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 2017 Nexenta Systems, Inc. All rights reserved.
27 * Copyright (c) 2012, 2020 by Delphix. All rights reserved.
28 * Copyright (c) 2014 Integros [integros.com]
29 * Copyright 2019 Joyent, Inc.
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_impl.h>
82 #include <sys/zvol.h>
83 #include <sys/dumphdr.h>
84 #include <sys/zil_impl.h>
85 #include <sys/dbuf.h>
86 #include <sys/dmu_tx.h>
87 #include <sys/zfeature.h>
88 #include <sys/zio_checksum.h>
89 #include <sys/zil_impl.h>
90 #include <sys/smt.h>
91 #include <sys/dkioc_free_util.h>
92 #include <sys/zfs_rlock.h>
93
94 #include "zfs_namecheck.h"
95
96 void *zfsdev_state;
97 static char *zvol_tag = "zvol_tag";
98
99 #define ZVOL_DUMPSIZE "dumpsize"
100
101 /*
102 * This lock protects the zfsdev_state structure from being modified
103 * while it's being used, e.g. an open that comes in before a create
104 * finishes. It also protects temporary opens of the dataset so that,
105 * e.g., an open doesn't get a spurious EBUSY.
106 */
107 kmutex_t zfsdev_state_lock;
108 static uint32_t zvol_minors;
109
110 typedef struct zvol_extent {
111 list_node_t ze_node;
112 dva_t ze_dva; /* dva associated with this extent */
113 uint64_t ze_nblks; /* number of blocks in extent */
114 } zvol_extent_t;
115
116 /*
117 * The in-core state of each volume.
118 */
119 typedef struct zvol_state {
120 char zv_name[MAXPATHLEN]; /* pool/dd name */
121 uint64_t zv_volsize; /* amount of space we advertise */
122 uint64_t zv_volblocksize; /* volume block size */
123 minor_t zv_minor; /* minor number */
124 uint8_t zv_min_bs; /* minimum addressable block shift */
125 uint8_t zv_flags; /* readonly, dumpified, etc. */
126 objset_t *zv_objset; /* objset handle */
127 uint32_t zv_open_count[OTYPCNT]; /* open counts */
128 uint32_t zv_total_opens; /* total open count */
129 zilog_t *zv_zilog; /* ZIL handle */
130 list_t zv_extents; /* List of extents for dump */
131 rangelock_t zv_rangelock;
132 dnode_t *zv_dn; /* dnode hold */
133 } zvol_state_t;
134
135 /*
136 * zvol specific flags
137 */
138 #define ZVOL_RDONLY 0x1
139 #define ZVOL_DUMPIFIED 0x2
140 #define ZVOL_EXCL 0x4
141 #define ZVOL_WCE 0x8
142
143 /*
144 * zvol maximum transfer in one DMU tx.
145 */
146 int zvol_maxphys = DMU_MAX_ACCESS/2;
147
148 /*
149 * Toggle unmap functionality.
150 */
151 boolean_t zvol_unmap_enabled = B_TRUE;
152
153 /*
154 * If true, unmaps requested as synchronous are executed synchronously,
155 * otherwise all unmaps are asynchronous.
156 */
157 boolean_t zvol_unmap_sync_enabled = B_FALSE;
158
159 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
160 nvlist_t *, nvlist_t *);
161 static int zvol_remove_zv(zvol_state_t *);
162 static int zvol_get_data(void *arg, lr_write_t *lr, char *buf,
163 struct lwb *lwb, zio_t *zio);
164 static int zvol_dumpify(zvol_state_t *zv);
165 static int zvol_dump_fini(zvol_state_t *zv);
166 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
167
168 static void
zvol_size_changed(zvol_state_t * zv,uint64_t volsize)169 zvol_size_changed(zvol_state_t *zv, uint64_t volsize)
170 {
171 dev_t dev = makedevice(ddi_driver_major(zfs_dip), zv->zv_minor);
172
173 zv->zv_volsize = volsize;
174 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
175 "Size", volsize) == DDI_SUCCESS);
176 VERIFY(ddi_prop_update_int64(dev, zfs_dip,
177 "Nblocks", lbtodb(volsize)) == DDI_SUCCESS);
178
179 /* Notify specfs to invalidate the cached size */
180 spec_size_invalidate(dev, VBLK);
181 spec_size_invalidate(dev, VCHR);
182 }
183
184 int
zvol_check_volsize(uint64_t volsize,uint64_t blocksize)185 zvol_check_volsize(uint64_t volsize, uint64_t blocksize)
186 {
187 if (volsize == 0)
188 return (SET_ERROR(EINVAL));
189
190 if (volsize % blocksize != 0)
191 return (SET_ERROR(EINVAL));
192
193 #ifdef _ILP32
194 if (volsize - 1 > SPEC_MAXOFFSET_T)
195 return (SET_ERROR(EOVERFLOW));
196 #endif
197 return (0);
198 }
199
200 int
zvol_check_volblocksize(uint64_t volblocksize)201 zvol_check_volblocksize(uint64_t volblocksize)
202 {
203 if (volblocksize < SPA_MINBLOCKSIZE ||
204 volblocksize > SPA_OLD_MAXBLOCKSIZE ||
205 !ISP2(volblocksize))
206 return (SET_ERROR(EDOM));
207
208 return (0);
209 }
210
211 int
zvol_get_stats(objset_t * os,nvlist_t * nv)212 zvol_get_stats(objset_t *os, nvlist_t *nv)
213 {
214 int error;
215 dmu_object_info_t doi;
216 uint64_t val;
217
218 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &val);
219 if (error)
220 return (error);
221
222 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLSIZE, val);
223
224 error = dmu_object_info(os, ZVOL_OBJ, &doi);
225
226 if (error == 0) {
227 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_VOLBLOCKSIZE,
228 doi.doi_data_block_size);
229 }
230
231 return (error);
232 }
233
234 static zvol_state_t *
zvol_minor_lookup(const char * name)235 zvol_minor_lookup(const char *name)
236 {
237 minor_t minor;
238 zvol_state_t *zv;
239
240 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
241
242 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
243 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
244 if (zv == NULL)
245 continue;
246 if (strcmp(zv->zv_name, name) == 0)
247 return (zv);
248 }
249
250 return (NULL);
251 }
252
253 /* extent mapping arg */
254 struct maparg {
255 zvol_state_t *ma_zv;
256 uint64_t ma_blks;
257 };
258
259 /*ARGSUSED*/
260 static int
zvol_map_block(spa_t * spa,zilog_t * zilog,const blkptr_t * bp,const zbookmark_phys_t * zb,const dnode_phys_t * dnp,void * arg)261 zvol_map_block(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
262 const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
263 {
264 struct maparg *ma = arg;
265 zvol_extent_t *ze;
266 int bs = ma->ma_zv->zv_volblocksize;
267
268 if (bp == NULL || BP_IS_HOLE(bp) ||
269 zb->zb_object != ZVOL_OBJ || zb->zb_level != 0)
270 return (0);
271
272 VERIFY(!BP_IS_EMBEDDED(bp));
273
274 VERIFY3U(ma->ma_blks, ==, zb->zb_blkid);
275 ma->ma_blks++;
276
277 /* Abort immediately if we have encountered gang blocks */
278 if (BP_IS_GANG(bp))
279 return (SET_ERROR(EFRAGS));
280
281 /*
282 * See if the block is at the end of the previous extent.
283 */
284 ze = list_tail(&ma->ma_zv->zv_extents);
285 if (ze &&
286 DVA_GET_VDEV(BP_IDENTITY(bp)) == DVA_GET_VDEV(&ze->ze_dva) &&
287 DVA_GET_OFFSET(BP_IDENTITY(bp)) ==
288 DVA_GET_OFFSET(&ze->ze_dva) + ze->ze_nblks * bs) {
289 ze->ze_nblks++;
290 return (0);
291 }
292
293 dprintf_bp(bp, "%s", "next blkptr:");
294
295 /* start a new extent */
296 ze = kmem_zalloc(sizeof (zvol_extent_t), KM_SLEEP);
297 ze->ze_dva = bp->blk_dva[0]; /* structure assignment */
298 ze->ze_nblks = 1;
299 list_insert_tail(&ma->ma_zv->zv_extents, ze);
300 return (0);
301 }
302
303 static void
zvol_free_extents(zvol_state_t * zv)304 zvol_free_extents(zvol_state_t *zv)
305 {
306 zvol_extent_t *ze;
307
308 while (ze = list_head(&zv->zv_extents)) {
309 list_remove(&zv->zv_extents, ze);
310 kmem_free(ze, sizeof (zvol_extent_t));
311 }
312 }
313
314 static int
zvol_get_lbas(zvol_state_t * zv)315 zvol_get_lbas(zvol_state_t *zv)
316 {
317 objset_t *os = zv->zv_objset;
318 struct maparg ma;
319 int err;
320
321 ma.ma_zv = zv;
322 ma.ma_blks = 0;
323 zvol_free_extents(zv);
324
325 /* commit any in-flight changes before traversing the dataset */
326 txg_wait_synced(dmu_objset_pool(os), 0);
327 err = traverse_dataset(dmu_objset_ds(os), 0,
328 TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA, zvol_map_block, &ma);
329 if (err || ma.ma_blks != (zv->zv_volsize / zv->zv_volblocksize)) {
330 zvol_free_extents(zv);
331 return (err ? err : EIO);
332 }
333
334 return (0);
335 }
336
337 /* ARGSUSED */
338 void
zvol_create_cb(objset_t * os,void * arg,cred_t * cr,dmu_tx_t * tx)339 zvol_create_cb(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx)
340 {
341 zfs_creat_t *zct = arg;
342 nvlist_t *nvprops = zct->zct_props;
343 int error;
344 uint64_t volblocksize, volsize;
345
346 VERIFY(nvlist_lookup_uint64(nvprops,
347 zfs_prop_to_name(ZFS_PROP_VOLSIZE), &volsize) == 0);
348 if (nvlist_lookup_uint64(nvprops,
349 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &volblocksize) != 0)
350 volblocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
351
352 /*
353 * These properties must be removed from the list so the generic
354 * property setting step won't apply to them.
