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, 2020 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>
31 #include <sys/vdev_trim.h>
33 #include <sys/fs/zfs.h>
34 #include <sys/fm/fs/zfs.h>
36 #include <sys/fcntl.h>
37 #include <sys/vnode.h>
38 #include <sys/zfs_file.h>
40 #include <linux/falloc.h>
43 * Virtual device vector for files.
46 static taskq_t *vdev_file_taskq;
49 * By default, the logical/physical ashift for file vdevs is set to
50 * SPA_MINBLOCKSHIFT (9). This allows all file vdevs to use 512B (1 << 9)
51 * blocksizes. Users may opt to change one or both of these for testing
52 * or performance reasons. Care should be taken as these values will
53 * impact the vdev_ashift setting which can only be set at vdev creation
56 unsigned long vdev_file_logical_ashift = SPA_MINBLOCKSHIFT;
57 unsigned long vdev_file_physical_ashift = SPA_MINBLOCKSHIFT;
60 vdev_file_hold(vdev_t *vd)
62 ASSERT(vd->vdev_path != NULL);
66 vdev_file_rele(vdev_t *vd)
68 ASSERT(vd->vdev_path != NULL);
72 vdev_file_open_mode(spa_mode_t spa_mode)
76 if ((spa_mode & SPA_MODE_READ) && (spa_mode & SPA_MODE_WRITE)) {
78 } else if (spa_mode & SPA_MODE_READ) {
80 } else if (spa_mode & SPA_MODE_WRITE) {
84 return (mode | O_LARGEFILE);
88 vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
89 uint64_t *logical_ashift, uint64_t *physical_ashift)
97 * Rotational optimizations only make sense on block devices.
99 vd->vdev_nonrot = B_TRUE;
102 * Allow TRIM on file based vdevs. This may not always be supported,
103 * since it depends on your kernel version and underlying filesystem
104 * type but it is always safe to attempt.
106 vd->vdev_has_trim = B_TRUE;
109 * Disable secure TRIM on file based vdevs. There is no way to
110 * request this behavior from the underlying filesystem.
112 vd->vdev_has_securetrim = B_FALSE;
115 * We must have a pathname, and it must be absolute.
117 if (vd->vdev_path == NULL || vd->vdev_path[0] != '/') {
118 vd->vdev_stat.vs_aux = VDEV_AUX_BAD_LABEL;
119 return (SET_ERROR(EINVAL));
123 * Reopen the device if it's not currently open. Otherwise,
124 * just update the physical size of the device.
126 if (vd->vdev_tsd != NULL) {
127 ASSERT(vd->vdev_reopening);
132 vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);
135 * We always open the files from the root of the global zone, even if
136 * we're in a local zone. If the user has gotten to this point, the
137 * administrator has already decided that the pool should be available
138 * to local zone users, so the underlying devices should be as well.
140 ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
142 error = zfs_file_open(vd->vdev_path,
143 vdev_file_open_mode(spa_mode(vd->vdev_spa)), 0, &fp);
145 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
153 * Make sure it's a regular file.
155 if (zfs_file_getattr(fp, &zfa)) {
156 return (SET_ERROR(ENODEV));
158 if (!S_ISREG(zfa.zfa_mode)) {
159 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
160 return (SET_ERROR(ENODEV));
166 error = zfs_file_getattr(vf->vf_file, &zfa);
168 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
172 *max_psize = *psize = zfa.zfa_size;
173 *logical_ashift = vdev_file_logical_ashift;
174 *physical_ashift = vdev_file_physical_ashift;
180 vdev_file_close(vdev_t *vd)
182 vdev_file_t *vf = vd->vdev_tsd;
184 if (vd->vdev_reopening || vf == NULL)
187 if (vf->vf_file != NULL) {
188 (void) zfs_file_close(vf->vf_file);
191 vd->vdev_delayed_close = B_FALSE;
192 kmem_free(vf, sizeof (vdev_file_t));
197 vdev_file_io_strategy(void *arg)
199 zio_t *zio = (zio_t *)arg;
200 vdev_t *vd = zio->io_vd;
201 vdev_file_t *vf = vd->vdev_tsd;
208 off = zio->io_offset;
212 if (zio->io_type == ZIO_TYPE_READ) {
213 buf = abd_borrow_buf(zio->io_abd, zio->io_size);
214 err = zfs_file_pread(vf->vf_file, buf, size, off, &resid);
215 abd_return_buf_copy(zio->io_abd, buf, size);
217 buf = abd_borrow_buf_copy(zio->io_abd, zio->io_size);
218 err = zfs_file_pwrite(vf->vf_file, buf, size, off, &resid);
219 abd_return_buf(zio->io_abd, buf, size);
222 if (resid != 0 && zio->io_error == 0)
223 zio->io_error = SET_ERROR(ENOSPC);
225 zio_delay_interrupt(zio);
229 vdev_file_io_fsync(void *arg)
231 zio_t *zio = (zio_t *)arg;
232 vdev_file_t *vf = zio->io_vd->vdev_tsd;
234 zio->io_error = zfs_file_fsync(vf->vf_file, O_SYNC | O_DSYNC);
240 vdev_file_io_start(zio_t *zio)
242 vdev_t *vd = zio->io_vd;
243 vdev_file_t *vf = vd->vdev_tsd;
245 if (zio->io_type == ZIO_TYPE_IOCTL) {
247 if (!vdev_readable(vd)) {
248 zio->io_error = SET_ERROR(ENXIO);
253 switch (zio->io_cmd) {
254 case DKIOCFLUSHWRITECACHE:
256 if (zfs_nocacheflush)
260 * We cannot safely call vfs_fsync() when PF_FSTRANS
261 * is set in the current context. Filesystems like
262 * XFS include sanity checks to verify it is not
263 * already set, see xfs_vm_writepage(). Therefore
264 * the sync must be dispatched to a different context.
