Line data Source code
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 0 : int ioprio_check_cap(int ioprio)
34 : {
35 0 : int class = IOPRIO_PRIO_CLASS(ioprio);
36 0 : int data = IOPRIO_PRIO_DATA(ioprio);
37 :
38 0 : 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 0 : 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 0 : 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 0 : if (data)
59 : return -EINVAL;
60 : break;
61 : default:
62 : return -EINVAL;
63 : }
64 :
65 : return 0;
66 : }
67 :
68 0 : 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 0 : ret = ioprio_check_cap(ioprio);
77 0 : if (ret)
78 0 : return ret;
79 :
80 0 : ret = -ESRCH;
81 : rcu_read_lock();
82 0 : switch (which) {
83 : case IOPRIO_WHO_PROCESS:
84 0 : if (!who)
85 0 : p = current;
86 : else
87 0 : p = find_task_by_vpid(who);
88 0 : if (p)
89 0 : ret = set_task_ioprio(p, ioprio);
90 : break;
91 : case IOPRIO_WHO_PGRP:
92 0 : if (!who)
93 0 : pgrp = task_pgrp(current);
94 : else
95 0 : pgrp = find_vpid(who);
96 :
97 0 : read_lock(&tasklist_lock);
98 0 : do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
99 0 : ret = set_task_ioprio(p, ioprio);
100 0 : if (ret) {
101 0 : read_unlock(&tasklist_lock);
102 0 : goto out;
103 : }
104 0 : } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
105 0 : read_unlock(&tasklist_lock);
106 :
107 0 : break;
108 : case IOPRIO_WHO_USER:
109 0 : uid = make_kuid(current_user_ns(), who);
110 0 : if (!uid_valid(uid))
111 : break;
112 0 : if (!who)
113 0 : user = current_user();
114 : else
115 0 : user = find_user(uid);
116 :
117 0 : if (!user)
118 : break;
119 :
120 0 : for_each_process_thread(g, p) {
121 0 : if (!uid_eq(task_uid(p), uid) ||
122 0 : !task_pid_vnr(p))
123 0 : continue;
124 0 : ret = set_task_ioprio(p, ioprio);
125 0 : if (ret)
126 : goto free_uid;
127 : }
128 : free_uid:
129 0 : if (who)
130 0 : free_uid(user);
131 : break;
132 : default:
133 : ret = -EINVAL;
134 : }
135 :
136 : out:
137 : rcu_read_unlock();
138 0 : return ret;
139 : }
140 :
141 : static int get_task_ioprio(struct task_struct *p)
142 : {
143 : int ret;
144 :
145 0 : ret = security_task_getioprio(p);
146 : if (ret)
147 : goto out;
148 0 : ret = IOPRIO_DEFAULT;
149 0 : task_lock(p);
150 0 : if (p->io_context)
151 0 : ret = p->io_context->ioprio;
152 0 : task_unlock(p);
153 : out:
154 : return ret;
155 : }
156 :
157 0 : int ioprio_best(unsigned short aprio, unsigned short bprio)
158 : {
159 0 : if (!ioprio_valid(aprio))
160 0 : aprio = IOPRIO_DEFAULT;
161 0 : if (!ioprio_valid(bprio))
162 0 : bprio = IOPRIO_DEFAULT;
163 :
164 0 : return min(aprio, bprio);
165 : }
166 :
167 0 : 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 0 : int ret = -ESRCH;
174 : int tmpio;
175 :
176 : rcu_read_lock();
177 0 : switch (which) {
178 : case IOPRIO_WHO_PROCESS:
179 0 : if (!who)
180 0 : p = current;
181 : else
182 0 : p = find_task_by_vpid(who);
183 0 : if (p)
184 0 : ret = get_task_ioprio(p);
185 : break;
186 : case IOPRIO_WHO_PGRP:
187 0 : if (!who)
188 0 : pgrp = task_pgrp(current);
189 : else
190 0 : pgrp = find_vpid(who);
191 0 : read_lock(&tasklist_lock);
192 0 : do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
193 0 : tmpio = get_task_ioprio(p);
194 : if (tmpio < 0)
195 : continue;
196 0 : if (ret == -ESRCH)
197 : ret = tmpio;
198 : else
199 0 : ret = ioprio_best(ret, tmpio);
200 0 : } while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
201 0 : read_unlock(&tasklist_lock);
202 :
203 0 : break;
204 : case IOPRIO_WHO_USER:
205 0 : uid = make_kuid(current_user_ns(), who);
206 0 : if (!who)
207 0 : user = current_user();
208 : else
209 0 : user = find_user(uid);
210 :
211 0 : if (!user)
212 : break;
213 :
214 0 : for_each_process_thread(g, p) {
215 0 : if (!uid_eq(task_uid(p), user->uid) ||
216 0 : !task_pid_vnr(p))
217 0 : continue;
218 0 : tmpio = get_task_ioprio(p);
219 : if (tmpio < 0)
220 : continue;
221 0 : if (ret == -ESRCH)
222 : ret = tmpio;
223 : else
224 0 : ret = ioprio_best(ret, tmpio);
225 : }
226 :
227 0 : if (who)
228 0 : free_uid(user);
229 : break;
230 : default:
231 : ret = -EINVAL;
232 : }
233 :
234 : rcu_read_unlock();
235 0 : return ret;
236 : }
|
