xref: /freebsd/stand/libsa/zfs/zfsimpl.c (revision 99db5849f7506e765c43f4e69a7105cc888e8d5e)
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 
769 static int
770 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
771     off_t offset, size_t bytes)
772 {
773 	vdev_t *kid;
774 	int rc;
775 
776 	rc = EIO;
777 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
778 		if (kid->v_state != VDEV_STATE_HEALTHY)
779 			continue;
780 		rc = kid->v_read(kid, bp, buf, offset, bytes);
781 		if (!rc)
782 			return (0);
783 	}
784 
785 	return (rc);
786 }
787 
788 static int
789 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
790     off_t offset, size_t bytes)
791 {
792 	vdev_t *kid;
793 
794 	/*
795 	 * Here we should have two kids:
796 	 * First one which is the one we are replacing and we can trust
797 	 * only this one to have valid data, but it might not be present.
798 	 * Second one is that one we are replacing with. It is most likely
799 	 * healthy, but we can't trust it has needed data, so we won't use it.
800 	 */
801 	kid = STAILQ_FIRST(&vdev->v_children);
802 	if (kid == NULL)
803 		return (EIO);
804 	if (kid->v_state != VDEV_STATE_HEALTHY)
805 		return (EIO);
806 	return (kid->v_read(kid, bp, buf, offset, bytes));
807 }
808 
809 static vdev_t *
810 vdev_find(uint64_t guid)
811 {
812 	vdev_t *vdev;
813 
814 	STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
815 		if (vdev->v_guid == guid)
816 			return (vdev);
817 
818 	return (0);
819 }
820 
821 static vdev_t *
822 vdev_create(uint64_t guid, vdev_read_t *_read)
823 {
824 	vdev_t *vdev;
825 	vdev_indirect_config_t *vic;
826 
827 	vdev = calloc(1, sizeof(vdev_t));
828 	if (vdev != NULL) {
829 		STAILQ_INIT(&vdev->v_children);
830 		vdev->v_guid = guid;
831 		vdev->v_read = _read;
832 
833 		/*
834 		 * root vdev has no read function, we use this fact to
835 		 * skip setting up data we do not need for root vdev.
836 		 * We only point root vdev from spa.
837 		 */
838 		if (_read != NULL) {
839 			vic = &vdev->vdev_indirect_config;
840 			vic->vic_prev_indirect_vdev = UINT64_MAX;
841 			STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
842 		}
843 	}
844 
845 	return (vdev);
846 }
847 
848 static void
849 vdev_set_initial_state(vdev_t *vdev, const nvlist_t *nvlist)
850 {
851 	uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
852 	uint64_t is_log;
853 
854 	is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
855 	is_log = 0;
856 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
857 	    &is_offline, NULL);
858 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
859 	    &is_removed, NULL);
860 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
861 	    &is_faulted, NULL);
862 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64,
863 	    NULL, &is_degraded, NULL);
864 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64,
865 	    NULL, &isnt_present, NULL);
866 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, NULL,
867 	    &is_log, NULL);
868 
869 	if (is_offline != 0)
870 		vdev->v_state = VDEV_STATE_OFFLINE;
871 	else if (is_removed != 0)
872 		vdev->v_state = VDEV_STATE_REMOVED;
873 	else if (is_faulted != 0)
874 		vdev->v_state = VDEV_STATE_FAULTED;
875 	else if (is_degraded != 0)
876 		vdev->v_state = VDEV_STATE_DEGRADED;
877 	else if (isnt_present != 0)
878 		vdev->v_state = VDEV_STATE_CANT_OPEN;
879 
880 	vdev->v_islog = is_log != 0;
881 }
882 
883 static int
884 vdev_init(uint64_t guid, const nvlist_t *nvlist, vdev_t **vdevp)
885 {
886 	uint64_t id, ashift, asize, nparity;
887 	const char *path;
888 	const char *type;
889 	int len, pathlen;
890 	char *name;
891 	vdev_t *vdev;
892 
893 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id,
894 	    NULL) ||
895 	    nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING, NULL,
896 	    &type, &len)) {
897 		return (ENOENT);
898 	}
899 
900 	if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 &&
901 	    memcmp(type, VDEV_TYPE_DISK, len) != 0 &&
902 #ifdef ZFS_TEST
903 	    memcmp(type, VDEV_TYPE_FILE, len) != 0 &&
904 #endif
905 	    memcmp(type, VDEV_TYPE_RAIDZ, len) != 0 &&
906 	    memcmp(type, VDEV_TYPE_INDIRECT, len) != 0 &&
907 	    memcmp(type, VDEV_TYPE_REPLACING, len) != 0) {
908 		printf("ZFS: can only boot from disk, mirror, raidz1, "
909 		    "raidz2 and raidz3 vdevs\n");
910 		return (EIO);
911 	}
912 
913 	if (memcmp(type, VDEV_TYPE_MIRROR, len) == 0)
914 		vdev = vdev_create(guid, vdev_mirror_read);
915 	else if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0)
916 		vdev = vdev_create(guid, vdev_raidz_read);
917 	else if (memcmp(type, VDEV_TYPE_REPLACING, len) == 0)
918 		vdev = vdev_create(guid, vdev_replacing_read);
919 	else if (memcmp(type, VDEV_TYPE_INDIRECT, len) == 0) {
920 		vdev_indirect_config_t *vic;
921 
922 		vdev = vdev_create(guid, vdev_indirect_read);
923 		if (vdev != NULL) {
924 			vdev->v_state = VDEV_STATE_HEALTHY;
925 			vic = &vdev->vdev_indirect_config;
926 
927 			nvlist_find(nvlist,
928 			    ZPOOL_CONFIG_INDIRECT_OBJECT,
929 			    DATA_TYPE_UINT64,
930 			    NULL, &vic->vic_mapping_object, NULL);
931 			nvlist_find(nvlist,
932 			    ZPOOL_CONFIG_INDIRECT_BIRTHS,
933 			    DATA_TYPE_UINT64,
934 			    NULL, &vic->vic_births_object, NULL);
935 			nvlist_find(nvlist,
936 			    ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
937 			    DATA_TYPE_UINT64,
938 			    NULL, &vic->vic_prev_indirect_vdev, NULL);
939 		}
940 	} else {
941 		vdev = vdev_create(guid, vdev_disk_read);
942 	}
943 
944 	if (vdev == NULL)
945 		return (ENOMEM);
946 
947 	vdev_set_initial_state(vdev, nvlist);
948 	vdev->v_id = id;
949 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
950 	    DATA_TYPE_UINT64, NULL, &ashift, NULL) == 0)
951 		vdev->v_ashift = ashift;
952 
953 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE,
954 	    DATA_TYPE_UINT64, NULL, &asize, NULL) == 0) {
955 		vdev->v_psize = asize +
956 		    VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
957 	}
958 
959 	if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
960 	    DATA_TYPE_UINT64, NULL, &nparity, NULL) == 0)
961 		vdev->v_nparity = nparity;
962 
963 	if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
964 	    DATA_TYPE_STRING, NULL, &path, &pathlen) == 0) {
965 		char prefix[] = "/dev/";
966 
967 		len = strlen(prefix);
968 		if (len < pathlen && memcmp(path, prefix, len) == 0) {
969 			path += len;
970 			pathlen -= len;
971 		}
972 		name = malloc(pathlen + 1);
973 		bcopy(path, name, pathlen);
974 		name[pathlen] = '\0';
975 		vdev->v_name = name;
976 	} else {
977 		name = NULL;
978 		if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) {
979 			if (vdev->v_nparity < 1 ||
980 			    vdev->v_nparity > 3) {
981 				printf("ZFS: invalid raidz parity: %d\n",
982 				    vdev->v_nparity);
983 				return (EIO);
984 			}
985 			(void) asprintf(&name, "%.*s%d-%" PRIu64, len, type,
986 			    vdev->v_nparity, id);
987 		} else {
988 			(void) asprintf(&name, "%.*s-%" PRIu64, len, type, id);
989 		}
990 		vdev->v_name = name;
991 	}
992 	*vdevp = vdev;
993 	return (0);
994 }
995 
996 /*
997  * Find slot for vdev. We return either NULL to signal to use
998  * STAILQ_INSERT_HEAD, or we return link element to be used with
999  * STAILQ_INSERT_AFTER.
