Christoph Hellwig | 3dcf60bc | 2019-04-30 14:42:43 -0400 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 2 | /* |
| 3 | * fs/ioprio.c |
| 4 | * |
Jens Axboe | 0fe2347 | 2006-09-04 15:41:16 +0200 | [diff] [blame] | 5 | * Copyright (C) 2004 Jens Axboe <axboe@kernel.dk> |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 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 | * |
Mauro Carvalho Chehab | 898bd37 | 2019-04-18 19:45:00 -0300 | [diff] [blame] | 20 | * See also Documentation/block/ioprio.rst |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 21 | * |
| 22 | */ |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 23 | #include <linux/gfp.h> |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 24 | #include <linux/kernel.h> |
| 25 | #include <linux/ioprio.h> |
Ingo Molnar | 5b825c3 | 2017-02-02 17:54:15 +0100 | [diff] [blame] | 26 | #include <linux/cred.h> |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 27 | #include <linux/blkdev.h> |
Randy Dunlap | 16f7e0f | 2006-01-11 12:17:46 -0800 | [diff] [blame] | 28 | #include <linux/capability.h> |
Adrian Bunk | 9abdc4c | 2005-11-08 16:57:02 +0100 | [diff] [blame] | 29 | #include <linux/syscalls.h> |
James Morris | 03e6806 | 2006-06-23 02:03:58 -0700 | [diff] [blame] | 30 | #include <linux/security.h> |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 31 | #include <linux/pid_namespace.h> |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 32 | |
Adam Manzanares | aa43457 | 2018-05-22 10:52:17 -0700 | [diff] [blame] | 33 | int ioprio_check_cap(int ioprio) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 34 | { |
| 35 | int class = IOPRIO_PRIO_CLASS(ioprio); |
| 36 | int data = IOPRIO_PRIO_DATA(ioprio); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 37 | |
| 38 | switch (class) { |
| 39 | case IOPRIO_CLASS_RT: |
Alistair Delva | 94c4b4fd | 2021-11-15 18:16:55 +0000 | [diff] [blame] | 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)) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 48 | return -EPERM; |
Gustavo A. R. Silva | df561f66 | 2020-08-23 17:36:59 -0500 | [diff] [blame] | 49 | fallthrough; |
Bart Van Assche | e29387e | 2017-06-21 09:40:11 -0700 | [diff] [blame] | 50 | /* rt has prio field too */ |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 51 | case IOPRIO_CLASS_BE: |
Damien Le Moal | 202bc94 | 2021-08-11 12:37:01 +0900 | [diff] [blame] | 52 | if (data >= IOPRIO_NR_LEVELS || data < 0) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 53 | return -EINVAL; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 54 | break; |
| 55 | case IOPRIO_CLASS_IDLE: |
| 56 | break; |
Jens Axboe | 8ec680e | 2007-11-07 13:54:07 +0100 | [diff] [blame] | 57 | case IOPRIO_CLASS_NONE: |
| 58 | if (data) |
| 59 | return -EINVAL; |
| 60 | break; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 61 | default: |
| 62 | return -EINVAL; |
| 63 | } |
| 64 | |
Adam Manzanares | aa43457 | 2018-05-22 10:52:17 -0700 | [diff] [blame] | 65 | return 0; |
| 66 | } |
| 67 | |
| 68 | SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio) |
| 69 | { |
| 70 | struct task_struct *p, *g; |
| 71 | struct user_struct *user; |
| 72 | struct pid *pgrp; |
| 73 | kuid_t uid; |
| 74 | int ret; |
| 75 | |
| 76 | ret = ioprio_check_cap(ioprio); |
| 77 | if (ret) |
| 78 | return ret; |
| 79 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 80 | ret = -ESRCH; |
Greg Thelen | d69b78b | 2010-11-15 10:20:52 +0100 | [diff] [blame] | 81 | rcu_read_lock(); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 82 | switch (which) { |
