1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3 * CDDL HEADER START
4 *
5 * The contents of this file are subject to the terms of the
6 * Common Development and Distribution License (the "License").
7 * You may not use this file except in compliance with the License.
8 *
9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10 * or https://opensource.org/licenses/CDDL-1.0.
11 * See the License for the specific language governing permissions
12 * and limitations under the License.
13 *
14 * When distributing Covered Code, include this CDDL HEADER in each
15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16 * If applicable, add the following below this CDDL HEADER, with the
17 * fields enclosed by brackets "[]" replaced with your own identifying
18 * information: Portions Copyright [yyyy] [name of copyright owner]
19 *
20 * CDDL HEADER END
21 */
22 /*
23 * Copyright (C) 2008-2010 Lawrence Livermore National Security, LLC.
24 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
25 * Rewritten for Linux by Brian Behlendorf <behlendorf1@llnl.gov>.
26 * LLNL-CODE-403049.
27 *
28 * ZFS volume emulation driver.
29 *
30 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
31 * Volumes are accessed through the symbolic links named:
32 *
33 * /dev/<pool_name>/<dataset_name>
34 *
35 * Volumes are persistent through reboot and module load. No user command
36 * needs to be run before opening and using a device.
37 *
38 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
39 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
40 * Copyright (c) 2012, 2019 by Delphix. All rights reserved.
41 * Copyright (c) 2024, Klara, Inc.
42 */
43
44 /*
45 * Note on locking of zvol state structures.
46 *
47 * These structures are used to maintain internal state used to emulate block
48 * devices on top of zvols. In particular, management of device minor number
49 * operations - create, remove, rename, and set_snapdev - involves access to
50 * these structures. The zvol_state_lock is primarily used to protect the
51 * zvol_state_list. The zv->zv_state_lock is used to protect the contents
52 * of the zvol_state_t structures, as well as to make sure that when the
53 * time comes to remove the structure from the list, it is not in use, and
54 * therefore, it can be taken off zvol_state_list and freed.
55 *
56 * The zv_suspend_lock was introduced to allow for suspending I/O to a zvol,
57 * e.g. for the duration of receive and rollback operations. This lock can be
58 * held for significant periods of time. Given that it is undesirable to hold
59 * mutexes for long periods of time, the following lock ordering applies:
60 * - take zvol_state_lock if necessary, to protect zvol_state_list
61 * - take zv_suspend_lock if necessary, by the code path in question
62 * - take zv_state_lock to protect zvol_state_t
63 *
64 * The minor operations are issued to spa->spa_zvol_taskq queues, that are
65 * single-threaded (to preserve order of minor operations), and are executed
66 * through the zvol_task_cb that dispatches the specific operations. Therefore,
67 * these operations are serialized per pool. Consequently, we can be certain
68 * that for a given zvol, there is only one operation at a time in progress.
69 * That is why one can be sure that first, zvol_state_t for a given zvol is
70 * allocated and placed on zvol_state_list, and then other minor operations
71 * for this zvol are going to proceed in the order of issue.
72 *
73 */
74
75 #include <sys/dataset_kstats.h>
76 #include <sys/dbuf.h>
77 #include <sys/dmu_traverse.h>
78 #include <sys/dsl_dataset.h>
79 #include <sys/dsl_prop.h>
80 #include <sys/dsl_dir.h>
81 #include <sys/zap.h>
82 #include <sys/zfeature.h>
83 #include <sys/zil_impl.h>
84 #include <sys/dmu_tx.h>
85 #include <sys/zio.h>
86 #include <sys/zfs_rlock.h>
87 #include <sys/spa_impl.h>
88 #include <sys/zvol.h>
89 #include <sys/zvol_impl.h>
90
91 unsigned int zvol_inhibit_dev = 0;
92 unsigned int zvol_volmode = ZFS_VOLMODE_GEOM;
93
94 struct hlist_head *zvol_htable;
95 static list_t zvol_state_list;
96 krwlock_t zvol_state_lock;
97 extern int zfs_bclone_wait_dirty;
98
99 typedef enum {
100 ZVOL_ASYNC_REMOVE_MINORS,
101 ZVOL_ASYNC_RENAME_MINORS,
102 ZVOL_ASYNC_SET_SNAPDEV,
103 ZVOL_ASYNC_SET_VOLMODE,
104 ZVOL_ASYNC_MAX
105 } zvol_async_op_t;
106
107 typedef struct {
108 zvol_async_op_t op;
109 char name1[MAXNAMELEN];
110 char name2[MAXNAMELEN];
111 uint64_t value;
112 } zvol_task_t;
113
114 uint64_t
zvol_name_hash(const char * name)115 zvol_name_hash(const char *name)
116 {
117 uint64_t crc = -1ULL;
118 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
119 for (const uint8_t *p = (const uint8_t *)name; *p != 0; p++)
120 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (*p)) & 0xFF];
121 return (crc);
122 }
123
124 /*
125 * Find a zvol_state_t given the name and hash generated by zvol_name_hash.
126 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
127 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
128 * before zv_state_lock. The mode argument indicates the mode (including none)
129 * for zv_suspend_lock to be taken.
130 */
131 zvol_state_t *
zvol_find_by_name_hash(const char * name,uint64_t hash,int mode)132 zvol_find_by_name_hash(const char *name, uint64_t hash, int mode)
133 {
134 zvol_state_t *zv;
135 struct hlist_node *p = NULL;
136
137 rw_enter(&zvol_state_lock, RW_READER);
138 hlist_for_each(p, ZVOL_HT_HEAD(hash)) {
139 zv = hlist_entry(p, zvol_state_t, zv_hlink);
140 mutex_enter(&zv->zv_state_lock);
141 if (zv->zv_hash == hash && strcmp(zv->zv_name, name) == 0) {
142 /*
143 * this is the right zvol, take the locks in the
144 * right order
145 */
146 if (mode != RW_NONE &&
147 !rw_tryenter(&zv->zv_suspend_lock, mode)) {
148 mutex_exit(&zv->zv_state_lock);
149 rw_enter(&zv->zv_suspend_lock, mode);
150 mutex_enter(&zv->zv_state_lock);
151 /*
152 * zvol cannot be renamed as we continue
153 * to hold zvol_state_lock
154 */
155 ASSERT(zv->zv_hash == hash &&
156 strcmp(zv->zv_name, name) == 0);
157 }
158 rw_exit(&zvol_state_lock);
159 return (zv);
160 }
161 mutex_exit(&zv->zv_state_lock);
162 }
163 rw_exit(&zvol_state_lock);
164
165 return (NULL);
166 }
167
168 /*
169 * Find a zvol_state_t given the name.
170 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
171 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
172 * before zv_state_lock. The mode argument indicates the mode (including none)
173 * for zv_suspend_lock to be taken.
174 */
175 static zvol_state_t *
zvol_find_by_name(const char * name,int mode)176 zvol_find_by_name(const char *name, int mode)
177 {
178 return (zvol_find_by_name_hash(name, zvol_name_hash(name), mode));
179 }
180
181 /*
182 * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation.
183 */
184 void
zvol_create_cb(objset_t * os,void * arg,cred_t * cr,dmu_tx_t * tx)185 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
186 {
187 zfs_creat_t *zct = arg;
188 nvlist_t *nvprops = zct->zct_props;
189 int error;
190 uint64_t volblocksize, volsize;
191
192 VERIFY(nvlist_lookup_uint64(nvprops,
193 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
194 if (nvlist_lookup_uint64(nvprops,
195 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
196 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
197
198 /*
199 * These properties must be removed from the list so the generic
200 * property setting step won't apply to them.
201 */
202 VERIFY(nvlist_remove_all(nvprops,
203 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
204 (void) nvlist_remove_all(nvprops,
205 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
206
207 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
208 DMU_OT_NONE, 0, tx);
209 ASSERT(error == 0);
210
211 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
212 DMU_OT_NONE, 0, tx);
213 ASSERT(error == 0);
214
215 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
216 ASSERT(error == 0);
217 }
218
219 /*
220 * ZFS_IOC_OBJSET_STATS entry point.
221 */
222 int
zvol_get_stats(objset_t * os,nvlist_t * nv)223 zvol_get_stats(objset_t *os, nvlist_t *nv)
224 {
225 int error;
226 dmu_object_info_t *doi;
227 uint64_t val;
228
229 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
230 if (error)
231 return (SET_ERROR(error));
232
233 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
234 doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP);
235 error = dmu_object_info(os, ZVOL_OBJ, doi);
236
237 if (error == 0) {
238 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
239 doi->doi_data_block_size);
240 }
241
242 kmem_free(doi, sizeof (dmu_object_info_t));
243
244 return (SET_ERROR(error));
245 }
246
247 /*
248 * Sanity check volume size.
249 */
250 int
zvol_check_volsize(uint64_t volsize,uint64_t blocksize)251 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
252 {
253 if (volsize == 0)
254 return (SET_ERROR(EINVAL));
255
256 if (volsize % blocksize != 0)
257 return (SET_ERROR(EINVAL));
258
259 #ifdef _ILP32
260 if (volsize - 1 > SPEC_MAXOFFSET_T)
261 return (SET_ERROR(EOVERFLOW));
262 #endif
263 return (0);
264 }
265
266 /*
267 * Ensure the zap is flushed then inform the VFS of the capacity change.
