?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
8585 improve batching done in zil_commit()
[unleashed.git] / usr / src / uts / common / fs / zfs / zvol.c
blob9a02d465ede0dd7a59ccbb9fbf30ed874715d417
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 *
24 * Portions Copyright 2010 Robert Milkowski
25 *
26 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
27 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
28 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 */
31
32 /*
33 * ZFS volume emulation driver.
34 *
35 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
36 * Volumes are accessed through the symbolic links named:
37 *
38 * /dev/zvol/dsk/<pool_name>/<dataset_name>
39 * /dev/zvol/rdsk/<pool_name>/<dataset_name>
40 *
41 * These links are created by the /dev filesystem (sdev_zvolops.c).
42 * Volumes are persistent through reboot. No user command needs to be
43 * run before opening and using a device.
44 */
45
46 #include <sys/types.h>
47 #include <sys/param.h>
48 #include <sys/errno.h>
49 #include <sys/uio.h>
50 #include <sys/buf.h>
51 #include <sys/modctl.h>
52 #include <sys/open.h>
53 #include <sys/kmem.h>
54 #include <sys/conf.h>
55 #include <sys/cmn_err.h>
56 #include <sys/stat.h>
57 #include <sys/zap.h>
58 #include <sys/spa.h>
59 #include <sys/spa_impl.h>
60 #include <sys/zio.h>
61 #include <sys/dmu_traverse.h>
62 #include <sys/dnode.h>
63 #include <sys/dsl_dataset.h>
64 #include <sys/dsl_prop.h>
65 #include <sys/dkio.h>
66 #include <sys/efi_partition.h>
67 #include <sys/byteorder.h>
68 #include <sys/pathname.h>
69 #include <sys/ddi.h>
70 #include <sys/sunddi.h>
71 #include <sys/crc32.h>
72 #include <sys/dirent.h>
73 #include <sys/policy.h>
74 #include <sys/fs/zfs.h>
75 #include <sys/zfs_ioctl.h>
76 #include <sys/mkdev.h>
77 #include <sys/zil.h>
78 #include <sys/refcount.h>
79 #include <sys/zfs_znode.h>
80 #include <sys/zfs_rlock.h>
81 #include <sys/vdev_disk.h>
82 #include <sys/vdev_impl.h>
83 #include <sys/vdev_raidz.h>
84 #include <sys/zvol.h>
85 #include <sys/dumphdr.h>
86 #include <sys/zil_impl.h>
87 #include <sys/dbuf.h>
88 #include <sys/dmu_tx.h>
89 #include <sys/zfeature.h>
90 #include <sys/zio_checksum.h>
91 #include <sys/zil_impl.h>
92
93 #include "zfs_namecheck.h"
94
95 void *zfsdev_state;
96 static char *zvol_tag = "zvol_tag";
97
98 #define ZVOL_DUMPSIZE "dumpsize"
99
100 /*
101 * This lock protects the zfsdev_state structure from being modified
102 * while it's being used, e.g. an open that comes in before a create
103 * finishes. It also protects temporary opens of the dataset so that,
104 * e.g., an open doesn't get a spurious EBUSY.
105 */
106 kmutex_t zfsdev_state_lock;
107 static uint32_t zvol_minors;
108
109 typedef struct zvol_extent {
110 list_node_t ze_node;
111 dva_t ze_dva; /* dva associated with this extent */
112 uint64_t ze_nblks; /* number of blocks in extent */
113 } zvol_extent_t;
114
115 /*
116 * The in-core state of each volume.
117 */
118 typedef struct zvol_state {
119 char zv_name[MAXPATHLEN]; /* pool/dd name */
120 uint64_t zv_volsize; /* amount of space we advertise */
121 uint64_t zv_volblocksize; /* volume block size */
122 minor_t zv_minor; /* minor number */
123 uint8_t zv_min_bs; /* minimum addressable block shift */
124 uint8_t zv_flags; /* readonly, dumpified, etc. */
125 objset_t *zv_objset; /* objset handle */
126 uint32_t zv_open_count[OTYPCNT]; /* open counts */
127 uint32_t zv_total_opens; /* total open count */
128 zilog_t *zv_zilog; /* ZIL handle */
129 list_t zv_extents; /* List of extents for dump */
130 znode_t zv_znode; /* for range locking */
131 dmu_buf_t *zv_dbuf; /* bonus handle */
132 } zvol_state_t;
133
134 /*
135 * zvol specific flags
136 */
137 #define ZVOL_RDONLY 0x1
138 #define ZVOL_DUMPIFIED 0x2
139 #define ZVOL_EXCL 0x4
140 #define ZVOL_WCE 0x8
141
142 /*
143 * zvol maximum transfer in one DMU tx.
144 */
145 int zvol_maxphys = DMU_MAX_ACCESS/2;
146
147 /*
148 * Toggle unmap functionality.
149 */
150 boolean_t zvol_unmap_enabled = B_TRUE;
151
152 /*
153 * If true, unmaps requested as synchronous are executed synchronously,
154 * otherwise all unmaps are asynchronous.
155 */
156 boolean_t zvol_unmap_sync_enabled = B_FALSE;
157
158 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
159 nvlist_t *, nvlist_t *);
160 static int zvol_remove_zv(zvol_state_t *);
161 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf,
162 struct lwb *lwb, zio_t *zio);
163 static int zvol_dumpify(zvol_state_t *zv);
164 static int zvol_dump_fini(zvol_state_t *zv);
165 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
166
167 static void
168 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
169 {
170 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
171
172 zv->zv_volsize = volsize;
173 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
174 "Size", volsize) == DDI_SUCCESS);
175 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
176 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
177
178 /* Notify specfs to invalidate the cached size */
179 spec_size_invalidate(dev, VBLK);
180 spec_size_invalidate(dev, VCHR);
181 }
182
183 int
184 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
185 {
186 if (volsize == 0)
187 return (SET_ERROR(EINVAL));
188
189 if (volsize % blocksize != 0)
190 return (SET_ERROR(EINVAL));
191
192 #ifdef _ILP32
193 if (volsize - 1 > SPEC_MAXOFFSET_T)
194 return (SET_ERROR(EOVERFLOW));
195 #endif
196 return (0);
197 }
198
199 int
200 zvol_check_volblocksize(uint64_t volblocksize)
201 {
202 if (volblocksize < SPA_MINBLOCKSIZE ||
203 volblocksize > SPA_OLD_MAXBLOCKSIZE ||
204 !ISP2(volblocksize))
205 return (SET_ERROR(EDOM));
206
207 return (0);
208 }
209
210 int
211 zvol_get_stats(objset_t *os, nvlist_t *nv)
212 {
213 int error;
214 dmu_object_info_t doi;
215 uint64_t val;
216
217 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
218 if (error)
219 return (error);
220
221 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
222
223 error = dmu_object_info(os, ZVOL_OBJ, &doi);
224
225 if (error == 0) {
226 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
227 doi.doi_data_block_size);
228 }
229
230 return (error);
231 }
232
233 static zvol_state_t *
234 zvol_minor_lookup(const char *name)
235 {
236 minor_t minor;
237 zvol_state_t *zv;
238
239 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
240
241 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
242 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
243 if (zv == NULL)
244 continue;
245 if (strcmp(zv->zv_name, name) == 0)
246 return (zv);
247 }
248
249 return (NULL);
250 }
251
252 /* extent mapping arg */
253 struct maparg {
254 zvol_state_t *ma_zv;
255 uint64_t ma_blks;
256 };
257
258 /*ARGSUSED*/
259 static int
260 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
261 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
262 {
263 struct maparg *ma = arg;
264 zvol_extent_t *ze;
265 int bs = ma->ma_zv->zv_volblocksize;
266
267 if (bp == NULL || BP_IS_HOLE(bp) ||
268 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
269 return (0);
270
271 VERIFY(!BP_IS_EMBEDDED(bp));
272
273 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
274 ma->ma_blks++;
275
276 /* Abort immediately if we have encountered gang blocks */
277 if (BP_IS_GANG(bp))
278 return (SET_ERROR(EFRAGS));
279
280 /*
281 * See if the block is at the end of the previous extent.
