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 <[email protected]>.
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.
36 *
37 * Copyright 2014 Nexenta Systems, Inc. All rights reserved.
38 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
39 * Copyright (c) 2012, 2019 by Delphix. All rights reserved.
40 */
41
42 /*
43 * Note on locking of zvol state structures.
44 *
45 * These structures are used to maintain internal state used to emulate block
46 * devices on top of zvols. In particular, management of device minor number
47 * operations - create, remove, rename, and set_snapdev - involves access to
48 * these structures. The zvol_state_lock is primarily used to protect the
49 * zvol_state_list. The zv->zv_state_lock is used to protect the contents
50 * of the zvol_state_t structures, as well as to make sure that when the
51 * time comes to remove the structure from the list, it is not in use, and
52 * therefore, it can be taken off zvol_state_list and freed.
53 *
54 * The zv_suspend_lock was introduced to allow for suspending I/O to a zvol,
55 * e.g. for the duration of receive and rollback operations. This lock can be
56 * held for significant periods of time. Given that it is undesirable to hold
57 * mutexes for long periods of time, the following lock ordering applies:
58 * - take zvol_state_lock if necessary, to protect zvol_state_list
59 * - take zv_suspend_lock if necessary, by the code path in question
60 * - take zv_state_lock to protect zvol_state_t
61 *
62 * The minor operations are issued to spa->spa_zvol_taskq queues, that are
63 * single-threaded (to preserve order of minor operations), and are executed
64 * through the zvol_task_cb that dispatches the specific operations. Therefore,
65 * these operations are serialized per pool. Consequently, we can be certain
66 * that for a given zvol, there is only one operation at a time in progress.
67 * That is why one can be sure that first, zvol_state_t for a given zvol is
68 * allocated and placed on zvol_state_list, and then other minor operations
69 * for this zvol are going to proceed in the order of issue.
70 *
71 */
72
73 #include <sys/dataset_kstats.h>
74 #include <sys/dbuf.h>
75 #include <sys/dmu_traverse.h>
76 #include <sys/dsl_dataset.h>
77 #include <sys/dsl_prop.h>
78 #include <sys/dsl_dir.h>
79 #include <sys/zap.h>
80 #include <sys/zfeature.h>
81 #include <sys/zil_impl.h>
82 #include <sys/dmu_tx.h>
83 #include <sys/zio.h>
84 #include <sys/zfs_rlock.h>
85 #include <sys/spa_impl.h>
86 #include <sys/zvol.h>
87
88 #include <sys/zvol_impl.h>
89
90
91 unsigned int zvol_inhibit_dev = 0;
92 unsigned int zvol_volmode = ZFS_VOLMODE_GEOM;
93
94 struct hlist_head *zvol_htable;
95 list_t zvol_state_list;
96 krwlock_t zvol_state_lock;
97 const zvol_platform_ops_t *ops;
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 pool[MAXNAMELEN];
110 char name1[MAXNAMELEN];
111 char name2[MAXNAMELEN];
112 zprop_source_t source;
113 uint64_t value;
114 } zvol_task_t;
115
116 uint64_t
zvol_name_hash(const char * name)117 zvol_name_hash(const char *name)
118 {
119 int i;
120 uint64_t crc = -1ULL;
121 const uint8_t *p = (const uint8_t *)name;
122 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
123 for (i = 0; i < MAXNAMELEN - 1 && *p; i++, p++) {
124 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (*p)) & 0xFF];
125 }
126 return (crc);
127 }
128
129 /*
130 * Find a zvol_state_t given the name and hash generated by zvol_name_hash.
131 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
132 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
133 * before zv_state_lock. The mode argument indicates the mode (including none)
134 * for zv_suspend_lock to be taken.
135 */
136 zvol_state_t *
zvol_find_by_name_hash(const char * name,uint64_t hash,int mode)137 zvol_find_by_name_hash(const char *name, uint64_t hash, int mode)
138 {
139 zvol_state_t *zv;
140 struct hlist_node *p = NULL;
141
142 rw_enter(&zvol_state_lock, RW_READER);
143 hlist_for_each(p, ZVOL_HT_HEAD(hash)) {
144 zv = hlist_entry(p, zvol_state_t, zv_hlink);
145 mutex_enter(&zv->zv_state_lock);
146 if (zv->zv_hash == hash &&
147 strncmp(zv->zv_name, name, MAXNAMELEN) == 0) {
148 /*
149 * this is the right zvol, take the locks in the
150 * right order
151 */
152 if (mode != RW_NONE &&
153 !rw_tryenter(&zv->zv_suspend_lock, mode)) {
154 mutex_exit(&zv->zv_state_lock);
155 rw_enter(&zv->zv_suspend_lock, mode);
156 mutex_enter(&zv->zv_state_lock);
157 /*
158 * zvol cannot be renamed as we continue
159 * to hold zvol_state_lock
160 */
161 ASSERT(zv->zv_hash == hash &&
162 strncmp(zv->zv_name, name, MAXNAMELEN)
163 == 0);
164 }
165 rw_exit(&zvol_state_lock);
166 return (zv);
167 }
168 mutex_exit(&zv->zv_state_lock);
169 }
170 rw_exit(&zvol_state_lock);
171
172 return (NULL);
173 }
174
175 /*
176 * Find a zvol_state_t given the name.
177 * If found, return with zv_suspend_lock and zv_state_lock taken, otherwise,
178 * return (NULL) without the taking locks. The zv_suspend_lock is always taken
179 * before zv_state_lock. The mode argument indicates the mode (including none)
180 * for zv_suspend_lock to be taken.
181 */
182 static zvol_state_t *
zvol_find_by_name(const char * name,int mode)183 zvol_find_by_name(const char *name, int mode)
184 {
185 return (zvol_find_by_name_hash(name, zvol_name_hash(name), mode));
186 }
187
188 /*
189 * ZFS_IOC_CREATE callback handles dmu zvol and zap object creation.
190 */
191 void
zvol_create_cb(objset_t * os,void * arg,cred_t * cr,dmu_tx_t * tx)192 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
193 {
194 zfs_creat_t *zct = arg;
195 nvlist_t *nvprops = zct->zct_props;
196 int error;
197 uint64_t volblocksize, volsize;
198
199 VERIFY(nvlist_lookup_uint64(nvprops,
200 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
201 if (nvlist_lookup_uint64(nvprops,
202 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
203 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
204
205 /*
206 * These properties must be removed from the list so the generic
207 * property setting step won't apply to them.
208 */
209 VERIFY(nvlist_remove_all(nvprops,
210 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
211 (void) nvlist_remove_all(nvprops,
212 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
213
214 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
215 DMU_OT_NONE, 0, tx);
216 ASSERT(error == 0);
217
218 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
219 DMU_OT_NONE, 0, tx);
220 ASSERT(error == 0);
221
222 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
223 ASSERT(error == 0);
224 }
225
226 /*
227 * ZFS_IOC_OBJSET_STATS entry point.
228 */
229 int
zvol_get_stats(objset_t * os,nvlist_t * nv)230 zvol_get_stats(objset_t *os, nvlist_t *nv)
231 {
232 int error;
233 dmu_object_info_t *doi;
234 uint64_t val;
235
236 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
237 if (error)
238 return (SET_ERROR(error));
239
240 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
241 doi = kmem_alloc(sizeof (dmu_object_info_t), KM_SLEEP);
242 error = dmu_object_info(os, ZVOL_OBJ, doi);
243
244 if (error == 0) {
245 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
246 doi->doi_data_block_size);
247 }
248
249 kmem_free(doi, sizeof (dmu_object_info_t));
250
251 return (SET_ERROR(error));
252 }
253
254 /*
255 * Sanity check volume size.
