xref: /freebsd/stand/libsa/zfs/zfsimpl.c (revision f157ca4696f5922275d5d451736005b9332eb136)
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 <sys/endian.h>
35 #include <sys/stat.h>
36 #include <sys/stdint.h>
37 #include <sys/list.h>
38 #include <machine/_inttypes.h>
39 
40 #include "zfsimpl.h"
41 #include "zfssubr.c"
42 
43 
44 struct zfsmount {
45 	const spa_t	*spa;
46 	objset_phys_t	objset;
47 	uint64_t	rootobj;
48 };
49 static struct zfsmount zfsmount __unused;
50 
51 /*
52  * The indirect_child_t represents the vdev that we will read from, when we
53  * need to read all copies of the data (e.g. for scrub or reconstruction).
54  * For plain (non-mirror) top-level vdevs (i.e. is_vdev is not a mirror),
55  * ic_vdev is the same as is_vdev.  However, for mirror top-level vdevs,
56  * ic_vdev is a child of the mirror.
57  */
58 typedef struct indirect_child {
59 	void *ic_data;
60 	vdev_t *ic_vdev;
61 } indirect_child_t;
62 
63 /*
64  * The indirect_split_t represents one mapped segment of an i/o to the
65  * indirect vdev. For non-split (contiguously-mapped) blocks, there will be
66  * only one indirect_split_t, with is_split_offset==0 and is_size==io_size.
67  * For split blocks, there will be several of these.
68  */
69 typedef struct indirect_split {
70 	list_node_t is_node; /* link on iv_splits */
71 
72 	/*
73 	 * is_split_offset is the offset into the i/o.
74 	 * This is the sum of the previous splits' is_size's.
75 	 */
76 	uint64_t is_split_offset;
77 
78 	vdev_t *is_vdev; /* top-level vdev */
79 	uint64_t is_target_offset; /* offset on is_vdev */
80 	uint64_t is_size;
81 	int is_children; /* number of entries in is_child[] */
82 
83 	/*
84 	 * is_good_child is the child that we are currently using to
85 	 * attempt reconstruction.
86 	 */
87 	int is_good_child;
88 
89 	indirect_child_t is_child[1]; /* variable-length */
90 } indirect_split_t;
91 
92 /*
93  * The indirect_vsd_t is associated with each i/o to the indirect vdev.
94  * It is the "Vdev-Specific Data" in the zio_t's io_vsd.
95  */
96 typedef struct indirect_vsd {
97 	boolean_t iv_split_block;
98 	boolean_t iv_reconstruct;
99 
100 	list_t iv_splits; /* list of indirect_split_t's */
101 } indirect_vsd_t;
102 
103 /*
104  * List of all vdevs, chained through v_alllink.
105  */
106 static vdev_list_t zfs_vdevs;
107 
108 /*
109  * List of ZFS features supported for read
110  */
111 static const char *features_for_read[] = {
112 	"org.illumos:lz4_compress",
113 	"com.delphix:hole_birth",
114 	"com.delphix:extensible_dataset",
115 	"com.delphix:embedded_data",
116 	"org.open-zfs:large_blocks",
117 	"org.illumos:sha512",
118 	"org.illumos:skein",
119 	"org.zfsonlinux:large_dnode",
120 	"com.joyent:multi_vdev_crash_dump",
121 	"com.delphix:spacemap_histogram",
122 	"com.delphix:zpool_checkpoint",
123 	"com.delphix:spacemap_v2",
124 	"com.datto:encryption",
125 	"org.zfsonlinux:allocation_classes",
126 	"com.datto:resilver_defer",
127 	"com.delphix:device_removal",
128 	"com.delphix:obsolete_counts",
129 	"com.intel:allocation_classes",
130 	NULL
131 };
132 
133 /*
134  * List of all pools, chained through spa_link.
135  */
136 static spa_list_t zfs_pools;
137 
138 static const dnode_phys_t *dnode_cache_obj;
139 static uint64_t dnode_cache_bn;
140 static char *dnode_cache_buf;
141 static char *zap_scratch;
142 static char *zfs_temp_buf, *zfs_temp_end, *zfs_temp_ptr;
143 
144 #define	TEMP_SIZE	(1024 * 1024)
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 	zfs_temp_buf = malloc(TEMP_SIZE);
172 	zfs_temp_end = zfs_temp_buf + TEMP_SIZE;
173 	zfs_temp_ptr = zfs_temp_buf;
174 	dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
175 	zap_scratch = malloc(SPA_MAXBLOCKSIZE);
176 
177 	zfs_init_crc();
178 }
179 
180 static void *
181 zfs_alloc(size_t size)
182 {
183 	char *ptr;
184 
185 	if (zfs_temp_ptr + size > zfs_temp_end) {
186 		panic("ZFS: out of temporary buffer space");
187 	}
188 	ptr = zfs_temp_ptr;
189 	zfs_temp_ptr += size;
190 
191 	return (ptr);
192 }
193 
194 static void
195 zfs_free(void *ptr, size_t size)
196 {
197 
198 	zfs_temp_ptr -= size;
199 	if (zfs_temp_ptr != ptr) {
200 		panic("ZFS: zfs_alloc()/zfs_free() mismatch");
201 	}
202 }
203 
204 static int
205 xdr_int(const unsigned char **xdr, int *ip)
206 {
207 	*ip = be32dec(*xdr);
208 	(*xdr) += 4;
209 	return (0);
210 }
211 
212 static int
213 xdr_u_int(const unsigned char **xdr, u_int *ip)
214 {
215 	*ip = be32dec(*xdr);
216 	(*xdr) += 4;
217 	return (0);
218 }
219 
220 static int
221 xdr_uint64_t(const unsigned char **xdr, uint64_t *lp)
222 {
223 	u_int hi, lo;
224 
225 	xdr_u_int(xdr, &hi);
226 	xdr_u_int(xdr, &lo);
227 	*lp = (((uint64_t)hi) << 32) | lo;
228 	return (0);
229 }
230 
231 static int
232 nvlist_find(const unsigned char *nvlist, const char *name, int type,
233     int *elementsp, void *valuep)
234 {
235 	const unsigned char *p, *pair;
236 	int junk;
237 	int encoded_size, decoded_size;
238 
239 	p = nvlist;
240 	xdr_int(&p, &junk);
241 	xdr_int(&p, &junk);
242 
243 	pair = p;
244 	xdr_int(&p, &encoded_size);
245 	xdr_int(&p, &decoded_size);
246 	while (encoded_size && decoded_size) {
247 		int namelen, pairtype, elements;
248 		const char *pairname;
249 
250 		xdr_int(&p, &namelen);
251 		pairname = (const char *)p;
252 		p += roundup(namelen, 4);
253 		xdr_int(&p, &pairtype);
254 
255 		if (memcmp(name, pairname, namelen) == 0 && type == pairtype) {
256 			xdr_int(&p, &elements);
257 			if (elementsp)
258 				*elementsp = elements;
259 			if (type == DATA_TYPE_UINT64) {
260 				xdr_uint64_t(&p, (uint64_t *)valuep);
261 				return (0);
262 			} else if (type == DATA_TYPE_STRING) {
263 				int len;
264 				xdr_int(&p, &len);
265 				(*(const char **)valuep) = (const char *)p;
266 				return (0);
267 			} else if (type == DATA_TYPE_NVLIST ||
268 			    type == DATA_TYPE_NVLIST_ARRAY) {
269 				(*(const unsigned char **)valuep) =
270 				    (const unsigned char *)p;
271 				return (0);
272 			} else {
273 				return (EIO);
274 			}
275 		} else {
276 			/*
277 			 * Not the pair we are looking for, skip to the
278 			 * next one.
279 			 */
280 			p = pair + encoded_size;
281 		}
282 
283 		pair = p;
284 		xdr_int(&p, &encoded_size);
285 		xdr_int(&p, &decoded_size);
286 	}
287 
288 	return (EIO);
289 }
290 
291 static int
292 nvlist_check_features_for_read(const unsigned char *nvlist)
293 {
294 	const unsigned char *p, *pair;
295 	int junk;
296 	int encoded_size, decoded_size;
297 	int rc;
298 
299 	rc = 0;
300 
301 	p = nvlist;
302 	xdr_int(&p, &junk);
303 	xdr_int(&p, &junk);
304 
305 	pair = p;
306 	xdr_int(&p, &encoded_size);
307 	xdr_int(&p, &decoded_size);
308 	while (encoded_size && decoded_size) {
309 		int namelen, pairtype;
310 		const char *pairname;
311 		int i, found;
312 
313 		found = 0;
314 
315 		xdr_int(&p, &namelen);
316 		pairname = (const char *)p;
317 		p += roundup(namelen, 4);
318 		xdr_int(&p, &pairtype);
319 
320 		for (i = 0; features_for_read[i] != NULL; i++) {
321 			if (memcmp(pairname, features_for_read[i],
322 			    namelen) == 0) {
323 				found = 1;
324 				break;
325 			}
326 		}
327 
328 		if (!found) {
329 			printf("ZFS: unsupported feature: %s\n", pairname);
330 			rc = EIO;
331 		}
332 
333 		p = pair + encoded_size;
334 
335 		pair = p;
336 		xdr_int(&p, &encoded_size);
337 		xdr_int(&p, &decoded_size);
338 	}
339 
340 	return (rc);
341 }
342 
343 /*
344  * Return the next nvlist in an nvlist array.
345  */
346 static const unsigned char *
347 nvlist_next(const unsigned char *nvlist)
348 {
349 	const unsigned char *p, *pair;
350 	int junk;
351 	int encoded_size, decoded_size;
352 
353 	p = nvlist;
354 	xdr_int(&p, &junk);
355 	xdr_int(&p, &junk);
356 
357 	pair = p;
358 	xdr_int(&p, &encoded_size);
359 	xdr_int(&p, &decoded_size);
360 	while (encoded_size && decoded_size) {
361 		p = pair + encoded_size;
362 
363 		pair = p;
364 		xdr_int(&p, &encoded_size);
365 		xdr_int(&p, &decoded_size);
366 	}
367 
368 	return (p);
369 }
370 
371 #ifdef TEST
372 
373 static const unsigned char *
374 nvlist_print(const unsigned char *nvlist, unsigned int indent)
375 {
376 	static const char *typenames[] = {
377 		"DATA_TYPE_UNKNOWN",
378 		"DATA_TYPE_BOOLEAN",
379 		"DATA_TYPE_BYTE",
380 		"DATA_TYPE_INT16",
381 		"DATA_TYPE_UINT16",
382 		"DATA_TYPE_INT32",
383 		"DATA_TYPE_UINT32",
384 		"DATA_TYPE_INT64",
385 		"DATA_TYPE_UINT64",
386 		"DATA_TYPE_STRING",
387 		"DATA_TYPE_BYTE_ARRAY",
388 		"DATA_TYPE_INT16_ARRAY",
389 		"DATA_TYPE_UINT16_ARRAY",
390 		"DATA_TYPE_INT32_ARRAY",
391 		"DATA_TYPE_UINT32_ARRAY",
392 		"DATA_TYPE_INT64_ARRAY",
393 		"DATA_TYPE_UINT64_ARRAY",
394 		"DATA_TYPE_STRING_ARRAY",
395 		"DATA_TYPE_HRTIME",
396 		"DATA_TYPE_NVLIST",
397 		"DATA_TYPE_NVLIST_ARRAY",
398 		"DATA_TYPE_BOOLEAN_VALUE",
399 		"DATA_TYPE_INT8",
400 		"DATA_TYPE_UINT8",
401 		"DATA_TYPE_BOOLEAN_ARRAY",
402 		"DATA_TYPE_INT8_ARRAY",
403 		"DATA_TYPE_UINT8_ARRAY"
404 	};
405 
406 	unsigned int i, j;
407 	const unsigned char *p, *pair;
408 	int junk;
409 	int encoded_size, decoded_size;
410 
411 	p = nvlist;
412 	xdr_int(&p, &junk);
413 	xdr_int(&p, &junk);
414 
415 	pair = p;
416 	xdr_int(&p, &encoded_size);
417 	xdr_int(&p, &decoded_size);
418 	while (encoded_size && decoded_size) {
419 		int namelen, pairtype, elements;
420 		const char *pairname;
421 
422 		xdr_int(&p, &namelen);
423 		pairname = (const char *)p;
424 		p += roundup(namelen, 4);
425 		xdr_int(&p, &pairtype);
426 
427 		for (i = 0; i < indent; i++)
428 			printf(" ");
429 		printf("%s %s", typenames[pairtype], pairname);
430 
431 		xdr_int(&p, &elements);
432 		switch (pairtype) {
433 		case DATA_TYPE_UINT64: {
434 			uint64_t val;
435 			xdr_uint64_t(&p, &val);
436 			printf(" = 0x%jx\n", (uintmax_t)val);
437 			break;
438 		}
439 
440 		case DATA_TYPE_STRING: {
441 			int len;
442 			xdr_int(&p, &len);
443 			printf(" = \"%s\"\n", p);
444 			break;
445 		}
446 
447 		case DATA_TYPE_NVLIST:
448 			printf("\n");
449 			nvlist_print(p, indent + 1);
450 			break;
451 
452 		case DATA_TYPE_NVLIST_ARRAY:
453 			for (j = 0; j < elements; j++) {
454 				printf("[%d]\n", j);
455 				p = nvlist_print(p, indent + 1);
456 				if (j != elements - 1) {
457 					for (i = 0; i < indent; i++)
458 						printf(" ");
459 					printf("%s %s", typenames[pairtype],
460 					    pairname);
461 				}
462 			}
463 			break;
464 
465 		default:
466 			printf("\n");
467 		}
468 
469 		p = pair + encoded_size;
470 
471 		pair = p;
472 		xdr_int(&p, &encoded_size);
473 		xdr_int(&p, &decoded_size);
474 	}
475 
476 	return (p);
477 }
478 
479 #endif
480 
481 static int
482 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
483     off_t offset, size_t size)
484 {
485 	size_t psize;
486 	int rc;
487 
488 	if (!vdev->v_phys_read)
489 		return (EIO);
490 
491 	if (bp) {
492 		psize = BP_GET_PSIZE(bp);
493 	} else {
494 		psize = size;
495 	}
496 
497 	rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
498 	if (rc == 0) {
499 		if (bp != NULL)
500 			rc = zio_checksum_verify(vdev->v_spa, bp, buf);
501 	}
502 
503 	return (rc);
504 }
505 
506 typedef struct remap_segment {
507 	vdev_t *rs_vd;
508 	uint64_t rs_offset;
509 	uint64_t rs_asize;
510 	uint64_t rs_split_offset;
511 	list_node_t rs_node;
512 } remap_segment_t;
513 
514 static remap_segment_t *
515 rs_alloc(vdev_t *vd, uint64_t offset, uint64_t asize, uint64_t split_offset)
516 {
517 	remap_segment_t *rs = malloc(sizeof (remap_segment_t));
518 
519 	if (rs != NULL) {
520 		rs->rs_vd = vd;
521 		rs->rs_offset = offset;
522 		rs->rs_asize = asize;
523 		rs->rs_split_offset = split_offset;
524 	}
525 
526 	return (rs);
527 }
528 
529 vdev_indirect_mapping_t *
530 vdev_indirect_mapping_open(spa_t *spa, objset_phys_t *os,
531     uint64_t mapping_object)
532 {
533 	vdev_indirect_mapping_t *vim;
534 	vdev_indirect_mapping_phys_t *vim_phys;
535 	int rc;
536 
537 	vim = calloc(1, sizeof (*vim));
538 	if (vim == NULL)
539 		return (NULL);
540 
541 	vim->vim_dn = calloc(1, sizeof (*vim->vim_dn));
542 	if (vim->vim_dn == NULL) {
543 		free(vim);
544 		return (NULL);
545 	}
546 
547 	rc = objset_get_dnode(spa, os, mapping_object, vim->vim_dn);
548 	if (rc != 0) {
549 		free(vim->vim_dn);
550 		free(vim);
551 		return (NULL);
552 	}
553 
554 	vim->vim_spa = spa;
555 	vim->vim_phys = malloc(sizeof (*vim->vim_phys));
556 	if (vim->vim_phys == NULL) {
557 		free(vim->vim_dn);
558 		free(vim);
559 		return (NULL);
560 	}
561 
562 	vim_phys = (vdev_indirect_mapping_phys_t *)DN_BONUS(vim->vim_dn);
563 	*vim->vim_phys = *vim_phys;
564 
565 	vim->vim_objset = os;
566 	vim->vim_object = mapping_object;
567 	vim->vim_entries = NULL;
568 
569 	vim->vim_havecounts =
570 	    (vim->vim_dn->dn_bonuslen > VDEV_INDIRECT_MAPPING_SIZE_V0);
571 
572 	return (vim);
573 }
574 
575 /*
576  * Compare an offset with an indirect mapping entry; there are three
577  * possible scenarios:
578  *
579  *     1. The offset is "less than" the mapping entry; meaning the
580  *        offset is less than the source offset of the mapping entry. In
581  *        this case, there is no overlap between the offset and the
582  *        mapping entry and -1 will be returned.
