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
33int 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 break;
50 case IOPRIO_CLASS_BE:
51 case IOPRIO_CLASS_IDLE:
52 break;
53 case IOPRIO_CLASS_NONE:
54 if (level)
55 return -EINVAL;
56 break;
57 case IOPRIO_CLASS_INVALID:
58 default:
59 return -EINVAL;
60 }
61
62 return 0;
63}
64
65SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio)
66{
67 struct task_struct *p, *g;
68 struct user_struct *user;
69 struct pid *pgrp;
70 kuid_t uid;
71 int ret;
72
73 ret = ioprio_check_cap(ioprio);
74 if (ret)
75 return ret;
76
77 ret = -ESRCH;
78 rcu_read_lock();
79 switch (which) {
80 case IOPRIO_WHO_PROCESS:
81 if (!who)
82 p = current;
83 else
84 p = find_task_by_vpid(nr: who);
85 if (p)
86 ret = set_task_ioprio(task: p, ioprio);
87 break;
88 case IOPRIO_WHO_PGRP:
89 if (!who)
90 pgrp = task_pgrp(current);
91 else
92 pgrp = find_vpid(nr: who);
93
94 read_lock(&tasklist_lock);
95 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
96 ret = set_task_ioprio(task: p, ioprio);
97 if (ret) {
98 read_unlock(&tasklist_lock);
99 goto out;
100 }
101 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
102 read_unlock(&tasklist_lock);
103
104 break;
105 case IOPRIO_WHO_USER:
106 uid = make_kuid(from: current_user_ns(), uid: who);
107 if (!uid_valid(uid))
108 break;
109 if (!who)
110 user = current_user();
111 else
112 user = find_user(uid);
113
114 if (!user)
115 break;
116
117 for_each_process_thread(g, p) {
118 if (!uid_eq(task_uid(p), right: uid) ||
119 !task_pid_vnr(tsk: p))
120 continue;
121 ret = set_task_ioprio(task: p, ioprio);
122 if (ret)
123 goto free_uid;
124 }
125free_uid:
126 if (who)
127 free_uid(user);
128 break;
129 default:
130 ret = -EINVAL;
131 }
132
133out:
134 rcu_read_unlock();
135 return ret;
136}
137
138static int get_task_ioprio(struct task_struct *p)
139{
140 int ret;
141
142 ret = security_task_getioprio(p);
143 if (ret)
144 goto out;
145 task_lock(p);
146 ret = __get_task_ioprio(p);
147 task_unlock(p);
148out:
149 return ret;
150}
151
152/*
153 * Return raw IO priority value as set by userspace. We use this for
154 * ioprio_get(pid, IOPRIO_WHO_PROCESS) so that we keep historical behavior and
155 * also so that userspace can distinguish unset IO priority (which just gets
156 * overriden based on task's nice value) from IO priority set to some value.
157 */
158static int get_task_raw_ioprio(struct task_struct *p)
159{
160 int ret;
161
162 ret = security_task_getioprio(p);
163 if (ret)
164 goto out;
165 task_lock(p);
166 if (p->io_context)
167 ret = p->io_context->ioprio;
168 else
169 ret = IOPRIO_DEFAULT;
170 task_unlock(p);
171out:
172 return ret;
173}
174
175static int ioprio_best(unsigned short aprio, unsigned short bprio)
176{
177 return min(aprio, bprio);
178}
179
180SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
181{
182 struct task_struct *g, *p;
183 struct user_struct *user;
184 struct pid *pgrp;
185 kuid_t uid;
186 int ret = -ESRCH;
187 int tmpio;
188
189 rcu_read_lock();
190 switch (which) {
191 case IOPRIO_WHO_PROCESS:
192 if (!who)
193 p = current;
194 else
195 p = find_task_by_vpid(nr: who);
196 if (p)
197 ret = get_task_raw_ioprio(p);
198 break;
199 case IOPRIO_WHO_PGRP:
200 if (!who)
201 pgrp = task_pgrp(current);
202 else
203 pgrp = find_vpid(nr: who);
204 read_lock(&tasklist_lock);
205 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
206 tmpio = get_task_ioprio(p);
207 if (tmpio < 0)
208 continue;
209 if (ret == -ESRCH)
210 ret = tmpio;
211 else
212 ret = ioprio_best(aprio: ret, bprio: tmpio);
213 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
214 read_unlock(&tasklist_lock);
215
216 break;
217 case IOPRIO_WHO_USER:
218 uid = make_kuid(from: current_user_ns(), uid: who);
219 if (!who)
220 user = current_user();
221 else
222 user = find_user(uid);
223
224 if (!user)
225 break;
226
227 for_each_process_thread(g, p) {
228 if (!uid_eq(task_uid(p), right: user->uid) ||
229 !task_pid_vnr(tsk: p))
230 continue;
231 tmpio = get_task_ioprio(p);
232 if (tmpio < 0)
233 continue;
234 if (ret == -ESRCH)
235 ret = tmpio;
236 else
237 ret = ioprio_best(aprio: ret, bprio: tmpio);
238 }
239
240 if (who)
241 free_uid(user);
242 break;
243 default:
244 ret = -EINVAL;
245 }
246
247 rcu_read_unlock();
248 return ret;
249}
250