282 */
283 ze = list_tail(&ma->ma_zv->zv_extents);
284 if (ze &&
285 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
286 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
287 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
288 ze->ze_nblks++;
289 return (0);
290 }
291
292 dprintf_bp(bp, "%s", "next blkptr:");
293
294 /* start a new extent */
295 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
296 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
297 ze->ze_nblks = 1;
298 list_insert_tail(&ma->ma_zv->zv_extents, ze);
299 return (0);
300 }
301
302 static void
303 zvol_free_extents(zvol_state_t *zv)
304 {
305 zvol_extent_t *ze;
306
307 while (ze = list_head(&zv->zv_extents)) {
308 list_remove(&zv->zv_extents, ze);
309 kmem_free(ze, sizeof (zvol_extent_t));
310 }
311 }
312
313 static int
314 zvol_get_lbas(zvol_state_t *zv)
315 {
316 objset_t *os = zv->zv_objset;
317 struct maparg ma;
318 int err;
319
320 ma.ma_zv = zv;
321 ma.ma_blks = 0;
322 zvol_free_extents(zv);
323
324 /* commit any in-flight changes before traversing the dataset */
325 txg_wait_synced(dmu_objset_pool(os), 0);
326 err = traverse_dataset(dmu_objset_ds(os), 0,
327 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
328 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
329 zvol_free_extents(zv);
330 return (err ? err : EIO);
331 }
332
333 return (0);
334 }
335
336 /* ARGSUSED */
337 void
338 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
339 {
340 zfs_creat_t *zct = arg;
341 nvlist_t *nvprops = zct->zct_props;
342 int error;
343 uint64_t volblocksize, volsize;
344
345 VERIFY(nvlist_lookup_uint64(nvprops,
346 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
347 if (nvlist_lookup_uint64(nvprops,
348 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
349 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
350
351 /*
352 * These properties must be removed from the list so the generic
353 * property setting step won't apply to them.
354 */
355 VERIFY(nvlist_remove_all(nvprops,
356 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
357 (void) nvlist_remove_all(nvprops,
358 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
359
360 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
361 DMU_OT_NONE, 0, tx);
362 ASSERT(error == 0);
363
364 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
365 DMU_OT_NONE, 0, tx);
366 ASSERT(error == 0);
367
368 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
369 ASSERT(error == 0);
370 }
371
372 /*
373 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
374 * implement DKIOCFREE/free-long-range.
375 */
376 static int
377 zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
378 {
379 uint64_t offset, length;
380
381 if (byteswap)
382 byteswap_uint64_array(lr, sizeof (*lr));
383
384 offset = lr->lr_offset;
385 length = lr->lr_length;
386
387 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
388 }
389
390 /*
391 * Replay a TX_WRITE ZIL transaction that didn't get committed
392 * after a system failure
393 */
394 static int
395 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
396 {
397 objset_t *os = zv->zv_objset;
398 char *data = (char *)(lr + 1); /* data follows lr_write_t */
399 uint64_t offset, length;
400 dmu_tx_t *tx;
401 int error;
402
403 if (byteswap)
404 byteswap_uint64_array(lr, sizeof (*lr));
405
406 offset = lr->lr_offset;
407 length = lr->lr_length;
408
409 /* If it's a dmu_sync() block, write the whole block */
410 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
411 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
412 if (length < blocksize) {
413 offset -= offset % blocksize;
414 length = blocksize;
415 }
416 }
417
418 tx = dmu_tx_create(os);
419 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
420 error = dmu_tx_assign(tx, TXG_WAIT);
421 if (error) {
422 dmu_tx_abort(tx);
423 } else {
424 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
425 dmu_tx_commit(tx);
426 }
427
428 return (error);
429 }
430
431 /* ARGSUSED */
432 static int
433 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
434 {
435 return (SET_ERROR(ENOTSUP));
436 }
437
438 /*
439 * Callback vectors for replaying records.
440 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
441 */
442 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
443 zvol_replay_err, /* 0 no such transaction type */
444 zvol_replay_err, /* TX_CREATE */
445 zvol_replay_err, /* TX_MKDIR */
446 zvol_replay_err, /* TX_MKXATTR */
447 zvol_replay_err, /* TX_SYMLINK */
448 zvol_replay_err, /* TX_REMOVE */
449 zvol_replay_err, /* TX_RMDIR */
450 zvol_replay_err, /* TX_LINK */
451 zvol_replay_err, /* TX_RENAME */
452 zvol_replay_write, /* TX_WRITE */
453 zvol_replay_truncate, /* TX_TRUNCATE */
454 zvol_replay_err, /* TX_SETATTR */
455 zvol_replay_err, /* TX_ACL */
456 zvol_replay_err, /* TX_CREATE_ACL */
457 zvol_replay_err, /* TX_CREATE_ATTR */
458 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
459 zvol_replay_err, /* TX_MKDIR_ACL */
460 zvol_replay_err, /* TX_MKDIR_ATTR */
461 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
462 zvol_replay_err, /* TX_WRITE2 */
463 };
464
465 int
466 zvol_name2minor(const char *name, minor_t *minor)
467 {
468 zvol_state_t *zv;
469
470 mutex_enter(&zfsdev_state_lock);
471 zv = zvol_minor_lookup(name);
472 if (minor && zv)
473 *minor = zv->zv_minor;
474 mutex_exit(&zfsdev_state_lock);
475 return (zv ? 0 : -1);
476 }
477
478 /*
479 * Create a minor node (plus a whole lot more) for the specified volume.
480 */
481 int
482 zvol_create_minor(const char *name)
483 {
484 zfs_soft_state_t *zs;
485 zvol_state_t *zv;
486 objset_t *os;
487 dmu_object_info_t doi;
488 minor_t minor = 0;
489 char chrbuf[30], blkbuf[30];
490 int error;
491
492 mutex_enter(&zfsdev_state_lock);
493
494 if (zvol_minor_lookup(name) != NULL) {
495 mutex_exit(&zfsdev_state_lock);
496 return (SET_ERROR(EEXIST));
497 }
498
499 /* lie and say we're read-only */
500 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
501
502 if (error) {
503 mutex_exit(&zfsdev_state_lock);
504 return (error);
505 }
506
507 if ((minor = zfsdev_minor_alloc()) == 0) {
508 dmu_objset_disown(os, FTAG);
509 mutex_exit(&zfsdev_state_lock);
510 return (SET_ERROR(ENXIO));
511 }
512
513 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
514 dmu_objset_disown(os, FTAG);
515 mutex_exit(&zfsdev_state_lock);
516 return (SET_ERROR(EAGAIN));
517 }
518 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
519 (char *)name);
520
521 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
522
523 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
524 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
525 ddi_soft_state_free(zfsdev_state, minor);
526 dmu_objset_disown(os, FTAG);
527 mutex_exit(&zfsdev_state_lock);
528 return (SET_ERROR(EAGAIN));
529 }
530
531 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
532
533 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
534 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
535 ddi_remove_minor_node(zfs_dip, chrbuf);
536 ddi_soft_state_free(zfsdev_state, minor);
537 dmu_objset_disown(os, FTAG);
538 mutex_exit(&zfsdev_state_lock);
539 return (SET_ERROR(EAGAIN));
540 }
541
542 zs = ddi_get_soft_state(zfsdev_state, minor);
543 zs->zss_type = ZSST_ZVOL;
544 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
545 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
546 zv->zv_min_bs = DEV_BSHIFT;
547 zv->zv_minor = minor;
548 zv->zv_objset = os;
549 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
550 zv->zv_flags |= ZVOL_RDONLY;
551 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
552 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
553 sizeof (rl_t), offsetof(rl_t, r_node));
554 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
555 offsetof(zvol_extent_t, ze_node));
556 /* get and cache the blocksize */
557 error = dmu_object_info(os, ZVOL_OBJ, &doi);
558 ASSERT(error == 0);
559 zv->zv_volblocksize = doi.doi_data_block_size;
560
561 if (spa_writeable(dmu_objset_spa(os))) {
562 if (zil_replay_disable)
563 zil_destroy(dmu_objset_zil(os), B_FALSE);
564 else
565 zil_replay(os, zv, zvol_replay_vector);
566 }
567 dmu_objset_disown(os, FTAG);
568 zv->zv_objset = NULL;
569
570 zvol_minors++;
571
572 mutex_exit(&zfsdev_state_lock);
573
574 return (0);
575 }
576
577 /*
578 * Remove minor node for the specified volume.
