• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * fs/ioprio.c
3  *
4  * Copyright (C) 2004 Jens Axboe <axboe@kernel.dk>
5  *
6  * Helper functions for setting/querying io priorities of processes. The
7  * system calls closely mimmick getpriority/setpriority, see the man page for
8  * those. The prio argument is a composite of prio class and prio data, where
9  * the data argument has meaning within that class. The standard scheduling
10  * classes have 8 distinct prio levels, with 0 being the highest prio and 7
11  * being the lowest.
12  *
13  * IOW, setting BE scheduling class with prio 2 is done ala:
14  *
15  * unsigned int prio = (IOPRIO_CLASS_BE << IOPRIO_CLASS_SHIFT) | 2;
16  *
17  * ioprio_set(PRIO_PROCESS, pid, prio);
18  *
19  * See also Documentation/block/ioprio.txt
20  *
21  */
22 #include <linux/gfp.h>
23 #include <linux/kernel.h>
24 #include <linux/export.h>
25 #include <linux/ioprio.h>
26 #include <linux/blkdev.h>
27 #include <linux/capability.h>
28 #include <linux/syscalls.h>
29 #include <linux/security.h>
30 #include <linux/pid_namespace.h>
31 
set_task_ioprio(struct task_struct * task,int ioprio)32 int set_task_ioprio(struct task_struct *task, int ioprio)
33 {
34 	int err;
35 	struct io_context *ioc;
36 	const struct cred *cred = current_cred(), *tcred;
37 
38 	rcu_read_lock();
39 	tcred = __task_cred(task);
40 	if (tcred->uid != cred->euid &&
41 	    tcred->uid != cred->uid && !capable(CAP_SYS_NICE)) {
42 		rcu_read_unlock();
43 		return -EPERM;
44 	}
45 	rcu_read_unlock();
46 
47 	err = security_task_setioprio(task, ioprio);
48 	if (err)
49 		return err;
50 
51 	ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE);
52 	if (ioc) {
53 		ioc_ioprio_changed(ioc, ioprio);
54 		put_io_context(ioc);
55 	}
56 
57 	return err;
58 }
59 EXPORT_SYMBOL_GPL(set_task_ioprio);
60 
SYSCALL_DEFINE3(ioprio_set,int,which,int,who,int,ioprio)61 SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio)
62 {
63 	int class = IOPRIO_PRIO_CLASS(ioprio);
64 	int data = IOPRIO_PRIO_DATA(ioprio);
65 	struct task_struct *p, *g;
66 	struct user_struct *user;
67 	struct pid *pgrp;
68 	int ret;
69 
70 	switch (class) {
71 		case IOPRIO_CLASS_RT:
72 			if (!capable(CAP_SYS_ADMIN))
73 				return -EPERM;
74 			/* fall through, rt has prio field too */
75 		case IOPRIO_CLASS_BE:
76 			if (data >= IOPRIO_BE_NR || data < 0)
77 				return -EINVAL;
78 
79 			break;
80 		case IOPRIO_CLASS_IDLE:
81 			break;
82 		case IOPRIO_CLASS_NONE:
83 			if (data)
84 				return -EINVAL;
85 			break;
86 		default:
87 			return -EINVAL;
88 	}
89 
90 	ret = -ESRCH;
91 	rcu_read_lock();
92 	switch (which) {
93 		case IOPRIO_WHO_PROCESS:
94 			if (!who)
95 				p = current;
96 			else
97 				p = find_task_by_vpid(who);
98 			if (p)
99 				ret = set_task_ioprio(p, ioprio);
100 			break;
101 		case IOPRIO_WHO_PGRP:
102 			if (!who)
103 				pgrp = task_pgrp(current);
104 			else
105 				pgrp = find_vpid(who);
106 			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
107 				ret = set_task_ioprio(p, ioprio);
108 				if (ret)
109 					break;
110 			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
111 			break;
112 		case IOPRIO_WHO_USER:
113 			if (!who)
114 				user = current_user();
115 			else
116 				user = find_user(who);
117 
118 			if (!user)
119 				break;
120 
121 			do_each_thread(g, p) {
122 				if (__task_cred(p)->uid != who)
123 					continue;
124 				ret = set_task_ioprio(p, ioprio);
125 				if (ret)
126 					goto free_uid;
127 			} while_each_thread(g, p);
128 free_uid:
129 			if (who)
130 				free_uid(user);
131 			break;
132 		default:
133 			ret = -EINVAL;
134 	}
135 
136 	rcu_read_unlock();
137 	return ret;
138 }
139 
get_task_ioprio(struct task_struct * p)140 static int get_task_ioprio(struct task_struct *p)
141 {
142 	int ret;
143 
144 	ret = security_task_getioprio(p);
145 	if (ret)
146 		goto out;
147 	ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
148 	if (p->io_context)
149 		ret = p->io_context->ioprio;
150 out:
151 	return ret;
152 }
153 
ioprio_best(unsigned short aprio,unsigned short bprio)154 int ioprio_best(unsigned short aprio, unsigned short bprio)
155 {
156 	unsigned short aclass = IOPRIO_PRIO_CLASS(aprio);
157 	unsigned short bclass = IOPRIO_PRIO_CLASS(bprio);
158 
159 	if (aclass == IOPRIO_CLASS_NONE)
160 		aclass = IOPRIO_CLASS_BE;
161 	if (bclass == IOPRIO_CLASS_NONE)
162 		bclass = IOPRIO_CLASS_BE;
163 
164 	if (aclass == bclass)
165 		return min(aprio, bprio);
166 	if (aclass > bclass)
167 		return bprio;
168 	else
169 		return aprio;
170 }
171 
SYSCALL_DEFINE2(ioprio_get,int,which,int,who)172 SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
173 {
174 	struct task_struct *g, *p;
175 	struct user_struct *user;
176 	struct pid *pgrp;
177 	int ret = -ESRCH;
178 	int tmpio;
179 
180 	rcu_read_lock();
181 	switch (which) {
182 		case IOPRIO_WHO_PROCESS:
183 			if (!who)
184 				p = current;
185 			else
186 				p = find_task_by_vpid(who);
187 			if (p)
188 				ret = get_task_ioprio(p);
189 			break;
190 		case IOPRIO_WHO_PGRP:
191 			if (!who)
192 				pgrp = task_pgrp(current);
193 			else
194 				pgrp = find_vpid(who);
195 			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
196 				tmpio = get_task_ioprio(p);
197 				if (tmpio < 0)
198 					continue;
199 				if (ret == -ESRCH)
200 					ret = tmpio;
201 				else
202 					ret = ioprio_best(ret, tmpio);
203 			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
204 			break;
205 		case IOPRIO_WHO_USER:
206 			if (!who)
207 				user = current_user();
208 			else
209 				user = find_user(who);
210 
211 			if (!user)
212 				break;
213 
214 			do_each_thread(g, p) {
215 				if (__task_cred(p)->uid != user->uid)
216 					continue;
217 				tmpio = get_task_ioprio(p);
218 				if (tmpio < 0)
219 					continue;
220 				if (ret == -ESRCH)
221 					ret = tmpio;
222 				else
223 					ret = ioprio_best(ret, tmpio);
224 			} while_each_thread(g, p);
225 
226 			if (who)
227 				free_uid(user);
228 			break;
229 		default:
230 			ret = -EINVAL;
231 	}
232 
233 	rcu_read_unlock();
234 	return ret;
235 }
236