268 */
269 static int
zvol_update_volsize(uint64_t volsize,objset_t * os)270 zvol_update_volsize(uint64_t volsize, objset_t *os)
271 {
272 dmu_tx_t *tx;
273 int error;
274 uint64_t txg;
275
276 tx = dmu_tx_create(os);
277 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
278 dmu_tx_mark_netfree(tx);
279 error = dmu_tx_assign(tx, DMU_TX_WAIT);
280 if (error) {
281 dmu_tx_abort(tx);
282 return (SET_ERROR(error));
283 }
284 txg = dmu_tx_get_txg(tx);
285
286 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
287 &volsize, tx);
288 dmu_tx_commit(tx);
289
290 txg_wait_synced(dmu_objset_pool(os), txg);
291
292 if (error == 0)
293 error = dmu_free_long_range(os,
294 ZVOL_OBJ, volsize, DMU_OBJECT_END);
295
296 return (error);
297 }
298
299 /*
300 * Set ZFS_PROP_VOLSIZE set entry point. Note that modifying the volume
301 * size will result in a udev "change" event being generated.
302 */
303 int
zvol_set_volsize(const char * name,uint64_t volsize)304 zvol_set_volsize(const char *name, uint64_t volsize)
305 {
306 objset_t *os = NULL;
307 uint64_t readonly;
308 int error;
309 boolean_t owned = B_FALSE;
310
311 error = dsl_prop_get_integer(name,
312 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
313 if (error != 0)
314 return (SET_ERROR(error));
315 if (readonly)
316 return (SET_ERROR(EROFS));
317
318 zvol_state_t *zv = zvol_find_by_name(name, RW_READER);
319
320 ASSERT(zv == NULL || (MUTEX_HELD(&zv->zv_state_lock) &&
321 RW_READ_HELD(&zv->zv_suspend_lock)));
322
323 if (zv == NULL || zv->zv_objset == NULL) {
324 if (zv != NULL)
325 rw_exit(&zv->zv_suspend_lock);
326 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, B_TRUE,
327 FTAG, &os)) != 0) {
328 if (zv != NULL)
329 mutex_exit(&zv->zv_state_lock);
330 return (SET_ERROR(error));
331 }
332 owned = B_TRUE;
333 if (zv != NULL)
334 zv->zv_objset = os;
335 } else {
336 os = zv->zv_objset;
337 }
338
339 dmu_object_info_t *doi = kmem_alloc(sizeof (*doi), KM_SLEEP);
340
341 if ((error = dmu_object_info(os, ZVOL_OBJ, doi)) ||
342 (error = zvol_check_volsize(volsize, doi->doi_data_block_size)))
343 goto out;
344
345 error = zvol_update_volsize(volsize, os);
346 if (error == 0 && zv != NULL) {
347 zv->zv_volsize = volsize;
348 zv->zv_changed = 1;
349 }
350 out:
351 kmem_free(doi, sizeof (dmu_object_info_t));
352
353 if (owned) {
354 dmu_objset_disown(os, B_TRUE, FTAG);
355 if (zv != NULL)
356 zv->zv_objset = NULL;
357 } else {
358 rw_exit(&zv->zv_suspend_lock);
359 }
360
361 if (zv != NULL)
362 mutex_exit(&zv->zv_state_lock);
363
364 if (error == 0 && zv != NULL)
365 zvol_os_update_volsize(zv, volsize);
366
367 return (SET_ERROR(error));
368 }
369
370 /*
371 * Update volthreading.
372 */
373 int
zvol_set_volthreading(const char * name,boolean_t value)374 zvol_set_volthreading(const char *name, boolean_t value)
375 {
376 zvol_state_t *zv = zvol_find_by_name(name, RW_NONE);
377 if (zv == NULL)
378 return (ENOENT);
379 zv->zv_threading = value;
380 mutex_exit(&zv->zv_state_lock);
381 return (0);
382 }
383
384 /*
385 * Update zvol ro property.
386 */
387 int
zvol_set_ro(const char * name,boolean_t value)388 zvol_set_ro(const char *name, boolean_t value)
389 {
390 zvol_state_t *zv = zvol_find_by_name(name, RW_NONE);
391 if (zv == NULL)
392 return (-1);
393 if (value) {
394 zvol_os_set_disk_ro(zv, 1);
395 zv->zv_flags |= ZVOL_RDONLY;
396 } else {
397 zvol_os_set_disk_ro(zv, 0);
398 zv->zv_flags &= ~ZVOL_RDONLY;
399 }
400 mutex_exit(&zv->zv_state_lock);
401 return (0);
402 }
403
404 /*
405 * Sanity check volume block size.
406 */
407 int
zvol_check_volblocksize(const char * name,uint64_t volblocksize)408 zvol_check_volblocksize(const char *name, uint64_t volblocksize)
409 {
410 /* Record sizes above 128k need the feature to be enabled */
411 if (volblocksize > SPA_OLD_MAXBLOCKSIZE) {
412 spa_t *spa;
413 int error;
414
415 if ((error = spa_open(name, &spa, FTAG)) != 0)
416 return (error);
417
418 if (!spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
419 spa_close(spa, FTAG);
420 return (SET_ERROR(ENOTSUP));
421 }
422
423 /*
424 * We don't allow setting the property above 1MB,
425 * unless the tunable has been changed.
426 */
427 if (volblocksize > zfs_max_recordsize)
428 return (SET_ERROR(EDOM));
429
430 spa_close(spa, FTAG);
431 }
432
433 if (volblocksize < SPA_MINBLOCKSIZE ||
434 volblocksize > SPA_MAXBLOCKSIZE ||
435 !ISP2(volblocksize))
436 return (SET_ERROR(EDOM));
437
438 return (0);
439 }
440
441 /*
442 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
443 * implement DKIOCFREE/free-long-range.
444 */
445 static int
zvol_replay_truncate(void * arg1,void * arg2,boolean_t byteswap)446 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
447 {
448 zvol_state_t *zv = arg1;
449 lr_truncate_t *lr = arg2;
450 uint64_t offset, length;
451
452 ASSERT3U(lr->lr_common.lrc_reclen, >=, sizeof (*lr));
453
454 if (byteswap)
455 byteswap_uint64_array(lr, sizeof (*lr));
456
457 offset = lr->lr_offset;
458 length = lr->lr_length;
459
460 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
461 dmu_tx_mark_netfree(tx);
462 int error = dmu_tx_assign(tx, DMU_TX_WAIT);
463 if (error != 0) {
464 dmu_tx_abort(tx);
465 } else {
466 (void) zil_replaying(zv->zv_zilog, tx);
467 dmu_tx_commit(tx);
468 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset,
469 length);
470 }
471
472 return (error);
473 }
474
475 /*
476 * Replay a TX_WRITE ZIL transaction that didn't get committed
477 * after a system failure
478 */
479 static int
zvol_replay_write(void * arg1,void * arg2,boolean_t byteswap)480 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap)
481 {
482 zvol_state_t *zv = arg1;
483 lr_write_t *lr = arg2;
484 objset_t *os = zv->zv_objset;
485 char *data = (char *)(lr + 1); /* data follows lr_write_t */
486 uint64_t offset, length;
487 dmu_tx_t *tx;
488 int error;
489
490 ASSERT3U(lr->lr_common.lrc_reclen, >=, sizeof (*lr));
491
492 if (byteswap)
493 byteswap_uint64_array(lr, sizeof (*lr));
494
495 offset = lr->lr_offset;
496 length = lr->lr_length;
497
498 /* If it's a dmu_sync() block, write the whole block */
499 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
500 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
501 if (length < blocksize) {
502 offset -= offset % blocksize;
503 length = blocksize;
504 }
505 }
506
507 tx = dmu_tx_create(os);
508 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
509 error = dmu_tx_assign(tx, DMU_TX_WAIT);
510 if (error) {
511 dmu_tx_abort(tx);
512 } else {
513 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
514 (void) zil_replaying(zv->zv_zilog, tx);
515 dmu_tx_commit(tx);
516 }
517
518 return (error);
519 }
520
521 /*
522 * Replay a TX_CLONE_RANGE ZIL transaction that didn't get committed
523 * after a system failure
524 */
525 static int
zvol_replay_clone_range(void * arg1,void * arg2,boolean_t byteswap)526 zvol_replay_clone_range(void *arg1, void *arg2, boolean_t byteswap)
527 {
528 zvol_state_t *zv = arg1;
529 lr_clone_range_t *lr = arg2;
530 objset_t *os = zv->zv_objset;
531 dmu_tx_t *tx;
532 int error;
533 uint64_t blksz;
534 uint64_t off;
535 uint64_t len;
536
537 ASSERT3U(lr->lr_common.lrc_reclen, >=, sizeof (*lr));
538 ASSERT3U(lr->lr_common.lrc_reclen, >=, offsetof(lr_clone_range_t,
539 lr_bps[lr->lr_nbps]));
540
541 if (byteswap)
542 byteswap_uint64_array(lr, sizeof (*lr));
543
544 ASSERT(spa_feature_is_enabled(dmu_objset_spa(os),
545 SPA_FEATURE_BLOCK_CLONING));
546
547 off = lr->lr_offset;
548 len = lr->lr_length;
549 blksz = lr->lr_blksz;
550
551 if ((off % blksz) != 0) {
552 return (SET_ERROR(EINVAL));
553 }
554
555 error = dnode_hold(os, ZVOL_OBJ, zv, &zv->zv_dn);
556 if (error != 0 || !zv->zv_dn)
557 return (error);
558 tx = dmu_tx_create(os);
559 dmu_tx_hold_clone_by_dnode(tx, zv->zv_dn, off, len);
560 error = dmu_tx_assign(tx, DMU_TX_WAIT);
561 if (error != 0) {
562 dmu_tx_abort(tx);
563 goto out;
564 }
565 error = dmu_brt_clone(zv->zv_objset, ZVOL_OBJ, off, len,
566 tx, lr->lr_bps, lr->lr_nbps);
567 if (error != 0) {
568 dmu_tx_commit(tx);
569 goto out;
570 }
571
572 /*
573 * zil_replaying() not only check if we are replaying ZIL, but also
574 * updates the ZIL header to record replay progress.