579 */
580 static int
581 zvol_remove_zv(zvol_state_t *zv)
582 {
583 char nmbuf[20];
584 minor_t minor = zv->zv_minor;
585
586 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
587 if (zv->zv_total_opens != 0)
588 return (SET_ERROR(EBUSY));
589
590 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
591 ddi_remove_minor_node(zfs_dip, nmbuf);
592
593 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
594 ddi_remove_minor_node(zfs_dip, nmbuf);
595
596 avl_destroy(&zv->zv_znode.z_range_avl);
597 mutex_destroy(&zv->zv_znode.z_range_lock);
598
599 kmem_free(zv, sizeof (zvol_state_t));
600
601 ddi_soft_state_free(zfsdev_state, minor);
602
603 zvol_minors--;
604 return (0);
605 }
606
607 int
608 zvol_remove_minor(const char *name)
609 {
610 zvol_state_t *zv;
611 int rc;
612
613 mutex_enter(&zfsdev_state_lock);
614 if ((zv = zvol_minor_lookup(name)) == NULL) {
615 mutex_exit(&zfsdev_state_lock);
616 return (SET_ERROR(ENXIO));
617 }
618 rc = zvol_remove_zv(zv);
619 mutex_exit(&zfsdev_state_lock);
620 return (rc);
621 }
622
623 int
624 zvol_first_open(zvol_state_t *zv)
625 {
626 objset_t *os;
627 uint64_t volsize;
628 int error;
629 uint64_t readonly;
630
631 /* lie and say we're read-only */
632 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
633 zvol_tag, &os);
634 if (error)
635 return (error);
636
637 zv->zv_objset = os;
638 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
639 if (error) {
640 ASSERT(error == 0);
641 dmu_objset_disown(os, zvol_tag);
642 return (error);
643 }
644
645 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
646 if (error) {
647 dmu_objset_disown(os, zvol_tag);
648 return (error);
649 }
650
651 zvol_size_changed(zv, volsize);
652 zv->zv_zilog = zil_open(os, zvol_get_data);
653
654 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
655 NULL) == 0);
656 if (readonly || dmu_objset_is_snapshot(os) ||
657 !spa_writeable(dmu_objset_spa(os)))
658 zv->zv_flags |= ZVOL_RDONLY;
659 else
660 zv->zv_flags &= ~ZVOL_RDONLY;
661 return (error);
662 }
663
664 void
665 zvol_last_close(zvol_state_t *zv)
666 {
667 zil_close(zv->zv_zilog);
668 zv->zv_zilog = NULL;
669
670 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
671 zv->zv_dbuf = NULL;
672
673 /*
674 * Evict cached data
675 */
676 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
677 !(zv->zv_flags & ZVOL_RDONLY))
678 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
679 dmu_objset_evict_dbufs(zv->zv_objset);
680
681 dmu_objset_disown(zv->zv_objset, zvol_tag);
682 zv->zv_objset = NULL;
683 }
684
685 int
686 zvol_prealloc(zvol_state_t *zv)
687 {
688 objset_t *os = zv->zv_objset;
689 dmu_tx_t *tx;
690 uint64_t refd, avail, usedobjs, availobjs;
691 uint64_t resid = zv->zv_volsize;
692 uint64_t off = 0;
693
694 /* Check the space usage before attempting to allocate the space */
695 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
696 if (avail < zv->zv_volsize)
697 return (SET_ERROR(ENOSPC));
698
699 /* Free old extents if they exist */
700 zvol_free_extents(zv);
701
702 while (resid != 0) {
703 int error;
704 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
705
706 tx = dmu_tx_create(os);
707 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
708 error = dmu_tx_assign(tx, TXG_WAIT);
709 if (error) {
710 dmu_tx_abort(tx);
711 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
712 return (error);
713 }
714 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
715 dmu_tx_commit(tx);
716 off += bytes;
717 resid -= bytes;
718 }
719 txg_wait_synced(dmu_objset_pool(os), 0);
720
721 return (0);
722 }
723
724 static int
725 zvol_update_volsize(objset_t *os, uint64_t volsize)
726 {
727 dmu_tx_t *tx;
728 int error;
729
730 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
731
732 tx = dmu_tx_create(os);
733 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
734 dmu_tx_mark_netfree(tx);
735 error = dmu_tx_assign(tx, TXG_WAIT);
736 if (error) {
737 dmu_tx_abort(tx);
738 return (error);
739 }
740
741 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
742 &volsize, tx);
743 dmu_tx_commit(tx);
744
745 if (error == 0)
746 error = dmu_free_long_range(os,
747 ZVOL_OBJ, volsize, DMU_OBJECT_END);
748 return (error);
749 }
750
751 void
752 zvol_remove_minors(const char *name)
753 {
754 zvol_state_t *zv;
755 char *namebuf;
756 minor_t minor;
757
758 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
759 (void) strncpy(namebuf, name, strlen(name));
760 (void) strcat(namebuf, "/");
761 mutex_enter(&zfsdev_state_lock);
762 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
763
764 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
765 if (zv == NULL)
766 continue;
767 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
768 (void) zvol_remove_zv(zv);
769 }
770 kmem_free(namebuf, strlen(name) + 2);
771
772 mutex_exit(&zfsdev_state_lock);
773 }
774
775 static int
776 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
777 {
778 uint64_t old_volsize = 0ULL;
779 int error = 0;
780
781 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
782
783 /*
784 * Reinitialize the dump area to the new size. If we
785 * failed to resize the dump area then restore it back to
786 * its original size. We must set the new volsize prior
787 * to calling dumpvp_resize() to ensure that the devices'
788 * size(9P) is not visible by the dump subsystem.
789 */
790 old_volsize = zv->zv_volsize;
791 zvol_size_changed(zv, volsize);
792
793 if (zv->zv_flags & ZVOL_DUMPIFIED) {
794 if ((error = zvol_dumpify(zv)) != 0 ||
795 (error = dumpvp_resize()) != 0) {
796 int dumpify_error;
797
798 (void) zvol_update_volsize(zv->zv_objset, old_volsize);
799 zvol_size_changed(zv, old_volsize);
800 dumpify_error = zvol_dumpify(zv);
801 error = dumpify_error ? dumpify_error : error;
802 }
803 }
804
805 /*
806 * Generate a LUN expansion event.
