[thirdparty/kernel/stable.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/export.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;
	rcu_read_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;
	}

	rcu_read_unlock();
	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;

	rcu_read_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;
	}

	rcu_read_unlock();
	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	23	#include <linux/kernel.h>
afeacc8c	24	#include <linux/export.h>
22e2c507 JA	25	#include <linux/ioprio.h>
22e2c507 JA	26	#include <linux/blkdev.h>
16f7e0fe	27	#include <linux/capability.h>
9abdc4cd	28	#include <linux/syscalls.h>
03e68060	29	#include <linux/security.h>
b488893a	30	#include <linux/pid_namespace.h>
22e2c507	31
b3881f74	32	int set_task_ioprio(struct task_struct *task, int ioprio)
22e2c507	33	{
03e68060	34	int err;
22e2c507	35	struct io_context *ioc;
c69e8d9c	36	const struct cred cred = current_cred(), tcred;
22e2c507	37
c69e8d9c DH	38	rcu_read_lock();
	39	tcred = __task_cred(task);
	40	if (tcred->uid != cred->euid &&
	41	tcred->uid != cred->uid && !capable(CAP_SYS_NICE)) {
	42	rcu_read_unlock();
22e2c507	43	return -EPERM;
c69e8d9c DH	44	}
c69e8d9c DH	45	rcu_read_unlock();
22e2c507	46
03e68060 JM	47	err = security_task_setioprio(task, ioprio);
	48	if (err)
	49	return err;
	50
22e2c507	51	task_lock(task);
fd0928df JA	52	do {
	53	ioc = task->io_context;
	54	/* see wmb() in current_io_context() */
	55	smp_read_barrier_depends();
	56	if (ioc)
	57	break;
22e2c507	58
fd0928df JA	59	ioc = alloc_io_context(GFP_ATOMIC, -1);
	60	if (!ioc) {
	61	err = -ENOMEM;
	62	break;
	63	}
	64	task->io_context = ioc;
fd0928df	65	} while (1);
9f83e45e	66
fd0928df JA	67	if (!err) {
fd0928df JA	68	ioc->ioprio = ioprio;
fc46379d	69	ioc->ioprio_changed = 1;
fd0928df	70	}
22e2c507 JA	71
22e2c507 JA	72	task_unlock(task);
fd0928df	73	return err;
22e2c507	74	}
b3881f74	75	EXPORT_SYMBOL_GPL(set_task_ioprio);
22e2c507	76
938bb9f5	77	SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio)
22e2c507 JA	78	{
	79	int class = IOPRIO_PRIO_CLASS(ioprio);
	80	int data = IOPRIO_PRIO_DATA(ioprio);
	81	struct task_struct p, g;
	82	struct user_struct *user;
41487c65	83	struct pid *pgrp;
22e2c507 JA	84	int ret;
	85
	86	switch (class) {
	87	case IOPRIO_CLASS_RT:
	88	if (!capable(CAP_SYS_ADMIN))
	89	return -EPERM;
	90	/* fall through, rt has prio field too */
	91	case IOPRIO_CLASS_BE:
	92	if (data >= IOPRIO_BE_NR \|\| data < 0)
	93	return -EINVAL;
	94
	95	break;
	96	case IOPRIO_CLASS_IDLE:
	97	break;
8ec680e4 JA	98	case IOPRIO_CLASS_NONE:
	99	if (data)
	100	return -EINVAL;
	101	break;
22e2c507 JA	102	default:
	103	return -EINVAL;
	104	}
	105
	106	ret = -ESRCH;
d69b78ba	107	rcu_read_lock();
22e2c507 JA	108	switch (which) {
	109	case IOPRIO_WHO_PROCESS:
	110	if (!who)
	111	p = current;
	112	else
228ebcbe	113	p = find_task_by_vpid(who);
22e2c507 JA	114	if (p)
	115	ret = set_task_ioprio(p, ioprio);
	116	break;
	117	case IOPRIO_WHO_PGRP:
	118	if (!who)
41487c65 EB	119	pgrp = task_pgrp(current);
41487c65 EB	120	else
b488893a	121	pgrp = find_vpid(who);
2d70b68d	122	do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
22e2c507 JA	123	ret = set_task_ioprio(p, ioprio);
