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
ioprio_check_cap(int ioprio)33 int ioprio_check_cap(int ioprio)
34 {
35 int class = IOPRIO_PRIO_CLASS(ioprio);
36 int data = IOPRIO_PRIO_DATA(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 (data >= IOPRIO_NR_LEVELS || data < 0)
53 return -EINVAL;
54 break;
55 case IOPRIO_CLASS_IDLE:
56 break;
57 case IOPRIO_CLASS_NONE:
58 if (data)
59 return -EINVAL;
60 break;
61 default:
62 return -EINVAL;
63 }
64
65 return 0;
66 }
67
SYSCALL_DEFINE3(ioprio_set,int,which,int,who,int,ioprio)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
80 ret = -ESRCH;
81 rcu_read_lock();
82 switch (which) {
83 case IOPRIO_WHO_PROCESS:
84 if (!who)
85 p = current;
86 else
87 p = find_task_by_vpid(who);
88 if (p)
89 ret = set_task_ioprio(p, ioprio);
90 break;
91 case IOPRIO_WHO_PGRP:
92 if (!who)
93 pgrp = task_pgrp(current);
94 else
95 pgrp = find_vpid(who);
96
97 read_lock(&tasklist_lock);
98 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
99 ret = set_task_ioprio(p, ioprio);
100 if (ret) {
101 read_unlock(&tasklist_lock);
102 goto out;
103 }
104 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
105 read_unlock(&tasklist_lock);
106
107 break;
108 case IOPRIO_WHO_USER:
109 uid = make_kuid(current_user_ns(), who);
110 if (!uid_valid(uid))
111 break;
112 if (!who)
113 user = current_user();
114 else
115 user = find_user(uid);
116
117 if (!user)
118 break;
119
120 for_each_process_thread(g, p) {
121 if (!uid_eq(task_uid(p), uid) ||
122 !task_pid_vnr(p))
123 continue;
124 ret = set_task_ioprio(p, ioprio);
125 if (ret)
126 goto free_uid;
127 }
128 free_uid:
129 if (who)
130 free_uid(user);
131 break;
132 default:
133 ret = -EINVAL;
134 }
135
136 out:
137 rcu_read_unlock();
138 return ret;
139 }
140
get_task_ioprio(struct task_struct * p)141 static int get_task_ioprio(struct task_struct *p)
142 {
143 int ret;
144
145 ret = security_task_getioprio(p);
146 if (ret)
147 goto out;
148 ret = IOPRIO_DEFAULT;
149 task_lock(p);
150 if (p->io_context)
151 ret = p->io_context->ioprio;
152 task_unlock(p);
153 out:
154 return ret;
155 }
156
ioprio_best(unsigned short aprio,unsigned short bprio)157 int ioprio_best(unsigned short aprio, unsigned short bprio)
158 {
159 if (!ioprio_valid(aprio))
160 aprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_BE_NORM);
161 if (!ioprio_valid(bprio))
162 bprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_BE_NORM);
163
164 return min(aprio, bprio);
165 }
166
SYSCALL_DEFINE2(ioprio_get,int,which,int,who)167 SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
168 {
169 struct task_struct *g, *p;
170 struct user_struct *user;
171 struct pid *pgrp;
172 kuid_t uid;
173 int ret = -ESRCH;
174 int tmpio;
175
176 rcu_read_lock();
177 switch (which) {
178 case IOPRIO_WHO_PROCESS:
179 if (!who)
180 p = current;
181 else
182 p = find_task_by_vpid(who);
183 if (p)
184 ret = get_task_ioprio(p);
185 break;
186 case IOPRIO_WHO_PGRP:
187 if (!who)
188 pgrp = task_pgrp(current);
189 else
190 pgrp = find_vpid(who);
191 read_lock(&tasklist_lock);
192 do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
193 tmpio = get_task_ioprio(p);
194 if (tmpio < 0)
195 continue;
196 if (ret == -ESRCH)
197 ret = tmpio;
198 else
199 ret = ioprio_best(ret, tmpio);
200 } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
201 read_unlock(&tasklist_lock);
202
203 break;
204 case IOPRIO_WHO_USER:
205 uid = make_kuid(current_user_ns(), who);
206 if (!who)
207 user = current_user();
208 else
209 user = find_user(uid);
210
211 if (!user)
212 break;
213
214 for_each_process_thread(g, p) {
215 if (!uid_eq(task_uid(p), user->uid) ||
216 !task_pid_vnr(p))
217 continue;
218 tmpio = get_task_ioprio(p);
219 if (tmpio < 0)
220 continue;
221 if (ret == -ESRCH)
222 ret = tmpio;
223 else
224 ret = ioprio_best(ret, tmpio);
225 }
226
227 if (who)
228 free_uid(user);
229 break;
230 default:
231 ret = -EINVAL;
232 }
233
234 rcu_read_unlock();
235 return ret;
236 }
237