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.
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.
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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
26 #include <sys/zfs_context.h>
28 #include <sys/vdev_file.h>
29 #include <sys/vdev_impl.h>
31 #include <sys/fs/zfs.h>
32 #include <sys/fm/fs/zfs.h>
36 * Virtual device vector for files.
39 static taskq_t *vdev_file_taskq;
44 vdev_file_taskq = taskq_create("z_vdev_file", MAX(max_ncpus, 16),
45 minclsyspri, max_ncpus, INT_MAX, 0);
51 taskq_destroy(vdev_file_taskq);
55 vdev_file_hold(vdev_t *vd)
57 ASSERT(vd->vdev_path != NULL);
61 vdev_file_rele(vdev_t *vd)
63 ASSERT(vd->vdev_path != NULL);
67 vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
68 uint64_t *logical_ashift, uint64_t *physical_ashift)
75 /* Rotational optimizations only make sense on block devices */
76 vd->vdev_nonrot = B_TRUE;
79 * We must have a pathname, and it must be absolute.
81 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
82 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
83 return (SET_ERROR(EINVAL));
87 * Reopen the device if it's not currently open. Otherwise,
88 * just update the physical size of the device.
90 if (vd->vdev_tsd != NULL) {
91 ASSERT(vd->vdev_reopening);
97 vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);
100 * We always open the files from the root of the global zone, even if
101 * we're in a local zone. If the user has gotten to this point, the
102 * administrator has already decided that the pool should be available
103 * to local zone users, so the underlying devices should be as well.
105 ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
106 error = vn_openat(vd->vdev_path + 1, UIO_SYSSPACE,
107 spa_mode(vd->vdev_spa) | FOFFMAX, 0, &vp, 0, 0, rootdir, -1);
110 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
111 kmem_free(vd->vdev_tsd, sizeof (vdev_file_t));
120 * Make sure it's a regular file.
122 if (vp->v_type != VREG) {
124 (void) VOP_CLOSE(vp, spa_mode(vd->vdev_spa), 1, 0, kcred, NULL);
126 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
128 kmem_free(vd->vdev_tsd, sizeof (vdev_file_t));
131 return (SET_ERROR(ENODEV));
137 * Determine the physical size of the file.
139 vattr.va_mask = AT_SIZE;
140 vn_lock(vp, LK_SHARED | LK_RETRY);
141 error = VOP_GETATTR(vp, &vattr, kcred);
144 (void) VOP_CLOSE(vp, spa_mode(vd->vdev_spa), 1, 0, kcred, NULL);
145 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
146 kmem_free(vd->vdev_tsd, sizeof (vdev_file_t));
151 vd->vdev_notrim = B_TRUE;
153 *max_psize = *psize = vattr.va_size;
154 *logical_ashift = SPA_MINBLOCKSHIFT;
155 *physical_ashift = SPA_MINBLOCKSHIFT;
161 vdev_file_close(vdev_t *vd)
163 vdev_file_t *vf = vd->vdev_tsd;
165 if (vd->vdev_reopening || vf == NULL)
168 if (vf->vf_vnode != NULL) {
169 (void) VOP_CLOSE(vf->vf_vnode, spa_mode(vd->vdev_spa), 1, 0,
173 vd->vdev_delayed_close = B_FALSE;
174 kmem_free(vf, sizeof (vdev_file_t));
179 * Implements the interrupt side for file vdev types. This routine will be
180 * called when the I/O completes allowing us to transfer the I/O to the
181 * interrupt taskqs. For consistency, the code structure mimics disk vdev
185 vdev_file_io_intr(zio_t *zio)
187 zio_delay_interrupt(zio);
191 vdev_file_io_strategy(void *arg)
194 vdev_t *vd = zio->io_vd;
203 ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
204 if (zio->io_type == ZIO_TYPE_READ) {
205 addr = abd_borrow_buf(zio->io_abd, zio->io_size);
207 addr = abd_borrow_buf_copy(zio->io_abd, zio->io_size);
210 zio->io_error = vn_rdwr(zio->io_type == ZIO_TYPE_READ ?
211 UIO_READ : UIO_WRITE, vp, addr, zio->io_size,
212 zio->io_offset, UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred, &resid);
214 if (zio->io_type == ZIO_TYPE_READ) {
215 abd_return_buf_copy(zio->io_abd, addr, zio->io_size);
217 abd_return_buf(zio->io_abd, addr, zio->io_size);
220 if (resid != 0 && zio->io_error == 0)
221 zio->io_error = ENOSPC;
223 vdev_file_io_intr(zio);
227 vdev_file_io_start(zio_t *zio)
229 vdev_t *vd = zio->io_vd;
230 vdev_file_t *vf = vd->vdev_tsd;
232 if (zio->io_type == ZIO_TYPE_IOCTL) {
234 if (!vdev_readable(vd)) {
235 zio->io_error = SET_ERROR(ENXIO);
240 switch (zio->io_cmd) {
241 case DKIOCFLUSHWRITECACHE:
243 if (zfs_nocacheflush)
246 zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
250 zio->io_error = SET_ERROR(ENOTSUP);
257 ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
258 zio->io_target_timestamp = zio_handle_io_delay(zio);
260 VERIFY3U(taskq_dispatch(vdev_file_taskq, vdev_file_io_strategy, zio,
266 vdev_file_io_done(zio_t *zio)
270 vdev_ops_t vdev_file_ops = {
282 VDEV_TYPE_FILE, /* name of this vdev type */
283 B_TRUE /* leaf vdev */
287 * From userland we access disks just like files.
291 vdev_ops_t vdev_disk_ops = {
303 VDEV_TYPE_DISK, /* name of this vdev type */
304 B_TRUE /* leaf vdev */