xref: /freebsd/stand/libsa/zfs/zfsimpl.c (revision 2fec3ae8964c8864a9e75d6a2f0ed137ede489a7)
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 extern int zstd_init(void);
46 
47 struct zfsmount {
48 	const spa_t	*spa;
49 	objset_phys_t	objset;
50 	uint64_t	rootobj;
51 };
52 static struct zfsmount zfsmount __unused;
53 
54 /*
55  * The indirect_child_t represents the vdev that we will read from, when we
56  * need to read all copies of the data (e.g. for scrub or reconstruction).
57  * For plain (non-mirror) top-level vdevs (i.e. is_vdev is not a mirror),
58  * ic_vdev is the same as is_vdev.  However, for mirror top-level vdevs,
59  * ic_vdev is a child of the mirror.
60  */
61 typedef struct indirect_child {
62 	void *ic_data;
63 	vdev_t *ic_vdev;
64 } indirect_child_t;
65 
66 /*
67  * The indirect_split_t represents one mapped segment of an i/o to the
68  * indirect vdev. For non-split (contiguously-mapped) blocks, there will be
69  * only one indirect_split_t, with is_split_offset==0 and is_size==io_size.
70  * For split blocks, there will be several of these.
71  */
72 typedef struct indirect_split {
73 	list_node_t is_node; /* link on iv_splits */
74 
75 	/*
76 	 * is_split_offset is the offset into the i/o.
77 	 * This is the sum of the previous splits' is_size's.
78 	 */
79 	uint64_t is_split_offset;
80 
81 	vdev_t *is_vdev; /* top-level vdev */
82 	uint64_t is_target_offset; /* offset on is_vdev */
83 	uint64_t is_size;
84 	int is_children; /* number of entries in is_child[] */
85 
86 	/*
87 	 * is_good_child is the child that we are currently using to
88 	 * attempt reconstruction.
89 	 */
90 	int is_good_child;
91 
92 	indirect_child_t is_child[1]; /* variable-length */
93 } indirect_split_t;
94 
95 /*
96  * The indirect_vsd_t is associated with each i/o to the indirect vdev.
97  * It is the "Vdev-Specific Data" in the zio_t's io_vsd.
98  */
99 typedef struct indirect_vsd {
100 	boolean_t iv_split_block;
101 	boolean_t iv_reconstruct;
102 
103 	list_t iv_splits; /* list of indirect_split_t's */
104 } indirect_vsd_t;
105 
106 /*
107  * List of all vdevs, chained through v_alllink.
108  */
109 static vdev_list_t zfs_vdevs;
110 
111 /*
112  * List of ZFS features supported for read
113  */
114 static const char *features_for_read[] = {
115 	"org.illumos:lz4_compress",
116 	"com.delphix:hole_birth",
117 	"com.delphix:extensible_dataset",
118 	"com.delphix:embedded_data",
119 	"org.open-zfs:large_blocks",
120 	"org.illumos:sha512",
121 	"org.illumos:skein",
122 	"org.zfsonlinux:large_dnode",
123 	"com.joyent:multi_vdev_crash_dump",
124 	"com.delphix:spacemap_histogram",
125 	"com.delphix:zpool_checkpoint",
126 	"com.delphix:spacemap_v2",
127 	"com.datto:encryption",
128 	"org.zfsonlinux:allocation_classes",
129 	"com.datto:resilver_defer",
130 	"com.delphix:device_removal",
131 	"com.delphix:obsolete_counts",
132 	"com.intel:allocation_classes",
133 	"org.freebsd:zstd_compress",
134 	NULL
135 };
136 
137 /*
138  * List of all pools, chained through spa_link.
139  */
140 static spa_list_t zfs_pools;
141 
142 static const dnode_phys_t *dnode_cache_obj;
143 static uint64_t dnode_cache_bn;
144 static char *dnode_cache_buf;
145 
146 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
147 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
148 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
149 static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode,
150     const char *name, uint64_t integer_size, uint64_t num_integers,
151     void *value);
152 static int objset_get_dnode(const spa_t *, const objset_phys_t *, uint64_t,
153     dnode_phys_t *);
154 static int dnode_read(const spa_t *, const dnode_phys_t *, off_t, void *,
155     size_t);
156 static int vdev_indirect_read(vdev_t *, const blkptr_t *, void *, off_t,
157     size_t);
158 static int vdev_mirror_read(vdev_t *, const blkptr_t *, void *, off_t, size_t);
159 vdev_indirect_mapping_t *vdev_indirect_mapping_open(spa_t *, objset_phys_t *,
160     uint64_t);
161 vdev_indirect_mapping_entry_phys_t *
162     vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *, uint64_t,
163     uint64_t, uint64_t *);
164 
165 static void
166 zfs_init(void)
167 {
168 	STAILQ_INIT(&zfs_vdevs);
169 	STAILQ_INIT(&zfs_pools);
170 
171 	dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
172 
173 	zfs_init_crc();
174 	zstd_init();
175 }
176 
177 static int
178 nvlist_check_features_for_read(nvlist_t *nvl)
179 {
180 	nvlist_t *features = NULL;
181 	nvs_data_t *data;
182 	nvp_header_t *nvp;
183 	nv_string_t *nvp_name;
184 	int rc;
185 
186 	rc = nvlist_find(nvl, ZPOOL_CONFIG_FEATURES_FOR_READ,
187 	    DATA_TYPE_NVLIST, NULL, &features, NULL);
188 	if (rc != 0)
189 		return (rc);
190 
191 	data = (nvs_data_t *)features->nv_data;
192 	nvp = &data->nvl_pair;	/* first pair in nvlist */
193 
194 	while (nvp->encoded_size != 0 && nvp->decoded_size != 0) {
195 		int i, found;
196 
197 		nvp_name = (nv_string_t *)((uintptr_t)nvp + sizeof(*nvp));
198 		found = 0;
199 
200 		for (i = 0; features_for_read[i] != NULL; i++) {
201 			if (memcmp(nvp_name->nv_data, features_for_read[i],
202 			    nvp_name->nv_size) == 0) {
203 				found = 1;
204 				break;
205 			}
206 		}
207 
208 		if (!found) {
209 			printf("ZFS: unsupported feature: %.*s\n",
210 			    nvp_name->nv_size, nvp_name->nv_data);
211 			rc = EIO;
212 		}
213 		nvp = (nvp_header_t *)((uint8_t *)nvp + nvp->encoded_size);
214 	}
215 	nvlist_destroy(features);
216 
217 	return (rc);
218 }
219 
220 static int
221 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
222     off_t offset, size_t size)
223 {
224 	size_t psize;
225 	int rc;
226 
227 	if (vdev->v_phys_read == NULL)
228 		return (ENOTSUP);
229 
230 	if (bp) {
231 		psize = BP_GET_PSIZE(bp);
232 	} else {
233 		psize = size;
234 	}
235 
236 	rc = vdev->v_phys_read(vdev, vdev->v_priv, offset, buf, psize);
237 	if (rc == 0) {
238 		if (bp != NULL)
239 			rc = zio_checksum_verify(vdev->v_spa, bp, buf);
240 	}
241 
242 	return (rc);
243 }
244 
245 static int
246 vdev_write_phys(vdev_t *vdev, void *buf, off_t offset, size_t size)
247 {
248 	if (vdev->v_phys_write == NULL)
249 		return (ENOTSUP);
250 
251 	return (vdev->v_phys_write(vdev, offset, buf, size));
252 }
253 
254 typedef struct remap_segment {
255 	vdev_t *rs_vd;
256 	uint64_t rs_offset;
257 	uint64_t rs_asize;
258 	uint64_t rs_split_offset;
259 	list_node_t rs_node;
260 } remap_segment_t;
261 
262 static remap_segment_t *
263 rs_alloc(vdev_t *vd, uint64_t offset, uint64_t asize, uint64_t split_offset)
264 {
265 	remap_segment_t *rs = malloc(sizeof (remap_segment_t));
266 
267 	if (rs != NULL) {
268 		rs->rs_vd = vd;
269 		rs->rs_offset = offset;
270 		rs->rs_asize = asize;
271 		rs->rs_split_offset = split_offset;
272 	}
273 
274 	return (rs);
275 }
276 
277 vdev_indirect_mapping_t *
278 vdev_indirect_mapping_open(spa_t *spa, objset_phys_t *os,
279     uint64_t mapping_object)
280 {
281 	vdev_indirect_mapping_t *vim;
282 	vdev_indirect_mapping_phys_t *vim_phys;
283 	int rc;
284 
285 	vim = calloc(1, sizeof (*vim));
286 	if (vim == NULL)
287 		return (NULL);
288 
289 	vim->vim_dn = calloc(1, sizeof (*vim->vim_dn));
290 	if (vim->vim_dn == NULL) {
291 		free(vim);
292 		return (NULL);
293 	}
294 
295 	rc = objset_get_dnode(spa, os, mapping_object, vim->vim_dn);
296 	if (rc != 0) {
297 		free(vim->vim_dn);
298 		free(vim);
299 		return (NULL);
300 	}
301 
302 	vim->vim_spa = spa;
303 	vim->vim_phys = malloc(sizeof (*vim->vim_phys));
304 	if (vim->vim_phys == NULL) {
305 		free(vim->vim_dn);
306 		free(vim);
307 		return (NULL);
308 	}
309 
310 	vim_phys = (vdev_indirect_mapping_phys_t *)DN_BONUS(vim->vim_dn);
311 	*vim->vim_phys = *vim_phys;
312 
313 	vim->vim_objset = os;
314 	vim->vim_object = mapping_object;
315 	vim->vim_entries = NULL;
316 
317 	vim->vim_havecounts =
318 	    (vim->vim_dn->dn_bonuslen > VDEV_INDIRECT_MAPPING_SIZE_V0);
319 
320 	return (vim);
321 }
322 
323 /*
324  * Compare an offset with an indirect mapping entry; there are three
325  * possible scenarios:
326  *
327  *     1. The offset is "less than" the mapping entry; meaning the
328  *        offset is less than the source offset of the mapping entry. In
329  *        this case, there is no overlap between the offset and the
330  *        mapping entry and -1 will be returned.
331  *
332  *     2. The offset is "greater than" the mapping entry; meaning the
333  *        offset is greater than the mapping entry's source offset plus
334  *        the entry's size. In this case, there is no overlap between
335  *        the offset and the mapping entry and 1 will be returned.
336  *
337  *        NOTE: If the offset is actually equal to the entry's offset
338  *        plus size, this is considered to be "greater" than the entry,
339  *        and this case applies (i.e. 1 will be returned). Thus, the
340  *        entry's "range" can be considered to be inclusive at its
341  *        start, but exclusive at its end: e.g. [src, src + size).
342  *
343  *     3. The last case to consider is if the offset actually falls
344  *        within the mapping entry's range. If this is the case, the
345  *        offset is considered to be "equal to" the mapping entry and
346  *        0 will be returned.
347  *
348  *        NOTE: If the offset is equal to the entry's source offset,
349  *        this case applies and 0 will be returned. If the offset is
350  *        equal to the entry's source plus its size, this case does
351  *        *not* apply (see "NOTE" above for scenario 2), and 1 will be
352  *        returned.
353  */
354 static int
355 dva_mapping_overlap_compare(const void *v_key, const void *v_array_elem)
356 {
357 	const uint64_t *key = v_key;
358 	const vdev_indirect_mapping_entry_phys_t *array_elem =
359 	    v_array_elem;
360 	uint64_t src_offset = DVA_MAPPING_GET_SRC_OFFSET(array_elem);
361 
362 	if (*key < src_offset) {
363 		return (-1);
364 	} else if (*key < src_offset + DVA_GET_ASIZE(&array_elem->vimep_dst)) {
365 		return (0);
366 	} else {
367 		return (1);
368 	}
369 }
370 
371 /*
372  * Return array entry.
373  */
374 static vdev_indirect_mapping_entry_phys_t *
375 vdev_indirect_mapping_entry(vdev_indirect_mapping_t *vim, uint64_t index)
376 {
377 	uint64_t size;
378 	off_t offset = 0;
379 	int rc;
380 
381 	if (vim->vim_phys->vimp_num_entries == 0)
382 		return (NULL);
383 
384 	if (vim->vim_entries == NULL) {
385 		uint64_t bsize;
386 
387 		bsize = vim->vim_dn->dn_datablkszsec << SPA_MINBLOCKSHIFT;
388 		size = vim->vim_phys->vimp_num_entries *
389 		    sizeof (*vim->vim_entries);
390 		if (size > bsize) {
391 			size = bsize / sizeof (*vim->vim_entries);
392 			size *= sizeof (*vim->vim_entries);
393 		}
394 		vim->vim_entries = malloc(size);
395 		if (vim->vim_entries == NULL)
396 			return (NULL);
397 		vim->vim_num_entries = size / sizeof (*vim->vim_entries);
398 		offset = index * sizeof (*vim->vim_entries);
399 	}
400 
401 	/* We have data in vim_entries */
402 	if (offset == 0) {
403 		if (index >= vim->vim_entry_offset &&
404 		    index <= vim->vim_entry_offset + vim->vim_num_entries) {
405 			index -= vim->vim_entry_offset;
406 			return (&vim->vim_entries[index]);
407 		}
408 		offset = index * sizeof (*vim->vim_entries);
409 	}
410 
411 	vim->vim_entry_offset = index;
412 	size = vim->vim_num_entries * sizeof (*vim->vim_entries);
413 	rc = dnode_read(vim->vim_spa, vim->vim_dn, offset, vim->vim_entries,
414 	    size);
415 	if (rc != 0) {
416 		/* Read error, invalidate vim_entries. */
417 		free(vim->vim_entries);
418 		vim->vim_entries = NULL;
419 		return (NULL);
420 	}
421 	index -= vim->vim_entry_offset;
422 	return (&vim->vim_entries[index]);
423 }
424 
425 /*
426  * Returns the mapping entry for the given offset.
427  *
428  * It's possible that the given offset will not be in the mapping table
429  * (i.e. no mapping entries contain this offset), in which case, the
430  * return value value depends on the "next_if_missing" parameter.
431  *
432  * If the offset is not found in the table and "next_if_missing" is
433  * B_FALSE, then NULL will always be returned. The behavior is intended
434  * to allow consumers to get the entry corresponding to the offset
435  * parameter, iff the offset overlaps with an entry in the table.
436  *
437  * If the offset is not found in the table and "next_if_missing" is
438  * B_TRUE, then the entry nearest to the given offset will be returned,
439  * such that the entry's source offset is greater than the offset
440  * passed in (i.e. the "next" mapping entry in the table is returned, if
441  * the offset is missing from the table). If there are no entries whose
442  * source offset is greater than the passed in offset, NULL is returned.
443  */
444 static vdev_indirect_mapping_entry_phys_t *
445 vdev_indirect_mapping_entry_for_offset(vdev_indirect_mapping_t *vim,
446     uint64_t offset)
447 {
448 	ASSERT(vim->vim_phys->vimp_num_entries > 0);
449 
450 	vdev_indirect_mapping_entry_phys_t *entry;
451 
452 	uint64_t last = vim->vim_phys->vimp_num_entries - 1;
453 	uint64_t base = 0;
454 
455 	/*
456 	 * We don't define these inside of the while loop because we use
457 	 * their value in the case that offset isn't in the mapping.
