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