]>
Commit | Line | Data |
---|---|---|
9f0a21e6 MM |
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 http://www.opensolaris.org/os/licensing. | |
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. | |
c494aa7f | 23 | * Copyright (c) 2011, 2020 by Delphix. All rights reserved. |
9f0a21e6 MM |
24 | */ |
25 | ||
26 | #include <sys/zfs_context.h> | |
27 | #include <sys/spa.h> | |
7ddb753d | 28 | #include <sys/file.h> |
9f0a21e6 MM |
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 | ||
c494aa7f GW |
43 | unsigned long vdev_file_logical_ashift = SPA_MINBLOCKSHIFT; |
44 | unsigned long vdev_file_physical_ashift = SPA_MINBLOCKSHIFT; | |
45 | ||
9f0a21e6 MM |
46 | 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 | |
54 | vdev_file_fini(void) | |
55 | { | |
56 | taskq_destroy(vdev_file_taskq); | |
57 | } | |
58 | ||
59 | static void | |
60 | vdev_file_hold(vdev_t *vd) | |
61 | { | |
62 | ASSERT(vd->vdev_path != NULL); | |
63 | } | |
64 | ||
65 | static void | |
66 | vdev_file_rele(vdev_t *vd) | |
67 | { | |
68 | ASSERT(vd->vdev_path != NULL); | |
69 | } | |
70 | ||
71 | static mode_t | |
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 | |
88 | vdev_file_open(vdev_t *vd, uint64_t *psize, uint64_t *max_psize, | |
6fe3498c | 89 | uint64_t *logical_ashift, uint64_t *physical_ashift) |
9f0a21e6 MM |
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; | |
c494aa7f GW |
173 | *logical_ashift = vdev_file_logical_ashift; |
174 | *physical_ashift = vdev_file_physical_ashift; | |
9f0a21e6 MM |
175 | |
176 | return (0); | |
177 | } | |
178 | ||
179 | static void | |
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 | 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 | /* | |
197 | * Implements the interrupt side for file vdev types. This routine will be | |
198 | * called when the I/O completes allowing us to transfer the I/O to the | |
199 | * interrupt taskqs. For consistency, the code structure mimics disk vdev | |
200 | * types. | |
201 | */ | |
202 | static void | |
203 | vdev_file_io_intr(zio_t *zio) | |
204 | { | |
205 | zio_delay_interrupt(zio); | |
206 | } | |
207 | ||
208 | static void | |
209 | vdev_file_io_strategy(void *arg) | |
210 | { | |
211 | zio_t *zio = arg; | |
212 | vdev_t *vd = zio->io_vd; | |
213 | vdev_file_t *vf; | |
214 | void *buf; | |
215 | ssize_t resid; | |
216 | loff_t off; | |
217 | ssize_t size; | |
218 | int err; | |
219 | ||
220 | off = zio->io_offset; | |
221 | size = zio->io_size; | |
222 | resid = 0; | |
223 | ||
224 | vf = vd->vdev_tsd; | |
225 | ||
226 | ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE); | |
227 | if (zio->io_type == ZIO_TYPE_READ) { | |
228 | buf = abd_borrow_buf(zio->io_abd, zio->io_size); | |
229 | err = zfs_file_pread(vf->vf_file, buf, size, off, &resid); | |
230 | abd_return_buf_copy(zio->io_abd, buf, size); | |
231 | } else { | |
232 | buf = abd_borrow_buf_copy(zio->io_abd, zio->io_size); | |
233 | err = zfs_file_pwrite(vf->vf_file, buf, size, off, &resid); | |
234 | abd_return_buf(zio->io_abd, buf, size); | |
235 | } | |
236 | if (resid != 0 && zio->io_error == 0) | |
237 | zio->io_error = ENOSPC; | |
238 | ||
239 | vdev_file_io_intr(zio); | |
240 | } | |
241 | ||
242 | static void | |
243 | vdev_file_io_start(zio_t *zio) | |
244 | { | |
245 | vdev_t *vd = zio->io_vd; | |
246 | vdev_file_t *vf = vd->vdev_tsd; | |
247 | ||
248 | if (zio->io_type == ZIO_TYPE_IOCTL) { | |
249 | /* XXPOLICY */ | |
250 | if (!vdev_readable(vd)) { | |
251 | zio->io_error = SET_ERROR(ENXIO); | |
252 | zio_interrupt(zio); | |
253 | return; | |
254 | } | |
255 | ||
256 | switch (zio->io_cmd) { | |