355 */
356 VERIFY(nvlist_remove_all(nvprops,
357 zfs_prop_to_name(ZFS_PROP_VOLSIZE)) == 0);
358 (void) nvlist_remove_all(nvprops,
359 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE));
360
361 error = dmu_object_claim(os, ZVOL_OBJ, DMU_OT_ZVOL, volblocksize,
362 DMU_OT_NONE, 0, tx);
363 ASSERT(error == 0);
364
365 error = zap_create_claim(os, ZVOL_ZAP_OBJ, DMU_OT_ZVOL_PROP,
366 DMU_OT_NONE, 0, tx);
367 ASSERT(error == 0);
368
369 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize, tx);
370 ASSERT(error == 0);
371 }
372
373 /*
374 * Replay a TX_TRUNCATE ZIL transaction if asked. TX_TRUNCATE is how we
375 * implement DKIOCFREE/free-long-range.
376 */
377 static int
zvol_replay_truncate(void * arg1,void * arg2,boolean_t byteswap)378 zvol_replay_truncate(void *arg1, void *arg2, boolean_t byteswap)
379 {
380 zvol_state_t *zv = arg1;
381 lr_truncate_t *lr = arg2;
382 uint64_t offset, length;
383
384 if (byteswap)
385 byteswap_uint64_array(lr, sizeof (*lr));
386
387 offset = lr->lr_offset;
388 length = lr->lr_length;
389
390 return (dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, offset, length));
391 }
392
393 /*
394 * Replay a TX_WRITE ZIL transaction that didn't get committed
395 * after a system failure
396 */
397 /* ARGSUSED */
398 static int
zvol_replay_write(void * arg1,void * arg2,boolean_t byteswap)399 zvol_replay_write(void *arg1, void *arg2, boolean_t byteswap)
400 {
401 zvol_state_t *zv = arg1;
402 lr_write_t *lr = arg2;
403 objset_t *os = zv->zv_objset;
404 char *data = (char *)(lr + 1); /* data follows lr_write_t */
405 uint64_t offset, length;
406 dmu_tx_t *tx;
407 int error;
408
409 if (byteswap)
410 byteswap_uint64_array(lr, sizeof (*lr));
411
412 offset = lr->lr_offset;
413 length = lr->lr_length;
414
415 /* If it's a dmu_sync() block, write the whole block */
416 if (lr->lr_common.lrc_reclen == sizeof (lr_write_t)) {
417 uint64_t blocksize = BP_GET_LSIZE(&lr->lr_blkptr);
418 if (length < blocksize) {
419 offset -= offset % blocksize;
420 length = blocksize;
421 }
422 }
423
424 tx = dmu_tx_create(os);
425 dmu_tx_hold_write(tx, ZVOL_OBJ, offset, length);
426 error = dmu_tx_assign(tx, TXG_WAIT);
427 if (error) {
428 dmu_tx_abort(tx);
429 } else {
430 dmu_write(os, ZVOL_OBJ, offset, length, data, tx);
431 dmu_tx_commit(tx);
432 }
433
434 return (error);
435 }
436
437 /* ARGSUSED */
438 static int
zvol_replay_err(void * arg1,void * arg2,boolean_t byteswap)439 zvol_replay_err(void *arg1, void *arg2, boolean_t byteswap)
440 {
441 return (SET_ERROR(ENOTSUP));
442 }
443
444 /*
445 * Callback vectors for replaying records.
446 * Only TX_WRITE and TX_TRUNCATE are needed for zvol.
447 */
448 zil_replay_func_t *zvol_replay_vector[TX_MAX_TYPE] = {
449 zvol_replay_err, /* 0 no such transaction type */
450 zvol_replay_err, /* TX_CREATE */
451 zvol_replay_err, /* TX_MKDIR */
452 zvol_replay_err, /* TX_MKXATTR */
453 zvol_replay_err, /* TX_SYMLINK */
454 zvol_replay_err, /* TX_REMOVE */
455 zvol_replay_err, /* TX_RMDIR */
456 zvol_replay_err, /* TX_LINK */
457 zvol_replay_err, /* TX_RENAME */
458 zvol_replay_write, /* TX_WRITE */
459 zvol_replay_truncate, /* TX_TRUNCATE */
460 zvol_replay_err, /* TX_SETATTR */
461 zvol_replay_err, /* TX_ACL */
462 zvol_replay_err, /* TX_CREATE_ACL */
463 zvol_replay_err, /* TX_CREATE_ATTR */
464 zvol_replay_err, /* TX_CREATE_ACL_ATTR */
465 zvol_replay_err, /* TX_MKDIR_ACL */
466 zvol_replay_err, /* TX_MKDIR_ATTR */
467 zvol_replay_err, /* TX_MKDIR_ACL_ATTR */
468 zvol_replay_err, /* TX_WRITE2 */
469 };
470
471 int
zvol_name2minor(const char * name,minor_t * minor)472 zvol_name2minor(const char *name, minor_t *minor)
473 {
474 zvol_state_t *zv;
475
476 mutex_enter(&zfsdev_state_lock);
477 zv = zvol_minor_lookup(name);
478 if (minor && zv)
479 *minor = zv->zv_minor;
480 mutex_exit(&zfsdev_state_lock);
481 return (zv ? 0 : -1);
482 }
483
484 /*
485 * Create a minor node (plus a whole lot more) for the specified volume.
486 */
487 int
zvol_create_minor(const char * name)488 zvol_create_minor(const char *name)
489 {
490 zfs_soft_state_t *zs;
491 zvol_state_t *zv;
492 objset_t *os;
493 dmu_object_info_t doi;
494 minor_t minor = 0;
495 char chrbuf[30], blkbuf[30];
496 int error;
497
498 mutex_enter(&zfsdev_state_lock);
499
500 if (zvol_minor_lookup(name) != NULL) {
501 mutex_exit(&zfsdev_state_lock);
502 return (SET_ERROR(EEXIST));
503 }
504
505 /* lie and say we're read-only */
506 error = dmu_objset_own(name, DMU_OST_ZVOL, B_TRUE, B_TRUE, FTAG, &os);
507
508 if (error) {
509 mutex_exit(&zfsdev_state_lock);
510 return (error);
511 }
512
513 if ((minor = zfsdev_minor_alloc()) == 0) {
514 dmu_objset_disown(os, 1, FTAG);
515 mutex_exit(&zfsdev_state_lock);
516 return (SET_ERROR(ENXIO));
517 }
518
519 if (ddi_soft_state_zalloc(zfsdev_state, minor) != DDI_SUCCESS) {
520 dmu_objset_disown(os, 1, FTAG);
521 mutex_exit(&zfsdev_state_lock);
522 return (SET_ERROR(EAGAIN));
523 }
524 (void) ddi_prop_update_string(minor, zfs_dip, ZVOL_PROP_NAME,
525 (char *)name);
526
527 (void) snprintf(chrbuf, sizeof (chrbuf), "%u,raw", minor);
528
529 if (ddi_create_minor_node(zfs_dip, chrbuf, S_IFCHR,
530 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
531 ddi_soft_state_free(zfsdev_state, minor);
532 dmu_objset_disown(os, 1, FTAG);
533 mutex_exit(&zfsdev_state_lock);
534 return (SET_ERROR(EAGAIN));
535 }
536
537 (void) snprintf(blkbuf, sizeof (blkbuf), "%u", minor);
538
539 if (ddi_create_minor_node(zfs_dip, blkbuf, S_IFBLK,
540 minor, DDI_PSEUDO, 0) == DDI_FAILURE) {
541 ddi_remove_minor_node(zfs_dip, chrbuf);
542 ddi_soft_state_free(zfsdev_state, minor);
543 dmu_objset_disown(os, 1, FTAG);
544 mutex_exit(&zfsdev_state_lock);
545 return (SET_ERROR(EAGAIN));
546 }
547
548 zs = ddi_get_soft_state(zfsdev_state, minor);
549 zs->zss_type = ZSST_ZVOL;
550 zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
551 (void) strlcpy(zv->zv_name, name, MAXPATHLEN);
552 zv->zv_min_bs = DEV_BSHIFT;
553 zv->zv_minor = minor;
554 zv->zv_objset = os;
555 if (dmu_objset_is_snapshot(os) || !spa_writeable(dmu_objset_spa(os)))
556 zv->zv_flags |= ZVOL_RDONLY;
557 rangelock_init(&zv->zv_rangelock, NULL, NULL);
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, 1, 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
zvol_remove_zv(zvol_state_t * zv)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 rangelock_fini(&zv->zv_rangelock);
601
602 kmem_free(zv, sizeof (zvol_state_t));
603
604 ddi_soft_state_free(zfsdev_state, minor);
605
606 zvol_minors--;
607 return (0);
608 }
609
610 int
zvol_remove_minor(const char * name)611 zvol_remove_minor(const char *name)
612 {
613 zvol_state_t *zv;
614 int rc;
615
616 mutex_enter(&zfsdev_state_lock);
617 if ((zv = zvol_minor_lookup(name)) == NULL) {
618 mutex_exit(&zfsdev_state_lock);
619 return (SET_ERROR(ENXIO));
620 }
621 rc = zvol_remove_zv(zv);
622 mutex_exit(&zfsdev_state_lock);
623 return (rc);
624 }
625
626 int
zvol_first_open(zvol_state_t * zv,boolean_t rdonly)627 zvol_first_open(zvol_state_t *zv, boolean_t rdonly)
628 {
629 objset_t *os;
630 uint64_t volsize;
631 int error;
632 uint64_t readonly;
633 boolean_t ro;
634
635 ro = (rdonly || (strchr(zv->zv_name, '@') != NULL));
636 error = dmu_objset_own(zv->zv_name, DMU_OST_ZVOL, ro, B_TRUE, zv, &os);
637 if (error)
638 return (error);
639
640 zv->zv_objset = os;
641 error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
642 if (error) {
643 ASSERT(error == 0);
644 dmu_objset_disown(os, 1, zv);
645 return (error);
646 }
647
648 error = dnode_hold(os, ZVOL_OBJ, zvol_tag, &zv->zv_dn);
649 if (error) {
650 dmu_objset_disown(os, 1, zv);
651 return (error);
652 }
653
654 zvol_size_changed(zv, volsize);
