xref: /freebsd/stand/libsa/zfs/zfsimpl.c (revision 8aac90f18aef7c9eea906c3ff9a001ca7b94f375)
1 /*-
2  * Copyright (c) 2007 Doug Rabson
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 /*
28  *	Stand-alone ZFS file reader.
29  */
30 
31 #include <stdbool.h>
32 #include <sys/endian.h>
33 #include <sys/stat.h>
34 #include <sys/stdint.h>
35 #include <sys/list.h>
36 #include <sys/zfs_bootenv.h>
37 #include <machine/_inttypes.h>
38 
39 #include "zfsimpl.h"
40 #include "zfssubr.c"
41 
42 #ifdef HAS_ZSTD_ZFS
43 extern int zstd_init(void);
44 #endif
45 
46 struct zfsmount {
47 	char			*path;
48 	const spa_t		*spa;
49 	objset_phys_t		objset;
50 	uint64_t		rootobj;
51 	STAILQ_ENTRY(zfsmount)	next;
52 };
53 
54 typedef STAILQ_HEAD(zfs_mnt_list, zfsmount) zfs_mnt_list_t;
55 static zfs_mnt_list_t zfsmount = STAILQ_HEAD_INITIALIZER(zfsmount);
56 
57 /*
58  * The indirect_child_t represents the vdev that we will read from, when we
59  * need to read all copies of the data (e.g. for scrub or reconstruction).
60  * For plain (non-mirror) top-level vdevs (i.e. is_vdev is not a mirror),
61  * ic_vdev is the same as is_vdev.  However, for mirror top-level vdevs,
62  * ic_vdev is a child of the mirror.
63  */
64 typedef struct indirect_child {
65 	void *ic_data;
66 	vdev_t *ic_vdev;
67 } indirect_child_t;
68 
69 /*
70  * The indirect_split_t represents one mapped segment of an i/o to the
71  * indirect vdev. For non-split (contiguously-mapped) blocks, there will be
72  * only one indirect_split_t, with is_split_offset==0 and is_size==io_size.
73  * For split blocks, there will be several of these.
74  */
75 typedef struct indirect_split {
76 	list_node_t is_node; /* link on iv_splits */
77 
78 	/*
79 	 * is_split_offset is the offset into the i/o.
80 	 * This is the sum of the previous splits' is_size's.
81 	 */
82 	uint64_t is_split_offset;
83 
84 	vdev_t *is_vdev; /* top-level vdev */
85 	uint64_t is_target_offset; /* offset on is_vdev */
86 	uint64_t is_size;
87 	int is_children; /* number of entries in is_child[] */
88 
89 	/*
90 	 * is_good_child is the child that we are currently using to
91 	 * attempt reconstruction.
92 	 */
93 	int is_good_child;
94 
95 	indirect_child_t is_child[1]; /* variable-length */
96 } indirect_split_t;
97 
98 /*
99  * The indirect_vsd_t is associated with each i/o to the indirect vdev.
100  * It is the "Vdev-Specific Data" in the zio_t's io_vsd.
101  */
102 typedef struct indirect_vsd {
103 	boolean_t iv_split_block;
104 	boolean_t iv_reconstruct;
105 
106 	list_t iv_splits; /* list of indirect_split_t's */
107 } indirect_vsd_t;
108 
109 /*
110  * List of all vdevs, chained through v_alllink.
111  */
112 static vdev_list_t zfs_vdevs;
113 
114 /*
115  * List of supported read-incompatible ZFS features.  Do not add here features
116  * marked as ZFEATURE_FLAG_READONLY_COMPAT, they are irrelevant for read-only!
117  */
118 static const char *features_for_read[] = {
119 	"com.datto:bookmark_v2",
120 	"com.datto:encryption",
121 	"com.delphix:bookmark_written",
122 	"com.delphix:device_removal",
123 	"com.delphix:embedded_data",
124 	"com.delphix:extensible_dataset",
125 	"com.delphix:head_errlog",
126 	"com.delphix:hole_birth",
127 	"com.joyent:multi_vdev_crash_dump",
128 	"com.klarasystems:vdev_zaps_v2",
129 	"org.freebsd:zstd_compress",
130 	"org.illumos:lz4_compress",
131 	"org.illumos:sha512",
132 	"org.illumos:skein",
133 	"org.open-zfs:large_blocks",
134 	"org.openzfs:blake3",
135 	"org.zfsonlinux:large_dnode",
136 	NULL
137 };
138 
139 /*
140  * List of all pools, chained through spa_link.
141  */
142 static spa_list_t zfs_pools;
143 
144 static const dnode_phys_t *dnode_cache_obj;
145 static uint64_t dnode_cache_bn;
146 static char *dnode_cache_buf;
147 
148 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
149 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
150 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
151 static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode,
152     const char *name, uint64_t integer_size, uint64_t num_integers,
153     void *value);
154 static int objset_get_dnode(const spa_t *, const objset_phys_t *, uint64_t,
155     dnode_phys_t *);
156 static int dnode_read(const spa_t *, const dnode_phys_t *, off_t, void *,
157     size_t);
158 static int vdev_indirect_read(vdev_t *, const blkptr_t *, void *, off_t,
159     size_t);
160 static int vdev_mirror_read(vdev_t *, const blkptr_t *, void *, off_t, size_t);
161 vdev_indirect_mapping_t *vdev_indirect_mapping_open(spa_t *, objset_phys_t *,
162     uint64_t);
163 vdev_indirect_mapping_entry_phys_t *
164     vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *, uint64_t,
165     uint64_t, uint64_t *);
166 
167 static void
168 zfs_init(void)
169 {
170 	STAILQ_INIT(&zfs_vdevs);
171 	STAILQ_INIT(&zfs_pools);
172 
173 	dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
174 
175 	zfs_init_crc();
176 #ifdef HAS_ZSTD_ZFS
177 	zstd_init();
178 #endif
179 }
180 
181 static int
182 nvlist_check_features_for_read(nvlist_t *nvl)
183 {
184 	nvlist_t *features = NULL;
185 	nvs_data_t *data;
186 	nvp_header_t *nvp;
187 	nv_string_t *nvp_name;
188 	int rc;
189 
190 	rc = nvlist_find(nvl, ZPOOL_CONFIG_FEATURES_FOR_READ,
191 	    DATA_TYPE_NVLIST, NULL, &features, NULL);
192 	switch (rc) {
193 	case 0:
194 		break;		/* Continue with checks */
195 
196 	case ENOENT:
197 		return (0);	/* All features are disabled */
198 
199 	default:
200 		return (rc);	/* Error while reading nvlist */
201 	}
202 
203 	data = (nvs_data_t *)features->nv_data;
204 	nvp = &data->nvl_pair;	/* first pair in nvlist */
205 
206 	while (nvp->encoded_size != 0 && nvp->decoded_size != 0) {
207 		int i, found;
208 
209 		nvp_name = (nv_string_t *)((uintptr_t)nvp + sizeof(*nvp));
210 		found = 0;
211 
212 		for (i = 0; features_for_read[i] != NULL; i++) {
213 			if (memcmp(nvp_name->nv_data, features_for_read[i],
214 			    nvp_name->nv_size) == 0) {
215 				found = 1;
216 				break;
217 			}
218 		}
219 
220 		if (!found) {
221 			printf("ZFS: unsupported feature: %.*s\n",
222 			    nvp_name->nv_size, nvp_name->nv_data);
223 			rc = EIO;
224 		}
225 		nvp = (nvp_header_t *)((uint8_t *)nvp + nvp->encoded_size);
226 	}
227 	nvlist_destroy(features);
228 
229 	return (rc);
230 }
231 
232 static int
233 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
234     off_t offset, size_t size)
235 {
236 	size_t psize;
237 	int rc;
238 
239 	if (vdev->v_phys_read == NULL)
240 		return (ENOTSUP);
241 
242 	if (bp) {
243 		psize = BP_GET_PSIZE(bp);
244 	} else {
245 		psize = size;
246 	}
247 
248 	rc = vdev->v_phys_read(vdev, vdev->v_priv, offset, buf, psize);
249 	if (rc == 0) {
250 		if (bp != NULL)
251 			rc = zio_checksum_verify(vdev->v_spa, bp, buf);
252 	}
253 
254 	return (rc);
255 }
256 
257 static int
258 vdev_write_phys(vdev_t *vdev, void *buf, off_t offset, size_t size)
259 {
260 	if (vdev->v_phys_write == NULL)
261 		return (ENOTSUP);
262 
263 	return (vdev->v_phys_write(vdev, offset, buf, size));
264 }
265 
266 typedef struct remap_segment {
267 	vdev_t *rs_vd;
268 	uint64_t rs_offset;
269 	uint64_t rs_asize;
270 	uint64_t rs_split_offset;
271 	list_node_t rs_node;
272 } remap_segment_t;
273 
274 static remap_segment_t *
275 rs_alloc(vdev_t *vd, uint64_t offset, uint64_t asize, uint64_t split_offset)
276 {
277 	remap_segment_t *rs = malloc(sizeof (remap_segment_t));
278 
279 	if (rs != NULL) {
280 		rs->rs_vd = vd;
281 		rs->rs_offset = offset;
282 		rs->rs_asize = asize;
283 		rs->rs_split_offset = split_offset;
284 	}
285 
286 	return (rs);
287 }
288 
289 vdev_indirect_mapping_t *
290 vdev_indirect_mapping_open(spa_t *spa, objset_phys_t *os,
291     uint64_t mapping_object)
292 {
293 	vdev_indirect_mapping_t *vim;
294 	vdev_indirect_mapping_phys_t *vim_phys;
295 	int rc;
296 
297 	vim = calloc(1, sizeof (*vim));
298 	if (vim == NULL)
299 		return (NULL);
300 
301 	vim->vim_dn = calloc(1, sizeof (*vim->vim_dn));
302 	if (vim->vim_dn == NULL) {
303 		free(vim);
304 		return (NULL);
305 	}
306 
307 	rc = objset_get_dnode(spa, os, mapping_object, vim->vim_dn);
308 	if (rc != 0) {
309 		free(vim->vim_dn);
310 		free(vim);
311 		return (NULL);
312 	}
313 
314 	vim->vim_spa = spa;
315 	vim->vim_phys = malloc(sizeof (*vim->vim_phys));
316 	if (vim->vim_phys == NULL) {
317 		free(vim->vim_dn);
318 		free(vim);
319 		return (NULL);
320 	}
321 
322 	vim_phys = (vdev_indirect_mapping_phys_t *)DN_BONUS(vim->vim_dn);
323 	*vim->vim_phys = *vim_phys;
324 
325 	vim->vim_objset = os;
326 	vim->vim_object = mapping_object;
327 	vim->vim_entries = NULL;
328 
329 	vim->vim_havecounts =
330 	    (vim->vim_dn->dn_bonuslen > VDEV_INDIRECT_MAPPING_SIZE_V0);
331 
332 	return (vim);
333 }
334 
335 /*
336  * Compare an offset with an indirect mapping entry; there are three
337  * possible scenarios:
338  *
339  *     1. The offset is "less than" the mapping entry; meaning the
340  *        offset is less than the source offset of the mapping entry. In
341  *        this case, there is no overlap between the offset and the
342  *        mapping entry and -1 will be returned.
343  *
344  *     2. The offset is "greater than" the mapping entry; meaning the
345  *        offset is greater than the mapping entry's source offset plus
346  *        the entry's size. In this case, there is no overlap between
347  *        the offset and the mapping entry and 1 will be returned.
348  *
349  *        NOTE: If the offset is actually equal to the entry's offset
350  *        plus size, this is considered to be "greater" than the entry,
351  *        and this case applies (i.e. 1 will be returned). Thus, the
352  *        entry's "range" can be considered to be inclusive at its
353  *        start, but exclusive at its end: e.g. [src, src + size).
354  *
355  *     3. The last case to consider is if the offset actually falls
356  *        within the mapping entry's range. If this is the case, the
357  *        offset is considered to be "equal to" the mapping entry and
358  *        0 will be returned.
359  *
360  *        NOTE: If the offset is equal to the entry's source offset,
361  *        this case applies and 0 will be returned. If the offset is
362  *        equal to the entry's source plus its size, this case does
363  *        *not* apply (see "NOTE" above for scenario 2), and 1 will be
364  *        returned.
365  */
366 static int
367 dva_mapping_overlap_compare(const void *v_key, const void *v_array_elem)
368 {
369 	const uint64_t *key = v_key;
370 	const vdev_indirect_mapping_entry_phys_t *array_elem =
371 	    v_array_elem;
372 	uint64_t src_offset = DVA_MAPPING_GET_SRC_OFFSET(array_elem);
373 
374 	if (*key < src_offset) {
375 		return (-1);
376 	} else if (*key < src_offset + DVA_GET_ASIZE(&array_elem->vimep_dst)) {
377 		return (0);
378 	} else {
379 		return (1);
380 	}
381 }
382 
383 /*
384  * Return array entry.
385  */
386 static vdev_indirect_mapping_entry_phys_t *
387 vdev_indirect_mapping_entry(vdev_indirect_mapping_t *vim, uint64_t index)
388 {
389 	uint64_t size;
390 	off_t offset = 0;
391 	int rc;
392 
393 	if (vim->vim_phys->vimp_num_entries == 0)
394 		return (NULL);
395 
396 	if (vim->vim_entries == NULL) {
397 		uint64_t bsize;
398 
399 		bsize = vim->vim_dn->dn_datablkszsec << SPA_MINBLOCKSHIFT;
400 		size = vim->vim_phys->vimp_num_entries *
401 		    sizeof (*vim->vim_entries);
402 		if (size > bsize) {
403 			size = bsize / sizeof (*vim->vim_entries);
404 			size *= sizeof (*vim->vim_entries);
405 		}
406 		vim->vim_entries = malloc(size);
407 		if (vim->vim_entries == NULL)
408 			return (NULL);
409 		vim->vim_num_entries = size / sizeof (*vim->vim_entries);
410 		offset = index * sizeof (*vim->vim_entries);
411 	}
412 
413 	/* We have data in vim_entries */
414 	if (offset == 0) {
415 		if (index >= vim->vim_entry_offset &&
416 		    index <= vim->vim_entry_offset + vim->vim_num_entries) {
417 			index -= vim->vim_entry_offset;
418 			return (&vim->vim_entries[index]);
419 		}
420 		offset = index * sizeof (*vim->vim_entries);
421 	}
422 
423 	vim->vim_entry_offset = index;
424 	size = vim->vim_num_entries * sizeof (*vim->vim_entries);
425 	rc = dnode_read(vim->vim_spa, vim->vim_dn, offset, vim->vim_entries,
426 	    size);
427 	if (rc != 0) {
428 		/* Read error, invalidate vim_entries. */
429 		free(vim->vim_entries);
430 		vim->vim_entries = NULL;
431 		return (NULL);
432 	}
433 	index -= vim->vim_entry_offset;
434 	return (&vim->vim_entries[index]);
435 }
436 
437 /*
438  * Returns the mapping entry for the given offset.
439  *
440  * It's possible that the given offset will not be in the mapping table
441  * (i.e. no mapping entries contain this offset), in which case, the
442  * return value depends on the "next_if_missing" parameter.
443  *
444  * If the offset is not found in the table and "next_if_missing" is
445  * B_FALSE, then NULL will always be returned. The behavior is intended
446  * to allow consumers to get the entry corresponding to the offset
447  * parameter, iff the offset overlaps with an entry in the table.
448  *
449  * If the offset is not found in the table and "next_if_missing" is
450  * B_TRUE, then the entry nearest to the given offset will be returned,
451  * such that the entry's source offset is greater than the offset
452  * passed in (i.e. the "next" mapping entry in the table is returned, if
453  * the offset is missing from the table). If there are no entries whose
454  * source offset is greater than the passed in offset, NULL is returned.
455  */
456 static vdev_indirect_mapping_entry_phys_t *
457 vdev_indirect_mapping_entry_for_offset(vdev_indirect_mapping_t *vim,
458     uint64_t offset)
459 {
460 	ASSERT(vim->vim_phys->vimp_num_entries > 0);
461 
462 	vdev_indirect_mapping_entry_phys_t *entry;
463 
464 	uint64_t last = vim->vim_phys->vimp_num_entries - 1;
465 	uint64_t base = 0;
466 
467 	/*
468 	 * We don't define these inside of the while loop because we use
469 	 * their value in the case that offset isn't in the mapping.
