[mirror_ubuntu-zesty-kernel.git] / block / 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 (!uid_eq(tcred->uid, cred->euid) &&
	    !uid_eq(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 = 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;
	kuid_t uid;
	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:
			uid = make_kuid(current_user_ns(), who);
			if (!uid_valid(uid))
				break;
			if (!who)
				user = current_user();
			else
				user = find_user(uid);

			if (!user)
				break;

			do_each_thread(g, p) {
				if (!uid_eq(task_uid(p), uid) ||
				    !task_pid_vnr(p))
					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);
	task_lock(p);
	if (p->io_context)
		ret = p->io_context->ioprio;
	task_unlock(p);
out:
	return ret;
}

int ioprio_best(unsigned short aprio, unsigned short bprio)
{
	unsigned short aclass;
	unsigned short bclass;

	if (!ioprio_valid(aprio))
		aprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM);
	if (!ioprio_valid(bprio))
		bprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM);

	aclass = IOPRIO_PRIO_CLASS(aprio);
	bclass = IOPRIO_PRIO_CLASS(bprio);
	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;
	kuid_t uid;
	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:
			uid = make_kuid(current_user_ns(), who);
			if (!who)
				user = current_user();
			else
				user = find_user(uid);

			if (!user)
				break;

			do_each_thread(g, p) {
				if (!uid_eq(task_uid(p), user->uid) ||
				    !task_pid_vnr(p))
					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();
c69e8d9c DH	39	tcred = __task_cred(task);
8e96e3b7 EB	40	if (!uid_eq(tcred->uid, cred->euid) &&
8e96e3b7 EB	41	!uid_eq(tcred->uid, cred->uid) && !capable(CAP_SYS_NICE)) {
c69e8d9c	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) {
2b566fa5	53	ioc->ioprio = 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;
7b44ab97	68	kuid_t uid;
22e2c507 JA	69	int ret;
	70
	71	switch (class) {
	72	case IOPRIO_CLASS_RT:
	73	if (!capable(CAP_SYS_ADMIN))
	74	return -EPERM;
	75	/* fall through, rt has prio field too */
	76	case IOPRIO_CLASS_BE:
	77	if (data >= IOPRIO_BE_NR \|\| data < 0)
	78	return -EINVAL;
	79
	80	break;
	81	case IOPRIO_CLASS_IDLE:
	82	break;
8ec680e4 JA	83	case IOPRIO_CLASS_NONE:
	84	if (data)
	85	return -EINVAL;
	86	break;
22e2c507 JA	87	default:
	88	return -EINVAL;
	89	}
	90
	91	ret = -ESRCH;
d69b78ba	92	rcu_read_lock();
22e2c507 JA	93	switch (which) {
	94	case IOPRIO_WHO_PROCESS:
	95	if (!who)
	96	p = current;
	97	else
228ebcbe	98	p = find_task_by_vpid(who);
22e2c507 JA	99	if (p)
	100	ret = set_task_ioprio(p, ioprio);
	101	break;
	102	case IOPRIO_WHO_PGRP:
	103	if (!who)
41487c65 EB	104	pgrp = task_pgrp(current);
41487c65 EB	105	else
b488893a	106	pgrp = find_vpid(who);
2d70b68d	107	do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
22e2c507 JA	108	ret = set_task_ioprio(p, ioprio);
	109	if (ret)
	110	break;
2d70b68d	111	} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	112	break;
22e2c507 JA	113	case IOPRIO_WHO_USER:
7b44ab97 EB	114	uid = make_kuid(current_user_ns(), who);
	115	if (!uid_valid(uid))
	116	break;
22e2c507	117	if (!who)
86a264ab	118	user = current_user();
22e2c507	119	else
7b44ab97	120	user = find_user(uid);
22e2c507 JA	121
	122	if (!user)
	123	break;
	124
	125	do_each_thread(g, p) {
