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/file.h>
29 #include <sys/vdev_file.h>
30 #include <sys/vdev_impl.h>
31 #include <sys/zio.h>
32 #include <sys/fs/zfs.h>
33 #include <sys/fm/fs/zfs.h>
34 #include <sys/abd.h>
35 #include <sys/stat.h>
36
37 /*
38 * Virtual device vector for files.
39 */
40
41 static taskq_t *vdev_file_taskq;
42
43 static uint_t vdev_file_logical_ashift = SPA_MINBLOCKSHIFT;
44 static uint_t vdev_file_physical_ashift = SPA_MINBLOCKSHIFT;
45
46 void
vdev_file_init(void)47 vdev_file_init(void)
48 {
49 vdev_file_taskq = taskq_create("z_vdev_file", MAX(max_ncpus, 16),
50 minclsyspri, max_ncpus, INT_MAX, 0);
51 }
52
53 void
vdev_file_fini(void)54 vdev_file_fini(void)
55 {
56 taskq_destroy(vdev_file_taskq);
57 }
58
59 static void
vdev_file_hold(vdev_t * vd)60 vdev_file_hold(vdev_t *vd)
61 {
62 ASSERT3P(vd->vdev_path, !=, NULL);
63 }
64
65 static void
vdev_file_rele(vdev_t * vd)66 vdev_file_rele(vdev_t *vd)
67 {
68 ASSERT3P(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 ASSERT3P(vd->vdev_path, !=, NULL);
141 ASSERT(vd->vdev_path[0] == '/');
142
143 error = zfs_file_open(vd->vdev_path,
144 vdev_file_open_mode(spa_mode(vd->vdev_spa)), 0, &fp);
145 if (error) {
146 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
147 return (error);
148 }
149
150 vf->vf_file = fp;
151
152 #ifdef _KERNEL
153 /*
154 * Make sure it's a regular file.
155 */
156 if (zfs_file_getattr(fp, &zfa)) {
157 return (SET_ERROR(ENODEV));
158 }
159 if (!S_ISREG(zfa.zfa_mode)) {
160 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
161 return (SET_ERROR(ENODEV));
162 }
163 #endif
164
165 skip_open:
166
167 error = zfs_file_getattr(vf->vf_file, &zfa);
168 if (error) {
169 vd->vdev_stat.vs_aux = VDEV_AUX_OPEN_FAILED;
170 return (error);
171 }
172
173 *max_psize = *psize = zfa.zfa_size;
174 *logical_ashift = vdev_file_logical_ashift;
175 *physical_ashift = vdev_file_physical_ashift;
176
177 return (0);
178 }
179
180 static void
vdev_file_close(vdev_t * vd)181 vdev_file_close(vdev_t *vd)
182 {
183 vdev_file_t *vf = vd->vdev_tsd;
184
185 if (vd->vdev_reopening || vf == NULL)
186 return;
187
188 if (vf->vf_file != NULL) {
189 zfs_file_close(vf->vf_file);
190 }
191
192 vd->vdev_delayed_close = B_FALSE;
193 kmem_free(vf, sizeof (vdev_file_t));
194 vd->vdev_tsd = NULL;
195 }
196
197 /*
198 * Implements the interrupt side for file vdev types. This routine will be
199 * called when the I/O completes allowing us to transfer the I/O to the
200 * interrupt taskqs. For consistency, the code structure mimics disk vdev
201 * types.