470 	 */
471 	uint64_t mid;
472 	int result;
473 
474 	while (last >= base) {
475 		mid = base + ((last - base) >> 1);
476 
477 		entry = vdev_indirect_mapping_entry(vim, mid);
478 		if (entry == NULL)
479 			break;
480 		result = dva_mapping_overlap_compare(&offset, entry);
481 
482 		if (result == 0) {
483 			break;
484 		} else if (result < 0) {
485 			last = mid - 1;
486 		} else {
487 			base = mid + 1;
488 		}
489 	}
490 	return (entry);
491 }
492 
493 /*
494  * Given an indirect vdev and an extent on that vdev, it duplicates the
495  * physical entries of the indirect mapping that correspond to the extent
496  * to a new array and returns a pointer to it. In addition, copied_entries
497  * is populated with the number of mapping entries that were duplicated.
498  *
499  * Finally, since we are doing an allocation, it is up to the caller to
500  * free the array allocated in this function.
501  */
502 vdev_indirect_mapping_entry_phys_t *
503 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *vd, uint64_t offset,
504     uint64_t asize, uint64_t *copied_entries)
505 {
506 	vdev_indirect_mapping_entry_phys_t *duplicate_mappings = NULL;
507 	vdev_indirect_mapping_t *vim = vd->v_mapping;
508 	uint64_t entries = 0;
509 
510 	vdev_indirect_mapping_entry_phys_t *first_mapping =
511 	    vdev_indirect_mapping_entry_for_offset(vim, offset);
512 	ASSERT3P(first_mapping, !=, NULL);
513 
514 	vdev_indirect_mapping_entry_phys_t *m = first_mapping;
515 	while (asize > 0) {
516 		uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
517 		uint64_t inner_offset = offset - DVA_MAPPING_GET_SRC_OFFSET(m);
518 		uint64_t inner_size = MIN(asize, size - inner_offset);
519 
520 		offset += inner_size;
521 		asize -= inner_size;
522 		entries++;
523 		m++;
524 	}
525 
526 	size_t copy_length = entries * sizeof (*first_mapping);
527 	duplicate_mappings = malloc(copy_length);
528 	if (duplicate_mappings != NULL)
529 		bcopy(first_mapping, duplicate_mappings, copy_length);
530 	else
531 		entries = 0;
532 
533 	*copied_entries = entries;
534 
535 	return (duplicate_mappings);
536 }
537 
538 static vdev_t *
539 vdev_lookup_top(spa_t *spa, uint64_t vdev)
540 {
541 	vdev_t *rvd;
542 	vdev_list_t *vlist;
543 
544 	vlist = &spa->spa_root_vdev->v_children;
545 	STAILQ_FOREACH(rvd, vlist, v_childlink)
546 		if (rvd->v_id == vdev)
547 			break;
548 
549 	return (rvd);
550 }
551 
552 /*
553  * This is a callback for vdev_indirect_remap() which allocates an
554  * indirect_split_t for each split segment and adds it to iv_splits.
555  */
556 static void
557 vdev_indirect_gather_splits(uint64_t split_offset, vdev_t *vd, uint64_t offset,
558     uint64_t size, void *arg)
559 {
560 	int n = 1;
561 	zio_t *zio = arg;
562 	indirect_vsd_t *iv = zio->io_vsd;
563 
564 	if (vd->v_read == vdev_indirect_read)
565 		return;
566 
567 	if (vd->v_read == vdev_mirror_read)
568 		n = vd->v_nchildren;
569 
570 	indirect_split_t *is =
571 	    malloc(offsetof(indirect_split_t, is_child[n]));
572 	if (is == NULL) {
573 		zio->io_error = ENOMEM;
574 		return;
575 	}
576 	bzero(is, offsetof(indirect_split_t, is_child[n]));
577 
578 	is->is_children = n;
579 	is->is_size = size;
580 	is->is_split_offset = split_offset;
581 	is->is_target_offset = offset;
582 	is->is_vdev = vd;
583 
584 	/*
585 	 * Note that we only consider multiple copies of the data for
586 	 * *mirror* vdevs.  We don't for "replacing" or "spare" vdevs, even
587 	 * though they use the same ops as mirror, because there's only one
588 	 * "good" copy under the replacing/spare.
589 	 */
590 	if (vd->v_read == vdev_mirror_read) {
591 		int i = 0;
592 		vdev_t *kid;
593 
594 		STAILQ_FOREACH(kid, &vd->v_children, v_childlink) {
595 			is->is_child[i++].ic_vdev = kid;
596 		}
597 	} else {
598 		is->is_child[0].ic_vdev = vd;
599 	}
600 
601 	list_insert_tail(&iv->iv_splits, is);
602 }
603 
604 static void
605 vdev_indirect_remap(vdev_t *vd, uint64_t offset, uint64_t asize, void *arg)
606 {
607 	list_t stack;
608 	spa_t *spa = vd->v_spa;
609 	zio_t *zio = arg;
610 	remap_segment_t *rs;
611 
612 	list_create(&stack, sizeof (remap_segment_t),
613 	    offsetof(remap_segment_t, rs_node));
614 
615 	rs = rs_alloc(vd, offset, asize, 0);
616 	if (rs == NULL) {
617 		printf("vdev_indirect_remap: out of memory.\n");
618 		zio->io_error = ENOMEM;
619 	}
620 	for (; rs != NULL; rs = list_remove_head(&stack)) {
621 		vdev_t *v = rs->rs_vd;
622 		uint64_t num_entries = 0;
623 		/* vdev_indirect_mapping_t *vim = v->v_mapping; */
624 		vdev_indirect_mapping_entry_phys_t *mapping =
625 		    vdev_indirect_mapping_duplicate_adjacent_entries(v,
626 		    rs->rs_offset, rs->rs_asize, &num_entries);
627 
628 		if (num_entries == 0)
629 			zio->io_error = ENOMEM;
630 
631 		for (uint64_t i = 0; i < num_entries; i++) {
632 			vdev_indirect_mapping_entry_phys_t *m = &mapping[i];
633 			uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
634 			uint64_t dst_offset = DVA_GET_OFFSET(&m->vimep_dst);
635 			uint64_t dst_vdev = DVA_GET_VDEV(&m->vimep_dst);
636 			uint64_t inner_offset = rs->rs_offset -
637 			    DVA_MAPPING_GET_SRC_OFFSET(m);
638 			uint64_t inner_size =
639 			    MIN(rs->rs_asize, size - inner_offset);
640 			vdev_t *dst_v = vdev_lookup_top(spa, dst_vdev);
641 
642 			if (dst_v->v_read == vdev_indirect_read) {
643 				remap_segment_t *o;
644 
645 				o = rs_alloc(dst_v, dst_offset + inner_offset,
646 				    inner_size, rs->rs_split_offset);
647 				if (o == NULL) {
648 					printf("vdev_indirect_remap: "
649 					    "out of memory.\n");
650 					zio->io_error = ENOMEM;
651 					break;
652 				}
653 
654 				list_insert_head(&stack, o);
655 			}
656 			vdev_indirect_gather_splits(rs->rs_split_offset, dst_v,
657 			    dst_offset + inner_offset,
658 			    inner_size, arg);
659 
660 			/*
661 			 * vdev_indirect_gather_splits can have memory
662 			 * allocation error, we can not recover from it.
663 			 */
664 			if (zio->io_error != 0)
665 				break;
666 			rs->rs_offset += inner_size;
667 			rs->rs_asize -= inner_size;
668 			rs->rs_split_offset += inner_size;
669 		}
670 
671 		free(mapping);
672 		free(rs);
673 		if (zio->io_error != 0)
674 			break;
675 	}
676 
677 	list_destroy(&stack);
678 }
679 
680 static void
681 vdev_indirect_map_free(zio_t *zio)
682 {
683 	indirect_vsd_t *iv = zio->io_vsd;
684 	indirect_split_t *is;
685 
686 	while ((is = list_head(&iv->iv_splits)) != NULL) {
687 		for (int c = 0; c < is->is_children; c++) {
688 			indirect_child_t *ic = &is->is_child[c];
689 			free(ic->ic_data);
690 		}
691 		list_remove(&iv->iv_splits, is);
692 		free(is);
693 	}
694 	free(iv);
695 }
696 
697 static int
698 vdev_indirect_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
699     off_t offset, size_t bytes)
700 {
701 	zio_t zio;
702 	spa_t *spa = vdev->v_spa;
703 	indirect_vsd_t *iv;
704 	indirect_split_t *first;
705 	int rc = EIO;
706 
707 	iv = calloc(1, sizeof(*iv));
708 	if (iv == NULL)
709 		return (ENOMEM);
710 
711 	list_create(&iv->iv_splits,
712 	    sizeof (indirect_split_t), offsetof(indirect_split_t, is_node));
713 
714 	bzero(&zio, sizeof(zio));
715 	zio.io_spa = spa;
716 	zio.io_bp = (blkptr_t *)bp;
717 	zio.io_data = buf;
718 	zio.io_size = bytes;
719 	zio.io_offset = offset;
720 	zio.io_vd = vdev;
721 	zio.io_vsd = iv;
722 
723 	if (vdev->v_mapping == NULL) {
724 		vdev_indirect_config_t *vic;
725 
726 		vic = &vdev->vdev_indirect_config;
727 		vdev->v_mapping = vdev_indirect_mapping_open(spa,
728 		    spa->spa_mos, vic->vic_mapping_object);
729 	}
730 
731 	vdev_indirect_remap(vdev, offset, bytes, &zio);
732 	if (zio.io_error != 0)
733 		return (zio.io_error);
734 
735 	first = list_head(&iv->iv_splits);
736 	if (first->is_size == zio.io_size) {
737 		/*
738 		 * This is not a split block; we are pointing to the entire
739 		 * data, which will checksum the same as the original data.
740 		 * Pass the BP down so that the child i/o can verify the
741 		 * checksum, and try a different location if available
742 		 * (e.g. on a mirror).
743 		 *
744 		 * While this special case could be handled the same as the
745 		 * general (split block) case, doing it this way ensures
746 		 * that the vast majority of blocks on indirect vdevs
747 		 * (which are not split) are handled identically to blocks
748 		 * on non-indirect vdevs.  This allows us to be less strict
749 		 * about performance in the general (but rare) case.
750 		 */
751 		rc = first->is_vdev->v_read(first->is_vdev, zio.io_bp,
752 		    zio.io_data, first->is_target_offset, bytes);
753 	} else {
754 		iv->iv_split_block = B_TRUE;
755 		/*
756 		 * Read one copy of each split segment, from the
757 		 * top-level vdev.  Since we don't know the
758 		 * checksum of each split individually, the child
759 		 * zio can't ensure that we get the right data.
760 		 * E.g. if it's a mirror, it will just read from a
761 		 * random (healthy) leaf vdev.  We have to verify
762 		 * the checksum in vdev_indirect_io_done().
763 		 */
764 		for (indirect_split_t *is = list_head(&iv->iv_splits);
765 		    is != NULL; is = list_next(&iv->iv_splits, is)) {
766 			char *ptr = zio.io_data;
767 
768 			rc = is->is_vdev->v_read(is->is_vdev, zio.io_bp,
769 			    ptr + is->is_split_offset, is->is_target_offset,
770 			    is->is_size);
771 		}
772 		if (zio_checksum_verify(spa, zio.io_bp, zio.io_data))
773 			rc = ECKSUM;
774 		else
775 			rc = 0;
776 	}
777 
778 	vdev_indirect_map_free(&zio);
779 	if (rc == 0)
780 		rc = zio.io_error;
781 
782 	return (rc);
783 }
784 
785 static int
786 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
787     off_t offset, size_t bytes)
788 {
789 
790 	return (vdev_read_phys(vdev, bp, buf,
791 	    offset + VDEV_LABEL_START_SIZE, bytes));
792 }
793 
794 static int
795 vdev_missing_read(vdev_t *vdev __unused, const blkptr_t *bp __unused,
796     void *buf __unused, off_t offset __unused, size_t bytes __unused)
797 {
798 
799 	return (ENOTSUP);
800 }
801 
802 static int
803 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
804     off_t offset, size_t bytes)
805 {
806 	vdev_t *kid;
807 	int rc;
808 
809 	rc = EIO;
810 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
811 		if (kid->v_state != VDEV_STATE_HEALTHY)
812 			continue;
813 		rc = kid->v_read(kid, bp, buf, offset, bytes);
814 		if (!rc)
815 			return (0);
816 	}
817 
818 	return (rc);
819 }
820 
821 static int
822 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
823     off_t offset, size_t bytes)
824 {
825 	vdev_t *kid;
826 
827 	/*
828 	 * Here we should have two kids:
829 	 * First one which is the one we are replacing and we can trust
830 	 * only this one to have valid data, but it might not be present.
831 	 * Second one is that one we are replacing with. It is most likely
832 	 * healthy, but we can't trust it has needed data, so we won't use it.
833 	 */
834 	kid = STAILQ_FIRST(&vdev->v_children);
835 	if (kid == NULL)
836 		return (EIO);
837 	if (kid->v_state != VDEV_STATE_HEALTHY)
838 		return (EIO);
839 	return (kid->v_read(kid, bp, buf, offset, bytes));
840 }
841 
842 static vdev_t *
843 vdev_find(uint64_t guid)
844 {
845 	vdev_t *vdev;
846 
847 	STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
848 		if (vdev->v_guid == guid)
849 			return (vdev);
850 
851 	return (0);
852 }
853 
854 static vdev_t *
855 vdev_create(uint64_t guid, vdev_read_t *_read)
856 {
857 	vdev_t *vdev;
858 	vdev_indirect_config_t *vic;
859 
860 	vdev = calloc(1, sizeof(vdev_t));
861 	if (vdev != NULL) {
862 		STAILQ_INIT(&vdev->v_children);
863 		vdev->v_guid = guid;
864 		vdev->v_read = _read;
865 
866 		/*
867 		 * root vdev has no read function, we use this fact to
868 		 * skip setting up data we do not need for root vdev.
869 		 * We only point root vdev from spa.
