]>
Commit | Line | Data |
---|---|---|
60101509 BB |
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) 2008-2010 Lawrence Livermore National Security, LLC. | |
23 | * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER). | |
24 | * Rewritten for Linux by Brian Behlendorf <behlendorf1@llnl.gov>. | |
25 | * LLNL-CODE-403049. | |
26 | * | |
27 | * ZFS volume emulation driver. | |
28 | * | |
29 | * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes. | |
30 | * Volumes are accessed through the symbolic links named: | |
31 | * | |
32 | * /dev/<pool_name>/<dataset_name> | |
33 | * | |
34 | * Volumes are persistent through reboot and module load. No user command | |
35 | * needs to be run before opening and using a device. | |
460a0213 DM |
36 | * |
37 | * Copyright 2014 Nexenta Systems, Inc. All rights reserved. | |
5428dc51 | 38 | * Copyright (c) 2016 Actifio, Inc. All rights reserved. |
60101509 BB |
39 | */ |
40 | ||
03c6040b | 41 | #include <sys/dbuf.h> |
60101509 BB |
42 | #include <sys/dmu_traverse.h> |
43 | #include <sys/dsl_dataset.h> | |
44 | #include <sys/dsl_prop.h> | |
a0bd735a | 45 | #include <sys/dsl_dir.h> |
60101509 | 46 | #include <sys/zap.h> |
4cb7b9c5 | 47 | #include <sys/zfeature.h> |
60101509 | 48 | #include <sys/zil_impl.h> |
460a0213 | 49 | #include <sys/dmu_tx.h> |
60101509 BB |
50 | #include <sys/zio.h> |
51 | #include <sys/zfs_rlock.h> | |
52 | #include <sys/zfs_znode.h> | |
a0bd735a | 53 | #include <sys/spa_impl.h> |
60101509 | 54 | #include <sys/zvol.h> |
61e90960 | 55 | #include <linux/blkdev_compat.h> |
60101509 | 56 | |
74497b7a | 57 | unsigned int zvol_inhibit_dev = 0; |
60101509 | 58 | unsigned int zvol_major = ZVOL_MAJOR; |
9965059a | 59 | unsigned int zvol_prefetch_bytes = (128 * 1024); |
7c0e5708 | 60 | unsigned long zvol_max_discard_blocks = 16384; |
60101509 | 61 | |
60101509 BB |
62 | static kmutex_t zvol_state_lock; |
63 | static list_t zvol_state_list; | |
47dfff3b | 64 | void *zvol_tag = "zvol_tag"; |
60101509 BB |
65 | |
66 | /* | |
67 | * The in-core state of each volume. | |
68 | */ | |
69 | typedef struct zvol_state { | |
4c0d8e50 | 70 | char zv_name[MAXNAMELEN]; /* name */ |
ce37ebd2 BB |
71 | uint64_t zv_volsize; /* advertised space */ |
72 | uint64_t zv_volblocksize; /* volume block size */ | |
60101509 BB |
73 | objset_t *zv_objset; /* objset handle */ |
74 | uint32_t zv_flags; /* ZVOL_* flags */ | |
75 | uint32_t zv_open_count; /* open counts */ | |
76 | uint32_t zv_changed; /* disk changed */ | |
77 | zilog_t *zv_zilog; /* ZIL handle */ | |
d88895a0 | 78 | zfs_rlock_t zv_range_lock; /* range lock */ |
60101509 BB |
79 | dmu_buf_t *zv_dbuf; /* bonus handle */ |
80 | dev_t zv_dev; /* device id */ | |
81 | struct gendisk *zv_disk; /* generic disk */ | |
82 | struct request_queue *zv_queue; /* request queue */ | |
60101509 BB |
83 | list_node_t zv_next; /* next zvol_state_t linkage */ |
84 | } zvol_state_t; | |
85 | ||
a0bd735a BP |
86 | typedef enum { |
87 | ZVOL_ASYNC_CREATE_MINORS, | |
88 | ZVOL_ASYNC_REMOVE_MINORS, | |
89 | ZVOL_ASYNC_RENAME_MINORS, | |
90 | ZVOL_ASYNC_SET_SNAPDEV, | |
91 | ZVOL_ASYNC_MAX | |
92 | } zvol_async_op_t; | |
93 | ||
94 | typedef struct { | |
95 | zvol_async_op_t op; | |
96 | char pool[MAXNAMELEN]; | |
97 | char name1[MAXNAMELEN]; | |
98 | char name2[MAXNAMELEN]; | |
99 | zprop_source_t source; | |
100 | uint64_t snapdev; | |
101 | } zvol_task_t; | |
102 | ||
60101509 BB |
103 | #define ZVOL_RDONLY 0x1 |
104 | ||
105 | /* | |
106 | * Find the next available range of ZVOL_MINORS minor numbers. The | |
107 | * zvol_state_list is kept in ascending minor order so we simply need | |
108 | * to scan the list for the first gap in the sequence. This allows us | |
109 | * to recycle minor number as devices are created and removed. | |
110 | */ | |
111 | static int | |
112 | zvol_find_minor(unsigned *minor) | |
113 | { | |
114 | zvol_state_t *zv; | |
115 | ||
116 | *minor = 0; | |
117 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
118 | for (zv = list_head(&zvol_state_list); zv != NULL; | |
ce37ebd2 | 119 | zv = list_next(&zvol_state_list, zv), *minor += ZVOL_MINORS) { |
60101509 BB |
120 | if (MINOR(zv->zv_dev) != MINOR(*minor)) |
121 | break; | |
122 | } | |
123 | ||
124 | /* All minors are in use */ | |
125 | if (*minor >= (1 << MINORBITS)) | |
ce37ebd2 | 126 | return (SET_ERROR(ENXIO)); |
60101509 | 127 | |
ce37ebd2 | 128 | return (0); |
60101509 BB |
129 | } |
130 | ||
131 | /* | |
132 | * Find a zvol_state_t given the full major+minor dev_t. | |
133 | */ | |
134 | static zvol_state_t * | |
135 | zvol_find_by_dev(dev_t dev) | |
136 | { | |
137 | zvol_state_t *zv; | |
138 | ||
139 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
140 | for (zv = list_head(&zvol_state_list); zv != NULL; | |
ce37ebd2 | 141 | zv = list_next(&zvol_state_list, zv)) { |
60101509 | 142 | if (zv->zv_dev == dev) |
ce37ebd2 | 143 | return (zv); |
60101509 BB |
144 | } |
145 | ||
ce37ebd2 | 146 | return (NULL); |
60101509 BB |
147 | } |
148 | ||
149 | /* | |
150 | * Find a zvol_state_t given the name provided at zvol_alloc() time. | |
151 | */ | |
152 | static zvol_state_t * | |
153 | zvol_find_by_name(const char *name) | |
154 | { | |
155 | zvol_state_t *zv; | |
156 | ||
157 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
158 | for (zv = list_head(&zvol_state_list); zv != NULL; | |
ce37ebd2 BB |
159 | zv = list_next(&zvol_state_list, zv)) { |
160 | if (strncmp(zv->zv_name, name, MAXNAMELEN) == 0) | |
161 | return (zv); | |
60101509 BB |
162 | } |
163 | ||
ce37ebd2 | 164 | return (NULL); |
60101509 BB |
165 | } |
166 | ||
6c285672 JL |
167 | |
168 | /* | |
169 | * Given a path, return TRUE if path is a ZVOL. | |
170 | */ | |
171 | boolean_t | |
172 | zvol_is_zvol(const char *device) | |
173 | { | |
174 | struct block_device *bdev; | |
175 | unsigned int major; | |
176 | ||
e02aaf17 | 177 | bdev = vdev_lookup_bdev(device); |
6c285672 JL |
178 | if (IS_ERR(bdev)) |
179 | return (B_FALSE); | |
180 | ||
181 | major = MAJOR(bdev->bd_dev); | |
182 | bdput(bdev); | |
183 | ||
184 | if (major == zvol_major) | |
ce37ebd2 | 185 | return (B_TRUE); |
6c285672 JL |
186 | |
187 | return (B_FALSE); | |
188 | } | |
189 | ||
60101509 BB |
190 | /* |
191 | * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation. | |
192 | */ | |
193 | void | |
194 | zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx) | |
195 | { | |
196 | zfs_creat_t *zct = arg; | |
197 | nvlist_t *nvprops = zct->zct_props; | |
198 | int error; | |
199 | uint64_t volblocksize, volsize; | |
200 | ||
201 | VERIFY(nvlist_lookup_uint64(nvprops, | |
202 | zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0); | |
203 | if (nvlist_lookup_uint64(nvprops, | |
204 | zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0) | |
205 | volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE); | |
206 | ||
207 | /* | |
208 | * These properties must be removed from the list so the generic | |
209 | * property setting step won't apply to them. | |
210 | */ | |
211 | VERIFY(nvlist_remove_all(nvprops, | |
212 | zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0); | |
213 | (void) nvlist_remove_all(nvprops, | |
214 | zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE)); | |
215 | ||
216 | error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize, | |
217 | DMU_OT_NONE, 0, tx); | |
218 | ASSERT(error == 0); | |
219 | ||
220 | error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP, | |
221 | DMU_OT_NONE, 0, tx); | |
222 | ASSERT(error == 0); | |
223 | ||
224 | error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx); | |
225 | ASSERT(error == 0); | |
226 | } | |
227 | ||
228 | /* | |
229 | * ZFS_IOC_OBJSET_STATS entry point. | |
230 | */ | |
231 | int | |
232 | zvol_get_stats(objset_t *os, nvlist_t *nv) | |
233 | { | |
234 | int error; | |
235 | dmu_object_info_t *doi; | |
236 | uint64_t val; | |
237 | ||
238 | error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val); | |
239 | if (error) | |
ce37ebd2 | 240 | return (SET_ERROR(error)); |
60101509 BB |
241 | |
242 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val); | |
ce37ebd2 | 243 | doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP); |
60101509 BB |
244 | error = dmu_object_info(os, ZVOL_OBJ, doi); |
245 | ||
246 | if (error == 0) { | |
247 | dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE, | |
248 | doi->doi_data_block_size); | |
249 | } | |
250 | ||
ce37ebd2 | 251 | kmem_free(doi, sizeof (dmu_object_info_t)); |
60101509 | 252 | |
ce37ebd2 | 253 | return (SET_ERROR(error)); |
60101509 BB |
254 | } |
255 | ||
35d3e322 BB |
256 | static void |
257 | zvol_size_changed(zvol_state_t *zv, uint64_t volsize) | |
258 | { | |
259 | struct block_device *bdev; | |
260 | ||
261 | bdev = bdget_disk(zv->zv_disk, 0); | |
262 | if (bdev == NULL) | |
263 | return; | |