575 */
576 VERIFY(zil_replaying(zv->zv_zilog, tx));
577 dmu_tx_commit(tx);
578
579 out:
580 dnode_rele(zv->zv_dn, zv);
581 zv->zv_dn = NULL;
582 return (error);
583 }
584
585 int
zvol_clone_range(zvol_state_t * zv_src,uint64_t inoff,zvol_state_t * zv_dst,uint64_t outoff,uint64_t len)586 zvol_clone_range(zvol_state_t *zv_src, uint64_t inoff, zvol_state_t *zv_dst,
587 uint64_t outoff, uint64_t len)
588 {
589 zilog_t *zilog_dst;
590 zfs_locked_range_t *inlr, *outlr;
591 objset_t *inos, *outos;
592 dmu_tx_t *tx;
593 blkptr_t *bps;
594 size_t maxblocks;
595 int error = EINVAL;
596
597 rw_enter(&zv_dst->zv_suspend_lock, RW_READER);
598 if (zv_dst->zv_zilog == NULL) {
599 rw_exit(&zv_dst->zv_suspend_lock);
600 rw_enter(&zv_dst->zv_suspend_lock, RW_WRITER);
601 if (zv_dst->zv_zilog == NULL) {
602 zv_dst->zv_zilog = zil_open(zv_dst->zv_objset,
603 zvol_get_data, &zv_dst->zv_kstat.dk_zil_sums);
604 zv_dst->zv_flags |= ZVOL_WRITTEN_TO;
605 VERIFY0((zv_dst->zv_zilog->zl_header->zh_flags &
606 ZIL_REPLAY_NEEDED));
607 }
608 rw_downgrade(&zv_dst->zv_suspend_lock);
609 }
610 if (zv_src != zv_dst)
611 rw_enter(&zv_src->zv_suspend_lock, RW_READER);
612
613 inos = zv_src->zv_objset;
614 outos = zv_dst->zv_objset;
615
616 /*
617 * Sanity checks
618 */
619 if (!spa_feature_is_enabled(dmu_objset_spa(outos),
620 SPA_FEATURE_BLOCK_CLONING)) {
621 error = EOPNOTSUPP;
622 goto out;
623 }
624 if (dmu_objset_spa(inos) != dmu_objset_spa(outos)) {
625 error = EXDEV;
626 goto out;
627 }
628 if (inos->os_encrypted != outos->os_encrypted) {
629 error = EXDEV;
630 goto out;
631 }
632 if (zv_src->zv_volblocksize != zv_dst->zv_volblocksize) {
633 error = EINVAL;
634 goto out;
635 }
636 if (inoff >= zv_src->zv_volsize || outoff >= zv_dst->zv_volsize) {
637 error = 0;
638 goto out;
639 }
640
641 /*
642 * Do not read beyond boundary
643 */
644 if (len > zv_src->zv_volsize - inoff)
645 len = zv_src->zv_volsize - inoff;
646 if (len > zv_dst->zv_volsize - outoff)
647 len = zv_dst->zv_volsize - outoff;
648 if (len == 0) {
649 error = 0;
650 goto out;
651 }
652
653 /*
654 * No overlapping if we are cloning within the same file
655 */
656 if (zv_src == zv_dst) {
657 if (inoff < outoff + len && outoff < inoff + len) {
658 error = EINVAL;
659 goto out;
660 }
661 }
662
663 /*
664 * Offsets and length must be at block boundaries
665 */
666 if ((inoff % zv_src->zv_volblocksize) != 0 ||
667 (outoff % zv_dst->zv_volblocksize) != 0) {
668 error = EINVAL;
669 goto out;
670 }
671
672 /*
673 * Length must be multiple of block size
674 */
675 if ((len % zv_src->zv_volblocksize) != 0) {
676 error = EINVAL;
677 goto out;
678 }
679
680 zilog_dst = zv_dst->zv_zilog;
681 maxblocks = zil_max_log_data(zilog_dst, sizeof (lr_clone_range_t)) /
682 sizeof (bps[0]);
683 bps = vmem_alloc(sizeof (bps[0]) * maxblocks, KM_SLEEP);
684 /*
685 * Maintain predictable lock order.
686 */
687 if (zv_src < zv_dst || (zv_src == zv_dst && inoff < outoff)) {
688 inlr = zfs_rangelock_enter(&zv_src->zv_rangelock, inoff, len,
689 RL_READER);
690 outlr = zfs_rangelock_enter(&zv_dst->zv_rangelock, outoff, len,
691 RL_WRITER);
692 } else {
693 outlr = zfs_rangelock_enter(&zv_dst->zv_rangelock, outoff, len,
694 RL_WRITER);
695 inlr = zfs_rangelock_enter(&zv_src->zv_rangelock, inoff, len,
696 RL_READER);
697 }
698
699 while (len > 0) {
700 uint64_t size, last_synced_txg;
701 size_t nbps = maxblocks;
702 size = MIN(zv_src->zv_volblocksize * maxblocks, len);
703 last_synced_txg = spa_last_synced_txg(
704 dmu_objset_spa(zv_src->zv_objset));
705 error = dmu_read_l0_bps(zv_src->zv_objset, ZVOL_OBJ, inoff,
706 size, bps, &nbps);
707 if (error != 0) {
708 /*
709 * If we are trying to clone a block that was created
710 * in the current transaction group, the error will be
711 * EAGAIN here. Based on zfs_bclone_wait_dirty either
712 * return a shortened range to the caller so it can
713 * fallback, or wait for the next TXG and check again.
714 */
715 if (error == EAGAIN && zfs_bclone_wait_dirty) {
716 txg_wait_synced(dmu_objset_pool
717 (zv_src->zv_objset), last_synced_txg + 1);
718 continue;
719 }
720 break;
721 }
722
723 tx = dmu_tx_create(zv_dst->zv_objset);
724 dmu_tx_hold_clone_by_dnode(tx, zv_dst->zv_dn, outoff, size);
725 error = dmu_tx_assign(tx, DMU_TX_WAIT);
726 if (error != 0) {
727 dmu_tx_abort(tx);
728 break;
729 }
730 error = dmu_brt_clone(zv_dst->zv_objset, ZVOL_OBJ, outoff, size,
731 tx, bps, nbps);
732 if (error != 0) {
733 dmu_tx_commit(tx);
734 break;
735 }
736 zvol_log_clone_range(zilog_dst, tx, TX_CLONE_RANGE, outoff,
737 size, zv_src->zv_volblocksize, bps, nbps);
738 dmu_tx_commit(tx);
739 inoff += size;
740 outoff += size;
741 len -= size;
742 }
743 vmem_free(bps, sizeof (bps[0]) * maxblocks);
744 zfs_rangelock_exit(outlr);
745 zfs_rangelock_exit(inlr);
746 if (error == 0 && zv_dst->zv_objset->os_sync == ZFS_SYNC_ALWAYS) {
747 zil_commit(zilog_dst, ZVOL_OBJ);
748 }
749 out:
750 if (zv_src != zv_dst)
751 rw_exit(&zv_src->zv_suspend_lock);
752 rw_exit(&zv_dst->zv_suspend_lock);
753 return (SET_ERROR(error));
754 }
755
756 /*
757 * Handles TX_CLONE_RANGE transactions.
758 */
759 void
zvol_log_clone_range(zilog_t * zilog,dmu_tx_t * tx,int txtype,uint64_t off,uint64_t len,uint64_t blksz,const blkptr_t * bps,size_t nbps)760 zvol_log_clone_range(zilog_t *zilog, dmu_tx_t *tx, int txtype, uint64_t off,
761 uint64_t len, uint64_t blksz, const blkptr_t *bps, size_t nbps)
762 {
763 itx_t *itx;
764 lr_clone_range_t *lr;
765 uint64_t partlen, max_log_data;
766 size_t partnbps;
767
768 if (zil_replaying(zilog, tx))
769 return;
770
771 max_log_data = zil_max_log_data(zilog, sizeof (lr_clone_range_t));
772
773 while (nbps > 0) {
774 partnbps = MIN(nbps, max_log_data / sizeof (bps[0]));
775 partlen = partnbps * blksz;
776 ASSERT3U(partlen, <, len + blksz);
777 partlen = MIN(partlen, len);
778
779 itx = zil_itx_create(txtype,
780 sizeof (*lr) + sizeof (bps[0]) * partnbps);
781 lr = (lr_clone_range_t *)&itx->itx_lr;
782 lr->lr_foid = ZVOL_OBJ;
783 lr->lr_offset = off;
784 lr->lr_length = partlen;
785 lr->lr_blksz = blksz;
786 lr->lr_nbps = partnbps;
787 memcpy(lr->lr_bps, bps, sizeof (bps[0]) * partnbps);
788
789 zil_itx_assign(zilog, itx, tx);
790
791 bps += partnbps;
792 ASSERT3U(nbps, >=, partnbps);
793 nbps -= partnbps;
794 off += partlen;
795 ASSERT3U(len, >=, partlen);
796 len -= partlen;
797 }
798 }
799
800 static int
zvol_replay_err(void * arg1,void * arg2,boolean_t byteswap)801 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap)
802 {
803 (void) arg1, (void) arg2, (void) byteswap;
804 return (SET_ERROR(ENOTSUP));
805 }
806
807 /*
808 * Callback vectors for replaying records.
