xref: /freebsd/stand/libsa/zfs/zfsimpl.c (revision f6385d921b2f354d71256d1d0392122597e0fd33)
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 	"org.freebsd:zstd_compress",
131 	"com.datto:encryption",
132 	NULL
133 };
134 
135 /*
136  * List of all pools, chained through spa_link.
137  */
138 static spa_list_t zfs_pools;
139 
140 static const dnode_phys_t *dnode_cache_obj;
141 static uint64_t dnode_cache_bn;
142 static char *dnode_cache_buf;
143 
144 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
145 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
146 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
147 static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode,
148     const char *name, uint64_t integer_size, uint64_t num_integers,
149     void *value);
150 static int objset_get_dnode(const spa_t *, const objset_phys_t *, uint64_t,
151     dnode_phys_t *);
152 static int dnode_read(const spa_t *, const dnode_phys_t *, off_t, void *,
153     size_t);
154 static int vdev_indirect_read(vdev_t *, const blkptr_t *, void *, off_t,
155     size_t);
156 static int vdev_mirror_read(vdev_t *, const blkptr_t *, void *, off_t, size_t);
157 vdev_indirect_mapping_t *vdev_indirect_mapping_open(spa_t *, objset_phys_t *,
158     uint64_t);
159 vdev_indirect_mapping_entry_phys_t *
160     vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *, uint64_t,
161     uint64_t, uint64_t *);
162 
163 static void
164 zfs_init(void)
165 {
166 	STAILQ_INIT(&zfs_vdevs);
167 	STAILQ_INIT(&zfs_pools);
168 
169 	dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
170 
171 	zfs_init_crc();
172 }
173 
174 static int
175 nvlist_check_features_for_read(nvlist_t *nvl)
176 {
177 	nvlist_t *features = NULL;
178 	nvs_data_t *data;
179 	nvp_header_t *nvp;
180 	nv_string_t *nvp_name;
181 	int rc;
182 
183 	rc = nvlist_find(nvl, ZPOOL_CONFIG_FEATURES_FOR_READ,
184 	    DATA_TYPE_NVLIST, NULL, &features, NULL);
185 	if (rc != 0)
186 		return (rc);
187 
188 	data = (nvs_data_t *)features->nv_data;
189 	nvp = &data->nvl_pair;	/* first pair in nvlist */
190 
191 	while (nvp->encoded_size != 0 && nvp->decoded_size != 0) {
192 		int i, found;
193 
194 		nvp_name = (nv_string_t *)((uintptr_t)nvp + sizeof(*nvp));
195 		found = 0;
196 
197 		for (i = 0; features_for_read[i] != NULL; i++) {
198 			if (memcmp(nvp_name->nv_data, features_for_read[i],
199 			    nvp_name->nv_size) == 0) {
200 				found = 1;
201 				break;
202 			}
203 		}
204 
205 		if (!found) {
206 			printf("ZFS: unsupported feature: %.*s\n",
207 			    nvp_name->nv_size, nvp_name->nv_data);
208 			rc = EIO;
209 		}
210 		nvp = (nvp_header_t *)((uint8_t *)nvp + nvp->encoded_size);
211 	}
212 	nvlist_destroy(features);
213 
214 	return (rc);
215 }
216 
217 static int
218 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
219     off_t offset, size_t size)
220 {
221 	size_t psize;
222 	int rc;
223 
224 	if (!vdev->v_phys_read)
225 		return (EIO);
226 
227 	if (bp) {
228 		psize = BP_GET_PSIZE(bp);
229 	} else {
230 		psize = size;
231 	}
232 
233 	rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
234 	if (rc == 0) {
235 		if (bp != NULL)
236 			rc = zio_checksum_verify(vdev->v_spa, bp, buf);
237 	}
238 
239 	return (rc);
240 }
241 
242 typedef struct remap_segment {
243 	vdev_t *rs_vd;
244 	uint64_t rs_offset;
245 	uint64_t rs_asize;
246 	uint64_t rs_split_offset;
247 	list_node_t rs_node;
248 } remap_segment_t;
249 
250 static remap_segment_t *
251 rs_alloc(vdev_t *vd, uint64_t offset, uint64_t asize, uint64_t split_offset)
252 {
253 	remap_segment_t *rs = malloc(sizeof (remap_segment_t));
254 
255 	if (rs != NULL) {
256 		rs->rs_vd = vd;
257 		rs->rs_offset = offset;
258 		rs->rs_asize = asize;
259 		rs->rs_split_offset = split_offset;
260 	}
261 
262 	return (rs);
263 }
264 
265 vdev_indirect_mapping_t *
266 vdev_indirect_mapping_open(spa_t *spa, objset_phys_t *os,
267     uint64_t mapping_object)
268 {
269 	vdev_indirect_mapping_t *vim;
270 	vdev_indirect_mapping_phys_t *vim_phys;
271 	int rc;
272 
273 	vim = calloc(1, sizeof (*vim));
274 	if (vim == NULL)
275 		return (NULL);
276 
277 	vim->vim_dn = calloc(1, sizeof (*vim->vim_dn));
278 	if (vim->vim_dn == NULL) {
279 		free(vim);
280 		return (NULL);
281 	}
282 
283 	rc = objset_get_dnode(spa, os, mapping_object, vim->vim_dn);
284 	if (rc != 0) {
285 		free(vim->vim_dn);
286 		free(vim);
287 		return (NULL);
288 	}
289 
290 	vim->vim_spa = spa;
291 	vim->vim_phys = malloc(sizeof (*vim->vim_phys));
292 	if (vim->vim_phys == NULL) {
293 		free(vim->vim_dn);
294 		free(vim);
295 		return (NULL);
296 	}
297 
298 	vim_phys = (vdev_indirect_mapping_phys_t *)DN_BONUS(vim->vim_dn);
299 	*vim->vim_phys = *vim_phys;
300 
301 	vim->vim_objset = os;
302 	vim->vim_object = mapping_object;
303 	vim->vim_entries = NULL;
304 
305 	vim->vim_havecounts =
306 	    (vim->vim_dn->dn_bonuslen > VDEV_INDIRECT_MAPPING_SIZE_V0);
307 
308 	return (vim);
309 }
310 
311 /*
312  * Compare an offset with an indirect mapping entry; there are three
313  * possible scenarios:
314  *
315  *     1. The offset is "less than" the mapping entry; meaning the
316  *        offset is less than the source offset of the mapping entry. In
317  *        this case, there is no overlap between the offset and the
318  *        mapping entry and -1 will be returned.
319  *
320  *     2. The offset is "greater than" the mapping entry; meaning the
321  *        offset is greater than the mapping entry's source offset plus
322  *        the entry's size. In this case, there is no overlap between
323  *        the offset and the mapping entry and 1 will be returned.
324  *
325  *        NOTE: If the offset is actually equal to the entry's offset
326  *        plus size, this is considered to be "greater" than the entry,
327  *        and this case applies (i.e. 1 will be returned). Thus, the
328  *        entry's "range" can be considered to be inclusive at its
329  *        start, but exclusive at its end: e.g. [src, src + size).
330  *
331  *     3. The last case to consider is if the offset actually falls
332  *        within the mapping entry's range. If this is the case, the
333  *        offset is considered to be "equal to" the mapping entry and
334  *        0 will be returned.
335  *
336  *        NOTE: If the offset is equal to the entry's source offset,
337  *        this case applies and 0 will be returned. If the offset is
338  *        equal to the entry's source plus its size, this case does
339  *        *not* apply (see "NOTE" above for scenario 2), and 1 will be
340  *        returned.
341  */
342 static int
343 dva_mapping_overlap_compare(const void *v_key, const void *v_array_elem)
344 {
345 	const uint64_t *key = v_key;
346 	const vdev_indirect_mapping_entry_phys_t *array_elem =
347 	    v_array_elem;
348 	uint64_t src_offset = DVA_MAPPING_GET_SRC_OFFSET(array_elem);
349 
350 	if (*key < src_offset) {
351 		return (-1);
352 	} else if (*key < src_offset + DVA_GET_ASIZE(&array_elem->vimep_dst)) {
353 		return (0);
354 	} else {
355 		return (1);
356 	}
357 }
358 
359 /*
360  * Return array entry.
361  */
362 static vdev_indirect_mapping_entry_phys_t *
363 vdev_indirect_mapping_entry(vdev_indirect_mapping_t *vim, uint64_t index)
364 {
365 	uint64_t size;
366 	off_t offset = 0;
367 	int rc;
368 
369 	if (vim->vim_phys->vimp_num_entries == 0)
370 		return (NULL);
371 
372 	if (vim->vim_entries == NULL) {
373 		uint64_t bsize;
374 
375 		bsize = vim->vim_dn->dn_datablkszsec << SPA_MINBLOCKSHIFT;
376 		size = vim->vim_phys->vimp_num_entries *
377 		    sizeof (*vim->vim_entries);
378 		if (size > bsize) {
379 			size = bsize / sizeof (*vim->vim_entries);
380 			size *= sizeof (*vim->vim_entries);
381 		}
382 		vim->vim_entries = malloc(size);
383 		if (vim->vim_entries == NULL)
384 			return (NULL);
385 		vim->vim_num_entries = size / sizeof (*vim->vim_entries);
386 		offset = index * sizeof (*vim->vim_entries);
387 	}
388 
389 	/* We have data in vim_entries */
390 	if (offset == 0) {
391 		if (index >= vim->vim_entry_offset &&
392 		    index <= vim->vim_entry_offset + vim->vim_num_entries) {
393 			index -= vim->vim_entry_offset;
394 			return (&vim->vim_entries[index]);
395 		}
396 		offset = index * sizeof (*vim->vim_entries);
397 	}
398 
399 	vim->vim_entry_offset = index;
400 	size = vim->vim_num_entries * sizeof (*vim->vim_entries);
401 	rc = dnode_read(vim->vim_spa, vim->vim_dn, offset, vim->vim_entries,
402 	    size);
403 	if (rc != 0) {
404 		/* Read error, invalidate vim_entries. */
405 		free(vim->vim_entries);
406 		vim->vim_entries = NULL;
407 		return (NULL);
408 	}
409 	index -= vim->vim_entry_offset;
410 	return (&vim->vim_entries[index]);
411 }
412 
413 /*
414  * Returns the mapping entry for the given offset.
415  *
416  * It's possible that the given offset will not be in the mapping table
417  * (i.e. no mapping entries contain this offset), in which case, the
418  * return value value depends on the "next_if_missing" parameter.
419  *
420  * If the offset is not found in the table and "next_if_missing" is
421  * B_FALSE, then NULL will always be returned. The behavior is intended
422  * to allow consumers to get the entry corresponding to the offset
423  * parameter, iff the offset overlaps with an entry in the table.
424  *
425  * If the offset is not found in the table and "next_if_missing" is
426  * B_TRUE, then the entry nearest to the given offset will be returned,
427  * such that the entry's source offset is greater than the offset
428  * passed in (i.e. the "next" mapping entry in the table is returned, if
429  * the offset is missing from the table). If there are no entries whose
430  * source offset is greater than the passed in offset, NULL is returned.
431  */
432 static vdev_indirect_mapping_entry_phys_t *
433 vdev_indirect_mapping_entry_for_offset(vdev_indirect_mapping_t *vim,
434     uint64_t offset)
435 {
436 	ASSERT(vim->vim_phys->vimp_num_entries > 0);
437 
438 	vdev_indirect_mapping_entry_phys_t *entry;
439 
440 	uint64_t last = vim->vim_phys->vimp_num_entries - 1;
441 	uint64_t base = 0;
442 
443 	/*
444 	 * We don't define these inside of the while loop because we use
445 	 * their value in the case that offset isn't in the mapping.
446 	 */
447 	uint64_t mid;
448 	int result;
449 
450 	while (last >= base) {
451 		mid = base + ((last - base) >> 1);
452 
453 		entry = vdev_indirect_mapping_entry(vim, mid);
454 		if (entry == NULL)
455 			break;
456 		result = dva_mapping_overlap_compare(&offset, entry);
457 
458 		if (result == 0) {
459 			break;
460 		} else if (result < 0) {
461 			last = mid - 1;
462 		} else {
463 			base = mid + 1;
464 		}
465 	}
466 	return (entry);
467 }
468 
469 /*
470  * Given an indirect vdev and an extent on that vdev, it duplicates the
471  * physical entries of the indirect mapping that correspond to the extent
472  * to a new array and returns a pointer to it. In addition, copied_entries
473  * is populated with the number of mapping entries that were duplicated.
474  *
475  * Finally, since we are doing an allocation, it is up to the caller to
476  * free the array allocated in this function.
477  */
478 vdev_indirect_mapping_entry_phys_t *
479 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *vd, uint64_t offset,
480     uint64_t asize, uint64_t *copied_entries)
481 {
482 	vdev_indirect_mapping_entry_phys_t *duplicate_mappings = NULL;
483 	vdev_indirect_mapping_t *vim = vd->v_mapping;
484 	uint64_t entries = 0;
485 
486 	vdev_indirect_mapping_entry_phys_t *first_mapping =
487 	    vdev_indirect_mapping_entry_for_offset(vim, offset);
488 	ASSERT3P(first_mapping, !=, NULL);
489 
490 	vdev_indirect_mapping_entry_phys_t *m = first_mapping;
491 	while (asize > 0) {
492 		uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
493 		uint64_t inner_offset = offset - DVA_MAPPING_GET_SRC_OFFSET(m);
494 		uint64_t inner_size = MIN(asize, size - inner_offset);
495 
496 		offset += inner_size;
497 		asize -= inner_size;
498 		entries++;
499 		m++;
500 	}
501 
502 	size_t copy_length = entries * sizeof (*first_mapping);
503 	duplicate_mappings = malloc(copy_length);
504 	if (duplicate_mappings != NULL)
505 		bcopy(first_mapping, duplicate_mappings, copy_length);
506 	else
507 		entries = 0;
508 
509 	*copied_entries = entries;
510 
511 	return (duplicate_mappings);
512 }
513 
514 static vdev_t *
515 vdev_lookup_top(spa_t *spa, uint64_t vdev)
516 {
517 	vdev_t *rvd;
518 	vdev_list_t *vlist;
519 
520 	vlist = &spa->spa_root_vdev->v_children;
521 	STAILQ_FOREACH(rvd, vlist, v_childlink)
522 		if (rvd->v_id == vdev)
523 			break;
524 
525 	return (rvd);
526 }
527 
528 /*
529  * This is a callback for vdev_indirect_remap() which allocates an
530  * indirect_split_t for each split segment and adds it to iv_splits.
