[mirror_zfs.git] / module / zfs / vdev_file.c

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
 */

#include <sys/zfs_context.h>
#include <sys/spa.h>
#include <sys/spa_impl.h>
#include <sys/vdev_file.h>
#include <sys/vdev_impl.h>
#include <sys/vdev_trim.h>
#include <sys/zio.h>
#include <sys/fs/zfs.h>
#include <sys/fm/fs/zfs.h>
#include <sys/abd.h>
#include <sys/fcntl.h>
#include <sys/vnode.h>

/*
 * Virtual device vector for files.
 */

static taskq_t *vdev_file_taskq;

static void
vdev_file_hold(vdev_t *vd)
{
	ASSERT(vd->vdev_path != NULL);
}

static void
vdev_file_rele(vdev_t *vd)
{
	ASSERT(vd->vdev_path != NULL);
}

static int
vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
    uint64_t *ashift)
{
	vdev_file_t *vf;
	vnode_t *vp;
	vattr_t vattr;
	int error;

	/*
	 * Rotational optimizations only make sense on block devices.
	 */
	vd->vdev_nonrot = B_TRUE;

	/*
	 * Allow TRIM on file based vdevs.  This may not always be supported,
	 * since it depends on your kernel version and underlying filesystem
	 * type but it is always safe to attempt.
	 */
	vd->vdev_has_trim = B_TRUE;

	/*
	 * Disable secure TRIM on file based vdevs.  There is no way to
	 * request this behavior from the underlying filesystem.
	 */
	vd->vdev_has_securetrim = B_FALSE;

	/*
	 * We must have a pathname, and it must be absolute.
	 */
	if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
		vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
		return (SET_ERROR(EINVAL));
	}

	/*
	 * Reopen the device if it's not currently open.  Otherwise,
	 * just update the physical size of the device.
	 */
	if (vd->vdev_tsd != NULL) {
		ASSERT(vd->vdev_reopening);
		vf = vd->vdev_tsd;
		goto skip_open;
	}

	vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);

	/*
	 * We always open the files from the root of the global zone, even if
	 * we're in a local zone.  If the user has gotten to this point, the
	 * administrator has already decided that the pool should be available
	 * to local zone users, so the underlying devices should be as well.
	 */
	ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
	error = vn_openat(vd->vdev_path + 1, UIO_SYSSPACE,
	    spa_mode(vd->vdev_spa) | FOFFMAX, 0, &vp, 0, 0, rootdir, -1);

	if (error) {
		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
		return (error);
	}

	vf->vf_vnode = vp;

#ifdef _KERNEL
	/*
	 * Make sure it's a regular file.
	 */
	if (vp->v_type != VREG) {
		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
		return (SET_ERROR(ENODEV));
	}
#endif

skip_open:
	/*
	 * Determine the physical size of the file.
	 */
	vattr.va_mask = AT_SIZE;
	error = VOP_GETATTR(vf->vf_vnode, &vattr, 0, kcred, NULL);
	if (error) {
		vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
		return (error);
	}

	*max_psize = *psize = vattr.va_size;
	*ashift = SPA_MINBLOCKSHIFT;

	return (0);
}

static void
vdev_file_close(vdev_t *vd)
{
	vdev_file_t *vf = vd->vdev_tsd;

	if (vd->vdev_reopening || vf == NULL)
		return;

	if (vf->vf_vnode != NULL) {
		(void) VOP_PUTPAGE(vf->vf_vnode, 0, 0, B_INVAL, kcred, NULL);
		(void) VOP_CLOSE(vf->vf_vnode, spa_mode(vd->vdev_spa), 1, 0,
		    kcred, NULL);
	}

	vd->vdev_delayed_close = B_FALSE;
	kmem_free(vf, sizeof (vdev_file_t));
	vd->vdev_tsd = NULL;
}

static void
vdev_file_io_strategy(void *arg)
{
	zio_t *zio = (zio_t *)arg;
	vdev_t *vd = zio->io_vd;
	vdev_file_t *vf = vd->vdev_tsd;
	ssize_t resid;
	void *buf;

	if (zio->io_type == ZIO_TYPE_READ)
		buf = abd_borrow_buf(zio->io_abd, zio->io_size);
	else
		buf = abd_borrow_buf_copy(zio->io_abd, zio->io_size);

	zio->io_error = vn_rdwr(zio->io_type == ZIO_TYPE_READ ?
	    UIO_READ : UIO_WRITE, vf->vf_vnode, buf, zio->io_size,
	    zio->io_offset, UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred, &resid);