458 	 */
459 	uint64_t mid;
460 	int result;
461 
462 	while (last >= base) {
463 		mid = base + ((last - base) >> 1);
464 
465 		entry = vdev_indirect_mapping_entry(vim, mid);
466 		if (entry == NULL)
467 			break;
468 		result = dva_mapping_overlap_compare(&offset, entry);
469 
470 		if (result == 0) {
471 			break;
472 		} else if (result < 0) {
473 			last = mid - 1;
474 		} else {
475 			base = mid + 1;
476 		}
477 	}
478 	return (entry);
479 }
480 
481 /*
482  * Given an indirect vdev and an extent on that vdev, it duplicates the
483  * physical entries of the indirect mapping that correspond to the extent
484  * to a new array and returns a pointer to it. In addition, copied_entries
485  * is populated with the number of mapping entries that were duplicated.
486  *
487  * Finally, since we are doing an allocation, it is up to the caller to
488  * free the array allocated in this function.
489  */
490 vdev_indirect_mapping_entry_phys_t *
491 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *vd, uint64_t offset,
492     uint64_t asize, uint64_t *copied_entries)
493 {
494 	vdev_indirect_mapping_entry_phys_t *duplicate_mappings = NULL;
495 	vdev_indirect_mapping_t *vim = vd->v_mapping;
496 	uint64_t entries = 0;
497 
498 	vdev_indirect_mapping_entry_phys_t *first_mapping =
499 	    vdev_indirect_mapping_entry_for_offset(vim, offset);
500 	ASSERT3P(first_mapping, !=, NULL);
501 
502 	vdev_indirect_mapping_entry_phys_t *m = first_mapping;
503 	while (asize > 0) {
504 		uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
505 		uint64_t inner_offset = offset - DVA_MAPPING_GET_SRC_OFFSET(m);
506 		uint64_t inner_size = MIN(asize, size - inner_offset);
507 
508 		offset += inner_size;
509 		asize -= inner_size;
510 		entries++;
511 		m++;
512 	}
513 
514 	size_t copy_length = entries * sizeof (*first_mapping);
515 	duplicate_mappings = malloc(copy_length);
516 	if (duplicate_mappings != NULL)
517 		bcopy(first_mapping, duplicate_mappings, copy_length);
518 	else
519 		entries = 0;
520 
521 	*copied_entries = entries;
522 
523 	return (duplicate_mappings);
524 }
525 
526 static vdev_t *
527 vdev_lookup_top(spa_t *spa, uint64_t vdev)
528 {
529 	vdev_t *rvd;
530 	vdev_list_t *vlist;
531 
532 	vlist = &spa->spa_root_vdev->v_children;
533 	STAILQ_FOREACH(rvd, vlist, v_childlink)
534 		if (rvd->v_id == vdev)
535 			break;
536 
537 	return (rvd);
538 }
539 
540 /*
541  * This is a callback for vdev_indirect_remap() which allocates an
542  * indirect_split_t for each split segment and adds it to iv_splits.
543  */
544 static void
545 vdev_indirect_gather_splits(uint64_t split_offset, vdev_t *vd, uint64_t offset,
546     uint64_t size, void *arg)
547 {
548 	int n = 1;
549 	zio_t *zio = arg;
550 	indirect_vsd_t *iv = zio->io_vsd;
551 
552 	if (vd->v_read == vdev_indirect_read)
553 		return;
554 
555 	if (vd->v_read == vdev_mirror_read)
556 		n = vd->v_nchildren;
557 
558 	indirect_split_t *is =
559 	    malloc(offsetof(indirect_split_t, is_child[n]));
560 	if (is == NULL) {
561 		zio->io_error = ENOMEM;
562 		return;
563 	}
564 	bzero(is, offsetof(indirect_split_t, is_child[n]));
565 
566 	is->is_children = n;
567 	is->is_size = size;
568 	is->is_split_offset = split_offset;
569 	is->is_target_offset = offset;
570 	is->is_vdev = vd;
571 
572 	/*
573 	 * Note that we only consider multiple copies of the data for
574 	 * *mirror* vdevs.  We don't for "replacing" or "spare" vdevs, even
575 	 * though they use the same ops as mirror, because there's only one
576 	 * "good" copy under the replacing/spare.
577 	 */
578 	if (vd->v_read == vdev_mirror_read) {
579 		int i = 0;
580 		vdev_t *kid;
581 
582 		STAILQ_FOREACH(kid, &vd->v_children, v_childlink) {
583 			is->is_child[i++].ic_vdev = kid;
584 		}
585 	} else {
586 		is->is_child[0].ic_vdev = vd;
587 	}
588 
589 	list_insert_tail(&iv->iv_splits, is);
590 }
591 
592 static void
593 vdev_indirect_remap(vdev_t *vd, uint64_t offset, uint64_t asize, void *arg)
594 {
595 	list_t stack;
596 	spa_t *spa = vd->v_spa;
597 	zio_t *zio = arg;
598 	remap_segment_t *rs;
599 
600 	list_create(&stack, sizeof (remap_segment_t),
601 	    offsetof(remap_segment_t, rs_node));
602 
603 	rs = rs_alloc(vd, offset, asize, 0);
604 	if (rs == NULL) {
605 		printf("vdev_indirect_remap: out of memory.\n");
606 		zio->io_error = ENOMEM;
607 	}
608 	for (; rs != NULL; rs = list_remove_head(&stack)) {
609 		vdev_t *v = rs->rs_vd;
610 		uint64_t num_entries = 0;
611 		/* vdev_indirect_mapping_t *vim = v->v_mapping; */
612 		vdev_indirect_mapping_entry_phys_t *mapping =
613 		    vdev_indirect_mapping_duplicate_adjacent_entries(v,
614 		    rs->rs_offset, rs->rs_asize, &num_entries);
615 
616 		if (num_entries == 0)
617 			zio->io_error = ENOMEM;
618 
619 		for (uint64_t i = 0; i < num_entries; i++) {
620 			vdev_indirect_mapping_entry_phys_t *m = &mapping[i];
621 			uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
622 			uint64_t dst_offset = DVA_GET_OFFSET(&m->vimep_dst);
623 			uint64_t dst_vdev = DVA_GET_VDEV(&m->vimep_dst);
624 			uint64_t inner_offset = rs->rs_offset -
625 			    DVA_MAPPING_GET_SRC_OFFSET(m);
626 			uint64_t inner_size =
627 			    MIN(rs->rs_asize, size - inner_offset);
628 			vdev_t *dst_v = vdev_lookup_top(spa, dst_vdev);
629 
630 			if (dst_v->v_read == vdev_indirect_read) {
631 				remap_segment_t *o;
632 
633 				o = rs_alloc(dst_v, dst_offset + inner_offset,
634 				    inner_size, rs->rs_split_offset);
635 				if (o == NULL) {
636 					printf("vdev_indirect_remap: "
637 					    "out of memory.\n");
638 					zio->io_error = ENOMEM;
639 					break;
640 				}
641 
642 				list_insert_head(&stack, o);
643 			}
644 			vdev_indirect_gather_splits(rs->rs_split_offset, dst_v,
645 			    dst_offset + inner_offset,
646 			    inner_size, arg);
647 
648 			/*
649 			 * vdev_indirect_gather_splits can have memory
650 			 * allocation error, we can not recover from it.
651 			 */
652 			if (zio->io_error != 0)
653 				break;
654 			rs->rs_offset += inner_size;
655 			rs->rs_asize -= inner_size;
656 			rs->rs_split_offset += inner_size;
657 		}
658 
659 		free(mapping);
660 		free(rs);
661 		if (zio->io_error != 0)
662 			break;
663 	}
664 
665 	list_destroy(&stack);
666 }
667 
668 static void
669 vdev_indirect_map_free(zio_t *zio)
670 {
671 	indirect_vsd_t *iv = zio->io_vsd;
672 	indirect_split_t *is;
673 
674 	while ((is = list_head(&iv->iv_splits)) != NULL) {
675 		for (int c = 0; c < is->is_children; c++) {
676 			indirect_child_t *ic = &is->is_child[c];
677 			free(ic->ic_data);
678 		}
679 		list_remove(&iv->iv_splits, is);
680 		free(is);
681 	}
682 	free(iv);
683 }
684 
685 static int
686 vdev_indirect_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
687     off_t offset, size_t bytes)
688 {
689 	zio_t zio;
690 	spa_t *spa = vdev->v_spa;
691 	indirect_vsd_t *iv;
692 	indirect_split_t *first;
693 	int rc = EIO;
694 
695 	iv = calloc(1, sizeof(*iv));
696 	if (iv == NULL)
697 		return (ENOMEM);
698 
699 	list_create(&iv->iv_splits,
700 	    sizeof (indirect_split_t), offsetof(indirect_split_t, is_node));
701 
702 	bzero(&zio, sizeof(zio));
703 	zio.io_spa = spa;
704 	zio.io_bp = (blkptr_t *)bp;
705 	zio.io_data = buf;
706 	zio.io_size = bytes;
707 	zio.io_offset = offset;
708 	zio.io_vd = vdev;
709 	zio.io_vsd = iv;
710 
711 	if (vdev->v_mapping == NULL) {
712 		vdev_indirect_config_t *vic;
713 
714 		vic = &vdev->vdev_indirect_config;
715 		vdev->v_mapping = vdev_indirect_mapping_open(spa,
716 		    spa->spa_mos, vic->vic_mapping_object);
717 	}
718 
719 	vdev_indirect_remap(vdev, offset, bytes, &zio);
720 	if (zio.io_error != 0)
721 		return (zio.io_error);
722 
723 	first = list_head(&iv->iv_splits);
724 	if (first->is_size == zio.io_size) {
725 		/*
726 		 * This is not a split block; we are pointing to the entire
727 		 * data, which will checksum the same as the original data.
728 		 * Pass the BP down so that the child i/o can verify the
729 		 * checksum, and try a different location if available
730 		 * (e.g. on a mirror).
731 		 *
732 		 * While this special case could be handled the same as the
733 		 * general (split block) case, doing it this way ensures
734 		 * that the vast majority of blocks on indirect vdevs
735 		 * (which are not split) are handled identically to blocks
736 		 * on non-indirect vdevs.  This allows us to be less strict
737 		 * about performance in the general (but rare) case.
738 		 */
739 		rc = first->is_vdev->v_read(first->is_vdev, zio.io_bp,
740 		    zio.io_data, first->is_target_offset, bytes);
741 	} else {
742 		iv->iv_split_block = B_TRUE;
743 		/*
744 		 * Read one copy of each split segment, from the
745 		 * top-level vdev.  Since we don't know the
746 		 * checksum of each split individually, the child
747 		 * zio can't ensure that we get the right data.
748 		 * E.g. if it's a mirror, it will just read from a
749 		 * random (healthy) leaf vdev.  We have to verify
750 		 * the checksum in vdev_indirect_io_done().
751 		 */
752 		for (indirect_split_t *is = list_head(&iv->iv_splits);
753 		    is != NULL; is = list_next(&iv->iv_splits, is)) {
754 			char *ptr = zio.io_data;
755 
756 			rc = is->is_vdev->v_read(is->is_vdev, zio.io_bp,
757 			    ptr + is->is_split_offset, is->is_target_offset,
758 			    is->is_size);
759 		}
760 		if (zio_checksum_verify(spa, zio.io_bp, zio.io_data))
761 			rc = ECKSUM;
762 		else
763 			rc = 0;
764 	}
765 
766 	vdev_indirect_map_free(&zio);
767 	if (rc == 0)
768 		rc = zio.io_error;
769 
770 	return (rc);
771 }
772 
773 static int
774 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
775     off_t offset, size_t bytes)
776 {
777 
778 	return (vdev_read_phys(vdev, bp, buf,
779 	    offset + VDEV_LABEL_START_SIZE, bytes));
780 }
781 
782 static int
783 vdev_missing_read(vdev_t *vdev __unused, const blkptr_t *bp __unused,
784     void *buf __unused, off_t offset __unused, size_t bytes __unused)
785 {
786 
787 	return (ENOTSUP);
788 }
789 
790 static int
791 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
792     off_t offset, size_t bytes)
793 {
794 	vdev_t *kid;
795 	int rc;
796 
797 	rc = EIO;
798 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
799 		if (kid->v_state != VDEV_STATE_HEALTHY)
800 			continue;
801 		rc = kid->v_read(kid, bp, buf, offset, bytes);
802 		if (!rc)
803 			return (0);
804 	}
805 
806 	return (rc);
807 }
808 
809 static int
810 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
811     off_t offset, size_t bytes)
812 {
813 	vdev_t *kid;
814 
815 	/*
816 	 * Here we should have two kids:
817 	 * First one which is the one we are replacing and we can trust
818 	 * only this one to have valid data, but it might not be present.
819 	 * Second one is that one we are replacing with. It is most likely
820 	 * healthy, but we can't trust it has needed data, so we won't use it.
821 	 */
822 	kid = STAILQ_FIRST(&vdev->v_children);
823 	if (kid == NULL)
824 		return (EIO);
825 	if (kid->v_state != VDEV_STATE_HEALTHY)
826 		return (EIO);
827 	return (kid->v_read(kid, bp, buf, offset, bytes));
828 }
829 
830 static vdev_t *
831 vdev_find(uint64_t guid)
832 {
833 	vdev_t *vdev;
834 
835 	STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
836 		if (vdev->v_guid == guid)
837 			return (vdev);
838 
839 	return (0);
840 }
841 
842 static vdev_t *
843 vdev_create(uint64_t guid, vdev_read_t *_read)
844 {
845 	vdev_t *vdev;
846 	vdev_indirect_config_t *vic;
847 
848 	vdev = calloc(1, sizeof(vdev_t));
849 	if (vdev != NULL) {
850 		STAILQ_INIT(&vdev->v_children);
851 		vdev->v_guid = guid;
852 		vdev->v_read = _read;
853 
854 		/*
855 		 * root vdev has no read function, we use this fact to
856 		 * skip setting up data we do not need for root vdev.
857 		 * We only point root vdev from spa.