870 		 */
871 		if (_read != NULL) {
872 			vic = &vdev->vdev_indirect_config;
873 			vic->vic_prev_indirect_vdev = UINT64_MAX;
874 			STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
875 		}
876 	}
877 
878 	return (vdev);
879 }
880 
881 static void
882 vdev_set_initial_state(vdev_t *vdev, const nvlist_t *nvlist)
883 {
884 	uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
885 	uint64_t is_log;
886 
887 	is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
888 	is_log = 0;
889 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
890 	    &is_offline, NULL);
891 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
892 	    &is_removed, NULL);
893 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
894 	    &is_faulted, NULL);
895 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64,
896 	    NULL, &is_degraded, NULL);
897 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64,
898 	    NULL, &isnt_present, NULL);
899 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, NULL,
900 	    &is_log, NULL);
901 
902 	if (is_offline != 0)
903 		vdev->v_state = VDEV_STATE_OFFLINE;
904 	else if (is_removed != 0)
905 		vdev->v_state = VDEV_STATE_REMOVED;
906 	else if (is_faulted != 0)
907 		vdev->v_state = VDEV_STATE_FAULTED;
908 	else if (is_degraded != 0)
909 		vdev->v_state = VDEV_STATE_DEGRADED;
910 	else if (isnt_present != 0)
911 		vdev->v_state = VDEV_STATE_CANT_OPEN;
912 
913 	vdev->v_islog = is_log != 0;
914 }
915 
916 static int
917 vdev_init(uint64_t guid, const nvlist_t *nvlist, vdev_t **vdevp)
918 {
919 	uint64_t id, ashift, asize, nparity;
920 	const char *path;
921 	const char *type;
922 	int len, pathlen;
923 	char *name;
924 	vdev_t *vdev;
925 
926 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id,
927 	    NULL) ||
928 	    nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING, NULL,
929 	    &type, &len)) {
930 		return (ENOENT);
931 	}
932 
933 	if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 &&
934 	    memcmp(type, VDEV_TYPE_DISK, len) != 0 &&
935 #ifdef ZFS_TEST
936 	    memcmp(type, VDEV_TYPE_FILE, len) != 0 &&
937 #endif
938 	    memcmp(type, VDEV_TYPE_RAIDZ, len) != 0 &&
939 	    memcmp(type, VDEV_TYPE_INDIRECT, len) != 0 &&
940 	    memcmp(type, VDEV_TYPE_REPLACING, len) != 0 &&
941 	    memcmp(type, VDEV_TYPE_HOLE, len) != 0) {
942 		printf("ZFS: can only boot from disk, mirror, raidz1, "
943 		    "raidz2 and raidz3 vdevs, got: %.*s\n", len, type);
944 		return (EIO);
945 	}
946 
947 	if (memcmp(type, VDEV_TYPE_MIRROR, len) == 0)
948 		vdev = vdev_create(guid, vdev_mirror_read);
949 	else if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0)
950 		vdev = vdev_create(guid, vdev_raidz_read);
951 	else if (memcmp(type, VDEV_TYPE_REPLACING, len) == 0)
952 		vdev = vdev_create(guid, vdev_replacing_read);
953 	else if (memcmp(type, VDEV_TYPE_INDIRECT, len) == 0) {
954 		vdev_indirect_config_t *vic;
955 
956 		vdev = vdev_create(guid, vdev_indirect_read);
957 		if (vdev != NULL) {
958 			vdev->v_state = VDEV_STATE_HEALTHY;
959 			vic = &vdev->vdev_indirect_config;
960 
961 			nvlist_find(nvlist,
962 			    ZPOOL_CONFIG_INDIRECT_OBJECT,
963 			    DATA_TYPE_UINT64,
964 			    NULL, &vic->vic_mapping_object, NULL);
965 			nvlist_find(nvlist,
966 			    ZPOOL_CONFIG_INDIRECT_BIRTHS,
967 			    DATA_TYPE_UINT64,
968 			    NULL, &vic->vic_births_object, NULL);
969 			nvlist_find(nvlist,
970 			    ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
971 			    DATA_TYPE_UINT64,
972 			    NULL, &vic->vic_prev_indirect_vdev, NULL);
973 		}
974 	} else if (memcmp(type, VDEV_TYPE_HOLE, len) == 0) {
975 		vdev = vdev_create(guid, vdev_missing_read);
976 	} else {
977 		vdev = vdev_create(guid, vdev_disk_read);
978 	}
979 
980 	if (vdev == NULL)
981 		return (ENOMEM);
982 
983 	vdev_set_initial_state(vdev, nvlist);
984 	vdev->v_id = id;
985 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
986 	    DATA_TYPE_UINT64, NULL, &ashift, NULL) == 0)
987 		vdev->v_ashift = ashift;
988 
989 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE,
990 	    DATA_TYPE_UINT64, NULL, &asize, NULL) == 0) {
991 		vdev->v_psize = asize +
992 		    VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
993 	}
994 
995 	if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
996 	    DATA_TYPE_UINT64, NULL, &nparity, NULL) == 0)
997 		vdev->v_nparity = nparity;
998 
999 	if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
1000 	    DATA_TYPE_STRING, NULL, &path, &pathlen) == 0) {
1001 		char prefix[] = "/dev/";
1002 
1003 		len = strlen(prefix);
1004 		if (len < pathlen && memcmp(path, prefix, len) == 0) {
1005 			path += len;
1006 			pathlen -= len;
1007 		}
1008 		name = malloc(pathlen + 1);
1009 		bcopy(path, name, pathlen);
1010 		name[pathlen] = '\0';
1011 		vdev->v_name = name;
1012 	} else {
1013 		name = NULL;
1014 		if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) {
1015 			if (vdev->v_nparity < 1 ||
1016 			    vdev->v_nparity > 3) {
1017 				printf("ZFS: invalid raidz parity: %d\n",
1018 				    vdev->v_nparity);
1019 				return (EIO);
1020 			}
1021 			(void) asprintf(&name, "%.*s%d-%" PRIu64, len, type,
1022 			    vdev->v_nparity, id);
1023 		} else {
1024 			(void) asprintf(&name, "%.*s-%" PRIu64, len, type, id);
1025 		}
1026 		vdev->v_name = name;
1027 	}
1028 	*vdevp = vdev;
1029 	return (0);
1030 }
1031 
1032 /*
1033  * Find slot for vdev. We return either NULL to signal to use
1034  * STAILQ_INSERT_HEAD, or we return link element to be used with
1035  * STAILQ_INSERT_AFTER.
1036  */
1037 static vdev_t *
1038 vdev_find_previous(vdev_t *top_vdev, vdev_t *vdev)
1039 {
1040 	vdev_t *v, *previous;
1041 
1042 	if (STAILQ_EMPTY(&top_vdev->v_children))
1043 		return (NULL);
1044 
1045 	previous = NULL;
1046 	STAILQ_FOREACH(v, &top_vdev->v_children, v_childlink) {
1047 		if (v->v_id > vdev->v_id)
1048 			return (previous);
1049 
1050 		if (v->v_id == vdev->v_id)
1051 			return (v);
1052 
1053 		if (v->v_id < vdev->v_id)
1054 			previous = v;
1055 	}
1056 	return (previous);
1057 }
1058 
1059 static size_t
1060 vdev_child_count(vdev_t *vdev)
1061 {
1062 	vdev_t *v;
1063 	size_t count;
1064 
1065 	count = 0;
1066 	STAILQ_FOREACH(v, &vdev->v_children, v_childlink) {
1067 		count++;
1068 	}
1069 	return (count);
1070 }
1071 
1072 /*
1073  * Insert vdev into top_vdev children list. List is ordered by v_id.
1074  */
1075 static void
1076 vdev_insert(vdev_t *top_vdev, vdev_t *vdev)
1077 {
1078 	vdev_t *previous;
1079 	size_t count;
1080 
1081 	/*
1082 	 * The top level vdev can appear in random order, depending how
1083 	 * the firmware is presenting the disk devices.
1084 	 * However, we will insert vdev to create list ordered by v_id,
1085 	 * so we can use either STAILQ_INSERT_HEAD or STAILQ_INSERT_AFTER
1086 	 * as STAILQ does not have insert before.
1087 	 */
1088 	previous = vdev_find_previous(top_vdev, vdev);
1089 
1090 	if (previous == NULL) {
1091 		STAILQ_INSERT_HEAD(&top_vdev->v_children, vdev, v_childlink);
1092 	} else if (previous->v_id == vdev->v_id) {
1093 		/*
1094 		 * This vdev was configured from label config,
1095 		 * do not insert duplicate.
1096 		 */
1097 		return;
1098 	} else {
1099 		STAILQ_INSERT_AFTER(&top_vdev->v_children, previous, vdev,
1100 		    v_childlink);
1101 	}
1102 
1103 	count = vdev_child_count(top_vdev);
1104 	if (top_vdev->v_nchildren < count)
1105 		top_vdev->v_nchildren = count;
1106 }
1107 
1108 static int
1109 vdev_from_nvlist(spa_t *spa, uint64_t top_guid, const nvlist_t *nvlist)
1110 {
1111 	vdev_t *top_vdev, *vdev;
1112 	nvlist_t **kids = NULL;
1113 	int rc, nkids;
1114 
1115 	/* Get top vdev. */
1116 	top_vdev = vdev_find(top_guid);
1117 	if (top_vdev == NULL) {
1118 		rc = vdev_init(top_guid, nvlist, &top_vdev);
1119 		if (rc != 0)
1120 			return (rc);
1121 		top_vdev->v_spa = spa;
1122 		top_vdev->v_top = top_vdev;
1123 		vdev_insert(spa->spa_root_vdev, top_vdev);
1124 	}
1125 
1126 	/* Add children if there are any. */
1127 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1128 	    &nkids, &kids, NULL);
1129 	if (rc == 0) {
1130 		for (int i = 0; i < nkids; i++) {
1131 			uint64_t guid;
1132 
1133 			rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID,
1134 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1135 			if (rc != 0)
1136 				goto done;
1137 
1138 			rc = vdev_init(guid, kids[i], &vdev);
1139 			if (rc != 0)
1140 				goto done;
1141 
1142 			vdev->v_spa = spa;
1143 			vdev->v_top = top_vdev;
1144 			vdev_insert(top_vdev, vdev);
1145 		}
1146 	} else {
1147 		/*
1148 		 * When there are no children, nvlist_find() does return
1149 		 * error, reset it because leaf devices have no children.
1150 		 */
1151 		rc = 0;
1152 	}
1153 done:
1154 	if (kids != NULL) {
1155 		for (int i = 0; i < nkids; i++)
1156 			nvlist_destroy(kids[i]);
1157 		free(kids);
1158 	}
1159 
1160 	return (rc);
1161 }
1162 
1163 static int
1164 vdev_init_from_label(spa_t *spa, const nvlist_t *nvlist)
1165 {
1166 	uint64_t pool_guid, top_guid;
1167 	nvlist_t *vdevs;
1168 	int rc;
1169 
1170 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1171 	    NULL, &pool_guid, NULL) ||
1172 	    nvlist_find(nvlist, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64,
1173 	    NULL, &top_guid, NULL) ||
1174 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1175 	    NULL, &vdevs, NULL)) {
1176 		printf("ZFS: can't find vdev details\n");
1177 		return (ENOENT);
1178 	}
1179 
1180 	rc = vdev_from_nvlist(spa, top_guid, vdevs);
1181 	nvlist_destroy(vdevs);
1182 	return (rc);
1183 }
1184 
1185 static void
1186 vdev_set_state(vdev_t *vdev)
1187 {
1188 	vdev_t *kid;
1189 	int good_kids;
1190 	int bad_kids;
1191 
1192 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1193 		vdev_set_state(kid);
1194 	}
1195 
1196 	/*
1197 	 * A mirror or raidz is healthy if all its kids are healthy. A
1198 	 * mirror is degraded if any of its kids is healthy; a raidz
1199 	 * is degraded if at most nparity kids are offline.
1200 	 */
1201 	if (STAILQ_FIRST(&vdev->v_children)) {
1202 		good_kids = 0;
1203 		bad_kids = 0;
1204 		STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1205 			if (kid->v_state == VDEV_STATE_HEALTHY)
1206 				good_kids++;
1207 			else
1208 				bad_kids++;
1209 		}
1210 		if (bad_kids == 0) {
1211 			vdev->v_state = VDEV_STATE_HEALTHY;
1212 		} else {
1213 			if (vdev->v_read == vdev_mirror_read) {
1214 				if (good_kids) {
1215 					vdev->v_state = VDEV_STATE_DEGRADED;
1216 				} else {
1217 					vdev->v_state = VDEV_STATE_OFFLINE;
1218 				}
1219 			} else if (vdev->v_read == vdev_raidz_read) {
1220 				if (bad_kids > vdev->v_nparity) {
1221 					vdev->v_state = VDEV_STATE_OFFLINE;
1222 				} else {
1223 					vdev->v_state = VDEV_STATE_DEGRADED;
1224 				}
1225 			}
1226 		}
1227 	}
1228 }
1229 
1230 static int
1231 vdev_update_from_nvlist(uint64_t top_guid, const nvlist_t *nvlist)
1232 {
1233 	vdev_t *vdev;
1234 	nvlist_t **kids = NULL;
1235 	int rc, nkids;
1236 
1237 	/* Update top vdev. */
1238 	vdev = vdev_find(top_guid);
1239 	if (vdev != NULL)
1240 		vdev_set_initial_state(vdev, nvlist);
1241 
1242 	/* Update children if there are any. */
1243 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1244 	    &nkids, &kids, NULL);
1245 	if (rc == 0) {
1246 		for (int i = 0; i < nkids; i++) {
1247 			uint64_t guid;
1248 
1249 			rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID,
1250 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1251 			if (rc != 0)
1252 				break;
1253 
1254 			vdev = vdev_find(guid);
1255 			if (vdev != NULL)
1256 				vdev_set_initial_state(vdev, kids[i]);
1257 		}
1258 	} else {
1259 		rc = 0;
1260 	}
1261 	if (kids != NULL) {
1262 		for (int i = 0; i < nkids; i++)
1263 			nvlist_destroy(kids[i]);
1264 		free(kids);
1265 	}
1266 
1267 	return (rc);
1268 }
1269 
1270 static int
1271 vdev_init_from_nvlist(spa_t *spa, const nvlist_t *nvlist)
1272 {
1273 	uint64_t pool_guid, vdev_children;
1274 	nvlist_t *vdevs = NULL, **kids = NULL;
1275 	int rc, nkids;
1276 
1277 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1278 	    NULL, &pool_guid, NULL) ||
1279 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_CHILDREN, DATA_TYPE_UINT64,
1280 	    NULL, &vdev_children, NULL) ||
1281 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1282 	    NULL, &vdevs, NULL)) {
1283 		printf("ZFS: can't find vdev details\n");
1284 		return (ENOENT);
1285 	}
1286 
1287 	/* Wrong guid?! */
1288 	if (spa->spa_guid != pool_guid) {
1289 		nvlist_destroy(vdevs);
1290 		return (EINVAL);
1291 	}
1292 
1293 	spa->spa_root_vdev->v_nchildren = vdev_children;
1294 
1295 	rc = nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1296 	    &nkids, &kids, NULL);
1297 	nvlist_destroy(vdevs);
1298 
1299 	/*
1300 	 * MOS config has at least one child for root vdev.
1301 	 */
1302 	if (rc != 0)
1303 		return (rc);
1304 
1305 	for (int i = 0; i < nkids; i++) {
1306 		uint64_t guid;
1307 		vdev_t *vdev;
1308 
1309 		rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1310 		    NULL, &guid, NULL);
1311 		if (rc != 0)
1312 			break;
1313 		vdev = vdev_find(guid);
1314 		/*
1315 		 * Top level vdev is missing, create it.
1316 		 */
1317 		if (vdev == NULL)
1318 			rc = vdev_from_nvlist(spa, guid, kids[i]);
1319 		else
1320 			rc = vdev_update_from_nvlist(guid, kids[i]);
1321 		if (rc != 0)
1322 			break;
1323 	}
1324 	if (kids != NULL) {
1325 		for (int i = 0; i < nkids; i++)
1326 			nvlist_destroy(kids[i]);
1327 		free(kids);
1328 	}
1329 
1330 	/*
1331 	 * Re-evaluate top-level vdev state.
1332 	 */
1333 	vdev_set_state(spa->spa_root_vdev);
1334 
1335 	return (rc);
1336 }
1337 
1338 static spa_t *
1339 spa_find_by_guid(uint64_t guid)
1340 {
1341 	spa_t *spa;
1342 
1343 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1344 		if (spa->spa_guid == guid)
1345 			return (spa);
1346 
1347 	return (NULL);
1348 }
1349 
1350 static spa_t *
1351 spa_find_by_name(const char *name)
1352 {
1353 	spa_t *spa;
1354 
1355 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1356 		if (strcmp(spa->spa_name, name) == 0)
1357 			return (spa);
1358 
1359 	return (NULL);
1360 }
1361 
1362 static spa_t *
1363 spa_create(uint64_t guid, const char *name)
1364 {
1365 	spa_t *spa;
1366 
1367 	if ((spa = calloc(1, sizeof(spa_t))) == NULL)
1368 		return (NULL);
1369 	if ((spa->spa_name = strdup(name)) == NULL) {
1370 		free(spa);
1371 		return (NULL);
1372 	}
1373 	spa->spa_uberblock = &spa->spa_uberblock_master;
1374 	spa->spa_mos = &spa->spa_mos_master;
1375 	spa->spa_guid = guid;
1376 	spa->spa_root_vdev = vdev_create(guid, NULL);
1377 	if (spa->spa_root_vdev == NULL) {
1378 		free(spa->spa_name);
1379 		free(spa);
1380 		return (NULL);
1381 	}
1382 	spa->spa_root_vdev->v_name = strdup("root");
1383 	STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
1384 
1385 	return (spa);
1386 }
1387 
1388 static const char *
1389 state_name(vdev_state_t state)
1390 {
1391 	static const char *names[] = {
1392 		"UNKNOWN",
1393 		"CLOSED",
1394 		"OFFLINE",
1395 		"REMOVED",
1396 		"CANT_OPEN",
1397 		"FAULTED",
1398 		"DEGRADED",
1399 		"ONLINE"
1400 	};
1401 	return (names[state]);
1402 }
1403 
1404 #ifdef BOOT2
1405 
1406 #define pager_printf printf
1407 
1408 #else
1409 
1410 static int
1411 pager_printf(const char *fmt, ...)