202 */
203 static void
vdev_file_io_intr(zio_t * zio)204 vdev_file_io_intr(zio_t *zio)
205 {
206 zio_delay_interrupt(zio);
207 }
208
209 static void
vdev_file_io_strategy(void * arg)210 vdev_file_io_strategy(void *arg)
211 {
212 zio_t *zio = arg;
213 vdev_t *vd = zio->io_vd;
214 vdev_file_t *vf;
215 void *buf;
216 ssize_t resid;
217 loff_t off;
218 ssize_t size;
219 int err;
220
221 off = zio->io_offset;
222 size = zio->io_size;
223 resid = 0;
224
225 vf = vd->vdev_tsd;
226
227 ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
228 if (zio->io_type == ZIO_TYPE_READ) {
229 buf = abd_borrow_buf(zio->io_abd, zio->io_size);
230 err = zfs_file_pread(vf->vf_file, buf, size, off, &resid);
231 abd_return_buf_copy(zio->io_abd, buf, size);
232 } else {
233 buf = abd_borrow_buf_copy(zio->io_abd, zio->io_size);
234 err = zfs_file_pwrite(vf->vf_file, buf, size, off, &resid);
235 abd_return_buf(zio->io_abd, buf, size);
236 }
237 zio->io_error = err;
238 if (resid != 0 && zio->io_error == 0)
239 zio->io_error = ENOSPC;
240
241 vdev_file_io_intr(zio);
242 }
243
244 static void
vdev_file_io_start(zio_t * zio)245 vdev_file_io_start(zio_t *zio)
246 {
247 vdev_t *vd = zio->io_vd;
248 vdev_file_t *vf = vd->vdev_tsd;
249
250 if (zio->io_type == ZIO_TYPE_FLUSH) {
251 /* XXPOLICY */
252 if (!vdev_readable(vd)) {
253 zio->io_error = SET_ERROR(ENXIO);
254 zio_interrupt(zio);
255 return;
256 }
257
258 zio->io_error = zfs_file_fsync(vf->vf_file, O_SYNC|O_DSYNC);
259
260 zio_execute(zio);
261 return;
262 } else if (zio->io_type == ZIO_TYPE_TRIM) {
263 ASSERT3U(zio->io_size, !=, 0);
264 zio->io_error = zfs_file_deallocate(vf->vf_file,
265 zio->io_offset, zio->io_size);
266 zio_execute(zio);
267 return;
268 }
269 ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
270 zio->io_target_timestamp = zio_handle_io_delay(zio);
271
272 VERIFY3U(taskq_dispatch(vdev_file_taskq, vdev_file_io_strategy, zio,
273 TQ_SLEEP), !=, 0);
274 }
275
276 static void
vdev_file_io_done(zio_t * zio)277 vdev_file_io_done(zio_t *zio)
278 {
279 (void) zio;
280 }
281
282 vdev_ops_t vdev_file_ops = {
283 .vdev_op_init = NULL,
284 .vdev_op_fini = NULL,
285 .vdev_op_open = vdev_file_open,
286 .vdev_op_close = vdev_file_close,
287 .vdev_op_asize = vdev_default_asize,
288 .vdev_op_min_asize = vdev_default_min_asize,
289 .vdev_op_min_alloc = NULL,
290 .vdev_op_io_start = vdev_file_io_start,
291 .vdev_op_io_done = vdev_file_io_done,
292 .vdev_op_state_change = NULL,
293 .vdev_op_need_resilver = NULL,
294 .vdev_op_hold = vdev_file_hold,
295 .vdev_op_rele = vdev_file_rele,
296 .vdev_op_remap = NULL,
297 .vdev_op_xlate = vdev_default_xlate,
298 .vdev_op_rebuild_asize = NULL,
299 .vdev_op_metaslab_init = NULL,
300 .vdev_op_config_generate = NULL,
301 .vdev_op_nparity = NULL,
302 .vdev_op_ndisks = NULL,
303 .vdev_op_type = VDEV_TYPE_FILE, /* name of this vdev type */
304 .vdev_op_leaf = B_TRUE /* leaf vdev */
305 };
306
307 /*
308 * From userland we access disks just like files.
309 */
310 #ifndef _KERNEL
311
312 vdev_ops_t vdev_disk_ops = {
313 .vdev_op_init = NULL,
314 .vdev_op_fini = NULL,
315 .vdev_op_open = vdev_file_open,
316 .vdev_op_close = vdev_file_close,
317 .vdev_op_asize = vdev_default_asize,
318 .vdev_op_min_asize = vdev_default_min_asize,
319 .vdev_op_min_alloc = NULL,
320 .vdev_op_io_start = vdev_file_io_start,
321 .vdev_op_io_done = vdev_file_io_done,
322 .vdev_op_state_change = NULL,
323 .vdev_op_need_resilver = NULL,
324 .vdev_op_hold = vdev_file_hold,
325 .vdev_op_rele = vdev_file_rele,
326 .vdev_op_remap = NULL,
327 .vdev_op_xlate = vdev_default_xlate,
328 .vdev_op_rebuild_asize = NULL,
329 .vdev_op_metaslab_init = NULL,
330 .vdev_op_config_generate = NULL,
331 .vdev_op_nparity = NULL,
332 .vdev_op_ndisks = NULL,
333 .vdev_op_type = VDEV_TYPE_DISK, /* name of this vdev type */
334 .vdev_op_leaf = B_TRUE /* leaf vdev */
335 };
336
337 #endif
338
339 ZFS_MODULE_PARAM(zfs_vdev_file, vdev_file_, logical_ashift, UINT, ZMOD_RW,
340 "Logical ashift for file-based devices");
341 ZFS_MODULE_PARAM(zfs_vdev_file, vdev_file_, physical_ashift, UINT, ZMOD_RW,
342 "Physical ashift for file-based devices");
343