| 83 | case IOPRIO_WHO_PROCESS: |
| 84 | if (!who) |
| 85 | p = current; |
| 86 | else |
Pavel Emelyanov | 228ebcb | 2007-10-18 23:40:16 -0700 | [diff] [blame] | 87 | p = find_task_by_vpid(who); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 88 | if (p) |
| 89 | ret = set_task_ioprio(p, ioprio); |
| 90 | break; |
| 91 | case IOPRIO_WHO_PGRP: |
| 92 | if (!who) |
Eric W. Biederman | 41487c6 | 2007-02-12 00:53:01 -0800 | [diff] [blame] | 93 | pgrp = task_pgrp(current); |
| 94 | else |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 95 | pgrp = find_vpid(who); |
Peter Zijlstra | 40c7fd3 | 2021-04-08 11:46:12 +0200 | [diff] [blame] | 96 | |
| 97 | read_lock(&tasklist_lock); |
Ken Chen | 2d70b68 | 2008-08-20 14:09:17 -0700 | [diff] [blame] | 98 | do_each_pid_thread(pgrp, PIDTYPE_PGID, p) { |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 99 | ret = set_task_ioprio(p, ioprio); |
Peter Zijlstra | 40c7fd3 | 2021-04-08 11:46:12 +0200 | [diff] [blame] | 100 | if (ret) { |
| 101 | read_unlock(&tasklist_lock); |
| 102 | goto out; |
| 103 | } |
Ken Chen | 2d70b68 | 2008-08-20 14:09:17 -0700 | [diff] [blame] | 104 | } while_each_pid_thread(pgrp, PIDTYPE_PGID, p); |
Peter Zijlstra | 40c7fd3 | 2021-04-08 11:46:12 +0200 | [diff] [blame] | 105 | read_unlock(&tasklist_lock); |
| 106 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 107 | break; |
| 108 | case IOPRIO_WHO_USER: |
Eric W. Biederman | 7b44ab9 | 2011-11-16 23:20:58 -0800 | [diff] [blame] | 109 | uid = make_kuid(current_user_ns(), who); |
| 110 | if (!uid_valid(uid)) |
| 111 | break; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 112 | if (!who) |
David Howells | 86a264a | 2008-11-14 10:39:18 +1100 | [diff] [blame] | 113 | user = current_user(); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 114 | else |
Eric W. Biederman | 7b44ab9 | 2011-11-16 23:20:58 -0800 | [diff] [blame] | 115 | user = find_user(uid); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 116 | |
| 117 | if (!user) |
| 118 | break; |
| 119 | |
Tetsuo Handa | 612dafa | 2017-02-22 15:40:50 -0800 | [diff] [blame] | 120 | for_each_process_thread(g, p) { |
Ben Segall | 8639b46 | 2015-11-06 16:32:48 -0800 | [diff] [blame] | 121 | if (!uid_eq(task_uid(p), uid) || |
| 122 | !task_pid_vnr(p)) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 123 | continue; |
| 124 | ret = set_task_ioprio(p, ioprio); |
| 125 | if (ret) |
Oleg Nesterov | 78bd4d4 | 2006-08-21 08:33:23 +0200 | [diff] [blame] | 126 | goto free_uid; |
Tetsuo Handa | 612dafa | 2017-02-22 15:40:50 -0800 | [diff] [blame] | 127 | } |
Oleg Nesterov | 78bd4d4 | 2006-08-21 08:33:23 +0200 | [diff] [blame] | 128 | free_uid: |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 129 | if (who) |
| 130 | free_uid(user); |
| 131 | break; |
| 132 | default: |
| 133 | ret = -EINVAL; |
| 134 | } |
| 135 | |
Peter Zijlstra | 40c7fd3 | 2021-04-08 11:46:12 +0200 | [diff] [blame] | 136 | out: |
Greg Thelen | d69b78b | 2010-11-15 10:20:52 +0100 | [diff] [blame] | 137 | rcu_read_unlock(); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 138 | return ret; |
| 139 | } |
| 140 | |
Jan Kara | 893e5d3 | 2022-06-23 09:48:28 +0200 | [diff] [blame] | 141 | /* |
| 142 | * If the task has set an I/O priority, use that. Otherwise, return |
| 143 | * the default I/O priority. |
| 144 | * |
| 145 | * Expected to be called for current task or with task_lock() held to keep |
| 146 | * io_context stable. |
| 147 | */ |