1412 {
1413 	char line[80];
1414 	va_list args;
1415 
1416 	va_start(args, fmt);
1417 	vsnprintf(line, sizeof(line), fmt, args);
1418 	va_end(args);
1419 	return (pager_output(line));
1420 }
1421 
1422 #endif
1423 
1424 #define	STATUS_FORMAT	"        %s %s\n"
1425 
1426 static int
1427 print_state(int indent, const char *name, vdev_state_t state)
1428 {
1429 	int i;
1430 	char buf[512];
1431 
1432 	buf[0] = 0;
1433 	for (i = 0; i < indent; i++)
1434 		strcat(buf, "  ");
1435 	strcat(buf, name);
1436 	return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
1437 }
1438 
1439 static int
1440 vdev_status(vdev_t *vdev, int indent)
1441 {
1442 	vdev_t *kid;
1443 	int ret;
1444 
1445 	if (vdev->v_islog) {
1446 		(void) pager_output("        logs\n");
1447 		indent++;
1448 	}
1449 
1450 	ret = print_state(indent, vdev->v_name, vdev->v_state);
1451 	if (ret != 0)
1452 		return (ret);
1453 
1454 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1455 		ret = vdev_status(kid, indent + 1);
1456 		if (ret != 0)
1457 			return (ret);
1458 	}
1459 	return (ret);
1460 }
1461 
1462 static int
1463 spa_status(spa_t *spa)
1464 {
1465 	static char bootfs[ZFS_MAXNAMELEN];
1466 	uint64_t rootid;
1467 	vdev_list_t *vlist;
1468 	vdev_t *vdev;
1469 	int good_kids, bad_kids, degraded_kids, ret;
1470 	vdev_state_t state;
1471 
1472 	ret = pager_printf("  pool: %s\n", spa->spa_name);
1473 	if (ret != 0)
1474 		return (ret);
1475 
1476 	if (zfs_get_root(spa, &rootid) == 0 &&
1477 	    zfs_rlookup(spa, rootid, bootfs) == 0) {
1478 		if (bootfs[0] == '\0')
1479 			ret = pager_printf("bootfs: %s\n", spa->spa_name);
1480 		else
1481 			ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
1482 			    bootfs);
1483 		if (ret != 0)
1484 			return (ret);
1485 	}
1486 	ret = pager_printf("config:\n\n");
1487 	if (ret != 0)
1488 		return (ret);
1489 	ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
1490 	if (ret != 0)
1491 		return (ret);
1492 
1493 	good_kids = 0;
1494 	degraded_kids = 0;
1495 	bad_kids = 0;
1496 	vlist = &spa->spa_root_vdev->v_children;
1497 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1498 		if (vdev->v_state == VDEV_STATE_HEALTHY)
1499 			good_kids++;
1500 		else if (vdev->v_state == VDEV_STATE_DEGRADED)
1501 			degraded_kids++;
1502 		else
1503 			bad_kids++;
1504 	}
1505 
1506 	state = VDEV_STATE_CLOSED;
1507 	if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
1508 		state = VDEV_STATE_HEALTHY;
1509 	else if ((good_kids + degraded_kids) > 0)
1510 		state = VDEV_STATE_DEGRADED;
1511 
1512 	ret = print_state(0, spa->spa_name, state);
1513 	if (ret != 0)
1514 		return (ret);
1515 
1516 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1517 		ret = vdev_status(vdev, 1);
1518 		if (ret != 0)
1519 			return (ret);
1520 	}
1521 	return (ret);
1522 }
1523 
1524 static int
1525 spa_all_status(void)
1526 {
1527 	spa_t *spa;
1528 	int first = 1, ret = 0;
1529 
1530 	STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
1531 		if (!first) {
1532 			ret = pager_printf("\n");
1533 			if (ret != 0)
1534 				return (ret);
1535 		}
1536 		first = 0;
1537 		ret = spa_status(spa);
1538 		if (ret != 0)
1539 			return (ret);
1540 	}
1541 	return (ret);
1542 }
1543 
1544 static uint64_t
1545 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
1546 {
1547 	uint64_t label_offset;
1548 
1549 	if (l < VDEV_LABELS / 2)
1550 		label_offset = 0;
1551 	else
1552 		label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
1553 
1554 	return (offset + l * sizeof (vdev_label_t) + label_offset);
1555 }
1556 
1557 static int
1558 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1559 {
1560 	unsigned int seq1 = 0;
1561 	unsigned int seq2 = 0;
1562 	int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
1563 
1564 	if (cmp != 0)
1565 		return (cmp);
1566 
1567 	cmp = AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1568 	if (cmp != 0)
1569 		return (cmp);
1570 
1571 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1572 		seq1 = MMP_SEQ(ub1);
1573 
1574 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1575 		seq2 = MMP_SEQ(ub2);
1576 
1577 	return (AVL_CMP(seq1, seq2));
1578 }
1579 
1580 static int
1581 uberblock_verify(uberblock_t *ub)
1582 {
1583 	if (ub->ub_magic == BSWAP_64((uint64_t)UBERBLOCK_MAGIC)) {
1584 		byteswap_uint64_array(ub, sizeof (uberblock_t));
1585 	}
1586 
1587 	if (ub->ub_magic != UBERBLOCK_MAGIC ||
1588 	    !SPA_VERSION_IS_SUPPORTED(ub->ub_version))
1589 		return (EINVAL);
1590 
1591 	return (0);
1592 }
1593 
1594 static int
1595 vdev_label_read(vdev_t *vd, int l, void *buf, uint64_t offset,
1596     size_t size)
1597 {
1598 	blkptr_t bp;
1599 	off_t off;
1600 
1601 	off = vdev_label_offset(vd->v_psize, l, offset);
1602 
1603 	BP_ZERO(&bp);
1604 	BP_SET_LSIZE(&bp, size);
1605 	BP_SET_PSIZE(&bp, size);
1606 	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1607 	BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1608 	DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
1609 	ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1610 
1611 	return (vdev_read_phys(vd, &bp, buf, off, size));
1612 }
1613 
1614 /*
1615  * We do need to be sure we write to correct location.
1616  * Our vdev label does consist of 4 fields:
1617  * pad1 (8k), reserved.
1618  * bootenv (8k), checksummed, previously reserved, may contian garbage.
1619  * vdev_phys (112k), checksummed
1620  * uberblock ring (128k), checksummed.
1621  *
1622  * Since bootenv area may contain garbage, we can not reliably read it, as
1623  * we can get checksum errors.
1624  * Next best thing is vdev_phys - it is just after bootenv. It still may
1625  * be corrupted, but in such case we will miss this one write.
1626  */
1627 static int
1628 vdev_label_write_validate(vdev_t *vd, int l, uint64_t offset)
1629 {
1630 	uint64_t off, o_phys;
1631 	void *buf;
1632 	size_t size = VDEV_PHYS_SIZE;
1633 	int rc;
1634 
1635 	o_phys = offsetof(vdev_label_t, vl_vdev_phys);
1636 	off = vdev_label_offset(vd->v_psize, l, o_phys);
1637 
1638 	/* off should be 8K from bootenv */
1639 	if (vdev_label_offset(vd->v_psize, l, offset) + VDEV_PAD_SIZE != off)
1640 		return (EINVAL);
1641 
1642 	buf = malloc(size);
1643 	if (buf == NULL)
1644 		return (ENOMEM);
1645 
1646 	/* Read vdev_phys */
1647 	rc = vdev_label_read(vd, l, buf, o_phys, size);
1648 	free(buf);
1649 	return (rc);
1650 }
1651 
1652 static int
1653 vdev_label_write(vdev_t *vd, int l, vdev_boot_envblock_t *be, uint64_t offset)
1654 {
1655 	zio_checksum_info_t *ci;
1656 	zio_cksum_t cksum;
1657 	off_t off;
1658 	size_t size = VDEV_PAD_SIZE;
1659 	int rc;
1660 
1661 	if (vd->v_phys_write == NULL)
1662 		return (ENOTSUP);
1663 
1664 	off = vdev_label_offset(vd->v_psize, l, offset);
1665 
1666 	rc = vdev_label_write_validate(vd, l, offset);
1667 	if (rc != 0) {
1668 		return (rc);
1669 	}
1670 
1671 	ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
1672 	be->vbe_zbt.zec_magic = ZEC_MAGIC;
1673 	zio_checksum_label_verifier(&be->vbe_zbt.zec_cksum, off);
1674 	ci->ci_func[0](be, size, NULL, &cksum);
1675 	be->vbe_zbt.zec_cksum = cksum;
1676 
1677 	return (vdev_write_phys(vd, be, off, size));
1678 }
1679 
1680 static int
1681 vdev_write_bootenv_impl(vdev_t *vdev, vdev_boot_envblock_t *be)
1682 {
1683 	vdev_t *kid;
1684 	int rv = 0, err;
1685 
1686 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1687 		if (kid->v_state != VDEV_STATE_HEALTHY)
1688 			continue;
1689 		err = vdev_write_bootenv_impl(kid, be);
1690 		if (err != 0)
1691 			rv = err;
1692 	}
1693 
1694 	/*
1695 	 * Non-leaf vdevs do not have v_phys_write.
1696 	 */
1697 	if (vdev->v_phys_write == NULL)
1698 		return (rv);
1699 
1700 	for (int l = 0; l < VDEV_LABELS; l++) {
1701 		err = vdev_label_write(vdev, l, be,
1702 		    offsetof(vdev_label_t, vl_be));
1703 		if (err != 0) {
1704 			printf("failed to write bootenv to %s label %d: %d\n",
1705 			    vdev->v_name ? vdev->v_name : "unknown", l, err);
1706 			rv = err;
1707 		}
1708 	}
1709 	return (rv);
1710 }
1711 
1712 int
1713 vdev_write_bootenv(vdev_t *vdev, nvlist_t *nvl)
1714 {
1715 	vdev_boot_envblock_t *be;
1716 	nvlist_t nv, *nvp;
1717 	uint64_t version;
1718 	int rv;
1719 
1720 	if (nvl->nv_size > sizeof(be->vbe_bootenv))
1721 		return (E2BIG);
1722 
1723 	version = VB_RAW;
1724 	nvp = vdev_read_bootenv(vdev);
1725 	if (nvp != NULL) {
1726 		nvlist_find(nvp, BOOTENV_VERSION, DATA_TYPE_UINT64, NULL,
1727 		    &version, NULL);
1728 		nvlist_destroy(nvp);
1729 	}
1730 
1731 	be = calloc(1, sizeof(*be));
1732 	if (be == NULL)
1733 		return (ENOMEM);
1734 
1735 	be->vbe_version = version;
1736 	switch (version) {
1737 	case VB_RAW:
1738 		/*
1739 		 * If there is no envmap, we will just wipe bootenv.
1740 		 */
1741 		nvlist_find(nvl, GRUB_ENVMAP, DATA_TYPE_STRING, NULL,
1742 		    be->vbe_bootenv, NULL);
1743 		rv = 0;
1744 		break;
1745 
1746 	case VB_NVLIST:
1747 		nv.nv_header = nvl->nv_header;
1748 		nv.nv_asize = nvl->nv_asize;
1749 		nv.nv_size = nvl->nv_size;
1750 
1751 		bcopy(&nv.nv_header, be->vbe_bootenv, sizeof(nv.nv_header));
1752 		nv.nv_data = be->vbe_bootenv + sizeof(nvs_header_t);
1753 		bcopy(nvl->nv_data, nv.nv_data, nv.nv_size);
1754 		rv = nvlist_export(&nv);
1755 		break;
1756 
1757 	default:
1758 		rv = EINVAL;
1759 		break;
1760 	}
1761 
1762 	if (rv == 0) {
1763 		be->vbe_version = htobe64(be->vbe_version);
1764 		rv = vdev_write_bootenv_impl(vdev, be);
1765 	}
1766 	free(be);
1767 	return (rv);
1768 }
1769 
1770 /*
1771  * Read the bootenv area from pool label, return the nvlist from it.
1772  * We return from first successful read.
1773  */
1774 nvlist_t *
1775 vdev_read_bootenv(vdev_t *vdev)
1776 {
1777 	vdev_t *kid;
1778 	nvlist_t *benv;
1779 	vdev_boot_envblock_t *be;
1780 	char *command;
1781 	bool ok;
1782 	int rv;
1783 
1784 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1785 		if (kid->v_state != VDEV_STATE_HEALTHY)
1786 			continue;
1787 
1788 		benv = vdev_read_bootenv(kid);
1789 		if (benv != NULL)
1790 			return (benv);
1791 	}
1792 
1793 	be = malloc(sizeof (*be));
1794 	if (be == NULL)
1795 		return (NULL);
1796 
1797 	rv = 0;
1798 	for (int l = 0; l < VDEV_LABELS; l++) {
1799 		rv = vdev_label_read(vdev, l, be,
1800 		    offsetof(vdev_label_t, vl_be),
1801 		    sizeof (*be));
1802 		if (rv == 0)
1803 			break;
1804 	}
1805 	if (rv != 0) {
1806 		free(be);
1807 		return (NULL);
1808 	}
1809 
1810 	be->vbe_version = be64toh(be->vbe_version);
1811 	switch (be->vbe_version) {
1812 	case VB_RAW:
1813 		/*
1814 		 * we have textual data in vbe_bootenv, create nvlist
1815 		 * with key "envmap".
1816 		 */
1817 		benv = nvlist_create(NV_UNIQUE_NAME);
1818 		if (benv != NULL) {
1819 			if (*be->vbe_bootenv == '\0') {
1820 				nvlist_add_uint64(benv, BOOTENV_VERSION,
1821 				    VB_NVLIST);
1822 				break;
1823 			}
1824 			nvlist_add_uint64(benv, BOOTENV_VERSION, VB_RAW);
1825 			be->vbe_bootenv[sizeof (be->vbe_bootenv) - 1] = '\0';
1826 			nvlist_add_string(benv, GRUB_ENVMAP, be->vbe_bootenv);
1827 		}
1828 		break;
1829 
1830 	case VB_NVLIST:
1831 		benv = nvlist_import(be->vbe_bootenv, sizeof(be->vbe_bootenv));
1832 		break;
1833 
1834 	default:
1835 		command = (char *)be;
1836 		ok = false;
1837 
1838 		/* Check for legacy zfsbootcfg command string */
1839 		for (int i = 0; command[i] != '\0'; i++) {
1840 			if (iscntrl(command[i])) {
1841 				ok = false;
1842 				break;
1843 			} else {
1844 				ok = true;
1845 			}
1846 		}
1847 		benv = nvlist_create(NV_UNIQUE_NAME);
1848 		if (benv != NULL) {
1849 			if (ok)
1850 				nvlist_add_string(benv, FREEBSD_BOOTONCE,
1851 				    command);
1852 			else
1853 				nvlist_add_uint64(benv, BOOTENV_VERSION,
1854 				    VB_NVLIST);
1855 		}
1856 		break;
1857 	}
1858 	free(be);
1859 	return (benv);
1860 }
1861 
1862 static uint64_t
1863 vdev_get_label_asize(nvlist_t *nvl)
1864 {
1865 	nvlist_t *vdevs;
1866 	uint64_t asize;
1867 	const char *type;
1868 	int len;
1869 
1870 	asize = 0;
1871 	/* Get vdev tree */
1872 	if (nvlist_find(nvl, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1873 	    NULL, &vdevs, NULL) != 0)
1874 		return (asize);
1875 
1876 	/*
1877 	 * Get vdev type. We will calculate asize for raidz, mirror and disk.
1878 	 * For raidz, the asize is raw size of all children.