583  *
584  *     2. The offset is "greater than" the mapping entry; meaning the
585  *        offset is greater than the mapping entry's source offset plus
586  *        the entry's size. In this case, there is no overlap between
587  *        the offset and the mapping entry and 1 will be returned.
588  *
589  *        NOTE: If the offset is actually equal to the entry's offset
590  *        plus size, this is considered to be "greater" than the entry,
591  *        and this case applies (i.e. 1 will be returned). Thus, the
592  *        entry's "range" can be considered to be inclusive at its
593  *        start, but exclusive at its end: e.g. [src, src + size).
594  *
595  *     3. The last case to consider is if the offset actually falls
596  *        within the mapping entry's range. If this is the case, the
597  *        offset is considered to be "equal to" the mapping entry and
598  *        0 will be returned.
599  *
600  *        NOTE: If the offset is equal to the entry's source offset,
601  *        this case applies and 0 will be returned. If the offset is
602  *        equal to the entry's source plus its size, this case does
603  *        *not* apply (see "NOTE" above for scenario 2), and 1 will be
604  *        returned.
605  */
606 static int
607 dva_mapping_overlap_compare(const void *v_key, const void *v_array_elem)
608 {
609 	const uint64_t *key = v_key;
610 	const vdev_indirect_mapping_entry_phys_t *array_elem =
611 	    v_array_elem;
612 	uint64_t src_offset = DVA_MAPPING_GET_SRC_OFFSET(array_elem);
613 
614 	if (*key < src_offset) {
615 		return (-1);
616 	} else if (*key < src_offset + DVA_GET_ASIZE(&array_elem->vimep_dst)) {
617 		return (0);
618 	} else {
619 		return (1);
620 	}
621 }
622 
623 /*
624  * Return array entry.
625  */
626 static vdev_indirect_mapping_entry_phys_t *
627 vdev_indirect_mapping_entry(vdev_indirect_mapping_t *vim, uint64_t index)
628 {
629 	uint64_t size;
630 	off_t offset = 0;
631 	int rc;
632 
633 	if (vim->vim_phys->vimp_num_entries == 0)
634 		return (NULL);
635 
636 	if (vim->vim_entries == NULL) {
637 		uint64_t bsize;
638 
639 		bsize = vim->vim_dn->dn_datablkszsec << SPA_MINBLOCKSHIFT;
640 		size = vim->vim_phys->vimp_num_entries *
641 		    sizeof (*vim->vim_entries);
642 		if (size > bsize) {
643 			size = bsize / sizeof (*vim->vim_entries);
644 			size *= sizeof (*vim->vim_entries);
645 		}
646 		vim->vim_entries = malloc(size);
647 		if (vim->vim_entries == NULL)
648 			return (NULL);
649 		vim->vim_num_entries = size / sizeof (*vim->vim_entries);
650 		offset = index * sizeof (*vim->vim_entries);
651 	}
652 
653 	/* We have data in vim_entries */
654 	if (offset == 0) {
655 		if (index >= vim->vim_entry_offset &&
656 		    index <= vim->vim_entry_offset + vim->vim_num_entries) {
657 			index -= vim->vim_entry_offset;
658 			return (&vim->vim_entries[index]);
659 		}
660 		offset = index * sizeof (*vim->vim_entries);
661 	}
662 
663 	vim->vim_entry_offset = index;
664 	size = vim->vim_num_entries * sizeof (*vim->vim_entries);
665 	rc = dnode_read(vim->vim_spa, vim->vim_dn, offset, vim->vim_entries,
666 	    size);
667 	if (rc != 0) {
668 		/* Read error, invalidate vim_entries. */
669 		free(vim->vim_entries);
670 		vim->vim_entries = NULL;
671 		return (NULL);
672 	}
673 	index -= vim->vim_entry_offset;
674 	return (&vim->vim_entries[index]);
675 }
676 
677 /*
678  * Returns the mapping entry for the given offset.
679  *
680  * It's possible that the given offset will not be in the mapping table
681  * (i.e. no mapping entries contain this offset), in which case, the
682  * return value value depends on the "next_if_missing" parameter.
683  *
684  * If the offset is not found in the table and "next_if_missing" is
685  * B_FALSE, then NULL will always be returned. The behavior is intended
686  * to allow consumers to get the entry corresponding to the offset
687  * parameter, iff the offset overlaps with an entry in the table.
688  *
689  * If the offset is not found in the table and "next_if_missing" is
690  * B_TRUE, then the entry nearest to the given offset will be returned,
691  * such that the entry's source offset is greater than the offset
692  * passed in (i.e. the "next" mapping entry in the table is returned, if
693  * the offset is missing from the table). If there are no entries whose
694  * source offset is greater than the passed in offset, NULL is returned.
695  */
696 static vdev_indirect_mapping_entry_phys_t *
697 vdev_indirect_mapping_entry_for_offset(vdev_indirect_mapping_t *vim,
698     uint64_t offset)
699 {
700 	ASSERT(vim->vim_phys->vimp_num_entries > 0);
701 
702 	vdev_indirect_mapping_entry_phys_t *entry;
703 
704 	uint64_t last = vim->vim_phys->vimp_num_entries - 1;
705 	uint64_t base = 0;
706 
707 	/*
708 	 * We don't define these inside of the while loop because we use
709 	 * their value in the case that offset isn't in the mapping.
710 	 */
711 	uint64_t mid;
712 	int result;
713 
714 	while (last >= base) {
715 		mid = base + ((last - base) >> 1);
716 
717 		entry = vdev_indirect_mapping_entry(vim, mid);
718 		if (entry == NULL)
719 			break;
720 		result = dva_mapping_overlap_compare(&offset, entry);
721 
722 		if (result == 0) {
723 			break;
724 		} else if (result < 0) {
725 			last = mid - 1;
726 		} else {
727 			base = mid + 1;
728 		}
729 	}
730 	return (entry);
731 }
732 
733 /*
734  * Given an indirect vdev and an extent on that vdev, it duplicates the
735  * physical entries of the indirect mapping that correspond to the extent
736  * to a new array and returns a pointer to it. In addition, copied_entries
737  * is populated with the number of mapping entries that were duplicated.
738  *
739  * Finally, since we are doing an allocation, it is up to the caller to
740  * free the array allocated in this function.
741  */
742 vdev_indirect_mapping_entry_phys_t *
743 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *vd, uint64_t offset,
744     uint64_t asize, uint64_t *copied_entries)
745 {
746 	vdev_indirect_mapping_entry_phys_t *duplicate_mappings = NULL;
747 	vdev_indirect_mapping_t *vim = vd->v_mapping;
748 	uint64_t entries = 0;
749 
750 	vdev_indirect_mapping_entry_phys_t *first_mapping =
751 	    vdev_indirect_mapping_entry_for_offset(vim, offset);
752 	ASSERT3P(first_mapping, !=, NULL);
753 
754 	vdev_indirect_mapping_entry_phys_t *m = first_mapping;
755 	while (asize > 0) {
756 		uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
757 		uint64_t inner_offset = offset - DVA_MAPPING_GET_SRC_OFFSET(m);
758 		uint64_t inner_size = MIN(asize, size - inner_offset);
759 
760 		offset += inner_size;
761 		asize -= inner_size;
762 		entries++;
763 		m++;
764 	}
765 
766 	size_t copy_length = entries * sizeof (*first_mapping);
767 	duplicate_mappings = malloc(copy_length);
768 	if (duplicate_mappings != NULL)
769 		bcopy(first_mapping, duplicate_mappings, copy_length);
770 	else
771 		entries = 0;
772 
773 	*copied_entries = entries;
774 
775 	return (duplicate_mappings);
776 }
777 
778 static vdev_t *
779 vdev_lookup_top(spa_t *spa, uint64_t vdev)
780 {
781 	vdev_t *rvd;
782 	vdev_list_t *vlist;
783 
784 	vlist = &spa->spa_root_vdev->v_children;
785 	STAILQ_FOREACH(rvd, vlist, v_childlink)
786 		if (rvd->v_id == vdev)
787 			break;
788 
789 	return (rvd);
790 }
791 
792 /*
793  * This is a callback for vdev_indirect_remap() which allocates an
794  * indirect_split_t for each split segment and adds it to iv_splits.
795  */
796 static void
797 vdev_indirect_gather_splits(uint64_t split_offset, vdev_t *vd, uint64_t offset,
798     uint64_t size, void *arg)
799 {
800 	int n = 1;
801 	zio_t *zio = arg;
802 	indirect_vsd_t *iv = zio->io_vsd;
803 
804 	if (vd->v_read == vdev_indirect_read)
805 		return;
806 
807 	if (vd->v_read == vdev_mirror_read)
808 		n = vd->v_nchildren;
809 
810 	indirect_split_t *is =
811 	    malloc(offsetof(indirect_split_t, is_child[n]));
812 	if (is == NULL) {
813 		zio->io_error = ENOMEM;
814 		return;
815 	}
816 	bzero(is, offsetof(indirect_split_t, is_child[n]));
817 
818 	is->is_children = n;
819 	is->is_size = size;
820 	is->is_split_offset = split_offset;
821 	is->is_target_offset = offset;
822 	is->is_vdev = vd;
823 
824 	/*
825 	 * Note that we only consider multiple copies of the data for
826 	 * *mirror* vdevs.  We don't for "replacing" or "spare" vdevs, even
827 	 * though they use the same ops as mirror, because there's only one
828 	 * "good" copy under the replacing/spare.
829 	 */
830 	if (vd->v_read == vdev_mirror_read) {
831 		int i = 0;
832 		vdev_t *kid;
833 
834 		STAILQ_FOREACH(kid, &vd->v_children, v_childlink) {
835 			is->is_child[i++].ic_vdev = kid;
836 		}
837 	} else {
838 		is->is_child[0].ic_vdev = vd;
839 	}
840 
841 	list_insert_tail(&iv->iv_splits, is);
842 }
843 
844 static void
845 vdev_indirect_remap(vdev_t *vd, uint64_t offset, uint64_t asize, void *arg)
846 {
847 	list_t stack;
848 	spa_t *spa = vd->v_spa;
849 	zio_t *zio = arg;
850 	remap_segment_t *rs;
851 
852 	list_create(&stack, sizeof (remap_segment_t),
853 	    offsetof(remap_segment_t, rs_node));
854 
855 	rs = rs_alloc(vd, offset, asize, 0);
856 	if (rs == NULL) {
857 		printf("vdev_indirect_remap: out of memory.\n");
858 		zio->io_error = ENOMEM;
859 	}
860 	for (; rs != NULL; rs = list_remove_head(&stack)) {
861 		vdev_t *v = rs->rs_vd;
862 		uint64_t num_entries = 0;
863 		/* vdev_indirect_mapping_t *vim = v->v_mapping; */
864 		vdev_indirect_mapping_entry_phys_t *mapping =
865 		    vdev_indirect_mapping_duplicate_adjacent_entries(v,
866 		    rs->rs_offset, rs->rs_asize, &num_entries);
867 
868 		if (num_entries == 0)
869 			zio->io_error = ENOMEM;
870 
871 		for (uint64_t i = 0; i < num_entries; i++) {
872 			vdev_indirect_mapping_entry_phys_t *m = &mapping[i];
873 			uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
874 			uint64_t dst_offset = DVA_GET_OFFSET(&m->vimep_dst);
875 			uint64_t dst_vdev = DVA_GET_VDEV(&m->vimep_dst);
876 			uint64_t inner_offset = rs->rs_offset -
877 			    DVA_MAPPING_GET_SRC_OFFSET(m);
878 			uint64_t inner_size =
879 			    MIN(rs->rs_asize, size - inner_offset);
880 			vdev_t *dst_v = vdev_lookup_top(spa, dst_vdev);
881 
882 			if (dst_v->v_read == vdev_indirect_read) {
883 				remap_segment_t *o;
884 
885 				o = rs_alloc(dst_v, dst_offset + inner_offset,
886 				    inner_size, rs->rs_split_offset);
887 				if (o == NULL) {
888 					printf("vdev_indirect_remap: "
889 					    "out of memory.\n");
890 					zio->io_error = ENOMEM;
891 					break;
892 				}
893 
894 				list_insert_head(&stack, o);
895 			}
896 			vdev_indirect_gather_splits(rs->rs_split_offset, dst_v,
897 			    dst_offset + inner_offset,
898 			    inner_size, arg);
899 
900 			/*
901 			 * vdev_indirect_gather_splits can have memory
902 			 * allocation error, we can not recover from it.
