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 
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  */
__get_task_ioprio(struct task_struct * p)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 
get_task_ioprio(struct task_struct * p)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;
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  */
get_task_raw_ioprio(struct task_struct * p)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;
194 	task_lock(p);
195 	if (p->io_context)
196 		ret = p->io_context->ioprio;
197 	else
198 		ret = IOPRIO_DEFAULT;
199 	task_unlock(p);
200 out:
201 	return ret;
202 }
203 
ioprio_best(unsigned short aprio,unsigned short bprio)204 static int ioprio_best(unsigned short aprio, unsigned short bprio)
205 {
206 	return min(aprio, bprio);
207 }
208 
SYSCALL_DEFINE2(ioprio_get,int,which,int,who)209 SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
210 {
211 	struct task_struct *g, *p;
212 	struct user_struct *user;
213 	struct pid *pgrp;
214 	kuid_t uid;
215 	int ret = -ESRCH;
216 	int tmpio;
217 
218 	rcu_read_lock();
219 	switch (which) {
220 		case IOPRIO_WHO_PROCESS:
221 			if (!who)
222 				p = current;
223 			else
224 				p = find_task_by_vpid(who);
225 			if (p)
226 				ret = get_task_raw_ioprio(p);
227 			break;
228 		case IOPRIO_WHO_PGRP:
229 			if (!who)
230 				pgrp = task_pgrp(current);
231 			else
232 				pgrp = find_vpid(who);
233 			read_lock(&tasklist_lock);
234 			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
235 				tmpio = get_task_ioprio(p);
236 				if (tmpio < 0)
237 					continue;
238 				if (ret == -ESRCH)
239 					ret = tmpio;
240 				else
241 					ret = ioprio_best(ret, tmpio);
242 			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
243 			read_unlock(&tasklist_lock);
244 
245 			break;
246 		case IOPRIO_WHO_USER:
247 			uid = make_kuid(current_user_ns(), who);
248 			if (!who)
249 				user = current_user();
250 			else
251 				user = find_user(uid);
252 
253 			if (!user)
254 				break;
255 
256 			for_each_process_thread(g, p) {
257 				if (!uid_eq(task_uid(p), user->uid) ||
258 				    !task_pid_vnr(p))
259 					continue;
260 				tmpio = get_task_ioprio(p);
261 				if (tmpio < 0)
262 					continue;
263 				if (ret == -ESRCH)
264 					ret = tmpio;
265 				else
266 					ret = ioprio_best(ret, tmpio);
267 			}
268 
269 			if (who)
270 				free_uid(user);
271 			break;
272 		default:
273 			ret = -EINVAL;
274 	}
275 
276 	rcu_read_unlock();
277 	return ret;
278 }
279