| 148 | int __get_task_ioprio(struct task_struct *p) |
| 149 | { |
| 150 | struct io_context *ioc = p->io_context; |
| 151 | int prio; |
| 152 | |
| 153 | if (p != current) |
| 154 | lockdep_assert_held(&p->alloc_lock); |
| 155 | if (ioc) |
| 156 | prio = ioc->ioprio; |
| 157 | else |
| 158 | prio = IOPRIO_DEFAULT; |
| 159 | |
| 160 | if (IOPRIO_PRIO_CLASS(prio) == IOPRIO_CLASS_NONE) |
| 161 | prio = IOPRIO_PRIO_VALUE(task_nice_ioclass(p), |
| 162 | task_nice_ioprio(p)); |
| 163 | return prio; |
| 164 | } |
| 165 | EXPORT_SYMBOL_GPL(__get_task_ioprio); |
| 166 | |
David Quigley | a1836a4 | 2006-06-30 01:55:49 -0700 | [diff] [blame] | 167 | static int get_task_ioprio(struct task_struct *p) |
| 168 | { |
| 169 | int ret; |
| 170 | |
| 171 | ret = security_task_getioprio(p); |
| 172 | if (ret) |
| 173 | goto out; |
Jan Kara | 4b838d9 | 2022-06-23 09:48:30 +0200 | [diff] [blame] | 174 | task_lock(p); |
| 175 | ret = __get_task_ioprio(p); |
| 176 | task_unlock(p); |
| 177 | out: |
| 178 | return ret; |
| 179 | } |
| 180 | |
| 181 | /* |
| 182 | * Return raw IO priority value as set by userspace. We use this for |
| 183 | * ioprio_get(pid, IOPRIO_WHO_PROCESS) so that we keep historical behavior and |
| 184 | * also so that userspace can distinguish unset IO priority (which just gets |
| 185 | * overriden based on task's nice value) from IO priority set to some value. |
| 186 | */ |
| 187 | static int get_task_raw_ioprio(struct task_struct *p) |
| 188 | { |
| 189 | int ret; |
| 190 | |
| 191 | ret = security_task_getioprio(p); |
| 192 | if (ret) |
| 193 | goto out; |
Omar Sandoval | 8ba8682 | 2016-07-01 00:39:35 -0700 | [diff] [blame] | 194 | task_lock(p); |
Jens Axboe | fd0928d | 2008-01-24 08:52:45 +0100 | [diff] [blame] | 195 | if (p->io_context) |
| 196 | ret = p->io_context->ioprio; |
Jan Kara | 4b838d9 | 2022-06-23 09:48:30 +0200 | [diff] [blame] | 197 | else |
| 198 | ret = IOPRIO_DEFAULT; |
Omar Sandoval | 8ba8682 | 2016-07-01 00:39:35 -0700 | [diff] [blame] | 199 | task_unlock(p); |
David Quigley | a1836a4 | 2006-06-30 01:55:49 -0700 | [diff] [blame] | 200 | out: |
| 201 | return ret; |
| 202 | } |
| 203 | |
Jan Kara | fc25545 | 2022-06-23 09:48:29 +0200 | [diff] [blame] | 204 | static int ioprio_best(unsigned short aprio, unsigned short bprio) |
Oleg Nesterov | e014ff8 | 2006-08-21 10:02:50 +0200 | [diff] [blame] | 205 | { |
Bart Van Assche | 9a87182 | 2017-04-19 14:01:28 -0700 | [diff] [blame] | 206 | return min(aprio, bprio); |
Oleg Nesterov | e014ff8 | 2006-08-21 10:02:50 +0200 | [diff] [blame] | 207 | } |
| 208 | |
Heiko Carstens | 938bb9f | 2009-01-14 14:14:30 +0100 | [diff] [blame] | 209 | SYSCALL_DEFINE2(ioprio_get, int, which, int, who) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 210 | { |
| 211 | struct task_struct *g, *p; |
| 212 | struct user_struct *user; |
Eric W. Biederman | 41487c6 | 2007-02-12 00:53:01 -0800 | [diff] [blame] | 213 | struct pid *pgrp; |
Eric W. Biederman | 7b44ab9 | 2011-11-16 23:20:58 -0800 | [diff] [blame] | 214 | kuid_t uid; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 215 | int ret = -ESRCH; |
David Quigley | a1836a4 | 2006-06-30 01:55:49 -0700 | [diff] [blame] | 216 | int tmpio; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 217 | |
Greg Thelen | d69b78b | 2010-11-15 10:20:52 +0100 | [diff] [blame] | 218 | rcu_read_lock(); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 219 | switch (which) { |
| 220 | case IOPRIO_WHO_PROCESS: |
| 221 | if (!who) |
| 222 | p = current; |
| 223 | else |
