?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge commit '7e3488dc6cdcb0c04e1ce167a1a3bfef83b5f2e0'
[unleashed.git] / kernel / fs / zfs / zvol.c
blobb10116dd7fc56959a9268526ec07a064eca95351
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 dnode_t *zv_dn; /* dnode hold */
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(void *arg1, void *arg2, boolean_t byteswap)
378 {
379 zvol_state_t *zv = arg1;
380 lr_truncate_t *lr = arg2;
381 uint64_t offset, length;
382
383 if (byteswap)
384 byteswap_uint64_array(lr, sizeof (*lr));
385
386 offset = lr->lr_offset;
387 length = lr->lr_length;
388
389 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
390 }
391
392 /*
393 * Replay a TX_WRITE ZIL transaction that didn't get committed
394 * after a system failure
395 */
396 static int
397 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap)
398 {
399 zvol_state_t *zv = arg1;
400 lr_write_t *lr = arg2;
401 objset_t *os = zv->zv_objset;
402 char *data = (char *)(lr + 1); /* data follows lr_write_t */
403 uint64_t offset, length;
404 dmu_tx_t *tx;
405 int error;
406
407 if (byteswap)
408 byteswap_uint64_array(lr, sizeof (*lr));
409
410 offset = lr->lr_offset;
411 length = lr->lr_length;
412
413 /* If it's a dmu_sync() block, write the whole block */
414 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
415 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
416 if (length < blocksize) {
417 offset -= offset % blocksize;
418 length = blocksize;
419 }
420 }
421
422 tx = dmu_tx_create(os);
423 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
424 error = dmu_tx_assign(tx, TXG_WAIT);
425 if (error) {
426 dmu_tx_abort(tx);
427 } else {
428 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
429 dmu_tx_commit(tx);
430 }
431
432 return (error);
433 }
434
435 /* ARGSUSED */
436 static int
437 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap)
438 {
439 return (SET_ERROR(ENOTSUP));
440 }
441
442 /*
443 * Callback vectors for replaying records.
444 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
445 */
446 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
447 zvol_replay_err, /* 0 no such transaction type */
448 zvol_replay_err, /* TX_CREATE */
449 zvol_replay_err, /* TX_MKDIR */
450 zvol_replay_err, /* TX_MKXATTR */
451 zvol_replay_err, /* TX_SYMLINK */
452 zvol_replay_err, /* TX_REMOVE */
453 zvol_replay_err, /* TX_RMDIR */
454 zvol_replay_err, /* TX_LINK */
455 zvol_replay_err, /* TX_RENAME */
456 zvol_replay_write, /* TX_WRITE */
457 zvol_replay_truncate, /* TX_TRUNCATE */
458 zvol_replay_err, /* TX_SETATTR */
459 zvol_replay_err, /* TX_ACL */
460 zvol_replay_err, /* TX_CREATE_ACL */
461 zvol_replay_err, /* TX_CREATE_ATTR */
462 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
463 zvol_replay_err, /* TX_MKDIR_ACL */
464 zvol_replay_err, /* TX_MKDIR_ATTR */
465 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
466 zvol_replay_err, /* TX_WRITE2 */
467 };
468
469 int
470 zvol_name2minor(const char *name, minor_t *minor)
471 {
472 zvol_state_t *zv;
473
474 mutex_enter(&zfsdev_state_lock);
475 zv = zvol_minor_lookup(name);
476 if (minor && zv)
477 *minor = zv->zv_minor;
478 mutex_exit(&zfsdev_state_lock);
479 return (zv ? 0 : -1);
480 }
481
482 /*
483 * Create a minor node (plus a whole lot more) for the specified volume.
484 */
485 int
486 zvol_create_minor(const char *name)
487 {
488 zfs_soft_state_t *zs;
489 zvol_state_t *zv;
490 objset_t *os;
491 dmu_object_info_t doi;
492 minor_t minor = 0;
493 char chrbuf[30], blkbuf[30];
494 int error;
495
496 mutex_enter(&zfsdev_state_lock);
497
498 if (zvol_minor_lookup(name) != NULL) {
499 mutex_exit(&zfsdev_state_lock);
500 return (SET_ERROR(EEXIST));
501 }
502
503 /* lie and say we're read-only */
504 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
505
506 if (error) {
507 mutex_exit(&zfsdev_state_lock);
508 return (error);
509 }
510
511 if ((minor = zfsdev_minor_alloc()) == 0) {
512 dmu_objset_disown(os, FTAG);
513 mutex_exit(&zfsdev_state_lock);
514 return (SET_ERROR(ENXIO));
515 }
516
517 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
518 dmu_objset_disown(os, FTAG);
519 mutex_exit(&zfsdev_state_lock);
520 return (SET_ERROR(EAGAIN));
521 }
522 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
523 (char *)name);
524
525 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
526
527 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
528 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
529 ddi_soft_state_free(zfsdev_state, minor);
530 dmu_objset_disown(os, FTAG);
531 mutex_exit(&zfsdev_state_lock);
532 return (SET_ERROR(EAGAIN));
533 }
534
535 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
536
537 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
538 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
539 ddi_remove_minor_node(zfs_dip, chrbuf);
540 ddi_soft_state_free(zfsdev_state, minor);
541 dmu_objset_disown(os, FTAG);
542 mutex_exit(&zfsdev_state_lock);
543 return (SET_ERROR(EAGAIN));
544 }
545
546 zs = ddi_get_soft_state(zfsdev_state, minor);
547 zs->zss_type = ZSST_ZVOL;
548 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
549 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
550 zv->zv_min_bs = DEV_BSHIFT;
551 zv->zv_minor = minor;
552 zv->zv_objset = os;
553 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
554 zv->zv_flags |= ZVOL_RDONLY;
555 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
556 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
557 sizeof (rl_t), offsetof(rl_t, r_node));
558 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
559 offsetof(zvol_extent_t, ze_node));
560 /* get and cache the blocksize */
561 error = dmu_object_info(os, ZVOL_OBJ, &doi);
562 ASSERT(error == 0);
563 zv->zv_volblocksize = doi.doi_data_block_size;
564
565 if (spa_writeable(dmu_objset_spa(os))) {
566 if (zil_replay_disable)
567 zil_destroy(dmu_objset_zil(os), B_FALSE);
568 else
569 zil_replay(os, zv, zvol_replay_vector);
570 }
571 dmu_objset_disown(os, FTAG);
572 zv->zv_objset = NULL;
573
574 zvol_minors++;
575
576 mutex_exit(&zfsdev_state_lock);
577
578 return (0);
579 }
580
581 /*
582 * Remove minor node for the specified volume.