1000  */
1001 static vdev_t *
1002 vdev_find_previous(vdev_t *top_vdev, vdev_t *vdev)
1003 {
1004 	vdev_t *v, *previous;
1005 
1006 	if (STAILQ_EMPTY(&top_vdev->v_children))
1007 		return (NULL);
1008 
1009 	previous = NULL;
1010 	STAILQ_FOREACH(v, &top_vdev->v_children, v_childlink) {
1011 		if (v->v_id > vdev->v_id)
1012 			return (previous);
1013 
1014 		if (v->v_id == vdev->v_id)
1015 			return (v);
1016 
1017 		if (v->v_id < vdev->v_id)
1018 			previous = v;
1019 	}
1020 	return (previous);
1021 }
1022 
1023 static size_t
1024 vdev_child_count(vdev_t *vdev)
1025 {
1026 	vdev_t *v;
1027 	size_t count;
1028 
1029 	count = 0;
1030 	STAILQ_FOREACH(v, &vdev->v_children, v_childlink) {
1031 		count++;
1032 	}
1033 	return (count);
1034 }
1035 
1036 /*
1037  * Insert vdev into top_vdev children list. List is ordered by v_id.
1038  */
1039 static void
1040 vdev_insert(vdev_t *top_vdev, vdev_t *vdev)
1041 {
1042 	vdev_t *previous;
1043 	size_t count;
1044 
1045 	/*
1046 	 * The top level vdev can appear in random order, depending how
1047 	 * the firmware is presenting the disk devices.
1048 	 * However, we will insert vdev to create list ordered by v_id,
1049 	 * so we can use either STAILQ_INSERT_HEAD or STAILQ_INSERT_AFTER
1050 	 * as STAILQ does not have insert before.
1051 	 */
1052 	previous = vdev_find_previous(top_vdev, vdev);
1053 
1054 	if (previous == NULL) {
1055 		STAILQ_INSERT_HEAD(&top_vdev->v_children, vdev, v_childlink);
1056 	} else if (previous->v_id == vdev->v_id) {
1057 		/*
1058 		 * This vdev was configured from label config,
1059 		 * do not insert duplicate.
1060 		 */
1061 		return;
1062 	} else {
1063 		STAILQ_INSERT_AFTER(&top_vdev->v_children, previous, vdev,
1064 		    v_childlink);
1065 	}
1066 
1067 	count = vdev_child_count(top_vdev);
1068 	if (top_vdev->v_nchildren < count)
1069 		top_vdev->v_nchildren = count;
1070 }
1071 
1072 static int
1073 vdev_from_nvlist(spa_t *spa, uint64_t top_guid, const nvlist_t *nvlist)
1074 {
1075 	vdev_t *top_vdev, *vdev;
1076 	nvlist_t *kids = NULL;
1077 	int rc, nkids;
1078 
1079 	/* Get top vdev. */
1080 	top_vdev = vdev_find(top_guid);
1081 	if (top_vdev == NULL) {
1082 		rc = vdev_init(top_guid, nvlist, &top_vdev);
1083 		if (rc != 0)
1084 			return (rc);
1085 		top_vdev->v_spa = spa;
1086 		top_vdev->v_top = top_vdev;
1087 		vdev_insert(spa->spa_root_vdev, top_vdev);
1088 	}
1089 
1090 	/* Add children if there are any. */
1091 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1092 	    &nkids, &kids, NULL);
1093 	if (rc == 0) {
1094 		for (int i = 0; i < nkids; i++) {
1095 			uint64_t guid;
1096 
1097 			rc = nvlist_find(kids, ZPOOL_CONFIG_GUID,
1098 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1099 			if (rc != 0) {
1100 				nvlist_destroy(kids);
1101 				return (rc);
1102 			}
1103 			rc = vdev_init(guid, kids, &vdev);
1104 			if (rc != 0)
1105 				return (rc);
1106 
1107 			vdev->v_spa = spa;
1108 			vdev->v_top = top_vdev;
1109 			vdev_insert(top_vdev, vdev);
1110 
1111 			rc = nvlist_next(kids);
1112 		}
1113 	} else {
1114 		/*
1115 		 * When there are no children, nvlist_find() does return
1116 		 * error, reset it because leaf devices have no children.
1117 		 */
1118 		rc = 0;
1119 	}
1120 	nvlist_destroy(kids);
1121 
1122 	return (rc);
1123 }
1124 
1125 static int
1126 vdev_init_from_label(spa_t *spa, const nvlist_t *nvlist)
1127 {
1128 	uint64_t pool_guid, top_guid;
1129 	nvlist_t *vdevs;
1130 	int rc;
1131 
1132 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1133 	    NULL, &pool_guid, NULL) ||
1134 	    nvlist_find(nvlist, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64,
1135 	    NULL, &top_guid, NULL) ||
1136 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1137 	    NULL, &vdevs, NULL)) {
1138 		printf("ZFS: can't find vdev details\n");
1139 		return (ENOENT);
1140 	}
1141 
1142 	rc = vdev_from_nvlist(spa, top_guid, vdevs);
1143 	nvlist_destroy(vdevs);
1144 	return (rc);
1145 }
1146 
1147 static void
1148 vdev_set_state(vdev_t *vdev)
1149 {
1150 	vdev_t *kid;
1151 	int good_kids;
1152 	int bad_kids;
1153 
1154 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1155 		vdev_set_state(kid);
1156 	}
1157 
1158 	/*
1159 	 * A mirror or raidz is healthy if all its kids are healthy. A
1160 	 * mirror is degraded if any of its kids is healthy; a raidz
1161 	 * is degraded if at most nparity kids are offline.
