[net-next-2.6.git] / fs / ioprio.c

/*
 * fs/ioprio.c
 *
 * Copyright (C) 2004 Jens Axboe <axboe@kernel.dk>
 *
 * Helper functions for setting/querying io priorities of processes. The
 * system calls closely mimmick getpriority/setpriority, see the man page for
 * those. The prio argument is a composite of prio class and prio data, where
 * the data argument has meaning within that class. The standard scheduling
 * classes have 8 distinct prio levels, with 0 being the highest prio and 7
 * being the lowest.
 *
 * IOW, setting BE scheduling class with prio 2 is done ala:
 *
 * unsigned int prio = (IOPRIO_CLASS_BE << IOPRIO_CLASS_SHIFT) | 2;
 *
 * ioprio_set(PRIO_PROCESS, pid, prio);
 *
 * See also Documentation/block/ioprio.txt
 *
 */
#include <linux/kernel.h>
#include <linux/ioprio.h>
#include <linux/blkdev.h>
#include <linux/capability.h>
#include <linux/syscalls.h>
#include <linux/security.h>
#include <linux/pid_namespace.h>

static int set_task_ioprio(struct task_struct *task, int ioprio)
{
	int err;
	struct io_context *ioc;

	if (task->uid != current->euid &&
	    task->uid != current->uid && !capable(CAP_SYS_NICE))
		return -EPERM;

	err = security_task_setioprio(task, ioprio);
	if (err)
		return err;

	task_lock(task);

	task->ioprio = ioprio;

	ioc = task->io_context;
	/* see wmb() in current_io_context() */
	smp_read_barrier_depends();

	if (ioc)
		ioc->ioprio_changed = 1;

	task_unlock(task);
	return 0;
}

asmlinkage long sys_ioprio_set(int which, int who, int ioprio)
{
	int class = IOPRIO_PRIO_CLASS(ioprio);
	int data = IOPRIO_PRIO_DATA(ioprio);
	struct task_struct *p, *g;
	struct user_struct *user;
	struct pid *pgrp;
	int ret;

	switch (class) {
		case IOPRIO_CLASS_RT:
			if (!capable(CAP_SYS_ADMIN))
				return -EPERM;
			/* fall through, rt has prio field too */
		case IOPRIO_CLASS_BE:
			if (data >= IOPRIO_BE_NR || data < 0)
				return -EINVAL;

			break;
		case IOPRIO_CLASS_IDLE:
			if (!capable(CAP_SYS_ADMIN))
				return -EPERM;
			break;
		default:
			return -EINVAL;
	}

	ret = -ESRCH;
	/*
	 * We want IOPRIO_WHO_PGRP/IOPRIO_WHO_USER to be "atomic",
	 * so we can't use rcu_read_lock(). See re-copy of ->ioprio
	 * in copy_process().
	 */
	read_lock(&tasklist_lock);
	switch (which) {
		case IOPRIO_WHO_PROCESS:
			if (!who)
				p = current;
			else
				p = find_task_by_pid_ns(who,
						current->nsproxy->pid_ns);
			if (p)
				ret = set_task_ioprio(p, ioprio);
			break;
		case IOPRIO_WHO_PGRP:
			if (!who)
				pgrp = task_pgrp(current);
			else
				pgrp = find_vpid(who);
			do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
				ret = set_task_ioprio(p, ioprio);
				if (ret)
					break;
			} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
			break;
		case IOPRIO_WHO_USER:
			if (!who)
				user = current->user;
			else
				user = find_user(who);

			if (!user)
				break;

			do_each_thread(g, p) {
				if (p->uid != who)
					continue;
				ret = set_task_ioprio(p, ioprio);
				if (ret)
					goto free_uid;
			} while_each_thread(g, p);
free_uid:
			if (who)
				free_uid(user);
			break;
		default:
			ret = -EINVAL;
	}

	read_unlock(&tasklist_lock);
	return ret;
}

static int get_task_ioprio(struct task_struct *p)
{
	int ret;

	ret = security_task_getioprio(p);
	if (ret)
		goto out;
	ret = p->ioprio;
out:
	return ret;
}

int ioprio_best(unsigned short aprio, unsigned short bprio)
{
	unsigned short aclass = IOPRIO_PRIO_CLASS(aprio);
	unsigned short bclass = IOPRIO_PRIO_CLASS(bprio);

	if (aclass == IOPRIO_CLASS_NONE)
		aclass = IOPRIO_CLASS_BE;
	if (bclass == IOPRIO_CLASS_NONE)
		bclass = IOPRIO_CLASS_BE;

