[mirror_ubuntu-bionic-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/cred.h>
#include <linux/blkdev.h>
#include <linux/capability.h>
#include <linux/sched/user.h>
#include <linux/sched/task.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;

			for_each_process_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;
			}
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;

			for_each_process_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);
			}

			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	25	#include <linux/ioprio.h>
5b825c3a	26	#include <linux/cred.h>
22e2c507	27	#include <linux/blkdev.h>
16f7e0fe	28	#include <linux/capability.h>
8703e8a4	29	#include <linux/sched/user.h>
f719ff9b	30	#include <linux/sched/task.h>
9abdc4cd	31	#include <linux/syscalls.h>
03e68060	32	#include <linux/security.h>
b488893a	33	#include <linux/pid_namespace.h>
22e2c507	34
b3881f74	35	int set_task_ioprio(struct task_struct *task, int ioprio)
22e2c507	36	{
03e68060	37	int err;
22e2c507	38	struct io_context *ioc;
c69e8d9c	39	const struct cred cred = current_cred(), tcred;
22e2c507	40
c69e8d9c DH	41	rcu_read_lock();
c69e8d9c DH	42	tcred = __task_cred(task);
8e96e3b7 EB	43	if (!uid_eq(tcred->uid, cred->euid) &&
8e96e3b7 EB	44	!uid_eq(tcred->uid, cred->uid) && !capable(CAP_SYS_NICE)) {
c69e8d9c	45	rcu_read_unlock();
22e2c507	46	return -EPERM;
c69e8d9c DH	47	}
c69e8d9c DH	48	rcu_read_unlock();
22e2c507	49
03e68060 JM	50	err = security_task_setioprio(task, ioprio);
	51	if (err)
	52	return err;
	53
6e736be7 TH	54	ioc = get_task_io_context(task, GFP_ATOMIC, NUMA_NO_NODE);
6e736be7 TH	55	if (ioc) {
2b566fa5	56	ioc->ioprio = ioprio;
11a3122f	57	put_io_context(ioc);
fd0928df	58	}
22e2c507	59
fd0928df	60	return err;
22e2c507	61	}
b3881f74	62	EXPORT_SYMBOL_GPL(set_task_ioprio);
22e2c507	63
938bb9f5	64	SYSCALL_DEFINE3(ioprio_set, int, which, int, who, int, ioprio)
22e2c507 JA	65	{
	66	int class = IOPRIO_PRIO_CLASS(ioprio);
	67	int data = IOPRIO_PRIO_DATA(ioprio);
	68	struct task_struct p, g;
	69	struct user_struct *user;
41487c65	70	struct pid *pgrp;
7b44ab97	71	kuid_t uid;
22e2c507 JA	72	int ret;
	73
	74	switch (class) {
	75	case IOPRIO_CLASS_RT:
	76	if (!capable(CAP_SYS_ADMIN))
	77	return -EPERM;
	78	/* fall through, rt has prio field too */
	79	case IOPRIO_CLASS_BE:
	80	if (data >= IOPRIO_BE_NR \|\| data < 0)
	81	return -EINVAL;
	82
	83	break;
	84	case IOPRIO_CLASS_IDLE:
	85	break;
8ec680e4 JA	86	case IOPRIO_CLASS_NONE:
	87	if (data)
	88	return -EINVAL;
	89	break;
22e2c507 JA	90	default:
	91	return -EINVAL;
	92	}
	93
	94	ret = -ESRCH;
d69b78ba	95	rcu_read_lock();
22e2c507 JA	96	switch (which) {
	97	case IOPRIO_WHO_PROCESS:
	98	if (!who)
	99	p = current;
	100	else
228ebcbe	101	p = find_task_by_vpid(who);
22e2c507 JA	102	if (p)
	103	ret = set_task_ioprio(p, ioprio);
	104	break;
	105	case IOPRIO_WHO_PGRP:
	106	if (!who)
41487c65 EB	107	pgrp = task_pgrp(current);
41487c65 EB	108	else
b488893a	109	pgrp = find_vpid(who);
2d70b68d	110	do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
22e2c507 JA	111	ret = set_task_ioprio(p, ioprio);
	112	if (ret)
	113	break;
2d70b68d	114	} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	115	break;
22e2c507 JA	116	case IOPRIO_WHO_USER:
7b44ab97 EB	117	uid = make_kuid(current_user_ns(), who);
	118	if (!uid_valid(uid))
	119	break;
22e2c507	120	if (!who)
86a264ab	121	user = current_user();
22e2c507	122	else
7b44ab97	123	user = find_user(uid);
