xref: /freebsd/stand/libsa/zfs/zfsimpl.c (revision d5e3895ea4fe4ef9db8823774e07b4368180a23e)
1 /*-
2  * Copyright (c) 2007 Doug Rabson
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 /*
31  *	Stand-alone ZFS file reader.
32  */
33 
34 #include <sys/endian.h>
35 #include <sys/stat.h>
36 #include <sys/stdint.h>
37 #include <sys/list.h>
38 #include <machine/_inttypes.h>
39 
40 #include "zfsimpl.h"
41 #include "zfssubr.c"
42 
43 
44 struct zfsmount {
45 	const spa_t	*spa;
46 	objset_phys_t	objset;
47 	uint64_t	rootobj;
48 };
49 static struct zfsmount zfsmount __unused;
50 
51 /*
52  * The indirect_child_t represents the vdev that we will read from, when we
53  * need to read all copies of the data (e.g. for scrub or reconstruction).
54  * For plain (non-mirror) top-level vdevs (i.e. is_vdev is not a mirror),
55  * ic_vdev is the same as is_vdev.  However, for mirror top-level vdevs,
56  * ic_vdev is a child of the mirror.
57  */
58 typedef struct indirect_child {
59 	void *ic_data;
60 	vdev_t *ic_vdev;
61 } indirect_child_t;
62 
63 /*
64  * The indirect_split_t represents one mapped segment of an i/o to the
65  * indirect vdev. For non-split (contiguously-mapped) blocks, there will be
66  * only one indirect_split_t, with is_split_offset==0 and is_size==io_size.
67  * For split blocks, there will be several of these.
68  */
69 typedef struct indirect_split {
70 	list_node_t is_node; /* link on iv_splits */
71 
72 	/*
73 	 * is_split_offset is the offset into the i/o.
74 	 * This is the sum of the previous splits' is_size's.
75 	 */
76 	uint64_t is_split_offset;
77 
78 	vdev_t *is_vdev; /* top-level vdev */
79 	uint64_t is_target_offset; /* offset on is_vdev */
80 	uint64_t is_size;
81 	int is_children; /* number of entries in is_child[] */
82 
83 	/*
84 	 * is_good_child is the child that we are currently using to
85 	 * attempt reconstruction.
86 	 */
87 	int is_good_child;
88 
89 	indirect_child_t is_child[1]; /* variable-length */
90 } indirect_split_t;
91 
92 /*
93  * The indirect_vsd_t is associated with each i/o to the indirect vdev.
94  * It is the "Vdev-Specific Data" in the zio_t's io_vsd.
95  */
96 typedef struct indirect_vsd {
97 	boolean_t iv_split_block;
98 	boolean_t iv_reconstruct;
99 
100 	list_t iv_splits; /* list of indirect_split_t's */
101 } indirect_vsd_t;
102 
103 /*
104  * List of all vdevs, chained through v_alllink.
105  */
106 static vdev_list_t zfs_vdevs;
107 
108 /*
109  * List of ZFS features supported for read
110  */
111 static const char *features_for_read[] = {
112 	"org.illumos:lz4_compress",
113 	"com.delphix:hole_birth",
114 	"com.delphix:extensible_dataset",
115 	"com.delphix:embedded_data",
116 	"org.open-zfs:large_blocks",
117 	"org.illumos:sha512",
118 	"org.illumos:skein",
119 	"org.zfsonlinux:large_dnode",
120 	"com.joyent:multi_vdev_crash_dump",
121 	"com.delphix:spacemap_histogram",
122 	"com.delphix:zpool_checkpoint",
123 	"com.delphix:spacemap_v2",
124 	"com.datto:encryption",
125 	"org.zfsonlinux:allocation_classes",
126 	"com.datto:resilver_defer",
127 	"com.delphix:device_removal",
128 	"com.delphix:obsolete_counts",
129 	"com.intel:allocation_classes",
130 	NULL
131 };
132 
133 /*
134  * List of all pools, chained through spa_link.
135  */
136 static spa_list_t zfs_pools;
137 
138 static const dnode_phys_t *dnode_cache_obj;
139 static uint64_t dnode_cache_bn;
140 static char *dnode_cache_buf;
141 
142 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
143 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
144 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
145 static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode,
146     const char *name, uint64_t integer_size, uint64_t num_integers,
147     void *value);
148 static int objset_get_dnode(const spa_t *, const objset_phys_t *, uint64_t,
149     dnode_phys_t *);
150 static int dnode_read(const spa_t *, const dnode_phys_t *, off_t, void *,
151     size_t);
152 static int vdev_indirect_read(vdev_t *, const blkptr_t *, void *, off_t,
153     size_t);
154 static int vdev_mirror_read(vdev_t *, const blkptr_t *, void *, off_t, size_t);
155 vdev_indirect_mapping_t *vdev_indirect_mapping_open(spa_t *, objset_phys_t *,
156     uint64_t);
157 vdev_indirect_mapping_entry_phys_t *
158     vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *, uint64_t,
159     uint64_t, uint64_t *);
160 
161 static void
162 zfs_init(void)
163 {
164 	STAILQ_INIT(&zfs_vdevs);
165 	STAILQ_INIT(&zfs_pools);
166 
167 	dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
168 
169 	zfs_init_crc();
170 }
171 
172 static int
173 nvlist_check_features_for_read(nvlist_t *nvl)
174 {
175 	nvlist_t *features = NULL;
176 	nvs_data_t *data;
177 	nvp_header_t *nvp;
178 	nv_string_t *nvp_name;
179 	int rc;
180 
181 	rc = nvlist_find(nvl, ZPOOL_CONFIG_FEATURES_FOR_READ,
182 	    DATA_TYPE_NVLIST, NULL, &features, NULL);
183 	if (rc != 0)
184 		return (rc);
185 
186 	data = (nvs_data_t *)features->nv_data;
187 	nvp = &data->nvl_pair;	/* first pair in nvlist */
188 
189 	while (nvp->encoded_size != 0 && nvp->decoded_size != 0) {
190 		int i, found;
191 
192 		nvp_name = (nv_string_t *)((uintptr_t)nvp + sizeof(*nvp));
193 		found = 0;
194 
195 		for (i = 0; features_for_read[i] != NULL; i++) {
196 			if (memcmp(nvp_name->nv_data, features_for_read[i],
197 			    nvp_name->nv_size) == 0) {
198 				found = 1;
199 				break;
200 			}
201 		}
202 
203 		if (!found) {
204 			printf("ZFS: unsupported feature: %.*s\n",
205 			    nvp_name->nv_size, nvp_name->nv_data);
206 			rc = EIO;
207 		}
208 		nvp = (nvp_header_t *)((uint8_t *)nvp + nvp->encoded_size);
209 	}
210 	nvlist_destroy(features);
211 
212 	return (rc);
213 }
214 
215 static int
216 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
217     off_t offset, size_t size)
218 {
219 	size_t psize;
220 	int rc;
221 
222 	if (!vdev->v_phys_read)
223 		return (EIO);
224 
225 	if (bp) {
226 		psize = BP_GET_PSIZE(bp);
227 	} else {
228 		psize = size;
229 	}
230 
231 	rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
232 	if (rc == 0) {
233 		if (bp != NULL)
234 			rc = zio_checksum_verify(vdev->v_spa, bp, buf);
235 	}
236 
237 	return (rc);
238 }
239 
240 typedef struct remap_segment {
241 	vdev_t *rs_vd;
242 	uint64_t rs_offset;
243 	uint64_t rs_asize;
244 	uint64_t rs_split_offset;
245 	list_node_t rs_node;
246 } remap_segment_t;
247 
248 static remap_segment_t *
249 rs_alloc(vdev_t *vd, uint64_t offset, uint64_t asize, uint64_t split_offset)
250 {
251 	remap_segment_t *rs = malloc(sizeof (remap_segment_t));
252 
253 	if (rs != NULL) {
254 		rs->rs_vd = vd;
255 		rs->rs_offset = offset;
256 		rs->rs_asize = asize;
257 		rs->rs_split_offset = split_offset;
258 	}
259 
260 	return (rs);
261 }
262 
263 vdev_indirect_mapping_t *
264 vdev_indirect_mapping_open(spa_t *spa, objset_phys_t *os,
265     uint64_t mapping_object)
266 {
267 	vdev_indirect_mapping_t *vim;
268 	vdev_indirect_mapping_phys_t *vim_phys;
269 	int rc;
270 
271 	vim = calloc(1, sizeof (*vim));
272 	if (vim == NULL)
273 		return (NULL);
274 
275 	vim->vim_dn = calloc(1, sizeof (*vim->vim_dn));
276 	if (vim->vim_dn == NULL) {
277 		free(vim);
278 		return (NULL);
279 	}
280 
281 	rc = objset_get_dnode(spa, os, mapping_object, vim->vim_dn);
282 	if (rc != 0) {
283 		free(vim->vim_dn);
284 		free(vim);
285 		return (NULL);
286 	}
287 
288 	vim->vim_spa = spa;
289 	vim->vim_phys = malloc(sizeof (*vim->vim_phys));
290 	if (vim->vim_phys == NULL) {
291 		free(vim->vim_dn);
292 		free(vim);
293 		return (NULL);
294 	}
295 
296 	vim_phys = (vdev_indirect_mapping_phys_t *)DN_BONUS(vim->vim_dn);
297 	*vim->vim_phys = *vim_phys;
298 
299 	vim->vim_objset = os;
300 	vim->vim_object = mapping_object;
301 	vim->vim_entries = NULL;
302 
303 	vim->vim_havecounts =
304 	    (vim->vim_dn->dn_bonuslen > VDEV_INDIRECT_MAPPING_SIZE_V0);
305 
306 	return (vim);
307 }
308 
309 /*
310  * Compare an offset with an indirect mapping entry; there are three
311  * possible scenarios:
312  *
313  *     1. The offset is "less than" the mapping entry; meaning the
314  *        offset is less than the source offset of the mapping entry. In
315  *        this case, there is no overlap between the offset and the
316  *        mapping entry and -1 will be returned.
317  *
318  *     2. The offset is "greater than" the mapping entry; meaning the
319  *        offset is greater than the mapping entry's source offset plus
320  *        the entry's size. In this case, there is no overlap between
321  *        the offset and the mapping entry and 1 will be returned.
322  *
323  *        NOTE: If the offset is actually equal to the entry's offset
324  *        plus size, this is considered to be "greater" than the entry,
325  *        and this case applies (i.e. 1 will be returned). Thus, the
326  *        entry's "range" can be considered to be inclusive at its
327  *        start, but exclusive at its end: e.g. [src, src + size).
328  *
329  *     3. The last case to consider is if the offset actually falls
330  *        within the mapping entry's range. If this is the case, the
331  *        offset is considered to be "equal to" the mapping entry and
332  *        0 will be returned.
333  *
334  *        NOTE: If the offset is equal to the entry's source offset,
335  *        this case applies and 0 will be returned. If the offset is
336  *        equal to the entry's source plus its size, this case does
337  *        *not* apply (see "NOTE" above for scenario 2), and 1 will be
338  *        returned.
339  */
340 static int
341 dva_mapping_overlap_compare(const void *v_key, const void *v_array_elem)
342 {
343 	const uint64_t *key = v_key;
344 	const vdev_indirect_mapping_entry_phys_t *array_elem =
345 	    v_array_elem;
346 	uint64_t src_offset = DVA_MAPPING_GET_SRC_OFFSET(array_elem);
347 
348 	if (*key < src_offset) {
349 		return (-1);
350 	} else if (*key < src_offset + DVA_GET_ASIZE(&array_elem->vimep_dst)) {
351 		return (0);
352 	} else {
353 		return (1);
354 	}
355 }
356 
357 /*
358  * Return array entry.
359  */
360 static vdev_indirect_mapping_entry_phys_t *
361 vdev_indirect_mapping_entry(vdev_indirect_mapping_t *vim, uint64_t index)
362 {
363 	uint64_t size;
364 	off_t offset = 0;
365 	int rc;
366 
367 	if (vim->vim_phys->vimp_num_entries == 0)
368 		return (NULL);
369 
370 	if (vim->vim_entries == NULL) {
371 		uint64_t bsize;
372 
373 		bsize = vim->vim_dn->dn_datablkszsec << SPA_MINBLOCKSHIFT;
374 		size = vim->vim_phys->vimp_num_entries *
375 		    sizeof (*vim->vim_entries);
376 		if (size > bsize) {
377 			size = bsize / sizeof (*vim->vim_entries);
378 			size *= sizeof (*vim->vim_entries);
379 		}
380 		vim->vim_entries = malloc(size);
381 		if (vim->vim_entries == NULL)
382 			return (NULL);
383 		vim->vim_num_entries = size / sizeof (*vim->vim_entries);
384 		offset = index * sizeof (*vim->vim_entries);
385 	}
386 
387 	/* We have data in vim_entries */
388 	if (offset == 0) {
389 		if (index >= vim->vim_entry_offset &&
390 		    index <= vim->vim_entry_offset + vim->vim_num_entries) {
391 			index -= vim->vim_entry_offset;
392 			return (&vim->vim_entries[index]);
393 		}
394 		offset = index * sizeof (*vim->vim_entries);
395 	}
396 
397 	vim->vim_entry_offset = index;
398 	size = vim->vim_num_entries * sizeof (*vim->vim_entries);
399 	rc = dnode_read(vim->vim_spa, vim->vim_dn, offset, vim->vim_entries,
400 	    size);
401 	if (rc != 0) {
402 		/* Read error, invalidate vim_entries. */
403 		free(vim->vim_entries);
404 		vim->vim_entries = NULL;
405 		return (NULL);
406 	}
407 	index -= vim->vim_entry_offset;
408 	return (&vim->vim_entries[index]);
409 }
410 
411 /*
412  * Returns the mapping entry for the given offset.
413  *
414  * It's possible that the given offset will not be in the mapping table
415  * (i.e. no mapping entries contain this offset), in which case, the
416  * return value value depends on the "next_if_missing" parameter.
417  *
418  * If the offset is not found in the table and "next_if_missing" is
419  * B_FALSE, then NULL will always be returned. The behavior is intended
420  * to allow consumers to get the entry corresponding to the offset
421  * parameter, iff the offset overlaps with an entry in the table.
422  *
423  * If the offset is not found in the table and "next_if_missing" is
424  * B_TRUE, then the entry nearest to the given offset will be returned,
425  * such that the entry's source offset is greater than the offset
426  * passed in (i.e. the "next" mapping entry in the table is returned, if
427  * the offset is missing from the table). If there are no entries whose
428  * source offset is greater than the passed in offset, NULL is returned.
