?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
4757 ZFS embedded-data block pointers ("zero block compression")
[unleashed.git] / usr / src / uts / common / fs / zfs / zvol.c
blob19f6bb97536cc0307ab4c9a330187e94134a746e
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) 2013 by Delphix. All rights reserved.
28 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
29 */
30
31 /*
32 * ZFS volume emulation driver.
33 *
34 * Makes a DMU object look like a volume of arbitrary size, up to 2^64 bytes.
35 * Volumes are accessed through the symbolic links named:
36 *
37 * /dev/zvol/dsk/<pool_name>/<dataset_name>
38 * /dev/zvol/rdsk/<pool_name>/<dataset_name>
39 *
40 * These links are created by the /dev filesystem (sdev_zvolops.c).
41 * Volumes are persistent through reboot. No user command needs to be
42 * run before opening and using a device.
43 */
44
45 #include <sys/types.h>
46 #include <sys/param.h>
47 #include <sys/errno.h>
48 #include <sys/uio.h>
49 #include <sys/buf.h>
50 #include <sys/modctl.h>
51 #include <sys/open.h>
52 #include <sys/kmem.h>
53 #include <sys/conf.h>
54 #include <sys/cmn_err.h>
55 #include <sys/stat.h>
56 #include <sys/zap.h>
57 #include <sys/spa.h>
58 #include <sys/spa_impl.h>
59 #include <sys/zio.h>
60 #include <sys/dmu_traverse.h>
61 #include <sys/dnode.h>
62 #include <sys/dsl_dataset.h>
63 #include <sys/dsl_prop.h>
64 #include <sys/dkio.h>
65 #include <sys/efi_partition.h>
66 #include <sys/byteorder.h>
67 #include <sys/pathname.h>
68 #include <sys/ddi.h>
69 #include <sys/sunddi.h>
70 #include <sys/crc32.h>
71 #include <sys/dirent.h>
72 #include <sys/policy.h>
73 #include <sys/fs/zfs.h>
74 #include <sys/zfs_ioctl.h>
75 #include <sys/mkdev.h>
76 #include <sys/zil.h>
77 #include <sys/refcount.h>
78 #include <sys/zfs_znode.h>
79 #include <sys/zfs_rlock.h>
80 #include <sys/vdev_disk.h>
81 #include <sys/vdev_impl.h>
82 #include <sys/vdev_raidz.h>
83 #include <sys/zvol.h>
84 #include <sys/dumphdr.h>
85 #include <sys/zil_impl.h>
86 #include <sys/dbuf.h>
87 #include <sys/dmu_tx.h>
88 #include <sys/zfeature.h>
89 #include <sys/zio_checksum.h>
90
91 #include "zfs_namecheck.h"
92
93 void *zfsdev_state;
94 static char *zvol_tag = "zvol_tag";
95
96 #define ZVOL_DUMPSIZE "dumpsize"
97
98 /*
99 * This lock protects the zfsdev_state structure from being modified
100 * while it's being used, e.g. an open that comes in before a create
101 * finishes. It also protects temporary opens of the dataset so that,
102 * e.g., an open doesn't get a spurious EBUSY.
103 */
104 kmutex_t zfsdev_state_lock;
105 static uint32_t zvol_minors;
106
107 typedef struct zvol_extent {
108 list_node_t ze_node;
109 dva_t ze_dva; /* dva associated with this extent */
110 uint64_t ze_nblks; /* number of blocks in extent */
111 } zvol_extent_t;
112
113 /*
114 * The in-core state of each volume.
115 */
116 typedef struct zvol_state {
117 char zv_name[MAXPATHLEN]; /* pool/dd name */
118 uint64_t zv_volsize; /* amount of space we advertise */
119 uint64_t zv_volblocksize; /* volume block size */
120 minor_t zv_minor; /* minor number */
121 uint8_t zv_min_bs; /* minimum addressable block shift */
122 uint8_t zv_flags; /* readonly, dumpified, etc. */
123 objset_t *zv_objset; /* objset handle */
124 uint32_t zv_open_count[OTYPCNT]; /* open counts */
125 uint32_t zv_total_opens; /* total open count */
126 zilog_t *zv_zilog; /* ZIL handle */
127 list_t zv_extents; /* List of extents for dump */
128 znode_t zv_znode; /* for range locking */
129 dmu_buf_t *zv_dbuf; /* bonus handle */
130 } zvol_state_t;
131
132 /*
133 * zvol specific flags
134 */
135 #define ZVOL_RDONLY 0x1
136 #define ZVOL_DUMPIFIED 0x2
137 #define ZVOL_EXCL 0x4
138 #define ZVOL_WCE 0x8
139
140 /*
141 * zvol maximum transfer in one DMU tx.
142 */
143 int zvol_maxphys = DMU_MAX_ACCESS/2;
144
145 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
146 nvlist_t *, nvlist_t *);
147 static int zvol_remove_zv(zvol_state_t *);
148 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio);
149 static int zvol_dumpify(zvol_state_t *zv);
150 static int zvol_dump_fini(zvol_state_t *zv);
151 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
152
153 static void
154 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
155 {
156 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
157
158 zv->zv_volsize = volsize;
159 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
160 "Size", volsize) == DDI_SUCCESS);
161 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
162 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
163
164 /* Notify specfs to invalidate the cached size */
165 spec_size_invalidate(dev, VBLK);
166 spec_size_invalidate(dev, VCHR);
167 }
168
169 int
170 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
171 {
172 if (volsize == 0)
173 return (SET_ERROR(EINVAL));
174
175 if (volsize % blocksize != 0)
176 return (SET_ERROR(EINVAL));
177
178 #ifdef _ILP32
179 if (volsize - 1 > SPEC_MAXOFFSET_T)
180 return (SET_ERROR(EOVERFLOW));
181 #endif
182 return (0);
183 }
184
185 int
186 zvol_check_volblocksize(uint64_t volblocksize)
187 {
188 if (volblocksize < SPA_MINBLOCKSIZE ||
189 volblocksize > SPA_MAXBLOCKSIZE ||
190 !ISP2(volblocksize))
191 return (SET_ERROR(EDOM));
192
193 return (0);
194 }
195
196 int
197 zvol_get_stats(objset_t *os, nvlist_t *nv)
198 {
199 int error;
200 dmu_object_info_t doi;
201 uint64_t val;
202
203 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
204 if (error)
205 return (error);
206
207 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
208
209 error = dmu_object_info(os, ZVOL_OBJ, &doi);
210
211 if (error == 0) {
212 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
213 doi.doi_data_block_size);
214 }
215
216 return (error);
217 }
218
219 static zvol_state_t *
220 zvol_minor_lookup(const char *name)
221 {
222 minor_t minor;
223 zvol_state_t *zv;
224
225 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
226
227 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
228 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
229 if (zv == NULL)
230 continue;
231 if (strcmp(zv->zv_name, name) == 0)
232 return (zv);
233 }
234
235 return (NULL);
236 }
237
238 /* extent mapping arg */
239 struct maparg {
240 zvol_state_t *ma_zv;
241 uint64_t ma_blks;
242 };
243
244 /*ARGSUSED*/
245 static int
246 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
247 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
248 {
249 struct maparg *ma = arg;
250 zvol_extent_t *ze;
251 int bs = ma->ma_zv->zv_volblocksize;
252
253 if (BP_IS_HOLE(bp) ||
254 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
255 return (0);
256
257 VERIFY(!BP_IS_EMBEDDED(bp));
258
259 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
260 ma->ma_blks++;
261
262 /* Abort immediately if we have encountered gang blocks */
263 if (BP_IS_GANG(bp))
264 return (SET_ERROR(EFRAGS));
265
266 /*
267 * See if the block is at the end of the previous extent.