	if (zio->io_type == ZIO_TYPE_READ)
		abd_return_buf_copy(zio->io_abd, buf, zio->io_size);
	else
		abd_return_buf(zio->io_abd, buf, zio->io_size);

	if (resid != 0 && zio->io_error == 0)
		zio->io_error = SET_ERROR(ENOSPC);

	zio_delay_interrupt(zio);
}

static void
vdev_file_io_fsync(void *arg)
{
	zio_t *zio = (zio_t *)arg;
	vdev_file_t *vf = zio->io_vd->vdev_tsd;

	zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC, kcred, NULL);

	zio_interrupt(zio);
}

static void
vdev_file_io_start(zio_t *zio)
{
	vdev_t *vd = zio->io_vd;
	vdev_file_t *vf = vd->vdev_tsd;

	if (zio->io_type == ZIO_TYPE_IOCTL) {
		/* XXPOLICY */
		if (!vdev_readable(vd)) {
			zio->io_error = SET_ERROR(ENXIO);
			zio_interrupt(zio);
			return;
		}

		switch (zio->io_cmd) {
		case DKIOCFLUSHWRITECACHE:

			if (zfs_nocacheflush)
				break;

			/*
			 * We cannot safely call vfs_fsync() when PF_FSTRANS
			 * is set in the current context.  Filesystems like
			 * XFS include sanity checks to verify it is not
			 * already set, see xfs_vm_writepage().  Therefore
			 * the sync must be dispatched to a different context.
			 */
			if (__spl_pf_fstrans_check()) {
				VERIFY3U(taskq_dispatch(vdev_file_taskq,
				    vdev_file_io_fsync, zio, TQ_SLEEP), !=,
				    TASKQID_INVALID);
				return;
			}

			zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
			    kcred, NULL);
			break;
		default:
			zio->io_error = SET_ERROR(ENOTSUP);
		}

		zio_execute(zio);
		return;
	} else if (zio->io_type == ZIO_TYPE_TRIM) {
		struct flock flck;

		ASSERT3U(zio->io_size, !=, 0);
		bzero(&flck, sizeof (flck));
		flck.l_type = F_FREESP;
		flck.l_start = zio->io_offset;
		flck.l_len = zio->io_size;
		flck.l_whence = SEEK_SET;

		zio->io_error = VOP_SPACE(vf->vf_vnode, F_FREESP, &flck,
		    0, 0, kcred, NULL);

		zio_execute(zio);
		return;
	}

	zio->io_target_timestamp = zio_handle_io_delay(zio);

	VERIFY3U(taskq_dispatch(vdev_file_taskq, vdev_file_io_strategy, zio,
	    TQ_SLEEP), !=, TASKQID_INVALID);
}

/* ARGSUSED */
static void
vdev_file_io_done(zio_t *zio)
{
}

vdev_ops_t vdev_file_ops = {
	vdev_file_open,
	vdev_file_close,
	vdev_default_asize,
	vdev_file_io_start,
	vdev_file_io_done,
	NULL,
	NULL,
	vdev_file_hold,
	vdev_file_rele,
	NULL,
	vdev_default_xlate,
	VDEV_TYPE_FILE,		/* name of this vdev type */
	B_TRUE			/* leaf vdev */
};

void
vdev_file_init(void)
{
	vdev_file_taskq = taskq_create("z_vdev_file", MAX(boot_ncpus, 16),
	    minclsyspri, boot_ncpus, INT_MAX, TASKQ_DYNAMIC);

	VERIFY(vdev_file_taskq);
}

void
vdev_file_fini(void)
{
	taskq_destroy(vdev_file_taskq);
}

/*
 * From userland we access disks just like files.
 */
#ifndef _KERNEL

vdev_ops_t vdev_disk_ops = {
	vdev_file_open,
	vdev_file_close,
	vdev_default_asize,
	vdev_file_io_start,
	vdev_file_io_done,
	NULL,
	NULL,
	vdev_file_hold,
	vdev_file_rele,
	NULL,
	vdev_default_xlate,
	VDEV_TYPE_DISK,		/* name of this vdev type */
	B_TRUE			/* leaf vdev */
};