429  */
430 static vdev_indirect_mapping_entry_phys_t *
431 vdev_indirect_mapping_entry_for_offset(vdev_indirect_mapping_t *vim,
432     uint64_t offset)
433 {
434 	ASSERT(vim->vim_phys->vimp_num_entries > 0);
435 
436 	vdev_indirect_mapping_entry_phys_t *entry;
437 
438 	uint64_t last = vim->vim_phys->vimp_num_entries - 1;
439 	uint64_t base = 0;
440 
441 	/*
442 	 * We don't define these inside of the while loop because we use
443 	 * their value in the case that offset isn't in the mapping.
444 	 */
445 	uint64_t mid;
446 	int result;
447 
448 	while (last >= base) {
449 		mid = base + ((last - base) >> 1);
450 
451 		entry = vdev_indirect_mapping_entry(vim, mid);
452 		if (entry == NULL)
453 			break;
454 		result = dva_mapping_overlap_compare(&offset, entry);
455 
456 		if (result == 0) {
457 			break;
458 		} else if (result < 0) {
459 			last = mid - 1;
460 		} else {
461 			base = mid + 1;
462 		}
463 	}
464 	return (entry);
465 }
466 
467 /*
468  * Given an indirect vdev and an extent on that vdev, it duplicates the
469  * physical entries of the indirect mapping that correspond to the extent
470  * to a new array and returns a pointer to it. In addition, copied_entries
471  * is populated with the number of mapping entries that were duplicated.
472  *
473  * Finally, since we are doing an allocation, it is up to the caller to
474  * free the array allocated in this function.
475  */
476 vdev_indirect_mapping_entry_phys_t *
477 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *vd, uint64_t offset,
478     uint64_t asize, uint64_t *copied_entries)
479 {
480 	vdev_indirect_mapping_entry_phys_t *duplicate_mappings = NULL;
481 	vdev_indirect_mapping_t *vim = vd->v_mapping;
482 	uint64_t entries = 0;
483 
484 	vdev_indirect_mapping_entry_phys_t *first_mapping =
485 	    vdev_indirect_mapping_entry_for_offset(vim, offset);
486 	ASSERT3P(first_mapping, !=, NULL);
487 
488 	vdev_indirect_mapping_entry_phys_t *m = first_mapping;
489 	while (asize > 0) {
490 		uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
491 		uint64_t inner_offset = offset - DVA_MAPPING_GET_SRC_OFFSET(m);
492 		uint64_t inner_size = MIN(asize, size - inner_offset);
493 
494 		offset += inner_size;
495 		asize -= inner_size;
496 		entries++;
497 		m++;
498 	}
499 
500 	size_t copy_length = entries * sizeof (*first_mapping);
501 	duplicate_mappings = malloc(copy_length);
502 	if (duplicate_mappings != NULL)
503 		bcopy(first_mapping, duplicate_mappings, copy_length);
504 	else
505 		entries = 0;
506 
507 	*copied_entries = entries;
508 
509 	return (duplicate_mappings);
510 }
511 
512 static vdev_t *
513 vdev_lookup_top(spa_t *spa, uint64_t vdev)
514 {
515 	vdev_t *rvd;
516 	vdev_list_t *vlist;
517 
518 	vlist = &spa->spa_root_vdev->v_children;
519 	STAILQ_FOREACH(rvd, vlist, v_childlink)
520 		if (rvd->v_id == vdev)
521 			break;
522 
523 	return (rvd);
524 }
525 
526 /*
527  * This is a callback for vdev_indirect_remap() which allocates an
528  * indirect_split_t for each split segment and adds it to iv_splits.
529  */
530 static void
531 vdev_indirect_gather_splits(uint64_t split_offset, vdev_t *vd, uint64_t offset,
532     uint64_t size, void *arg)
533 {
534 	int n = 1;
535 	zio_t *zio = arg;
536 	indirect_vsd_t *iv = zio->io_vsd;
537 
538 	if (vd->v_read == vdev_indirect_read)
539 		return;
540 
541 	if (vd->v_read == vdev_mirror_read)
542 		n = vd->v_nchildren;
543 
544 	indirect_split_t *is =
545 	    malloc(offsetof(indirect_split_t, is_child[n]));
546 	if (is == NULL) {
547 		zio->io_error = ENOMEM;
548 		return;
549 	}
550 	bzero(is, offsetof(indirect_split_t, is_child[n]));
551 
552 	is->is_children = n;
553 	is->is_size = size;
554 	is->is_split_offset = split_offset;
555 	is->is_target_offset = offset;
556 	is->is_vdev = vd;
557 
558 	/*
559 	 * Note that we only consider multiple copies of the data for
560 	 * *mirror* vdevs.  We don't for "replacing" or "spare" vdevs, even
561 	 * though they use the same ops as mirror, because there's only one
562 	 * "good" copy under the replacing/spare.
563 	 */
564 	if (vd->v_read == vdev_mirror_read) {
565 		int i = 0;
566 		vdev_t *kid;
567 
568 		STAILQ_FOREACH(kid, &vd->v_children, v_childlink) {
569 			is->is_child[i++].ic_vdev = kid;
570 		}
571 	} else {
572 		is->is_child[0].ic_vdev = vd;
573 	}
574 
575 	list_insert_tail(&iv->iv_splits, is);
576 }
577 
578 static void
579 vdev_indirect_remap(vdev_t *vd, uint64_t offset, uint64_t asize, void *arg)
580 {
581 	list_t stack;
582 	spa_t *spa = vd->v_spa;
583 	zio_t *zio = arg;
584 	remap_segment_t *rs;
585 
586 	list_create(&stack, sizeof (remap_segment_t),
587 	    offsetof(remap_segment_t, rs_node));
588 
589 	rs = rs_alloc(vd, offset, asize, 0);
590 	if (rs == NULL) {
591 		printf("vdev_indirect_remap: out of memory.\n");
592 		zio->io_error = ENOMEM;
593 	}
594 	for (; rs != NULL; rs = list_remove_head(&stack)) {
595 		vdev_t *v = rs->rs_vd;
596 		uint64_t num_entries = 0;
597 		/* vdev_indirect_mapping_t *vim = v->v_mapping; */
598 		vdev_indirect_mapping_entry_phys_t *mapping =
599 		    vdev_indirect_mapping_duplicate_adjacent_entries(v,
600 		    rs->rs_offset, rs->rs_asize, &num_entries);
601 
602 		if (num_entries == 0)
603 			zio->io_error = ENOMEM;
604 
605 		for (uint64_t i = 0; i < num_entries; i++) {
606 			vdev_indirect_mapping_entry_phys_t *m = &mapping[i];
607 			uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
608 			uint64_t dst_offset = DVA_GET_OFFSET(&m->vimep_dst);
609 			uint64_t dst_vdev = DVA_GET_VDEV(&m->vimep_dst);
610 			uint64_t inner_offset = rs->rs_offset -
611 			    DVA_MAPPING_GET_SRC_OFFSET(m);
612 			uint64_t inner_size =
613 			    MIN(rs->rs_asize, size - inner_offset);
614 			vdev_t *dst_v = vdev_lookup_top(spa, dst_vdev);
615 
616 			if (dst_v->v_read == vdev_indirect_read) {
617 				remap_segment_t *o;
618 
619 				o = rs_alloc(dst_v, dst_offset + inner_offset,
620 				    inner_size, rs->rs_split_offset);
621 				if (o == NULL) {
622 					printf("vdev_indirect_remap: "
623 					    "out of memory.\n");
624 					zio->io_error = ENOMEM;
625 					break;
626 				}
627 
628 				list_insert_head(&stack, o);
629 			}
630 			vdev_indirect_gather_splits(rs->rs_split_offset, dst_v,
631 			    dst_offset + inner_offset,
632 			    inner_size, arg);
633 
634 			/*
635 			 * vdev_indirect_gather_splits can have memory
636 			 * allocation error, we can not recover from it.
637 			 */
638 			if (zio->io_error != 0)
639 				break;
640 			rs->rs_offset += inner_size;
641 			rs->rs_asize -= inner_size;
642 			rs->rs_split_offset += inner_size;
643 		}
644 
645 		free(mapping);
646 		free(rs);
647 		if (zio->io_error != 0)
648 			break;
649 	}
650 
651 	list_destroy(&stack);
652 }
653 
654 static void
655 vdev_indirect_map_free(zio_t *zio)
656 {
657 	indirect_vsd_t *iv = zio->io_vsd;
658 	indirect_split_t *is;
659 
660 	while ((is = list_head(&iv->iv_splits)) != NULL) {
661 		for (int c = 0; c < is->is_children; c++) {
662 			indirect_child_t *ic = &is->is_child[c];
663 			free(ic->ic_data);
664 		}
665 		list_remove(&iv->iv_splits, is);
666 		free(is);
667 	}
668 	free(iv);
669 }
670 
671 static int
672 vdev_indirect_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
673     off_t offset, size_t bytes)
674 {
675 	zio_t zio;
676 	spa_t *spa = vdev->v_spa;
677 	indirect_vsd_t *iv;
678 	indirect_split_t *first;
679 	int rc = EIO;
680 
681 	iv = calloc(1, sizeof(*iv));
682 	if (iv == NULL)
683 		return (ENOMEM);
684 
685 	list_create(&iv->iv_splits,
686 	    sizeof (indirect_split_t), offsetof(indirect_split_t, is_node));
687 
688 	bzero(&zio, sizeof(zio));
689 	zio.io_spa = spa;
690 	zio.io_bp = (blkptr_t *)bp;
691 	zio.io_data = buf;
692 	zio.io_size = bytes;
693 	zio.io_offset = offset;
694 	zio.io_vd = vdev;
695 	zio.io_vsd = iv;
696 
697 	if (vdev->v_mapping == NULL) {
698 		vdev_indirect_config_t *vic;
699 
700 		vic = &vdev->vdev_indirect_config;
701 		vdev->v_mapping = vdev_indirect_mapping_open(spa,
702 		    spa->spa_mos, vic->vic_mapping_object);
703 	}
704 
705 	vdev_indirect_remap(vdev, offset, bytes, &zio);
706 	if (zio.io_error != 0)
707 		return (zio.io_error);
708 
709 	first = list_head(&iv->iv_splits);
710 	if (first->is_size == zio.io_size) {
711 		/*
712 		 * This is not a split block; we are pointing to the entire
713 		 * data, which will checksum the same as the original data.
714 		 * Pass the BP down so that the child i/o can verify the
715 		 * checksum, and try a different location if available
716 		 * (e.g. on a mirror).
717 		 *
718 		 * While this special case could be handled the same as the
719 		 * general (split block) case, doing it this way ensures
720 		 * that the vast majority of blocks on indirect vdevs
721 		 * (which are not split) are handled identically to blocks
722 		 * on non-indirect vdevs.  This allows us to be less strict
723 		 * about performance in the general (but rare) case.
724 		 */
725 		rc = first->is_vdev->v_read(first->is_vdev, zio.io_bp,
726 		    zio.io_data, first->is_target_offset, bytes);
727 	} else {
728 		iv->iv_split_block = B_TRUE;
729 		/*
730 		 * Read one copy of each split segment, from the
731 		 * top-level vdev.  Since we don't know the
732 		 * checksum of each split individually, the child
733 		 * zio can't ensure that we get the right data.
734 		 * E.g. if it's a mirror, it will just read from a
735 		 * random (healthy) leaf vdev.  We have to verify
736 		 * the checksum in vdev_indirect_io_done().
737 		 */
738 		for (indirect_split_t *is = list_head(&iv->iv_splits);
739 		    is != NULL; is = list_next(&iv->iv_splits, is)) {
740 			char *ptr = zio.io_data;
741 
742 			rc = is->is_vdev->v_read(is->is_vdev, zio.io_bp,
743 			    ptr + is->is_split_offset, is->is_target_offset,
744 			    is->is_size);
745 		}
746 		if (zio_checksum_verify(spa, zio.io_bp, zio.io_data))
747 			rc = ECKSUM;
748 		else
749 			rc = 0;
750 	}
751 
752 	vdev_indirect_map_free(&zio);
753 	if (rc == 0)
754 		rc = zio.io_error;
755 
756 	return (rc);
757 }
758 
759 static int
760 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
761     off_t offset, size_t bytes)
762 {
763 
764 	return (vdev_read_phys(vdev, bp, buf,
765 	    offset + VDEV_LABEL_START_SIZE, bytes));
766 }
767 
768 static int
769 vdev_missing_read(vdev_t *vdev __unused, const blkptr_t *bp __unused,
770     void *buf __unused, off_t offset __unused, size_t bytes __unused)
771 {
772 
773 	return (ENOTSUP);
774 }
775 
776 static int
777 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
778     off_t offset, size_t bytes)
779 {
780 	vdev_t *kid;
781 	int rc;
782 
783 	rc = EIO;
784 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
785 		if (kid->v_state != VDEV_STATE_HEALTHY)
786 			continue;
787 		rc = kid->v_read(kid, bp, buf, offset, bytes);
788 		if (!rc)
789 			return (0);
790 	}
791 
792 	return (rc);
793 }
794 
795 static int
796 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
797     off_t offset, size_t bytes)
798 {
799 	vdev_t *kid;
800 
801 	/*
802 	 * Here we should have two kids:
803 	 * First one which is the one we are replacing and we can trust
804 	 * only this one to have valid data, but it might not be present.
805 	 * Second one is that one we are replacing with. It is most likely
806 	 * healthy, but we can't trust it has needed data, so we won't use it.
807 	 */
808 	kid = STAILQ_FIRST(&vdev->v_children);
809 	if (kid == NULL)
810 		return (EIO);
811 	if (kid->v_state != VDEV_STATE_HEALTHY)
812 		return (EIO);
813 	return (kid->v_read(kid, bp, buf, offset, bytes));
814 }
815 
816 static vdev_t *
817 vdev_find(uint64_t guid)
818 {
819 	vdev_t *vdev;
820 
821 	STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
822 		if (vdev->v_guid == guid)
823 			return (vdev);
824 
825 	return (0);
826 }
827 
828 static vdev_t *
829 vdev_create(uint64_t guid, vdev_read_t *_read)
830 {
831 	vdev_t *vdev;
832 	vdev_indirect_config_t *vic;
833 
834 	vdev = calloc(1, sizeof(vdev_t));
835 	if (vdev != NULL) {
836 		STAILQ_INIT(&vdev->v_children);
837 		vdev->v_guid = guid;
838 		vdev->v_read = _read;
839 
840 		/*
841 		 * root vdev has no read function, we use this fact to
842 		 * skip setting up data we do not need for root vdev.