22e2c507 JA	124
	125	if (!user)
	126	break;
	127
612dafab	128	for_each_process_thread(g, p) {
8639b461 BS	129	if (!uid_eq(task_uid(p), uid) \|\|
8639b461 BS	130	!task_pid_vnr(p))
22e2c507 JA	131	continue;
	132	ret = set_task_ioprio(p, ioprio);
	133	if (ret)
78bd4d48	134	goto free_uid;
612dafab	135	}
78bd4d48	136	free_uid:
22e2c507 JA	137	if (who)
	138	free_uid(user);
	139	break;
	140	default:
	141	ret = -EINVAL;
	142	}
	143
d69b78ba	144	rcu_read_unlock();
22e2c507 JA	145	return ret;
	146	}
	147
a1836a42 DQ	148	static int get_task_ioprio(struct task_struct *p)
	149	{
	150	int ret;
	151
	152	ret = security_task_getioprio(p);
	153	if (ret)
	154	goto out;
fd0928df	155	ret = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_NONE, IOPRIO_NORM);
8ba86821	156	task_lock(p);
fd0928df JA	157	if (p->io_context)
fd0928df JA	158	ret = p->io_context->ioprio;
8ba86821	159	task_unlock(p);
a1836a42 DQ	160	out:
	161	return ret;
	162	}
	163
e014ff8d ON	164	int ioprio_best(unsigned short aprio, unsigned short bprio)
e014ff8d ON	165	{
ece9c72a JK	166	unsigned short aclass;
ece9c72a JK	167	unsigned short bclass;
e014ff8d	168
ece9c72a JK	169	if (!ioprio_valid(aprio))
	170	aprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM);
	171	if (!ioprio_valid(bprio))
	172	bprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, IOPRIO_NORM);
e014ff8d	173
ece9c72a JK	174	aclass = IOPRIO_PRIO_CLASS(aprio);
ece9c72a JK	175	bclass = IOPRIO_PRIO_CLASS(bprio);
e014ff8d ON	176	if (aclass == bclass)
	177	return min(aprio, bprio);
	178	if (aclass > bclass)
	179	return bprio;
	180	else
	181	return aprio;
	182	}
	183
938bb9f5	184	SYSCALL_DEFINE2(ioprio_get, int, which, int, who)
22e2c507 JA	185	{
	186	struct task_struct g, p;
	187	struct user_struct *user;
41487c65	188	struct pid *pgrp;
7b44ab97	189	kuid_t uid;
22e2c507	190	int ret = -ESRCH;
a1836a42	191	int tmpio;
22e2c507	192
d69b78ba	193	rcu_read_lock();
22e2c507 JA	194	switch (which) {
	195	case IOPRIO_WHO_PROCESS:
	196	if (!who)
	197	p = current;
	198	else
228ebcbe	199	p = find_task_by_vpid(who);
22e2c507	200	if (p)
a1836a42	201	ret = get_task_ioprio(p);
22e2c507 JA	202	break;
	203	case IOPRIO_WHO_PGRP:
	204	if (!who)
41487c65 EB	205	pgrp = task_pgrp(current);
41487c65 EB	206	else
b488893a	207	pgrp = find_vpid(who);
2d70b68d	208	do_each_pid_thread(pgrp, PIDTYPE_PGID, p) {
a1836a42 DQ	209	tmpio = get_task_ioprio(p);
	210	if (tmpio < 0)
	211	continue;
22e2c507	212	if (ret == -ESRCH)
a1836a42	213	ret = tmpio;
22e2c507	214	else
a1836a42	215	ret = ioprio_best(ret, tmpio);
2d70b68d	216	} while_each_pid_thread(pgrp, PIDTYPE_PGID, p);
22e2c507 JA	217	break;
22e2c507 JA	218	case IOPRIO_WHO_USER:
7b44ab97	219	uid = make_kuid(current_user_ns(), who);
22e2c507	220	if (!who)
86a264ab	221	user = current_user();
22e2c507	222	else
7b44ab97	223	user = find_user(uid);
22e2c507 JA	224
	225	if (!user)
	226	break;
	227
612dafab	228	for_each_process_thread(g, p) {
8639b461 BS	229	if (!uid_eq(task_uid(p), user->uid) \|\|
8639b461 BS	230	!task_pid_vnr(p))
22e2c507	231	continue;
a1836a42 DQ	232	tmpio = get_task_ioprio(p);
	233	if (tmpio < 0)
	234	continue;
22e2c507	235	if (ret == -ESRCH)
a1836a42	236	ret = tmpio;
22e2c507	237	else
a1836a42	238	ret = ioprio_best(ret, tmpio);
612dafab	239	}
22e2c507 JA	240
	241	if (who)
	242	free_uid(user);
	243	break;
	244	default:
	245	ret = -EINVAL;
	246	}
	247
d69b78ba	248	rcu_read_unlock();
22e2c507 JA	249	return ret;
22e2c507 JA	250	}