531  */
532 static void
533 vdev_indirect_gather_splits(uint64_t split_offset, vdev_t *vd, uint64_t offset,
534     uint64_t size, void *arg)
535 {
536 	int n = 1;
537 	zio_t *zio = arg;
538 	indirect_vsd_t *iv = zio->io_vsd;
539 
540 	if (vd->v_read == vdev_indirect_read)
541 		return;
542 
543 	if (vd->v_read == vdev_mirror_read)
544 		n = vd->v_nchildren;
545 
546 	indirect_split_t *is =
547 	    malloc(offsetof(indirect_split_t, is_child[n]));
548 	if (is == NULL) {
549 		zio->io_error = ENOMEM;
550 		return;
551 	}
552 	bzero(is, offsetof(indirect_split_t, is_child[n]));
553 
554 	is->is_children = n;
555 	is->is_size = size;
556 	is->is_split_offset = split_offset;
557 	is->is_target_offset = offset;
558 	is->is_vdev = vd;
559 
560 	/*
561 	 * Note that we only consider multiple copies of the data for
562 	 * *mirror* vdevs.  We don't for "replacing" or "spare" vdevs, even
563 	 * though they use the same ops as mirror, because there's only one
564 	 * "good" copy under the replacing/spare.
565 	 */
566 	if (vd->v_read == vdev_mirror_read) {
567 		int i = 0;
568 		vdev_t *kid;
569 
570 		STAILQ_FOREACH(kid, &vd->v_children, v_childlink) {
571 			is->is_child[i++].ic_vdev = kid;
572 		}
573 	} else {
574 		is->is_child[0].ic_vdev = vd;
575 	}
576 
577 	list_insert_tail(&iv->iv_splits, is);
578 }
579 
580 static void
581 vdev_indirect_remap(vdev_t *vd, uint64_t offset, uint64_t asize, void *arg)
582 {
583 	list_t stack;
584 	spa_t *spa = vd->v_spa;
585 	zio_t *zio = arg;
586 	remap_segment_t *rs;
587 
588 	list_create(&stack, sizeof (remap_segment_t),
589 	    offsetof(remap_segment_t, rs_node));
590 
591 	rs = rs_alloc(vd, offset, asize, 0);
592 	if (rs == NULL) {
593 		printf("vdev_indirect_remap: out of memory.\n");
594 		zio->io_error = ENOMEM;
595 	}
596 	for (; rs != NULL; rs = list_remove_head(&stack)) {
597 		vdev_t *v = rs->rs_vd;
598 		uint64_t num_entries = 0;
599 		/* vdev_indirect_mapping_t *vim = v->v_mapping; */
600 		vdev_indirect_mapping_entry_phys_t *mapping =
601 		    vdev_indirect_mapping_duplicate_adjacent_entries(v,
602 		    rs->rs_offset, rs->rs_asize, &num_entries);
603 
604 		if (num_entries == 0)
605 			zio->io_error = ENOMEM;
606 
607 		for (uint64_t i = 0; i < num_entries; i++) {
608 			vdev_indirect_mapping_entry_phys_t *m = &mapping[i];
609 			uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
610 			uint64_t dst_offset = DVA_GET_OFFSET(&m->vimep_dst);
611 			uint64_t dst_vdev = DVA_GET_VDEV(&m->vimep_dst);
612 			uint64_t inner_offset = rs->rs_offset -
613 			    DVA_MAPPING_GET_SRC_OFFSET(m);
614 			uint64_t inner_size =
615 			    MIN(rs->rs_asize, size - inner_offset);
616 			vdev_t *dst_v = vdev_lookup_top(spa, dst_vdev);
617 
618 			if (dst_v->v_read == vdev_indirect_read) {
619 				remap_segment_t *o;
620 
621 				o = rs_alloc(dst_v, dst_offset + inner_offset,
622 				    inner_size, rs->rs_split_offset);
623 				if (o == NULL) {
624 					printf("vdev_indirect_remap: "
625 					    "out of memory.\n");
626 					zio->io_error = ENOMEM;
627 					break;
628 				}
629 
630 				list_insert_head(&stack, o);
631 			}
632 			vdev_indirect_gather_splits(rs->rs_split_offset, dst_v,
633 			    dst_offset + inner_offset,
634 			    inner_size, arg);
635 
636 			/*
637 			 * vdev_indirect_gather_splits can have memory
638 			 * allocation error, we can not recover from it.
639 			 */
640 			if (zio->io_error != 0)
641 				break;
642 			rs->rs_offset += inner_size;
643 			rs->rs_asize -= inner_size;
644 			rs->rs_split_offset += inner_size;
645 		}
646 
647 		free(mapping);
648 		free(rs);
649 		if (zio->io_error != 0)
650 			break;
651 	}
652 
653 	list_destroy(&stack);
654 }
655 
656 static void
657 vdev_indirect_map_free(zio_t *zio)
658 {
659 	indirect_vsd_t *iv = zio->io_vsd;
660 	indirect_split_t *is;
661 
662 	while ((is = list_head(&iv->iv_splits)) != NULL) {
663 		for (int c = 0; c < is->is_children; c++) {
664 			indirect_child_t *ic = &is->is_child[c];
665 			free(ic->ic_data);
666 		}
667 		list_remove(&iv->iv_splits, is);
668 		free(is);
669 	}
670 	free(iv);
671 }
672 
673 static int
674 vdev_indirect_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
675     off_t offset, size_t bytes)
676 {
677 	zio_t zio;
678 	spa_t *spa = vdev->v_spa;
679 	indirect_vsd_t *iv;
680 	indirect_split_t *first;
681 	int rc = EIO;
682 
683 	iv = calloc(1, sizeof(*iv));
684 	if (iv == NULL)
685 		return (ENOMEM);
686 
687 	list_create(&iv->iv_splits,
688 	    sizeof (indirect_split_t), offsetof(indirect_split_t, is_node));
689 
690 	bzero(&zio, sizeof(zio));
691 	zio.io_spa = spa;
692 	zio.io_bp = (blkptr_t *)bp;
693 	zio.io_data = buf;
694 	zio.io_size = bytes;
695 	zio.io_offset = offset;
696 	zio.io_vd = vdev;
697 	zio.io_vsd = iv;
698 
699 	if (vdev->v_mapping == NULL) {
700 		vdev_indirect_config_t *vic;
701 
702 		vic = &vdev->vdev_indirect_config;
703 		vdev->v_mapping = vdev_indirect_mapping_open(spa,
704 		    spa->spa_mos, vic->vic_mapping_object);
705 	}
706 
707 	vdev_indirect_remap(vdev, offset, bytes, &zio);
708 	if (zio.io_error != 0)
709 		return (zio.io_error);
710 
711 	first = list_head(&iv->iv_splits);
712 	if (first->is_size == zio.io_size) {
713 		/*
714 		 * This is not a split block; we are pointing to the entire
715 		 * data, which will checksum the same as the original data.
716 		 * Pass the BP down so that the child i/o can verify the
717 		 * checksum, and try a different location if available
718 		 * (e.g. on a mirror).
719 		 *
720 		 * While this special case could be handled the same as the
721 		 * general (split block) case, doing it this way ensures
722 		 * that the vast majority of blocks on indirect vdevs
723 		 * (which are not split) are handled identically to blocks
724 		 * on non-indirect vdevs.  This allows us to be less strict
725 		 * about performance in the general (but rare) case.
726 		 */
727 		rc = first->is_vdev->v_read(first->is_vdev, zio.io_bp,
728 		    zio.io_data, first->is_target_offset, bytes);
729 	} else {
730 		iv->iv_split_block = B_TRUE;
731 		/*
732 		 * Read one copy of each split segment, from the
733 		 * top-level vdev.  Since we don't know the
734 		 * checksum of each split individually, the child
735 		 * zio can't ensure that we get the right data.
736 		 * E.g. if it's a mirror, it will just read from a
737 		 * random (healthy) leaf vdev.  We have to verify
738 		 * the checksum in vdev_indirect_io_done().
739 		 */
740 		for (indirect_split_t *is = list_head(&iv->iv_splits);
741 		    is != NULL; is = list_next(&iv->iv_splits, is)) {
742 			char *ptr = zio.io_data;
743 
744 			rc = is->is_vdev->v_read(is->is_vdev, zio.io_bp,
745 			    ptr + is->is_split_offset, is->is_target_offset,
746 			    is->is_size);
747 		}
748 		if (zio_checksum_verify(spa, zio.io_bp, zio.io_data))
749 			rc = ECKSUM;
750 		else
751 			rc = 0;
752 	}
753 
754 	vdev_indirect_map_free(&zio);
755 	if (rc == 0)
756 		rc = zio.io_error;
757 
758 	return (rc);
759 }
760 
761 static int
762 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
763     off_t offset, size_t bytes)
764 {
765 
766 	return (vdev_read_phys(vdev, bp, buf,
767 	    offset + VDEV_LABEL_START_SIZE, bytes));
768 }
769 
770 static int
771 vdev_missing_read(vdev_t *vdev __unused, const blkptr_t *bp __unused,
772     void *buf __unused, off_t offset __unused, size_t bytes __unused)
773 {
774 
775 	return (ENOTSUP);
776 }
777 
778 static int
779 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
780     off_t offset, size_t bytes)
781 {
782 	vdev_t *kid;
783 	int rc;
784 
785 	rc = EIO;
786 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
787 		if (kid->v_state != VDEV_STATE_HEALTHY)
788 			continue;
789 		rc = kid->v_read(kid, bp, buf, offset, bytes);
790 		if (!rc)
791 			return (0);
792 	}
793 
794 	return (rc);
795 }
796 
797 static int
798 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
799     off_t offset, size_t bytes)
800 {
801 	vdev_t *kid;
802 
803 	/*
804 	 * Here we should have two kids:
805 	 * First one which is the one we are replacing and we can trust
806 	 * only this one to have valid data, but it might not be present.
807 	 * Second one is that one we are replacing with. It is most likely
808 	 * healthy, but we can't trust it has needed data, so we won't use it.
809 	 */
810 	kid = STAILQ_FIRST(&vdev->v_children);
811 	if (kid == NULL)
812 		return (EIO);
813 	if (kid->v_state != VDEV_STATE_HEALTHY)
814 		return (EIO);
815 	return (kid->v_read(kid, bp, buf, offset, bytes));
816 }
817 
818 static vdev_t *
819 vdev_find(uint64_t guid)
820 {
821 	vdev_t *vdev;
822 
823 	STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
824 		if (vdev->v_guid == guid)
825 			return (vdev);
826 
827 	return (0);
828 }
829 
830 static vdev_t *
831 vdev_create(uint64_t guid, vdev_read_t *_read)
832 {
833 	vdev_t *vdev;
834 	vdev_indirect_config_t *vic;
835 
836 	vdev = calloc(1, sizeof(vdev_t));
837 	if (vdev != NULL) {
838 		STAILQ_INIT(&vdev->v_children);
839 		vdev->v_guid = guid;
840 		vdev->v_read = _read;
841 
842 		/*
843 		 * root vdev has no read function, we use this fact to
844 		 * skip setting up data we do not need for root vdev.
845 		 * We only point root vdev from spa.