8639b461 BS	126	if (!uid_eq(task_uid(p), uid) \|\|
8639b461 BS	127	!task_pid_vnr(p))
22e2c507 JA	128	continue;
	129	ret = set_task_ioprio(p, ioprio);
	130	if (ret)
78bd4d48	131	goto free_uid;
22e2c507	132	} while_each_thread(g, p);
78bd4d48	133	free_uid:
22e2c507 JA	134	if (who)
	135	free_uid(user);
	136	break;
	137	default:
	138	ret = -EINVAL;
	139	}
	140
d69b78ba	141	rcu_read_unlock();
22e2c507 JA	142	return ret;
	143	}
	144
a1836a42 DQ	145	static int get_task_ioprio(struct task_struct *p)
	146	{
	147	int ret;
	148
	149	ret = security_task_getioprio(p);
	150	if (ret)
	151	goto out;
fd0928df	152	ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
8ba86821	153	task_lock(p);
fd0928df JA	154	if (p->io_context)
fd0928df JA	155	ret = p->io_context->ioprio;
8ba86821	156	task_unlock(p);
a1836a42 DQ	157	out:
	158	return ret;
	159	}
	160
e014ff8d ON	161	int ioprio_best(unsigned short aprio, unsigned short bprio)
e014ff8d ON	162	{
ece9c72a JK	163	unsigned short aclass;
ece9c72a JK	164	unsigned short bclass;
e014ff8d	165
ece9c72a JK	166	if (!ioprio_valid(aprio))
	167	aprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM);
	168	if (!ioprio_valid(bprio))
	169	bprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM);
e014ff8d	170
ece9c72a JK	171	aclass = IOPRIO_PRIO_CLASS(aprio);
ece9c72a JK	172	bclass = IOPRIO_PRIO_CLASS(bprio);
e014ff8d ON	173	if (aclass == bclass)
	174	return min(aprio, bprio);
	175	if (aclass > bclass)
	176	return bprio;
	177	else
	178	return aprio;
	179	}
	180
938bb9f5	181	SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
22e2c507 JA	182	{
	183	struct task_struct g, p;
	184	struct user_struct *user;
41487c65	185	struct pid *pgrp;
7b44ab97	186	kuid_t uid;
22e2c507	187	int ret = -ESRCH;
a1836a42	188	int tmpio;
22e2c507	189
d69b78ba	190	rcu_read_lock();
22e2c507 JA	191	switch (which) {
	192	case IOPRIO_WHO_PROCESS:
	193	if (!who)
	194	p = current;
	195	else
228ebcbe	196	p = find_task_by_vpid(who);
22e2c507	197	if (p)
a1836a42	198	ret = get_task_ioprio(p);
22e2c507 JA	199	break;
	200	case IOPRIO_WHO_PGRP:
	201	if (!who)
41487c65 EB	202	pgrp = task_pgrp(current);
41487c65 EB	203	else
b488893a	204	pgrp = find_vpid(who);
2d70b68d	205	do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
a1836a42 DQ	206	tmpio = get_task_ioprio(p);
	207	if (tmpio < 0)
	208	continue;
22e2c507	209	if (ret == -ESRCH)
a1836a42	210	ret = tmpio;
22e2c507	211	else
a1836a42	212	ret = ioprio_best(ret, tmpio);
2d70b68d	213	} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	214	break;
22e2c507 JA	215	case IOPRIO_WHO_USER:
7b44ab97	216	uid = make_kuid(current_user_ns(), who);
22e2c507	217	if (!who)
86a264ab	218	user = current_user();
22e2c507	219	else
7b44ab97	220	user = find_user(uid);
22e2c507 JA	221
	222	if (!user)
	223	break;
	224
	225	do_each_thread(g, p) {
8639b461 BS	226	if (!uid_eq(task_uid(p), user->uid) \|\|
8639b461 BS	227	!task_pid_vnr(p))
22e2c507	228	continue;
a1836a42 DQ	229	tmpio = get_task_ioprio(p);
	230	if (tmpio < 0)
	231	continue;
22e2c507	232	if (ret == -ESRCH)
a1836a42	233	ret = tmpio;
22e2c507	234	else
a1836a42	235	ret = ioprio_best(ret, tmpio);
22e2c507 JA	236	} while_each_thread(g, p);
	237
	238	if (who)
	239	free_uid(user);
	240	break;
	241	default:
	242	ret = -EINVAL;
	243	}
	244
d69b78ba	245	rcu_read_unlock();
22e2c507 JA	246	return ret;
22e2c507 JA	247	}