903 			 */
904 			if (zio->io_error != 0)
905 				break;
906 			rs->rs_offset += inner_size;
907 			rs->rs_asize -= inner_size;
908 			rs->rs_split_offset += inner_size;
909 		}
910 
911 		free(mapping);
912 		free(rs);
913 		if (zio->io_error != 0)
914 			break;
915 	}
916 
917 	list_destroy(&stack);
918 }
919 
920 static void
921 vdev_indirect_map_free(zio_t *zio)
922 {
923 	indirect_vsd_t *iv = zio->io_vsd;
924 	indirect_split_t *is;
925 
926 	while ((is = list_head(&iv->iv_splits)) != NULL) {
927 		for (int c = 0; c < is->is_children; c++) {
928 			indirect_child_t *ic = &is->is_child[c];
929 			free(ic->ic_data);
930 		}
931 		list_remove(&iv->iv_splits, is);
932 		free(is);
933 	}
934 	free(iv);
935 }
936 
937 static int
938 vdev_indirect_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
939     off_t offset, size_t bytes)
940 {
941 	zio_t zio;
942 	spa_t *spa = vdev->v_spa;
943 	indirect_vsd_t *iv;
944 	indirect_split_t *first;
945 	int rc = EIO;
946 
947 	iv = calloc(1, sizeof(*iv));
948 	if (iv == NULL)
949 		return (ENOMEM);
950 
951 	list_create(&iv->iv_splits,
952 	    sizeof (indirect_split_t), offsetof(indirect_split_t, is_node));
953 
954 	bzero(&zio, sizeof(zio));
955 	zio.io_spa = spa;
956 	zio.io_bp = (blkptr_t *)bp;
957 	zio.io_data = buf;
958 	zio.io_size = bytes;
959 	zio.io_offset = offset;
960 	zio.io_vd = vdev;
961 	zio.io_vsd = iv;
962 
963 	if (vdev->v_mapping == NULL) {
964 		vdev_indirect_config_t *vic;
965 
966 		vic = &vdev->vdev_indirect_config;
967 		vdev->v_mapping = vdev_indirect_mapping_open(spa,
968 		    &spa->spa_mos, vic->vic_mapping_object);
969 	}
970 
971 	vdev_indirect_remap(vdev, offset, bytes, &zio);
972 	if (zio.io_error != 0)
973 		return (zio.io_error);
974 
975 	first = list_head(&iv->iv_splits);
976 	if (first->is_size == zio.io_size) {
977 		/*
978 		 * This is not a split block; we are pointing to the entire
979 		 * data, which will checksum the same as the original data.
980 		 * Pass the BP down so that the child i/o can verify the
981 		 * checksum, and try a different location if available
982 		 * (e.g. on a mirror).
983 		 *
984 		 * While this special case could be handled the same as the
985 		 * general (split block) case, doing it this way ensures
986 		 * that the vast majority of blocks on indirect vdevs
987 		 * (which are not split) are handled identically to blocks
988 		 * on non-indirect vdevs.  This allows us to be less strict
989 		 * about performance in the general (but rare) case.
990 		 */
991 		rc = first->is_vdev->v_read(first->is_vdev, zio.io_bp,
992 		    zio.io_data, first->is_target_offset, bytes);
993 	} else {
994 		iv->iv_split_block = B_TRUE;
995 		/*
996 		 * Read one copy of each split segment, from the
997 		 * top-level vdev.  Since we don't know the
998 		 * checksum of each split individually, the child
999 		 * zio can't ensure that we get the right data.
1000 		 * E.g. if it's a mirror, it will just read from a
1001 		 * random (healthy) leaf vdev.  We have to verify
1002 		 * the checksum in vdev_indirect_io_done().
1003 		 */
1004 		for (indirect_split_t *is = list_head(&iv->iv_splits);
1005 		    is != NULL; is = list_next(&iv->iv_splits, is)) {
1006 			char *ptr = zio.io_data;
1007 
1008 			rc = is->is_vdev->v_read(is->is_vdev, zio.io_bp,
1009 			    ptr + is->is_split_offset, is->is_target_offset,
1010 			    is->is_size);
1011 		}
1012 		if (zio_checksum_verify(spa, zio.io_bp, zio.io_data))
1013 			rc = ECKSUM;
1014 		else
1015 			rc = 0;
1016 	}
1017 
1018 	vdev_indirect_map_free(&zio);
1019 	if (rc == 0)
1020 		rc = zio.io_error;
1021 
1022 	return (rc);
1023 }
1024 
1025 static int
1026 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
1027     off_t offset, size_t bytes)
1028 {
1029 
1030 	return (vdev_read_phys(vdev, bp, buf,
1031 	    offset + VDEV_LABEL_START_SIZE, bytes));
1032 }
1033 
1034 
1035 static int
1036 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
1037     off_t offset, size_t bytes)
1038 {
1039 	vdev_t *kid;
1040 	int rc;
1041 
1042 	rc = EIO;
1043 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1044 		if (kid->v_state != VDEV_STATE_HEALTHY)
1045 			continue;
1046 		rc = kid->v_read(kid, bp, buf, offset, bytes);
1047 		if (!rc)
1048 			return (0);
1049 	}
1050 
1051 	return (rc);
1052 }
1053 
1054 static int
1055 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
1056     off_t offset, size_t bytes)
1057 {
1058 	vdev_t *kid;
1059 
1060 	/*
1061 	 * Here we should have two kids:
1062 	 * First one which is the one we are replacing and we can trust
1063 	 * only this one to have valid data, but it might not be present.
1064 	 * Second one is that one we are replacing with. It is most likely
1065 	 * healthy, but we can't trust it has needed data, so we won't use it.
1066 	 */
1067 	kid = STAILQ_FIRST(&vdev->v_children);
1068 	if (kid == NULL)
1069 		return (EIO);
1070 	if (kid->v_state != VDEV_STATE_HEALTHY)
1071 		return (EIO);
1072 	return (kid->v_read(kid, bp, buf, offset, bytes));
1073 }
1074 
1075 static vdev_t *
1076 vdev_find(uint64_t guid)
1077 {
1078 	vdev_t *vdev;
1079 
1080 	STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
1081 		if (vdev->v_guid == guid)
1082 			return (vdev);
1083 
1084 	return (0);
1085 }
1086 
1087 static vdev_t *
1088 vdev_create(uint64_t guid, vdev_read_t *_read)
1089 {
1090 	vdev_t *vdev;
1091 	vdev_indirect_config_t *vic;
1092 
1093 	vdev = calloc(1, sizeof(vdev_t));
1094 	if (vdev != NULL) {
1095 		STAILQ_INIT(&vdev->v_children);
1096 		vdev->v_guid = guid;
1097 		vdev->v_read = _read;
1098 
1099 		/*
1100 		 * root vdev has no read function.
1101 		 * We only point root vdev from spa.
1102 		 */
1103 		if (_read != NULL) {
1104 			vic = &vdev->vdev_indirect_config;
1105 			vic->vic_prev_indirect_vdev = UINT64_MAX;
1106 			STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
1107 		}
1108 	}
1109 
1110 	return (vdev);
1111 }
1112 
1113 static void
1114 vdev_set_initial_state(vdev_t *vdev, const unsigned char *nvlist)
1115 {
1116 	uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
1117 	uint64_t is_log;
1118 
1119 	is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
1120 	is_log = 0;
1121 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
1122 	    &is_offline);
1123 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
1124 	    &is_removed);
1125 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
1126 	    &is_faulted);
1127 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64,
1128 	    NULL, &is_degraded);
1129 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64,
1130 	    NULL, &isnt_present);
1131 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, NULL,
1132 	    &is_log);
1133 
1134 	if (is_offline != 0)
1135 		vdev->v_state = VDEV_STATE_OFFLINE;
1136 	else if (is_removed != 0)
1137 		vdev->v_state = VDEV_STATE_REMOVED;
1138 	else if (is_faulted != 0)
1139 		vdev->v_state = VDEV_STATE_FAULTED;
1140 	else if (is_degraded != 0)
1141 		vdev->v_state = VDEV_STATE_DEGRADED;
1142 	else if (isnt_present != 0)
1143 		vdev->v_state = VDEV_STATE_CANT_OPEN;
1144 
1145 	vdev->v_islog = is_log == 1;
1146 }
1147 
1148 static int
1149 vdev_init(uint64_t guid, const unsigned char *nvlist, vdev_t **vdevp)
1150 {
1151 	uint64_t id, ashift, asize, nparity;
1152 	const char *path;
1153 	const char *type;
1154 	vdev_t *vdev;
1155 
1156 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id) ||
1157 	    nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
1158 	    NULL, &type)) {
1159 		return (ENOENT);
1160 	}
1161 
1162 	if (strcmp(type, VDEV_TYPE_MIRROR) != 0 &&
1163 	    strcmp(type, VDEV_TYPE_DISK) != 0 &&
1164 #ifdef ZFS_TEST
1165 	    strcmp(type, VDEV_TYPE_FILE) != 0 &&
1166 #endif
1167 	    strcmp(type, VDEV_TYPE_RAIDZ) != 0 &&
1168 	    strcmp(type, VDEV_TYPE_INDIRECT) != 0 &&
1169 	    strcmp(type, VDEV_TYPE_REPLACING) != 0) {
1170 		printf("ZFS: can only boot from disk, mirror, raidz1, "
1171 		    "raidz2 and raidz3 vdevs\n");
1172 		return (EIO);
1173 	}
1174 
1175 	if (strcmp(type, VDEV_TYPE_MIRROR) == 0)
1176 		vdev = vdev_create(guid, vdev_mirror_read);
1177 	else if (strcmp(type, VDEV_TYPE_RAIDZ) == 0)
1178 		vdev = vdev_create(guid, vdev_raidz_read);
1179 	else if (strcmp(type, VDEV_TYPE_REPLACING) == 0)
1180 		vdev = vdev_create(guid, vdev_replacing_read);
1181 	else if (strcmp(type, VDEV_TYPE_INDIRECT) == 0) {
1182 		vdev_indirect_config_t *vic;
1183 
1184 		vdev = vdev_create(guid, vdev_indirect_read);
1185 		if (vdev != NULL) {
1186 			vdev->v_state = VDEV_STATE_HEALTHY;
1187 			vic = &vdev->vdev_indirect_config;
1188 
1189 			nvlist_find(nvlist,
1190 			    ZPOOL_CONFIG_INDIRECT_OBJECT,
1191 			    DATA_TYPE_UINT64,
1192 			    NULL, &vic->vic_mapping_object);
1193 			nvlist_find(nvlist,
1194 			    ZPOOL_CONFIG_INDIRECT_BIRTHS,
1195 			    DATA_TYPE_UINT64,
1196 			    NULL, &vic->vic_births_object);
1197 			nvlist_find(nvlist,
1198 			    ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
1199 			    DATA_TYPE_UINT64,
1200 			    NULL, &vic->vic_prev_indirect_vdev);
1201 		}
1202 	} else {
1203 		vdev = vdev_create(guid, vdev_disk_read);
1204 	}
1205 
1206 	if (vdev == NULL)
1207 		return (ENOMEM);
1208 
1209 	vdev_set_initial_state(vdev, nvlist);
1210 	vdev->v_id = id;
1211 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
1212 	    DATA_TYPE_UINT64, NULL, &ashift) == 0)
1213 		vdev->v_ashift = ashift;
1214 
1215 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE,
1216 	    DATA_TYPE_UINT64, NULL, &asize) == 0) {
1217 		vdev->v_psize = asize +
1218 		    VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
1219 	}
1220 
1221 	if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
1222 	    DATA_TYPE_UINT64, NULL, &nparity) == 0)
1223 		vdev->v_nparity = nparity;
1224 
1225 	if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
1226 	    DATA_TYPE_STRING, NULL, &path) == 0) {
1227 		if (strncmp(path, "/dev/", 5) == 0)
1228 			path += 5;
1229 		vdev->v_name = strdup(path);
1230 	} else {
1231 		char *name;
1232 
1233 		name = NULL;
1234 		if (strcmp(type, "raidz") == 0) {
1235 			if (vdev->v_nparity < 1 ||
1236 			    vdev->v_nparity > 3) {
1237 				printf("ZFS: can only boot from disk, "
1238 				    "mirror, raidz1, raidz2 and raidz3 "
1239 				    "vdevs\n");
1240 				return (EIO);
1241 			}
1242 			(void) asprintf(&name, "%s%d-%" PRIu64, type,
1243 			    vdev->v_nparity, id);
1244 		} else {
1245 			(void) asprintf(&name, "%s-%" PRIu64, type, id);
1246 		}
1247 		vdev->v_name = name;
1248 	}
1249 	*vdevp = vdev;
1250 	return (0);
1251 }
1252 
1253 /*
1254  * Find slot for vdev. We return either NULL to signal to use
1255  * STAILQ_INSERT_HEAD, or we return link element to be used with
1256  * STAILQ_INSERT_AFTER.
1257  */
1258 static vdev_t *
1259 vdev_find_previous(vdev_t *top_vdev, vdev_t *vdev)
1260 {
1261 	vdev_t *v, *previous;
1262 
1263 	if (STAILQ_EMPTY(&top_vdev->v_children))
1264 		return (NULL);
1265 
1266 	previous = NULL;
1267 	STAILQ_FOREACH(v, &top_vdev->v_children, v_childlink) {
1268 		if (v->v_id > vdev->v_id)
1269 			return (previous);
1270 
1271 		if (v->v_id == vdev->v_id)
1272 			return (v);
1273 
1274 		if (v->v_id < vdev->v_id)
1275 			previous = v;
1276 	}
1277 	return (previous);
1278 }
1279 
1280 static size_t
1281 vdev_child_count(vdev_t *vdev)
1282 {
1283 	vdev_t *v;
1284 	size_t count;
1285 
1286 	count = 0;
1287 	STAILQ_FOREACH(v, &vdev->v_children, v_childlink) {
1288 		count++;
1289 	}
1290 	return (count);
1291 }
1292 
1293 /*
1294  * Insert vdev into top_vdev children list. List is ordered by v_id.