583 */
584 static int
585 zvol_remove_zv(zvol_state_t *zv)
586 {
587 char nmbuf[20];
588 minor_t minor = zv->zv_minor;
589
590 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
591 if (zv->zv_total_opens != 0)
592 return (SET_ERROR(EBUSY));
593
594 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
595 ddi_remove_minor_node(zfs_dip, nmbuf);
596
597 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
598 ddi_remove_minor_node(zfs_dip, nmbuf);
599
600 avl_destroy(&zv->zv_znode.z_range_avl);
601 mutex_destroy(&zv->zv_znode.z_range_lock);
602
603 kmem_free(zv, sizeof (zvol_state_t));
604
605 ddi_soft_state_free(zfsdev_state, minor);
606
607 zvol_minors--;
608 return (0);
609 }
610
611 int
612 zvol_remove_minor(const char *name)
613 {
614 zvol_state_t *zv;
615 int rc;
616
617 mutex_enter(&zfsdev_state_lock);
618 if ((zv = zvol_minor_lookup(name)) == NULL) {
619 mutex_exit(&zfsdev_state_lock);
620 return (SET_ERROR(ENXIO));
621 }
622 rc = zvol_remove_zv(zv);
623 mutex_exit(&zfsdev_state_lock);
624 return (rc);
625 }
626
627 int
628 zvol_first_open(zvol_state_t *zv)
629 {
630 objset_t *os;
631 uint64_t volsize;
632 int error;
633 uint64_t readonly;
634
635 /* lie and say we're read-only */
636 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
637 zvol_tag, &os);
638 if (error)
639 return (error);
640
641 zv->zv_objset = os;
642 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
643 if (error) {
644 ASSERT(error == 0);
645 dmu_objset_disown(os, zvol_tag);
646 return (error);
647 }
648
649 error = dnode_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dn);
650 if (error) {
651 dmu_objset_disown(os, zvol_tag);
652 return (error);
653 }
654
655 zvol_size_changed(zv, volsize);
656 zv->zv_zilog = zil_open(os, zvol_get_data);
657
658 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
659 NULL) == 0);
660 if (readonly || dmu_objset_is_snapshot(os) ||
661 !spa_writeable(dmu_objset_spa(os)))
662 zv->zv_flags |= ZVOL_RDONLY;
663 else
664 zv->zv_flags &= ~ZVOL_RDONLY;
665 return (error);
666 }
667
668 void
669 zvol_last_close(zvol_state_t *zv)
670 {
671 zil_close(zv->zv_zilog);
672 zv->zv_zilog = NULL;
673
674 dnode_rele(zv->zv_dn, zvol_tag);
675 zv->zv_dn = NULL;
676
677 /*
678 * Evict cached data
679 */
680 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
681 !(zv->zv_flags & ZVOL_RDONLY))
682 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
683 dmu_objset_evict_dbufs(zv->zv_objset);
684
685 dmu_objset_disown(zv->zv_objset, zvol_tag);
686 zv->zv_objset = NULL;
687 }
688
689 int
690 zvol_prealloc(zvol_state_t *zv)
691 {
692 objset_t *os = zv->zv_objset;
693 dmu_tx_t *tx;
694 uint64_t refd, avail, usedobjs, availobjs;
695 uint64_t resid = zv->zv_volsize;
696 uint64_t off = 0;
697
698 /* Check the space usage before attempting to allocate the space */
699 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
700 if (avail < zv->zv_volsize)
701 return (SET_ERROR(ENOSPC));
702
703 /* Free old extents if they exist */
704 zvol_free_extents(zv);
705
706 while (resid != 0) {
707 int error;
708 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
709
710 tx = dmu_tx_create(os);
711 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
712 error = dmu_tx_assign(tx, TXG_WAIT);
713 if (error) {
714 dmu_tx_abort(tx);
715 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
716 return (error);
717 }
718 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
719 dmu_tx_commit(tx);
720 off += bytes;
721 resid -= bytes;
722 }
723 txg_wait_synced(dmu_objset_pool(os), 0);
724
725 return (0);
726 }
727
728 static int
729 zvol_update_volsize(objset_t *os, uint64_t volsize)
730 {
731 dmu_tx_t *tx;
732 int error;
733
734 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
735
736 tx = dmu_tx_create(os);
737 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
738 dmu_tx_mark_netfree(tx);
739 error = dmu_tx_assign(tx, TXG_WAIT);
740 if (error) {
741 dmu_tx_abort(tx);
742 return (error);
743 }
744
745 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
746 &volsize, tx);
747 dmu_tx_commit(tx);
748
749 if (error == 0)
750 error = dmu_free_long_range(os,
751 ZVOL_OBJ, volsize, DMU_OBJECT_END);
752 return (error);
753 }
754
755 void
756 zvol_remove_minors(const char *name)
757 {
758 zvol_state_t *zv;
759 char *namebuf;
760 minor_t minor;
761
762 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
763 (void) strncpy(namebuf, name, strlen(name));
764 (void) strcat(namebuf, "/");
765 mutex_enter(&zfsdev_state_lock);
766 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
767
768 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
769 if (zv == NULL)
770 continue;
771 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
772 (void) zvol_remove_zv(zv);
773 }
774 kmem_free(namebuf, strlen(name) + 2);
775
776 mutex_exit(&zfsdev_state_lock);
777 }
778
779 static int
780 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
781 {
782 uint64_t old_volsize = 0ULL;
783 int error = 0;
784
785 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
786
787 /*
788 * Reinitialize the dump area to the new size. If we
789 * failed to resize the dump area then restore it back to
790 * its original size. We must set the new volsize prior
791 * to calling dumpvp_resize() to ensure that the devices'
792 * size(9P) is not visible by the dump subsystem.
793 */
794 old_volsize = zv->zv_volsize;
795 zvol_size_changed(zv, volsize);
796
797 if (zv->zv_flags & ZVOL_DUMPIFIED) {
798 if ((error = zvol_dumpify(zv)) != 0 ||
799 (error = dumpvp_resize()) != 0) {
800 int dumpify_error;
801
802 (void) zvol_update_volsize(zv->zv_objset, old_volsize);
803 zvol_size_changed(zv, old_volsize);
804 dumpify_error = zvol_dumpify(zv);
805 error = dumpify_error ? dumpify_error : error;
806 }
807 }
808
809 /*
810 * Generate a LUN expansion event.