35d3e322 BB |
264 | set_capacity(zv->zv_disk, volsize >> 9); |
265 | zv->zv_volsize = volsize; | |
266 | check_disk_size_change(zv->zv_disk, bdev); | |
35d3e322 BB |
267 | |
268 | bdput(bdev); | |
269 | } | |
270 | ||
60101509 BB |
271 | /* |
272 | * Sanity check volume size. | |
273 | */ | |
274 | int | |
275 | zvol_check_volsize(uint64_t volsize, uint64_t blocksize) | |
276 | { | |
277 | if (volsize == 0) | |
2e528b49 | 278 | return (SET_ERROR(EINVAL)); |
60101509 BB |
279 | |
280 | if (volsize % blocksize != 0) | |
2e528b49 | 281 | return (SET_ERROR(EINVAL)); |
60101509 BB |
282 | |
283 | #ifdef _ILP32 | |
82ec9d41 | 284 | if (volsize - 1 > SPEC_MAXOFFSET_T) |
2e528b49 | 285 | return (SET_ERROR(EOVERFLOW)); |
60101509 BB |
286 | #endif |
287 | return (0); | |
288 | } | |
289 | ||
290 | /* | |
291 | * Ensure the zap is flushed then inform the VFS of the capacity change. | |
292 | */ | |
293 | static int | |
35d3e322 | 294 | zvol_update_volsize(uint64_t volsize, objset_t *os) |
60101509 | 295 | { |
60101509 BB |
296 | dmu_tx_t *tx; |
297 | int error; | |
513168ab | 298 | uint64_t txg; |
60101509 BB |
299 | |
300 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
301 | ||
df554c14 | 302 | tx = dmu_tx_create(os); |
60101509 | 303 | dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL); |
19d55079 | 304 | dmu_tx_mark_netfree(tx); |
60101509 BB |
305 | error = dmu_tx_assign(tx, TXG_WAIT); |
306 | if (error) { | |
307 | dmu_tx_abort(tx); | |
ce37ebd2 | 308 | return (SET_ERROR(error)); |
60101509 | 309 | } |
513168ab | 310 | txg = dmu_tx_get_txg(tx); |
60101509 | 311 | |
df554c14 | 312 | error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, |
60101509 BB |
313 | &volsize, tx); |
314 | dmu_tx_commit(tx); | |
315 | ||
513168ab | 316 | txg_wait_synced(dmu_objset_pool(os), txg); |
317 | ||
35d3e322 BB |
318 | if (error == 0) |
319 | error = dmu_free_long_range(os, | |
320 | ZVOL_OBJ, volsize, DMU_OBJECT_END); | |
60101509 | 321 | |
35d3e322 BB |
322 | return (error); |
323 | } | |
60101509 | 324 | |
35d3e322 BB |
325 | static int |
326 | zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize) | |
327 | { | |
328 | zvol_size_changed(zv, volsize); | |
60101509 | 329 | |
35d3e322 BB |
330 | /* |
331 | * We should post a event here describing the expansion. However, | |
332 | * the zfs_ereport_post() interface doesn't nicely support posting | |
333 | * events for zvols, it assumes events relate to vdevs or zios. | |
334 | */ | |
60101509 BB |
335 | |
336 | return (0); | |
337 | } | |
338 | ||
339 | /* | |
340 | * Set ZFS_PROP_VOLSIZE set entry point. | |
341 | */ | |
342 | int | |
343 | zvol_set_volsize(const char *name, uint64_t volsize) | |
344 | { | |
35d3e322 | 345 | zvol_state_t *zv = NULL; |
60101509 | 346 | objset_t *os = NULL; |
60101509 | 347 | int error; |
35d3e322 BB |
348 | dmu_object_info_t *doi; |
349 | uint64_t readonly; | |
350 | boolean_t owned = B_FALSE; | |
60101509 | 351 | |
13fe0198 MA |
352 | error = dsl_prop_get_integer(name, |
353 | zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL); | |
354 | if (error != 0) | |
ce37ebd2 | 355 | return (SET_ERROR(error)); |
13fe0198 | 356 | if (readonly) |
2e528b49 | 357 | return (SET_ERROR(EROFS)); |
13fe0198 | 358 | |
60101509 | 359 | mutex_enter(&zvol_state_lock); |
60101509 | 360 | zv = zvol_find_by_name(name); |
35d3e322 BB |
361 | |
362 | if (zv == NULL || zv->zv_objset == NULL) { | |
363 | if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, | |
364 | FTAG, &os)) != 0) { | |
365 | mutex_exit(&zvol_state_lock); | |
366 | return (SET_ERROR(error)); | |
367 | } | |
368 | owned = B_TRUE; | |
369 | if (zv != NULL) | |
370 | zv->zv_objset = os; | |
371 | } else { | |
372 | os = zv->zv_objset; | |
60101509 BB |
373 | } |
374 | ||
ce37ebd2 | 375 | doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP); |
60101509 | 376 | |
ce37ebd2 BB |
377 | if ((error = dmu_object_info(os, ZVOL_OBJ, doi)) || |
378 | (error = zvol_check_volsize(volsize, doi->doi_data_block_size))) | |
35d3e322 | 379 | goto out; |
60101509 | 380 | |
35d3e322 | 381 | error = zvol_update_volsize(volsize, os); |
ce37ebd2 | 382 | kmem_free(doi, sizeof (dmu_object_info_t)); |
60101509 | 383 | |
35d3e322 BB |
384 | if (error == 0 && zv != NULL) |
385 | error = zvol_update_live_volsize(zv, volsize); | |
386 | out: | |
387 | if (owned) { | |
388 | dmu_objset_disown(os, FTAG); | |
389 | if (zv != NULL) | |
390 | zv->zv_objset = NULL; | |
391 | } | |
60101509 | 392 | mutex_exit(&zvol_state_lock); |
35d3e322 | 393 | return (error); |
60101509 BB |
394 | } |
395 | ||
396 | /* | |
397 | * Sanity check volume block size. | |
398 | */ | |
399 | int | |
4cb7b9c5 | 400 | zvol_check_volblocksize(const char *name, uint64_t volblocksize) |
60101509 | 401 | { |
4cb7b9c5 BB |
402 | /* Record sizes above 128k need the feature to be enabled */ |
403 | if (volblocksize > SPA_OLD_MAXBLOCKSIZE) { | |
404 | spa_t *spa; | |
405 | int error; | |
406 | ||
407 | if ((error = spa_open(name, &spa, FTAG)) != 0) | |
408 | return (error); | |
409 | ||
410 | if (!spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) { | |
411 | spa_close(spa, FTAG); | |
412 | return (SET_ERROR(ENOTSUP)); | |
413 | } | |
414 | ||
415 | /* | |
416 | * We don't allow setting the property above 1MB, | |
417 | * unless the tunable has been changed. | |
418 | */ | |
419 | if (volblocksize > zfs_max_recordsize) | |
420 | return (SET_ERROR(EDOM)); | |
421 | ||
422 | spa_close(spa, FTAG); | |
423 | } | |
424 | ||
60101509 BB |
425 | if (volblocksize < SPA_MINBLOCKSIZE || |
426 | volblocksize > SPA_MAXBLOCKSIZE || | |
427 | !ISP2(volblocksize)) | |
2e528b49 | 428 | return (SET_ERROR(EDOM)); |
60101509 BB |
429 | |
430 | return (0); | |
431 | } | |
432 | ||
433 | /* | |
434 | * Set ZFS_PROP_VOLBLOCKSIZE set entry point. | |
435 | */ | |
436 | int | |
437 | zvol_set_volblocksize(const char *name, uint64_t volblocksize) | |
438 | { | |
439 | zvol_state_t *zv; | |
440 | dmu_tx_t *tx; | |
441 | int error; | |
442 | ||
443 | mutex_enter(&zvol_state_lock); | |
444 | ||
445 | zv = zvol_find_by_name(name); | |
446 | if (zv == NULL) { | |
2e528b49 | 447 | error = SET_ERROR(ENXIO); |
60101509 BB |
448 | goto out; |
449 | } | |
450 | ||
ba6a2402 | 451 | if (zv->zv_flags & ZVOL_RDONLY) { |
2e528b49 | 452 | error = SET_ERROR(EROFS); |
60101509 BB |
453 | goto out; |
454 | } | |
455 | ||
456 | tx = dmu_tx_create(zv->zv_objset); | |
457 | dmu_tx_hold_bonus(tx, ZVOL_OBJ); | |
458 | error = dmu_tx_assign(tx, TXG_WAIT); | |
459 | if (error) { | |
460 | dmu_tx_abort(tx); | |
461 | } else { | |
462 | error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ, | |
463 | volblocksize, 0, tx); | |
464 | if (error == ENOTSUP) | |
2e528b49 | 465 | error = SET_ERROR(EBUSY); |
60101509 BB |
466 | dmu_tx_commit(tx); |
467 | if (error == 0) | |
468 | zv->zv_volblocksize = volblocksize; | |
469 | } | |
470 | out: | |
471 | mutex_exit(&zvol_state_lock); | |
472 | ||
ce37ebd2 | 473 | return (SET_ERROR(error)); |
60101509 BB |
474 | } |
475 | ||
460a0213 DM |
476 | /* |
477 | * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we | |
478 | * implement DKIOCFREE/free-long-range. | |
479 | */ | |
480 | static int | |
481 | zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap) | |
482 | { | |
483 | uint64_t offset, length; | |
484 | ||
485 | if (byteswap) | |
486 | byteswap_uint64_array(lr, sizeof (*lr)); | |
487 | ||
488 | offset = lr->lr_offset; | |
489 | length = lr->lr_length; | |
490 | ||
491 | return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length)); | |
492 | } | |
493 | ||
60101509 BB |
494 | /* |
495 | * Replay a TX_WRITE ZIL transaction that didn't get committed | |
496 | * after a system failure | |
497 | */ | |
498 | static int | |
499 | zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap) | |
500 | { | |
501 | objset_t *os = zv->zv_objset; | |
502 | char *data = (char *)(lr + 1); /* data follows lr_write_t */ | |
503 | uint64_t off = lr->lr_offset; | |
504 | uint64_t len = lr->lr_length; | |
505 | dmu_tx_t *tx; | |
506 | int error; | |
507 | ||
508 | if (byteswap) | |
509 | byteswap_uint64_array(lr, sizeof (*lr)); | |
510 | ||
511 | tx = dmu_tx_create(os); | |
512 | dmu_tx_hold_write(tx, ZVOL_OBJ, off, len); | |
513 | error = dmu_tx_assign(tx, TXG_WAIT); | |
514 | if (error) { | |
515 | dmu_tx_abort(tx); | |
516 | } else { | |
517 | dmu_write(os, ZVOL_OBJ, off, len, data, tx); | |
518 | dmu_tx_commit(tx); | |
519 | } | |
520 | ||
ce37ebd2 | 521 | return (SET_ERROR(error)); |
60101509 BB |
522 | } |
523 | ||
524 | static int | |
525 | zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap) | |
526 | { | |
2e528b49 | 527 | return (SET_ERROR(ENOTSUP)); |
60101509 BB |
528 | } |
529 | ||
530 | /* | |
531 | * Callback vectors for replaying records. | |