256 */
257 int
zvol_check_volsize(uint64_t volsize,uint64_t blocksize)258 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
259 {
260 if (volsize == 0)
261 return (SET_ERROR(EINVAL));
262
263 if (volsize % blocksize != 0)
264 return (SET_ERROR(EINVAL));
265
266 #ifdef _ILP32
267 if (volsize - 1 > SPEC_MAXOFFSET_T)
268 return (SET_ERROR(EOVERFLOW));
269 #endif
270 return (0);
271 }
272
273 /*
274 * Ensure the zap is flushed then inform the VFS of the capacity change.
275 */
276 static int
zvol_update_volsize(uint64_t volsize,objset_t * os)277 zvol_update_volsize(uint64_t volsize, objset_t *os)
278 {
279 dmu_tx_t *tx;
280 int error;
281 uint64_t txg;
282
283 tx = dmu_tx_create(os);
284 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
285 dmu_tx_mark_netfree(tx);
286 error = dmu_tx_assign(tx, TXG_WAIT);
287 if (error) {
288 dmu_tx_abort(tx);
289 return (SET_ERROR(error));
290 }
291 txg = dmu_tx_get_txg(tx);
292
293 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
294 &volsize, tx);
295 dmu_tx_commit(tx);
296
297 txg_wait_synced(dmu_objset_pool(os), txg);
298
299 if (error == 0)
300 error = dmu_free_long_range(os,
301 ZVOL_OBJ, volsize, DMU_OBJECT_END);
302
303 return (error);
304 }
305
306 /*
307 * Set ZFS_PROP_VOLSIZE set entry point. Note that modifying the volume
308 * size will result in a udev "change" event being generated.
309 */
310 int
zvol_set_volsize(const char * name,uint64_t volsize)311 zvol_set_volsize(const char *name, uint64_t volsize)
312 {
313 objset_t *os = NULL;
314 uint64_t readonly;
315 int error;
316 boolean_t owned = B_FALSE;
317
318 error = dsl_prop_get_integer(name,
319 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
320 if (error != 0)
321 return (SET_ERROR(error));
322 if (readonly)
323 return (SET_ERROR(EROFS));
324
325 zvol_state_t *zv = zvol_find_by_name(name, RW_READER);
326
327 ASSERT(zv == NULL || (MUTEX_HELD(&zv->zv_state_lock) &&
328 RW_READ_HELD(&zv->zv_suspend_lock)));
329
330 if (zv == NULL || zv->zv_objset == NULL) {
331 if (zv != NULL)
332 rw_exit(&zv->zv_suspend_lock);
333 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, B_TRUE,
334 FTAG, &os)) != 0) {
335 if (zv != NULL)
336 mutex_exit(&zv->zv_state_lock);
337 return (SET_ERROR(error));
338 }
339 owned = B_TRUE;
340 if (zv != NULL)
341 zv->zv_objset = os;
342 } else {
343 os = zv->zv_objset;
344 }
345
346 dmu_object_info_t *doi = kmem_alloc(sizeof (*doi), KM_SLEEP);
347
348 if ((error = dmu_object_info(os, ZVOL_OBJ, doi)) ||
349 (error = zvol_check_volsize(volsize, doi->doi_data_block_size)))
350 goto out;
351
352 error = zvol_update_volsize(volsize, os);
353 if (error == 0 && zv != NULL) {
354 zv->zv_volsize = volsize;
355 zv->zv_changed = 1;
356 }
357 out:
358 kmem_free(doi, sizeof (dmu_object_info_t));
359
360 if (owned) {
361 dmu_objset_disown(os, B_TRUE, FTAG);
362 if (zv != NULL)
363 zv->zv_objset = NULL;
364 } else {
365 rw_exit(&zv->zv_suspend_lock);
366 }
367
368 if (zv != NULL)
369 mutex_exit(&zv->zv_state_lock);
370
371 if (error == 0 && zv != NULL)
372 ops->zv_update_volsize(zv, volsize);
373
374 return (SET_ERROR(error));
375 }
376
377 /*
378 * Sanity check volume block size.
379 */
380 int
zvol_check_volblocksize(const char * name,uint64_t volblocksize)381 zvol_check_volblocksize(const char *name, uint64_t volblocksize)
382 {
383 /* Record sizes above 128k need the feature to be enabled */
384 if (volblocksize > SPA_OLD_MAXBLOCKSIZE) {
385 spa_t *spa;
386 int error;
387
388 if ((error = spa_open(name, &spa, FTAG)) != 0)
389 return (error);
390
391 if (!spa_feature_is_enabled(spa, SPA_FEATURE_LARGE_BLOCKS)) {
392 spa_close(spa, FTAG);
393 return (SET_ERROR(ENOTSUP));
394 }
395
396 /*
397 * We don't allow setting the property above 1MB,
398 * unless the tunable has been changed.
399 */
400 if (volblocksize > zfs_max_recordsize)
401 return (SET_ERROR(EDOM));
402
403 spa_close(spa, FTAG);
404 }
405
406 if (volblocksize < SPA_MINBLOCKSIZE ||
407 volblocksize > SPA_MAXBLOCKSIZE ||
408 !ISP2(volblocksize))
409 return (SET_ERROR(EDOM));
410
411 return (0);
412 }
413
414 /*
415 * Set ZFS_PROP_VOLBLOCKSIZE set entry point.
416 */
417 int
zvol_set_volblocksize(const char * name,uint64_t volblocksize)418 zvol_set_volblocksize(const char *name, uint64_t volblocksize)
419 {
420 zvol_state_t *zv;
421 dmu_tx_t *tx;
422 int error;
423
424 zv = zvol_find_by_name(name, RW_READER);
425
426 if (zv == NULL)
427 return (SET_ERROR(ENXIO));
428
429 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
430 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock));
431
432 if (zv->zv_flags & ZVOL_RDONLY) {
433 mutex_exit(&zv->zv_state_lock);
434 rw_exit(&zv->zv_suspend_lock);
435 return (SET_ERROR(EROFS));
436 }
437
438 tx = dmu_tx_create(zv->zv_objset);
439 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
440 error = dmu_tx_assign(tx, TXG_WAIT);
441 if (error) {
442 dmu_tx_abort(tx);
443 } else {
444 error = dmu_object_set_blocksize(zv->zv_objset, ZVOL_OBJ,
445 volblocksize, 0, tx);
446 if (error == ENOTSUP)
447 error = SET_ERROR(EBUSY);
448 dmu_tx_commit(tx);
449 if (error == 0)
450 zv->zv_volblocksize = volblocksize;
451 }
452
453 mutex_exit(&zv->zv_state_lock);
454 rw_exit(&zv->zv_suspend_lock);
455
456 return (SET_ERROR(error));
457 }
458
459 /*
460 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
461 * implement DKIOCFREE/free-long-range.
462 */
463 static int
zvol_replay_truncate(void * arg1,void * arg2,boolean_t byteswap)464 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
465 {
466 zvol_state_t *zv = arg1;
467 lr_truncate_t *lr = arg2;
468 uint64_t offset, length;
469
470 if (byteswap)
471 byteswap_uint64_array(lr, sizeof (*lr));
472
473 offset = lr->lr_offset;
474 length = lr->lr_length;
475
476 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
477 }
478
479 /*
480 * Replay a TX_WRITE ZIL transaction that didn't get committed
481 * after a system failure
482 */
483 static int
zvol_replay_write(void * arg1,void * arg2,boolean_t byteswap)484 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap)
485 {
486 zvol_state_t *zv = arg1;
487 lr_write_t *lr = arg2;
488 objset_t *os = zv->zv_objset;
489 char *data = (char *)(lr + 1); /* data follows lr_write_t */
490 uint64_t offset, length;
491 dmu_tx_t *tx;
492 int error;
493
494 if (byteswap)
495 byteswap_uint64_array(lr, sizeof (*lr));
496
497 offset = lr->lr_offset;
498 length = lr->lr_length;
499
500 /* If it's a dmu_sync() block, write the whole block */
501 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
502 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
503 if (length < blocksize) {
504 offset -= offset % blocksize;
505 length = blocksize;
506 }
507 }
508
509 tx = dmu_tx_create(os);
510 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
511 error = dmu_tx_assign(tx, TXG_WAIT);
512 if (error) {
513 dmu_tx_abort(tx);
514 } else {
515 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
516 dmu_tx_commit(tx);
517 }
518
519 return (error);
520 }
521
522 static int
zvol_replay_err(void * arg1,void * arg2,boolean_t byteswap)523 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap)
524 {
525 return (SET_ERROR(ENOTSUP));
526 }
527
528 /*
529 * Callback vectors for replaying records.
