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 https://opensource.org/licenses/CDDL-1.0.
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 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2020 by Delphix. All rights reserved.
24 */
25
26 #include <sys/zfs_context.h>
27 #include <sys/spa.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>
32 #include <sys/zio.h>
33 #include <sys/fs/zfs.h>
34 #include <sys/fm/fs/zfs.h>
35 #include <sys/abd.h>
36 #include <sys/fcntl.h>
37 #include <sys/vnode.h>
38 #include <sys/zfs_file.h>
39 #ifdef _KERNEL
40 #include <linux/falloc.h>
41 #endif
42 /*
43 * Virtual device vector for files.
44 */
45
46 static taskq_t *vdev_file_taskq;
47
48 /*
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
54 * time.
55 */
56 static uint_t vdev_file_logical_ashift = SPA_MINBLOCKSHIFT;
57 static uint_t vdev_file_physical_ashift = SPA_MINBLOCKSHIFT;
58
59 static void
vdev_file_hold(vdev_t * vd)60 vdev_file_hold(vdev_t *vd)
61 {
62 ASSERT(vd->vdev_path != NULL);
63 }
64
65 static void
vdev_file_rele(vdev_t * vd)66 vdev_file_rele(vdev_t *vd)
67 {
68 ASSERT(vd->vdev_path != NULL);
69 }
70
71 static mode_t
vdev_file_open_mode(spa_mode_t spa_mode)72 vdev_file_open_mode(spa_mode_t spa_mode)
73 {
74 mode_t mode = 0;
75
76 if ((spa_mode & SPA_MODE_READ) && (spa_mode & SPA_MODE_WRITE)) {
77 mode = O_RDWR;
78 } else if (spa_mode & SPA_MODE_READ) {
79 mode = O_RDONLY;
80 } else if (spa_mode & SPA_MODE_WRITE) {
81 mode = O_WRONLY;
82 }
83
84 return (mode | O_LARGEFILE);
85 }
86
87 static int
vdev_file_open(vdev_t * vd,uint64_t * psize,uint64_t * max_psize,uint64_t * logical_ashift,uint64_t * physical_ashift)88 vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize,
89 uint64_t *logical_ashift, uint64_t *physical_ashift)
90 {
91 vdev_file_t *vf;
92 zfs_file_t *fp;
93 zfs_file_attr_t zfa;
94 int error;
95
96 /*
97 * Rotational optimizations only make sense on block devices.
98 */
99 vd->vdev_nonrot = B_TRUE;
100
101 /*
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.
105 */
106 vd->vdev_has_trim = B_TRUE;
107
108 /*
109 * Disable secure TRIM on file based vdevs. There is no way to
110 * request this behavior from the underlying filesystem.
111 */
112 vd->vdev_has_securetrim = B_FALSE;
113
114 /*
115 * We must have a pathname, and it must be absolute.
116 */
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));
120 }
121
122 /*
123 * Reopen the device if it's not currently open. Otherwise,
124 * just update the physical size of the device.
125 */
126 if (vd->vdev_tsd != NULL) {
127 ASSERT(vd->vdev_reopening);
128 vf = vd->vdev_tsd;
129 goto skip_open;
130 }
131
132 vf = vd->vdev_tsd = kmem_zalloc(sizeof (vdev_file_t), KM_SLEEP);
133
134 /*
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.
139 */
140 ASSERT(vd->vdev_path != NULL && vd->vdev_path[0] == '/');
141
142 error = zfs_file_open(vd->vdev_path,
143 vdev_file_open_mode(spa_mode(vd->vdev_spa)), 0, &fp);
144 if (error) {
145 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
146 return (error);
147 }
148
149 vf->vf_file = fp;
150
151 #ifdef _KERNEL
152 /*
153 * Make sure it's a regular file.
