xref: /freebsd/stand/libsa/zfs/zfsimpl.c (revision 5e801ac66d24704442eba426ed13c3effb8a34e7)
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 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 /*
31  *	Stand-alone ZFS file reader.
32  */
33 
34 #include <stdbool.h>
35 #include <sys/endian.h>
36 #include <sys/stat.h>
37 #include <sys/stdint.h>
38 #include <sys/list.h>
39 #include <sys/zfs_bootenv.h>
40 #include <machine/_inttypes.h>
41 
42 #include "zfsimpl.h"
43 #include "zfssubr.c"
44 
45 #ifdef HAS_ZSTD_ZFS
46 extern int zstd_init(void);
47 #endif
48 
49 struct zfsmount {
50 	char			*path;
51 	const spa_t		*spa;
52 	objset_phys_t		objset;
53 	uint64_t		rootobj;
54 	STAILQ_ENTRY(zfsmount)	next;
55 };
56 
57 typedef STAILQ_HEAD(zfs_mnt_list, zfsmount) zfs_mnt_list_t;
58 static zfs_mnt_list_t zfsmount = STAILQ_HEAD_INITIALIZER(zfsmount);
59 
60 /*
61  * The indirect_child_t represents the vdev that we will read from, when we
62  * need to read all copies of the data (e.g. for scrub or reconstruction).
63  * For plain (non-mirror) top-level vdevs (i.e. is_vdev is not a mirror),
64  * ic_vdev is the same as is_vdev.  However, for mirror top-level vdevs,
65  * ic_vdev is a child of the mirror.
66  */
67 typedef struct indirect_child {
68 	void *ic_data;
69 	vdev_t *ic_vdev;
70 } indirect_child_t;
71 
72 /*
73  * The indirect_split_t represents one mapped segment of an i/o to the
74  * indirect vdev. For non-split (contiguously-mapped) blocks, there will be
75  * only one indirect_split_t, with is_split_offset==0 and is_size==io_size.
76  * For split blocks, there will be several of these.
77  */
78 typedef struct indirect_split {
79 	list_node_t is_node; /* link on iv_splits */
80 
81 	/*
82 	 * is_split_offset is the offset into the i/o.
83 	 * This is the sum of the previous splits' is_size's.
84 	 */
85 	uint64_t is_split_offset;
86 
87 	vdev_t *is_vdev; /* top-level vdev */
88 	uint64_t is_target_offset; /* offset on is_vdev */
89 	uint64_t is_size;
90 	int is_children; /* number of entries in is_child[] */
91 
92 	/*
93 	 * is_good_child is the child that we are currently using to
94 	 * attempt reconstruction.
95 	 */
96 	int is_good_child;
97 
98 	indirect_child_t is_child[1]; /* variable-length */
99 } indirect_split_t;
100 
101 /*
102  * The indirect_vsd_t is associated with each i/o to the indirect vdev.
103  * It is the "Vdev-Specific Data" in the zio_t's io_vsd.
104  */
105 typedef struct indirect_vsd {
106 	boolean_t iv_split_block;
107 	boolean_t iv_reconstruct;
108 
109 	list_t iv_splits; /* list of indirect_split_t's */
110 } indirect_vsd_t;
111 
112 /*
113  * List of all vdevs, chained through v_alllink.
114  */
115 static vdev_list_t zfs_vdevs;
116 
117 /*
118  * List of ZFS features supported for read
119  */
120 static const char *features_for_read[] = {
121 	"org.illumos:lz4_compress",
122 	"com.delphix:hole_birth",
123 	"com.delphix:extensible_dataset",
124 	"com.delphix:embedded_data",
125 	"org.open-zfs:large_blocks",
126 	"org.illumos:sha512",
127 	"org.illumos:skein",
128 	"org.zfsonlinux:large_dnode",
129 	"com.joyent:multi_vdev_crash_dump",
130 	"com.delphix:spacemap_histogram",
131 	"com.delphix:zpool_checkpoint",
132 	"com.delphix:spacemap_v2",
133 	"com.datto:encryption",
134 	"com.datto:bookmark_v2",
135 	"org.zfsonlinux:allocation_classes",
136 	"com.datto:resilver_defer",
137 	"com.delphix:device_removal",
138 	"com.delphix:obsolete_counts",
139 	"com.intel:allocation_classes",
140 	"org.freebsd:zstd_compress",
141 	"com.delphix:bookmark_written",
142 	NULL
143 };
144 
145 /*
146  * List of all pools, chained through spa_link.
147  */
148 static spa_list_t zfs_pools;
149 
150 static const dnode_phys_t *dnode_cache_obj;
151 static uint64_t dnode_cache_bn;
152 static char *dnode_cache_buf;
153 
154 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
155 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
156 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
157 static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode,
158     const char *name, uint64_t integer_size, uint64_t num_integers,
159     void *value);
160 static int objset_get_dnode(const spa_t *, const objset_phys_t *, uint64_t,
161     dnode_phys_t *);
162 static int dnode_read(const spa_t *, const dnode_phys_t *, off_t, void *,
163     size_t);
164 static int vdev_indirect_read(vdev_t *, const blkptr_t *, void *, off_t,
165     size_t);
166 static int vdev_mirror_read(vdev_t *, const blkptr_t *, void *, off_t, size_t);
167 vdev_indirect_mapping_t *vdev_indirect_mapping_open(spa_t *, objset_phys_t *,
168     uint64_t);
169 vdev_indirect_mapping_entry_phys_t *
170     vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *, uint64_t,
171     uint64_t, uint64_t *);
172 
173 static void
174 zfs_init(void)
175 {
176 	STAILQ_INIT(&zfs_vdevs);
177 	STAILQ_INIT(&zfs_pools);
178 
179 	dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
180 
181 	zfs_init_crc();
182 #ifdef HAS_ZSTD_ZFS
183 	zstd_init();
184 #endif
185 }
186 
187 static int
188 nvlist_check_features_for_read(nvlist_t *nvl)
189 {
190 	nvlist_t *features = NULL;
191 	nvs_data_t *data;
192 	nvp_header_t *nvp;
193 	nv_string_t *nvp_name;
194 	int rc;
195 
196 	rc = nvlist_find(nvl, ZPOOL_CONFIG_FEATURES_FOR_READ,
197 	    DATA_TYPE_NVLIST, NULL, &features, NULL);
198 	switch (rc) {
199 	case 0:
200 		break;		/* Continue with checks */
201 
202 	case ENOENT:
203 		return (0);	/* All features are disabled */
204 
205 	default:
206 		return (rc);	/* Error while reading nvlist */
207 	}
208 
209 	data = (nvs_data_t *)features->nv_data;
210 	nvp = &data->nvl_pair;	/* first pair in nvlist */
211 
212 	while (nvp->encoded_size != 0 && nvp->decoded_size != 0) {
213 		int i, found;
214 
215 		nvp_name = (nv_string_t *)((uintptr_t)nvp + sizeof(*nvp));
216 		found = 0;
217 
218 		for (i = 0; features_for_read[i] != NULL; i++) {
219 			if (memcmp(nvp_name->nv_data, features_for_read[i],
220 			    nvp_name->nv_size) == 0) {
221 				found = 1;
222 				break;
223 			}
224 		}
225 
226 		if (!found) {
227 			printf("ZFS: unsupported feature: %.*s\n",
228 			    nvp_name->nv_size, nvp_name->nv_data);
229 			rc = EIO;
230 		}
231 		nvp = (nvp_header_t *)((uint8_t *)nvp + nvp->encoded_size);
232 	}
233 	nvlist_destroy(features);
234 
235 	return (rc);
236 }
237 
238 static int
239 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
240     off_t offset, size_t size)
241 {
242 	size_t psize;
243 	int rc;
244 
245 	if (vdev->v_phys_read == NULL)
246 		return (ENOTSUP);
247 
248 	if (bp) {
249 		psize = BP_GET_PSIZE(bp);
250 	} else {
251 		psize = size;
252 	}
253 
254 	rc = vdev->v_phys_read(vdev, vdev->v_priv, offset, buf, psize);
255 	if (rc == 0) {
256 		if (bp != NULL)
257 			rc = zio_checksum_verify(vdev->v_spa, bp, buf);
258 	}
259 
260 	return (rc);
261 }
262 
263 static int
264 vdev_write_phys(vdev_t *vdev, void *buf, off_t offset, size_t size)
265 {
266 	if (vdev->v_phys_write == NULL)
267 		return (ENOTSUP);
268 
269 	return (vdev->v_phys_write(vdev, offset, buf, size));
270 }
271 
272 typedef struct remap_segment {
273 	vdev_t *rs_vd;
274 	uint64_t rs_offset;
275 	uint64_t rs_asize;
276 	uint64_t rs_split_offset;
277 	list_node_t rs_node;
278 } remap_segment_t;
279 
280 static remap_segment_t *
281 rs_alloc(vdev_t *vd, uint64_t offset, uint64_t asize, uint64_t split_offset)
282 {
283 	remap_segment_t *rs = malloc(sizeof (remap_segment_t));
284 
285 	if (rs != NULL) {
286 		rs->rs_vd = vd;
287 		rs->rs_offset = offset;
288 		rs->rs_asize = asize;
289 		rs->rs_split_offset = split_offset;
290 	}
291 
292 	return (rs);
293 }
294 
295 vdev_indirect_mapping_t *
296 vdev_indirect_mapping_open(spa_t *spa, objset_phys_t *os,
297     uint64_t mapping_object)
298 {
299 	vdev_indirect_mapping_t *vim;
300 	vdev_indirect_mapping_phys_t *vim_phys;
301 	int rc;
302 
303 	vim = calloc(1, sizeof (*vim));
304 	if (vim == NULL)
305 		return (NULL);
306 
307 	vim->vim_dn = calloc(1, sizeof (*vim->vim_dn));
308 	if (vim->vim_dn == NULL) {
309 		free(vim);
310 		return (NULL);
311 	}
312 
313 	rc = objset_get_dnode(spa, os, mapping_object, vim->vim_dn);
314 	if (rc != 0) {
315 		free(vim->vim_dn);
316 		free(vim);
317 		return (NULL);
318 	}
319 
320 	vim->vim_spa = spa;
321 	vim->vim_phys = malloc(sizeof (*vim->vim_phys));
322 	if (vim->vim_phys == NULL) {
323 		free(vim->vim_dn);
324 		free(vim);
325 		return (NULL);
326 	}
327 
328 	vim_phys = (vdev_indirect_mapping_phys_t *)DN_BONUS(vim->vim_dn);
329 	*vim->vim_phys = *vim_phys;
330 
331 	vim->vim_objset = os;
332 	vim->vim_object = mapping_object;
333 	vim->vim_entries = NULL;
334 
335 	vim->vim_havecounts =
336 	    (vim->vim_dn->dn_bonuslen > VDEV_INDIRECT_MAPPING_SIZE_V0);
337 
338 	return (vim);
339 }
340 
341 /*
342  * Compare an offset with an indirect mapping entry; there are three
343  * possible scenarios:
344  *
345  *     1. The offset is "less than" the mapping entry; meaning the
346  *        offset is less than the source offset of the mapping entry. In
347  *        this case, there is no overlap between the offset and the
348  *        mapping entry and -1 will be returned.
349  *
350  *     2. The offset is "greater than" the mapping entry; meaning the
351  *        offset is greater than the mapping entry's source offset plus
352  *        the entry's size. In this case, there is no overlap between
353  *        the offset and the mapping entry and 1 will be returned.
354  *
355  *        NOTE: If the offset is actually equal to the entry's offset
356  *        plus size, this is considered to be "greater" than the entry,
357  *        and this case applies (i.e. 1 will be returned). Thus, the
358  *        entry's "range" can be considered to be inclusive at its
359  *        start, but exclusive at its end: e.g. [src, src + size).
360  *
361  *     3. The last case to consider is if the offset actually falls
362  *        within the mapping entry's range. If this is the case, the
363  *        offset is considered to be "equal to" the mapping entry and
364  *        0 will be returned.
365  *
366  *        NOTE: If the offset is equal to the entry's source offset,
367  *        this case applies and 0 will be returned. If the offset is
368  *        equal to the entry's source plus its size, this case does
369  *        *not* apply (see "NOTE" above for scenario 2), and 1 will be
370  *        returned.
371  */
372 static int
373 dva_mapping_overlap_compare(const void *v_key, const void *v_array_elem)
374 {
375 	const uint64_t *key = v_key;
376 	const vdev_indirect_mapping_entry_phys_t *array_elem =
377 	    v_array_elem;
378 	uint64_t src_offset = DVA_MAPPING_GET_SRC_OFFSET(array_elem);
379 
380 	if (*key < src_offset) {
381 		return (-1);
382 	} else if (*key < src_offset + DVA_GET_ASIZE(&array_elem->vimep_dst)) {
383 		return (0);
384 	} else {
385 		return (1);
386 	}
387 }
388 
389 /*
390  * Return array entry.
391  */
392 static vdev_indirect_mapping_entry_phys_t *
393 vdev_indirect_mapping_entry(vdev_indirect_mapping_t *vim, uint64_t index)
394 {
395 	uint64_t size;
396 	off_t offset = 0;
397 	int rc;
398 
399 	if (vim->vim_phys->vimp_num_entries == 0)
400 		return (NULL);
401 
402 	if (vim->vim_entries == NULL) {
403 		uint64_t bsize;
404 
405 		bsize = vim->vim_dn->dn_datablkszsec << SPA_MINBLOCKSHIFT;
406 		size = vim->vim_phys->vimp_num_entries *
407 		    sizeof (*vim->vim_entries);
408 		if (size > bsize) {
409 			size = bsize / sizeof (*vim->vim_entries);
410 			size *= sizeof (*vim->vim_entries);
411 		}
412 		vim->vim_entries = malloc(size);
413 		if (vim->vim_entries == NULL)
414 			return (NULL);
415 		vim->vim_num_entries = size / sizeof (*vim->vim_entries);
416 		offset = index * sizeof (*vim->vim_entries);
417 	}
418 
419 	/* We have data in vim_entries */
420 	if (offset == 0) {
421 		if (index >= vim->vim_entry_offset &&
422 		    index <= vim->vim_entry_offset + vim->vim_num_entries) {
423 			index -= vim->vim_entry_offset;
424 			return (&vim->vim_entries[index]);
425 		}
426 		offset = index * sizeof (*vim->vim_entries);
427 	}
428 
429 	vim->vim_entry_offset = index;
430 	size = vim->vim_num_entries * sizeof (*vim->vim_entries);
431 	rc = dnode_read(vim->vim_spa, vim->vim_dn, offset, vim->vim_entries,
432 	    size);
433 	if (rc != 0) {
434 		/* Read error, invalidate vim_entries. */
435 		free(vim->vim_entries);
436 		vim->vim_entries = NULL;
437 		return (NULL);
438 	}
439 	index -= vim->vim_entry_offset;
440 	return (&vim->vim_entries[index]);
441 }
442 
443 /*
444  * Returns the mapping entry for the given offset.
445  *
446  * It's possible that the given offset will not be in the mapping table
447  * (i.e. no mapping entries contain this offset), in which case, the
448  * return value value depends on the "next_if_missing" parameter.
449  *
450  * If the offset is not found in the table and "next_if_missing" is
451  * B_FALSE, then NULL will always be returned. The behavior is intended
452  * to allow consumers to get the entry corresponding to the offset
453  * parameter, iff the offset overlaps with an entry in the table.
454  *
455  * If the offset is not found in the table and "next_if_missing" is
456  * B_TRUE, then the entry nearest to the given offset will be returned,
457  * such that the entry's source offset is greater than the offset
458  * passed in (i.e. the "next" mapping entry in the table is returned, if
459  * the offset is missing from the table). If there are no entries whose
460  * source offset is greater than the passed in offset, NULL is returned.
