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/ioprio.h>
25#include <linux/blkdev.h>
26#include <linux/capability.h>
27#include <linux/syscalls.h>
28#include <linux/security.h>
29#include <linux/pid_namespace.h>
30
31int set_task_ioprio(struct task_struct *task, int ioprio)
32{
33	int err;
34	struct io_context *ioc;
35	const struct cred *cred = current_cred(), *tcred;
36
37	rcu_read_lock();
38	tcred = __task_cred(task);
39	if (tcred->uid != cred->euid &&
40	    tcred->uid != cred->uid && !capable(CAP_SYS_NICE)) {
41		rcu_read_unlock();
42		return -EPERM;
43	}
44	rcu_read_unlock();
45
46	err = security_task_setioprio(task, ioprio);
47	if (err)
48		return err;
49
50	task_lock(task);
51	do {
52		ioc = task->io_context;
53		/* see wmb() in current_io_context() */
54		smp_read_barrier_depends();
55		if (ioc)
56			break;
57
58		ioc = alloc_io_context(GFP_ATOMIC, -1);
59		if (!ioc) {
60			err = -ENOMEM;
61			break;
62		}
63		task->io_context = ioc;
64	} while (1);
65
66	if (!err) {
67		ioc->ioprio = ioprio;
68		ioc->ioprio_changed = 1;
69	}
70
71	task_unlock(task);
72	return err;
73}
74EXPORT_SYMBOL_GPL(set_task_ioprio);
75
76SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio)
77{
78	int class = IOPRIO_PRIO_CLASS(ioprio);
79	int data = IOPRIO_PRIO_DATA(ioprio);
80	struct task_struct *p, *g;
81	struct user_struct *user;
82	struct pid *pgrp;
83	int ret;
84
85	switch (class) {
86		case IOPRIO_CLASS_RT:
87			if (!capable(CAP_SYS_ADMIN))
88				return -EPERM;
89			/* fall through, rt has prio field too */
90		case IOPRIO_CLASS_BE:
91			if (data >= IOPRIO_BE_NR || data < 0)
92				return -EINVAL;
93
94			break;
95		case IOPRIO_CLASS_IDLE:
96			break;
97		case IOPRIO_CLASS_NONE:
98			if (data)
99				return -EINVAL;
100			break;
101		default:
102			return -EINVAL;
103	}
104
105	ret = -ESRCH;
106	/*
107	 * We want IOPRIO_WHO_PGRP/IOPRIO_WHO_USER to be "atomic",
108	 * so we can't use rcu_read_lock(). See re-copy of ->ioprio
109	 * in copy_process().
110	 */
111	read_lock(&tasklist_lock);
112	switch (which) {
113		case IOPRIO_WHO_PROCESS:
114			if (!who)
115				p = current;
116			else
117				p = find_task_by_vpid(who);
118			if (p)
119				ret = set_task_ioprio(p, ioprio);
120			break;
121		case IOPRIO_WHO_PGRP:
122			if (!who)
123				pgrp = task_pgrp(current);
124			else
125				pgrp = find_vpid(who);
126			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
127				ret = set_task_ioprio(p, ioprio);
128				if (ret)
129					break;
130			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
131			break;
132		case IOPRIO_WHO_USER:
133			if (!who)
134				user = current_user();
135			else
136				user = find_user(who);
137
138			if (!user)
139				break;
140
141			do_each_thread(g, p) {
142				if (__task_cred(p)->uid != who)
143					continue;
144				ret = set_task_ioprio(p, ioprio);
145				if (ret)
146					goto free_uid;
147			} while_each_thread(g, p);
148free_uid:
149			if (who)
150				free_uid(user);
151			break;
152		default:
153			ret = -EINVAL;
154	}
155
156	read_unlock(&tasklist_lock);
157	return ret;
158}
159
160static int get_task_ioprio(struct task_struct *p)
161{
162	int ret;
163
164	ret = security_task_getioprio(p);
165	if (ret)
166		goto out;
167	ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
168	if (p->io_context)
169		ret = p->io_context->ioprio;
170out:
171	return ret;
172}
173
174int ioprio_best(unsigned short aprio, unsigned short bprio)
175{
176	unsigned short aclass = IOPRIO_PRIO_CLASS(aprio);
177	unsigned short bclass = IOPRIO_PRIO_CLASS(bprio);
178
179	if (aclass == IOPRIO_CLASS_NONE)
180		aclass = IOPRIO_CLASS_BE;
181	if (bclass == IOPRIO_CLASS_NONE)
182		bclass = IOPRIO_CLASS_BE;
183
184	if (aclass == bclass)
185		return min(aprio, bprio);
186	if (aclass > bclass)
187		return bprio;
188	else
189		return aprio;
190}
191
192SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
193{
194	struct task_struct *g, *p;
195	struct user_struct *user;
196	struct pid *pgrp;
197	int ret = -ESRCH;
198	int tmpio;
199
200	read_lock(&tasklist_lock);
201	switch (which) {
202		case IOPRIO_WHO_PROCESS:
203			if (!who)
204				p = current;
205			else
206				p = find_task_by_vpid(who);
207			if (p)
208				ret = get_task_ioprio(p);
209			break;
210		case IOPRIO_WHO_PGRP:
211			if (!who)
212				pgrp = task_pgrp(current);
213			else
214				pgrp = find_vpid(who);
215			do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
216				tmpio = get_task_ioprio(p);
217				if (tmpio < 0)
218					continue;
219				if (ret == -ESRCH)
220					ret = tmpio;
221				else
222					ret = ioprio_best(ret, tmpio);
223			} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
224			break;
225		case IOPRIO_WHO_USER:
226			if (!who)
227				user = current_user();
228			else
229				user = find_user(who);
230
231			if (!user)
232				break;
233
234			do_each_thread(g, p) {
235				if (__task_cred(p)->uid != user->uid)
236					continue;
237				tmpio = get_task_ioprio(p);
238				if (tmpio < 0)
239					continue;
240				if (ret == -ESRCH)
241					ret = tmpio;
242				else
243					ret = ioprio_best(ret, tmpio);
244			} while_each_thread(g, p);
245
246			if (who)
247				free_uid(user);
248			break;
249		default:
250			ret = -EINVAL;
251	}
252
253	read_unlock(&tasklist_lock);
254	return ret;
255}
256