460a0213 | 532 | * Only TX_WRITE and TX_TRUNCATE are needed for zvol. |
60101509 | 533 | */ |
b01615d5 RY |
534 | zil_replay_func_t zvol_replay_vector[TX_MAX_TYPE] = { |
535 | (zil_replay_func_t)zvol_replay_err, /* no such transaction type */ | |
536 | (zil_replay_func_t)zvol_replay_err, /* TX_CREATE */ | |
537 | (zil_replay_func_t)zvol_replay_err, /* TX_MKDIR */ | |
538 | (zil_replay_func_t)zvol_replay_err, /* TX_MKXATTR */ | |
539 | (zil_replay_func_t)zvol_replay_err, /* TX_SYMLINK */ | |
540 | (zil_replay_func_t)zvol_replay_err, /* TX_REMOVE */ | |
541 | (zil_replay_func_t)zvol_replay_err, /* TX_RMDIR */ | |
542 | (zil_replay_func_t)zvol_replay_err, /* TX_LINK */ | |
543 | (zil_replay_func_t)zvol_replay_err, /* TX_RENAME */ | |
544 | (zil_replay_func_t)zvol_replay_write, /* TX_WRITE */ | |
460a0213 | 545 | (zil_replay_func_t)zvol_replay_truncate, /* TX_TRUNCATE */ |
b01615d5 RY |
546 | (zil_replay_func_t)zvol_replay_err, /* TX_SETATTR */ |
547 | (zil_replay_func_t)zvol_replay_err, /* TX_ACL */ | |
60101509 BB |
548 | }; |
549 | ||
550 | /* | |
551 | * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions. | |
552 | * | |
553 | * We store data in the log buffers if it's small enough. | |
554 | * Otherwise we will later flush the data out via dmu_sync(). | |
555 | */ | |
556 | ssize_t zvol_immediate_write_sz = 32768; | |
557 | ||
558 | static void | |
ce37ebd2 BB |
559 | zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, uint64_t offset, |
560 | uint64_t size, int sync) | |
60101509 BB |
561 | { |
562 | uint32_t blocksize = zv->zv_volblocksize; | |
563 | zilog_t *zilog = zv->zv_zilog; | |
564 | boolean_t slogging; | |
ab85f845 | 565 | ssize_t immediate_write_sz; |
60101509 BB |
566 | |
567 | if (zil_replaying(zilog, tx)) | |
568 | return; | |
569 | ||
ab85f845 ED |
570 | immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT) |
571 | ? 0 : zvol_immediate_write_sz; | |
572 | slogging = spa_has_slogs(zilog->zl_spa) && | |
573 | (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY); | |
60101509 BB |
574 | |
575 | while (size) { | |
576 | itx_t *itx; | |
577 | lr_write_t *lr; | |
578 | ssize_t len; | |
579 | itx_wr_state_t write_state; | |
580 | ||
581 | /* | |
582 | * Unlike zfs_log_write() we can be called with | |
583 | * up to DMU_MAX_ACCESS/2 (5MB) writes. | |
584 | */ | |
ab85f845 | 585 | if (blocksize > immediate_write_sz && !slogging && |
60101509 BB |
586 | size >= blocksize && offset % blocksize == 0) { |
587 | write_state = WR_INDIRECT; /* uses dmu_sync */ | |
588 | len = blocksize; | |
589 | } else if (sync) { | |
590 | write_state = WR_COPIED; | |
591 | len = MIN(ZIL_MAX_LOG_DATA, size); | |
592 | } else { | |
593 | write_state = WR_NEED_COPY; | |
594 | len = MIN(ZIL_MAX_LOG_DATA, size); | |
595 | } | |
596 | ||
597 | itx = zil_itx_create(TX_WRITE, sizeof (*lr) + | |
598 | (write_state == WR_COPIED ? len : 0)); | |
599 | lr = (lr_write_t *)&itx->itx_lr; | |
600 | if (write_state == WR_COPIED && dmu_read(zv->zv_objset, | |
601 | ZVOL_OBJ, offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) { | |
602 | zil_itx_destroy(itx); | |
603 | itx = zil_itx_create(TX_WRITE, sizeof (*lr)); | |
604 | lr = (lr_write_t *)&itx->itx_lr; | |
605 | write_state = WR_NEED_COPY; | |
606 | } | |
607 | ||
608 | itx->itx_wr_state = write_state; | |
609 | if (write_state == WR_NEED_COPY) | |
610 | itx->itx_sod += len; | |
611 | lr->lr_foid = ZVOL_OBJ; | |
612 | lr->lr_offset = offset; | |
613 | lr->lr_length = len; | |
614 | lr->lr_blkoff = 0; | |
615 | BP_ZERO(&lr->lr_blkptr); | |
616 | ||
617 | itx->itx_private = zv; | |
618 | itx->itx_sync = sync; | |
619 | ||
620 | (void) zil_itx_assign(zilog, itx, tx); | |
621 | ||
622 | offset += len; | |
623 | size -= len; | |
624 | } | |
625 | } | |
626 | ||
37f9dac5 | 627 | static int |
a765a34a | 628 | zvol_write(zvol_state_t *zv, uio_t *uio, boolean_t sync) |
60101509 | 629 | { |
a765a34a | 630 | uint64_t volsize = zv->zv_volsize; |
60101509 | 631 | rl_t *rl; |
a765a34a | 632 | int error = 0; |
60101509 | 633 | |
5428dc51 BP |
634 | ASSERT(zv && zv->zv_open_count > 0); |
635 | ||
d88895a0 CC |
636 | rl = zfs_range_lock(&zv->zv_range_lock, uio->uio_loffset, |
637 | uio->uio_resid, RL_WRITER); | |
b18019d2 | 638 | |
a765a34a RY |
639 | while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { |
640 | uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); | |
641 | uint64_t off = uio->uio_loffset; | |
642 | dmu_tx_t *tx = dmu_tx_create(zv->zv_objset); | |
b18019d2 | 643 | |
a765a34a RY |
644 | if (bytes > volsize - off) /* don't write past the end */ |
645 | bytes = volsize - off; | |
2727b9d3 | 646 | |
a765a34a | 647 | dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes); |
60101509 | 648 | |
a765a34a RY |
649 | /* This will only fail for ENOSPC */ |
650 | error = dmu_tx_assign(tx, TXG_WAIT); | |
651 | if (error) { | |
652 | dmu_tx_abort(tx); | |
653 | break; | |
654 | } | |
655 | error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx); | |
656 | if (error == 0) | |
657 | zvol_log_write(zv, tx, off, bytes, sync); | |
658 | dmu_tx_commit(tx); | |
60101509 | 659 | |
a765a34a RY |
660 | if (error) |
661 | break; | |
60101509 | 662 | } |
60101509 | 663 | zfs_range_unlock(rl); |
a765a34a | 664 | if (sync) |
60101509 | 665 | zil_commit(zv->zv_zilog, ZVOL_OBJ); |
37f9dac5 | 666 | return (error); |
60101509 BB |
667 | } |
668 | ||
460a0213 DM |
669 | /* |
670 | * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE. | |
671 | */ | |
672 | static void | |
673 | zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len, | |
674 | boolean_t sync) | |
675 | { | |
676 | itx_t *itx; | |
677 | lr_truncate_t *lr; | |
678 | zilog_t *zilog = zv->zv_zilog; | |
679 | ||
680 | if (zil_replaying(zilog, tx)) | |
681 | return; | |
682 | ||
683 | itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr)); | |
684 | lr = (lr_truncate_t *)&itx->itx_lr; | |
685 | lr->lr_foid = ZVOL_OBJ; | |
686 | lr->lr_offset = off; | |
687 | lr->lr_length = len; | |
688 | ||
689 | itx->itx_sync = sync; | |
690 | zil_itx_assign(zilog, itx, tx); | |
691 | } | |
692 | ||
37f9dac5 RY |
693 | static int |
694 | zvol_discard(struct bio *bio) | |
30930fba | 695 | { |
37f9dac5 RY |
696 | zvol_state_t *zv = bio->bi_bdev->bd_disk->private_data; |
697 | uint64_t start = BIO_BI_SECTOR(bio) << 9; | |
698 | uint64_t size = BIO_BI_SIZE(bio); | |
699 | uint64_t end = start + size; | |
30930fba ED |
700 | int error; |
701 | rl_t *rl; | |
460a0213 | 702 | dmu_tx_t *tx; |
30930fba | 703 | |
5428dc51 BP |
704 | ASSERT(zv && zv->zv_open_count > 0); |
705 | ||
37f9dac5 RY |
706 | if (end > zv->zv_volsize) |
707 | return (SET_ERROR(EIO)); | |
30930fba | 708 | |
089fa91b | 709 | /* |
cf41432c BB |
710 | * Align the request to volume block boundaries when a secure erase is |
711 | * not required. This will prevent dnode_free_range() from zeroing out | |
712 | * the unaligned parts which is slow (read-modify-write) and useless | |
713 | * since we are not freeing any space by doing so. | |
089fa91b | 714 | */ |
cf41432c | 715 | if (!bio_is_secure_erase(bio)) { |
fa565676 RY |
716 | start = P2ROUNDUP(start, zv->zv_volblocksize); |
717 | end = P2ALIGN(end, zv->zv_volblocksize); | |
f52ebcb3 | 718 | size = end - start; |
fa565676 | 719 | } |
089fa91b | 720 | |
37f9dac5 RY |
721 | if (start >= end) |
722 | return (0); | |
30930fba | 723 | |
d88895a0 | 724 | rl = zfs_range_lock(&zv->zv_range_lock, start, size, RL_WRITER); |
460a0213 DM |
725 | tx = dmu_tx_create(zv->zv_objset); |
726 | dmu_tx_mark_netfree(tx); | |
727 | error = dmu_tx_assign(tx, TXG_WAIT); | |
728 | if (error != 0) { | |
729 | dmu_tx_abort(tx); | |
730 | } else { | |
731 | zvol_log_truncate(zv, tx, start, size, B_TRUE); | |
732 | dmu_tx_commit(tx); | |
733 | error = dmu_free_long_range(zv->zv_objset, | |
734 | ZVOL_OBJ, start, size); | |
735 | } | |
30930fba ED |
736 | |
737 | zfs_range_unlock(rl); | |
37f9dac5 RY |
738 | |
739 | return (error); | |
30930fba | 740 | } |
30930fba | 741 | |
37f9dac5 | 742 | static int |
a765a34a | 743 | zvol_read(zvol_state_t *zv, uio_t *uio) |
60101509 | 744 | { |
a765a34a | 745 | uint64_t volsize = zv->zv_volsize; |
60101509 | 746 | rl_t *rl; |
a765a34a | 747 | int error = 0; |
b18019d2 | 748 | |
5428dc51 BP |
749 | ASSERT(zv && zv->zv_open_count > 0); |
750 | ||
d88895a0 CC |
751 | rl = zfs_range_lock(&zv->zv_range_lock, uio->uio_loffset, |
752 | uio->uio_resid, RL_READER); | |
a765a34a RY |
753 | while (uio->uio_resid > 0 && uio->uio_loffset < volsize) { |
754 | uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1); | |
60101509 | 755 | |
a765a34a RY |
756 | /* don't read past the end */ |
757 | if (bytes > volsize - uio->uio_loffset) | |
758 | bytes = volsize - uio->uio_loffset; | |
60101509 | 759 | |
19a47cb1 | 760 | error = dmu_read_uio_dbuf(zv->zv_dbuf, uio, bytes); |
a765a34a RY |
761 | if (error) { |