268 */
269 ze = list_tail(&ma->ma_zv->zv_extents);
270 if (ze &&
271 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
272 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
273 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
274 ze->ze_nblks++;
275 return (0);
276 }
277
278 dprintf_bp(bp, "%s", "next blkptr:");
279
280 /* start a new extent */
281 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
282 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
283 ze->ze_nblks = 1;
284 list_insert_tail(&ma->ma_zv->zv_extents, ze);
285 return (0);
286 }
287
288 static void
289 zvol_free_extents(zvol_state_t *zv)
290 {
291 zvol_extent_t *ze;
292
293 while (ze = list_head(&zv->zv_extents)) {
294 list_remove(&zv->zv_extents, ze);
295 kmem_free(ze, sizeof (zvol_extent_t));
296 }
297 }
298
299 static int
300 zvol_get_lbas(zvol_state_t *zv)
301 {
302 objset_t *os = zv->zv_objset;
303 struct maparg ma;
304 int err;
305
306 ma.ma_zv = zv;
307 ma.ma_blks = 0;
308 zvol_free_extents(zv);
309
310 /* commit any in-flight changes before traversing the dataset */
311 txg_wait_synced(dmu_objset_pool(os), 0);
312 err = traverse_dataset(dmu_objset_ds(os), 0,
313 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
314 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
315 zvol_free_extents(zv);
316 return (err ? err : EIO);
317 }
318
319 return (0);
320 }
321
322 /* ARGSUSED */
323 void
324 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
325 {
326 zfs_creat_t *zct = arg;
327 nvlist_t *nvprops = zct->zct_props;
328 int error;
329 uint64_t volblocksize, volsize;
330
331 VERIFY(nvlist_lookup_uint64(nvprops,
332 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
333 if (nvlist_lookup_uint64(nvprops,
334 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
335 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
336
337 /*
338 * These properties must be removed from the list so the generic
339 * property setting step won't apply to them.
340 */
341 VERIFY(nvlist_remove_all(nvprops,
342 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
343 (void) nvlist_remove_all(nvprops,
344 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
345
346 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
347 DMU_OT_NONE, 0, tx);
348 ASSERT(error == 0);
349
350 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
351 DMU_OT_NONE, 0, tx);
352 ASSERT(error == 0);
353
354 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
355 ASSERT(error == 0);
356 }
357
358 /*
359 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
360 * implement DKIOCFREE/free-long-range.
361 */
362 static int
363 zvol_replay_truncate(zvol_state_t *zv, lr_truncate_t *lr, boolean_t byteswap)
364 {
365 uint64_t offset, length;
366
367 if (byteswap)
368 byteswap_uint64_array(lr, sizeof (*lr));
369
370 offset = lr->lr_offset;
371 length = lr->lr_length;
372
373 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
374 }
375
376 /*
377 * Replay a TX_WRITE ZIL transaction that didn't get committed
378 * after a system failure
379 */
380 static int
381 zvol_replay_write(zvol_state_t *zv, lr_write_t *lr, boolean_t byteswap)
382 {
383 objset_t *os = zv->zv_objset;
384 char *data = (char *)(lr + 1); /* data follows lr_write_t */
385 uint64_t offset, length;
386 dmu_tx_t *tx;
387 int error;
388
389 if (byteswap)
390 byteswap_uint64_array(lr, sizeof (*lr));
391
392 offset = lr->lr_offset;
393 length = lr->lr_length;
394
395 /* If it's a dmu_sync() block, write the whole block */
396 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
397 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
398 if (length < blocksize) {
399 offset -= offset % blocksize;
400 length = blocksize;
401 }
402 }
403
404 tx = dmu_tx_create(os);
405 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
406 error = dmu_tx_assign(tx, TXG_WAIT);
407 if (error) {
408 dmu_tx_abort(tx);
409 } else {
410 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
411 dmu_tx_commit(tx);
412 }
413
414 return (error);
415 }
416
417 /* ARGSUSED */
418 static int
419 zvol_replay_err(zvol_state_t *zv, lr_t *lr, boolean_t byteswap)
420 {
421 return (SET_ERROR(ENOTSUP));
422 }
423
424 /*
425 * Callback vectors for replaying records.
426 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
427 */
428 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
429 zvol_replay_err, /* 0 no such transaction type */
430 zvol_replay_err, /* TX_CREATE */
431 zvol_replay_err, /* TX_MKDIR */
432 zvol_replay_err, /* TX_MKXATTR */
433 zvol_replay_err, /* TX_SYMLINK */
434 zvol_replay_err, /* TX_REMOVE */
435 zvol_replay_err, /* TX_RMDIR */
436 zvol_replay_err, /* TX_LINK */
437 zvol_replay_err, /* TX_RENAME */
438 zvol_replay_write, /* TX_WRITE */
439 zvol_replay_truncate, /* TX_TRUNCATE */
440 zvol_replay_err, /* TX_SETATTR */
441 zvol_replay_err, /* TX_ACL */
442 zvol_replay_err, /* TX_CREATE_ACL */
443 zvol_replay_err, /* TX_CREATE_ATTR */
444 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
445 zvol_replay_err, /* TX_MKDIR_ACL */
446 zvol_replay_err, /* TX_MKDIR_ATTR */
447 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
448 zvol_replay_err, /* TX_WRITE2 */
449 };
450
451 int
452 zvol_name2minor(const char *name, minor_t *minor)
453 {
454 zvol_state_t *zv;
455
456 mutex_enter(&zfsdev_state_lock);
457 zv = zvol_minor_lookup(name);
458 if (minor && zv)
459 *minor = zv->zv_minor;
460 mutex_exit(&zfsdev_state_lock);
461 return (zv ? 0 : -1);
462 }
463
464 /*
465 * Create a minor node (plus a whole lot more) for the specified volume.
466 */
467 int
468 zvol_create_minor(const char *name)
469 {
470 zfs_soft_state_t *zs;
471 zvol_state_t *zv;
472 objset_t *os;
473 dmu_object_info_t doi;
474 minor_t minor = 0;
475 char chrbuf[30], blkbuf[30];
476 int error;
477
478 mutex_enter(&zfsdev_state_lock);
479
480 if (zvol_minor_lookup(name) != NULL) {
481 mutex_exit(&zfsdev_state_lock);
482 return (SET_ERROR(EEXIST));
483 }
484
485 /* lie and say we're read-only */
486 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, FTAG, &os);
487
488 if (error) {
489 mutex_exit(&zfsdev_state_lock);
490 return (error);
491 }
492
493 if ((minor = zfsdev_minor_alloc()) == 0) {
494 dmu_objset_disown(os, FTAG);
495 mutex_exit(&zfsdev_state_lock);
496 return (SET_ERROR(ENXIO));
497 }
498
499 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
500 dmu_objset_disown(os, FTAG);
501 mutex_exit(&zfsdev_state_lock);
502 return (SET_ERROR(EAGAIN));
503 }
504 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
505 (char *)name);
506
507 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
508
509 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
510 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
511 ddi_soft_state_free(zfsdev_state, minor);
512 dmu_objset_disown(os, FTAG);
513 mutex_exit(&zfsdev_state_lock);
514 return (SET_ERROR(EAGAIN));
515 }
516
517 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
518
519 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
520 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
521 ddi_remove_minor_node(zfs_dip, chrbuf);
522 ddi_soft_state_free(zfsdev_state, minor);
523 dmu_objset_disown(os, FTAG);
524 mutex_exit(&zfsdev_state_lock);
525 return (SET_ERROR(EAGAIN));
526 }
527
528 zs = ddi_get_soft_state(zfsdev_state, minor);
529 zs->zss_type = ZSST_ZVOL;
530 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
531 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
532 zv->zv_min_bs = DEV_BSHIFT;
533 zv->zv_minor = minor;
534 zv->zv_objset = os;
535 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
536 zv->zv_flags |= ZVOL_RDONLY;
537 mutex_init(&zv->zv_znode.z_range_lock, NULL, MUTEX_DEFAULT, NULL);
538 avl_create(&zv->zv_znode.z_range_avl, zfs_range_compare,
539 sizeof (rl_t), offsetof(rl_t, r_node));
540 list_create(&zv->zv_extents, sizeof (zvol_extent_t),
541 offsetof(zvol_extent_t, ze_node));
542 /* get and cache the blocksize */
543 error = dmu_object_info(os, ZVOL_OBJ, &doi);
544 ASSERT(error == 0);
545 zv->zv_volblocksize = doi.doi_data_block_size;
546
547 if (spa_writeable(dmu_objset_spa(os))) {
548 if (zil_replay_disable)
549 zil_destroy(dmu_objset_zil(os), B_FALSE);
550 else
551 zil_replay(os, zv, zvol_replay_vector);
552 }
553 dmu_objset_disown(os, FTAG);
554 zv->zv_objset = NULL;
555
556 zvol_minors++;
557
558 mutex_exit(&zfsdev_state_lock);
559
560 return (0);
561 }
562
563 /*
564 * Remove minor node for the specified volume.