846 		 */
847 		if (_read != NULL) {
848 			vic = &vdev->vdev_indirect_config;
849 			vic->vic_prev_indirect_vdev = UINT64_MAX;
850 			STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
851 		}
852 	}
853 
854 	return (vdev);
855 }
856 
857 static void
858 vdev_set_initial_state(vdev_t *vdev, const nvlist_t *nvlist)
859 {
860 	uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
861 	uint64_t is_log;
862 
863 	is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
864 	is_log = 0;
865 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
866 	    &is_offline, NULL);
867 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
868 	    &is_removed, NULL);
869 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
870 	    &is_faulted, NULL);
871 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64,
872 	    NULL, &is_degraded, NULL);
873 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64,
874 	    NULL, &isnt_present, NULL);
875 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, NULL,
876 	    &is_log, NULL);
877 
878 	if (is_offline != 0)
879 		vdev->v_state = VDEV_STATE_OFFLINE;
880 	else if (is_removed != 0)
881 		vdev->v_state = VDEV_STATE_REMOVED;
882 	else if (is_faulted != 0)
883 		vdev->v_state = VDEV_STATE_FAULTED;
884 	else if (is_degraded != 0)
885 		vdev->v_state = VDEV_STATE_DEGRADED;
886 	else if (isnt_present != 0)
887 		vdev->v_state = VDEV_STATE_CANT_OPEN;
888 
889 	vdev->v_islog = is_log != 0;
890 }
891 
892 static int
893 vdev_init(uint64_t guid, const nvlist_t *nvlist, vdev_t **vdevp)
894 {
895 	uint64_t id, ashift, asize, nparity;
896 	const char *path;
897 	const char *type;
898 	int len, pathlen;
899 	char *name;
900 	vdev_t *vdev;
901 
902 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id,
903 	    NULL) ||
904 	    nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING, NULL,
905 	    &type, &len)) {
906 		return (ENOENT);
907 	}
908 
909 	if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 &&
910 	    memcmp(type, VDEV_TYPE_DISK, len) != 0 &&
911 #ifdef ZFS_TEST
912 	    memcmp(type, VDEV_TYPE_FILE, len) != 0 &&
913 #endif
914 	    memcmp(type, VDEV_TYPE_RAIDZ, len) != 0 &&
915 	    memcmp(type, VDEV_TYPE_INDIRECT, len) != 0 &&
916 	    memcmp(type, VDEV_TYPE_REPLACING, len) != 0 &&
917 	    memcmp(type, VDEV_TYPE_HOLE, len) != 0) {
918 		printf("ZFS: can only boot from disk, mirror, raidz1, "
919 		    "raidz2 and raidz3 vdevs, got: %.*s\n", len, type);
920 		return (EIO);
921 	}
922 
923 	if (memcmp(type, VDEV_TYPE_MIRROR, len) == 0)
924 		vdev = vdev_create(guid, vdev_mirror_read);
925 	else if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0)
926 		vdev = vdev_create(guid, vdev_raidz_read);
927 	else if (memcmp(type, VDEV_TYPE_REPLACING, len) == 0)
928 		vdev = vdev_create(guid, vdev_replacing_read);
929 	else if (memcmp(type, VDEV_TYPE_INDIRECT, len) == 0) {
930 		vdev_indirect_config_t *vic;
931 
932 		vdev = vdev_create(guid, vdev_indirect_read);
933 		if (vdev != NULL) {
934 			vdev->v_state = VDEV_STATE_HEALTHY;
935 			vic = &vdev->vdev_indirect_config;
936 
937 			nvlist_find(nvlist,
938 			    ZPOOL_CONFIG_INDIRECT_OBJECT,
939 			    DATA_TYPE_UINT64,
940 			    NULL, &vic->vic_mapping_object, NULL);
941 			nvlist_find(nvlist,
942 			    ZPOOL_CONFIG_INDIRECT_BIRTHS,
943 			    DATA_TYPE_UINT64,
944 			    NULL, &vic->vic_births_object, NULL);
945 			nvlist_find(nvlist,
946 			    ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
947 			    DATA_TYPE_UINT64,
948 			    NULL, &vic->vic_prev_indirect_vdev, NULL);
949 		}
950 	} else if (memcmp(type, VDEV_TYPE_HOLE, len) == 0) {
951 		vdev = vdev_create(guid, vdev_missing_read);
952 	} else {
953 		vdev = vdev_create(guid, vdev_disk_read);
954 	}
955 
956 	if (vdev == NULL)
957 		return (ENOMEM);
958 
959 	vdev_set_initial_state(vdev, nvlist);
960 	vdev->v_id = id;
961 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
962 	    DATA_TYPE_UINT64, NULL, &ashift, NULL) == 0)
963 		vdev->v_ashift = ashift;
964 
965 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE,
966 	    DATA_TYPE_UINT64, NULL, &asize, NULL) == 0) {
967 		vdev->v_psize = asize +
968 		    VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
969 	}
970 
971 	if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
972 	    DATA_TYPE_UINT64, NULL, &nparity, NULL) == 0)
973 		vdev->v_nparity = nparity;
974 
975 	if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
976 	    DATA_TYPE_STRING, NULL, &path, &pathlen) == 0) {
977 		char prefix[] = "/dev/";
978 
979 		len = strlen(prefix);
980 		if (len < pathlen && memcmp(path, prefix, len) == 0) {
981 			path += len;
982 			pathlen -= len;
983 		}
984 		name = malloc(pathlen + 1);
985 		bcopy(path, name, pathlen);
986 		name[pathlen] = '\0';
987 		vdev->v_name = name;
988 	} else {
989 		name = NULL;
990 		if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) {
991 			if (vdev->v_nparity < 1 ||
992 			    vdev->v_nparity > 3) {
993 				printf("ZFS: invalid raidz parity: %d\n",
994 				    vdev->v_nparity);
995 				return (EIO);
996 			}
997 			(void) asprintf(&name, "%.*s%d-%" PRIu64, len, type,
998 			    vdev->v_nparity, id);
999 		} else {
1000 			(void) asprintf(&name, "%.*s-%" PRIu64, len, type, id);
1001 		}
1002 		vdev->v_name = name;
1003 	}
1004 	*vdevp = vdev;
1005 	return (0);
1006 }
1007 
1008 /*
1009  * Find slot for vdev. We return either NULL to signal to use
1010  * STAILQ_INSERT_HEAD, or we return link element to be used with
1011  * STAILQ_INSERT_AFTER.
1012  */
1013 static vdev_t *
1014 vdev_find_previous(vdev_t *top_vdev, vdev_t *vdev)
1015 {
1016 	vdev_t *v, *previous;
1017 
1018 	if (STAILQ_EMPTY(&top_vdev->v_children))
1019 		return (NULL);
1020 
1021 	previous = NULL;
1022 	STAILQ_FOREACH(v, &top_vdev->v_children, v_childlink) {
1023 		if (v->v_id > vdev->v_id)
1024 			return (previous);
1025 
1026 		if (v->v_id == vdev->v_id)
1027 			return (v);
1028 
1029 		if (v->v_id < vdev->v_id)
1030 			previous = v;
1031 	}
1032 	return (previous);
1033 }
1034 
1035 static size_t
1036 vdev_child_count(vdev_t *vdev)
1037 {
1038 	vdev_t *v;
1039 	size_t count;
1040 
1041 	count = 0;
1042 	STAILQ_FOREACH(v, &vdev->v_children, v_childlink) {
1043 		count++;
1044 	}
1045 	return (count);
1046 }
1047 
1048 /*
1049  * Insert vdev into top_vdev children list. List is ordered by v_id.
1050  */
1051 static void
1052 vdev_insert(vdev_t *top_vdev, vdev_t *vdev)
1053 {
1054 	vdev_t *previous;
1055 	size_t count;
1056 
1057 	/*
1058 	 * The top level vdev can appear in random order, depending how
1059 	 * the firmware is presenting the disk devices.
1060 	 * However, we will insert vdev to create list ordered by v_id,
1061 	 * so we can use either STAILQ_INSERT_HEAD or STAILQ_INSERT_AFTER
1062 	 * as STAILQ does not have insert before.
1063 	 */
1064 	previous = vdev_find_previous(top_vdev, vdev);
1065 
1066 	if (previous == NULL) {
1067 		STAILQ_INSERT_HEAD(&top_vdev->v_children, vdev, v_childlink);
1068 	} else if (previous->v_id == vdev->v_id) {
1069 		/*
1070 		 * This vdev was configured from label config,
1071 		 * do not insert duplicate.
1072 		 */
1073 		return;
1074 	} else {
1075 		STAILQ_INSERT_AFTER(&top_vdev->v_children, previous, vdev,
1076 		    v_childlink);
1077 	}
1078 
1079 	count = vdev_child_count(top_vdev);
1080 	if (top_vdev->v_nchildren < count)
1081 		top_vdev->v_nchildren = count;
1082 }
1083 
1084 static int
1085 vdev_from_nvlist(spa_t *spa, uint64_t top_guid, const nvlist_t *nvlist)
1086 {
1087 	vdev_t *top_vdev, *vdev;
1088 	nvlist_t *kids = NULL;
1089 	int rc, nkids;
1090 
1091 	/* Get top vdev. */
1092 	top_vdev = vdev_find(top_guid);
1093 	if (top_vdev == NULL) {
1094 		rc = vdev_init(top_guid, nvlist, &top_vdev);
1095 		if (rc != 0)
1096 			return (rc);
1097 		top_vdev->v_spa = spa;
1098 		top_vdev->v_top = top_vdev;
1099 		vdev_insert(spa->spa_root_vdev, top_vdev);
1100 	}
1101 
1102 	/* Add children if there are any. */
1103 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1104 	    &nkids, &kids, NULL);
1105 	if (rc == 0) {
1106 		for (int i = 0; i < nkids; i++) {
1107 			uint64_t guid;
1108 
1109 			rc = nvlist_find(kids, ZPOOL_CONFIG_GUID,
1110 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1111 			if (rc != 0) {
1112 				nvlist_destroy(kids);
1113 				return (rc);
1114 			}
1115 			rc = vdev_init(guid, kids, &vdev);
1116 			if (rc != 0) {
1117 				nvlist_destroy(kids);
1118 				return (rc);
1119 			}
1120 
1121 			vdev->v_spa = spa;
1122 			vdev->v_top = top_vdev;
1123 			vdev_insert(top_vdev, vdev);
1124 
1125 			rc = nvlist_next(kids);
1126 			if (rc != 0) {
1127 				nvlist_destroy(kids);
1128 				return (rc);
1129 			}
1130 		}
1131 	} else {
1132 		/*
1133 		 * When there are no children, nvlist_find() does return
1134 		 * error, reset it because leaf devices have no children.
1135 		 */
1136 		rc = 0;
1137 	}
1138 	nvlist_destroy(kids);
1139 
1140 	return (rc);
1141 }
1142 
1143 static int
1144 vdev_init_from_label(spa_t *spa, const nvlist_t *nvlist)
1145 {
1146 	uint64_t pool_guid, top_guid;
1147 	nvlist_t *vdevs;
1148 	int rc;
1149 
1150 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1151 	    NULL, &pool_guid, NULL) ||
1152 	    nvlist_find(nvlist, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64,
1153 	    NULL, &top_guid, NULL) ||
1154 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1155 	    NULL, &vdevs, NULL)) {
1156 		printf("ZFS: can't find vdev details\n");
1157 		return (ENOENT);
1158 	}
1159 
1160 	rc = vdev_from_nvlist(spa, top_guid, vdevs);
1161 	nvlist_destroy(vdevs);
1162 	return (rc);
1163 }
1164 
1165 static void
1166 vdev_set_state(vdev_t *vdev)
1167 {
1168 	vdev_t *kid;
1169 	int good_kids;
1170 	int bad_kids;
1171 
1172 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1173 		vdev_set_state(kid);
1174 	}
1175 
1176 	/*
1177 	 * A mirror or raidz is healthy if all its kids are healthy. A
1178 	 * mirror is degraded if any of its kids is healthy; a raidz
1179 	 * is degraded if at most nparity kids are offline.
1180 	 */
1181 	if (STAILQ_FIRST(&vdev->v_children)) {
1182 		good_kids = 0;
1183 		bad_kids = 0;
1184 		STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1185 			if (kid->v_state == VDEV_STATE_HEALTHY)
1186 				good_kids++;
1187 			else
1188 				bad_kids++;
1189 		}
1190 		if (bad_kids == 0) {
1191 			vdev->v_state = VDEV_STATE_HEALTHY;
1192 		} else {
1193 			if (vdev->v_read == vdev_mirror_read) {
1194 				if (good_kids) {
1195 					vdev->v_state = VDEV_STATE_DEGRADED;
1196 				} else {
1197 					vdev->v_state = VDEV_STATE_OFFLINE;
1198 				}
1199 			} else if (vdev->v_read == vdev_raidz_read) {
1200 				if (bad_kids > vdev->v_nparity) {
1201 					vdev->v_state = VDEV_STATE_OFFLINE;
1202 				} else {
1203 					vdev->v_state = VDEV_STATE_DEGRADED;
1204 				}
1205 			}
1206 		}
1207 	}
1208 }
1209 
1210 static int
1211 vdev_update_from_nvlist(uint64_t top_guid, const nvlist_t *nvlist)
1212 {
1213 	vdev_t *vdev;
1214 	nvlist_t *kids = NULL;
1215 	int rc, nkids;
1216 
1217 	/* Update top vdev. */
1218 	vdev = vdev_find(top_guid);
1219 	if (vdev != NULL)
1220 		vdev_set_initial_state(vdev, nvlist);
1221 
1222 	/* Update children if there are any. */
1223 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1224 	    &nkids, &kids, NULL);
1225 	if (rc == 0) {
1226 		for (int i = 0; i < nkids; i++) {
1227 			uint64_t guid;
1228 
1229 			rc = nvlist_find(kids, ZPOOL_CONFIG_GUID,
1230 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1231 			if (rc != 0)
1232 				break;
1233 
1234 			vdev = vdev_find(guid);
1235 			if (vdev != NULL)
1236 				vdev_set_initial_state(vdev, kids);
1237 
1238 			rc = nvlist_next(kids);
1239 			if (rc != 0)
1240 				break;
1241 		}
1242 	} else {
1243 		rc = 0;
1244 	}
1245 	nvlist_destroy(kids);
1246 
1247 	return (rc);
1248 }
1249 
1250 static int
1251 vdev_init_from_nvlist(spa_t *spa, const nvlist_t *nvlist)
1252 {
1253 	uint64_t pool_guid, vdev_children;
1254 	nvlist_t *vdevs = NULL, *kids = NULL;
1255 	int rc, nkids;
1256 
1257 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1258 	    NULL, &pool_guid, NULL) ||
1259 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_CHILDREN, DATA_TYPE_UINT64,
1260 	    NULL, &vdev_children, NULL) ||
1261 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1262 	    NULL, &vdevs, NULL)) {
1263 		printf("ZFS: can't find vdev details\n");
1264 		return (ENOENT);
1265 	}
1266 
1267 	/* Wrong guid?! */
1268 	if (spa->spa_guid != pool_guid) {
1269 		nvlist_destroy(vdevs);
1270 		return (EINVAL);
1271 	}
1272 
1273 	spa->spa_root_vdev->v_nchildren = vdev_children;
1274 
1275 	rc = nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1276 	    &nkids, &kids, NULL);
1277 	nvlist_destroy(vdevs);
1278 
1279 	/*
1280 	 * MOS config has at least one child for root vdev.
1281 	 */
1282 	if (rc != 0)
1283 		return (rc);
1284 
1285 	for (int i = 0; i < nkids; i++) {
1286 		uint64_t guid;
1287 		vdev_t *vdev;
1288 
1289 		rc = nvlist_find(kids, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1290 		    NULL, &guid, NULL);
1291 		if (rc != 0)
1292 			break;
1293 		vdev = vdev_find(guid);
1294 		/*
1295 		 * Top level vdev is missing, create it.
1296 		 */
1297 		if (vdev == NULL)
1298 			rc = vdev_from_nvlist(spa, guid, kids);
1299 		else
1300 			rc = vdev_update_from_nvlist(guid, kids);
1301 		if (rc != 0)
1302 			break;
1303 		rc = nvlist_next(kids);
1304 		if (rc != 0)
1305 			break;
1306 	}
1307 	nvlist_destroy(kids);
1308 
1309 	/*
1310 	 * Re-evaluate top-level vdev state.
