[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>

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;
	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(who);
			if (p)
				ret = set_task_ioprio(p, ioprio);
			break;
		case IOPRIO_WHO_PGRP:
			if (!who)
				who = process_group(current);
			do_each_task_pid(who, PIDTYPE_PGID, p) {
				ret = set_task_ioprio(p, ioprio);
				if (ret)
					break;
			} while_each_task_pid(who, 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;
	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(who);
			if (p)
				ret = get_task_ioprio(p);
			break;
		case IOPRIO_WHO_PGRP:
			if (!who)
				who = process_group(current);
			do_each_task_pid(who, 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_task_pid(who, 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>
22e2c507 JA	28
	29	static int set_task_ioprio(struct task_struct *task, int ioprio)
	30	{
03e68060	31	int err;
22e2c507 JA	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
03e68060 JM	38	err = security_task_setioprio(task, ioprio);
	39	if (err)
	40	return err;
	41
22e2c507 JA	42	task_lock(task);
	43
	44	task->ioprio = ioprio;
	45
	46	ioc = task->io_context;
9f83e45e ON	47	/* see wmb() in current_io_context() */
	48	smp_read_barrier_depends();
	49
fc46379d JA	50	if (ioc)
fc46379d JA	51	ioc->ioprio_changed = 1;
22e2c507 JA	52
	53	task_unlock(task);
	54	return 0;
	55	}
	56
cf366808	57	asmlinkage long sys_ioprio_set(int which, int who, int ioprio)
22e2c507 JA	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	int ret;
	64
	65	switch (class) {
	66	case IOPRIO_CLASS_RT:
	67	if (!capable(CAP_SYS_ADMIN))
	68	return -EPERM;
	69	/* fall through, rt has prio field too */
	70	case IOPRIO_CLASS_BE:
	71	if (data >= IOPRIO_BE_NR \|\| data < 0)
	72	return -EINVAL;
	73
	74	break;
	75	case IOPRIO_CLASS_IDLE:
f6fdd7d9 LT	76	if (!capable(CAP_SYS_ADMIN))
f6fdd7d9 LT	77	return -EPERM;
22e2c507 JA	78	break;
	79	default:
	80	return -EINVAL;
	81	}
	82
	83	ret = -ESRCH;
cf342e52 ON	84	/*
	85	* We want IOPRIO_WHO_PGRP/IOPRIO_WHO_USER to be "atomic",
	86	* so we can't use rcu_read_lock(). See re-copy of ->ioprio
	87	* in copy_process().
	88	*/
	89	read_lock(&tasklist_lock);
22e2c507 JA	90	switch (which) {
	91	case IOPRIO_WHO_PROCESS:
	92	if (!who)
	93	p = current;
	94	else
	95	p = find_task_by_pid(who);
	96	if (p)
	97	ret = set_task_ioprio(p, ioprio);
	98	break;
	99	case IOPRIO_WHO_PGRP:
	100	if (!who)
	101	who = process_group(current);
	102	do_each_task_pid(who, PIDTYPE_PGID, p) {
	103	ret = set_task_ioprio(p, ioprio);
	104	if (ret)
	105	break;
	106	} while_each_task_pid(who, PIDTYPE_PGID, p);
	107	break;
	108	case IOPRIO_WHO_USER:
	109	if (!who)
	110	user = current->user;
	111	else
	112	user = find_user(who);
	113
	114	if (!user)
	115	break;
	116
	117	do_each_thread(g, p) {
	118	if (p->uid != who)
	119	continue;
	120	ret = set_task_ioprio(p, ioprio);
	121	if (ret)
78bd4d48	122	goto free_uid;
22e2c507	123	} while_each_thread(g, p);
78bd4d48	124	free_uid:
22e2c507 JA	125	if (who)
	126	free_uid(user);
	127	break;
	128	default:
	129	ret = -EINVAL;
	130	}
	131
cf342e52	132	read_unlock(&tasklist_lock);
22e2c507 JA	133	return ret;
	134	}
	135
a1836a42 DQ	136	static int get_task_ioprio(struct task_struct *p)
	137	{
	138	int ret;
	139
	140	ret = security_task_getioprio(p);
	141	if (ret)
	142	goto out;
	143	ret = p->ioprio;
	144	out:
	145	return ret;
	146	}
	147
e014ff8d ON	148	int ioprio_best(unsigned short aprio, unsigned short bprio)
	149	{
	150	unsigned short aclass = IOPRIO_PRIO_CLASS(aprio);
	151	unsigned short bclass = IOPRIO_PRIO_CLASS(bprio);
	152
e014ff8d ON	153	if (aclass == IOPRIO_CLASS_NONE)
	154	aclass = IOPRIO_CLASS_BE;
	155	if (bclass == IOPRIO_CLASS_NONE)
	156	bclass = IOPRIO_CLASS_BE;
	157
	158	if (aclass == bclass)
	159	return min(aprio, bprio);
	160	if (aclass > bclass)
	161	return bprio;
	162	else
	163	return aprio;
	164	}
	165
cf366808	166	asmlinkage long sys_ioprio_get(int which, int who)
22e2c507 JA	167	{
	168	struct task_struct g, p;
	169	struct user_struct *user;
	170	int ret = -ESRCH;
a1836a42	171	int tmpio;
22e2c507	172
cf342e52	173	read_lock(&tasklist_lock);
22e2c507 JA	174	switch (which) {
	175	case IOPRIO_WHO_PROCESS:
	176	if (!who)
	177	p = current;
	178	else
	179	p = find_task_by_pid(who);
	180	if (p)
a1836a42	181	ret = get_task_ioprio(p);
22e2c507 JA	182	break;
	183	case IOPRIO_WHO_PGRP:
	184	if (!who)
	185	who = process_group(current);
	186	do_each_task_pid(who, PIDTYPE_PGID, p) {
a1836a42 DQ	187	tmpio = get_task_ioprio(p);
	188	if (tmpio < 0)
	189	continue;
22e2c507	190	if (ret == -ESRCH)
a1836a42	191	ret = tmpio;
22e2c507	192	else
a1836a42	193	ret = ioprio_best(ret, tmpio);
22e2c507 JA	194	} while_each_task_pid(who, PIDTYPE_PGID, p);
	195	break;
	196	case IOPRIO_WHO_USER:
	197	if (!who)
	198	user = current->user;
	199	else
	200	user = find_user(who);
	201
	202	if (!user)
	203	break;
	204
	205	do_each_thread(g, p) {
	206	if (p->uid != user->uid)
	207	continue;
a1836a42 DQ	208	tmpio = get_task_ioprio(p);
	209	if (tmpio < 0)
	210	continue;
22e2c507	211	if (ret == -ESRCH)
a1836a42	212	ret = tmpio;
22e2c507	213	else
a1836a42	214	ret = ioprio_best(ret, tmpio);
22e2c507 JA	215	} while_each_thread(g, p);
	216
	217	if (who)
	218	free_uid(user);
	219	break;
	220	default:
	221	ret = -EINVAL;
	222	}
	223
cf342e52	224	read_unlock(&tasklist_lock);
22e2c507 JA	225	return ret;
	226	}
	227