#endif
Commit	Line	Data
34dc7c2f BB	1	/*
	2	* CDDL HEADER START
	3	*
	4	* The contents of this file are subject to the terms of the
	5	* Common Development and Distribution License (the "License").
	6	* You may not use this file except in compliance with the License.
	7	*
	8	* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
	9	* or http://www.opensolaris.org/os/licensing.
	10	* See the License for the specific language governing permissions
	11	* and limitations under the License.
	12	*
	13	* When distributing Covered Code, include this CDDL HEADER in each
	14	* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
	15	* If applicable, add the following below this CDDL HEADER, with the
	16	* fields enclosed by brackets "[]" replaced with your own identifying
	17	* information: Portions Copyright [yyyy] [name of copyright owner]
	18	*
	19	* CDDL HEADER END
	20	*/
	21	/*
428870ff	22	* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
619f0976	23	* Copyright (c) 2011, 2016 by Delphix. All rights reserved.
34dc7c2f BB	24	*/
34dc7c2f BB	25
34dc7c2f BB	26	#include <sys/zfs_context.h>
34dc7c2f BB	27	#include <sys/spa.h>
5853fe79	28	#include <sys/spa_impl.h>
34dc7c2f BB	29	#include <sys/vdev_file.h>
34dc7c2f BB	30	#include <sys/vdev_impl.h>
1b939560	31	#include <sys/vdev_trim.h>
34dc7c2f BB	32	#include <sys/zio.h>
34dc7c2f BB	33	#include <sys/fs/zfs.h>
b128c09f	34	#include <sys/fm/fs/zfs.h>
a6255b7f	35	#include <sys/abd.h>
1b939560 BB	36	#include <sys/fcntl.h>
1b939560 BB	37	#include <sys/vnode.h>
34dc7c2f BB	38
	39	/*
	40	* Virtual device vector for files.
	41	*/
	42
da8f51e1 CC	43	static taskq_t *vdev_file_taskq;
da8f51e1 CC	44
428870ff BB	45	static void
	46	vdev_file_hold(vdev_t *vd)
	47	{
	48	ASSERT(vd->vdev_path != NULL);
	49	}
	50
	51	static void
	52	vdev_file_rele(vdev_t *vd)
	53	{
	54	ASSERT(vd->vdev_path != NULL);
	55	}
	56
34dc7c2f	57	static int
1bd201e7 CS	58	vdev_file_open(vdev_t vd, uint64_t psize, uint64_t *max_psize,
1bd201e7 CS	59	uint64_t *ashift)
34dc7c2f BB	60	{
	61	vdev_file_t *vf;
	62	vnode_t *vp;
b128c09f	63	vattr_t vattr;
34dc7c2f BB	64	int error;
34dc7c2f BB	65
1b939560 BB	66	/*
	67	* Rotational optimizations only make sense on block devices.
	68	*/
fb40095f RY	69	vd->vdev_nonrot = B_TRUE;
fb40095f RY	70
1b939560 BB	71	/*
	72	* Allow TRIM on file based vdevs. This may not always be supported,
	73	* since it depends on your kernel version and underlying filesystem
	74	* type but it is always safe to attempt.
	75	*/
	76	vd->vdev_has_trim = B_TRUE;
	77
	78	/*
	79	* Disable secure TRIM on file based vdevs. There is no way to
	80	* request this behavior from the underlying filesystem.
	81	*/
	82	vd->vdev_has_securetrim = B_FALSE;
	83
34dc7c2f BB	84	/*
	85	* We must have a pathname, and it must be absolute.
	86	*/
	87	if (vd->vdev_path == NULL \|\| vd->vdev_path[0] != '/') {
	88	vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
2e528b49	89	return (SET_ERROR(EINVAL));
34dc7c2f BB	90	}
34dc7c2f BB	91
428870ff BB	92	/*
	93	* Reopen the device if it's not currently open. Otherwise,
	94	* just update the physical size of the device.
	95	*/
	96	if (vd->vdev_tsd != NULL) {
	97	ASSERT(vd->vdev_reopening);
	98	vf = vd->vdev_tsd;
	99	goto skip_open;
	100	}
	101
79c76d5b	102	vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);
34dc7c2f BB	103
	104	/*
	105	* We always open the files from the root of the global zone, even if