858 		 */
859 		if (_read != NULL) {
860 			vic = &vdev->vdev_indirect_config;
861 			vic->vic_prev_indirect_vdev = UINT64_MAX;
862 			STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
863 		}
864 	}
865 
866 	return (vdev);
867 }
868 
869 static void
870 vdev_set_initial_state(vdev_t *vdev, const nvlist_t *nvlist)
871 {
872 	uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
873 	uint64_t is_log;
874 
875 	is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
876 	is_log = 0;
877 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
878 	    &is_offline, NULL);
879 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
880 	    &is_removed, NULL);
881 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
882 	    &is_faulted, NULL);
883 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64,
884 	    NULL, &is_degraded, NULL);
885 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64,
886 	    NULL, &isnt_present, NULL);
887 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, NULL,
888 	    &is_log, NULL);
889 
890 	if (is_offline != 0)
891 		vdev->v_state = VDEV_STATE_OFFLINE;
892 	else if (is_removed != 0)
893 		vdev->v_state = VDEV_STATE_REMOVED;
894 	else if (is_faulted != 0)
895 		vdev->v_state = VDEV_STATE_FAULTED;
896 	else if (is_degraded != 0)
897 		vdev->v_state = VDEV_STATE_DEGRADED;
898 	else if (isnt_present != 0)
899 		vdev->v_state = VDEV_STATE_CANT_OPEN;
900 
901 	vdev->v_islog = is_log != 0;
902 }
903 
904 static int
905 vdev_init(uint64_t guid, const nvlist_t *nvlist, vdev_t **vdevp)
906 {
907 	uint64_t id, ashift, asize, nparity;
908 	const char *path;
909 	const char *type;
910 	int len, pathlen;
911 	char *name;
912 	vdev_t *vdev;
913 
914 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id,
915 	    NULL) ||
916 	    nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING, NULL,
917 	    &type, &len)) {
918 		return (ENOENT);
919 	}
920 
921 	if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 &&
922 	    memcmp(type, VDEV_TYPE_DISK, len) != 0 &&
923 #ifdef ZFS_TEST
924 	    memcmp(type, VDEV_TYPE_FILE, len) != 0 &&
925 #endif
926 	    memcmp(type, VDEV_TYPE_RAIDZ, len) != 0 &&
927 	    memcmp(type, VDEV_TYPE_INDIRECT, len) != 0 &&
928 	    memcmp(type, VDEV_TYPE_REPLACING, len) != 0 &&
929 	    memcmp(type, VDEV_TYPE_HOLE, len) != 0) {
930 		printf("ZFS: can only boot from disk, mirror, raidz1, "
931 		    "raidz2 and raidz3 vdevs, got: %.*s\n", len, type);
932 		return (EIO);
933 	}
934 
935 	if (memcmp(type, VDEV_TYPE_MIRROR, len) == 0)
936 		vdev = vdev_create(guid, vdev_mirror_read);
937 	else if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0)
938 		vdev = vdev_create(guid, vdev_raidz_read);
939 	else if (memcmp(type, VDEV_TYPE_REPLACING, len) == 0)
940 		vdev = vdev_create(guid, vdev_replacing_read);
941 	else if (memcmp(type, VDEV_TYPE_INDIRECT, len) == 0) {
942 		vdev_indirect_config_t *vic;
943 
944 		vdev = vdev_create(guid, vdev_indirect_read);
945 		if (vdev != NULL) {
946 			vdev->v_state = VDEV_STATE_HEALTHY;
947 			vic = &vdev->vdev_indirect_config;
948 
949 			nvlist_find(nvlist,
950 			    ZPOOL_CONFIG_INDIRECT_OBJECT,
951 			    DATA_TYPE_UINT64,
952 			    NULL, &vic->vic_mapping_object, NULL);
953 			nvlist_find(nvlist,
954 			    ZPOOL_CONFIG_INDIRECT_BIRTHS,
955 			    DATA_TYPE_UINT64,
956 			    NULL, &vic->vic_births_object, NULL);
957 			nvlist_find(nvlist,
958 			    ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
959 			    DATA_TYPE_UINT64,
960 			    NULL, &vic->vic_prev_indirect_vdev, NULL);
961 		}
962 	} else if (memcmp(type, VDEV_TYPE_HOLE, len) == 0) {
963 		vdev = vdev_create(guid, vdev_missing_read);
964 	} else {
965 		vdev = vdev_create(guid, vdev_disk_read);
966 	}
967 
968 	if (vdev == NULL)
969 		return (ENOMEM);
970 
971 	vdev_set_initial_state(vdev, nvlist);
972 	vdev->v_id = id;
973 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
974 	    DATA_TYPE_UINT64, NULL, &ashift, NULL) == 0)
975 		vdev->v_ashift = ashift;
976 
977 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE,
978 	    DATA_TYPE_UINT64, NULL, &asize, NULL) == 0) {
979 		vdev->v_psize = asize +
980 		    VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
981 	}
982 
983 	if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
984 	    DATA_TYPE_UINT64, NULL, &nparity, NULL) == 0)
985 		vdev->v_nparity = nparity;
986 
987 	if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
988 	    DATA_TYPE_STRING, NULL, &path, &pathlen) == 0) {
989 		char prefix[] = "/dev/";
990 
991 		len = strlen(prefix);
992 		if (len < pathlen && memcmp(path, prefix, len) == 0) {
993 			path += len;
994 			pathlen -= len;
995 		}
996 		name = malloc(pathlen + 1);
997 		bcopy(path, name, pathlen);
998 		name[pathlen] = '\0';
999 		vdev->v_name = name;
1000 	} else {
1001 		name = NULL;
1002 		if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) {
1003 			if (vdev->v_nparity < 1 ||
1004 			    vdev->v_nparity > 3) {
1005 				printf("ZFS: invalid raidz parity: %d\n",
1006 				    vdev->v_nparity);
1007 				return (EIO);
1008 			}
1009 			(void) asprintf(&name, "%.*s%d-%" PRIu64, len, type,
1010 			    vdev->v_nparity, id);
1011 		} else {
1012 			(void) asprintf(&name, "%.*s-%" PRIu64, len, type, id);
1013 		}
1014 		vdev->v_name = name;
1015 	}
1016 	*vdevp = vdev;
1017 	return (0);
1018 }
1019 
1020 /*
1021  * Find slot for vdev. We return either NULL to signal to use
1022  * STAILQ_INSERT_HEAD, or we return link element to be used with
1023  * STAILQ_INSERT_AFTER.
1024  */
1025 static vdev_t *
1026 vdev_find_previous(vdev_t *top_vdev, vdev_t *vdev)
1027 {
1028 	vdev_t *v, *previous;
1029 
1030 	if (STAILQ_EMPTY(&top_vdev->v_children))
1031 		return (NULL);
1032 
1033 	previous = NULL;
1034 	STAILQ_FOREACH(v, &top_vdev->v_children, v_childlink) {
1035 		if (v->v_id > vdev->v_id)
1036 			return (previous);
1037 
1038 		if (v->v_id == vdev->v_id)
1039 			return (v);
1040 
1041 		if (v->v_id < vdev->v_id)
1042 			previous = v;
1043 	}
1044 	return (previous);
1045 }
1046 
1047 static size_t
1048 vdev_child_count(vdev_t *vdev)
1049 {
1050 	vdev_t *v;
1051 	size_t count;
1052 
1053 	count = 0;
1054 	STAILQ_FOREACH(v, &vdev->v_children, v_childlink) {
1055 		count++;
1056 	}
1057 	return (count);
1058 }
1059 
1060 /*
1061  * Insert vdev into top_vdev children list. List is ordered by v_id.
1062  */
1063 static void
1064 vdev_insert(vdev_t *top_vdev, vdev_t *vdev)
1065 {
1066 	vdev_t *previous;
1067 	size_t count;
1068 
1069 	/*
1070 	 * The top level vdev can appear in random order, depending how
1071 	 * the firmware is presenting the disk devices.
1072 	 * However, we will insert vdev to create list ordered by v_id,
1073 	 * so we can use either STAILQ_INSERT_HEAD or STAILQ_INSERT_AFTER
1074 	 * as STAILQ does not have insert before.
1075 	 */
1076 	previous = vdev_find_previous(top_vdev, vdev);
1077 
1078 	if (previous == NULL) {
1079 		STAILQ_INSERT_HEAD(&top_vdev->v_children, vdev, v_childlink);
1080 	} else if (previous->v_id == vdev->v_id) {
1081 		/*
1082 		 * This vdev was configured from label config,
1083 		 * do not insert duplicate.
1084 		 */
1085 		return;
1086 	} else {
1087 		STAILQ_INSERT_AFTER(&top_vdev->v_children, previous, vdev,
1088 		    v_childlink);
1089 	}
1090 
1091 	count = vdev_child_count(top_vdev);
1092 	if (top_vdev->v_nchildren < count)
1093 		top_vdev->v_nchildren = count;
1094 }
1095 
1096 static int
1097 vdev_from_nvlist(spa_t *spa, uint64_t top_guid, const nvlist_t *nvlist)
1098 {
1099 	vdev_t *top_vdev, *vdev;
1100 	nvlist_t **kids = NULL;
1101 	int rc, nkids;
1102 
1103 	/* Get top vdev. */
1104 	top_vdev = vdev_find(top_guid);
1105 	if (top_vdev == NULL) {
1106 		rc = vdev_init(top_guid, nvlist, &top_vdev);
1107 		if (rc != 0)
1108 			return (rc);
1109 		top_vdev->v_spa = spa;
1110 		top_vdev->v_top = top_vdev;
1111 		vdev_insert(spa->spa_root_vdev, top_vdev);
1112 	}
1113 
1114 	/* Add children if there are any. */
1115 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1116 	    &nkids, &kids, NULL);
1117 	if (rc == 0) {
1118 		for (int i = 0; i < nkids; i++) {
1119 			uint64_t guid;
1120 
1121 			rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID,
1122 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1123 			if (rc != 0)
1124 				goto done;
1125 
1126 			rc = vdev_init(guid, kids[i], &vdev);
1127 			if (rc != 0)
1128 				goto done;
1129 
1130 			vdev->v_spa = spa;
1131 			vdev->v_top = top_vdev;
1132 			vdev_insert(top_vdev, vdev);
1133 		}
1134 	} else {
1135 		/*
1136 		 * When there are no children, nvlist_find() does return
1137 		 * error, reset it because leaf devices have no children.
1138 		 */
1139 		rc = 0;
1140 	}
1141 done:
1142 	if (kids != NULL) {
1143 		for (int i = 0; i < nkids; i++)
1144 			nvlist_destroy(kids[i]);
1145 		free(kids);
1146 	}
1147 
1148 	return (rc);
1149 }
1150 
1151 static int
1152 vdev_init_from_label(spa_t *spa, const nvlist_t *nvlist)
1153 {
1154 	uint64_t pool_guid, top_guid;
1155 	nvlist_t *vdevs;
1156 	int rc;
1157 
1158 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1159 	    NULL, &pool_guid, NULL) ||
1160 	    nvlist_find(nvlist, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64,
1161 	    NULL, &top_guid, NULL) ||
1162 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1163 	    NULL, &vdevs, NULL)) {
1164 		printf("ZFS: can't find vdev details\n");
1165 		return (ENOENT);
1166 	}
1167 
1168 	rc = vdev_from_nvlist(spa, top_guid, vdevs);
1169 	nvlist_destroy(vdevs);
1170 	return (rc);
1171 }
1172 
1173 static void
1174 vdev_set_state(vdev_t *vdev)
1175 {
1176 	vdev_t *kid;
1177 	int good_kids;
1178 	int bad_kids;
1179 
1180 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1181 		vdev_set_state(kid);
1182 	}
1183 
1184 	/*
1185 	 * A mirror or raidz is healthy if all its kids are healthy. A
1186 	 * mirror is degraded if any of its kids is healthy; a raidz
1187 	 * is degraded if at most nparity kids are offline.
1188 	 */
1189 	if (STAILQ_FIRST(&vdev->v_children)) {
1190 		good_kids = 0;
1191 		bad_kids = 0;
1192 		STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1193 			if (kid->v_state == VDEV_STATE_HEALTHY)
1194 				good_kids++;
1195 			else
1196 				bad_kids++;
1197 		}
1198 		if (bad_kids == 0) {
1199 			vdev->v_state = VDEV_STATE_HEALTHY;
1200 		} else {
1201 			if (vdev->v_read == vdev_mirror_read) {
1202 				if (good_kids) {
1203 					vdev->v_state = VDEV_STATE_DEGRADED;
1204 				} else {
1205 					vdev->v_state = VDEV_STATE_OFFLINE;
1206 				}
1207 			} else if (vdev->v_read == vdev_raidz_read) {
1208 				if (bad_kids > vdev->v_nparity) {
1209 					vdev->v_state = VDEV_STATE_OFFLINE;
1210 				} else {
1211 					vdev->v_state = VDEV_STATE_DEGRADED;
1212 				}
1213 			}
1214 		}
1215 	}
1216 }
1217 
1218 static int
1219 vdev_update_from_nvlist(uint64_t top_guid, const nvlist_t *nvlist)
1220 {
1221 	vdev_t *vdev;
1222 	nvlist_t **kids = NULL;
1223 	int rc, nkids;
1224 
1225 	/* Update top vdev. */
1226 	vdev = vdev_find(top_guid);
1227 	if (vdev != NULL)
1228 		vdev_set_initial_state(vdev, nvlist);
1229 
1230 	/* Update children if there are any. */
1231 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1232 	    &nkids, &kids, NULL);
1233 	if (rc == 0) {
1234 		for (int i = 0; i < nkids; i++) {
1235 			uint64_t guid;
1236 
1237 			rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID,
1238 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1239 			if (rc != 0)
1240 				break;
1241 
1242 			vdev = vdev_find(guid);
1243 			if (vdev != NULL)
1244 				vdev_set_initial_state(vdev, kids[i]);
1245 		}
1246 	} else {
1247 		rc = 0;
1248 	}
1249 	if (kids != NULL) {
1250 		for (int i = 0; i < nkids; i++)
1251 			nvlist_destroy(kids[i]);
1252 		free(kids);
1253 	}
1254 
1255 	return (rc);
1256 }
1257 
1258 static int
1259 vdev_init_from_nvlist(spa_t *spa, const nvlist_t *nvlist)
1260 {
1261 	uint64_t pool_guid, vdev_children;
1262 	nvlist_t *vdevs = NULL, **kids = NULL;
1263 	int rc, nkids;
1264 
1265 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1266 	    NULL, &pool_guid, NULL) ||
1267 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_CHILDREN, DATA_TYPE_UINT64,
1268 	    NULL, &vdev_children, NULL) ||
1269 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1270 	    NULL, &vdevs, NULL)) {
1271 		printf("ZFS: can't find vdev details\n");
1272 		return (ENOENT);
1273 	}
1274 
1275 	/* Wrong guid?! */
1276 	if (spa->spa_guid != pool_guid) {
1277 		nvlist_destroy(vdevs);
1278 		return (EINVAL);
1279 	}
1280 
1281 	spa->spa_root_vdev->v_nchildren = vdev_children;
1282 
1283 	rc = nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1284 	    &nkids, &kids, NULL);
1285 	nvlist_destroy(vdevs);
1286 
1287 	/*
1288 	 * MOS config has at least one child for root vdev.
1289 	 */
1290 	if (rc != 0)
1291 		return (rc);
1292 
1293 	for (int i = 0; i < nkids; i++) {
1294 		uint64_t guid;
1295 		vdev_t *vdev;
1296 
1297 		rc = nvlist_find(kids[i], ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1298 		    NULL, &guid, NULL);
1299 		if (rc != 0)
1300 			break;
1301 		vdev = vdev_find(guid);
1302 		/*
1303 		 * Top level vdev is missing, create it.
1304 		 */
1305 		if (vdev == NULL)
1306 			rc = vdev_from_nvlist(spa, guid, kids[i]);
1307 		else
1308 			rc = vdev_update_from_nvlist(guid, kids[i]);
1309 		if (rc != 0)
1310 			break;
1311 	}
1312 	if (kids != NULL) {
1313 		for (int i = 0; i < nkids; i++)
1314 			nvlist_destroy(kids[i]);
1315 		free(kids);
1316 	}
1317 
1318 	/*
1319 	 * Re-evaluate top-level vdev state.
