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
32int 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 (!uid_eq(tcred->uid, cred->euid) &&
41	    !uid_eq(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 = ioprio;
54		put_io_context(ioc);
55	}
56
57	return err;
58}
59EXPORT_SYMBOL_GPL(set_task_ioprio);
60
61SYSCALL_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	kuid_t uid;
69	int ret;
70
71	switch (class) {
72		case IOPRIO_CLASS_RT:
73			if (!capable(CAP_SYS_ADMIN))
74				return -EPERM;
75			/* fall through, rt has prio field too */
76		case IOPRIO_CLASS_BE:
77			if (data >= IOPRIO_BE_NR || data < 0)
78				return -EINVAL;
79
80			break;
81		case IOPRIO_CLASS_IDLE:
82			break;
83		case IOPRIO_CLASS_NONE:
84			if (data)
85				return -EINVAL;
86			break;
87		default:
88			return -EINVAL;
89	}
90
91	ret = -ESRCH;
92	rcu_read_lock();
93	switch (which) {
94		case IOPRIO_WHO_PROCESS:
95			if (!who)
96				p = current;
97			else
98				p = find_task_by_vpid(who);
99			if (p)
100				ret = set_task_ioprio(p, ioprio);
101			break;
102		case IOPRIO_WHO_PGRP:
103			if (!who)
104				pgrp = task_pgrp(current);
105			else
106				pgrp = find_vpid(who);
107			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
108				ret = set_task_ioprio(p, ioprio);
109				if (ret)
110					break;
111			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
112			break;
113		case IOPRIO_WHO_USER:
114			uid = make_kuid(current_user_ns(), who);
115			if (!uid_valid(uid))
116				break;
117			if (!who)
118				user = current_user();
119			else
120				user = find_user(uid);
121
122			if (!user)
123				break;
124
125			do_each_thread(g, p) {
126				if (!uid_eq(task_uid(p), uid))
127					continue;
128				ret = set_task_ioprio(p, ioprio);
129				if (ret)
130					goto free_uid;
131			} while_each_thread(g, p);
132free_uid:
133			if (who)
134				free_uid(user);
135			break;
136		default:
137			ret = -EINVAL;
138	}
139
140	rcu_read_unlock();
141	return ret;
142}
143
144static int get_task_ioprio(struct task_struct *p)
145{
146	int ret;
147
148	ret = security_task_getioprio(p);
149	if (ret)
150		goto out;
151	ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
152	if (p->io_context)
153		ret = p->io_context->ioprio;
154out:
155	return ret;
156}
157
158int ioprio_best(unsigned short aprio, unsigned short bprio)
159{
160	unsigned short aclass;
161	unsigned short bclass;
162
163	if (!ioprio_valid(aprio))
164		aprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM);
165	if (!ioprio_valid(bprio))
166		bprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM);
167
168	aclass = IOPRIO_PRIO_CLASS(aprio);
169	bclass = IOPRIO_PRIO_CLASS(bprio);
170	if (aclass == bclass)
171		return min(aprio, bprio);
172	if (aclass > bclass)
173		return bprio;
174	else
175		return aprio;
176}
177
178SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
179{
180	struct task_struct *g, *p;
181	struct user_struct *user;
182	struct pid *pgrp;
183	kuid_t uid;
184	int ret = -ESRCH;
185	int tmpio;
186
187	rcu_read_lock();
188	switch (which) {
189		case IOPRIO_WHO_PROCESS:
190			if (!who)
191				p = current;
192			else
193				p = find_task_by_vpid(who);
194			if (p)
195				ret = get_task_ioprio(p);
196			break;
197		case IOPRIO_WHO_PGRP:
198			if (!who)
199				pgrp = task_pgrp(current);
200			else
201				pgrp = find_vpid(who);
202			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
203				tmpio = get_task_ioprio(p);
204				if (tmpio < 0)
205					continue;
206				if (ret == -ESRCH)
207					ret = tmpio;
208				else
209					ret = ioprio_best(ret, tmpio);
210			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
211			break;
212		case IOPRIO_WHO_USER:
213			uid = make_kuid(current_user_ns(), who);
214			if (!who)
215				user = current_user();
216			else
217				user = find_user(uid);
218
219			if (!user)
220				break;
221
222			do_each_thread(g, p) {
223				if (!uid_eq(task_uid(p), user->uid))
224					continue;
225				tmpio = get_task_ioprio(p);
226				if (tmpio < 0)
227					continue;
228				if (ret == -ESRCH)
229					ret = tmpio;
230				else
231					ret = ioprio_best(ret, tmpio);
232			} while_each_thread(g, p);
233
234			if (who)
235				free_uid(user);
236			break;
237		default:
238			ret = -EINVAL;
239	}
240
241	rcu_read_unlock();
242	return ret;
243}
244