1311 	 */
1312 	vdev_set_state(spa->spa_root_vdev);
1313 
1314 	return (rc);
1315 }
1316 
1317 static spa_t *
1318 spa_find_by_guid(uint64_t guid)
1319 {
1320 	spa_t *spa;
1321 
1322 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1323 		if (spa->spa_guid == guid)
1324 			return (spa);
1325 
1326 	return (NULL);
1327 }
1328 
1329 static spa_t *
1330 spa_find_by_name(const char *name)
1331 {
1332 	spa_t *spa;
1333 
1334 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1335 		if (strcmp(spa->spa_name, name) == 0)
1336 			return (spa);
1337 
1338 	return (NULL);
1339 }
1340 
1341 static spa_t *
1342 spa_create(uint64_t guid, const char *name)
1343 {
1344 	spa_t *spa;
1345 
1346 	if ((spa = calloc(1, sizeof(spa_t))) == NULL)
1347 		return (NULL);
1348 	if ((spa->spa_name = strdup(name)) == NULL) {
1349 		free(spa);
1350 		return (NULL);
1351 	}
1352 	spa->spa_uberblock = &spa->spa_uberblock_master;
1353 	spa->spa_mos = &spa->spa_mos_master;
1354 	spa->spa_guid = guid;
1355 	spa->spa_root_vdev = vdev_create(guid, NULL);
1356 	if (spa->spa_root_vdev == NULL) {
1357 		free(spa->spa_name);
1358 		free(spa);
1359 		return (NULL);
1360 	}
1361 	spa->spa_root_vdev->v_name = strdup("root");
1362 	STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
1363 
1364 	return (spa);
1365 }
1366 
1367 static const char *
1368 state_name(vdev_state_t state)
1369 {
1370 	static const char *names[] = {
1371 		"UNKNOWN",
1372 		"CLOSED",
1373 		"OFFLINE",
1374 		"REMOVED",
1375 		"CANT_OPEN",
1376 		"FAULTED",
1377 		"DEGRADED",
1378 		"ONLINE"
1379 	};
1380 	return (names[state]);
1381 }
1382 
1383 #ifdef BOOT2
1384 
1385 #define pager_printf printf
1386 
1387 #else
1388 
1389 static int
1390 pager_printf(const char *fmt, ...)
1391 {
1392 	char line[80];
1393 	va_list args;
1394 
1395 	va_start(args, fmt);
1396 	vsnprintf(line, sizeof(line), fmt, args);
1397 	va_end(args);
1398 	return (pager_output(line));
1399 }
1400 
1401 #endif
1402 
1403 #define	STATUS_FORMAT	"        %s %s\n"
1404 
1405 static int
1406 print_state(int indent, const char *name, vdev_state_t state)
1407 {
1408 	int i;
1409 	char buf[512];
1410 
1411 	buf[0] = 0;
1412 	for (i = 0; i < indent; i++)
1413 		strcat(buf, "  ");
1414 	strcat(buf, name);
1415 	return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
1416 }
1417 
1418 static int
1419 vdev_status(vdev_t *vdev, int indent)
1420 {
1421 	vdev_t *kid;
1422 	int ret;
1423 
1424 	if (vdev->v_islog) {
1425 		(void) pager_output("        logs\n");
1426 		indent++;
1427 	}
1428 
1429 	ret = print_state(indent, vdev->v_name, vdev->v_state);
1430 	if (ret != 0)
1431 		return (ret);
1432 
1433 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1434 		ret = vdev_status(kid, indent + 1);
1435 		if (ret != 0)
1436 			return (ret);
1437 	}
1438 	return (ret);
1439 }
1440 
1441 static int
1442 spa_status(spa_t *spa)
1443 {
1444 	static char bootfs[ZFS_MAXNAMELEN];
1445 	uint64_t rootid;
1446 	vdev_list_t *vlist;
1447 	vdev_t *vdev;
1448 	int good_kids, bad_kids, degraded_kids, ret;
1449 	vdev_state_t state;
1450 
1451 	ret = pager_printf("  pool: %s\n", spa->spa_name);
1452 	if (ret != 0)
1453 		return (ret);
1454 
1455 	if (zfs_get_root(spa, &rootid) == 0 &&
1456 	    zfs_rlookup(spa, rootid, bootfs) == 0) {
1457 		if (bootfs[0] == '\0')
1458 			ret = pager_printf("bootfs: %s\n", spa->spa_name);
1459 		else
1460 			ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
1461 			    bootfs);
1462 		if (ret != 0)
1463 			return (ret);
1464 	}
1465 	ret = pager_printf("config:\n\n");
1466 	if (ret != 0)
1467 		return (ret);
1468 	ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
1469 	if (ret != 0)
1470 		return (ret);
1471 
1472 	good_kids = 0;
1473 	degraded_kids = 0;
1474 	bad_kids = 0;
1475 	vlist = &spa->spa_root_vdev->v_children;
1476 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1477 		if (vdev->v_state == VDEV_STATE_HEALTHY)
1478 			good_kids++;
1479 		else if (vdev->v_state == VDEV_STATE_DEGRADED)
1480 			degraded_kids++;
1481 		else
1482 			bad_kids++;
1483 	}
1484 
1485 	state = VDEV_STATE_CLOSED;
1486 	if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
1487 		state = VDEV_STATE_HEALTHY;
1488 	else if ((good_kids + degraded_kids) > 0)
1489 		state = VDEV_STATE_DEGRADED;
1490 
1491 	ret = print_state(0, spa->spa_name, state);
1492 	if (ret != 0)
1493 		return (ret);
1494 
1495 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1496 		ret = vdev_status(vdev, 1);
1497 		if (ret != 0)
1498 			return (ret);
1499 	}
1500 	return (ret);
1501 }
1502 
1503 static int
1504 spa_all_status(void)
1505 {
1506 	spa_t *spa;
1507 	int first = 1, ret = 0;
1508 
1509 	STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
1510 		if (!first) {
1511 			ret = pager_printf("\n");
1512 			if (ret != 0)
1513 				return (ret);
1514 		}
1515 		first = 0;
1516 		ret = spa_status(spa);
1517 		if (ret != 0)
1518 			return (ret);
1519 	}
1520 	return (ret);
1521 }
1522 
1523 static uint64_t
1524 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
1525 {
1526 	uint64_t label_offset;
1527 
1528 	if (l < VDEV_LABELS / 2)
1529 		label_offset = 0;
1530 	else
1531 		label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
1532 
1533 	return (offset + l * sizeof (vdev_label_t) + label_offset);
1534 }
1535 
1536 static int
1537 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1538 {
1539 	unsigned int seq1 = 0;
1540 	unsigned int seq2 = 0;
1541 	int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
1542 
1543 	if (cmp != 0)
1544 		return (cmp);
1545 
1546 	cmp = AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1547 	if (cmp != 0)
1548 		return (cmp);
1549 
1550 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1551 		seq1 = MMP_SEQ(ub1);
1552 
1553 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1554 		seq2 = MMP_SEQ(ub2);
1555 
1556 	return (AVL_CMP(seq1, seq2));
1557 }
1558 
1559 static int
1560 uberblock_verify(uberblock_t *ub)
1561 {
1562 	if (ub->ub_magic == BSWAP_64((uint64_t)UBERBLOCK_MAGIC)) {
1563 		byteswap_uint64_array(ub, sizeof (uberblock_t));
1564 	}
1565 
1566 	if (ub->ub_magic != UBERBLOCK_MAGIC ||
1567 	    !SPA_VERSION_IS_SUPPORTED(ub->ub_version))
1568 		return (EINVAL);
1569 
1570 	return (0);
1571 }
1572 
1573 static int
1574 vdev_label_read(vdev_t *vd, int l, void *buf, uint64_t offset,
1575     size_t size)
1576 {
1577 	blkptr_t bp;
1578 	off_t off;
1579 
1580 	off = vdev_label_offset(vd->v_psize, l, offset);
1581 
1582 	BP_ZERO(&bp);
1583 	BP_SET_LSIZE(&bp, size);
1584 	BP_SET_PSIZE(&bp, size);
1585 	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1586 	BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1587 	DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
1588 	ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1589 
1590 	return (vdev_read_phys(vd, &bp, buf, off, size));
1591 }
1592 
1593 static uint64_t
1594 vdev_get_label_asize(nvlist_t *nvl)
1595 {
1596 	nvlist_t *vdevs;
1597 	uint64_t asize;
1598 	const char *type;
1599 	int len;
1600 
1601 	asize = 0;
1602 	/* Get vdev tree */
1603 	if (nvlist_find(nvl, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1604 	    NULL, &vdevs, NULL) != 0)
1605 		return (asize);
1606 
1607 	/*
1608 	 * Get vdev type. We will calculate asize for raidz, mirror and disk.
1609 	 * For raidz, the asize is raw size of all children.
1610 	 */
1611 	if (nvlist_find(vdevs, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
1612 	    NULL, &type, &len) != 0)
1613 		goto done;
1614 
1615 	if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 &&
1616 	    memcmp(type, VDEV_TYPE_DISK, len) != 0 &&
1617 	    memcmp(type, VDEV_TYPE_RAIDZ, len) != 0)
1618 		goto done;
1619 
1620 	if (nvlist_find(vdevs, ZPOOL_CONFIG_ASIZE, DATA_TYPE_UINT64,
1621 	    NULL, &asize, NULL) != 0)
1622 		goto done;
1623 
1624 	if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) {
1625 		nvlist_t *kids;
1626 		int nkids;
1627 
1628 		if (nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN,
1629 		    DATA_TYPE_NVLIST_ARRAY, &nkids, &kids, NULL) != 0) {
1630 			asize = 0;
1631 			goto done;
1632 		}
1633 
1634 		asize /= nkids;
1635 		nvlist_destroy(kids);
1636 	}
1637 
1638 	asize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
1639 done:
1640 	nvlist_destroy(vdevs);
1641 	return (asize);
1642 }
1643 
1644 static nvlist_t *
1645 vdev_label_read_config(vdev_t *vd, uint64_t txg)
1646 {
1647 	vdev_phys_t *label;
1648 	uint64_t best_txg = 0;
1649 	uint64_t label_txg = 0;
1650 	uint64_t asize;
1651 	nvlist_t *nvl = NULL, *tmp;
1652 	int error;
1653 
1654 	label = malloc(sizeof (vdev_phys_t));
1655 	if (label == NULL)
1656 		return (NULL);
1657 
1658 	for (int l = 0; l < VDEV_LABELS; l++) {
1659 		const unsigned char *nvlist;
1660 
1661 		if (vdev_label_read(vd, l, label,
1662 		    offsetof(vdev_label_t, vl_vdev_phys),
1663 		    sizeof (vdev_phys_t)))
1664 			continue;
1665 
1666 		nvlist = (const unsigned char *) label->vp_nvlist;
1667 		tmp = nvlist_import(nvlist + 4, nvlist[0], nvlist[1]);
1668 		if (tmp == NULL)
1669 			continue;
1670 
1671 		error = nvlist_find(tmp, ZPOOL_CONFIG_POOL_TXG,
1672 		    DATA_TYPE_UINT64, NULL, &label_txg, NULL);
1673 		if (error != 0 || label_txg == 0) {
1674 			nvlist_destroy(nvl);
1675 			nvl = tmp;
1676 			goto done;
1677 		}
1678 
1679 		if (label_txg <= txg && label_txg > best_txg) {
1680 			best_txg = label_txg;
1681 			nvlist_destroy(nvl);
1682 			nvl = tmp;
1683 			tmp = NULL;
1684 
1685 			/*
1686 			 * Use asize from pool config. We need this
1687 			 * because we can get bad value from BIOS.
1688 			 */
1689 			asize = vdev_get_label_asize(nvl);
1690 			if (asize != 0) {
1691 				vd->v_psize = asize;
1692 			}
1693 		}
1694 		nvlist_destroy(tmp);
1695 	}
1696 
1697 	if (best_txg == 0) {
1698 		nvlist_destroy(nvl);
1699 		nvl = NULL;
1700 	}
1701 done:
1702 	free(label);
1703 	return (nvl);
1704 }
1705 
1706 static void
1707 vdev_uberblock_load(vdev_t *vd, uberblock_t *ub)
1708 {
1709 	uberblock_t *buf;
1710 
1711 	buf = malloc(VDEV_UBERBLOCK_SIZE(vd));
1712 	if (buf == NULL)
1713 		return;
1714 
1715 	for (int l = 0; l < VDEV_LABELS; l++) {
1716 		for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1717 			if (vdev_label_read(vd, l, buf,
1718 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1719 			    VDEV_UBERBLOCK_SIZE(vd)))
1720 				continue;
1721 			if (uberblock_verify(buf) != 0)
1722 				continue;
1723 
1724 			if (vdev_uberblock_compare(buf, ub) > 0)
1725 				*ub = *buf;
1726 		}
1727 	}
1728 	free(buf);
1729 }
1730 
1731 static int
1732 vdev_probe(vdev_phys_read_t *_read, void *read_priv, spa_t **spap)
1733 {
1734 	vdev_t vtmp;
1735 	spa_t *spa;
1736 	vdev_t *vdev;
1737 	nvlist_t *nvl;
1738 	uint64_t val;
1739 	uint64_t guid, vdev_children;
1740 	uint64_t pool_txg, pool_guid;
1741 	const char *pool_name;
1742 	int rc, namelen;
1743 
1744 	/*
1745 	 * Load the vdev label and figure out which
1746 	 * uberblock is most current.