1879 	 */
1880 	if (nvlist_find(vdevs, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
1881 	    NULL, &type, &len) != 0)
1882 		goto done;
1883 
1884 	if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 &&
1885 	    memcmp(type, VDEV_TYPE_DISK, len) != 0 &&
1886 	    memcmp(type, VDEV_TYPE_RAIDZ, len) != 0)
1887 		goto done;
1888 
1889 	if (nvlist_find(vdevs, ZPOOL_CONFIG_ASIZE, DATA_TYPE_UINT64,
1890 	    NULL, &asize, NULL) != 0)
1891 		goto done;
1892 
1893 	if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) {
1894 		nvlist_t **kids;
1895 		int nkids;
1896 
1897 		if (nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN,
1898 		    DATA_TYPE_NVLIST_ARRAY, &nkids, &kids, NULL) != 0) {
1899 			asize = 0;
1900 			goto done;
1901 		}
1902 
1903 		asize /= nkids;
1904 		for (int i = 0; i < nkids; i++)
1905 			nvlist_destroy(kids[i]);
1906 		free(kids);
1907 	}
1908 
1909 	asize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
1910 done:
1911 	nvlist_destroy(vdevs);
1912 	return (asize);
1913 }
1914 
1915 static nvlist_t *
1916 vdev_label_read_config(vdev_t *vd, uint64_t txg)
1917 {
1918 	vdev_phys_t *label;
1919 	uint64_t best_txg = 0;
1920 	uint64_t label_txg = 0;
1921 	uint64_t asize;
1922 	nvlist_t *nvl = NULL, *tmp;
1923 	int error;
1924 
1925 	label = malloc(sizeof (vdev_phys_t));
1926 	if (label == NULL)
1927 		return (NULL);
1928 
1929 	for (int l = 0; l < VDEV_LABELS; l++) {
1930 		if (vdev_label_read(vd, l, label,
1931 		    offsetof(vdev_label_t, vl_vdev_phys),
1932 		    sizeof (vdev_phys_t)))
1933 			continue;
1934 
1935 		tmp = nvlist_import(label->vp_nvlist,
1936 		    sizeof(label->vp_nvlist));
1937 		if (tmp == NULL)
1938 			continue;
1939 
1940 		error = nvlist_find(tmp, ZPOOL_CONFIG_POOL_TXG,
1941 		    DATA_TYPE_UINT64, NULL, &label_txg, NULL);
1942 		if (error != 0 || label_txg == 0) {
1943 			nvlist_destroy(nvl);
1944 			nvl = tmp;
1945 			goto done;
1946 		}
1947 
1948 		if (label_txg <= txg && label_txg > best_txg) {
1949 			best_txg = label_txg;
1950 			nvlist_destroy(nvl);
1951 			nvl = tmp;
1952 			tmp = NULL;
1953 
1954 			/*
1955 			 * Use asize from pool config. We need this
1956 			 * because we can get bad value from BIOS.
1957 			 */
1958 			asize = vdev_get_label_asize(nvl);
1959 			if (asize != 0) {
1960 				vd->v_psize = asize;
1961 			}
1962 		}
1963 		nvlist_destroy(tmp);
1964 	}
1965 
1966 	if (best_txg == 0) {
1967 		nvlist_destroy(nvl);
1968 		nvl = NULL;
1969 	}
1970 done:
1971 	free(label);
1972 	return (nvl);
1973 }
1974 
1975 static void
1976 vdev_uberblock_load(vdev_t *vd, uberblock_t *ub)
1977 {
1978 	uberblock_t *buf;
1979 
1980 	buf = malloc(VDEV_UBERBLOCK_SIZE(vd));
1981 	if (buf == NULL)
1982 		return;
1983 
1984 	for (int l = 0; l < VDEV_LABELS; l++) {
1985 		for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1986 			if (vdev_label_read(vd, l, buf,
1987 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1988 			    VDEV_UBERBLOCK_SIZE(vd)))
1989 				continue;
1990 			if (uberblock_verify(buf) != 0)
1991 				continue;
1992 
1993 			if (vdev_uberblock_compare(buf, ub) > 0)
1994 				*ub = *buf;
1995 		}
1996 	}
1997 	free(buf);
1998 }
1999 
2000 static int
2001 vdev_probe(vdev_phys_read_t *_read, vdev_phys_write_t *_write, void *priv,
2002     spa_t **spap)
2003 {
2004 	vdev_t vtmp;
2005 	spa_t *spa;
2006 	vdev_t *vdev;
2007 	nvlist_t *nvl;
2008 	uint64_t val;
2009 	uint64_t guid, vdev_children;
2010 	uint64_t pool_txg, pool_guid;
2011 	const char *pool_name;
2012 	int rc, namelen;
2013 
2014 	/*
2015 	 * Load the vdev label and figure out which
2016 	 * uberblock is most current.
2017 	 */
2018 	memset(&vtmp, 0, sizeof(vtmp));
2019 	vtmp.v_phys_read = _read;
2020 	vtmp.v_phys_write = _write;
2021 	vtmp.v_priv = priv;
2022 	vtmp.v_psize = P2ALIGN(ldi_get_size(priv),
2023 	    (uint64_t)sizeof (vdev_label_t));
2024 
2025 	/* Test for minimum device size. */
2026 	if (vtmp.v_psize < SPA_MINDEVSIZE)
2027 		return (EIO);
2028 
2029 	nvl = vdev_label_read_config(&vtmp, UINT64_MAX);
2030 	if (nvl == NULL)
2031 		return (EIO);
2032 
2033 	if (nvlist_find(nvl, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
2034 	    NULL, &val, NULL) != 0) {
2035 		nvlist_destroy(nvl);
2036 		return (EIO);
2037 	}
2038 
2039 	if (!SPA_VERSION_IS_SUPPORTED(val)) {
2040 		printf("ZFS: unsupported ZFS version %u (should be %u)\n",
2041 		    (unsigned)val, (unsigned)SPA_VERSION);
2042 		nvlist_destroy(nvl);
2043 		return (EIO);
2044 	}
2045 
2046 	/* Check ZFS features for read */
2047 	rc = nvlist_check_features_for_read(nvl);
2048 	if (rc != 0) {
2049 		nvlist_destroy(nvl);
2050 		return (EIO);
2051 	}
2052 
2053 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
2054 	    NULL, &val, NULL) != 0) {
2055 		nvlist_destroy(nvl);
2056 		return (EIO);
2057 	}
2058 
2059 	if (val == POOL_STATE_DESTROYED) {
2060 		/* We don't boot only from destroyed pools. */
2061 		nvlist_destroy(nvl);
2062 		return (EIO);
2063 	}
2064 
2065 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
2066 	    NULL, &pool_txg, NULL) != 0 ||
2067 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
2068 	    NULL, &pool_guid, NULL) != 0 ||
2069 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
2070 	    NULL, &pool_name, &namelen) != 0) {
2071 		/*
2072 		 * Cache and spare devices end up here - just ignore
2073 		 * them.
2074 		 */
2075 		nvlist_destroy(nvl);
2076 		return (EIO);
2077 	}
2078 
2079 	/*
2080 	 * Create the pool if this is the first time we've seen it.
2081 	 */
2082 	spa = spa_find_by_guid(pool_guid);
2083 	if (spa == NULL) {
2084 		char *name;
2085 
2086 		nvlist_find(nvl, ZPOOL_CONFIG_VDEV_CHILDREN,
2087 		    DATA_TYPE_UINT64, NULL, &vdev_children, NULL);
2088 		name = malloc(namelen + 1);
2089 		if (name == NULL) {
2090 			nvlist_destroy(nvl);
2091 			return (ENOMEM);
2092 		}
2093 		bcopy(pool_name, name, namelen);
2094 		name[namelen] = '\0';
2095 		spa = spa_create(pool_guid, name);
2096 		free(name);
2097 		if (spa == NULL) {
2098 			nvlist_destroy(nvl);
2099 			return (ENOMEM);
2100 		}
2101 		spa->spa_root_vdev->v_nchildren = vdev_children;
2102 	}
2103 	if (pool_txg > spa->spa_txg)
2104 		spa->spa_txg = pool_txg;
2105 
2106 	/*
2107 	 * Get the vdev tree and create our in-core copy of it.
2108 	 * If we already have a vdev with this guid, this must
2109 	 * be some kind of alias (overlapping slices, dangerously dedicated
2110 	 * disks etc).
2111 	 */
2112 	if (nvlist_find(nvl, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
2113 	    NULL, &guid, NULL) != 0) {
2114 		nvlist_destroy(nvl);
2115 		return (EIO);
2116 	}
2117 	vdev = vdev_find(guid);
2118 	/* Has this vdev already been inited? */
2119 	if (vdev && vdev->v_phys_read) {
2120 		nvlist_destroy(nvl);
2121 		return (EIO);
2122 	}
2123 
2124 	rc = vdev_init_from_label(spa, nvl);
2125 	nvlist_destroy(nvl);
2126 	if (rc != 0)
2127 		return (rc);
2128 
2129 	/*
2130 	 * We should already have created an incomplete vdev for this
2131 	 * vdev. Find it and initialise it with our read proc.
2132 	 */
2133 	vdev = vdev_find(guid);
2134 	if (vdev != NULL) {
2135 		vdev->v_phys_read = _read;
2136 		vdev->v_phys_write = _write;
2137 		vdev->v_priv = priv;
2138 		vdev->v_psize = vtmp.v_psize;
2139 		/*
2140 		 * If no other state is set, mark vdev healthy.
2141 		 */
2142 		if (vdev->v_state == VDEV_STATE_UNKNOWN)
2143 			vdev->v_state = VDEV_STATE_HEALTHY;
2144 	} else {
2145 		printf("ZFS: inconsistent nvlist contents\n");
2146 		return (EIO);
2147 	}
2148 
2149 	if (vdev->v_islog)
2150 		spa->spa_with_log = vdev->v_islog;
2151 
2152 	/*
2153 	 * Re-evaluate top-level vdev state.
2154 	 */
2155 	vdev_set_state(vdev->v_top);
2156 
2157 	/*
2158 	 * Ok, we are happy with the pool so far. Lets find
2159 	 * the best uberblock and then we can actually access
2160 	 * the contents of the pool.
2161 	 */
2162 	vdev_uberblock_load(vdev, spa->spa_uberblock);
2163 
2164 	if (spap != NULL)
2165 		*spap = spa;
2166 	return (0);
2167 }
2168 
2169 static int
2170 ilog2(int n)
2171 {
2172 	int v;
2173 
2174 	for (v = 0; v < 32; v++)
2175 		if (n == (1 << v))
2176 			return (v);
2177 	return (-1);
2178 }
2179 
2180 static int
2181 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
2182 {
2183 	blkptr_t gbh_bp;
2184 	zio_gbh_phys_t zio_gb;
2185 	char *pbuf;
2186 	int i;
2187 
2188 	/* Artificial BP for gang block header. */
2189 	gbh_bp = *bp;
2190 	BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
2191 	BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
2192 	BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
2193 	BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
2194 	for (i = 0; i < SPA_DVAS_PER_BP; i++)
2195 		DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
2196 
2197 	/* Read gang header block using the artificial BP. */
2198 	if (zio_read(spa, &gbh_bp, &zio_gb))
2199 		return (EIO);
2200 
2201 	pbuf = buf;
2202 	for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
2203 		blkptr_t *gbp = &zio_gb.zg_blkptr[i];
2204 
2205 		if (BP_IS_HOLE(gbp))
2206 			continue;
2207 		if (zio_read(spa, gbp, pbuf))
2208 			return (EIO);
2209 		pbuf += BP_GET_PSIZE(gbp);
2210 	}
2211 
2212 	if (zio_checksum_verify(spa, bp, buf))
2213 		return (EIO);
2214 	return (0);
2215 }
2216 
2217 static int
2218 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
2219 {
2220 	int cpfunc = BP_GET_COMPRESS(bp);
2221 	uint64_t align, size;
2222 	void *pbuf;
2223 	int i, error;
2224 
2225 	/*
2226 	 * Process data embedded in block pointer
2227 	 */
2228 	if (BP_IS_EMBEDDED(bp)) {
2229 		ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
2230 
2231 		size = BPE_GET_PSIZE(bp);
2232 		ASSERT(size <= BPE_PAYLOAD_SIZE);
2233 
2234 		if (cpfunc != ZIO_COMPRESS_OFF)
2235 			pbuf = malloc(size);
2236 		else
2237 			pbuf = buf;
2238 
2239 		if (pbuf == NULL)
2240 			return (ENOMEM);
2241 
2242 		decode_embedded_bp_compressed(bp, pbuf);
2243 		error = 0;
2244 
2245 		if (cpfunc != ZIO_COMPRESS_OFF) {
2246 			error = zio_decompress_data(cpfunc, pbuf,
2247 			    size, buf, BP_GET_LSIZE(bp));
2248 			free(pbuf);
2249 		}
2250 		if (error != 0)
2251 			printf("ZFS: i/o error - unable to decompress "
2252 			    "block pointer data, error %d\n", error);
2253 		return (error);
2254 	}
2255 
2256 	error = EIO;
2257 
2258 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
2259 		const dva_t *dva = &bp->blk_dva[i];
2260 		vdev_t *vdev;
2261 		vdev_list_t *vlist;
2262 		uint64_t vdevid;
2263 		off_t offset;
2264 
2265 		if (!dva->dva_word[0] && !dva->dva_word[1])
2266 			continue;
2267 
2268 		vdevid = DVA_GET_VDEV(dva);
2269 		offset = DVA_GET_OFFSET(dva);
2270 		vlist = &spa->spa_root_vdev->v_children;
2271 		STAILQ_FOREACH(vdev, vlist, v_childlink) {
2272 			if (vdev->v_id == vdevid)
2273 				break;
2274 		}
2275 		if (!vdev || !vdev->v_read)
2276 			continue;
2277 
2278 		size = BP_GET_PSIZE(bp);
2279 		if (vdev->v_read == vdev_raidz_read) {
2280 			align = 1ULL << vdev->v_ashift;
2281 			if (P2PHASE(size, align) != 0)
2282 				size = P2ROUNDUP(size, align);
2283 		}
2284 		if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
2285 			pbuf = malloc(size);
2286 		else
2287 			pbuf = buf;
2288 
2289 		if (pbuf == NULL) {
2290 			error = ENOMEM;
2291 			break;
2292 		}
2293 
2294 		if (DVA_GET_GANG(dva))
2295 			error = zio_read_gang(spa, bp, pbuf);
2296 		else
2297 			error = vdev->v_read(vdev, bp, pbuf, offset, size);
2298 		if (error == 0) {
2299 			if (cpfunc != ZIO_COMPRESS_OFF)
2300 				error = zio_decompress_data(cpfunc, pbuf,
2301 				    BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
2302 			else if (size != BP_GET_PSIZE(bp))
2303 				bcopy(pbuf, buf, BP_GET_PSIZE(bp));
2304 		} else {
2305 			printf("zio_read error: %d\n", error);
2306 		}
2307 		if (buf != pbuf)
2308 			free(pbuf);
2309 		if (error == 0)
2310 			break;
2311 	}
2312 	if (error != 0)
2313 		printf("ZFS: i/o error - all block copies unavailable\n");
2314 
2315 	return (error);
2316 }
2317 
2318 static int
2319 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset,
2320     void *buf, size_t buflen)
2321 {
2322 	int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
2323 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2324 	int nlevels = dnode->dn_nlevels;
2325 	int i, rc;
2326 
2327 	if (bsize > SPA_MAXBLOCKSIZE) {
2328 		printf("ZFS: I/O error - blocks larger than %llu are not "
2329 		    "supported\n", SPA_MAXBLOCKSIZE);
2330 		return (EIO);
2331 	}
2332 
2333 	/*
2334 	 * Handle odd block sizes, mirrors dmu_read_impl().  Data can't exist
2335 	 * past the first block, so we'll clip the read to the portion of the
2336 	 * buffer within bsize and zero out the remainder.
2337 	 */
2338 	if (dnode->dn_maxblkid == 0) {
2339 		size_t newbuflen;
2340 
2341 		newbuflen = offset > bsize ? 0 : MIN(buflen, bsize - offset);
2342 		bzero((char *)buf + newbuflen, buflen - newbuflen);
2343 		buflen = newbuflen;
2344 	}
2345 
2346 	/*
2347 	 * Note: bsize may not be a power of two here so we need to do an
2348 	 * actual divide rather than a bitshift.
2349 	 */
2350 	while (buflen > 0) {
2351 		uint64_t bn = offset / bsize;
2352 		int boff = offset % bsize;
2353 		int ibn;
2354 		const blkptr_t *indbp;
2355 		blkptr_t bp;
2356 
2357 		if (bn > dnode->dn_maxblkid)
2358 			return (EIO);
2359 
2360 		if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
2361 			goto cached;
2362 
2363 		indbp = dnode->dn_blkptr;
2364 		for (i = 0; i < nlevels; i++) {
2365 			/*
2366 			 * Copy the bp from the indirect array so that
2367 			 * we can re-use the scratch buffer for multi-level
2368 			 * objects.
2369 			 */
2370 			ibn = bn >> ((nlevels - i - 1) * ibshift);
2371 			ibn &= ((1 << ibshift) - 1);
2372 			bp = indbp[ibn];
2373 			if (BP_IS_HOLE(&bp)) {
2374 				memset(dnode_cache_buf, 0, bsize);
2375 				break;
2376 			}
2377 			rc = zio_read(spa, &bp, dnode_cache_buf);
2378 			if (rc)
2379 				return (rc);
2380 			indbp = (const blkptr_t *) dnode_cache_buf;
2381 		}
2382 		dnode_cache_obj = dnode;
2383 		dnode_cache_bn = bn;
2384 	cached:
2385 
2386 		/*
2387 		 * The buffer contains our data block. Copy what we
2388 		 * need from it and loop.