762 | /* convert checksum errors into IO errors */ | |
763 | if (error == ECKSUM) | |
764 | error = SET_ERROR(EIO); | |
765 | break; | |
766 | } | |
767 | } | |
60101509 | 768 | zfs_range_unlock(rl); |
37f9dac5 | 769 | return (error); |
60101509 BB |
770 | } |
771 | ||
37f9dac5 RY |
772 | static MAKE_REQUEST_FN_RET |
773 | zvol_request(struct request_queue *q, struct bio *bio) | |
60101509 | 774 | { |
a765a34a | 775 | uio_t uio; |
60101509 | 776 | zvol_state_t *zv = q->queuedata; |
37f9dac5 | 777 | fstrans_cookie_t cookie = spl_fstrans_mark(); |
8198d18c RY |
778 | int rw = bio_data_dir(bio); |
779 | #ifdef HAVE_GENERIC_IO_ACCT | |
780 | unsigned long start = jiffies; | |
781 | #endif | |
37f9dac5 | 782 | int error = 0; |
60101509 | 783 | |
a765a34a RY |
784 | uio.uio_bvec = &bio->bi_io_vec[BIO_BI_IDX(bio)]; |
785 | uio.uio_skip = BIO_BI_SKIP(bio); | |
786 | uio.uio_resid = BIO_BI_SIZE(bio); | |
787 | uio.uio_iovcnt = bio->bi_vcnt - BIO_BI_IDX(bio); | |
788 | uio.uio_loffset = BIO_BI_SECTOR(bio) << 9; | |
789 | uio.uio_limit = MAXOFFSET_T; | |
790 | uio.uio_segflg = UIO_BVEC; | |
791 | ||
792 | if (bio_has_data(bio) && uio.uio_loffset + uio.uio_resid > | |
793 | zv->zv_volsize) { | |
37f9dac5 | 794 | printk(KERN_INFO |
a765a34a | 795 | "%s: bad access: offset=%llu, size=%lu\n", |
37f9dac5 | 796 | zv->zv_disk->disk_name, |
a765a34a RY |
797 | (long long unsigned)uio.uio_loffset, |
798 | (long unsigned)uio.uio_resid); | |
37f9dac5 | 799 | error = SET_ERROR(EIO); |
8198d18c | 800 | goto out1; |
37f9dac5 RY |
801 | } |
802 | ||
a765a34a | 803 | generic_start_io_acct(rw, bio_sectors(bio), &zv->zv_disk->part0); |
8198d18c RY |
804 | |
805 | if (rw == WRITE) { | |
37f9dac5 RY |
806 | if (unlikely(zv->zv_flags & ZVOL_RDONLY)) { |
807 | error = SET_ERROR(EROFS); | |
8198d18c | 808 | goto out2; |
60101509 BB |
809 | } |
810 | ||
cf41432c | 811 | if (bio_is_discard(bio) || bio_is_secure_erase(bio)) { |
37f9dac5 | 812 | error = zvol_discard(bio); |
8198d18c | 813 | goto out2; |
37f9dac5 | 814 | } |
60101509 | 815 | |
a765a34a RY |
816 | /* |
817 | * Some requests are just for flush and nothing else. | |
818 | */ | |
819 | if (uio.uio_resid == 0) { | |
cf41432c | 820 | if (bio_is_flush(bio)) |
a765a34a RY |
821 | zil_commit(zv->zv_zilog, ZVOL_OBJ); |
822 | goto out2; | |
823 | } | |
824 | ||
825 | error = zvol_write(zv, &uio, | |
cf41432c BB |
826 | bio_is_flush(bio) || bio_is_fua(bio) || |
827 | zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS); | |
37f9dac5 | 828 | } else |
a765a34a | 829 | error = zvol_read(zv, &uio); |
30930fba | 830 | |
8198d18c RY |
831 | out2: |
832 | generic_end_io_acct(rw, &zv->zv_disk->part0, start); | |
833 | out1: | |
784a7fe5 | 834 | BIO_END_IO(bio, -error); |
37f9dac5 RY |
835 | spl_fstrans_unmark(cookie); |
836 | #ifdef HAVE_MAKE_REQUEST_FN_RET_INT | |
837 | return (0); | |
1a093716 CC |
838 | #elif defined(HAVE_MAKE_REQUEST_FN_RET_QC) |
839 | return (BLK_QC_T_NONE); | |
37f9dac5 | 840 | #endif |
60101509 BB |
841 | } |
842 | ||
843 | static void | |
844 | zvol_get_done(zgd_t *zgd, int error) | |
845 | { | |
846 | if (zgd->zgd_db) | |
847 | dmu_buf_rele(zgd->zgd_db, zgd); | |
848 | ||
849 | zfs_range_unlock(zgd->zgd_rl); | |
850 | ||
851 | if (error == 0 && zgd->zgd_bp) | |
852 | zil_add_block(zgd->zgd_zilog, zgd->zgd_bp); | |
853 | ||
854 | kmem_free(zgd, sizeof (zgd_t)); | |
855 | } | |
856 | ||
857 | /* | |
858 | * Get data to generate a TX_WRITE intent log record. | |
859 | */ | |
860 | static int | |
861 | zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) | |
862 | { | |
863 | zvol_state_t *zv = arg; | |
864 | objset_t *os = zv->zv_objset; | |
03c6040b | 865 | uint64_t object = ZVOL_OBJ; |
60101509 BB |
866 | uint64_t offset = lr->lr_offset; |
867 | uint64_t size = lr->lr_length; | |
03c6040b | 868 | blkptr_t *bp = &lr->lr_blkptr; |
60101509 BB |
869 | dmu_buf_t *db; |
870 | zgd_t *zgd; | |
871 | int error; | |
872 | ||
873 | ASSERT(zio != NULL); | |
874 | ASSERT(size != 0); | |
875 | ||
79c76d5b | 876 | zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); |
60101509 | 877 | zgd->zgd_zilog = zv->zv_zilog; |
d88895a0 CC |
878 | zgd->zgd_rl = zfs_range_lock(&zv->zv_range_lock, offset, size, |
879 | RL_READER); | |
60101509 BB |
880 | |
881 | /* | |
882 | * Write records come in two flavors: immediate and indirect. | |
883 | * For small writes it's cheaper to store the data with the | |
884 | * log record (immediate); for large writes it's cheaper to | |
885 | * sync the data and get a pointer to it (indirect) so that | |
886 | * we don't have to write the data twice. | |
887 | */ | |
888 | if (buf != NULL) { /* immediate write */ | |
03c6040b | 889 | error = dmu_read(os, object, offset, size, buf, |
60101509 BB |
890 | DMU_READ_NO_PREFETCH); |
891 | } else { | |
892 | size = zv->zv_volblocksize; | |
893 | offset = P2ALIGN_TYPED(offset, size, uint64_t); | |
03c6040b | 894 | error = dmu_buf_hold(os, object, offset, zgd, &db, |
60101509 BB |
895 | DMU_READ_NO_PREFETCH); |
896 | if (error == 0) { | |
03c6040b GW |
897 | blkptr_t *obp = dmu_buf_get_blkptr(db); |
898 | if (obp) { | |
899 | ASSERT(BP_IS_HOLE(bp)); | |
900 | *bp = *obp; | |
901 | } | |
902 | ||
60101509 BB |
903 | zgd->zgd_db = db; |
904 | zgd->zgd_bp = &lr->lr_blkptr; | |
905 | ||
906 | ASSERT(db != NULL); | |
907 | ASSERT(db->db_offset == offset); | |
908 | ASSERT(db->db_size == size); | |
909 | ||
910 | error = dmu_sync(zio, lr->lr_common.lrc_txg, | |
911 | zvol_get_done, zgd); | |
912 | ||
913 | if (error == 0) | |
914 | return (0); | |
915 | } | |
916 | } | |
917 | ||
918 | zvol_get_done(zgd, error); | |
919 | ||
ce37ebd2 | 920 | return (SET_ERROR(error)); |
60101509 BB |
921 | } |
922 | ||
923 | /* | |
924 | * The zvol_state_t's are inserted in increasing MINOR(dev_t) order. | |
925 | */ | |
926 | static void | |
927 | zvol_insert(zvol_state_t *zv_insert) | |
928 | { | |
929 | zvol_state_t *zv = NULL; | |
930 | ||
931 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
932 | ASSERT3U(MINOR(zv_insert->zv_dev) & ZVOL_MINOR_MASK, ==, 0); | |
933 | for (zv = list_head(&zvol_state_list); zv != NULL; | |
ce37ebd2 | 934 | zv = list_next(&zvol_state_list, zv)) { |
60101509 BB |
935 | if (MINOR(zv->zv_dev) > MINOR(zv_insert->zv_dev)) |
936 | break; | |
937 | } | |
938 | ||
939 | list_insert_before(&zvol_state_list, zv, zv_insert); | |
940 | } | |
941 | ||
942 | /* | |
943 | * Simply remove the zvol from to list of zvols. | |
944 | */ | |
945 | static void | |
946 | zvol_remove(zvol_state_t *zv_remove) | |
947 | { | |
948 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
949 | list_remove(&zvol_state_list, zv_remove); | |
950 | } | |
951 | ||
952 | static int | |
953 | zvol_first_open(zvol_state_t *zv) | |
954 | { | |
955 | objset_t *os; | |
956 | uint64_t volsize; | |
957 | int error; | |
958 | uint64_t ro; | |
959 | ||
960 | /* lie and say we're read-only */ | |
961 | error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, 1, zvol_tag, &os); | |
962 | if (error) | |
1ee159f4 BP |
963 | return (SET_ERROR(-error)); |
964 | ||
965 | zv->zv_objset = os; | |
966 | ||
967 | error = dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL); | |
968 | if (error) | |
969 | goto out_owned; | |
60101509 BB |
970 | |
971 | error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); | |
1ee159f4 BP |
972 | if (error) |
973 | goto out_owned; | |
60101509 | 974 | |
60101509 | 975 | error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf); |
1ee159f4 BP |
976 | if (error) |
977 | goto out_owned; | |
60101509 BB |
978 | |
979 | set_capacity(zv->zv_disk, volsize >> 9); | |
980 | zv->zv_volsize = volsize; | |
981 | zv->zv_zilog = zil_open(os, zvol_get_data); | |
982 | ||
a4430fce GW |
983 | if (ro || dmu_objset_is_snapshot(os) || |
984 | !spa_writeable(dmu_objset_spa(os))) { | |
babf3f9b MM |
985 | set_disk_ro(zv->zv_disk, 1); |
986 | zv->zv_flags |= ZVOL_RDONLY; | |
60101509 | 987 | } else { |
babf3f9b MM |
988 | set_disk_ro(zv->zv_disk, 0); |
989 | zv->zv_flags &= ~ZVOL_RDONLY; | |
60101509 BB |
990 | } |
991 | ||
1ee159f4 BP |
992 | out_owned: |
993 | if (error) { | |
994 | dmu_objset_disown(os, zvol_tag); | |
995 | zv->zv_objset = NULL; | |
996 | } | |
babf3f9b | 997 | |
ce37ebd2 | 998 | return (SET_ERROR(-error)); |
60101509 BB |
999 | } |
1000 | ||
1001 | static void | |
1002 | zvol_last_close(zvol_state_t *zv) | |
1003 | { | |
1004 | zil_close(zv->zv_zilog); | |
1005 | zv->zv_zilog = NULL; | |
04434775 | 1006 | |
60101509 BB |
1007 | dmu_buf_rele(zv->zv_dbuf, zvol_tag); |
1008 | zv->zv_dbuf = NULL; | |
04434775 MA |
1009 | |
1010 | /* | |
1011 | * Evict cached data | |
1012 | */ | |
1013 | if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) && | |
1014 | !(zv->zv_flags & ZVOL_RDONLY)) | |
1015 | txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0); | |