	if (aclass == bclass)
		return min(aprio, bprio);
	if (aclass > bclass)
		return bprio;
	else
		return aprio;
}

asmlinkage long sys_ioprio_get(int which, int who)
{
	struct task_struct *g, *p;
	struct user_struct *user;
	struct pid *pgrp;
	int ret = -ESRCH;
	int tmpio;

	read_lock(&tasklist_lock);
	switch (which) {
		case IOPRIO_WHO_PROCESS:
			if (!who)
				p = current;
			else
				p = find_task_by_pid_ns(who,
						current->nsproxy->pid_ns);
			if (p)
				ret = get_task_ioprio(p);
			break;
		case IOPRIO_WHO_PGRP:
			if (!who)
				pgrp = task_pgrp(current);
			else
				pgrp = find_vpid(who);
			do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
				tmpio = get_task_ioprio(p);
				if (tmpio < 0)
					continue;
				if (ret == -ESRCH)
					ret = tmpio;
				else
					ret = ioprio_best(ret, tmpio);
			} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
			break;
		case IOPRIO_WHO_USER:
			if (!who)
				user = current->user;
			else
				user = find_user(who);

			if (!user)
				break;

			do_each_thread(g, p) {
				if (p->uid != user->uid)
					continue;
				tmpio = get_task_ioprio(p);
				if (tmpio < 0)
					continue;
				if (ret == -ESRCH)
					ret = tmpio;
				else
					ret = ioprio_best(ret, tmpio);
			} while_each_thread(g, p);