565 */
566 static int
567 zvol_remove_zv(zvol_state_t *zv)
568 {
569 char nmbuf[20];
570 minor_t minor = zv->zv_minor;
571
572 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
573 if (zv->zv_total_opens != 0)
574 return (SET_ERROR(EBUSY));
575
576 (void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
577 ddi_remove_minor_node(zfs_dip, nmbuf);
578
579 (void) snprintf(nmbuf, sizeof (nmbuf), "%u", minor);
580 ddi_remove_minor_node(zfs_dip, nmbuf);
581
582 avl_destroy(&zv->zv_znode.z_range_avl);
583 mutex_destroy(&zv->zv_znode.z_range_lock);
584
585 kmem_free(zv, sizeof (zvol_state_t));
586
587 ddi_soft_state_free(zfsdev_state, minor);
588
589 zvol_minors--;
590 return (0);
591 }
592
593 int
594 zvol_remove_minor(const char *name)
595 {
596 zvol_state_t *zv;
597 int rc;
598
599 mutex_enter(&zfsdev_state_lock);
600 if ((zv = zvol_minor_lookup(name)) == NULL) {
601 mutex_exit(&zfsdev_state_lock);
602 return (SET_ERROR(ENXIO));
603 }
604 rc = zvol_remove_zv(zv);
605 mutex_exit(&zfsdev_state_lock);
606 return (rc);
607 }
608
609 int
610 zvol_first_open(zvol_state_t *zv)
611 {
612 objset_t *os;
613 uint64_t volsize;
614 int error;
615 uint64_t readonly;
616
617 /* lie and say we're read-only */
618 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, B_TRUE,
619 zvol_tag, &os);
620 if (error)
621 return (error);
622
623 zv->zv_objset = os;
624 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
625 if (error) {
626 ASSERT(error == 0);
627 dmu_objset_disown(os, zvol_tag);
628 return (error);
629 }
630
631 error = dmu_bonus_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dbuf);
632 if (error) {
633 dmu_objset_disown(os, zvol_tag);
634 return (error);
635 }
636
637 zvol_size_changed(zv, volsize);
638 zv->zv_zilog = zil_open(os, zvol_get_data);
639
640 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
641 NULL) == 0);
642 if (readonly || dmu_objset_is_snapshot(os) ||
643 !spa_writeable(dmu_objset_spa(os)))
644 zv->zv_flags |= ZVOL_RDONLY;
645 else
646 zv->zv_flags &= ~ZVOL_RDONLY;
647 return (error);
648 }
649
650 void
651 zvol_last_close(zvol_state_t *zv)
652 {
653 zil_close(zv->zv_zilog);
654 zv->zv_zilog = NULL;
655
656 dmu_buf_rele(zv->zv_dbuf, zvol_tag);
657 zv->zv_dbuf = NULL;
658
659 /*
660 * Evict cached data
661 */
662 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
663 !(zv->zv_flags & ZVOL_RDONLY))
664 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
665 dmu_objset_evict_dbufs(zv->zv_objset);
666
667 dmu_objset_disown(zv->zv_objset, zvol_tag);
668 zv->zv_objset = NULL;
669 }
670
671 int
672 zvol_prealloc(zvol_state_t *zv)
673 {
674 objset_t *os = zv->zv_objset;
675 dmu_tx_t *tx;
676 uint64_t refd, avail, usedobjs, availobjs;
677 uint64_t resid = zv->zv_volsize;
678 uint64_t off = 0;
679
680 /* Check the space usage before attempting to allocate the space */
681 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
682 if (avail < zv->zv_volsize)
683 return (SET_ERROR(ENOSPC));
684
685 /* Free old extents if they exist */
686 zvol_free_extents(zv);
687
688 while (resid != 0) {
689 int error;
690 uint64_t bytes = MIN(resid, SPA_MAXBLOCKSIZE);
691
692 tx = dmu_tx_create(os);
693 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
694 error = dmu_tx_assign(tx, TXG_WAIT);
695 if (error) {
696 dmu_tx_abort(tx);
697 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
698 return (error);
699 }
700 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
701 dmu_tx_commit(tx);
702 off += bytes;
703 resid -= bytes;
704 }
705 txg_wait_synced(dmu_objset_pool(os), 0);
706
707 return (0);
708 }
709
710 static int
711 zvol_update_volsize(objset_t *os, uint64_t volsize)
712 {
713 dmu_tx_t *tx;
714 int error;
715
716 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
717
718 tx = dmu_tx_create(os);
719 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
720 error = dmu_tx_assign(tx, TXG_WAIT);
721 if (error) {
722 dmu_tx_abort(tx);
723 return (error);
724 }
725
726 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
727 &volsize, tx);
728 dmu_tx_commit(tx);
729
730 if (error == 0)
731 error = dmu_free_long_range(os,
732 ZVOL_OBJ, volsize, DMU_OBJECT_END);
733 return (error);
734 }
735
736 void
737 zvol_remove_minors(const char *name)
738 {
739 zvol_state_t *zv;
740 char *namebuf;
741 minor_t minor;
742
743 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
744 (void) strncpy(namebuf, name, strlen(name));
745 (void) strcat(namebuf, "/");
746 mutex_enter(&zfsdev_state_lock);
747 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
748
749 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
750 if (zv == NULL)
751 continue;
752 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
753 (void) zvol_remove_zv(zv);
754 }
755 kmem_free(namebuf, strlen(name) + 2);
756
757 mutex_exit(&zfsdev_state_lock);
758 }
759
760 static int
761 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
762 {
763 uint64_t old_volsize = 0ULL;
764 int error = 0;
765
766 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
767
768 /*
769 * Reinitialize the dump area to the new size. If we
770 * failed to resize the dump area then restore it back to
771 * its original size. We must set the new volsize prior
772 * to calling dumpvp_resize() to ensure that the devices'
773 * size(9P) is not visible by the dump subsystem.
774 */
775 old_volsize = zv->zv_volsize;
776 zvol_size_changed(zv, volsize);
777
778 if (zv->zv_flags & ZVOL_DUMPIFIED) {
779 if ((error = zvol_dumpify(zv)) != 0 ||
780 (error = dumpvp_resize()) != 0) {
781 int dumpify_error;
782
783 (void) zvol_update_volsize(zv->zv_objset, old_volsize);
784 zvol_size_changed(zv, old_volsize);
785 dumpify_error = zvol_dumpify(zv);
786 error = dumpify_error ? dumpify_error : error;
787 }
788 }
789
790 /*
791 * Generate a LUN expansion event.