655 zv->zv_zilog = zil_open(os, zvol_get_data);
656
657 VERIFY(dsl_prop_get_integer(zv->zv_name, "readonly", &readonly,
658 NULL) == 0);
659 if (readonly || dmu_objset_is_snapshot(os) ||
660 !spa_writeable(dmu_objset_spa(os)))
661 zv->zv_flags |= ZVOL_RDONLY;
662 else
663 zv->zv_flags &= ~ZVOL_RDONLY;
664 return (error);
665 }
666
667 void
zvol_last_close(zvol_state_t * zv)668 zvol_last_close(zvol_state_t *zv)
669 {
670 zil_close(zv->zv_zilog);
671 zv->zv_zilog = NULL;
672
673 dnode_rele(zv->zv_dn, zvol_tag);
674 zv->zv_dn = NULL;
675
676 /*
677 * Evict cached data
678 */
679 if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
680 !(zv->zv_flags & ZVOL_RDONLY))
681 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
682 dmu_objset_evict_dbufs(zv->zv_objset);
683
684 dmu_objset_disown(zv->zv_objset, 1, zv);
685 zv->zv_objset = NULL;
686 }
687
688 int
zvol_prealloc(zvol_state_t * zv)689 zvol_prealloc(zvol_state_t *zv)
690 {
691 objset_t *os = zv->zv_objset;
692 dmu_tx_t *tx;
693 uint64_t refd, avail, usedobjs, availobjs;
694 uint64_t resid = zv->zv_volsize;
695 uint64_t off = 0;
696
697 /* Check the space usage before attempting to allocate the space */
698 dmu_objset_space(os, &refd, &avail, &usedobjs, &availobjs);
699 if (avail < zv->zv_volsize)
700 return (SET_ERROR(ENOSPC));
701
702 /* Free old extents if they exist */
703 zvol_free_extents(zv);
704
705 while (resid != 0) {
706 int error;
707 uint64_t bytes = MIN(resid, SPA_OLD_MAXBLOCKSIZE);
708
709 tx = dmu_tx_create(os);
710 dmu_tx_hold_write(tx, ZVOL_OBJ, off, bytes);
711 error = dmu_tx_assign(tx, TXG_WAIT);
712 if (error) {
713 dmu_tx_abort(tx);
714 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, off);
715 return (error);
716 }
717 dmu_prealloc(os, ZVOL_OBJ, off, bytes, tx);
718 dmu_tx_commit(tx);
719 off += bytes;
720 resid -= bytes;
721 }
722 txg_wait_synced(dmu_objset_pool(os), 0);
723
724 return (0);
725 }
726
727 static int
zvol_update_volsize(objset_t * os,uint64_t volsize)728 zvol_update_volsize(objset_t *os, uint64_t volsize)
729 {
730 dmu_tx_t *tx;
731 int error;
732 uint64_t txg;
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 txg = dmu_tx_get_txg(tx);
745
746 error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
747 &volsize, tx);
748 dmu_tx_commit(tx);
749
750 txg_wait_synced(dmu_objset_pool(os), txg);
751
752 if (error == 0)
753 error = dmu_free_long_range(os,
754 ZVOL_OBJ, volsize, DMU_OBJECT_END);
755 return (error);
756 }
757
758 void
zvol_remove_minors(const char * name)759 zvol_remove_minors(const char *name)
760 {
761 zvol_state_t *zv;
762 char *namebuf;
763 minor_t minor;
764
765 namebuf = kmem_zalloc(strlen(name) + 2, KM_SLEEP);
766 (void) strncpy(namebuf, name, strlen(name));
767 (void) strcat(namebuf, "/");
768 mutex_enter(&zfsdev_state_lock);
769 for (minor = 1; minor <= ZFSDEV_MAX_MINOR; minor++) {
770
771 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
772 if (zv == NULL)
773 continue;
774 if (strncmp(namebuf, zv->zv_name, strlen(namebuf)) == 0)
775 (void) zvol_remove_zv(zv);
776 }
777 kmem_free(namebuf, strlen(name) + 2);
778
779 mutex_exit(&zfsdev_state_lock);
780 }
781
782 static int
zvol_update_live_volsize(zvol_state_t * zv,uint64_t volsize)783 zvol_update_live_volsize(zvol_state_t *zv, uint64_t volsize)
784 {
785 uint64_t old_volsize = 0ULL;
786 int error = 0;
787
788 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
789
790 /*
791 * Reinitialize the dump area to the new size. If we
792 * failed to resize the dump area then restore it back to
793 * its original size. We must set the new volsize prior
794 * to calling dumpvp_resize() to ensure that the devices'
795 * size(9P) is not visible by the dump subsystem.
796 */
797 old_volsize = zv->zv_volsize;
798 zvol_size_changed(zv, volsize);
799
800 if (zv->zv_flags & ZVOL_DUMPIFIED) {
801 if ((error = zvol_dumpify(zv)) != 0 ||
802 (error = dumpvp_resize()) != 0) {
803 int dumpify_error;
804
805 (void) zvol_update_volsize(zv->zv_objset, old_volsize);
806 zvol_size_changed(zv, old_volsize);
807 dumpify_error = zvol_dumpify(zv);
808 error = dumpify_error ? dumpify_error : error;
809 }
810 }
811
812 /*
813 * Generate a LUN expansion event.
814 */
815 if (error == 0) {
816 char *physpath = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
817
818 (void) snprintf(physpath, MAXPATHLEN, "%s%u", ZVOL_PSEUDO_DEV,
819 zv->zv_minor);
820
821 zfs_post_dle_sysevent(physpath);
822 kmem_free(physpath, MAXPATHLEN);
823 }
824 return (error);
825 }
826
827 int
zvol_set_volsize(const char * name,uint64_t volsize)828 zvol_set_volsize(const char *name, uint64_t volsize)
829 {
830 zvol_state_t *zv = NULL;
831 objset_t *os;
832 int error;
833 dmu_object_info_t doi;
834 uint64_t readonly;
835 boolean_t owned = B_FALSE;
836
837 error = dsl_prop_get_integer(name,
838 zfs_prop_to_name(ZFS_PROP_READONLY), &readonly, NULL);
839 if (error != 0)
840 return (error);
841 if (readonly)
842 return (SET_ERROR(EROFS));
843
844 mutex_enter(&zfsdev_state_lock);
845 zv = zvol_minor_lookup(name);
846
847 if (zv == NULL || zv->zv_objset == NULL) {
848 if ((error = dmu_objset_own(name, DMU_OST_ZVOL, B_FALSE, B_TRUE,
849 FTAG, &os)) != 0) {
850 mutex_exit(&zfsdev_state_lock);
851 return (error);
852 }
853 owned = B_TRUE;
854 if (zv != NULL)
855 zv->zv_objset = os;
856 } else {
857 os = zv->zv_objset;
858 }
859
860 if ((error = dmu_object_info(os, ZVOL_OBJ, &doi)) != 0 ||
861 (error = zvol_check_volsize(volsize, doi.doi_data_block_size)) != 0)
862 goto out;
863
864 error = zvol_update_volsize(os, volsize);
865
866 if (error == 0 && zv != NULL)
867 error = zvol_update_live_volsize(zv, volsize);
868 out:
869 if (owned) {
870 dmu_objset_disown(os, B_TRUE, FTAG);
871 if (zv != NULL)
872 zv->zv_objset = NULL;
873 }
874 mutex_exit(&zfsdev_state_lock);
875 return (error);
876 }
877
878 /*ARGSUSED*/
879 int
zvol_open(dev_t * devp,int flag,int otyp,cred_t * cr)880 zvol_open(dev_t *devp, int flag, int otyp, cred_t *cr)
881 {
882 zvol_state_t *zv;
883 int err = 0;
884
885 mutex_enter(&zfsdev_state_lock);
886
887 zv = zfsdev_get_soft_state(getminor(*devp), ZSST_ZVOL);
888 if (zv == NULL) {
889 mutex_exit(&zfsdev_state_lock);
890 return (SET_ERROR(ENXIO));
891 }
892
893 if (zv->zv_total_opens == 0)
894 err = zvol_first_open(zv, !(flag & FWRITE));
895 if (err) {
896 mutex_exit(&zfsdev_state_lock);
897 return (err);
898 }
899
900 if ((flag & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
901 err = SET_ERROR(EROFS);
902 goto out;
903 }
904 if (zv->zv_flags & ZVOL_EXCL) {
905 err = SET_ERROR(EBUSY);
906 goto out;
907 }
908 if (flag & FEXCL) {
909 if (zv->zv_total_opens != 0) {
910 err = SET_ERROR(EBUSY);
911 goto out;
912 }
913 zv->zv_flags |= ZVOL_EXCL;
914 }
915
916 if (zv->zv_open_count[otyp] == 0 || otyp == OTYP_LYR) {
917 zv->zv_open_count[otyp]++;
918 zv->zv_total_opens++;
919 }
920 mutex_exit(&zfsdev_state_lock);
921
922 return (err);
923 out:
924 if (zv->zv_total_opens == 0)
925 zvol_last_close(zv);
926 mutex_exit(&zfsdev_state_lock);
927 return (err);
928 }
929
930 /*ARGSUSED*/
931 int
zvol_close(dev_t dev,int flag,int otyp,cred_t * cr)932 zvol_close(dev_t dev, int flag, int otyp, cred_t *cr)
933 {
934 minor_t minor = getminor(dev);
935 zvol_state_t *zv;
936 int error = 0;
937
938 mutex_enter(&zfsdev_state_lock);
939
940 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
941 if (zv == NULL) {
942 mutex_exit(&zfsdev_state_lock);
943 return (SET_ERROR(ENXIO));
944 }
945
946 if (zv->zv_flags & ZVOL_EXCL) {
947 ASSERT(zv->zv_total_opens == 1);
948 zv->zv_flags &= ~ZVOL_EXCL;
949 }
950
951 /*
952 * If the open count is zero, this is a spurious close.