1295  */
1296 static void
1297 vdev_insert(vdev_t *top_vdev, vdev_t *vdev)
1298 {
1299 	vdev_t *previous;
1300 	size_t count;
1301 
1302 	/*
1303 	 * The top level vdev can appear in random order, depending how
1304 	 * the firmware is presenting the disk devices.
1305 	 * However, we will insert vdev to create list ordered by v_id,
1306 	 * so we can use either STAILQ_INSERT_HEAD or STAILQ_INSERT_AFTER
1307 	 * as STAILQ does not have insert before.
1308 	 */
1309 	previous = vdev_find_previous(top_vdev, vdev);
1310 
1311 	if (previous == NULL) {
1312 		STAILQ_INSERT_HEAD(&top_vdev->v_children, vdev, v_childlink);
1313 		count = vdev_child_count(top_vdev);
1314 		if (top_vdev->v_nchildren < count)
1315 			top_vdev->v_nchildren = count;
1316 		return;
1317 	}
1318 
1319 	if (previous->v_id == vdev->v_id)
1320 		return;
1321 
1322 	STAILQ_INSERT_AFTER(&top_vdev->v_children, previous, vdev, v_childlink);
1323 	count = vdev_child_count(top_vdev);
1324 	if (top_vdev->v_nchildren < count)
1325 		top_vdev->v_nchildren = count;
1326 }
1327 
1328 static int
1329 vdev_from_nvlist(spa_t *spa, uint64_t top_guid, const unsigned char *nvlist)
1330 {
1331 	vdev_t *top_vdev, *vdev;
1332 	const unsigned char *kids;
1333 	int rc, nkids;
1334 
1335 	/* Get top vdev. */
1336 	top_vdev = vdev_find(top_guid);
1337 	if (top_vdev == NULL) {
1338 		rc = vdev_init(top_guid, nvlist, &top_vdev);
1339 		if (rc != 0)
1340 			return (rc);
1341 		top_vdev->v_spa = spa;
1342 		top_vdev->v_top = top_vdev;
1343 		vdev_insert(spa->spa_root_vdev, top_vdev);
1344 	}
1345 
1346 	/* Add children if there are any. */
1347 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1348 	    &nkids, &kids);
1349 	if (rc == 0) {
1350 		for (int i = 0; i < nkids; i++) {
1351 			uint64_t guid;
1352 
1353 			rc = nvlist_find(kids, ZPOOL_CONFIG_GUID,
1354 			    DATA_TYPE_UINT64, NULL, &guid);
1355 			if (rc != 0)
1356 				return (rc);
1357 			rc = vdev_init(guid, kids, &vdev);
1358 			if (rc != 0)
1359 				return (rc);
1360 
1361 			vdev->v_spa = spa;
1362 			vdev->v_top = top_vdev;
1363 			vdev_insert(top_vdev, vdev);
1364 
1365 			kids = nvlist_next(kids);
1366 		}
1367 	} else {
1368 		rc = 0;
1369 	}
1370 
1371 	return (rc);
1372 }
1373 
1374 static int
1375 vdev_init_from_label(spa_t *spa, const unsigned char *nvlist)
1376 {
1377 	uint64_t pool_guid, top_guid;
1378 	const unsigned char *vdevs;
1379 
1380 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1381 	    NULL, &pool_guid) ||
1382 	    nvlist_find(nvlist, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64,
1383 	    NULL, &top_guid) ||
1384 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1385 	    NULL, &vdevs)) {
1386 		printf("ZFS: can't find vdev details\n");
1387 		return (ENOENT);
1388 	}
1389 
1390 	return (vdev_from_nvlist(spa, top_guid, vdevs));
1391 }
1392 
1393 static void
1394 vdev_set_state(vdev_t *vdev)
1395 {
1396 	vdev_t *kid;
1397 	int good_kids;
1398 	int bad_kids;
1399 
1400 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1401 		vdev_set_state(kid);
1402 	}
1403 
1404 	/*
1405 	 * A mirror or raidz is healthy if all its kids are healthy. A
1406 	 * mirror is degraded if any of its kids is healthy; a raidz
1407 	 * is degraded if at most nparity kids are offline.
1408 	 */
1409 	if (STAILQ_FIRST(&vdev->v_children)) {
1410 		good_kids = 0;
1411 		bad_kids = 0;
1412 		STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1413 			if (kid->v_state == VDEV_STATE_HEALTHY)
1414 				good_kids++;
1415 			else
1416 				bad_kids++;
1417 		}
1418 		if (bad_kids == 0) {
1419 			vdev->v_state = VDEV_STATE_HEALTHY;
1420 		} else {
1421 			if (vdev->v_read == vdev_mirror_read) {
1422 				if (good_kids) {
1423 					vdev->v_state = VDEV_STATE_DEGRADED;
1424 				} else {
1425 					vdev->v_state = VDEV_STATE_OFFLINE;
1426 				}
1427 			} else if (vdev->v_read == vdev_raidz_read) {
1428 				if (bad_kids > vdev->v_nparity) {
1429 					vdev->v_state = VDEV_STATE_OFFLINE;
1430 				} else {
1431 					vdev->v_state = VDEV_STATE_DEGRADED;
1432 				}
1433 			}
1434 		}
1435 	}
1436 }
1437 
1438 static int
1439 vdev_update_from_nvlist(uint64_t top_guid, const unsigned char *nvlist)
1440 {
1441 	vdev_t *vdev;
1442 	const unsigned char *kids;
1443 	int rc, nkids;
1444 
1445 	/* Update top vdev. */
1446 	vdev = vdev_find(top_guid);
1447 	if (vdev != NULL)
1448 		vdev_set_initial_state(vdev, nvlist);
1449 
1450 	/* Update children if there are any. */
1451 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1452 	    &nkids, &kids);
1453 	if (rc == 0) {
1454 		for (int i = 0; i < nkids; i++) {
1455 			uint64_t guid;
1456 
1457 			rc = nvlist_find(kids, ZPOOL_CONFIG_GUID,
1458 			    DATA_TYPE_UINT64, NULL, &guid);
1459 			if (rc != 0)
1460 				break;
1461 
1462 			vdev = vdev_find(guid);
1463 			if (vdev != NULL)
1464 				vdev_set_initial_state(vdev, kids);
1465 
1466 			kids = nvlist_next(kids);
1467 		}
1468 	} else {
1469 		rc = 0;
1470 	}
1471 
1472 	return (rc);
1473 }
1474 
1475 static int
1476 vdev_init_from_nvlist(spa_t *spa, const unsigned char *nvlist)
1477 {
1478 	uint64_t pool_guid, vdev_children;
1479 	const unsigned char *vdevs, *kids;
1480 	int rc, nkids;
1481 
1482 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1483 	    NULL, &pool_guid) ||
1484 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_CHILDREN, DATA_TYPE_UINT64,
1485 	    NULL, &vdev_children) ||
1486 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1487 	    NULL, &vdevs)) {
1488 		printf("ZFS: can't find vdev details\n");
1489 		return (ENOENT);
1490 	}
1491 
1492 	/* Wrong guid?! */
1493 	if (spa->spa_guid != pool_guid)
1494 		return (EIO);
1495 
1496 	spa->spa_root_vdev->v_nchildren = vdev_children;
1497 
1498 	rc = nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1499 	    &nkids, &kids);
1500 
1501 	/*
1502 	 * MOS config has at least one child for root vdev.
1503 	 */
1504 	if (rc != 0)
1505 		return (EIO);
1506 
1507 	for (int i = 0; i < nkids; i++) {
1508 		uint64_t guid;
1509 		vdev_t *vdev;
1510 
1511 		rc = nvlist_find(kids, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1512 		    NULL, &guid);
1513 		if (rc != 0)
1514 			break;
1515 		vdev = vdev_find(guid);
1516 		/*
1517 		 * Top level vdev is missing, create it.
1518 		 */
1519 		if (vdev == NULL)
1520 			rc = vdev_from_nvlist(spa, guid, kids);
1521 		else
1522 			rc = vdev_update_from_nvlist(guid, kids);
1523 		if (rc != 0)
1524 			break;
1525 		kids = nvlist_next(kids);
1526 	}
1527 
1528 	/*
1529 	 * Re-evaluate top-level vdev state.
1530 	 */
1531 	vdev_set_state(spa->spa_root_vdev);
1532 
1533 	return (rc);
1534 }
1535 
1536 static spa_t *
1537 spa_find_by_guid(uint64_t guid)
1538 {
1539 	spa_t *spa;
1540 
1541 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1542 		if (spa->spa_guid == guid)
1543 			return (spa);
1544 
1545 	return (NULL);
1546 }
1547 
1548 static spa_t *
1549 spa_find_by_name(const char *name)
1550 {
1551 	spa_t *spa;
1552 
1553 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1554 		if (strcmp(spa->spa_name, name) == 0)
1555 			return (spa);
1556 
1557 	return (NULL);
1558 }
1559 
1560 #ifdef BOOT2
1561 static spa_t *
1562 spa_get_primary(void)
1563 {
1564 
1565 	return (STAILQ_FIRST(&zfs_pools));
1566 }
1567 
1568 static vdev_t *
1569 spa_get_primary_vdev(const spa_t *spa)
1570 {
1571 	vdev_t *vdev;
1572 	vdev_t *kid;
1573 
1574 	if (spa == NULL)
1575 		spa = spa_get_primary();
1576 	if (spa == NULL)
1577 		return (NULL);
1578 	vdev = spa->spa_root_vdev;
1579 	if (vdev == NULL)
1580 		return (NULL);
1581 	for (kid = STAILQ_FIRST(&vdev->v_children); kid != NULL;
1582 	    kid = STAILQ_FIRST(&vdev->v_children))
1583 		vdev = kid;
1584 	return (vdev);
1585 }
1586 #endif
1587 
1588 static spa_t *
1589 spa_create(uint64_t guid, const char *name)
1590 {
1591 	spa_t *spa;
1592 
1593 	if ((spa = calloc(1, sizeof(spa_t))) == NULL)
1594 		return (NULL);
1595 	if ((spa->spa_name = strdup(name)) == NULL) {
1596 		free(spa);
1597 		return (NULL);
1598 	}
1599 	spa->spa_guid = guid;
1600 	spa->spa_root_vdev = vdev_create(guid, NULL);
1601 	if (spa->spa_root_vdev == NULL) {
1602 		free(spa->spa_name);
1603 		free(spa);
1604 		return (NULL);
1605 	}
1606 	spa->spa_root_vdev->v_name = strdup("root");
1607 	STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
1608 
1609 	return (spa);
1610 }
1611 
1612 static const char *
1613 state_name(vdev_state_t state)
1614 {
1615 	static const char *names[] = {
1616 		"UNKNOWN",
1617 		"CLOSED",
1618 		"OFFLINE",
1619 		"REMOVED",
1620 		"CANT_OPEN",
1621 		"FAULTED",
1622 		"DEGRADED",
1623 		"ONLINE"
1624 	};
1625 	return (names[state]);
1626 }
1627 
1628 #ifdef BOOT2
1629 
1630 #define pager_printf printf
1631 
1632 #else
1633 
1634 static int
1635 pager_printf(const char *fmt, ...)