530 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
531 */
532 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
533 zvol_replay_err, /* no such transaction type */
534 zvol_replay_err, /* TX_CREATE */
535 zvol_replay_err, /* TX_MKDIR */
536 zvol_replay_err, /* TX_MKXATTR */
537 zvol_replay_err, /* TX_SYMLINK */
538 zvol_replay_err, /* TX_REMOVE */
539 zvol_replay_err, /* TX_RMDIR */
540 zvol_replay_err, /* TX_LINK */
541 zvol_replay_err, /* TX_RENAME */
542 zvol_replay_write, /* TX_WRITE */
543 zvol_replay_truncate, /* TX_TRUNCATE */
544 zvol_replay_err, /* TX_SETATTR */
545 zvol_replay_err, /* TX_ACL */
546 zvol_replay_err, /* TX_CREATE_ATTR */
547 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
548 zvol_replay_err, /* TX_MKDIR_ACL */
549 zvol_replay_err, /* TX_MKDIR_ATTR */
550 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
551 zvol_replay_err, /* TX_WRITE2 */
552 };
553
554 /*
555 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
556 *
557 * We store data in the log buffers if it's small enough.
558 * Otherwise we will later flush the data out via dmu_sync().
559 */
560 ssize_t zvol_immediate_write_sz = 32768;
561
562 void
zvol_log_write(zvol_state_t * zv,dmu_tx_t * tx,uint64_t offset,uint64_t size,int sync)563 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, uint64_t offset,
564 uint64_t size, int sync)
565 {
566 uint32_t blocksize = zv->zv_volblocksize;
567 zilog_t *zilog = zv->zv_zilog;
568 itx_wr_state_t write_state;
569
570 if (zil_replaying(zilog, tx))
571 return;
572
573 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
574 write_state = WR_INDIRECT;
575 else if (!spa_has_slogs(zilog->zl_spa) &&
576 size >= blocksize && blocksize > zvol_immediate_write_sz)
577 write_state = WR_INDIRECT;
578 else if (sync)
579 write_state = WR_COPIED;
580 else
581 write_state = WR_NEED_COPY;
582
583 while (size) {
584 itx_t *itx;
585 lr_write_t *lr;
586 itx_wr_state_t wr_state = write_state;
587 ssize_t len = size;
588
589 if (wr_state == WR_COPIED && size > zil_max_copied_data(zilog))
590 wr_state = WR_NEED_COPY;
591 else if (wr_state == WR_INDIRECT)
592 len = MIN(blocksize - P2PHASE(offset, blocksize), size);
593
594 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
595 (wr_state == WR_COPIED ? len : 0));
596 lr = (lr_write_t *)&itx->itx_lr;
597 if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
598 offset, len, lr+1, DMU_READ_NO_PREFETCH) != 0) {
599 zil_itx_destroy(itx);
600 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
601 lr = (lr_write_t *)&itx->itx_lr;
602 wr_state = WR_NEED_COPY;
603 }
604
605 itx->itx_wr_state = wr_state;
606 lr->lr_foid = ZVOL_OBJ;
607 lr->lr_offset = offset;
608 lr->lr_length = len;
609 lr->lr_blkoff = 0;
610 BP_ZERO(&lr->lr_blkptr);
611
612 itx->itx_private = zv;
613 itx->itx_sync = sync;
614
615 (void) zil_itx_assign(zilog, itx, tx);
616
617 offset += len;
618 size -= len;
619 }
620 }
621
622 /*
623 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
624 */
625 void
zvol_log_truncate(zvol_state_t * zv,dmu_tx_t * tx,uint64_t off,uint64_t len,boolean_t sync)626 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
627 boolean_t sync)
628 {
629 itx_t *itx;
630 lr_truncate_t *lr;
631 zilog_t *zilog = zv->zv_zilog;
632
633 if (zil_replaying(zilog, tx))
634 return;
635
636 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
637 lr = (lr_truncate_t *)&itx->itx_lr;
638 lr->lr_foid = ZVOL_OBJ;
639 lr->lr_offset = off;
640 lr->lr_length = len;
641
642 itx->itx_sync = sync;
643 zil_itx_assign(zilog, itx, tx);
644 }
645
646
647 /* ARGSUSED */
648 static void
zvol_get_done(zgd_t * zgd,int error)649 zvol_get_done(zgd_t *zgd, int error)
650 {
651 if (zgd->zgd_db)
652 dmu_buf_rele(zgd->zgd_db, zgd);
653
654 zfs_rangelock_exit(zgd->zgd_lr);
655
656 kmem_free(zgd, sizeof (zgd_t));
657 }
658
659 /*
660 * Get data to generate a TX_WRITE intent log record.
661 */
662 int
zvol_get_data(void * arg,lr_write_t * lr,char * buf,struct lwb * lwb,zio_t * zio)663 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
664 {
665 zvol_state_t *zv = arg;
666 uint64_t offset = lr->lr_offset;
667 uint64_t size = lr->lr_length;
668 dmu_buf_t *db;
669 zgd_t *zgd;
670 int error;
671
672 ASSERT3P(lwb, !=, NULL);
673 ASSERT3P(zio, !=, NULL);
674 ASSERT3U(size, !=, 0);
675
676 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
677 zgd->zgd_lwb = lwb;
678
679 /*
680 * Write records come in two flavors: immediate and indirect.
681 * For small writes it's cheaper to store the data with the
682 * log record (immediate); for large writes it's cheaper to
683 * sync the data and get a pointer to it (indirect) so that
684 * we don't have to write the data twice.
685 */
686 if (buf != NULL) { /* immediate write */
687 zgd->zgd_lr = zfs_rangelock_enter(&zv->zv_rangelock, offset,
688 size, RL_READER);
689 error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
690 DMU_READ_NO_PREFETCH);
691 } else { /* indirect write */
692 /*
693 * Have to lock the whole block to ensure when it's written out
694 * and its checksum is being calculated that no one can change
695 * the data. Contrarily to zfs_get_data we need not re-check
696 * blocksize after we get the lock because it cannot be changed.
697 */
698 size = zv->zv_volblocksize;
699 offset = P2ALIGN_TYPED(offset, size, uint64_t);
700 zgd->zgd_lr = zfs_rangelock_enter(&zv->zv_rangelock, offset,
701 size, RL_READER);
702 error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db,
703 DMU_READ_NO_PREFETCH);
704 if (error == 0) {
705 blkptr_t *bp = &lr->lr_blkptr;
706
707 zgd->zgd_db = db;
708 zgd->zgd_bp = bp;
709
710 ASSERT(db != NULL);
711 ASSERT(db->db_offset == offset);
712 ASSERT(db->db_size == size);
713
714 error = dmu_sync(zio, lr->lr_common.lrc_txg,
715 zvol_get_done, zgd);
716
717 if (error == 0)
718 return (0);
719 }
720 }
721
722 zvol_get_done(zgd, error);
723
724 return (SET_ERROR(error));
725 }
726
727 /*
728 * The zvol_state_t's are inserted into zvol_state_list and zvol_htable.