792 */
793 if (error == 0) {
794 sysevent_id_t eid;
795 nvlist_t *attr;
796 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
797
798 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
799 zv->zv_minor);
800
801 VERIFY(nvlist_alloc(&attr, NV_UNIQUE_NAME, KM_SLEEP) == 0);
802 VERIFY(nvlist_add_string(attr, DEV_PHYS_PATH, physpath) == 0);
803
804 (void) ddi_log_sysevent(zfs_dip, SUNW_VENDOR, EC_DEV_STATUS,
805 ESC_DEV_DLE, attr, &eid, DDI_SLEEP);
806
807 nvlist_free(attr);
808 kmem_free(physpath, MAXPATHLEN);
809 }
810 return (error);
811 }
812
813 int
814 zvol_set_volsize(const char *name, uint64_t volsize)
815 {
816 zvol_state_t *zv = NULL;
817 objset_t *os;
818 int error;
819 dmu_object_info_t doi;
820 uint64_t readonly;
821 boolean_t owned = B_FALSE;
822
823 error = dsl_prop_get_integer(name,
824 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
825 if (error != 0)
826 return (error);
827 if (readonly)
828 return (SET_ERROR(EROFS));
829
830 mutex_enter(&zfsdev_state_lock);
831 zv = zvol_minor_lookup(name);
832
833 if (zv == NULL || zv->zv_objset == NULL) {
834 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE,
835 FTAG, &os)) != 0) {
836 mutex_exit(&zfsdev_state_lock);
837 return (error);
838 }
839 owned = B_TRUE;
840 if (zv != NULL)
841 zv->zv_objset = os;
842 } else {
843 os = zv->zv_objset;
844 }
845
846 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
847 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
848 goto out;
849
850 error = zvol_update_volsize(os, volsize);
851
852 if (error == 0 && zv != NULL)
853 error = zvol_update_live_volsize(zv, volsize);
854 out:
855 if (owned) {
856 dmu_objset_disown(os, FTAG);
857 if (zv != NULL)
858 zv->zv_objset = NULL;
859 }
860 mutex_exit(&zfsdev_state_lock);
861 return (error);
862 }
863
864 /*ARGSUSED*/
865 int
866 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
867 {
868 zvol_state_t *zv;
869 int err = 0;
870
871 mutex_enter(&zfsdev_state_lock);
872
873 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
874 if (zv == NULL) {
875 mutex_exit(&zfsdev_state_lock);
876 return (SET_ERROR(ENXIO));
877 }
878
879 if (zv->zv_total_opens == 0)
880 err = zvol_first_open(zv);
881 if (err) {
882 mutex_exit(&zfsdev_state_lock);
883 return (err);
884 }
885 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
886 err = SET_ERROR(EROFS);
887 goto out;
888 }
889 if (zv->zv_flags & ZVOL_EXCL) {
890 err = SET_ERROR(EBUSY);
891 goto out;
892 }
893 if (flag & FEXCL) {
894 if (zv->zv_total_opens != 0) {
895 err = SET_ERROR(EBUSY);
896 goto out;
897 }
898 zv->zv_flags |= ZVOL_EXCL;
899 }
900
901 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
902 zv->zv_open_count[otyp]++;
903 zv->zv_total_opens++;
904 }
905 mutex_exit(&zfsdev_state_lock);
906
907 return (err);
908 out:
909 if (zv->zv_total_opens == 0)
910 zvol_last_close(zv);
911 mutex_exit(&zfsdev_state_lock);
912 return (err);
913 }
914
915 /*ARGSUSED*/
916 int
917 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
918 {
919 minor_t minor = getminor(dev);
920 zvol_state_t *zv;
921 int error = 0;
922
923 mutex_enter(&zfsdev_state_lock);
924
925 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
926 if (zv == NULL) {
927 mutex_exit(&zfsdev_state_lock);
928 return (SET_ERROR(ENXIO));
929 }
930
931 if (zv->zv_flags & ZVOL_EXCL) {
932 ASSERT(zv->zv_total_opens == 1);
933 zv->zv_flags &= ~ZVOL_EXCL;
934 }
935
936 /*
937 * If the open count is zero, this is a spurious close.
938 * That indicates a bug in the kernel / DDI framework.
939 */
940 ASSERT(zv->zv_open_count[otyp] != 0);
941 ASSERT(zv->zv_total_opens != 0);
942
943 /*
944 * You may get multiple opens, but only one close.
945 */
946 zv->zv_open_count[otyp]--;
947 zv->zv_total_opens--;
948
949 if (zv->zv_total_opens == 0)
950 zvol_last_close(zv);
951
952 mutex_exit(&zfsdev_state_lock);
953 return (error);
954 }
955
956 static void
957 zvol_get_done(zgd_t *zgd, int error)
958 {
959 if (zgd->zgd_db)
960 dmu_buf_rele(zgd->zgd_db, zgd);
961
962 zfs_range_unlock(zgd->zgd_rl);
963
964 if (error == 0 && zgd->zgd_bp)
965 zil_add_block(zgd->zgd_zilog, zgd->zgd_bp);
966
967 kmem_free(zgd, sizeof (zgd_t));
968 }
969
970 /*
971 * Get data to generate a TX_WRITE intent log record.
972 */
973 static int
974 zvol_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio)
975 {
976 zvol_state_t *zv = arg;
977 objset_t *os = zv->zv_objset;
978 uint64_t object = ZVOL_OBJ;
979 uint64_t offset = lr->lr_offset;
980 uint64_t size = lr->lr_length; /* length of user data */
981 blkptr_t *bp = &lr->lr_blkptr;
982 dmu_buf_t *db;
983 zgd_t *zgd;
984 int error;
985
986 ASSERT(zio != NULL);
987 ASSERT(size != 0);
988
989 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
990 zgd->zgd_zilog = zv->zv_zilog;
991 zgd->zgd_rl = zfs_range_lock(&zv->zv_znode, offset, size, RL_READER);
992
993 /*
994 * Write records come in two flavors: immediate and indirect.
995 * For small writes it's cheaper to store the data with the
996 * log record (immediate); for large writes it's cheaper to
997 * sync the data and get a pointer to it (indirect) so that
998 * we don't have to write the data twice.
999 */
1000 if (buf != NULL) { /* immediate write */
1001 error = dmu_read(os, object, offset, size, buf,
1002 DMU_READ_NO_PREFETCH);
1003 } else {
1004 size = zv->zv_volblocksize;
1005 offset = P2ALIGN(offset, size);
1006 error = dmu_buf_hold(os, object, offset, zgd, &db,
1007 DMU_READ_NO_PREFETCH);
1008 if (error == 0) {
1009 blkptr_t *obp = dmu_buf_get_blkptr(db);
1010 if (obp) {
1011 ASSERT(BP_IS_HOLE(bp));
1012 *bp = *obp;
1013 }
1014
1015 zgd->zgd_db = db;
1016 zgd->zgd_bp = bp;
1017
1018 ASSERT(db->db_offset == offset);
1019 ASSERT(db->db_size == size);
1020
1021 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1022 zvol_get_done, zgd);
1023
1024 if (error == 0)
1025 return (0);
1026 }
1027 }
1028
1029 zvol_get_done(zgd, error);
1030
1031 return (error);
1032 }
1033
1034 /*
1035 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1036 *
1037 * We store data in the log buffers if it's small enough.
1038 * Otherwise we will later flush the data out via dmu_sync().
1039 */
1040 ssize_t zvol_immediate_write_sz = 32768;
1041
1042 static void
1043 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1044 boolean_t sync)
1045 {
1046 uint32_t blocksize = zv->zv_volblocksize;
1047 zilog_t *zilog = zv->zv_zilog;
1048 boolean_t slogging;
1049 ssize_t immediate_write_sz;
1050
1051 if (zil_replaying(zilog, tx))
1052 return;
1053
1054 immediate_write_sz = (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1055 ? 0 : zvol_immediate_write_sz;
1056
1057 slogging = spa_has_slogs(zilog->zl_spa) &&
1058 (zilog->zl_logbias == ZFS_LOGBIAS_LATENCY);
1059
1060 while (resid) {
1061 itx_t *itx;
1062 lr_write_t *lr;
1063 ssize_t len;
1064 itx_wr_state_t write_state;
1065
1066 /*
1067 * Unlike zfs_log_write() we can be called with
1068 * upto DMU_MAX_ACCESS/2 (5MB) writes.