1162 	 */
1163 	if (STAILQ_FIRST(&vdev->v_children)) {
1164 		good_kids = 0;
1165 		bad_kids = 0;
1166 		STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1167 			if (kid->v_state == VDEV_STATE_HEALTHY)
1168 				good_kids++;
1169 			else
1170 				bad_kids++;
1171 		}
1172 		if (bad_kids == 0) {
1173 			vdev->v_state = VDEV_STATE_HEALTHY;
1174 		} else {
1175 			if (vdev->v_read == vdev_mirror_read) {
1176 				if (good_kids) {
1177 					vdev->v_state = VDEV_STATE_DEGRADED;
1178 				} else {
1179 					vdev->v_state = VDEV_STATE_OFFLINE;
1180 				}
1181 			} else if (vdev->v_read == vdev_raidz_read) {
1182 				if (bad_kids > vdev->v_nparity) {
1183 					vdev->v_state = VDEV_STATE_OFFLINE;
1184 				} else {
1185 					vdev->v_state = VDEV_STATE_DEGRADED;
1186 				}
1187 			}
1188 		}
1189 	}
1190 }
1191 
1192 static int
1193 vdev_update_from_nvlist(uint64_t top_guid, const nvlist_t *nvlist)
1194 {
1195 	vdev_t *vdev;
1196 	nvlist_t *kids = NULL;
1197 	int rc, nkids;
1198 
1199 	/* Update top vdev. */
1200 	vdev = vdev_find(top_guid);
1201 	if (vdev != NULL)
1202 		vdev_set_initial_state(vdev, nvlist);
1203 
1204 	/* Update children if there are any. */
1205 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1206 	    &nkids, &kids, NULL);
1207 	if (rc == 0) {
1208 		for (int i = 0; i < nkids; i++) {
1209 			uint64_t guid;
1210 
1211 			rc = nvlist_find(kids, ZPOOL_CONFIG_GUID,
1212 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1213 			if (rc != 0)
1214 				break;
1215 
1216 			vdev = vdev_find(guid);
1217 			if (vdev != NULL)
1218 				vdev_set_initial_state(vdev, kids);
1219 
1220 			rc = nvlist_next(kids);
1221 		}
1222 	} else {
1223 		rc = 0;
1224 	}
1225 	nvlist_destroy(kids);
1226 
1227 	return (rc);
1228 }
1229 
1230 static int
1231 vdev_init_from_nvlist(spa_t *spa, const nvlist_t *nvlist)
1232 {
1233 	uint64_t pool_guid, vdev_children;
1234 	nvlist_t *vdevs = NULL, *kids = NULL;
1235 	int rc, nkids;
1236 
1237 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1238 	    NULL, &pool_guid, NULL) ||
1239 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_CHILDREN, DATA_TYPE_UINT64,
1240 	    NULL, &vdev_children, NULL) ||
1241 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1242 	    NULL, &vdevs, NULL)) {
1243 		printf("ZFS: can't find vdev details\n");
1244 		return (ENOENT);
1245 	}
1246 
1247 	/* Wrong guid?! */
1248 	if (spa->spa_guid != pool_guid) {
1249 		nvlist_destroy(vdevs);
1250 		return (EINVAL);
1251 	}
1252 
1253 	spa->spa_root_vdev->v_nchildren = vdev_children;
1254 
1255 	rc = nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1256 	    &nkids, &kids, NULL);
1257 	nvlist_destroy(vdevs);
1258 
1259 	/*
1260 	 * MOS config has at least one child for root vdev.
1261 	 */
1262 	if (rc != 0)
1263 		return (rc);
1264 
1265 	for (int i = 0; i < nkids; i++) {
1266 		uint64_t guid;
1267 		vdev_t *vdev;
1268 
1269 		rc = nvlist_find(kids, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1270 		    NULL, &guid, NULL);
1271 		if (rc != 0)
1272 			break;
1273 		vdev = vdev_find(guid);
1274 		/*
1275 		 * Top level vdev is missing, create it.
1276 		 */
1277 		if (vdev == NULL)
1278 			rc = vdev_from_nvlist(spa, guid, kids);
1279 		else
1280 			rc = vdev_update_from_nvlist(guid, kids);
1281 		if (rc != 0)
1282 			break;
1283 		nvlist_next(kids);
1284 	}
1285 	nvlist_destroy(kids);
1286 
1287 	/*
1288 	 * Re-evaluate top-level vdev state.
1289 	 */
1290 	vdev_set_state(spa->spa_root_vdev);
1291 
1292 	return (rc);
1293 }
1294 
1295 static spa_t *
1296 spa_find_by_guid(uint64_t guid)
1297 {
1298 	spa_t *spa;
1299 
1300 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1301 		if (spa->spa_guid == guid)
1302 			return (spa);
1303 
1304 	return (NULL);
1305 }
1306 
1307 static spa_t *
1308 spa_find_by_name(const char *name)
1309 {
1310 	spa_t *spa;
1311 
1312 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1313 		if (strcmp(spa->spa_name, name) == 0)
1314 			return (spa);
1315 
1316 	return (NULL);
1317 }
1318 
1319 #ifdef BOOT2
1320 static spa_t *
1321 spa_get_primary(void)
1322 {
1323 
1324 	return (STAILQ_FIRST(&zfs_pools));
1325 }
1326 
1327 static vdev_t *
1328 spa_get_primary_vdev(const spa_t *spa)
1329 {
1330 	vdev_t *vdev;
1331 	vdev_t *kid;
1332 
1333 	if (spa == NULL)
1334 		spa = spa_get_primary();
1335 	if (spa == NULL)
1336 		return (NULL);
1337 	vdev = spa->spa_root_vdev;
1338 	if (vdev == NULL)
1339 		return (NULL);
1340 	for (kid = STAILQ_FIRST(&vdev->v_children); kid != NULL;
1341 	    kid = STAILQ_FIRST(&vdev->v_children))
1342 		vdev = kid;
1343 	return (vdev);
1344 }
1345 #endif
1346 
1347 static spa_t *
1348 spa_create(uint64_t guid, const char *name)
1349 {
1350 	spa_t *spa;
1351 
1352 	if ((spa = calloc(1, sizeof(spa_t))) == NULL)
1353 		return (NULL);
1354 	if ((spa->spa_name = strdup(name)) == NULL) {
1355 		free(spa);
1356 		return (NULL);
1357 	}
1358 	spa->spa_guid = guid;
1359 	spa->spa_root_vdev = vdev_create(guid, NULL);
1360 	if (spa->spa_root_vdev == NULL) {
1361 		free(spa->spa_name);
1362 		free(spa);
1363 		return (NULL);
1364 	}
1365 	spa->spa_root_vdev->v_name = strdup("root");
1366 	STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
1367 
1368 	return (spa);
1369 }
1370 
1371 static const char *
1372 state_name(vdev_state_t state)
1373 {
1374 	static const char *names[] = {
1375 		"UNKNOWN",
1376 		"CLOSED",
1377 		"OFFLINE",
1378 		"REMOVED",
1379 		"CANT_OPEN",
1380 		"FAULTED",
1381 		"DEGRADED",
1382 		"ONLINE"
1383 	};
1384 	return (names[state]);
1385 }
1386 
1387 #ifdef BOOT2
1388 
1389 #define pager_printf printf
1390 
1391 #else
1392 
1393 static int
1394 pager_printf(const char *fmt, ...)