729 */
730
731 void
zvol_insert(zvol_state_t * zv)732 zvol_insert(zvol_state_t *zv)
733 {
734 ASSERT(RW_WRITE_HELD(&zvol_state_lock));
735 list_insert_head(&zvol_state_list, zv);
736 hlist_add_head(&zv->zv_hlink, ZVOL_HT_HEAD(zv->zv_hash));
737 }
738
739 /*
740 * Simply remove the zvol from to list of zvols.
741 */
742 static void
zvol_remove(zvol_state_t * zv)743 zvol_remove(zvol_state_t *zv)
744 {
745 ASSERT(RW_WRITE_HELD(&zvol_state_lock));
746 list_remove(&zvol_state_list, zv);
747 hlist_del(&zv->zv_hlink);
748 }
749
750 /*
751 * Setup zv after we just own the zv->objset
752 */
753 static int
zvol_setup_zv(zvol_state_t * zv)754 zvol_setup_zv(zvol_state_t *zv)
755 {
756 uint64_t volsize;
757 int error;
758 uint64_t ro;
759 objset_t *os = zv->zv_objset;
760
761 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
762 ASSERT(RW_LOCK_HELD(&zv->zv_suspend_lock));
763
764 zv->zv_zilog = NULL;
765 zv->zv_flags &= ~ZVOL_WRITTEN_TO;
766
767 error = dsl_prop_get_integer(zv->zv_name, "readonly", &ro, NULL);
768 if (error)
769 return (SET_ERROR(error));
770
771 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
772 if (error)
773 return (SET_ERROR(error));
774
775 error = dnode_hold(os, ZVOL_OBJ, zv, &zv->zv_dn);
776 if (error)
777 return (SET_ERROR(error));
778
779 ops->zv_set_capacity(zv, volsize >> 9);
780 zv->zv_volsize = volsize;
781
782 if (ro || dmu_objset_is_snapshot(os) ||
783 !spa_writeable(dmu_objset_spa(os))) {
784 ops->zv_set_disk_ro(zv, 1);
785 zv->zv_flags |= ZVOL_RDONLY;
786 } else {
787 ops->zv_set_disk_ro(zv, 0);
788 zv->zv_flags &= ~ZVOL_RDONLY;
789 }
790 return (0);
791 }
792
793 /*
794 * Shutdown every zv_objset related stuff except zv_objset itself.
795 * The is the reverse of zvol_setup_zv.
796 */
797 static void
zvol_shutdown_zv(zvol_state_t * zv)798 zvol_shutdown_zv(zvol_state_t *zv)
799 {
800 ASSERT(MUTEX_HELD(&zv->zv_state_lock) &&
801 RW_LOCK_HELD(&zv->zv_suspend_lock));
802
803 if (zv->zv_flags & ZVOL_WRITTEN_TO) {
804 ASSERT(zv->zv_zilog != NULL);
805 zil_close(zv->zv_zilog);
806 }
807
808 zv->zv_zilog = NULL;
809
810 dnode_rele(zv->zv_dn, zv);
811 zv->zv_dn = NULL;
812
813 /*
814 * Evict cached data. We must write out any dirty data before
815 * disowning the dataset.
816 */
817 if (zv->zv_flags & ZVOL_WRITTEN_TO)
818 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
819 (void) dmu_objset_evict_dbufs(zv->zv_objset);
820 }
821
822 /*
823 * return the proper tag for rollback and recv
824 */
825 void *
zvol_tag(zvol_state_t * zv)826 zvol_tag(zvol_state_t *zv)
827 {
828 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
829 return (zv->zv_open_count > 0 ? zv : NULL);
830 }
831
832 /*
833 * Suspend the zvol for recv and rollback.
834 */
835 zvol_state_t *
zvol_suspend(const char * name)836 zvol_suspend(const char *name)
837 {
838 zvol_state_t *zv;
839
840 zv = zvol_find_by_name(name, RW_WRITER);
841
842 if (zv == NULL)
843 return (NULL);
844
845 /* block all I/O, release in zvol_resume. */
846 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
847 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
848
849 atomic_inc(&zv->zv_suspend_ref);
850
851 if (zv->zv_open_count > 0)
852 zvol_shutdown_zv(zv);
853
854 /*
855 * do not hold zv_state_lock across suspend/resume to
856 * avoid locking up zvol lookups
857 */
858 mutex_exit(&zv->zv_state_lock);
859
860 /* zv_suspend_lock is released in zvol_resume() */
861 return (zv);
862 }
863
864 int
zvol_resume(zvol_state_t * zv)865 zvol_resume(zvol_state_t *zv)
866 {
867 int error = 0;
868
869 ASSERT(RW_WRITE_HELD(&zv->zv_suspend_lock));
870
871 mutex_enter(&zv->zv_state_lock);
872
873 if (zv->zv_open_count > 0) {
874 VERIFY0(dmu_objset_hold(zv->zv_name, zv, &zv->zv_objset));
875 VERIFY3P(zv->zv_objset->os_dsl_dataset->ds_owner, ==, zv);
876 VERIFY(dsl_dataset_long_held(zv->zv_objset->os_dsl_dataset));
877 dmu_objset_rele(zv->zv_objset, zv);
878
879 error = zvol_setup_zv(zv);
880 }
881
882 mutex_exit(&zv->zv_state_lock);
883
884 rw_exit(&zv->zv_suspend_lock);
885 /*
886 * We need this because we don't hold zvol_state_lock while releasing
887 * zv_suspend_lock. zvol_remove_minors_impl thus cannot check
888 * zv_suspend_lock to determine it is safe to free because rwlock is
889 * not inherent atomic.
890 */
891 atomic_dec(&zv->zv_suspend_ref);
892
893 return (SET_ERROR(error));
894 }
895
896 int
zvol_first_open(zvol_state_t * zv,boolean_t readonly)897 zvol_first_open(zvol_state_t *zv, boolean_t readonly)
898 {
899 objset_t *os;
900 int error, locked = 0;
901 boolean_t ro;
902
903 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock));
904 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
905
906 /*
907 * In all other cases the spa_namespace_lock is taken before the
908 * bdev->bd_mutex lock. But in this case the Linux __blkdev_get()
909 * function calls fops->open() with the bdev->bd_mutex lock held.
910 * This deadlock can be easily observed with zvols used as vdevs.
911 *
912 * To avoid a potential lock inversion deadlock we preemptively
913 * try to take the spa_namespace_lock(). Normally it will not
914 * be contended and this is safe because spa_open_common() handles
915 * the case where the caller already holds the spa_namespace_lock.
916 *
917 * When it is contended we risk a lock inversion if we were to
918 * block waiting for the lock. Luckily, the __blkdev_get()
919 * function allows us to return -ERESTARTSYS which will result in
920 * bdev->bd_mutex being dropped, reacquired, and fops->open() being
921 * called again. This process can be repeated safely until both
922 * locks are acquired.