461  */
462 static vdev_indirect_mapping_entry_phys_t *
463 vdev_indirect_mapping_entry_for_offset(vdev_indirect_mapping_t *vim,
464     uint64_t offset)
465 {
466 	ASSERT(vim->vim_phys->vimp_num_entries > 0);
467 
468 	vdev_indirect_mapping_entry_phys_t *entry;
469 
470 	uint64_t last = vim->vim_phys->vimp_num_entries - 1;
471 	uint64_t base = 0;
472 
473 	/*
474 	 * We don't define these inside of the while loop because we use
475 	 * their value in the case that offset isn't in the mapping.
476 	 */
477 	uint64_t mid;
478 	int result;
479 
480 	while (last >= base) {
481 		mid = base + ((last - base) >> 1);
482 
483 		entry = vdev_indirect_mapping_entry(vim, mid);
484 		if (entry == NULL)
485 			break;
486 		result = dva_mapping_overlap_compare(&offset, entry);
487 
488 		if (result == 0) {
489 			break;
490 		} else if (result < 0) {
491 			last = mid - 1;
492 		} else {
493 			base = mid + 1;
494 		}
495 	}
496 	return (entry);
497 }
498 
499 /*
500  * Given an indirect vdev and an extent on that vdev, it duplicates the
501  * physical entries of the indirect mapping that correspond to the extent
502  * to a new array and returns a pointer to it. In addition, copied_entries
503  * is populated with the number of mapping entries that were duplicated.
504  *
505  * Finally, since we are doing an allocation, it is up to the caller to
506  * free the array allocated in this function.
507  */
508 vdev_indirect_mapping_entry_phys_t *
509 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *vd, uint64_t offset,
510     uint64_t asize, uint64_t *copied_entries)
511 {
512 	vdev_indirect_mapping_entry_phys_t *duplicate_mappings = NULL;
513 	vdev_indirect_mapping_t *vim = vd->v_mapping;
514 	uint64_t entries = 0;
515 
516 	vdev_indirect_mapping_entry_phys_t *first_mapping =
517 	    vdev_indirect_mapping_entry_for_offset(vim, offset);
518 	ASSERT3P(first_mapping, !=, NULL);
519 
520 	vdev_indirect_mapping_entry_phys_t *m = first_mapping;
521 	while (asize > 0) {
522 		uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
523 		uint64_t inner_offset = offset - DVA_MAPPING_GET_SRC_OFFSET(m);
524 		uint64_t inner_size = MIN(asize, size - inner_offset);
525 
526 		offset += inner_size;
527 		asize -= inner_size;
528 		entries++;
529 		m++;
530 	}
531 
532 	size_t copy_length = entries * sizeof (*first_mapping);
533 	duplicate_mappings = malloc(copy_length);
534 	if (duplicate_mappings != NULL)
535 		bcopy(first_mapping, duplicate_mappings, copy_length);
536 	else
537 		entries = 0;
538 
539 	*copied_entries = entries;
540 
541 	return (duplicate_mappings);
542 }
543 
544 static vdev_t *
545 vdev_lookup_top(spa_t *spa, uint64_t vdev)
546 {
547 	vdev_t *rvd;
548 	vdev_list_t *vlist;
549 
550 	vlist = &spa->spa_root_vdev->v_children;
551 	STAILQ_FOREACH(rvd, vlist, v_childlink)
552 		if (rvd->v_id == vdev)
553 			break;
554 
555 	return (rvd);
556 }
557 
558 /*
559  * This is a callback for vdev_indirect_remap() which allocates an
560  * indirect_split_t for each split segment and adds it to iv_splits.
561  */
562 static void
563 vdev_indirect_gather_splits(uint64_t split_offset, vdev_t *vd, uint64_t offset,
564     uint64_t size, void *arg)
565 {
566 	int n = 1;
567 	zio_t *zio = arg;
568 	indirect_vsd_t *iv = zio->io_vsd;
569 
570 	if (vd->v_read == vdev_indirect_read)
571 		return;
572 
573 	if (vd->v_read == vdev_mirror_read)
574 		n = vd->v_nchildren;
575 
576 	indirect_split_t *is =
577 	    malloc(offsetof(indirect_split_t, is_child[n]));
578 	if (is == NULL) {
579 		zio->io_error = ENOMEM;
580 		return;
581 	}
582 	bzero(is, offsetof(indirect_split_t, is_child[n]));
583 
584 	is->is_children = n;
585 	is->is_size = size;
586 	is->is_split_offset = split_offset;
587 	is->is_target_offset = offset;
588 	is->is_vdev = vd;
589 
590 	/*
591 	 * Note that we only consider multiple copies of the data for
592 	 * *mirror* vdevs.  We don't for "replacing" or "spare" vdevs, even
593 	 * though they use the same ops as mirror, because there's only one
594 	 * "good" copy under the replacing/spare.
595 	 */
596 	if (vd->v_read == vdev_mirror_read) {
597 		int i = 0;
598 		vdev_t *kid;
599 
600 		STAILQ_FOREACH(kid, &vd->v_children, v_childlink) {
601 			is->is_child[i++].ic_vdev = kid;
602 		}
603 	} else {
604 		is->is_child[0].ic_vdev = vd;
605 	}
606 
607 	list_insert_tail(&iv->iv_splits, is);
608 }
609 
610 static void
611 vdev_indirect_remap(vdev_t *vd, uint64_t offset, uint64_t asize, void *arg)
612 {
613 	list_t stack;
614 	spa_t *spa = vd->v_spa;
615 	zio_t *zio = arg;
616 	remap_segment_t *rs;
617 
618 	list_create(&stack, sizeof (remap_segment_t),
619 	    offsetof(remap_segment_t, rs_node));
620 
621 	rs = rs_alloc(vd, offset, asize, 0);
622 	if (rs == NULL) {
623 		printf("vdev_indirect_remap: out of memory.\n");
624 		zio->io_error = ENOMEM;
625 	}
626 	for (; rs != NULL; rs = list_remove_head(&stack)) {
627 		vdev_t *v = rs->rs_vd;
628 		uint64_t num_entries = 0;
629 		/* vdev_indirect_mapping_t *vim = v->v_mapping; */
630 		vdev_indirect_mapping_entry_phys_t *mapping =
631 		    vdev_indirect_mapping_duplicate_adjacent_entries(v,
632 		    rs->rs_offset, rs->rs_asize, &num_entries);
633 
634 		if (num_entries == 0)
635 			zio->io_error = ENOMEM;
636 
637 		for (uint64_t i = 0; i < num_entries; i++) {
638 			vdev_indirect_mapping_entry_phys_t *m = &mapping[i];
639 			uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
640 			uint64_t dst_offset = DVA_GET_OFFSET(&m->vimep_dst);
641 			uint64_t dst_vdev = DVA_GET_VDEV(&m->vimep_dst);
642 			uint64_t inner_offset = rs->rs_offset -
643 			    DVA_MAPPING_GET_SRC_OFFSET(m);
644 			uint64_t inner_size =
645 			    MIN(rs->rs_asize, size - inner_offset);
646 			vdev_t *dst_v = vdev_lookup_top(spa, dst_vdev);
647 
648 			if (dst_v->v_read == vdev_indirect_read) {
649 				remap_segment_t *o;
650 
651 				o = rs_alloc(dst_v, dst_offset + inner_offset,
652 				    inner_size, rs->rs_split_offset);
653 				if (o == NULL) {
654 					printf("vdev_indirect_remap: "
655 					    "out of memory.\n");
656 					zio->io_error = ENOMEM;
657 					break;
658 				}
659 
660 				list_insert_head(&stack, o);
661 			}
662 			vdev_indirect_gather_splits(rs->rs_split_offset, dst_v,
663 			    dst_offset + inner_offset,
664 			    inner_size, arg);
665 
666 			/*
667 			 * vdev_indirect_gather_splits can have memory
668 			 * allocation error, we can not recover from it.
669 			 */
670 			if (zio->io_error != 0)
671 				break;
672 			rs->rs_offset += inner_size;
673 			rs->rs_asize -= inner_size;
674 			rs->rs_split_offset += inner_size;
675 		}
676 
677 		free(mapping);
678 		free(rs);
679 		if (zio->io_error != 0)
680 			break;
681 	}
682 
683 	list_destroy(&stack);
684 }
685 
686 static void
687 vdev_indirect_map_free(zio_t *zio)
688 {
689 	indirect_vsd_t *iv = zio->io_vsd;
690 	indirect_split_t *is;
691 
692 	while ((is = list_head(&iv->iv_splits)) != NULL) {
693 		for (int c = 0; c < is->is_children; c++) {
694 			indirect_child_t *ic = &is->is_child[c];
695 			free(ic->ic_data);
696 		}
697 		list_remove(&iv->iv_splits, is);
698 		free(is);
699 	}
700 	free(iv);
701 }
702 
703 static int
704 vdev_indirect_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
705     off_t offset, size_t bytes)
706 {
707 	zio_t zio;
708 	spa_t *spa = vdev->v_spa;
709 	indirect_vsd_t *iv;
710 	indirect_split_t *first;
711 	int rc = EIO;
712 
713 	iv = calloc(1, sizeof(*iv));
714 	if (iv == NULL)
715 		return (ENOMEM);
716 
717 	list_create(&iv->iv_splits,
718 	    sizeof (indirect_split_t), offsetof(indirect_split_t, is_node));
719 
720 	bzero(&zio, sizeof(zio));
721 	zio.io_spa = spa;
722 	zio.io_bp = (blkptr_t *)bp;
723 	zio.io_data = buf;
724 	zio.io_size = bytes;
725 	zio.io_offset = offset;
726 	zio.io_vd = vdev;
727 	zio.io_vsd = iv;
728 
729 	if (vdev->v_mapping == NULL) {
730 		vdev_indirect_config_t *vic;
731 
732 		vic = &vdev->vdev_indirect_config;
733 		vdev->v_mapping = vdev_indirect_mapping_open(spa,
734 		    spa->spa_mos, vic->vic_mapping_object);
735 	}
736 
737 	vdev_indirect_remap(vdev, offset, bytes, &zio);
738 	if (zio.io_error != 0)
739 		return (zio.io_error);
740 
741 	first = list_head(&iv->iv_splits);
742 	if (first->is_size == zio.io_size) {
743 		/*
744 		 * This is not a split block; we are pointing to the entire
745 		 * data, which will checksum the same as the original data.
746 		 * Pass the BP down so that the child i/o can verify the
747 		 * checksum, and try a different location if available
748 		 * (e.g. on a mirror).
749 		 *
750 		 * While this special case could be handled the same as the
751 		 * general (split block) case, doing it this way ensures
752 		 * that the vast majority of blocks on indirect vdevs
753 		 * (which are not split) are handled identically to blocks
754 		 * on non-indirect vdevs.  This allows us to be less strict
755 		 * about performance in the general (but rare) case.
756 		 */
757 		rc = first->is_vdev->v_read(first->is_vdev, zio.io_bp,
758 		    zio.io_data, first->is_target_offset, bytes);
759 	} else {
760 		iv->iv_split_block = B_TRUE;
761 		/*
762 		 * Read one copy of each split segment, from the
763 		 * top-level vdev.  Since we don't know the
764 		 * checksum of each split individually, the child
765 		 * zio can't ensure that we get the right data.
766 		 * E.g. if it's a mirror, it will just read from a
767 		 * random (healthy) leaf vdev.  We have to verify
768 		 * the checksum in vdev_indirect_io_done().
769 		 */
770 		for (indirect_split_t *is = list_head(&iv->iv_splits);
771 		    is != NULL; is = list_next(&iv->iv_splits, is)) {
772 			char *ptr = zio.io_data;
773 
774 			rc = is->is_vdev->v_read(is->is_vdev, zio.io_bp,
775 			    ptr + is->is_split_offset, is->is_target_offset,
776 			    is->is_size);
777 		}
778 		if (zio_checksum_verify(spa, zio.io_bp, zio.io_data))
779 			rc = ECKSUM;
780 		else
781 			rc = 0;
782 	}
783 
784 	vdev_indirect_map_free(&zio);
785 	if (rc == 0)
786 		rc = zio.io_error;
787 
788 	return (rc);
789 }
790 
791 static int
792 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
793     off_t offset, size_t bytes)
794 {
795 
796 	return (vdev_read_phys(vdev, bp, buf,
797 	    offset + VDEV_LABEL_START_SIZE, bytes));
798 }
799 
800 static int
801 vdev_missing_read(vdev_t *vdev __unused, const blkptr_t *bp __unused,
802     void *buf __unused, off_t offset __unused, size_t bytes __unused)
803 {
804 
805 	return (ENOTSUP);
806 }
807 
808 static int
809 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
810     off_t offset, size_t bytes)
811 {
812 	vdev_t *kid;
813 	int rc;
814 
815 	rc = EIO;
816 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
817 		if (kid->v_state != VDEV_STATE_HEALTHY)
818 			continue;
819 		rc = kid->v_read(kid, bp, buf, offset, bytes);
820 		if (!rc)
821 			return (0);
822 	}
823 
824 	return (rc);
825 }
826 
827 static int
828 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
829     off_t offset, size_t bytes)
830 {
831 	vdev_t *kid;
832 
833 	/*
834 	 * Here we should have two kids:
835 	 * First one which is the one we are replacing and we can trust
836 	 * only this one to have valid data, but it might not be present.
837 	 * Second one is that one we are replacing with. It is most likely
838 	 * healthy, but we can't trust it has needed data, so we won't use it.
839 	 */
840 	kid = STAILQ_FIRST(&vdev->v_children);
841 	if (kid == NULL)
842 		return (EIO);
843 	if (kid->v_state != VDEV_STATE_HEALTHY)
844 		return (EIO);
845 	return (kid->v_read(kid, bp, buf, offset, bytes));
846 }
847 
848 static vdev_t *
849 vdev_find(uint64_t guid)
850 {
851 	vdev_t *vdev;
852 
853 	STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
854 		if (vdev->v_guid == guid)
855 			return (vdev);
856 
857 	return (0);
858 }
859 
860 static vdev_t *
861 vdev_create(uint64_t guid, vdev_read_t *_read)
862 {
863 	vdev_t *vdev;
864 	vdev_indirect_config_t *vic;
865 
866 	vdev = calloc(1, sizeof(vdev_t));
867 	if (vdev != NULL) {
868 		STAILQ_INIT(&vdev->v_children);
869 		vdev->v_guid = guid;
870 		vdev->v_read = _read;
871 
872 		/*
873 		 * root vdev has no read function, we use this fact to
874 		 * skip setting up data we do not need for root vdev.
875 		 * We only point root vdev from spa.
