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, 2015 by Delphix. All rights reserved.
26 #include <sys/zfs_context.h>
28 #include <sys/spa_impl.h>
29 #include <sys/vdev_file.h>
30 #include <sys/vdev_impl.h>
32 #include <sys/fs/zfs.h>
33 #include <sys/fm/fs/zfs.h>
37 * Virtual device vector for files.
40 static taskq_t *vdev_file_taskq;
43 vdev_file_hold(vdev_t *vd)
45 ASSERT(vd->vdev_path != NULL);
49 vdev_file_rele(vdev_t *vd)
51 ASSERT(vd->vdev_path != NULL);
55 vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
63 /* Rotational optimizations only make sense on block devices */
64 vd->vdev_nonrot = B_TRUE;
67 * We must have a pathname, and it must be absolute.
69 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
70 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
71 return (SET_ERROR(EINVAL));
75 * Reopen the device if it's not currently open. Otherwise,
76 * just update the physical size of the device.
78 if (vd->vdev_tsd != NULL) {
79 ASSERT(vd->vdev_reopening);
84 vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);
87 * We always open the files from the root of the global zone, even if
88 * we're in a local zone. If the user has gotten to this point, the
89 * administrator has already decided that the pool should be available
90 * to local zone users, so the underlying devices should be as well.
92 ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
93 error = vn_openat(vd->vdev_path + 1, UIO_SYSSPACE,
94 spa_mode(vd->vdev_spa) | FOFFMAX, 0, &vp, 0, 0, rootdir, -1);
97 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
105 * Make sure it's a regular file.
107 if (vp->v_type != VREG) {
108 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
109 return (SET_ERROR(ENODEV));
115 * Determine the physical size of the file.
117 vattr.va_mask = AT_SIZE;
118 error = VOP_GETATTR(vf->vf_vnode, &vattr, 0, kcred, NULL);
120 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
124 *max_psize = *psize = vattr.va_size;
125 *ashift = SPA_MINBLOCKSHIFT;
131 vdev_file_close(vdev_t *vd)
133 vdev_file_t *vf = vd->vdev_tsd;
135 if (vd->vdev_reopening || vf == NULL)
138 if (vf->vf_vnode != NULL) {
139 (void) VOP_PUTPAGE(vf->vf_vnode, 0, 0, B_INVAL, kcred, NULL);
140 (void) VOP_CLOSE(vf->vf_vnode, spa_mode(vd->vdev_spa), 1, 0,
144 vd->vdev_delayed_close = B_FALSE;
145 kmem_free(vf, sizeof (vdev_file_t));
150 vdev_file_io_strategy(void *arg)
152 zio_t *zio = (zio_t *)arg;
153 vdev_t *vd = zio->io_vd;
154 vdev_file_t *vf = vd->vdev_tsd;
158 if (zio->io_type == ZIO_TYPE_READ)
159 buf = abd_borrow_buf(zio->io_abd, zio->io_size);
161 buf = abd_borrow_buf_copy(zio->io_abd, zio->io_size);
163 zio->io_error = vn_rdwr(zio->io_type == ZIO_TYPE_READ ?
164 UIO_READ : UIO_WRITE, vf->vf_vnode, buf, zio->io_size,
165 zio->io_offset, UIO_SYSSPACE, 0, RLIM64_INFINITY, kcred, &resid);
167 if (zio->io_type == ZIO_TYPE_READ)
168 abd_return_buf_copy(zio->io_abd, buf, zio->io_size);
170 abd_return_buf(zio->io_abd, buf, zio->io_size);
172 if (resid != 0 && zio->io_error == 0)
173 zio->io_error = SET_ERROR(ENOSPC);
175 zio_delay_interrupt(zio);
179 vdev_file_io_fsync(void *arg)
181 zio_t *zio = (zio_t *)arg;
182 vdev_file_t *vf = zio->io_vd->vdev_tsd;
184 zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC, kcred, NULL);
190 vdev_file_io_start(zio_t *zio)
192 vdev_t *vd = zio->io_vd;
193 vdev_file_t *vf = vd->vdev_tsd;
195 if (zio->io_type == ZIO_TYPE_IOCTL) {
197 if (!vdev_readable(vd)) {
198 zio->io_error = SET_ERROR(ENXIO);
203 switch (zio->io_cmd) {
204 case DKIOCFLUSHWRITECACHE:
206 if (zfs_nocacheflush)
210 * We cannot safely call vfs_fsync() when PF_FSTRANS
211 * is set in the current context. Filesystems like
212 * XFS include sanity checks to verify it is not
213 * already set, see xfs_vm_writepage(). Therefore
214 * the sync must be dispatched to a different context.
216 if (__spl_pf_fstrans_check()) {
217 VERIFY3U(taskq_dispatch(vdev_file_taskq,
218 vdev_file_io_fsync, zio, TQ_SLEEP), !=,
223 zio->io_error = VOP_FSYNC(vf->vf_vnode, FSYNC | FDSYNC,
227 zio->io_error = SET_ERROR(ENOTSUP);
234 zio->io_target_timestamp = zio_handle_io_delay(zio);
236 VERIFY3U(taskq_dispatch(vdev_file_taskq, vdev_file_io_strategy, zio,
237 TQ_SLEEP), !=, TASKQID_INVALID);
242 vdev_file_io_done(zio_t *zio)
246 vdev_ops_t vdev_file_ops = {
257 VDEV_TYPE_FILE, /* name of this vdev type */
258 B_TRUE /* leaf vdev */
264 vdev_file_taskq = taskq_create("z_vdev_file", MAX(boot_ncpus, 16),
265 minclsyspri, boot_ncpus, INT_MAX, TASKQ_DYNAMIC);
267 VERIFY(vdev_file_taskq);
273 taskq_destroy(vdev_file_taskq);
277 * From userland we access disks just like files.
281 vdev_ops_t vdev_disk_ops = {
292 VDEV_TYPE_DISK, /* name of this vdev type */
293 B_TRUE /* leaf vdev */