843 		 * We only point root vdev from spa.
844 		 */
845 		if (_read != NULL) {
846 			vic = &vdev->vdev_indirect_config;
847 			vic->vic_prev_indirect_vdev = UINT64_MAX;
848 			STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
849 		}
850 	}
851 
852 	return (vdev);
853 }
854 
855 static void
856 vdev_set_initial_state(vdev_t *vdev, const nvlist_t *nvlist)
857 {
858 	uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
859 	uint64_t is_log;
860 
861 	is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
862 	is_log = 0;
863 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
864 	    &is_offline, NULL);
865 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
866 	    &is_removed, NULL);
867 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
868 	    &is_faulted, NULL);
869 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64,
870 	    NULL, &is_degraded, NULL);
871 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64,
872 	    NULL, &isnt_present, NULL);
873 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, NULL,
874 	    &is_log, NULL);
875 
876 	if (is_offline != 0)
877 		vdev->v_state = VDEV_STATE_OFFLINE;
878 	else if (is_removed != 0)
879 		vdev->v_state = VDEV_STATE_REMOVED;
880 	else if (is_faulted != 0)
881 		vdev->v_state = VDEV_STATE_FAULTED;
882 	else if (is_degraded != 0)
883 		vdev->v_state = VDEV_STATE_DEGRADED;
884 	else if (isnt_present != 0)
885 		vdev->v_state = VDEV_STATE_CANT_OPEN;
886 
887 	vdev->v_islog = is_log != 0;
888 }
889 
890 static int
891 vdev_init(uint64_t guid, const nvlist_t *nvlist, vdev_t **vdevp)
892 {
893 	uint64_t id, ashift, asize, nparity;
894 	const char *path;
895 	const char *type;
896 	int len, pathlen;
897 	char *name;
898 	vdev_t *vdev;
899 
900 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id,
901 	    NULL) ||
902 	    nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING, NULL,
903 	    &type, &len)) {
904 		return (ENOENT);
905 	}
906 
907 	if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 &&
908 	    memcmp(type, VDEV_TYPE_DISK, len) != 0 &&
909 #ifdef ZFS_TEST
910 	    memcmp(type, VDEV_TYPE_FILE, len) != 0 &&
911 #endif
912 	    memcmp(type, VDEV_TYPE_RAIDZ, len) != 0 &&
913 	    memcmp(type, VDEV_TYPE_INDIRECT, len) != 0 &&
914 	    memcmp(type, VDEV_TYPE_REPLACING, len) != 0 &&
915 	    memcmp(type, VDEV_TYPE_HOLE, len) != 0) {
916 		printf("ZFS: can only boot from disk, mirror, raidz1, "
917 		    "raidz2 and raidz3 vdevs, got: %.*s\n", len, type);
918 		return (EIO);
919 	}
920 
921 	if (memcmp(type, VDEV_TYPE_MIRROR, len) == 0)
922 		vdev = vdev_create(guid, vdev_mirror_read);
923 	else if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0)
924 		vdev = vdev_create(guid, vdev_raidz_read);
925 	else if (memcmp(type, VDEV_TYPE_REPLACING, len) == 0)
926 		vdev = vdev_create(guid, vdev_replacing_read);
927 	else if (memcmp(type, VDEV_TYPE_INDIRECT, len) == 0) {
928 		vdev_indirect_config_t *vic;
929 
930 		vdev = vdev_create(guid, vdev_indirect_read);
931 		if (vdev != NULL) {
932 			vdev->v_state = VDEV_STATE_HEALTHY;
933 			vic = &vdev->vdev_indirect_config;
934 
935 			nvlist_find(nvlist,
936 			    ZPOOL_CONFIG_INDIRECT_OBJECT,
937 			    DATA_TYPE_UINT64,
938 			    NULL, &vic->vic_mapping_object, NULL);
939 			nvlist_find(nvlist,
940 			    ZPOOL_CONFIG_INDIRECT_BIRTHS,
941 			    DATA_TYPE_UINT64,
942 			    NULL, &vic->vic_births_object, NULL);
943 			nvlist_find(nvlist,
944 			    ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
945 			    DATA_TYPE_UINT64,
946 			    NULL, &vic->vic_prev_indirect_vdev, NULL);
947 		}
948 	} else if (memcmp(type, VDEV_TYPE_HOLE, len) == 0) {
949 		vdev = vdev_create(guid, vdev_missing_read);
950 	} else {
951 		vdev = vdev_create(guid, vdev_disk_read);
952 	}
953 
954 	if (vdev == NULL)
955 		return (ENOMEM);
956 
957 	vdev_set_initial_state(vdev, nvlist);
958 	vdev->v_id = id;
959 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
960 	    DATA_TYPE_UINT64, NULL, &ashift, NULL) == 0)
961 		vdev->v_ashift = ashift;
962 
963 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE,
964 	    DATA_TYPE_UINT64, NULL, &asize, NULL) == 0) {
965 		vdev->v_psize = asize +
966 		    VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
967 	}
968 
969 	if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
970 	    DATA_TYPE_UINT64, NULL, &nparity, NULL) == 0)
971 		vdev->v_nparity = nparity;
972 
973 	if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
974 	    DATA_TYPE_STRING, NULL, &path, &pathlen) == 0) {
975 		char prefix[] = "/dev/";
976 
977 		len = strlen(prefix);
978 		if (len < pathlen && memcmp(path, prefix, len) == 0) {
979 			path += len;
980 			pathlen -= len;
981 		}
982 		name = malloc(pathlen + 1);
983 		bcopy(path, name, pathlen);
984 		name[pathlen] = '\0';
985 		vdev->v_name = name;
986 	} else {
987 		name = NULL;
988 		if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) {
989 			if (vdev->v_nparity < 1 ||
990 			    vdev->v_nparity > 3) {
991 				printf("ZFS: invalid raidz parity: %d\n",
992 				    vdev->v_nparity);
993 				return (EIO);
994 			}
995 			(void) asprintf(&name, "%.*s%d-%" PRIu64, len, type,
996 			    vdev->v_nparity, id);
997 		} else {
998 			(void) asprintf(&name, "%.*s-%" PRIu64, len, type, id);
999 		}
1000 		vdev->v_name = name;
1001 	}
1002 	*vdevp = vdev;
1003 	return (0);
1004 }
1005 
1006 /*
1007  * Find slot for vdev. We return either NULL to signal to use
1008  * STAILQ_INSERT_HEAD, or we return link element to be used with
1009  * STAILQ_INSERT_AFTER.
1010  */
1011 static vdev_t *
1012 vdev_find_previous(vdev_t *top_vdev, vdev_t *vdev)
1013 {
1014 	vdev_t *v, *previous;
1015 
1016 	if (STAILQ_EMPTY(&top_vdev->v_children))
1017 		return (NULL);
1018 
1019 	previous = NULL;
1020 	STAILQ_FOREACH(v, &top_vdev->v_children, v_childlink) {
1021 		if (v->v_id > vdev->v_id)
1022 			return (previous);
1023 
1024 		if (v->v_id == vdev->v_id)
1025 			return (v);
1026 
1027 		if (v->v_id < vdev->v_id)
1028 			previous = v;
1029 	}
1030 	return (previous);
1031 }
1032 
1033 static size_t
1034 vdev_child_count(vdev_t *vdev)
1035 {
1036 	vdev_t *v;
1037 	size_t count;
1038 
1039 	count = 0;
1040 	STAILQ_FOREACH(v, &vdev->v_children, v_childlink) {
1041 		count++;
1042 	}
1043 	return (count);
1044 }
1045 
1046 /*
1047  * Insert vdev into top_vdev children list. List is ordered by v_id.
1048  */
1049 static void
1050 vdev_insert(vdev_t *top_vdev, vdev_t *vdev)
1051 {
1052 	vdev_t *previous;
1053 	size_t count;
1054 
1055 	/*
1056 	 * The top level vdev can appear in random order, depending how
1057 	 * the firmware is presenting the disk devices.
1058 	 * However, we will insert vdev to create list ordered by v_id,
1059 	 * so we can use either STAILQ_INSERT_HEAD or STAILQ_INSERT_AFTER
1060 	 * as STAILQ does not have insert before.
1061 	 */
1062 	previous = vdev_find_previous(top_vdev, vdev);
1063 
1064 	if (previous == NULL) {
1065 		STAILQ_INSERT_HEAD(&top_vdev->v_children, vdev, v_childlink);
1066 	} else if (previous->v_id == vdev->v_id) {
1067 		/*
1068 		 * This vdev was configured from label config,
1069 		 * do not insert duplicate.
1070 		 */
1071 		return;
1072 	} else {
1073 		STAILQ_INSERT_AFTER(&top_vdev->v_children, previous, vdev,
1074 		    v_childlink);
1075 	}
1076 
1077 	count = vdev_child_count(top_vdev);
1078 	if (top_vdev->v_nchildren < count)
1079 		top_vdev->v_nchildren = count;
1080 }
1081 
1082 static int
1083 vdev_from_nvlist(spa_t *spa, uint64_t top_guid, const nvlist_t *nvlist)
1084 {
1085 	vdev_t *top_vdev, *vdev;
1086 	nvlist_t *kids = NULL;
1087 	int rc, nkids;
1088 
1089 	/* Get top vdev. */
1090 	top_vdev = vdev_find(top_guid);
1091 	if (top_vdev == NULL) {
1092 		rc = vdev_init(top_guid, nvlist, &top_vdev);
1093 		if (rc != 0)
1094 			return (rc);
1095 		top_vdev->v_spa = spa;
1096 		top_vdev->v_top = top_vdev;
1097 		vdev_insert(spa->spa_root_vdev, top_vdev);
1098 	}
1099 
1100 	/* Add children if there are any. */
1101 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1102 	    &nkids, &kids, NULL);
1103 	if (rc == 0) {
1104 		for (int i = 0; i < nkids; i++) {
1105 			uint64_t guid;
1106 
1107 			rc = nvlist_find(kids, ZPOOL_CONFIG_GUID,
1108 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1109 			if (rc != 0) {
1110 				nvlist_destroy(kids);
1111 				return (rc);
1112 			}
1113 			rc = vdev_init(guid, kids, &vdev);
1114 			if (rc != 0) {
1115 				nvlist_destroy(kids);
1116 				return (rc);
1117 			}
1118 
1119 			vdev->v_spa = spa;
1120 			vdev->v_top = top_vdev;
1121 			vdev_insert(top_vdev, vdev);
1122 
1123 			rc = nvlist_next(kids);
1124 			if (rc != 0) {
1125 				nvlist_destroy(kids);
1126 				return (rc);
1127 			}
1128 		}
1129 	} else {
1130 		/*
1131 		 * When there are no children, nvlist_find() does return
1132 		 * error, reset it because leaf devices have no children.
1133 		 */
1134 		rc = 0;
1135 	}
1136 	nvlist_destroy(kids);
1137 
1138 	return (rc);
1139 }
1140 
1141 static int
1142 vdev_init_from_label(spa_t *spa, const nvlist_t *nvlist)
1143 {
1144 	uint64_t pool_guid, top_guid;
1145 	nvlist_t *vdevs;
1146 	int rc;
1147 
1148 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1149 	    NULL, &pool_guid, NULL) ||
1150 	    nvlist_find(nvlist, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64,
1151 	    NULL, &top_guid, NULL) ||
1152 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1153 	    NULL, &vdevs, NULL)) {
1154 		printf("ZFS: can't find vdev details\n");
1155 		return (ENOENT);
1156 	}
1157 
1158 	rc = vdev_from_nvlist(spa, top_guid, vdevs);
1159 	nvlist_destroy(vdevs);
1160 	return (rc);
1161 }
1162 
1163 static void
1164 vdev_set_state(vdev_t *vdev)
1165 {
1166 	vdev_t *kid;
1167 	int good_kids;
1168 	int bad_kids;
1169 
1170 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1171 		vdev_set_state(kid);
1172 	}
1173 
1174 	/*
1175 	 * A mirror or raidz is healthy if all its kids are healthy. A
1176 	 * mirror is degraded if any of its kids is healthy; a raidz
1177 	 * is degraded if at most nparity kids are offline.
1178 	 */
1179 	if (STAILQ_FIRST(&vdev->v_children)) {
1180 		good_kids = 0;
1181 		bad_kids = 0;
1182 		STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1183 			if (kid->v_state == VDEV_STATE_HEALTHY)
1184 				good_kids++;
1185 			else
1186 				bad_kids++;
1187 		}
1188 		if (bad_kids == 0) {
1189 			vdev->v_state = VDEV_STATE_HEALTHY;
1190 		} else {
1191 			if (vdev->v_read == vdev_mirror_read) {
1192 				if (good_kids) {
1193 					vdev->v_state = VDEV_STATE_DEGRADED;
1194 				} else {
1195 					vdev->v_state = VDEV_STATE_OFFLINE;
1196 				}
1197 			} else if (vdev->v_read == vdev_raidz_read) {
1198 				if (bad_kids > vdev->v_nparity) {
1199 					vdev->v_state = VDEV_STATE_OFFLINE;
1200 				} else {
1201 					vdev->v_state = VDEV_STATE_DEGRADED;
1202 				}
1203 			}
1204 		}
1205 	}
1206 }
1207 
1208 static int
1209 vdev_update_from_nvlist(uint64_t top_guid, const nvlist_t *nvlist)
1210 {
1211 	vdev_t *vdev;
1212 	nvlist_t *kids = NULL;
1213 	int rc, nkids;
1214 
1215 	/* Update top vdev. */
1216 	vdev = vdev_find(top_guid);
1217 	if (vdev != NULL)
1218 		vdev_set_initial_state(vdev, nvlist);
1219 
1220 	/* Update children if there are any. */
1221 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1222 	    &nkids, &kids, NULL);
1223 	if (rc == 0) {
1224 		for (int i = 0; i < nkids; i++) {
1225 			uint64_t guid;
1226 
1227 			rc = nvlist_find(kids, ZPOOL_CONFIG_GUID,
1228 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1229 			if (rc != 0)
1230 				break;
1231 
1232 			vdev = vdev_find(guid);
1233 			if (vdev != NULL)
1234 				vdev_set_initial_state(vdev, kids);
1235 
1236 			rc = nvlist_next(kids);
1237 			if (rc != 0)
1238 				break;
1239 		}
1240 	} else {
1241 		rc = 0;
1242 	}
1243 	nvlist_destroy(kids);
1244 
1245 	return (rc);
1246 }
1247 
1248 static int
1249 vdev_init_from_nvlist(spa_t *spa, const nvlist_t *nvlist)
1250 {
1251 	uint64_t pool_guid, vdev_children;
1252 	nvlist_t *vdevs = NULL, *kids = NULL;
1253 	int rc, nkids;
1254 
1255 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1256 	    NULL, &pool_guid, NULL) ||
1257 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_CHILDREN, DATA_TYPE_UINT64,
1258 	    NULL, &vdev_children, NULL) ||
1259 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1260 	    NULL, &vdevs, NULL)) {
1261 		printf("ZFS: can't find vdev details\n");
1262 		return (ENOENT);
1263 	}
1264 
1265 	/* Wrong guid?! */
1266 	if (spa->spa_guid != pool_guid) {
1267 		nvlist_destroy(vdevs);
1268 		return (EINVAL);
1269 	}
1270 
1271 	spa->spa_root_vdev->v_nchildren = vdev_children;
1272 
1273 	rc = nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1274 	    &nkids, &kids, NULL);
1275 	nvlist_destroy(vdevs);
1276 
1277 	/*
1278 	 * MOS config has at least one child for root vdev.