811 */
812 if (error == 0) {
813 sysevent_id_t eid;
814 nvlist_t *attr;
815 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
816
817 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
818 zv->zv_minor);
819
820 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
821 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
822
823 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
824 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
825
826 nvlist_free(attr);
827 kmem_free(physpath, MAXPATHLEN);
828 }
829 return (error);
830 }
831
832 int
833 zvol_set_volsize(const char *name, uint64_t volsize)
834 {
835 zvol_state_t *zv = NULL;
836 objset_t *os;
837 int error;
838 dmu_object_info_t doi;
839 uint64_t readonly;
840 boolean_t owned = B_FALSE;
841
842 error = dsl_prop_get_integer(name,
843 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
844 if (error != 0)
845 return (error);
846 if (readonly)
847 return (SET_ERROR(EROFS));
848
849 mutex_enter(&zfsdev_state_lock);
850 zv = zvol_minor_lookup(name);
851
852 if (zv == NULL || zv->zv_objset == NULL) {
853 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
854 FTAG, &os)) != 0) {
855 mutex_exit(&zfsdev_state_lock);
856 return (error);
857 }
858 owned = B_TRUE;
859 if (zv != NULL)
860 zv->zv_objset = os;
861 } else {
862 os = zv->zv_objset;
863 }
864
865 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
866 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
867 goto out;
868
869 error = zvol_update_volsize(os, volsize);
870
871 if (error == 0 && zv != NULL)
872 error = zvol_update_live_volsize(zv, volsize);
873 out:
874 if (owned) {
875 dmu_objset_disown(os, FTAG);
876 if (zv != NULL)
877 zv->zv_objset = NULL;
878 }
879 mutex_exit(&zfsdev_state_lock);
880 return (error);
881 }
882
883 /*ARGSUSED*/
884 int
885 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
886 {
887 zvol_state_t *zv;
888 int err = 0;
889
890 mutex_enter(&zfsdev_state_lock);
891
892 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
893 if (zv == NULL) {
894 mutex_exit(&zfsdev_state_lock);
895 return (SET_ERROR(ENXIO));
896 }
897
898 if (zv->zv_total_opens == 0)
899 err = zvol_first_open(zv);
900 if (err) {
901 mutex_exit(&zfsdev_state_lock);
902 return (err);
903 }
904 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
905 err = SET_ERROR(EROFS);
906 goto out;
907 }
908 if (zv->zv_flags & ZVOL_EXCL) {
909 err = SET_ERROR(EBUSY);
910 goto out;
911 }
912 if (flag & FEXCL) {
913 if (zv->zv_total_opens != 0) {
914 err = SET_ERROR(EBUSY);
915 goto out;
916 }
917 zv->zv_flags |= ZVOL_EXCL;
918 }
919
920 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
921 zv->zv_open_count[otyp]++;
922 zv->zv_total_opens++;
923 }
924 mutex_exit(&zfsdev_state_lock);
925
926 return (err);
927 out:
928 if (zv->zv_total_opens == 0)
929 zvol_last_close(zv);
930 mutex_exit(&zfsdev_state_lock);
931 return (err);
932 }
933
934 /*ARGSUSED*/
935 int
936 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
937 {
938 minor_t minor = getminor(dev);
939 zvol_state_t *zv;
940 int error = 0;
941
942 mutex_enter(&zfsdev_state_lock);
943
944 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
945 if (zv == NULL) {
946 mutex_exit(&zfsdev_state_lock);
947 return (SET_ERROR(ENXIO));
948 }
949
950 if (zv->zv_flags & ZVOL_EXCL) {
951 ASSERT(zv->zv_total_opens == 1);
952 zv->zv_flags &= ~ZVOL_EXCL;
953 }
954
955 /*
956 * If the open count is zero, this is a spurious close.
957 * That indicates a bug in the kernel / DDI framework.
958 */
959 ASSERT(zv->zv_open_count[otyp] != 0);
960 ASSERT(zv->zv_total_opens != 0);
961
962 /*
963 * You may get multiple opens, but only one close.
964 */
965 zv->zv_open_count[otyp]--;
966 zv->zv_total_opens--;
967
968 if (zv->zv_total_opens == 0)
969 zvol_last_close(zv);
970
971 mutex_exit(&zfsdev_state_lock);
972 return (error);
973 }
974
975 static void
976 zvol_get_done(zgd_t *zgd, int error)
977 {
978 if (zgd->zgd_db)
979 dmu_buf_rele(zgd->zgd_db, zgd);
980
981 zfs_range_unlock(zgd->zgd_rl);
982
983 if (error == 0 && zgd->zgd_bp)
984 zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp);
985
986 kmem_free(zgd, sizeof (zgd_t));
987 }
988
989 /*
990 * Get data to generate a TX_WRITE intent log record.
991 */
992 static int
993 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
994 {
995 zvol_state_t *zv = arg;
996 uint64_t offset = lr->lr_offset;
997 uint64_t size = lr->lr_length; /* length of user data */
998 dmu_buf_t *db;
999 zgd_t *zgd;
1000 int error;
1001
1002 ASSERT3P(lwb, !=, NULL);
1003 ASSERT3P(zio, !=, NULL);
1004 ASSERT3U(size, !=, 0);
1005
1006 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1007 zgd->zgd_lwb = lwb;
1008
1009 /*
1010 * Write records come in two flavors: immediate and indirect.
1011 * For small writes it's cheaper to store the data with the
1012 * log record (immediate); for large writes it's cheaper to
1013 * sync the data and get a pointer to it (indirect) so that
1014 * we don't have to write the data twice.
1015 */
1016 if (buf != NULL) { /* immediate write */
1017 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size,
1018 RL_READER);
1019 error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
1020 DMU_READ_NO_PREFETCH);
1021 } else { /* indirect write */
1022 /*
1023 * Have to lock the whole block to ensure when it's written out
1024 * and its checksum is being calculated that no one can change
1025 * the data. Contrarily to zfs_get_data we need not re-check
1026 * blocksize after we get the lock because it cannot be changed.
1027 */
1028 size = zv->zv_volblocksize;
1029 offset = P2ALIGN(offset, size);
1030 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size,
1031 RL_READER);
1032 error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db,
1033 DMU_READ_NO_PREFETCH);
1034 if (error == 0) {
1035 blkptr_t *bp = &lr->lr_blkptr;
1036
1037 zgd->zgd_db = db;
1038 zgd->zgd_bp = bp;
1039
1040 ASSERT(db->db_offset == offset);
1041 ASSERT(db->db_size == size);
1042
1043 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1044 zvol_get_done, zgd);
1045
1046 if (error == 0)
1047 return (0);
1048 }
1049 }
1050
1051 zvol_get_done(zgd, error);
1052
1053 return (error);
1054 }
1055
1056 /*
1057 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1058 *
1059 * We store data in the log buffers if it's small enough.