1395 {
1396 	char line[80];
1397 	va_list args;
1398 
1399 	va_start(args, fmt);
1400 	vsnprintf(line, sizeof(line), fmt, args);
1401 	va_end(args);
1402 	return (pager_output(line));
1403 }
1404 
1405 #endif
1406 
1407 #define	STATUS_FORMAT	"        %s %s\n"
1408 
1409 static int
1410 print_state(int indent, const char *name, vdev_state_t state)
1411 {
1412 	int i;
1413 	char buf[512];
1414 
1415 	buf[0] = 0;
1416 	for (i = 0; i < indent; i++)
1417 		strcat(buf, "  ");
1418 	strcat(buf, name);
1419 	return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
1420 }
1421 
1422 static int
1423 vdev_status(vdev_t *vdev, int indent)
1424 {
1425 	vdev_t *kid;
1426 	int ret;
1427 
1428 	if (vdev->v_islog) {
1429 		(void) pager_output("        logs\n");
1430 		indent++;
1431 	}
1432 
1433 	ret = print_state(indent, vdev->v_name, vdev->v_state);
1434 	if (ret != 0)
1435 		return (ret);
1436 
1437 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1438 		ret = vdev_status(kid, indent + 1);
1439 		if (ret != 0)
1440 			return (ret);
1441 	}
1442 	return (ret);
1443 }
1444 
1445 static int
1446 spa_status(spa_t *spa)
1447 {
1448 	static char bootfs[ZFS_MAXNAMELEN];
1449 	uint64_t rootid;
1450 	vdev_list_t *vlist;
1451 	vdev_t *vdev;
1452 	int good_kids, bad_kids, degraded_kids, ret;
1453 	vdev_state_t state;
1454 
1455 	ret = pager_printf("  pool: %s\n", spa->spa_name);
1456 	if (ret != 0)
1457 		return (ret);
1458 
1459 	if (zfs_get_root(spa, &rootid) == 0 &&
1460 	    zfs_rlookup(spa, rootid, bootfs) == 0) {
1461 		if (bootfs[0] == '\0')
1462 			ret = pager_printf("bootfs: %s\n", spa->spa_name);
1463 		else
1464 			ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
1465 			    bootfs);
1466 		if (ret != 0)
1467 			return (ret);
1468 	}
1469 	ret = pager_printf("config:\n\n");
1470 	if (ret != 0)
1471 		return (ret);
1472 	ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
1473 	if (ret != 0)
1474 		return (ret);
1475 
1476 	good_kids = 0;
1477 	degraded_kids = 0;
1478 	bad_kids = 0;
1479 	vlist = &spa->spa_root_vdev->v_children;
1480 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1481 		if (vdev->v_state == VDEV_STATE_HEALTHY)
1482 			good_kids++;
1483 		else if (vdev->v_state == VDEV_STATE_DEGRADED)
1484 			degraded_kids++;
1485 		else
1486 			bad_kids++;
1487 	}
1488 
1489 	state = VDEV_STATE_CLOSED;
1490 	if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
1491 		state = VDEV_STATE_HEALTHY;
1492 	else if ((good_kids + degraded_kids) > 0)
1493 		state = VDEV_STATE_DEGRADED;
1494 
1495 	ret = print_state(0, spa->spa_name, state);
1496 	if (ret != 0)
1497 		return (ret);
1498 
1499 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1500 		ret = vdev_status(vdev, 1);
1501 		if (ret != 0)
1502 			return (ret);
1503 	}
1504 	return (ret);
1505 }
1506 
1507 static int
1508 spa_all_status(void)
1509 {
1510 	spa_t *spa;
1511 	int first = 1, ret = 0;
1512 
1513 	STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
1514 		if (!first) {
1515 			ret = pager_printf("\n");
1516 			if (ret != 0)
1517 				return (ret);
1518 		}
1519 		first = 0;
1520 		ret = spa_status(spa);
1521 		if (ret != 0)
1522 			return (ret);
1523 	}
1524 	return (ret);
1525 }
1526 
1527 static uint64_t
1528 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
1529 {
1530 	uint64_t label_offset;
1531 
1532 	if (l < VDEV_LABELS / 2)
1533 		label_offset = 0;
1534 	else
1535 		label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
1536 
1537 	return (offset + l * sizeof (vdev_label_t) + label_offset);
1538 }
1539 
1540 static int
1541 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1542 {
1543 	unsigned int seq1 = 0;
1544 	unsigned int seq2 = 0;
1545 	int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
1546 
1547 	if (cmp != 0)
1548 		return (cmp);
1549 
1550 	cmp = AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1551 	if (cmp != 0)
1552 		return (cmp);
1553 
1554 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1555 		seq1 = MMP_SEQ(ub1);
1556 
1557 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1558 		seq2 = MMP_SEQ(ub2);
1559 
1560 	return (AVL_CMP(seq1, seq2));
1561 }
1562 
1563 static int
1564 uberblock_verify(uberblock_t *ub)
1565 {
1566 	if (ub->ub_magic == BSWAP_64((uint64_t)UBERBLOCK_MAGIC)) {
1567 		byteswap_uint64_array(ub, sizeof (uberblock_t));
1568 	}
1569 
1570 	if (ub->ub_magic != UBERBLOCK_MAGIC ||
1571 	    !SPA_VERSION_IS_SUPPORTED(ub->ub_version))
1572 		return (EINVAL);
1573 
1574 	return (0);
1575 }
1576 
1577 static int
1578 vdev_label_read(vdev_t *vd, int l, void *buf, uint64_t offset,
1579     size_t size)
1580 {
1581 	blkptr_t bp;
1582 	off_t off;
1583 
1584 	off = vdev_label_offset(vd->v_psize, l, offset);
1585 
1586 	BP_ZERO(&bp);
1587 	BP_SET_LSIZE(&bp, size);
1588 	BP_SET_PSIZE(&bp, size);
1589 	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1590 	BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1591 	DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
1592 	ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1593 
1594 	return (vdev_read_phys(vd, &bp, buf, off, size));
1595 }
1596 
1597 static nvlist_t *
1598 vdev_label_read_config(vdev_t *vd, uint64_t txg)
1599 {
1600 	vdev_phys_t *label;
1601 	uint64_t best_txg = 0;
1602 	uint64_t label_txg = 0;
1603 	uint64_t asize;
1604 	nvlist_t *nvl = NULL, *tmp;
1605 	int error;
1606 
1607 	label = malloc(sizeof (vdev_phys_t));
1608 	if (label == NULL)
1609 		return (NULL);
1610 
1611 	for (int l = 0; l < VDEV_LABELS; l++) {
1612 		const unsigned char *nvlist;
1613 
1614 		if (vdev_label_read(vd, l, label,
1615 		    offsetof(vdev_label_t, vl_vdev_phys),
1616 		    sizeof (vdev_phys_t)))
1617 			continue;
1618 
1619 		nvlist = (const unsigned char *) label->vp_nvlist;
1620 		tmp = nvlist_import(nvlist + 4, nvlist[0], nvlist[1]);
1621 		if (tmp == NULL)
1622 			continue;
1623 
1624 		error = nvlist_find(tmp, ZPOOL_CONFIG_POOL_TXG,
1625 		    DATA_TYPE_UINT64, NULL, &label_txg, NULL);
1626 		if (error != 0 || label_txg == 0) {
1627 			nvlist_destroy(nvl);
1628 			nvl = tmp;
1629 			goto done;
1630 		}
1631 
1632 		if (label_txg <= txg && label_txg > best_txg) {
1633 			best_txg = label_txg;
1634 			nvlist_destroy(nvl);
1635 			nvl = tmp;
1636 			tmp = NULL;
1637 
1638 			/*
1639 			 * Use asize from pool config. We need this
1640 			 * because we can get bad value from BIOS.