876 		 */
877 		if (_read != NULL) {
878 			vic = &vdev->vdev_indirect_config;
879 			vic->vic_prev_indirect_vdev = UINT64_MAX;
880 			STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
881 		}
882 	}
883 
884 	return (vdev);
885 }
886 
887 static void
888 vdev_set_initial_state(vdev_t *vdev, const nvlist_t *nvlist)
889 {
890 	uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
891 	uint64_t is_log;
892 
893 	is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
894 	is_log = 0;
895 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
896 	    &is_offline, NULL);
897 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
898 	    &is_removed, NULL);
899 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
900 	    &is_faulted, NULL);
901 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64,
902 	    NULL, &is_degraded, NULL);
903 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64,
904 	    NULL, &isnt_present, NULL);
905 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, NULL,
906 	    &is_log, NULL);
907 
908 	if (is_offline != 0)
909 		vdev->v_state = VDEV_STATE_OFFLINE;
910 	else if (is_removed != 0)
911 		vdev->v_state = VDEV_STATE_REMOVED;
912 	else if (is_faulted != 0)
913 		vdev->v_state = VDEV_STATE_FAULTED;
914 	else if (is_degraded != 0)
915 		vdev->v_state = VDEV_STATE_DEGRADED;
916 	else if (isnt_present != 0)
917 		vdev->v_state = VDEV_STATE_CANT_OPEN;
918 
919 	vdev->v_islog = is_log != 0;
920 }
921 
922 static int
923 vdev_init(uint64_t guid, const nvlist_t *nvlist, vdev_t **vdevp)
924 {
925 	uint64_t id, ashift, asize, nparity;
926 	const char *path;
927 	const char *type;
928 	int len, pathlen;
929 	char *name;
930 	vdev_t *vdev;
931 
932 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id,
933 	    NULL) ||
934 	    nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING, NULL,
935 	    &type, &len)) {
936 		return (ENOENT);
937 	}
938 
939 	if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 &&
940 	    memcmp(type, VDEV_TYPE_DISK, len) != 0 &&
941 #ifdef ZFS_TEST
942 	    memcmp(type, VDEV_TYPE_FILE, len) != 0 &&
943 #endif
944 	    memcmp(type, VDEV_TYPE_RAIDZ, len) != 0 &&
945 	    memcmp(type, VDEV_TYPE_INDIRECT, len) != 0 &&
946 	    memcmp(type, VDEV_TYPE_REPLACING, len) != 0 &&
947 	    memcmp(type, VDEV_TYPE_HOLE, len) != 0) {
948 		printf("ZFS: can only boot from disk, mirror, raidz1, "
949 		    "raidz2 and raidz3 vdevs, got: %.*s\n", len, type);
950 		return (EIO);
951 	}
952 
953 	if (memcmp(type, VDEV_TYPE_MIRROR, len) == 0)
954 		vdev = vdev_create(guid, vdev_mirror_read);
955 	else if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0)
956 		vdev = vdev_create(guid, vdev_raidz_read);
957 	else if (memcmp(type, VDEV_TYPE_REPLACING, len) == 0)
958 		vdev = vdev_create(guid, vdev_replacing_read);
959 	else if (memcmp(type, VDEV_TYPE_INDIRECT, len) == 0) {
960 		vdev_indirect_config_t *vic;
961 
962 		vdev = vdev_create(guid, vdev_indirect_read);
963 		if (vdev != NULL) {
964 			vdev->v_state = VDEV_STATE_HEALTHY;
965 			vic = &vdev->vdev_indirect_config;
966 
967 			nvlist_find(nvlist,
968 			    ZPOOL_CONFIG_INDIRECT_OBJECT,
969 			    DATA_TYPE_UINT64,
970 			    NULL, &vic->vic_mapping_object, NULL);
971 			nvlist_find(nvlist,
972 			    ZPOOL_CONFIG_INDIRECT_BIRTHS,
973 			    DATA_TYPE_UINT64,
974 			    NULL, &vic->vic_births_object, NULL);
975 			nvlist_find(nvlist,
976 			    ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
977 			    DATA_TYPE_UINT64,
978 			    NULL, &vic->vic_prev_indirect_vdev, NULL);
979 		}
980 	} else if (memcmp(type, VDEV_TYPE_HOLE, len) == 0) {
981 		vdev = vdev_create(guid, vdev_missing_read);
982 	} else {
983 		vdev = vdev_create(guid, vdev_disk_read);
984 	}
985 
986 	if (vdev == NULL)
987 		return (ENOMEM);
988 
989 	vdev_set_initial_state(vdev, nvlist);
990 	vdev->v_id = id;
991 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
992 	    DATA_TYPE_UINT64, NULL, &ashift, NULL) == 0)
993 		vdev->v_ashift = ashift;
994 
995 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE,
996 	    DATA_TYPE_UINT64, NULL, &asize, NULL) == 0) {
997 		vdev->v_psize = asize +
998 		    VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
999 	}
1000 
1001 	if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
1002 	    DATA_TYPE_UINT64, NULL, &nparity, NULL) == 0)
1003 		vdev->v_nparity = nparity;
1004 
1005 	if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
1006 	    DATA_TYPE_STRING, NULL, &path, &pathlen) == 0) {
1007 		char prefix[] = "/dev/";
1008 
1009 		len = strlen(prefix);
1010 		if (len < pathlen && memcmp(path, prefix, len) == 0) {
1011 			path += len;
1012 			pathlen -= len;
1013 		}
1014 		name = malloc(pathlen + 1);
1015 		bcopy(path, name, pathlen);
1016 		name[pathlen] = '\0';
1017 		vdev->v_name = name;
1018 	} else {
1019 		name = NULL;
1020 		if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) {
1021 			if (vdev->v_nparity < 1 ||
1022 			    vdev->v_nparity > 3) {
1023 				printf("ZFS: invalid raidz parity: %d\n",
1024 				    vdev->v_nparity);
1025 				return (EIO);
1026 			}
1027 			(void) asprintf(&name, "%.*s%d-%" PRIu64, len, type,
1028 			    vdev->v_nparity, id);
1029 		} else {
1030 			(void) asprintf(&name, "%.*s-%" PRIu64, len, type, id);
1031 		}
1032 		vdev->v_name = name;
1033 	}
1034 	*vdevp = vdev;
1035 	return (0);
1036 }
1037 
1038 /*
1039  * Find slot for vdev. We return either NULL to signal to use
1040  * STAILQ_INSERT_HEAD, or we return link element to be used with
1041  * STAILQ_INSERT_AFTER.
1042  */
1043 static vdev_t *
1044 vdev_find_previous(vdev_t *top_vdev, vdev_t *vdev)
1045 {
1046 	vdev_t *v, *previous;
1047 
1048 	if (STAILQ_EMPTY(&top_vdev->v_children))
1049 		return (NULL);
1050 
1051 	previous = NULL;
1052 	STAILQ_FOREACH(v, &top_vdev->v_children, v_childlink) {
1053 		if (v->v_id > vdev->v_id)
1054 			return (previous);
1055 
1056 		if (v->v_id == vdev->v_id)
1057 			return (v);
1058 
1059 		if (v->v_id < vdev->v_id)
1060 			previous = v;
1061 	}
1062 	return (previous);
1063 }
1064 
1065 static size_t
1066 vdev_child_count(vdev_t *vdev)
1067 {
1068 	vdev_t *v;
1069 	size_t count;
1070 
1071 	count = 0;
1072 	STAILQ_FOREACH(v, &vdev->v_children, v_childlink) {
1073 		count++;
1074 	}
1075 	return (count);
1076 }
1077 
1078 /*
1079  * Insert vdev into top_vdev children list. List is ordered by v_id.
1080  */
1081 static void
1082 vdev_insert(vdev_t *top_vdev, vdev_t *vdev)
1083 {
1084 	vdev_t *previous;
1085 	size_t count;
1086 
1087 	/*
1088 	 * The top level vdev can appear in random order, depending how
1089 	 * the firmware is presenting the disk devices.
1090 	 * However, we will insert vdev to create list ordered by v_id,
1091 	 * so we can use either STAILQ_INSERT_HEAD or STAILQ_INSERT_AFTER
1092 	 * as STAILQ does not have insert before.
1093 	 */
1094 	previous = vdev_find_previous(top_vdev, vdev);
1095 
1096 	if (previous == NULL) {
1097 		STAILQ_INSERT_HEAD(&top_vdev->v_children, vdev, v_childlink);
1098 	} else if (previous->v_id == vdev->v_id) {
1099 		/*
1100 		 * This vdev was configured from label config,
1101 		 * do not insert duplicate.
1102 		 */
1103 		return;
1104 	} else {
1105 		STAILQ_INSERT_AFTER(&top_vdev->v_children, previous, vdev,
1106 		    v_childlink);
1107 	}
1108 
1109 	count = vdev_child_count(top_vdev);
1110 	if (top_vdev->v_nchildren < count)
1111 		top_vdev->v_nchildren = count;
1112 }
1113 
1114 static int
1115 vdev_from_nvlist(spa_t *spa, uint64_t top_guid, const nvlist_t *nvlist)
1116 {
1117 	vdev_t *top_vdev, *vdev;
1118 	nvlist_t **kids = NULL;
1119 	int rc, nkids;
1120 
1121 	/* Get top vdev. */
1122 	top_vdev = vdev_find(top_guid);
1123 	if (top_vdev == NULL) {
1124 		rc = vdev_init(top_guid, nvlist, &top_vdev);
1125 		if (rc != 0)
1126 			return (rc);
1127 		top_vdev->v_spa = spa;
1128 		top_vdev->v_top = top_vdev;
1129 		vdev_insert(spa->spa_root_vdev, top_vdev);
1130 	}
1131 
1132 	/* Add children if there are any. */
1133 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1134 	    &nkids, &kids, NULL);
1135 	if (rc == 0) {
1136 		for (int i = 0; i < nkids; i++) {
1137 			uint64_t guid;
1138 
1139 			rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID,
1140 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1141 			if (rc != 0)
1142 				goto done;
1143 
1144 			rc = vdev_init(guid, kids[i], &vdev);
1145 			if (rc != 0)
1146 				goto done;
1147 
1148 			vdev->v_spa = spa;
1149 			vdev->v_top = top_vdev;
1150 			vdev_insert(top_vdev, vdev);
1151 		}
1152 	} else {
1153 		/*
1154 		 * When there are no children, nvlist_find() does return
1155 		 * error, reset it because leaf devices have no children.
1156 		 */
1157 		rc = 0;
1158 	}
1159 done:
1160 	if (kids != NULL) {
1161 		for (int i = 0; i < nkids; i++)
1162 			nvlist_destroy(kids[i]);
1163 		free(kids);
1164 	}
1165 
1166 	return (rc);
1167 }
1168 
1169 static int
1170 vdev_init_from_label(spa_t *spa, const nvlist_t *nvlist)
1171 {
1172 	uint64_t pool_guid, top_guid;
1173 	nvlist_t *vdevs;
1174 	int rc;
1175 
1176 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1177 	    NULL, &pool_guid, NULL) ||
1178 	    nvlist_find(nvlist, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64,
1179 	    NULL, &top_guid, NULL) ||
1180 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1181 	    NULL, &vdevs, NULL)) {
1182 		printf("ZFS: can't find vdev details\n");
1183 		return (ENOENT);
1184 	}
1185 
1186 	rc = vdev_from_nvlist(spa, top_guid, vdevs);
1187 	nvlist_destroy(vdevs);
1188 	return (rc);
1189 }
1190 
1191 static void
1192 vdev_set_state(vdev_t *vdev)
1193 {
1194 	vdev_t *kid;
1195 	int good_kids;
1196 	int bad_kids;
1197 
1198 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1199 		vdev_set_state(kid);
1200 	}
1201 
1202 	/*
1203 	 * A mirror or raidz is healthy if all its kids are healthy. A
1204 	 * mirror is degraded if any of its kids is healthy; a raidz
1205 	 * is degraded if at most nparity kids are offline.
1206 	 */
1207 	if (STAILQ_FIRST(&vdev->v_children)) {
1208 		good_kids = 0;
1209 		bad_kids = 0;
1210 		STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1211 			if (kid->v_state == VDEV_STATE_HEALTHY)
1212 				good_kids++;
1213 			else
1214 				bad_kids++;
1215 		}
1216 		if (bad_kids == 0) {
1217 			vdev->v_state = VDEV_STATE_HEALTHY;
1218 		} else {
1219 			if (vdev->v_read == vdev_mirror_read) {
1220 				if (good_kids) {
1221 					vdev->v_state = VDEV_STATE_DEGRADED;
1222 				} else {
1223 					vdev->v_state = VDEV_STATE_OFFLINE;
1224 				}
1225 			} else if (vdev->v_read == vdev_raidz_read) {
1226 				if (bad_kids > vdev->v_nparity) {
1227 					vdev->v_state = VDEV_STATE_OFFLINE;
1228 				} else {
1229 					vdev->v_state = VDEV_STATE_DEGRADED;
1230 				}
1231 			}
1232 		}
1233 	}
1234 }
1235 
1236 static int
1237 vdev_update_from_nvlist(uint64_t top_guid, const nvlist_t *nvlist)
1238 {
1239 	vdev_t *vdev;
1240 	nvlist_t **kids = NULL;
1241 	int rc, nkids;
1242 
1243 	/* Update top vdev. */
1244 	vdev = vdev_find(top_guid);
1245 	if (vdev != NULL)
1246 		vdev_set_initial_state(vdev, nvlist);
1247 
1248 	/* Update children if there are any. */
1249 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1250 	    &nkids, &kids, NULL);
1251 	if (rc == 0) {
1252 		for (int i = 0; i < nkids; i++) {
1253 			uint64_t guid;
1254 
1255 			rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID,
1256 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1257 			if (rc != 0)
1258 				break;
1259 
1260 			vdev = vdev_find(guid);
1261 			if (vdev != NULL)
1262 				vdev_set_initial_state(vdev, kids[i]);
1263 		}
1264 	} else {
1265 		rc = 0;
1266 	}
1267 	if (kids != NULL) {
1268 		for (int i = 0; i < nkids; i++)
1269 			nvlist_destroy(kids[i]);
1270 		free(kids);
1271 	}
1272 
1273 	return (rc);
1274 }
1275 
1276 static int
1277 vdev_init_from_nvlist(spa_t *spa, const nvlist_t *nvlist)
1278 {
1279 	uint64_t pool_guid, vdev_children;
1280 	nvlist_t *vdevs = NULL, **kids = NULL;
1281 	int rc, nkids;
1282 
1283 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1284 	    NULL, &pool_guid, NULL) ||
1285 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_CHILDREN, DATA_TYPE_UINT64,
1286 	    NULL, &vdev_children, NULL) ||
1287 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1288 	    NULL, &vdevs, NULL)) {
1289 		printf("ZFS: can't find vdev details\n");
1290 		return (ENOENT);
1291 	}
1292 
1293 	/* Wrong guid?! */
1294 	if (spa->spa_guid != pool_guid) {
1295 		nvlist_destroy(vdevs);
1296 		return (EINVAL);
1297 	}
1298 
1299 	spa->spa_root_vdev->v_nchildren = vdev_children;
1300 
1301 	rc = nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1302 	    &nkids, &kids, NULL);
1303 	nvlist_destroy(vdevs);
1304 
1305 	/*
1306 	 * MOS config has at least one child for root vdev.
1307 	 */
1308 	if (rc != 0)
1309 		return (rc);
1310 
1311 	for (int i = 0; i < nkids; i++) {
1312 		uint64_t guid;
1313 		vdev_t *vdev;
1314 
1315 		rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1316 		    NULL, &guid, NULL);
1317 		if (rc != 0)
1318 			break;
1319 		vdev = vdev_find(guid);
1320 		/*
1321 		 * Top level vdev is missing, create it.
1322 		 */
1323 		if (vdev == NULL)
1324 			rc = vdev_from_nvlist(spa, guid, kids[i]);
1325 		else
1326 			rc = vdev_update_from_nvlist(guid, kids[i]);
1327 		if (rc != 0)
1328 			break;
1329 	}
1330 	if (kids != NULL) {
1331 		for (int i = 0; i < nkids; i++)
1332 			nvlist_destroy(kids[i]);
1333 		free(kids);
1334 	}
1335 
1336 	/*
1337 	 * Re-evaluate top-level vdev state.