809 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
810 */
811 zil_replay_func_t *const zvol_replay_vector[TX_MAX_TYPE] = {
812 zvol_replay_err, /* no such transaction type */
813 zvol_replay_err, /* TX_CREATE */
814 zvol_replay_err, /* TX_MKDIR */
815 zvol_replay_err, /* TX_MKXATTR */
816 zvol_replay_err, /* TX_SYMLINK */
817 zvol_replay_err, /* TX_REMOVE */
818 zvol_replay_err, /* TX_RMDIR */
819 zvol_replay_err, /* TX_LINK */
820 zvol_replay_err, /* TX_RENAME */
821 zvol_replay_write, /* TX_WRITE */
822 zvol_replay_truncate, /* TX_TRUNCATE */
823 zvol_replay_err, /* TX_SETATTR */
824 zvol_replay_err, /* TX_ACL_V0 */
825 zvol_replay_err, /* TX_ACL */
826 zvol_replay_err, /* TX_CREATE_ACL */
827 zvol_replay_err, /* TX_CREATE_ATTR */
828 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
829 zvol_replay_err, /* TX_MKDIR_ACL */
830 zvol_replay_err, /* TX_MKDIR_ATTR */
831 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
832 zvol_replay_err, /* TX_WRITE2 */
833 zvol_replay_err, /* TX_SETSAXATTR */
834 zvol_replay_err, /* TX_RENAME_EXCHANGE */
835 zvol_replay_err, /* TX_RENAME_WHITEOUT */
836 zvol_replay_clone_range, /* TX_CLONE_RANGE */
837 };
838
839 /*
840 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
841 *
842 * We store data in the log buffers if it's small enough.
843 * Otherwise we will later flush the data out via dmu_sync().
844 */
845 static const ssize_t zvol_immediate_write_sz = 32768;
846
847 void
zvol_log_write(zvol_state_t * zv,dmu_tx_t * tx,uint64_t offset,uint64_t size,boolean_t commit)848 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, uint64_t offset,
849 uint64_t size, boolean_t commit)
850 {
851 uint32_t blocksize = zv->zv_volblocksize;
852 zilog_t *zilog = zv->zv_zilog;
853 itx_wr_state_t write_state;
854 uint64_t sz = size;
855
856 if (zil_replaying(zilog, tx))
857 return;
858
859 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
860 write_state = WR_INDIRECT;
861 else if (!spa_has_slogs(zilog->zl_spa) &&
862 size >= blocksize && blocksize > zvol_immediate_write_sz)
863 write_state = WR_INDIRECT;
864 else if (commit)
865 write_state = WR_COPIED;
866 else
867 write_state = WR_NEED_COPY;
868
869 while (size) {
870 itx_t *itx;
871 lr_write_t *lr;
872 itx_wr_state_t wr_state = write_state;
873 ssize_t len = size;
874
875 if (wr_state == WR_COPIED && size > zil_max_copied_data(zilog))
876 wr_state = WR_NEED_COPY;
877 else if (wr_state == WR_INDIRECT)
878 len = MIN(blocksize - P2PHASE(offset, blocksize), size);
879
880 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
881 (wr_state == WR_COPIED ? len : 0));
882 lr = (lr_write_t *)&itx->itx_lr;
883 if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
884 offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) {
885 zil_itx_destroy(itx);
886 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
887 lr = (lr_write_t *)&itx->itx_lr;
888 wr_state = WR_NEED_COPY;
889 }
890
891 itx->itx_wr_state = wr_state;
892 lr->lr_foid = ZVOL_OBJ;
893 lr->lr_offset = offset;
894 lr->lr_length = len;
895 lr->lr_blkoff = 0;
896 BP_ZERO(&lr->lr_blkptr);
897
898 itx->itx_private = zv;
899
900 (void) zil_itx_assign(zilog, itx, tx);
901
902 offset += len;
903 size -= len;
904 }
905
906 if (write_state == WR_COPIED || write_state == WR_NEED_COPY) {
907 dsl_pool_wrlog_count(zilog->zl_dmu_pool, sz, tx->tx_txg);
908 }
909 }
910
911 /*
912 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
913 */
914 void
zvol_log_truncate(zvol_state_t * zv,dmu_tx_t * tx,uint64_t off,uint64_t len)915 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len)
916 {
917 itx_t *itx;
918 lr_truncate_t *lr;
919 zilog_t *zilog = zv->zv_zilog;
920
921 if (zil_replaying(zilog, tx))
922 return;
923
924 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
925 lr = (lr_truncate_t *)&itx->itx_lr;
926 lr->lr_foid = ZVOL_OBJ;
927 lr->lr_offset = off;
928 lr->lr_length = len;
929
930 zil_itx_assign(zilog, itx, tx);
931 }
932
933
934 static void
zvol_get_done(zgd_t * zgd,int error)935 zvol_get_done(zgd_t *zgd, int error)
936 {
937 (void) error;
938 if (zgd->zgd_db)
939 dmu_buf_rele(zgd->zgd_db, zgd);
940
941 zfs_rangelock_exit(zgd->zgd_lr);
942
943 kmem_free(zgd, sizeof (zgd_t));
944 }
945
946 /*
947 * Get data to generate a TX_WRITE intent log record.
948 */
949 int
zvol_get_data(void * arg,uint64_t arg2,lr_write_t * lr,char * buf,struct lwb * lwb,zio_t * zio)950 zvol_get_data(void *arg, uint64_t arg2, lr_write_t *lr, char *buf,
951 struct lwb *lwb, zio_t *zio)
952 {
953 zvol_state_t *zv = arg;
954 uint64_t offset = lr->lr_offset;
955 uint64_t size = lr->lr_length;
956 dmu_buf_t *db;
957 zgd_t *zgd;
958 int error;
959
960 ASSERT3P(lwb, !=, NULL);
961 ASSERT3U(size, !=, 0);
962
963 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
964 zgd->zgd_lwb = lwb;
965
966 /*
967 * Write records come in two flavors: immediate and indirect.
968 * For small writes it's cheaper to store the data with the
969 * log record (immediate); for large writes it's cheaper to
970 * sync the data and get a pointer to it (indirect) so that
971 * we don't have to write the data twice.
972 */
973 if (buf != NULL) { /* immediate write */
974 zgd->zgd_lr = zfs_rangelock_enter(&zv->zv_rangelock, offset,
975 size, RL_READER);
976 error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
977 DMU_READ_NO_PREFETCH);
978 } else { /* indirect write */
979 ASSERT3P(zio, !=, NULL);
980 /*
981 * Have to lock the whole block to ensure when it's written out
982 * and its checksum is being calculated that no one can change
983 * the data. Contrarily to zfs_get_data we need not re-check
984 * blocksize after we get the lock because it cannot be changed.
985 */
986 size = zv->zv_volblocksize;
987 offset = P2ALIGN_TYPED(offset, size, uint64_t);
988 zgd->zgd_lr = zfs_rangelock_enter(&zv->zv_rangelock, offset,
989 size, RL_READER);
990 error = dmu_buf_hold_noread_by_dnode(zv->zv_dn, offset, zgd,
991 &db);
992 if (error == 0) {
993 blkptr_t *bp = &lr->lr_blkptr;
994
995 zgd->zgd_db = db;
996 zgd->zgd_bp = bp;
997
998 ASSERT(db != NULL);
999 ASSERT(db->db_offset == offset);
1000 ASSERT(db->db_size == size);
1001
1002 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1003 zvol_get_done, zgd);
1004
1005 if (error == 0)
1006 return (0);
1007 }
1008 }
1009
1010 zvol_get_done(zgd, error);
1011
1012 return (SET_ERROR(error));
1013 }
1014
1015 /*
1016 * The zvol_state_t's are inserted into zvol_state_list and zvol_htable.
1017 */
1018
1019 void
zvol_insert(zvol_state_t * zv)1020 zvol_insert(zvol_state_t *zv)
1021 {
1022 ASSERT(RW_WRITE_HELD(&zvol_state_lock));
1023 list_insert_head(&zvol_state_list, zv);
1024 hlist_add_head(&zv->zv_hlink, ZVOL_HT_HEAD(zv->zv_hash));
1025 }
1026
1027 /*
1028 * Simply remove the zvol from to list of zvols.