1636 {
1637 	char line[80];
1638 	va_list args;
1639 
1640 	va_start(args, fmt);
1641 	vsnprintf(line, sizeof(line), fmt, args);
1642 	va_end(args);
1643 	return (pager_output(line));
1644 }
1645 
1646 #endif
1647 
1648 #define	STATUS_FORMAT	"        %s %s\n"
1649 
1650 static int
1651 print_state(int indent, const char *name, vdev_state_t state)
1652 {
1653 	int i;
1654 	char buf[512];
1655 
1656 	buf[0] = 0;
1657 	for (i = 0; i < indent; i++)
1658 		strcat(buf, "  ");
1659 	strcat(buf, name);
1660 	return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
1661 }
1662 
1663 static int
1664 vdev_status(vdev_t *vdev, int indent)
1665 {
1666 	vdev_t *kid;
1667 	int ret;
1668 
1669 	if (vdev->v_islog) {
1670 		(void) pager_output("        logs\n");
1671 		indent++;
1672 	}
1673 
1674 	ret = print_state(indent, vdev->v_name, vdev->v_state);
1675 	if (ret != 0)
1676 		return (ret);
1677 
1678 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1679 		ret = vdev_status(kid, indent + 1);
1680 		if (ret != 0)
1681 			return (ret);
1682 	}
1683 	return (ret);
1684 }
1685 
1686 static int
1687 spa_status(spa_t *spa)
1688 {
1689 	static char bootfs[ZFS_MAXNAMELEN];
1690 	uint64_t rootid;
1691 	vdev_list_t *vlist;
1692 	vdev_t *vdev;
1693 	int good_kids, bad_kids, degraded_kids, ret;
1694 	vdev_state_t state;
1695 
1696 	ret = pager_printf("  pool: %s\n", spa->spa_name);
1697 	if (ret != 0)
1698 		return (ret);
1699 
1700 	if (zfs_get_root(spa, &rootid) == 0 &&
1701 	    zfs_rlookup(spa, rootid, bootfs) == 0) {
1702 		if (bootfs[0] == '\0')
1703 			ret = pager_printf("bootfs: %s\n", spa->spa_name);
1704 		else
1705 			ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
1706 			    bootfs);
1707 		if (ret != 0)
1708 			return (ret);
1709 	}
1710 	ret = pager_printf("config:\n\n");
1711 	if (ret != 0)
1712 		return (ret);
1713 	ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
1714 	if (ret != 0)
1715 		return (ret);
1716 
1717 	good_kids = 0;
1718 	degraded_kids = 0;
1719 	bad_kids = 0;
1720 	vlist = &spa->spa_root_vdev->v_children;
1721 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1722 		if (vdev->v_state == VDEV_STATE_HEALTHY)
1723 			good_kids++;
1724 		else if (vdev->v_state == VDEV_STATE_DEGRADED)
1725 			degraded_kids++;
1726 		else
1727 			bad_kids++;
1728 	}
1729 
1730 	state = VDEV_STATE_CLOSED;
1731 	if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
1732 		state = VDEV_STATE_HEALTHY;
1733 	else if ((good_kids + degraded_kids) > 0)
1734 		state = VDEV_STATE_DEGRADED;
1735 
1736 	ret = print_state(0, spa->spa_name, state);
1737 	if (ret != 0)
1738 		return (ret);
1739 
1740 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1741 		ret = vdev_status(vdev, 1);
1742 		if (ret != 0)
1743 			return (ret);
1744 	}
1745 	return (ret);
1746 }
1747 
1748 static int
1749 spa_all_status(void)
1750 {
1751 	spa_t *spa;
1752 	int first = 1, ret = 0;
1753 
1754 	STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
1755 		if (!first) {
1756 			ret = pager_printf("\n");
1757 			if (ret != 0)
1758 				return (ret);
1759 		}
1760 		first = 0;
1761 		ret = spa_status(spa);
1762 		if (ret != 0)
1763 			return (ret);
1764 	}
1765 	return (ret);
1766 }
1767 
1768 static uint64_t
1769 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
1770 {
1771 	uint64_t label_offset;
1772 
1773 	if (l < VDEV_LABELS / 2)
1774 		label_offset = 0;
1775 	else
1776 		label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
1777 
1778 	return (offset + l * sizeof (vdev_label_t) + label_offset);
1779 }
1780 
1781 static int
1782 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1783 {
1784 	unsigned int seq1 = 0;
1785 	unsigned int seq2 = 0;
1786 	int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
1787 
1788 	if (cmp != 0)
1789 		return (cmp);
1790 
1791 	cmp = AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1792 	if (cmp != 0)
1793 		return (cmp);
1794 
1795 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1796 		seq1 = MMP_SEQ(ub1);
1797 
1798 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1799 		seq2 = MMP_SEQ(ub2);
1800 
1801 	return (AVL_CMP(seq1, seq2));
1802 }
1803 
1804 static int
1805 uberblock_verify(uberblock_t *ub)
1806 {
1807 	if (ub->ub_magic == BSWAP_64((uint64_t)UBERBLOCK_MAGIC)) {
1808 		byteswap_uint64_array(ub, sizeof (uberblock_t));
1809 	}
1810 
1811 	if (ub->ub_magic != UBERBLOCK_MAGIC ||
1812 	    !SPA_VERSION_IS_SUPPORTED(ub->ub_version))
1813 		return (EINVAL);
1814 
1815 	return (0);
1816 }
1817 
1818 static int
1819 vdev_label_read(vdev_t *vd, int l, void *buf, uint64_t offset,
1820     size_t size)
1821 {
1822 	blkptr_t bp;
1823 	off_t off;
1824 
1825 	off = vdev_label_offset(vd->v_psize, l, offset);
1826 
1827 	BP_ZERO(&bp);
1828 	BP_SET_LSIZE(&bp, size);
1829 	BP_SET_PSIZE(&bp, size);
1830 	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1831 	BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1832 	DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
1833 	ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1834 
1835 	return (vdev_read_phys(vd, &bp, buf, off, size));
1836 }
1837 
1838 static unsigned char *
1839 vdev_label_read_config(vdev_t *vd, uint64_t txg)
1840 {
1841 	vdev_phys_t *label;
1842 	uint64_t best_txg = 0;
1843 	uint64_t label_txg = 0;
1844 	uint64_t asize;
1845 	unsigned char *nvl;
1846 	size_t nvl_size;
1847 	int error;
1848 
1849 	label = malloc(sizeof (vdev_phys_t));
1850 	if (label == NULL)
1851 		return (NULL);
1852 
1853 	nvl_size = VDEV_PHYS_SIZE - sizeof (zio_eck_t) - 4;
1854 	nvl = malloc(nvl_size);
1855 	if (nvl == NULL)
1856 		goto done;
1857 
1858 	for (int l = 0; l < VDEV_LABELS; l++) {
1859 		const unsigned char *nvlist;
1860 
1861 		if (vdev_label_read(vd, l, label,
1862 		    offsetof(vdev_label_t, vl_vdev_phys),
1863 		    sizeof (vdev_phys_t)))
1864 			continue;
1865 
1866 		if (label->vp_nvlist[0] != NV_ENCODE_XDR)
1867 			continue;
1868 
1869 		nvlist = (const unsigned char *) label->vp_nvlist + 4;
1870 		error = nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG,
1871 		    DATA_TYPE_UINT64, NULL, &label_txg);
1872 		if (error != 0 || label_txg == 0) {
1873 			memcpy(nvl, nvlist, nvl_size);
1874 			goto done;
1875 		}
1876 
1877 		if (label_txg <= txg && label_txg > best_txg) {
1878 			best_txg = label_txg;
1879 			memcpy(nvl, nvlist, nvl_size);
1880 
1881 			/*
1882 			 * Use asize from pool config. We need this
1883 			 * because we can get bad value from BIOS.
1884 			 */
1885 			if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE,
1886 			    DATA_TYPE_UINT64, NULL, &asize) == 0) {
1887 				vd->v_psize = asize +
1888 				    VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
1889 			}
1890 		}
1891 	}
1892 
1893 	if (best_txg == 0) {
1894 		free(nvl);
1895 		nvl = NULL;
1896 	}
1897 done:
1898 	free(label);
1899 	return (nvl);
1900 }
1901 
1902 static void
1903 vdev_uberblock_load(vdev_t *vd, uberblock_t *ub)
1904 {
1905 	uberblock_t *buf;
1906 
1907 	buf = malloc(VDEV_UBERBLOCK_SIZE(vd));
1908 	if (buf == NULL)
1909 		return;
1910 
1911 	for (int l = 0; l < VDEV_LABELS; l++) {
1912 		for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1913 			if (vdev_label_read(vd, l, buf,
1914 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1915 			    VDEV_UBERBLOCK_SIZE(vd)))
1916 				continue;
1917 			if (uberblock_verify(buf) != 0)
1918 				continue;
1919 
1920 			if (vdev_uberblock_compare(buf, ub) > 0)
1921 				*ub = *buf;
1922 		}
1923 	}
1924 	free(buf);
1925 }
1926 
1927 static int
1928 vdev_probe(vdev_phys_read_t *_read, void *read_priv, spa_t **spap)
1929 {
1930 	vdev_t vtmp;
1931 	spa_t *spa;
1932 	vdev_t *vdev;
1933 	unsigned char *nvlist;
1934 	uint64_t val;
1935 	uint64_t guid, vdev_children;
1936 	uint64_t pool_txg, pool_guid;
1937 	const char *pool_name;
1938 	const unsigned char *features;
1939 	int rc;
1940 
1941 	/*
1942 	 * Load the vdev label and figure out which
1943 	 * uberblock is most current.
1944 	 */
1945 	memset(&vtmp, 0, sizeof(vtmp));
1946 	vtmp.v_phys_read = _read;
1947 	vtmp.v_read_priv = read_priv;
1948 	vtmp.v_psize = P2ALIGN(ldi_get_size(read_priv),
1949 	    (uint64_t)sizeof (vdev_label_t));
1950 
1951 	/* Test for minimum device size. */
1952 	if (vtmp.v_psize < SPA_MINDEVSIZE)
1953 		return (EIO);
1954 
1955 	nvlist = vdev_label_read_config(&vtmp, UINT64_MAX);
1956 	if (nvlist == NULL)
1957 		return (EIO);
1958 
1959 	if (nvlist_find(nvlist, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
1960 	    NULL, &val) != 0) {
1961 		free(nvlist);
1962 		return (EIO);
1963 	}
1964 
1965 	if (!SPA_VERSION_IS_SUPPORTED(val)) {
1966 		printf("ZFS: unsupported ZFS version %u (should be %u)\n",
1967 		    (unsigned)val, (unsigned)SPA_VERSION);
1968 		free(nvlist);
1969 		return (EIO);
1970 	}
1971 
1972 	/* Check ZFS features for read */
1973 	if (nvlist_find(nvlist, ZPOOL_CONFIG_FEATURES_FOR_READ,
1974 	    DATA_TYPE_NVLIST, NULL, &features) == 0 &&
1975 	    nvlist_check_features_for_read(features) != 0) {
1976 		free(nvlist);
1977 		return (EIO);
1978 	}
1979 
1980 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
1981 	    NULL, &val) != 0) {
1982 		free(nvlist);
1983 		return (EIO);
1984 	}
1985 
1986 	if (val == POOL_STATE_DESTROYED) {
1987 		/* We don't boot only from destroyed pools. */
1988 		free(nvlist);
1989 		return (EIO);
1990 	}
1991 
1992 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
1993 	    NULL, &pool_txg) != 0 ||
1994 	    nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1995 	    NULL, &pool_guid) != 0 ||
1996 	    nvlist_find(nvlist, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
1997 	    NULL, &pool_name) != 0) {
1998 		/*
1999 		 * Cache and spare devices end up here - just ignore
2000 		 * them.
2001 		 */
2002 		free(nvlist);
2003 		return (EIO);
2004 	}
2005 
2006 	/*
2007 	 * Create the pool if this is the first time we've seen it.
2008 	 */
2009 	spa = spa_find_by_guid(pool_guid);
2010 	if (spa == NULL) {
2011 		nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_CHILDREN,
2012 		    DATA_TYPE_UINT64, NULL, &vdev_children);
2013 		spa = spa_create(pool_guid, pool_name);
2014 		if (spa == NULL) {
2015 			free(nvlist);
2016 			return (ENOMEM);
2017 		}
2018 		spa->spa_root_vdev->v_nchildren = vdev_children;
2019 	}
2020 	if (pool_txg > spa->spa_txg)
2021 		spa->spa_txg = pool_txg;
2022 
2023 	/*
2024 	 * Get the vdev tree and create our in-core copy of it.
2025 	 * If we already have a vdev with this guid, this must
2026 	 * be some kind of alias (overlapping slices, dangerously dedicated
2027 	 * disks etc).
2028 	 */
2029 	if (nvlist_find(nvlist, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
2030 	    NULL, &guid) != 0) {
2031 		free(nvlist);
2032 		return (EIO);
2033 	}
2034 	vdev = vdev_find(guid);
2035 	/* Has this vdev already been inited? */
2036 	if (vdev && vdev->v_phys_read) {
2037 		free(nvlist);
2038 		return (EIO);
2039 	}
2040 
2041 	rc = vdev_init_from_label(spa, nvlist);
2042 	free(nvlist);
2043 	if (rc != 0)
2044 		return (rc);
2045 
2046 	/*
2047 	 * We should already have created an incomplete vdev for this
2048 	 * vdev. Find it and initialise it with our read proc.
2049 	 */
2050 	vdev = vdev_find(guid);
2051 	if (vdev != NULL) {
2052 		vdev->v_phys_read = _read;
2053 		vdev->v_read_priv = read_priv;
2054 		vdev->v_psize = vtmp.v_psize;
2055 		/*
2056 		 * If no other state is set, mark vdev healthy.
2057 		 */
2058 		if (vdev->v_state == VDEV_STATE_UNKNOWN)
2059 			vdev->v_state = VDEV_STATE_HEALTHY;
2060 	} else {
2061 		printf("ZFS: inconsistent nvlist contents\n");
2062 		return (EIO);
2063 	}
2064 
2065 	if (vdev->v_islog)
2066 		spa->spa_with_log = vdev->v_islog;
2067 
2068 	/*
2069 	 * Re-evaluate top-level vdev state.
2070 	 */
2071 	vdev_set_state(vdev->v_top);
2072 
2073 	/*
2074 	 * Ok, we are happy with the pool so far. Lets find
2075 	 * the best uberblock and then we can actually access
2076 	 * the contents of the pool.
2077 	 */
2078 	vdev_uberblock_load(vdev, &spa->spa_uberblock);
2079 
2080 	if (spap != NULL)
2081 		*spap = spa;
2082 	return (0);
2083 }
2084 
2085 static int
2086 ilog2(int n)
2087 {
2088 	int v;
2089 
2090 	for (v = 0; v < 32; v++)
2091 		if (n == (1 << v))
2092 			return (v);
2093 	return (-1);
2094 }
2095 
2096 static int
2097 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
2098 {
2099 	blkptr_t gbh_bp;
2100 	zio_gbh_phys_t zio_gb;
2101 	char *pbuf;
2102 	int i;
2103 
2104 	/* Artificial BP for gang block header. */
2105 	gbh_bp = *bp;
2106 	BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
2107 	BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
2108 	BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
2109 	BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
2110 	for (i = 0; i < SPA_DVAS_PER_BP; i++)
2111 		DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
2112 
2113 	/* Read gang header block using the artificial BP. */
2114 	if (zio_read(spa, &gbh_bp, &zio_gb))
2115 		return (EIO);
2116 
2117 	pbuf = buf;
2118 	for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
2119 		blkptr_t *gbp = &zio_gb.zg_blkptr[i];
2120 
2121 		if (BP_IS_HOLE(gbp))
2122 			continue;
2123 		if (zio_read(spa, gbp, pbuf))
2124 			return (EIO);
2125 		pbuf += BP_GET_PSIZE(gbp);
2126 	}
2127 
2128 	if (zio_checksum_verify(spa, bp, buf))
2129 		return (EIO);
2130 	return (0);
2131 }
2132 
2133 static int
2134 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
2135 {
2136 	int cpfunc = BP_GET_COMPRESS(bp);
2137 	uint64_t align, size;
2138 	void *pbuf;
2139 	int i, error;
2140 
2141 	/*
2142 	 * Process data embedded in block pointer
2143 	 */
2144 	if (BP_IS_EMBEDDED(bp)) {
2145 		ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
2146 
2147 		size = BPE_GET_PSIZE(bp);
2148 		ASSERT(size <= BPE_PAYLOAD_SIZE);
2149 
2150 		if (cpfunc != ZIO_COMPRESS_OFF)
2151 			pbuf = zfs_alloc(size);
2152 		else
2153 			pbuf = buf;
2154 
2155 		decode_embedded_bp_compressed(bp, pbuf);
2156 		error = 0;
2157 
2158 		if (cpfunc != ZIO_COMPRESS_OFF) {
2159 			error = zio_decompress_data(cpfunc, pbuf,
2160 			    size, buf, BP_GET_LSIZE(bp));
2161 			zfs_free(pbuf, size);
2162 		}
2163 		if (error != 0)
2164 			printf("ZFS: i/o error - unable to decompress "
2165 			    "block pointer data, error %d\n", error);
2166 		return (error);
2167 	}
2168 
2169 	error = EIO;
2170 
2171 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
2172 		const dva_t *dva = &bp->blk_dva[i];
2173 		vdev_t *vdev;
2174 		vdev_list_t *vlist;
2175 		uint64_t vdevid;
2176 		off_t offset;
2177 
2178 		if (!dva->dva_word[0] && !dva->dva_word[1])
2179 			continue;
2180 
2181 		vdevid = DVA_GET_VDEV(dva);
2182 		offset = DVA_GET_OFFSET(dva);
2183 		vlist = &spa->spa_root_vdev->v_children;
2184 		STAILQ_FOREACH(vdev, vlist, v_childlink) {
2185 			if (vdev->v_id == vdevid)
2186 				break;
2187 		}
2188 		if (!vdev || !vdev->v_read)
2189 			continue;
2190 
2191 		size = BP_GET_PSIZE(bp);
2192 		if (vdev->v_read == vdev_raidz_read) {
2193 			align = 1ULL << vdev->v_ashift;
2194 			if (P2PHASE(size, align) != 0)
2195 				size = P2ROUNDUP(size, align);
2196 		}
2197 		if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
2198 			pbuf = zfs_alloc(size);
2199 		else
2200 			pbuf = buf;
2201 
2202 		if (DVA_GET_GANG(dva))
2203 			error = zio_read_gang(spa, bp, pbuf);
2204 		else
2205 			error = vdev->v_read(vdev, bp, pbuf, offset, size);
2206 		if (error == 0) {
2207 			if (cpfunc != ZIO_COMPRESS_OFF)
2208 				error = zio_decompress_data(cpfunc, pbuf,
2209 				    BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
2210 			else if (size != BP_GET_PSIZE(bp))
2211 				bcopy(pbuf, buf, BP_GET_PSIZE(bp));
2212 		}
2213 		if (buf != pbuf)
2214 			zfs_free(pbuf, size);
2215 		if (error == 0)
2216 			break;
2217 	}
2218 	if (error != 0)
2219 		printf("ZFS: i/o error - all block copies unavailable\n");
2220 	return (error);
2221 }
2222 
2223 static int
2224 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset,
2225     void *buf, size_t buflen)
2226 {
2227 	int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
2228 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2229 	int nlevels = dnode->dn_nlevels;
2230 	int i, rc;
2231 
2232 	if (bsize > SPA_MAXBLOCKSIZE) {
2233 		printf("ZFS: I/O error - blocks larger than %llu are not "
2234 		    "supported\n", SPA_MAXBLOCKSIZE);
2235 		return (EIO);
2236 	}
2237 
2238 	/*
2239 	 * Note: bsize may not be a power of two here so we need to do an
2240 	 * actual divide rather than a bitshift.