1338 	 */
1339 	vdev_set_state(spa->spa_root_vdev);
1340 
1341 	return (rc);
1342 }
1343 
1344 static spa_t *
1345 spa_find_by_guid(uint64_t guid)
1346 {
1347 	spa_t *spa;
1348 
1349 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1350 		if (spa->spa_guid == guid)
1351 			return (spa);
1352 
1353 	return (NULL);
1354 }
1355 
1356 static spa_t *
1357 spa_find_by_name(const char *name)
1358 {
1359 	spa_t *spa;
1360 
1361 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1362 		if (strcmp(spa->spa_name, name) == 0)
1363 			return (spa);
1364 
1365 	return (NULL);
1366 }
1367 
1368 static spa_t *
1369 spa_find_by_dev(struct zfs_devdesc *dev)
1370 {
1371 
1372 	if (dev->dd.d_dev->dv_type != DEVT_ZFS)
1373 		return (NULL);
1374 
1375 	if (dev->pool_guid == 0)
1376 		return (STAILQ_FIRST(&zfs_pools));
1377 
1378 	return (spa_find_by_guid(dev->pool_guid));
1379 }
1380 
1381 static spa_t *
1382 spa_create(uint64_t guid, const char *name)
1383 {
1384 	spa_t *spa;
1385 
1386 	if ((spa = calloc(1, sizeof(spa_t))) == NULL)
1387 		return (NULL);
1388 	if ((spa->spa_name = strdup(name)) == NULL) {
1389 		free(spa);
1390 		return (NULL);
1391 	}
1392 	spa->spa_uberblock = &spa->spa_uberblock_master;
1393 	spa->spa_mos = &spa->spa_mos_master;
1394 	spa->spa_guid = guid;
1395 	spa->spa_root_vdev = vdev_create(guid, NULL);
1396 	if (spa->spa_root_vdev == NULL) {
1397 		free(spa->spa_name);
1398 		free(spa);
1399 		return (NULL);
1400 	}
1401 	spa->spa_root_vdev->v_name = strdup("root");
1402 	STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
1403 
1404 	return (spa);
1405 }
1406 
1407 static const char *
1408 state_name(vdev_state_t state)
1409 {
1410 	static const char *names[] = {
1411 		"UNKNOWN",
1412 		"CLOSED",
1413 		"OFFLINE",
1414 		"REMOVED",
1415 		"CANT_OPEN",
1416 		"FAULTED",
1417 		"DEGRADED",
1418 		"ONLINE"
1419 	};
1420 	return (names[state]);
1421 }
1422 
1423 #ifdef BOOT2
1424 
1425 #define pager_printf printf
1426 
1427 #else
1428 
1429 static int
1430 pager_printf(const char *fmt, ...)
1431 {
1432 	char line[80];
1433 	va_list args;
1434 
1435 	va_start(args, fmt);
1436 	vsnprintf(line, sizeof(line), fmt, args);
1437 	va_end(args);
1438 	return (pager_output(line));
1439 }
1440 
1441 #endif
1442 
1443 #define	STATUS_FORMAT	"        %s %s\n"
1444 
1445 static int
1446 print_state(int indent, const char *name, vdev_state_t state)
1447 {
1448 	int i;
1449 	char buf[512];
1450 
1451 	buf[0] = 0;
1452 	for (i = 0; i < indent; i++)
1453 		strcat(buf, "  ");
1454 	strcat(buf, name);
1455 	return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
1456 }
1457 
1458 static int
1459 vdev_status(vdev_t *vdev, int indent)
1460 {
1461 	vdev_t *kid;
1462 	int ret;
1463 
1464 	if (vdev->v_islog) {
1465 		(void) pager_output("        logs\n");
1466 		indent++;
1467 	}
1468 
1469 	ret = print_state(indent, vdev->v_name, vdev->v_state);
1470 	if (ret != 0)
1471 		return (ret);
1472 
1473 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1474 		ret = vdev_status(kid, indent + 1);
1475 		if (ret != 0)
1476 			return (ret);
1477 	}
1478 	return (ret);
1479 }
1480 
1481 static int
1482 spa_status(spa_t *spa)
1483 {
1484 	static char bootfs[ZFS_MAXNAMELEN];
1485 	uint64_t rootid;
1486 	vdev_list_t *vlist;
1487 	vdev_t *vdev;
1488 	int good_kids, bad_kids, degraded_kids, ret;
1489 	vdev_state_t state;
1490 
1491 	ret = pager_printf("  pool: %s\n", spa->spa_name);
1492 	if (ret != 0)
1493 		return (ret);
1494 
1495 	if (zfs_get_root(spa, &rootid) == 0 &&
1496 	    zfs_rlookup(spa, rootid, bootfs) == 0) {
1497 		if (bootfs[0] == '\0')
1498 			ret = pager_printf("bootfs: %s\n", spa->spa_name);
1499 		else
1500 			ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
1501 			    bootfs);
1502 		if (ret != 0)
1503 			return (ret);
1504 	}
1505 	ret = pager_printf("config:\n\n");
1506 	if (ret != 0)
1507 		return (ret);
1508 	ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
1509 	if (ret != 0)
1510 		return (ret);
1511 
1512 	good_kids = 0;
1513 	degraded_kids = 0;
1514 	bad_kids = 0;
1515 	vlist = &spa->spa_root_vdev->v_children;
1516 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1517 		if (vdev->v_state == VDEV_STATE_HEALTHY)
1518 			good_kids++;
1519 		else if (vdev->v_state == VDEV_STATE_DEGRADED)
1520 			degraded_kids++;
1521 		else
1522 			bad_kids++;
1523 	}
1524 
1525 	state = VDEV_STATE_CLOSED;
1526 	if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
1527 		state = VDEV_STATE_HEALTHY;
1528 	else if ((good_kids + degraded_kids) > 0)
1529 		state = VDEV_STATE_DEGRADED;
1530 
1531 	ret = print_state(0, spa->spa_name, state);
1532 	if (ret != 0)
1533 		return (ret);
1534 
1535 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1536 		ret = vdev_status(vdev, 1);
1537 		if (ret != 0)
1538 			return (ret);
1539 	}
1540 	return (ret);
1541 }
1542 
1543 static int
1544 spa_all_status(void)
1545 {
1546 	spa_t *spa;
1547 	int first = 1, ret = 0;
1548 
1549 	STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
1550 		if (!first) {
1551 			ret = pager_printf("\n");
1552 			if (ret != 0)
1553 				return (ret);
1554 		}
1555 		first = 0;
1556 		ret = spa_status(spa);
1557 		if (ret != 0)
1558 			return (ret);
1559 	}
1560 	return (ret);
1561 }
1562 
1563 static uint64_t
1564 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
1565 {
1566 	uint64_t label_offset;
1567 
1568 	if (l < VDEV_LABELS / 2)
1569 		label_offset = 0;
1570 	else
1571 		label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
1572 
1573 	return (offset + l * sizeof (vdev_label_t) + label_offset);
1574 }
1575 
1576 static int
1577 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1578 {
1579 	unsigned int seq1 = 0;
1580 	unsigned int seq2 = 0;
1581 	int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
1582 
1583 	if (cmp != 0)
1584 		return (cmp);
1585 
1586 	cmp = AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1587 	if (cmp != 0)
1588 		return (cmp);
1589 
1590 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1591 		seq1 = MMP_SEQ(ub1);
1592 
1593 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1594 		seq2 = MMP_SEQ(ub2);
1595 
1596 	return (AVL_CMP(seq1, seq2));
1597 }
1598 
1599 static int
1600 uberblock_verify(uberblock_t *ub)
1601 {
1602 	if (ub->ub_magic == BSWAP_64((uint64_t)UBERBLOCK_MAGIC)) {
1603 		byteswap_uint64_array(ub, sizeof (uberblock_t));
1604 	}
1605 
1606 	if (ub->ub_magic != UBERBLOCK_MAGIC ||
1607 	    !SPA_VERSION_IS_SUPPORTED(ub->ub_version))
1608 		return (EINVAL);
1609 
1610 	return (0);
1611 }
1612 
1613 static int
1614 vdev_label_read(vdev_t *vd, int l, void *buf, uint64_t offset,
1615     size_t size)
1616 {
1617 	blkptr_t bp;
1618 	off_t off;
1619 
1620 	off = vdev_label_offset(vd->v_psize, l, offset);
1621 
1622 	BP_ZERO(&bp);
1623 	BP_SET_LSIZE(&bp, size);
1624 	BP_SET_PSIZE(&bp, size);
1625 	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1626 	BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1627 	DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
1628 	ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1629 
1630 	return (vdev_read_phys(vd, &bp, buf, off, size));
1631 }
1632 
1633 /*
1634  * We do need to be sure we write to correct location.
1635  * Our vdev label does consist of 4 fields:
1636  * pad1 (8k), reserved.
1637  * bootenv (8k), checksummed, previously reserved, may contian garbage.
1638  * vdev_phys (112k), checksummed
1639  * uberblock ring (128k), checksummed.
1640  *
1641  * Since bootenv area may contain garbage, we can not reliably read it, as
1642  * we can get checksum errors.
1643  * Next best thing is vdev_phys - it is just after bootenv. It still may
1644  * be corrupted, but in such case we will miss this one write.
1645  */
1646 static int
1647 vdev_label_write_validate(vdev_t *vd, int l, uint64_t offset)
1648 {
1649 	uint64_t off, o_phys;
1650 	void *buf;
1651 	size_t size = VDEV_PHYS_SIZE;
1652 	int rc;
1653 
1654 	o_phys = offsetof(vdev_label_t, vl_vdev_phys);
1655 	off = vdev_label_offset(vd->v_psize, l, o_phys);
1656 
1657 	/* off should be 8K from bootenv */
1658 	if (vdev_label_offset(vd->v_psize, l, offset) + VDEV_PAD_SIZE != off)
1659 		return (EINVAL);
1660 
1661 	buf = malloc(size);
1662 	if (buf == NULL)
1663 		return (ENOMEM);
1664 
1665 	/* Read vdev_phys */
1666 	rc = vdev_label_read(vd, l, buf, o_phys, size);
1667 	free(buf);
1668 	return (rc);
1669 }
1670 
1671 static int
1672 vdev_label_write(vdev_t *vd, int l, vdev_boot_envblock_t *be, uint64_t offset)
1673 {
1674 	zio_checksum_info_t *ci;
1675 	zio_cksum_t cksum;
1676 	off_t off;
1677 	size_t size = VDEV_PAD_SIZE;
1678 	int rc;
1679 
1680 	if (vd->v_phys_write == NULL)
1681 		return (ENOTSUP);
1682 
1683 	off = vdev_label_offset(vd->v_psize, l, offset);
1684 
1685 	rc = vdev_label_write_validate(vd, l, offset);
1686 	if (rc != 0) {
1687 		return (rc);
1688 	}
1689 
1690 	ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
1691 	be->vbe_zbt.zec_magic = ZEC_MAGIC;
1692 	zio_checksum_label_verifier(&be->vbe_zbt.zec_cksum, off);
1693 	ci->ci_func[0](be, size, NULL, &cksum);
1694 	be->vbe_zbt.zec_cksum = cksum;
1695 
1696 	return (vdev_write_phys(vd, be, off, size));
1697 }
1698 
1699 static int
1700 vdev_write_bootenv_impl(vdev_t *vdev, vdev_boot_envblock_t *be)
1701 {
1702 	vdev_t *kid;
1703 	int rv = 0, rc;
1704 
1705 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1706 		if (kid->v_state != VDEV_STATE_HEALTHY)
1707 			continue;
1708 		rc = vdev_write_bootenv_impl(kid, be);
1709 		if (rv == 0)
1710 			rv = rc;
1711 	}
1712 
1713 	/*
1714 	 * Non-leaf vdevs do not have v_phys_write.
1715 	 */
1716 	if (vdev->v_phys_write == NULL)
1717 		return (rv);
1718 
1719 	for (int l = 0; l < VDEV_LABELS; l++) {
1720 		rc = vdev_label_write(vdev, l, be,
1721 		    offsetof(vdev_label_t, vl_be));
1722 		if (rc != 0) {
1723 			printf("failed to write bootenv to %s label %d: %d\n",
1724 			    vdev->v_name ? vdev->v_name : "unknown", l, rc);
1725 			rv = rc;
1726 		}
1727 	}
1728 	return (rv);
1729 }
1730 
1731 int
1732 vdev_write_bootenv(vdev_t *vdev, nvlist_t *nvl)
1733 {
1734 	vdev_boot_envblock_t *be;
1735 	nvlist_t nv, *nvp;
1736 	uint64_t version;
1737 	int rv;
1738 
1739 	if (nvl->nv_size > sizeof(be->vbe_bootenv))
1740 		return (E2BIG);
1741 
1742 	version = VB_RAW;
1743 	nvp = vdev_read_bootenv(vdev);
1744 	if (nvp != NULL) {
1745 		nvlist_find(nvp, BOOTENV_VERSION, DATA_TYPE_UINT64, NULL,
1746 		    &version, NULL);
1747 		nvlist_destroy(nvp);
1748 	}
1749 
1750 	be = calloc(1, sizeof(*be));
1751 	if (be == NULL)
1752 		return (ENOMEM);
1753 
1754 	be->vbe_version = version;
1755 	switch (version) {
1756 	case VB_RAW:
1757 		/*
1758 		 * If there is no envmap, we will just wipe bootenv.
1759 		 */
1760 		nvlist_find(nvl, GRUB_ENVMAP, DATA_TYPE_STRING, NULL,
1761 		    be->vbe_bootenv, NULL);
1762 		rv = 0;
1763 		break;
1764 
1765 	case VB_NVLIST:
1766 		nv.nv_header = nvl->nv_header;
1767 		nv.nv_asize = nvl->nv_asize;
1768 		nv.nv_size = nvl->nv_size;
1769 
1770 		bcopy(&nv.nv_header, be->vbe_bootenv, sizeof(nv.nv_header));
1771 		nv.nv_data = be->vbe_bootenv + sizeof(nvs_header_t);
1772 		bcopy(nvl->nv_data, nv.nv_data, nv.nv_size);
1773 		rv = nvlist_export(&nv);
1774 		break;
1775 
1776 	default:
1777 		rv = EINVAL;
1778 		break;
1779 	}
1780 
1781 	if (rv == 0) {
1782 		be->vbe_version = htobe64(be->vbe_version);
1783 		rv = vdev_write_bootenv_impl(vdev, be);
1784 	}
1785 	free(be);
1786 	return (rv);
1787 }
1788 
1789 /*
1790  * Read the bootenv area from pool label, return the nvlist from it.
1791  * We return from first successful read.
1792  */
1793 nvlist_t *
1794 vdev_read_bootenv(vdev_t *vdev)
1795 {
1796 	vdev_t *kid;
1797 	nvlist_t *benv;
1798 	vdev_boot_envblock_t *be;
1799 	char *command;
1800 	bool ok;
1801 	int rv;
1802 
1803 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1804 		if (kid->v_state != VDEV_STATE_HEALTHY)
1805 			continue;
1806 
1807 		benv = vdev_read_bootenv(kid);
1808 		if (benv != NULL)
1809 			return (benv);
1810 	}
1811 
1812 	be = malloc(sizeof (*be));
1813 	if (be == NULL)
1814 		return (NULL);
1815 
1816 	rv = 0;
1817 	for (int l = 0; l < VDEV_LABELS; l++) {
1818 		rv = vdev_label_read(vdev, l, be,
1819 		    offsetof(vdev_label_t, vl_be),
1820 		    sizeof (*be));
1821 		if (rv == 0)
1822 			break;
1823 	}
1824 	if (rv != 0) {
1825 		free(be);
1826 		return (NULL);
1827 	}
1828 
1829 	be->vbe_version = be64toh(be->vbe_version);
1830 	switch (be->vbe_version) {
1831 	case VB_RAW:
1832 		/*
1833 		 * we have textual data in vbe_bootenv, create nvlist
1834 		 * with key "envmap".