923 */
924 if (!mutex_owned(&spa_namespace_lock)) {
925 locked = mutex_tryenter(&spa_namespace_lock);
926 if (!locked)
927 return (SET_ERROR(EINTR));
928 }
929
930 ro = (readonly || (strchr(zv->zv_name, '@') != NULL));
931 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, ro, B_TRUE, zv, &os);
932 if (error)
933 goto out_mutex;
934
935 zv->zv_objset = os;
936
937 error = zvol_setup_zv(zv);
938
939 if (error) {
940 dmu_objset_disown(os, 1, zv);
941 zv->zv_objset = NULL;
942 }
943
944 out_mutex:
945 if (locked)
946 mutex_exit(&spa_namespace_lock);
947 return (SET_ERROR(error));
948 }
949
950 void
zvol_last_close(zvol_state_t * zv)951 zvol_last_close(zvol_state_t *zv)
952 {
953 ASSERT(RW_READ_HELD(&zv->zv_suspend_lock));
954 ASSERT(MUTEX_HELD(&zv->zv_state_lock));
955
956 zvol_shutdown_zv(zv);
957
958 dmu_objset_disown(zv->zv_objset, 1, zv);
959 zv->zv_objset = NULL;
960 }
961
962 typedef struct minors_job {
963 list_t *list;
964 list_node_t link;
965 /* input */
966 char *name;
967 /* output */
968 int error;
969 } minors_job_t;
970
971 /*
972 * Prefetch zvol dnodes for the minors_job
973 */
974 static void
zvol_prefetch_minors_impl(void * arg)975 zvol_prefetch_minors_impl(void *arg)
976 {
977 minors_job_t *job = arg;
978 char *dsname = job->name;
979 objset_t *os = NULL;
980
981 job->error = dmu_objset_own(dsname, DMU_OST_ZVOL, B_TRUE, B_TRUE,
982 FTAG, &os);
983 if (job->error == 0) {
984 dmu_prefetch(os, ZVOL_OBJ, 0, 0, 0, ZIO_PRIORITY_SYNC_READ);
985 dmu_objset_disown(os, B_TRUE, FTAG);
986 }
987 }
988
989 /*
990 * Mask errors to continue dmu_objset_find() traversal
991 */
992 static int
zvol_create_snap_minor_cb(const char * dsname,void * arg)993 zvol_create_snap_minor_cb(const char *dsname, void *arg)
994 {
995 minors_job_t *j = arg;
996 list_t *minors_list = j->list;
997 const char *name = j->name;
998
999 ASSERT0(MUTEX_HELD(&spa_namespace_lock));
1000
1001 /* skip the designated dataset */
1002 if (name && strcmp(dsname, name) == 0)
1003 return (0);
1004
1005 /* at this point, the dsname should name a snapshot */
1006 if (strchr(dsname, '@') == 0) {
1007 dprintf("zvol_create_snap_minor_cb(): "
1008 "%s is not a snapshot name\n", dsname);
1009 } else {
1010 minors_job_t *job;
1011 char *n = kmem_strdup(dsname);
1012 if (n == NULL)
1013 return (0);
1014
1015 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP);
1016 job->name = n;
1017 job->list = minors_list;
1018 job->error = 0;
1019 list_insert_tail(minors_list, job);
1020 /* don't care if dispatch fails, because job->error is 0 */
1021 taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job,
1022 TQ_SLEEP);
1023 }
1024
1025 return (0);
1026 }
1027
1028 /*
1029 * Mask errors to continue dmu_objset_find() traversal
1030 */
1031 static int
zvol_create_minors_cb(const char * dsname,void * arg)1032 zvol_create_minors_cb(const char *dsname, void *arg)
1033 {
1034 uint64_t snapdev;
1035 int error;
1036 list_t *minors_list = arg;
1037
1038 ASSERT0(MUTEX_HELD(&spa_namespace_lock));
1039
1040 error = dsl_prop_get_integer(dsname, "snapdev", &snapdev, NULL);
1041 if (error)
1042 return (0);
1043
1044 /*
1045 * Given the name and the 'snapdev' property, create device minor nodes
1046 * with the linkages to zvols/snapshots as needed.
1047 * If the name represents a zvol, create a minor node for the zvol, then
1048 * check if its snapshots are 'visible', and if so, iterate over the
1049 * snapshots and create device minor nodes for those.
1050 */
1051 if (strchr(dsname, '@') == 0) {
1052 minors_job_t *job;
1053 char *n = kmem_strdup(dsname);
1054 if (n == NULL)
1055 return (0);
1056
1057 job = kmem_alloc(sizeof (minors_job_t), KM_SLEEP);
1058 job->name = n;
1059 job->list = minors_list;
1060 job->error = 0;
1061 list_insert_tail(minors_list, job);
1062 /* don't care if dispatch fails, because job->error is 0 */
1063 taskq_dispatch(system_taskq, zvol_prefetch_minors_impl, job,
1064 TQ_SLEEP);
1065
1066 if (snapdev == ZFS_SNAPDEV_VISIBLE) {
1067 /*
1068 * traverse snapshots only, do not traverse children,
1069 * and skip the 'dsname'
1070 */
1071 error = dmu_objset_find(dsname,
1072 zvol_create_snap_minor_cb, (void *)job,
1073 DS_FIND_SNAPSHOTS);
1074 }
1075 } else {
1076 dprintf("zvol_create_minors_cb(): %s is not a zvol name\n",
1077 dsname);
1078 }
1079
1080 return (0);
1081 }
1082
1083 /*
1084 * Create minors for the specified dataset, including children and snapshots.
1085 * Pay attention to the 'snapdev' property and iterate over the snapshots
1086 * only if they are 'visible'. This approach allows one to assure that the
1087 * snapshot metadata is read from disk only if it is needed.
1088 *
1089 * The name can represent a dataset to be recursively scanned for zvols and
1090 * their snapshots, or a single zvol snapshot. If the name represents a
1091 * dataset, the scan is performed in two nested stages:
1092 * - scan the dataset for zvols, and
1093 * - for each zvol, create a minor node, then check if the zvol's snapshots
1094 * are 'visible', and only then iterate over the snapshots if needed
1095 *
1096 * If the name represents a snapshot, a check is performed if the snapshot is
1097 * 'visible' (which also verifies that the parent is a zvol), and if so,
1098 * a minor node for that snapshot is created.
1099 */
1100 void
zvol_create_minors_recursive(const char * name)1101 zvol_create_minors_recursive(const char *name)
1102 {
1103 list_t minors_list;
1104 minors_job_t *job;
1105
1106 if (zvol_inhibit_dev)
1107 return;
1108
1109 /*
1110 * This is the list for prefetch jobs. Whenever we found a match
1111 * during dmu_objset_find, we insert a minors_job to the list and do
1112 * taskq_dispatch to parallel prefetch zvol dnodes. Note we don't need
1113 * any lock because all list operation is done on the current thread.
1114 *
1115 * We will use this list to do zvol_create_minor_impl after prefetch
1116 * so we don't have to traverse using dmu_objset_find again.
1117 */
1118 list_create(&minors_list, sizeof (minors_job_t),
1119 offsetof(minors_job_t, link));
1120
1121
1122 if (strchr(name, '@') != NULL) {
1123 uint64_t snapdev;
1124
1125 int error = dsl_prop_get_integer(name, "snapdev",
1126 &snapdev, NULL);
1127
1128 if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE)
1129 (void) ops->zv_create_minor(name);
1130 } else {
1131 fstrans_cookie_t cookie = spl_fstrans_mark();
1132 (void) dmu_objset_find(name, zvol_create_minors_cb,
1133 &minors_list, DS_FIND_CHILDREN);
1134 spl_fstrans_unmark(cookie);
1135 }
1136
1137 taskq_wait_outstanding(system_taskq, 0);
1138
1139 /*
1140 * Prefetch is completed, we can do zvol_create_minor_impl
1141 * sequentially.
1142 */
1143 while ((job = list_head(&minors_list)) != NULL) {
1144 list_remove(&minors_list, job);
1145 if (!job->error)
1146 (void) ops->zv_create_minor(job->name);
1147 kmem_strfree(job->name);
1148 kmem_free(job, sizeof (minors_job_t));
1149 }
1150
1151 list_destroy(&minors_list);
1152 }
1153
1154 void
zvol_create_minor(const char * name)1155 zvol_create_minor(const char *name)
1156 {
1157 /*
1158 * Note: the dsl_pool_config_lock must not be held.
