[linux.git] / fs / ioprio.c

/*
 * fs/ioprio.c
 *
 * Copyright (C) 2004 Jens Axboe <[email protected]>
 *
 * 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;

	ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE);
	if (ioc) {
		ioc_ioprio_changed(ioc, ioprio);
		put_io_context(ioc);
	}

	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 <[email protected]>
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
6e736be7 TH	51	ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE);
6e736be7 TH	52	if (ioc) {
dc86900e	53	ioc_ioprio_changed(ioc, ioprio);
11a3122f	54	put_io_context(ioc);
fd0928df	55	}
22e2c507	56
fd0928df	57	return err;
22e2c507	58	}
b3881f74	59	EXPORT_SYMBOL_GPL(set_task_ioprio);
22e2c507	60
938bb9f5	61	SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio)
22e2c507 JA	62	{
	63	int class = IOPRIO_PRIO_CLASS(ioprio);
	64	int data = IOPRIO_PRIO_DATA(ioprio);
	65	struct task_struct p, g;
	66	struct user_struct *user;
41487c65	67	struct pid *pgrp;
22e2c507 JA	68	int ret;
	69
	70	switch (class) {
	71	case IOPRIO_CLASS_RT:
	72	if (!capable(CAP_SYS_ADMIN))
	73	return -EPERM;
	74	/* fall through, rt has prio field too */
	75	case IOPRIO_CLASS_BE:
	76	if (data >= IOPRIO_BE_NR \|\| data < 0)
	77	return -EINVAL;
	78
	79	break;
	80	case IOPRIO_CLASS_IDLE:
	81	break;
8ec680e4 JA	82	case IOPRIO_CLASS_NONE:
	83	if (data)
	84	return -EINVAL;
	85	break;
22e2c507 JA	86	default:
	87	return -EINVAL;
	88	}
	89
	90	ret = -ESRCH;
d69b78ba	91	rcu_read_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);
2d70b68d	106	do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
22e2c507 JA	107	ret = set_task_ioprio(p, ioprio);
	108	if (ret)
	109	break;
2d70b68d	110	} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	111	break;
	112	case IOPRIO_WHO_USER:
	113	if (!who)
86a264ab	114	user = current_user();
22e2c507 JA	115	else
	116	user = find_user(who);
	117
	118	if (!user)
	119	break;
	120
	121	do_each_thread(g, p) {
d69b78ba	122	if (__task_cred(p)->uid != who)
22e2c507 JA	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
d69b78ba	136	rcu_read_unlock();
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;
fd0928df JA	147	ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
	148	if (p->io_context)
	149	ret = p->io_context->ioprio;
a1836a42 DQ	150	out:
	151	return ret;
	152	}
	153
e014ff8d ON	154	int ioprio_best(unsigned short aprio, unsigned short bprio)
	155	{
	156	unsigned short aclass = IOPRIO_PRIO_CLASS(aprio);
	157	unsigned short bclass = IOPRIO_PRIO_CLASS(bprio);
	158
e014ff8d ON	159	if (aclass == IOPRIO_CLASS_NONE)
	160	aclass = IOPRIO_CLASS_BE;
	161	if (bclass == IOPRIO_CLASS_NONE)
	162	bclass = IOPRIO_CLASS_BE;
	163
	164	if (aclass == bclass)
	165	return min(aprio, bprio);
	166	if (aclass > bclass)
	167	return bprio;
	168	else
	169	return aprio;
	170	}
	171
938bb9f5	172	SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
22e2c507 JA	173	{
	174	struct task_struct g, p;
	175	struct user_struct *user;
41487c65	176	struct pid *pgrp;
22e2c507	177	int ret = -ESRCH;
a1836a42	178	int tmpio;
22e2c507	179
d69b78ba	180	rcu_read_lock();
22e2c507 JA	181	switch (which) {
	182	case IOPRIO_WHO_PROCESS:
	183	if (!who)
	184	p = current;
	185	else
228ebcbe	186	p = find_task_by_vpid(who);
22e2c507	187	if (p)
a1836a42	188	ret = get_task_ioprio(p);
22e2c507 JA	189	break;
	190	case IOPRIO_WHO_PGRP:
	191	if (!who)
41487c65 EB	192	pgrp = task_pgrp(current);
41487c65 EB	193	else
b488893a	194	pgrp = find_vpid(who);
2d70b68d	195	do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
a1836a42 DQ	196	tmpio = get_task_ioprio(p);
	197	if (tmpio < 0)
	198	continue;
22e2c507	199	if (ret == -ESRCH)
a1836a42	200	ret = tmpio;
22e2c507	201	else
a1836a42	202	ret = ioprio_best(ret, tmpio);
2d70b68d	203	} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	204	break;
	205	case IOPRIO_WHO_USER:
	206	if (!who)
86a264ab	207	user = current_user();
22e2c507 JA	208	else
	209	user = find_user(who);
	210
	211	if (!user)
	212	break;
	213
	214	do_each_thread(g, p) {
d69b78ba	215	if (__task_cred(p)->uid != user->uid)
22e2c507	216	continue;
a1836a42 DQ	217	tmpio = get_task_ioprio(p);
	218	if (tmpio < 0)
	219	continue;
22e2c507	220	if (ret == -ESRCH)
a1836a42	221	ret = tmpio;
22e2c507	222	else
a1836a42	223	ret = ioprio_best(ret, tmpio);
22e2c507 JA	224	} while_each_thread(g, p);
	225
	226	if (who)
	227	free_uid(user);
	228	break;
	229	default:
	230	ret = -EINVAL;
	231	}
	232
d69b78ba	233	rcu_read_unlock();
22e2c507 JA	234	return ret;
22e2c507 JA	235	}