257 | case DKIOCFLUSHWRITECACHE: | |
258 | zio->io_error = zfs_file_fsync(vf->vf_file, | |
259 | O_SYNC|O_DSYNC); | |
260 | break; | |
261 | default: | |
262 | zio->io_error = SET_ERROR(ENOTSUP); | |
263 | } | |
264 | ||
265 | zio_execute(zio); | |
266 | return; | |
267 | } else if (zio->io_type == ZIO_TYPE_TRIM) { | |
268 | #ifdef notyet | |
269 | int mode = 0; | |
270 | ||
271 | ASSERT3U(zio->io_size, !=, 0); | |
272 | ||
273 | /* XXX FreeBSD has no fallocate routine in file ops */ | |
274 | zio->io_error = zfs_file_fallocate(vf->vf_file, | |
275 | mode, zio->io_offset, zio->io_size); | |
276 | #endif | |
277 | zio->io_error = SET_ERROR(ENOTSUP); | |
278 | zio_execute(zio); | |
279 | return; | |
280 | } | |
281 | ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE); | |
282 | zio->io_target_timestamp = zio_handle_io_delay(zio); | |
283 | ||
284 | VERIFY3U(taskq_dispatch(vdev_file_taskq, vdev_file_io_strategy, zio, | |
285 | TQ_SLEEP), !=, 0); | |
286 | } | |
287 | ||
288 | /* ARGSUSED */ | |
289 | static void | |
290 | vdev_file_io_done(zio_t *zio) | |
291 | { | |
292 | } | |
293 | ||
294 | vdev_ops_t vdev_file_ops = { | |
b2255edc BB |
295 | .vdev_op_init = NULL, |
296 | .vdev_op_fini = NULL, | |
297 | .vdev_op_open = vdev_file_open, | |
298 | .vdev_op_close = vdev_file_close, | |
299 | .vdev_op_asize = vdev_default_asize, | |
300 | .vdev_op_min_asize = vdev_default_min_asize, | |
301 | .vdev_op_min_alloc = NULL, | |
302 | .vdev_op_io_start = vdev_file_io_start, | |
303 | .vdev_op_io_done = vdev_file_io_done, | |
304 | .vdev_op_state_change = NULL, | |
305 | .vdev_op_need_resilver = NULL, | |
306 | .vdev_op_hold = vdev_file_hold, | |
307 | .vdev_op_rele = vdev_file_rele, | |
308 | .vdev_op_remap = NULL, | |
309 | .vdev_op_xlate = vdev_default_xlate, | |
310 | .vdev_op_rebuild_asize = NULL, | |
311 | .vdev_op_metaslab_init = NULL, | |
312 | .vdev_op_config_generate = NULL, | |
313 | .vdev_op_nparity = NULL, | |
314 | .vdev_op_ndisks = NULL, | |
315 | .vdev_op_type = VDEV_TYPE_FILE, /* name of this vdev type */ | |
316 | .vdev_op_leaf = B_TRUE /* leaf vdev */ | |
9f0a21e6 MM |
317 | }; |
318 | ||
319 | /* | |
320 | * From userland we access disks just like files. | |
321 | */ | |
322 | #ifndef _KERNEL | |
323 | ||
324 | vdev_ops_t vdev_disk_ops = { | |
b2255edc BB |
325 | .vdev_op_init = NULL, |
326 | .vdev_op_fini = NULL, | |
327 | .vdev_op_open = vdev_file_open, | |
328 | .vdev_op_close = vdev_file_close, | |
329 | .vdev_op_asize = vdev_default_asize, | |
330 | .vdev_op_min_asize = vdev_default_min_asize, | |
331 | .vdev_op_min_alloc = NULL, | |
332 | .vdev_op_io_start = vdev_file_io_start, | |
333 | .vdev_op_io_done = vdev_file_io_done, | |
334 | .vdev_op_state_change = NULL, | |
335 | .vdev_op_need_resilver = NULL, | |
336 | .vdev_op_hold = vdev_file_hold, | |
337 | .vdev_op_rele = vdev_file_rele, | |
338 | .vdev_op_remap = NULL, | |
339 | .vdev_op_xlate = vdev_default_xlate, | |
340 | .vdev_op_rebuild_asize = NULL, | |
341 | .vdev_op_metaslab_init = NULL, | |
342 | .vdev_op_config_generate = NULL, | |
343 | .vdev_op_nparity = NULL, | |
344 | .vdev_op_ndisks = NULL, | |
345 | .vdev_op_type = VDEV_TYPE_DISK, /* name of this vdev type */ | |
346 | .vdev_op_leaf = B_TRUE /* leaf vdev */ | |
9f0a21e6 MM |
347 | }; |
348 | ||
349 | #endif | |
c494aa7f GW |
350 | |
351 | ZFS_MODULE_PARAM(zfs_vdev_file, vdev_file_, logical_ashift, ULONG, ZMOD_RW, | |
352 | "Logical ashift for file-based devices"); | |
353 | ZFS_MODULE_PARAM(zfs_vdev_file, vdev_file_, physical_ashift, ULONG, ZMOD_RW, | |
354 | "Physical ashift for file-based devices"); |