953 * That indicates a bug in the kernel / DDI framework.
954 */
955 ASSERT(zv->zv_open_count[otyp] != 0);
956 ASSERT(zv->zv_total_opens != 0);
957
958 /*
959 * You may get multiple opens, but only one close.
960 */
961 zv->zv_open_count[otyp]--;
962 zv->zv_total_opens--;
963
964 if (zv->zv_total_opens == 0)
965 zvol_last_close(zv);
966
967 mutex_exit(&zfsdev_state_lock);
968 return (error);
969 }
970
971 /* ARGSUSED */
972 static void
zvol_get_done(zgd_t * zgd,int error)973 zvol_get_done(zgd_t *zgd, int error)
974 {
975 if (zgd->zgd_db)
976 dmu_buf_rele(zgd->zgd_db, zgd);
977
978 rangelock_exit(zgd->zgd_lr);
979
980 kmem_free(zgd, sizeof (zgd_t));
981 }
982
983 /*
984 * Get data to generate a TX_WRITE intent log record.
985 */
986 static int
zvol_get_data(void * arg,lr_write_t * lr,char * buf,struct lwb * lwb,zio_t * zio)987 zvol_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio)
988 {
989 zvol_state_t *zv = arg;
990 uint64_t offset = lr->lr_offset;
991 uint64_t size = lr->lr_length; /* length of user data */
992 dmu_buf_t *db;
993 zgd_t *zgd;
994 int error;
995
996 ASSERT3P(lwb, !=, NULL);
997 ASSERT3P(zio, !=, NULL);
998 ASSERT3U(size, !=, 0);
999
1000 zgd = kmem_zalloc(sizeof (zgd_t), KM_SLEEP);
1001 zgd->zgd_lwb = lwb;
1002
1003 /*
1004 * Write records come in two flavors: immediate and indirect.
1005 * For small writes it's cheaper to store the data with the
1006 * log record (immediate); for large writes it's cheaper to
1007 * sync the data and get a pointer to it (indirect) so that
1008 * we don't have to write the data twice.
1009 */
1010 if (buf != NULL) { /* immediate write */
1011 zgd->zgd_lr = rangelock_enter(&zv->zv_rangelock, offset, size,
1012 RL_READER);
1013 error = dmu_read_by_dnode(zv->zv_dn, offset, size, buf,
1014 DMU_READ_NO_PREFETCH);
1015 } else { /* indirect write */
1016 /*
1017 * Have to lock the whole block to ensure when it's written out
1018 * and its checksum is being calculated that no one can change
1019 * the data. Contrarily to zfs_get_data we need not re-check
1020 * blocksize after we get the lock because it cannot be changed.
1021 */
1022 size = zv->zv_volblocksize;
1023 offset = P2ALIGN(offset, size);
1024 zgd->zgd_lr = rangelock_enter(&zv->zv_rangelock, offset, size,
1025 RL_READER);
1026 error = dmu_buf_hold_by_dnode(zv->zv_dn, offset, zgd, &db,
1027 DMU_READ_NO_PREFETCH);
1028 if (error == 0) {
1029 blkptr_t *bp = &lr->lr_blkptr;
1030
1031 zgd->zgd_db = db;
1032 zgd->zgd_bp = bp;
1033
1034 ASSERT(db->db_offset == offset);
1035 ASSERT(db->db_size == size);
1036
1037 error = dmu_sync(zio, lr->lr_common.lrc_txg,
1038 zvol_get_done, zgd);
1039
1040 if (error == 0)
1041 return (0);
1042 }
1043 }
1044
1045 zvol_get_done(zgd, error);
1046
1047 return (error);
1048 }
1049
1050 /*
1051 * zvol_log_write() handles synchronous writes using TX_WRITE ZIL transactions.
1052 *
1053 * We store data in the log buffers if it's small enough.
1054 * Otherwise we will later flush the data out via dmu_sync().
1055 */
1056 ssize_t zvol_immediate_write_sz = 32768;
1057
1058 static void
zvol_log_write(zvol_state_t * zv,dmu_tx_t * tx,offset_t off,ssize_t resid,boolean_t sync)1059 zvol_log_write(zvol_state_t *zv, dmu_tx_t *tx, offset_t off, ssize_t resid,
1060 boolean_t sync)
1061 {
1062 uint32_t blocksize = zv->zv_volblocksize;
1063 zilog_t *zilog = zv->zv_zilog;
1064 itx_wr_state_t write_state;
1065
1066 if (zil_replaying(zilog, tx))
1067 return;
1068
1069 if (zilog->zl_logbias == ZFS_LOGBIAS_THROUGHPUT)
1070 write_state = WR_INDIRECT;
1071 else if (!spa_has_slogs(zilog->zl_spa) &&
1072 resid >= blocksize && blocksize > zvol_immediate_write_sz)
1073 write_state = WR_INDIRECT;
1074 else if (sync)
1075 write_state = WR_COPIED;
1076 else
1077 write_state = WR_NEED_COPY;
1078
1079 while (resid) {
1080 itx_t *itx;
1081 lr_write_t *lr;
1082 itx_wr_state_t wr_state = write_state;
1083 ssize_t len = resid;
1084
1085 if (wr_state == WR_COPIED && resid > ZIL_MAX_COPIED_DATA)
1086 wr_state = WR_NEED_COPY;
1087 else if (wr_state == WR_INDIRECT)
1088 len = MIN(blocksize - P2PHASE(off, blocksize), resid);
1089
1090 itx = zil_itx_create(TX_WRITE, sizeof (*lr) +
1091 (wr_state == WR_COPIED ? len : 0));
1092 lr = (lr_write_t *)&itx->itx_lr;
1093 if (wr_state == WR_COPIED && dmu_read_by_dnode(zv->zv_dn,
1094 off, len, lr + 1, DMU_READ_NO_PREFETCH) != 0) {
1095 zil_itx_destroy(itx);
1096 itx = zil_itx_create(TX_WRITE, sizeof (*lr));
1097 lr = (lr_write_t *)&itx->itx_lr;
1098 wr_state = WR_NEED_COPY;
1099 }
1100
1101 itx->itx_wr_state = wr_state;
1102 lr->lr_foid = ZVOL_OBJ;
1103 lr->lr_offset = off;
1104 lr->lr_length = len;
1105 lr->lr_blkoff = 0;
1106 BP_ZERO(&lr->lr_blkptr);
1107
1108 itx->itx_private = zv;
1109 itx->itx_sync = sync;
1110
1111 zil_itx_assign(zilog, itx, tx);
1112
1113 off += len;
1114 resid -= len;
1115 }
1116 }
1117
1118 static int
zvol_dumpio_vdev(vdev_t * vd,void * addr,uint64_t offset,uint64_t origoffset,uint64_t size,boolean_t doread,boolean_t isdump)1119 zvol_dumpio_vdev(vdev_t *vd, void *addr, uint64_t offset, uint64_t origoffset,
1120 uint64_t size, boolean_t doread, boolean_t isdump)
1121 {
1122 if (doread && !vdev_readable(vd))
1123 return (SET_ERROR(EIO));
1124 if (!doread && !vdev_writeable(vd))
1125 return (SET_ERROR(EIO));
1126 if (vd->vdev_ops->vdev_op_dumpio == NULL)
1127 return (SET_ERROR(EIO));
1128
1129 return (vd->vdev_ops->vdev_op_dumpio(vd, addr, size,
1130 offset, origoffset, doread, isdump));
1131 }
1132
1133 static int
zvol_dumpio(zvol_state_t * zv,void * addr,uint64_t offset,uint64_t size,boolean_t doread,boolean_t isdump)1134 zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
1135 boolean_t doread, boolean_t isdump)
1136 {
1137 vdev_t *vd;
1138 int error;
1139 zvol_extent_t *ze;
1140 spa_t *spa = dmu_objset_spa(zv->zv_objset);
1141
1142 /* Must be sector aligned, and not stradle a block boundary. */
1143 if (P2PHASE(offset, DEV_BSIZE) || P2PHASE(size, DEV_BSIZE) ||
1144 P2BOUNDARY(offset, size, zv->zv_volblocksize)) {
1145 return (SET_ERROR(EINVAL));
1146 }
1147 VERIFY3U(size, <=, zv->zv_volblocksize);
1148
1149 /* Locate the extent this belongs to */
1150 for (ze = list_head(&zv->zv_extents);
1151 ze != NULL && offset >= ze->ze_nblks * zv->zv_volblocksize;
1152 ze = list_next(&zv->zv_extents, ze)) {
1153 offset -= ze->ze_nblks * zv->zv_volblocksize;
1154 }
1155
1156 if (ze == NULL)
1157 return (SET_ERROR(EINVAL));
1158
1159 if (!ddi_in_panic())
1160 spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
1161
1162 vd = vdev_lookup_top(spa, DVA_GET_VDEV(&ze->ze_dva));
1163 offset += DVA_GET_OFFSET(&ze->ze_dva);
1164 error = zvol_dumpio_vdev(vd, addr, offset, DVA_GET_OFFSET(&ze->ze_dva),
1165 size, doread, isdump);
1166
1167 if (!ddi_in_panic())
1168 spa_config_exit(spa, SCL_STATE, FTAG);
1169
1170 return (error);
1171 }
1172
1173 int
zvol_strategy(buf_t * bp)1174 zvol_strategy(buf_t *bp)
1175 {
1176 zfs_soft_state_t *zs = NULL;
1177 zvol_state_t *zv;
1178 uint64_t off, volsize;
1179 size_t resid;
1180 char *addr;
1181 objset_t *os;
1182 int error = 0;
1183 boolean_t doread = !!(bp->b_flags & B_READ);
1184 boolean_t is_dumpified;
1185 boolean_t sync;
1186
1187 if (getminor(bp->b_edev) == 0) {
1188 error = SET_ERROR(EINVAL);
1189 } else {
1190 zs = ddi_get_soft_state(zfsdev_state, getminor(bp->b_edev));
1191 if (zs == NULL)
1192 error = SET_ERROR(ENXIO);
1193 else if (zs->zss_type != ZSST_ZVOL)
1194 error = SET_ERROR(EINVAL);
1195 }
1196
1197 if (error) {
1198 bioerror(bp, error);
1199 biodone(bp);
1200 return (0);
1201 }
1202
1203 zv = zs->zss_data;
1204
1205 if (!(bp->b_flags & B_READ) && (zv->zv_flags & ZVOL_RDONLY)) {
1206 bioerror(bp, EROFS);
1207 biodone(bp);
1208 return (0);
1209 }
1210
1211 off = ldbtob(bp->b_blkno);
1212 volsize = zv->zv_volsize;
1213
1214 os = zv->zv_objset;
1215 ASSERT(os != NULL);
1216
1217 bp_mapin(bp);
1218 addr = bp->b_un.b_addr;
1219 resid = bp->b_bcount;
1220
1221 if (resid > 0 && off >= volsize) {
1222 bioerror(bp, EIO);
1223 biodone(bp);
1224 return (0);
1225 }
1226
1227 is_dumpified = zv->zv_flags & ZVOL_DUMPIFIED;
1228 sync = ((!(bp->b_flags & B_ASYNC) &&
1229 !(zv->zv_flags & ZVOL_WCE)) ||
1230 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)) &&
1231 !doread && !is_dumpified;
1232
1233 smt_begin_unsafe();
1234
1235 /*
1236 * There must be no buffer changes when doing a dmu_sync() because
1237 * we can't change the data whilst calculating the checksum.