1069 */
1070 if (blocksize > immediate_write_sz && !slogging &&
1071 resid >= blocksize && off % blocksize == 0) {
1072 write_state = WR_INDIRECT; /* uses dmu_sync */
1073 len = blocksize;
1074 } else if (sync) {
1075 write_state = WR_COPIED;
1076 len = MIN(ZIL_MAX_LOG_DATA, resid);
1077 } else {
1078 write_state = WR_NEED_COPY;
1079 len = MIN(ZIL_MAX_LOG_DATA, resid);
1080 }
1081
1082 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1083 (write_state == WR_COPIED ? len : 0));
1084 lr = (lr_write_t *)&itx->itx_lr;
1085 if (write_state == WR_COPIED && dmu_read(zv->zv_objset,
1086 ZVOL_OBJ, off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1087 zil_itx_destroy(itx);
1088 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1089 lr = (lr_write_t *)&itx->itx_lr;
1090 write_state = WR_NEED_COPY;
1091 }
1092
1093 itx->itx_wr_state = write_state;
1094 if (write_state == WR_NEED_COPY)
1095 itx->itx_sod += len;
1096 lr->lr_foid = ZVOL_OBJ;
1097 lr->lr_offset = off;
1098 lr->lr_length = len;
1099 lr->lr_blkoff = 0;
1100 BP_ZERO(&lr->lr_blkptr);
1101
1102 itx->itx_private = zv;
1103 itx->itx_sync = sync;
1104
1105 zil_itx_assign(zilog, itx, tx);
1106
1107 off += len;
1108 resid -= len;
1109 }
1110 }
1111
1112 static int
1113 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1114 uint64_t size, boolean_t doread, boolean_t isdump)
1115 {
1116 vdev_disk_t *dvd;
1117 int c;
1118 int numerrors = 0;
1119
1120 if (vd->vdev_ops == &vdev_mirror_ops ||
1121 vd->vdev_ops == &vdev_replacing_ops ||
1122 vd->vdev_ops == &vdev_spare_ops) {
1123 for (c = 0; c < vd->vdev_children; c++) {
1124 int err = zvol_dumpio_vdev(vd->vdev_child[c],
1125 addr, offset, origoffset, size, doread, isdump);
1126 if (err != 0) {
1127 numerrors++;
1128 } else if (doread) {
1129 break;
1130 }
1131 }
1132 }
1133
1134 if (!vd->vdev_ops->vdev_op_leaf && vd->vdev_ops != &vdev_raidz_ops)
1135 return (numerrors < vd->vdev_children ? 0 : EIO);
1136
1137 if (doread && !vdev_readable(vd))
1138 return (SET_ERROR(EIO));
1139 else if (!doread && !vdev_writeable(vd))
1140 return (SET_ERROR(EIO));
1141
1142 if (vd->vdev_ops == &vdev_raidz_ops) {
1143 return (vdev_raidz_physio(vd,
1144 addr, size, offset, origoffset, doread, isdump));
1145 }
1146
1147 offset += VDEV_LABEL_START_SIZE;
1148
1149 if (ddi_in_panic() || isdump) {
1150 ASSERT(!doread);
1151 if (doread)
1152 return (SET_ERROR(EIO));
1153 dvd = vd->vdev_tsd;
1154 ASSERT3P(dvd, !=, NULL);
1155 return (ldi_dump(dvd->vd_lh, addr, lbtodb(offset),
1156 lbtodb(size)));
1157 } else {
1158 dvd = vd->vdev_tsd;
1159 ASSERT3P(dvd, !=, NULL);
1160 return (vdev_disk_ldi_physio(dvd->vd_lh, addr, size,
1161 offset, doread ? B_READ : B_WRITE));
1162 }
1163 }
1164
1165 static int
1166 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1167 boolean_t doread, boolean_t isdump)
1168 {
1169 vdev_t *vd;
1170 int error;
1171 zvol_extent_t *ze;
1172 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1173
1174 /* Must be sector aligned, and not stradle a block boundary. */
1175 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1176 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1177 return (SET_ERROR(EINVAL));
1178 }
1179 ASSERT(size <= zv->zv_volblocksize);
1180
1181 /* Locate the extent this belongs to */
1182 ze = list_head(&zv->zv_extents);
1183 while (offset >= ze->ze_nblks * zv->zv_volblocksize) {
1184 offset -= ze->ze_nblks * zv->zv_volblocksize;
1185 ze = list_next(&zv->zv_extents, ze);
1186 }
1187
1188 if (ze == NULL)
1189 return (SET_ERROR(EINVAL));
1190
1191 if (!ddi_in_panic())
1192 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1193
1194 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1195 offset += DVA_GET_OFFSET(&ze->ze_dva);
1196 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1197 size, doread, isdump);
1198
1199 if (!ddi_in_panic())
1200 spa_config_exit(spa, SCL_STATE, FTAG);
1201
1202 return (error);
1203 }
1204
1205 int
1206 zvol_strategy(buf_t *bp)
1207 {
1208 zfs_soft_state_t *zs = NULL;
1209 zvol_state_t *zv;
1210 uint64_t off, volsize;
1211 size_t resid;
1212 char *addr;
1213 objset_t *os;
1214 rl_t *rl;
1215 int error = 0;
1216 boolean_t doread = bp->b_flags & B_READ;
1217 boolean_t is_dumpified;
1218 boolean_t sync;
1219
1220 if (getminor(bp->b_edev) == 0) {
1221 error = SET_ERROR(EINVAL);
1222 } else {
1223 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1224 if (zs == NULL)
1225 error = SET_ERROR(ENXIO);
1226 else if (zs->zss_type != ZSST_ZVOL)
1227 error = SET_ERROR(EINVAL);
1228 }
1229
1230 if (error) {
1231 bioerror(bp, error);
1232 biodone(bp);
1233 return (0);
1234 }
1235
1236 zv = zs->zss_data;
1237
1238 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1239 bioerror(bp, EROFS);
1240 biodone(bp);
1241 return (0);
1242 }
1243
1244 off = ldbtob(bp->b_blkno);
1245 volsize = zv->zv_volsize;
1246
1247 os = zv->zv_objset;
1248 ASSERT(os != NULL);
1249
1250 bp_mapin(bp);
1251 addr = bp->b_un.b_addr;
1252 resid = bp->b_bcount;
1253
1254 if (resid > 0 && (off < 0 || off >= volsize)) {
1255 bioerror(bp, EIO);
1256 biodone(bp);
1257 return (0);
1258 }
1259
1260 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1261 sync = ((!(bp->b_flags & B_ASYNC) &&
1262 !(zv->zv_flags & ZVOL_WCE)) ||
1263 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1264 !doread && !is_dumpified;
1265
1266 /*
1267 * There must be no buffer changes when doing a dmu_sync() because
1268 * we can't change the data whilst calculating the checksum.
1269 */
1270 rl = zfs_range_lock(&zv->zv_znode, off, resid,
1271 doread ? RL_READER : RL_WRITER);
1272
1273 while (resid != 0 && off < volsize) {
1274 size_t size = MIN(resid, zvol_maxphys);
1275 if (is_dumpified) {
1276 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1277 error = zvol_dumpio(zv, addr, off, size,
1278 doread, B_FALSE);
1279 } else if (doread) {
1280 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1281 DMU_READ_PREFETCH);
1282 } else {
1283 dmu_tx_t *tx = dmu_tx_create(os);
1284 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1285 error = dmu_tx_assign(tx, TXG_WAIT);
1286 if (error) {
1287 dmu_tx_abort(tx);
1288 } else {
1289 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1290 zvol_log_write(zv, tx, off, size, sync);
1291 dmu_tx_commit(tx);
1292 }
1293 }
1294 if (error) {
1295 /* convert checksum errors into IO errors */
1296 if (error == ECKSUM)
1297 error = SET_ERROR(EIO);
1298 break;
1299 }
1300 off += size;
1301 addr += size;
1302 resid -= size;
1303 }
1304 zfs_range_unlock(rl);
1305
1306 if ((bp->b_resid = resid) == bp->b_bcount)
1307 bioerror(bp, off > volsize ? EINVAL : error);
1308
1309 if (sync)
1310 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1311 biodone(bp);
1312
1313 return (0);
1314 }
1315
1316 /*
1317 * Set the buffer count to the zvol maximum transfer.