1029 */
1030 static void
zvol_remove(zvol_state_t * zv)1031 zvol_remove(zvol_state_t *zv)
1032 {
1033 ASSERT(RW_WRITE_HELD(&zvol_state_lock));
1034 list_remove(&zvol_state_list, zv);
1035 hlist_del(&zv->zv_hlink);
1036 }
1037
1038 /*
1039 * Setup zv after we just own the zv->objset
1040 */
1041 static int
zvol_setup_zv(zvol_state_t * zv)1042 zvol_setup_zv(zvol_state_t *zv)
1043 {
1044 uint64_t volsize;
1045 int error;
1046 uint64_t ro;
1047 objset_t *os = zv->zv_objset;
1048
1049 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1050 ASSERT(RW_LOCK_HELD(&zv->zv_suspend_lock));
1051
1052 zv->zv_zilog = NULL;
1053 zv->zv_flags &= ~ZVOL_WRITTEN_TO;
1054
1055 error = dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL);
1056 if (error)
1057 return (SET_ERROR(error));
1058
1059 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
1060 if (error)
1061 return (SET_ERROR(error));
1062
1063 error = dnode_hold(os, ZVOL_OBJ, zv, &zv->zv_dn);
1064 if (error)
1065 return (SET_ERROR(error));
1066
1067 zvol_os_set_capacity(zv, volsize >> 9);
1068 zv->zv_volsize = volsize;
1069
1070 if (ro || dmu_objset_is_snapshot(os) ||
1071 !spa_writeable(dmu_objset_spa(os))) {
1072 zvol_os_set_disk_ro(zv, 1);
1073 zv->zv_flags |= ZVOL_RDONLY;
1074 } else {
1075 zvol_os_set_disk_ro(zv, 0);
1076 zv->zv_flags &= ~ZVOL_RDONLY;
1077 }
1078 return (0);
1079 }
1080
1081 /*
1082 * Shutdown every zv_objset related stuff except zv_objset itself.
1083 * The is the reverse of zvol_setup_zv.
1084 */
1085 static void
zvol_shutdown_zv(zvol_state_t * zv)1086 zvol_shutdown_zv(zvol_state_t *zv)
1087 {
1088 ASSERT(MUTEX_HELD(&zv->zv_state_lock) &&
1089 RW_LOCK_HELD(&zv->zv_suspend_lock));
1090
1091 if (zv->zv_flags & ZVOL_WRITTEN_TO) {
1092 ASSERT(zv->zv_zilog != NULL);
1093 zil_close(zv->zv_zilog);
1094 }
1095
1096 zv->zv_zilog = NULL;
1097
1098 dnode_rele(zv->zv_dn, zv);
1099 zv->zv_dn = NULL;
1100
1101 /*
1102 * Evict cached data. We must write out any dirty data before
1103 * disowning the dataset.
1104 */
1105 if (zv->zv_flags & ZVOL_WRITTEN_TO)
1106 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1107 (void) dmu_objset_evict_dbufs(zv->zv_objset);
1108 }
1109
1110 /*
1111 * return the proper tag for rollback and recv
1112 */
1113 void *
zvol_tag(zvol_state_t * zv)1114 zvol_tag(zvol_state_t *zv)
1115 {
1116 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
1117 return (zv->zv_open_count > 0 ? zv : NULL);
1118 }
1119
1120 /*
1121 * Suspend the zvol for recv and rollback.
1122 */
1123 zvol_state_t *
zvol_suspend(const char * name)1124 zvol_suspend(const char *name)
1125 {
1126 zvol_state_t *zv;
1127
1128 zv = zvol_find_by_name(name, RW_WRITER);
1129
1130 if (zv == NULL)
1131 return (NULL);
1132
1133 /* block all I/O, release in zvol_resume. */
1134 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1135 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
1136
1137 atomic_inc(&zv->zv_suspend_ref);
1138
1139 if (zv->zv_open_count > 0)
1140 zvol_shutdown_zv(zv);
1141
1142 /*
1143 * do not hold zv_state_lock across suspend/resume to
1144 * avoid locking up zvol lookups
1145 */
1146 mutex_exit(&zv->zv_state_lock);
1147
1148 /* zv_suspend_lock is released in zvol_resume() */
1149 return (zv);
1150 }
1151
1152 int
zvol_resume(zvol_state_t * zv)1153 zvol_resume(zvol_state_t *zv)
1154 {
1155 int error = 0;
1156
1157 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
1158
1159 mutex_enter(&zv->zv_state_lock);
1160
1161 if (zv->zv_open_count > 0) {
1162 VERIFY0(dmu_objset_hold(zv->zv_name, zv, &zv->zv_objset));
1163 VERIFY3P(zv->zv_objset->os_dsl_dataset->ds_owner, ==, zv);
1164 VERIFY(dsl_dataset_long_held(zv->zv_objset->os_dsl_dataset));
1165 dmu_objset_rele(zv->zv_objset, zv);
1166
1167 error = zvol_setup_zv(zv);
1168 }
1169
1170 mutex_exit(&zv->zv_state_lock);
1171
1172 rw_exit(&zv->zv_suspend_lock);
1173 /*
1174 * We need this because we don't hold zvol_state_lock while releasing
1175 * zv_suspend_lock. zvol_remove_minors_impl thus cannot check
1176 * zv_suspend_lock to determine it is safe to free because rwlock is
1177 * not inherent atomic.
1178 */
1179 atomic_dec(&zv->zv_suspend_ref);
1180
1181 if (zv->zv_flags & ZVOL_REMOVING)
1182 cv_broadcast(&zv->zv_removing_cv);
1183
1184 return (SET_ERROR(error));
1185 }
1186
1187 int
zvol_first_open(zvol_state_t * zv,boolean_t readonly)1188 zvol_first_open(zvol_state_t *zv, boolean_t readonly)
1189 {
1190 objset_t *os;
1191 int error;
1192
1193 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock));
1194 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1195 ASSERT(mutex_owned(&spa_namespace_lock));
1196
1197 boolean_t ro = (readonly || (strchr(zv->zv_name, '@') != NULL));
1198 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, ro, B_TRUE, zv, &os);
1199 if (error)
1200 return (SET_ERROR(error));
1201
1202 zv->zv_objset = os;
1203
1204 error = zvol_setup_zv(zv);
1205 if (error) {
1206 dmu_objset_disown(os, 1, zv);
1207 zv->zv_objset = NULL;
1208 }
1209
1210 return (error);
1211 }
1212
1213 void
zvol_last_close(zvol_state_t * zv)1214 zvol_last_close(zvol_state_t *zv)
1215 {
1216 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock));
1217 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1218
1219 if (zv->zv_flags & ZVOL_REMOVING)
1220 cv_broadcast(&zv->zv_removing_cv);
1221
1222 zvol_shutdown_zv(zv);
1223
1224 dmu_objset_disown(zv->zv_objset, 1, zv);
1225 zv->zv_objset = NULL;
1226 }
1227
1228 typedef struct minors_job {
1229 list_t *list;
1230 list_node_t link;
1231 /* input */
1232 char *name;
1233 /* output */
1234 int error;
1235 } minors_job_t;
1236
1237 /*
1238 * Prefetch zvol dnodes for the minors_job
1239 */
1240 static void
zvol_prefetch_minors_impl(void * arg)1241 zvol_prefetch_minors_impl(void *arg)
1242 {
1243 minors_job_t *job = arg;
1244 char *dsname = job->name;
1245 objset_t *os = NULL;
1246
1247 job->error = dmu_objset_own(dsname, DMU_OST_ZVOL, B_TRUE, B_TRUE,
1248 FTAG, &os);
1249 if (job->error == 0) {
1250 dmu_prefetch_dnode(os, ZVOL_OBJ, ZIO_PRIORITY_SYNC_READ);
1251 dmu_objset_disown(os, B_TRUE, FTAG);
1252 }
1253 }
1254
1255 /*
1256 * Mask errors to continue dmu_objset_find() traversal
1257 */
1258 static int
zvol_create_snap_minor_cb(const char * dsname,void * arg)1259 zvol_create_snap_minor_cb(const char *dsname, void *arg)
1260 {
1261 minors_job_t *j = arg;
1262 list_t *minors_list = j->list;
1263 const char *name = j->name;
1264
1265 ASSERT0(MUTEX_HELD(&spa_namespace_lock));
1266
1267 /* skip the designated dataset */
1268 if (name && strcmp(dsname, name) == 0)
1269 return (0);
1270
1271 /* at this point, the dsname should name a snapshot */
1272 if (strchr(dsname, '@') == 0) {
1273 dprintf("zvol_create_snap_minor_cb(): "
1274 "%s is not a snapshot name\n", dsname);
1275 } else {
1276 minors_job_t *job;
1277 char *n = kmem_strdup(dsname);
1278 if (n == NULL)
1279 return (0);
1280
1281 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP);
1282 job->name = n;
1283 job->list = minors_list;
1284 job->error = 0;
1285 list_insert_tail(minors_list, job);
1286 /* don't care if dispatch fails, because job->error is 0 */
1287 taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job,
1288 TQ_SLEEP);
1289 }
1290
1291 return (0);
1292 }
1293
1294 /*
1295 * If spa_keystore_load_wkey() is called for an encrypted zvol,
1296 * we need to look for any clones also using the key. This function
1297 * is "best effort" - so we just skip over it if there are failures.