1320 	 */
1321 	vdev_set_state(spa->spa_root_vdev);
1322 
1323 	return (rc);
1324 }
1325 
1326 static spa_t *
1327 spa_find_by_guid(uint64_t guid)
1328 {
1329 	spa_t *spa;
1330 
1331 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1332 		if (spa->spa_guid == guid)
1333 			return (spa);
1334 
1335 	return (NULL);
1336 }
1337 
1338 static spa_t *
1339 spa_find_by_name(const char *name)
1340 {
1341 	spa_t *spa;
1342 
1343 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1344 		if (strcmp(spa->spa_name, name) == 0)
1345 			return (spa);
1346 
1347 	return (NULL);
1348 }
1349 
1350 static spa_t *
1351 spa_find_by_dev(struct zfs_devdesc *dev)
1352 {
1353 
1354 	if (dev->dd.d_dev->dv_type != DEVT_ZFS)
1355 		return (NULL);
1356 
1357 	if (dev->pool_guid == 0)
1358 		return (STAILQ_FIRST(&zfs_pools));
1359 
1360 	return (spa_find_by_guid(dev->pool_guid));
1361 }
1362 
1363 static spa_t *
1364 spa_create(uint64_t guid, const char *name)
1365 {
1366 	spa_t *spa;
1367 
1368 	if ((spa = calloc(1, sizeof(spa_t))) == NULL)
1369 		return (NULL);
1370 	if ((spa->spa_name = strdup(name)) == NULL) {
1371 		free(spa);
1372 		return (NULL);
1373 	}
1374 	spa->spa_uberblock = &spa->spa_uberblock_master;
1375 	spa->spa_mos = &spa->spa_mos_master;
1376 	spa->spa_guid = guid;
1377 	spa->spa_root_vdev = vdev_create(guid, NULL);
1378 	if (spa->spa_root_vdev == NULL) {
1379 		free(spa->spa_name);
1380 		free(spa);
1381 		return (NULL);
1382 	}
1383 	spa->spa_root_vdev->v_name = strdup("root");
1384 	STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
1385 
1386 	return (spa);
1387 }
1388 
1389 static const char *
1390 state_name(vdev_state_t state)
1391 {
1392 	static const char *names[] = {
1393 		"UNKNOWN",
1394 		"CLOSED",
1395 		"OFFLINE",
1396 		"REMOVED",
1397 		"CANT_OPEN",
1398 		"FAULTED",
1399 		"DEGRADED",
1400 		"ONLINE"
1401 	};
1402 	return (names[state]);
1403 }
1404 
1405 #ifdef BOOT2
1406 
1407 #define pager_printf printf
1408 
1409 #else
1410 
1411 static int
1412 pager_printf(const char *fmt, ...)
1413 {
1414 	char line[80];
1415 	va_list args;
1416 
1417 	va_start(args, fmt);
1418 	vsnprintf(line, sizeof(line), fmt, args);
1419 	va_end(args);
1420 	return (pager_output(line));
1421 }
1422 
1423 #endif
1424 
1425 #define	STATUS_FORMAT	"        %s %s\n"
1426 
1427 static int
1428 print_state(int indent, const char *name, vdev_state_t state)
1429 {
1430 	int i;
1431 	char buf[512];
1432 
1433 	buf[0] = 0;
1434 	for (i = 0; i < indent; i++)
1435 		strcat(buf, "  ");
1436 	strcat(buf, name);
1437 	return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
1438 }
1439 
1440 static int
1441 vdev_status(vdev_t *vdev, int indent)
1442 {
1443 	vdev_t *kid;
1444 	int ret;
1445 
1446 	if (vdev->v_islog) {
1447 		(void) pager_output("        logs\n");
1448 		indent++;
1449 	}
1450 
1451 	ret = print_state(indent, vdev->v_name, vdev->v_state);
1452 	if (ret != 0)
1453 		return (ret);
1454 
1455 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1456 		ret = vdev_status(kid, indent + 1);
1457 		if (ret != 0)
1458 			return (ret);
1459 	}
1460 	return (ret);
1461 }
1462 
1463 static int
1464 spa_status(spa_t *spa)
1465 {
1466 	static char bootfs[ZFS_MAXNAMELEN];
1467 	uint64_t rootid;
1468 	vdev_list_t *vlist;
1469 	vdev_t *vdev;
1470 	int good_kids, bad_kids, degraded_kids, ret;
1471 	vdev_state_t state;
1472 
1473 	ret = pager_printf("  pool: %s\n", spa->spa_name);
1474 	if (ret != 0)
1475 		return (ret);
1476 
1477 	if (zfs_get_root(spa, &rootid) == 0 &&
1478 	    zfs_rlookup(spa, rootid, bootfs) == 0) {
1479 		if (bootfs[0] == '\0')
1480 			ret = pager_printf("bootfs: %s\n", spa->spa_name);
1481 		else
1482 			ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
1483 			    bootfs);
1484 		if (ret != 0)
1485 			return (ret);
1486 	}
1487 	ret = pager_printf("config:\n\n");
1488 	if (ret != 0)
1489 		return (ret);
1490 	ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
1491 	if (ret != 0)
1492 		return (ret);
1493 
1494 	good_kids = 0;
1495 	degraded_kids = 0;
1496 	bad_kids = 0;
1497 	vlist = &spa->spa_root_vdev->v_children;
1498 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1499 		if (vdev->v_state == VDEV_STATE_HEALTHY)
1500 			good_kids++;
1501 		else if (vdev->v_state == VDEV_STATE_DEGRADED)
1502 			degraded_kids++;
1503 		else
1504 			bad_kids++;
1505 	}
1506 
1507 	state = VDEV_STATE_CLOSED;
1508 	if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
1509 		state = VDEV_STATE_HEALTHY;
1510 	else if ((good_kids + degraded_kids) > 0)
1511 		state = VDEV_STATE_DEGRADED;
1512 
1513 	ret = print_state(0, spa->spa_name, state);
1514 	if (ret != 0)
1515 		return (ret);
1516 
1517 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1518 		ret = vdev_status(vdev, 1);
1519 		if (ret != 0)
1520 			return (ret);
1521 	}
1522 	return (ret);
1523 }
1524 
1525 static int
1526 spa_all_status(void)
1527 {
1528 	spa_t *spa;
1529 	int first = 1, ret = 0;
1530 
1531 	STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
1532 		if (!first) {
1533 			ret = pager_printf("\n");
1534 			if (ret != 0)
1535 				return (ret);
1536 		}
1537 		first = 0;
1538 		ret = spa_status(spa);
1539 		if (ret != 0)
1540 			return (ret);
1541 	}
1542 	return (ret);
1543 }
1544 
1545 static uint64_t
1546 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
1547 {
1548 	uint64_t label_offset;
1549 
1550 	if (l < VDEV_LABELS / 2)
1551 		label_offset = 0;
1552 	else
1553 		label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
1554 
1555 	return (offset + l * sizeof (vdev_label_t) + label_offset);
1556 }
1557 
1558 static int
1559 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1560 {
1561 	unsigned int seq1 = 0;
1562 	unsigned int seq2 = 0;
1563 	int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
1564 
1565 	if (cmp != 0)
1566 		return (cmp);
1567 
1568 	cmp = AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1569 	if (cmp != 0)
1570 		return (cmp);
1571 
1572 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1573 		seq1 = MMP_SEQ(ub1);
1574 
1575 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1576 		seq2 = MMP_SEQ(ub2);
1577 
1578 	return (AVL_CMP(seq1, seq2));
1579 }
1580 
1581 static int
1582 uberblock_verify(uberblock_t *ub)
1583 {
1584 	if (ub->ub_magic == BSWAP_64((uint64_t)UBERBLOCK_MAGIC)) {
1585 		byteswap_uint64_array(ub, sizeof (uberblock_t));
1586 	}
1587 
1588 	if (ub->ub_magic != UBERBLOCK_MAGIC ||
1589 	    !SPA_VERSION_IS_SUPPORTED(ub->ub_version))
1590 		return (EINVAL);
1591 
1592 	return (0);
1593 }
1594 
1595 static int
1596 vdev_label_read(vdev_t *vd, int l, void *buf, uint64_t offset,
1597     size_t size)
1598 {
1599 	blkptr_t bp;
1600 	off_t off;
1601 
1602 	off = vdev_label_offset(vd->v_psize, l, offset);
1603 
1604 	BP_ZERO(&bp);
1605 	BP_SET_LSIZE(&bp, size);
1606 	BP_SET_PSIZE(&bp, size);
1607 	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1608 	BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1609 	DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
1610 	ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1611 
1612 	return (vdev_read_phys(vd, &bp, buf, off, size));
1613 }
1614 
1615 /*
1616  * We do need to be sure we write to correct location.
1617  * Our vdev label does consist of 4 fields:
1618  * pad1 (8k), reserved.
1619  * bootenv (8k), checksummed, previously reserved, may contian garbage.
1620  * vdev_phys (112k), checksummed
1621  * uberblock ring (128k), checksummed.
1622  *
1623  * Since bootenv area may contain garbage, we can not reliably read it, as
1624  * we can get checksum errors.
1625  * Next best thing is vdev_phys - it is just after bootenv. It still may
1626  * be corrupted, but in such case we will miss this one write.
1627  */
1628 static int
1629 vdev_label_write_validate(vdev_t *vd, int l, uint64_t offset)
1630 {
1631 	uint64_t off, o_phys;
1632 	void *buf;
1633 	size_t size = VDEV_PHYS_SIZE;
1634 	int rc;
1635 
1636 	o_phys = offsetof(vdev_label_t, vl_vdev_phys);
1637 	off = vdev_label_offset(vd->v_psize, l, o_phys);
1638 
1639 	/* off should be 8K from bootenv */
1640 	if (vdev_label_offset(vd->v_psize, l, offset) + VDEV_PAD_SIZE != off)
1641 		return (EINVAL);
1642 
1643 	buf = malloc(size);
1644 	if (buf == NULL)
1645 		return (ENOMEM);
1646 
1647 	/* Read vdev_phys */
1648 	rc = vdev_label_read(vd, l, buf, o_phys, size);
1649 	free(buf);
1650 	return (rc);
1651 }
1652 
1653 static int
1654 vdev_label_write(vdev_t *vd, int l, vdev_boot_envblock_t *be, uint64_t offset)
1655 {
1656 	zio_checksum_info_t *ci;
1657 	zio_cksum_t cksum;
1658 	off_t off;
1659 	size_t size = VDEV_PAD_SIZE;
1660 	int rc;
1661 
1662 	if (vd->v_phys_write == NULL)
1663 		return (ENOTSUP);
1664 
1665 	off = vdev_label_offset(vd->v_psize, l, offset);
1666 
1667 	rc = vdev_label_write_validate(vd, l, offset);
1668 	if (rc != 0) {
1669 		return (rc);
1670 	}
1671 
1672 	ci = &zio_checksum_table[ZIO_CHECKSUM_LABEL];
1673 	be->vbe_zbt.zec_magic = ZEC_MAGIC;
1674 	zio_checksum_label_verifier(&be->vbe_zbt.zec_cksum, off);
1675 	ci->ci_func[0](be, size, NULL, &cksum);
1676 	be->vbe_zbt.zec_cksum = cksum;
1677 
1678 	return (vdev_write_phys(vd, be, off, size));
1679 }
1680 
1681 static int
1682 vdev_write_bootenv_impl(vdev_t *vdev, vdev_boot_envblock_t *be)
1683 {
1684 	vdev_t *kid;
1685 	int rv = 0, rc;
1686 
1687 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1688 		if (kid->v_state != VDEV_STATE_HEALTHY)
1689 			continue;
1690 		rc = vdev_write_bootenv_impl(kid, be);
1691 		if (rv == 0)
1692 			rv = rc;
1693 	}
1694 
1695 	/*
1696 	 * Non-leaf vdevs do not have v_phys_write.
1697 	 */
1698 	if (vdev->v_phys_write == NULL)
1699 		return (rv);
1700 
1701 	for (int l = 0; l < VDEV_LABELS; l++) {
1702 		rc = vdev_label_write(vdev, l, be,
1703 		    offsetof(vdev_label_t, vl_be));
1704 		if (rc != 0) {
1705 			printf("failed to write bootenv to %s label %d: %d\n",
1706 			    vdev->v_name ? vdev->v_name : "unknown", l, rc);
1707 			rv = rc;
1708 		}
1709 	}
1710 	return (rv);
1711 }
1712 
1713 int
1714 vdev_write_bootenv(vdev_t *vdev, nvlist_t *nvl)
1715 {
1716 	vdev_boot_envblock_t *be;
1717 	nvlist_t nv, *nvp;
1718 	uint64_t version;
1719 	int rv;
1720 
1721 	if (nvl->nv_size > sizeof(be->vbe_bootenv))
1722 		return (E2BIG);
1723 
1724 	version = VB_RAW;
1725 	nvp = vdev_read_bootenv(vdev);
1726 	if (nvp != NULL) {
1727 		nvlist_find(nvp, BOOTENV_VERSION, DATA_TYPE_UINT64, NULL,
1728 		    &version, NULL);
1729 		nvlist_destroy(nvp);
1730 	}
1731 
1732 	be = calloc(1, sizeof(*be));
1733 	if (be == NULL)
1734 		return (ENOMEM);
1735 
1736 	be->vbe_version = version;
1737 	switch (version) {
1738 	case VB_RAW:
1739 		/*
1740 		 * If there is no envmap, we will just wipe bootenv.
1741 		 */
1742 		nvlist_find(nvl, GRUB_ENVMAP, DATA_TYPE_STRING, NULL,
1743 		    be->vbe_bootenv, NULL);
1744 		rv = 0;
1745 		break;
1746 
1747 	case VB_NVLIST:
1748 		nv.nv_header = nvl->nv_header;
1749 		nv.nv_asize = nvl->nv_asize;
1750 		nv.nv_size = nvl->nv_size;
1751 
1752 		bcopy(&nv.nv_header, be->vbe_bootenv, sizeof(nv.nv_header));
1753 		nv.nv_data = be->vbe_bootenv + sizeof(nvs_header_t);
1754 		bcopy(nvl->nv_data, nv.nv_data, nv.nv_size);
1755 		rv = nvlist_export(&nv);
1756 		break;
1757 
1758 	default:
1759 		rv = EINVAL;
1760 		break;
1761 	}
1762 
1763 	if (rv == 0) {
1764 		be->vbe_version = htobe64(be->vbe_version);
1765 		rv = vdev_write_bootenv_impl(vdev, be);
1766 	}
1767 	free(be);
1768 	return (rv);
1769 }
1770 
1771 /*
1772  * Read the bootenv area from pool label, return the nvlist from it.
1773  * We return from first successful read.
1774  */
1775 nvlist_t *
1776 vdev_read_bootenv(vdev_t *vdev)
1777 {
1778 	vdev_t *kid;
1779 	nvlist_t *benv;
1780 	vdev_boot_envblock_t *be;
1781 	char *command;
1782 	bool ok;
1783 	int rv;
1784 
1785 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1786 		if (kid->v_state != VDEV_STATE_HEALTHY)
1787 			continue;
1788 
1789 		benv = vdev_read_bootenv(kid);
1790 		if (benv != NULL)
1791 			return (benv);
1792 	}
1793 
1794 	be = malloc(sizeof (*be));
1795 	if (be == NULL)
1796 		return (NULL);
1797 
1798 	rv = 0;
1799 	for (int l = 0; l < VDEV_LABELS; l++) {
1800 		rv = vdev_label_read(vdev, l, be,
1801 		    offsetof(vdev_label_t, vl_be),
1802 		    sizeof (*be));
1803 		if (rv == 0)
1804 			break;
1805 	}
1806 	if (rv != 0) {
1807 		free(be);
1808 		return (NULL);
1809 	}
1810 
1811 	be->vbe_version = be64toh(be->vbe_version);
1812 	switch (be->vbe_version) {
1813 	case VB_RAW:
1814 		/*
1815 		 * we have textual data in vbe_bootenv, create nvlist
1816 		 * with key "envmap".