1747 	 */
1748 	memset(&vtmp, 0, sizeof(vtmp));
1749 	vtmp.v_phys_read = _read;
1750 	vtmp.v_read_priv = read_priv;
1751 	vtmp.v_psize = P2ALIGN(ldi_get_size(read_priv),
1752 	    (uint64_t)sizeof (vdev_label_t));
1753 
1754 	/* Test for minimum device size. */
1755 	if (vtmp.v_psize < SPA_MINDEVSIZE)
1756 		return (EIO);
1757 
1758 	nvl = vdev_label_read_config(&vtmp, UINT64_MAX);
1759 	if (nvl == NULL)
1760 		return (EIO);
1761 
1762 	if (nvlist_find(nvl, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
1763 	    NULL, &val, NULL) != 0) {
1764 		nvlist_destroy(nvl);
1765 		return (EIO);
1766 	}
1767 
1768 	if (!SPA_VERSION_IS_SUPPORTED(val)) {
1769 		printf("ZFS: unsupported ZFS version %u (should be %u)\n",
1770 		    (unsigned)val, (unsigned)SPA_VERSION);
1771 		nvlist_destroy(nvl);
1772 		return (EIO);
1773 	}
1774 
1775 	/* Check ZFS features for read */
1776 	rc = nvlist_check_features_for_read(nvl);
1777 	if (rc != 0) {
1778 		nvlist_destroy(nvl);
1779 		return (EIO);
1780 	}
1781 
1782 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
1783 	    NULL, &val, NULL) != 0) {
1784 		nvlist_destroy(nvl);
1785 		return (EIO);
1786 	}
1787 
1788 	if (val == POOL_STATE_DESTROYED) {
1789 		/* We don't boot only from destroyed pools. */
1790 		nvlist_destroy(nvl);
1791 		return (EIO);
1792 	}
1793 
1794 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
1795 	    NULL, &pool_txg, NULL) != 0 ||
1796 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1797 	    NULL, &pool_guid, NULL) != 0 ||
1798 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
1799 	    NULL, &pool_name, &namelen) != 0) {
1800 		/*
1801 		 * Cache and spare devices end up here - just ignore
1802 		 * them.
1803 		 */
1804 		nvlist_destroy(nvl);
1805 		return (EIO);
1806 	}
1807 
1808 	/*
1809 	 * Create the pool if this is the first time we've seen it.
1810 	 */
1811 	spa = spa_find_by_guid(pool_guid);
1812 	if (spa == NULL) {
1813 		char *name;
1814 
1815 		nvlist_find(nvl, ZPOOL_CONFIG_VDEV_CHILDREN,
1816 		    DATA_TYPE_UINT64, NULL, &vdev_children, NULL);
1817 		name = malloc(namelen + 1);
1818 		if (name == NULL) {
1819 			nvlist_destroy(nvl);
1820 			return (ENOMEM);
1821 		}
1822 		bcopy(pool_name, name, namelen);
1823 		name[namelen] = '\0';
1824 		spa = spa_create(pool_guid, name);
1825 		free(name);
1826 		if (spa == NULL) {
1827 			nvlist_destroy(nvl);
1828 			return (ENOMEM);
1829 		}
1830 		spa->spa_root_vdev->v_nchildren = vdev_children;
1831 	}
1832 	if (pool_txg > spa->spa_txg)
1833 		spa->spa_txg = pool_txg;
1834 
1835 	/*
1836 	 * Get the vdev tree and create our in-core copy of it.
1837 	 * If we already have a vdev with this guid, this must
1838 	 * be some kind of alias (overlapping slices, dangerously dedicated
1839 	 * disks etc).
1840 	 */
1841 	if (nvlist_find(nvl, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1842 	    NULL, &guid, NULL) != 0) {
1843 		nvlist_destroy(nvl);
1844 		return (EIO);
1845 	}
1846 	vdev = vdev_find(guid);
1847 	/* Has this vdev already been inited? */
1848 	if (vdev && vdev->v_phys_read) {
1849 		nvlist_destroy(nvl);
1850 		return (EIO);
1851 	}
1852 
1853 	rc = vdev_init_from_label(spa, nvl);
1854 	nvlist_destroy(nvl);
1855 	if (rc != 0)
1856 		return (rc);
1857 
1858 	/*
1859 	 * We should already have created an incomplete vdev for this
1860 	 * vdev. Find it and initialise it with our read proc.
1861 	 */
1862 	vdev = vdev_find(guid);
1863 	if (vdev != NULL) {
1864 		vdev->v_phys_read = _read;
1865 		vdev->v_read_priv = read_priv;
1866 		vdev->v_psize = vtmp.v_psize;
1867 		/*
1868 		 * If no other state is set, mark vdev healthy.
1869 		 */
1870 		if (vdev->v_state == VDEV_STATE_UNKNOWN)
1871 			vdev->v_state = VDEV_STATE_HEALTHY;
1872 	} else {
1873 		printf("ZFS: inconsistent nvlist contents\n");
1874 		return (EIO);
1875 	}
1876 
1877 	if (vdev->v_islog)
1878 		spa->spa_with_log = vdev->v_islog;
1879 
1880 	/*
1881 	 * Re-evaluate top-level vdev state.
1882 	 */
1883 	vdev_set_state(vdev->v_top);
1884 
1885 	/*
1886 	 * Ok, we are happy with the pool so far. Lets find
1887 	 * the best uberblock and then we can actually access
1888 	 * the contents of the pool.
1889 	 */
1890 	vdev_uberblock_load(vdev, spa->spa_uberblock);
1891 
1892 	if (spap != NULL)
1893 		*spap = spa;
1894 	return (0);
1895 }
1896 
1897 static int
1898 ilog2(int n)
1899 {
1900 	int v;
1901 
1902 	for (v = 0; v < 32; v++)
1903 		if (n == (1 << v))
1904 			return (v);
1905 	return (-1);
1906 }
1907 
1908 static int
1909 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
1910 {
1911 	blkptr_t gbh_bp;
1912 	zio_gbh_phys_t zio_gb;
1913 	char *pbuf;
1914 	int i;
1915 
1916 	/* Artificial BP for gang block header. */
1917 	gbh_bp = *bp;
1918 	BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1919 	BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1920 	BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
1921 	BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
1922 	for (i = 0; i < SPA_DVAS_PER_BP; i++)
1923 		DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
1924 
1925 	/* Read gang header block using the artificial BP. */
1926 	if (zio_read(spa, &gbh_bp, &zio_gb))
1927 		return (EIO);
1928 
1929 	pbuf = buf;
1930 	for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1931 		blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1932 
1933 		if (BP_IS_HOLE(gbp))
1934 			continue;
1935 		if (zio_read(spa, gbp, pbuf))
1936 			return (EIO);
1937 		pbuf += BP_GET_PSIZE(gbp);
1938 	}
1939 
1940 	if (zio_checksum_verify(spa, bp, buf))
1941 		return (EIO);
1942 	return (0);
1943 }
1944 
1945 static int
1946 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1947 {
1948 	int cpfunc = BP_GET_COMPRESS(bp);
1949 	uint64_t align, size;
1950 	void *pbuf;
1951 	int i, error;
1952 
1953 	/*
1954 	 * Process data embedded in block pointer
1955 	 */
1956 	if (BP_IS_EMBEDDED(bp)) {
1957 		ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
1958 
1959 		size = BPE_GET_PSIZE(bp);
1960 		ASSERT(size <= BPE_PAYLOAD_SIZE);
1961 
1962 		if (cpfunc != ZIO_COMPRESS_OFF)
1963 			pbuf = malloc(size);
1964 		else
1965 			pbuf = buf;
1966 
1967 		if (pbuf == NULL)
1968 			return (ENOMEM);
1969 
1970 		decode_embedded_bp_compressed(bp, pbuf);
1971 		error = 0;
1972 
1973 		if (cpfunc != ZIO_COMPRESS_OFF) {
1974 			error = zio_decompress_data(cpfunc, pbuf,
1975 			    size, buf, BP_GET_LSIZE(bp));
1976 			free(pbuf);
1977 		}
1978 		if (error != 0)
1979 			printf("ZFS: i/o error - unable to decompress "
1980 			    "block pointer data, error %d\n", error);
1981 		return (error);
1982 	}
1983 
1984 	error = EIO;
1985 
1986 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1987 		const dva_t *dva = &bp->blk_dva[i];
1988 		vdev_t *vdev;
1989 		vdev_list_t *vlist;
1990 		uint64_t vdevid;
1991 		off_t offset;
1992 
1993 		if (!dva->dva_word[0] && !dva->dva_word[1])
1994 			continue;
1995 
1996 		vdevid = DVA_GET_VDEV(dva);
1997 		offset = DVA_GET_OFFSET(dva);
1998 		vlist = &spa->spa_root_vdev->v_children;
1999 		STAILQ_FOREACH(vdev, vlist, v_childlink) {
2000 			if (vdev->v_id == vdevid)
2001 				break;
2002 		}
2003 		if (!vdev || !vdev->v_read)
2004 			continue;
2005 
2006 		size = BP_GET_PSIZE(bp);
2007 		if (vdev->v_read == vdev_raidz_read) {
2008 			align = 1ULL << vdev->v_ashift;
2009 			if (P2PHASE(size, align) != 0)
2010 				size = P2ROUNDUP(size, align);
2011 		}
2012 		if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
2013 			pbuf = malloc(size);
2014 		else
2015 			pbuf = buf;
2016 
2017 		if (pbuf == NULL) {
2018 			error = ENOMEM;
2019 			break;
2020 		}
2021 
2022 		if (DVA_GET_GANG(dva))
2023 			error = zio_read_gang(spa, bp, pbuf);
2024 		else
2025 			error = vdev->v_read(vdev, bp, pbuf, offset, size);
2026 		if (error == 0) {
2027 			if (cpfunc != ZIO_COMPRESS_OFF)
2028 				error = zio_decompress_data(cpfunc, pbuf,
2029 				    BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
2030 			else if (size != BP_GET_PSIZE(bp))
2031 				bcopy(pbuf, buf, BP_GET_PSIZE(bp));
2032 		} else {
2033 			printf("zio_read error: %d\n", error);
2034 		}
2035 		if (buf != pbuf)
2036 			free(pbuf);
2037 		if (error == 0)
2038 			break;
2039 	}
2040 	if (error != 0)
2041 		printf("ZFS: i/o error - all block copies unavailable\n");
2042 
2043 	return (error);
2044 }
2045 
2046 static int
2047 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset,
2048     void *buf, size_t buflen)
2049 {
2050 	int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
2051 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2052 	int nlevels = dnode->dn_nlevels;
2053 	int i, rc;
2054 
2055 	if (bsize > SPA_MAXBLOCKSIZE) {
2056 		printf("ZFS: I/O error - blocks larger than %llu are not "
2057 		    "supported\n", SPA_MAXBLOCKSIZE);
2058 		return (EIO);
2059 	}
2060 
2061 	/*
2062 	 * Note: bsize may not be a power of two here so we need to do an
2063 	 * actual divide rather than a bitshift.
2064 	 */
2065 	while (buflen > 0) {
2066 		uint64_t bn = offset / bsize;
2067 		int boff = offset % bsize;
2068 		int ibn;
2069 		const blkptr_t *indbp;
2070 		blkptr_t bp;
2071 
2072 		if (bn > dnode->dn_maxblkid)
2073 			return (EIO);
2074 
2075 		if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
2076 			goto cached;
2077 
2078 		indbp = dnode->dn_blkptr;
2079 		for (i = 0; i < nlevels; i++) {
2080 			/*
2081 			 * Copy the bp from the indirect array so that
2082 			 * we can re-use the scratch buffer for multi-level
2083 			 * objects.
2084 			 */
2085 			ibn = bn >> ((nlevels - i - 1) * ibshift);
2086 			ibn &= ((1 << ibshift) - 1);
2087 			bp = indbp[ibn];
2088 			if (BP_IS_HOLE(&bp)) {
2089 				memset(dnode_cache_buf, 0, bsize);
2090 				break;
2091 			}
2092 			rc = zio_read(spa, &bp, dnode_cache_buf);
2093 			if (rc)
2094 				return (rc);
2095 			indbp = (const blkptr_t *) dnode_cache_buf;
2096 		}
2097 		dnode_cache_obj = dnode;
2098 		dnode_cache_bn = bn;
2099 	cached:
2100 
2101 		/*
2102 		 * The buffer contains our data block. Copy what we
2103 		 * need from it and loop.
2104 		 */
2105 		i = bsize - boff;
2106 		if (i > buflen) i = buflen;
2107 		memcpy(buf, &dnode_cache_buf[boff], i);
2108 		buf = ((char *)buf) + i;
2109 		offset += i;
2110 		buflen -= i;
2111 	}
2112 
2113 	return (0);
2114 }
2115 
2116 /*
2117  * Lookup a value in a microzap directory.
2118  */
2119 static int
2120 mzap_lookup(const mzap_phys_t *mz, size_t size, const char *name,
2121     uint64_t *value)
2122 {
2123 	const mzap_ent_phys_t *mze;
2124 	int chunks, i;
2125 
2126 	/*
2127 	 * Microzap objects use exactly one block. Read the whole
2128 	 * thing.
2129 	 */
2130 	chunks = size / MZAP_ENT_LEN - 1;
2131 	for (i = 0; i < chunks; i++) {
2132 		mze = &mz->mz_chunk[i];
2133 		if (strcmp(mze->mze_name, name) == 0) {
2134 			*value = mze->mze_value;
2135 			return (0);
2136 		}
2137 	}
2138 
2139 	return (ENOENT);
2140 }
2141 
2142 /*
2143  * Compare a name with a zap leaf entry. Return non-zero if the name
2144  * matches.
2145  */
2146 static int
2147 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2148     const char *name)
2149 {
2150 	size_t namelen;
2151 	const zap_leaf_chunk_t *nc;
2152 	const char *p;
2153 
2154 	namelen = zc->l_entry.le_name_numints;
2155 
2156 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2157 	p = name;
2158 	while (namelen > 0) {
2159 		size_t len;
2160 
2161 		len = namelen;
2162 		if (len > ZAP_LEAF_ARRAY_BYTES)
2163 			len = ZAP_LEAF_ARRAY_BYTES;
2164 		if (memcmp(p, nc->l_array.la_array, len))
2165 			return (0);
2166 		p += len;
2167 		namelen -= len;
2168 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2169 	}
2170 
2171 	return (1);
2172 }
2173 
2174 /*
2175  * Extract a uint64_t value from a zap leaf entry.