1060 * Otherwise we will later flush the data out via dmu_sync().
1061 */
1062 ssize_t zvol_immediate_write_sz = 32768;
1063
1064 static void
1065 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1066 boolean_t sync)
1067 {
1068 uint32_t blocksize = zv->zv_volblocksize;
1069 zilog_t *zilog = zv->zv_zilog;
1070 itx_wr_state_t write_state;
1071
1072 if (zil_replaying(zilog, tx))
1073 return;
1074
1075 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1076 write_state = WR_INDIRECT;
1077 else if (!spa_has_slogs(zilog->zl_spa) &&
1078 resid >= blocksize && blocksize > zvol_immediate_write_sz)
1079 write_state = WR_INDIRECT;
1080 else if (sync)
1081 write_state = WR_COPIED;
1082 else
1083 write_state = WR_NEED_COPY;
1084
1085 while (resid) {
1086 itx_t *itx;
1087 lr_write_t *lr;
1088 itx_wr_state_t wr_state = write_state;
1089 ssize_t len = resid;
1090
1091 if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA)
1092 wr_state = WR_NEED_COPY;
1093 else if (wr_state == WR_INDIRECT)
1094 len = MIN(blocksize - P2PHASE(off, blocksize), resid);
1095
1096 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1097 (wr_state == WR_COPIED ? len : 0));
1098 lr = (lr_write_t *)&itx->itx_lr;
1099 if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
1100 off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1101 zil_itx_destroy(itx);
1102 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1103 lr = (lr_write_t *)&itx->itx_lr;
1104 wr_state = WR_NEED_COPY;
1105 }
1106
1107 itx->itx_wr_state = wr_state;
1108 lr->lr_foid = ZVOL_OBJ;
1109 lr->lr_offset = off;
1110 lr->lr_length = len;
1111 lr->lr_blkoff = 0;
1112 BP_ZERO(&lr->lr_blkptr);
1113
1114 itx->itx_private = zv;
1115 itx->itx_sync = sync;
1116
1117 zil_itx_assign(zilog, itx, tx);
1118
1119 off += len;
1120 resid -= len;
1121 }
1122 }
1123
1124 static int
1125 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1126 uint64_t size, boolean_t doread, boolean_t isdump)
1127 {
1128 vdev_disk_t *dvd;
1129 int c;
1130 int numerrors = 0;
1131
1132 if (vd->vdev_ops == &vdev_mirror_ops ||
1133 vd->vdev_ops == &vdev_replacing_ops ||
1134 vd->vdev_ops == &vdev_spare_ops) {
1135 for (c = 0; c < vd->vdev_children; c++) {
1136 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1137 addr, offset, origoffset, size, doread, isdump);
1138 if (err != 0) {
1139 numerrors++;
1140 } else if (doread) {
1141 break;
1142 }
1143 }
1144 }
1145
1146 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1147 return (numerrors < vd->vdev_children ? 0 : EIO);
1148
1149 if (doread && !vdev_readable(vd))
1150 return (SET_ERROR(EIO));
1151 else if (!doread && !vdev_writeable(vd))
1152 return (SET_ERROR(EIO));
1153
1154 if (vd->vdev_ops == &vdev_raidz_ops) {
1155 return (vdev_raidz_physio(vd,
1156 addr, size, offset, origoffset, doread, isdump));
1157 }
1158
1159 offset += VDEV_LABEL_START_SIZE;
1160
1161 if (ddi_in_panic() || isdump) {
1162 ASSERT(!doread);
1163 if (doread)
1164 return (SET_ERROR(EIO));
1165 dvd = vd->vdev_tsd;
1166 ASSERT3P(dvd, !=, NULL);
1167 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1168 lbtodb(size)));
1169 } else {
1170 dvd = vd->vdev_tsd;
1171 ASSERT3P(dvd, !=, NULL);
1172 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1173 offset, doread ? B_READ : B_WRITE));
1174 }
1175 }
1176
1177 static int
1178 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1179 boolean_t doread, boolean_t isdump)
1180 {
1181 vdev_t *vd;
1182 int error;
1183 zvol_extent_t *ze;
1184 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1185
1186 /* Must be sector aligned, and not stradle a block boundary. */
1187 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1188 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1189 return (SET_ERROR(EINVAL));
1190 }
1191 ASSERT(size <= zv->zv_volblocksize);
1192
1193 /* Locate the extent this belongs to */
1194 ze = list_head(&zv->zv_extents);
1195 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1196 offset -= ze->ze_nblks * zv->zv_volblocksize;
1197 ze = list_next(&zv->zv_extents, ze);
1198 }
1199
1200 if (ze == NULL)
1201 return (SET_ERROR(EINVAL));
1202
1203 if (!ddi_in_panic())
1204 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1205
1206 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1207 offset += DVA_GET_OFFSET(&ze->ze_dva);
1208 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1209 size, doread, isdump);
1210
1211 if (!ddi_in_panic())
1212 spa_config_exit(spa, SCL_STATE, FTAG);
1213
1214 return (error);
1215 }
1216
1217 int
1218 zvol_strategy(buf_t *bp)
1219 {
1220 zfs_soft_state_t *zs = NULL;
1221 zvol_state_t *zv;
1222 uint64_t off, volsize;
1223 size_t resid;
1224 char *addr;
1225 objset_t *os;
1226 rl_t *rl;
1227 int error = 0;
1228 boolean_t doread = bp->b_flags & B_READ;
1229 boolean_t is_dumpified;
1230 boolean_t sync;
1231
1232 if (getminor(bp->b_edev) == 0) {
1233 error = SET_ERROR(EINVAL);
1234 } else {
1235 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1236 if (zs == NULL)
1237 error = SET_ERROR(ENXIO);
1238 else if (zs->zss_type != ZSST_ZVOL)
1239 error = SET_ERROR(EINVAL);
1240 }
1241
1242 if (error) {
1243 bioerror(bp, error);
1244 biodone(bp);
1245 return (0);
1246 }
1247
1248 zv = zs->zss_data;
1249
1250 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1251 bioerror(bp, EROFS);
1252 biodone(bp);
1253 return (0);
1254 }
1255
1256 off = ldbtob(bp->b_blkno);
1257 volsize = zv->zv_volsize;
1258
1259 os = zv->zv_objset;
1260 ASSERT(os != NULL);
1261
1262 bp_mapin(bp);
1263 addr = bp->b_un.b_addr;
1264 resid = bp->b_bcount;
1265
1266 if (resid > 0 && (off < 0 || off >= volsize)) {
1267 bioerror(bp, EIO);
1268 biodone(bp);
1269 return (0);
1270 }
1271
1272 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1273 sync = ((!(bp->b_flags & B_ASYNC) &&
1274 !(zv->zv_flags & ZVOL_WCE)) ||
1275 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1276 !doread && !is_dumpified;
1277
1278 /*
1279 * There must be no buffer changes when doing a dmu_sync() because
1280 * we can't change the data whilst calculating the checksum.
