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