2176  */
2177 static uint64_t
2178 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
2179 {
2180 	const zap_leaf_chunk_t *vc;
2181 	int i;
2182 	uint64_t value;
2183 	const uint8_t *p;
2184 
2185 	vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
2186 	for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
2187 		value = (value << 8) | p[i];
2188 	}
2189 
2190 	return (value);
2191 }
2192 
2193 static void
2194 stv(int len, void *addr, uint64_t value)
2195 {
2196 	switch (len) {
2197 	case 1:
2198 		*(uint8_t *)addr = value;
2199 		return;
2200 	case 2:
2201 		*(uint16_t *)addr = value;
2202 		return;
2203 	case 4:
2204 		*(uint32_t *)addr = value;
2205 		return;
2206 	case 8:
2207 		*(uint64_t *)addr = value;
2208 		return;
2209 	}
2210 }
2211 
2212 /*
2213  * Extract a array from a zap leaf entry.
2214  */
2215 static void
2216 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2217     uint64_t integer_size, uint64_t num_integers, void *buf)
2218 {
2219 	uint64_t array_int_len = zc->l_entry.le_value_intlen;
2220 	uint64_t value = 0;
2221 	uint64_t *u64 = buf;
2222 	char *p = buf;
2223 	int len = MIN(zc->l_entry.le_value_numints, num_integers);
2224 	int chunk = zc->l_entry.le_value_chunk;
2225 	int byten = 0;
2226 
2227 	if (integer_size == 8 && len == 1) {
2228 		*u64 = fzap_leaf_value(zl, zc);
2229 		return;
2230 	}
2231 
2232 	while (len > 0) {
2233 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
2234 		int i;
2235 
2236 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
2237 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
2238 			value = (value << 8) | la->la_array[i];
2239 			byten++;
2240 			if (byten == array_int_len) {
2241 				stv(integer_size, p, value);
2242 				byten = 0;
2243 				len--;
2244 				if (len == 0)
2245 					return;
2246 				p += integer_size;
2247 			}
2248 		}
2249 		chunk = la->la_next;
2250 	}
2251 }
2252 
2253 static int
2254 fzap_check_size(uint64_t integer_size, uint64_t num_integers)
2255 {
2256 
2257 	switch (integer_size) {
2258 	case 1:
2259 	case 2:
2260 	case 4:
2261 	case 8:
2262 		break;
2263 	default:
2264 		return (EINVAL);
2265 	}
2266 
2267 	if (integer_size * num_integers > ZAP_MAXVALUELEN)
2268 		return (E2BIG);
2269 
2270 	return (0);
2271 }
2272 
2273 static void
2274 zap_leaf_free(zap_leaf_t *leaf)
2275 {
2276 	free(leaf->l_phys);
2277 	free(leaf);
2278 }
2279 
2280 static int
2281 zap_get_leaf_byblk(fat_zap_t *zap, uint64_t blk, zap_leaf_t **lp)
2282 {
2283 	int bs = FZAP_BLOCK_SHIFT(zap);
2284 	int err;
2285 
2286 	*lp = malloc(sizeof(**lp));
2287 	if (*lp == NULL)
2288 		return (ENOMEM);
2289 
2290 	(*lp)->l_bs = bs;
2291 	(*lp)->l_phys = malloc(1 << bs);
2292 
2293 	if ((*lp)->l_phys == NULL) {
2294 		free(*lp);
2295 		return (ENOMEM);
2296 	}
2297 	err = dnode_read(zap->zap_spa, zap->zap_dnode, blk << bs, (*lp)->l_phys,
2298 	    1 << bs);
2299 	if (err != 0) {
2300 		zap_leaf_free(*lp);
2301 	}
2302 	return (err);
2303 }
2304 
2305 static int
2306 zap_table_load(fat_zap_t *zap, zap_table_phys_t *tbl, uint64_t idx,
2307     uint64_t *valp)
2308 {
2309 	int bs = FZAP_BLOCK_SHIFT(zap);
2310 	uint64_t blk = idx >> (bs - 3);
2311 	uint64_t off = idx & ((1 << (bs - 3)) - 1);
2312 	uint64_t *buf;
2313 	int rc;
2314 
2315 	buf = malloc(1 << zap->zap_block_shift);
2316 	if (buf == NULL)
2317 		return (ENOMEM);
2318 	rc = dnode_read(zap->zap_spa, zap->zap_dnode, (tbl->zt_blk + blk) << bs,
2319 	    buf, 1 << zap->zap_block_shift);
2320 	if (rc == 0)
2321 		*valp = buf[off];
2322 	free(buf);
2323 	return (rc);
2324 }
2325 
2326 static int
2327 zap_idx_to_blk(fat_zap_t *zap, uint64_t idx, uint64_t *valp)
2328 {
2329 	if (zap->zap_phys->zap_ptrtbl.zt_numblks == 0) {
2330 		*valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
2331 		return (0);
2332 	} else {
2333 		return (zap_table_load(zap, &zap->zap_phys->zap_ptrtbl,
2334 		    idx, valp));
2335 	}
2336 }
2337 
2338 #define	ZAP_HASH_IDX(hash, n)	(((n) == 0) ? 0 : ((hash) >> (64 - (n))))
2339 static int
2340 zap_deref_leaf(fat_zap_t *zap, uint64_t h, zap_leaf_t **lp)
2341 {
2342 	uint64_t idx, blk;
2343 	int err;
2344 
2345 	idx = ZAP_HASH_IDX(h, zap->zap_phys->zap_ptrtbl.zt_shift);
2346 	err = zap_idx_to_blk(zap, idx, &blk);
2347 	if (err != 0)
2348 		return (err);
2349 	return (zap_get_leaf_byblk(zap, blk, lp));
2350 }
2351 
2352 #define	CHAIN_END	0xffff	/* end of the chunk chain */
2353 #define	LEAF_HASH(l, h) \
2354 	((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \
2355 	((h) >> \
2356 	(64 - ZAP_LEAF_HASH_SHIFT(l) - (l)->l_phys->l_hdr.lh_prefix_len)))
2357 #define	LEAF_HASH_ENTPTR(l, h)	(&(l)->l_phys->l_hash[LEAF_HASH(l, h)])
2358 
2359 static int
2360 zap_leaf_lookup(zap_leaf_t *zl, uint64_t hash, const char *name,
2361     uint64_t integer_size, uint64_t num_integers, void *value)
2362 {
2363 	int rc;
2364 	uint16_t *chunkp;
2365 	struct zap_leaf_entry *le;
2366 
2367 	/*
2368 	 * Make sure this chunk matches our hash.
2369 	 */
2370 	if (zl->l_phys->l_hdr.lh_prefix_len > 0 &&
2371 	    zl->l_phys->l_hdr.lh_prefix !=
2372 	    hash >> (64 - zl->l_phys->l_hdr.lh_prefix_len))
2373 		return (EIO);
2374 
2375 	rc = ENOENT;
2376 	for (chunkp = LEAF_HASH_ENTPTR(zl, hash);
2377 	    *chunkp != CHAIN_END; chunkp = &le->le_next) {
2378 		zap_leaf_chunk_t *zc;
2379 		uint16_t chunk = *chunkp;
2380 
2381 		le = ZAP_LEAF_ENTRY(zl, chunk);
2382 		if (le->le_hash != hash)
2383 			continue;
2384 		zc = &ZAP_LEAF_CHUNK(zl, chunk);
2385 		if (fzap_name_equal(zl, zc, name)) {
2386 			if (zc->l_entry.le_value_intlen > integer_size) {
2387 				rc = EINVAL;
2388 			} else {
2389 				fzap_leaf_array(zl, zc, integer_size,
2390 				    num_integers, value);
2391 				rc = 0;
2392 			}
2393 			break;
2394 		}
2395 	}
2396 	return (rc);
2397 }
2398 
2399 /*
2400  * Lookup a value in a fatzap directory.
2401  */
2402 static int
2403 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2404     const char *name, uint64_t integer_size, uint64_t num_integers,
2405     void *value)
2406 {
2407 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2408 	fat_zap_t z;
2409 	zap_leaf_t *zl;
2410 	uint64_t hash;
2411 	int rc;
2412 
2413 	if (zh->zap_magic != ZAP_MAGIC)
2414 		return (EIO);
2415 
2416 	if ((rc = fzap_check_size(integer_size, num_integers)) != 0) {
2417 		return (rc);
2418 	}
2419 
2420 	z.zap_block_shift = ilog2(bsize);
2421 	z.zap_phys = zh;
2422 	z.zap_spa = spa;
2423 	z.zap_dnode = dnode;
2424 
2425 	hash = zap_hash(zh->zap_salt, name);
2426 	rc = zap_deref_leaf(&z, hash, &zl);
2427 	if (rc != 0)
2428 		return (rc);
2429 
2430 	rc = zap_leaf_lookup(zl, hash, name, integer_size, num_integers, value);
2431 
2432 	zap_leaf_free(zl);
2433 	return (rc);
2434 }
2435 
2436 /*
2437  * Lookup a name in a zap object and return its value as a uint64_t.
2438  */
2439 static int
2440 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2441     uint64_t integer_size, uint64_t num_integers, void *value)
2442 {
2443 	int rc;
2444 	zap_phys_t *zap;
2445 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2446 
2447 	zap = malloc(size);
2448 	if (zap == NULL)
2449 		return (ENOMEM);
2450 
2451 	rc = dnode_read(spa, dnode, 0, zap, size);
2452 	if (rc)
2453 		goto done;
2454 
2455 	switch (zap->zap_block_type) {
2456 	case ZBT_MICRO:
2457 		rc = mzap_lookup((const mzap_phys_t *)zap, size, name, value);
2458 		break;
2459 	case ZBT_HEADER:
2460 		rc = fzap_lookup(spa, dnode, zap, name, integer_size,
2461 		    num_integers, value);
2462 		break;
2463 	default:
2464 		printf("ZFS: invalid zap_type=%" PRIx64 "\n",
2465 		    zap->zap_block_type);
2466 		rc = EIO;
2467 	}
2468 done:
2469 	free(zap);
2470 	return (rc);
2471 }
2472 
2473 /*
2474  * List a microzap directory.
2475  */
2476 static int
2477 mzap_list(const mzap_phys_t *mz, size_t size,
2478     int (*callback)(const char *, uint64_t))
2479 {
2480 	const mzap_ent_phys_t *mze;
2481 	int chunks, i, rc;
2482 
2483 	/*
2484 	 * Microzap objects use exactly one block. Read the whole
2485 	 * thing.
2486 	 */
2487 	rc = 0;
2488 	chunks = size / MZAP_ENT_LEN - 1;
2489 	for (i = 0; i < chunks; i++) {
2490 		mze = &mz->mz_chunk[i];
2491 		if (mze->mze_name[0]) {
2492 			rc = callback(mze->mze_name, mze->mze_value);
2493 			if (rc != 0)
2494 				break;
2495 		}
2496 	}
2497 
2498 	return (rc);
2499 }
2500 
2501 /*
2502  * List a fatzap directory.
2503  */
2504 static int
2505 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2506     int (*callback)(const char *, uint64_t))
2507 {
2508 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2509 	fat_zap_t z;
2510 	uint64_t i;
2511 	int j, rc;
2512 
2513 	if (zh->zap_magic != ZAP_MAGIC)
2514 		return (EIO);
2515 
2516 	z.zap_block_shift = ilog2(bsize);
2517 	z.zap_phys = zh;
2518 
2519 	/*
2520 	 * This assumes that the leaf blocks start at block 1. The
2521 	 * documentation isn't exactly clear on this.
2522 	 */
2523 	zap_leaf_t zl;
2524 	zl.l_bs = z.zap_block_shift;
2525 	zl.l_phys = malloc(bsize);
2526 	if (zl.l_phys == NULL)
2527 		return (ENOMEM);
2528 
2529 	for (i = 0; i < zh->zap_num_leafs; i++) {
2530 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2531 		char name[256], *p;
2532 		uint64_t value;
2533 
2534 		if (dnode_read(spa, dnode, off, zl.l_phys, bsize)) {
2535 			free(zl.l_phys);
2536 			return (EIO);
2537 		}
2538 
2539 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2540 			zap_leaf_chunk_t *zc, *nc;
2541 			int namelen;
2542 
2543 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2544 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2545 				continue;
2546 			namelen = zc->l_entry.le_name_numints;
2547 			if (namelen > sizeof(name))
2548 				namelen = sizeof(name);
2549 
2550 			/*
2551 			 * Paste the name back together.
2552 			 */
2553 			nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
2554 			p = name;
2555 			while (namelen > 0) {
2556 				int len;
2557 				len = namelen;
2558 				if (len > ZAP_LEAF_ARRAY_BYTES)
2559 					len = ZAP_LEAF_ARRAY_BYTES;
2560 				memcpy(p, nc->l_array.la_array, len);
2561 				p += len;
2562 				namelen -= len;
2563 				nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
2564 			}
2565 
2566 			/*
2567 			 * Assume the first eight bytes of the value are
2568 			 * a uint64_t.
2569 			 */
2570 			value = fzap_leaf_value(&zl, zc);
2571 
2572 			/* printf("%s 0x%jx\n", name, (uintmax_t)value); */
2573 			rc = callback((const char *)name, value);
2574 			if (rc != 0) {
2575 				free(zl.l_phys);
2576 				return (rc);
2577 			}
2578 		}
2579 	}
2580 
2581 	free(zl.l_phys);
2582 	return (0);
2583 }
2584 
2585 static int zfs_printf(const char *name, uint64_t value __unused)
2586 {
2587 
2588 	printf("%s\n", name);
2589 
2590 	return (0);
2591 }
2592 
2593 /*
2594  * List a zap directory.