Pavel Emelyanov | 228ebcb | 2007-10-18 23:40:16 -0700 | [diff] [blame] | 224 | p = find_task_by_vpid(who); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 225 | if (p) |
Jan Kara | 4b838d9 | 2022-06-23 09:48:30 +0200 | [diff] [blame] | 226 | ret = get_task_raw_ioprio(p); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 227 | break; |
| 228 | case IOPRIO_WHO_PGRP: |
| 229 | if (!who) |
Eric W. Biederman | 41487c6 | 2007-02-12 00:53:01 -0800 | [diff] [blame] | 230 | pgrp = task_pgrp(current); |
| 231 | else |
Pavel Emelyanov | b488893 | 2007-10-18 23:40:14 -0700 | [diff] [blame] | 232 | pgrp = find_vpid(who); |
Davidlohr Bueso | e6a59aa | 2021-12-10 10:20:58 -0800 | [diff] [blame] | 233 | read_lock(&tasklist_lock); |
Ken Chen | 2d70b68 | 2008-08-20 14:09:17 -0700 | [diff] [blame] | 234 | do_each_pid_thread(pgrp, PIDTYPE_PGID, p) { |
David Quigley | a1836a4 | 2006-06-30 01:55:49 -0700 | [diff] [blame] | 235 | tmpio = get_task_ioprio(p); |
| 236 | if (tmpio < 0) |
| 237 | continue; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 238 | if (ret == -ESRCH) |
David Quigley | a1836a4 | 2006-06-30 01:55:49 -0700 | [diff] [blame] | 239 | ret = tmpio; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 240 | else |
David Quigley | a1836a4 | 2006-06-30 01:55:49 -0700 | [diff] [blame] | 241 | ret = ioprio_best(ret, tmpio); |
Ken Chen | 2d70b68 | 2008-08-20 14:09:17 -0700 | [diff] [blame] | 242 | } while_each_pid_thread(pgrp, PIDTYPE_PGID, p); |
Davidlohr Bueso | e6a59aa | 2021-12-10 10:20:58 -0800 | [diff] [blame] | 243 | read_unlock(&tasklist_lock); |
| 244 | |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 245 | break; |
| 246 | case IOPRIO_WHO_USER: |
Eric W. Biederman | 7b44ab9 | 2011-11-16 23:20:58 -0800 | [diff] [blame] | 247 | uid = make_kuid(current_user_ns(), who); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 248 | if (!who) |
David Howells | 86a264a | 2008-11-14 10:39:18 +1100 | [diff] [blame] | 249 | user = current_user(); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 250 | else |
Eric W. Biederman | 7b44ab9 | 2011-11-16 23:20:58 -0800 | [diff] [blame] | 251 | user = find_user(uid); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 252 | |
| 253 | if (!user) |
| 254 | break; |
| 255 | |
Tetsuo Handa | 612dafa | 2017-02-22 15:40:50 -0800 | [diff] [blame] | 256 | for_each_process_thread(g, p) { |
Ben Segall | 8639b46 | 2015-11-06 16:32:48 -0800 | [diff] [blame] | 257 | if (!uid_eq(task_uid(p), user->uid) || |
| 258 | !task_pid_vnr(p)) |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 259 | continue; |
David Quigley | a1836a4 | 2006-06-30 01:55:49 -0700 | [diff] [blame] | 260 | tmpio = get_task_ioprio(p); |
| 261 | if (tmpio < 0) |
| 262 | continue; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 263 | if (ret == -ESRCH) |
David Quigley | a1836a4 | 2006-06-30 01:55:49 -0700 | [diff] [blame] | 264 | ret = tmpio; |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 265 | else |
David Quigley | a1836a4 | 2006-06-30 01:55:49 -0700 | [diff] [blame] | 266 | ret = ioprio_best(ret, tmpio); |
Tetsuo Handa | 612dafa | 2017-02-22 15:40:50 -0800 | [diff] [blame] | 267 | } |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 268 | |
| 269 | if (who) |
| 270 | free_uid(user); |
| 271 | break; |
| 272 | default: |
| 273 | ret = -EINVAL; |
| 274 | } |
| 275 | |
Greg Thelen | d69b78b | 2010-11-15 10:20:52 +0100 | [diff] [blame] | 276 | rcu_read_unlock(); |
Jens Axboe | 22e2c50 | 2005-06-27 10:55:12 +0200 | [diff] [blame] | 277 | return ret; |
| 278 | } |