1159 * Minor node creation needs to obtain the zvol_state_lock.
1160 * zvol_open() obtains the zvol_state_lock and then the dsl pool
1161 * config lock. Therefore, we can't have the config lock now if
1162 * we are going to wait for the zvol_state_lock, because it
1163 * would be a lock order inversion which could lead to deadlock.
1164 */
1165
1166 if (zvol_inhibit_dev)
1167 return;
1168
1169 if (strchr(name, '@') != NULL) {
1170 uint64_t snapdev;
1171
1172 int error = dsl_prop_get_integer(name,
1173 "snapdev", &snapdev, NULL);
1174
1175 if (error == 0 && snapdev == ZFS_SNAPDEV_VISIBLE)
1176 (void) ops->zv_create_minor(name);
1177 } else {
1178 (void) ops->zv_create_minor(name);
1179 }
1180 }
1181
1182 /*
1183 * Remove minors for specified dataset including children and snapshots.
1184 */
1185
1186 void
zvol_remove_minors_impl(const char * name)1187 zvol_remove_minors_impl(const char *name)
1188 {
1189 zvol_state_t *zv, *zv_next;
1190 int namelen = ((name) ? strlen(name) : 0);
1191 taskqid_t t;
1192 list_t free_list;
1193
1194 if (zvol_inhibit_dev)
1195 return;
1196
1197 list_create(&free_list, sizeof (zvol_state_t),
1198 offsetof(zvol_state_t, zv_next));
1199
1200 rw_enter(&zvol_state_lock, RW_WRITER);
1201
1202 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1203 zv_next = list_next(&zvol_state_list, zv);
1204
1205 mutex_enter(&zv->zv_state_lock);
1206 if (name == NULL || strcmp(zv->zv_name, name) == 0 ||
1207 (strncmp(zv->zv_name, name, namelen) == 0 &&
1208 (zv->zv_name[namelen] == '/' ||
1209 zv->zv_name[namelen] == '@'))) {
1210 /*
1211 * By holding zv_state_lock here, we guarantee that no
1212 * one is currently using this zv
1213 */
1214
1215 /* If in use, leave alone */
1216 if (zv->zv_open_count > 0 ||
1217 atomic_read(&zv->zv_suspend_ref)) {
1218 mutex_exit(&zv->zv_state_lock);
1219 continue;
1220 }
1221
1222 zvol_remove(zv);
1223
1224 /*
1225 * Cleared while holding zvol_state_lock as a writer
1226 * which will prevent zvol_open() from opening it.
1227 */
1228 ops->zv_clear_private(zv);
1229
1230 /* Drop zv_state_lock before zvol_free() */
1231 mutex_exit(&zv->zv_state_lock);
1232
1233 /* Try parallel zv_free, if failed do it in place */
1234 t = taskq_dispatch(system_taskq,
1235 (task_func_t *)ops->zv_free, zv, TQ_SLEEP);
1236 if (t == TASKQID_INVALID)
1237 list_insert_head(&free_list, zv);
1238 } else {
1239 mutex_exit(&zv->zv_state_lock);
1240 }
1241 }
1242 rw_exit(&zvol_state_lock);
1243
1244 /* Drop zvol_state_lock before calling zvol_free() */
1245 while ((zv = list_head(&free_list)) != NULL) {
1246 list_remove(&free_list, zv);
1247 ops->zv_free(zv);
1248 }
1249 }
1250
1251 /* Remove minor for this specific volume only */
1252 static void
zvol_remove_minor_impl(const char * name)1253 zvol_remove_minor_impl(const char *name)
1254 {
1255 zvol_state_t *zv = NULL, *zv_next;
1256
1257 if (zvol_inhibit_dev)
1258 return;
1259
1260 rw_enter(&zvol_state_lock, RW_WRITER);
1261
1262 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1263 zv_next = list_next(&zvol_state_list, zv);
1264
1265 mutex_enter(&zv->zv_state_lock);
1266 if (strcmp(zv->zv_name, name) == 0) {
1267 /*
1268 * By holding zv_state_lock here, we guarantee that no
1269 * one is currently using this zv
1270 */
1271
1272 /* If in use, leave alone */
1273 if (zv->zv_open_count > 0 ||
1274 atomic_read(&zv->zv_suspend_ref)) {
1275 mutex_exit(&zv->zv_state_lock);
1276 continue;
1277 }
1278 zvol_remove(zv);
1279
1280 ops->zv_clear_private(zv);
1281 mutex_exit(&zv->zv_state_lock);
1282 break;
1283 } else {
1284 mutex_exit(&zv->zv_state_lock);
1285 }
1286 }
1287
1288 /* Drop zvol_state_lock before calling zvol_free() */
1289 rw_exit(&zvol_state_lock);
1290
1291 if (zv != NULL)
1292 ops->zv_free(zv);
1293 }
1294
1295 /*
1296 * Rename minors for specified dataset including children and snapshots.
1297 */
1298 static void
zvol_rename_minors_impl(const char * oldname,const char * newname)1299 zvol_rename_minors_impl(const char *oldname, const char *newname)
1300 {
1301 zvol_state_t *zv, *zv_next;
1302 int oldnamelen, newnamelen;
1303
1304 if (zvol_inhibit_dev)
1305 return;
1306
1307 oldnamelen = strlen(oldname);
1308 newnamelen = strlen(newname);
1309
1310 rw_enter(&zvol_state_lock, RW_READER);
1311
1312 for (zv = list_head(&zvol_state_list); zv != NULL; zv = zv_next) {
1313 zv_next = list_next(&zvol_state_list, zv);
1314
1315 mutex_enter(&zv->zv_state_lock);
1316
1317 if (strcmp(zv->zv_name, oldname) == 0) {
1318 ops->zv_rename_minor(zv, newname);
1319 } else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
1320 (zv->zv_name[oldnamelen] == '/' ||
1321 zv->zv_name[oldnamelen] == '@')) {
1322 char *name = kmem_asprintf("%s%c%s", newname,
1323 zv->zv_name[oldnamelen],
1324 zv->zv_name + oldnamelen + 1);
1325 ops->zv_rename_minor(zv, name);
1326 kmem_strfree(name);
1327 }
1328
1329 mutex_exit(&zv->zv_state_lock);
1330 }
1331
1332 rw_exit(&zvol_state_lock);
1333 }
1334
1335 typedef struct zvol_snapdev_cb_arg {
1336 uint64_t snapdev;
1337 } zvol_snapdev_cb_arg_t;
1338
1339 static int
zvol_set_snapdev_cb(const char * dsname,void * param)1340 zvol_set_snapdev_cb(const char *dsname, void *param)
1341 {
1342 zvol_snapdev_cb_arg_t *arg = param;
1343
1344 if (strchr(dsname, '@') == NULL)
1345 return (0);
1346
1347 switch (arg->snapdev) {
1348 case ZFS_SNAPDEV_VISIBLE:
1349 (void) ops->zv_create_minor(dsname);
1350 break;
1351 case ZFS_SNAPDEV_HIDDEN:
1352 (void) zvol_remove_minor_impl(dsname);
1353 break;
1354 }
1355
1356 return (0);
1357 }
1358
1359 static void
zvol_set_snapdev_impl(char * name,uint64_t snapdev)1360 zvol_set_snapdev_impl(char *name, uint64_t snapdev)
1361 {
1362 zvol_snapdev_cb_arg_t arg = {snapdev};
1363 fstrans_cookie_t cookie = spl_fstrans_mark();
1364 /*
1365 * The zvol_set_snapdev_sync() sets snapdev appropriately
1366 * in the dataset hierarchy. Here, we only scan snapshots.