1279 	 */
1280 	if (rc != 0)
1281 		return (rc);
1282 
1283 	for (int i = 0; i < nkids; i++) {
1284 		uint64_t guid;
1285 		vdev_t *vdev;
1286 
1287 		rc = nvlist_find(kids, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1288 		    NULL, &guid, NULL);
1289 		if (rc != 0)
1290 			break;
1291 		vdev = vdev_find(guid);
1292 		/*
1293 		 * Top level vdev is missing, create it.
1294 		 */
1295 		if (vdev == NULL)
1296 			rc = vdev_from_nvlist(spa, guid, kids);
1297 		else
1298 			rc = vdev_update_from_nvlist(guid, kids);
1299 		if (rc != 0)
1300 			break;
1301 		rc = nvlist_next(kids);
1302 		if (rc != 0)
1303 			break;
1304 	}
1305 	nvlist_destroy(kids);
1306 
1307 	/*
1308 	 * Re-evaluate top-level vdev state.
1309 	 */
1310 	vdev_set_state(spa->spa_root_vdev);
1311 
1312 	return (rc);
1313 }
1314 
1315 static spa_t *
1316 spa_find_by_guid(uint64_t guid)
1317 {
1318 	spa_t *spa;
1319 
1320 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1321 		if (spa->spa_guid == guid)
1322 			return (spa);
1323 
1324 	return (NULL);
1325 }
1326 
1327 static spa_t *
1328 spa_find_by_name(const char *name)
1329 {
1330 	spa_t *spa;
1331 
1332 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1333 		if (strcmp(spa->spa_name, name) == 0)
1334 			return (spa);
1335 
1336 	return (NULL);
1337 }
1338 
1339 static spa_t *
1340 spa_create(uint64_t guid, const char *name)
1341 {
1342 	spa_t *spa;
1343 
1344 	if ((spa = calloc(1, sizeof(spa_t))) == NULL)
1345 		return (NULL);
1346 	if ((spa->spa_name = strdup(name)) == NULL) {
1347 		free(spa);
1348 		return (NULL);
1349 	}
1350 	spa->spa_uberblock = &spa->spa_uberblock_master;
1351 	spa->spa_mos = &spa->spa_mos_master;
1352 	spa->spa_guid = guid;
1353 	spa->spa_root_vdev = vdev_create(guid, NULL);
1354 	if (spa->spa_root_vdev == NULL) {
1355 		free(spa->spa_name);
1356 		free(spa);
1357 		return (NULL);
1358 	}
1359 	spa->spa_root_vdev->v_name = strdup("root");
1360 	STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
1361 
1362 	return (spa);
1363 }
1364 
1365 static const char *
1366 state_name(vdev_state_t state)
1367 {
1368 	static const char *names[] = {
1369 		"UNKNOWN",
1370 		"CLOSED",
1371 		"OFFLINE",
1372 		"REMOVED",
1373 		"CANT_OPEN",
1374 		"FAULTED",
1375 		"DEGRADED",
1376 		"ONLINE"
1377 	};
1378 	return (names[state]);
1379 }
1380 
1381 #ifdef BOOT2
1382 
1383 #define pager_printf printf
1384 
1385 #else
1386 
1387 static int
1388 pager_printf(const char *fmt, ...)
1389 {
1390 	char line[80];
1391 	va_list args;
1392 
1393 	va_start(args, fmt);
1394 	vsnprintf(line, sizeof(line), fmt, args);
1395 	va_end(args);
1396 	return (pager_output(line));
1397 }
1398 
1399 #endif
1400 
1401 #define	STATUS_FORMAT	"        %s %s\n"
1402 
1403 static int
1404 print_state(int indent, const char *name, vdev_state_t state)
1405 {
1406 	int i;
1407 	char buf[512];
1408 
1409 	buf[0] = 0;
1410 	for (i = 0; i < indent; i++)
1411 		strcat(buf, "  ");
1412 	strcat(buf, name);
1413 	return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
1414 }
1415 
1416 static int
1417 vdev_status(vdev_t *vdev, int indent)
1418 {
1419 	vdev_t *kid;
1420 	int ret;
1421 
1422 	if (vdev->v_islog) {
1423 		(void) pager_output("        logs\n");
1424 		indent++;
1425 	}
1426 
1427 	ret = print_state(indent, vdev->v_name, vdev->v_state);
1428 	if (ret != 0)
1429 		return (ret);
1430 
1431 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1432 		ret = vdev_status(kid, indent + 1);
1433 		if (ret != 0)
1434 			return (ret);
1435 	}
1436 	return (ret);
1437 }
1438 
1439 static int
1440 spa_status(spa_t *spa)
1441 {
1442 	static char bootfs[ZFS_MAXNAMELEN];
1443 	uint64_t rootid;
1444 	vdev_list_t *vlist;
1445 	vdev_t *vdev;
1446 	int good_kids, bad_kids, degraded_kids, ret;
1447 	vdev_state_t state;
1448 
1449 	ret = pager_printf("  pool: %s\n", spa->spa_name);
1450 	if (ret != 0)
1451 		return (ret);
1452 
1453 	if (zfs_get_root(spa, &rootid) == 0 &&
1454 	    zfs_rlookup(spa, rootid, bootfs) == 0) {
1455 		if (bootfs[0] == '\0')
1456 			ret = pager_printf("bootfs: %s\n", spa->spa_name);
1457 		else
1458 			ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
1459 			    bootfs);
1460 		if (ret != 0)
1461 			return (ret);
1462 	}
1463 	ret = pager_printf("config:\n\n");
1464 	if (ret != 0)
1465 		return (ret);
1466 	ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
1467 	if (ret != 0)
1468 		return (ret);
1469 
1470 	good_kids = 0;
1471 	degraded_kids = 0;
1472 	bad_kids = 0;
1473 	vlist = &spa->spa_root_vdev->v_children;
1474 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1475 		if (vdev->v_state == VDEV_STATE_HEALTHY)
1476 			good_kids++;
1477 		else if (vdev->v_state == VDEV_STATE_DEGRADED)
1478 			degraded_kids++;
1479 		else
1480 			bad_kids++;
1481 	}
1482 
1483 	state = VDEV_STATE_CLOSED;
1484 	if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
1485 		state = VDEV_STATE_HEALTHY;
1486 	else if ((good_kids + degraded_kids) > 0)
1487 		state = VDEV_STATE_DEGRADED;
1488 
1489 	ret = print_state(0, spa->spa_name, state);
1490 	if (ret != 0)
1491 		return (ret);
1492 
1493 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1494 		ret = vdev_status(vdev, 1);
1495 		if (ret != 0)
1496 			return (ret);
1497 	}
1498 	return (ret);
1499 }
1500 
1501 static int
1502 spa_all_status(void)
1503 {
1504 	spa_t *spa;
1505 	int first = 1, ret = 0;
1506 
1507 	STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
1508 		if (!first) {
1509 			ret = pager_printf("\n");
1510 			if (ret != 0)
1511 				return (ret);
1512 		}
1513 		first = 0;
1514 		ret = spa_status(spa);
1515 		if (ret != 0)
1516 			return (ret);
1517 	}
1518 	return (ret);
1519 }
1520 
1521 static uint64_t
1522 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
1523 {
1524 	uint64_t label_offset;
1525 
1526 	if (l < VDEV_LABELS / 2)
1527 		label_offset = 0;
1528 	else
1529 		label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
1530 
1531 	return (offset + l * sizeof (vdev_label_t) + label_offset);
1532 }
1533 
1534 static int
1535 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1536 {
1537 	unsigned int seq1 = 0;
1538 	unsigned int seq2 = 0;
1539 	int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
1540 
1541 	if (cmp != 0)
1542 		return (cmp);
1543 
1544 	cmp = AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1545 	if (cmp != 0)
1546 		return (cmp);
1547 
1548 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1549 		seq1 = MMP_SEQ(ub1);
1550 
1551 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1552 		seq2 = MMP_SEQ(ub2);
1553 
1554 	return (AVL_CMP(seq1, seq2));
1555 }
1556 
1557 static int
1558 uberblock_verify(uberblock_t *ub)
1559 {
1560 	if (ub->ub_magic == BSWAP_64((uint64_t)UBERBLOCK_MAGIC)) {
1561 		byteswap_uint64_array(ub, sizeof (uberblock_t));
1562 	}
1563 
1564 	if (ub->ub_magic != UBERBLOCK_MAGIC ||
1565 	    !SPA_VERSION_IS_SUPPORTED(ub->ub_version))
1566 		return (EINVAL);
1567 
1568 	return (0);
1569 }
1570 
1571 static int
1572 vdev_label_read(vdev_t *vd, int l, void *buf, uint64_t offset,
1573     size_t size)
1574 {
1575 	blkptr_t bp;
1576 	off_t off;
1577 
1578 	off = vdev_label_offset(vd->v_psize, l, offset);
1579 
1580 	BP_ZERO(&bp);
1581 	BP_SET_LSIZE(&bp, size);
1582 	BP_SET_PSIZE(&bp, size);
1583 	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1584 	BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1585 	DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
1586 	ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1587 
1588 	return (vdev_read_phys(vd, &bp, buf, off, size));
1589 }
1590 
1591 static uint64_t
1592 vdev_get_label_asize(nvlist_t *nvl)
1593 {
1594 	nvlist_t *vdevs;
1595 	uint64_t asize;
1596 	const char *type;
1597 	int len;
1598 
1599 	asize = 0;
1600 	/* Get vdev tree */
1601 	if (nvlist_find(nvl, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1602 	    NULL, &vdevs, NULL) != 0)
1603 		return (asize);
1604 
1605 	/*
1606 	 * Get vdev type. We will calculate asize for raidz, mirror and disk.
1607 	 * For raidz, the asize is raw size of all children.
1608 	 */
1609 	if (nvlist_find(vdevs, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
1610 	    NULL, &type, &len) != 0)
1611 		goto done;
1612 
1613 	if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 &&
1614 	    memcmp(type, VDEV_TYPE_DISK, len) != 0 &&
1615 	    memcmp(type, VDEV_TYPE_RAIDZ, len) != 0)
1616 		goto done;
1617 
1618 	if (nvlist_find(vdevs, ZPOOL_CONFIG_ASIZE, DATA_TYPE_UINT64,
1619 	    NULL, &asize, NULL) != 0)
1620 		goto done;
1621 
1622 	if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) {
1623 		nvlist_t *kids;
1624 		int nkids;
1625 
1626 		if (nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN,
1627 		    DATA_TYPE_NVLIST_ARRAY, &nkids, &kids, NULL) != 0) {
1628 			asize = 0;
1629 			goto done;
1630 		}
1631 
1632 		asize /= nkids;
1633 		nvlist_destroy(kids);
1634 	}
1635 
1636 	asize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
1637 done:
1638 	nvlist_destroy(vdevs);
1639 	return (asize);
1640 }
1641 
1642 static nvlist_t *
1643 vdev_label_read_config(vdev_t *vd, uint64_t txg)
1644 {
1645 	vdev_phys_t *label;
1646 	uint64_t best_txg = 0;
1647 	uint64_t label_txg = 0;
1648 	uint64_t asize;
1649 	nvlist_t *nvl = NULL, *tmp;
1650 	int error;
1651 
1652 	label = malloc(sizeof (vdev_phys_t));
1653 	if (label == NULL)
1654 		return (NULL);
1655 
1656 	for (int l = 0; l < VDEV_LABELS; l++) {
1657 		const unsigned char *nvlist;
1658 
1659 		if (vdev_label_read(vd, l, label,
1660 		    offsetof(vdev_label_t, vl_vdev_phys),
1661 		    sizeof (vdev_phys_t)))
1662 			continue;
1663 
1664 		nvlist = (const unsigned char *) label->vp_nvlist;
1665 		tmp = nvlist_import(nvlist + 4, nvlist[0], nvlist[1]);
1666 		if (tmp == NULL)
1667 			continue;
1668 
1669 		error = nvlist_find(tmp, ZPOOL_CONFIG_POOL_TXG,
1670 		    DATA_TYPE_UINT64, NULL, &label_txg, NULL);
1671 		if (error != 0 || label_txg == 0) {
1672 			nvlist_destroy(nvl);
1673 			nvl = tmp;
1674 			goto done;
1675 		}
1676 
1677 		if (label_txg <= txg && label_txg > best_txg) {
1678 			best_txg = label_txg;
1679 			nvlist_destroy(nvl);
1680 			nvl = tmp;
1681 			tmp = NULL;
1682 
1683 			/*
1684 			 * Use asize from pool config. We need this
1685 			 * because we can get bad value from BIOS.