2389 		 */
2390 		i = bsize - boff;
2391 		if (i > buflen) i = buflen;
2392 		memcpy(buf, &dnode_cache_buf[boff], i);
2393 		buf = ((char *)buf) + i;
2394 		offset += i;
2395 		buflen -= i;
2396 	}
2397 
2398 	return (0);
2399 }
2400 
2401 /*
2402  * Lookup a value in a microzap directory.
2403  */
2404 static int
2405 mzap_lookup(const mzap_phys_t *mz, size_t size, const char *name,
2406     uint64_t *value)
2407 {
2408 	const mzap_ent_phys_t *mze;
2409 	int chunks, i;
2410 
2411 	/*
2412 	 * Microzap objects use exactly one block. Read the whole
2413 	 * thing.
2414 	 */
2415 	chunks = size / MZAP_ENT_LEN - 1;
2416 	for (i = 0; i < chunks; i++) {
2417 		mze = &mz->mz_chunk[i];
2418 		if (strcmp(mze->mze_name, name) == 0) {
2419 			*value = mze->mze_value;
2420 			return (0);
2421 		}
2422 	}
2423 
2424 	return (ENOENT);
2425 }
2426 
2427 /*
2428  * Compare a name with a zap leaf entry. Return non-zero if the name
2429  * matches.
2430  */
2431 static int
2432 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2433     const char *name)
2434 {
2435 	size_t namelen;
2436 	const zap_leaf_chunk_t *nc;
2437 	const char *p;
2438 
2439 	namelen = zc->l_entry.le_name_numints;
2440 
2441 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2442 	p = name;
2443 	while (namelen > 0) {
2444 		size_t len;
2445 
2446 		len = namelen;
2447 		if (len > ZAP_LEAF_ARRAY_BYTES)
2448 			len = ZAP_LEAF_ARRAY_BYTES;
2449 		if (memcmp(p, nc->l_array.la_array, len))
2450 			return (0);
2451 		p += len;
2452 		namelen -= len;
2453 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2454 	}
2455 
2456 	return (1);
2457 }
2458 
2459 /*
2460  * Extract a uint64_t value from a zap leaf entry.
2461  */
2462 static uint64_t
2463 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
2464 {
2465 	const zap_leaf_chunk_t *vc;
2466 	int i;
2467 	uint64_t value;
2468 	const uint8_t *p;
2469 
2470 	vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
2471 	for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
2472 		value = (value << 8) | p[i];
2473 	}
2474 
2475 	return (value);
2476 }
2477 
2478 static void
2479 stv(int len, void *addr, uint64_t value)
2480 {
2481 	switch (len) {
2482 	case 1:
2483 		*(uint8_t *)addr = value;
2484 		return;
2485 	case 2:
2486 		*(uint16_t *)addr = value;
2487 		return;
2488 	case 4:
2489 		*(uint32_t *)addr = value;
2490 		return;
2491 	case 8:
2492 		*(uint64_t *)addr = value;
2493 		return;
2494 	}
2495 }
2496 
2497 /*
2498  * Extract a array from a zap leaf entry.
2499  */
2500 static void
2501 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2502     uint64_t integer_size, uint64_t num_integers, void *buf)
2503 {
2504 	uint64_t array_int_len = zc->l_entry.le_value_intlen;
2505 	uint64_t value = 0;
2506 	uint64_t *u64 = buf;
2507 	char *p = buf;
2508 	int len = MIN(zc->l_entry.le_value_numints, num_integers);
2509 	int chunk = zc->l_entry.le_value_chunk;
2510 	int byten = 0;
2511 
2512 	if (integer_size == 8 && len == 1) {
2513 		*u64 = fzap_leaf_value(zl, zc);
2514 		return;
2515 	}
2516 
2517 	while (len > 0) {
2518 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
2519 		int i;
2520 
2521 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
2522 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
2523 			value = (value << 8) | la->la_array[i];
2524 			byten++;
2525 			if (byten == array_int_len) {
2526 				stv(integer_size, p, value);
2527 				byten = 0;
2528 				len--;
2529 				if (len == 0)
2530 					return;
2531 				p += integer_size;
2532 			}
2533 		}
2534 		chunk = la->la_next;
2535 	}
2536 }
2537 
2538 static int
2539 fzap_check_size(uint64_t integer_size, uint64_t num_integers)
2540 {
2541 
2542 	switch (integer_size) {
2543 	case 1:
2544 	case 2:
2545 	case 4:
2546 	case 8:
2547 		break;
2548 	default:
2549 		return (EINVAL);
2550 	}
2551 
2552 	if (integer_size * num_integers > ZAP_MAXVALUELEN)
2553 		return (E2BIG);
2554 
2555 	return (0);
2556 }
2557 
2558 static void
2559 zap_leaf_free(zap_leaf_t *leaf)
2560 {
2561 	free(leaf->l_phys);
2562 	free(leaf);
2563 }
2564 
2565 static int
2566 zap_get_leaf_byblk(fat_zap_t *zap, uint64_t blk, zap_leaf_t **lp)
2567 {
2568 	int bs = FZAP_BLOCK_SHIFT(zap);
2569 	int err;
2570 
2571 	*lp = malloc(sizeof(**lp));
2572 	if (*lp == NULL)
2573 		return (ENOMEM);
2574 
2575 	(*lp)->l_bs = bs;
2576 	(*lp)->l_phys = malloc(1 << bs);
2577 
2578 	if ((*lp)->l_phys == NULL) {
2579 		free(*lp);
2580 		return (ENOMEM);
2581 	}
2582 	err = dnode_read(zap->zap_spa, zap->zap_dnode, blk << bs, (*lp)->l_phys,
2583 	    1 << bs);
2584 	if (err != 0) {
2585 		zap_leaf_free(*lp);
2586 	}
2587 	return (err);
2588 }
2589 
2590 static int
2591 zap_table_load(fat_zap_t *zap, zap_table_phys_t *tbl, uint64_t idx,
2592     uint64_t *valp)
2593 {
2594 	int bs = FZAP_BLOCK_SHIFT(zap);
2595 	uint64_t blk = idx >> (bs - 3);
2596 	uint64_t off = idx & ((1 << (bs - 3)) - 1);
2597 	uint64_t *buf;
2598 	int rc;
2599 
2600 	buf = malloc(1 << zap->zap_block_shift);
2601 	if (buf == NULL)
2602 		return (ENOMEM);
2603 	rc = dnode_read(zap->zap_spa, zap->zap_dnode, (tbl->zt_blk + blk) << bs,
2604 	    buf, 1 << zap->zap_block_shift);
2605 	if (rc == 0)
2606 		*valp = buf[off];
2607 	free(buf);
2608 	return (rc);
2609 }
2610 
2611 static int
2612 zap_idx_to_blk(fat_zap_t *zap, uint64_t idx, uint64_t *valp)
2613 {
2614 	if (zap->zap_phys->zap_ptrtbl.zt_numblks == 0) {
2615 		*valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
2616 		return (0);
2617 	} else {
2618 		return (zap_table_load(zap, &zap->zap_phys->zap_ptrtbl,
2619 		    idx, valp));
2620 	}
2621 }
2622 
2623 #define	ZAP_HASH_IDX(hash, n)	(((n) == 0) ? 0 : ((hash) >> (64 - (n))))
2624 static int
2625 zap_deref_leaf(fat_zap_t *zap, uint64_t h, zap_leaf_t **lp)
2626 {
2627 	uint64_t idx, blk;
2628 	int err;
2629 
2630 	idx = ZAP_HASH_IDX(h, zap->zap_phys->zap_ptrtbl.zt_shift);
2631 	err = zap_idx_to_blk(zap, idx, &blk);
2632 	if (err != 0)
2633 		return (err);
2634 	return (zap_get_leaf_byblk(zap, blk, lp));
2635 }
2636 
2637 #define	CHAIN_END	0xffff	/* end of the chunk chain */
2638 #define	LEAF_HASH(l, h) \
2639 	((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \
2640 	((h) >> \
2641 	(64 - ZAP_LEAF_HASH_SHIFT(l) - (l)->l_phys->l_hdr.lh_prefix_len)))
2642 #define	LEAF_HASH_ENTPTR(l, h)	(&(l)->l_phys->l_hash[LEAF_HASH(l, h)])
2643 
2644 static int
2645 zap_leaf_lookup(zap_leaf_t *zl, uint64_t hash, const char *name,
2646     uint64_t integer_size, uint64_t num_integers, void *value)
2647 {
2648 	int rc;
2649 	uint16_t *chunkp;
2650 	struct zap_leaf_entry *le;
2651 
2652 	/*
2653 	 * Make sure this chunk matches our hash.
2654 	 */
2655 	if (zl->l_phys->l_hdr.lh_prefix_len > 0 &&
2656 	    zl->l_phys->l_hdr.lh_prefix !=
2657 	    hash >> (64 - zl->l_phys->l_hdr.lh_prefix_len))
2658 		return (EIO);
2659 
2660 	rc = ENOENT;
2661 	for (chunkp = LEAF_HASH_ENTPTR(zl, hash);
2662 	    *chunkp != CHAIN_END; chunkp = &le->le_next) {
2663 		zap_leaf_chunk_t *zc;
2664 		uint16_t chunk = *chunkp;
2665 
2666 		le = ZAP_LEAF_ENTRY(zl, chunk);
2667 		if (le->le_hash != hash)
2668 			continue;
2669 		zc = &ZAP_LEAF_CHUNK(zl, chunk);
2670 		if (fzap_name_equal(zl, zc, name)) {
2671 			if (zc->l_entry.le_value_intlen > integer_size) {
2672 				rc = EINVAL;
2673 			} else {
2674 				fzap_leaf_array(zl, zc, integer_size,
2675 				    num_integers, value);
2676 				rc = 0;
2677 			}
2678 			break;
2679 		}
2680 	}
2681 	return (rc);
2682 }
2683 
2684 /*
2685  * Lookup a value in a fatzap directory.
2686  */
2687 static int
2688 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2689     const char *name, uint64_t integer_size, uint64_t num_integers,
2690     void *value)
2691 {
2692 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2693 	fat_zap_t z;
2694 	zap_leaf_t *zl;
2695 	uint64_t hash;
2696 	int rc;
2697 
2698 	if (zh->zap_magic != ZAP_MAGIC)
2699 		return (EIO);
2700 
2701 	if ((rc = fzap_check_size(integer_size, num_integers)) != 0) {
2702 		return (rc);
2703 	}
2704 
2705 	z.zap_block_shift = ilog2(bsize);
2706 	z.zap_phys = zh;
2707 	z.zap_spa = spa;
2708 	z.zap_dnode = dnode;
2709 
2710 	hash = zap_hash(zh->zap_salt, name);
2711 	rc = zap_deref_leaf(&z, hash, &zl);
2712 	if (rc != 0)
2713 		return (rc);
2714 
2715 	rc = zap_leaf_lookup(zl, hash, name, integer_size, num_integers, value);
2716 
2717 	zap_leaf_free(zl);
2718 	return (rc);
2719 }
2720 
2721 /*
2722  * Lookup a name in a zap object and return its value as a uint64_t.
2723  */
2724 static int
2725 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2726     uint64_t integer_size, uint64_t num_integers, void *value)
2727 {
2728 	int rc;
2729 	zap_phys_t *zap;
2730 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2731 
2732 	zap = malloc(size);
2733 	if (zap == NULL)
2734 		return (ENOMEM);
2735 
2736 	rc = dnode_read(spa, dnode, 0, zap, size);
2737 	if (rc)
2738 		goto done;
2739 
2740 	switch (zap->zap_block_type) {
2741 	case ZBT_MICRO:
2742 		rc = mzap_lookup((const mzap_phys_t *)zap, size, name, value);
2743 		break;
2744 	case ZBT_HEADER:
2745 		rc = fzap_lookup(spa, dnode, zap, name, integer_size,
2746 		    num_integers, value);
2747 		break;
2748 	default:
2749 		printf("ZFS: invalid zap_type=%" PRIx64 "\n",
2750 		    zap->zap_block_type);
2751 		rc = EIO;
2752 	}
2753 done:
2754 	free(zap);
2755 	return (rc);
2756 }
2757 
2758 /*
2759  * List a microzap directory.
2760  */
2761 static int
2762 mzap_list(const mzap_phys_t *mz, size_t size,
2763     int (*callback)(const char *, uint64_t))
2764 {
2765 	const mzap_ent_phys_t *mze;
2766 	int chunks, i, rc;
2767 
2768 	/*
2769 	 * Microzap objects use exactly one block. Read the whole
2770 	 * thing.
2771 	 */
2772 	rc = 0;
2773 	chunks = size / MZAP_ENT_LEN - 1;
2774 	for (i = 0; i < chunks; i++) {
2775 		mze = &mz->mz_chunk[i];
2776 		if (mze->mze_name[0]) {
2777 			rc = callback(mze->mze_name, mze->mze_value);
2778 			if (rc != 0)
2779 				break;
2780 		}
2781 	}
2782 
2783 	return (rc);
2784 }
2785 
2786 /*
2787  * List a fatzap directory.
2788  */
2789 static int
2790 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2791     int (*callback)(const char *, uint64_t))
2792 {
2793 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2794 	fat_zap_t z;
2795 	uint64_t i;
2796 	int j, rc;
2797 
2798 	if (zh->zap_magic != ZAP_MAGIC)
2799 		return (EIO);
2800 
2801 	z.zap_block_shift = ilog2(bsize);
2802 	z.zap_phys = zh;
2803 
2804 	/*
2805 	 * This assumes that the leaf blocks start at block 1. The
2806 	 * documentation isn't exactly clear on this.
2807 	 */
2808 	zap_leaf_t zl;
2809 	zl.l_bs = z.zap_block_shift;
2810 	zl.l_phys = malloc(bsize);
2811 	if (zl.l_phys == NULL)
2812 		return (ENOMEM);
2813 
2814 	for (i = 0; i < zh->zap_num_leafs; i++) {
2815 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2816 		char name[256], *p;
2817 		uint64_t value;
2818 
2819 		if (dnode_read(spa, dnode, off, zl.l_phys, bsize)) {
2820 			free(zl.l_phys);
2821 			return (EIO);
2822 		}
2823 
2824 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2825 			zap_leaf_chunk_t *zc, *nc;
2826 			int namelen;
2827 
2828 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2829 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2830 				continue;
2831 			namelen = zc->l_entry.le_name_numints;
2832 			if (namelen > sizeof(name))
2833 				namelen = sizeof(name);
2834 
2835 			/*
2836 			 * Paste the name back together.
2837 			 */
2838 			nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
2839 			p = name;
2840 			while (namelen > 0) {
2841 				int len;
2842 				len = namelen;
2843 				if (len > ZAP_LEAF_ARRAY_BYTES)
2844 					len = ZAP_LEAF_ARRAY_BYTES;
2845 				memcpy(p, nc->l_array.la_array, len);
2846 				p += len;
2847 				namelen -= len;
2848 				nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
2849 			}
2850 
2851 			/*
2852 			 * Assume the first eight bytes of the value are
2853 			 * a uint64_t.
2854 			 */
2855 			value = fzap_leaf_value(&zl, zc);
2856 
2857 			/* printf("%s 0x%jx\n", name, (uintmax_t)value); */
2858 			rc = callback((const char *)name, value);
2859 			if (rc != 0) {
2860 				free(zl.l_phys);
2861 				return (rc);
2862 			}
2863 		}
2864 	}
2865 
2866 	free(zl.l_phys);
2867 	return (0);
2868 }
2869 
2870 static int zfs_printf(const char *name, uint64_t value __unused)
2871 {
2872 
2873 	printf("%s\n", name);
2874 
2875 	return (0);
2876 }
2877 
2878 /*
2879  * List a zap directory.