1817 		 */
1818 		benv = nvlist_create(NV_UNIQUE_NAME);
1819 		if (benv != NULL) {
1820 			if (*be->vbe_bootenv == '\0') {
1821 				nvlist_add_uint64(benv, BOOTENV_VERSION,
1822 				    VB_NVLIST);
1823 				break;
1824 			}
1825 			nvlist_add_uint64(benv, BOOTENV_VERSION, VB_RAW);
1826 			be->vbe_bootenv[sizeof (be->vbe_bootenv) - 1] = '\0';
1827 			nvlist_add_string(benv, GRUB_ENVMAP, be->vbe_bootenv);
1828 		}
1829 		break;
1830 
1831 	case VB_NVLIST:
1832 		benv = nvlist_import(be->vbe_bootenv, sizeof(be->vbe_bootenv));
1833 		break;
1834 
1835 	default:
1836 		command = (char *)be;
1837 		ok = false;
1838 
1839 		/* Check for legacy zfsbootcfg command string */
1840 		for (int i = 0; command[i] != '\0'; i++) {
1841 			if (iscntrl(command[i])) {
1842 				ok = false;
1843 				break;
1844 			} else {
1845 				ok = true;
1846 			}
1847 		}
1848 		benv = nvlist_create(NV_UNIQUE_NAME);
1849 		if (benv != NULL) {
1850 			if (ok)
1851 				nvlist_add_string(benv, FREEBSD_BOOTONCE,
1852 				    command);
1853 			else
1854 				nvlist_add_uint64(benv, BOOTENV_VERSION,
1855 				    VB_NVLIST);
1856 		}
1857 		break;
1858 	}
1859 	free(be);
1860 	return (benv);
1861 }
1862 
1863 static uint64_t
1864 vdev_get_label_asize(nvlist_t *nvl)
1865 {
1866 	nvlist_t *vdevs;
1867 	uint64_t asize;
1868 	const char *type;
1869 	int len;
1870 
1871 	asize = 0;
1872 	/* Get vdev tree */
1873 	if (nvlist_find(nvl, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1874 	    NULL, &vdevs, NULL) != 0)
1875 		return (asize);
1876 
1877 	/*
1878 	 * Get vdev type. We will calculate asize for raidz, mirror and disk.
1879 	 * For raidz, the asize is raw size of all children.
1880 	 */
1881 	if (nvlist_find(vdevs, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
1882 	    NULL, &type, &len) != 0)
1883 		goto done;
1884 
1885 	if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 &&
1886 	    memcmp(type, VDEV_TYPE_DISK, len) != 0 &&
1887 	    memcmp(type, VDEV_TYPE_RAIDZ, len) != 0)
1888 		goto done;
1889 
1890 	if (nvlist_find(vdevs, ZPOOL_CONFIG_ASIZE, DATA_TYPE_UINT64,
1891 	    NULL, &asize, NULL) != 0)
1892 		goto done;
1893 
1894 	if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) {
1895 		nvlist_t **kids;
1896 		int nkids;
1897 
1898 		if (nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN,
1899 		    DATA_TYPE_NVLIST_ARRAY, &nkids, &kids, NULL) != 0) {
1900 			asize = 0;
1901 			goto done;
1902 		}
1903 
1904 		asize /= nkids;
1905 		for (int i = 0; i < nkids; i++)
1906 			nvlist_destroy(kids[i]);
1907 		free(kids);
1908 	}
1909 
1910 	asize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
1911 done:
1912 	nvlist_destroy(vdevs);
1913 	return (asize);
1914 }
1915 
1916 static nvlist_t *
1917 vdev_label_read_config(vdev_t *vd, uint64_t txg)
1918 {
1919 	vdev_phys_t *label;
1920 	uint64_t best_txg = 0;
1921 	uint64_t label_txg = 0;
1922 	uint64_t asize;
1923 	nvlist_t *nvl = NULL, *tmp;
1924 	int error;
1925 
1926 	label = malloc(sizeof (vdev_phys_t));
1927 	if (label == NULL)
1928 		return (NULL);
1929 
1930 	for (int l = 0; l < VDEV_LABELS; l++) {
1931 		if (vdev_label_read(vd, l, label,
1932 		    offsetof(vdev_label_t, vl_vdev_phys),
1933 		    sizeof (vdev_phys_t)))
1934 			continue;
1935 
1936 		tmp = nvlist_import(label->vp_nvlist,
1937 		    sizeof(label->vp_nvlist));
1938 		if (tmp == NULL)
1939 			continue;
1940 
1941 		error = nvlist_find(tmp, ZPOOL_CONFIG_POOL_TXG,
1942 		    DATA_TYPE_UINT64, NULL, &label_txg, NULL);
1943 		if (error != 0 || label_txg == 0) {
1944 			nvlist_destroy(nvl);
1945 			nvl = tmp;
1946 			goto done;
1947 		}
1948 
1949 		if (label_txg <= txg && label_txg > best_txg) {
1950 			best_txg = label_txg;
1951 			nvlist_destroy(nvl);
1952 			nvl = tmp;
1953 			tmp = NULL;
1954 
1955 			/*
1956 			 * Use asize from pool config. We need this
1957 			 * because we can get bad value from BIOS.
1958 			 */
1959 			asize = vdev_get_label_asize(nvl);
1960 			if (asize != 0) {
1961 				vd->v_psize = asize;
1962 			}
1963 		}
1964 		nvlist_destroy(tmp);
1965 	}
1966 
1967 	if (best_txg == 0) {
1968 		nvlist_destroy(nvl);
1969 		nvl = NULL;
1970 	}
1971 done:
1972 	free(label);
1973 	return (nvl);
1974 }
1975 
1976 static void
1977 vdev_uberblock_load(vdev_t *vd, uberblock_t *ub)
1978 {
1979 	uberblock_t *buf;
1980 
1981 	buf = malloc(VDEV_UBERBLOCK_SIZE(vd));
1982 	if (buf == NULL)
1983 		return;
1984 
1985 	for (int l = 0; l < VDEV_LABELS; l++) {
1986 		for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1987 			if (vdev_label_read(vd, l, buf,
1988 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1989 			    VDEV_UBERBLOCK_SIZE(vd)))
1990 				continue;
1991 			if (uberblock_verify(buf) != 0)
1992 				continue;
1993 
1994 			if (vdev_uberblock_compare(buf, ub) > 0)
1995 				*ub = *buf;
1996 		}
1997 	}
1998 	free(buf);
1999 }
2000 
2001 static int
2002 vdev_probe(vdev_phys_read_t *_read, vdev_phys_write_t *_write, void *priv,
2003     spa_t **spap)
2004 {
2005 	vdev_t vtmp;
2006 	spa_t *spa;
2007 	vdev_t *vdev;
2008 	nvlist_t *nvl;
2009 	uint64_t val;
2010 	uint64_t guid, vdev_children;
2011 	uint64_t pool_txg, pool_guid;
2012 	const char *pool_name;
2013 	int rc, namelen;
2014 
2015 	/*
2016 	 * Load the vdev label and figure out which
2017 	 * uberblock is most current.
2018 	 */
2019 	memset(&vtmp, 0, sizeof(vtmp));
2020 	vtmp.v_phys_read = _read;
2021 	vtmp.v_phys_write = _write;
2022 	vtmp.v_priv = priv;
2023 	vtmp.v_psize = P2ALIGN(ldi_get_size(priv),
2024 	    (uint64_t)sizeof (vdev_label_t));
2025 
2026 	/* Test for minimum device size. */
2027 	if (vtmp.v_psize < SPA_MINDEVSIZE)
2028 		return (EIO);
2029 
2030 	nvl = vdev_label_read_config(&vtmp, UINT64_MAX);
2031 	if (nvl == NULL)
2032 		return (EIO);
2033 
2034 	if (nvlist_find(nvl, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
2035 	    NULL, &val, NULL) != 0) {
2036 		nvlist_destroy(nvl);
2037 		return (EIO);
2038 	}
2039 
2040 	if (!SPA_VERSION_IS_SUPPORTED(val)) {
2041 		printf("ZFS: unsupported ZFS version %u (should be %u)\n",
2042 		    (unsigned)val, (unsigned)SPA_VERSION);
2043 		nvlist_destroy(nvl);
2044 		return (EIO);
2045 	}
2046 
2047 	/* Check ZFS features for read */
2048 	rc = nvlist_check_features_for_read(nvl);
2049 	if (rc != 0) {
2050 		nvlist_destroy(nvl);
2051 		return (EIO);
2052 	}
2053 
2054 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
2055 	    NULL, &val, NULL) != 0) {
2056 		nvlist_destroy(nvl);
2057 		return (EIO);
2058 	}
2059 
2060 	if (val == POOL_STATE_DESTROYED) {
2061 		/* We don't boot only from destroyed pools. */
2062 		nvlist_destroy(nvl);
2063 		return (EIO);
2064 	}
2065 
2066 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
2067 	    NULL, &pool_txg, NULL) != 0 ||
2068 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
2069 	    NULL, &pool_guid, NULL) != 0 ||
2070 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
2071 	    NULL, &pool_name, &namelen) != 0) {
2072 		/*
2073 		 * Cache and spare devices end up here - just ignore
2074 		 * them.
2075 		 */
2076 		nvlist_destroy(nvl);
2077 		return (EIO);
2078 	}
2079 
2080 	/*
2081 	 * Create the pool if this is the first time we've seen it.
2082 	 */
2083 	spa = spa_find_by_guid(pool_guid);
2084 	if (spa == NULL) {
2085 		char *name;
2086 
2087 		nvlist_find(nvl, ZPOOL_CONFIG_VDEV_CHILDREN,
2088 		    DATA_TYPE_UINT64, NULL, &vdev_children, NULL);
2089 		name = malloc(namelen + 1);
2090 		if (name == NULL) {
2091 			nvlist_destroy(nvl);
2092 			return (ENOMEM);
2093 		}
2094 		bcopy(pool_name, name, namelen);
2095 		name[namelen] = '\0';
2096 		spa = spa_create(pool_guid, name);
2097 		free(name);
2098 		if (spa == NULL) {
2099 			nvlist_destroy(nvl);
2100 			return (ENOMEM);
2101 		}
2102 		spa->spa_root_vdev->v_nchildren = vdev_children;
2103 	}
2104 	if (pool_txg > spa->spa_txg)
2105 		spa->spa_txg = pool_txg;
2106 
2107 	/*
2108 	 * Get the vdev tree and create our in-core copy of it.
2109 	 * If we already have a vdev with this guid, this must
2110 	 * be some kind of alias (overlapping slices, dangerously dedicated
2111 	 * disks etc).
2112 	 */
2113 	if (nvlist_find(nvl, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
2114 	    NULL, &guid, NULL) != 0) {
2115 		nvlist_destroy(nvl);
2116 		return (EIO);
2117 	}
2118 	vdev = vdev_find(guid);
2119 	/* Has this vdev already been inited? */
2120 	if (vdev && vdev->v_phys_read) {
2121 		nvlist_destroy(nvl);
2122 		return (EIO);
2123 	}
2124 
2125 	rc = vdev_init_from_label(spa, nvl);
2126 	nvlist_destroy(nvl);
2127 	if (rc != 0)
2128 		return (rc);
2129 
2130 	/*
2131 	 * We should already have created an incomplete vdev for this
2132 	 * vdev. Find it and initialise it with our read proc.
2133 	 */
2134 	vdev = vdev_find(guid);
2135 	if (vdev != NULL) {
2136 		vdev->v_phys_read = _read;
2137 		vdev->v_phys_write = _write;
2138 		vdev->v_priv = priv;
2139 		vdev->v_psize = vtmp.v_psize;
2140 		/*
2141 		 * If no other state is set, mark vdev healthy.
2142 		 */
2143 		if (vdev->v_state == VDEV_STATE_UNKNOWN)
2144 			vdev->v_state = VDEV_STATE_HEALTHY;
2145 	} else {
2146 		printf("ZFS: inconsistent nvlist contents\n");
2147 		return (EIO);
2148 	}
2149 
2150 	if (vdev->v_islog)
2151 		spa->spa_with_log = vdev->v_islog;
2152 
2153 	/*
2154 	 * Re-evaluate top-level vdev state.
2155 	 */
2156 	vdev_set_state(vdev->v_top);
2157 
2158 	/*
2159 	 * Ok, we are happy with the pool so far. Lets find
2160 	 * the best uberblock and then we can actually access
2161 	 * the contents of the pool.
2162 	 */
2163 	vdev_uberblock_load(vdev, spa->spa_uberblock);
2164 
2165 	if (spap != NULL)
2166 		*spap = spa;
2167 	return (0);
2168 }
2169 
2170 static int
2171 ilog2(int n)
2172 {
2173 	int v;
2174 
2175 	for (v = 0; v < 32; v++)
2176 		if (n == (1 << v))
2177 			return (v);
2178 	return (-1);
2179 }
2180 
2181 static int
2182 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
2183 {
2184 	blkptr_t gbh_bp;
2185 	zio_gbh_phys_t zio_gb;
2186 	char *pbuf;
2187 	int i;
2188 
2189 	/* Artificial BP for gang block header. */
2190 	gbh_bp = *bp;
2191 	BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
2192 	BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
2193 	BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
2194 	BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
2195 	for (i = 0; i < SPA_DVAS_PER_BP; i++)
2196 		DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
2197 
2198 	/* Read gang header block using the artificial BP. */
2199 	if (zio_read(spa, &gbh_bp, &zio_gb))
2200 		return (EIO);
2201 
2202 	pbuf = buf;
2203 	for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
2204 		blkptr_t *gbp = &zio_gb.zg_blkptr[i];
2205 
2206 		if (BP_IS_HOLE(gbp))
2207 			continue;
2208 		if (zio_read(spa, gbp, pbuf))
2209 			return (EIO);
2210 		pbuf += BP_GET_PSIZE(gbp);
2211 	}
2212 
2213 	if (zio_checksum_verify(spa, bp, buf))
2214 		return (EIO);
2215 	return (0);
2216 }
2217 
2218 static int
2219 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
2220 {
2221 	int cpfunc = BP_GET_COMPRESS(bp);
2222 	uint64_t align, size;
2223 	void *pbuf;
2224 	int i, error;
2225 
2226 	/*
2227 	 * Process data embedded in block pointer
2228 	 */
2229 	if (BP_IS_EMBEDDED(bp)) {
2230 		ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
2231 
2232 		size = BPE_GET_PSIZE(bp);
2233 		ASSERT(size <= BPE_PAYLOAD_SIZE);
2234 
2235 		if (cpfunc != ZIO_COMPRESS_OFF)
2236 			pbuf = malloc(size);
2237 		else
2238 			pbuf = buf;
2239 
2240 		if (pbuf == NULL)
2241 			return (ENOMEM);
2242 
2243 		decode_embedded_bp_compressed(bp, pbuf);
2244 		error = 0;
2245 
2246 		if (cpfunc != ZIO_COMPRESS_OFF) {
2247 			error = zio_decompress_data(cpfunc, pbuf,
2248 			    size, buf, BP_GET_LSIZE(bp));
2249 			free(pbuf);
2250 		}
2251 		if (error != 0)
2252 			printf("ZFS: i/o error - unable to decompress "
2253 			    "block pointer data, error %d\n", error);
2254 		return (error);
2255 	}
2256 
2257 	error = EIO;
2258 
2259 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
2260 		const dva_t *dva = &bp->blk_dva[i];
2261 		vdev_t *vdev;
2262 		vdev_list_t *vlist;
2263 		uint64_t vdevid;
2264 		off_t offset;
2265 
2266 		if (!dva->dva_word[0] && !dva->dva_word[1])
2267 			continue;
2268 
2269 		vdevid = DVA_GET_VDEV(dva);
2270 		offset = DVA_GET_OFFSET(dva);
2271 		vlist = &spa->spa_root_vdev->v_children;
2272 		STAILQ_FOREACH(vdev, vlist, v_childlink) {
2273 			if (vdev->v_id == vdevid)
2274 				break;
2275 		}
2276 		if (!vdev || !vdev->v_read)
2277 			continue;
2278 
2279 		size = BP_GET_PSIZE(bp);
2280 		if (vdev->v_read == vdev_raidz_read) {
2281 			align = 1ULL << vdev->v_ashift;
2282 			if (P2PHASE(size, align) != 0)
2283 				size = P2ROUNDUP(size, align);
2284 		}
2285 		if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
2286 			pbuf = malloc(size);
2287 		else
2288 			pbuf = buf;
2289 
2290 		if (pbuf == NULL) {
2291 			error = ENOMEM;
2292 			break;
2293 		}
2294 
2295 		if (DVA_GET_GANG(dva))
2296 			error = zio_read_gang(spa, bp, pbuf);
2297 		else
2298 			error = vdev->v_read(vdev, bp, pbuf, offset, size);
2299 		if (error == 0) {
2300 			if (cpfunc != ZIO_COMPRESS_OFF)
2301 				error = zio_decompress_data(cpfunc, pbuf,
2302 				    BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
2303 			else if (size != BP_GET_PSIZE(bp))
2304 				bcopy(pbuf, buf, BP_GET_PSIZE(bp));
2305 		} else {
2306 			printf("zio_read error: %d\n", error);
2307 		}
2308 		if (buf != pbuf)
2309 			free(pbuf);
2310 		if (error == 0)
2311 			break;
2312 	}
2313 	if (error != 0)
2314 		printf("ZFS: i/o error - all block copies unavailable\n");
2315 
2316 	return (error);
2317 }
2318 
2319 static int
2320 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset,
2321     void *buf, size_t buflen)
2322 {
2323 	int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
2324 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2325 	int nlevels = dnode->dn_nlevels;
2326 	int i, rc;
2327 
2328 	if (bsize > SPA_MAXBLOCKSIZE) {
2329 		printf("ZFS: I/O error - blocks larger than %llu are not "
2330 		    "supported\n", SPA_MAXBLOCKSIZE);
2331 		return (EIO);
2332 	}
2333 
2334 	/*
2335 	 * Note: bsize may not be a power of two here so we need to do an
2336 	 * actual divide rather than a bitshift.