266 if (__spl_pf_fstrans_check()) {
267 VERIFY3U(taskq_dispatch(vdev_file_taskq,
268 vdev_file_io_fsync, zio, TQ_SLEEP), !=,
273 zio->io_error = zfs_file_fsync(vf->vf_file,
277 zio->io_error = SET_ERROR(ENOTSUP);
282 } else if (zio->io_type == ZIO_TYPE_TRIM) {
285 ASSERT3U(zio->io_size, !=, 0);
287 mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
289 zio->io_error = zfs_file_fallocate(vf->vf_file,
290 mode, zio->io_offset, zio->io_size);
295 zio->io_target_timestamp = zio_handle_io_delay(zio);
297 VERIFY3U(taskq_dispatch(vdev_file_taskq, vdev_file_io_strategy, zio,
298 TQ_SLEEP), !=, TASKQID_INVALID);
303 vdev_file_io_done(zio_t *zio)
307 vdev_ops_t vdev_file_ops = {
308 .vdev_op_open = vdev_file_open,
309 .vdev_op_close = vdev_file_close,
310 .vdev_op_asize = vdev_default_asize,
311 .vdev_op_io_start = vdev_file_io_start,
312 .vdev_op_io_done = vdev_file_io_done,
313 .vdev_op_state_change = NULL,
314 .vdev_op_need_resilver = NULL,
315 .vdev_op_hold = vdev_file_hold,
316 .vdev_op_rele = vdev_file_rele,
317 .vdev_op_remap = NULL,
318 .vdev_op_xlate = vdev_default_xlate,
319 .vdev_op_type = VDEV_TYPE_FILE, /* name of this vdev type */
320 .vdev_op_leaf = B_TRUE /* leaf vdev */
326 vdev_file_taskq = taskq_create("z_vdev_file", MAX(boot_ncpus, 16),
327 minclsyspri, boot_ncpus, INT_MAX, TASKQ_DYNAMIC);
329 VERIFY(vdev_file_taskq);
335 taskq_destroy(vdev_file_taskq);
339 * From userland we access disks just like files.
343 vdev_ops_t vdev_disk_ops = {
344 .vdev_op_open = vdev_file_open,
345 .vdev_op_close = vdev_file_close,
346 .vdev_op_asize = vdev_default_asize,
347 .vdev_op_io_start = vdev_file_io_start,
348 .vdev_op_io_done = vdev_file_io_done,
349 .vdev_op_state_change = NULL,
350 .vdev_op_need_resilver = NULL,
351 .vdev_op_hold = vdev_file_hold,
352 .vdev_op_rele = vdev_file_rele,
353 .vdev_op_remap = NULL,
354 .vdev_op_xlate = vdev_default_xlate,
355 .vdev_op_type = VDEV_TYPE_DISK, /* name of this vdev type */
356 .vdev_op_leaf = B_TRUE /* leaf vdev */
361 ZFS_MODULE_PARAM(zfs_vdev_file, vdev_file_, logical_ashift, ULONG, ZMOD_RW,
362 "Logical ashift for file-based devices");
363 ZFS_MODULE_PARAM(zfs_vdev_file, vdev_file_, physical_ashift, ULONG, ZMOD_RW,
364 "Physical ashift for file-based devices");