1298 */
1299 static void
zvol_add_clones(const char * dsname,list_t * minors_list)1300 zvol_add_clones(const char *dsname, list_t *minors_list)
1301 {
1302 /* Also check if it has clones */
1303 dsl_dir_t *dd = NULL;
1304 dsl_pool_t *dp = NULL;
1305
1306 if (dsl_pool_hold(dsname, FTAG, &dp) != 0)
1307 return;
1308
1309 if (!spa_feature_is_enabled(dp->dp_spa,
1310 SPA_FEATURE_ENCRYPTION))
1311 goto out;
1312
1313 if (dsl_dir_hold(dp, dsname, FTAG, &dd, NULL) != 0)
1314 goto out;
1315
1316 if (dsl_dir_phys(dd)->dd_clones == 0)
1317 goto out;
1318
1319 zap_cursor_t *zc = kmem_alloc(sizeof (zap_cursor_t), KM_SLEEP);
1320 zap_attribute_t *za = zap_attribute_alloc();
1321 objset_t *mos = dd->dd_pool->dp_meta_objset;
1322
1323 for (zap_cursor_init(zc, mos, dsl_dir_phys(dd)->dd_clones);
1324 zap_cursor_retrieve(zc, za) == 0;
1325 zap_cursor_advance(zc)) {
1326 dsl_dataset_t *clone;
1327 minors_job_t *job;
1328
1329 if (dsl_dataset_hold_obj(dd->dd_pool,
1330 za->za_first_integer, FTAG, &clone) == 0) {
1331
1332 char name[ZFS_MAX_DATASET_NAME_LEN];
1333 dsl_dataset_name(clone, name);
1334
1335 char *n = kmem_strdup(name);
1336 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP);
1337 job->name = n;
1338 job->list = minors_list;
1339 job->error = 0;
1340 list_insert_tail(minors_list, job);
1341
1342 dsl_dataset_rele(clone, FTAG);
1343 }
1344 }
1345 zap_cursor_fini(zc);
1346 zap_attribute_free(za);
1347 kmem_free(zc, sizeof (zap_cursor_t));
1348
1349 out:
1350 if (dd != NULL)
1351 dsl_dir_rele(dd, FTAG);
1352 dsl_pool_rele(dp, FTAG);
1353 }
1354
1355 /*
1356 * Mask errors to continue dmu_objset_find() traversal
1357 */
1358 static int
zvol_create_minors_cb(const char * dsname,void * arg)1359 zvol_create_minors_cb(const char *dsname, void *arg)
1360 {
1361 uint64_t snapdev;
1362 int error;
1363 list_t *minors_list = arg;
1364
1365 ASSERT0(MUTEX_HELD(&spa_namespace_lock));
1366
1367 error = dsl_prop_get_integer(dsname, "snapdev", &snapdev, NULL);
1368 if (error)
1369 return (0);
1370
1371 /*
1372 * Given the name and the 'snapdev' property, create device minor nodes
1373 * with the linkages to zvols/snapshots as needed.
1374 * If the name represents a zvol, create a minor node for the zvol, then
1375 * check if its snapshots are 'visible', and if so, iterate over the
1376 * snapshots and create device minor nodes for those.
1377 */
1378 if (strchr(dsname, '@') == 0) {
1379 minors_job_t *job;
1380 char *n = kmem_strdup(dsname);
1381 if (n == NULL)
1382 return (0);
1383
1384 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP);
1385 job->name = n;
1386 job->list = minors_list;
1387 job->error = 0;
1388 list_insert_tail(minors_list, job);
1389 /* don't care if dispatch fails, because job->error is 0 */
1390 taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job,
1391 TQ_SLEEP);
1392
1393 zvol_add_clones(dsname, minors_list);
1394
1395 if (snapdev == ZFS_SNAPDEV_VISIBLE) {
1396 /*
1397 * traverse snapshots only, do not traverse children,
1398 * and skip the 'dsname'
1399 */
1400 (void) dmu_objset_find(dsname,
1401 zvol_create_snap_minor_cb, (void *)job,
1402 DS_FIND_SNAPSHOTS);
1403 }
1404 } else {
1405 dprintf("zvol_create_minors_cb(): %s is not a zvol name\n",
1406 dsname);
1407 }
1408
1409 return (0);
1410 }
1411
1412 /*
1413 * Create minors for the specified dataset, including children and snapshots.
1414 * Pay attention to the 'snapdev' property and iterate over the snapshots
1415 * only if they are 'visible'. This approach allows one to assure that the
1416 * snapshot metadata is read from disk only if it is needed.
1417 *
1418 * The name can represent a dataset to be recursively scanned for zvols and
1419 * their snapshots, or a single zvol snapshot. If the name represents a
1420 * dataset, the scan is performed in two nested stages:
1421 * - scan the dataset for zvols, and
1422 * - for each zvol, create a minor node, then check if the zvol's snapshots
1423 * are 'visible', and only then iterate over the snapshots if needed
1424 *
1425 * If the name represents a snapshot, a check is performed if the snapshot is
1426 * 'visible' (which also verifies that the parent is a zvol), and if so,
1427 * a minor node for that snapshot is created.
1428 */
1429 void
zvol_create_minors_recursive(const char * name)1430 zvol_create_minors_recursive(const char *name)
1431 {
1432 list_t minors_list;
1433 minors_job_t *job;
1434
1435 if (zvol_inhibit_dev)
1436 return;
1437
1438 /*
1439 * This is the list for prefetch jobs. Whenever we found a match
1440 * during dmu_objset_find, we insert a minors_job to the list and do
1441 * taskq_dispatch to parallel prefetch zvol dnodes. Note we don't need
1442 * any lock because all list operation is done on the current thread.
1443 *
1444 * We will use this list to do zvol_os_create_minor after prefetch
1445 * so we don't have to traverse using dmu_objset_find again.
1446 */
1447 list_create(&minors_list, sizeof (minors_job_t),
1448 offsetof(minors_job_t, link));
1449
1450
1451 if (strchr(name, '@') != NULL) {
1452 uint64_t snapdev;
1453
1454 int error = dsl_prop_get_integer(name, "snapdev",
1455 &snapdev, NULL);
1456
1457 if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE)
1458 (void) zvol_os_create_minor(name);
1459 } else {
1460 fstrans_cookie_t cookie = spl_fstrans_mark();
1461 (void) dmu_objset_find(name, zvol_create_minors_cb,
1462 &minors_list, DS_FIND_CHILDREN);
1463 spl_fstrans_unmark(cookie);
1464 }
1465
1466 taskq_wait_outstanding(system_taskq, 0);
1467
1468 /*
1469 * Prefetch is completed, we can do zvol_os_create_minor
1470 * sequentially.
1471 */
1472 while ((job = list_remove_head(&minors_list)) != NULL) {
1473 if (!job->error)
1474 (void) zvol_os_create_minor(job->name);
1475 kmem_strfree(job->name);
1476 kmem_free(job, sizeof (minors_job_t));
1477 }
1478
1479 list_destroy(&minors_list);
1480 }
1481
1482 void
zvol_create_minor(const char * name)1483 zvol_create_minor(const char *name)
1484 {
1485 /*
1486 * Note: the dsl_pool_config_lock must not be held.
1487 * Minor node creation needs to obtain the zvol_state_lock.
1488 * zvol_open() obtains the zvol_state_lock and then the dsl pool
1489 * config lock. Therefore, we can't have the config lock now if
1490 * we are going to wait for the zvol_state_lock, because it
1491 * would be a lock order inversion which could lead to deadlock.
1492 */
1493
1494 if (zvol_inhibit_dev)
1495 return;
1496
1497 if (strchr(name, '@') != NULL) {
1498 uint64_t snapdev;
1499
1500 int error = dsl_prop_get_integer(name,
1501 "snapdev", &snapdev, NULL);
1502
1503 if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE)
1504 (void) zvol_os_create_minor(name);
1505 } else {
1506 (void) zvol_os_create_minor(name);
1507 }
1508 }
1509
1510 /*
1511 * Remove minors for specified dataset including children and snapshots.
1512 */
1513
1514 /*
1515 * Remove the minor for a given zvol. This will do it all:
1516 * - flag the zvol for removal, so new requests are rejected
1517 * - wait until outstanding requests are completed
1518 * - remove it from lists
1519 * - free it
1520 * It's also usable as a taskq task, and smells nice too.
1521 */
1522 static void
zvol_remove_minor_task(void * arg)1523 zvol_remove_minor_task(void *arg)
1524 {
1525 zvol_state_t *zv = (zvol_state_t *)arg;
1526
1527 ASSERT(!RW_LOCK_HELD(&zvol_state_lock));
1528 ASSERT(!MUTEX_HELD(&zv->zv_state_lock));
1529
1530 mutex_enter(&zv->zv_state_lock);
1531 while (zv->zv_open_count > 0 || atomic_read(&zv->zv_suspend_ref)) {
1532 zv->zv_flags |= ZVOL_REMOVING;
1533 cv_wait(&zv->zv_removing_cv, &zv->zv_state_lock);
1534 }
1535 mutex_exit(&zv->zv_state_lock);
1536
1537 rw_enter(&zvol_state_lock, RW_WRITER);
1538 mutex_enter(&zv->zv_state_lock);
1539
1540 zvol_remove(zv);
1541 zvol_os_clear_private(zv);
1542
1543 mutex_exit(&zv->zv_state_lock);
1544 rw_exit(&zvol_state_lock);
1545
1546 zvol_os_free(zv);
1547 }
1548
1549 static void
zvol_free_task(void * arg)1550 zvol_free_task(void *arg)
1551 {
1552 zvol_os_free(arg);
1553 }
1554
1555 void
zvol_remove_minors_impl(const char * name)1556 zvol_remove_minors_impl(const char *name)
1557 {
1558 zvol_state_t *zv, *zv_next;
1559 int namelen = ((name) ? strlen(name) : 0);
1560 taskqid_t t;
1561 list_t delay_list, free_list;
1562
1563 if (zvol_inhibit_dev)
1564 return;
1565
1566 list_create(&delay_list, sizeof (zvol_state_t),
1567 offsetof(zvol_state_t, zv_next));
1568 list_create(&free_list, sizeof (zvol_state_t),
1569 offsetof(zvol_state_t, zv_next));
1570
1571 rw_enter(&zvol_state_lock, RW_WRITER);
1572
1573 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1574 zv_next = list_next(&zvol_state_list, zv);
1575
1576 mutex_enter(&zv->zv_state_lock);
1577 if (name == NULL || strcmp(zv->zv_name, name) == 0 ||
1578 (strncmp(zv->zv_name, name, namelen) == 0 &&
1579 (zv->zv_name[namelen] == '/' ||
1580 zv->zv_name[namelen] == '@'))) {
1581 /*
1582 * By holding zv_state_lock here, we guarantee that no
1583 * one is currently using this zv
1584 */
1585
1586 /*
1587 * If in use, try to throw everyone off and try again
1588 * later.