807 */
808 if (error == 0) {
809 sysevent_id_t eid;
810 nvlist_t *attr;
811 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
812
813 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
814 zv->zv_minor);
815
816 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
817 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
818
819 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
820 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
821
822 nvlist_free(attr);
823 kmem_free(physpath, MAXPATHLEN);
824 }
825 return (error);
826 }
827
828 int
829 zvol_set_volsize(const char *name, uint64_t volsize)
830 {
831 zvol_state_t *zv = NULL;
832 objset_t *os;
833 int error;
834 dmu_object_info_t doi;
835 uint64_t readonly;
836 boolean_t owned = B_FALSE;
837
838 error = dsl_prop_get_integer(name,
839 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
840 if (error != 0)
841 return (error);
842 if (readonly)
843 return (SET_ERROR(EROFS));
844
845 mutex_enter(&zfsdev_state_lock);
846 zv = zvol_minor_lookup(name);
847
848 if (zv == NULL || zv->zv_objset == NULL) {
849 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
850 FTAG, &os)) != 0) {
851 mutex_exit(&zfsdev_state_lock);
852 return (error);
853 }
854 owned = B_TRUE;
855 if (zv != NULL)
856 zv->zv_objset = os;
857 } else {
858 os = zv->zv_objset;
859 }
860
861 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
862 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
863 goto out;
864
865 error = zvol_update_volsize(os, volsize);
866
867 if (error == 0 && zv != NULL)
868 error = zvol_update_live_volsize(zv, volsize);
869 out:
870 if (owned) {
871 dmu_objset_disown(os, FTAG);
872 if (zv != NULL)
873 zv->zv_objset = NULL;
874 }
875 mutex_exit(&zfsdev_state_lock);
876 return (error);
877 }
878
879 /*ARGSUSED*/
880 int
881 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
882 {
883 zvol_state_t *zv;
884 int err = 0;
885
886 mutex_enter(&zfsdev_state_lock);
887
888 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
889 if (zv == NULL) {
890 mutex_exit(&zfsdev_state_lock);
891 return (SET_ERROR(ENXIO));
892 }
893
894 if (zv->zv_total_opens == 0)
895 err = zvol_first_open(zv);
896 if (err) {
897 mutex_exit(&zfsdev_state_lock);
898 return (err);
899 }
900 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
901 err = SET_ERROR(EROFS);
902 goto out;
903 }
904 if (zv->zv_flags & ZVOL_EXCL) {
905 err = SET_ERROR(EBUSY);
906 goto out;
907 }
908 if (flag & FEXCL) {
909 if (zv->zv_total_opens != 0) {
910 err = SET_ERROR(EBUSY);
911 goto out;
912 }
913 zv->zv_flags |= ZVOL_EXCL;
914 }
915
916 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
917 zv->zv_open_count[otyp]++;
918 zv->zv_total_opens++;
919 }
920 mutex_exit(&zfsdev_state_lock);
921
922 return (err);
923 out:
924 if (zv->zv_total_opens == 0)
925 zvol_last_close(zv);
926 mutex_exit(&zfsdev_state_lock);
927 return (err);
928 }
929
930 /*ARGSUSED*/
931 int
932 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
933 {
934 minor_t minor = getminor(dev);
935 zvol_state_t *zv;
936 int error = 0;
937
938 mutex_enter(&zfsdev_state_lock);
939
940 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
941 if (zv == NULL) {
942 mutex_exit(&zfsdev_state_lock);
943 return (SET_ERROR(ENXIO));
944 }
945
946 if (zv->zv_flags & ZVOL_EXCL) {
947 ASSERT(zv->zv_total_opens == 1);
948 zv->zv_flags &= ~ZVOL_EXCL;
949 }
950
951 /*
952 * If the open count is zero, this is a spurious close.
953 * That indicates a bug in the kernel / DDI framework.
954 */
955 ASSERT(zv->zv_open_count[otyp] != 0);
956 ASSERT(zv->zv_total_opens != 0);
957
958 /*
959 * You may get multiple opens, but only one close.
960 */
961 zv->zv_open_count[otyp]--;
962 zv->zv_total_opens--;
963
964 if (zv->zv_total_opens == 0)
965 zvol_last_close(zv);
966
967 mutex_exit(&zfsdev_state_lock);
968 return (error);
969 }
970
971 static void
972 zvol_get_done(zgd_t *zgd, int error)
973 {
974 if (zgd->zgd_db)
975 dmu_buf_rele(zgd->zgd_db, zgd);
976
977 zfs_range_unlock(zgd->zgd_rl);
978
979 if (error == 0 && zgd->zgd_bp)
980 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
981
982 kmem_free(zgd, sizeof (zgd_t));
983 }
984
985 /*
986 * Get data to generate a TX_WRITE intent log record.
987 */
988 static int
989 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
990 {
991 zvol_state_t *zv = arg;
992 objset_t *os = zv->zv_objset;
993 uint64_t object = ZVOL_OBJ;
994 uint64_t offset = lr->lr_offset;
995 uint64_t size = lr->lr_length; /* length of user data */
996 dmu_buf_t *db;
997 zgd_t *zgd;
998 int error;
999
1000 ASSERT3P(lwb, !=, NULL);
1001 ASSERT3P(zio, !=, NULL);
1002 ASSERT3U(size, !=, 0);
1003
1004 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1005 zgd->zgd_lwb = lwb;
1006 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
1007
1008 /*
1009 * Write records come in two flavors: immediate and indirect.
1010 * For small writes it's cheaper to store the data with the
1011 * log record (immediate); for large writes it's cheaper to
1012 * sync the data and get a pointer to it (indirect) so that
1013 * we don't have to write the data twice.
1014 */
1015 if (buf != NULL) { /* immediate write */
1016 error = dmu_read(os, object, offset, size, buf,
1017 DMU_READ_NO_PREFETCH);
1018 } else {
1019 size = zv->zv_volblocksize;
1020 offset = P2ALIGN(offset, size);
1021 error = dmu_buf_hold(os, object, offset, zgd, &db,
1022 DMU_READ_NO_PREFETCH);
1023 if (error == 0) {
1024 blkptr_t *bp = &lr->lr_blkptr;
1025
1026 zgd->zgd_db = db;
1027 zgd->zgd_bp = bp;
1028
1029 ASSERT(db->db_offset == offset);
1030 ASSERT(db->db_size == size);
1031
1032 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1033 zvol_get_done, zgd);
1034
1035 if (error == 0)
1036 return (0);
1037 }
1038 }
1039
1040 zvol_get_done(zgd, error);
1041
1042 return (error);
1043 }
1044
1045 /*
1046 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1047 *
1048 * We store data in the log buffers if it's small enough.
1049 * Otherwise we will later flush the data out via dmu_sync().