1367 */
1368 dmu_objset_find(name, zvol_set_snapdev_cb, &arg, DS_FIND_SNAPSHOTS);
1369 spl_fstrans_unmark(cookie);
1370 }
1371
1372 typedef struct zvol_volmode_cb_arg {
1373 uint64_t volmode;
1374 } zvol_volmode_cb_arg_t;
1375
1376 static void
zvol_set_volmode_impl(char * name,uint64_t volmode)1377 zvol_set_volmode_impl(char *name, uint64_t volmode)
1378 {
1379 fstrans_cookie_t cookie;
1380 uint64_t old_volmode;
1381 zvol_state_t *zv;
1382
1383 if (strchr(name, '@') != NULL)
1384 return;
1385
1386 /*
1387 * It's unfortunate we need to remove minors before we create new ones:
1388 * this is necessary because our backing gendisk (zvol_state->zv_disk)
1389 * could be different when we set, for instance, volmode from "geom"
1390 * to "dev" (or vice versa).
1391 */
1392 zv = zvol_find_by_name(name, RW_NONE);
1393 if (zv == NULL && volmode == ZFS_VOLMODE_NONE)
1394 return;
1395 if (zv != NULL) {
1396 old_volmode = zv->zv_volmode;
1397 mutex_exit(&zv->zv_state_lock);
1398 if (old_volmode == volmode)
1399 return;
1400 zvol_wait_close(zv);
1401 }
1402 cookie = spl_fstrans_mark();
1403 switch (volmode) {
1404 case ZFS_VOLMODE_NONE:
1405 (void) zvol_remove_minor_impl(name);
1406 break;
1407 case ZFS_VOLMODE_GEOM:
1408 case ZFS_VOLMODE_DEV:
1409 (void) zvol_remove_minor_impl(name);
1410 (void) ops->zv_create_minor(name);
1411 break;
1412 case ZFS_VOLMODE_DEFAULT:
1413 (void) zvol_remove_minor_impl(name);
1414 if (zvol_volmode == ZFS_VOLMODE_NONE)
1415 break;
1416 else /* if zvol_volmode is invalid defaults to "geom" */
1417 (void) ops->zv_create_minor(name);
1418 break;
1419 }
1420 spl_fstrans_unmark(cookie);
1421 }
1422
1423 static zvol_task_t *
zvol_task_alloc(zvol_async_op_t op,const char * name1,const char * name2,uint64_t value)1424 zvol_task_alloc(zvol_async_op_t op, const char *name1, const char *name2,
1425 uint64_t value)
1426 {
1427 zvol_task_t *task;
1428 char *delim;
1429
1430 /* Never allow tasks on hidden names. */
1431 if (name1[0] == '$')
1432 return (NULL);
1433
1434 task = kmem_zalloc(sizeof (zvol_task_t), KM_SLEEP);
1435 task->op = op;
1436 task->value = value;
1437 delim = strchr(name1, '/');
1438 strlcpy(task->pool, name1, delim ? (delim - name1 + 1) : MAXNAMELEN);
1439
1440 strlcpy(task->name1, name1, MAXNAMELEN);
1441 if (name2 != NULL)
1442 strlcpy(task->name2, name2, MAXNAMELEN);
1443
1444 return (task);
1445 }
1446
1447 static void
zvol_task_free(zvol_task_t * task)1448 zvol_task_free(zvol_task_t *task)
1449 {
1450 kmem_free(task, sizeof (zvol_task_t));
1451 }
1452
1453 /*
1454 * The worker thread function performed asynchronously.
1455 */
1456 static void
zvol_task_cb(void * arg)1457 zvol_task_cb(void *arg)
1458 {
1459 zvol_task_t *task = arg;
1460
1461 switch (task->op) {
1462 case ZVOL_ASYNC_REMOVE_MINORS:
1463 zvol_remove_minors_impl(task->name1);
1464 break;
1465 case ZVOL_ASYNC_RENAME_MINORS:
1466 zvol_rename_minors_impl(task->name1, task->name2);
1467 break;
1468 case ZVOL_ASYNC_SET_SNAPDEV:
1469 zvol_set_snapdev_impl(task->name1, task->value);
1470 break;
1471 case ZVOL_ASYNC_SET_VOLMODE:
1472 zvol_set_volmode_impl(task->name1, task->value);
1473 break;
1474 default:
1475 VERIFY(0);
1476 break;
1477 }
1478
1479 zvol_task_free(task);
1480 }
1481
1482 typedef struct zvol_set_prop_int_arg {
1483 const char *zsda_name;
1484 uint64_t zsda_value;
1485 zprop_source_t zsda_source;
1486 dmu_tx_t *zsda_tx;
1487 } zvol_set_prop_int_arg_t;
1488
1489 /*
1490 * Sanity check the dataset for safe use by the sync task. No additional
1491 * conditions are imposed.
1492 */
1493 static int
zvol_set_snapdev_check(void * arg,dmu_tx_t * tx)1494 zvol_set_snapdev_check(void *arg, dmu_tx_t *tx)
1495 {
1496 zvol_set_prop_int_arg_t *zsda = arg;
1497 dsl_pool_t *dp = dmu_tx_pool(tx);
1498 dsl_dir_t *dd;
1499 int error;
1500
1501 error = dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL);
1502 if (error != 0)
1503 return (error);
1504
1505 dsl_dir_rele(dd, FTAG);
1506
1507 return (error);
1508 }
1509
1510 /* ARGSUSED */
1511 static int
zvol_set_snapdev_sync_cb(dsl_pool_t * dp,dsl_dataset_t * ds,void * arg)1512 zvol_set_snapdev_sync_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1513 {
1514 char dsname[MAXNAMELEN];
1515 zvol_task_t *task;
1516 uint64_t snapdev;
1517
1518 dsl_dataset_name(ds, dsname);
1519 if (dsl_prop_get_int_ds(ds, "snapdev", &snapdev) != 0)
1520 return (0);
1521 task = zvol_task_alloc(ZVOL_ASYNC_SET_SNAPDEV, dsname, NULL, snapdev);
1522 if (task == NULL)
1523 return (0);
1524
1525 (void) taskq_dispatch(dp->dp_spa->spa_zvol_taskq, zvol_task_cb,
1526 task, TQ_SLEEP);
1527 return (0);
1528 }
1529
1530 /*
1531 * Traverse all child datasets and apply snapdev appropriately.
1532 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
1533 * dataset and read the effective "snapdev" on every child in the callback
1534 * function: this is because the value is not guaranteed to be the same in the
1535 * whole dataset hierarchy.
1536 */
1537 static void
zvol_set_snapdev_sync(void * arg,dmu_tx_t * tx)1538 zvol_set_snapdev_sync(void *arg, dmu_tx_t *tx)
1539 {
1540 zvol_set_prop_int_arg_t *zsda = arg;
1541 dsl_pool_t *dp = dmu_tx_pool(tx);
1542 dsl_dir_t *dd;
1543 dsl_dataset_t *ds;
1544 int error;
1545
1546 VERIFY0(dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL));
1547 zsda->zsda_tx = tx;
1548
1549 error = dsl_dataset_hold(dp, zsda->zsda_name, FTAG, &ds);
1550 if (error == 0) {
1551 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_SNAPDEV),
1552 zsda->zsda_source, sizeof (zsda->zsda_value), 1,
1553 &zsda->zsda_value, zsda->zsda_tx);
1554 dsl_dataset_rele(ds, FTAG);
1555 }
1556 dmu_objset_find_dp(dp, dd->dd_object, zvol_set_snapdev_sync_cb,
1557 zsda, DS_FIND_CHILDREN);
1558
1559 dsl_dir_rele(dd, FTAG);
1560 }
1561
1562 int
zvol_set_snapdev(const char * ddname,zprop_source_t source,uint64_t snapdev)1563 zvol_set_snapdev(const char *ddname, zprop_source_t source, uint64_t snapdev)
1564 {
1565 zvol_set_prop_int_arg_t zsda;
1566
1567 zsda.zsda_name = ddname;
1568 zsda.zsda_source = source;
1569 zsda.zsda_value = snapdev;
1570
1571 return (dsl_sync_task(ddname, zvol_set_snapdev_check,
1572 zvol_set_snapdev_sync, &zsda, 0, ZFS_SPACE_CHECK_NONE));
1573 }
1574
1575 /*
1576 * Sanity check the dataset for safe use by the sync task. No additional
1577 * conditions are imposed.