1686 			 */
1687 			asize = vdev_get_label_asize(nvl);
1688 			if (asize != 0) {
1689 				vd->v_psize = asize;
1690 			}
1691 		}
1692 		nvlist_destroy(tmp);
1693 	}
1694 
1695 	if (best_txg == 0) {
1696 		nvlist_destroy(nvl);
1697 		nvl = NULL;
1698 	}
1699 done:
1700 	free(label);
1701 	return (nvl);
1702 }
1703 
1704 static void
1705 vdev_uberblock_load(vdev_t *vd, uberblock_t *ub)
1706 {
1707 	uberblock_t *buf;
1708 
1709 	buf = malloc(VDEV_UBERBLOCK_SIZE(vd));
1710 	if (buf == NULL)
1711 		return;
1712 
1713 	for (int l = 0; l < VDEV_LABELS; l++) {
1714 		for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1715 			if (vdev_label_read(vd, l, buf,
1716 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1717 			    VDEV_UBERBLOCK_SIZE(vd)))
1718 				continue;
1719 			if (uberblock_verify(buf) != 0)
1720 				continue;
1721 
1722 			if (vdev_uberblock_compare(buf, ub) > 0)
1723 				*ub = *buf;
1724 		}
1725 	}
1726 	free(buf);
1727 }
1728 
1729 static int
1730 vdev_probe(vdev_phys_read_t *_read, void *read_priv, spa_t **spap)
1731 {
1732 	vdev_t vtmp;
1733 	spa_t *spa;
1734 	vdev_t *vdev;
1735 	nvlist_t *nvl;
1736 	uint64_t val;
1737 	uint64_t guid, vdev_children;
1738 	uint64_t pool_txg, pool_guid;
1739 	const char *pool_name;
1740 	int rc, namelen;
1741 
1742 	/*
1743 	 * Load the vdev label and figure out which
1744 	 * uberblock is most current.
1745 	 */
1746 	memset(&vtmp, 0, sizeof(vtmp));
1747 	vtmp.v_phys_read = _read;
1748 	vtmp.v_read_priv = read_priv;
1749 	vtmp.v_psize = P2ALIGN(ldi_get_size(read_priv),
1750 	    (uint64_t)sizeof (vdev_label_t));
1751 
1752 	/* Test for minimum device size. */
1753 	if (vtmp.v_psize < SPA_MINDEVSIZE)
1754 		return (EIO);
1755 
1756 	nvl = vdev_label_read_config(&vtmp, UINT64_MAX);
1757 	if (nvl == NULL)
1758 		return (EIO);
1759 
1760 	if (nvlist_find(nvl, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
1761 	    NULL, &val, NULL) != 0) {
1762 		nvlist_destroy(nvl);
1763 		return (EIO);
1764 	}
1765 
1766 	if (!SPA_VERSION_IS_SUPPORTED(val)) {
1767 		printf("ZFS: unsupported ZFS version %u (should be %u)\n",
1768 		    (unsigned)val, (unsigned)SPA_VERSION);
1769 		nvlist_destroy(nvl);
1770 		return (EIO);
1771 	}
1772 
1773 	/* Check ZFS features for read */
1774 	rc = nvlist_check_features_for_read(nvl);
1775 	if (rc != 0) {
1776 		nvlist_destroy(nvl);
1777 		return (EIO);
1778 	}
1779 
1780 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
1781 	    NULL, &val, NULL) != 0) {
1782 		nvlist_destroy(nvl);
1783 		return (EIO);
1784 	}
1785 
1786 	if (val == POOL_STATE_DESTROYED) {
1787 		/* We don't boot only from destroyed pools. */
1788 		nvlist_destroy(nvl);
1789 		return (EIO);
1790 	}
1791 
1792 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
1793 	    NULL, &pool_txg, NULL) != 0 ||
1794 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1795 	    NULL, &pool_guid, NULL) != 0 ||
1796 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
1797 	    NULL, &pool_name, &namelen) != 0) {
1798 		/*
1799 		 * Cache and spare devices end up here - just ignore
1800 		 * them.
1801 		 */
1802 		nvlist_destroy(nvl);
1803 		return (EIO);
1804 	}
1805 
1806 	/*
1807 	 * Create the pool if this is the first time we've seen it.
1808 	 */
1809 	spa = spa_find_by_guid(pool_guid);
1810 	if (spa == NULL) {
1811 		char *name;
1812 
1813 		nvlist_find(nvl, ZPOOL_CONFIG_VDEV_CHILDREN,
1814 		    DATA_TYPE_UINT64, NULL, &vdev_children, NULL);
1815 		name = malloc(namelen + 1);
1816 		if (name == NULL) {
1817 			nvlist_destroy(nvl);
1818 			return (ENOMEM);
1819 		}
1820 		bcopy(pool_name, name, namelen);
1821 		name[namelen] = '\0';
1822 		spa = spa_create(pool_guid, name);
1823 		free(name);
1824 		if (spa == NULL) {
1825 			nvlist_destroy(nvl);
1826 			return (ENOMEM);
1827 		}
1828 		spa->spa_root_vdev->v_nchildren = vdev_children;
1829 	}
1830 	if (pool_txg > spa->spa_txg)
1831 		spa->spa_txg = pool_txg;
1832 
1833 	/*
1834 	 * Get the vdev tree and create our in-core copy of it.
1835 	 * If we already have a vdev with this guid, this must
1836 	 * be some kind of alias (overlapping slices, dangerously dedicated
1837 	 * disks etc).
1838 	 */
1839 	if (nvlist_find(nvl, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1840 	    NULL, &guid, NULL) != 0) {
1841 		nvlist_destroy(nvl);
1842 		return (EIO);
1843 	}
1844 	vdev = vdev_find(guid);
1845 	/* Has this vdev already been inited? */
1846 	if (vdev && vdev->v_phys_read) {
1847 		nvlist_destroy(nvl);
1848 		return (EIO);
1849 	}
1850 
1851 	rc = vdev_init_from_label(spa, nvl);
1852 	nvlist_destroy(nvl);
1853 	if (rc != 0)
1854 		return (rc);
1855 
1856 	/*
1857 	 * We should already have created an incomplete vdev for this
1858 	 * vdev. Find it and initialise it with our read proc.
1859 	 */
1860 	vdev = vdev_find(guid);
1861 	if (vdev != NULL) {
1862 		vdev->v_phys_read = _read;
1863 		vdev->v_read_priv = read_priv;
1864 		vdev->v_psize = vtmp.v_psize;
1865 		/*
1866 		 * If no other state is set, mark vdev healthy.
1867 		 */
1868 		if (vdev->v_state == VDEV_STATE_UNKNOWN)
1869 			vdev->v_state = VDEV_STATE_HEALTHY;
1870 	} else {
1871 		printf("ZFS: inconsistent nvlist contents\n");
1872 		return (EIO);
1873 	}
1874 
1875 	if (vdev->v_islog)
1876 		spa->spa_with_log = vdev->v_islog;
1877 
1878 	/*
1879 	 * Re-evaluate top-level vdev state.
1880 	 */
1881 	vdev_set_state(vdev->v_top);
1882 
1883 	/*
1884 	 * Ok, we are happy with the pool so far. Lets find
1885 	 * the best uberblock and then we can actually access
1886 	 * the contents of the pool.
1887 	 */
1888 	vdev_uberblock_load(vdev, spa->spa_uberblock);
1889 
1890 	if (spap != NULL)
1891 		*spap = spa;
1892 	return (0);
1893 }
1894 
1895 static int
1896 ilog2(int n)
1897 {
1898 	int v;
1899 
1900 	for (v = 0; v < 32; v++)
1901 		if (n == (1 << v))
1902 			return (v);
1903 	return (-1);
1904 }
1905 
1906 static int
1907 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
1908 {
1909 	blkptr_t gbh_bp;
1910 	zio_gbh_phys_t zio_gb;
1911 	char *pbuf;
1912 	int i;
1913 
1914 	/* Artificial BP for gang block header. */
1915 	gbh_bp = *bp;
1916 	BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1917 	BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1918 	BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
1919 	BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
1920 	for (i = 0; i < SPA_DVAS_PER_BP; i++)
1921 		DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
1922 
1923 	/* Read gang header block using the artificial BP. */
1924 	if (zio_read(spa, &gbh_bp, &zio_gb))
1925 		return (EIO);
1926 
1927 	pbuf = buf;
1928 	for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1929 		blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1930 
1931 		if (BP_IS_HOLE(gbp))
1932 			continue;
1933 		if (zio_read(spa, gbp, pbuf))
1934 			return (EIO);
1935 		pbuf += BP_GET_PSIZE(gbp);
1936 	}
1937 
1938 	if (zio_checksum_verify(spa, bp, buf))
1939 		return (EIO);
1940 	return (0);
1941 }
1942 
1943 static int
1944 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1945 {
1946 	int cpfunc = BP_GET_COMPRESS(bp);
1947 	uint64_t align, size;
1948 	void *pbuf;
1949 	int i, error;
1950 
1951 	/*
1952 	 * Process data embedded in block pointer
1953 	 */
1954 	if (BP_IS_EMBEDDED(bp)) {
1955 		ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
1956 
1957 		size = BPE_GET_PSIZE(bp);
1958 		ASSERT(size <= BPE_PAYLOAD_SIZE);
1959 
1960 		if (cpfunc != ZIO_COMPRESS_OFF)
1961 			pbuf = malloc(size);
1962 		else
1963 			pbuf = buf;
1964 
1965 		if (pbuf == NULL)
1966 			return (ENOMEM);
1967 
1968 		decode_embedded_bp_compressed(bp, pbuf);
1969 		error = 0;
1970 
1971 		if (cpfunc != ZIO_COMPRESS_OFF) {
1972 			error = zio_decompress_data(cpfunc, pbuf,
1973 			    size, buf, BP_GET_LSIZE(bp));
1974 			free(pbuf);
1975 		}
1976 		if (error != 0)
1977 			printf("ZFS: i/o error - unable to decompress "
1978 			    "block pointer data, error %d\n", error);
1979 		return (error);
1980 	}
1981 
1982 	error = EIO;
1983 
1984 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1985 		const dva_t *dva = &bp->blk_dva[i];
1986 		vdev_t *vdev;
1987 		vdev_list_t *vlist;
1988 		uint64_t vdevid;
1989 		off_t offset;
1990 
1991 		if (!dva->dva_word[0] && !dva->dva_word[1])
1992 			continue;
1993 
1994 		vdevid = DVA_GET_VDEV(dva);
1995 		offset = DVA_GET_OFFSET(dva);
1996 		vlist = &spa->spa_root_vdev->v_children;
1997 		STAILQ_FOREACH(vdev, vlist, v_childlink) {
1998 			if (vdev->v_id == vdevid)
1999 				break;
2000 		}
2001 		if (!vdev || !vdev->v_read)
2002 			continue;
2003 
2004 		size = BP_GET_PSIZE(bp);
2005 		if (vdev->v_read == vdev_raidz_read) {
2006 			align = 1ULL << vdev->v_ashift;
2007 			if (P2PHASE(size, align) != 0)
2008 				size = P2ROUNDUP(size, align);
2009 		}
2010 		if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
2011 			pbuf = malloc(size);
2012 		else
2013 			pbuf = buf;
2014 
2015 		if (pbuf == NULL) {
2016 			error = ENOMEM;
2017 			break;
2018 		}
2019 
2020 		if (DVA_GET_GANG(dva))
2021 			error = zio_read_gang(spa, bp, pbuf);
2022 		else
2023 			error = vdev->v_read(vdev, bp, pbuf, offset, size);
2024 		if (error == 0) {
2025 			if (cpfunc != ZIO_COMPRESS_OFF)
2026 				error = zio_decompress_data(cpfunc, pbuf,
2027 				    BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
2028 			else if (size != BP_GET_PSIZE(bp))
2029 				bcopy(pbuf, buf, BP_GET_PSIZE(bp));
2030 		} else {
2031 			printf("zio_read error: %d\n", error);
2032 		}
2033 		if (buf != pbuf)
2034 			free(pbuf);
2035 		if (error == 0)
2036 			break;
2037 	}
2038 	if (error != 0)
2039 		printf("ZFS: i/o error - all block copies unavailable\n");
2040 
2041 	return (error);
2042 }
2043 
2044 static int
2045 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset,
2046     void *buf, size_t buflen)
2047 {
2048 	int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
2049 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2050 	int nlevels = dnode->dn_nlevels;
2051 	int i, rc;
2052 
2053 	if (bsize > SPA_MAXBLOCKSIZE) {
2054 		printf("ZFS: I/O error - blocks larger than %llu are not "
2055 		    "supported\n", SPA_MAXBLOCKSIZE);
2056 		return (EIO);
2057 	}
2058 
2059 	/*
2060 	 * Note: bsize may not be a power of two here so we need to do an
2061 	 * actual divide rather than a bitshift.
2062 	 */
2063 	while (buflen > 0) {
2064 		uint64_t bn = offset / bsize;
2065 		int boff = offset % bsize;
2066 		int ibn;
2067 		const blkptr_t *indbp;
2068 		blkptr_t bp;
2069 
2070 		if (bn > dnode->dn_maxblkid)
2071 			return (EIO);
2072 
2073 		if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
2074 			goto cached;
2075 
2076 		indbp = dnode->dn_blkptr;
2077 		for (i = 0; i < nlevels; i++) {
2078 			/*
2079 			 * Copy the bp from the indirect array so that
2080 			 * we can re-use the scratch buffer for multi-level
2081 			 * objects.
2082 			 */
2083 			ibn = bn >> ((nlevels - i - 1) * ibshift);
2084 			ibn &= ((1 << ibshift) - 1);
2085 			bp = indbp[ibn];
2086 			if (BP_IS_HOLE(&bp)) {
2087 				memset(dnode_cache_buf, 0, bsize);
2088 				break;
2089 			}
2090 			rc = zio_read(spa, &bp, dnode_cache_buf);
2091 			if (rc)
2092 				return (rc);
2093 			indbp = (const blkptr_t *) dnode_cache_buf;
2094 		}
2095 		dnode_cache_obj = dnode;
2096 		dnode_cache_bn = bn;
2097 	cached:
2098 
2099 		/*
2100 		 * The buffer contains our data block. Copy what we
2101 		 * need from it and loop.
2102 		 */
2103 		i = bsize - boff;
2104 		if (i > buflen) i = buflen;
2105 		memcpy(buf, &dnode_cache_buf[boff], i);
2106 		buf = ((char *)buf) + i;
2107 		offset += i;
2108 		buflen -= i;
2109 	}
2110 
2111 	return (0);
2112 }
2113 
2114 /*
2115  * Lookup a value in a microzap directory.
2116  */
2117 static int
2118 mzap_lookup(const mzap_phys_t *mz, size_t size, const char *name,
2119     uint64_t *value)
2120 {
2121 	const mzap_ent_phys_t *mze;
2122 	int chunks, i;
2123 
2124 	/*
2125 	 * Microzap objects use exactly one block. Read the whole
2126 	 * thing.