1238 */
1239 locked_range_t *lr = rangelock_enter(&zv->zv_rangelock, off, resid,
1240 doread ? RL_READER : RL_WRITER);
1241
1242 while (resid != 0 && off < volsize) {
1243 size_t size = MIN(resid, zvol_maxphys);
1244 if (is_dumpified) {
1245 size = MIN(size, P2END(off, zv->zv_volblocksize) - off);
1246 error = zvol_dumpio(zv, addr, off, size,
1247 doread, B_FALSE);
1248 } else if (doread) {
1249 error = dmu_read(os, ZVOL_OBJ, off, size, addr,
1250 DMU_READ_PREFETCH);
1251 } else {
1252 dmu_tx_t *tx = dmu_tx_create(os);
1253 dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
1254 error = dmu_tx_assign(tx, TXG_WAIT);
1255 if (error) {
1256 dmu_tx_abort(tx);
1257 } else {
1258 dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
1259 zvol_log_write(zv, tx, off, size, sync);
1260 dmu_tx_commit(tx);
1261 }
1262 }
1263 if (error) {
1264 /* convert checksum errors into IO errors */
1265 if (error == ECKSUM)
1266 error = SET_ERROR(EIO);
1267 break;
1268 }
1269 off += size;
1270 addr += size;
1271 resid -= size;
1272 }
1273 rangelock_exit(lr);
1274
1275 if ((bp->b_resid = resid) == bp->b_bcount)
1276 bioerror(bp, off > volsize ? EINVAL : error);
1277
1278 if (sync)
1279 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1280 biodone(bp);
1281
1282 smt_end_unsafe();
1283
1284 return (0);
1285 }
1286
1287 /*
1288 * Set the buffer count to the zvol maximum transfer.
1289 * Using our own routine instead of the default minphys()
1290 * means that for larger writes we write bigger buffers on X86
1291 * (128K instead of 56K) and flush the disk write cache less often
1292 * (every zvol_maxphys - currently 1MB) instead of minphys (currently
1293 * 56K on X86 and 128K on sparc).
1294 */
1295 void
zvol_minphys(struct buf * bp)1296 zvol_minphys(struct buf *bp)
1297 {
1298 if (bp->b_bcount > zvol_maxphys)
1299 bp->b_bcount = zvol_maxphys;
1300 }
1301
1302 int
zvol_dump(dev_t dev,caddr_t addr,daddr_t blkno,int nblocks)1303 zvol_dump(dev_t dev, caddr_t addr, daddr_t blkno, int nblocks)
1304 {
1305 minor_t minor = getminor(dev);
1306 zvol_state_t *zv;
1307 int error = 0;
1308 uint64_t size;
1309 uint64_t boff;
1310 uint64_t resid;
1311
1312 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1313 if (zv == NULL)
1314 return (SET_ERROR(ENXIO));
1315
1316 if ((zv->zv_flags & ZVOL_DUMPIFIED) == 0)
1317 return (SET_ERROR(EINVAL));
1318
1319 boff = ldbtob(blkno);
1320 resid = ldbtob(nblocks);
1321
1322 VERIFY3U(boff + resid, <=, zv->zv_volsize);
1323
1324 while (resid) {
1325 size = MIN(resid, P2END(boff, zv->zv_volblocksize) - boff);
1326 error = zvol_dumpio(zv, addr, boff, size, B_FALSE, B_TRUE);
1327 if (error)
1328 break;
1329 boff += size;
1330 addr += size;
1331 resid -= size;
1332 }
1333
1334 return (error);
1335 }
1336
1337 /*ARGSUSED*/
1338 int
zvol_read(dev_t dev,uio_t * uio,cred_t * cr)1339 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
1340 {
1341 minor_t minor = getminor(dev);
1342 zvol_state_t *zv;
1343 uint64_t volsize;
1344 int error = 0;
1345
1346 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1347 if (zv == NULL)
1348 return (SET_ERROR(ENXIO));
1349
1350 volsize = zv->zv_volsize;
1351 if (uio->uio_resid > 0 &&
1352 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1353 return (SET_ERROR(EIO));
1354
1355 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1356 error = physio(zvol_strategy, NULL, dev, B_READ,
1357 zvol_minphys, uio);
1358 return (error);
1359 }
1360
1361 smt_begin_unsafe();
1362
1363 locked_range_t *lr = rangelock_enter(&zv->zv_rangelock,
1364 uio->uio_loffset, uio->uio_resid, RL_READER);
1365 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1366 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1367
1368 /* don't read past the end */
1369 if (bytes > volsize - uio->uio_loffset)
1370 bytes = volsize - uio->uio_loffset;
1371
1372 error = dmu_read_uio(zv->zv_objset, ZVOL_OBJ, uio, bytes);
1373 if (error) {
1374 /* convert checksum errors into IO errors */
1375 if (error == ECKSUM)
1376 error = SET_ERROR(EIO);
1377 break;
1378 }
1379 }
1380 rangelock_exit(lr);
1381
1382 smt_end_unsafe();
1383
1384 return (error);
1385 }
1386
1387 /*ARGSUSED*/
1388 int
zvol_write(dev_t dev,uio_t * uio,cred_t * cr)1389 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
1390 {
1391 minor_t minor = getminor(dev);
1392 zvol_state_t *zv;
1393 uint64_t volsize;
1394 int error = 0;
1395 boolean_t sync;
1396
1397 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1398 if (zv == NULL)
1399 return (SET_ERROR(ENXIO));
1400
1401 volsize = zv->zv_volsize;
1402 if (uio->uio_resid > 0 &&
1403 (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
1404 return (SET_ERROR(EIO));
1405
1406 if (zv->zv_flags & ZVOL_DUMPIFIED) {
1407 error = physio(zvol_strategy, NULL, dev, B_WRITE,
1408 zvol_minphys, uio);
1409 return (error);
1410 }
1411
1412 smt_begin_unsafe();
1413
1414 sync = !(zv->zv_flags & ZVOL_WCE) ||
1415 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
1416
1417 locked_range_t *lr = rangelock_enter(&zv->zv_rangelock,
1418 uio->uio_loffset, uio->uio_resid, RL_WRITER);
1419 while (uio->uio_resid > 0 && uio->uio_loffset < volsize) {
1420 uint64_t bytes = MIN(uio->uio_resid, DMU_MAX_ACCESS >> 1);
1421 uint64_t off = uio->uio_loffset;
1422 dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
1423
1424 if (bytes > volsize - off) /* don't write past the end */
1425 bytes = volsize - off;
1426
1427 dmu_tx_hold_write_by_dnode(tx, zv->zv_dn, off, bytes);
1428 error = dmu_tx_assign(tx, TXG_WAIT);
1429 if (error) {
1430 dmu_tx_abort(tx);
1431 break;
1432 }
1433 error = dmu_write_uio_dnode(zv->zv_dn, uio, bytes, tx);
1434 if (error == 0)
1435 zvol_log_write(zv, tx, off, bytes, sync);
1436 dmu_tx_commit(tx);
1437
1438 if (error)
1439 break;
1440 }
1441 rangelock_exit(lr);
1442
1443 if (sync)
1444 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1445
1446 smt_end_unsafe();
1447
1448 return (error);
1449 }
1450
1451 int
zvol_getefi(void * arg,int flag,uint64_t vs,uint8_t bs)1452 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
1453 {
1454 struct uuid uuid = EFI_RESERVED;
1455 efi_gpe_t gpe = { 0 };
1456 uint32_t crc;
1457 dk_efi_t efi;
1458 int length;
1459 char *ptr;
1460
1461 if (ddi_copyin(arg, &efi, sizeof (dk_efi_t), flag))
1462 return (SET_ERROR(EFAULT));
1463 ptr = (char *)(uintptr_t)efi.dki_data_64;
1464 length = efi.dki_length;
1465 /*
1466 * Some clients may attempt to request a PMBR for the
1467 * zvol. Currently this interface will return EINVAL to
1468 * such requests. These requests could be supported by
1469 * adding a check for lba == 0 and consing up an appropriate
1470 * PMBR.