			if (who)
				free_uid(user);
			break;
		default:
			ret = -EINVAL;
	}

	read_unlock(&tasklist_lock);
	return ret;
}
Commit	Line	Data
22e2c507 JA	1	/*
	2	* fs/ioprio.c
	3	*
0fe23479	4	* Copyright (C) 2004 Jens Axboe <axboe@kernel.dk>
22e2c507 JA	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>
16f7e0fe	25	#include <linux/capability.h>
9abdc4cd	26	#include <linux/syscalls.h>
03e68060	27	#include <linux/security.h>
b488893a	28	#include <linux/pid_namespace.h>
22e2c507 JA	29
	30	static int set_task_ioprio(struct task_struct *task, int ioprio)
	31	{
03e68060	32	int err;
22e2c507 JA	33	struct io_context *ioc;
	34
	35	if (task->uid != current->euid &&
	36	task->uid != current->uid && !capable(CAP_SYS_NICE))
	37	return -EPERM;
	38
03e68060 JM	39	err = security_task_setioprio(task, ioprio);
	40	if (err)
	41	return err;
	42
22e2c507 JA	43	task_lock(task);
	44
	45	task->ioprio = ioprio;
	46
	47	ioc = task->io_context;
9f83e45e ON	48	/* see wmb() in current_io_context() */
	49	smp_read_barrier_depends();
	50
fc46379d JA	51	if (ioc)
fc46379d JA	52	ioc->ioprio_changed = 1;
22e2c507 JA	53
	54	task_unlock(task);
	55	return 0;
	56	}
	57
cf366808	58	asmlinkage long sys_ioprio_set(int which, int who, int ioprio)
22e2c507 JA	59	{
	60	int class = IOPRIO_PRIO_CLASS(ioprio);
	61	int data = IOPRIO_PRIO_DATA(ioprio);
	62	struct task_struct p, g;
	63	struct user_struct *user;
41487c65	64	struct pid *pgrp;
22e2c507 JA	65	int ret;
	66
	67	switch (class) {
	68	case IOPRIO_CLASS_RT:
	69	if (!capable(CAP_SYS_ADMIN))
	70	return -EPERM;
	71	/* fall through, rt has prio field too */
	72	case IOPRIO_CLASS_BE:
	73	if (data >= IOPRIO_BE_NR \|\| data < 0)
	74	return -EINVAL;
	75
	76	break;
	77	case IOPRIO_CLASS_IDLE:
f6fdd7d9 LT	78	if (!capable(CAP_SYS_ADMIN))
f6fdd7d9 LT	79	return -EPERM;
22e2c507 JA	80	break;
	81	default:
	82	return -EINVAL;
	83	}
	84
	85	ret = -ESRCH;
cf342e52 ON	86	/*
	87	* We want IOPRIO_WHO_PGRP/IOPRIO_WHO_USER to be "atomic",
	88	* so we can't use rcu_read_lock(). See re-copy of ->ioprio
	89	* in copy_process().
	90	*/
	91	read_lock(&tasklist_lock);
22e2c507 JA	92	switch (which) {
	93	case IOPRIO_WHO_PROCESS:
	94	if (!who)
	95	p = current;
	96	else
b488893a PE	97	p = find_task_by_pid_ns(who,
b488893a PE	98	current->nsproxy->pid_ns);
22e2c507 JA	99	if (p)
	100	ret = set_task_ioprio(p, ioprio);
	101	break;
	102	case IOPRIO_WHO_PGRP:
	103	if (!who)
41487c65 EB	104	pgrp = task_pgrp(current);
41487c65 EB	105	else
b488893a	106	pgrp = find_vpid(who);
41487c65	107	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
22e2c507 JA	108	ret = set_task_ioprio(p, ioprio);
	109	if (ret)
	110	break;
41487c65	111	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	112	break;
	113	case IOPRIO_WHO_USER:
	114	if (!who)
	115	user = current->user;
	116	else
	117	user = find_user(who);
	118
	119	if (!user)
	120	break;
	121
	122	do_each_thread(g, p) {
	123	if (p->uid != who)
	124	continue;
	125	ret = set_task_ioprio(p, ioprio);
	126	if (ret)
78bd4d48	127	goto free_uid;
22e2c507	128	} while_each_thread(g, p);
78bd4d48	129	free_uid:
22e2c507 JA	130	if (who)
	131	free_uid(user);
	132	break;
	133	default:
	134	ret = -EINVAL;
	135	}
	136
cf342e52	137	read_unlock(&tasklist_lock);
22e2c507 JA	138	return ret;
	139	}
	140
a1836a42 DQ	141	static int get_task_ioprio(struct task_struct *p)
	142	{
	143	int ret;
	144
	145	ret = security_task_getioprio(p);
	146	if (ret)
	147	goto out;
	148	ret = p->ioprio;
	149	out:
	150	return ret;
	151	}
	152
e014ff8d ON	153	int ioprio_best(unsigned short aprio, unsigned short bprio)
	154	{
	155	unsigned short aclass = IOPRIO_PRIO_CLASS(aprio);
	156	unsigned short bclass = IOPRIO_PRIO_CLASS(bprio);
	157
e014ff8d ON	158	if (aclass == IOPRIO_CLASS_NONE)
	159	aclass = IOPRIO_CLASS_BE;
	160	if (bclass == IOPRIO_CLASS_NONE)
	161	bclass = IOPRIO_CLASS_BE;
	162
	163	if (aclass == bclass)
	164	return min(aprio, bprio);
	165	if (aclass > bclass)
	166	return bprio;
	167	else
	168	return aprio;
	169	}
	170
cf366808	171	asmlinkage long sys_ioprio_get(int which, int who)
22e2c507 JA	172	{
	173	struct task_struct g, p;
	174	struct user_struct *user;
41487c65	175	struct pid *pgrp;
22e2c507	176	int ret = -ESRCH;
a1836a42	177	int tmpio;
22e2c507	178
cf342e52	179	read_lock(&tasklist_lock);
22e2c507 JA	180	switch (which) {
	181	case IOPRIO_WHO_PROCESS:
	182	if (!who)
	183	p = current;
	184	else
b488893a PE	185	p = find_task_by_pid_ns(who,
b488893a PE	186	current->nsproxy->pid_ns);
22e2c507	187	if (p)
a1836a42	188	ret = get_task_ioprio(p);
22e2c507 JA	189	break;
	190	case IOPRIO_WHO_PGRP:
	191	if (!who)
41487c65 EB	192	pgrp = task_pgrp(current);
41487c65 EB	193	else
b488893a	194	pgrp = find_vpid(who);
41487c65	195	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
a1836a42 DQ	196	tmpio = get_task_ioprio(p);
	197	if (tmpio < 0)
	198	continue;
22e2c507	199	if (ret == -ESRCH)
a1836a42	200	ret = tmpio;
22e2c507	201	else
a1836a42	202	ret = ioprio_best(ret, tmpio);
41487c65	203	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	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 (p->uid != user->uid)
	216	continue;
a1836a42 DQ	217	tmpio = get_task_ioprio(p);
	218	if (tmpio < 0)
	219	continue;
22e2c507	220	if (ret == -ESRCH)
a1836a42	221	ret = tmpio;
22e2c507	222	else
a1836a42	223	ret = ioprio_best(ret, tmpio);
22e2c507 JA	224	} while_each_thread(g, p);
	225
	226	if (who)
	227	free_uid(user);
	228	break;
	229	default:
	230	ret = -EINVAL;
	231	}
	232
cf342e52	233	read_unlock(&tasklist_lock);
22e2c507 JA	234	return ret;
	235	}
	236