[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_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 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
228ebcbe	97	p = find_task_by_vpid(who);
22e2c507 JA	98	if (p)
	99	ret = set_task_ioprio(p, ioprio);
	100	break;
	101	case IOPRIO_WHO_PGRP:
	102	if (!who)
41487c65 EB	103	pgrp = task_pgrp(current);
41487c65 EB	104	else
b488893a	105	pgrp = find_vpid(who);
41487c65	106	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
22e2c507 JA	107	ret = set_task_ioprio(p, ioprio);
	108	if (ret)
	109	break;
41487c65	110	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	111	break;
	112	case IOPRIO_WHO_USER:
	113	if (!who)
	114	user = current->user;
	115	else
	116	user = find_user(who);
	117
	118	if (!user)
	119	break;
	120
	121	do_each_thread(g, p) {
	122	if (p->uid != who)
	123	continue;
	124	ret = set_task_ioprio(p, ioprio);
	125	if (ret)
78bd4d48	126	goto free_uid;
22e2c507	127	} while_each_thread(g, p);
78bd4d48	128	free_uid:
22e2c507 JA	129	if (who)
	130	free_uid(user);
	131	break;
	132	default:
	133	ret = -EINVAL;
	134	}
	135
cf342e52	136	read_unlock(&tasklist_lock);
22e2c507 JA	137	return ret;
	138	}
	139
a1836a42 DQ	140	static int get_task_ioprio(struct task_struct *p)
	141	{
	142	int ret;
	143
	144	ret = security_task_getioprio(p);
	145	if (ret)
	146	goto out;
	147	ret = p->ioprio;
	148	out:
	149	return ret;
	150	}
	151
e014ff8d ON	152	int ioprio_best(unsigned short aprio, unsigned short bprio)
	153	{
	154	unsigned short aclass = IOPRIO_PRIO_CLASS(aprio);
	155	unsigned short bclass = IOPRIO_PRIO_CLASS(bprio);
	156
e014ff8d ON	157	if (aclass == IOPRIO_CLASS_NONE)
	158	aclass = IOPRIO_CLASS_BE;
	159	if (bclass == IOPRIO_CLASS_NONE)
	160	bclass = IOPRIO_CLASS_BE;
	161
	162	if (aclass == bclass)
	163	return min(aprio, bprio);
	164	if (aclass > bclass)
	165	return bprio;
	166	else
	167	return aprio;
	168	}
	169
cf366808	170	asmlinkage long sys_ioprio_get(int which, int who)
22e2c507 JA	171	{
	172	struct task_struct g, p;
	173	struct user_struct *user;
41487c65	174	struct pid *pgrp;
22e2c507	175	int ret = -ESRCH;
a1836a42	176	int tmpio;
22e2c507	177
cf342e52	178	read_lock(&tasklist_lock);
22e2c507 JA	179	switch (which) {
	180	case IOPRIO_WHO_PROCESS:
	181	if (!who)
	182	p = current;
	183	else
228ebcbe	184	p = find_task_by_vpid(who);
22e2c507	185	if (p)
a1836a42	186	ret = get_task_ioprio(p);
22e2c507 JA	187	break;
	188	case IOPRIO_WHO_PGRP:
	189	if (!who)
41487c65 EB	190	pgrp = task_pgrp(current);
41487c65 EB	191	else
b488893a	192	pgrp = find_vpid(who);
41487c65	193	do_each_pid_task(pgrp, PIDTYPE_PGID, p) {
a1836a42 DQ	194	tmpio = get_task_ioprio(p);
	195	if (tmpio < 0)
	196	continue;
22e2c507	197	if (ret == -ESRCH)
a1836a42	198	ret = tmpio;
22e2c507	199	else
a1836a42	200	ret = ioprio_best(ret, tmpio);
41487c65	201	} while_each_pid_task(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	202	break;
	203	case IOPRIO_WHO_USER:
	204	if (!who)
	205	user = current->user;
	206	else
	207	user = find_user(who);
	208
	209	if (!user)
	210	break;
	211
	212	do_each_thread(g, p) {
	213	if (p->uid != user->uid)
	214	continue;
a1836a42 DQ	215	tmpio = get_task_ioprio(p);
	216	if (tmpio < 0)
	217	continue;
22e2c507	218	if (ret == -ESRCH)
a1836a42	219	ret = tmpio;
22e2c507	220	else
a1836a42	221	ret = ioprio_best(ret, tmpio);
22e2c507 JA	222	} while_each_thread(g, p);
	223
	224	if (who)
	225	free_uid(user);
	226	break;
	227	default:
	228	ret = -EINVAL;
	229	}
	230
cf342e52	231	read_unlock(&tasklist_lock);
22e2c507 JA	232	return ret;
	233	}
	234