1050 */
1051 ssize_t zvol_immediate_write_sz = 32768;
1052
1053 static void
1054 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1055 boolean_t sync)
1056 {
1057 uint32_t blocksize = zv->zv_volblocksize;
1058 zilog_t *zilog = zv->zv_zilog;
1059 itx_wr_state_t write_state;
1060
1061 if (zil_replaying(zilog, tx))
1062 return;
1063
1064 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1065 write_state = WR_INDIRECT;
1066 else if (!spa_has_slogs(zilog->zl_spa) &&
1067 resid >= blocksize && blocksize > zvol_immediate_write_sz)
1068 write_state = WR_INDIRECT;
1069 else if (sync)
1070 write_state = WR_COPIED;
1071 else
1072 write_state = WR_NEED_COPY;
1073
1074 while (resid) {
1075 itx_t *itx;
1076 lr_write_t *lr;
1077 itx_wr_state_t wr_state = write_state;
1078 ssize_t len = resid;
1079
1080 if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA)
1081 wr_state = WR_NEED_COPY;
1082 else if (wr_state == WR_INDIRECT)
1083 len = MIN(blocksize - P2PHASE(off, blocksize), resid);
1084
1085 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1086 (wr_state == WR_COPIED ? len : 0));
1087 lr = (lr_write_t *)&itx->itx_lr;
1088 if (wr_state == WR_COPIED && dmu_read(zv->zv_objset,
1089 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1090 zil_itx_destroy(itx);
1091 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1092 lr = (lr_write_t *)&itx->itx_lr;
1093 wr_state = WR_NEED_COPY;
1094 }
1095
1096 itx->itx_wr_state = wr_state;
1097 lr->lr_foid = ZVOL_OBJ;
1098 lr->lr_offset = off;
1099 lr->lr_length = len;
1100 lr->lr_blkoff = 0;
1101 BP_ZERO(&lr->lr_blkptr);
1102
1103 itx->itx_private = zv;
1104 itx->itx_sync = sync;
1105
1106 zil_itx_assign(zilog, itx, tx);
1107
1108 off += len;
1109 resid -= len;
1110 }
1111 }
1112
1113 static int
1114 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1115 uint64_t size, boolean_t doread, boolean_t isdump)
1116 {
1117 vdev_disk_t *dvd;
1118 int c;
1119 int numerrors = 0;
1120
1121 if (vd->vdev_ops == &vdev_mirror_ops ||
1122 vd->vdev_ops == &vdev_replacing_ops ||
1123 vd->vdev_ops == &vdev_spare_ops) {
1124 for (c = 0; c < vd->vdev_children; c++) {
1125 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1126 addr, offset, origoffset, size, doread, isdump);
1127 if (err != 0) {
1128 numerrors++;
1129 } else if (doread) {
1130 break;
1131 }
1132 }
1133 }
1134
1135 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1136 return (numerrors < vd->vdev_children ? 0 : EIO);
1137
1138 if (doread && !vdev_readable(vd))
1139 return (SET_ERROR(EIO));
1140 else if (!doread && !vdev_writeable(vd))
1141 return (SET_ERROR(EIO));
1142
1143 if (vd->vdev_ops == &vdev_raidz_ops) {
1144 return (vdev_raidz_physio(vd,
1145 addr, size, offset, origoffset, doread, isdump));
1146 }
1147
1148 offset += VDEV_LABEL_START_SIZE;
1149
1150 if (ddi_in_panic() || isdump) {
1151 ASSERT(!doread);
1152 if (doread)
1153 return (SET_ERROR(EIO));
1154 dvd = vd->vdev_tsd;
1155 ASSERT3P(dvd, !=, NULL);
1156 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1157 lbtodb(size)));
1158 } else {
1159 dvd = vd->vdev_tsd;
1160 ASSERT3P(dvd, !=, NULL);
1161 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1162 offset, doread ? B_READ : B_WRITE));
1163 }
1164 }
1165
1166 static int
1167 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1168 boolean_t doread, boolean_t isdump)
1169 {
1170 vdev_t *vd;
1171 int error;
1172 zvol_extent_t *ze;
1173 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1174
1175 /* Must be sector aligned, and not stradle a block boundary. */
1176 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1177 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1178 return (SET_ERROR(EINVAL));
1179 }
1180 ASSERT(size <= zv->zv_volblocksize);
1181
1182 /* Locate the extent this belongs to */
1183 ze = list_head(&zv->zv_extents);
1184 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1185 offset -= ze->ze_nblks * zv->zv_volblocksize;
1186 ze = list_next(&zv->zv_extents, ze);
1187 }
1188
1189 if (ze == NULL)
1190 return (SET_ERROR(EINVAL));
1191
1192 if (!ddi_in_panic())
1193 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1194
1195 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1196 offset += DVA_GET_OFFSET(&ze->ze_dva);
1197 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1198 size, doread, isdump);
1199
1200 if (!ddi_in_panic())
1201 spa_config_exit(spa, SCL_STATE, FTAG);
1202
1203 return (error);
1204 }
1205
1206 int
1207 zvol_strategy(buf_t *bp)
1208 {
1209 zfs_soft_state_t *zs = NULL;
1210 zvol_state_t *zv;
1211 uint64_t off, volsize;
1212 size_t resid;
1213 char *addr;
1214 objset_t *os;
1215 rl_t *rl;
1216 int error = 0;
1217 boolean_t doread = bp->b_flags & B_READ;
1218 boolean_t is_dumpified;
1219 boolean_t sync;
1220
1221 if (getminor(bp->b_edev) == 0) {
1222 error = SET_ERROR(EINVAL);
1223 } else {
1224 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1225 if (zs == NULL)
1226 error = SET_ERROR(ENXIO);
1227 else if (zs->zss_type != ZSST_ZVOL)
1228 error = SET_ERROR(EINVAL);
1229 }
1230
1231 if (error) {
1232 bioerror(bp, error);
1233 biodone(bp);
1234 return (0);
1235 }
1236
1237 zv = zs->zss_data;
1238
1239 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1240 bioerror(bp, EROFS);
1241 biodone(bp);
1242 return (0);
1243 }
1244
1245 off = ldbtob(bp->b_blkno);
1246 volsize = zv->zv_volsize;
1247
1248 os = zv->zv_objset;
1249 ASSERT(os != NULL);
1250
1251 bp_mapin(bp);
1252 addr = bp->b_un.b_addr;
1253 resid = bp->b_bcount;
1254
1255 if (resid > 0 && (off < 0 || off >= volsize)) {
1256 bioerror(bp, EIO);
1257 biodone(bp);
1258 return (0);
1259 }
1260
1261 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1262 sync = ((!(bp->b_flags & B_ASYNC) &&
1263 !(zv->zv_flags & ZVOL_WCE)) ||
1264 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1265 !doread && !is_dumpified;
1266
1267 /*
1268 * There must be no buffer changes when doing a dmu_sync() because
1269 * we can't change the data whilst calculating the checksum.
1270 */
1271 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1272 doread ? RL_READER : RL_WRITER);
1273
1274 while (resid != 0 && off < volsize) {
1275 size_t size = MIN(resid, zvol_maxphys);
1276 if (is_dumpified) {
1277 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1278 error = zvol_dumpio(zv, addr, off, size,
1279 doread, B_FALSE);
1280 } else if (doread) {
1281 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1282 DMU_READ_PREFETCH);
1283 } else {
1284 dmu_tx_t *tx = dmu_tx_create(os);
1285 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1286 error = dmu_tx_assign(tx, TXG_WAIT);
1287 if (error) {
1288 dmu_tx_abort(tx);
1289 } else {
1290 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1291 zvol_log_write(zv, tx, off, size, sync);
1292 dmu_tx_commit(tx);
1293 }
1294 }
1295 if (error) {
1296 /* convert checksum errors into IO errors */
1297 if (error == ECKSUM)
1298 error = SET_ERROR(EIO);
1299 break;
1300 }
1301 off += size;
1302 addr += size;
1303 resid -= size;
1304 }
1305 zfs_range_unlock(rl);
1306
1307 if ((bp->b_resid = resid) == bp->b_bcount)
1308 bioerror(bp, off > volsize ? EINVAL : error);
1309
1310 if (sync)
1311 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1312 biodone(bp);
1313
1314 return (0);
1315 }
1316
1317 /*
1318 * Set the buffer count to the zvol maximum transfer.
1319 * Using our own routine instead of the default minphys()
1320 * means that for larger writes we write bigger buffers on X86
1321 * (128K instead of 56K) and flush the disk write cache less often
1322 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1323 * 56K on X86 and 128K on sparc).