1318 * Using our own routine instead of the default minphys()
1319 * means that for larger writes we write bigger buffers on X86
1320 * (128K instead of 56K) and flush the disk write cache less often
1321 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1322 * 56K on X86 and 128K on sparc).
1323 */
1324 void
1325 zvol_minphys(struct buf *bp)
1326 {
1327 if (bp->b_bcount > zvol_maxphys)
1328 bp->b_bcount = zvol_maxphys;
1329 }
1330
1331 int
1332 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1333 {
1334 minor_t minor = getminor(dev);
1335 zvol_state_t *zv;
1336 int error = 0;
1337 uint64_t size;
1338 uint64_t boff;
1339 uint64_t resid;
1340
1341 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1342 if (zv == NULL)
1343 return (SET_ERROR(ENXIO));
1344
1345 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1346 return (SET_ERROR(EINVAL));
1347
1348 boff = ldbtob(blkno);
1349 resid = ldbtob(nblocks);
1350
1351 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1352
1353 while (resid) {
1354 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1355 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1356 if (error)
1357 break;
1358 boff += size;
1359 addr += size;
1360 resid -= size;
1361 }
1362
1363 return (error);
1364 }
1365
1366 /*ARGSUSED*/
1367 int
1368 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1369 {
1370 minor_t minor = getminor(dev);
1371 zvol_state_t *zv;
1372 uint64_t volsize;
1373 rl_t *rl;
1374 int error = 0;
1375
1376 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1377 if (zv == NULL)
1378 return (SET_ERROR(ENXIO));
1379
1380 volsize = zv->zv_volsize;
1381 if (uio->uio_resid > 0 &&
1382 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1383 return (SET_ERROR(EIO));
1384
1385 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1386 error = physio(zvol_strategy, NULL, dev, B_READ,
1387 zvol_minphys, uio);
1388 return (error);
1389 }
1390
1391 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1392 RL_READER);
1393 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1394 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1395
1396 /* don't read past the end */
1397 if (bytes > volsize - uio->uio_loffset)
1398 bytes = volsize - uio->uio_loffset;
1399
1400 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1401 if (error) {
1402 /* convert checksum errors into IO errors */
1403 if (error == ECKSUM)
1404 error = SET_ERROR(EIO);
1405 break;
1406 }
1407 }
1408 zfs_range_unlock(rl);
1409 return (error);
1410 }
1411
1412 /*ARGSUSED*/
1413 int
1414 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1415 {
1416 minor_t minor = getminor(dev);
1417 zvol_state_t *zv;
1418 uint64_t volsize;
1419 rl_t *rl;
1420 int error = 0;
1421 boolean_t sync;
1422
1423 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1424 if (zv == NULL)
1425 return (SET_ERROR(ENXIO));
1426
1427 volsize = zv->zv_volsize;
1428 if (uio->uio_resid > 0 &&
1429 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1430 return (SET_ERROR(EIO));
1431
1432 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1433 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1434 zvol_minphys, uio);
1435 return (error);
1436 }
1437
1438 sync = !(zv->zv_flags & ZVOL_WCE) ||
1439 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1440
1441 rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
1442 RL_WRITER);
1443 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1444 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1445 uint64_t off = uio->uio_loffset;
1446 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1447
1448 if (bytes > volsize - off) /* don't write past the end */
1449 bytes = volsize - off;
1450
1451 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
1452 error = dmu_tx_assign(tx, TXG_WAIT);
1453 if (error) {
1454 dmu_tx_abort(tx);
1455 break;
1456 }
1457 error = dmu_write_uio_dbuf(zv->zv_dbuf, uio, bytes, tx);
1458 if (error == 0)
1459 zvol_log_write(zv, tx, off, bytes, sync);
1460 dmu_tx_commit(tx);
1461
1462 if (error)
1463 break;
1464 }
1465 zfs_range_unlock(rl);
1466 if (sync)
1467 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1468 return (error);
1469 }
1470
1471 int
1472 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1473 {
1474 struct uuid uuid = EFI_RESERVED;
1475 efi_gpe_t gpe = { 0 };
1476 uint32_t crc;
1477 dk_efi_t efi;
1478 int length;
1479 char *ptr;
1480
1481 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1482 return (SET_ERROR(EFAULT));
1483 ptr = (char *)(uintptr_t)efi.dki_data_64;
1484 length = efi.dki_length;
1485 /*
1486 * Some clients may attempt to request a PMBR for the
1487 * zvol. Currently this interface will return EINVAL to
1488 * such requests. These requests could be supported by
1489 * adding a check for lba == 0 and consing up an appropriate
1490 * PMBR.
1491 */
1492 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1493 return (SET_ERROR(EINVAL));
1494
1495 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1496 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1497 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1498
1499 if (efi.dki_lba == 1) {
1500 efi_gpt_t gpt = { 0 };
1501
1502 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1503 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1504 gpt.efi_gpt_HeaderSize = LE_32(sizeof (gpt));
1505 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1506 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1507 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1508 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1509 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1510 gpt.efi_gpt_SizeOfPartitionEntry =
1511 LE_32(sizeof (efi_gpe_t));
1512 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1513 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1514 CRC32(crc, &gpt, sizeof (gpt), -1U, crc32_table);
1515 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1516 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1517 flag))
1518 return (SET_ERROR(EFAULT));
1519 ptr += sizeof (gpt);
1520 length -= sizeof (gpt);
1521 }
1522 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1523 length), flag))
1524 return (SET_ERROR(EFAULT));
1525 return (0);
1526 }
1527
1528 /*
1529 * BEGIN entry points to allow external callers access to the volume.
1530 */
1531 /*
1532 * Return the volume parameters needed for access from an external caller.
1533 * These values are invariant as long as the volume is held open.
1534 */
1535 int
1536 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1537 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1538 void **rl_hdl, void **bonus_hdl)
1539 {
1540 zvol_state_t *zv;
1541
1542 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1543 if (zv == NULL)
1544 return (SET_ERROR(ENXIO));
1545 if (zv->zv_flags & ZVOL_DUMPIFIED)
1546 return (SET_ERROR(ENXIO));
1547
1548 ASSERT(blksize && max_xfer_len && minor_hdl &&
1549 objset_hdl && zil_hdl && rl_hdl && bonus_hdl);
1550
1551 *blksize = zv->zv_volblocksize;
1552 *max_xfer_len = (uint64_t)zvol_maxphys;
1553 *minor_hdl = zv;
1554 *objset_hdl = zv->zv_objset;
1555 *zil_hdl = zv->zv_zilog;
1556 *rl_hdl = &zv->zv_znode;
1557 *bonus_hdl = zv->zv_dbuf;
1558 return (0);
1559 }
1560
1561 /*
1562 * Return the current volume size to an external caller.
1563 * The size can change while the volume is open.
1564 */
1565 uint64_t
1566 zvol_get_volume_size(void *minor_hdl)
1567 {
1568 zvol_state_t *zv = minor_hdl;
1569
1570 return (zv->zv_volsize);
1571 }
1572
1573 /*
1574 * Return the current WCE setting to an external caller.
1575 * The WCE setting can change while the volume is open.
1576 */
1577 int
1578 zvol_get_volume_wce(void *minor_hdl)
1579 {
1580 zvol_state_t *zv = minor_hdl;
1581
1582 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1583 }
1584
1585 /*
1586 * Entry point for external callers to zvol_log_write
1587 */
1588 void
1589 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1590 boolean_t sync)
1591 {
1592 zvol_state_t *zv = minor_hdl;
1593
1594 zvol_log_write(zv, tx, off, resid, sync);
1595 }
1596 /*
1597 * END entry points to allow external callers access to the volume.