1471 */
1472 if (efi.dki_lba < 1 || efi.dki_lba > 2 || length <= 0)
1473 return (SET_ERROR(EINVAL));
1474
1475 gpe.efi_gpe_StartingLBA = LE_64(34ULL);
1476 gpe.efi_gpe_EndingLBA = LE_64((vs >> bs) - 1);
1477 UUID_LE_CONVERT(gpe.efi_gpe_PartitionTypeGUID, uuid);
1478
1479 if (efi.dki_lba == 1) {
1480 efi_gpt_t gpt = { 0 };
1481
1482 gpt.efi_gpt_Signature = LE_64(EFI_SIGNATURE);
1483 gpt.efi_gpt_Revision = LE_32(EFI_VERSION_CURRENT);
1484 gpt.efi_gpt_HeaderSize = LE_32(EFI_HEADER_SIZE);
1485 gpt.efi_gpt_MyLBA = LE_64(1ULL);
1486 gpt.efi_gpt_FirstUsableLBA = LE_64(34ULL);
1487 gpt.efi_gpt_LastUsableLBA = LE_64((vs >> bs) - 1);
1488 gpt.efi_gpt_PartitionEntryLBA = LE_64(2ULL);
1489 gpt.efi_gpt_NumberOfPartitionEntries = LE_32(1);
1490 gpt.efi_gpt_SizeOfPartitionEntry =
1491 LE_32(sizeof (efi_gpe_t));
1492 CRC32(crc, &gpe, sizeof (gpe), -1U, crc32_table);
1493 gpt.efi_gpt_PartitionEntryArrayCRC32 = LE_32(~crc);
1494 CRC32(crc, &gpt, EFI_HEADER_SIZE, -1U, crc32_table);
1495 gpt.efi_gpt_HeaderCRC32 = LE_32(~crc);
1496 if (ddi_copyout(&gpt, ptr, MIN(sizeof (gpt), length),
1497 flag))
1498 return (SET_ERROR(EFAULT));
1499 ptr += sizeof (gpt);
1500 length -= sizeof (gpt);
1501 }
1502 if (length > 0 && ddi_copyout(&gpe, ptr, MIN(sizeof (gpe),
1503 length), flag))
1504 return (SET_ERROR(EFAULT));
1505 return (0);
1506 }
1507
1508 /*
1509 * BEGIN entry points to allow external callers access to the volume.
1510 */
1511 /*
1512 * Return the volume parameters needed for access from an external caller.
1513 * These values are invariant as long as the volume is held open.
1514 */
1515 int
zvol_get_volume_params(minor_t minor,uint64_t * blksize,uint64_t * max_xfer_len,void ** minor_hdl,void ** objset_hdl,void ** zil_hdl,void ** rl_hdl,void ** dnode_hdl)1516 zvol_get_volume_params(minor_t minor, uint64_t *blksize,
1517 uint64_t *max_xfer_len, void **minor_hdl, void **objset_hdl, void **zil_hdl,
1518 void **rl_hdl, void **dnode_hdl)
1519 {
1520 zvol_state_t *zv;
1521
1522 zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
1523 if (zv == NULL)
1524 return (SET_ERROR(ENXIO));
1525 if (zv->zv_flags & ZVOL_DUMPIFIED)
1526 return (SET_ERROR(ENXIO));
1527
1528 ASSERT(blksize && max_xfer_len && minor_hdl &&
1529 objset_hdl && zil_hdl && rl_hdl && dnode_hdl);
1530
1531 *blksize = zv->zv_volblocksize;
1532 *max_xfer_len = (uint64_t)zvol_maxphys;
1533 *minor_hdl = zv;
1534 *objset_hdl = zv->zv_objset;
1535 *zil_hdl = zv->zv_zilog;
1536 *rl_hdl = &zv->zv_rangelock;
1537 *dnode_hdl = zv->zv_dn;
1538 return (0);
1539 }
1540
1541 /*
1542 * Return the current volume size to an external caller.
1543 * The size can change while the volume is open.
1544 */
1545 uint64_t
zvol_get_volume_size(void * minor_hdl)1546 zvol_get_volume_size(void *minor_hdl)
1547 {
1548 zvol_state_t *zv = minor_hdl;
1549
1550 return (zv->zv_volsize);
1551 }
1552
1553 /*
1554 * Return the current WCE setting to an external caller.
1555 * The WCE setting can change while the volume is open.
1556 */
1557 int
zvol_get_volume_wce(void * minor_hdl)1558 zvol_get_volume_wce(void *minor_hdl)
1559 {
1560 zvol_state_t *zv = minor_hdl;
1561
1562 return ((zv->zv_flags & ZVOL_WCE) ? 1 : 0);
1563 }
1564
1565 /*
1566 * Entry point for external callers to zvol_log_write
1567 */
1568 void
zvol_log_write_minor(void * minor_hdl,dmu_tx_t * tx,offset_t off,ssize_t resid,boolean_t sync)1569 zvol_log_write_minor(void *minor_hdl, dmu_tx_t *tx, offset_t off, ssize_t resid,
1570 boolean_t sync)
1571 {
1572 zvol_state_t *zv = minor_hdl;
1573
1574 zvol_log_write(zv, tx, off, resid, sync);
1575 }
1576 /*
1577 * END entry points to allow external callers access to the volume.
1578 */
1579
1580 /*
1581 * Log a DKIOCFREE/free-long-range to the ZIL with TX_TRUNCATE.
1582 */
1583 static void
zvol_log_truncate(zvol_state_t * zv,dmu_tx_t * tx,uint64_t off,uint64_t len,boolean_t sync)1584 zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off, uint64_t len,
1585 boolean_t sync)
1586 {
1587 itx_t *itx;
1588 lr_truncate_t *lr;
1589 zilog_t *zilog = zv->zv_zilog;
1590
1591 if (zil_replaying(zilog, tx))
1592 return;
1593
1594 itx = zil_itx_create(TX_TRUNCATE, sizeof (*lr));
1595 lr = (lr_truncate_t *)&itx->itx_lr;
1596 lr->lr_foid = ZVOL_OBJ;
1597 lr->lr_offset = off;
1598 lr->lr_length = len;
1599
1600 itx->itx_sync = sync;
1601 zil_itx_assign(zilog, itx, tx);
1602 }
1603
1604 /*
1605 * Dirtbag ioctls to support mkfs(8) for UFS filesystems. See dkio(4I).
1606 * Also a dirtbag dkio ioctl for unmap/free-block functionality.