1641 			 */
1642 			if (nvlist_find(nvl, ZPOOL_CONFIG_ASIZE,
1643 			    DATA_TYPE_UINT64, NULL, &asize, NULL) == 0) {
1644 				vd->v_psize = asize +
1645 				    VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
1646 			}
1647 		}
1648 		nvlist_destroy(tmp);
1649 	}
1650 
1651 	if (best_txg == 0) {
1652 		nvlist_destroy(nvl);
1653 		nvl = NULL;
1654 	}
1655 done:
1656 	free(label);
1657 	return (nvl);
1658 }
1659 
1660 static void
1661 vdev_uberblock_load(vdev_t *vd, uberblock_t *ub)
1662 {
1663 	uberblock_t *buf;
1664 
1665 	buf = malloc(VDEV_UBERBLOCK_SIZE(vd));
1666 	if (buf == NULL)
1667 		return;
1668 
1669 	for (int l = 0; l < VDEV_LABELS; l++) {
1670 		for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1671 			if (vdev_label_read(vd, l, buf,
1672 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1673 			    VDEV_UBERBLOCK_SIZE(vd)))
1674 				continue;
1675 			if (uberblock_verify(buf) != 0)
1676 				continue;
1677 
1678 			if (vdev_uberblock_compare(buf, ub) > 0)
1679 				*ub = *buf;
1680 		}
1681 	}
1682 	free(buf);
1683 }
1684 
1685 static int
1686 vdev_probe(vdev_phys_read_t *_read, void *read_priv, spa_t **spap)
1687 {
1688 	vdev_t vtmp;
1689 	spa_t *spa;
1690 	vdev_t *vdev;
1691 	nvlist_t *nvl;
1692 	uint64_t val;
1693 	uint64_t guid, vdev_children;
1694 	uint64_t pool_txg, pool_guid;
1695 	const char *pool_name;
1696 	int rc, namelen;
1697 
1698 	/*
1699 	 * Load the vdev label and figure out which
1700 	 * uberblock is most current.
1701 	 */
1702 	memset(&vtmp, 0, sizeof(vtmp));
1703 	vtmp.v_phys_read = _read;
1704 	vtmp.v_read_priv = read_priv;
1705 	vtmp.v_psize = P2ALIGN(ldi_get_size(read_priv),
1706 	    (uint64_t)sizeof (vdev_label_t));
1707 
1708 	/* Test for minimum device size. */
1709 	if (vtmp.v_psize < SPA_MINDEVSIZE)
1710 		return (EIO);
1711 
1712 	nvl = vdev_label_read_config(&vtmp, UINT64_MAX);
1713 	if (nvl == NULL)
1714 		return (EIO);
1715 
1716 	if (nvlist_find(nvl, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
1717 	    NULL, &val, NULL) != 0) {
1718 		nvlist_destroy(nvl);
1719 		return (EIO);
1720 	}
1721 
1722 	if (!SPA_VERSION_IS_SUPPORTED(val)) {
1723 		printf("ZFS: unsupported ZFS version %u (should be %u)\n",
1724 		    (unsigned)val, (unsigned)SPA_VERSION);
1725 		nvlist_destroy(nvl);
1726 		return (EIO);
1727 	}
1728 
1729 	/* Check ZFS features for read */
1730 	rc = nvlist_check_features_for_read(nvl);
1731 	if (rc != 0) {
1732 		nvlist_destroy(nvl);
1733 		return (EIO);
1734 	}
1735 
1736 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
1737 	    NULL, &val, NULL) != 0) {
1738 		nvlist_destroy(nvl);
1739 		return (EIO);
1740 	}
1741 
1742 	if (val == POOL_STATE_DESTROYED) {
1743 		/* We don't boot only from destroyed pools. */
1744 		nvlist_destroy(nvl);
1745 		return (EIO);
1746 	}
1747 
1748 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
1749 	    NULL, &pool_txg, NULL) != 0 ||
1750 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1751 	    NULL, &pool_guid, NULL) != 0 ||
1752 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
1753 	    NULL, &pool_name, &namelen) != 0) {
1754 		/*
1755 		 * Cache and spare devices end up here - just ignore
1756 		 * them.
1757 		 */
1758 		nvlist_destroy(nvl);
1759 		return (EIO);
1760 	}
1761 
1762 	/*
1763 	 * Create the pool if this is the first time we've seen it.
1764 	 */
1765 	spa = spa_find_by_guid(pool_guid);
1766 	if (spa == NULL) {
1767 		char *name;
1768 
1769 		nvlist_find(nvl, ZPOOL_CONFIG_VDEV_CHILDREN,
1770 		    DATA_TYPE_UINT64, NULL, &vdev_children, NULL);
1771 		name = malloc(namelen + 1);
1772 		if (name == NULL) {
1773 			nvlist_destroy(nvl);
1774 			return (ENOMEM);
1775 		}
1776 		bcopy(pool_name, name, namelen);
1777 		name[namelen] = '\0';
1778 		spa = spa_create(pool_guid, name);
1779 		free(name);
1780 		if (spa == NULL) {
1781 			nvlist_destroy(nvl);
1782 			return (ENOMEM);
1783 		}
1784 		spa->spa_root_vdev->v_nchildren = vdev_children;
1785 	}
1786 	if (pool_txg > spa->spa_txg)
1787 		spa->spa_txg = pool_txg;
1788 
1789 	/*
1790 	 * Get the vdev tree and create our in-core copy of it.
1791 	 * If we already have a vdev with this guid, this must
1792 	 * be some kind of alias (overlapping slices, dangerously dedicated
1793 	 * disks etc).
1794 	 */
1795 	if (nvlist_find(nvl, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1796 	    NULL, &guid, NULL) != 0) {
1797 		nvlist_destroy(nvl);
1798 		return (EIO);
1799 	}
1800 	vdev = vdev_find(guid);
1801 	/* Has this vdev already been inited? */
1802 	if (vdev && vdev->v_phys_read) {
1803 		nvlist_destroy(nvl);
1804 		return (EIO);
1805 	}
1806 
1807 	rc = vdev_init_from_label(spa, nvl);
1808 	nvlist_destroy(nvl);
1809 	if (rc != 0)
1810 		return (rc);
1811 
1812 	/*
1813 	 * We should already have created an incomplete vdev for this
1814 	 * vdev. Find it and initialise it with our read proc.
1815 	 */
1816 	vdev = vdev_find(guid);
1817 	if (vdev != NULL) {
1818 		vdev->v_phys_read = _read;
1819 		vdev->v_read_priv = read_priv;
1820 		vdev->v_psize = vtmp.v_psize;
1821 		/*
1822 		 * If no other state is set, mark vdev healthy.
1823 		 */
1824 		if (vdev->v_state == VDEV_STATE_UNKNOWN)
1825 			vdev->v_state = VDEV_STATE_HEALTHY;
1826 	} else {
1827 		printf("ZFS: inconsistent nvlist contents\n");
1828 		return (EIO);
1829 	}
1830 
1831 	if (vdev->v_islog)
1832 		spa->spa_with_log = vdev->v_islog;
1833 
1834 	/*
1835 	 * Re-evaluate top-level vdev state.
1836 	 */
1837 	vdev_set_state(vdev->v_top);
1838 
1839 	/*
1840 	 * Ok, we are happy with the pool so far. Lets find
1841 	 * the best uberblock and then we can actually access
1842 	 * the contents of the pool.