2127 	 */
2128 	chunks = size / MZAP_ENT_LEN - 1;
2129 	for (i = 0; i < chunks; i++) {
2130 		mze = &mz->mz_chunk[i];
2131 		if (strcmp(mze->mze_name, name) == 0) {
2132 			*value = mze->mze_value;
2133 			return (0);
2134 		}
2135 	}
2136 
2137 	return (ENOENT);
2138 }
2139 
2140 /*
2141  * Compare a name with a zap leaf entry. Return non-zero if the name
2142  * matches.
2143  */
2144 static int
2145 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2146     const char *name)
2147 {
2148 	size_t namelen;
2149 	const zap_leaf_chunk_t *nc;
2150 	const char *p;
2151 
2152 	namelen = zc->l_entry.le_name_numints;
2153 
2154 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2155 	p = name;
2156 	while (namelen > 0) {
2157 		size_t len;
2158 
2159 		len = namelen;
2160 		if (len > ZAP_LEAF_ARRAY_BYTES)
2161 			len = ZAP_LEAF_ARRAY_BYTES;
2162 		if (memcmp(p, nc->l_array.la_array, len))
2163 			return (0);
2164 		p += len;
2165 		namelen -= len;
2166 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2167 	}
2168 
2169 	return (1);
2170 }
2171 
2172 /*
2173  * Extract a uint64_t value from a zap leaf entry.
2174  */
2175 static uint64_t
2176 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
2177 {
2178 	const zap_leaf_chunk_t *vc;
2179 	int i;
2180 	uint64_t value;
2181 	const uint8_t *p;
2182 
2183 	vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
2184 	for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
2185 		value = (value << 8) | p[i];
2186 	}
2187 
2188 	return (value);
2189 }
2190 
2191 static void
2192 stv(int len, void *addr, uint64_t value)
2193 {
2194 	switch (len) {
2195 	case 1:
2196 		*(uint8_t *)addr = value;
2197 		return;
2198 	case 2:
2199 		*(uint16_t *)addr = value;
2200 		return;
2201 	case 4:
2202 		*(uint32_t *)addr = value;
2203 		return;
2204 	case 8:
2205 		*(uint64_t *)addr = value;
2206 		return;
2207 	}
2208 }
2209 
2210 /*
2211  * Extract a array from a zap leaf entry.
2212  */
2213 static void
2214 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2215     uint64_t integer_size, uint64_t num_integers, void *buf)
2216 {
2217 	uint64_t array_int_len = zc->l_entry.le_value_intlen;
2218 	uint64_t value = 0;
2219 	uint64_t *u64 = buf;
2220 	char *p = buf;
2221 	int len = MIN(zc->l_entry.le_value_numints, num_integers);
2222 	int chunk = zc->l_entry.le_value_chunk;
2223 	int byten = 0;
2224 
2225 	if (integer_size == 8 && len == 1) {
2226 		*u64 = fzap_leaf_value(zl, zc);
2227 		return;
2228 	}
2229 
2230 	while (len > 0) {
2231 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
2232 		int i;
2233 
2234 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
2235 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
2236 			value = (value << 8) | la->la_array[i];
2237 			byten++;
2238 			if (byten == array_int_len) {
2239 				stv(integer_size, p, value);
2240 				byten = 0;
2241 				len--;
2242 				if (len == 0)
2243 					return;
2244 				p += integer_size;
2245 			}
2246 		}
2247 		chunk = la->la_next;
2248 	}
2249 }
2250 
2251 static int
2252 fzap_check_size(uint64_t integer_size, uint64_t num_integers)
2253 {
2254 
2255 	switch (integer_size) {
2256 	case 1:
2257 	case 2:
2258 	case 4:
2259 	case 8:
2260 		break;
2261 	default:
2262 		return (EINVAL);
2263 	}
2264 
2265 	if (integer_size * num_integers > ZAP_MAXVALUELEN)
2266 		return (E2BIG);
2267 
2268 	return (0);
2269 }
2270 
2271 static void
2272 zap_leaf_free(zap_leaf_t *leaf)
2273 {
2274 	free(leaf->l_phys);
2275 	free(leaf);
2276 }
2277 
2278 static int
2279 zap_get_leaf_byblk(fat_zap_t *zap, uint64_t blk, zap_leaf_t **lp)
2280 {
2281 	int bs = FZAP_BLOCK_SHIFT(zap);
2282 	int err;
2283 
2284 	*lp = malloc(sizeof(**lp));
2285 	if (*lp == NULL)
2286 		return (ENOMEM);
2287 
2288 	(*lp)->l_bs = bs;
2289 	(*lp)->l_phys = malloc(1 << bs);
2290 
2291 	if ((*lp)->l_phys == NULL) {
2292 		free(*lp);
2293 		return (ENOMEM);
2294 	}
2295 	err = dnode_read(zap->zap_spa, zap->zap_dnode, blk << bs, (*lp)->l_phys,
2296 	    1 << bs);
2297 	if (err != 0) {
2298 		zap_leaf_free(*lp);
2299 	}
2300 	return (err);
2301 }
2302 
2303 static int
2304 zap_table_load(fat_zap_t *zap, zap_table_phys_t *tbl, uint64_t idx,
2305     uint64_t *valp)
2306 {
2307 	int bs = FZAP_BLOCK_SHIFT(zap);
2308 	uint64_t blk = idx >> (bs - 3);
2309 	uint64_t off = idx & ((1 << (bs - 3)) - 1);
2310 	uint64_t *buf;
2311 	int rc;
2312 
2313 	buf = malloc(1 << zap->zap_block_shift);
2314 	if (buf == NULL)
2315 		return (ENOMEM);
2316 	rc = dnode_read(zap->zap_spa, zap->zap_dnode, (tbl->zt_blk + blk) << bs,
2317 	    buf, 1 << zap->zap_block_shift);
2318 	if (rc == 0)
2319 		*valp = buf[off];
2320 	free(buf);
2321 	return (rc);
2322 }
2323 
2324 static int
2325 zap_idx_to_blk(fat_zap_t *zap, uint64_t idx, uint64_t *valp)
2326 {
2327 	if (zap->zap_phys->zap_ptrtbl.zt_numblks == 0) {
2328 		*valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
2329 		return (0);
2330 	} else {
2331 		return (zap_table_load(zap, &zap->zap_phys->zap_ptrtbl,
2332 		    idx, valp));
2333 	}
2334 }
2335 
2336 #define	ZAP_HASH_IDX(hash, n)	(((n) == 0) ? 0 : ((hash) >> (64 - (n))))
2337 static int
2338 zap_deref_leaf(fat_zap_t *zap, uint64_t h, zap_leaf_t **lp)
2339 {
2340 	uint64_t idx, blk;
2341 	int err;
2342 
2343 	idx = ZAP_HASH_IDX(h, zap->zap_phys->zap_ptrtbl.zt_shift);
2344 	err = zap_idx_to_blk(zap, idx, &blk);
2345 	if (err != 0)
2346 		return (err);
2347 	return (zap_get_leaf_byblk(zap, blk, lp));
2348 }
2349 
2350 #define	CHAIN_END	0xffff	/* end of the chunk chain */
2351 #define	LEAF_HASH(l, h) \
2352 	((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \
2353 	((h) >> \
2354 	(64 - ZAP_LEAF_HASH_SHIFT(l) - (l)->l_phys->l_hdr.lh_prefix_len)))
2355 #define	LEAF_HASH_ENTPTR(l, h)	(&(l)->l_phys->l_hash[LEAF_HASH(l, h)])
2356 
2357 static int
2358 zap_leaf_lookup(zap_leaf_t *zl, uint64_t hash, const char *name,
2359     uint64_t integer_size, uint64_t num_integers, void *value)
2360 {
2361 	int rc;
2362 	uint16_t *chunkp;
2363 	struct zap_leaf_entry *le;
2364 
2365 	/*
2366 	 * Make sure this chunk matches our hash.
2367 	 */
2368 	if (zl->l_phys->l_hdr.lh_prefix_len > 0 &&
2369 	    zl->l_phys->l_hdr.lh_prefix !=
2370 	    hash >> (64 - zl->l_phys->l_hdr.lh_prefix_len))
2371 		return (EIO);
2372 
2373 	rc = ENOENT;
2374 	for (chunkp = LEAF_HASH_ENTPTR(zl, hash);
2375 	    *chunkp != CHAIN_END; chunkp = &le->le_next) {
2376 		zap_leaf_chunk_t *zc;
2377 		uint16_t chunk = *chunkp;
2378 
2379 		le = ZAP_LEAF_ENTRY(zl, chunk);
2380 		if (le->le_hash != hash)
2381 			continue;
2382 		zc = &ZAP_LEAF_CHUNK(zl, chunk);
2383 		if (fzap_name_equal(zl, zc, name)) {
2384 			if (zc->l_entry.le_value_intlen > integer_size) {
2385 				rc = EINVAL;
2386 			} else {
2387 				fzap_leaf_array(zl, zc, integer_size,
2388 				    num_integers, value);
2389 				rc = 0;
2390 			}
2391 			break;
2392 		}
2393 	}
2394 	return (rc);
2395 }
2396 
2397 /*
2398  * Lookup a value in a fatzap directory.
2399  */
2400 static int
2401 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2402     const char *name, uint64_t integer_size, uint64_t num_integers,
2403     void *value)
2404 {
2405 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2406 	fat_zap_t z;
2407 	zap_leaf_t *zl;
2408 	uint64_t hash;
2409 	int rc;
2410 
2411 	if (zh->zap_magic != ZAP_MAGIC)
2412 		return (EIO);
2413 
2414 	if ((rc = fzap_check_size(integer_size, num_integers)) != 0) {
2415 		return (rc);
2416 	}
2417 
2418 	z.zap_block_shift = ilog2(bsize);
2419 	z.zap_phys = zh;
2420 	z.zap_spa = spa;
2421 	z.zap_dnode = dnode;
2422 
2423 	hash = zap_hash(zh->zap_salt, name);
2424 	rc = zap_deref_leaf(&z, hash, &zl);
2425 	if (rc != 0)
2426 		return (rc);
2427 
2428 	rc = zap_leaf_lookup(zl, hash, name, integer_size, num_integers, value);
2429 
2430 	zap_leaf_free(zl);
2431 	return (rc);
2432 }
2433 
2434 /*
2435  * Lookup a name in a zap object and return its value as a uint64_t.
2436  */
2437 static int
2438 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2439     uint64_t integer_size, uint64_t num_integers, void *value)
2440 {
2441 	int rc;
2442 	zap_phys_t *zap;
2443 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2444 
2445 	zap = malloc(size);
2446 	if (zap == NULL)
2447 		return (ENOMEM);
2448 
2449 	rc = dnode_read(spa, dnode, 0, zap, size);
2450 	if (rc)
2451 		goto done;
2452 
2453 	switch (zap->zap_block_type) {
2454 	case ZBT_MICRO:
2455 		rc = mzap_lookup((const mzap_phys_t *)zap, size, name, value);
2456 		break;
2457 	case ZBT_HEADER:
2458 		rc = fzap_lookup(spa, dnode, zap, name, integer_size,
2459 		    num_integers, value);
2460 		break;
2461 	default:
2462 		printf("ZFS: invalid zap_type=%" PRIx64 "\n",
2463 		    zap->zap_block_type);
2464 		rc = EIO;
2465 	}
2466 done:
2467 	free(zap);
2468 	return (rc);
2469 }
2470 
2471 /*
2472  * List a microzap directory.
2473  */
2474 static int
2475 mzap_list(const mzap_phys_t *mz, size_t size,
2476     int (*callback)(const char *, uint64_t))
2477 {
2478 	const mzap_ent_phys_t *mze;
2479 	int chunks, i, rc;
2480 
2481 	/*
2482 	 * Microzap objects use exactly one block. Read the whole
2483 	 * thing.
2484 	 */
2485 	rc = 0;
2486 	chunks = size / MZAP_ENT_LEN - 1;
2487 	for (i = 0; i < chunks; i++) {
2488 		mze = &mz->mz_chunk[i];
2489 		if (mze->mze_name[0]) {
2490 			rc = callback(mze->mze_name, mze->mze_value);
2491 			if (rc != 0)
2492 				break;
2493 		}
2494 	}
2495 
2496 	return (rc);
2497 }
2498 
2499 /*
2500  * List a fatzap directory.
2501  */
2502 static int
2503 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2504     int (*callback)(const char *, uint64_t))
2505 {
2506 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2507 	fat_zap_t z;
2508 	uint64_t i;
2509 	int j, rc;
2510 
2511 	if (zh->zap_magic != ZAP_MAGIC)
2512 		return (EIO);
2513 
2514 	z.zap_block_shift = ilog2(bsize);
2515 	z.zap_phys = zh;
2516 
2517 	/*
2518 	 * This assumes that the leaf blocks start at block 1. The
2519 	 * documentation isn't exactly clear on this.
2520 	 */
2521 	zap_leaf_t zl;
2522 	zl.l_bs = z.zap_block_shift;
2523 	zl.l_phys = malloc(bsize);
2524 	if (zl.l_phys == NULL)
2525 		return (ENOMEM);
2526 
2527 	for (i = 0; i < zh->zap_num_leafs; i++) {
2528 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2529 		char name[256], *p;
2530 		uint64_t value;
2531 
2532 		if (dnode_read(spa, dnode, off, zl.l_phys, bsize)) {
2533 			free(zl.l_phys);
2534 			return (EIO);
2535 		}
2536 
2537 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2538 			zap_leaf_chunk_t *zc, *nc;
2539 			int namelen;
2540 
2541 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2542 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2543 				continue;
2544 			namelen = zc->l_entry.le_name_numints;
2545 			if (namelen > sizeof(name))
2546 				namelen = sizeof(name);
2547 
2548 			/*
2549 			 * Paste the name back together.
2550 			 */
2551 			nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
2552 			p = name;
2553 			while (namelen > 0) {
2554 				int len;
2555 				len = namelen;
2556 				if (len > ZAP_LEAF_ARRAY_BYTES)
2557 					len = ZAP_LEAF_ARRAY_BYTES;
2558 				memcpy(p, nc->l_array.la_array, len);
2559 				p += len;
2560 				namelen -= len;
2561 				nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
2562 			}
2563 
2564 			/*
2565 			 * Assume the first eight bytes of the value are
2566 			 * a uint64_t.