2880  */
2881 static int
2882 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
2883 {
2884 	zap_phys_t *zap;
2885 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2886 	int rc;
2887 
2888 	zap = malloc(size);
2889 	if (zap == NULL)
2890 		return (ENOMEM);
2891 
2892 	rc = dnode_read(spa, dnode, 0, zap, size);
2893 	if (rc == 0) {
2894 		if (zap->zap_block_type == ZBT_MICRO)
2895 			rc = mzap_list((const mzap_phys_t *)zap, size,
2896 			    zfs_printf);
2897 		else
2898 			rc = fzap_list(spa, dnode, zap, zfs_printf);
2899 	}
2900 	free(zap);
2901 	return (rc);
2902 }
2903 
2904 static int
2905 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum,
2906     dnode_phys_t *dnode)
2907 {
2908 	off_t offset;
2909 
2910 	offset = objnum * sizeof(dnode_phys_t);
2911 	return dnode_read(spa, &os->os_meta_dnode, offset,
2912 		dnode, sizeof(dnode_phys_t));
2913 }
2914 
2915 /*
2916  * Lookup a name in a microzap directory.
2917  */
2918 static int
2919 mzap_rlookup(const mzap_phys_t *mz, size_t size, char *name, uint64_t value)
2920 {
2921 	const mzap_ent_phys_t *mze;
2922 	int chunks, i;
2923 
2924 	/*
2925 	 * Microzap objects use exactly one block. Read the whole
2926 	 * thing.
2927 	 */
2928 	chunks = size / MZAP_ENT_LEN - 1;
2929 	for (i = 0; i < chunks; i++) {
2930 		mze = &mz->mz_chunk[i];
2931 		if (value == mze->mze_value) {
2932 			strcpy(name, mze->mze_name);
2933 			return (0);
2934 		}
2935 	}
2936 
2937 	return (ENOENT);
2938 }
2939 
2940 static void
2941 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
2942 {
2943 	size_t namelen;
2944 	const zap_leaf_chunk_t *nc;
2945 	char *p;
2946 
2947 	namelen = zc->l_entry.le_name_numints;
2948 
2949 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2950 	p = name;
2951 	while (namelen > 0) {
2952 		size_t len;
2953 		len = namelen;
2954 		if (len > ZAP_LEAF_ARRAY_BYTES)
2955 			len = ZAP_LEAF_ARRAY_BYTES;
2956 		memcpy(p, nc->l_array.la_array, len);
2957 		p += len;
2958 		namelen -= len;
2959 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2960 	}
2961 
2962 	*p = '\0';
2963 }
2964 
2965 static int
2966 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2967     char *name, uint64_t value)
2968 {
2969 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2970 	fat_zap_t z;
2971 	uint64_t i;
2972 	int j, rc;
2973 
2974 	if (zh->zap_magic != ZAP_MAGIC)
2975 		return (EIO);
2976 
2977 	z.zap_block_shift = ilog2(bsize);
2978 	z.zap_phys = zh;
2979 
2980 	/*
2981 	 * This assumes that the leaf blocks start at block 1. The
2982 	 * documentation isn't exactly clear on this.
2983 	 */
2984 	zap_leaf_t zl;
2985 	zl.l_bs = z.zap_block_shift;
2986 	zl.l_phys = malloc(bsize);
2987 	if (zl.l_phys == NULL)
2988 		return (ENOMEM);
2989 
2990 	for (i = 0; i < zh->zap_num_leafs; i++) {
2991 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2992 
2993 		rc = dnode_read(spa, dnode, off, zl.l_phys, bsize);
2994 		if (rc != 0)
2995 			goto done;
2996 
2997 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2998 			zap_leaf_chunk_t *zc;
2999 
3000 			zc = &ZAP_LEAF_CHUNK(&zl, j);
3001 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
3002 				continue;
3003 			if (zc->l_entry.le_value_intlen != 8 ||
3004 			    zc->l_entry.le_value_numints != 1)
3005 				continue;
3006 
3007 			if (fzap_leaf_value(&zl, zc) == value) {
3008 				fzap_name_copy(&zl, zc, name);
3009 				goto done;
3010 			}
3011 		}
3012 	}
3013 
3014 	rc = ENOENT;
3015 done:
3016 	free(zl.l_phys);
3017 	return (rc);
3018 }
3019 
3020 static int
3021 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name,
3022     uint64_t value)
3023 {
3024 	zap_phys_t *zap;
3025 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
3026 	int rc;
3027 
3028 	zap = malloc(size);
3029 	if (zap == NULL)
3030 		return (ENOMEM);
3031 
3032 	rc = dnode_read(spa, dnode, 0, zap, size);
3033 	if (rc == 0) {
3034 		if (zap->zap_block_type == ZBT_MICRO)
3035 			rc = mzap_rlookup((const mzap_phys_t *)zap, size,
3036 			    name, value);
3037 		else
3038 			rc = fzap_rlookup(spa, dnode, zap, name, value);
3039 	}
3040 	free(zap);
3041 	return (rc);
3042 }
3043 
3044 static int
3045 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
3046 {
3047 	char name[256];
3048 	char component[256];
3049 	uint64_t dir_obj, parent_obj, child_dir_zapobj;
3050 	dnode_phys_t child_dir_zap, snapnames_zap, dataset, dir, parent;
3051 	dsl_dir_phys_t *dd;
3052 	dsl_dataset_phys_t *ds;
3053 	char *p;
3054 	int len;
3055 	boolean_t issnap = B_FALSE;
3056 
3057 	p = &name[sizeof(name) - 1];
3058 	*p = '\0';
3059 
3060 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
3061 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3062 		return (EIO);
3063 	}
3064 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3065 	dir_obj = ds->ds_dir_obj;
3066 	if (ds->ds_snapnames_zapobj == 0)
3067 		issnap = B_TRUE;
3068 
3069 	for (;;) {
3070 		if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir) != 0)
3071 			return (EIO);
3072 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3073 
3074 		/* Actual loop condition. */
3075 		parent_obj = dd->dd_parent_obj;
3076 		if (parent_obj == 0)
3077 			break;
3078 
3079 		if (objset_get_dnode(spa, spa->spa_mos, parent_obj,
3080 		    &parent) != 0)
3081 			return (EIO);
3082 		dd = (dsl_dir_phys_t *)&parent.dn_bonus;
3083 		if (issnap == B_TRUE) {
3084 			/*
3085 			 * The dataset we are looking up is a snapshot
3086 			 * the dir_obj is the parent already, we don't want
3087 			 * the grandparent just yet. Reset to the parent.
3088 			 */
3089 			dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3090 			/* Lookup the dataset to get the snapname ZAP */
3091 			if (objset_get_dnode(spa, spa->spa_mos,
3092 			    dd->dd_head_dataset_obj, &dataset))
3093 				return (EIO);
3094 			ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3095 			if (objset_get_dnode(spa, spa->spa_mos,
3096 			    ds->ds_snapnames_zapobj, &snapnames_zap) != 0)
3097 				return (EIO);
3098 			/* Get the name of the snapshot */
3099 			if (zap_rlookup(spa, &snapnames_zap, component,
3100 			    objnum) != 0)
3101 				return (EIO);
3102 			len = strlen(component);
3103 			p -= len;
3104 			memcpy(p, component, len);
3105 			--p;
3106 			*p = '@';
3107 			issnap = B_FALSE;
3108 			continue;
3109 		}
3110 
3111 		child_dir_zapobj = dd->dd_child_dir_zapobj;
3112 		if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
3113 		    &child_dir_zap) != 0)
3114 			return (EIO);
3115 		if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
3116 			return (EIO);
3117 
3118 		len = strlen(component);
3119 		p -= len;
3120 		memcpy(p, component, len);
3121 		--p;
3122 		*p = '/';
3123 
3124 		/* Actual loop iteration. */
3125 		dir_obj = parent_obj;
3126 	}
3127 
3128 	if (*p != '\0')
3129 		++p;
3130 	strcpy(result, p);
3131 
3132 	return (0);
3133 }
3134 
3135 static int
3136 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
3137 {
3138 	char element[256];
3139 	uint64_t dir_obj, child_dir_zapobj;
3140 	dnode_phys_t child_dir_zap, snapnames_zap, dir, dataset;
3141 	dsl_dir_phys_t *dd;
3142 	dsl_dataset_phys_t *ds;
3143 	const char *p, *q;
3144 	boolean_t issnap = B_FALSE;
3145 
3146 	if (objset_get_dnode(spa, spa->spa_mos,
3147 	    DMU_POOL_DIRECTORY_OBJECT, &dir))
3148 		return (EIO);
3149 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
3150 	    1, &dir_obj))
3151 		return (EIO);
3152 
3153 	p = name;
3154 	for (;;) {
3155 		if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir))
3156 			return (EIO);
3157 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3158 
3159 		while (*p == '/')
3160 			p++;
3161 		/* Actual loop condition #1. */
3162 		if (*p == '\0')
3163 			break;
3164 
3165 		q = strchr(p, '/');
3166 		if (q) {
3167 			memcpy(element, p, q - p);
3168 			element[q - p] = '\0';
3169 			p = q + 1;
3170 		} else {
3171 			strcpy(element, p);
3172 			p += strlen(p);
3173 		}
3174 
3175 		if (issnap == B_TRUE) {
3176 		        if (objset_get_dnode(spa, spa->spa_mos,
3177 			    dd->dd_head_dataset_obj, &dataset))
3178 		                return (EIO);
3179 			ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3180 			if (objset_get_dnode(spa, spa->spa_mos,
3181 			    ds->ds_snapnames_zapobj, &snapnames_zap) != 0)
3182 				return (EIO);
3183 			/* Actual loop condition #2. */
3184 			if (zap_lookup(spa, &snapnames_zap, element,
3185 			    sizeof (dir_obj), 1, &dir_obj) != 0)
3186 				return (ENOENT);
3187 			*objnum = dir_obj;
3188 			return (0);
3189 		} else if ((q = strchr(element, '@')) != NULL) {
3190 			issnap = B_TRUE;
3191 			element[q - element] = '\0';
3192 			p = q + 1;
3193 		}
3194 		child_dir_zapobj = dd->dd_child_dir_zapobj;
3195 		if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
3196 		    &child_dir_zap) != 0)
3197 			return (EIO);
3198 
3199 		/* Actual loop condition #2. */
3200 		if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
3201 		    1, &dir_obj) != 0)
3202 			return (ENOENT);
3203 	}
3204 
3205 	*objnum = dd->dd_head_dataset_obj;
3206 	return (0);
3207 }
3208 
3209 #ifndef BOOT2
3210 static int
3211 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
3212 {
3213 	uint64_t dir_obj, child_dir_zapobj;
3214 	dnode_phys_t child_dir_zap, dir, dataset;
3215 	dsl_dataset_phys_t *ds;
3216 	dsl_dir_phys_t *dd;
3217 
3218 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
3219 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3220 		return (EIO);
3221 	}
3222 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3223 	dir_obj = ds->ds_dir_obj;
3224 
3225 	if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir)) {
3226 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
3227 		return (EIO);
3228 	}
3229 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3230 
3231 	child_dir_zapobj = dd->dd_child_dir_zapobj;
3232 	if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
3233 	    &child_dir_zap) != 0) {
3234 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
3235 		return (EIO);
3236 	}
3237 
3238 	return (zap_list(spa, &child_dir_zap) != 0);
3239 }
3240 
3241 int
3242 zfs_callback_dataset(const spa_t *spa, uint64_t objnum,
3243     int (*callback)(const char *, uint64_t))
3244 {
3245 	uint64_t dir_obj, child_dir_zapobj;
3246 	dnode_phys_t child_dir_zap, dir, dataset;
3247 	dsl_dataset_phys_t *ds;
3248 	dsl_dir_phys_t *dd;
3249 	zap_phys_t *zap;
3250 	size_t size;
3251 	int err;
3252 
3253 	err = objset_get_dnode(spa, spa->spa_mos, objnum, &dataset);
3254 	if (err != 0) {
3255 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3256 		return (err);
3257 	}
3258 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3259 	dir_obj = ds->ds_dir_obj;
3260 
3261 	err = objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir);
3262 	if (err != 0) {
3263 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
3264 		return (err);
3265 	}
3266 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3267 
3268 	child_dir_zapobj = dd->dd_child_dir_zapobj;
3269 	err = objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
3270 	    &child_dir_zap);
3271 	if (err != 0) {
3272 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
3273 		return (err);
3274 	}
3275 
3276 	size = child_dir_zap.dn_datablkszsec << SPA_MINBLOCKSHIFT;
3277 	zap = malloc(size);
3278 	if (zap != NULL) {
3279 		err = dnode_read(spa, &child_dir_zap, 0, zap, size);
3280 		if (err != 0)
3281 			goto done;
3282 
3283 		if (zap->zap_block_type == ZBT_MICRO)
3284 			err = mzap_list((const mzap_phys_t *)zap, size,
3285 			    callback);
3286 		else
3287 			err = fzap_list(spa, &child_dir_zap, zap, callback);
3288 	} else {
3289 		err = ENOMEM;
3290 	}
3291 done:
3292 	free(zap);
3293 	return (err);
3294 }
3295 #endif
3296 
3297 /*
3298  * Find the object set given the object number of its dataset object
3299  * and return its details in *objset
3300  */
3301 static int
3302 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
3303 {
3304 	dnode_phys_t dataset;
3305 	dsl_dataset_phys_t *ds;
3306 
3307 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
3308 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3309 		return (EIO);
3310 	}
3311 
3312 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3313 	if (zio_read(spa, &ds->ds_bp, objset)) {
3314 		printf("ZFS: can't read object set for dataset %ju\n",
3315 		    (uintmax_t)objnum);
3316 		return (EIO);
3317 	}
3318 
3319 	return (0);
3320 }
3321 
3322 /*
3323  * Find the object set pointed to by the BOOTFS property or the root
3324  * dataset if there is none and return its details in *objset
3325  */
3326 static int
3327 zfs_get_root(const spa_t *spa, uint64_t *objid)
3328 {
3329 	dnode_phys_t dir, propdir;
3330 	uint64_t props, bootfs, root;
3331 
3332 	*objid = 0;
3333 
3334 	/*
3335 	 * Start with the MOS directory object.
3336 	 */
3337 	if (objset_get_dnode(spa, spa->spa_mos,
3338 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3339 		printf("ZFS: can't read MOS object directory\n");
3340 		return (EIO);
3341 	}
3342 
3343 	/*
3344 	 * Lookup the pool_props and see if we can find a bootfs.
3345 	 */
3346 	if (zap_lookup(spa, &dir, DMU_POOL_PROPS,
3347 	    sizeof(props), 1, &props) == 0 &&
3348 	    objset_get_dnode(spa, spa->spa_mos, props, &propdir) == 0 &&
3349 	    zap_lookup(spa, &propdir, "bootfs",
3350 	    sizeof(bootfs), 1, &bootfs) == 0 && bootfs != 0) {
3351 		*objid = bootfs;
3352 		return (0);
3353 	}
3354 	/*
3355 	 * Lookup the root dataset directory
3356 	 */
3357 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET,
3358 	    sizeof(root), 1, &root) ||
3359 	    objset_get_dnode(spa, spa->spa_mos, root, &dir)) {
3360 		printf("ZFS: can't find root dsl_dir\n");
3361 		return (EIO);
3362 	}
3363 
3364 	/*
3365 	 * Use the information from the dataset directory's bonus buffer
3366 	 * to find the dataset object and from that the object set itself.
3367 	 */
3368 	dsl_dir_phys_t *dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3369 	*objid = dd->dd_head_dataset_obj;
3370 	return (0);
3371 }
3372 
3373 static int
3374 zfs_mount_impl(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
3375 {
3376 
3377 	mount->spa = spa;
3378 
3379 	/*
3380 	 * Find the root object set if not explicitly provided
3381 	 */
3382 	if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
3383 		printf("ZFS: can't find root filesystem\n");
3384 		return (EIO);
3385 	}
3386 
3387 	if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
3388 		printf("ZFS: can't open root filesystem\n");
3389 		return (EIO);
3390 	}
3391 
3392 	mount->rootobj = rootobj;
3393 
3394 	return (0);
3395 }
3396 
3397 /*
3398  * callback function for feature name checks.
3399  */
3400 static int
3401 check_feature(const char *name, uint64_t value)
3402 {
3403 	int i;
3404 
3405 	if (value == 0)
3406 		return (0);
3407 	if (name[0] == '\0')
3408 		return (0);
3409 
3410 	for (i = 0; features_for_read[i] != NULL; i++) {
3411 		if (strcmp(name, features_for_read[i]) == 0)
3412 			return (0);
3413 	}
3414 	printf("ZFS: unsupported feature: %s\n", name);
3415 	return (EIO);
3416 }
3417 
3418 /*
3419  * Checks whether the MOS features that are active are supported.