154 */
155 if (zfs_file_getattr(fp, &zfa)) {
156 return (SET_ERROR(ENODEV));
157 }
158 if (!S_ISREG(zfa.zfa_mode)) {
159 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
160 return (SET_ERROR(ENODEV));
161 }
162 #endif
163
164 skip_open:
165
166 error = zfs_file_getattr(vf->vf_file, &zfa);
167 if (error) {
168 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
169 return (error);
170 }
171
172 *max_psize = *psize = zfa.zfa_size;
173 *logical_ashift = vdev_file_logical_ashift;
174 *physical_ashift = vdev_file_physical_ashift;
175
176 return (0);
177 }
178
179 static void
vdev_file_close(vdev_t * vd)180 vdev_file_close(vdev_t *vd)
181 {
182 vdev_file_t *vf = vd->vdev_tsd;
183
184 if (vd->vdev_reopening || vf == NULL)
185 return;
186
187 if (vf->vf_file != NULL) {
188 (void) zfs_file_close(vf->vf_file);
189 }
190
191 vd->vdev_delayed_close = B_FALSE;
192 kmem_free(vf, sizeof (vdev_file_t));
193 vd->vdev_tsd = NULL;
194 }
195
196 static void
vdev_file_io_strategy(void * arg)197 vdev_file_io_strategy(void *arg)
198 {
199 zio_t *zio = (zio_t *)arg;
200 vdev_t *vd = zio->io_vd;
201 vdev_file_t *vf = vd->vdev_tsd;
202 ssize_t resid;
203 void *buf;
204 loff_t off;
205 ssize_t size;
206 int err;
207
208 off = zio->io_offset;
209 size = zio->io_size;
210 resid = 0;
211
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);
216 } else {
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);
220 }
221 zio->io_error = err;
222 if (resid != 0 && zio->io_error == 0)
223 zio->io_error = SET_ERROR(ENOSPC);
224
225 zio_delay_interrupt(zio);
226 }
227
228 static void
vdev_file_io_fsync(void * arg)229 vdev_file_io_fsync(void *arg)
230 {
231 zio_t *zio = (zio_t *)arg;
232 vdev_file_t *vf = zio->io_vd->vdev_tsd;
233
234 zio->io_error = zfs_file_fsync(vf->vf_file, O_SYNC | O_DSYNC);
235
236 zio_interrupt(zio);
237 }
238
239 static void
vdev_file_io_start(zio_t * zio)240 vdev_file_io_start(zio_t *zio)
241 {
242 vdev_t *vd = zio->io_vd;
243 vdev_file_t *vf = vd->vdev_tsd;
244
245 if (zio->io_type == ZIO_TYPE_FLUSH) {
246 /* XXPOLICY */
247 if (!vdev_readable(vd)) {
248 zio->io_error = SET_ERROR(ENXIO);
249 zio_interrupt(zio);
250 return;
251 }
252
253 if (zfs_nocacheflush) {
254 zio_execute(zio);
255 return;
256 }
257
258 /*
259 * We cannot safely call vfs_fsync() when PF_FSTRANS
260 * is set in the current context. Filesystems like
261 * XFS include sanity checks to verify it is not
262 * already set, see xfs_vm_writepage(). Therefore
263 * the sync must be dispatched to a different context.