1843 	 */
1844 	vdev_uberblock_load(vdev, &spa->spa_uberblock);
1845 
1846 	if (spap != NULL)
1847 		*spap = spa;
1848 	return (0);
1849 }
1850 
1851 static int
1852 ilog2(int n)
1853 {
1854 	int v;
1855 
1856 	for (v = 0; v < 32; v++)
1857 		if (n == (1 << v))
1858 			return (v);
1859 	return (-1);
1860 }
1861 
1862 static int
1863 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
1864 {
1865 	blkptr_t gbh_bp;
1866 	zio_gbh_phys_t zio_gb;
1867 	char *pbuf;
1868 	int i;
1869 
1870 	/* Artificial BP for gang block header. */
1871 	gbh_bp = *bp;
1872 	BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1873 	BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1874 	BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
1875 	BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
1876 	for (i = 0; i < SPA_DVAS_PER_BP; i++)
1877 		DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
1878 
1879 	/* Read gang header block using the artificial BP. */
1880 	if (zio_read(spa, &gbh_bp, &zio_gb))
1881 		return (EIO);
1882 
1883 	pbuf = buf;
1884 	for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1885 		blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1886 
1887 		if (BP_IS_HOLE(gbp))
1888 			continue;
1889 		if (zio_read(spa, gbp, pbuf))
1890 			return (EIO);
1891 		pbuf += BP_GET_PSIZE(gbp);
1892 	}
1893 
1894 	if (zio_checksum_verify(spa, bp, buf))
1895 		return (EIO);
1896 	return (0);
1897 }
1898 
1899 static int
1900 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1901 {
1902 	int cpfunc = BP_GET_COMPRESS(bp);
1903 	uint64_t align, size;
1904 	void *pbuf;
1905 	int i, error;
1906 
1907 	/*
1908 	 * Process data embedded in block pointer
1909 	 */
1910 	if (BP_IS_EMBEDDED(bp)) {
1911 		ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
1912 
1913 		size = BPE_GET_PSIZE(bp);
1914 		ASSERT(size <= BPE_PAYLOAD_SIZE);
1915 
1916 		if (cpfunc != ZIO_COMPRESS_OFF)
1917 			pbuf = malloc(size);
1918 		else
1919 			pbuf = buf;
1920 
1921 		if (pbuf == NULL)
1922 			return (ENOMEM);
1923 
1924 		decode_embedded_bp_compressed(bp, pbuf);
1925 		error = 0;
1926 
1927 		if (cpfunc != ZIO_COMPRESS_OFF) {
1928 			error = zio_decompress_data(cpfunc, pbuf,
1929 			    size, buf, BP_GET_LSIZE(bp));
1930 			free(pbuf);
1931 		}
1932 		if (error != 0)
1933 			printf("ZFS: i/o error - unable to decompress "
1934 			    "block pointer data, error %d\n", error);
1935 		return (error);
1936 	}
1937 
1938 	error = EIO;
1939 
1940 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1941 		const dva_t *dva = &bp->blk_dva[i];
1942 		vdev_t *vdev;
1943 		vdev_list_t *vlist;
1944 		uint64_t vdevid;
1945 		off_t offset;
1946 
1947 		if (!dva->dva_word[0] && !dva->dva_word[1])
1948 			continue;
1949 
1950 		vdevid = DVA_GET_VDEV(dva);
1951 		offset = DVA_GET_OFFSET(dva);
1952 		vlist = &spa->spa_root_vdev->v_children;
1953 		STAILQ_FOREACH(vdev, vlist, v_childlink) {
1954 			if (vdev->v_id == vdevid)
1955 				break;
1956 		}
1957 		if (!vdev || !vdev->v_read)
1958 			continue;
1959 
1960 		size = BP_GET_PSIZE(bp);
1961 		if (vdev->v_read == vdev_raidz_read) {
1962 			align = 1ULL << vdev->v_ashift;
1963 			if (P2PHASE(size, align) != 0)
1964 				size = P2ROUNDUP(size, align);
1965 		}
1966 		if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
1967 			pbuf = malloc(size);
1968 		else
1969 			pbuf = buf;
1970 
1971 		if (pbuf == NULL) {
1972 			error = ENOMEM;
1973 			break;
1974 		}
1975 
1976 		if (DVA_GET_GANG(dva))
1977 			error = zio_read_gang(spa, bp, pbuf);
1978 		else
1979 			error = vdev->v_read(vdev, bp, pbuf, offset, size);
1980 		if (error == 0) {
1981 			if (cpfunc != ZIO_COMPRESS_OFF)
1982 				error = zio_decompress_data(cpfunc, pbuf,
1983 				    BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
1984 			else if (size != BP_GET_PSIZE(bp))
1985 				bcopy(pbuf, buf, BP_GET_PSIZE(bp));
1986 		} else {
1987 			printf("zio_read error: %d\n", error);
1988 		}
1989 		if (buf != pbuf)
1990 			free(pbuf);
1991 		if (error == 0)
1992 			break;
1993 	}
1994 	if (error != 0)
1995 		printf("ZFS: i/o error - all block copies unavailable\n");
1996 
1997 	return (error);
1998 }
1999 
2000 static int
2001 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset,
2002     void *buf, size_t buflen)
2003 {
2004 	int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
2005 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2006 	int nlevels = dnode->dn_nlevels;
2007 	int i, rc;
2008 
2009 	if (bsize > SPA_MAXBLOCKSIZE) {
2010 		printf("ZFS: I/O error - blocks larger than %llu are not "
2011 		    "supported\n", SPA_MAXBLOCKSIZE);
2012 		return (EIO);
2013 	}
2014 
2015 	/*
2016 	 * Note: bsize may not be a power of two here so we need to do an
2017 	 * actual divide rather than a bitshift.
2018 	 */
2019 	while (buflen > 0) {
2020 		uint64_t bn = offset / bsize;
2021 		int boff = offset % bsize;
2022 		int ibn;
2023 		const blkptr_t *indbp;
2024 		blkptr_t bp;
2025 
2026 		if (bn > dnode->dn_maxblkid)
2027 			return (EIO);
2028 
2029 		if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
2030 			goto cached;
2031 
2032 		indbp = dnode->dn_blkptr;
2033 		for (i = 0; i < nlevels; i++) {
2034 			/*
2035 			 * Copy the bp from the indirect array so that
2036 			 * we can re-use the scratch buffer for multi-level
2037 			 * objects.
2038 			 */
2039 			ibn = bn >> ((nlevels - i - 1) * ibshift);
2040 			ibn &= ((1 << ibshift) - 1);
2041 			bp = indbp[ibn];
2042 			if (BP_IS_HOLE(&bp)) {
2043 				memset(dnode_cache_buf, 0, bsize);
2044 				break;
2045 			}
2046 			rc = zio_read(spa, &bp, dnode_cache_buf);
2047 			if (rc)
2048 				return (rc);
2049 			indbp = (const blkptr_t *) dnode_cache_buf;
2050 		}
2051 		dnode_cache_obj = dnode;
2052 		dnode_cache_bn = bn;
2053 	cached:
2054 
2055 		/*
2056 		 * The buffer contains our data block. Copy what we
2057 		 * need from it and loop.
2058 		 */
2059 		i = bsize - boff;
2060 		if (i > buflen) i = buflen;
2061 		memcpy(buf, &dnode_cache_buf[boff], i);
2062 		buf = ((char *)buf) + i;
2063 		offset += i;
2064 		buflen -= i;
2065 	}
2066 
2067 	return (0);
2068 }
2069 
2070 /*
2071  * Lookup a value in a microzap directory.