1281 */
1282 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1283 doread ? RL_READER : RL_WRITER);
1284
1285 while (resid != 0 && off < volsize) {
1286 size_t size = MIN(resid, zvol_maxphys);
1287 if (is_dumpified) {
1288 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1289 error = zvol_dumpio(zv, addr, off, size,
1290 doread, B_FALSE);
1291 } else if (doread) {
1292 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1293 DMU_READ_PREFETCH);
1294 } else {
1295 dmu_tx_t *tx = dmu_tx_create(os);
1296 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1297 error = dmu_tx_assign(tx, TXG_WAIT);
1298 if (error) {
1299 dmu_tx_abort(tx);
1300 } else {
1301 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1302 zvol_log_write(zv, tx, off, size, sync);
1303 dmu_tx_commit(tx);
1304 }
1305 }
1306 if (error) {
1307 /* convert checksum errors into IO errors */
1308 if (error == ECKSUM)
1309 error = SET_ERROR(EIO);
1310 break;
1311 }
1312 off += size;
1313 addr += size;
1314 resid -= size;
1315 }
1316 zfs_range_unlock(rl);
1317
1318 if ((bp->b_resid = resid) == bp->b_bcount)
1319 bioerror(bp, off > volsize ? EINVAL : error);
1320
1321 if (sync)
1322 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1323 biodone(bp);
1324
1325 return (0);
1326 }
1327
1328 /*
1329 * Set the buffer count to the zvol maximum transfer.
1330 * Using our own routine instead of the default minphys()
1331 * means that for larger writes we write bigger buffers on X86
1332 * (128K instead of 56K) and flush the disk write cache less often
1333 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1334 * 56K on X86 and 128K on sparc).
1335 */
1336 void
1337 zvol_minphys(struct buf *bp)
1338 {
1339 if (bp->b_bcount > zvol_maxphys)
1340 bp->b_bcount = zvol_maxphys;
1341 }
1342
1343 int
1344 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1345 {
1346 minor_t minor = getminor(dev);
1347 zvol_state_t *zv;
1348 int error = 0;
1349 uint64_t size;
1350 uint64_t boff;
1351 uint64_t resid;
1352
1353 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1354 if (zv == NULL)
1355 return (SET_ERROR(ENXIO));
1356
1357 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1358 return (SET_ERROR(EINVAL));
1359
1360 boff = ldbtob(blkno);
1361 resid = ldbtob(nblocks);
1362
1363 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1364
1365 while (resid) {
1366 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1367 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1368 if (error)
1369 break;
1370 boff += size;
1371 addr += size;
1372 resid -= size;
1373 }
1374
1375 return (error);
1376 }
1377
1378 /*ARGSUSED*/
1379 int
1380 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1381 {
1382 minor_t minor = getminor(dev);
1383 zvol_state_t *zv;
1384 uint64_t volsize;
1385 rl_t *rl;
1386 int error = 0;
1387
1388 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1389 if (zv == NULL)
1390 return (SET_ERROR(ENXIO));
1391
1392 volsize = zv->zv_volsize;
1393 if (uio->uio_resid > 0 &&
1394 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1395 return (SET_ERROR(EIO));
1396
1397 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1398 error = physio(zvol_strategy, NULL, dev, B_READ,
1399 zvol_minphys, uio);
1400 return (error);
1401 }
1402
1403 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1404 RL_READER);
1405 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1406 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1407
1408 /* don't read past the end */
1409 if (bytes > volsize - uio->uio_loffset)
1410 bytes = volsize - uio->uio_loffset;
1411
1412 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1413 if (error) {
1414 /* convert checksum errors into IO errors */
1415 if (error == ECKSUM)
1416 error = SET_ERROR(EIO);
1417 break;
1418 }
1419 }
1420 zfs_range_unlock(rl);
1421 return (error);
1422 }
1423
1424 /*ARGSUSED*/
1425 int
1426 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1427 {
1428 minor_t minor = getminor(dev);
1429 zvol_state_t *zv;
1430 uint64_t volsize;
1431 rl_t *rl;
1432 int error = 0;
1433 boolean_t sync;
1434
1435 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1436 if (zv == NULL)
1437 return (SET_ERROR(ENXIO));
1438
1439 volsize = zv->zv_volsize;
1440 if (uio->uio_resid > 0 &&
1441 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1442 return (SET_ERROR(EIO));
1443
1444 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1445 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1446 zvol_minphys, uio);
1447 return (error);
1448 }
1449
1450 sync = !(zv->zv_flags & ZVOL_WCE) ||
1451 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1452
1453 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1454 RL_WRITER);
1455 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1456 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1457 uint64_t off = uio->uio_loffset;
1458 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1459
1460 if (bytes > volsize - off) /* don't write past the end */
1461 bytes = volsize - off;
1462
1463 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1464 error = dmu_tx_assign(tx, TXG_WAIT);
1465 if (error) {
1466 dmu_tx_abort(tx);
1467 break;
1468 }
1469 error = dmu_write_uio_dnode(zv->zv_dn, uio, bytes, tx);
1470 if (error == 0)
1471 zvol_log_write(zv, tx, off, bytes, sync);
1472 dmu_tx_commit(tx);
1473
1474 if (error)
1475 break;
1476 }
1477 zfs_range_unlock(rl);
1478 if (sync)
1479 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1480 return (error);
1481 }
1482
1483 int
1484 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1485 {
1486 struct uuid uuid = EFI_RESERVED;
1487 efi_gpe_t gpe = { 0 };
1488 uint32_t crc;
1489 dk_efi_t efi;
1490 int length;
1491 char *ptr;
1492
1493 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1494 return (SET_ERROR(EFAULT));
1495 ptr = (char *)(uintptr_t)efi.dki_data_64;
1496 length = efi.dki_length;
1497 /*
1498 * Some clients may attempt to request a PMBR for the
1499 * zvol. Currently this interface will return EINVAL to
1500 * such requests. These requests could be supported by
1501 * adding a check for lba == 0 and consing up an appropriate
1502 * PMBR.