1835 		 */
1836 		benv = nvlist_create(NV_UNIQUE_NAME);
1837 		if (benv != NULL) {
1838 			if (*be->vbe_bootenv == '\0') {
1839 				nvlist_add_uint64(benv, BOOTENV_VERSION,
1840 				    VB_NVLIST);
1841 				break;
1842 			}
1843 			nvlist_add_uint64(benv, BOOTENV_VERSION, VB_RAW);
1844 			be->vbe_bootenv[sizeof (be->vbe_bootenv) - 1] = '\0';
1845 			nvlist_add_string(benv, GRUB_ENVMAP, be->vbe_bootenv);
1846 		}
1847 		break;
1848 
1849 	case VB_NVLIST:
1850 		benv = nvlist_import(be->vbe_bootenv, sizeof(be->vbe_bootenv));
1851 		break;
1852 
1853 	default:
1854 		command = (char *)be;
1855 		ok = false;
1856 
1857 		/* Check for legacy zfsbootcfg command string */
1858 		for (int i = 0; command[i] != '\0'; i++) {
1859 			if (iscntrl(command[i])) {
1860 				ok = false;
1861 				break;
1862 			} else {
1863 				ok = true;
1864 			}
1865 		}
1866 		benv = nvlist_create(NV_UNIQUE_NAME);
1867 		if (benv != NULL) {
1868 			if (ok)
1869 				nvlist_add_string(benv, FREEBSD_BOOTONCE,
1870 				    command);
1871 			else
1872 				nvlist_add_uint64(benv, BOOTENV_VERSION,
1873 				    VB_NVLIST);
1874 		}
1875 		break;
1876 	}
1877 	free(be);
1878 	return (benv);
1879 }
1880 
1881 static uint64_t
1882 vdev_get_label_asize(nvlist_t *nvl)
1883 {
1884 	nvlist_t *vdevs;
1885 	uint64_t asize;
1886 	const char *type;
1887 	int len;
1888 
1889 	asize = 0;
1890 	/* Get vdev tree */
1891 	if (nvlist_find(nvl, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1892 	    NULL, &vdevs, NULL) != 0)
1893 		return (asize);
1894 
1895 	/*
1896 	 * Get vdev type. We will calculate asize for raidz, mirror and disk.
1897 	 * For raidz, the asize is raw size of all children.
1898 	 */
1899 	if (nvlist_find(vdevs, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
1900 	    NULL, &type, &len) != 0)
1901 		goto done;
1902 
1903 	if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 &&
1904 	    memcmp(type, VDEV_TYPE_DISK, len) != 0 &&
1905 	    memcmp(type, VDEV_TYPE_RAIDZ, len) != 0)
1906 		goto done;
1907 
1908 	if (nvlist_find(vdevs, ZPOOL_CONFIG_ASIZE, DATA_TYPE_UINT64,
1909 	    NULL, &asize, NULL) != 0)
1910 		goto done;
1911 
1912 	if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) {
1913 		nvlist_t **kids;
1914 		int nkids;
1915 
1916 		if (nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN,
1917 		    DATA_TYPE_NVLIST_ARRAY, &nkids, &kids, NULL) != 0) {
1918 			asize = 0;
1919 			goto done;
1920 		}
1921 
1922 		asize /= nkids;
1923 		for (int i = 0; i < nkids; i++)
1924 			nvlist_destroy(kids[i]);
1925 		free(kids);
1926 	}
1927 
1928 	asize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
1929 done:
1930 	nvlist_destroy(vdevs);
1931 	return (asize);
1932 }
1933 
1934 static nvlist_t *
1935 vdev_label_read_config(vdev_t *vd, uint64_t txg)
1936 {
1937 	vdev_phys_t *label;
1938 	uint64_t best_txg = 0;
1939 	uint64_t label_txg = 0;
1940 	uint64_t asize;
1941 	nvlist_t *nvl = NULL, *tmp;
1942 	int error;
1943 
1944 	label = malloc(sizeof (vdev_phys_t));
1945 	if (label == NULL)
1946 		return (NULL);
1947 
1948 	for (int l = 0; l < VDEV_LABELS; l++) {
1949 		if (vdev_label_read(vd, l, label,
1950 		    offsetof(vdev_label_t, vl_vdev_phys),
1951 		    sizeof (vdev_phys_t)))
1952 			continue;
1953 
1954 		tmp = nvlist_import(label->vp_nvlist,
1955 		    sizeof(label->vp_nvlist));
1956 		if (tmp == NULL)
1957 			continue;
1958 
1959 		error = nvlist_find(tmp, ZPOOL_CONFIG_POOL_TXG,
1960 		    DATA_TYPE_UINT64, NULL, &label_txg, NULL);
1961 		if (error != 0 || label_txg == 0) {
1962 			nvlist_destroy(nvl);
1963 			nvl = tmp;
1964 			goto done;
1965 		}
1966 
1967 		if (label_txg <= txg && label_txg > best_txg) {
1968 			best_txg = label_txg;
1969 			nvlist_destroy(nvl);
1970 			nvl = tmp;
1971 			tmp = NULL;
1972 
1973 			/*
1974 			 * Use asize from pool config. We need this
1975 			 * because we can get bad value from BIOS.
1976 			 */
1977 			asize = vdev_get_label_asize(nvl);
1978 			if (asize != 0) {
1979 				vd->v_psize = asize;
1980 			}
1981 		}
1982 		nvlist_destroy(tmp);
1983 	}
1984 
1985 	if (best_txg == 0) {
1986 		nvlist_destroy(nvl);
1987 		nvl = NULL;
1988 	}
1989 done:
1990 	free(label);
1991 	return (nvl);
1992 }
1993 
1994 static void
1995 vdev_uberblock_load(vdev_t *vd, uberblock_t *ub)
1996 {
1997 	uberblock_t *buf;
1998 
1999 	buf = malloc(VDEV_UBERBLOCK_SIZE(vd));
2000 	if (buf == NULL)
2001 		return;
2002 
2003 	for (int l = 0; l < VDEV_LABELS; l++) {
2004 		for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
2005 			if (vdev_label_read(vd, l, buf,
2006 			    VDEV_UBERBLOCK_OFFSET(vd, n),
2007 			    VDEV_UBERBLOCK_SIZE(vd)))
2008 				continue;
2009 			if (uberblock_verify(buf) != 0)
2010 				continue;
2011 
2012 			if (vdev_uberblock_compare(buf, ub) > 0)
2013 				*ub = *buf;
2014 		}
2015 	}
2016 	free(buf);
2017 }
2018 
2019 static int
2020 vdev_probe(vdev_phys_read_t *_read, vdev_phys_write_t *_write, void *priv,
2021     spa_t **spap)
2022 {
2023 	vdev_t vtmp;
2024 	spa_t *spa;
2025 	vdev_t *vdev;
2026 	nvlist_t *nvl;
2027 	uint64_t val;
2028 	uint64_t guid, vdev_children;
2029 	uint64_t pool_txg, pool_guid;
2030 	const char *pool_name;
2031 	int rc, namelen;
2032 
2033 	/*
2034 	 * Load the vdev label and figure out which
2035 	 * uberblock is most current.
2036 	 */
2037 	memset(&vtmp, 0, sizeof(vtmp));
2038 	vtmp.v_phys_read = _read;
2039 	vtmp.v_phys_write = _write;
2040 	vtmp.v_priv = priv;
2041 	vtmp.v_psize = P2ALIGN(ldi_get_size(priv),
2042 	    (uint64_t)sizeof (vdev_label_t));
2043 
2044 	/* Test for minimum device size. */
2045 	if (vtmp.v_psize < SPA_MINDEVSIZE)
2046 		return (EIO);
2047 
2048 	nvl = vdev_label_read_config(&vtmp, UINT64_MAX);
2049 	if (nvl == NULL)
2050 		return (EIO);
2051 
2052 	if (nvlist_find(nvl, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
2053 	    NULL, &val, NULL) != 0) {
2054 		nvlist_destroy(nvl);
2055 		return (EIO);
2056 	}
2057 
2058 	if (!SPA_VERSION_IS_SUPPORTED(val)) {
2059 		printf("ZFS: unsupported ZFS version %u (should be %u)\n",
2060 		    (unsigned)val, (unsigned)SPA_VERSION);
2061 		nvlist_destroy(nvl);
2062 		return (EIO);
2063 	}
2064 
2065 	/* Check ZFS features for read */
2066 	rc = nvlist_check_features_for_read(nvl);
2067 	if (rc != 0) {
2068 		nvlist_destroy(nvl);
2069 		return (EIO);
2070 	}
2071 
2072 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
2073 	    NULL, &val, NULL) != 0) {
2074 		nvlist_destroy(nvl);
2075 		return (EIO);
2076 	}
2077 
2078 	if (val == POOL_STATE_DESTROYED) {
2079 		/* We don't boot only from destroyed pools. */
2080 		nvlist_destroy(nvl);
2081 		return (EIO);
2082 	}
2083 
2084 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
2085 	    NULL, &pool_txg, NULL) != 0 ||
2086 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
2087 	    NULL, &pool_guid, NULL) != 0 ||
2088 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
2089 	    NULL, &pool_name, &namelen) != 0) {
2090 		/*
2091 		 * Cache and spare devices end up here - just ignore
2092 		 * them.
2093 		 */
2094 		nvlist_destroy(nvl);
2095 		return (EIO);
2096 	}
2097 
2098 	/*
2099 	 * Create the pool if this is the first time we've seen it.
2100 	 */
2101 	spa = spa_find_by_guid(pool_guid);
2102 	if (spa == NULL) {
2103 		char *name;
2104 
2105 		nvlist_find(nvl, ZPOOL_CONFIG_VDEV_CHILDREN,
2106 		    DATA_TYPE_UINT64, NULL, &vdev_children, NULL);
2107 		name = malloc(namelen + 1);
2108 		if (name == NULL) {
2109 			nvlist_destroy(nvl);
2110 			return (ENOMEM);
2111 		}
2112 		bcopy(pool_name, name, namelen);
2113 		name[namelen] = '\0';
2114 		spa = spa_create(pool_guid, name);
2115 		free(name);
2116 		if (spa == NULL) {
2117 			nvlist_destroy(nvl);
2118 			return (ENOMEM);
2119 		}
2120 		spa->spa_root_vdev->v_nchildren = vdev_children;
2121 	}
2122 	if (pool_txg > spa->spa_txg)
2123 		spa->spa_txg = pool_txg;
2124 
2125 	/*
2126 	 * Get the vdev tree and create our in-core copy of it.
2127 	 * If we already have a vdev with this guid, this must
2128 	 * be some kind of alias (overlapping slices, dangerously dedicated
2129 	 * disks etc).
2130 	 */
2131 	if (nvlist_find(nvl, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
2132 	    NULL, &guid, NULL) != 0) {
2133 		nvlist_destroy(nvl);
2134 		return (EIO);
2135 	}
2136 	vdev = vdev_find(guid);
2137 	/* Has this vdev already been inited? */
2138 	if (vdev && vdev->v_phys_read) {
2139 		nvlist_destroy(nvl);
2140 		return (EIO);
2141 	}
2142 
2143 	rc = vdev_init_from_label(spa, nvl);
2144 	nvlist_destroy(nvl);
2145 	if (rc != 0)
2146 		return (rc);
2147 
2148 	/*
2149 	 * We should already have created an incomplete vdev for this
2150 	 * vdev. Find it and initialise it with our read proc.
2151 	 */
2152 	vdev = vdev_find(guid);
2153 	if (vdev != NULL) {
2154 		vdev->v_phys_read = _read;
2155 		vdev->v_phys_write = _write;
2156 		vdev->v_priv = priv;
2157 		vdev->v_psize = vtmp.v_psize;
2158 		/*
2159 		 * If no other state is set, mark vdev healthy.
2160 		 */
2161 		if (vdev->v_state == VDEV_STATE_UNKNOWN)
2162 			vdev->v_state = VDEV_STATE_HEALTHY;
2163 	} else {
2164 		printf("ZFS: inconsistent nvlist contents\n");
2165 		return (EIO);
2166 	}
2167 
2168 	if (vdev->v_islog)
2169 		spa->spa_with_log = vdev->v_islog;
2170 
2171 	/*
2172 	 * Re-evaluate top-level vdev state.
2173 	 */
2174 	vdev_set_state(vdev->v_top);
2175 
2176 	/*
2177 	 * Ok, we are happy with the pool so far. Lets find
2178 	 * the best uberblock and then we can actually access
2179 	 * the contents of the pool.
2180 	 */
2181 	vdev_uberblock_load(vdev, spa->spa_uberblock);
2182 
2183 	if (spap != NULL)
2184 		*spap = spa;
2185 	return (0);
2186 }
2187 
2188 static int
2189 ilog2(int n)
2190 {
2191 	int v;
2192 
2193 	for (v = 0; v < 32; v++)
2194 		if (n == (1 << v))
2195 			return (v);
2196 	return (-1);
2197 }
2198 
2199 static int
2200 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
2201 {
2202 	blkptr_t gbh_bp;
2203 	zio_gbh_phys_t zio_gb;
2204 	char *pbuf;
2205 	int i;
2206 
2207 	/* Artificial BP for gang block header. */
2208 	gbh_bp = *bp;
2209 	BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
2210 	BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
2211 	BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
2212 	BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
2213 	for (i = 0; i < SPA_DVAS_PER_BP; i++)
2214 		DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
2215 
2216 	/* Read gang header block using the artificial BP. */
2217 	if (zio_read(spa, &gbh_bp, &zio_gb))
2218 		return (EIO);
2219 
2220 	pbuf = buf;
2221 	for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
2222 		blkptr_t *gbp = &zio_gb.zg_blkptr[i];
2223 
2224 		if (BP_IS_HOLE(gbp))
2225 			continue;
2226 		if (zio_read(spa, gbp, pbuf))
2227 			return (EIO);
2228 		pbuf += BP_GET_PSIZE(gbp);
2229 	}
2230 
2231 	if (zio_checksum_verify(spa, bp, buf))
2232 		return (EIO);
2233 	return (0);
2234 }
2235 
2236 static int
2237 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
2238 {
2239 	int cpfunc = BP_GET_COMPRESS(bp);
2240 	uint64_t align, size;
2241 	void *pbuf;
2242 	int i, error;
2243 
2244 	/*
2245 	 * Process data embedded in block pointer
2246 	 */
2247 	if (BP_IS_EMBEDDED(bp)) {
2248 		ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
2249 
2250 		size = BPE_GET_PSIZE(bp);
2251 		ASSERT(size <= BPE_PAYLOAD_SIZE);
2252 
2253 		if (cpfunc != ZIO_COMPRESS_OFF)
2254 			pbuf = malloc(size);
2255 		else
2256 			pbuf = buf;
2257 
2258 		if (pbuf == NULL)
2259 			return (ENOMEM);
2260 
2261 		decode_embedded_bp_compressed(bp, pbuf);
2262 		error = 0;
2263 
2264 		if (cpfunc != ZIO_COMPRESS_OFF) {
2265 			error = zio_decompress_data(cpfunc, pbuf,
2266 			    size, buf, BP_GET_LSIZE(bp));
2267 			free(pbuf);
2268 		}
2269 		if (error != 0)
2270 			printf("ZFS: i/o error - unable to decompress "
2271 			    "block pointer data, error %d\n", error);
2272 		return (error);
2273 	}
2274 
2275 	error = EIO;
2276 
2277 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
2278 		const dva_t *dva = &bp->blk_dva[i];
2279 		vdev_t *vdev;
2280 		vdev_list_t *vlist;
2281 		uint64_t vdevid;
2282 		off_t offset;
2283 
2284 		if (!dva->dva_word[0] && !dva->dva_word[1])
2285 			continue;
2286 
2287 		vdevid = DVA_GET_VDEV(dva);
2288 		offset = DVA_GET_OFFSET(dva);
2289 		vlist = &spa->spa_root_vdev->v_children;
2290 		STAILQ_FOREACH(vdev, vlist, v_childlink) {
2291 			if (vdev->v_id == vdevid)
2292 				break;
2293 		}
2294 		if (!vdev || !vdev->v_read)
2295 			continue;
2296 
2297 		size = BP_GET_PSIZE(bp);
2298 		if (vdev->v_read == vdev_raidz_read) {
2299 			align = 1ULL << vdev->v_ashift;
2300 			if (P2PHASE(size, align) != 0)
2301 				size = P2ROUNDUP(size, align);
2302 		}
2303 		if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
2304 			pbuf = malloc(size);
2305 		else
2306 			pbuf = buf;
2307 
2308 		if (pbuf == NULL) {
2309 			error = ENOMEM;
2310 			break;
2311 		}
2312 
2313 		if (DVA_GET_GANG(dva))
2314 			error = zio_read_gang(spa, bp, pbuf);
2315 		else
2316 			error = vdev->v_read(vdev, bp, pbuf, offset, size);
2317 		if (error == 0) {
2318 			if (cpfunc != ZIO_COMPRESS_OFF)
2319 				error = zio_decompress_data(cpfunc, pbuf,
2320 				    BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
2321 			else if (size != BP_GET_PSIZE(bp))
2322 				bcopy(pbuf, buf, BP_GET_PSIZE(bp));
2323 		} else {
2324 			printf("zio_read error: %d\n", error);
2325 		}
2326 		if (buf != pbuf)
2327 			free(pbuf);
2328 		if (error == 0)
2329 			break;
2330 	}
2331 	if (error != 0)
2332 		printf("ZFS: i/o error - all block copies unavailable\n");
2333 
2334 	return (error);
2335 }
2336 
2337 static int
2338 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset,
2339     void *buf, size_t buflen)
2340 {
2341 	int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
2342 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2343 	int nlevels = dnode->dn_nlevels;
2344 	int i, rc;
2345 
2346 	if (bsize > SPA_MAXBLOCKSIZE) {
2347 		printf("ZFS: I/O error - blocks larger than %llu are not "
2348 		    "supported\n", SPA_MAXBLOCKSIZE);
2349 		return (EIO);
2350 	}
2351 
2352 	/*
2353 	 * Handle odd block sizes, mirrors dmu_read_impl().  Data can't exist
2354 	 * past the first block, so we'll clip the read to the portion of the
2355 	 * buffer within bsize and zero out the remainder.