1589 */
1590 if (zv->zv_open_count > 0 ||
1591 atomic_read(&zv->zv_suspend_ref)) {
1592 zv->zv_flags |= ZVOL_REMOVING;
1593 t = taskq_dispatch(
1594 zv->zv_objset->os_spa->spa_zvol_taskq,
1595 zvol_remove_minor_task, zv, TQ_SLEEP);
1596 if (t == TASKQID_INVALID) {
1597 /*
1598 * Couldn't create the task, so we'll
1599 * do it in place once the loop is
1600 * finished.
1601 */
1602 list_insert_head(&delay_list, zv);
1603 }
1604 mutex_exit(&zv->zv_state_lock);
1605 continue;
1606 }
1607
1608 zvol_remove(zv);
1609
1610 /*
1611 * Cleared while holding zvol_state_lock as a writer
1612 * which will prevent zvol_open() from opening it.
1613 */
1614 zvol_os_clear_private(zv);
1615
1616 /* Drop zv_state_lock before zvol_free() */
1617 mutex_exit(&zv->zv_state_lock);
1618
1619 /* Try parallel zv_free, if failed do it in place */
1620 t = taskq_dispatch(system_taskq, zvol_free_task, zv,
1621 TQ_SLEEP);
1622 if (t == TASKQID_INVALID)
1623 list_insert_head(&free_list, zv);
1624 } else {
1625 mutex_exit(&zv->zv_state_lock);
1626 }
1627 }
1628 rw_exit(&zvol_state_lock);
1629
1630 /* Wait for zvols that we couldn't create a remove task for */
1631 while ((zv = list_remove_head(&delay_list)) != NULL)
1632 zvol_remove_minor_task(zv);
1633
1634 /* Free any that we couldn't free in parallel earlier */
1635 while ((zv = list_remove_head(&free_list)) != NULL)
1636 zvol_os_free(zv);
1637 }
1638
1639 /* Remove minor for this specific volume only */
1640 static void
zvol_remove_minor_impl(const char * name)1641 zvol_remove_minor_impl(const char *name)
1642 {
1643 zvol_state_t *zv = NULL, *zv_next;
1644
1645 if (zvol_inhibit_dev)
1646 return;
1647
1648 rw_enter(&zvol_state_lock, RW_WRITER);
1649
1650 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1651 zv_next = list_next(&zvol_state_list, zv);
1652
1653 mutex_enter(&zv->zv_state_lock);
1654 if (strcmp(zv->zv_name, name) == 0)
1655 /* Found, leave the the loop with zv_lock held */
1656 break;
1657 mutex_exit(&zv->zv_state_lock);
1658 }
1659
1660 if (zv == NULL) {
1661 rw_exit(&zvol_state_lock);
1662 return;
1663 }
1664
1665 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
1666
1667 if (zv->zv_open_count > 0 || atomic_read(&zv->zv_suspend_ref)) {
1668 /*
1669 * In use, so try to throw everyone off, then wait
1670 * until finished.
1671 */
1672 zv->zv_flags |= ZVOL_REMOVING;
1673 mutex_exit(&zv->zv_state_lock);
1674 rw_exit(&zvol_state_lock);
1675 zvol_remove_minor_task(zv);
1676 return;
1677 }
1678
1679 zvol_remove(zv);
1680 zvol_os_clear_private(zv);
1681
1682 mutex_exit(&zv->zv_state_lock);
1683 rw_exit(&zvol_state_lock);
1684
1685 zvol_os_free(zv);
1686 }
1687
1688 /*
1689 * Rename minors for specified dataset including children and snapshots.
1690 */
1691 static void
zvol_rename_minors_impl(const char * oldname,const char * newname)1692 zvol_rename_minors_impl(const char *oldname, const char *newname)
1693 {
1694 zvol_state_t *zv, *zv_next;
1695 int oldnamelen;
1696
1697 if (zvol_inhibit_dev)
1698 return;
1699
1700 oldnamelen = strlen(oldname);
1701
1702 rw_enter(&zvol_state_lock, RW_READER);
1703
1704 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1705 zv_next = list_next(&zvol_state_list, zv);
1706
1707 mutex_enter(&zv->zv_state_lock);
1708
1709 if (strcmp(zv->zv_name, oldname) == 0) {
1710 zvol_os_rename_minor(zv, newname);
1711 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
1712 (zv->zv_name[oldnamelen] == '/' ||
1713 zv->zv_name[oldnamelen] == '@')) {
1714 char *name = kmem_asprintf("%s%c%s", newname,
1715 zv->zv_name[oldnamelen],
1716 zv->zv_name + oldnamelen + 1);
1717 zvol_os_rename_minor(zv, name);
1718 kmem_strfree(name);
1719 }
1720
1721 mutex_exit(&zv->zv_state_lock);
1722 }
1723
1724 rw_exit(&zvol_state_lock);
1725 }
1726
1727 typedef struct zvol_snapdev_cb_arg {
1728 uint64_t snapdev;
1729 } zvol_snapdev_cb_arg_t;
1730
1731 static int
zvol_set_snapdev_cb(const char * dsname,void * param)1732 zvol_set_snapdev_cb(const char *dsname, void *param)
1733 {
1734 zvol_snapdev_cb_arg_t *arg = param;
1735
1736 if (strchr(dsname, '@') == NULL)
1737 return (0);
1738
1739 switch (arg->snapdev) {
1740 case ZFS_SNAPDEV_VISIBLE:
1741 (void) zvol_os_create_minor(dsname);
1742 break;
1743 case ZFS_SNAPDEV_HIDDEN:
1744 (void) zvol_remove_minor_impl(dsname);
1745 break;
1746 }
1747
1748 return (0);
1749 }
1750
1751 static void
zvol_set_snapdev_impl(char * name,uint64_t snapdev)1752 zvol_set_snapdev_impl(char *name, uint64_t snapdev)
1753 {
1754 zvol_snapdev_cb_arg_t arg = {snapdev};
1755 fstrans_cookie_t cookie = spl_fstrans_mark();
1756 /*
1757 * The zvol_set_snapdev_sync() sets snapdev appropriately
1758 * in the dataset hierarchy. Here, we only scan snapshots.
1759 */
1760 dmu_objset_find(name, zvol_set_snapdev_cb, &arg, DS_FIND_SNAPSHOTS);
1761 spl_fstrans_unmark(cookie);
1762 }
1763
1764 static void
zvol_set_volmode_impl(char * name,uint64_t volmode)1765 zvol_set_volmode_impl(char *name, uint64_t volmode)
1766 {
1767 fstrans_cookie_t cookie;
1768 uint64_t old_volmode;
1769 zvol_state_t *zv;
1770
1771 if (strchr(name, '@') != NULL)
1772 return;
1773
1774 /*
1775 * It's unfortunate we need to remove minors before we create new ones:
1776 * this is necessary because our backing gendisk (zvol_state->zv_disk)
1777 * could be different when we set, for instance, volmode from "geom"
1778 * to "dev" (or vice versa).