1503 */
1504 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1505 return (SET_ERROR(EINVAL));
1506
1507 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1508 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1509 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1510
1511 if (efi.dki_lba == 1) {
1512 efi_gpt_t gpt = { 0 };
1513
1514 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1515 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1516 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1517 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1518 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1519 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1520 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1521 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1522 gpt.efi_gpt_SizeOfPartitionEntry =
1523 LE_32(sizeof (efi_gpe_t));
1524 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1525 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1526 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1527 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1528 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1529 flag))
1530 return (SET_ERROR(EFAULT));
1531 ptr += sizeof (gpt);
1532 length -= sizeof (gpt);
1533 }
1534 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1535 length), flag))
1536 return (SET_ERROR(EFAULT));
1537 return (0);
1538 }
1539
1540 /*
1541 * BEGIN entry points to allow external callers access to the volume.
1542 */
1543 /*
1544 * Return the volume parameters needed for access from an external caller.
1545 * These values are invariant as long as the volume is held open.
1546 */
1547 int
1548 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1549 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1550 void **rl_hdl, void **dnode_hdl)
1551 {
1552 zvol_state_t *zv;
1553
1554 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1555 if (zv == NULL)
1556 return (SET_ERROR(ENXIO));
1557 if (zv->zv_flags & ZVOL_DUMPIFIED)
1558 return (SET_ERROR(ENXIO));
1559
1560 ASSERT(blksize && max_xfer_len && minor_hdl &&
1561 objset_hdl && zil_hdl && rl_hdl && dnode_hdl);
1562
1563 *blksize = zv->zv_volblocksize;
1564 *max_xfer_len = (uint64_t)zvol_maxphys;
1565 *minor_hdl = zv;
1566 *objset_hdl = zv->zv_objset;
1567 *zil_hdl = zv->zv_zilog;
1568 *rl_hdl = &zv->zv_znode;
1569 *dnode_hdl = zv->zv_dn;
1570 return (0);
1571 }
1572
1573 /*
1574 * Return the current volume size to an external caller.
1575 * The size can change while the volume is open.
1576 */
1577 uint64_t
1578 zvol_get_volume_size(void *minor_hdl)
1579 {
1580 zvol_state_t *zv = minor_hdl;
1581
1582 return (zv->zv_volsize);
1583 }
1584
1585 /*
1586 * Return the current WCE setting to an external caller.
1587 * The WCE setting can change while the volume is open.
1588 */
1589 int
1590 zvol_get_volume_wce(void *minor_hdl)
1591 {
1592 zvol_state_t *zv = minor_hdl;
1593
1594 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1595 }
1596
1597 /*
1598 * Entry point for external callers to zvol_log_write
1599 */
1600 void
1601 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1602 boolean_t sync)
1603 {
1604 zvol_state_t *zv = minor_hdl;
1605
1606 zvol_log_write(zv, tx, off, resid, sync);
1607 }
1608 /*
1609 * END entry points to allow external callers access to the volume.
1610 */
1611
1612 /*
1613 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1614 */
1615 static void
1616 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1617 boolean_t sync)
1618 {
1619 itx_t *itx;
1620 lr_truncate_t *lr;
1621 zilog_t *zilog = zv->zv_zilog;
1622
1623 if (zil_replaying(zilog, tx))
1624 return;
1625
1626 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1627 lr = (lr_truncate_t *)&itx->itx_lr;
1628 lr->lr_foid = ZVOL_OBJ;
1629 lr->lr_offset = off;
1630 lr->lr_length = len;
1631
1632 itx->itx_sync = sync;
1633 zil_itx_assign(zilog, itx, tx);
1634 }
1635
1636 /*
1637 * Dirtbag ioctls to support mkfs(8) for UFS filesystems. See dkio(7I).
1638 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1639 */
1640 /*ARGSUSED*/
1641 int
1642 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1643 {
1644 zvol_state_t *zv;
1645 struct dk_callback *dkc;
1646 int error = 0;
1647 rl_t *rl;
1648
1649 mutex_enter(&zfsdev_state_lock);
1650
1651 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1652
1653 if (zv == NULL) {
1654 mutex_exit(&zfsdev_state_lock);
1655 return (SET_ERROR(ENXIO));
1656 }
1657 ASSERT(zv->zv_total_opens > 0);
1658
1659 switch (cmd) {
1660
1661 case DKIOCINFO:
1662 {
1663 struct dk_cinfo dki;
1664
1665 bzero(&dki, sizeof (dki));
1666 (void) strcpy(dki.dki_cname, "zvol");
1667 (void) strcpy(dki.dki_dname, "zvol");
1668 dki.dki_ctype = DKC_UNKNOWN;
1669 dki.dki_unit = getminor(dev);
1670 dki.dki_maxtransfer =
1671 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
1672 mutex_exit(&zfsdev_state_lock);
1673 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1674 error = SET_ERROR(EFAULT);
1675 return (error);
1676 }
1677
1678 case DKIOCGMEDIAINFO:
1679 {
1680 struct dk_minfo dkm;
1681
1682 bzero(&dkm, sizeof (dkm));
1683 dkm.dki_lbsize = 1U << zv->zv_min_bs;
1684 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1685 dkm.dki_media_type = DK_UNKNOWN;
1686 mutex_exit(&zfsdev_state_lock);
1687 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1688 error = SET_ERROR(EFAULT);
1689 return (error);
1690 }
1691
1692 case DKIOCGMEDIAINFOEXT:
1693 {
1694 struct dk_minfo_ext dkmext;
1695
1696 bzero(&dkmext, sizeof (dkmext));
1697 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
1698 dkmext.dki_pbsize = zv->zv_volblocksize;
1699 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1700 dkmext.dki_media_type = DK_UNKNOWN;
1701 mutex_exit(&zfsdev_state_lock);
1702 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
1703 error = SET_ERROR(EFAULT);
1704 return (error);
1705 }
1706
1707 case DKIOCGETEFI:
1708 {
1709 uint64_t vs = zv->zv_volsize;
1710 uint8_t bs = zv->zv_min_bs;
1711
1712 mutex_exit(&zfsdev_state_lock);
1713 error = zvol_getefi((void *)arg, flag, vs, bs);
1714 return (error);
1715 }
1716
1717 case DKIOCFLUSHWRITECACHE:
1718 dkc = (struct dk_callback *)arg;
1719 mutex_exit(&zfsdev_state_lock);
1720 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1721 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1722 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
1723 error = 0;
1724 }
1725 return (error);
1726
1727 case DKIOCGETWCE:
1728 {
1729 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1730 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1731 flag))
1732 error = SET_ERROR(EFAULT);