2356 	 */
2357 	if (dnode->dn_maxblkid == 0) {
2358 		size_t newbuflen;
2359 
2360 		newbuflen = offset > bsize ? 0 : MIN(buflen, bsize - offset);
2361 		bzero((char *)buf + newbuflen, buflen - newbuflen);
2362 		buflen = newbuflen;
2363 	}
2364 
2365 	/*
2366 	 * Note: bsize may not be a power of two here so we need to do an
2367 	 * actual divide rather than a bitshift.
2368 	 */
2369 	while (buflen > 0) {
2370 		uint64_t bn = offset / bsize;
2371 		int boff = offset % bsize;
2372 		int ibn;
2373 		const blkptr_t *indbp;
2374 		blkptr_t bp;
2375 
2376 		if (bn > dnode->dn_maxblkid)
2377 			return (EIO);
2378 
2379 		if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
2380 			goto cached;
2381 
2382 		indbp = dnode->dn_blkptr;
2383 		for (i = 0; i < nlevels; i++) {
2384 			/*
2385 			 * Copy the bp from the indirect array so that
2386 			 * we can re-use the scratch buffer for multi-level
2387 			 * objects.
2388 			 */
2389 			ibn = bn >> ((nlevels - i - 1) * ibshift);
2390 			ibn &= ((1 << ibshift) - 1);
2391 			bp = indbp[ibn];
2392 			if (BP_IS_HOLE(&bp)) {
2393 				memset(dnode_cache_buf, 0, bsize);
2394 				break;
2395 			}
2396 			rc = zio_read(spa, &bp, dnode_cache_buf);
2397 			if (rc)
2398 				return (rc);
2399 			indbp = (const blkptr_t *) dnode_cache_buf;
2400 		}
2401 		dnode_cache_obj = dnode;
2402 		dnode_cache_bn = bn;
2403 	cached:
2404 
2405 		/*
2406 		 * The buffer contains our data block. Copy what we
2407 		 * need from it and loop.
2408 		 */
2409 		i = bsize - boff;
2410 		if (i > buflen) i = buflen;
2411 		memcpy(buf, &dnode_cache_buf[boff], i);
2412 		buf = ((char *)buf) + i;
2413 		offset += i;
2414 		buflen -= i;
2415 	}
2416 
2417 	return (0);
2418 }
2419 
2420 /*
2421  * Lookup a value in a microzap directory.
2422  */
2423 static int
2424 mzap_lookup(const mzap_phys_t *mz, size_t size, const char *name,
2425     uint64_t *value)
2426 {
2427 	const mzap_ent_phys_t *mze;
2428 	int chunks, i;
2429 
2430 	/*
2431 	 * Microzap objects use exactly one block. Read the whole
2432 	 * thing.
2433 	 */
2434 	chunks = size / MZAP_ENT_LEN - 1;
2435 	for (i = 0; i < chunks; i++) {
2436 		mze = &mz->mz_chunk[i];
2437 		if (strcmp(mze->mze_name, name) == 0) {
2438 			*value = mze->mze_value;
2439 			return (0);
2440 		}
2441 	}
2442 
2443 	return (ENOENT);
2444 }
2445 
2446 /*
2447  * Compare a name with a zap leaf entry. Return non-zero if the name
2448  * matches.
2449  */
2450 static int
2451 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2452     const char *name)
2453 {
2454 	size_t namelen;
2455 	const zap_leaf_chunk_t *nc;
2456 	const char *p;
2457 
2458 	namelen = zc->l_entry.le_name_numints;
2459 
2460 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2461 	p = name;
2462 	while (namelen > 0) {
2463 		size_t len;
2464 
2465 		len = namelen;
2466 		if (len > ZAP_LEAF_ARRAY_BYTES)
2467 			len = ZAP_LEAF_ARRAY_BYTES;
2468 		if (memcmp(p, nc->l_array.la_array, len))
2469 			return (0);
2470 		p += len;
2471 		namelen -= len;
2472 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2473 	}
2474 
2475 	return (1);
2476 }
2477 
2478 /*
2479  * Extract a uint64_t value from a zap leaf entry.
2480  */
2481 static uint64_t
2482 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
2483 {
2484 	const zap_leaf_chunk_t *vc;
2485 	int i;
2486 	uint64_t value;
2487 	const uint8_t *p;
2488 
2489 	vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
2490 	for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
2491 		value = (value << 8) | p[i];
2492 	}
2493 
2494 	return (value);
2495 }
2496 
2497 static void
2498 stv(int len, void *addr, uint64_t value)
2499 {
2500 	switch (len) {
2501 	case 1:
2502 		*(uint8_t *)addr = value;
2503 		return;
2504 	case 2:
2505 		*(uint16_t *)addr = value;
2506 		return;
2507 	case 4:
2508 		*(uint32_t *)addr = value;
2509 		return;
2510 	case 8:
2511 		*(uint64_t *)addr = value;
2512 		return;
2513 	}
2514 }
2515 
2516 /*
2517  * Extract a array from a zap leaf entry.
2518  */
2519 static void
2520 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2521     uint64_t integer_size, uint64_t num_integers, void *buf)
2522 {
2523 	uint64_t array_int_len = zc->l_entry.le_value_intlen;
2524 	uint64_t value = 0;
2525 	uint64_t *u64 = buf;
2526 	char *p = buf;
2527 	int len = MIN(zc->l_entry.le_value_numints, num_integers);
2528 	int chunk = zc->l_entry.le_value_chunk;
2529 	int byten = 0;
2530 
2531 	if (integer_size == 8 && len == 1) {
2532 		*u64 = fzap_leaf_value(zl, zc);
2533 		return;
2534 	}
2535 
2536 	while (len > 0) {
2537 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
2538 		int i;
2539 
2540 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
2541 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
2542 			value = (value << 8) | la->la_array[i];
2543 			byten++;
2544 			if (byten == array_int_len) {
2545 				stv(integer_size, p, value);
2546 				byten = 0;
2547 				len--;
2548 				if (len == 0)
2549 					return;
2550 				p += integer_size;
2551 			}
2552 		}
2553 		chunk = la->la_next;
2554 	}
2555 }
2556 
2557 static int
2558 fzap_check_size(uint64_t integer_size, uint64_t num_integers)
2559 {
2560 
2561 	switch (integer_size) {
2562 	case 1:
2563 	case 2:
2564 	case 4:
2565 	case 8:
2566 		break;
2567 	default:
2568 		return (EINVAL);
2569 	}
2570 
2571 	if (integer_size * num_integers > ZAP_MAXVALUELEN)
2572 		return (E2BIG);
2573 
2574 	return (0);
2575 }
2576 
2577 static void
2578 zap_leaf_free(zap_leaf_t *leaf)
2579 {
2580 	free(leaf->l_phys);
2581 	free(leaf);
2582 }
2583 
2584 static int
2585 zap_get_leaf_byblk(fat_zap_t *zap, uint64_t blk, zap_leaf_t **lp)
2586 {
2587 	int bs = FZAP_BLOCK_SHIFT(zap);
2588 	int err;
2589 
2590 	*lp = malloc(sizeof(**lp));
2591 	if (*lp == NULL)
2592 		return (ENOMEM);
2593 
2594 	(*lp)->l_bs = bs;
2595 	(*lp)->l_phys = malloc(1 << bs);
2596 
2597 	if ((*lp)->l_phys == NULL) {
2598 		free(*lp);
2599 		return (ENOMEM);
2600 	}
2601 	err = dnode_read(zap->zap_spa, zap->zap_dnode, blk << bs, (*lp)->l_phys,
2602 	    1 << bs);
2603 	if (err != 0) {
2604 		zap_leaf_free(*lp);
2605 	}
2606 	return (err);
2607 }
2608 
2609 static int
2610 zap_table_load(fat_zap_t *zap, zap_table_phys_t *tbl, uint64_t idx,
2611     uint64_t *valp)
2612 {
2613 	int bs = FZAP_BLOCK_SHIFT(zap);
2614 	uint64_t blk = idx >> (bs - 3);
2615 	uint64_t off = idx & ((1 << (bs - 3)) - 1);
2616 	uint64_t *buf;
2617 	int rc;
2618 
2619 	buf = malloc(1 << zap->zap_block_shift);
2620 	if (buf == NULL)
2621 		return (ENOMEM);
2622 	rc = dnode_read(zap->zap_spa, zap->zap_dnode, (tbl->zt_blk + blk) << bs,
2623 	    buf, 1 << zap->zap_block_shift);
2624 	if (rc == 0)
2625 		*valp = buf[off];
2626 	free(buf);
2627 	return (rc);
2628 }
2629 
2630 static int
2631 zap_idx_to_blk(fat_zap_t *zap, uint64_t idx, uint64_t *valp)
2632 {
2633 	if (zap->zap_phys->zap_ptrtbl.zt_numblks == 0) {
2634 		*valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
2635 		return (0);
2636 	} else {
2637 		return (zap_table_load(zap, &zap->zap_phys->zap_ptrtbl,
2638 		    idx, valp));
2639 	}
2640 }
2641 
2642 #define	ZAP_HASH_IDX(hash, n)	(((n) == 0) ? 0 : ((hash) >> (64 - (n))))
2643 static int
2644 zap_deref_leaf(fat_zap_t *zap, uint64_t h, zap_leaf_t **lp)
2645 {
2646 	uint64_t idx, blk;
2647 	int err;
2648 
2649 	idx = ZAP_HASH_IDX(h, zap->zap_phys->zap_ptrtbl.zt_shift);
2650 	err = zap_idx_to_blk(zap, idx, &blk);
2651 	if (err != 0)
2652 		return (err);
2653 	return (zap_get_leaf_byblk(zap, blk, lp));
2654 }
2655 
2656 #define	CHAIN_END	0xffff	/* end of the chunk chain */
2657 #define	LEAF_HASH(l, h) \
2658 	((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \
2659 	((h) >> \
2660 	(64 - ZAP_LEAF_HASH_SHIFT(l) - (l)->l_phys->l_hdr.lh_prefix_len)))
2661 #define	LEAF_HASH_ENTPTR(l, h)	(&(l)->l_phys->l_hash[LEAF_HASH(l, h)])
2662 
2663 static int
2664 zap_leaf_lookup(zap_leaf_t *zl, uint64_t hash, const char *name,
2665     uint64_t integer_size, uint64_t num_integers, void *value)
2666 {
2667 	int rc;
2668 	uint16_t *chunkp;
2669 	struct zap_leaf_entry *le;
2670 
2671 	/*
2672 	 * Make sure this chunk matches our hash.
2673 	 */
2674 	if (zl->l_phys->l_hdr.lh_prefix_len > 0 &&
2675 	    zl->l_phys->l_hdr.lh_prefix !=
2676 	    hash >> (64 - zl->l_phys->l_hdr.lh_prefix_len))
2677 		return (EIO);
2678 
2679 	rc = ENOENT;
2680 	for (chunkp = LEAF_HASH_ENTPTR(zl, hash);
2681 	    *chunkp != CHAIN_END; chunkp = &le->le_next) {
2682 		zap_leaf_chunk_t *zc;
2683 		uint16_t chunk = *chunkp;
2684 
2685 		le = ZAP_LEAF_ENTRY(zl, chunk);
2686 		if (le->le_hash != hash)
2687 			continue;
2688 		zc = &ZAP_LEAF_CHUNK(zl, chunk);
2689 		if (fzap_name_equal(zl, zc, name)) {
2690 			if (zc->l_entry.le_value_intlen > integer_size) {
2691 				rc = EINVAL;
2692 			} else {
2693 				fzap_leaf_array(zl, zc, integer_size,
2694 				    num_integers, value);
2695 				rc = 0;
2696 			}
2697 			break;
2698 		}
2699 	}
2700 	return (rc);
2701 }
2702 
2703 /*
2704  * Lookup a value in a fatzap directory.
2705  */
2706 static int
2707 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2708     const char *name, uint64_t integer_size, uint64_t num_integers,
2709     void *value)
2710 {
2711 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2712 	fat_zap_t z;
2713 	zap_leaf_t *zl;
2714 	uint64_t hash;
2715 	int rc;
2716 
2717 	if (zh->zap_magic != ZAP_MAGIC)
2718 		return (EIO);
2719 
2720 	if ((rc = fzap_check_size(integer_size, num_integers)) != 0) {
2721 		return (rc);
2722 	}
2723 
2724 	z.zap_block_shift = ilog2(bsize);
2725 	z.zap_phys = zh;
2726 	z.zap_spa = spa;
2727 	z.zap_dnode = dnode;
2728 
2729 	hash = zap_hash(zh->zap_salt, name);
2730 	rc = zap_deref_leaf(&z, hash, &zl);
2731 	if (rc != 0)
2732 		return (rc);
2733 
2734 	rc = zap_leaf_lookup(zl, hash, name, integer_size, num_integers, value);
2735 
2736 	zap_leaf_free(zl);
2737 	return (rc);
2738 }
2739 
2740 /*
2741  * Lookup a name in a zap object and return its value as a uint64_t.
2742  */
2743 static int
2744 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2745     uint64_t integer_size, uint64_t num_integers, void *value)
2746 {
2747 	int rc;
2748 	zap_phys_t *zap;
2749 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2750 
2751 	zap = malloc(size);
2752 	if (zap == NULL)
2753 		return (ENOMEM);
2754 
2755 	rc = dnode_read(spa, dnode, 0, zap, size);
2756 	if (rc)
2757 		goto done;
2758 
2759 	switch (zap->zap_block_type) {
2760 	case ZBT_MICRO:
2761 		rc = mzap_lookup((const mzap_phys_t *)zap, size, name, value);
2762 		break;
2763 	case ZBT_HEADER:
2764 		rc = fzap_lookup(spa, dnode, zap, name, integer_size,
2765 		    num_integers, value);
2766 		break;
2767 	default:
2768 		printf("ZFS: invalid zap_type=%" PRIx64 "\n",
2769 		    zap->zap_block_type);
2770 		rc = EIO;
2771 	}
2772 done:
2773 	free(zap);
2774 	return (rc);
2775 }
2776 
2777 /*
2778  * List a microzap directory.
