[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/gfp.h>
#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>

int set_task_ioprio(struct task_struct *task, int ioprio)
{
	int err;
	struct io_context *ioc;
	const struct cred *cred = current_cred(), *tcred;

	rcu_read_lock();
	tcred = __task_cred(task);
	if (tcred->uid != cred->euid &&
	    tcred->uid != cred->uid && !capable(CAP_SYS_NICE)) {
		rcu_read_unlock();
		return -EPERM;
	}
	rcu_read_unlock();

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

	task_lock(task);
	do {
		ioc = task->io_context;
		/* see wmb() in current_io_context() */
		smp_read_barrier_depends();
		if (ioc)
			break;

		ioc = alloc_io_context(GFP_ATOMIC, -1);
		if (!ioc) {
			err = -ENOMEM;
			break;
		}
		task->io_context = ioc;
	} while (1);

	if (!err) {
		ioc->ioprio = ioprio;
		ioc->ioprio_changed = 1;
	}

	task_unlock(task);
	return err;
}
EXPORT_SYMBOL_GPL(set_task_ioprio);

SYSCALL_DEFINE3(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:
			break;
		case IOPRIO_CLASS_NONE:
			if (data)
				return -EINVAL;
			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_vpid(who);
			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_thread(pgrp, PIDTYPE_PGID, p) {
				ret = set_task_ioprio(p, ioprio);
				if (ret)
					break;
			} while_each_pid_thread(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 (__task_cred(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 = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
	if (p->io_context)
		ret = p->io_context->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;
}