2567 			 */
2568 			value = fzap_leaf_value(&zl, zc);
2569 
2570 			/* printf("%s 0x%jx\n", name, (uintmax_t)value); */
2571 			rc = callback((const char *)name, value);
2572 			if (rc != 0) {
2573 				free(zl.l_phys);
2574 				return (rc);
2575 			}
2576 		}
2577 	}
2578 
2579 	free(zl.l_phys);
2580 	return (0);
2581 }
2582 
2583 static int zfs_printf(const char *name, uint64_t value __unused)
2584 {
2585 
2586 	printf("%s\n", name);
2587 
2588 	return (0);
2589 }
2590 
2591 /*
2592  * List a zap directory.
2593  */
2594 static int
2595 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
2596 {
2597 	zap_phys_t *zap;
2598 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2599 	int rc;
2600 
2601 	zap = malloc(size);
2602 	if (zap == NULL)
2603 		return (ENOMEM);
2604 
2605 	rc = dnode_read(spa, dnode, 0, zap, size);
2606 	if (rc == 0) {
2607 		if (zap->zap_block_type == ZBT_MICRO)
2608 			rc = mzap_list((const mzap_phys_t *)zap, size,
2609 			    zfs_printf);
2610 		else
2611 			rc = fzap_list(spa, dnode, zap, zfs_printf);
2612 	}
2613 	free(zap);
2614 	return (rc);
2615 }
2616 
2617 static int
2618 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum,
2619     dnode_phys_t *dnode)
2620 {
2621 	off_t offset;
2622 
2623 	offset = objnum * sizeof(dnode_phys_t);
2624 	return dnode_read(spa, &os->os_meta_dnode, offset,
2625 		dnode, sizeof(dnode_phys_t));
2626 }
2627 
2628 /*
2629  * Lookup a name in a microzap directory.
2630  */
2631 static int
2632 mzap_rlookup(const mzap_phys_t *mz, size_t size, char *name, uint64_t value)
2633 {
2634 	const mzap_ent_phys_t *mze;
2635 	int chunks, i;
2636 
2637 	/*
2638 	 * Microzap objects use exactly one block. Read the whole
2639 	 * thing.
2640 	 */
2641 	chunks = size / MZAP_ENT_LEN - 1;
2642 	for (i = 0; i < chunks; i++) {
2643 		mze = &mz->mz_chunk[i];
2644 		if (value == mze->mze_value) {
2645 			strcpy(name, mze->mze_name);
2646 			return (0);
2647 		}
2648 	}
2649 
2650 	return (ENOENT);
2651 }
2652 
2653 static void
2654 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
2655 {
2656 	size_t namelen;
2657 	const zap_leaf_chunk_t *nc;
2658 	char *p;
2659 
2660 	namelen = zc->l_entry.le_name_numints;
2661 
2662 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2663 	p = name;
2664 	while (namelen > 0) {
2665 		size_t len;
2666 		len = namelen;
2667 		if (len > ZAP_LEAF_ARRAY_BYTES)
2668 			len = ZAP_LEAF_ARRAY_BYTES;
2669 		memcpy(p, nc->l_array.la_array, len);
2670 		p += len;
2671 		namelen -= len;
2672 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2673 	}
2674 
2675 	*p = '\0';
2676 }
2677 
2678 static int
2679 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2680     char *name, uint64_t value)
2681 {
2682 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2683 	fat_zap_t z;
2684 	uint64_t i;
2685 	int j, rc;
2686 
2687 	if (zh->zap_magic != ZAP_MAGIC)
2688 		return (EIO);
2689 
2690 	z.zap_block_shift = ilog2(bsize);
2691 	z.zap_phys = zh;
2692 
2693 	/*
2694 	 * This assumes that the leaf blocks start at block 1. The
2695 	 * documentation isn't exactly clear on this.
2696 	 */
2697 	zap_leaf_t zl;
2698 	zl.l_bs = z.zap_block_shift;
2699 	zl.l_phys = malloc(bsize);
2700 	if (zl.l_phys == NULL)
2701 		return (ENOMEM);
2702 
2703 	for (i = 0; i < zh->zap_num_leafs; i++) {
2704 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2705 
2706 		rc = dnode_read(spa, dnode, off, zl.l_phys, bsize);
2707 		if (rc != 0)
2708 			goto done;
2709 
2710 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2711 			zap_leaf_chunk_t *zc;
2712 
2713 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2714 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2715 				continue;
2716 			if (zc->l_entry.le_value_intlen != 8 ||
2717 			    zc->l_entry.le_value_numints != 1)
2718 				continue;
2719 
2720 			if (fzap_leaf_value(&zl, zc) == value) {
2721 				fzap_name_copy(&zl, zc, name);
2722 				goto done;
2723 			}
2724 		}
2725 	}
2726 
2727 	rc = ENOENT;
2728 done:
2729 	free(zl.l_phys);
2730 	return (rc);
2731 }
2732 
2733 static int
2734 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name,
2735     uint64_t value)
2736 {
2737 	zap_phys_t *zap;
2738 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2739 	int rc;
2740 
2741 	zap = malloc(size);
2742 	if (zap == NULL)
2743 		return (ENOMEM);
2744 
2745 	rc = dnode_read(spa, dnode, 0, zap, size);
2746 	if (rc == 0) {
2747 		if (zap->zap_block_type == ZBT_MICRO)
2748 			rc = mzap_rlookup((const mzap_phys_t *)zap, size,
2749 			    name, value);
2750 		else
2751 			rc = fzap_rlookup(spa, dnode, zap, name, value);
2752 	}
2753 	free(zap);
2754 	return (rc);
2755 }
2756 
2757 static int
2758 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
2759 {
2760 	char name[256];
2761 	char component[256];
2762 	uint64_t dir_obj, parent_obj, child_dir_zapobj;
2763 	dnode_phys_t child_dir_zap, dataset, dir, parent;
2764 	dsl_dir_phys_t *dd;
2765 	dsl_dataset_phys_t *ds;
2766 	char *p;
2767 	int len;
2768 
2769 	p = &name[sizeof(name) - 1];
2770 	*p = '\0';
2771 
2772 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
2773 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2774 		return (EIO);
2775 	}
2776 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2777 	dir_obj = ds->ds_dir_obj;
2778 
2779 	for (;;) {
2780 		if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir) != 0)
2781 			return (EIO);
2782 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2783 
2784 		/* Actual loop condition. */
2785 		parent_obj = dd->dd_parent_obj;
2786 		if (parent_obj == 0)
2787 			break;
2788 
2789 		if (objset_get_dnode(spa, spa->spa_mos, parent_obj,
2790 		    &parent) != 0)
2791 			return (EIO);
2792 		dd = (dsl_dir_phys_t *)&parent.dn_bonus;
2793 		child_dir_zapobj = dd->dd_child_dir_zapobj;
2794 		if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
2795 		    &child_dir_zap) != 0)
2796 			return (EIO);
2797 		if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
2798 			return (EIO);
2799 
2800 		len = strlen(component);
2801 		p -= len;
2802 		memcpy(p, component, len);
2803 		--p;
2804 		*p = '/';
2805 
2806 		/* Actual loop iteration. */
2807 		dir_obj = parent_obj;
2808 	}
2809 
2810 	if (*p != '\0')
2811 		++p;
2812 	strcpy(result, p);
2813 
2814 	return (0);
2815 }
2816 
2817 static int
2818 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
2819 {
2820 	char element[256];
2821 	uint64_t dir_obj, child_dir_zapobj;
2822 	dnode_phys_t child_dir_zap, dir;
2823 	dsl_dir_phys_t *dd;
2824 	const char *p, *q;
2825 
2826 	if (objset_get_dnode(spa, spa->spa_mos,
2827 	    DMU_POOL_DIRECTORY_OBJECT, &dir))
2828 		return (EIO);
2829 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
2830 	    1, &dir_obj))
2831 		return (EIO);
2832 
2833 	p = name;
2834 	for (;;) {
2835 		if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir))
2836 			return (EIO);
2837 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2838 
2839 		while (*p == '/')
2840 			p++;
2841 		/* Actual loop condition #1. */
2842 		if (*p == '\0')
2843 			break;
2844 
2845 		q = strchr(p, '/');
2846 		if (q) {
2847 			memcpy(element, p, q - p);
2848 			element[q - p] = '\0';
2849 			p = q + 1;
2850 		} else {
2851 			strcpy(element, p);
2852 			p += strlen(p);
2853 		}
2854 
2855 		child_dir_zapobj = dd->dd_child_dir_zapobj;
2856 		if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
2857 		    &child_dir_zap) != 0)
2858 			return (EIO);
2859 
2860 		/* Actual loop condition #2. */
2861 		if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
2862 		    1, &dir_obj) != 0)
2863 			return (ENOENT);
2864 	}
2865 
2866 	*objnum = dd->dd_head_dataset_obj;
2867 	return (0);
2868 }
2869 
2870 #ifndef BOOT2
2871 static int
2872 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
2873 {
2874 	uint64_t dir_obj, child_dir_zapobj;
2875 	dnode_phys_t child_dir_zap, dir, dataset;
2876 	dsl_dataset_phys_t *ds;
2877 	dsl_dir_phys_t *dd;
2878 
2879 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
2880 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2881 		return (EIO);
2882 	}
2883 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2884 	dir_obj = ds->ds_dir_obj;
2885 
2886 	if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir)) {
2887 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2888 		return (EIO);
2889 	}
2890 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2891 
2892 	child_dir_zapobj = dd->dd_child_dir_zapobj;
2893 	if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
2894 	    &child_dir_zap) != 0) {
2895 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2896 		return (EIO);
2897 	}
2898 
2899 	return (zap_list(spa, &child_dir_zap) != 0);
2900 }
2901 
2902 int
2903 zfs_callback_dataset(const spa_t *spa, uint64_t objnum,
2904     int (*callback)(const char *, uint64_t))
2905 {
2906 	uint64_t dir_obj, child_dir_zapobj;
2907 	dnode_phys_t child_dir_zap, dir, dataset;
2908 	dsl_dataset_phys_t *ds;
2909 	dsl_dir_phys_t *dd;
2910 	zap_phys_t *zap;
2911 	size_t size;
2912 	int err;
2913 
2914 	err = objset_get_dnode(spa, spa->spa_mos, objnum, &dataset);
2915 	if (err != 0) {
2916 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2917 		return (err);
2918 	}
2919 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2920 	dir_obj = ds->ds_dir_obj;
2921 
2922 	err = objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir);
2923 	if (err != 0) {
2924 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2925 		return (err);
2926 	}
2927 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2928 
2929 	child_dir_zapobj = dd->dd_child_dir_zapobj;
2930 	err = objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
2931 	    &child_dir_zap);
2932 	if (err != 0) {
2933 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2934 		return (err);
2935 	}
2936 
2937 	size = child_dir_zap.dn_datablkszsec << SPA_MINBLOCKSHIFT;
2938 	zap = malloc(size);
2939 	if (zap != NULL) {
2940 		err = dnode_read(spa, &child_dir_zap, 0, zap, size);
2941 		if (err != 0)
2942 			goto done;
2943 
2944 		if (zap->zap_block_type == ZBT_MICRO)
2945 			err = mzap_list((const mzap_phys_t *)zap, size,
2946 			    callback);
2947 		else
2948 			err = fzap_list(spa, &child_dir_zap, zap, callback);
2949 	} else {
2950 		err = ENOMEM;
2951 	}
2952 done:
2953 	free(zap);
2954 	return (err);
2955 }
2956 #endif
2957 
2958 /*
2959  * Find the object set given the object number of its dataset object
2960  * and return its details in *objset
2961  */
2962 static int
2963 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
2964 {
2965 	dnode_phys_t dataset;
2966 	dsl_dataset_phys_t *ds;
2967 
2968 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
2969 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2970 		return (EIO);
2971 	}
2972 
2973 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2974 	if (zio_read(spa, &ds->ds_bp, objset)) {
2975 		printf("ZFS: can't read object set for dataset %ju\n",
2976 		    (uintmax_t)objnum);
2977 		return (EIO);
2978 	}
2979 
2980 	return (0);
2981 }
2982 
2983 /*
2984  * Find the object set pointed to by the BOOTFS property or the root
2985  * dataset if there is none and return its details in *objset
2986  */
2987 static int
2988 zfs_get_root(const spa_t *spa, uint64_t *objid)
2989 {
2990 	dnode_phys_t dir, propdir;
2991 	uint64_t props, bootfs, root;
2992 
2993 	*objid = 0;
2994 
2995 	/*
2996 	 * Start with the MOS directory object.
2997 	 */
2998 	if (objset_get_dnode(spa, spa->spa_mos,
2999 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3000 		printf("ZFS: can't read MOS object directory\n");
3001 		return (EIO);
3002 	}
3003 
3004 	/*
3005 	 * Lookup the pool_props and see if we can find a bootfs.
3006 	 */
3007 	if (zap_lookup(spa, &dir, DMU_POOL_PROPS,
3008 	    sizeof(props), 1, &props) == 0 &&
3009 	    objset_get_dnode(spa, spa->spa_mos, props, &propdir) == 0 &&
3010 	    zap_lookup(spa, &propdir, "bootfs",
3011 	    sizeof(bootfs), 1, &bootfs) == 0 && bootfs != 0) {
3012 		*objid = bootfs;
3013 		return (0);
3014 	}
3015 	/*
3016 	 * Lookup the root dataset directory
3017 	 */
3018 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET,
3019 	    sizeof(root), 1, &root) ||
3020 	    objset_get_dnode(spa, spa->spa_mos, root, &dir)) {
3021 		printf("ZFS: can't find root dsl_dir\n");
3022 		return (EIO);
3023 	}
3024 
3025 	/*
3026 	 * Use the information from the dataset directory's bonus buffer
3027 	 * to find the dataset object and from that the object set itself.
3028 	 */
3029 	dsl_dir_phys_t *dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3030 	*objid = dd->dd_head_dataset_obj;
3031 	return (0);
3032 }
3033 
3034 static int
3035 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
3036 {
3037 
3038 	mount->spa = spa;
3039 
3040 	/*
3041 	 * Find the root object set if not explicitly provided
3042 	 */
3043 	if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
3044 		printf("ZFS: can't find root filesystem\n");
3045 		return (EIO);
3046 	}
3047 
3048 	if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
3049 		printf("ZFS: can't open root filesystem\n");
3050 		return (EIO);
3051 	}
3052 
3053 	mount->rootobj = rootobj;
3054 
3055 	return (0);
3056 }
3057 
3058 /*
3059  * callback function for feature name checks.