1324 */
1325 void
1326 zvol_minphys(struct buf *bp)
1327 {
1328 if (bp->b_bcount > zvol_maxphys)
1329 bp->b_bcount = zvol_maxphys;
1330 }
1331
1332 int
1333 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1334 {
1335 minor_t minor = getminor(dev);
1336 zvol_state_t *zv;
1337 int error = 0;
1338 uint64_t size;
1339 uint64_t boff;
1340 uint64_t resid;
1341
1342 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1343 if (zv == NULL)
1344 return (SET_ERROR(ENXIO));
1345
1346 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1347 return (SET_ERROR(EINVAL));
1348
1349 boff = ldbtob(blkno);
1350 resid = ldbtob(nblocks);
1351
1352 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1353
1354 while (resid) {
1355 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1356 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1357 if (error)
1358 break;
1359 boff += size;
1360 addr += size;
1361 resid -= size;
1362 }
1363
1364 return (error);
1365 }
1366
1367 /*ARGSUSED*/
1368 int
1369 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1370 {
1371 minor_t minor = getminor(dev);
1372 zvol_state_t *zv;
1373 uint64_t volsize;
1374 rl_t *rl;
1375 int error = 0;
1376
1377 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1378 if (zv == NULL)
1379 return (SET_ERROR(ENXIO));
1380
1381 volsize = zv->zv_volsize;
1382 if (uio->uio_resid > 0 &&
1383 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1384 return (SET_ERROR(EIO));
1385
1386 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1387 error = physio(zvol_strategy, NULL, dev, B_READ,
1388 zvol_minphys, uio);
1389 return (error);
1390 }
1391
1392 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1393 RL_READER);
1394 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1395 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1396
1397 /* don't read past the end */
1398 if (bytes > volsize - uio->uio_loffset)
1399 bytes = volsize - uio->uio_loffset;
1400
1401 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1402 if (error) {
1403 /* convert checksum errors into IO errors */
1404 if (error == ECKSUM)
1405 error = SET_ERROR(EIO);
1406 break;
1407 }
1408 }
1409 zfs_range_unlock(rl);
1410 return (error);
1411 }
1412
1413 /*ARGSUSED*/
1414 int
1415 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1416 {
1417 minor_t minor = getminor(dev);
1418 zvol_state_t *zv;
1419 uint64_t volsize;
1420 rl_t *rl;
1421 int error = 0;
1422 boolean_t sync;
1423
1424 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1425 if (zv == NULL)
1426 return (SET_ERROR(ENXIO));
1427
1428 volsize = zv->zv_volsize;
1429 if (uio->uio_resid > 0 &&
1430 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1431 return (SET_ERROR(EIO));
1432
1433 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1434 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1435 zvol_minphys, uio);
1436 return (error);
1437 }
1438
1439 sync = !(zv->zv_flags & ZVOL_WCE) ||
1440 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1441
1442 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1443 RL_WRITER);
1444 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1445 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1446 uint64_t off = uio->uio_loffset;
1447 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1448
1449 if (bytes > volsize - off) /* don't write past the end */
1450 bytes = volsize - off;
1451
1452 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1453 error = dmu_tx_assign(tx, TXG_WAIT);
1454 if (error) {
1455 dmu_tx_abort(tx);
1456 break;
1457 }
1458 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1459 if (error == 0)
1460 zvol_log_write(zv, tx, off, bytes, sync);
1461 dmu_tx_commit(tx);
1462
1463 if (error)
1464 break;
1465 }
1466 zfs_range_unlock(rl);
1467 if (sync)
1468 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1469 return (error);
1470 }
1471
1472 int
1473 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1474 {
1475 struct uuid uuid = EFI_RESERVED;
1476 efi_gpe_t gpe = { 0 };
1477 uint32_t crc;
1478 dk_efi_t efi;
1479 int length;
1480 char *ptr;
1481
1482 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1483 return (SET_ERROR(EFAULT));
1484 ptr = (char *)(uintptr_t)efi.dki_data_64;
1485 length = efi.dki_length;
1486 /*
1487 * Some clients may attempt to request a PMBR for the
1488 * zvol. Currently this interface will return EINVAL to
1489 * such requests. These requests could be supported by
1490 * adding a check for lba == 0 and consing up an appropriate
1491 * PMBR.
1492 */
1493 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1494 return (SET_ERROR(EINVAL));
1495
1496 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1497 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1498 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1499
1500 if (efi.dki_lba == 1) {
1501 efi_gpt_t gpt = { 0 };
1502
1503 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1504 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1505 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1506 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1507 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1508 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1509 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1510 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1511 gpt.efi_gpt_SizeOfPartitionEntry =
1512 LE_32(sizeof (efi_gpe_t));
1513 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1514 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1515 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1516 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1517 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1518 flag))
1519 return (SET_ERROR(EFAULT));
1520 ptr += sizeof (gpt);
1521 length -= sizeof (gpt);
1522 }
1523 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1524 length), flag))
1525 return (SET_ERROR(EFAULT));
1526 return (0);
1527 }
1528
1529 /*
1530 * BEGIN entry points to allow external callers access to the volume.
1531 */
1532 /*
1533 * Return the volume parameters needed for access from an external caller.
1534 * These values are invariant as long as the volume is held open.
1535 */
1536 int
1537 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1538 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1539 void **rl_hdl, void **bonus_hdl)
1540 {
1541 zvol_state_t *zv;
1542
1543 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1544 if (zv == NULL)
1545 return (SET_ERROR(ENXIO));
1546 if (zv->zv_flags & ZVOL_DUMPIFIED)
1547 return (SET_ERROR(ENXIO));
1548
1549 ASSERT(blksize && max_xfer_len && minor_hdl &&
1550 objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1551
1552 *blksize = zv->zv_volblocksize;
1553 *max_xfer_len = (uint64_t)zvol_maxphys;
1554 *minor_hdl = zv;
1555 *objset_hdl = zv->zv_objset;
1556 *zil_hdl = zv->zv_zilog;
1557 *rl_hdl = &zv->zv_znode;
1558 *bonus_hdl = zv->zv_dbuf;
1559 return (0);
1560 }
1561
1562 /*
1563 * Return the current volume size to an external caller.
1564 * The size can change while the volume is open.
1565 */
1566 uint64_t
1567 zvol_get_volume_size(void *minor_hdl)
1568 {
1569 zvol_state_t *zv = minor_hdl;
1570
1571 return (zv->zv_volsize);
1572 }
1573
1574 /*
1575 * Return the current WCE setting to an external caller.
1576 * The WCE setting can change while the volume is open.
1577 */
1578 int
1579 zvol_get_volume_wce(void *minor_hdl)
1580 {
1581 zvol_state_t *zv = minor_hdl;
1582
1583 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1584 }
1585
1586 /*
1587 * Entry point for external callers to zvol_log_write
1588 */
1589 void
1590 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1591 boolean_t sync)
1592 {
1593 zvol_state_t *zv = minor_hdl;
1594
1595 zvol_log_write(zv, tx, off, resid, sync);
1596 }
1597 /*
1598 * END entry points to allow external callers access to the volume.