2337 	 */
2338 	while (buflen > 0) {
2339 		uint64_t bn = offset / bsize;
2340 		int boff = offset % bsize;
2341 		int ibn;
2342 		const blkptr_t *indbp;
2343 		blkptr_t bp;
2344 
2345 		if (bn > dnode->dn_maxblkid)
2346 			return (EIO);
2347 
2348 		if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
2349 			goto cached;
2350 
2351 		indbp = dnode->dn_blkptr;
2352 		for (i = 0; i < nlevels; i++) {
2353 			/*
2354 			 * Copy the bp from the indirect array so that
2355 			 * we can re-use the scratch buffer for multi-level
2356 			 * objects.
2357 			 */
2358 			ibn = bn >> ((nlevels - i - 1) * ibshift);
2359 			ibn &= ((1 << ibshift) - 1);
2360 			bp = indbp[ibn];
2361 			if (BP_IS_HOLE(&bp)) {
2362 				memset(dnode_cache_buf, 0, bsize);
2363 				break;
2364 			}
2365 			rc = zio_read(spa, &bp, dnode_cache_buf);
2366 			if (rc)
2367 				return (rc);
2368 			indbp = (const blkptr_t *) dnode_cache_buf;
2369 		}
2370 		dnode_cache_obj = dnode;
2371 		dnode_cache_bn = bn;
2372 	cached:
2373 
2374 		/*
2375 		 * The buffer contains our data block. Copy what we
2376 		 * need from it and loop.
2377 		 */
2378 		i = bsize - boff;
2379 		if (i > buflen) i = buflen;
2380 		memcpy(buf, &dnode_cache_buf[boff], i);
2381 		buf = ((char *)buf) + i;
2382 		offset += i;
2383 		buflen -= i;
2384 	}
2385 
2386 	return (0);
2387 }
2388 
2389 /*
2390  * Lookup a value in a microzap directory.
2391  */
2392 static int
2393 mzap_lookup(const mzap_phys_t *mz, size_t size, const char *name,
2394     uint64_t *value)
2395 {
2396 	const mzap_ent_phys_t *mze;
2397 	int chunks, i;
2398 
2399 	/*
2400 	 * Microzap objects use exactly one block. Read the whole
2401 	 * thing.
2402 	 */
2403 	chunks = size / MZAP_ENT_LEN - 1;
2404 	for (i = 0; i < chunks; i++) {
2405 		mze = &mz->mz_chunk[i];
2406 		if (strcmp(mze->mze_name, name) == 0) {
2407 			*value = mze->mze_value;
2408 			return (0);
2409 		}
2410 	}
2411 
2412 	return (ENOENT);
2413 }
2414 
2415 /*
2416  * Compare a name with a zap leaf entry. Return non-zero if the name
2417  * matches.
2418  */
2419 static int
2420 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2421     const char *name)
2422 {
2423 	size_t namelen;
2424 	const zap_leaf_chunk_t *nc;
2425 	const char *p;
2426 
2427 	namelen = zc->l_entry.le_name_numints;
2428 
2429 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2430 	p = name;
2431 	while (namelen > 0) {
2432 		size_t len;
2433 
2434 		len = namelen;
2435 		if (len > ZAP_LEAF_ARRAY_BYTES)
2436 			len = ZAP_LEAF_ARRAY_BYTES;
2437 		if (memcmp(p, nc->l_array.la_array, len))
2438 			return (0);
2439 		p += len;
2440 		namelen -= len;
2441 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2442 	}
2443 
2444 	return (1);
2445 }
2446 
2447 /*
2448  * Extract a uint64_t value from a zap leaf entry.
2449  */
2450 static uint64_t
2451 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
2452 {
2453 	const zap_leaf_chunk_t *vc;
2454 	int i;
2455 	uint64_t value;
2456 	const uint8_t *p;
2457 
2458 	vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
2459 	for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
2460 		value = (value << 8) | p[i];
2461 	}
2462 
2463 	return (value);
2464 }
2465 
2466 static void
2467 stv(int len, void *addr, uint64_t value)
2468 {
2469 	switch (len) {
2470 	case 1:
2471 		*(uint8_t *)addr = value;
2472 		return;
2473 	case 2:
2474 		*(uint16_t *)addr = value;
2475 		return;
2476 	case 4:
2477 		*(uint32_t *)addr = value;
2478 		return;
2479 	case 8:
2480 		*(uint64_t *)addr = value;
2481 		return;
2482 	}
2483 }
2484 
2485 /*
2486  * Extract a array from a zap leaf entry.
2487  */
2488 static void
2489 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2490     uint64_t integer_size, uint64_t num_integers, void *buf)
2491 {
2492 	uint64_t array_int_len = zc->l_entry.le_value_intlen;
2493 	uint64_t value = 0;
2494 	uint64_t *u64 = buf;
2495 	char *p = buf;
2496 	int len = MIN(zc->l_entry.le_value_numints, num_integers);
2497 	int chunk = zc->l_entry.le_value_chunk;
2498 	int byten = 0;
2499 
2500 	if (integer_size == 8 && len == 1) {
2501 		*u64 = fzap_leaf_value(zl, zc);
2502 		return;
2503 	}
2504 
2505 	while (len > 0) {
2506 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
2507 		int i;
2508 
2509 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
2510 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
2511 			value = (value << 8) | la->la_array[i];
2512 			byten++;
2513 			if (byten == array_int_len) {
2514 				stv(integer_size, p, value);
2515 				byten = 0;
2516 				len--;
2517 				if (len == 0)
2518 					return;
2519 				p += integer_size;
2520 			}
2521 		}
2522 		chunk = la->la_next;
2523 	}
2524 }
2525 
2526 static int
2527 fzap_check_size(uint64_t integer_size, uint64_t num_integers)
2528 {
2529 
2530 	switch (integer_size) {
2531 	case 1:
2532 	case 2:
2533 	case 4:
2534 	case 8:
2535 		break;
2536 	default:
2537 		return (EINVAL);
2538 	}
2539 
2540 	if (integer_size * num_integers > ZAP_MAXVALUELEN)
2541 		return (E2BIG);
2542 
2543 	return (0);
2544 }
2545 
2546 static void
2547 zap_leaf_free(zap_leaf_t *leaf)
2548 {
2549 	free(leaf->l_phys);
2550 	free(leaf);
2551 }
2552 
2553 static int
2554 zap_get_leaf_byblk(fat_zap_t *zap, uint64_t blk, zap_leaf_t **lp)
2555 {
2556 	int bs = FZAP_BLOCK_SHIFT(zap);
2557 	int err;
2558 
2559 	*lp = malloc(sizeof(**lp));
2560 	if (*lp == NULL)
2561 		return (ENOMEM);
2562 
2563 	(*lp)->l_bs = bs;
2564 	(*lp)->l_phys = malloc(1 << bs);
2565 
2566 	if ((*lp)->l_phys == NULL) {
2567 		free(*lp);
2568 		return (ENOMEM);
2569 	}
2570 	err = dnode_read(zap->zap_spa, zap->zap_dnode, blk << bs, (*lp)->l_phys,
2571 	    1 << bs);
2572 	if (err != 0) {
2573 		zap_leaf_free(*lp);
2574 	}
2575 	return (err);
2576 }
2577 
2578 static int
2579 zap_table_load(fat_zap_t *zap, zap_table_phys_t *tbl, uint64_t idx,
2580     uint64_t *valp)
2581 {
2582 	int bs = FZAP_BLOCK_SHIFT(zap);
2583 	uint64_t blk = idx >> (bs - 3);
2584 	uint64_t off = idx & ((1 << (bs - 3)) - 1);
2585 	uint64_t *buf;
2586 	int rc;
2587 
2588 	buf = malloc(1 << zap->zap_block_shift);
2589 	if (buf == NULL)
2590 		return (ENOMEM);
2591 	rc = dnode_read(zap->zap_spa, zap->zap_dnode, (tbl->zt_blk + blk) << bs,
2592 	    buf, 1 << zap->zap_block_shift);
2593 	if (rc == 0)
2594 		*valp = buf[off];
2595 	free(buf);
2596 	return (rc);
2597 }
2598 
2599 static int
2600 zap_idx_to_blk(fat_zap_t *zap, uint64_t idx, uint64_t *valp)
2601 {
2602 	if (zap->zap_phys->zap_ptrtbl.zt_numblks == 0) {
2603 		*valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
2604 		return (0);
2605 	} else {
2606 		return (zap_table_load(zap, &zap->zap_phys->zap_ptrtbl,
2607 		    idx, valp));
2608 	}
2609 }
2610 
2611 #define	ZAP_HASH_IDX(hash, n)	(((n) == 0) ? 0 : ((hash) >> (64 - (n))))
2612 static int
2613 zap_deref_leaf(fat_zap_t *zap, uint64_t h, zap_leaf_t **lp)
2614 {
2615 	uint64_t idx, blk;
2616 	int err;
2617 
2618 	idx = ZAP_HASH_IDX(h, zap->zap_phys->zap_ptrtbl.zt_shift);
2619 	err = zap_idx_to_blk(zap, idx, &blk);
2620 	if (err != 0)
2621 		return (err);
2622 	return (zap_get_leaf_byblk(zap, blk, lp));
2623 }
2624 
2625 #define	CHAIN_END	0xffff	/* end of the chunk chain */
2626 #define	LEAF_HASH(l, h) \
2627 	((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \
2628 	((h) >> \
2629 	(64 - ZAP_LEAF_HASH_SHIFT(l) - (l)->l_phys->l_hdr.lh_prefix_len)))
2630 #define	LEAF_HASH_ENTPTR(l, h)	(&(l)->l_phys->l_hash[LEAF_HASH(l, h)])
2631 
2632 static int
2633 zap_leaf_lookup(zap_leaf_t *zl, uint64_t hash, const char *name,
2634     uint64_t integer_size, uint64_t num_integers, void *value)
2635 {
2636 	int rc;
2637 	uint16_t *chunkp;
2638 	struct zap_leaf_entry *le;
2639 
2640 	/*
2641 	 * Make sure this chunk matches our hash.
2642 	 */
2643 	if (zl->l_phys->l_hdr.lh_prefix_len > 0 &&
2644 	    zl->l_phys->l_hdr.lh_prefix !=
2645 	    hash >> (64 - zl->l_phys->l_hdr.lh_prefix_len))
2646 		return (EIO);
2647 
2648 	rc = ENOENT;
2649 	for (chunkp = LEAF_HASH_ENTPTR(zl, hash);
2650 	    *chunkp != CHAIN_END; chunkp = &le->le_next) {
2651 		zap_leaf_chunk_t *zc;
2652 		uint16_t chunk = *chunkp;
2653 
2654 		le = ZAP_LEAF_ENTRY(zl, chunk);
2655 		if (le->le_hash != hash)
2656 			continue;
2657 		zc = &ZAP_LEAF_CHUNK(zl, chunk);
2658 		if (fzap_name_equal(zl, zc, name)) {
2659 			if (zc->l_entry.le_value_intlen > integer_size) {
2660 				rc = EINVAL;
2661 			} else {
2662 				fzap_leaf_array(zl, zc, integer_size,
2663 				    num_integers, value);
2664 				rc = 0;
2665 			}
2666 			break;
2667 		}
2668 	}
2669 	return (rc);
2670 }
2671 
2672 /*
2673  * Lookup a value in a fatzap directory.
2674  */
2675 static int
2676 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2677     const char *name, uint64_t integer_size, uint64_t num_integers,
2678     void *value)
2679 {
2680 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2681 	fat_zap_t z;
2682 	zap_leaf_t *zl;
2683 	uint64_t hash;
2684 	int rc;
2685 
2686 	if (zh->zap_magic != ZAP_MAGIC)
2687 		return (EIO);
2688 
2689 	if ((rc = fzap_check_size(integer_size, num_integers)) != 0) {
2690 		return (rc);
2691 	}
2692 
2693 	z.zap_block_shift = ilog2(bsize);
2694 	z.zap_phys = zh;
2695 	z.zap_spa = spa;
2696 	z.zap_dnode = dnode;
2697 
2698 	hash = zap_hash(zh->zap_salt, name);
2699 	rc = zap_deref_leaf(&z, hash, &zl);
2700 	if (rc != 0)
2701 		return (rc);
2702 
2703 	rc = zap_leaf_lookup(zl, hash, name, integer_size, num_integers, value);
2704 
2705 	zap_leaf_free(zl);
2706 	return (rc);
2707 }
2708 
2709 /*
2710  * Lookup a name in a zap object and return its value as a uint64_t.
2711  */
2712 static int
2713 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2714     uint64_t integer_size, uint64_t num_integers, void *value)
2715 {
2716 	int rc;
2717 	zap_phys_t *zap;
2718 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2719 
2720 	zap = malloc(size);
2721 	if (zap == NULL)
2722 		return (ENOMEM);
2723 
2724 	rc = dnode_read(spa, dnode, 0, zap, size);
2725 	if (rc)
2726 		goto done;
2727 
2728 	switch (zap->zap_block_type) {
2729 	case ZBT_MICRO:
2730 		rc = mzap_lookup((const mzap_phys_t *)zap, size, name, value);
2731 		break;
2732 	case ZBT_HEADER:
2733 		rc = fzap_lookup(spa, dnode, zap, name, integer_size,
2734 		    num_integers, value);
2735 		break;
2736 	default:
2737 		printf("ZFS: invalid zap_type=%" PRIx64 "\n",
2738 		    zap->zap_block_type);
2739 		rc = EIO;
2740 	}
2741 done:
2742 	free(zap);
2743 	return (rc);
2744 }
2745 
2746 /*
2747  * List a microzap directory.
2748  */
2749 static int
2750 mzap_list(const mzap_phys_t *mz, size_t size,
2751     int (*callback)(const char *, uint64_t))
2752 {
2753 	const mzap_ent_phys_t *mze;
2754 	int chunks, i, rc;
2755 
2756 	/*
2757 	 * Microzap objects use exactly one block. Read the whole
2758 	 * thing.