3060  */
3061 static int
3062 check_feature(const char *name, uint64_t value)
3063 {
3064 	int i;
3065 
3066 	if (value == 0)
3067 		return (0);
3068 	if (name[0] == '\0')
3069 		return (0);
3070 
3071 	for (i = 0; features_for_read[i] != NULL; i++) {
3072 		if (strcmp(name, features_for_read[i]) == 0)
3073 			return (0);
3074 	}
3075 	printf("ZFS: unsupported feature: %s\n", name);
3076 	return (EIO);
3077 }
3078 
3079 /*
3080  * Checks whether the MOS features that are active are supported.
3081  */
3082 static int
3083 check_mos_features(const spa_t *spa)
3084 {
3085 	dnode_phys_t dir;
3086 	zap_phys_t *zap;
3087 	uint64_t objnum;
3088 	size_t size;
3089 	int rc;
3090 
3091 	if ((rc = objset_get_dnode(spa, spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
3092 	    &dir)) != 0)
3093 		return (rc);
3094 	if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
3095 	    sizeof (objnum), 1, &objnum)) != 0) {
3096 		/*
3097 		 * It is older pool without features. As we have already
3098 		 * tested the label, just return without raising the error.
3099 		 */
3100 		return (0);
3101 	}
3102 
3103 	if ((rc = objset_get_dnode(spa, spa->spa_mos, objnum, &dir)) != 0)
3104 		return (rc);
3105 
3106 	if (dir.dn_type != DMU_OTN_ZAP_METADATA)
3107 		return (EIO);
3108 
3109 	size = dir.dn_datablkszsec << SPA_MINBLOCKSHIFT;
3110 	zap = malloc(size);
3111 	if (zap == NULL)
3112 		return (ENOMEM);
3113 
3114 	if (dnode_read(spa, &dir, 0, zap, size)) {
3115 		free(zap);
3116 		return (EIO);
3117 	}
3118 
3119 	if (zap->zap_block_type == ZBT_MICRO)
3120 		rc = mzap_list((const mzap_phys_t *)zap, size, check_feature);
3121 	else
3122 		rc = fzap_list(spa, &dir, zap, check_feature);
3123 
3124 	free(zap);
3125 	return (rc);
3126 }
3127 
3128 static int
3129 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
3130 {
3131 	dnode_phys_t dir;
3132 	size_t size;
3133 	int rc;
3134 	unsigned char *nv;
3135 
3136 	*value = NULL;
3137 	if ((rc = objset_get_dnode(spa, spa->spa_mos, obj, &dir)) != 0)
3138 		return (rc);
3139 	if (dir.dn_type != DMU_OT_PACKED_NVLIST &&
3140 	    dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) {
3141 		return (EIO);
3142 	}
3143 
3144 	if (dir.dn_bonuslen != sizeof (uint64_t))
3145 		return (EIO);
3146 
3147 	size = *(uint64_t *)DN_BONUS(&dir);
3148 	nv = malloc(size);
3149 	if (nv == NULL)
3150 		return (ENOMEM);
3151 
3152 	rc = dnode_read(spa, &dir, 0, nv, size);
3153 	if (rc != 0) {
3154 		free(nv);
3155 		nv = NULL;
3156 		return (rc);
3157 	}
3158 	*value = nvlist_import(nv + 4, nv[0], nv[1]);
3159 	free(nv);
3160 	return (rc);
3161 }
3162 
3163 static int
3164 zfs_spa_init(spa_t *spa)
3165 {
3166 	struct uberblock checkpoint;
3167 	dnode_phys_t dir;
3168 	uint64_t config_object;
3169 	nvlist_t *nvlist;
3170 	int rc;
3171 
3172 	if (zio_read(spa, &spa->spa_uberblock->ub_rootbp, spa->spa_mos)) {
3173 		printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
3174 		return (EIO);
3175 	}
3176 	if (spa->spa_mos->os_type != DMU_OST_META) {
3177 		printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
3178 		return (EIO);
3179 	}
3180 
3181 	if (objset_get_dnode(spa, &spa->spa_mos_master,
3182 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3183 		printf("ZFS: failed to read pool %s directory object\n",
3184 		    spa->spa_name);
3185 		return (EIO);
3186 	}
3187 	/* this is allowed to fail, older pools do not have salt */
3188 	rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
3189 	    sizeof (spa->spa_cksum_salt.zcs_bytes),
3190 	    spa->spa_cksum_salt.zcs_bytes);
3191 
3192 	rc = check_mos_features(spa);
3193 	if (rc != 0) {
3194 		printf("ZFS: pool %s is not supported\n", spa->spa_name);
3195 		return (rc);
3196 	}
3197 
3198 	rc = zap_lookup(spa, &dir, DMU_POOL_CONFIG,
3199 	    sizeof (config_object), 1, &config_object);
3200 	if (rc != 0) {
3201 		printf("ZFS: can not read MOS %s\n", DMU_POOL_CONFIG);
3202 		return (EIO);
3203 	}
3204 	rc = load_nvlist(spa, config_object, &nvlist);
3205 	if (rc != 0)
3206 		return (rc);
3207 
3208 	rc = zap_lookup(spa, &dir, DMU_POOL_ZPOOL_CHECKPOINT,
3209 	    sizeof(uint64_t), sizeof(checkpoint) / sizeof(uint64_t),
3210 	    &checkpoint);
3211 	if (rc == 0 && checkpoint.ub_checkpoint_txg != 0) {
3212 		memcpy(&spa->spa_uberblock_checkpoint, &checkpoint,
3213 		    sizeof(checkpoint));
3214 		if (zio_read(spa, &spa->spa_uberblock_checkpoint.ub_rootbp,
3215 		    &spa->spa_mos_checkpoint)) {
3216 			printf("ZFS: can not read checkpoint data.\n");
3217 			return (EIO);
3218 		}
3219 	}
3220 
3221 	/*
3222 	 * Update vdevs from MOS config. Note, we do skip encoding bytes
3223 	 * here. See also vdev_label_read_config().
3224 	 */
3225 	rc = vdev_init_from_nvlist(spa, nvlist);
3226 	nvlist_destroy(nvlist);
3227 	return (rc);
3228 }
3229 
3230 static int
3231 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
3232 {
3233 
3234 	if (dn->dn_bonustype != DMU_OT_SA) {
3235 		znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
3236 
3237 		sb->st_mode = zp->zp_mode;
3238 		sb->st_uid = zp->zp_uid;
3239 		sb->st_gid = zp->zp_gid;
3240 		sb->st_size = zp->zp_size;
3241 	} else {
3242 		sa_hdr_phys_t *sahdrp;
3243 		int hdrsize;
3244 		size_t size = 0;
3245 		void *buf = NULL;
3246 
3247 		if (dn->dn_bonuslen != 0)
3248 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3249 		else {
3250 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
3251 				blkptr_t *bp = DN_SPILL_BLKPTR(dn);
3252 				int error;
3253 
3254 				size = BP_GET_LSIZE(bp);
3255 				buf = malloc(size);
3256 				if (buf == NULL)
3257 					error = ENOMEM;
3258 				else
3259 					error = zio_read(spa, bp, buf);
3260 
3261 				if (error != 0) {
3262 					free(buf);
3263 					return (error);
3264 				}
3265 				sahdrp = buf;
3266 			} else {
3267 				return (EIO);
3268 			}
3269 		}
3270 		hdrsize = SA_HDR_SIZE(sahdrp);
3271 		sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
3272 		    SA_MODE_OFFSET);
3273 		sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
3274 		    SA_UID_OFFSET);
3275 		sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
3276 		    SA_GID_OFFSET);
3277 		sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
3278 		    SA_SIZE_OFFSET);
3279 		free(buf);
3280 	}
3281 
3282 	return (0);
3283 }
3284 
3285 static int
3286 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
3287 {
3288 	int rc = 0;
3289 
3290 	if (dn->dn_bonustype == DMU_OT_SA) {
3291 		sa_hdr_phys_t *sahdrp = NULL;
3292 		size_t size = 0;
3293 		void *buf = NULL;
3294 		int hdrsize;
3295 		char *p;
3296 
3297 		if (dn->dn_bonuslen != 0) {
3298 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3299 		} else {
3300 			blkptr_t *bp;
3301 
3302 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
3303 				return (EIO);
3304 			bp = DN_SPILL_BLKPTR(dn);
3305 
3306 			size = BP_GET_LSIZE(bp);
3307 			buf = malloc(size);
3308 			if (buf == NULL)
3309 				rc = ENOMEM;
3310 			else
3311 				rc = zio_read(spa, bp, buf);
3312 			if (rc != 0) {
3313 				free(buf);
3314 				return (rc);
3315 			}
3316 			sahdrp = buf;
3317 		}
3318 		hdrsize = SA_HDR_SIZE(sahdrp);
3319 		p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
3320 		memcpy(path, p, psize);
3321 		free(buf);
3322 		return (0);
3323 	}
3324 	/*
3325 	 * Second test is purely to silence bogus compiler
3326 	 * warning about accessing past the end of dn_bonus.
3327 	 */
3328 	if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
3329 	    sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
3330 		memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
3331 	} else {
3332 		rc = dnode_read(spa, dn, 0, path, psize);
3333 	}
3334 	return (rc);
3335 }
3336 
3337 struct obj_list {
3338 	uint64_t		objnum;
3339 	STAILQ_ENTRY(obj_list)	entry;
3340 };
3341 
3342 /*
3343  * Lookup a file and return its dnode.
3344  */
3345 static int
3346 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
3347 {
3348 	int rc;
3349 	uint64_t objnum;
3350 	const spa_t *spa;
3351 	dnode_phys_t dn;
3352 	const char *p, *q;
3353 	char element[256];
3354 	char path[1024];
3355 	int symlinks_followed = 0;
3356 	struct stat sb;
3357 	struct obj_list *entry, *tentry;
3358 	STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
3359 
3360 	spa = mount->spa;
3361 	if (mount->objset.os_type != DMU_OST_ZFS) {
3362 		printf("ZFS: unexpected object set type %ju\n",
3363 		    (uintmax_t)mount->objset.os_type);
3364 		return (EIO);
3365 	}
3366 
3367 	if ((entry = malloc(sizeof(struct obj_list))) == NULL)
3368 		return (ENOMEM);
3369 
3370 	/*
3371 	 * Get the root directory dnode.
3372 	 */
3373 	rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
3374 	if (rc) {
3375 		free(entry);
3376 		return (rc);
3377 	}
3378 
3379 	rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof(objnum), 1, &objnum);
3380 	if (rc) {
3381 		free(entry);
3382 		return (rc);
3383 	}
3384 	entry->objnum = objnum;
3385 	STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3386 
3387 	rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3388 	if (rc != 0)
3389 		goto done;
3390 
3391 	p = upath;
3392 	while (p && *p) {
3393 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3394 		if (rc != 0)
3395 			goto done;
3396 
3397 		while (*p == '/')
3398 			p++;
3399 		if (*p == '\0')
3400 			break;
3401 		q = p;
3402 		while (*q != '\0' && *q != '/')
3403 			q++;
3404 
3405 		/* skip dot */
3406 		if (p + 1 == q && p[0] == '.') {
3407 			p++;
3408 			continue;
3409 		}
3410 		/* double dot */
3411 		if (p + 2 == q && p[0] == '.' && p[1] == '.') {
3412 			p += 2;
3413 			if (STAILQ_FIRST(&on_cache) ==
3414 			    STAILQ_LAST(&on_cache, obj_list, entry)) {
3415 				rc = ENOENT;
3416 				goto done;
3417 			}
3418 			entry = STAILQ_FIRST(&on_cache);
3419 			STAILQ_REMOVE_HEAD(&on_cache, entry);
3420 			free(entry);
3421 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3422 			continue;
3423 		}
3424 		if (q - p + 1 > sizeof(element)) {
3425 			rc = ENAMETOOLONG;
3426 			goto done;
3427 		}
3428 		memcpy(element, p, q - p);
3429 		element[q - p] = 0;
3430 		p = q;
3431 
3432 		if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
3433 			goto done;
3434 		if (!S_ISDIR(sb.st_mode)) {
3435 			rc = ENOTDIR;
3436 			goto done;
3437 		}
3438 
3439 		rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum);
3440 		if (rc)
3441 			goto done;
3442 		objnum = ZFS_DIRENT_OBJ(objnum);
3443 
3444 		if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
3445 			rc = ENOMEM;
3446 			goto done;
3447 		}
3448 		entry->objnum = objnum;
3449 		STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3450 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3451 		if (rc)
3452 			goto done;
3453 
3454 		/*
3455 		 * Check for symlink.
3456 		 */
3457 		rc = zfs_dnode_stat(spa, &dn, &sb);
3458 		if (rc)
3459 			goto done;
3460 		if (S_ISLNK(sb.st_mode)) {
3461 			if (symlinks_followed > 10) {
3462 				rc = EMLINK;
3463 				goto done;
3464 			}
3465 			symlinks_followed++;
3466 
3467 			/*
3468 			 * Read the link value and copy the tail of our
3469 			 * current path onto the end.
3470 			 */
3471 			if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
3472 				rc = ENAMETOOLONG;
3473 				goto done;
3474 			}
3475 			strcpy(&path[sb.st_size], p);
3476 
3477 			rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
3478 			if (rc != 0)
3479 				goto done;
3480 
3481 			/*
3482 			 * Restart with the new path, starting either at
3483 			 * the root or at the parent depending whether or
3484 			 * not the link is relative.
3485 			 */
3486 			p = path;
3487 			if (*p == '/') {
3488 				while (STAILQ_FIRST(&on_cache) !=
3489 				    STAILQ_LAST(&on_cache, obj_list, entry)) {
3490 					entry = STAILQ_FIRST(&on_cache);
3491 					STAILQ_REMOVE_HEAD(&on_cache, entry);
3492 					free(entry);
3493 				}
3494 			} else {
3495 				entry = STAILQ_FIRST(&on_cache);
3496 				STAILQ_REMOVE_HEAD(&on_cache, entry);
3497 				free(entry);
3498 			}
3499 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3500 		}
3501 	}
3502 
3503 	*dnode = dn;
3504 done:
3505 	STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)
3506 		free(entry);
3507 	return (rc);
3508 }
3509