1016 | (void) dmu_objset_evict_dbufs(zv->zv_objset); | |
1017 | ||
60101509 BB |
1018 | dmu_objset_disown(zv->zv_objset, zvol_tag); |
1019 | zv->zv_objset = NULL; | |
1020 | } | |
1021 | ||
1022 | static int | |
1023 | zvol_open(struct block_device *bdev, fmode_t flag) | |
1024 | { | |
5428dc51 | 1025 | zvol_state_t *zv; |
60101509 BB |
1026 | int error = 0, drop_mutex = 0; |
1027 | ||
1028 | /* | |
1029 | * If the caller is already holding the mutex do not take it | |
a0bd735a | 1030 | * again, this will happen as part of zvol_create_minor_impl(). |
60101509 BB |
1031 | * Once add_disk() is called the device is live and the kernel |
1032 | * will attempt to open it to read the partition information. | |
1033 | */ | |
1034 | if (!mutex_owned(&zvol_state_lock)) { | |
1035 | mutex_enter(&zvol_state_lock); | |
1036 | drop_mutex = 1; | |
1037 | } | |
1038 | ||
5428dc51 BP |
1039 | /* |
1040 | * Obtain a copy of private_data under the lock to make sure | |
1041 | * that either the result of zvol_freeg() setting | |
1042 | * bdev->bd_disk->private_data to NULL is observed, or zvol_free() | |
1043 | * is not called on this zv because of the positive zv_open_count. | |
1044 | */ | |
1045 | zv = bdev->bd_disk->private_data; | |
1046 | if (zv == NULL) { | |
1047 | error = -ENXIO; | |
1048 | goto out_mutex; | |
1049 | } | |
60101509 BB |
1050 | |
1051 | if (zv->zv_open_count == 0) { | |
1052 | error = zvol_first_open(zv); | |
1053 | if (error) | |
1054 | goto out_mutex; | |
1055 | } | |
1056 | ||
ba6a2402 | 1057 | if ((flag & FMODE_WRITE) && (zv->zv_flags & ZVOL_RDONLY)) { |
60101509 BB |
1058 | error = -EROFS; |
1059 | goto out_open_count; | |
1060 | } | |
1061 | ||
1062 | zv->zv_open_count++; | |
1063 | ||
5428dc51 BP |
1064 | check_disk_change(bdev); |
1065 | ||
60101509 BB |
1066 | out_open_count: |
1067 | if (zv->zv_open_count == 0) | |
1068 | zvol_last_close(zv); | |
1069 | ||
1070 | out_mutex: | |
1071 | if (drop_mutex) | |
1072 | mutex_exit(&zvol_state_lock); | |
1073 | ||
ce37ebd2 | 1074 | return (SET_ERROR(error)); |
60101509 BB |
1075 | } |
1076 | ||
a1d9543a CD |
1077 | #ifdef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID |
1078 | static void | |
1079 | #else | |
60101509 | 1080 | static int |
a1d9543a | 1081 | #endif |
60101509 BB |
1082 | zvol_release(struct gendisk *disk, fmode_t mode) |
1083 | { | |
1084 | zvol_state_t *zv = disk->private_data; | |
1085 | int drop_mutex = 0; | |
1086 | ||
5428dc51 BP |
1087 | ASSERT(zv && zv->zv_open_count > 0); |
1088 | ||
60101509 BB |
1089 | if (!mutex_owned(&zvol_state_lock)) { |
1090 | mutex_enter(&zvol_state_lock); | |
1091 | drop_mutex = 1; | |
1092 | } | |
1093 | ||
5428dc51 BP |
1094 | zv->zv_open_count--; |
1095 | if (zv->zv_open_count == 0) | |
1096 | zvol_last_close(zv); | |
60101509 BB |
1097 | |
1098 | if (drop_mutex) | |
1099 | mutex_exit(&zvol_state_lock); | |
1100 | ||
a1d9543a | 1101 | #ifndef HAVE_BLOCK_DEVICE_OPERATIONS_RELEASE_VOID |
60101509 | 1102 | return (0); |
a1d9543a | 1103 | #endif |
60101509 BB |
1104 | } |
1105 | ||
1106 | static int | |
1107 | zvol_ioctl(struct block_device *bdev, fmode_t mode, | |
ce37ebd2 | 1108 | unsigned int cmd, unsigned long arg) |
60101509 BB |
1109 | { |
1110 | zvol_state_t *zv = bdev->bd_disk->private_data; | |
1111 | int error = 0; | |
1112 | ||
5428dc51 | 1113 | ASSERT(zv && zv->zv_open_count > 0); |
60101509 BB |
1114 | |
1115 | switch (cmd) { | |
1116 | case BLKFLSBUF: | |
1117 | zil_commit(zv->zv_zilog, ZVOL_OBJ); | |
1118 | break; | |
4c0d8e50 FN |
1119 | case BLKZNAME: |
1120 | error = copy_to_user((void *)arg, zv->zv_name, MAXNAMELEN); | |
1121 | break; | |
60101509 BB |
1122 | |
1123 | default: | |
1124 | error = -ENOTTY; | |
1125 | break; | |
1126 | ||
1127 | } | |
1128 | ||
ce37ebd2 | 1129 | return (SET_ERROR(error)); |
60101509 BB |
1130 | } |
1131 | ||
1132 | #ifdef CONFIG_COMPAT | |
1133 | static int | |
1134 | zvol_compat_ioctl(struct block_device *bdev, fmode_t mode, | |
ce37ebd2 | 1135 | unsigned cmd, unsigned long arg) |
60101509 | 1136 | { |
ce37ebd2 | 1137 | return (zvol_ioctl(bdev, mode, cmd, arg)); |
60101509 BB |
1138 | } |
1139 | #else | |
ce37ebd2 | 1140 | #define zvol_compat_ioctl NULL |
60101509 BB |
1141 | #endif |
1142 | ||
1143 | static int zvol_media_changed(struct gendisk *disk) | |
1144 | { | |
1145 | zvol_state_t *zv = disk->private_data; | |
1146 | ||
5428dc51 BP |
1147 | ASSERT(zv && zv->zv_open_count > 0); |
1148 | ||
ce37ebd2 | 1149 | return (zv->zv_changed); |
60101509 BB |
1150 | } |
1151 | ||
1152 | static int zvol_revalidate_disk(struct gendisk *disk) | |
1153 | { | |
1154 | zvol_state_t *zv = disk->private_data; | |
1155 | ||
5428dc51 BP |
1156 | ASSERT(zv && zv->zv_open_count > 0); |
1157 | ||
60101509 BB |
1158 | zv->zv_changed = 0; |
1159 | set_capacity(zv->zv_disk, zv->zv_volsize >> 9); | |
1160 | ||
ce37ebd2 | 1161 | return (0); |
60101509 BB |
1162 | } |
1163 | ||
1164 | /* | |
1165 | * Provide a simple virtual geometry for legacy compatibility. For devices | |
1166 | * smaller than 1 MiB a small head and sector count is used to allow very | |
1167 | * tiny devices. For devices over 1 Mib a standard head and sector count | |
1168 | * is used to keep the cylinders count reasonable. | |
1169 | */ | |
1170 | static int | |
1171 | zvol_getgeo(struct block_device *bdev, struct hd_geometry *geo) | |
1172 | { | |
1173 | zvol_state_t *zv = bdev->bd_disk->private_data; | |
5428dc51 BP |
1174 | sector_t sectors; |
1175 | ||
1176 | ASSERT(zv && zv->zv_open_count > 0); | |
1177 | ||
1178 | sectors = get_capacity(zv->zv_disk); | |
60101509 BB |
1179 | |
1180 | if (sectors > 2048) { | |
1181 | geo->heads = 16; | |
1182 | geo->sectors = 63; | |
1183 | } else { | |
1184 | geo->heads = 2; | |
1185 | geo->sectors = 4; | |
1186 | } | |
1187 | ||
1188 | geo->start = 0; | |
1189 | geo->cylinders = sectors / (geo->heads * geo->sectors); | |
1190 | ||
ce37ebd2 | 1191 | return (0); |
60101509 BB |
1192 | } |
1193 | ||
1194 | static struct kobject * | |
1195 | zvol_probe(dev_t dev, int *part, void *arg) | |
1196 | { | |
1197 | zvol_state_t *zv; | |
1198 | struct kobject *kobj; | |
1199 | ||
1200 | mutex_enter(&zvol_state_lock); | |
1201 | zv = zvol_find_by_dev(dev); | |
23a61ccc | 1202 | kobj = zv ? get_disk(zv->zv_disk) : NULL; |
60101509 BB |
1203 | mutex_exit(&zvol_state_lock); |
1204 | ||
ce37ebd2 | 1205 | return (kobj); |
60101509 BB |
1206 | } |
1207 | ||
1208 | #ifdef HAVE_BDEV_BLOCK_DEVICE_OPERATIONS | |
1209 | static struct block_device_operations zvol_ops = { | |
ce37ebd2 BB |
1210 | .open = zvol_open, |
1211 | .release = zvol_release, | |
1212 | .ioctl = zvol_ioctl, | |
1213 | .compat_ioctl = zvol_compat_ioctl, | |
1214 | .media_changed = zvol_media_changed, | |
1215 | .revalidate_disk = zvol_revalidate_disk, | |
1216 | .getgeo = zvol_getgeo, | |
1217 | .owner = THIS_MODULE, | |
60101509 BB |
1218 | }; |
1219 | ||
1220 | #else /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */ | |
1221 | ||
1222 | static int | |
1223 | zvol_open_by_inode(struct inode *inode, struct file *file) | |
1224 | { | |
ce37ebd2 | 1225 | return (zvol_open(inode->i_bdev, file->f_mode)); |
60101509 BB |
1226 | } |
1227 | ||
1228 | static int | |
1229 | zvol_release_by_inode(struct inode *inode, struct file *file) | |
1230 | { | |
ce37ebd2 | 1231 | return (zvol_release(inode->i_bdev->bd_disk, file->f_mode)); |
60101509 BB |
1232 | } |
1233 | ||
1234 | static int | |
1235 | zvol_ioctl_by_inode(struct inode *inode, struct file *file, | |
ce37ebd2 | 1236 | unsigned int cmd, unsigned long arg) |
60101509 | 1237 | { |
b1c58213 | 1238 | if (file == NULL || inode == NULL) |
ce37ebd2 BB |
1239 | return (SET_ERROR(-EINVAL)); |
1240 | ||
1241 | return (zvol_ioctl(inode->i_bdev, file->f_mode, cmd, arg)); | |
60101509 BB |
1242 | } |
1243 | ||
ce37ebd2 | 1244 | #ifdef CONFIG_COMPAT |
60101509 BB |
1245 | static long |
1246 | zvol_compat_ioctl_by_inode(struct file *file, | |
ce37ebd2 | 1247 | unsigned int cmd, unsigned long arg) |
60101509 | 1248 | { |
b1c58213 | 1249 | if (file == NULL) |
ce37ebd2 BB |
1250 | return (SET_ERROR(-EINVAL)); |
1251 | ||
1252 | return (zvol_compat_ioctl(file->f_dentry->d_inode->i_bdev, | |
1253 | file->f_mode, cmd, arg)); | |
60101509 | 1254 | } |
ce37ebd2 BB |
1255 | #else |
1256 | #define zvol_compat_ioctl_by_inode NULL | |
1257 | #endif | |
60101509 BB |
1258 | |
1259 | static struct block_device_operations zvol_ops = { | |
ce37ebd2 BB |
1260 | .open = zvol_open_by_inode, |
1261 | .release = zvol_release_by_inode, | |
1262 | .ioctl = zvol_ioctl_by_inode, | |
1263 | .compat_ioctl = zvol_compat_ioctl_by_inode, | |
1264 | .media_changed = zvol_media_changed, | |
1265 | .revalidate_disk = zvol_revalidate_disk, | |
1266 | .getgeo = zvol_getgeo, | |
1267 | .owner = THIS_MODULE, | |
60101509 BB |
1268 | }; |