2779  */
2780 static int
2781 mzap_list(const mzap_phys_t *mz, size_t size,
2782     int (*callback)(const char *, uint64_t))
2783 {
2784 	const mzap_ent_phys_t *mze;
2785 	int chunks, i, rc;
2786 
2787 	/*
2788 	 * Microzap objects use exactly one block. Read the whole
2789 	 * thing.
2790 	 */
2791 	rc = 0;
2792 	chunks = size / MZAP_ENT_LEN - 1;
2793 	for (i = 0; i < chunks; i++) {
2794 		mze = &mz->mz_chunk[i];
2795 		if (mze->mze_name[0]) {
2796 			rc = callback(mze->mze_name, mze->mze_value);
2797 			if (rc != 0)
2798 				break;
2799 		}
2800 	}
2801 
2802 	return (rc);
2803 }
2804 
2805 /*
2806  * List a fatzap directory.
2807  */
2808 static int
2809 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2810     int (*callback)(const char *, uint64_t))
2811 {
2812 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2813 	fat_zap_t z;
2814 	uint64_t i;
2815 	int j, rc;
2816 
2817 	if (zh->zap_magic != ZAP_MAGIC)
2818 		return (EIO);
2819 
2820 	z.zap_block_shift = ilog2(bsize);
2821 	z.zap_phys = zh;
2822 
2823 	/*
2824 	 * This assumes that the leaf blocks start at block 1. The
2825 	 * documentation isn't exactly clear on this.
2826 	 */
2827 	zap_leaf_t zl;
2828 	zl.l_bs = z.zap_block_shift;
2829 	zl.l_phys = malloc(bsize);
2830 	if (zl.l_phys == NULL)
2831 		return (ENOMEM);
2832 
2833 	for (i = 0; i < zh->zap_num_leafs; i++) {
2834 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2835 		char name[256], *p;
2836 		uint64_t value;
2837 
2838 		if (dnode_read(spa, dnode, off, zl.l_phys, bsize)) {
2839 			free(zl.l_phys);
2840 			return (EIO);
2841 		}
2842 
2843 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2844 			zap_leaf_chunk_t *zc, *nc;
2845 			int namelen;
2846 
2847 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2848 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2849 				continue;
2850 			namelen = zc->l_entry.le_name_numints;
2851 			if (namelen > sizeof(name))
2852 				namelen = sizeof(name);
2853 
2854 			/*
2855 			 * Paste the name back together.
2856 			 */
2857 			nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
2858 			p = name;
2859 			while (namelen > 0) {
2860 				int len;
2861 				len = namelen;
2862 				if (len > ZAP_LEAF_ARRAY_BYTES)
2863 					len = ZAP_LEAF_ARRAY_BYTES;
2864 				memcpy(p, nc->l_array.la_array, len);
2865 				p += len;
2866 				namelen -= len;
2867 				nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
2868 			}
2869 
2870 			/*
2871 			 * Assume the first eight bytes of the value are
2872 			 * a uint64_t.
2873 			 */
2874 			value = fzap_leaf_value(&zl, zc);
2875 
2876 			/* printf("%s 0x%jx\n", name, (uintmax_t)value); */
2877 			rc = callback((const char *)name, value);
2878 			if (rc != 0) {
2879 				free(zl.l_phys);
2880 				return (rc);
2881 			}
2882 		}
2883 	}
2884 
2885 	free(zl.l_phys);
2886 	return (0);
2887 }
2888 
2889 static int zfs_printf(const char *name, uint64_t value __unused)
2890 {
2891 
2892 	printf("%s\n", name);
2893 
2894 	return (0);
2895 }
2896 
2897 /*
2898  * List a zap directory.
2899  */
2900 static int
2901 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
2902 {
2903 	zap_phys_t *zap;
2904 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2905 	int rc;
2906 
2907 	zap = malloc(size);
2908 	if (zap == NULL)
2909 		return (ENOMEM);
2910 
2911 	rc = dnode_read(spa, dnode, 0, zap, size);
2912 	if (rc == 0) {
2913 		if (zap->zap_block_type == ZBT_MICRO)
2914 			rc = mzap_list((const mzap_phys_t *)zap, size,
2915 			    zfs_printf);
2916 		else
2917 			rc = fzap_list(spa, dnode, zap, zfs_printf);
2918 	}
2919 	free(zap);
2920 	return (rc);
2921 }
2922 
2923 static int
2924 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum,
2925     dnode_phys_t *dnode)
2926 {
2927 	off_t offset;
2928 
2929 	offset = objnum * sizeof(dnode_phys_t);
2930 	return dnode_read(spa, &os->os_meta_dnode, offset,
2931 		dnode, sizeof(dnode_phys_t));
2932 }
2933 
2934 /*
2935  * Lookup a name in a microzap directory.
2936  */
2937 static int
2938 mzap_rlookup(const mzap_phys_t *mz, size_t size, char *name, uint64_t value)
2939 {
2940 	const mzap_ent_phys_t *mze;
2941 	int chunks, i;
2942 
2943 	/*
2944 	 * Microzap objects use exactly one block. Read the whole
2945 	 * thing.
2946 	 */
2947 	chunks = size / MZAP_ENT_LEN - 1;
2948 	for (i = 0; i < chunks; i++) {
2949 		mze = &mz->mz_chunk[i];
2950 		if (value == mze->mze_value) {
2951 			strcpy(name, mze->mze_name);
2952 			return (0);
2953 		}
2954 	}
2955 
2956 	return (ENOENT);
2957 }
2958 
2959 static void
2960 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
2961 {
2962 	size_t namelen;
2963 	const zap_leaf_chunk_t *nc;
2964 	char *p;
2965 
2966 	namelen = zc->l_entry.le_name_numints;
2967 
2968 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2969 	p = name;
2970 	while (namelen > 0) {
2971 		size_t len;
2972 		len = namelen;
2973 		if (len > ZAP_LEAF_ARRAY_BYTES)
2974 			len = ZAP_LEAF_ARRAY_BYTES;
2975 		memcpy(p, nc->l_array.la_array, len);
2976 		p += len;
2977 		namelen -= len;
2978 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2979 	}
2980 
2981 	*p = '\0';
2982 }
2983 
2984 static int
2985 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2986     char *name, uint64_t value)
2987 {
2988 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2989 	fat_zap_t z;
2990 	uint64_t i;
2991 	int j, rc;
2992 
2993 	if (zh->zap_magic != ZAP_MAGIC)
2994 		return (EIO);
2995 
2996 	z.zap_block_shift = ilog2(bsize);
2997 	z.zap_phys = zh;
2998 
2999 	/*
3000 	 * This assumes that the leaf blocks start at block 1. The
3001 	 * documentation isn't exactly clear on this.
3002 	 */
3003 	zap_leaf_t zl;
3004 	zl.l_bs = z.zap_block_shift;
3005 	zl.l_phys = malloc(bsize);
3006 	if (zl.l_phys == NULL)
3007 		return (ENOMEM);
3008 
3009 	for (i = 0; i < zh->zap_num_leafs; i++) {
3010 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
3011 
3012 		rc = dnode_read(spa, dnode, off, zl.l_phys, bsize);
3013 		if (rc != 0)
3014 			goto done;
3015 
3016 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
3017 			zap_leaf_chunk_t *zc;
3018 
3019 			zc = &ZAP_LEAF_CHUNK(&zl, j);
3020 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
3021 				continue;
3022 			if (zc->l_entry.le_value_intlen != 8 ||
3023 			    zc->l_entry.le_value_numints != 1)
3024 				continue;
3025 
3026 			if (fzap_leaf_value(&zl, zc) == value) {
3027 				fzap_name_copy(&zl, zc, name);
3028 				goto done;
3029 			}
3030 		}
3031 	}
3032 
3033 	rc = ENOENT;
3034 done:
3035 	free(zl.l_phys);
3036 	return (rc);
3037 }
3038 
3039 static int
3040 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name,
3041     uint64_t value)
3042 {
3043 	zap_phys_t *zap;
3044 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
3045 	int rc;
3046 
3047 	zap = malloc(size);
3048 	if (zap == NULL)
3049 		return (ENOMEM);
3050 
3051 	rc = dnode_read(spa, dnode, 0, zap, size);
3052 	if (rc == 0) {
3053 		if (zap->zap_block_type == ZBT_MICRO)
3054 			rc = mzap_rlookup((const mzap_phys_t *)zap, size,
3055 			    name, value);
3056 		else
3057 			rc = fzap_rlookup(spa, dnode, zap, name, value);
3058 	}
3059 	free(zap);
3060 	return (rc);
3061 }
3062 
3063 static int
3064 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
3065 {
3066 	char name[256];
3067 	char component[256];
3068 	uint64_t dir_obj, parent_obj, child_dir_zapobj;
3069 	dnode_phys_t child_dir_zap, dataset, dir, parent;
3070 	dsl_dir_phys_t *dd;
3071 	dsl_dataset_phys_t *ds;
3072 	char *p;
3073 	int len;
3074 
3075 	p = &name[sizeof(name) - 1];
3076 	*p = '\0';
3077 
3078 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
3079 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3080 		return (EIO);
3081 	}
3082 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3083 	dir_obj = ds->ds_dir_obj;
3084 
3085 	for (;;) {
3086 		if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir) != 0)
3087 			return (EIO);
3088 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3089 
3090 		/* Actual loop condition. */
3091 		parent_obj = dd->dd_parent_obj;
3092 		if (parent_obj == 0)
3093 			break;
3094 
3095 		if (objset_get_dnode(spa, spa->spa_mos, parent_obj,
3096 		    &parent) != 0)
3097 			return (EIO);
3098 		dd = (dsl_dir_phys_t *)&parent.dn_bonus;
3099 		child_dir_zapobj = dd->dd_child_dir_zapobj;
3100 		if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
3101 		    &child_dir_zap) != 0)
3102 			return (EIO);
3103 		if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
3104 			return (EIO);
3105 
3106 		len = strlen(component);
3107 		p -= len;
3108 		memcpy(p, component, len);
3109 		--p;
3110 		*p = '/';
3111 
3112 		/* Actual loop iteration. */
3113 		dir_obj = parent_obj;
3114 	}
3115 
3116 	if (*p != '\0')
3117 		++p;
3118 	strcpy(result, p);
3119 
3120 	return (0);
3121 }
3122 
3123 static int
3124 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
3125 {
3126 	char element[256];
3127 	uint64_t dir_obj, child_dir_zapobj;
3128 	dnode_phys_t child_dir_zap, dir;
3129 	dsl_dir_phys_t *dd;
3130 	const char *p, *q;
3131 
3132 	if (objset_get_dnode(spa, spa->spa_mos,
3133 	    DMU_POOL_DIRECTORY_OBJECT, &dir))
3134 		return (EIO);
3135 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
3136 	    1, &dir_obj))
3137 		return (EIO);
3138 
3139 	p = name;
3140 	for (;;) {
3141 		if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir))
3142 			return (EIO);
3143 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3144 
3145 		while (*p == '/')
3146 			p++;
3147 		/* Actual loop condition #1. */
3148 		if (*p == '\0')
3149 			break;
3150 
3151 		q = strchr(p, '/');
3152 		if (q) {
3153 			memcpy(element, p, q - p);
3154 			element[q - p] = '\0';
3155 			p = q + 1;
3156 		} else {
3157 			strcpy(element, p);
3158 			p += strlen(p);
3159 		}
3160 
3161 		child_dir_zapobj = dd->dd_child_dir_zapobj;
3162 		if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
3163 		    &child_dir_zap) != 0)
3164 			return (EIO);
3165 
3166 		/* Actual loop condition #2. */
3167 		if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
3168 		    1, &dir_obj) != 0)
3169 			return (ENOENT);
3170 	}
3171 
3172 	*objnum = dd->dd_head_dataset_obj;
3173 	return (0);
3174 }
3175 
3176 #ifndef BOOT2
3177 static int
3178 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
3179 {
3180 	uint64_t dir_obj, child_dir_zapobj;
3181 	dnode_phys_t child_dir_zap, dir, dataset;
3182 	dsl_dataset_phys_t *ds;
3183 	dsl_dir_phys_t *dd;
3184 
3185 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
3186 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3187 		return (EIO);
3188 	}
3189 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3190 	dir_obj = ds->ds_dir_obj;
3191 
3192 	if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir)) {
3193 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
3194 		return (EIO);
3195 	}
3196 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3197 
3198 	child_dir_zapobj = dd->dd_child_dir_zapobj;
3199 	if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
3200 	    &child_dir_zap) != 0) {
3201 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
3202 		return (EIO);
3203 	}
3204 
3205 	return (zap_list(spa, &child_dir_zap) != 0);
3206 }
3207 
3208 int
3209 zfs_callback_dataset(const spa_t *spa, uint64_t objnum,
3210     int (*callback)(const char *, uint64_t))
3211 {
3212 	uint64_t dir_obj, child_dir_zapobj;
3213 	dnode_phys_t child_dir_zap, dir, dataset;
3214 	dsl_dataset_phys_t *ds;
3215 	dsl_dir_phys_t *dd;
3216 	zap_phys_t *zap;
3217 	size_t size;
3218 	int err;
3219 
3220 	err = objset_get_dnode(spa, spa->spa_mos, objnum, &dataset);
3221 	if (err != 0) {
3222 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3223 		return (err);
3224 	}
3225 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3226 	dir_obj = ds->ds_dir_obj;
3227 
3228 	err = objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir);
3229 	if (err != 0) {
3230 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
3231 		return (err);
3232 	}
3233 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3234 
3235 	child_dir_zapobj = dd->dd_child_dir_zapobj;
3236 	err = objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
3237 	    &child_dir_zap);
3238 	if (err != 0) {
3239 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
3240 		return (err);
3241 	}
3242 
3243 	size = child_dir_zap.dn_datablkszsec << SPA_MINBLOCKSHIFT;
3244 	zap = malloc(size);
3245 	if (zap != NULL) {
3246 		err = dnode_read(spa, &child_dir_zap, 0, zap, size);
3247 		if (err != 0)
3248 			goto done;
3249 
3250 		if (zap->zap_block_type == ZBT_MICRO)
3251 			err = mzap_list((const mzap_phys_t *)zap, size,
3252 			    callback);
3253 		else
3254 			err = fzap_list(spa, &child_dir_zap, zap, callback);
3255 	} else {
3256 		err = ENOMEM;
3257 	}
3258 done:
3259 	free(zap);
3260 	return (err);
3261 }
3262 #endif
3263 
3264 /*
3265  * Find the object set given the object number of its dataset object
3266  * and return its details in *objset
3267  */
3268 static int
3269 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
3270 {
3271 	dnode_phys_t dataset;
3272 	dsl_dataset_phys_t *ds;
3273 
3274 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
3275 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3276 		return (EIO);
3277 	}
3278 
3279 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3280 	if (zio_read(spa, &ds->ds_bp, objset)) {
3281 		printf("ZFS: can't read object set for dataset %ju\n",
3282 		    (uintmax_t)objnum);
3283 		return (EIO);
3284 	}
3285 
3286 	return (0);
3287 }
3288 
3289 /*
3290  * Find the object set pointed to by the BOOTFS property or the root
3291  * dataset if there is none and return its details in *objset
3292  */
3293 static int
3294 zfs_get_root(const spa_t *spa, uint64_t *objid)
3295 {
3296 	dnode_phys_t dir, propdir;
3297 	uint64_t props, bootfs, root;
3298 
3299 	*objid = 0;
3300 
3301 	/*
3302 	 * Start with the MOS directory object.
3303 	 */
3304 	if (objset_get_dnode(spa, spa->spa_mos,
3305 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3306 		printf("ZFS: can't read MOS object directory\n");
3307 		return (EIO);
3308 	}
3309 
3310 	/*
3311 	 * Lookup the pool_props and see if we can find a bootfs.