SYSCALL_DEFINE2(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_vpid(who);
			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_thread(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_thread(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 (__task_cred(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	*/
5a0e3ad6	22	#include <linux/gfp.h>
22e2c507 JA	23	#include <linux/kernel.h>
	24	#include <linux/ioprio.h>
	25	#include <linux/blkdev.h>
16f7e0fe	26	#include <linux/capability.h>
9abdc4cd	27	#include <linux/syscalls.h>
03e68060	28	#include <linux/security.h>
b488893a	29	#include <linux/pid_namespace.h>
22e2c507	30
b3881f74	31	int set_task_ioprio(struct task_struct *task, int ioprio)
22e2c507	32	{
03e68060	33	int err;
22e2c507	34	struct io_context *ioc;
c69e8d9c	35	const struct cred cred = current_cred(), tcred;
22e2c507	36
c69e8d9c DH	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();
22e2c507	42	return -EPERM;
c69e8d9c DH	43	}
c69e8d9c DH	44	rcu_read_unlock();
22e2c507	45
03e68060 JM	46	err = security_task_setioprio(task, ioprio);
	47	if (err)
	48	return err;
	49
22e2c507	50	task_lock(task);
fd0928df JA	51	do {
	52	ioc = task->io_context;
	53	/* see wmb() in current_io_context() */
	54	smp_read_barrier_depends();
	55	if (ioc)
	56	break;
22e2c507	57
fd0928df JA	58	ioc = alloc_io_context(GFP_ATOMIC, -1);
	59	if (!ioc) {
	60	err = -ENOMEM;
	61	break;
	62	}
	63	task->io_context = ioc;
fd0928df	64	} while (1);
9f83e45e	65
fd0928df JA	66	if (!err) {
fd0928df JA	67	ioc->ioprio = ioprio;
fc46379d	68	ioc->ioprio_changed = 1;
fd0928df	69	}
22e2c507 JA	70
22e2c507 JA	71	task_unlock(task);
fd0928df	72	return err;
22e2c507	73	}
b3881f74	74	EXPORT_SYMBOL_GPL(set_task_ioprio);
22e2c507	75
938bb9f5	76	SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio)
22e2c507 JA	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;
41487c65	82	struct pid *pgrp;
22e2c507 JA	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;
8ec680e4 JA	97	case IOPRIO_CLASS_NONE:
	98	if (data)
	99	return -EINVAL;
	100	break;
22e2c507 JA	101	default:
	102	return -EINVAL;
	103	}
	104
	105	ret = -ESRCH;
cf342e52 ON	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);
22e2c507 JA	112	switch (which) {
	113	case IOPRIO_WHO_PROCESS:
	114	if (!who)
	115	p = current;
	116	else
228ebcbe	117	p = find_task_by_vpid(who);
22e2c507 JA	118	if (p)
	119	ret = set_task_ioprio(p, ioprio);
	120	break;
	121	case IOPRIO_WHO_PGRP:
	122	if (!who)
41487c65 EB	123	pgrp = task_pgrp(current);
41487c65 EB	124	else
b488893a	125	pgrp = find_vpid(who);
2d70b68d	126	do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
22e2c507 JA	127	ret = set_task_ioprio(p, ioprio);
	128	if (ret)
	129	break;
2d70b68d	130	} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	131	break;
	132	case IOPRIO_WHO_USER:
	133	if (!who)
86a264ab	134	user = current_user();
22e2c507 JA	135	else
	136	user = find_user(who);
	137
	138	if (!user)
	139	break;
	140
	141	do_each_thread(g, p) {
c69e8d9c	142	if (__task_cred(p)->uid != who)
22e2c507 JA	143	continue;
	144	ret = set_task_ioprio(p, ioprio);
	145	if (ret)
78bd4d48	146	goto free_uid;
22e2c507	147	} while_each_thread(g, p);
78bd4d48	148	free_uid:
22e2c507 JA	149	if (who)
	150	free_uid(user);
	151	break;
	152	default:
	153	ret = -EINVAL;
	154	}
	155
cf342e52	156	read_unlock(&tasklist_lock);
22e2c507 JA	157	return ret;
	158	}
	159
a1836a42 DQ	160	static 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;
fd0928df JA	167	ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
	168	if (p->io_context)
	169	ret = p->io_context->ioprio;
a1836a42 DQ	170	out:
	171	return ret;
	172	}
	173
e014ff8d ON	174	int 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
e014ff8d ON	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
938bb9f5	192	SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
22e2c507 JA	193	{
	194	struct task_struct g, p;
	195	struct user_struct *user;
41487c65	196	struct pid *pgrp;
22e2c507	197	int ret = -ESRCH;
a1836a42	198	int tmpio;
22e2c507	199
cf342e52	200	read_lock(&tasklist_lock);
22e2c507 JA	201	switch (which) {
	202	case IOPRIO_WHO_PROCESS:
	203	if (!who)
	204	p = current;
	205	else
228ebcbe	206	p = find_task_by_vpid(who);
22e2c507	207	if (p)
a1836a42	208	ret = get_task_ioprio(p);
22e2c507 JA	209	break;
	210	case IOPRIO_WHO_PGRP:
	211	if (!who)
41487c65 EB	212	pgrp = task_pgrp(current);
41487c65 EB	213	else
b488893a	214	pgrp = find_vpid(who);
2d70b68d	215	do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
a1836a42 DQ	216	tmpio = get_task_ioprio(p);
	217	if (tmpio < 0)
	218	continue;
22e2c507	219	if (ret == -ESRCH)
a1836a42	220	ret = tmpio;
22e2c507	221	else
a1836a42	222	ret = ioprio_best(ret, tmpio);
2d70b68d	223	} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	224	break;
	225	case IOPRIO_WHO_USER:
	226	if (!who)
86a264ab	227	user = current_user();
22e2c507 JA	228	else
	229	user = find_user(who);
	230
	231	if (!user)
	232	break;
	233
	234	do_each_thread(g, p) {
c69e8d9c	235	if (__task_cred(p)->uid != user->uid)
22e2c507	236	continue;
a1836a42 DQ	237	tmpio = get_task_ioprio(p);
	238	if (tmpio < 0)
	239	continue;
22e2c507	240	if (ret == -ESRCH)
a1836a42	241	ret = tmpio;
22e2c507	242	else
a1836a42	243	ret = ioprio_best(ret, tmpio);
22e2c507 JA	244	} while_each_thread(g, p);
	245
	246	if (who)
	247	free_uid(user);
	248	break;
	249	default:
	250	ret = -EINVAL;
	251	}
	252
cf342e52	253	read_unlock(&tasklist_lock);
22e2c507 JA	254	return ret;
22e2c507 JA	255	}