2241 	 */
2242 	while (buflen > 0) {
2243 		uint64_t bn = offset / bsize;
2244 		int boff = offset % bsize;
2245 		int ibn;
2246 		const blkptr_t *indbp;
2247 		blkptr_t bp;
2248 
2249 		if (bn > dnode->dn_maxblkid)
2250 			return (EIO);
2251 
2252 		if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
2253 			goto cached;
2254 
2255 		indbp = dnode->dn_blkptr;
2256 		for (i = 0; i < nlevels; i++) {
2257 			/*
2258 			 * Copy the bp from the indirect array so that
2259 			 * we can re-use the scratch buffer for multi-level
2260 			 * objects.
2261 			 */
2262 			ibn = bn >> ((nlevels - i - 1) * ibshift);
2263 			ibn &= ((1 << ibshift) - 1);
2264 			bp = indbp[ibn];
2265 			if (BP_IS_HOLE(&bp)) {
2266 				memset(dnode_cache_buf, 0, bsize);
2267 				break;
2268 			}
2269 			rc = zio_read(spa, &bp, dnode_cache_buf);
2270 			if (rc)
2271 				return (rc);
2272 			indbp = (const blkptr_t *) dnode_cache_buf;
2273 		}
2274 		dnode_cache_obj = dnode;
2275 		dnode_cache_bn = bn;
2276 	cached:
2277 
2278 		/*
2279 		 * The buffer contains our data block. Copy what we
2280 		 * need from it and loop.
2281 		 */
2282 		i = bsize - boff;
2283 		if (i > buflen) i = buflen;
2284 		memcpy(buf, &dnode_cache_buf[boff], i);
2285 		buf = ((char *)buf) + i;
2286 		offset += i;
2287 		buflen -= i;
2288 	}
2289 
2290 	return (0);
2291 }
2292 
2293 /*
2294  * Lookup a value in a microzap directory. Assumes that the zap
2295  * scratch buffer contains the directory contents.
2296  */
2297 static int
2298 mzap_lookup(const dnode_phys_t *dnode, const char *name, uint64_t *value)
2299 {
2300 	const mzap_phys_t *mz;
2301 	const mzap_ent_phys_t *mze;
2302 	size_t size;
2303 	int chunks, i;
2304 
2305 	/*
2306 	 * Microzap objects use exactly one block. Read the whole
2307 	 * thing.
2308 	 */
2309 	size = dnode->dn_datablkszsec * 512;
2310 
2311 	mz = (const mzap_phys_t *) zap_scratch;
2312 	chunks = size / MZAP_ENT_LEN - 1;
2313 
2314 	for (i = 0; i < chunks; i++) {
2315 		mze = &mz->mz_chunk[i];
2316 		if (strcmp(mze->mze_name, name) == 0) {
2317 			*value = mze->mze_value;
2318 			return (0);
2319 		}
2320 	}
2321 
2322 	return (ENOENT);
2323 }
2324 
2325 /*
2326  * Compare a name with a zap leaf entry. Return non-zero if the name
2327  * matches.
2328  */
2329 static int
2330 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2331     const char *name)
2332 {
2333 	size_t namelen;
2334 	const zap_leaf_chunk_t *nc;
2335 	const char *p;
2336 
2337 	namelen = zc->l_entry.le_name_numints;
2338 
2339 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2340 	p = name;
2341 	while (namelen > 0) {
2342 		size_t len;
2343 
2344 		len = namelen;
2345 		if (len > ZAP_LEAF_ARRAY_BYTES)
2346 			len = ZAP_LEAF_ARRAY_BYTES;
2347 		if (memcmp(p, nc->l_array.la_array, len))
2348 			return (0);
2349 		p += len;
2350 		namelen -= len;
2351 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2352 	}
2353 
2354 	return (1);
2355 }
2356 
2357 /*
2358  * Extract a uint64_t value from a zap leaf entry.
2359  */
2360 static uint64_t
2361 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
2362 {
2363 	const zap_leaf_chunk_t *vc;
2364 	int i;
2365 	uint64_t value;
2366 	const uint8_t *p;
2367 
2368 	vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
2369 	for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
2370 		value = (value << 8) | p[i];
2371 	}
2372 
2373 	return (value);
2374 }
2375 
2376 static void
2377 stv(int len, void *addr, uint64_t value)
2378 {
2379 	switch (len) {
2380 	case 1:
2381 		*(uint8_t *)addr = value;
2382 		return;
2383 	case 2:
2384 		*(uint16_t *)addr = value;
2385 		return;
2386 	case 4:
2387 		*(uint32_t *)addr = value;
2388 		return;
2389 	case 8:
2390 		*(uint64_t *)addr = value;
2391 		return;
2392 	}
2393 }
2394 
2395 /*
2396  * Extract a array from a zap leaf entry.
2397  */
2398 static void
2399 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2400     uint64_t integer_size, uint64_t num_integers, void *buf)
2401 {
2402 	uint64_t array_int_len = zc->l_entry.le_value_intlen;
2403 	uint64_t value = 0;
2404 	uint64_t *u64 = buf;
2405 	char *p = buf;
2406 	int len = MIN(zc->l_entry.le_value_numints, num_integers);
2407 	int chunk = zc->l_entry.le_value_chunk;
2408 	int byten = 0;
2409 
2410 	if (integer_size == 8 && len == 1) {
2411 		*u64 = fzap_leaf_value(zl, zc);
2412 		return;
2413 	}
2414 
2415 	while (len > 0) {
2416 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
2417 		int i;
2418 
2419 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
2420 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
2421 			value = (value << 8) | la->la_array[i];
2422 			byten++;
2423 			if (byten == array_int_len) {
2424 				stv(integer_size, p, value);
2425 				byten = 0;
2426 				len--;
2427 				if (len == 0)
2428 					return;
2429 				p += integer_size;
2430 			}
2431 		}
2432 		chunk = la->la_next;
2433 	}
2434 }
2435 
2436 static int
2437 fzap_check_size(uint64_t integer_size, uint64_t num_integers)
2438 {
2439 
2440 	switch (integer_size) {
2441 	case 1:
2442 	case 2:
2443 	case 4:
2444 	case 8:
2445 		break;
2446 	default:
2447 		return (EINVAL);
2448 	}
2449 
2450 	if (integer_size * num_integers > ZAP_MAXVALUELEN)
2451 		return (E2BIG);
2452 
2453 	return (0);
2454 }
2455 
2456 /*
2457  * Lookup a value in a fatzap directory. Assumes that the zap scratch
2458  * buffer contains the directory header.
2459  */
2460 static int
2461 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2462     uint64_t integer_size, uint64_t num_integers, void *value)
2463 {
2464 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2465 	zap_phys_t zh = *(zap_phys_t *)zap_scratch;
2466 	fat_zap_t z;
2467 	uint64_t *ptrtbl;
2468 	uint64_t hash;
2469 	int rc;
2470 
2471 	if (zh.zap_magic != ZAP_MAGIC)
2472 		return (EIO);
2473 
2474 	if ((rc = fzap_check_size(integer_size, num_integers)) != 0)
2475 		return (rc);
2476 
2477 	z.zap_block_shift = ilog2(bsize);
2478 	z.zap_phys = (zap_phys_t *)zap_scratch;
2479 
2480 	/*
2481 	 * Figure out where the pointer table is and read it in if necessary.
2482 	 */
2483 	if (zh.zap_ptrtbl.zt_blk) {
2484 		rc = dnode_read(spa, dnode, zh.zap_ptrtbl.zt_blk * bsize,
2485 		    zap_scratch, bsize);
2486 		if (rc)
2487 			return (rc);
2488 		ptrtbl = (uint64_t *)zap_scratch;
2489 	} else {
2490 		ptrtbl = &ZAP_EMBEDDED_PTRTBL_ENT(&z, 0);
2491 	}
2492 
2493 	hash = zap_hash(zh.zap_salt, name);
2494 
2495 	zap_leaf_t zl;
2496 	zl.l_bs = z.zap_block_shift;
2497 
2498 	off_t off = ptrtbl[hash >> (64 - zh.zap_ptrtbl.zt_shift)] << zl.l_bs;
2499 	zap_leaf_chunk_t *zc;
2500 
2501 	rc = dnode_read(spa, dnode, off, zap_scratch, bsize);
2502 	if (rc)
2503 		return (rc);
2504 
2505 	zl.l_phys = (zap_leaf_phys_t *)zap_scratch;
2506 
2507 	/*
2508 	 * Make sure this chunk matches our hash.
2509 	 */
2510 	if (zl.l_phys->l_hdr.lh_prefix_len > 0 &&
2511 	    zl.l_phys->l_hdr.lh_prefix !=
2512 	    hash >> (64 - zl.l_phys->l_hdr.lh_prefix_len))
2513 		return (ENOENT);
2514 
2515 	/*
2516 	 * Hash within the chunk to find our entry.
2517 	 */
2518 	int shift = (64 - ZAP_LEAF_HASH_SHIFT(&zl) -
2519 	    zl.l_phys->l_hdr.lh_prefix_len);
2520 	int h = (hash >> shift) & ((1 << ZAP_LEAF_HASH_SHIFT(&zl)) - 1);
2521 	h = zl.l_phys->l_hash[h];
2522 	if (h == 0xffff)
2523 		return (ENOENT);
2524 	zc = &ZAP_LEAF_CHUNK(&zl, h);
2525 	while (zc->l_entry.le_hash != hash) {
2526 		if (zc->l_entry.le_next == 0xffff)
2527 			return (ENOENT);
2528 		zc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_next);
2529 	}
2530 	if (fzap_name_equal(&zl, zc, name)) {
2531 		if (zc->l_entry.le_value_intlen * zc->l_entry.le_value_numints >
2532 		    integer_size * num_integers)
2533 			return (E2BIG);
2534 		fzap_leaf_array(&zl, zc, integer_size, num_integers, value);
2535 		return (0);
2536 	}
2537 
2538 	return (ENOENT);
2539 }
2540 
2541 /*
2542  * Lookup a name in a zap object and return its value as a uint64_t.
2543  */
2544 static int
2545 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2546     uint64_t integer_size, uint64_t num_integers, void *value)
2547 {
2548 	int rc;
2549 	uint64_t zap_type;
2550 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2551 
2552 	rc = dnode_read(spa, dnode, 0, zap_scratch, size);
2553 	if (rc)
2554 		return (rc);
2555 
2556 	zap_type = *(uint64_t *)zap_scratch;
2557 	if (zap_type == ZBT_MICRO)
2558 		return (mzap_lookup(dnode, name, value));
2559 	else if (zap_type == ZBT_HEADER) {
2560 		return (fzap_lookup(spa, dnode, name, integer_size,
2561 		    num_integers, value));
2562 	}
2563 	printf("ZFS: invalid zap_type=%d\n", (int)zap_type);
2564 	return (EIO);
2565 }
2566 
2567 /*
2568  * List a microzap directory. Assumes that the zap scratch buffer contains
2569  * the directory contents.
2570  */
2571 static int
2572 mzap_list(const dnode_phys_t *dnode, int (*callback)(const char *, uint64_t))
2573 {
2574 	const mzap_phys_t *mz;
2575 	const mzap_ent_phys_t *mze;
2576 	size_t size;
2577 	int chunks, i, rc;
2578 
2579 	/*
2580 	 * Microzap objects use exactly one block. Read the whole
2581 	 * thing.
2582 	 */
2583 	size = dnode->dn_datablkszsec * 512;
2584 	mz = (const mzap_phys_t *) zap_scratch;
2585 	chunks = size / MZAP_ENT_LEN - 1;
2586 
2587 	for (i = 0; i < chunks; i++) {
2588 		mze = &mz->mz_chunk[i];
2589 		if (mze->mze_name[0]) {
2590 			rc = callback(mze->mze_name, mze->mze_value);
2591 			if (rc != 0)
2592 				return (rc);
2593 		}
2594 	}
2595 
2596 	return (0);
2597 }
2598 
2599 /*
2600  * List a fatzap directory. Assumes that the zap scratch buffer contains
2601  * the directory header.