2072  */
2073 static int
2074 mzap_lookup(const mzap_phys_t *mz, size_t size, const char *name,
2075     uint64_t *value)
2076 {
2077 	const mzap_ent_phys_t *mze;
2078 	int chunks, i;
2079 
2080 	/*
2081 	 * Microzap objects use exactly one block. Read the whole
2082 	 * thing.
2083 	 */
2084 	chunks = size / MZAP_ENT_LEN - 1;
2085 	for (i = 0; i < chunks; i++) {
2086 		mze = &mz->mz_chunk[i];
2087 		if (strcmp(mze->mze_name, name) == 0) {
2088 			*value = mze->mze_value;
2089 			return (0);
2090 		}
2091 	}
2092 
2093 	return (ENOENT);
2094 }
2095 
2096 /*
2097  * Compare a name with a zap leaf entry. Return non-zero if the name
2098  * matches.
2099  */
2100 static int
2101 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2102     const char *name)
2103 {
2104 	size_t namelen;
2105 	const zap_leaf_chunk_t *nc;
2106 	const char *p;
2107 
2108 	namelen = zc->l_entry.le_name_numints;
2109 
2110 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2111 	p = name;
2112 	while (namelen > 0) {
2113 		size_t len;
2114 
2115 		len = namelen;
2116 		if (len > ZAP_LEAF_ARRAY_BYTES)
2117 			len = ZAP_LEAF_ARRAY_BYTES;
2118 		if (memcmp(p, nc->l_array.la_array, len))
2119 			return (0);
2120 		p += len;
2121 		namelen -= len;
2122 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2123 	}
2124 
2125 	return (1);
2126 }
2127 
2128 /*
2129  * Extract a uint64_t value from a zap leaf entry.
2130  */
2131 static uint64_t
2132 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
2133 {
2134 	const zap_leaf_chunk_t *vc;
2135 	int i;
2136 	uint64_t value;
2137 	const uint8_t *p;
2138 
2139 	vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
2140 	for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
2141 		value = (value << 8) | p[i];
2142 	}
2143 
2144 	return (value);
2145 }
2146 
2147 static void
2148 stv(int len, void *addr, uint64_t value)
2149 {
2150 	switch (len) {
2151 	case 1:
2152 		*(uint8_t *)addr = value;
2153 		return;
2154 	case 2:
2155 		*(uint16_t *)addr = value;
2156 		return;
2157 	case 4:
2158 		*(uint32_t *)addr = value;
2159 		return;
2160 	case 8:
2161 		*(uint64_t *)addr = value;
2162 		return;
2163 	}
2164 }
2165 
2166 /*
2167  * Extract a array from a zap leaf entry.
2168  */
2169 static void
2170 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2171     uint64_t integer_size, uint64_t num_integers, void *buf)
2172 {
2173 	uint64_t array_int_len = zc->l_entry.le_value_intlen;
2174 	uint64_t value = 0;
2175 	uint64_t *u64 = buf;
2176 	char *p = buf;
2177 	int len = MIN(zc->l_entry.le_value_numints, num_integers);
2178 	int chunk = zc->l_entry.le_value_chunk;
2179 	int byten = 0;
2180 
2181 	if (integer_size == 8 && len == 1) {
2182 		*u64 = fzap_leaf_value(zl, zc);
2183 		return;
2184 	}
2185 
2186 	while (len > 0) {
2187 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
2188 		int i;
2189 
2190 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
2191 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
2192 			value = (value << 8) | la->la_array[i];
2193 			byten++;
2194 			if (byten == array_int_len) {
2195 				stv(integer_size, p, value);
2196 				byten = 0;
2197 				len--;
2198 				if (len == 0)
2199 					return;
2200 				p += integer_size;
2201 			}
2202 		}
2203 		chunk = la->la_next;
2204 	}
2205 }
2206 
2207 static int
2208 fzap_check_size(uint64_t integer_size, uint64_t num_integers)
2209 {
2210 
2211 	switch (integer_size) {
2212 	case 1:
2213 	case 2:
2214 	case 4:
2215 	case 8:
2216 		break;
2217 	default:
2218 		return (EINVAL);
2219 	}
2220 
2221 	if (integer_size * num_integers > ZAP_MAXVALUELEN)
2222 		return (E2BIG);
2223 
2224 	return (0);
2225 }
2226 
2227 static void
2228 zap_leaf_free(zap_leaf_t *leaf)
2229 {
2230 	free(leaf->l_phys);
2231 	free(leaf);
2232 }
2233 
2234 static int
2235 zap_get_leaf_byblk(fat_zap_t *zap, uint64_t blk, zap_leaf_t **lp)
2236 {
2237 	int bs = FZAP_BLOCK_SHIFT(zap);
2238 	int err;
2239 
2240 	*lp = malloc(sizeof(**lp));
2241 	if (*lp == NULL)
2242 		return (ENOMEM);
2243 
2244 	(*lp)->l_bs = bs;
2245 	(*lp)->l_phys = malloc(1 << bs);
2246 
2247 	if ((*lp)->l_phys == NULL) {
2248 		free(*lp);
2249 		return (ENOMEM);
2250 	}
2251 	err = dnode_read(zap->zap_spa, zap->zap_dnode, blk << bs, (*lp)->l_phys,
2252 	    1 << bs);
2253 	if (err != 0) {
2254 		zap_leaf_free(*lp);
2255 	}
2256 	return (err);
2257 }
2258 
2259 static int
2260 zap_table_load(fat_zap_t *zap, zap_table_phys_t *tbl, uint64_t idx,
2261     uint64_t *valp)
2262 {
2263 	int bs = FZAP_BLOCK_SHIFT(zap);
2264 	uint64_t blk = idx >> (bs - 3);
2265 	uint64_t off = idx & ((1 << (bs - 3)) - 1);
2266 	uint64_t *buf;
2267 	int rc;
2268 
2269 	buf = malloc(1 << zap->zap_block_shift);
2270 	if (buf == NULL)
2271 		return (ENOMEM);
2272 	rc = dnode_read(zap->zap_spa, zap->zap_dnode, (tbl->zt_blk + blk) << bs,
2273 	    buf, 1 << zap->zap_block_shift);
2274 	if (rc == 0)
2275 		*valp = buf[off];
2276 	free(buf);
2277 	return (rc);
2278 }
2279 
2280 static int
2281 zap_idx_to_blk(fat_zap_t *zap, uint64_t idx, uint64_t *valp)
2282 {
2283 	if (zap->zap_phys->zap_ptrtbl.zt_numblks == 0) {
2284 		*valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
2285 		return (0);
2286 	} else {
2287 		return (zap_table_load(zap, &zap->zap_phys->zap_ptrtbl,
2288 		    idx, valp));
2289 	}
2290 }
2291 
2292 #define	ZAP_HASH_IDX(hash, n)	(((n) == 0) ? 0 : ((hash) >> (64 - (n))))
2293 static int
2294 zap_deref_leaf(fat_zap_t *zap, uint64_t h, zap_leaf_t **lp)
2295 {
2296 	uint64_t idx, blk;
2297 	int err;
2298 
2299 	idx = ZAP_HASH_IDX(h, zap->zap_phys->zap_ptrtbl.zt_shift);
2300 	err = zap_idx_to_blk(zap, idx, &blk);
2301 	if (err != 0)
2302 		return (err);
2303 	return (zap_get_leaf_byblk(zap, blk, lp));
2304 }
2305 
2306 #define	CHAIN_END	0xffff	/* end of the chunk chain */
2307 #define	LEAF_HASH(l, h) \
2308 	((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \
2309 	((h) >> \
2310 	(64 - ZAP_LEAF_HASH_SHIFT(l) - (l)->l_phys->l_hdr.lh_prefix_len)))
2311 #define	LEAF_HASH_ENTPTR(l, h)	(&(l)->l_phys->l_hash[LEAF_HASH(l, h)])
2312 
2313 static int
2314 zap_leaf_lookup(zap_leaf_t *zl, uint64_t hash, const char *name,
2315     uint64_t integer_size, uint64_t num_integers, void *value)
2316 {
2317 	int rc;
2318 	uint16_t *chunkp;
2319 	struct zap_leaf_entry *le;
2320 
2321 	/*
2322 	 * Make sure this chunk matches our hash.