1269 | #endif /* HAVE_BDEV_BLOCK_DEVICE_OPERATIONS */ | |
1270 | ||
1271 | /* | |
1272 | * Allocate memory for a new zvol_state_t and setup the required | |
1273 | * request queue and generic disk structures for the block device. | |
1274 | */ | |
1275 | static zvol_state_t * | |
1276 | zvol_alloc(dev_t dev, const char *name) | |
1277 | { | |
1278 | zvol_state_t *zv; | |
1279 | ||
79c76d5b | 1280 | zv = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP); |
60101509 | 1281 | |
2a3871d4 RY |
1282 | list_link_init(&zv->zv_next); |
1283 | ||
37f9dac5 | 1284 | zv->zv_queue = blk_alloc_queue(GFP_ATOMIC); |
60101509 BB |
1285 | if (zv->zv_queue == NULL) |
1286 | goto out_kmem; | |
1287 | ||
37f9dac5 | 1288 | blk_queue_make_request(zv->zv_queue, zvol_request); |
cf41432c | 1289 | blk_queue_set_write_cache(zv->zv_queue, B_TRUE, B_TRUE); |
b18019d2 | 1290 | |
60101509 BB |
1291 | zv->zv_disk = alloc_disk(ZVOL_MINORS); |
1292 | if (zv->zv_disk == NULL) | |
1293 | goto out_queue; | |
1294 | ||
1295 | zv->zv_queue->queuedata = zv; | |
1296 | zv->zv_dev = dev; | |
1297 | zv->zv_open_count = 0; | |
4c0d8e50 | 1298 | strlcpy(zv->zv_name, name, MAXNAMELEN); |
60101509 | 1299 | |
d88895a0 | 1300 | zfs_rlock_init(&zv->zv_range_lock); |
3c4988c8 | 1301 | |
60101509 BB |
1302 | zv->zv_disk->major = zvol_major; |
1303 | zv->zv_disk->first_minor = (dev & MINORMASK); | |
1304 | zv->zv_disk->fops = &zvol_ops; | |
1305 | zv->zv_disk->private_data = zv; | |
1306 | zv->zv_disk->queue = zv->zv_queue; | |
4c0d8e50 FN |
1307 | snprintf(zv->zv_disk->disk_name, DISK_NAME_LEN, "%s%d", |
1308 | ZVOL_DEV_NAME, (dev & MINORMASK)); | |
60101509 | 1309 | |
ce37ebd2 | 1310 | return (zv); |
60101509 BB |
1311 | |
1312 | out_queue: | |
1313 | blk_cleanup_queue(zv->zv_queue); | |
1314 | out_kmem: | |
1315 | kmem_free(zv, sizeof (zvol_state_t)); | |
0a6bef26 | 1316 | |
ce37ebd2 | 1317 | return (NULL); |
60101509 BB |
1318 | } |
1319 | ||
1320 | /* | |
1321 | * Cleanup then free a zvol_state_t which was created by zvol_alloc(). | |
1322 | */ | |
1323 | static void | |
1324 | zvol_free(zvol_state_t *zv) | |
1325 | { | |
5428dc51 BP |
1326 | ASSERT(MUTEX_HELD(&zvol_state_lock)); |
1327 | ASSERT(zv->zv_open_count == 0); | |
1328 | ||
d88895a0 | 1329 | zfs_rlock_destroy(&zv->zv_range_lock); |
60101509 | 1330 | |
5428dc51 BP |
1331 | zv->zv_disk->private_data = NULL; |
1332 | ||
60101509 BB |
1333 | del_gendisk(zv->zv_disk); |
1334 | blk_cleanup_queue(zv->zv_queue); | |
1335 | put_disk(zv->zv_disk); | |
1336 | ||
1337 | kmem_free(zv, sizeof (zvol_state_t)); | |
1338 | } | |
1339 | ||
a0bd735a BP |
1340 | /* |
1341 | * Create a block device minor node and setup the linkage between it | |
1342 | * and the specified volume. Once this function returns the block | |
1343 | * device is live and ready for use. | |
1344 | */ | |
60101509 | 1345 | static int |
a0bd735a | 1346 | zvol_create_minor_impl(const char *name) |
60101509 BB |
1347 | { |
1348 | zvol_state_t *zv; | |
1349 | objset_t *os; | |
1350 | dmu_object_info_t *doi; | |
1351 | uint64_t volsize; | |
9965059a | 1352 | uint64_t len; |
60101509 BB |
1353 | unsigned minor = 0; |
1354 | int error = 0; | |
1355 | ||
a0bd735a | 1356 | mutex_enter(&zvol_state_lock); |
60101509 BB |
1357 | |
1358 | zv = zvol_find_by_name(name); | |
1359 | if (zv) { | |
2e528b49 | 1360 | error = SET_ERROR(EEXIST); |
60101509 BB |
1361 | goto out; |
1362 | } | |
1363 | ||
79c76d5b | 1364 | doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP); |
60101509 BB |
1365 | |
1366 | error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, zvol_tag, &os); | |
1367 | if (error) | |
1368 | goto out_doi; | |
1369 | ||
1370 | error = dmu_object_info(os, ZVOL_OBJ, doi); | |
1371 | if (error) | |
1372 | goto out_dmu_objset_disown; | |
1373 | ||
1374 | error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize); | |
1375 | if (error) | |
1376 | goto out_dmu_objset_disown; | |
1377 | ||
1378 | error = zvol_find_minor(&minor); | |
1379 | if (error) | |
1380 | goto out_dmu_objset_disown; | |
1381 | ||
1382 | zv = zvol_alloc(MKDEV(zvol_major, minor), name); | |
1383 | if (zv == NULL) { | |
2e528b49 | 1384 | error = SET_ERROR(EAGAIN); |
60101509 BB |
1385 | goto out_dmu_objset_disown; |
1386 | } | |
1387 | ||
1388 | if (dmu_objset_is_snapshot(os)) | |
1389 | zv->zv_flags |= ZVOL_RDONLY; | |
1390 | ||
1391 | zv->zv_volblocksize = doi->doi_data_block_size; | |
1392 | zv->zv_volsize = volsize; | |
1393 | zv->zv_objset = os; | |
1394 | ||
1395 | set_capacity(zv->zv_disk, zv->zv_volsize >> 9); | |
1396 | ||
c495fe2c | 1397 | blk_queue_max_hw_sectors(zv->zv_queue, (DMU_MAX_ACCESS / 4) >> 9); |
34037afe ED |
1398 | blk_queue_max_segments(zv->zv_queue, UINT16_MAX); |
1399 | blk_queue_max_segment_size(zv->zv_queue, UINT_MAX); | |
1400 | blk_queue_physical_block_size(zv->zv_queue, zv->zv_volblocksize); | |
1401 | blk_queue_io_opt(zv->zv_queue, zv->zv_volblocksize); | |
7c0e5708 ED |
1402 | blk_queue_max_discard_sectors(zv->zv_queue, |
1403 | (zvol_max_discard_blocks * zv->zv_volblocksize) >> 9); | |
ee5fd0bb | 1404 | blk_queue_discard_granularity(zv->zv_queue, zv->zv_volblocksize); |
30930fba | 1405 | queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zv->zv_queue); |
37f9dac5 | 1406 | #ifdef QUEUE_FLAG_NONROT |
34037afe ED |
1407 | queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zv->zv_queue); |
1408 | #endif | |
c6a3a222 RY |
1409 | #ifdef QUEUE_FLAG_ADD_RANDOM |
1410 | queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zv->zv_queue); | |
1411 | #endif | |
34037afe | 1412 | |
a4430fce GW |
1413 | if (spa_writeable(dmu_objset_spa(os))) { |
1414 | if (zil_replay_disable) | |
1415 | zil_destroy(dmu_objset_zil(os), B_FALSE); | |
1416 | else | |
1417 | zil_replay(os, zv, zvol_replay_vector); | |
1418 | } | |
60101509 | 1419 | |
9965059a BB |
1420 | /* |
1421 | * When udev detects the addition of the device it will immediately | |
1422 | * invoke blkid(8) to determine the type of content on the device. | |
1423 | * Prefetching the blocks commonly scanned by blkid(8) will speed | |
1424 | * up this process. | |
1425 | */ | |
1426 | len = MIN(MAX(zvol_prefetch_bytes, 0), SPA_MAXBLOCKSIZE); | |
1427 | if (len > 0) { | |
fcff0f35 PD |
1428 | dmu_prefetch(os, ZVOL_OBJ, 0, 0, len, ZIO_PRIORITY_SYNC_READ); |
1429 | dmu_prefetch(os, ZVOL_OBJ, 0, volsize - len, len, | |
1430 | ZIO_PRIORITY_SYNC_READ); | |
9965059a BB |
1431 | } |
1432 | ||
f74a147c | 1433 | zv->zv_objset = NULL; |
60101509 BB |
1434 | out_dmu_objset_disown: |
1435 | dmu_objset_disown(os, zvol_tag); | |
60101509 | 1436 | out_doi: |
ce37ebd2 | 1437 | kmem_free(doi, sizeof (dmu_object_info_t)); |
60101509 BB |
1438 | out: |
1439 | ||
1440 | if (error == 0) { | |
1441 | zvol_insert(zv); | |
5428dc51 BP |
1442 | /* |
1443 | * Drop the lock to prevent deadlock with sys_open() -> | |
1444 | * zvol_open(), which first takes bd_disk->bd_mutex and then | |
1445 | * takes zvol_state_lock, whereas this code path first takes | |
1446 | * zvol_state_lock, and then takes bd_disk->bd_mutex. | |
1447 | */ | |
1448 | mutex_exit(&zvol_state_lock); | |
60101509 | 1449 | add_disk(zv->zv_disk); |
a0bd735a BP |
1450 | } else { |
1451 | mutex_exit(&zvol_state_lock); | |
60101509 BB |
1452 | } |
1453 | ||
ce37ebd2 | 1454 | return (SET_ERROR(error)); |
60101509 BB |
1455 | } |
1456 | ||
ba6a2402 BB |
1457 | /* |
1458 | * Rename a block device minor mode for the specified volume. | |
1459 | */ | |
1460 | static void | |
a0bd735a | 1461 | zvol_rename_minor(zvol_state_t *zv, const char *newname) |
ba6a2402 BB |
1462 | { |
1463 | int readonly = get_disk_ro(zv->zv_disk); | |
1464 | ||
1465 | ASSERT(MUTEX_HELD(&zvol_state_lock)); | |
1466 | ||
1467 | strlcpy(zv->zv_name, newname, sizeof (zv->zv_name)); | |
1468 | ||
1469 | /* | |
1470 | * The block device's read-only state is briefly changed causing | |
1471 | * a KOBJ_CHANGE uevent to be issued. This ensures udev detects | |
1472 | * the name change and fixes the symlinks. This does not change | |
1473 | * ZVOL_RDONLY in zv->zv_flags so the actual read-only state never | |
1474 | * changes. This would normally be done using kobject_uevent() but | |
1475 | * that is a GPL-only symbol which is why we need this workaround. | |
1476 | */ | |
1477 | set_disk_ro(zv->zv_disk, !readonly); | |
1478 | set_disk_ro(zv->zv_disk, readonly); | |
1479 | } | |
1480 | ||
a0bd735a BP |
1481 | |
1482 | /* | |
1483 | * Mask errors to continue dmu_objset_find() traversal | |
1484 | */ | |
1485 | static int | |
1486 | zvol_create_snap_minor_cb(const char *dsname, void *arg) | |
1487 | { | |
1488 | const char *name = (const char *)arg; | |
1489 | ||
1ee159f4 BP |
1490 | ASSERT0(MUTEX_HELD(&spa_namespace_lock)); |
1491 | ||
a0bd735a BP |
1492 | /* skip the designated dataset */ |
1493 | if (name && strcmp(dsname, name) == 0) | |
1494 | return (0); | |
1495 | ||