2602  */
2603 static int
2604 fzap_list(const spa_t *spa, const dnode_phys_t *dnode,
2605     int (*callback)(const char *, uint64_t))
2606 {
2607 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2608 	zap_phys_t zh = *(zap_phys_t *)zap_scratch;
2609 	fat_zap_t z;
2610 	int i, j, rc;
2611 
2612 	if (zh.zap_magic != ZAP_MAGIC)
2613 		return (EIO);
2614 
2615 	z.zap_block_shift = ilog2(bsize);
2616 	z.zap_phys = (zap_phys_t *)zap_scratch;
2617 
2618 	/*
2619 	 * This assumes that the leaf blocks start at block 1. The
2620 	 * documentation isn't exactly clear on this.
2621 	 */
2622 	zap_leaf_t zl;
2623 	zl.l_bs = z.zap_block_shift;
2624 	for (i = 0; i < zh.zap_num_leafs; i++) {
2625 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2626 		char name[256], *p;
2627 		uint64_t value;
2628 
2629 		if (dnode_read(spa, dnode, off, zap_scratch, bsize))
2630 			return (EIO);
2631 
2632 		zl.l_phys = (zap_leaf_phys_t *)zap_scratch;
2633 
2634 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2635 			zap_leaf_chunk_t *zc, *nc;
2636 			int namelen;
2637 
2638 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2639 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2640 				continue;
2641 			namelen = zc->l_entry.le_name_numints;
2642 			if (namelen > sizeof(name))
2643 				namelen = sizeof(name);
2644 
2645 			/*
2646 			 * Paste the name back together.
2647 			 */
2648 			nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
2649 			p = name;
2650 			while (namelen > 0) {
2651 				int len;
2652 				len = namelen;
2653 				if (len > ZAP_LEAF_ARRAY_BYTES)
2654 					len = ZAP_LEAF_ARRAY_BYTES;
2655 				memcpy(p, nc->l_array.la_array, len);
2656 				p += len;
2657 				namelen -= len;
2658 				nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
2659 			}
2660 
2661 			/*
2662 			 * Assume the first eight bytes of the value are
2663 			 * a uint64_t.
2664 			 */
2665 			value = fzap_leaf_value(&zl, zc);
2666 
2667 			/* printf("%s 0x%jx\n", name, (uintmax_t)value); */
2668 			rc = callback((const char *)name, value);
2669 			if (rc != 0)
2670 				return (rc);
2671 		}
2672 	}
2673 
2674 	return (0);
2675 }
2676 
2677 static int zfs_printf(const char *name, uint64_t value __unused)
2678 {
2679 
2680 	printf("%s\n", name);
2681 
2682 	return (0);
2683 }
2684 
2685 /*
2686  * List a zap directory.
2687  */
2688 static int
2689 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
2690 {
2691 	uint64_t zap_type;
2692 	size_t size = dnode->dn_datablkszsec * 512;
2693 
2694 	if (dnode_read(spa, dnode, 0, zap_scratch, size))
2695 		return (EIO);
2696 
2697 	zap_type = *(uint64_t *)zap_scratch;
2698 	if (zap_type == ZBT_MICRO)
2699 		return (mzap_list(dnode, zfs_printf));
2700 	else
2701 		return (fzap_list(spa, dnode, zfs_printf));
2702 }
2703 
2704 static int
2705 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum,
2706     dnode_phys_t *dnode)
2707 {
2708 	off_t offset;
2709 
2710 	offset = objnum * sizeof(dnode_phys_t);
2711 	return dnode_read(spa, &os->os_meta_dnode, offset,
2712 		dnode, sizeof(dnode_phys_t));
2713 }
2714 
2715 static int
2716 mzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name,
2717     uint64_t value)
2718 {
2719 	const mzap_phys_t *mz;
2720 	const mzap_ent_phys_t *mze;
2721 	size_t size;
2722 	int chunks, i;
2723 
2724 	/*
2725 	 * Microzap objects use exactly one block. Read the whole
2726 	 * thing.
2727 	 */
2728 	size = dnode->dn_datablkszsec * 512;
2729 
2730 	mz = (const mzap_phys_t *)zap_scratch;
2731 	chunks = size / MZAP_ENT_LEN - 1;
2732 
2733 	for (i = 0; i < chunks; i++) {
2734 		mze = &mz->mz_chunk[i];
2735 		if (value == mze->mze_value) {
2736 			strcpy(name, mze->mze_name);
2737 			return (0);
2738 		}
2739 	}
2740 
2741 	return (ENOENT);
2742 }
2743 
2744 static void
2745 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
2746 {
2747 	size_t namelen;
2748 	const zap_leaf_chunk_t *nc;
2749 	char *p;
2750 
2751 	namelen = zc->l_entry.le_name_numints;
2752 
2753 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2754 	p = name;
2755 	while (namelen > 0) {
2756 		size_t len;
2757 		len = namelen;
2758 		if (len > ZAP_LEAF_ARRAY_BYTES)
2759 			len = ZAP_LEAF_ARRAY_BYTES;
2760 		memcpy(p, nc->l_array.la_array, len);
2761 		p += len;
2762 		namelen -= len;
2763 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2764 	}
2765 
2766 	*p = '\0';
2767 }
2768 
2769 static int
2770 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name,
2771     uint64_t value)
2772 {
2773 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2774 	zap_phys_t zh = *(zap_phys_t *)zap_scratch;
2775 	fat_zap_t z;
2776 	int i, j;
2777 
2778 	if (zh.zap_magic != ZAP_MAGIC)
2779 		return (EIO);
2780 
2781 	z.zap_block_shift = ilog2(bsize);
2782 	z.zap_phys = (zap_phys_t *)zap_scratch;
2783 
2784 	/*
2785 	 * This assumes that the leaf blocks start at block 1. The
2786 	 * documentation isn't exactly clear on this.
2787 	 */
2788 	zap_leaf_t zl;
2789 	zl.l_bs = z.zap_block_shift;
2790 	for (i = 0; i < zh.zap_num_leafs; i++) {
2791 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2792 
2793 		if (dnode_read(spa, dnode, off, zap_scratch, bsize))
2794 			return (EIO);
2795 
2796 		zl.l_phys = (zap_leaf_phys_t *)zap_scratch;
2797 
2798 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2799 			zap_leaf_chunk_t *zc;
2800 
2801 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2802 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2803 				continue;
2804 			if (zc->l_entry.le_value_intlen != 8 ||
2805 			    zc->l_entry.le_value_numints != 1)
2806 				continue;
2807 
2808 			if (fzap_leaf_value(&zl, zc) == value) {
2809 				fzap_name_copy(&zl, zc, name);
2810 				return (0);
2811 			}
2812 		}
2813 	}
2814 
2815 	return (ENOENT);
2816 }
2817 
2818 static int
2819 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name,
2820     uint64_t value)
2821 {
2822 	int rc;
2823 	uint64_t zap_type;
2824 	size_t size = dnode->dn_datablkszsec * 512;
2825 
2826 	rc = dnode_read(spa, dnode, 0, zap_scratch, size);
2827 	if (rc)
2828 		return (rc);
2829 
2830 	zap_type = *(uint64_t *)zap_scratch;
2831 	if (zap_type == ZBT_MICRO)
2832 		return (mzap_rlookup(spa, dnode, name, value));
2833 	else
2834 		return (fzap_rlookup(spa, dnode, name, value));
2835 }
2836 
2837 static int
2838 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
2839 {
2840 	char name[256];
2841 	char component[256];
2842 	uint64_t dir_obj, parent_obj, child_dir_zapobj;
2843 	dnode_phys_t child_dir_zap, dataset, dir, parent;
2844 	dsl_dir_phys_t *dd;
2845 	dsl_dataset_phys_t *ds;
2846 	char *p;
2847 	int len;
2848 
2849 	p = &name[sizeof(name) - 1];
2850 	*p = '\0';
2851 
2852 	if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2853 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2854 		return (EIO);
2855 	}
2856 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2857 	dir_obj = ds->ds_dir_obj;
2858 
2859 	for (;;) {
2860 		if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir) != 0)
2861 			return (EIO);
2862 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2863 
2864 		/* Actual loop condition. */
2865 		parent_obj = dd->dd_parent_obj;
2866 		if (parent_obj == 0)
2867 			break;
2868 
2869 		if (objset_get_dnode(spa, &spa->spa_mos, parent_obj,
2870 		    &parent) != 0)
2871 			return (EIO);
2872 		dd = (dsl_dir_phys_t *)&parent.dn_bonus;
2873 		child_dir_zapobj = dd->dd_child_dir_zapobj;
2874 		if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj,
2875 		    &child_dir_zap) != 0)
2876 			return (EIO);
2877 		if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
2878 			return (EIO);
2879 
2880 		len = strlen(component);
2881 		p -= len;
2882 		memcpy(p, component, len);
2883 		--p;
2884 		*p = '/';
2885 
2886 		/* Actual loop iteration. */
2887 		dir_obj = parent_obj;
2888 	}
2889 
2890 	if (*p != '\0')
2891 		++p;
2892 	strcpy(result, p);
2893 
2894 	return (0);
2895 }
2896 
2897 static int
2898 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
2899 {
2900 	char element[256];
2901 	uint64_t dir_obj, child_dir_zapobj;
2902 	dnode_phys_t child_dir_zap, dir;
2903 	dsl_dir_phys_t *dd;
2904 	const char *p, *q;
2905 
2906 	if (objset_get_dnode(spa, &spa->spa_mos,
2907 	    DMU_POOL_DIRECTORY_OBJECT, &dir))
2908 		return (EIO);
2909 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
2910 	    1, &dir_obj))
2911 		return (EIO);
2912 
2913 	p = name;
2914 	for (;;) {
2915 		if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir))
2916 			return (EIO);
2917 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2918 
2919 		while (*p == '/')
2920 			p++;
2921 		/* Actual loop condition #1. */
2922 		if (*p == '\0')
2923 			break;
2924 
2925 		q = strchr(p, '/');
2926 		if (q) {
2927 			memcpy(element, p, q - p);
2928 			element[q - p] = '\0';
2929 			p = q + 1;
2930 		} else {
2931 			strcpy(element, p);
2932 			p += strlen(p);
2933 		}
2934 
2935 		child_dir_zapobj = dd->dd_child_dir_zapobj;
2936 		if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj,
2937 		    &child_dir_zap) != 0)
2938 			return (EIO);
2939 
2940 		/* Actual loop condition #2. */
2941 		if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
2942 		    1, &dir_obj) != 0)
2943 			return (ENOENT);
2944 	}
2945 
2946 	*objnum = dd->dd_head_dataset_obj;
2947 	return (0);
2948 }
2949 
2950 #ifndef BOOT2
2951 static int
2952 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
2953 {
2954 	uint64_t dir_obj, child_dir_zapobj;
2955 	dnode_phys_t child_dir_zap, dir, dataset;
2956 	dsl_dataset_phys_t *ds;
2957 	dsl_dir_phys_t *dd;
2958 
2959 	if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2960 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2961 		return (EIO);
2962 	}
2963 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2964 	dir_obj = ds->ds_dir_obj;
2965 
2966 	if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir)) {
2967 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2968 		return (EIO);
2969 	}
2970 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2971 
2972 	child_dir_zapobj = dd->dd_child_dir_zapobj;
2973 	if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj,
2974 	    &child_dir_zap) != 0) {
2975 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2976 		return (EIO);
2977 	}
2978 
2979 	return (zap_list(spa, &child_dir_zap) != 0);
2980 }
2981 
2982 int
2983 zfs_callback_dataset(const spa_t *spa, uint64_t objnum,
2984     int (*callback)(const char *, uint64_t))
2985 {
2986 	uint64_t dir_obj, child_dir_zapobj, zap_type;
2987 	dnode_phys_t child_dir_zap, dir, dataset;
2988 	dsl_dataset_phys_t *ds;
2989 	dsl_dir_phys_t *dd;
2990 	int err;
2991 
2992 	err = objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset);
2993 	if (err != 0) {
2994 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2995 		return (err);
2996 	}
2997 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2998 	dir_obj = ds->ds_dir_obj;
2999 
3000 	err = objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir);
3001 	if (err != 0) {
3002 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
3003 		return (err);
3004 	}
3005 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3006 
3007 	child_dir_zapobj = dd->dd_child_dir_zapobj;
3008 	err = objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj,
3009 	    &child_dir_zap);
3010 	if (err != 0) {
3011 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
3012 		return (err);
3013 	}
3014 
3015 	err = dnode_read(spa, &child_dir_zap, 0, zap_scratch,
3016 	    child_dir_zap.dn_datablkszsec * 512);
3017 	if (err != 0)
3018 		return (err);
3019 
3020 	zap_type = *(uint64_t *)zap_scratch;
3021 	if (zap_type == ZBT_MICRO)
3022 		return (mzap_list(&child_dir_zap, callback));
3023 	else
3024 		return (fzap_list(spa, &child_dir_zap, callback));
3025 }
3026 #endif
3027 
3028 /*
3029  * Find the object set given the object number of its dataset object
3030  * and return its details in *objset
3031  */
3032 static int
3033 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
3034 {
3035 	dnode_phys_t dataset;
3036 	dsl_dataset_phys_t *ds;
3037 
3038 	if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
3039 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
3040 		return (EIO);
3041 	}
3042 
3043 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
3044 	if (zio_read(spa, &ds->ds_bp, objset)) {
3045 		printf("ZFS: can't read object set for dataset %ju\n",
3046 		    (uintmax_t)objnum);
3047 		return (EIO);
3048 	}
3049 
3050 	return (0);
3051 }
3052 
3053 /*
3054  * Find the object set pointed to by the BOOTFS property or the root
3055  * dataset if there is none and return its details in *objset
3056  */
3057 static int
3058 zfs_get_root(const spa_t *spa, uint64_t *objid)
3059 {
3060 	dnode_phys_t dir, propdir;
3061 	uint64_t props, bootfs, root;
3062 
3063 	*objid = 0;
3064 
3065 	/*
3066 	 * Start with the MOS directory object.
3067 	 */
3068 	if (objset_get_dnode(spa, &spa->spa_mos,
3069 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3070 		printf("ZFS: can't read MOS object directory\n");
3071 		return (EIO);
3072 	}
3073 
3074 	/*
3075 	 * Lookup the pool_props and see if we can find a bootfs.