3420  */
3421 static int
3422 check_mos_features(const spa_t *spa)
3423 {
3424 	dnode_phys_t dir;
3425 	zap_phys_t *zap;
3426 	uint64_t objnum;
3427 	size_t size;
3428 	int rc;
3429 
3430 	if ((rc = objset_get_dnode(spa, spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
3431 	    &dir)) != 0)
3432 		return (rc);
3433 	if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
3434 	    sizeof (objnum), 1, &objnum)) != 0) {
3435 		/*
3436 		 * It is older pool without features. As we have already
3437 		 * tested the label, just return without raising the error.
3438 		 */
3439 		return (0);
3440 	}
3441 
3442 	if ((rc = objset_get_dnode(spa, spa->spa_mos, objnum, &dir)) != 0)
3443 		return (rc);
3444 
3445 	if (dir.dn_type != DMU_OTN_ZAP_METADATA)
3446 		return (EIO);
3447 
3448 	size = dir.dn_datablkszsec << SPA_MINBLOCKSHIFT;
3449 	zap = malloc(size);
3450 	if (zap == NULL)
3451 		return (ENOMEM);
3452 
3453 	if (dnode_read(spa, &dir, 0, zap, size)) {
3454 		free(zap);
3455 		return (EIO);
3456 	}
3457 
3458 	if (zap->zap_block_type == ZBT_MICRO)
3459 		rc = mzap_list((const mzap_phys_t *)zap, size, check_feature);
3460 	else
3461 		rc = fzap_list(spa, &dir, zap, check_feature);
3462 
3463 	free(zap);
3464 	return (rc);
3465 }
3466 
3467 static int
3468 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
3469 {
3470 	dnode_phys_t dir;
3471 	size_t size;
3472 	int rc;
3473 	char *nv;
3474 
3475 	*value = NULL;
3476 	if ((rc = objset_get_dnode(spa, spa->spa_mos, obj, &dir)) != 0)
3477 		return (rc);
3478 	if (dir.dn_type != DMU_OT_PACKED_NVLIST &&
3479 	    dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) {
3480 		return (EIO);
3481 	}
3482 
3483 	if (dir.dn_bonuslen != sizeof (uint64_t))
3484 		return (EIO);
3485 
3486 	size = *(uint64_t *)DN_BONUS(&dir);
3487 	nv = malloc(size);
3488 	if (nv == NULL)
3489 		return (ENOMEM);
3490 
3491 	rc = dnode_read(spa, &dir, 0, nv, size);
3492 	if (rc != 0) {
3493 		free(nv);
3494 		nv = NULL;
3495 		return (rc);
3496 	}
3497 	*value = nvlist_import(nv, size);
3498 	free(nv);
3499 	return (rc);
3500 }
3501 
3502 static int
3503 zfs_spa_init(spa_t *spa)
3504 {
3505 	struct uberblock checkpoint;
3506 	dnode_phys_t dir;
3507 	uint64_t config_object;
3508 	nvlist_t *nvlist;
3509 	int rc;
3510 
3511 	if (zio_read(spa, &spa->spa_uberblock->ub_rootbp, spa->spa_mos)) {
3512 		printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
3513 		return (EIO);
3514 	}
3515 	if (spa->spa_mos->os_type != DMU_OST_META) {
3516 		printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
3517 		return (EIO);
3518 	}
3519 
3520 	if (objset_get_dnode(spa, &spa->spa_mos_master,
3521 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3522 		printf("ZFS: failed to read pool %s directory object\n",
3523 		    spa->spa_name);
3524 		return (EIO);
3525 	}
3526 	/* this is allowed to fail, older pools do not have salt */
3527 	rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
3528 	    sizeof (spa->spa_cksum_salt.zcs_bytes),
3529 	    spa->spa_cksum_salt.zcs_bytes);
3530 
3531 	rc = check_mos_features(spa);
3532 	if (rc != 0) {
3533 		printf("ZFS: pool %s is not supported\n", spa->spa_name);
3534 		return (rc);
3535 	}
3536 
3537 	rc = zap_lookup(spa, &dir, DMU_POOL_CONFIG,
3538 	    sizeof (config_object), 1, &config_object);
3539 	if (rc != 0) {
3540 		printf("ZFS: can not read MOS %s\n", DMU_POOL_CONFIG);
3541 		return (EIO);
3542 	}
3543 	rc = load_nvlist(spa, config_object, &nvlist);
3544 	if (rc != 0)
3545 		return (rc);
3546 
3547 	rc = zap_lookup(spa, &dir, DMU_POOL_ZPOOL_CHECKPOINT,
3548 	    sizeof(uint64_t), sizeof(checkpoint) / sizeof(uint64_t),
3549 	    &checkpoint);
3550 	if (rc == 0 && checkpoint.ub_checkpoint_txg != 0) {
3551 		memcpy(&spa->spa_uberblock_checkpoint, &checkpoint,
3552 		    sizeof(checkpoint));
3553 		if (zio_read(spa, &spa->spa_uberblock_checkpoint.ub_rootbp,
3554 		    &spa->spa_mos_checkpoint)) {
3555 			printf("ZFS: can not read checkpoint data.\n");
3556 			return (EIO);
3557 		}
3558 	}
3559 
3560 	/*
3561 	 * Update vdevs from MOS config. Note, we do skip encoding bytes
3562 	 * here. See also vdev_label_read_config().
3563 	 */
3564 	rc = vdev_init_from_nvlist(spa, nvlist);
3565 	nvlist_destroy(nvlist);
3566 	return (rc);
3567 }
3568 
3569 static int
3570 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
3571 {
3572 
3573 	if (dn->dn_bonustype != DMU_OT_SA) {
3574 		znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
3575 
3576 		sb->st_mode = zp->zp_mode;
3577 		sb->st_uid = zp->zp_uid;
3578 		sb->st_gid = zp->zp_gid;
3579 		sb->st_size = zp->zp_size;
3580 	} else {
3581 		sa_hdr_phys_t *sahdrp;
3582 		int hdrsize;
3583 		size_t size = 0;
3584 		void *buf = NULL;
3585 
3586 		if (dn->dn_bonuslen != 0)
3587 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3588 		else {
3589 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
3590 				blkptr_t *bp = DN_SPILL_BLKPTR(dn);
3591 				int error;
3592 
3593 				size = BP_GET_LSIZE(bp);
3594 				buf = malloc(size);
3595 				if (buf == NULL)
3596 					error = ENOMEM;
3597 				else
3598 					error = zio_read(spa, bp, buf);
3599 
3600 				if (error != 0) {
3601 					free(buf);
3602 					return (error);
3603 				}
3604 				sahdrp = buf;
3605 			} else {
3606 				return (EIO);
3607 			}
3608 		}
3609 		hdrsize = SA_HDR_SIZE(sahdrp);
3610 		sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
3611 		    SA_MODE_OFFSET);
3612 		sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
3613 		    SA_UID_OFFSET);
3614 		sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
3615 		    SA_GID_OFFSET);
3616 		sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
3617 		    SA_SIZE_OFFSET);
3618 		free(buf);
3619 	}
3620 
3621 	return (0);
3622 }
3623 
3624 static int
3625 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
3626 {
3627 	int rc = 0;
3628 
3629 	if (dn->dn_bonustype == DMU_OT_SA) {
3630 		sa_hdr_phys_t *sahdrp = NULL;
3631 		size_t size = 0;
3632 		void *buf = NULL;
3633 		int hdrsize;
3634 		char *p;
3635 
3636 		if (dn->dn_bonuslen != 0) {
3637 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3638 		} else {
3639 			blkptr_t *bp;
3640 
3641 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
3642 				return (EIO);
3643 			bp = DN_SPILL_BLKPTR(dn);
3644 
3645 			size = BP_GET_LSIZE(bp);
3646 			buf = malloc(size);
3647 			if (buf == NULL)
3648 				rc = ENOMEM;
3649 			else
3650 				rc = zio_read(spa, bp, buf);
3651 			if (rc != 0) {
3652 				free(buf);
3653 				return (rc);
3654 			}
3655 			sahdrp = buf;
3656 		}
3657 		hdrsize = SA_HDR_SIZE(sahdrp);
3658 		p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
3659 		memcpy(path, p, psize);
3660 		free(buf);
3661 		return (0);
3662 	}
3663 	/*
3664 	 * Second test is purely to silence bogus compiler
3665 	 * warning about accessing past the end of dn_bonus.
3666 	 */
3667 	if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
3668 	    sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
3669 		memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
3670 	} else {
3671 		rc = dnode_read(spa, dn, 0, path, psize);
3672 	}
3673 	return (rc);
3674 }
3675 
3676 struct obj_list {
3677 	uint64_t		objnum;
3678 	STAILQ_ENTRY(obj_list)	entry;
3679 };
3680 
3681 /*
3682  * Lookup a file and return its dnode.
3683  */
3684 static int
3685 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
3686 {
3687 	int rc;
3688 	uint64_t objnum;
3689 	const spa_t *spa;
3690 	dnode_phys_t dn;
3691 	const char *p, *q;
3692 	char element[256];
3693 	char path[1024];
3694 	int symlinks_followed = 0;
3695 	struct stat sb;
3696 	struct obj_list *entry, *tentry;
3697 	STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
3698 
3699 	spa = mount->spa;
3700 	if (mount->objset.os_type != DMU_OST_ZFS) {
3701 		printf("ZFS: unexpected object set type %ju\n",
3702 		    (uintmax_t)mount->objset.os_type);
3703 		return (EIO);
3704 	}
3705 
3706 	if ((entry = malloc(sizeof(struct obj_list))) == NULL)
3707 		return (ENOMEM);
3708 
3709 	/*
3710 	 * Get the root directory dnode.
3711 	 */
3712 	rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
3713 	if (rc) {
3714 		free(entry);
3715 		return (rc);
3716 	}
3717 
3718 	rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof(objnum), 1, &objnum);
3719 	if (rc) {
3720 		free(entry);
3721 		return (rc);
3722 	}
3723 	entry->objnum = objnum;
3724 	STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3725 
3726 	rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3727 	if (rc != 0)
3728 		goto done;
3729 
3730 	p = upath;
3731 	while (p && *p) {
3732 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3733 		if (rc != 0)
3734 			goto done;
3735 
3736 		while (*p == '/')
3737 			p++;
3738 		if (*p == '\0')
3739 			break;
3740 		q = p;
3741 		while (*q != '\0' && *q != '/')
3742 			q++;
3743 
3744 		/* skip dot */
3745 		if (p + 1 == q && p[0] == '.') {
3746 			p++;
3747 			continue;
3748 		}
3749 		/* double dot */
3750 		if (p + 2 == q && p[0] == '.' && p[1] == '.') {
3751 			p += 2;
3752 			if (STAILQ_FIRST(&on_cache) ==
3753 			    STAILQ_LAST(&on_cache, obj_list, entry)) {
3754 				rc = ENOENT;
3755 				goto done;
3756 			}
3757 			entry = STAILQ_FIRST(&on_cache);
3758 			STAILQ_REMOVE_HEAD(&on_cache, entry);
3759 			free(entry);
3760 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3761 			continue;
3762 		}
3763 		if (q - p + 1 > sizeof(element)) {
3764 			rc = ENAMETOOLONG;
3765 			goto done;
3766 		}
3767 		memcpy(element, p, q - p);
3768 		element[q - p] = 0;
3769 		p = q;
3770 
3771 		if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
3772 			goto done;
3773 		if (!S_ISDIR(sb.st_mode)) {
3774 			rc = ENOTDIR;
3775 			goto done;
3776 		}
3777 
3778 		rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum);
3779 		if (rc)
3780 			goto done;
3781 		objnum = ZFS_DIRENT_OBJ(objnum);
3782 
3783 		if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
3784 			rc = ENOMEM;
3785 			goto done;
3786 		}
3787 		entry->objnum = objnum;
3788 		STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3789 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3790 		if (rc)
3791 			goto done;
3792 
3793 		/*
3794 		 * Check for symlink.
3795 		 */
3796 		rc = zfs_dnode_stat(spa, &dn, &sb);
3797 		if (rc)
3798 			goto done;
3799 		if (S_ISLNK(sb.st_mode)) {
3800 			if (symlinks_followed > 10) {
3801 				rc = EMLINK;
3802 				goto done;
3803 			}
3804 			symlinks_followed++;
3805 
3806 			/*
3807 			 * Read the link value and copy the tail of our
3808 			 * current path onto the end.
3809 			 */
3810 			if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
3811 				rc = ENAMETOOLONG;
3812 				goto done;
3813 			}
3814 			strcpy(&path[sb.st_size], p);
3815 
3816 			rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
3817 			if (rc != 0)
3818 				goto done;
3819 
3820 			/*
3821 			 * Restart with the new path, starting either at
3822 			 * the root or at the parent depending whether or
3823 			 * not the link is relative.
3824 			 */
3825 			p = path;
3826 			if (*p == '/') {
3827 				while (STAILQ_FIRST(&on_cache) !=
3828 				    STAILQ_LAST(&on_cache, obj_list, entry)) {
3829 					entry = STAILQ_FIRST(&on_cache);
3830 					STAILQ_REMOVE_HEAD(&on_cache, entry);
3831 					free(entry);
3832 				}
3833 			} else {
3834 				entry = STAILQ_FIRST(&on_cache);
3835 				STAILQ_REMOVE_HEAD(&on_cache, entry);
3836 				free(entry);
3837 			}
3838 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3839 		}
3840 	}
3841 
3842 	*dnode = dn;
3843 done:
3844 	STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)
3845 		free(entry);
3846 	return (rc);
3847 }
3848 
3849 /*
3850  * Return either a cached copy of the bootenv, or read each of the vdev children
3851  * looking for the bootenv. Cache what's found and return the results. Returns 0
3852  * when benvp is filled in, and some errno when not.
3853  */
3854 static int
3855 zfs_get_bootenv_spa(spa_t *spa, nvlist_t **benvp)
3856 {
3857 	vdev_t *vd;
3858 	nvlist_t *benv = NULL;
3859 
3860 	if (spa->spa_bootenv == NULL) {
3861 		STAILQ_FOREACH(vd, &spa->spa_root_vdev->v_children,
3862 		    v_childlink) {
3863 			benv = vdev_read_bootenv(vd);
3864 
3865 			if (benv != NULL)
3866 				break;
3867 		}
3868 		spa->spa_bootenv = benv;
3869 	}
3870 	benv = spa->spa_bootenv;
3871 
3872 	if (benv == NULL)
3873 		return (ENOENT);
3874 
3875 	*benvp = benv;
3876 	return (0);
3877 }
3878 
3879 /*
3880  * Store nvlist to pool label bootenv area. Also updates cached pointer in spa.
3881  */
3882 static int
3883 zfs_set_bootenv_spa(spa_t *spa, nvlist_t *benv)
3884 {
3885 	vdev_t *vd;
3886 
3887 	STAILQ_FOREACH(vd, &spa->spa_root_vdev->v_children, v_childlink) {
3888 		vdev_write_bootenv(vd, benv);
3889 	}
3890 
3891 	spa->spa_bootenv = benv;
3892 	return (0);
3893 }
3894 
3895 /*
3896  * Get bootonce value by key. The bootonce <key, value> pair is removed from the
3897  * bootenv nvlist and the remaining nvlist is committed back to disk. This process
3898  * the bootonce flag since we've reached the point in the boot that we've 'used'
3899  * the BE. For chained boot scenarios, we may reach this point multiple times (but
3900  * only remove it and return 0 the first time).
3901  */
3902 static int
3903 zfs_get_bootonce_spa(spa_t *spa, const char *key, char *buf, size_t size)
3904 {
3905 	nvlist_t *benv;
3906 	char *result = NULL;
3907 	int result_size, rv;
3908 
3909 	if ((rv = zfs_get_bootenv_spa(spa, &benv)) != 0)
3910 		return (rv);
3911 
3912 	if ((rv = nvlist_find(benv, key, DATA_TYPE_STRING, NULL,
3913 	    &result, &result_size)) == 0) {
3914 		if (result_size == 0) {
3915 			/* ignore empty string */
3916 			rv = ENOENT;
3917 		} else if (buf != NULL) {
3918 			size = MIN((size_t)result_size + 1, size);
3919 			strlcpy(buf, result, size);
3920 		}
3921 		(void)nvlist_remove(benv, key, DATA_TYPE_STRING);
3922 		(void)zfs_set_bootenv_spa(spa, benv);
3923 	}
3924 
3925 	return (rv);
3926 }
3927