1599 */
1600
1601 /*
1602 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1603 */
1604 static void
1605 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1606 boolean_t sync)
1607 {
1608 itx_t *itx;
1609 lr_truncate_t *lr;
1610 zilog_t *zilog = zv->zv_zilog;
1611
1612 if (zil_replaying(zilog, tx))
1613 return;
1614
1615 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1616 lr = (lr_truncate_t *)&itx->itx_lr;
1617 lr->lr_foid = ZVOL_OBJ;
1618 lr->lr_offset = off;
1619 lr->lr_length = len;
1620
1621 itx->itx_sync = sync;
1622 zil_itx_assign(zilog, itx, tx);
1623 }
1624
1625 /*
1626 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
1627 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1628 */
1629 /*ARGSUSED*/
1630 int
1631 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1632 {
1633 zvol_state_t *zv;
1634 struct dk_callback *dkc;
1635 int error = 0;
1636 rl_t *rl;
1637
1638 mutex_enter(&zfsdev_state_lock);
1639
1640 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1641
1642 if (zv == NULL) {
1643 mutex_exit(&zfsdev_state_lock);
1644 return (SET_ERROR(ENXIO));
1645 }
1646 ASSERT(zv->zv_total_opens > 0);
1647
1648 switch (cmd) {
1649
1650 case DKIOCINFO:
1651 {
1652 struct dk_cinfo dki;
1653
1654 bzero(&dki, sizeof (dki));
1655 (void) strcpy(dki.dki_cname, "zvol");
1656 (void) strcpy(dki.dki_dname, "zvol");
1657 dki.dki_ctype = DKC_UNKNOWN;
1658 dki.dki_unit = getminor(dev);
1659 dki.dki_maxtransfer =
1660 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
1661 mutex_exit(&zfsdev_state_lock);
1662 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1663 error = SET_ERROR(EFAULT);
1664 return (error);
1665 }
1666
1667 case DKIOCGMEDIAINFO:
1668 {
1669 struct dk_minfo dkm;
1670
1671 bzero(&dkm, sizeof (dkm));
1672 dkm.dki_lbsize = 1U << zv->zv_min_bs;
1673 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1674 dkm.dki_media_type = DK_UNKNOWN;
1675 mutex_exit(&zfsdev_state_lock);
1676 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1677 error = SET_ERROR(EFAULT);
1678 return (error);
1679 }
1680
1681 case DKIOCGMEDIAINFOEXT:
1682 {
1683 struct dk_minfo_ext dkmext;
1684
1685 bzero(&dkmext, sizeof (dkmext));
1686 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
1687 dkmext.dki_pbsize = zv->zv_volblocksize;
1688 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1689 dkmext.dki_media_type = DK_UNKNOWN;
1690 mutex_exit(&zfsdev_state_lock);
1691 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
1692 error = SET_ERROR(EFAULT);
1693 return (error);
1694 }
1695
1696 case DKIOCGETEFI:
1697 {
1698 uint64_t vs = zv->zv_volsize;
1699 uint8_t bs = zv->zv_min_bs;
1700
1701 mutex_exit(&zfsdev_state_lock);
1702 error = zvol_getefi((void *)arg, flag, vs, bs);
1703 return (error);
1704 }
1705
1706 case DKIOCFLUSHWRITECACHE:
1707 dkc = (struct dk_callback *)arg;
1708 mutex_exit(&zfsdev_state_lock);
1709 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1710 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1711 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
1712 error = 0;
1713 }
1714 return (error);
1715
1716 case DKIOCGETWCE:
1717 {
1718 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1719 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1720 flag))
1721 error = SET_ERROR(EFAULT);
1722 break;
1723 }
1724 case DKIOCSETWCE:
1725 {
1726 int wce;
1727 if (ddi_copyin((void *)arg, &wce, sizeof (int),
1728 flag)) {
1729 error = SET_ERROR(EFAULT);
1730 break;
1731 }
1732 if (wce) {
1733 zv->zv_flags |= ZVOL_WCE;
1734 mutex_exit(&zfsdev_state_lock);
1735 } else {
1736 zv->zv_flags &= ~ZVOL_WCE;
1737 mutex_exit(&zfsdev_state_lock);
1738 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1739 }
1740 return (0);
1741 }
1742
1743 case DKIOCGGEOM:
1744 case DKIOCGVTOC:
1745 /*
1746 * commands using these (like prtvtoc) expect ENOTSUP
1747 * since we're emulating an EFI label
1748 */
1749 error = SET_ERROR(ENOTSUP);
1750 break;
1751
1752 case DKIOCDUMPINIT:
1753 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1754 RL_WRITER);
1755 error = zvol_dumpify(zv);
1756 zfs_range_unlock(rl);
1757 break;
1758
1759 case DKIOCDUMPFINI:
1760 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1761 break;
1762 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1763 RL_WRITER);
1764 error = zvol_dump_fini(zv);
1765 zfs_range_unlock(rl);
1766 break;
1767
1768 case DKIOCFREE:
1769 {
1770 dkioc_free_t df;
1771 dmu_tx_t *tx;
1772
1773 if (!zvol_unmap_enabled)
1774 break;
1775
1776 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
1777 error = SET_ERROR(EFAULT);
1778 break;
1779 }
1780
1781 /*
1782 * Apply Postel's Law to length-checking. If they overshoot,
1783 * just blank out until the end, if there's a need to blank
1784 * out anything.
1785 */
1786 if (df.df_start >= zv->zv_volsize)
1787 break; /* No need to do anything... */
1788
1789 mutex_exit(&zfsdev_state_lock);
1790
1791 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
1792 RL_WRITER);
1793 tx = dmu_tx_create(zv->zv_objset);
1794 dmu_tx_mark_netfree(tx);
1795 error = dmu_tx_assign(tx, TXG_WAIT);
1796 if (error != 0) {
1797 dmu_tx_abort(tx);
1798 } else {
1799 zvol_log_truncate(zv, tx, df.df_start,
1800 df.df_length, B_TRUE);
1801 dmu_tx_commit(tx);
1802 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1803 df.df_start, df.df_length);
1804 }
1805
1806 zfs_range_unlock(rl);
1807
1808 /*
1809 * If the write-cache is disabled, 'sync' property
1810 * is set to 'always', or if the caller is asking for
1811 * a synchronous free, commit this operation to the zil.
1812 * This will sync any previous uncommitted writes to the
1813 * zvol object.
1814 * Can be overridden by the zvol_unmap_sync_enabled tunable.
1815 */
1816 if ((error == 0) && zvol_unmap_sync_enabled &&
1817 (!(zv->zv_flags & ZVOL_WCE) ||
1818 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) ||
1819 (df.df_flags & DF_WAIT_SYNC))) {
1820 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1821 }
1822
1823 return (error);
1824 }
1825
1826 default:
1827 error = SET_ERROR(ENOTTY);
1828 break;
1829
1830 }
1831 mutex_exit(&zfsdev_state_lock);
1832 return (error);
1833 }
1834
1835 int
1836 zvol_busy(void)
1837 {
1838 return (zvol_minors != 0);
1839 }
1840
1841 void
1842 zvol_init(void)
1843 {
1844 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1845 1) == 0);
1846 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
1847 }
1848
1849 void
1850 zvol_fini(void)
1851 {
1852 mutex_destroy(&zfsdev_state_lock);
1853 ddi_soft_state_fini(&zfsdev_state);
1854 }
1855
1856 /*ARGSUSED*/
1857 static int
1858 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
1859 {
1860 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1861
1862 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1863 return (1);
1864 return (0);
1865 }
1866
1867 /*ARGSUSED*/
1868 static void
1869 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
1870 {
1871 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1872
1873 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
1874 }
1875
1876 static int
1877 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1878 {
1879 dmu_tx_t *tx;
1880 int error;
1881 objset_t *os = zv->zv_objset;
1882 spa_t *spa = dmu_objset_spa(os);
1883 vdev_t *vd = spa->spa_root_vdev;
1884 nvlist_t *nv = NULL;
1885 uint64_t version = spa_version(spa);
1886 uint64_t checksum, compress, refresrv, vbs, dedup;
1887
1888 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1889 ASSERT(vd->vdev_ops == &vdev_root_ops);
1890
1891 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1892 DMU_OBJECT_END);
1893 if (error != 0)
1894 return (error);
1895 /* wait for dmu_free_long_range to actually free the blocks */
1896 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1897
1898 /*
1899 * If the pool on which the dump device is being initialized has more
1900 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
1901 * enabled. If so, bump that feature's counter to indicate that the
1902 * feature is active. We also check the vdev type to handle the
1903 * following case:
1904 * # zpool create test raidz disk1 disk2 disk3
1905 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
1906 * the raidz vdev itself has 3 children.