	106	* we're in a local zone. If the user has gotten to this point, the
	107	* administrator has already decided that the pool should be available
	108	* to local zone users, so the underlying devices should be as well.
	109	*/
	110	ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
	111	error = vn_openat(vd->vdev_path + 1, UIO_SYSSPACE,
fb5f0bc8	112	spa_mode(vd->vdev_spa) \| FOFFMAX, 0, &vp, 0, 0, rootdir, -1);
34dc7c2f BB	113
	114	if (error) {
	115	vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
	116	return (error);
	117	}
	118
	119	vf->vf_vnode = vp;
	120
	121	#ifdef _KERNEL
	122	/*
	123	* Make sure it's a regular file.
	124	*/
	125	if (vp->v_type != VREG) {
	126	vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
2e528b49	127	return (SET_ERROR(ENODEV));
34dc7c2f BB	128	}
34dc7c2f BB	129	#endif
428870ff BB	130
428870ff BB	131	skip_open:
34dc7c2f BB	132	/*
	133	* Determine the physical size of the file.
	134	*/
	135	vattr.va_mask = AT_SIZE;
	136	error = VOP_GETATTR(vf->vf_vnode, &vattr, 0, kcred, NULL);
	137	if (error) {
	138	vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
	139	return (error);
	140	}
	141
1bd201e7	142	max_psize = psize = vattr.va_size;
34dc7c2f BB	143	*ashift = SPA_MINBLOCKSHIFT;
	144
	145	return (0);
	146	}
	147
	148	static void
	149	vdev_file_close(vdev_t *vd)
	150	{
	151	vdev_file_t *vf = vd->vdev_tsd;
	152
428870ff	153	if (vd->vdev_reopening \|\| vf == NULL)
34dc7c2f BB	154	return;
	155
	156	if (vf->vf_vnode != NULL) {
	157	(void) VOP_PUTPAGE(vf->vf_vnode, 0, 0, B_INVAL, kcred, NULL);
fb5f0bc8 BB	158	(void) VOP_CLOSE(vf->vf_vnode, spa_mode(vd->vdev_spa), 1, 0,
fb5f0bc8 BB	159	kcred, NULL);
34dc7c2f BB	160	}
34dc7c2f BB	161
428870ff	162	vd->vdev_delayed_close = B_FALSE;
34dc7c2f BB	163	kmem_free(vf, sizeof (vdev_file_t));
	164	vd->vdev_tsd = NULL;
	165	}
	166
5853fe79 GW	167	static void
5853fe79 GW	168	vdev_file_io_strategy(void *arg)
34dc7c2f	169	{
5853fe79	170	zio_t zio = (zio_t )arg;
34dc7c2f	171	vdev_t *vd = zio->io_vd;
5853fe79 GW	172	vdev_file_t *vf = vd->vdev_tsd;
5853fe79 GW	173	ssize_t resid;
a6255b7f DQ	174	void *buf;
	175
	176	if (zio->io_type == ZIO_TYPE_READ)
	177	buf = abd_borrow_buf(zio->io_abd, zio->io_size);
	178	else
	179	buf = abd_borrow_buf_copy(zio->io_abd, zio->io_size);
34dc7c2f	180
5853fe79	181	zio->io_error = vn_rdwr(zio->io_type == ZIO_TYPE_READ ?
a6255b7f DQ	182	UIO_READ : UIO_WRITE, vf->vf_vnode, buf, zio->io_size,
	183	zio->io_offset, UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred, &resid);
	184
	185	if (zio->io_type == ZIO_TYPE_READ)
	186	abd_return_buf_copy(zio->io_abd, buf, zio->io_size);
	187	else
	188	abd_return_buf(zio->io_abd, buf, zio->io_size);
5853fe79 GW	189
5853fe79 GW	190	if (resid != 0 && zio->io_error == 0)
2e528b49	191	zio->io_error = SET_ERROR(ENOSPC);
3adfc400	192
26ef0cc7	193	zio_delay_interrupt(zio);
5853fe79 GW	194	}
5853fe79 GW	195
92119cc2 BB	196	static void
	197	vdev_file_io_fsync(void *arg)
	198	{
	199	zio_t zio = (zio_t )arg;
	200	vdev_file_t *vf = zio->io_vd->vdev_tsd;
	201
	202	zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC \| FDSYNC, kcred, NULL);
	203
	204	zio_interrupt(zio);
	205	}
	206
98b25418	207	static void
5853fe79 GW	208	vdev_file_io_start(zio_t *zio)
5853fe79 GW	209	{
5853fe79 GW	210	vdev_t *vd = zio->io_vd;
5853fe79 GW	211	vdev_file_t *vf = vd->vdev_tsd;
34dc7c2f	212