3076 	 */
3077 	if (zap_lookup(spa, &dir, DMU_POOL_PROPS,
3078 	    sizeof(props), 1, &props) == 0 &&
3079 	    objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0 &&
3080 	    zap_lookup(spa, &propdir, "bootfs",
3081 	    sizeof(bootfs), 1, &bootfs) == 0 && bootfs != 0) {
3082 		*objid = bootfs;
3083 		return (0);
3084 	}
3085 	/*
3086 	 * Lookup the root dataset directory
3087 	 */
3088 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET,
3089 	    sizeof(root), 1, &root) ||
3090 	    objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
3091 		printf("ZFS: can't find root dsl_dir\n");
3092 		return (EIO);
3093 	}
3094 
3095 	/*
3096 	 * Use the information from the dataset directory's bonus buffer
3097 	 * to find the dataset object and from that the object set itself.
3098 	 */
3099 	dsl_dir_phys_t *dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3100 	*objid = dd->dd_head_dataset_obj;
3101 	return (0);
3102 }
3103 
3104 static int
3105 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
3106 {
3107 
3108 	mount->spa = spa;
3109 
3110 	/*
3111 	 * Find the root object set if not explicitly provided
3112 	 */
3113 	if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
3114 		printf("ZFS: can't find root filesystem\n");
3115 		return (EIO);
3116 	}
3117 
3118 	if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
3119 		printf("ZFS: can't open root filesystem\n");
3120 		return (EIO);
3121 	}
3122 
3123 	mount->rootobj = rootobj;
3124 
3125 	return (0);
3126 }
3127 
3128 /*
3129  * callback function for feature name checks.
3130  */
3131 static int
3132 check_feature(const char *name, uint64_t value)
3133 {
3134 	int i;
3135 
3136 	if (value == 0)
3137 		return (0);
3138 	if (name[0] == '\0')
3139 		return (0);
3140 
3141 	for (i = 0; features_for_read[i] != NULL; i++) {
3142 		if (strcmp(name, features_for_read[i]) == 0)
3143 			return (0);
3144 	}
3145 	printf("ZFS: unsupported feature: %s\n", name);
3146 	return (EIO);
3147 }
3148 
3149 /*
3150  * Checks whether the MOS features that are active are supported.
3151  */
3152 static int
3153 check_mos_features(const spa_t *spa)
3154 {
3155 	dnode_phys_t dir;
3156 	uint64_t objnum, zap_type;
3157 	size_t size;
3158 	int rc;
3159 
3160 	if ((rc = objset_get_dnode(spa, &spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
3161 	    &dir)) != 0)
3162 		return (rc);
3163 	if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
3164 	    sizeof (objnum), 1, &objnum)) != 0) {
3165 		/*
3166 		 * It is older pool without features. As we have already
3167 		 * tested the label, just return without raising the error.
3168 		 */
3169 		return (0);
3170 	}
3171 
3172 	if ((rc = objset_get_dnode(spa, &spa->spa_mos, objnum, &dir)) != 0)
3173 		return (rc);
3174 
3175 	if (dir.dn_type != DMU_OTN_ZAP_METADATA)
3176 		return (EIO);
3177 
3178 	size = dir.dn_datablkszsec * 512;
3179 	if (dnode_read(spa, &dir, 0, zap_scratch, size))
3180 		return (EIO);
3181 
3182 	zap_type = *(uint64_t *)zap_scratch;
3183 	if (zap_type == ZBT_MICRO)
3184 		rc = mzap_list(&dir, check_feature);
3185 	else
3186 		rc = fzap_list(spa, &dir, check_feature);
3187 
3188 	return (rc);
3189 }
3190 
3191 static int
3192 load_nvlist(spa_t *spa, uint64_t obj, unsigned char **value)
3193 {
3194 	dnode_phys_t dir;
3195 	size_t size;
3196 	int rc;
3197 	unsigned char *nv;
3198 
3199 	*value = NULL;
3200 	if ((rc = objset_get_dnode(spa, &spa->spa_mos, obj, &dir)) != 0)
3201 		return (rc);
3202 	if (dir.dn_type != DMU_OT_PACKED_NVLIST &&
3203 	    dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) {
3204 		return (EIO);
3205 	}
3206 
3207 	if (dir.dn_bonuslen != sizeof (uint64_t))
3208 		return (EIO);
3209 
3210 	size = *(uint64_t *)DN_BONUS(&dir);
3211 	nv = malloc(size);
3212 	if (nv == NULL)
3213 		return (ENOMEM);
3214 
3215 	rc = dnode_read(spa, &dir, 0, nv, size);
3216 	if (rc != 0) {
3217 		free(nv);
3218 		nv = NULL;
3219 		return (rc);
3220 	}
3221 	*value = nv;
3222 	return (rc);
3223 }
3224 
3225 static int
3226 zfs_spa_init(spa_t *spa)
3227 {
3228 	dnode_phys_t dir;
3229 	uint64_t config_object;
3230 	unsigned char *nvlist;
3231 	int rc;
3232 
3233 	if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
3234 		printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
3235 		return (EIO);
3236 	}
3237 	if (spa->spa_mos.os_type != DMU_OST_META) {
3238 		printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
3239 		return (EIO);
3240 	}
3241 
3242 	if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT,
3243 	    &dir)) {
3244 		printf("ZFS: failed to read pool %s directory object\n",
3245 		    spa->spa_name);
3246 		return (EIO);
3247 	}
3248 	/* this is allowed to fail, older pools do not have salt */
3249 	rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
3250 	    sizeof (spa->spa_cksum_salt.zcs_bytes),
3251 	    spa->spa_cksum_salt.zcs_bytes);
3252 
3253 	rc = check_mos_features(spa);
3254 	if (rc != 0) {
3255 		printf("ZFS: pool %s is not supported\n", spa->spa_name);
3256 		return (rc);
3257 	}
3258 
3259 	rc = zap_lookup(spa, &dir, DMU_POOL_CONFIG,
3260 	    sizeof (config_object), 1, &config_object);
3261 	if (rc != 0) {
3262 		printf("ZFS: can not read MOS %s\n", DMU_POOL_CONFIG);
3263 		return (EIO);
3264 	}
3265 	rc = load_nvlist(spa, config_object, &nvlist);
3266 	if (rc != 0)
3267 		return (rc);
3268 
3269 	/* Update vdevs from MOS config. */
3270 	rc = vdev_init_from_nvlist(spa, nvlist + 4);
3271 	free(nvlist);
3272 	return (rc);
3273 }
3274 
3275 static int
3276 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
3277 {
3278 
3279 	if (dn->dn_bonustype != DMU_OT_SA) {
3280 		znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
3281 
3282 		sb->st_mode = zp->zp_mode;
3283 		sb->st_uid = zp->zp_uid;
3284 		sb->st_gid = zp->zp_gid;
3285 		sb->st_size = zp->zp_size;
3286 	} else {
3287 		sa_hdr_phys_t *sahdrp;
3288 		int hdrsize;
3289 		size_t size = 0;
3290 		void *buf = NULL;
3291 
3292 		if (dn->dn_bonuslen != 0)
3293 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3294 		else {
3295 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
3296 				blkptr_t *bp = DN_SPILL_BLKPTR(dn);
3297 				int error;
3298 
3299 				size = BP_GET_LSIZE(bp);
3300 				buf = zfs_alloc(size);
3301 				error = zio_read(spa, bp, buf);
3302 				if (error != 0) {
3303 					zfs_free(buf, size);
3304 					return (error);
3305 				}
3306 				sahdrp = buf;
3307 			} else {
3308 				return (EIO);
3309 			}
3310 		}
3311 		hdrsize = SA_HDR_SIZE(sahdrp);
3312 		sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
3313 		    SA_MODE_OFFSET);
3314 		sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
3315 		    SA_UID_OFFSET);
3316 		sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
3317 		    SA_GID_OFFSET);
3318 		sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
3319 		    SA_SIZE_OFFSET);
3320 		if (buf != NULL)
3321 			zfs_free(buf, size);
3322 	}
3323 
3324 	return (0);
3325 }
3326 
3327 static int
3328 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
3329 {
3330 	int rc = 0;
3331 
3332 	if (dn->dn_bonustype == DMU_OT_SA) {
3333 		sa_hdr_phys_t *sahdrp = NULL;
3334 		size_t size = 0;
3335 		void *buf = NULL;
3336 		int hdrsize;
3337 		char *p;
3338 
3339 		if (dn->dn_bonuslen != 0)
3340 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3341 		else {
3342 			blkptr_t *bp;
3343 
3344 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
3345 				return (EIO);
3346 			bp = DN_SPILL_BLKPTR(dn);
3347 
3348 			size = BP_GET_LSIZE(bp);
3349 			buf = zfs_alloc(size);
3350 			rc = zio_read(spa, bp, buf);
3351 			if (rc != 0) {
3352 				zfs_free(buf, size);
3353 				return (rc);
3354 			}
3355 			sahdrp = buf;
3356 		}
3357 		hdrsize = SA_HDR_SIZE(sahdrp);
3358 		p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
3359 		memcpy(path, p, psize);
3360 		if (buf != NULL)
3361 			zfs_free(buf, size);
3362 		return (0);
3363 	}
3364 	/*
3365 	 * Second test is purely to silence bogus compiler
3366 	 * warning about accessing past the end of dn_bonus.
3367 	 */
3368 	if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
3369 	    sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
3370 		memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
3371 	} else {
3372 		rc = dnode_read(spa, dn, 0, path, psize);
3373 	}
3374 	return (rc);
3375 }
3376 
3377 struct obj_list {
3378 	uint64_t		objnum;
3379 	STAILQ_ENTRY(obj_list)	entry;
3380 };
3381 
3382 /*
3383  * Lookup a file and return its dnode.
3384  */
3385 static int
3386 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
3387 {
3388 	int rc;
3389 	uint64_t objnum;
3390 	const spa_t *spa;
3391 	dnode_phys_t dn;
3392 	const char *p, *q;
3393 	char element[256];
3394 	char path[1024];
3395 	int symlinks_followed = 0;
3396 	struct stat sb;
3397 	struct obj_list *entry, *tentry;
3398 	STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
3399 
3400 	spa = mount->spa;
3401 	if (mount->objset.os_type != DMU_OST_ZFS) {
3402 		printf("ZFS: unexpected object set type %ju\n",
3403 		    (uintmax_t)mount->objset.os_type);
3404 		return (EIO);
3405 	}
3406 
3407 	if ((entry = malloc(sizeof(struct obj_list))) == NULL)
3408 		return (ENOMEM);
3409 
3410 	/*
3411 	 * Get the root directory dnode.
3412 	 */
3413 	rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
3414 	if (rc) {
3415 		free(entry);
3416 		return (rc);
3417 	}
3418 
3419 	rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof(objnum), 1, &objnum);
3420 	if (rc) {
3421 		free(entry);
3422 		return (rc);
3423 	}
3424 	entry->objnum = objnum;
3425 	STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3426 
3427 	rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3428 	if (rc != 0)
3429 		goto done;
3430 
3431 	p = upath;
3432 	while (p && *p) {
3433 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3434 		if (rc != 0)
3435 			goto done;
3436 
3437 		while (*p == '/')
3438 			p++;
3439 		if (*p == '\0')
3440 			break;
3441 		q = p;
3442 		while (*q != '\0' && *q != '/')
3443 			q++;
3444 
3445 		/* skip dot */
3446 		if (p + 1 == q && p[0] == '.') {
3447 			p++;
3448 			continue;
3449 		}
3450 		/* double dot */
3451 		if (p + 2 == q && p[0] == '.' && p[1] == '.') {
3452 			p += 2;
3453 			if (STAILQ_FIRST(&on_cache) ==
3454 			    STAILQ_LAST(&on_cache, obj_list, entry)) {
3455 				rc = ENOENT;
3456 				goto done;
3457 			}
3458 			entry = STAILQ_FIRST(&on_cache);
3459 			STAILQ_REMOVE_HEAD(&on_cache, entry);
3460 			free(entry);
3461 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3462 			continue;
3463 		}
3464 		if (q - p + 1 > sizeof(element)) {
3465 			rc = ENAMETOOLONG;
3466 			goto done;
3467 		}
3468 		memcpy(element, p, q - p);
3469 		element[q - p] = 0;
3470 		p = q;
3471 
3472 		if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
3473 			goto done;
3474 		if (!S_ISDIR(sb.st_mode)) {
3475 			rc = ENOTDIR;
3476 			goto done;
3477 		}
3478 
3479 		rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum);
3480 		if (rc)
3481 			goto done;
3482 		objnum = ZFS_DIRENT_OBJ(objnum);
3483 
3484 		if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
3485 			rc = ENOMEM;
3486 			goto done;
3487 		}
3488 		entry->objnum = objnum;
3489 		STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3490 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3491 		if (rc)
3492 			goto done;
3493 
3494 		/*
3495 		 * Check for symlink.
3496 		 */
3497 		rc = zfs_dnode_stat(spa, &dn, &sb);
3498 		if (rc)
3499 			goto done;
3500 		if (S_ISLNK(sb.st_mode)) {
3501 			if (symlinks_followed > 10) {
3502 				rc = EMLINK;
3503 				goto done;
3504 			}
3505 			symlinks_followed++;
3506 
3507 			/*
3508 			 * Read the link value and copy the tail of our
3509 			 * current path onto the end.
3510 			 */
3511 			if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
3512 				rc = ENAMETOOLONG;
3513 				goto done;
3514 			}
3515 			strcpy(&path[sb.st_size], p);
3516 
3517 			rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
3518 			if (rc != 0)
3519 				goto done;
3520 
3521 			/*
3522 			 * Restart with the new path, starting either at
3523 			 * the root or at the parent depending whether or
3524 			 * not the link is relative.
3525 			 */
3526 			p = path;
3527 			if (*p == '/') {
3528 				while (STAILQ_FIRST(&on_cache) !=
3529 				    STAILQ_LAST(&on_cache, obj_list, entry)) {
3530 					entry = STAILQ_FIRST(&on_cache);
3531 					STAILQ_REMOVE_HEAD(&on_cache, entry);
3532 					free(entry);
3533 				}
3534 			} else {
3535 				entry = STAILQ_FIRST(&on_cache);
3536 				STAILQ_REMOVE_HEAD(&on_cache, entry);
3537 				free(entry);
3538 			}
3539 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3540 		}
3541 	}
3542 
3543 	*dnode = dn;
3544 done:
3545 	STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)
3546 		free(entry);
3547 	return (rc);
3548 }
3549