2595  */
2596 static int
2597 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
2598 {
2599 	zap_phys_t *zap;
2600 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2601 	int rc;
2602 
2603 	zap = malloc(size);
2604 	if (zap == NULL)
2605 		return (ENOMEM);
2606 
2607 	rc = dnode_read(spa, dnode, 0, zap, size);
2608 	if (rc == 0) {
2609 		if (zap->zap_block_type == ZBT_MICRO)
2610 			rc = mzap_list((const mzap_phys_t *)zap, size,
2611 			    zfs_printf);
2612 		else
2613 			rc = fzap_list(spa, dnode, zap, zfs_printf);
2614 	}
2615 	free(zap);
2616 	return (rc);
2617 }
2618 
2619 static int
2620 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum,
2621     dnode_phys_t *dnode)
2622 {
2623 	off_t offset;
2624 
2625 	offset = objnum * sizeof(dnode_phys_t);
2626 	return dnode_read(spa, &os->os_meta_dnode, offset,
2627 		dnode, sizeof(dnode_phys_t));
2628 }
2629 
2630 /*
2631  * Lookup a name in a microzap directory.
2632  */
2633 static int
2634 mzap_rlookup(const mzap_phys_t *mz, size_t size, char *name, uint64_t value)
2635 {
2636 	const mzap_ent_phys_t *mze;
2637 	int chunks, i;
2638 
2639 	/*
2640 	 * Microzap objects use exactly one block. Read the whole
2641 	 * thing.
2642 	 */
2643 	chunks = size / MZAP_ENT_LEN - 1;
2644 	for (i = 0; i < chunks; i++) {
2645 		mze = &mz->mz_chunk[i];
2646 		if (value == mze->mze_value) {
2647 			strcpy(name, mze->mze_name);
2648 			return (0);
2649 		}
2650 	}
2651 
2652 	return (ENOENT);
2653 }
2654 
2655 static void
2656 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
2657 {
2658 	size_t namelen;
2659 	const zap_leaf_chunk_t *nc;
2660 	char *p;
2661 
2662 	namelen = zc->l_entry.le_name_numints;
2663 
2664 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2665 	p = name;
2666 	while (namelen > 0) {
2667 		size_t len;
2668 		len = namelen;
2669 		if (len > ZAP_LEAF_ARRAY_BYTES)
2670 			len = ZAP_LEAF_ARRAY_BYTES;
2671 		memcpy(p, nc->l_array.la_array, len);
2672 		p += len;
2673 		namelen -= len;
2674 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2675 	}
2676 
2677 	*p = '\0';
2678 }
2679 
2680 static int
2681 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2682     char *name, uint64_t value)
2683 {
2684 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2685 	fat_zap_t z;
2686 	uint64_t i;
2687 	int j, rc;
2688 
2689 	if (zh->zap_magic != ZAP_MAGIC)
2690 		return (EIO);
2691 
2692 	z.zap_block_shift = ilog2(bsize);
2693 	z.zap_phys = zh;
2694 
2695 	/*
2696 	 * This assumes that the leaf blocks start at block 1. The
2697 	 * documentation isn't exactly clear on this.
2698 	 */
2699 	zap_leaf_t zl;
2700 	zl.l_bs = z.zap_block_shift;
2701 	zl.l_phys = malloc(bsize);
2702 	if (zl.l_phys == NULL)
2703 		return (ENOMEM);
2704 
2705 	for (i = 0; i < zh->zap_num_leafs; i++) {
2706 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2707 
2708 		rc = dnode_read(spa, dnode, off, zl.l_phys, bsize);
2709 		if (rc != 0)
2710 			goto done;
2711 
2712 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2713 			zap_leaf_chunk_t *zc;
2714 
2715 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2716 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2717 				continue;
2718 			if (zc->l_entry.le_value_intlen != 8 ||
2719 			    zc->l_entry.le_value_numints != 1)
2720 				continue;
2721 
2722 			if (fzap_leaf_value(&zl, zc) == value) {
2723 				fzap_name_copy(&zl, zc, name);
2724 				goto done;
2725 			}
2726 		}
2727 	}
2728 
2729 	rc = ENOENT;
2730 done:
2731 	free(zl.l_phys);
2732 	return (rc);
2733 }
2734 
2735 static int
2736 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name,
2737     uint64_t value)
2738 {
2739 	zap_phys_t *zap;
2740 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2741 	int rc;
2742 
2743 	zap = malloc(size);
2744 	if (zap == NULL)
2745 		return (ENOMEM);
2746 
2747 	rc = dnode_read(spa, dnode, 0, zap, size);
2748 	if (rc == 0) {
2749 		if (zap->zap_block_type == ZBT_MICRO)
2750 			rc = mzap_rlookup((const mzap_phys_t *)zap, size,
2751 			    name, value);
2752 		else
2753 			rc = fzap_rlookup(spa, dnode, zap, name, value);
2754 	}
2755 	free(zap);
2756 	return (rc);
2757 }
2758 
2759 static int
2760 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
2761 {
2762 	char name[256];
2763 	char component[256];
2764 	uint64_t dir_obj, parent_obj, child_dir_zapobj;
2765 	dnode_phys_t child_dir_zap, dataset, dir, parent;
2766 	dsl_dir_phys_t *dd;
2767 	dsl_dataset_phys_t *ds;
2768 	char *p;
2769 	int len;
2770 
2771 	p = &name[sizeof(name) - 1];
2772 	*p = '\0';
2773 
2774 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
2775 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2776 		return (EIO);
2777 	}
2778 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2779 	dir_obj = ds->ds_dir_obj;
2780 
2781 	for (;;) {
2782 		if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir) != 0)
2783 			return (EIO);
2784 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2785 
2786 		/* Actual loop condition. */
2787 		parent_obj = dd->dd_parent_obj;
2788 		if (parent_obj == 0)
2789 			break;
2790 
2791 		if (objset_get_dnode(spa, spa->spa_mos, parent_obj,
2792 		    &parent) != 0)
2793 			return (EIO);
2794 		dd = (dsl_dir_phys_t *)&parent.dn_bonus;
2795 		child_dir_zapobj = dd->dd_child_dir_zapobj;
2796 		if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
2797 		    &child_dir_zap) != 0)
2798 			return (EIO);
2799 		if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
2800 			return (EIO);
2801 
2802 		len = strlen(component);
2803 		p -= len;
2804 		memcpy(p, component, len);
2805 		--p;
2806 		*p = '/';
2807 
2808 		/* Actual loop iteration. */
2809 		dir_obj = parent_obj;
2810 	}
2811 
2812 	if (*p != '\0')
2813 		++p;
2814 	strcpy(result, p);
2815 
2816 	return (0);
2817 }
2818 
2819 static int
2820 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
2821 {
2822 	char element[256];
2823 	uint64_t dir_obj, child_dir_zapobj;
2824 	dnode_phys_t child_dir_zap, dir;
2825 	dsl_dir_phys_t *dd;
2826 	const char *p, *q;
2827 
2828 	if (objset_get_dnode(spa, spa->spa_mos,
2829 	    DMU_POOL_DIRECTORY_OBJECT, &dir))
2830 		return (EIO);
2831 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
2832 	    1, &dir_obj))
2833 		return (EIO);
2834 
2835 	p = name;
2836 	for (;;) {
2837 		if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir))
2838 			return (EIO);
2839 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2840 
2841 		while (*p == '/')
2842 			p++;
2843 		/* Actual loop condition #1. */
2844 		if (*p == '\0')
2845 			break;
2846 
2847 		q = strchr(p, '/');
2848 		if (q) {
2849 			memcpy(element, p, q - p);
2850 			element[q - p] = '\0';
2851 			p = q + 1;
2852 		} else {
2853 			strcpy(element, p);
2854 			p += strlen(p);
2855 		}
2856 
2857 		child_dir_zapobj = dd->dd_child_dir_zapobj;
2858 		if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
2859 		    &child_dir_zap) != 0)
2860 			return (EIO);
2861 
2862 		/* Actual loop condition #2. */
2863 		if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
2864 		    1, &dir_obj) != 0)
2865 			return (ENOENT);
2866 	}
2867 
2868 	*objnum = dd->dd_head_dataset_obj;
2869 	return (0);
2870 }
2871 
2872 #ifndef BOOT2
2873 static int
2874 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
2875 {
2876 	uint64_t dir_obj, child_dir_zapobj;
2877 	dnode_phys_t child_dir_zap, dir, dataset;
2878 	dsl_dataset_phys_t *ds;
2879 	dsl_dir_phys_t *dd;
2880 
2881 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
2882 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2883 		return (EIO);
2884 	}
2885 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2886 	dir_obj = ds->ds_dir_obj;
2887 
2888 	if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir)) {
2889 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2890 		return (EIO);
2891 	}
2892 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2893 
2894 	child_dir_zapobj = dd->dd_child_dir_zapobj;
2895 	if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
2896 	    &child_dir_zap) != 0) {
2897 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2898 		return (EIO);
2899 	}
2900 
2901 	return (zap_list(spa, &child_dir_zap) != 0);
2902 }
2903 
2904 int
2905 zfs_callback_dataset(const spa_t *spa, uint64_t objnum,
2906     int (*callback)(const char *, uint64_t))
2907 {
2908 	uint64_t dir_obj, child_dir_zapobj;
2909 	dnode_phys_t child_dir_zap, dir, dataset;
2910 	dsl_dataset_phys_t *ds;
2911 	dsl_dir_phys_t *dd;
2912 	zap_phys_t *zap;
2913 	size_t size;
2914 	int err;
2915 
2916 	err = objset_get_dnode(spa, spa->spa_mos, objnum, &dataset);
2917 	if (err != 0) {
2918 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2919 		return (err);
2920 	}
2921 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2922 	dir_obj = ds->ds_dir_obj;
2923 
2924 	err = objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir);
2925 	if (err != 0) {
2926 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2927 		return (err);
2928 	}
2929 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2930 
2931 	child_dir_zapobj = dd->dd_child_dir_zapobj;
2932 	err = objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
2933 	    &child_dir_zap);
2934 	if (err != 0) {
2935 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2936 		return (err);
2937 	}
2938 
2939 	size = child_dir_zap.dn_datablkszsec << SPA_MINBLOCKSHIFT;
2940 	zap = malloc(size);
2941 	if (zap != NULL) {
2942 		err = dnode_read(spa, &child_dir_zap, 0, zap, size);
2943 		if (err != 0)
2944 			goto done;
2945 
2946 		if (zap->zap_block_type == ZBT_MICRO)
2947 			err = mzap_list((const mzap_phys_t *)zap, size,
2948 			    callback);
2949 		else
2950 			err = fzap_list(spa, &child_dir_zap, zap, callback);
2951 	} else {
2952 		err = ENOMEM;
2953 	}
2954 done:
2955 	free(zap);
2956 	return (err);
2957 }
2958 #endif
2959 
2960 /*
2961  * Find the object set given the object number of its dataset object
2962  * and return its details in *objset
2963  */
2964 static int
2965 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
2966 {
2967 	dnode_phys_t dataset;
2968 	dsl_dataset_phys_t *ds;
2969 
2970 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
2971 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2972 		return (EIO);
2973 	}
2974 
2975 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2976 	if (zio_read(spa, &ds->ds_bp, objset)) {
2977 		printf("ZFS: can't read object set for dataset %ju\n",
2978 		    (uintmax_t)objnum);
2979 		return (EIO);
2980 	}
2981 
2982 	return (0);
2983 }
2984 
2985 /*
2986  * Find the object set pointed to by the BOOTFS property or the root
2987  * dataset if there is none and return its details in *objset
2988  */
2989 static int
2990 zfs_get_root(const spa_t *spa, uint64_t *objid)
2991 {
2992 	dnode_phys_t dir, propdir;
2993 	uint64_t props, bootfs, root;
2994 
2995 	*objid = 0;
2996 
2997 	/*
2998 	 * Start with the MOS directory object.
2999 	 */
3000 	if (objset_get_dnode(spa, spa->spa_mos,
3001 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3002 		printf("ZFS: can't read MOS object directory\n");
3003 		return (EIO);
3004 	}
3005 
3006 	/*
3007 	 * Lookup the pool_props and see if we can find a bootfs.
3008 	 */
3009 	if (zap_lookup(spa, &dir, DMU_POOL_PROPS,
3010 	    sizeof(props), 1, &props) == 0 &&
3011 	    objset_get_dnode(spa, spa->spa_mos, props, &propdir) == 0 &&
3012 	    zap_lookup(spa, &propdir, "bootfs",
3013 	    sizeof(bootfs), 1, &bootfs) == 0 && bootfs != 0) {
3014 		*objid = bootfs;
3015 		return (0);
3016 	}
3017 	/*
3018 	 * Lookup the root dataset directory
3019 	 */
3020 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET,
3021 	    sizeof(root), 1, &root) ||
3022 	    objset_get_dnode(spa, spa->spa_mos, root, &dir)) {
3023 		printf("ZFS: can't find root dsl_dir\n");
3024 		return (EIO);
3025 	}
3026 
3027 	/*
3028 	 * Use the information from the dataset directory's bonus buffer
3029 	 * to find the dataset object and from that the object set itself.
3030 	 */
3031 	dsl_dir_phys_t *dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3032 	*objid = dd->dd_head_dataset_obj;
3033 	return (0);
3034 }
3035 
3036 static int
3037 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
3038 {
3039 
3040 	mount->spa = spa;
3041 
3042 	/*
3043 	 * Find the root object set if not explicitly provided
3044 	 */
3045 	if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
3046 		printf("ZFS: can't find root filesystem\n");
3047 		return (EIO);
3048 	}
3049 
3050 	if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
3051 		printf("ZFS: can't open root filesystem\n");
3052 		return (EIO);
3053 	}
3054 
3055 	mount->rootobj = rootobj;
3056 
3057 	return (0);
3058 }
3059 
3060 /*
3061  * callback function for feature name checks.