1598 */
1599
1600 /*
1601 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1602 */
1603 static void
1604 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1605 boolean_t sync)
1606 {
1607 itx_t *itx;
1608 lr_truncate_t *lr;
1609 zilog_t *zilog = zv->zv_zilog;
1610
1611 if (zil_replaying(zilog, tx))
1612 return;
1613
1614 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1615 lr = (lr_truncate_t *)&itx->itx_lr;
1616 lr->lr_foid = ZVOL_OBJ;
1617 lr->lr_offset = off;
1618 lr->lr_length = len;
1619
1620 itx->itx_sync = sync;
1621 zil_itx_assign(zilog, itx, tx);
1622 }
1623
1624 /*
1625 * Dirtbag ioctls to support mkfs(1M) for UFS filesystems. See dkio(7I).
1626 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1627 */
1628 /*ARGSUSED*/
1629 int
1630 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1631 {
1632 zvol_state_t *zv;
1633 struct dk_callback *dkc;
1634 int error = 0;
1635 rl_t *rl;
1636
1637 mutex_enter(&zfsdev_state_lock);
1638
1639 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1640
1641 if (zv == NULL) {
1642 mutex_exit(&zfsdev_state_lock);
1643 return (SET_ERROR(ENXIO));
1644 }
1645 ASSERT(zv->zv_total_opens > 0);
1646
1647 switch (cmd) {
1648
1649 case DKIOCINFO:
1650 {
1651 struct dk_cinfo dki;
1652
1653 bzero(&dki, sizeof (dki));
1654 (void) strcpy(dki.dki_cname, "zvol");
1655 (void) strcpy(dki.dki_dname, "zvol");
1656 dki.dki_ctype = DKC_UNKNOWN;
1657 dki.dki_unit = getminor(dev);
1658 dki.dki_maxtransfer = 1 << (SPA_MAXBLOCKSHIFT - zv->zv_min_bs);
1659 mutex_exit(&zfsdev_state_lock);
1660 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1661 error = SET_ERROR(EFAULT);
1662 return (error);
1663 }
1664
1665 case DKIOCGMEDIAINFO:
1666 {
1667 struct dk_minfo dkm;
1668
1669 bzero(&dkm, sizeof (dkm));
1670 dkm.dki_lbsize = 1U << zv->zv_min_bs;
1671 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1672 dkm.dki_media_type = DK_UNKNOWN;
1673 mutex_exit(&zfsdev_state_lock);
1674 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1675 error = SET_ERROR(EFAULT);
1676 return (error);
1677 }
1678
1679 case DKIOCGMEDIAINFOEXT:
1680 {
1681 struct dk_minfo_ext dkmext;
1682
1683 bzero(&dkmext, sizeof (dkmext));
1684 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
1685 dkmext.dki_pbsize = zv->zv_volblocksize;
1686 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1687 dkmext.dki_media_type = DK_UNKNOWN;
1688 mutex_exit(&zfsdev_state_lock);
1689 if (ddi_copyout(&dkmext, (void *)arg, sizeof (dkmext), flag))
1690 error = SET_ERROR(EFAULT);
1691 return (error);
1692 }
1693
1694 case DKIOCGETEFI:
1695 {
1696 uint64_t vs = zv->zv_volsize;
1697 uint8_t bs = zv->zv_min_bs;
1698
1699 mutex_exit(&zfsdev_state_lock);
1700 error = zvol_getefi((void *)arg, flag, vs, bs);
1701 return (error);
1702 }
1703
1704 case DKIOCFLUSHWRITECACHE:
1705 dkc = (struct dk_callback *)arg;
1706 mutex_exit(&zfsdev_state_lock);
1707 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1708 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1709 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
1710 error = 0;
1711 }
1712 return (error);
1713
1714 case DKIOCGETWCE:
1715 {
1716 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1717 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1718 flag))
1719 error = SET_ERROR(EFAULT);
1720 break;
1721 }
1722 case DKIOCSETWCE:
1723 {
1724 int wce;
1725 if (ddi_copyin((void *)arg, &wce, sizeof (int),
1726 flag)) {
1727 error = SET_ERROR(EFAULT);
1728 break;
1729 }
1730 if (wce) {
1731 zv->zv_flags |= ZVOL_WCE;
1732 mutex_exit(&zfsdev_state_lock);
1733 } else {
1734 zv->zv_flags &= ~ZVOL_WCE;
1735 mutex_exit(&zfsdev_state_lock);
1736 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1737 }
1738 return (0);
1739 }
1740
1741 case DKIOCGGEOM:
1742 case DKIOCGVTOC:
1743 /*
1744 * commands using these (like prtvtoc) expect ENOTSUP
1745 * since we're emulating an EFI label
1746 */
1747 error = SET_ERROR(ENOTSUP);
1748 break;
1749
1750 case DKIOCDUMPINIT:
1751 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1752 RL_WRITER);
1753 error = zvol_dumpify(zv);
1754 zfs_range_unlock(rl);
1755 break;
1756
1757 case DKIOCDUMPFINI:
1758 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1759 break;
1760 rl = zfs_range_lock(&zv->zv_znode, 0, zv->zv_volsize,
1761 RL_WRITER);
1762 error = zvol_dump_fini(zv);
1763 zfs_range_unlock(rl);
1764 break;
1765
1766 case DKIOCFREE:
1767 {
1768 dkioc_free_t df;
1769 dmu_tx_t *tx;
1770
1771 if (ddi_copyin((void *)arg, &df, sizeof (df), flag)) {
1772 error = SET_ERROR(EFAULT);
1773 break;
1774 }
1775
1776 /*
1777 * Apply Postel's Law to length-checking. If they overshoot,
1778 * just blank out until the end, if there's a need to blank
1779 * out anything.
1780 */
1781 if (df.df_start >= zv->zv_volsize)
1782 break; /* No need to do anything... */
1783 if (df.df_start + df.df_length > zv->zv_volsize)
1784 df.df_length = DMU_OBJECT_END;
1785
1786 rl = zfs_range_lock(&zv->zv_znode, df.df_start, df.df_length,
1787 RL_WRITER);
1788 tx = dmu_tx_create(zv->zv_objset);
1789 error = dmu_tx_assign(tx, TXG_WAIT);
1790 if (error != 0) {
1791 dmu_tx_abort(tx);
1792 } else {
1793 zvol_log_truncate(zv, tx, df.df_start,
1794 df.df_length, B_TRUE);
1795 dmu_tx_commit(tx);
1796 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
1797 df.df_start, df.df_length);
1798 }
1799
1800 zfs_range_unlock(rl);
1801
1802 if (error == 0) {
1803 /*
1804 * If the write-cache is disabled or 'sync' property
1805 * is set to 'always' then treat this as a synchronous
1806 * operation (i.e. commit to zil).
1807 */
1808 if (!(zv->zv_flags & ZVOL_WCE) ||
1809 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS))
1810 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1811
1812 /*
1813 * If the caller really wants synchronous writes, and
1814 * can't wait for them, don't return until the write
1815 * is done.
1816 */
1817 if (df.df_flags & DF_WAIT_SYNC) {
1818 txg_wait_synced(
1819 dmu_objset_pool(zv->zv_objset), 0);
1820 }
1821 }
1822 break;
1823 }
1824
1825 default:
1826 error = SET_ERROR(ENOTTY);
1827 break;
1828
1829 }
1830 mutex_exit(&zfsdev_state_lock);
1831 return (error);
1832 }
1833
1834 int
1835 zvol_busy(void)
1836 {
1837 return (zvol_minors != 0);
1838 }
1839
1840 void
1841 zvol_init(void)
1842 {
1843 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1844 1) == 0);
1845 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
1846 }
1847
1848 void
1849 zvol_fini(void)
1850 {
1851 mutex_destroy(&zfsdev_state_lock);
1852 ddi_soft_state_fini(&zfsdev_state);
1853 }
1854
1855 /*ARGSUSED*/
1856 static int
1857 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
1858 {
1859 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1860
1861 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1862 return (1);
1863 return (0);
1864 }
1865
1866 /*ARGSUSED*/
1867 static void
1868 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
1869 {
1870 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1871
1872 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
1873 }
1874
1875 static int
1876 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1877 {
1878 dmu_tx_t *tx;
1879 int error;
1880 objset_t *os = zv->zv_objset;
1881 spa_t *spa = dmu_objset_spa(os);
1882 vdev_t *vd = spa->spa_root_vdev;
1883 nvlist_t *nv = NULL;
1884 uint64_t version = spa_version(spa);
1885 enum zio_checksum checksum;
1886
1887 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1888 ASSERT(vd->vdev_ops == &vdev_root_ops);
1889
1890 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1891 DMU_OBJECT_END);
1892 /* wait for dmu_free_long_range to actually free the blocks */
1893 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1894
1895 /*
1896 * If the pool on which the dump device is being initialized has more
1897 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
1898 * enabled. If so, bump that feature's counter to indicate that the
1899 * feature is active. We also check the vdev type to handle the
1900 * following case:
1901 * # zpool create test raidz disk1 disk2 disk3
1902 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
1903 * the raidz vdev itself has 3 children.