1779 */
1780 zv = zvol_find_by_name(name, RW_NONE);
1781 if (zv == NULL && volmode == ZFS_VOLMODE_NONE)
1782 return;
1783 if (zv != NULL) {
1784 old_volmode = zv->zv_volmode;
1785 mutex_exit(&zv->zv_state_lock);
1786 if (old_volmode == volmode)
1787 return;
1788 zvol_wait_close(zv);
1789 }
1790 cookie = spl_fstrans_mark();
1791 switch (volmode) {
1792 case ZFS_VOLMODE_NONE:
1793 (void) zvol_remove_minor_impl(name);
1794 break;
1795 case ZFS_VOLMODE_GEOM:
1796 case ZFS_VOLMODE_DEV:
1797 (void) zvol_remove_minor_impl(name);
1798 (void) zvol_os_create_minor(name);
1799 break;
1800 case ZFS_VOLMODE_DEFAULT:
1801 (void) zvol_remove_minor_impl(name);
1802 if (zvol_volmode == ZFS_VOLMODE_NONE)
1803 break;
1804 else /* if zvol_volmode is invalid defaults to "geom" */
1805 (void) zvol_os_create_minor(name);
1806 break;
1807 }
1808 spl_fstrans_unmark(cookie);
1809 }
1810
1811 static zvol_task_t *
zvol_task_alloc(zvol_async_op_t op,const char * name1,const char * name2,uint64_t value)1812 zvol_task_alloc(zvol_async_op_t op, const char *name1, const char *name2,
1813 uint64_t value)
1814 {
1815 zvol_task_t *task;
1816
1817 /* Never allow tasks on hidden names. */
1818 if (name1[0] == '$')
1819 return (NULL);
1820
1821 task = kmem_zalloc(sizeof (zvol_task_t), KM_SLEEP);
1822 task->op = op;
1823 task->value = value;
1824
1825 strlcpy(task->name1, name1, sizeof (task->name1));
1826 if (name2 != NULL)
1827 strlcpy(task->name2, name2, sizeof (task->name2));
1828
1829 return (task);
1830 }
1831
1832 static void
zvol_task_free(zvol_task_t * task)1833 zvol_task_free(zvol_task_t *task)
1834 {
1835 kmem_free(task, sizeof (zvol_task_t));
1836 }
1837
1838 /*
1839 * The worker thread function performed asynchronously.
1840 */
1841 static void
zvol_task_cb(void * arg)1842 zvol_task_cb(void *arg)
1843 {
1844 zvol_task_t *task = arg;
1845
1846 switch (task->op) {
1847 case ZVOL_ASYNC_REMOVE_MINORS:
1848 zvol_remove_minors_impl(task->name1);
1849 break;
1850 case ZVOL_ASYNC_RENAME_MINORS:
1851 zvol_rename_minors_impl(task->name1, task->name2);
1852 break;
1853 case ZVOL_ASYNC_SET_SNAPDEV:
1854 zvol_set_snapdev_impl(task->name1, task->value);
1855 break;
1856 case ZVOL_ASYNC_SET_VOLMODE:
1857 zvol_set_volmode_impl(task->name1, task->value);
1858 break;
1859 default:
1860 VERIFY(0);
1861 break;
1862 }
1863
1864 zvol_task_free(task);
1865 }
1866
1867 typedef struct zvol_set_prop_int_arg {
1868 const char *zsda_name;
1869 uint64_t zsda_value;
1870 zprop_source_t zsda_source;
1871 zfs_prop_t zsda_prop;
1872 } zvol_set_prop_int_arg_t;
1873
1874 /*
1875 * Sanity check the dataset for safe use by the sync task. No additional
1876 * conditions are imposed.
1877 */
1878 static int
zvol_set_common_check(void * arg,dmu_tx_t * tx)1879 zvol_set_common_check(void *arg, dmu_tx_t *tx)
1880 {
1881 zvol_set_prop_int_arg_t *zsda = arg;
1882 dsl_pool_t *dp = dmu_tx_pool(tx);
1883 dsl_dir_t *dd;
1884 int error;
1885
1886 error = dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL);
1887 if (error != 0)
1888 return (error);
1889
1890 dsl_dir_rele(dd, FTAG);
1891
1892 return (error);
1893 }
1894
1895 static int
zvol_set_common_sync_cb(dsl_pool_t * dp,dsl_dataset_t * ds,void * arg)1896 zvol_set_common_sync_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1897 {
1898 zvol_set_prop_int_arg_t *zsda = arg;
1899 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1900 zvol_task_t *task;
1901 uint64_t prop;
1902
1903 const char *prop_name = zfs_prop_to_name(zsda->zsda_prop);
1904 dsl_dataset_name(ds, dsname);
1905
1906 if (dsl_prop_get_int_ds(ds, prop_name, &prop) != 0)
1907 return (0);
1908
1909 switch (zsda->zsda_prop) {
1910 case ZFS_PROP_VOLMODE:
1911 task = zvol_task_alloc(ZVOL_ASYNC_SET_VOLMODE, dsname,
1912 NULL, prop);
1913 break;
1914 case ZFS_PROP_SNAPDEV:
1915 task = zvol_task_alloc(ZVOL_ASYNC_SET_SNAPDEV, dsname,
1916 NULL, prop);
1917 break;
1918 default:
1919 task = NULL;
1920 break;
1921 }
1922
1923 if (task == NULL)
1924 return (0);
1925
1926 (void) taskq_dispatch(dp->dp_spa->spa_zvol_taskq, zvol_task_cb,
1927 task, TQ_SLEEP);
1928 return (0);
1929 }
1930
1931 /*
1932 * Traverse all child datasets and apply the property appropriately.
1933 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
1934 * dataset and read the effective "property" on every child in the callback
1935 * function: this is because the value is not guaranteed to be the same in the
1936 * whole dataset hierarchy.
1937 */
1938 static void
zvol_set_common_sync(void * arg,dmu_tx_t * tx)1939 zvol_set_common_sync(void *arg, dmu_tx_t *tx)
1940 {
1941 zvol_set_prop_int_arg_t *zsda = arg;
1942 dsl_pool_t *dp = dmu_tx_pool(tx);
1943 dsl_dir_t *dd;
1944 dsl_dataset_t *ds;
1945 int error;
1946
1947 VERIFY0(dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL));
1948
1949 error = dsl_dataset_hold(dp, zsda->zsda_name, FTAG, &ds);
1950 if (error == 0) {
1951 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(zsda->zsda_prop),
1952 zsda->zsda_source, sizeof (zsda->zsda_value), 1,
1953 &zsda->zsda_value, tx);
1954 dsl_dataset_rele(ds, FTAG);
1955 }
1956
1957 dmu_objset_find_dp(dp, dd->dd_object, zvol_set_common_sync_cb,
1958 zsda, DS_FIND_CHILDREN);
1959
1960 dsl_dir_rele(dd, FTAG);
1961 }
1962
1963 int
zvol_set_common(const char * ddname,zfs_prop_t prop,zprop_source_t source,uint64_t val)1964 zvol_set_common(const char *ddname, zfs_prop_t prop, zprop_source_t source,
1965 uint64_t val)
1966 {
1967 zvol_set_prop_int_arg_t zsda;
1968
1969 zsda.zsda_name = ddname;
1970 zsda.zsda_source = source;
1971 zsda.zsda_value = val;
1972 zsda.zsda_prop = prop;
1973
1974 return (dsl_sync_task(ddname, zvol_set_common_check,
1975 zvol_set_common_sync, &zsda, 0, ZFS_SPACE_CHECK_NONE));
1976 }
1977
1978 void
zvol_remove_minors(spa_t * spa,const char * name,boolean_t async)1979 zvol_remove_minors(spa_t *spa, const char *name, boolean_t async)
1980 {
1981 zvol_task_t *task;
1982 taskqid_t id;
1983
1984 task = zvol_task_alloc(ZVOL_ASYNC_REMOVE_MINORS, name, NULL, ~0ULL);
1985 if (task == NULL)
1986 return;
1987
1988 id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP);
1989 if ((async == B_FALSE) && (id != TASKQID_INVALID))
1990 taskq_wait_id(spa->spa_zvol_taskq, id);
1991 }
1992
1993 void
zvol_rename_minors(spa_t * spa,const char * name1,const char * name2,boolean_t async)1994 zvol_rename_minors(spa_t *spa, const char *name1, const char *name2,
1995 boolean_t async)
1996 {
1997 zvol_task_t *task;
1998 taskqid_t id;
1999
2000 task = zvol_task_alloc(ZVOL_ASYNC_RENAME_MINORS, name1, name2, ~0ULL);
2001 if (task == NULL)
2002 return;
2003
2004 id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP);
2005 if ((async == B_FALSE) && (id != TASKQID_INVALID))
2006 taskq_wait_id(spa->spa_zvol_taskq, id);
2007 }
2008
2009 boolean_t
zvol_is_zvol(const char * name)2010 zvol_is_zvol(const char *name)
2011 {
2012
2013 return (zvol_os_is_zvol(name));
2014 }
2015
2016 int
zvol_init_impl(void)2017 zvol_init_impl(void)
2018 {
2019 int i;
2020
2021 list_create(&zvol_state_list, sizeof (zvol_state_t),
2022 offsetof(zvol_state_t, zv_next));
2023 rw_init(&zvol_state_lock, NULL, RW_DEFAULT, NULL);
2024
2025 zvol_htable = kmem_alloc(ZVOL_HT_SIZE * sizeof (struct hlist_head),
2026 KM_SLEEP);
2027 for (i = 0; i < ZVOL_HT_SIZE; i++)
2028 INIT_HLIST_HEAD(&zvol_htable[i]);
2029
2030 return (0);
2031 }
2032
2033 void
zvol_fini_impl(void)2034 zvol_fini_impl(void)
2035 {
2036 zvol_remove_minors_impl(NULL);
2037
2038 /*
2039 * The call to "zvol_remove_minors_impl" may dispatch entries to
2040 * the system_taskq, but it doesn't wait for those entries to
2041 * complete before it returns. Thus, we must wait for all of the
2042 * removals to finish, before we can continue.
2043 */
2044 taskq_wait_outstanding(system_taskq, 0);
2045
2046 kmem_free(zvol_htable, ZVOL_HT_SIZE * sizeof (struct hlist_head));
2047 list_destroy(&zvol_state_list);
2048 rw_destroy(&zvol_state_lock);
2049 }
2050