xref: /openbmc/linux/block/ioprio.c (revision cf21eb6a)
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 /*
143  * If the task has set an I/O priority, use that. Otherwise, return
144  * the default I/O priority.
145  *
146  * Expected to be called for current task or with task_lock() held to keep
147  * io_context stable.
148  */
149 int __get_task_ioprio(struct task_struct *p)
150 {
151 	struct io_context *ioc = p->io_context;
152 	int prio;
153 
154 	if (p != current)
155 		lockdep_assert_held(&p->alloc_lock);
156 	if (ioc)
157 		prio = ioc->ioprio;
158 	else
159 		prio = IOPRIO_DEFAULT;
160 
161 	if (IOPRIO_PRIO_CLASS(prio) == IOPRIO_CLASS_NONE)
162 		prio = IOPRIO_PRIO_VALUE(task_nice_ioclass(p),
163 					 task_nice_ioprio(p));
164 	return prio;
165 }
166 EXPORT_SYMBOL_GPL(__get_task_ioprio);
167 
168 static int get_task_ioprio(struct task_struct *p)
169 {
170 	int ret;
171 
172 	ret = security_task_getioprio(p);
173 	if (ret)
174 		goto out;
175 	task_lock(p);
176 	ret = __get_task_ioprio(p);
177 	task_unlock(p);
178 out:
179 	return ret;
180 }
181 
182 /*
183  * Return raw IO priority value as set by userspace. We use this for
184  * ioprio_get(pid, IOPRIO_WHO_PROCESS) so that we keep historical behavior and
185  * also so that userspace can distinguish unset IO priority (which just gets
186  * overriden based on task's nice value) from IO priority set to some value.
187  */
188 static int get_task_raw_ioprio(struct task_struct *p)
189 {
190 	int ret;
191 
192 	ret = security_task_getioprio(p);
193 	if (ret)
194 		goto out;
195 	task_lock(p);
196 	if (p->io_context)
197 		ret = p->io_context->ioprio;
198 	else
199 		ret = IOPRIO_DEFAULT;
200 	task_unlock(p);
201 out:
202 	return ret;
203 }
204 
205 static int ioprio_best(unsigned short aprio, unsigned short bprio)
206 {
207 	return min(aprio, bprio);
208 }
209 
210 SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
211 {
212 	struct task_struct *g, *p;
213 	struct user_struct *user;
214 	struct pid *pgrp;
215 	kuid_t uid;
216 	int ret = -ESRCH;
217 	int tmpio;
218 
219 	rcu_read_lock();
220 	switch (which) {
221 		case IOPRIO_WHO_PROCESS:
222 			if (!who)
223 				p = current;
224 			else
225 				p = find_task_by_vpid(who);
226 			if (p)
227 				ret = get_task_raw_ioprio(p);
228 			break;
229 		case IOPRIO_WHO_PGRP:
230 			if (!who)
231 				pgrp = task_pgrp(current);
232 			else
233 				pgrp = find_vpid(who);
234 			read_lock(&tasklist_lock);
235 			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
236 				tmpio = get_task_ioprio(p);
237 				if (tmpio < 0)
238 					continue;
239 				if (ret == -ESRCH)
240 					ret = tmpio;
241 				else
242 					ret = ioprio_best(ret, tmpio);
243 			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
244 			read_unlock(&tasklist_lock);
245 
246 			break;
247 		case IOPRIO_WHO_USER:
248 			uid = make_kuid(current_user_ns(), who);
249 			if (!who)
250 				user = current_user();
251 			else
252 				user = find_user(uid);
253 
254 			if (!user)
255 				break;
256 
257 			for_each_process_thread(g, p) {
258 				if (!uid_eq(task_uid(p), user->uid) ||
259 				    !task_pid_vnr(p))
260 					continue;
261 				tmpio = get_task_ioprio(p);
262 				if (tmpio < 0)
263 					continue;
264 				if (ret == -ESRCH)
265 					ret = tmpio;
266 				else
267 					ret = ioprio_best(ret, tmpio);
268 			}
269 
270 			if (who)
271 				free_uid(user);
272 			break;
273 		default:
274 			ret = -EINVAL;
275 	}
276 
277 	rcu_read_unlock();
278 	return ret;
279 }
280