrp;
 189         kuid_t uid;
 190         int ret = -ESRCH;
 191         int tmpio;
 192
 193         rcu_read_lock();
 194         switch (which) {
 195                 case IOPRIO_WHO_PROCESS:
 196                         if (!who)
 197                                 p = current;
 198                         else
 199                                 p = find_task_by_vpid(who);
 200                         if (p)
 201                                 ret = get_task_raw_ioprio(p);
 202                         break;
 203                 case IOPRIO_WHO_PGRP:
 204                         if (!who)
 205                                 pgrp = task_pgrp(current);
 206                         else
 207                                 pgrp = find_vpid(who);
 208                         read_lock(&tasklist_lock);
 209                         do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
 210                                 tmpio = get_task_ioprio(p);
 211                                 if (tmpio < 0)
 212                                         continue;
 213                                 if (ret == -ESRCH)
 214                                         ret = tmpio;
 215                                 else
 216                                         ret = ioprio_best(ret, tmpio);
 217                         } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
 218                         read_unlock(&tasklist_lock);
 219
 220                         break;
 221                 case IOPRIO_WHO_USER:
 222                         uid = make_kuid(current_user_ns(), who);
 223                         if (!who)
 224                                 user = current_user();
 225                         else
 226                                 user = find_user(uid);
 227
 228                         if (!user)
 229                                 break;
 230
 231                         for_each_process_thread(g, p) {
 232                                 if (!uid_eq(task_uid(p), user->uid) ||
 233                                     !task_pid_vnr(p))
 234                                         continue;
 235                                 tmpio = get_task_ioprio(p);
 236                                 if (tmpio < 0)
 237                                         continue;
 238                                 if (ret == -ESRCH)
 239                                         ret = tmpio;
 240                                 else
 241                                         ret = ioprio_best(ret, tmpio);
 242                         }
 243
 244                         if (who)
 245                                 free_uid(user);
 246                         break;
 247                 default:
 248                         ret = -EINVAL;
 249         }
 250
 251         rcu_read_unlock();
 252         return ret;
 253 }