2759 	 */
2760 	rc = 0;
2761 	chunks = size / MZAP_ENT_LEN - 1;
2762 	for (i = 0; i < chunks; i++) {
2763 		mze = &mz->mz_chunk[i];
2764 		if (mze->mze_name[0]) {
2765 			rc = callback(mze->mze_name, mze->mze_value);
2766 			if (rc != 0)
2767 				break;
2768 		}
2769 	}
2770 
2771 	return (rc);
2772 }
2773 
2774 /*
2775  * List a fatzap directory.
2776  */
2777 static int
2778 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2779     int (*callback)(const char *, uint64_t))
2780 {
2781 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2782 	fat_zap_t z;
2783 	uint64_t i;
2784 	int j, rc;
2785 
2786 	if (zh->zap_magic != ZAP_MAGIC)
2787 		return (EIO);
2788 
2789 	z.zap_block_shift = ilog2(bsize);
2790 	z.zap_phys = zh;
2791 
2792 	/*
2793 	 * This assumes that the leaf blocks start at block 1. The
2794 	 * documentation isn't exactly clear on this.
2795 	 */
2796 	zap_leaf_t zl;
2797 	zl.l_bs = z.zap_block_shift;
2798 	zl.l_phys = malloc(bsize);
2799 	if (zl.l_phys == NULL)
2800 		return (ENOMEM);
2801 
2802 	for (i = 0; i < zh->zap_num_leafs; i++) {
2803 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2804 		char name[256], *p;
2805 		uint64_t value;
2806 
2807 		if (dnode_read(spa, dnode, off, zl.l_phys, bsize)) {
2808 			free(zl.l_phys);
2809 			return (EIO);
2810 		}
2811 
2812 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2813 			zap_leaf_chunk_t *zc, *nc;
2814 			int namelen;
2815 
2816 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2817 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2818 				continue;
2819 			namelen = zc->l_entry.le_name_numints;
2820 			if (namelen > sizeof(name))
2821 				namelen = sizeof(name);
2822 
2823 			/*
2824 			 * Paste the name back together.
2825 			 */
2826 			nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
2827 			p = name;
2828 			while (namelen > 0) {
2829 				int len;
2830 				len = namelen;
2831 				if (len > ZAP_LEAF_ARRAY_BYTES)
2832 					len = ZAP_LEAF_ARRAY_BYTES;
2833 				memcpy(p, nc->l_array.la_array, len);
2834 				p += len;
2835 				namelen -= len;
2836 				nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
2837 			}
2838 
2839 			/*
2840 			 * Assume the first eight bytes of the value are
2841 			 * a uint64_t.
2842 			 */
2843 			value = fzap_leaf_value(&zl, zc);
2844 
2845 			/* printf("%s 0x%jx\n", name, (uintmax_t)value); */
2846 			rc = callback((const char *)name, value);
2847 			if (rc != 0) {
2848 				free(zl.l_phys);
2849 				return (rc);
2850 			}
2851 		}
2852 	}
2853 
2854 	free(zl.l_phys);
2855 	return (0);
2856 }
2857 
2858 static int zfs_printf(const char *name, uint64_t value __unused)
2859 {
2860 
2861 	printf("%s\n", name);
2862 
2863 	return (0);
2864 }
2865 
2866 /*
2867  * List a zap directory.
2868  */
2869 static int
2870 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
2871 {
2872 	zap_phys_t *zap;
2873 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2874 	int rc;
2875 
2876 	zap = malloc(size);
2877 	if (zap == NULL)
2878 		return (ENOMEM);
2879 
2880 	rc = dnode_read(spa, dnode, 0, zap, size);
2881 	if (rc == 0) {
2882 		if (zap->zap_block_type == ZBT_MICRO)
2883 			rc = mzap_list((const mzap_phys_t *)zap, size,
2884 			    zfs_printf);
2885 		else
2886 			rc = fzap_list(spa, dnode, zap, zfs_printf);
2887 	}
2888 	free(zap);
2889 	return (rc);
2890 }
2891 
2892 static int
2893 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum,
2894     dnode_phys_t *dnode)
2895 {
2896 	off_t offset;
2897 
2898 	offset = objnum * sizeof(dnode_phys_t);
2899 	return dnode_read(spa, &os->os_meta_dnode, offset,
2900 		dnode, sizeof(dnode_phys_t));
2901 }
2902 
2903 /*
2904  * Lookup a name in a microzap directory.
2905  */
2906 static int
2907 mzap_rlookup(const mzap_phys_t *mz, size_t size, char *name, uint64_t value)
2908 {
2909 	const mzap_ent_phys_t *mze;
2910 	int chunks, i;
2911 
2912 	/*
2913 	 * Microzap objects use exactly one block. Read the whole
2914 	 * thing.
2915 	 */
2916 	chunks = size / MZAP_ENT_LEN - 1;
2917 	for (i = 0; i < chunks; i++) {
2918 		mze = &mz->mz_chunk[i];
2919 		if (value == mze->mze_value) {
2920 			strcpy(name, mze->mze_name);
2921 			return (0);
2922 		}
2923 	}
2924 
2925 	return (ENOENT);
2926 }
2927 
2928 static void
2929 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
2930 {
2931 	size_t namelen;
2932 	const zap_leaf_chunk_t *nc;
2933 	char *p;
2934 
2935 	namelen = zc->l_entry.le_name_numints;
2936 
2937 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2938 	p = name;
2939 	while (namelen > 0) {
2940 		size_t len;
2941 		len = namelen;
2942 		if (len > ZAP_LEAF_ARRAY_BYTES)
2943 			len = ZAP_LEAF_ARRAY_BYTES;
2944 		memcpy(p, nc->l_array.la_array, len);
2945 		p += len;
2946 		namelen -= len;
2947 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2948 	}
2949 
2950 	*p = '\0';
2951 }
2952 
2953 static int
2954 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2955     char *name, uint64_t value)
2956 {
2957 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2958 	fat_zap_t z;
2959 	uint64_t i;
2960 	int j, rc;
2961 
2962 	if (zh->zap_magic != ZAP_MAGIC)
2963 		return (EIO);
2964 
2965 	z.zap_block_shift = ilog2(bsize);
2966 	z.zap_phys = zh;
2967 
2968 	/*
2969 	 * This assumes that the leaf blocks start at block 1. The
2970 	 * documentation isn't exactly clear on this.
2971 	 */
2972 	zap_leaf_t zl;
2973 	zl.l_bs = z.zap_block_shift;
2974 	zl.l_phys = malloc(bsize);
2975 	if (zl.l_phys == NULL)
2976 		return (ENOMEM);
2977 
2978 	for (i = 0; i < zh->zap_num_leafs; i++) {
2979 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2980 
2981 		rc = dnode_read(spa, dnode, off, zl.l_phys, bsize);
2982 		if (rc != 0)
2983 			goto done;
2984 
2985 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2986 			zap_leaf_chunk_t *zc;
2987 
2988 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2989 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2990 				continue;
2991 			if (zc->l_entry.le_value_intlen != 8 ||
2992 			    zc->l_entry.le_value_numints != 1)
2993 				continue;
2994 
2995 			if (fzap_leaf_value(&zl, zc) == value) {
2996 				fzap_name_copy(&zl, zc, name);
2997 				goto done;
2998 			}
2999 		}
3000 	}
3001 
3002 	rc = ENOENT;
3003 done:
3004 	free(zl.l_phys);
3005 	return (rc);
3006 }
3007 
3008 static int
3009 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name,
3010     uint64_t value)
3011 {
3012 	zap_phys_t *zap;
3013 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
3014 	int rc;
3015 
3016 	zap = malloc(size);
3017 	if (zap == NULL)
3018 		return (ENOMEM);
3019 
3020 	rc = dnode_read(spa, dnode, 0, zap, size);
3021 	if (rc == 0) {
3022 		if (zap->zap_block_type == ZBT_MICRO)
3023 			rc = mzap_rlookup((const mzap_phys_t *)zap, size,
3024 			    name, value);
3025 		else
3026 			rc = fzap_rlookup(spa, dnode, zap, name, value);
3027 	}
3028 	free(zap);
3029 	return (rc);
3030 }
3031 
3032 static int
3033 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
3034 {
3035 	char name[256];
3036 	char component[256];
3037 	uint64_t dir_obj, parent_obj, child_dir_zapobj;
3038 	dnode_phys_t child_dir_zap, dataset, dir, parent;
3039 	dsl_dir_phys_t *dd;
3040 	dsl_dataset_phys_t *ds;
3041 	char *p;
3042 	int len;
3043 
3044 	p = &name[sizeof(name) - 1];
3045 	*p = '\0';
3046 
3047 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
3048 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3049 		return (EIO);
3050 	}
3051 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3052 	dir_obj = ds->ds_dir_obj;
3053 
3054 	for (;;) {
3055 		if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir) != 0)
3056 			return (EIO);
3057 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3058 
3059 		/* Actual loop condition. */
3060 		parent_obj = dd->dd_parent_obj;
3061 		if (parent_obj == 0)
3062 			break;
3063 
3064 		if (objset_get_dnode(spa, spa->spa_mos, parent_obj,
3065 		    &parent) != 0)
3066 			return (EIO);
3067 		dd = (dsl_dir_phys_t *)&parent.dn_bonus;
3068 		child_dir_zapobj = dd->dd_child_dir_zapobj;
3069 		if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
3070 		    &child_dir_zap) != 0)
3071 			return (EIO);
3072 		if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
3073 			return (EIO);
3074 
3075 		len = strlen(component);
3076 		p -= len;
3077 		memcpy(p, component, len);
3078 		--p;
3079 		*p = '/';
3080 
3081 		/* Actual loop iteration. */
3082 		dir_obj = parent_obj;
3083 	}
3084 
3085 	if (*p != '\0')
3086 		++p;
3087 	strcpy(result, p);
3088 
3089 	return (0);
3090 }
3091 
3092 static int
3093 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
3094 {
3095 	char element[256];
3096 	uint64_t dir_obj, child_dir_zapobj;
3097 	dnode_phys_t child_dir_zap, dir;
3098 	dsl_dir_phys_t *dd;
3099 	const char *p, *q;
3100 
3101 	if (objset_get_dnode(spa, spa->spa_mos,
3102 	    DMU_POOL_DIRECTORY_OBJECT, &dir))
3103 		return (EIO);
3104 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
3105 	    1, &dir_obj))
3106 		return (EIO);
3107 
3108 	p = name;
3109 	for (;;) {
3110 		if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir))
3111 			return (EIO);
3112 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3113 
3114 		while (*p == '/')
3115 			p++;
3116 		/* Actual loop condition #1. */
3117 		if (*p == '\0')
3118 			break;
3119 
3120 		q = strchr(p, '/');
3121 		if (q) {
3122 			memcpy(element, p, q - p);
3123 			element[q - p] = '\0';
3124 			p = q + 1;
3125 		} else {
3126 			strcpy(element, p);
3127 			p += strlen(p);
3128 		}
3129 
3130 		child_dir_zapobj = dd->dd_child_dir_zapobj;
3131 		if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
3132 		    &child_dir_zap) != 0)
3133 			return (EIO);
3134 
3135 		/* Actual loop condition #2. */
3136 		if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
3137 		    1, &dir_obj) != 0)
3138 			return (ENOENT);
3139 	}
3140 
3141 	*objnum = dd->dd_head_dataset_obj;
3142 	return (0);
3143 }
3144 
3145 #ifndef BOOT2
3146 static int
3147 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
3148 {
3149 	uint64_t dir_obj, child_dir_zapobj;
3150 	dnode_phys_t child_dir_zap, dir, dataset;
3151 	dsl_dataset_phys_t *ds;
3152 	dsl_dir_phys_t *dd;
3153 
3154 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
3155 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3156 		return (EIO);
3157 	}
3158 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3159 	dir_obj = ds->ds_dir_obj;
3160 
3161 	if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir)) {
3162 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
3163 		return (EIO);
3164 	}
3165 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3166 
3167 	child_dir_zapobj = dd->dd_child_dir_zapobj;
3168 	if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
3169 	    &child_dir_zap) != 0) {
3170 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
3171 		return (EIO);
3172 	}
3173 
3174 	return (zap_list(spa, &child_dir_zap) != 0);
3175 }
3176 
3177 int
3178 zfs_callback_dataset(const spa_t *spa, uint64_t objnum,
3179     int (*callback)(const char *, uint64_t))
3180 {
3181 	uint64_t dir_obj, child_dir_zapobj;
3182 	dnode_phys_t child_dir_zap, dir, dataset;
3183 	dsl_dataset_phys_t *ds;
3184 	dsl_dir_phys_t *dd;
3185 	zap_phys_t *zap;
3186 	size_t size;
3187 	int err;
3188 
3189 	err = objset_get_dnode(spa, spa->spa_mos, objnum, &dataset);
3190 	if (err != 0) {
3191 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3192 		return (err);
3193 	}
3194 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3195 	dir_obj = ds->ds_dir_obj;
3196 
3197 	err = objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir);
3198 	if (err != 0) {
3199 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
3200 		return (err);
3201 	}
3202 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3203 
3204 	child_dir_zapobj = dd->dd_child_dir_zapobj;
3205 	err = objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
3206 	    &child_dir_zap);
3207 	if (err != 0) {
3208 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
3209 		return (err);
3210 	}
3211 
3212 	size = child_dir_zap.dn_datablkszsec << SPA_MINBLOCKSHIFT;
3213 	zap = malloc(size);
3214 	if (zap != NULL) {
3215 		err = dnode_read(spa, &child_dir_zap, 0, zap, size);
3216 		if (err != 0)
3217 			goto done;
3218 
3219 		if (zap->zap_block_type == ZBT_MICRO)
3220 			err = mzap_list((const mzap_phys_t *)zap, size,
3221 			    callback);
3222 		else
3223 			err = fzap_list(spa, &child_dir_zap, zap, callback);
3224 	} else {
3225 		err = ENOMEM;
3226 	}
3227 done:
3228 	free(zap);
3229 	return (err);
3230 }
3231 #endif
3232 
3233 /*
3234  * Find the object set given the object number of its dataset object
3235  * and return its details in *objset
3236  */
3237 static int
3238 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
3239 {
3240 	dnode_phys_t dataset;
3241 	dsl_dataset_phys_t *ds;
3242 
3243 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
3244 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3245 		return (EIO);
3246 	}
3247 
3248 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3249 	if (zio_read(spa, &ds->ds_bp, objset)) {
3250 		printf("ZFS: can't read object set for dataset %ju\n",
3251 		    (uintmax_t)objnum);
3252 		return (EIO);
3253 	}
3254 
3255 	return (0);
3256 }
3257 
3258 /*
3259  * Find the object set pointed to by the BOOTFS property or the root
3260  * dataset if there is none and return its details in *objset
3261  */
3262 static int
3263 zfs_get_root(const spa_t *spa, uint64_t *objid)
3264 {
3265 	dnode_phys_t dir, propdir;
3266 	uint64_t props, bootfs, root;
3267 
3268 	*objid = 0;
3269 
3270 	/*
3271 	 * Start with the MOS directory object.
3272 	 */
3273 	if (objset_get_dnode(spa, spa->spa_mos,
3274 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3275 		printf("ZFS: can't read MOS object directory\n");
3276 		return (EIO);
3277 	}
3278 
3279 	/*
3280 	 * Lookup the pool_props and see if we can find a bootfs.