1904 */
1905 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
1906 if (!spa_feature_is_enabled(spa,
1907 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1908 return (SET_ERROR(ENOTSUP));
1909 (void) dsl_sync_task(spa_name(spa),
1910 zfs_mvdev_dump_feature_check,
1911 zfs_mvdev_dump_activate_feature_sync, NULL, 2);
1912 }
1913
1914 tx = dmu_tx_create(os);
1915 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1916 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1917 error = dmu_tx_assign(tx, TXG_WAIT);
1918 if (error) {
1919 dmu_tx_abort(tx);
1920 return (error);
1921 }
1922
1923 /*
1924 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
1925 * function. Otherwise, use the old default -- OFF.
1926 */
1927 checksum = spa_feature_is_active(spa,
1928 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
1929 ZIO_CHECKSUM_OFF;
1930
1931 /*
1932 * If we are resizing the dump device then we only need to
1933 * update the refreservation to match the newly updated
1934 * zvolsize. Otherwise, we save off the original state of the
1935 * zvol so that we can restore them if the zvol is ever undumpified.
1936 */
1937 if (resize) {
1938 error = zap_update(os, ZVOL_ZAP_OBJ,
1939 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1940 &zv->zv_volsize, tx);
1941 } else {
1942 uint64_t checksum, compress, refresrv, vbs, dedup;
1943
1944 error = dsl_prop_get_integer(zv->zv_name,
1945 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1946 error = error ? error : dsl_prop_get_integer(zv->zv_name,
1947 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum, NULL);
1948 error = error ? error : dsl_prop_get_integer(zv->zv_name,
1949 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &refresrv, NULL);
1950 error = error ? error : dsl_prop_get_integer(zv->zv_name,
1951 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs, NULL);
1952 if (version >= SPA_VERSION_DEDUP) {
1953 error = error ? error :
1954 dsl_prop_get_integer(zv->zv_name,
1955 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1956 }
1957
1958 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1959 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
1960 &compress, tx);
1961 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1962 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum, tx);
1963 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1964 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1965 &refresrv, tx);
1966 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1967 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
1968 &vbs, tx);
1969 error = error ? error : dmu_object_set_blocksize(
1970 os, ZVOL_OBJ, SPA_MAXBLOCKSIZE, 0, tx);
1971 if (version >= SPA_VERSION_DEDUP) {
1972 error = error ? error : zap_update(os, ZVOL_ZAP_OBJ,
1973 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
1974 &dedup, tx);
1975 }
1976 if (error == 0)
1977 zv->zv_volblocksize = SPA_MAXBLOCKSIZE;
1978 }
1979 dmu_tx_commit(tx);
1980
1981 /*
1982 * We only need update the zvol's property if we are initializing
1983 * the dump area for the first time.
1984 */
1985 if (!resize) {
1986 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
1987 VERIFY(nvlist_add_uint64(nv,
1988 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
1989 VERIFY(nvlist_add_uint64(nv,
1990 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
1991 ZIO_COMPRESS_OFF) == 0);
1992 VERIFY(nvlist_add_uint64(nv,
1993 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
1994 checksum) == 0);
1995 if (version >= SPA_VERSION_DEDUP) {
1996 VERIFY(nvlist_add_uint64(nv,
1997 zfs_prop_to_name(ZFS_PROP_DEDUP),
1998 ZIO_CHECKSUM_OFF) == 0);
1999 }
2000
2001 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2002 nv, NULL);
2003 nvlist_free(nv);
2004
2005 if (error)
2006 return (error);
2007 }
2008
2009 /* Allocate the space for the dump */
2010 error = zvol_prealloc(zv);
2011 return (error);
2012 }
2013
2014 static int
2015 zvol_dumpify(zvol_state_t *zv)
2016 {
2017 int error = 0;
2018 uint64_t dumpsize = 0;
2019 dmu_tx_t *tx;
2020 objset_t *os = zv->zv_objset;
2021
2022 if (zv->zv_flags & ZVOL_RDONLY)
2023 return (SET_ERROR(EROFS));
2024
2025 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2026 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2027 boolean_t resize = (dumpsize > 0);
2028
2029 if ((error = zvol_dump_init(zv, resize)) != 0) {
2030 (void) zvol_dump_fini(zv);
2031 return (error);
2032 }
2033 }
2034
2035 /*
2036 * Build up our lba mapping.
2037 */
2038 error = zvol_get_lbas(zv);
2039 if (error) {
2040 (void) zvol_dump_fini(zv);
2041 return (error);
2042 }
2043
2044 tx = dmu_tx_create(os);
2045 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2046 error = dmu_tx_assign(tx, TXG_WAIT);
2047 if (error) {
2048 dmu_tx_abort(tx);
2049 (void) zvol_dump_fini(zv);
2050 return (error);
2051 }
2052
2053 zv->zv_flags |= ZVOL_DUMPIFIED;
2054 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2055 &zv->zv_volsize, tx);
2056 dmu_tx_commit(tx);
2057
2058 if (error) {
2059 (void) zvol_dump_fini(zv);
2060 return (error);
2061 }
2062
2063 txg_wait_synced(dmu_objset_pool(os), 0);
2064 return (0);
2065 }
2066
2067 static int
2068 zvol_dump_fini(zvol_state_t *zv)
2069 {
2070 dmu_tx_t *tx;
2071 objset_t *os = zv->zv_objset;
2072 nvlist_t *nv;
2073 int error = 0;
2074 uint64_t checksum, compress, refresrv, vbs, dedup;
2075 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2076
2077 /*
2078 * Attempt to restore the zvol back to its pre-dumpified state.
2079 * This is a best-effort attempt as it's possible that not all
2080 * of these properties were initialized during the dumpify process
2081 * (i.e. error during zvol_dump_init).
2082 */
2083
2084 tx = dmu_tx_create(os);
2085 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2086 error = dmu_tx_assign(tx, TXG_WAIT);
2087 if (error) {
2088 dmu_tx_abort(tx);
2089 return (error);
2090 }
2091 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2092 dmu_tx_commit(tx);
2093
2094 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2095 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2096 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2097 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2098 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2099 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2100 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2101 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2102
2103 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2104 (void) nvlist_add_uint64(nv,
2105 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2106 (void) nvlist_add_uint64(nv,
2107 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2108 (void) nvlist_add_uint64(nv,
2109 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2110 if (version >= SPA_VERSION_DEDUP &&
2111 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2112 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2113 (void) nvlist_add_uint64(nv,
2114 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2115 }
2116 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2117 nv, NULL);
2118 nvlist_free(nv);
2119
2120 zvol_free_extents(zv);
2121 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2122 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2123 /* wait for dmu_free_long_range to actually free the blocks */
2124 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2125 tx = dmu_tx_create(os);
2126 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2127 error = dmu_tx_assign(tx, TXG_WAIT);
2128 if (error) {
2129 dmu_tx_abort(tx);
2130 return (error);
2131 }
2132 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2133 zv->zv_volblocksize = vbs;
2134 dmu_tx_commit(tx);
2135
2136 return (0);
2137 }
Unleashed OS