1578 */
1579 static int
zvol_set_volmode_check(void * arg,dmu_tx_t * tx)1580 zvol_set_volmode_check(void *arg, dmu_tx_t *tx)
1581 {
1582 zvol_set_prop_int_arg_t *zsda = arg;
1583 dsl_pool_t *dp = dmu_tx_pool(tx);
1584 dsl_dir_t *dd;
1585 int error;
1586
1587 error = dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL);
1588 if (error != 0)
1589 return (error);
1590
1591 dsl_dir_rele(dd, FTAG);
1592
1593 return (error);
1594 }
1595
1596 /* ARGSUSED */
1597 static int
zvol_set_volmode_sync_cb(dsl_pool_t * dp,dsl_dataset_t * ds,void * arg)1598 zvol_set_volmode_sync_cb(dsl_pool_t *dp, dsl_dataset_t *ds, void *arg)
1599 {
1600 char dsname[MAXNAMELEN];
1601 zvol_task_t *task;
1602 uint64_t volmode;
1603
1604 dsl_dataset_name(ds, dsname);
1605 if (dsl_prop_get_int_ds(ds, "volmode", &volmode) != 0)
1606 return (0);
1607 task = zvol_task_alloc(ZVOL_ASYNC_SET_VOLMODE, dsname, NULL, volmode);
1608 if (task == NULL)
1609 return (0);
1610
1611 (void) taskq_dispatch(dp->dp_spa->spa_zvol_taskq, zvol_task_cb,
1612 task, TQ_SLEEP);
1613 return (0);
1614 }
1615
1616 /*
1617 * Traverse all child datasets and apply volmode appropriately.
1618 * We call dsl_prop_set_sync_impl() here to set the value only on the toplevel
1619 * dataset and read the effective "volmode" on every child in the callback
1620 * function: this is because the value is not guaranteed to be the same in the
1621 * whole dataset hierarchy.
1622 */
1623 static void
zvol_set_volmode_sync(void * arg,dmu_tx_t * tx)1624 zvol_set_volmode_sync(void *arg, dmu_tx_t *tx)
1625 {
1626 zvol_set_prop_int_arg_t *zsda = arg;
1627 dsl_pool_t *dp = dmu_tx_pool(tx);
1628 dsl_dir_t *dd;
1629 dsl_dataset_t *ds;
1630 int error;
1631
1632 VERIFY0(dsl_dir_hold(dp, zsda->zsda_name, FTAG, &dd, NULL));
1633 zsda->zsda_tx = tx;
1634
1635 error = dsl_dataset_hold(dp, zsda->zsda_name, FTAG, &ds);
1636 if (error == 0) {
1637 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_VOLMODE),
1638 zsda->zsda_source, sizeof (zsda->zsda_value), 1,
1639 &zsda->zsda_value, zsda->zsda_tx);
1640 dsl_dataset_rele(ds, FTAG);
1641 }
1642
1643 dmu_objset_find_dp(dp, dd->dd_object, zvol_set_volmode_sync_cb,
1644 zsda, DS_FIND_CHILDREN);
1645
1646 dsl_dir_rele(dd, FTAG);
1647 }
1648
1649 int
zvol_set_volmode(const char * ddname,zprop_source_t source,uint64_t volmode)1650 zvol_set_volmode(const char *ddname, zprop_source_t source, uint64_t volmode)
1651 {
1652 zvol_set_prop_int_arg_t zsda;
1653
1654 zsda.zsda_name = ddname;
1655 zsda.zsda_source = source;
1656 zsda.zsda_value = volmode;
1657
1658 return (dsl_sync_task(ddname, zvol_set_volmode_check,
1659 zvol_set_volmode_sync, &zsda, 0, ZFS_SPACE_CHECK_NONE));
1660 }
1661
1662 void
zvol_remove_minors(spa_t * spa,const char * name,boolean_t async)1663 zvol_remove_minors(spa_t *spa, const char *name, boolean_t async)
1664 {
1665 zvol_task_t *task;
1666 taskqid_t id;
1667
1668 task = zvol_task_alloc(ZVOL_ASYNC_REMOVE_MINORS, name, NULL, ~0ULL);
1669 if (task == NULL)
1670 return;
1671
1672 id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP);
1673 if ((async == B_FALSE) && (id != TASKQID_INVALID))
1674 taskq_wait_id(spa->spa_zvol_taskq, id);
1675 }
1676
1677 void
zvol_rename_minors(spa_t * spa,const char * name1,const char * name2,boolean_t async)1678 zvol_rename_minors(spa_t *spa, const char *name1, const char *name2,
1679 boolean_t async)
1680 {
1681 zvol_task_t *task;
1682 taskqid_t id;
1683
1684 task = zvol_task_alloc(ZVOL_ASYNC_RENAME_MINORS, name1, name2, ~0ULL);
1685 if (task == NULL)
1686 return;
1687
1688 id = taskq_dispatch(spa->spa_zvol_taskq, zvol_task_cb, task, TQ_SLEEP);
1689 if ((async == B_FALSE) && (id != TASKQID_INVALID))
1690 taskq_wait_id(spa->spa_zvol_taskq, id);
1691 }
1692
1693 boolean_t
zvol_is_zvol(const char * name)1694 zvol_is_zvol(const char *name)
1695 {
1696
1697 return (ops->zv_is_zvol(name));
1698 }
1699
1700 void
zvol_register_ops(const zvol_platform_ops_t * zvol_ops)1701 zvol_register_ops(const zvol_platform_ops_t *zvol_ops)
1702 {
1703 ops = zvol_ops;
1704 }
1705
1706 int
zvol_init_impl(void)1707 zvol_init_impl(void)
1708 {
1709 int i;
1710
1711 list_create(&zvol_state_list, sizeof (zvol_state_t),
1712 offsetof(zvol_state_t, zv_next));
1713 rw_init(&zvol_state_lock, NULL, RW_DEFAULT, NULL);
1714
1715 zvol_htable = kmem_alloc(ZVOL_HT_SIZE * sizeof (struct hlist_head),
1716 KM_SLEEP);
1717 for (i = 0; i < ZVOL_HT_SIZE; i++)
1718 INIT_HLIST_HEAD(&zvol_htable[i]);
1719
1720 return (0);
1721 }
1722
1723 void
zvol_fini_impl(void)1724 zvol_fini_impl(void)
1725 {
1726 zvol_remove_minors_impl(NULL);
1727
1728 /*
1729 * The call to "zvol_remove_minors_impl" may dispatch entries to
1730 * the system_taskq, but it doesn't wait for those entries to
1731 * complete before it returns. Thus, we must wait for all of the
1732 * removals to finish, before we can continue.
1733 */
1734 taskq_wait_outstanding(system_taskq, 0);
1735
1736 kmem_free(zvol_htable, ZVOL_HT_SIZE * sizeof (struct hlist_head));
1737 list_destroy(&zvol_state_list);
1738 rw_destroy(&zvol_state_lock);
1739 }
1740