1607 */
1608 /*ARGSUSED*/
1609 int
zvol_ioctl(dev_t dev,int cmd,intptr_t arg,int flag,cred_t * cr,int * rvalp)1610 zvol_ioctl(dev_t dev, int cmd, intptr_t arg, int flag, cred_t *cr, int *rvalp)
1611 {
1612 zvol_state_t *zv;
1613 struct dk_callback *dkc;
1614 int i, error = 0;
1615 locked_range_t *lr;
1616
1617 mutex_enter(&zfsdev_state_lock);
1618
1619 zv = zfsdev_get_soft_state(getminor(dev), ZSST_ZVOL);
1620
1621 if (zv == NULL) {
1622 mutex_exit(&zfsdev_state_lock);
1623 return (SET_ERROR(ENXIO));
1624 }
1625 ASSERT(zv->zv_total_opens > 0);
1626
1627 switch (cmd) {
1628
1629 case DKIOCINFO:
1630 {
1631 struct dk_cinfo dki;
1632
1633 bzero(&dki, sizeof (dki));
1634 (void) strcpy(dki.dki_cname, "zvol");
1635 (void) strcpy(dki.dki_dname, "zvol");
1636 dki.dki_ctype = DKC_UNKNOWN;
1637 dki.dki_unit = getminor(dev);
1638 dki.dki_maxtransfer =
1639 1 << (SPA_OLD_MAXBLOCKSHIFT - zv->zv_min_bs);
1640 mutex_exit(&zfsdev_state_lock);
1641 if (ddi_copyout(&dki, (void *)arg, sizeof (dki), flag))
1642 error = SET_ERROR(EFAULT);
1643 return (error);
1644 }
1645
1646 case DKIOCGMEDIAINFO:
1647 {
1648 struct dk_minfo dkm;
1649
1650 bzero(&dkm, sizeof (dkm));
1651 dkm.dki_lbsize = 1U << zv->zv_min_bs;
1652 dkm.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1653 dkm.dki_media_type = DK_UNKNOWN;
1654 mutex_exit(&zfsdev_state_lock);
1655 if (ddi_copyout(&dkm, (void *)arg, sizeof (dkm), flag))
1656 error = SET_ERROR(EFAULT);
1657 return (error);
1658 }
1659
1660 case DKIOCGMEDIAINFOEXT:
1661 {
1662 struct dk_minfo_ext dkmext;
1663 size_t len;
1664
1665 bzero(&dkmext, sizeof (dkmext));
1666 dkmext.dki_lbsize = 1U << zv->zv_min_bs;
1667 dkmext.dki_pbsize = zv->zv_volblocksize;
1668 dkmext.dki_capacity = zv->zv_volsize >> zv->zv_min_bs;
1669 dkmext.dki_media_type = DK_UNKNOWN;
1670 mutex_exit(&zfsdev_state_lock);
1671
1672 switch (ddi_model_convert_from(flag & FMODELS)) {
1673 case DDI_MODEL_ILP32:
1674 len = sizeof (struct dk_minfo_ext32);
1675 break;
1676 default:
1677 len = sizeof (struct dk_minfo_ext);
1678 break;
1679 }
1680
1681 if (ddi_copyout(&dkmext, (void *)arg, len, flag))
1682 error = SET_ERROR(EFAULT);
1683 return (error);
1684 }
1685
1686 case DKIOCGETEFI:
1687 {
1688 uint64_t vs = zv->zv_volsize;
1689 uint8_t bs = zv->zv_min_bs;
1690
1691 mutex_exit(&zfsdev_state_lock);
1692 error = zvol_getefi((void *)arg, flag, vs, bs);
1693 return (error);
1694 }
1695
1696 case DKIOCFLUSHWRITECACHE:
1697 dkc = (struct dk_callback *)arg;
1698 mutex_exit(&zfsdev_state_lock);
1699
1700 smt_begin_unsafe();
1701
1702 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1703 if ((flag & FKIOCTL) && dkc != NULL && dkc->dkc_callback) {
1704 (*dkc->dkc_callback)(dkc->dkc_cookie, error);
1705 error = 0;
1706 }
1707
1708 smt_end_unsafe();
1709
1710 return (error);
1711
1712 case DKIOCGETWCE:
1713 {
1714 int wce = (zv->zv_flags & ZVOL_WCE) ? 1 : 0;
1715 if (ddi_copyout(&wce, (void *)arg, sizeof (int),
1716 flag))
1717 error = SET_ERROR(EFAULT);
1718 break;
1719 }
1720 case DKIOCSETWCE:
1721 {
1722 int wce;
1723 if (ddi_copyin((void *)arg, &wce, sizeof (int),
1724 flag)) {
1725 error = SET_ERROR(EFAULT);
1726 break;
1727 }
1728 if (wce) {
1729 zv->zv_flags |= ZVOL_WCE;
1730 mutex_exit(&zfsdev_state_lock);
1731 } else {
1732 zv->zv_flags &= ~ZVOL_WCE;
1733 mutex_exit(&zfsdev_state_lock);
1734 smt_begin_unsafe();
1735 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1736 smt_end_unsafe();
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 lr = rangelock_enter(&zv->zv_rangelock, 0, zv->zv_volsize,
1752 RL_WRITER);
1753 error = zvol_dumpify(zv);
1754 rangelock_exit(lr);
1755 break;
1756
1757 case DKIOCDUMPFINI:
1758 if (!(zv->zv_flags & ZVOL_DUMPIFIED))
1759 break;
1760 lr = rangelock_enter(&zv->zv_rangelock, 0, zv->zv_volsize,
1761 RL_WRITER);
1762 error = zvol_dump_fini(zv);
1763 rangelock_exit(lr);
1764 break;
1765
1766 case DKIOCFREE:
1767 {
1768 dkioc_free_list_t *dfl;
1769 dmu_tx_t *tx;
1770
1771 if (!zvol_unmap_enabled)
1772 break;
1773
1774 if (!(flag & FKIOCTL)) {
1775 error = dfl_copyin((void *)arg, &dfl, flag, KM_SLEEP);
1776 if (error != 0)
1777 break;
1778 } else {
1779 dfl = (dkioc_free_list_t *)arg;
1780 ASSERT3U(dfl->dfl_num_exts, <=, DFL_COPYIN_MAX_EXTS);
1781 if (dfl->dfl_num_exts > DFL_COPYIN_MAX_EXTS) {
1782 error = SET_ERROR(EINVAL);
1783 break;
1784 }
1785 }
1786
1787 mutex_exit(&zfsdev_state_lock);
1788
1789 smt_begin_unsafe();
1790
1791 for (int i = 0; i < dfl->dfl_num_exts; i++) {
1792 uint64_t start = dfl->dfl_exts[i].dfle_start,
1793 length = dfl->dfl_exts[i].dfle_length,
1794 end = start + length;
1795
1796 /*
1797 * Apply Postel's Law to length-checking. If they
1798 * overshoot, just blank out until the end, if there's
1799 * a need to blank out anything.
1800 */
1801 if (start >= zv->zv_volsize)
1802 continue; /* No need to do anything... */
1803 if (end > zv->zv_volsize) {
1804 end = DMU_OBJECT_END;
1805 length = end - start;
1806 }
1807
1808 lr = rangelock_enter(&zv->zv_rangelock, start, length,
1809 RL_WRITER);
1810 tx = dmu_tx_create(zv->zv_objset);
1811 error = dmu_tx_assign(tx, TXG_WAIT);
1812 if (error != 0) {
1813 dmu_tx_abort(tx);
1814 } else {
1815 zvol_log_truncate(zv, tx, start, length,
1816 B_TRUE);
1817 dmu_tx_commit(tx);
1818 error = dmu_free_long_range(zv->zv_objset,
1819 ZVOL_OBJ, start, length);
1820 }
1821
1822 rangelock_exit(lr);
1823
1824 if (error != 0)
1825 break;
1826 }
1827
1828 /*
1829 * If the write-cache is disabled, 'sync' property
1830 * is set to 'always', or if the caller is asking for
1831 * a synchronous free, commit this operation to the zil.
1832 * This will sync any previous uncommitted writes to the
1833 * zvol object.
1834 * Can be overridden by the zvol_unmap_sync_enabled tunable.
1835 */
1836 if ((error == 0) && zvol_unmap_sync_enabled &&
1837 (!(zv->zv_flags & ZVOL_WCE) ||
1838 (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS) ||
1839 (dfl->dfl_flags & DF_WAIT_SYNC))) {
1840 zil_commit(zv->zv_zilog, ZVOL_OBJ);
1841 }
1842
1843 if (!(flag & FKIOCTL))
1844 dfl_free(dfl);
1845
1846 smt_end_unsafe();
1847
1848 return (error);
1849 }
1850
1851 case DKIOC_CANFREE:
1852 i = zvol_unmap_enabled ? 1 : 0;
1853 if (ddi_copyout(&i, (void *)arg, sizeof (int), flag) != 0) {
1854 error = EFAULT;
1855 } else {
1856 error = 0;
1857 }
1858 break;
1859
1860 default:
1861 error = SET_ERROR(ENOTTY);
1862 break;
1863
1864 }
1865 mutex_exit(&zfsdev_state_lock);
1866 return (error);
1867 }
1868
1869 int
zvol_busy(void)1870 zvol_busy(void)
1871 {
1872 return (zvol_minors != 0);
1873 }
1874
1875 void
zvol_init(void)1876 zvol_init(void)
1877 {
1878 VERIFY(ddi_soft_state_init(&zfsdev_state, sizeof (zfs_soft_state_t),
1879 1) == 0);
1880 mutex_init(&zfsdev_state_lock, NULL, MUTEX_DEFAULT, NULL);
1881 }
1882
1883 void
zvol_fini(void)1884 zvol_fini(void)
1885 {
1886 mutex_destroy(&zfsdev_state_lock);
1887 ddi_soft_state_fini(&zfsdev_state);
1888 }
1889
1890 /*ARGSUSED*/
1891 static int
zfs_mvdev_dump_feature_check(void * arg,dmu_tx_t * tx)1892 zfs_mvdev_dump_feature_check(void *arg, dmu_tx_t *tx)
1893 {
1894 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1895
1896 if (spa_feature_is_active(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1897 return (1);
1898 return (0);
1899 }
1900
1901 /*ARGSUSED*/
1902 static void
zfs_mvdev_dump_activate_feature_sync(void * arg,dmu_tx_t * tx)1903 zfs_mvdev_dump_activate_feature_sync(void *arg, dmu_tx_t *tx)
1904 {
1905 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1906
1907 spa_feature_incr(spa, SPA_FEATURE_MULTI_VDEV_CRASH_DUMP, tx);
1908 }
1909
1910 static int
zvol_dump_init(zvol_state_t * zv,boolean_t resize)1911 zvol_dump_init(zvol_state_t *zv, boolean_t resize)
1912 {
1913 dmu_tx_t *tx;
1914 int error;
1915 objset_t *os = zv->zv_objset;
1916 spa_t *spa = dmu_objset_spa(os);
1917 vdev_t *vd = spa->spa_root_vdev;
1918 nvlist_t *nv = NULL;
1919 uint64_t version = spa_version(spa);
1920 uint64_t checksum, compress, refresrv, vbs, dedup;
1921
1922 ASSERT(MUTEX_HELD(&zfsdev_state_lock));
1923 ASSERT(vd->vdev_ops == &vdev_root_ops);
1924
1925 error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ, 0,
1926 DMU_OBJECT_END);
1927 if (error != 0)
1928 return (error);
1929 /* wait for dmu_free_long_range to actually free the blocks */
1930 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
1931
1932 /*
1933 * If the pool on which the dump device is being initialized has more
1934 * than one child vdev, check that the MULTI_VDEV_CRASH_DUMP feature is
1935 * enabled. If so, bump that feature's counter to indicate that the
1936 * feature is active. We also check the vdev type to handle the
1937 * following case:
1938 * # zpool create test raidz disk1 disk2 disk3
1939 * Now have spa_root_vdev->vdev_children == 1 (the raidz vdev),
1940 * the raidz vdev itself has 3 children.