1496 | /* at this point, the dsname should name a snapshot */ | |
1497 | if (strchr(dsname, '@') == 0) { | |
1498 | dprintf("zvol_create_snap_minor_cb(): " | |
1499 | "%s is not a shapshot name\n", dsname); | |
1500 | } else { | |
1501 | (void) zvol_create_minor_impl(dsname); | |
1502 | } | |
1503 | ||
1504 | return (0); | |
1505 | } | |
1506 | ||
1507 | /* | |
1508 | * Mask errors to continue dmu_objset_find() traversal | |
1509 | */ | |
60101509 | 1510 | static int |
13fe0198 | 1511 | zvol_create_minors_cb(const char *dsname, void *arg) |
60101509 | 1512 | { |
a0bd735a BP |
1513 | uint64_t snapdev; |
1514 | int error; | |
1515 | ||
1ee159f4 BP |
1516 | ASSERT0(MUTEX_HELD(&spa_namespace_lock)); |
1517 | ||
a0bd735a BP |
1518 | error = dsl_prop_get_integer(dsname, "snapdev", &snapdev, NULL); |
1519 | if (error) | |
1520 | return (0); | |
1521 | ||
1522 | /* | |
1523 | * Given the name and the 'snapdev' property, create device minor nodes | |
1524 | * with the linkages to zvols/snapshots as needed. | |
1525 | * If the name represents a zvol, create a minor node for the zvol, then | |
1526 | * check if its snapshots are 'visible', and if so, iterate over the | |
1527 | * snapshots and create device minor nodes for those. | |
1528 | */ | |
1529 | if (strchr(dsname, '@') == 0) { | |
1530 | /* create minor for the 'dsname' explicitly */ | |
1531 | error = zvol_create_minor_impl(dsname); | |
1532 | if ((error == 0 || error == EEXIST) && | |
1533 | (snapdev == ZFS_SNAPDEV_VISIBLE)) { | |
1534 | fstrans_cookie_t cookie = spl_fstrans_mark(); | |
1535 | /* | |
1536 | * traverse snapshots only, do not traverse children, | |
1537 | * and skip the 'dsname' | |
1538 | */ | |
1539 | error = dmu_objset_find((char *)dsname, | |
1540 | zvol_create_snap_minor_cb, (void *)dsname, | |
1541 | DS_FIND_SNAPSHOTS); | |
1542 | spl_fstrans_unmark(cookie); | |
1543 | } | |
1544 | } else { | |
1545 | dprintf("zvol_create_minors_cb(): %s is not a zvol name\n", | |
1546 | dsname); | |
1547 | } | |
60101509 | 1548 | |
d5674448 | 1549 | return (0); |
60101509 BB |
1550 | } |
1551 | ||
1552 | /* | |
a0bd735a BP |
1553 | * Create minors for the specified dataset, including children and snapshots. |
1554 | * Pay attention to the 'snapdev' property and iterate over the snapshots | |
1555 | * only if they are 'visible'. This approach allows one to assure that the | |
1556 | * snapshot metadata is read from disk only if it is needed. | |
1557 | * | |
1558 | * The name can represent a dataset to be recursively scanned for zvols and | |
1559 | * their snapshots, or a single zvol snapshot. If the name represents a | |
1560 | * dataset, the scan is performed in two nested stages: | |
1561 | * - scan the dataset for zvols, and | |
1562 | * - for each zvol, create a minor node, then check if the zvol's snapshots | |
1563 | * are 'visible', and only then iterate over the snapshots if needed | |
1564 | * | |
1565 | * If the name represents a snapshot, a check is perfromed if the snapshot is | |
1566 | * 'visible' (which also verifies that the parent is a zvol), and if so, | |
1567 | * a minor node for that snapshot is created. | |
60101509 | 1568 | */ |
a0bd735a BP |
1569 | static int |
1570 | zvol_create_minors_impl(const char *name) | |
60101509 | 1571 | { |
60101509 | 1572 | int error = 0; |
5428dc51 | 1573 | fstrans_cookie_t cookie; |
a0bd735a | 1574 | char *atp, *parent; |
60101509 | 1575 | |
5428dc51 BP |
1576 | if (zvol_inhibit_dev) |
1577 | return (0); | |
1578 | ||
a0bd735a BP |
1579 | parent = kmem_alloc(MAXPATHLEN, KM_SLEEP); |
1580 | (void) strlcpy(parent, name, MAXPATHLEN); | |
1581 | ||
1582 | if ((atp = strrchr(parent, '@')) != NULL) { | |
1583 | uint64_t snapdev; | |
1584 | ||
1585 | *atp = '\0'; | |
1586 | error = dsl_prop_get_integer(parent, "snapdev", | |
1587 | &snapdev, NULL); | |
1588 | ||
1589 | if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE) | |
1590 | error = zvol_create_minor_impl(name); | |
1591 | } else { | |
1592 | cookie = spl_fstrans_mark(); | |
1593 | error = dmu_objset_find(parent, zvol_create_minors_cb, | |
1594 | NULL, DS_FIND_CHILDREN); | |
1595 | spl_fstrans_unmark(cookie); | |
1596 | } | |
1597 | ||
1598 | kmem_free(parent, MAXPATHLEN); | |
ba6a2402 BB |
1599 | |
1600 | return (SET_ERROR(error)); | |
1601 | } | |
1602 | ||
1603 | /* | |
1604 | * Remove minors for specified dataset including children and snapshots. | |
1605 | */ | |
a0bd735a BP |
1606 | static void |
1607 | zvol_remove_minors_impl(const char *name) | |
ba6a2402 BB |
1608 | { |
1609 | zvol_state_t *zv, *zv_next; | |
1610 | int namelen = ((name) ? strlen(name) : 0); | |
1611 | ||
74497b7a | 1612 | if (zvol_inhibit_dev) |
ba6a2402 | 1613 | return; |
74497b7a | 1614 | |
60101509 | 1615 | mutex_enter(&zvol_state_lock); |
ba6a2402 BB |
1616 | |
1617 | for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { | |
1618 | zv_next = list_next(&zvol_state_list, zv); | |
1619 | ||
1620 | if (name == NULL || strcmp(zv->zv_name, name) == 0 || | |
1621 | (strncmp(zv->zv_name, name, namelen) == 0 && | |
5428dc51 BP |
1622 | (zv->zv_name[namelen] == '/' || |
1623 | zv->zv_name[namelen] == '@'))) { | |
1624 | ||
1625 | /* If in use, leave alone */ | |
1626 | if (zv->zv_open_count > 0) | |
1627 | continue; | |
1628 | ||
ba6a2402 BB |
1629 | zvol_remove(zv); |
1630 | zvol_free(zv); | |
60101509 | 1631 | } |
60101509 | 1632 | } |
60101509 | 1633 | |
ba6a2402 | 1634 | mutex_exit(&zvol_state_lock); |
60101509 BB |
1635 | } |
1636 | ||
a0bd735a BP |
1637 | /* Remove minor for this specific snapshot only */ |
1638 | static void | |
1639 | zvol_remove_minor_impl(const char *name) | |
1640 | { | |
1641 | zvol_state_t *zv, *zv_next; | |
1642 | ||
1643 | if (zvol_inhibit_dev) | |
1644 | return; | |
1645 | ||
1646 | if (strchr(name, '@') == NULL) | |
1647 | return; | |
1648 | ||
1649 | mutex_enter(&zvol_state_lock); | |
1650 | ||
1651 | for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { | |
1652 | zv_next = list_next(&zvol_state_list, zv); | |
1653 | ||
1654 | if (strcmp(zv->zv_name, name) == 0) { | |
1655 | /* If in use, leave alone */ | |
1656 | if (zv->zv_open_count > 0) | |
1657 | continue; | |
1658 | zvol_remove(zv); | |
1659 | zvol_free(zv); | |
1660 | break; | |
1661 | } | |
1662 | } | |
1663 | ||
1664 | mutex_exit(&zvol_state_lock); | |
1665 | } | |
1666 | ||
60101509 | 1667 | /* |
ba6a2402 | 1668 | * Rename minors for specified dataset including children and snapshots. |
60101509 | 1669 | */ |
a0bd735a BP |
1670 | static void |
1671 | zvol_rename_minors_impl(const char *oldname, const char *newname) | |
60101509 BB |
1672 | { |
1673 | zvol_state_t *zv, *zv_next; | |
ba6a2402 BB |
1674 | int oldnamelen, newnamelen; |
1675 | char *name; | |
60101509 | 1676 | |
74497b7a DH |
1677 | if (zvol_inhibit_dev) |
1678 | return; | |
1679 | ||
ba6a2402 BB |
1680 | oldnamelen = strlen(oldname); |
1681 | newnamelen = strlen(newname); | |
79c76d5b | 1682 | name = kmem_alloc(MAXNAMELEN, KM_SLEEP); |
60101509 BB |
1683 | |
1684 | mutex_enter(&zvol_state_lock); | |
ba6a2402 | 1685 | |
60101509 BB |
1686 | for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) { |
1687 | zv_next = list_next(&zvol_state_list, zv); | |
1688 | ||
5428dc51 BP |
1689 | /* If in use, leave alone */ |
1690 | if (zv->zv_open_count > 0) | |
1691 | continue; | |
1692 | ||
ba6a2402 | 1693 | if (strcmp(zv->zv_name, oldname) == 0) { |
a0bd735a | 1694 | zvol_rename_minor(zv, newname); |
ba6a2402 BB |
1695 | } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 && |
1696 | (zv->zv_name[oldnamelen] == '/' || | |
1697 | zv->zv_name[oldnamelen] == '@')) { | |
1698 | snprintf(name, MAXNAMELEN, "%s%c%s", newname, | |
1699 | zv->zv_name[oldnamelen], | |
1700 | zv->zv_name + oldnamelen + 1); | |
a0bd735a | 1701 | zvol_rename_minor(zv, name); |
60101509 BB |
1702 | } |
1703 | } | |
ba6a2402 | 1704 | |
60101509 | 1705 | mutex_exit(&zvol_state_lock); |
ba6a2402 BB |
1706 | |
1707 | kmem_free(name, MAXNAMELEN); | |
60101509 BB |
1708 | } |
1709 | ||
a0bd735a BP |
1710 | typedef struct zvol_snapdev_cb_arg { |
1711 | uint64_t snapdev; | |
1712 | } zvol_snapdev_cb_arg_t; | |
1713 | ||
0b4d1b58 | 1714 | static int |
a0bd735a BP |
1715 | zvol_set_snapdev_cb(const char *dsname, void *param) { |
1716 | zvol_snapdev_cb_arg_t *arg = param; | |
0b4d1b58 ED |
1717 | |
1718 | if (strchr(dsname, '@') == NULL) | |
ba6a2402 | 1719 | return (0); |
0b4d1b58 | 1720 | |
a0bd735a | 1721 | switch (arg->snapdev) { |
0b4d1b58 | 1722 | case ZFS_SNAPDEV_VISIBLE: |
a0bd735a | 1723 | (void) zvol_create_minor_impl(dsname); |
0b4d1b58 ED |
1724 | break; |
1725 | case ZFS_SNAPDEV_HIDDEN: | |
a0bd735a | 1726 | (void) zvol_remove_minor_impl(dsname); |
0b4d1b58 ED |
1727 | break; |
1728 | } | |
ba6a2402 BB |
1729 | |
1730 | return (0); | |
0b4d1b58 ED |
1731 | } |
1732 | ||
a0bd735a BP |
