[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;
		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_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_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_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	80	break;
8ec680e4 JA	81	case IOPRIO_CLASS_NONE:
	82	if (data)
	83	return -EINVAL;
	84	break;
22e2c507 JA	85	default:
	86	return -EINVAL;
	87	}
	88
	89	ret = -ESRCH;
cf342e52 ON	90	/*
	91	* We want IOPRIO_WHO_PGRP/IOPRIO_WHO_USER to be "atomic",
	92	* so we can't use rcu_read_lock(). See re-copy of ->ioprio
	93	* in copy_process().
	94	*/
	95	read_lock(&tasklist_lock);
22e2c507 JA	96	switch (which) {
	97	case IOPRIO_WHO_PROCESS:
	98	if (!who)
	99	p = current;
	100	else
228ebcbe	101	p = find_task_by_vpid(who);
22e2c507 JA	102	if (p)
	103	ret = set_task_ioprio(p, ioprio);
	104	break;
	105	case IOPRIO_WHO_PGRP:
	106	if (!who)
41487c65 EB	107	pgrp = task_pgrp(current);
41487c65 EB	108	else
b488893a	109	pgrp = find_vpid(who);
41487c65	110	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
22e2c507 JA	111	ret = set_task_ioprio(p, ioprio);
	112	if (ret)
	113	break;
41487c65	114	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	115	break;
	116	case IOPRIO_WHO_USER:
	117	if (!who)
	118	user = current->user;
	119	else
	120	user = find_user(who);
	121
	122	if (!user)
	123	break;
	124
	125	do_each_thread(g, p) {
	126	if (p->uid != who)
	127	continue;
	128	ret = set_task_ioprio(p, ioprio);
	129	if (ret)
78bd4d48	130	goto free_uid;
22e2c507	131	} while_each_thread(g, p);
78bd4d48	132	free_uid:
22e2c507 JA	133	if (who)
	134	free_uid(user);
	135	break;
	136	default:
	137	ret = -EINVAL;
	138	}
	139
cf342e52	140	read_unlock(&tasklist_lock);
22e2c507 JA	141	return ret;
	142	}
	143
a1836a42 DQ	144	static 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 = p->ioprio;
	152	out:
	153	return ret;
	154	}
	155
e014ff8d ON	156	int ioprio_best(unsigned short aprio, unsigned short bprio)
	157	{
	158	unsigned short aclass = IOPRIO_PRIO_CLASS(aprio);
	159	unsigned short bclass = IOPRIO_PRIO_CLASS(bprio);
	160
e014ff8d ON	161	if (aclass == IOPRIO_CLASS_NONE)
	162	aclass = IOPRIO_CLASS_BE;
	163	if (bclass == IOPRIO_CLASS_NONE)
	164	bclass = IOPRIO_CLASS_BE;
	165
	166	if (aclass == bclass)
	167	return min(aprio, bprio);
	168	if (aclass > bclass)
	169	return bprio;
	170	else
	171	return aprio;
	172	}
	173
cf366808	174	asmlinkage long sys_ioprio_get(int which, int who)
22e2c507 JA	175	{
	176	struct task_struct g, p;
	177	struct user_struct *user;
41487c65	178	struct pid *pgrp;
22e2c507	179	int ret = -ESRCH;
a1836a42	180	int tmpio;
22e2c507	181
cf342e52	182	read_lock(&tasklist_lock);
22e2c507 JA	183	switch (which) {
	184	case IOPRIO_WHO_PROCESS:
	185	if (!who)
	186	p = current;
	187	else
228ebcbe	188	p = find_task_by_vpid(who);
22e2c507	189	if (p)
a1836a42	190	ret = get_task_ioprio(p);
22e2c507 JA	191	break;
	192	case IOPRIO_WHO_PGRP:
	193	if (!who)
41487c65 EB	194	pgrp = task_pgrp(current);
41487c65 EB	195	else
b488893a	196	pgrp = find_vpid(who);
41487c65	197	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
a1836a42 DQ	198	tmpio = get_task_ioprio(p);
	199	if (tmpio < 0)
	200	continue;
22e2c507	201	if (ret == -ESRCH)
a1836a42	202	ret = tmpio;
22e2c507	203	else
a1836a42	204	ret = ioprio_best(ret, tmpio);
41487c65	205	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	206	break;
	207	case IOPRIO_WHO_USER:
	208	if (!who)
	209	user = current->user;
	210	else
	211	user = find_user(who);
	212
	213	if (!user)
	214	break;
	215
	216	do_each_thread(g, p) {
	217	if (p->uid != user->uid)
	218	continue;
a1836a42 DQ	219	tmpio = get_task_ioprio(p);
	220	if (tmpio < 0)
	221	continue;
22e2c507	222	if (ret == -ESRCH)
a1836a42	223	ret = tmpio;
22e2c507	224	else
a1836a42	225	ret = ioprio_best(ret, tmpio);
22e2c507 JA	226	} while_each_thread(g, p);
	227
	228	if (who)
	229	free_uid(user);
	230	break;
	231	default:
	232	ret = -EINVAL;
	233	}
	234
cf342e52	235	read_unlock(&tasklist_lock);
22e2c507 JA	236	return ret;
	237	}
	238