264 */
265 if (__spl_pf_fstrans_check()) {
266 VERIFY3U(taskq_dispatch(vdev_file_taskq,
267 vdev_file_io_fsync, zio, TQ_SLEEP), !=,
268 TASKQID_INVALID);
269 return;
270 }
271
272 zio->io_error = zfs_file_fsync(vf->vf_file, O_SYNC | O_DSYNC);
273
274 zio_execute(zio);
275 return;
276 } else if (zio->io_type == ZIO_TYPE_TRIM) {
277 int mode = 0;
278
279 ASSERT3U(zio->io_size, !=, 0);
280 #ifdef __linux__
281 mode = FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE;
282 #endif
283 zio->io_error = zfs_file_fallocate(vf->vf_file,
284 mode, zio->io_offset, zio->io_size);
285 zio_execute(zio);
286 return;
287 }
288
289 zio->io_target_timestamp = zio_handle_io_delay(zio);
290
291 VERIFY3U(taskq_dispatch(vdev_file_taskq, vdev_file_io_strategy, zio,
292 TQ_SLEEP), !=, TASKQID_INVALID);
293 }
294
295 static void
vdev_file_io_done(zio_t * zio)296 vdev_file_io_done(zio_t *zio)
297 {
298 (void) zio;
299 }
300
301 vdev_ops_t vdev_file_ops = {
302 .vdev_op_init = NULL,
303 .vdev_op_fini = NULL,
304 .vdev_op_open = vdev_file_open,
305 .vdev_op_close = vdev_file_close,
306 .vdev_op_asize = vdev_default_asize,
307 .vdev_op_min_asize = vdev_default_min_asize,
308 .vdev_op_min_alloc = NULL,
309 .vdev_op_io_start = vdev_file_io_start,
310 .vdev_op_io_done = vdev_file_io_done,
311 .vdev_op_state_change = NULL,
312 .vdev_op_need_resilver = NULL,
313 .vdev_op_hold = vdev_file_hold,
314 .vdev_op_rele = vdev_file_rele,
315 .vdev_op_remap = NULL,
316 .vdev_op_xlate = vdev_default_xlate,
317 .vdev_op_rebuild_asize = NULL,
318 .vdev_op_metaslab_init = NULL,
319 .vdev_op_config_generate = NULL,
320 .vdev_op_nparity = NULL,
321 .vdev_op_ndisks = NULL,
322 .vdev_op_type = VDEV_TYPE_FILE, /* name of this vdev type */
323 .vdev_op_leaf = B_TRUE /* leaf vdev */
324 };
325
326 void
vdev_file_init(void)327 vdev_file_init(void)
328 {
329 vdev_file_taskq = taskq_create("z_vdev_file", MAX(boot_ncpus, 16),
330 minclsyspri, boot_ncpus, INT_MAX, TASKQ_DYNAMIC);
331
332 VERIFY(vdev_file_taskq);
333 }
334
335 void
vdev_file_fini(void)336 vdev_file_fini(void)
337 {
338 taskq_destroy(vdev_file_taskq);
339 }
340
341 /*
342 * From userland we access disks just like files.
343 */
344 #ifndef _KERNEL
345
346 vdev_ops_t vdev_disk_ops = {
347 .vdev_op_init = NULL,
348 .vdev_op_fini = NULL,
349 .vdev_op_open = vdev_file_open,
350 .vdev_op_close = vdev_file_close,
351 .vdev_op_asize = vdev_default_asize,
352 .vdev_op_min_asize = vdev_default_min_asize,
353 .vdev_op_min_alloc = NULL,
354 .vdev_op_io_start = vdev_file_io_start,
355 .vdev_op_io_done = vdev_file_io_done,
356 .vdev_op_state_change = NULL,
357 .vdev_op_need_resilver = NULL,
358 .vdev_op_hold = vdev_file_hold,
359 .vdev_op_rele = vdev_file_rele,
360 .vdev_op_remap = NULL,
361 .vdev_op_xlate = vdev_default_xlate,
362 .vdev_op_rebuild_asize = NULL,
363 .vdev_op_metaslab_init = NULL,
364 .vdev_op_config_generate = NULL,
365 .vdev_op_nparity = NULL,
366 .vdev_op_ndisks = NULL,
367 .vdev_op_type = VDEV_TYPE_DISK, /* name of this vdev type */
368 .vdev_op_leaf = B_TRUE /* leaf vdev */
369 };
370
371 #endif
372
373 ZFS_MODULE_PARAM(zfs_vdev_file, vdev_file_, logical_ashift, UINT, ZMOD_RW,
374 "Logical ashift for file-based devices");
375 ZFS_MODULE_PARAM(zfs_vdev_file, vdev_file_, physical_ashift, UINT, ZMOD_RW,
376 "Physical ashift for file-based devices");
377