3312 	 */
3313 	if (zap_lookup(spa, &dir, DMU_POOL_PROPS,
3314 	    sizeof(props), 1, &props) == 0 &&
3315 	    objset_get_dnode(spa, spa->spa_mos, props, &propdir) == 0 &&
3316 	    zap_lookup(spa, &propdir, "bootfs",
3317 	    sizeof(bootfs), 1, &bootfs) == 0 && bootfs != 0) {
3318 		*objid = bootfs;
3319 		return (0);
3320 	}
3321 	/*
3322 	 * Lookup the root dataset directory
3323 	 */
3324 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET,
3325 	    sizeof(root), 1, &root) ||
3326 	    objset_get_dnode(spa, spa->spa_mos, root, &dir)) {
3327 		printf("ZFS: can't find root dsl_dir\n");
3328 		return (EIO);
3329 	}
3330 
3331 	/*
3332 	 * Use the information from the dataset directory's bonus buffer
3333 	 * to find the dataset object and from that the object set itself.
3334 	 */
3335 	dsl_dir_phys_t *dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3336 	*objid = dd->dd_head_dataset_obj;
3337 	return (0);
3338 }
3339 
3340 static int
3341 zfs_mount_impl(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
3342 {
3343 
3344 	mount->spa = spa;
3345 
3346 	/*
3347 	 * Find the root object set if not explicitly provided
3348 	 */
3349 	if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
3350 		printf("ZFS: can't find root filesystem\n");
3351 		return (EIO);
3352 	}
3353 
3354 	if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
3355 		printf("ZFS: can't open root filesystem\n");
3356 		return (EIO);
3357 	}
3358 
3359 	mount->rootobj = rootobj;
3360 
3361 	return (0);
3362 }
3363 
3364 /*
3365  * callback function for feature name checks.
3366  */
3367 static int
3368 check_feature(const char *name, uint64_t value)
3369 {
3370 	int i;
3371 
3372 	if (value == 0)
3373 		return (0);
3374 	if (name[0] == '\0')
3375 		return (0);
3376 
3377 	for (i = 0; features_for_read[i] != NULL; i++) {
3378 		if (strcmp(name, features_for_read[i]) == 0)
3379 			return (0);
3380 	}
3381 	printf("ZFS: unsupported feature: %s\n", name);
3382 	return (EIO);
3383 }
3384 
3385 /*
3386  * Checks whether the MOS features that are active are supported.
3387  */
3388 static int
3389 check_mos_features(const spa_t *spa)
3390 {
3391 	dnode_phys_t dir;
3392 	zap_phys_t *zap;
3393 	uint64_t objnum;
3394 	size_t size;
3395 	int rc;
3396 
3397 	if ((rc = objset_get_dnode(spa, spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
3398 	    &dir)) != 0)
3399 		return (rc);
3400 	if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
3401 	    sizeof (objnum), 1, &objnum)) != 0) {
3402 		/*
3403 		 * It is older pool without features. As we have already
3404 		 * tested the label, just return without raising the error.
3405 		 */
3406 		return (0);
3407 	}
3408 
3409 	if ((rc = objset_get_dnode(spa, spa->spa_mos, objnum, &dir)) != 0)
3410 		return (rc);
3411 
3412 	if (dir.dn_type != DMU_OTN_ZAP_METADATA)
3413 		return (EIO);
3414 
3415 	size = dir.dn_datablkszsec << SPA_MINBLOCKSHIFT;
3416 	zap = malloc(size);
3417 	if (zap == NULL)
3418 		return (ENOMEM);
3419 
3420 	if (dnode_read(spa, &dir, 0, zap, size)) {
3421 		free(zap);
3422 		return (EIO);
3423 	}
3424 
3425 	if (zap->zap_block_type == ZBT_MICRO)
3426 		rc = mzap_list((const mzap_phys_t *)zap, size, check_feature);
3427 	else
3428 		rc = fzap_list(spa, &dir, zap, check_feature);
3429 
3430 	free(zap);
3431 	return (rc);
3432 }
3433 
3434 static int
3435 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
3436 {
3437 	dnode_phys_t dir;
3438 	size_t size;
3439 	int rc;
3440 	char *nv;
3441 
3442 	*value = NULL;
3443 	if ((rc = objset_get_dnode(spa, spa->spa_mos, obj, &dir)) != 0)
3444 		return (rc);
3445 	if (dir.dn_type != DMU_OT_PACKED_NVLIST &&
3446 	    dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) {
3447 		return (EIO);
3448 	}
3449 
3450 	if (dir.dn_bonuslen != sizeof (uint64_t))
3451 		return (EIO);
3452 
3453 	size = *(uint64_t *)DN_BONUS(&dir);
3454 	nv = malloc(size);
3455 	if (nv == NULL)
3456 		return (ENOMEM);
3457 
3458 	rc = dnode_read(spa, &dir, 0, nv, size);
3459 	if (rc != 0) {
3460 		free(nv);
3461 		nv = NULL;
3462 		return (rc);
3463 	}
3464 	*value = nvlist_import(nv, size);
3465 	free(nv);
3466 	return (rc);
3467 }
3468 
3469 static int
3470 zfs_spa_init(spa_t *spa)
3471 {
3472 	struct uberblock checkpoint;
3473 	dnode_phys_t dir;
3474 	uint64_t config_object;
3475 	nvlist_t *nvlist;
3476 	int rc;
3477 
3478 	if (zio_read(spa, &spa->spa_uberblock->ub_rootbp, spa->spa_mos)) {
3479 		printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
3480 		return (EIO);
3481 	}
3482 	if (spa->spa_mos->os_type != DMU_OST_META) {
3483 		printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
3484 		return (EIO);
3485 	}
3486 
3487 	if (objset_get_dnode(spa, &spa->spa_mos_master,
3488 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3489 		printf("ZFS: failed to read pool %s directory object\n",
3490 		    spa->spa_name);
3491 		return (EIO);
3492 	}
3493 	/* this is allowed to fail, older pools do not have salt */
3494 	rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
3495 	    sizeof (spa->spa_cksum_salt.zcs_bytes),
3496 	    spa->spa_cksum_salt.zcs_bytes);
3497 
3498 	rc = check_mos_features(spa);
3499 	if (rc != 0) {
3500 		printf("ZFS: pool %s is not supported\n", spa->spa_name);
3501 		return (rc);
3502 	}
3503 
3504 	rc = zap_lookup(spa, &dir, DMU_POOL_CONFIG,
3505 	    sizeof (config_object), 1, &config_object);
3506 	if (rc != 0) {
3507 		printf("ZFS: can not read MOS %s\n", DMU_POOL_CONFIG);
3508 		return (EIO);
3509 	}
3510 	rc = load_nvlist(spa, config_object, &nvlist);
3511 	if (rc != 0)
3512 		return (rc);
3513 
3514 	rc = zap_lookup(spa, &dir, DMU_POOL_ZPOOL_CHECKPOINT,
3515 	    sizeof(uint64_t), sizeof(checkpoint) / sizeof(uint64_t),
3516 	    &checkpoint);
3517 	if (rc == 0 && checkpoint.ub_checkpoint_txg != 0) {
3518 		memcpy(&spa->spa_uberblock_checkpoint, &checkpoint,
3519 		    sizeof(checkpoint));
3520 		if (zio_read(spa, &spa->spa_uberblock_checkpoint.ub_rootbp,
3521 		    &spa->spa_mos_checkpoint)) {
3522 			printf("ZFS: can not read checkpoint data.\n");
3523 			return (EIO);
3524 		}
3525 	}
3526 
3527 	/*
3528 	 * Update vdevs from MOS config. Note, we do skip encoding bytes
3529 	 * here. See also vdev_label_read_config().
3530 	 */
3531 	rc = vdev_init_from_nvlist(spa, nvlist);
3532 	nvlist_destroy(nvlist);
3533 	return (rc);
3534 }
3535 
3536 static int
3537 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
3538 {
3539 
3540 	if (dn->dn_bonustype != DMU_OT_SA) {
3541 		znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
3542 
3543 		sb->st_mode = zp->zp_mode;
3544 		sb->st_uid = zp->zp_uid;
3545 		sb->st_gid = zp->zp_gid;
3546 		sb->st_size = zp->zp_size;
3547 	} else {
3548 		sa_hdr_phys_t *sahdrp;
3549 		int hdrsize;
3550 		size_t size = 0;
3551 		void *buf = NULL;
3552 
3553 		if (dn->dn_bonuslen != 0)
3554 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3555 		else {
3556 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
3557 				blkptr_t *bp = DN_SPILL_BLKPTR(dn);
3558 				int error;
3559 
3560 				size = BP_GET_LSIZE(bp);
3561 				buf = malloc(size);
3562 				if (buf == NULL)
3563 					error = ENOMEM;
3564 				else
3565 					error = zio_read(spa, bp, buf);
3566 
3567 				if (error != 0) {
3568 					free(buf);
3569 					return (error);
3570 				}
3571 				sahdrp = buf;
3572 			} else {
3573 				return (EIO);
3574 			}
3575 		}
3576 		hdrsize = SA_HDR_SIZE(sahdrp);
3577 		sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
3578 		    SA_MODE_OFFSET);
3579 		sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
3580 		    SA_UID_OFFSET);
3581 		sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
3582 		    SA_GID_OFFSET);
3583 		sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
3584 		    SA_SIZE_OFFSET);
3585 		free(buf);
3586 	}
3587 
3588 	return (0);
3589 }
3590 
3591 static int
3592 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
3593 {
3594 	int rc = 0;
3595 
3596 	if (dn->dn_bonustype == DMU_OT_SA) {
3597 		sa_hdr_phys_t *sahdrp = NULL;
3598 		size_t size = 0;
3599 		void *buf = NULL;
3600 		int hdrsize;
3601 		char *p;
3602 
3603 		if (dn->dn_bonuslen != 0) {
3604 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3605 		} else {
3606 			blkptr_t *bp;
3607 
3608 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
3609 				return (EIO);
3610 			bp = DN_SPILL_BLKPTR(dn);
3611 
3612 			size = BP_GET_LSIZE(bp);
3613 			buf = malloc(size);
3614 			if (buf == NULL)
3615 				rc = ENOMEM;
3616 			else
3617 				rc = zio_read(spa, bp, buf);
3618 			if (rc != 0) {
3619 				free(buf);
3620 				return (rc);
3621 			}
3622 			sahdrp = buf;
3623 		}
3624 		hdrsize = SA_HDR_SIZE(sahdrp);
3625 		p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
3626 		memcpy(path, p, psize);
3627 		free(buf);
3628 		return (0);
3629 	}
3630 	/*
3631 	 * Second test is purely to silence bogus compiler
3632 	 * warning about accessing past the end of dn_bonus.
3633 	 */
3634 	if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
3635 	    sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
3636 		memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
3637 	} else {
3638 		rc = dnode_read(spa, dn, 0, path, psize);
3639 	}
3640 	return (rc);
3641 }
3642 
3643 struct obj_list {
3644 	uint64_t		objnum;
3645 	STAILQ_ENTRY(obj_list)	entry;
3646 };
3647 
3648 /*
3649  * Lookup a file and return its dnode.
3650  */
3651 static int
3652 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
3653 {
3654 	int rc;
3655 	uint64_t objnum;
3656 	const spa_t *spa;
3657 	dnode_phys_t dn;
3658 	const char *p, *q;
3659 	char element[256];
3660 	char path[1024];
3661 	int symlinks_followed = 0;
3662 	struct stat sb;
3663 	struct obj_list *entry, *tentry;
3664 	STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
3665 
3666 	spa = mount->spa;
3667 	if (mount->objset.os_type != DMU_OST_ZFS) {
3668 		printf("ZFS: unexpected object set type %ju\n",
3669 		    (uintmax_t)mount->objset.os_type);
3670 		return (EIO);
3671 	}
3672 
3673 	if ((entry = malloc(sizeof(struct obj_list))) == NULL)
3674 		return (ENOMEM);
3675 
3676 	/*
3677 	 * Get the root directory dnode.
3678 	 */
3679 	rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
3680 	if (rc) {
3681 		free(entry);
3682 		return (rc);
3683 	}
3684 
3685 	rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof(objnum), 1, &objnum);
3686 	if (rc) {
3687 		free(entry);
3688 		return (rc);
3689 	}
3690 	entry->objnum = objnum;
3691 	STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3692 
3693 	rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3694 	if (rc != 0)
3695 		goto done;
3696 
3697 	p = upath;
3698 	while (p && *p) {
3699 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3700 		if (rc != 0)
3701 			goto done;
3702 
3703 		while (*p == '/')
3704 			p++;
3705 		if (*p == '\0')
3706 			break;
3707 		q = p;
3708 		while (*q != '\0' && *q != '/')
3709 			q++;
3710 
3711 		/* skip dot */
3712 		if (p + 1 == q && p[0] == '.') {
3713 			p++;
3714 			continue;
3715 		}
3716 		/* double dot */
3717 		if (p + 2 == q && p[0] == '.' && p[1] == '.') {
3718 			p += 2;
3719 			if (STAILQ_FIRST(&on_cache) ==
3720 			    STAILQ_LAST(&on_cache, obj_list, entry)) {
3721 				rc = ENOENT;
3722 				goto done;
3723 			}
3724 			entry = STAILQ_FIRST(&on_cache);
3725 			STAILQ_REMOVE_HEAD(&on_cache, entry);
3726 			free(entry);
3727 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3728 			continue;
3729 		}
3730 		if (q - p + 1 > sizeof(element)) {
3731 			rc = ENAMETOOLONG;
3732 			goto done;
3733 		}
3734 		memcpy(element, p, q - p);
3735 		element[q - p] = 0;
3736 		p = q;
3737 
3738 		if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
3739 			goto done;
3740 		if (!S_ISDIR(sb.st_mode)) {
3741 			rc = ENOTDIR;
3742 			goto done;
3743 		}
3744 
3745 		rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum);
3746 		if (rc)
3747 			goto done;
3748 		objnum = ZFS_DIRENT_OBJ(objnum);
3749 
3750 		if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
3751 			rc = ENOMEM;
3752 			goto done;
3753 		}
3754 		entry->objnum = objnum;
3755 		STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3756 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3757 		if (rc)
3758 			goto done;
3759 
3760 		/*
3761 		 * Check for symlink.
3762 		 */
3763 		rc = zfs_dnode_stat(spa, &dn, &sb);
3764 		if (rc)
3765 			goto done;
3766 		if (S_ISLNK(sb.st_mode)) {
3767 			if (symlinks_followed > 10) {
3768 				rc = EMLINK;
3769 				goto done;
3770 			}
3771 			symlinks_followed++;
3772 
3773 			/*
3774 			 * Read the link value and copy the tail of our
3775 			 * current path onto the end.
3776 			 */
3777 			if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
3778 				rc = ENAMETOOLONG;
3779 				goto done;
3780 			}
3781 			strcpy(&path[sb.st_size], p);
3782 
3783 			rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
3784 			if (rc != 0)
3785 				goto done;
3786 
3787 			/*
3788 			 * Restart with the new path, starting either at
3789 			 * the root or at the parent depending whether or
3790 			 * not the link is relative.
3791 			 */
3792 			p = path;
3793 			if (*p == '/') {
3794 				while (STAILQ_FIRST(&on_cache) !=
3795 				    STAILQ_LAST(&on_cache, obj_list, entry)) {
3796 					entry = STAILQ_FIRST(&on_cache);
3797 					STAILQ_REMOVE_HEAD(&on_cache, entry);
3798 					free(entry);
3799 				}
3800 			} else {
3801 				entry = STAILQ_FIRST(&on_cache);
3802 				STAILQ_REMOVE_HEAD(&on_cache, entry);
3803 				free(entry);
3804 			}
3805 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3806 		}
3807 	}
3808 
3809 	*dnode = dn;
3810 done:
3811 	STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)
3812 		free(entry);
3813 	return (rc);
3814 }
3815