3281 	 */
3282 	if (zap_lookup(spa, &dir, DMU_POOL_PROPS,
3283 	    sizeof(props), 1, &props) == 0 &&
3284 	    objset_get_dnode(spa, spa->spa_mos, props, &propdir) == 0 &&
3285 	    zap_lookup(spa, &propdir, "bootfs",
3286 	    sizeof(bootfs), 1, &bootfs) == 0 && bootfs != 0) {
3287 		*objid = bootfs;
3288 		return (0);
3289 	}
3290 	/*
3291 	 * Lookup the root dataset directory
3292 	 */
3293 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET,
3294 	    sizeof(root), 1, &root) ||
3295 	    objset_get_dnode(spa, spa->spa_mos, root, &dir)) {
3296 		printf("ZFS: can't find root dsl_dir\n");
3297 		return (EIO);
3298 	}
3299 
3300 	/*
3301 	 * Use the information from the dataset directory's bonus buffer
3302 	 * to find the dataset object and from that the object set itself.
3303 	 */
3304 	dsl_dir_phys_t *dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3305 	*objid = dd->dd_head_dataset_obj;
3306 	return (0);
3307 }
3308 
3309 static int
3310 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
3311 {
3312 
3313 	mount->spa = spa;
3314 
3315 	/*
3316 	 * Find the root object set if not explicitly provided
3317 	 */
3318 	if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
3319 		printf("ZFS: can't find root filesystem\n");
3320 		return (EIO);
3321 	}
3322 
3323 	if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
3324 		printf("ZFS: can't open root filesystem\n");
3325 		return (EIO);
3326 	}
3327 
3328 	mount->rootobj = rootobj;
3329 
3330 	return (0);
3331 }
3332 
3333 /*
3334  * callback function for feature name checks.
3335  */
3336 static int
3337 check_feature(const char *name, uint64_t value)
3338 {
3339 	int i;
3340 
3341 	if (value == 0)
3342 		return (0);
3343 	if (name[0] == '\0')
3344 		return (0);
3345 
3346 	for (i = 0; features_for_read[i] != NULL; i++) {
3347 		if (strcmp(name, features_for_read[i]) == 0)
3348 			return (0);
3349 	}
3350 	printf("ZFS: unsupported feature: %s\n", name);
3351 	return (EIO);
3352 }
3353 
3354 /*
3355  * Checks whether the MOS features that are active are supported.
3356  */
3357 static int
3358 check_mos_features(const spa_t *spa)
3359 {
3360 	dnode_phys_t dir;
3361 	zap_phys_t *zap;
3362 	uint64_t objnum;
3363 	size_t size;
3364 	int rc;
3365 
3366 	if ((rc = objset_get_dnode(spa, spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
3367 	    &dir)) != 0)
3368 		return (rc);
3369 	if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
3370 	    sizeof (objnum), 1, &objnum)) != 0) {
3371 		/*
3372 		 * It is older pool without features. As we have already
3373 		 * tested the label, just return without raising the error.
3374 		 */
3375 		return (0);
3376 	}
3377 
3378 	if ((rc = objset_get_dnode(spa, spa->spa_mos, objnum, &dir)) != 0)
3379 		return (rc);
3380 
3381 	if (dir.dn_type != DMU_OTN_ZAP_METADATA)
3382 		return (EIO);
3383 
3384 	size = dir.dn_datablkszsec << SPA_MINBLOCKSHIFT;
3385 	zap = malloc(size);
3386 	if (zap == NULL)
3387 		return (ENOMEM);
3388 
3389 	if (dnode_read(spa, &dir, 0, zap, size)) {
3390 		free(zap);
3391 		return (EIO);
3392 	}
3393 
3394 	if (zap->zap_block_type == ZBT_MICRO)
3395 		rc = mzap_list((const mzap_phys_t *)zap, size, check_feature);
3396 	else
3397 		rc = fzap_list(spa, &dir, zap, check_feature);
3398 
3399 	free(zap);
3400 	return (rc);
3401 }
3402 
3403 static int
3404 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
3405 {
3406 	dnode_phys_t dir;
3407 	size_t size;
3408 	int rc;
3409 	char *nv;
3410 
3411 	*value = NULL;
3412 	if ((rc = objset_get_dnode(spa, spa->spa_mos, obj, &dir)) != 0)
3413 		return (rc);
3414 	if (dir.dn_type != DMU_OT_PACKED_NVLIST &&
3415 	    dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) {
3416 		return (EIO);
3417 	}
3418 
3419 	if (dir.dn_bonuslen != sizeof (uint64_t))
3420 		return (EIO);
3421 
3422 	size = *(uint64_t *)DN_BONUS(&dir);
3423 	nv = malloc(size);
3424 	if (nv == NULL)
3425 		return (ENOMEM);
3426 
3427 	rc = dnode_read(spa, &dir, 0, nv, size);
3428 	if (rc != 0) {
3429 		free(nv);
3430 		nv = NULL;
3431 		return (rc);
3432 	}
3433 	*value = nvlist_import(nv, size);
3434 	free(nv);
3435 	return (rc);
3436 }
3437 
3438 static int
3439 zfs_spa_init(spa_t *spa)
3440 {
3441 	struct uberblock checkpoint;
3442 	dnode_phys_t dir;
3443 	uint64_t config_object;
3444 	nvlist_t *nvlist;
3445 	int rc;
3446 
3447 	if (zio_read(spa, &spa->spa_uberblock->ub_rootbp, spa->spa_mos)) {
3448 		printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
3449 		return (EIO);
3450 	}
3451 	if (spa->spa_mos->os_type != DMU_OST_META) {
3452 		printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
3453 		return (EIO);
3454 	}
3455 
3456 	if (objset_get_dnode(spa, &spa->spa_mos_master,
3457 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3458 		printf("ZFS: failed to read pool %s directory object\n",
3459 		    spa->spa_name);
3460 		return (EIO);
3461 	}
3462 	/* this is allowed to fail, older pools do not have salt */
3463 	rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
3464 	    sizeof (spa->spa_cksum_salt.zcs_bytes),
3465 	    spa->spa_cksum_salt.zcs_bytes);
3466 
3467 	rc = check_mos_features(spa);
3468 	if (rc != 0) {
3469 		printf("ZFS: pool %s is not supported\n", spa->spa_name);
3470 		return (rc);
3471 	}
3472 
3473 	rc = zap_lookup(spa, &dir, DMU_POOL_CONFIG,
3474 	    sizeof (config_object), 1, &config_object);
3475 	if (rc != 0) {
3476 		printf("ZFS: can not read MOS %s\n", DMU_POOL_CONFIG);
3477 		return (EIO);
3478 	}
3479 	rc = load_nvlist(spa, config_object, &nvlist);
3480 	if (rc != 0)
3481 		return (rc);
3482 
3483 	rc = zap_lookup(spa, &dir, DMU_POOL_ZPOOL_CHECKPOINT,
3484 	    sizeof(uint64_t), sizeof(checkpoint) / sizeof(uint64_t),
3485 	    &checkpoint);
3486 	if (rc == 0 && checkpoint.ub_checkpoint_txg != 0) {
3487 		memcpy(&spa->spa_uberblock_checkpoint, &checkpoint,
3488 		    sizeof(checkpoint));
3489 		if (zio_read(spa, &spa->spa_uberblock_checkpoint.ub_rootbp,
3490 		    &spa->spa_mos_checkpoint)) {
3491 			printf("ZFS: can not read checkpoint data.\n");
3492 			return (EIO);
3493 		}
3494 	}
3495 
3496 	/*
3497 	 * Update vdevs from MOS config. Note, we do skip encoding bytes
3498 	 * here. See also vdev_label_read_config().
3499 	 */
3500 	rc = vdev_init_from_nvlist(spa, nvlist);
3501 	nvlist_destroy(nvlist);
3502 	return (rc);
3503 }
3504 
3505 static int
3506 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
3507 {
3508 
3509 	if (dn->dn_bonustype != DMU_OT_SA) {
3510 		znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
3511 
3512 		sb->st_mode = zp->zp_mode;
3513 		sb->st_uid = zp->zp_uid;
3514 		sb->st_gid = zp->zp_gid;
3515 		sb->st_size = zp->zp_size;
3516 	} else {
3517 		sa_hdr_phys_t *sahdrp;
3518 		int hdrsize;
3519 		size_t size = 0;
3520 		void *buf = NULL;
3521 
3522 		if (dn->dn_bonuslen != 0)
3523 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3524 		else {
3525 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
3526 				blkptr_t *bp = DN_SPILL_BLKPTR(dn);
3527 				int error;
3528 
3529 				size = BP_GET_LSIZE(bp);
3530 				buf = malloc(size);
3531 				if (buf == NULL)
3532 					error = ENOMEM;
3533 				else
3534 					error = zio_read(spa, bp, buf);
3535 
3536 				if (error != 0) {
3537 					free(buf);
3538 					return (error);
3539 				}
3540 				sahdrp = buf;
3541 			} else {
3542 				return (EIO);
3543 			}
3544 		}
3545 		hdrsize = SA_HDR_SIZE(sahdrp);
3546 		sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
3547 		    SA_MODE_OFFSET);
3548 		sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
3549 		    SA_UID_OFFSET);
3550 		sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
3551 		    SA_GID_OFFSET);
3552 		sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
3553 		    SA_SIZE_OFFSET);
3554 		free(buf);
3555 	}
3556 
3557 	return (0);
3558 }
3559 
3560 static int
3561 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
3562 {
3563 	int rc = 0;
3564 
3565 	if (dn->dn_bonustype == DMU_OT_SA) {
3566 		sa_hdr_phys_t *sahdrp = NULL;
3567 		size_t size = 0;
3568 		void *buf = NULL;
3569 		int hdrsize;
3570 		char *p;
3571 
3572 		if (dn->dn_bonuslen != 0) {
3573 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3574 		} else {
3575 			blkptr_t *bp;
3576 
3577 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
3578 				return (EIO);
3579 			bp = DN_SPILL_BLKPTR(dn);
3580 
3581 			size = BP_GET_LSIZE(bp);
3582 			buf = malloc(size);
3583 			if (buf == NULL)
3584 				rc = ENOMEM;
3585 			else
3586 				rc = zio_read(spa, bp, buf);
3587 			if (rc != 0) {
3588 				free(buf);
3589 				return (rc);
3590 			}
3591 			sahdrp = buf;
3592 		}
3593 		hdrsize = SA_HDR_SIZE(sahdrp);
3594 		p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
3595 		memcpy(path, p, psize);
3596 		free(buf);
3597 		return (0);
3598 	}
3599 	/*
3600 	 * Second test is purely to silence bogus compiler
3601 	 * warning about accessing past the end of dn_bonus.
3602 	 */
3603 	if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
3604 	    sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
3605 		memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
3606 	} else {
3607 		rc = dnode_read(spa, dn, 0, path, psize);
3608 	}
3609 	return (rc);
3610 }
3611 
3612 struct obj_list {
3613 	uint64_t		objnum;
3614 	STAILQ_ENTRY(obj_list)	entry;
3615 };
3616 
3617 /*
3618  * Lookup a file and return its dnode.
3619  */
3620 static int
3621 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
3622 {
3623 	int rc;
3624 	uint64_t objnum;
3625 	const spa_t *spa;
3626 	dnode_phys_t dn;
3627 	const char *p, *q;
3628 	char element[256];
3629 	char path[1024];
3630 	int symlinks_followed = 0;
3631 	struct stat sb;
3632 	struct obj_list *entry, *tentry;
3633 	STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
3634 
3635 	spa = mount->spa;
3636 	if (mount->objset.os_type != DMU_OST_ZFS) {
3637 		printf("ZFS: unexpected object set type %ju\n",
3638 		    (uintmax_t)mount->objset.os_type);
3639 		return (EIO);
3640 	}
3641 
3642 	if ((entry = malloc(sizeof(struct obj_list))) == NULL)
3643 		return (ENOMEM);
3644 
3645 	/*
3646 	 * Get the root directory dnode.
3647 	 */
3648 	rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
3649 	if (rc) {
3650 		free(entry);
3651 		return (rc);
3652 	}
3653 
3654 	rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof(objnum), 1, &objnum);
3655 	if (rc) {
3656 		free(entry);
3657 		return (rc);
3658 	}
3659 	entry->objnum = objnum;
3660 	STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3661 
3662 	rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3663 	if (rc != 0)
3664 		goto done;
3665 
3666 	p = upath;
3667 	while (p && *p) {
3668 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3669 		if (rc != 0)
3670 			goto done;
3671 
3672 		while (*p == '/')
3673 			p++;
3674 		if (*p == '\0')
3675 			break;
3676 		q = p;
3677 		while (*q != '\0' && *q != '/')
3678 			q++;
3679 
3680 		/* skip dot */
3681 		if (p + 1 == q && p[0] == '.') {
3682 			p++;
3683 			continue;
3684 		}
3685 		/* double dot */
3686 		if (p + 2 == q && p[0] == '.' && p[1] == '.') {
3687 			p += 2;
3688 			if (STAILQ_FIRST(&on_cache) ==
3689 			    STAILQ_LAST(&on_cache, obj_list, entry)) {
3690 				rc = ENOENT;
3691 				goto done;
3692 			}
3693 			entry = STAILQ_FIRST(&on_cache);
3694 			STAILQ_REMOVE_HEAD(&on_cache, entry);
3695 			free(entry);
3696 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3697 			continue;
3698 		}
3699 		if (q - p + 1 > sizeof(element)) {
3700 			rc = ENAMETOOLONG;
3701 			goto done;
3702 		}
3703 		memcpy(element, p, q - p);
3704 		element[q - p] = 0;
3705 		p = q;
3706 
3707 		if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
3708 			goto done;
3709 		if (!S_ISDIR(sb.st_mode)) {
3710 			rc = ENOTDIR;
3711 			goto done;
3712 		}
3713 
3714 		rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum);
3715 		if (rc)
3716 			goto done;
3717 		objnum = ZFS_DIRENT_OBJ(objnum);
3718 
3719 		if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
3720 			rc = ENOMEM;
3721 			goto done;
3722 		}
3723 		entry->objnum = objnum;
3724 		STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3725 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3726 		if (rc)
3727 			goto done;
3728 
3729 		/*
3730 		 * Check for symlink.
3731 		 */
3732 		rc = zfs_dnode_stat(spa, &dn, &sb);
3733 		if (rc)
3734 			goto done;
3735 		if (S_ISLNK(sb.st_mode)) {
3736 			if (symlinks_followed > 10) {
3737 				rc = EMLINK;
3738 				goto done;
3739 			}
3740 			symlinks_followed++;
3741 
3742 			/*
3743 			 * Read the link value and copy the tail of our
3744 			 * current path onto the end.
3745 			 */
3746 			if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
3747 				rc = ENAMETOOLONG;
3748 				goto done;
3749 			}
3750 			strcpy(&path[sb.st_size], p);
3751 
3752 			rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
3753 			if (rc != 0)
3754 				goto done;
3755 
3756 			/*
3757 			 * Restart with the new path, starting either at
3758 			 * the root or at the parent depending whether or
3759 			 * not the link is relative.
3760 			 */
3761 			p = path;
3762 			if (*p == '/') {
3763 				while (STAILQ_FIRST(&on_cache) !=
3764 				    STAILQ_LAST(&on_cache, obj_list, entry)) {
3765 					entry = STAILQ_FIRST(&on_cache);
3766 					STAILQ_REMOVE_HEAD(&on_cache, entry);
3767 					free(entry);
3768 				}
3769 			} else {
3770 				entry = STAILQ_FIRST(&on_cache);
3771 				STAILQ_REMOVE_HEAD(&on_cache, entry);
3772 				free(entry);
3773 			}
3774 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3775 		}
3776 	}
3777 
3778 	*dnode = dn;
3779 done:
3780 	STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)
3781 		free(entry);
3782 	return (rc);
3783 }
3784