1733 break;
1734 }
1735 case DKIOCSETWCE:
1736 {
1737 int wce;
1738 if (ddi_copyin((void *)arg, &wce, sizeof (int),
1739 flag)) {
1740 error = SET_ERROR(EFAULT);
1741 break;
1742 }
1743 if (wce) {
1744 zv->zv_flags |= ZVOL_WCE;
1745 mutex_exit(&zfsdev_state_lock);
1746 } else {
1747 zv->zv_flags &= ~ZVOL_WCE;
1748 mutex_exit(&zfsdev_state_lock);
1749 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1750 }
1751 return (0);
1752 }
1753
1754 case DKIOCGGEOM:
1755 case DKIOCGVTOC:
1756 /*
1757 * commands using these (like prtvtoc) expect ENOTSUP
1758 * since we're emulating an EFI label
1759 */
1760 error = SET_ERROR(ENOTSUP);
1761 break;
1762
1763 case DKIOCDUMPINIT:
1764 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1765 RL_WRITER);
1766 error = zvol_dumpify(zv);
1767 zfs_range_unlock(rl);
1768 break;
1769
1770 case DKIOCDUMPFINI:
1771 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1772 break;
1773 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1774 RL_WRITER);
1775 error = zvol_dump_fini(zv);
1776 zfs_range_unlock(rl);
1777 break;
1778
1779 case DKIOCFREE:
1780 {
1781 dkioc_free_t df;
1782 dmu_tx_t *tx;
1783
1784 if (!zvol_unmap_enabled)
1785 break;
1786
1787 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
1788 error = SET_ERROR(EFAULT);
1789 break;
1790 }
1791
1792 /*
1793 * Apply Postel's Law to length-checking. If they overshoot,
1794 * just blank out until the end, if there's a need to blank
1795 * out anything.
1796 */
1797 if (df.df_start >= zv->zv_volsize)
1798 break; /* No need to do anything... */
1799
1800 mutex_exit(&zfsdev_state_lock);
1801
1802 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
1803 RL_WRITER);
1804 tx = dmu_tx_create(zv->zv_objset);
1805 dmu_tx_mark_netfree(tx);
1806 error = dmu_tx_assign(tx, TXG_WAIT);
1807 if (error != 0) {
1808 dmu_tx_abort(tx);
1809 } else {
1810 zvol_log_truncate(zv, tx, df.df_start,
1811 df.df_length, B_TRUE);
1812 dmu_tx_commit(tx);
1813 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1814 df.df_start, df.df_length);
1815 }
1816
1817 zfs_range_unlock(rl);
1818
1819 /*
1820 * If the write-cache is disabled, 'sync' property
1821 * is set to 'always', or if the caller is asking for
1822 * a synchronous free, commit this operation to the zil.
1823 * This will sync any previous uncommitted writes to the
1824 * zvol object.
1825 * Can be overridden by the zvol_unmap_sync_enabled tunable.
1826 */
1827 if ((error == 0) && zvol_unmap_sync_enabled &&
1828 (!(zv->zv_flags & ZVOL_WCE) ||
1829 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) ||
1830 (df.df_flags & DF_WAIT_SYNC))) {
1831 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1832 }
1833
1834 return (error);
1835 }
1836
1837 default:
1838 error = SET_ERROR(ENOTTY);
1839 break;
1840
1841 }
1842 mutex_exit(&zfsdev_state_lock);
1843 return (error);
1844 }
1845
1846 int
1847 zvol_busy(void)
1848 {
1849 return (zvol_minors != 0);
1850 }
1851
1852 void
1853 zvol_init(void)
1854 {
1855 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1856 1) == 0);
1857 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
1858 }
1859
1860 void
1861 zvol_fini(void)
1862 {
1863 mutex_destroy(&zfsdev_state_lock);
1864 ddi_soft_state_fini(&zfsdev_state);
1865 }
1866
1867 /*ARGSUSED*/
1868 static int
1869 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
1870 {
1871 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1872
1873 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1874 return (1);
1875 return (0);
1876 }
1877
1878 /*ARGSUSED*/
1879 static void
1880 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
1881 {
1882 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1883
1884 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
1885 }
1886
1887 static int
1888 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1889 {
1890 dmu_tx_t *tx;
1891 int error;
1892 objset_t *os = zv->zv_objset;
1893 spa_t *spa = dmu_objset_spa(os);
1894 vdev_t *vd = spa->spa_root_vdev;
1895 nvlist_t *nv = NULL;
1896 uint64_t version = spa_version(spa);
1897 uint64_t checksum, compress, refresrv, vbs, dedup;
1898
1899 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1900 ASSERT(vd->vdev_ops == &vdev_root_ops);
1901
1902 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1903 DMU_OBJECT_END);
1904 if (error != 0)
1905 return (error);
1906 /* wait for dmu_free_long_range to actually free the blocks */
1907 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1908
1909 /*
1910 * If the pool on which the dump device is being initialized has more
1911 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
1912 * enabled. If so, bump that feature's counter to indicate that the
1913 * feature is active. We also check the vdev type to handle the
1914 * following case:
1915 * # zpool create test raidz disk1 disk2 disk3
1916 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
1917 * the raidz vdev itself has 3 children.
1918 */
1919 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
1920 if (!spa_feature_is_enabled(spa,
1921 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1922 return (SET_ERROR(ENOTSUP));
1923 (void) dsl_sync_task(spa_name(spa),
1924 zfs_mvdev_dump_feature_check,
1925 zfs_mvdev_dump_activate_feature_sync, NULL,
1926 2, ZFS_SPACE_CHECK_RESERVED);
1927 }
1928
1929 if (!resize) {
1930 error = dsl_prop_get_integer(zv->zv_name,
1931 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1932 if (error == 0) {
1933 error = dsl_prop_get_integer(zv->zv_name,
1934 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
1935 NULL);
1936 }
1937 if (error == 0) {
1938 error = dsl_prop_get_integer(zv->zv_name,
1939 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
1940 &refresrv, NULL);
1941 }
1942 if (error == 0) {
1943 error = dsl_prop_get_integer(zv->zv_name,
1944 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
1945 NULL);
1946 }
1947 if (version >= SPA_VERSION_DEDUP && error == 0) {
1948 error = dsl_prop_get_integer(zv->zv_name,
1949 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1950 }
1951 }
1952 if (error != 0)
1953 return (error);
1954
1955 tx = dmu_tx_create(os);
1956 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1957 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1958 error = dmu_tx_assign(tx, TXG_WAIT);
1959 if (error != 0) {
1960 dmu_tx_abort(tx);
1961 return (error);
1962 }
1963
1964 /*
1965 * If we are resizing the dump device then we only need to
1966 * update the refreservation to match the newly updated
1967 * zvolsize. Otherwise, we save off the original state of the
1968 * zvol so that we can restore them if the zvol is ever undumpified.