	124	if (ret)
	125	break;
2d70b68d	126	} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	127	break;
	128	case IOPRIO_WHO_USER:
	129	if (!who)
86a264ab	130	user = current_user();
22e2c507 JA	131	else
	132	user = find_user(who);
	133
	134	if (!user)
	135	break;
	136
	137	do_each_thread(g, p) {
d69b78ba	138	if (__task_cred(p)->uid != who)
22e2c507 JA	139	continue;
	140	ret = set_task_ioprio(p, ioprio);
	141	if (ret)
78bd4d48	142	goto free_uid;
22e2c507	143	} while_each_thread(g, p);
78bd4d48	144	free_uid:
22e2c507 JA	145	if (who)
	146	free_uid(user);
	147	break;
	148	default:
	149	ret = -EINVAL;
	150	}
	151
d69b78ba	152	rcu_read_unlock();
22e2c507 JA	153	return ret;
	154	}
	155
a1836a42 DQ	156	static int get_task_ioprio(struct task_struct *p)
	157	{
	158	int ret;
	159
	160	ret = security_task_getioprio(p);
	161	if (ret)
	162	goto out;
fd0928df JA	163	ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
	164	if (p->io_context)
	165	ret = p->io_context->ioprio;
a1836a42 DQ	166	out:
	167	return ret;
	168	}
	169
e014ff8d ON	170	int ioprio_best(unsigned short aprio, unsigned short bprio)
	171	{
	172	unsigned short aclass = IOPRIO_PRIO_CLASS(aprio);
	173	unsigned short bclass = IOPRIO_PRIO_CLASS(bprio);
	174
e014ff8d ON	175	if (aclass == IOPRIO_CLASS_NONE)
	176	aclass = IOPRIO_CLASS_BE;
	177	if (bclass == IOPRIO_CLASS_NONE)
	178	bclass = IOPRIO_CLASS_BE;
	179
	180	if (aclass == bclass)
	181	return min(aprio, bprio);
	182	if (aclass > bclass)
	183	return bprio;
	184	else
	185	return aprio;
	186	}
	187
938bb9f5	188	SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
22e2c507 JA	189	{
	190	struct task_struct g, p;
	191	struct user_struct *user;
41487c65	192	struct pid *pgrp;
22e2c507	193	int ret = -ESRCH;
a1836a42	194	int tmpio;
22e2c507	195
d69b78ba	196	rcu_read_lock();
22e2c507 JA	197	switch (which) {
	198	case IOPRIO_WHO_PROCESS:
	199	if (!who)
	200	p = current;
	201	else
228ebcbe	202	p = find_task_by_vpid(who);
22e2c507	203	if (p)
a1836a42	204	ret = get_task_ioprio(p);
22e2c507 JA	205	break;
	206	case IOPRIO_WHO_PGRP:
	207	if (!who)
41487c65 EB	208	pgrp = task_pgrp(current);
41487c65 EB	209	else
b488893a	210	pgrp = find_vpid(who);
2d70b68d	211	do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
a1836a42 DQ	212	tmpio = get_task_ioprio(p);
	213	if (tmpio < 0)
	214	continue;
22e2c507	215	if (ret == -ESRCH)
a1836a42	216	ret = tmpio;
22e2c507	217	else
a1836a42	218	ret = ioprio_best(ret, tmpio);
2d70b68d	219	} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	220	break;
	221	case IOPRIO_WHO_USER:
	222	if (!who)
86a264ab	223	user = current_user();
22e2c507 JA	224	else
	225	user = find_user(who);
	226
	227	if (!user)
	228	break;
	229
	230	do_each_thread(g, p) {
d69b78ba	231	if (__task_cred(p)->uid != user->uid)
22e2c507	232	continue;
a1836a42 DQ	233	tmpio = get_task_ioprio(p);
	234	if (tmpio < 0)
	235	continue;
22e2c507	236	if (ret == -ESRCH)
a1836a42	237	ret = tmpio;
22e2c507	238	else
a1836a42	239	ret = ioprio_best(ret, tmpio);
22e2c507 JA	240	} while_each_thread(g, p);
	241
	242	if (who)
	243	free_uid(user);
	244	break;
	245	default:
	246	ret = -EINVAL;
	247	}
	248
d69b78ba	249	rcu_read_unlock();
22e2c507 JA	250	return ret;
22e2c507 JA	251	}