1907 */
1908 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
1909 if (!spa_feature_is_enabled(spa,
1910 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1911 return (SET_ERROR(ENOTSUP));
1912 (void) dsl_sync_task(spa_name(spa),
1913 zfs_mvdev_dump_feature_check,
1914 zfs_mvdev_dump_activate_feature_sync, NULL,
1915 2, ZFS_SPACE_CHECK_RESERVED);
1916 }
1917
1918 if (!resize) {
1919 error = dsl_prop_get_integer(zv->zv_name,
1920 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1921 if (error == 0) {
1922 error = dsl_prop_get_integer(zv->zv_name,
1923 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
1924 NULL);
1925 }
1926 if (error == 0) {
1927 error = dsl_prop_get_integer(zv->zv_name,
1928 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
1929 &refresrv, NULL);
1930 }
1931 if (error == 0) {
1932 error = dsl_prop_get_integer(zv->zv_name,
1933 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
1934 NULL);
1935 }
1936 if (version >= SPA_VERSION_DEDUP && error == 0) {
1937 error = dsl_prop_get_integer(zv->zv_name,
1938 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1939 }
1940 }
1941 if (error != 0)
1942 return (error);
1943
1944 tx = dmu_tx_create(os);
1945 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1946 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1947 error = dmu_tx_assign(tx, TXG_WAIT);
1948 if (error != 0) {
1949 dmu_tx_abort(tx);
1950 return (error);
1951 }
1952
1953 /*
1954 * If we are resizing the dump device then we only need to
1955 * update the refreservation to match the newly updated
1956 * zvolsize. Otherwise, we save off the original state of the
1957 * zvol so that we can restore them if the zvol is ever undumpified.
1958 */
1959 if (resize) {
1960 error = zap_update(os, ZVOL_ZAP_OBJ,
1961 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1962 &zv->zv_volsize, tx);
1963 } else {
1964 error = zap_update(os, ZVOL_ZAP_OBJ,
1965 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1966 &compress, tx);
1967 if (error == 0) {
1968 error = zap_update(os, ZVOL_ZAP_OBJ,
1969 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
1970 &checksum, tx);
1971 }
1972 if (error == 0) {
1973 error = zap_update(os, ZVOL_ZAP_OBJ,
1974 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1975 &refresrv, tx);
1976 }
1977 if (error == 0) {
1978 error = zap_update(os, ZVOL_ZAP_OBJ,
1979 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1980 &vbs, tx);
1981 }
1982 if (error == 0) {
1983 error = dmu_object_set_blocksize(
1984 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
1985 }
1986 if (version >= SPA_VERSION_DEDUP && error == 0) {
1987 error = zap_update(os, ZVOL_ZAP_OBJ,
1988 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
1989 &dedup, tx);
1990 }
1991 if (error == 0)
1992 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
1993 }
1994 dmu_tx_commit(tx);
1995
1996 /*
1997 * We only need update the zvol's property if we are initializing
1998 * the dump area for the first time.
1999 */
2000 if (error == 0 && !resize) {
2001 /*
2002 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2003 * function. Otherwise, use the old default -- OFF.
2004 */
2005 checksum = spa_feature_is_active(spa,
2006 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2007 ZIO_CHECKSUM_OFF;
2008
2009 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2010 VERIFY(nvlist_add_uint64(nv,
2011 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2012 VERIFY(nvlist_add_uint64(nv,
2013 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2014 ZIO_COMPRESS_OFF) == 0);
2015 VERIFY(nvlist_add_uint64(nv,
2016 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2017 checksum) == 0);
2018 if (version >= SPA_VERSION_DEDUP) {
2019 VERIFY(nvlist_add_uint64(nv,
2020 zfs_prop_to_name(ZFS_PROP_DEDUP),
2021 ZIO_CHECKSUM_OFF) == 0);
2022 }
2023
2024 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2025 nv, NULL);
2026 nvlist_free(nv);
2027 }
2028
2029 /* Allocate the space for the dump */
2030 if (error == 0)
2031 error = zvol_prealloc(zv);
2032 return (error);
2033 }
2034
2035 static int
2036 zvol_dumpify(zvol_state_t *zv)
2037 {
2038 int error = 0;
2039 uint64_t dumpsize = 0;
2040 dmu_tx_t *tx;
2041 objset_t *os = zv->zv_objset;
2042
2043 if (zv->zv_flags & ZVOL_RDONLY)
2044 return (SET_ERROR(EROFS));
2045
2046 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2047 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2048 boolean_t resize = (dumpsize > 0);
2049
2050 if ((error = zvol_dump_init(zv, resize)) != 0) {
2051 (void) zvol_dump_fini(zv);
2052 return (error);
2053 }
2054 }
2055
2056 /*
2057 * Build up our lba mapping.
2058 */
2059 error = zvol_get_lbas(zv);
2060 if (error) {
2061 (void) zvol_dump_fini(zv);
2062 return (error);
2063 }
2064
2065 tx = dmu_tx_create(os);
2066 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2067 error = dmu_tx_assign(tx, TXG_WAIT);
2068 if (error) {
2069 dmu_tx_abort(tx);
2070 (void) zvol_dump_fini(zv);
2071 return (error);
2072 }
2073
2074 zv->zv_flags |= ZVOL_DUMPIFIED;
2075 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2076 &zv->zv_volsize, tx);
2077 dmu_tx_commit(tx);
2078
2079 if (error) {
2080 (void) zvol_dump_fini(zv);
2081 return (error);
2082 }
2083
2084 txg_wait_synced(dmu_objset_pool(os), 0);
2085 return (0);
2086 }
2087
2088 static int
2089 zvol_dump_fini(zvol_state_t *zv)
2090 {
2091 dmu_tx_t *tx;
2092 objset_t *os = zv->zv_objset;
2093 nvlist_t *nv;
2094 int error = 0;
2095 uint64_t checksum, compress, refresrv, vbs, dedup;
2096 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2097
2098 /*
2099 * Attempt to restore the zvol back to its pre-dumpified state.
2100 * This is a best-effort attempt as it's possible that not all
2101 * of these properties were initialized during the dumpify process
2102 * (i.e. error during zvol_dump_init).
2103 */
2104
2105 tx = dmu_tx_create(os);
2106 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2107 error = dmu_tx_assign(tx, TXG_WAIT);
2108 if (error) {
2109 dmu_tx_abort(tx);
2110 return (error);
2111 }
2112 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2113 dmu_tx_commit(tx);
2114
2115 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2116 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2117 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2118 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2119 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2120 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2121 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2122 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2123
2124 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2125 (void) nvlist_add_uint64(nv,
2126 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2127 (void) nvlist_add_uint64(nv,
2128 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2129 (void) nvlist_add_uint64(nv,
2130 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2131 if (version >= SPA_VERSION_DEDUP &&
2132 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2133 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2134 (void) nvlist_add_uint64(nv,
2135 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2136 }
2137 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2138 nv, NULL);
2139 nvlist_free(nv);
2140
2141 zvol_free_extents(zv);
2142 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2143 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2144 /* wait for dmu_free_long_range to actually free the blocks */
2145 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2146 tx = dmu_tx_create(os);
2147 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2148 error = dmu_tx_assign(tx, TXG_WAIT);
2149 if (error) {
2150 dmu_tx_abort(tx);
2151 return (error);
2152 }
2153 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2154 zv->zv_volblocksize = vbs;
2155 dmu_tx_commit(tx);
2156
2157 return (0);
2158 }
Unleashed OS