3062  */
3063 static int
3064 check_feature(const char *name, uint64_t value)
3065 {
3066 	int i;
3067 
3068 	if (value == 0)
3069 		return (0);
3070 	if (name[0] == '\0')
3071 		return (0);
3072 
3073 	for (i = 0; features_for_read[i] != NULL; i++) {
3074 		if (strcmp(name, features_for_read[i]) == 0)
3075 			return (0);
3076 	}
3077 	printf("ZFS: unsupported feature: %s\n", name);
3078 	return (EIO);
3079 }
3080 
3081 /*
3082  * Checks whether the MOS features that are active are supported.
3083  */
3084 static int
3085 check_mos_features(const spa_t *spa)
3086 {
3087 	dnode_phys_t dir;
3088 	zap_phys_t *zap;
3089 	uint64_t objnum;
3090 	size_t size;
3091 	int rc;
3092 
3093 	if ((rc = objset_get_dnode(spa, spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
3094 	    &dir)) != 0)
3095 		return (rc);
3096 	if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
3097 	    sizeof (objnum), 1, &objnum)) != 0) {
3098 		/*
3099 		 * It is older pool without features. As we have already
3100 		 * tested the label, just return without raising the error.
3101 		 */
3102 		return (0);
3103 	}
3104 
3105 	if ((rc = objset_get_dnode(spa, spa->spa_mos, objnum, &dir)) != 0)
3106 		return (rc);
3107 
3108 	if (dir.dn_type != DMU_OTN_ZAP_METADATA)
3109 		return (EIO);
3110 
3111 	size = dir.dn_datablkszsec << SPA_MINBLOCKSHIFT;
3112 	zap = malloc(size);
3113 	if (zap == NULL)
3114 		return (ENOMEM);
3115 
3116 	if (dnode_read(spa, &dir, 0, zap, size)) {
3117 		free(zap);
3118 		return (EIO);
3119 	}
3120 
3121 	if (zap->zap_block_type == ZBT_MICRO)
3122 		rc = mzap_list((const mzap_phys_t *)zap, size, check_feature);
3123 	else
3124 		rc = fzap_list(spa, &dir, zap, check_feature);
3125 
3126 	free(zap);
3127 	return (rc);
3128 }
3129 
3130 static int
3131 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
3132 {
3133 	dnode_phys_t dir;
3134 	size_t size;
3135 	int rc;
3136 	unsigned char *nv;
3137 
3138 	*value = NULL;
3139 	if ((rc = objset_get_dnode(spa, spa->spa_mos, obj, &dir)) != 0)
3140 		return (rc);
3141 	if (dir.dn_type != DMU_OT_PACKED_NVLIST &&
3142 	    dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) {
3143 		return (EIO);
3144 	}
3145 
3146 	if (dir.dn_bonuslen != sizeof (uint64_t))
3147 		return (EIO);
3148 
3149 	size = *(uint64_t *)DN_BONUS(&dir);
3150 	nv = malloc(size);
3151 	if (nv == NULL)
3152 		return (ENOMEM);
3153 
3154 	rc = dnode_read(spa, &dir, 0, nv, size);
3155 	if (rc != 0) {
3156 		free(nv);
3157 		nv = NULL;
3158 		return (rc);
3159 	}
3160 	*value = nvlist_import(nv + 4, nv[0], nv[1]);
3161 	free(nv);
3162 	return (rc);
3163 }
3164 
3165 static int
3166 zfs_spa_init(spa_t *spa)
3167 {
3168 	struct uberblock checkpoint;
3169 	dnode_phys_t dir;
3170 	uint64_t config_object;
3171 	nvlist_t *nvlist;
3172 	int rc;
3173 
3174 	if (zio_read(spa, &spa->spa_uberblock->ub_rootbp, spa->spa_mos)) {
3175 		printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
3176 		return (EIO);
3177 	}
3178 	if (spa->spa_mos->os_type != DMU_OST_META) {
3179 		printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
3180 		return (EIO);
3181 	}
3182 
3183 	if (objset_get_dnode(spa, &spa->spa_mos_master,
3184 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3185 		printf("ZFS: failed to read pool %s directory object\n",
3186 		    spa->spa_name);
3187 		return (EIO);
3188 	}
3189 	/* this is allowed to fail, older pools do not have salt */
3190 	rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
3191 	    sizeof (spa->spa_cksum_salt.zcs_bytes),
3192 	    spa->spa_cksum_salt.zcs_bytes);
3193 
3194 	rc = check_mos_features(spa);
3195 	if (rc != 0) {
3196 		printf("ZFS: pool %s is not supported\n", spa->spa_name);
3197 		return (rc);
3198 	}
3199 
3200 	rc = zap_lookup(spa, &dir, DMU_POOL_CONFIG,
3201 	    sizeof (config_object), 1, &config_object);
3202 	if (rc != 0) {
3203 		printf("ZFS: can not read MOS %s\n", DMU_POOL_CONFIG);
3204 		return (EIO);
3205 	}
3206 	rc = load_nvlist(spa, config_object, &nvlist);
3207 	if (rc != 0)
3208 		return (rc);
3209 
3210 	rc = zap_lookup(spa, &dir, DMU_POOL_ZPOOL_CHECKPOINT,
3211 	    sizeof(uint64_t), sizeof(checkpoint) / sizeof(uint64_t),
3212 	    &checkpoint);
3213 	if (rc == 0 && checkpoint.ub_checkpoint_txg != 0) {
3214 		memcpy(&spa->spa_uberblock_checkpoint, &checkpoint,
3215 		    sizeof(checkpoint));
3216 		if (zio_read(spa, &spa->spa_uberblock_checkpoint.ub_rootbp,
3217 		    &spa->spa_mos_checkpoint)) {
3218 			printf("ZFS: can not read checkpoint data.\n");
3219 			return (EIO);
3220 		}
3221 	}
3222 
3223 	/*
3224 	 * Update vdevs from MOS config. Note, we do skip encoding bytes
3225 	 * here. See also vdev_label_read_config().
3226 	 */
3227 	rc = vdev_init_from_nvlist(spa, nvlist);
3228 	nvlist_destroy(nvlist);
3229 	return (rc);
3230 }
3231 
3232 static int
3233 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
3234 {
3235 
3236 	if (dn->dn_bonustype != DMU_OT_SA) {
3237 		znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
3238 
3239 		sb->st_mode = zp->zp_mode;
3240 		sb->st_uid = zp->zp_uid;
3241 		sb->st_gid = zp->zp_gid;
3242 		sb->st_size = zp->zp_size;
3243 	} else {
3244 		sa_hdr_phys_t *sahdrp;
3245 		int hdrsize;
3246 		size_t size = 0;
3247 		void *buf = NULL;
3248 
3249 		if (dn->dn_bonuslen != 0)
3250 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3251 		else {
3252 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
3253 				blkptr_t *bp = DN_SPILL_BLKPTR(dn);
3254 				int error;
3255 
3256 				size = BP_GET_LSIZE(bp);
3257 				buf = malloc(size);
3258 				if (buf == NULL)
3259 					error = ENOMEM;
3260 				else
3261 					error = zio_read(spa, bp, buf);
3262 
3263 				if (error != 0) {
3264 					free(buf);
3265 					return (error);
3266 				}
3267 				sahdrp = buf;
3268 			} else {
3269 				return (EIO);
3270 			}
3271 		}
3272 		hdrsize = SA_HDR_SIZE(sahdrp);
3273 		sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
3274 		    SA_MODE_OFFSET);
3275 		sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
3276 		    SA_UID_OFFSET);
3277 		sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
3278 		    SA_GID_OFFSET);
3279 		sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
3280 		    SA_SIZE_OFFSET);
3281 		free(buf);
3282 	}
3283 
3284 	return (0);
3285 }
3286 
3287 static int
3288 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
3289 {
3290 	int rc = 0;
3291 
3292 	if (dn->dn_bonustype == DMU_OT_SA) {
3293 		sa_hdr_phys_t *sahdrp = NULL;
3294 		size_t size = 0;
3295 		void *buf = NULL;
3296 		int hdrsize;
3297 		char *p;
3298 
3299 		if (dn->dn_bonuslen != 0) {
3300 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3301 		} else {
3302 			blkptr_t *bp;
3303 
3304 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
3305 				return (EIO);
3306 			bp = DN_SPILL_BLKPTR(dn);
3307 
3308 			size = BP_GET_LSIZE(bp);
3309 			buf = malloc(size);
3310 			if (buf == NULL)
3311 				rc = ENOMEM;
3312 			else
3313 				rc = zio_read(spa, bp, buf);
3314 			if (rc != 0) {
3315 				free(buf);
3316 				return (rc);
3317 			}
3318 			sahdrp = buf;
3319 		}
3320 		hdrsize = SA_HDR_SIZE(sahdrp);
3321 		p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
3322 		memcpy(path, p, psize);
3323 		free(buf);
3324 		return (0);
3325 	}
3326 	/*
3327 	 * Second test is purely to silence bogus compiler
3328 	 * warning about accessing past the end of dn_bonus.
3329 	 */
3330 	if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
3331 	    sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
3332 		memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
3333 	} else {
3334 		rc = dnode_read(spa, dn, 0, path, psize);
3335 	}
3336 	return (rc);
3337 }
3338 
3339 struct obj_list {
3340 	uint64_t		objnum;
3341 	STAILQ_ENTRY(obj_list)	entry;
3342 };
3343 
3344 /*
3345  * Lookup a file and return its dnode.
3346  */
3347 static int
3348 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
3349 {
3350 	int rc;
3351 	uint64_t objnum;
3352 	const spa_t *spa;
3353 	dnode_phys_t dn;
3354 	const char *p, *q;
3355 	char element[256];
3356 	char path[1024];
3357 	int symlinks_followed = 0;
3358 	struct stat sb;
3359 	struct obj_list *entry, *tentry;
3360 	STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
3361 
3362 	spa = mount->spa;
3363 	if (mount->objset.os_type != DMU_OST_ZFS) {
3364 		printf("ZFS: unexpected object set type %ju\n",
3365 		    (uintmax_t)mount->objset.os_type);
3366 		return (EIO);
3367 	}
3368 
3369 	if ((entry = malloc(sizeof(struct obj_list))) == NULL)
3370 		return (ENOMEM);
3371 
3372 	/*
3373 	 * Get the root directory dnode.
3374 	 */
3375 	rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
3376 	if (rc) {
3377 		free(entry);
3378 		return (rc);
3379 	}
3380 
3381 	rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof(objnum), 1, &objnum);
3382 	if (rc) {
3383 		free(entry);
3384 		return (rc);
3385 	}
3386 	entry->objnum = objnum;
3387 	STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3388 
3389 	rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3390 	if (rc != 0)
3391 		goto done;
3392 
3393 	p = upath;
3394 	while (p && *p) {
3395 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3396 		if (rc != 0)
3397 			goto done;
3398 
3399 		while (*p == '/')
3400 			p++;
3401 		if (*p == '\0')
3402 			break;
3403 		q = p;
3404 		while (*q != '\0' && *q != '/')
3405 			q++;
3406 
3407 		/* skip dot */
3408 		if (p + 1 == q && p[0] == '.') {
3409 			p++;
3410 			continue;
3411 		}
3412 		/* double dot */
3413 		if (p + 2 == q && p[0] == '.' && p[1] == '.') {
3414 			p += 2;
3415 			if (STAILQ_FIRST(&on_cache) ==
3416 			    STAILQ_LAST(&on_cache, obj_list, entry)) {
3417 				rc = ENOENT;
3418 				goto done;
3419 			}
3420 			entry = STAILQ_FIRST(&on_cache);
3421 			STAILQ_REMOVE_HEAD(&on_cache, entry);
3422 			free(entry);
3423 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3424 			continue;
3425 		}
3426 		if (q - p + 1 > sizeof(element)) {
3427 			rc = ENAMETOOLONG;
3428 			goto done;
3429 		}
3430 		memcpy(element, p, q - p);
3431 		element[q - p] = 0;
3432 		p = q;
3433 
3434 		if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
3435 			goto done;
3436 		if (!S_ISDIR(sb.st_mode)) {
3437 			rc = ENOTDIR;
3438 			goto done;
3439 		}
3440 
3441 		rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum);
3442 		if (rc)
3443 			goto done;
3444 		objnum = ZFS_DIRENT_OBJ(objnum);
3445 
3446 		if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
3447 			rc = ENOMEM;
3448 			goto done;
3449 		}
3450 		entry->objnum = objnum;
3451 		STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3452 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3453 		if (rc)
3454 			goto done;
3455 
3456 		/*
3457 		 * Check for symlink.
3458 		 */
3459 		rc = zfs_dnode_stat(spa, &dn, &sb);
3460 		if (rc)
3461 			goto done;
3462 		if (S_ISLNK(sb.st_mode)) {
3463 			if (symlinks_followed > 10) {
3464 				rc = EMLINK;
3465 				goto done;
3466 			}
3467 			symlinks_followed++;
3468 
3469 			/*
3470 			 * Read the link value and copy the tail of our
3471 			 * current path onto the end.
3472 			 */
3473 			if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
3474 				rc = ENAMETOOLONG;
3475 				goto done;
3476 			}
3477 			strcpy(&path[sb.st_size], p);
3478 
3479 			rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
3480 			if (rc != 0)
3481 				goto done;
3482 
3483 			/*
3484 			 * Restart with the new path, starting either at
3485 			 * the root or at the parent depending whether or
3486 			 * not the link is relative.
3487 			 */
3488 			p = path;
3489 			if (*p == '/') {
3490 				while (STAILQ_FIRST(&on_cache) !=
3491 				    STAILQ_LAST(&on_cache, obj_list, entry)) {
3492 					entry = STAILQ_FIRST(&on_cache);
3493 					STAILQ_REMOVE_HEAD(&on_cache, entry);
3494 					free(entry);
3495 				}
3496 			} else {
3497 				entry = STAILQ_FIRST(&on_cache);
3498 				STAILQ_REMOVE_HEAD(&on_cache, entry);
3499 				free(entry);
3500 			}
3501 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3502 		}
3503 	}
3504 
3505 	*dnode = dn;
3506 done:
3507 	STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)
3508 		free(entry);
3509 	return (rc);
3510 }
3511