1941 */
1942 if (vd->vdev_children > 1 || vd->vdev_ops == &vdev_raidz_ops) {
1943 if (!spa_feature_is_enabled(spa,
1944 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP))
1945 return (SET_ERROR(ENOTSUP));
1946 (void) dsl_sync_task(spa_name(spa),
1947 zfs_mvdev_dump_feature_check,
1948 zfs_mvdev_dump_activate_feature_sync, NULL,
1949 2, ZFS_SPACE_CHECK_RESERVED);
1950 }
1951
1952 if (!resize) {
1953 error = dsl_prop_get_integer(zv->zv_name,
1954 zfs_prop_to_name(ZFS_PROP_COMPRESSION), &compress, NULL);
1955 if (error == 0) {
1956 error = dsl_prop_get_integer(zv->zv_name,
1957 zfs_prop_to_name(ZFS_PROP_CHECKSUM), &checksum,
1958 NULL);
1959 }
1960 if (error == 0) {
1961 error = dsl_prop_get_integer(zv->zv_name,
1962 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
1963 &refresrv, NULL);
1964 }
1965 if (error == 0) {
1966 error = dsl_prop_get_integer(zv->zv_name,
1967 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), &vbs,
1968 NULL);
1969 }
1970 if (version >= SPA_VERSION_DEDUP && error == 0) {
1971 error = dsl_prop_get_integer(zv->zv_name,
1972 zfs_prop_to_name(ZFS_PROP_DEDUP), &dedup, NULL);
1973 }
1974 }
1975 if (error != 0)
1976 return (error);
1977
1978 tx = dmu_tx_create(os);
1979 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
1980 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
1981 error = dmu_tx_assign(tx, TXG_WAIT);
1982 if (error != 0) {
1983 dmu_tx_abort(tx);
1984 return (error);
1985 }
1986
1987 /*
1988 * If we are resizing the dump device then we only need to
1989 * update the refreservation to match the newly updated
1990 * zvolsize. Otherwise, we save off the original state of the
1991 * zvol so that we can restore them if the zvol is ever undumpified.
1992 */
1993 if (resize) {
1994 error = zap_update(os, ZVOL_ZAP_OBJ,
1995 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
1996 &zv->zv_volsize, tx);
1997 } else {
1998 error = zap_update(os, ZVOL_ZAP_OBJ,
1999 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1,
2000 &compress, tx);
2001 if (error == 0) {
2002 error = zap_update(os, ZVOL_ZAP_OBJ,
2003 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1,
2004 &checksum, tx);
2005 }
2006 if (error == 0) {
2007 error = zap_update(os, ZVOL_ZAP_OBJ,
2008 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1,
2009 &refresrv, tx);
2010 }
2011 if (error == 0) {
2012 error = zap_update(os, ZVOL_ZAP_OBJ,
2013 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1,
2014 &vbs, tx);
2015 }
2016 if (error == 0) {
2017 error = dmu_object_set_blocksize(
2018 os, ZVOL_OBJ, SPA_OLD_MAXBLOCKSIZE, 0, tx);
2019 }
2020 if (version >= SPA_VERSION_DEDUP && error == 0) {
2021 error = zap_update(os, ZVOL_ZAP_OBJ,
2022 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1,
2023 &dedup, tx);
2024 }
2025 if (error == 0)
2026 zv->zv_volblocksize = SPA_OLD_MAXBLOCKSIZE;
2027 }
2028 dmu_tx_commit(tx);
2029
2030 /*
2031 * We only need update the zvol's property if we are initializing
2032 * the dump area for the first time.
2033 */
2034 if (error == 0 && !resize) {
2035 /*
2036 * If MULTI_VDEV_CRASH_DUMP is active, use the NOPARITY checksum
2037 * function. Otherwise, use the old default -- OFF.
2038 */
2039 checksum = spa_feature_is_active(spa,
2040 SPA_FEATURE_MULTI_VDEV_CRASH_DUMP) ? ZIO_CHECKSUM_NOPARITY :
2041 ZIO_CHECKSUM_OFF;
2042
2043 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2044 VERIFY(nvlist_add_uint64(nv,
2045 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 0) == 0);
2046 VERIFY(nvlist_add_uint64(nv,
2047 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
2048 ZIO_COMPRESS_OFF) == 0);
2049 VERIFY(nvlist_add_uint64(nv,
2050 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
2051 checksum) == 0);
2052 if (version >= SPA_VERSION_DEDUP) {
2053 VERIFY(nvlist_add_uint64(nv,
2054 zfs_prop_to_name(ZFS_PROP_DEDUP),
2055 ZIO_CHECKSUM_OFF) == 0);
2056 }
2057
2058 error = zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2059 nv, NULL);
2060 nvlist_free(nv);
2061 }
2062
2063 /* Allocate the space for the dump */
2064 if (error == 0)
2065 error = zvol_prealloc(zv);
2066 return (error);
2067 }
2068
2069 static int
zvol_dumpify(zvol_state_t * zv)2070 zvol_dumpify(zvol_state_t *zv)
2071 {
2072 int error = 0;
2073 uint64_t dumpsize = 0;
2074 dmu_tx_t *tx;
2075 objset_t *os = zv->zv_objset;
2076
2077 if (zv->zv_flags & ZVOL_RDONLY)
2078 return (SET_ERROR(EROFS));
2079
2080 if (os->os_encrypted)
2081 return (SET_ERROR(ENOTSUP));
2082
2083 if (zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE,
2084 8, 1, &dumpsize) != 0 || dumpsize != zv->zv_volsize) {
2085 boolean_t resize = (dumpsize > 0);
2086
2087 if ((error = zvol_dump_init(zv, resize)) != 0) {
2088 (void) zvol_dump_fini(zv);
2089 return (error);
2090 }
2091 }
2092
2093 /*
2094 * Build up our lba mapping.
2095 */
2096 error = zvol_get_lbas(zv);
2097 if (error) {
2098 (void) zvol_dump_fini(zv);
2099 return (error);
2100 }
2101
2102 tx = dmu_tx_create(os);
2103 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2104 error = dmu_tx_assign(tx, TXG_WAIT);
2105 if (error) {
2106 dmu_tx_abort(tx);
2107 (void) zvol_dump_fini(zv);
2108 return (error);
2109 }
2110
2111 zv->zv_flags |= ZVOL_DUMPIFIED;
2112 error = zap_update(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, 8, 1,
2113 &zv->zv_volsize, tx);
2114 dmu_tx_commit(tx);
2115
2116 if (error) {
2117 (void) zvol_dump_fini(zv);
2118 return (error);
2119 }
2120
2121 txg_wait_synced(dmu_objset_pool(os), 0);
2122 return (0);
2123 }
2124
2125 static int
zvol_dump_fini(zvol_state_t * zv)2126 zvol_dump_fini(zvol_state_t *zv)
2127 {
2128 dmu_tx_t *tx;
2129 objset_t *os = zv->zv_objset;
2130 nvlist_t *nv;
2131 int error = 0;
2132 uint64_t checksum, compress, refresrv, vbs, dedup;
2133 uint64_t version = spa_version(dmu_objset_spa(zv->zv_objset));
2134
2135 /*
2136 * Attempt to restore the zvol back to its pre-dumpified state.
2137 * This is a best-effort attempt as it's possible that not all
2138 * of these properties were initialized during the dumpify process
2139 * (i.e. error during zvol_dump_init).
2140 */
2141
2142 tx = dmu_tx_create(os);
2143 dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
2144 error = dmu_tx_assign(tx, TXG_WAIT);
2145 if (error) {
2146 dmu_tx_abort(tx);
2147 return (error);
2148 }
2149 (void) zap_remove(os, ZVOL_ZAP_OBJ, ZVOL_DUMPSIZE, tx);
2150 dmu_tx_commit(tx);
2151
2152 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2153 zfs_prop_to_name(ZFS_PROP_CHECKSUM), 8, 1, &checksum);
2154 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2155 zfs_prop_to_name(ZFS_PROP_COMPRESSION), 8, 1, &compress);
2156 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2157 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), 8, 1, &refresrv);
2158 (void) zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2159 zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE), 8, 1, &vbs);
2160
2161 VERIFY(nvlist_alloc(&nv, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2162 (void) nvlist_add_uint64(nv,
2163 zfs_prop_to_name(ZFS_PROP_CHECKSUM), checksum);
2164 (void) nvlist_add_uint64(nv,
2165 zfs_prop_to_name(ZFS_PROP_COMPRESSION), compress);
2166 (void) nvlist_add_uint64(nv,
2167 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), refresrv);
2168 if (version >= SPA_VERSION_DEDUP &&
2169 zap_lookup(zv->zv_objset, ZVOL_ZAP_OBJ,
2170 zfs_prop_to_name(ZFS_PROP_DEDUP), 8, 1, &dedup) == 0) {
2171 (void) nvlist_add_uint64(nv,
2172 zfs_prop_to_name(ZFS_PROP_DEDUP), dedup);
2173 }
2174 (void) zfs_set_prop_nvlist(zv->zv_name, ZPROP_SRC_LOCAL,
2175 nv, NULL);
2176 nvlist_free(nv);
2177
2178 zvol_free_extents(zv);
2179 zv->zv_flags &= ~ZVOL_DUMPIFIED;
2180 (void) dmu_free_long_range(os, ZVOL_OBJ, 0, DMU_OBJECT_END);
2181 /* wait for dmu_free_long_range to actually free the blocks */
2182 txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
2183 tx = dmu_tx_create(os);
2184 dmu_tx_hold_bonus(tx, ZVOL_OBJ);
2185 error = dmu_tx_assign(tx, TXG_WAIT);
2186 if (error) {
2187 dmu_tx_abort(tx);
2188 return (error);
2189 }
2190 if (dmu_object_set_blocksize(os, ZVOL_OBJ, vbs, 0, tx) == 0)
2191 zv->zv_volblocksize = vbs;
2192 dmu_tx_commit(tx);
2193
2194 return (0);
2195 }
2196