1969 */
1970 if (resize) {
1971 error = zap_update(os, ZVOL_ZAP_OBJ,
1972 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1973 &zv->zv_volsize, tx);
1974 } else {
1975 error = zap_update(os, ZVOL_ZAP_OBJ,
1976 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1977 &compress, tx);
1978 if (error == 0) {
1979 error = zap_update(os, ZVOL_ZAP_OBJ,
1980 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
1981 &checksum, tx);
1982 }
1983 if (error == 0) {
1984 error = zap_update(os, ZVOL_ZAP_OBJ,
1985 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1986 &refresrv, tx);
1987 }
1988 if (error == 0) {
1989 error = zap_update(os, ZVOL_ZAP_OBJ,
1990 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1991 &vbs, tx);
1992 }
1993 if (error == 0) {
1994 error = dmu_object_set_blocksize(
1995 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
1996 }
1997 if (version >= SPA_VERSION_DEDUP && error == 0) {
1998 error = zap_update(os, ZVOL_ZAP_OBJ,
1999 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2000 &dedup, tx);
2001 }
2002 if (error == 0)
2003 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2004 }
2005 dmu_tx_commit(tx);
2006
2007 /*
2008 * We only need update the zvol's property if we are initializing
2009 * the dump area for the first time.
2010 */
2011 if (error == 0 && !resize) {
2012 /*
2013 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2014 * function. Otherwise, use the old default -- OFF.
2015 */
2016 checksum = spa_feature_is_active(spa,
2017 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2018 ZIO_CHECKSUM_OFF;
2019
2020 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2021 VERIFY(nvlist_add_uint64(nv,
2022 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2023 VERIFY(nvlist_add_uint64(nv,
2024 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2025 ZIO_COMPRESS_OFF) == 0);
2026 VERIFY(nvlist_add_uint64(nv,
2027 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2028 checksum) == 0);
2029 if (version >= SPA_VERSION_DEDUP) {
2030 VERIFY(nvlist_add_uint64(nv,
2031 zfs_prop_to_name(ZFS_PROP_DEDUP),
2032 ZIO_CHECKSUM_OFF) == 0);
2033 }
2034
2035 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2036 nv, NULL);
2037 nvlist_free(nv);
2038 }
2039
2040 /* Allocate the space for the dump */
2041 if (error == 0)
2042 error = zvol_prealloc(zv);
2043 return (error);
2044 }
2045
2046 static int
2047 zvol_dumpify(zvol_state_t *zv)
2048 {
2049 int error = 0;
2050 uint64_t dumpsize = 0;
2051 dmu_tx_t *tx;
2052 objset_t *os = zv->zv_objset;
2053
2054 if (zv->zv_flags & ZVOL_RDONLY)
2055 return (SET_ERROR(EROFS));
2056
2057 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2058 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2059 boolean_t resize = (dumpsize > 0);
2060
2061 if ((error = zvol_dump_init(zv, resize)) != 0) {
2062 (void) zvol_dump_fini(zv);
2063 return (error);
2064 }
2065 }
2066
2067 /*
2068 * Build up our lba mapping.
2069 */
2070 error = zvol_get_lbas(zv);
2071 if (error) {
2072 (void) zvol_dump_fini(zv);
2073 return (error);
2074 }
2075
2076 tx = dmu_tx_create(os);
2077 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2078 error = dmu_tx_assign(tx, TXG_WAIT);
2079 if (error) {
2080 dmu_tx_abort(tx);
2081 (void) zvol_dump_fini(zv);
2082 return (error);
2083 }
2084
2085 zv->zv_flags |= ZVOL_DUMPIFIED;
2086 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2087 &zv->zv_volsize, tx);
2088 dmu_tx_commit(tx);
2089
2090 if (error) {
2091 (void) zvol_dump_fini(zv);
2092 return (error);
2093 }
2094
2095 txg_wait_synced(dmu_objset_pool(os), 0);
2096 return (0);
2097 }
2098
2099 static int
2100 zvol_dump_fini(zvol_state_t *zv)
2101 {
2102 dmu_tx_t *tx;
2103 objset_t *os = zv->zv_objset;
2104 nvlist_t *nv;
2105 int error = 0;
2106 uint64_t checksum, compress, refresrv, vbs, dedup;
2107 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2108
2109 /*
2110 * Attempt to restore the zvol back to its pre-dumpified state.
2111 * This is a best-effort attempt as it's possible that not all
2112 * of these properties were initialized during the dumpify process
2113 * (i.e. error during zvol_dump_init).
2114 */
2115
2116 tx = dmu_tx_create(os);
2117 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2118 error = dmu_tx_assign(tx, TXG_WAIT);
2119 if (error) {
2120 dmu_tx_abort(tx);
2121 return (error);
2122 }
2123 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2124 dmu_tx_commit(tx);
2125
2126 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2127 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2128 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2129 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2130 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2131 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2132 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2133 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2134
2135 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2136 (void) nvlist_add_uint64(nv,
2137 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2138 (void) nvlist_add_uint64(nv,
2139 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2140 (void) nvlist_add_uint64(nv,
2141 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2142 if (version >= SPA_VERSION_DEDUP &&
2143 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2144 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2145 (void) nvlist_add_uint64(nv,
2146 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2147 }
2148 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2149 nv, NULL);
2150 nvlist_free(nv);
2151
2152 zvol_free_extents(zv);
2153 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2154 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2155 /* wait for dmu_free_long_range to actually free the blocks */
2156 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2157 tx = dmu_tx_create(os);
2158 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2159 error = dmu_tx_assign(tx, TXG_WAIT);
2160 if (error) {
2161 dmu_tx_abort(tx);
2162 return (error);
2163 }
2164 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2165 zv->zv_volblocksize = vbs;
2166 dmu_tx_commit(tx);
2167
2168 return (0);
2169 }
Unleashed OS