1733 | static void |
1734 | zvol_set_snapdev_impl(char *name, uint64_t snapdev) | |
1735 | { | |
1736 | zvol_snapdev_cb_arg_t arg = {snapdev}; | |
1737 | fstrans_cookie_t cookie = spl_fstrans_mark(); | |
1738 | /* | |
1739 | * The zvol_set_snapdev_sync() sets snapdev appropriately | |
1740 | * in the dataset hierarchy. Here, we only scan snapshots. | |
1741 | */ | |
1742 | dmu_objset_find(name, zvol_set_snapdev_cb, &arg, DS_FIND_SNAPSHOTS); | |
1743 | spl_fstrans_unmark(cookie); | |
1744 | } | |
1745 | ||
1746 | static zvol_task_t * | |
1747 | zvol_task_alloc(zvol_async_op_t op, const char *name1, const char *name2, | |
1748 | uint64_t snapdev) | |
1749 | { | |
1750 | zvol_task_t *task; | |
1751 | char *delim; | |
1752 | ||
1753 | /* Never allow tasks on hidden names. */ | |
1754 | if (name1[0] == '$') | |
1755 | return (NULL); | |
1756 | ||
1757 | task = kmem_zalloc(sizeof (zvol_task_t), KM_SLEEP); | |
1758 | task->op = op; | |
1759 | task->snapdev = snapdev; | |
1760 | delim = strchr(name1, '/'); | |
1761 | strlcpy(task->pool, name1, delim ? (delim - name1 + 1) : MAXNAMELEN); | |
1762 | ||
1763 | strlcpy(task->name1, name1, MAXNAMELEN); | |
1764 | if (name2 != NULL) | |
1765 | strlcpy(task->name2, name2, MAXNAMELEN); | |
1766 | ||
1767 | return (task); | |
1768 | } | |
1769 | ||
1770 | static void | |
1771 | zvol_task_free(zvol_task_t *task) | |
1772 | { | |
1773 | kmem_free(task, sizeof (zvol_task_t)); | |
1774 | } | |
1775 | ||
1776 | /* | |
1777 | * The worker thread function performed asynchronously. | |
1778 | */ | |
1779 | static void | |
1780 | zvol_task_cb(void *param) | |
1781 | { | |
1782 | zvol_task_t *task = (zvol_task_t *)param; | |
1783 | ||
1784 | switch (task->op) { | |
1785 | case ZVOL_ASYNC_CREATE_MINORS: | |
1786 | (void) zvol_create_minors_impl(task->name1); | |
1787 | break; | |
1788 | case ZVOL_ASYNC_REMOVE_MINORS: | |
1789 | zvol_remove_minors_impl(task->name1); | |
1790 | break; | |
1791 | case ZVOL_ASYNC_RENAME_MINORS: | |
1792 | zvol_rename_minors_impl(task->name1, task->name2); | |
1793 | break; | |
1794 | case ZVOL_ASYNC_SET_SNAPDEV: | |
1795 | zvol_set_snapdev_impl(task->name1, task->snapdev); | |
1796 | break; | |
1797 | default: | |
1798 | VERIFY(0); | |
1799 | break; | |
1800 | } | |
1801 | ||
1802 | zvol_task_free(task); | |
1803 | } | |
1804 | ||
1805 | typedef struct zvol_set_snapdev_arg { | |
1806 | const char *zsda_name; | |
1807 | uint64_t zsda_value; | |
1808 | zprop_source_t zsda_source; | |
1809 | dmu_tx_t *zsda_tx; | |
1810 | } zvol_set_snapdev_arg_t; | |
1811 | ||
1812 | /* | |
1813 | * Sanity check the dataset for safe use by the sync task. No additional | |
1814 | * conditions are imposed. | |
1815 | */ | |
1816 | static int | |
1817 | zvol_set_snapdev_check(void *arg, dmu_tx_t *tx) | |
1818 | { | |
1819 | zvol_set_snapdev_arg_t *zsda = arg; | |
1820 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1821 | dsl_dir_t *dd; | |
1822 | int error; | |
1823 | ||
1824 | error = dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL); | |
1825 | if (error != 0) | |
1826 | return (error); | |
1827 | ||
1828 | dsl_dir_rele(dd, FTAG); | |
1829 | ||
1830 | return (error); | |
1831 | } | |
1832 | ||
1833 | static int | |
1834 | zvol_set_snapdev_sync_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg) | |
1835 | { | |
1836 | zvol_set_snapdev_arg_t *zsda = arg; | |
1837 | char dsname[MAXNAMELEN]; | |
1838 | zvol_task_t *task; | |
1839 | ||
1840 | dsl_dataset_name(ds, dsname); | |
1841 | dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_SNAPDEV), | |
1842 | zsda->zsda_source, sizeof (zsda->zsda_value), 1, | |
1843 | &zsda->zsda_value, zsda->zsda_tx); | |
1844 | ||
1845 | task = zvol_task_alloc(ZVOL_ASYNC_SET_SNAPDEV, dsname, | |
1846 | NULL, zsda->zsda_value); | |
1847 | if (task == NULL) | |
1848 | return (0); | |
1849 | ||
1850 | (void) taskq_dispatch(dp->dp_spa->spa_zvol_taskq, zvol_task_cb, | |
1851 | task, TQ_SLEEP); | |
1852 | return (0); | |
1853 | } | |
1854 | ||
1855 | /* | |
1856 | * Traverse all child snapshot datasets and apply snapdev appropriately. | |
1857 | */ | |
1858 | static void | |
1859 | zvol_set_snapdev_sync(void *arg, dmu_tx_t *tx) | |
1860 | { | |
1861 | zvol_set_snapdev_arg_t *zsda = arg; | |
1862 | dsl_pool_t *dp = dmu_tx_pool(tx); | |
1863 | dsl_dir_t *dd; | |
1864 | ||
1865 | VERIFY0(dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL)); | |
1866 | zsda->zsda_tx = tx; | |
1867 | ||
1868 | dmu_objset_find_dp(dp, dd->dd_object, zvol_set_snapdev_sync_cb, | |
1869 | zsda, DS_FIND_CHILDREN); | |
1870 | ||
1871 | dsl_dir_rele(dd, FTAG); | |
1872 | } | |
1873 | ||
0b4d1b58 | 1874 | int |
a0bd735a BP |
1875 | zvol_set_snapdev(const char *ddname, zprop_source_t source, uint64_t snapdev) |
1876 | { | |
1877 | zvol_set_snapdev_arg_t zsda; | |
5428dc51 | 1878 | |
a0bd735a BP |
1879 | zsda.zsda_name = ddname; |
1880 | zsda.zsda_source = source; | |
1881 | zsda.zsda_value = snapdev; | |
5428dc51 | 1882 | |
a0bd735a BP |
1883 | return (dsl_sync_task(ddname, zvol_set_snapdev_check, |
1884 | zvol_set_snapdev_sync, &zsda, 0, ZFS_SPACE_CHECK_NONE)); | |
1885 | } | |
1886 | ||
1887 | void | |
1888 | zvol_create_minors(spa_t *spa, const char *name, boolean_t async) | |
1889 | { | |
1890 | zvol_task_t *task; | |
1891 | taskqid_t id; | |
1892 | ||
1893 | task = zvol_task_alloc(ZVOL_ASYNC_CREATE_MINORS, name, NULL, ~0ULL); | |
1894 | if (task == NULL) | |
1895 | return; | |
1896 | ||
1897 | id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP); | |
48d3eb40 | 1898 | if ((async == B_FALSE) && (id != TASKQID_INVALID)) |
a0bd735a BP |
1899 | taskq_wait_id(spa->spa_zvol_taskq, id); |
1900 | } | |
1901 | ||
1902 | void | |
1903 | zvol_remove_minors(spa_t *spa, const char *name, boolean_t async) | |
1904 | { | |
1905 | zvol_task_t *task; | |
1906 | taskqid_t id; | |
1907 | ||
1908 | task = zvol_task_alloc(ZVOL_ASYNC_REMOVE_MINORS, name, NULL, ~0ULL); | |
1909 | if (task == NULL) | |
1910 | return; | |
5428dc51 | 1911 | |
a0bd735a | 1912 | id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP); |
48d3eb40 | 1913 | if ((async == B_FALSE) && (id != TASKQID_INVALID)) |
a0bd735a BP |
1914 | taskq_wait_id(spa->spa_zvol_taskq, id); |
1915 | } | |
1916 | ||
1917 | void | |
1918 | zvol_rename_minors(spa_t *spa, const char *name1, const char *name2, | |
1919 | boolean_t async) | |
1920 | { | |
1921 | zvol_task_t *task; | |
1922 | taskqid_t id; | |
1923 | ||
1924 | task = zvol_task_alloc(ZVOL_ASYNC_RENAME_MINORS, name1, name2, ~0ULL); | |
1925 | if (task == NULL) | |
1926 | return; | |
1927 | ||
1928 | id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP); | |
48d3eb40 | 1929 | if ((async == B_FALSE) && (id != TASKQID_INVALID)) |
a0bd735a | 1930 | taskq_wait_id(spa->spa_zvol_taskq, id); |
0b4d1b58 ED |
1931 | } |
1932 | ||
60101509 BB |
1933 | int |
1934 | zvol_init(void) | |
1935 | { | |
1936 | int error; | |
1937 | ||
2a3871d4 | 1938 | list_create(&zvol_state_list, sizeof (zvol_state_t), |
ce37ebd2 | 1939 | offsetof(zvol_state_t, zv_next)); |
2a3871d4 RY |
1940 | mutex_init(&zvol_state_lock, NULL, MUTEX_DEFAULT, NULL); |
1941 | ||
60101509 BB |
1942 | error = register_blkdev(zvol_major, ZVOL_DRIVER); |
1943 | if (error) { | |
1944 | printk(KERN_INFO "ZFS: register_blkdev() failed %d\n", error); | |
37f9dac5 | 1945 | goto out; |
60101509 BB |
1946 | } |
1947 | ||
1948 | blk_register_region(MKDEV(zvol_major, 0), 1UL << MINORBITS, | |
ce37ebd2 | 1949 | THIS_MODULE, zvol_probe, NULL, NULL); |
60101509 | 1950 | |
60101509 | 1951 | return (0); |
2a3871d4 | 1952 | |
37f9dac5 | 1953 | out: |
2a3871d4 RY |
1954 | mutex_destroy(&zvol_state_lock); |
1955 | list_destroy(&zvol_state_list); | |
1956 | ||
ce37ebd2 | 1957 | return (SET_ERROR(error)); |
60101509 BB |
1958 | } |
1959 | ||
1960 | void | |
1961 | zvol_fini(void) | |
1962 | { | |
a0bd735a BP |
1963 | zvol_remove_minors_impl(NULL); |
1964 | ||
60101509 BB |
1965 | blk_unregister_region(MKDEV(zvol_major, 0), 1UL << MINORBITS); |
1966 | unregister_blkdev(zvol_major, ZVOL_DRIVER); | |
a0bd735a | 1967 | |
60101509 | 1968 | list_destroy(&zvol_state_list); |
a0bd735a | 1969 | mutex_destroy(&zvol_state_lock); |
60101509 BB |
1970 | } |
1971 | ||
74497b7a DH |
1972 | module_param(zvol_inhibit_dev, uint, 0644); |
1973 | MODULE_PARM_DESC(zvol_inhibit_dev, "Do not create zvol device nodes"); | |
1974 | ||
30a9524e | 1975 | module_param(zvol_major, uint, 0444); |
60101509 BB |
1976 | MODULE_PARM_DESC(zvol_major, "Major number for zvol device"); |
1977 | ||
7c0e5708 | 1978 | module_param(zvol_max_discard_blocks, ulong, 0444); |
ce37ebd2 | 1979 | MODULE_PARM_DESC(zvol_max_discard_blocks, "Max number of blocks to discard"); |
9965059a BB |
1980 | |
1981 | module_param(zvol_prefetch_bytes, uint, 0644); | |
1982 | MODULE_PARM_DESC(zvol_prefetch_bytes, "Prefetch N bytes at zvol start+end"); |