3adfc400	213	if (zio->io_type == ZIO_TYPE_IOCTL) {
5853fe79 GW	214	/* XXPOLICY */
5853fe79 GW	215	if (!vdev_readable(vd)) {
2e528b49	216	zio->io_error = SET_ERROR(ENXIO);
98b25418 GW	217	zio_interrupt(zio);
98b25418 GW	218	return;
5853fe79 GW	219	}
5853fe79 GW	220
34dc7c2f BB	221	switch (zio->io_cmd) {
34dc7c2f BB	222	case DKIOCFLUSHWRITECACHE:
f9a1ac4d H	223
	224	if (zfs_nocacheflush)
	225	break;
	226
92119cc2 BB	227	/*
	228	* We cannot safely call vfs_fsync() when PF_FSTRANS
	229	* is set in the current context. Filesystems like
	230	* XFS include sanity checks to verify it is not
	231	* already set, see xfs_vm_writepage(). Therefore
	232	* the sync must be dispatched to a different context.
	233	*/
e624cd19	234	if (__spl_pf_fstrans_check()) {
da8f51e1	235	VERIFY3U(taskq_dispatch(vdev_file_taskq,
48d3eb40 BB	236	vdev_file_io_fsync, zio, TQ_SLEEP), !=,
48d3eb40 BB	237	TASKQID_INVALID);
98b25418	238	return;
92119cc2 BB	239	}
92119cc2 BB	240
34dc7c2f BB	241	zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC \| FDSYNC,
34dc7c2f BB	242	kcred, NULL);
34dc7c2f BB	243	break;
34dc7c2f BB	244	default:
2e528b49	245	zio->io_error = SET_ERROR(ENOTSUP);
34dc7c2f BB	246	}
34dc7c2f BB	247
1b939560 BB	248	zio_execute(zio);
	249	return;
	250	} else if (zio->io_type == ZIO_TYPE_TRIM) {
	251	struct flock flck;
	252
	253	ASSERT3U(zio->io_size, !=, 0);
	254	bzero(&flck, sizeof (flck));
	255	flck.l_type = F_FREESP;
	256	flck.l_start = zio->io_offset;
	257	flck.l_len = zio->io_size;
126d0fa7	258	flck.l_whence = SEEK_SET;
1b939560 BB	259
	260	zio->io_error = VOP_SPACE(vf->vf_vnode, F_FREESP, &flck,
	261	0, 0, kcred, NULL);
	262
98b25418 GW	263	zio_execute(zio);
98b25418 GW	264	return;
34dc7c2f BB	265	}
34dc7c2f BB	266
26ef0cc7 TH	267	zio->io_target_timestamp = zio_handle_io_delay(zio);
26ef0cc7 TH	268
da8f51e1	269	VERIFY3U(taskq_dispatch(vdev_file_taskq, vdev_file_io_strategy, zio,
48d3eb40	270	TQ_SLEEP), !=, TASKQID_INVALID);
34dc7c2f BB	271	}
34dc7c2f BB	272
b128c09f BB	273	/* ARGSUSED */
b128c09f BB	274	static void
34dc7c2f BB	275	vdev_file_io_done(zio_t *zio)
34dc7c2f BB	276	{
34dc7c2f BB	277	}
	278
	279	vdev_ops_t vdev_file_ops = {
	280	vdev_file_open,
	281	vdev_file_close,
34dc7c2f BB	282	vdev_default_asize,
	283	vdev_file_io_start,
	284	vdev_file_io_done,
	285	NULL,
3d6da72d	286	NULL,
428870ff BB	287	vdev_file_hold,
428870ff BB	288	vdev_file_rele,
a1d477c2	289	NULL,
619f0976	290	vdev_default_xlate,
34dc7c2f BB	291	VDEV_TYPE_FILE, /* name of this vdev type */
	292	B_TRUE /* leaf vdev */
	293	};
	294
da8f51e1 CC	295	void
	296	vdev_file_init(void)
	297	{
	298	vdev_file_taskq = taskq_create("z_vdev_file", MAX(boot_ncpus, 16),
	299	minclsyspri, boot_ncpus, INT_MAX, TASKQ_DYNAMIC);
	300
	301	VERIFY(vdev_file_taskq);
	302	}
	303
	304	void
	305	vdev_file_fini(void)
	306	{
	307	taskq_destroy(vdev_file_taskq);
	308	}
	309
34dc7c2f BB	310	/*
	311	* From userland we access disks just like files.
	312	*/
	313	#ifndef _KERNEL
	314
	315	vdev_ops_t vdev_disk_ops = {
	316	vdev_file_open,
	317	vdev_file_close,
34dc7c2f BB	318	vdev_default_asize,
	319	vdev_file_io_start,
	320	vdev_file_io_done,
	321	NULL,
3d6da72d	322	NULL,
428870ff BB	323	vdev_file_hold,
428870ff BB	324	vdev_file_rele,
a1d477c2	325	NULL,
619f0976	326	vdev_default_xlate,
34dc7c2f BB	327	VDEV_TYPE_DISK, /* name of this vdev type */
	328	B_TRUE /* leaf vdev */
	329	};
	330
	331	#endif