2323 	 */
2324 	if (zl->l_phys->l_hdr.lh_prefix_len > 0 &&
2325 	    zl->l_phys->l_hdr.lh_prefix !=
2326 	    hash >> (64 - zl->l_phys->l_hdr.lh_prefix_len))
2327 		return (EIO);
2328 
2329 	rc = ENOENT;
2330 	for (chunkp = LEAF_HASH_ENTPTR(zl, hash);
2331 	    *chunkp != CHAIN_END; chunkp = &le->le_next) {
2332 		zap_leaf_chunk_t *zc;
2333 		uint16_t chunk = *chunkp;
2334 
2335 		le = ZAP_LEAF_ENTRY(zl, chunk);
2336 		if (le->le_hash != hash)
2337 			continue;
2338 		zc = &ZAP_LEAF_CHUNK(zl, chunk);
2339 		if (fzap_name_equal(zl, zc, name)) {
2340 			if (zc->l_entry.le_value_intlen > integer_size) {
2341 				rc = EINVAL;
2342 			} else {
2343 				fzap_leaf_array(zl, zc, integer_size,
2344 				    num_integers, value);
2345 				rc = 0;
2346 			}
2347 			break;
2348 		}
2349 	}
2350 	return (rc);
2351 }
2352 
2353 /*
2354  * Lookup a value in a fatzap directory.
2355  */
2356 static int
2357 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2358     const char *name, uint64_t integer_size, uint64_t num_integers,
2359     void *value)
2360 {
2361 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2362 	fat_zap_t z;
2363 	zap_leaf_t *zl;
2364 	uint64_t hash;
2365 	int rc;
2366 
2367 	if (zh->zap_magic != ZAP_MAGIC)
2368 		return (EIO);
2369 
2370 	if ((rc = fzap_check_size(integer_size, num_integers)) != 0)
2371 		return (rc);
2372 
2373 	z.zap_block_shift = ilog2(bsize);
2374 	z.zap_phys = zh;
2375 	z.zap_spa = spa;
2376 	z.zap_dnode = dnode;
2377 
2378 	hash = zap_hash(zh->zap_salt, name);
2379 	rc = zap_deref_leaf(&z, hash, &zl);
2380 	if (rc != 0)
2381 		return (rc);
2382 
2383 	rc = zap_leaf_lookup(zl, hash, name, integer_size, num_integers, value);
2384 
2385 	zap_leaf_free(zl);
2386 	return (rc);
2387 }
2388 
2389 /*
2390  * Lookup a name in a zap object and return its value as a uint64_t.
2391  */
2392 static int
2393 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2394     uint64_t integer_size, uint64_t num_integers, void *value)
2395 {
2396 	int rc;
2397 	zap_phys_t *zap;
2398 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2399 
2400 	zap = malloc(size);
2401 	if (zap == NULL)
2402 		return (ENOMEM);
2403 
2404 	rc = dnode_read(spa, dnode, 0, zap, size);
2405 	if (rc)
2406 		goto done;
2407 
2408 	switch (zap->zap_block_type) {
2409 	case ZBT_MICRO:
2410 		rc = mzap_lookup((const mzap_phys_t *)zap, size, name, value);
2411 		break;
2412 	case ZBT_HEADER:
2413 		rc = fzap_lookup(spa, dnode, zap, name, integer_size,
2414 		    num_integers, value);
2415 		break;
2416 	default:
2417 		printf("ZFS: invalid zap_type=%" PRIx64 "\n",
2418 		    zap->zap_block_type);
2419 		rc = EIO;
2420 	}
2421 done:
2422 	free(zap);
2423 	return (rc);
2424 }
2425 
2426 /*
2427  * List a microzap directory.
2428  */
2429 static int
2430 mzap_list(const mzap_phys_t *mz, size_t size,
2431     int (*callback)(const char *, uint64_t))
2432 {
2433 	const mzap_ent_phys_t *mze;
2434 	int chunks, i, rc;
2435 
2436 	/*
2437 	 * Microzap objects use exactly one block. Read the whole
2438 	 * thing.
2439 	 */
2440 	rc = 0;
2441 	chunks = size / MZAP_ENT_LEN - 1;
2442 	for (i = 0; i < chunks; i++) {
2443 		mze = &mz->mz_chunk[i];
2444 		if (mze->mze_name[0]) {
2445 			rc = callback(mze->mze_name, mze->mze_value);
2446 			if (rc != 0)
2447 				break;
2448 		}
2449 	}
2450 
2451 	return (rc);
2452 }
2453 
2454 /*
2455  * List a fatzap directory.
2456  */
2457 static int
2458 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2459     int (*callback)(const char *, uint64_t))
2460 {
2461 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2462 	fat_zap_t z;
2463 	uint64_t i;
2464 	int j, rc;
2465 
2466 	if (zh->zap_magic != ZAP_MAGIC)
2467 		return (EIO);
2468 
2469 	z.zap_block_shift = ilog2(bsize);
2470 	z.zap_phys = zh;
2471 
2472 	/*
2473 	 * This assumes that the leaf blocks start at block 1. The
2474 	 * documentation isn't exactly clear on this.
2475 	 */
2476 	zap_leaf_t zl;
2477 	zl.l_bs = z.zap_block_shift;
2478 	zl.l_phys = malloc(bsize);
2479 	if (zl.l_phys == NULL)
2480 		return (ENOMEM);
2481 
2482 	for (i = 0; i < zh->zap_num_leafs; i++) {
2483 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2484 		char name[256], *p;
2485 		uint64_t value;
2486 
2487 		if (dnode_read(spa, dnode, off, zl.l_phys, bsize)) {
2488 			free(zl.l_phys);
2489 			return (EIO);
2490 		}
2491 
2492 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2493 			zap_leaf_chunk_t *zc, *nc;
2494 			int namelen;
2495 
2496 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2497 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2498 				continue;
2499 			namelen = zc->l_entry.le_name_numints;
2500 			if (namelen > sizeof(name))
2501 				namelen = sizeof(name);
2502 
2503 			/*
2504 			 * Paste the name back together.
2505 			 */
2506 			nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
2507 			p = name;
2508 			while (namelen > 0) {
2509 				int len;
2510 				len = namelen;
2511 				if (len > ZAP_LEAF_ARRAY_BYTES)
2512 					len = ZAP_LEAF_ARRAY_BYTES;
2513 				memcpy(p, nc->l_array.la_array, len);
2514 				p += len;
2515 				namelen -= len;
2516 				nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
2517 			}
2518 
2519 			/*
2520 			 * Assume the first eight bytes of the value are
2521 			 * a uint64_t.
2522 			 */
2523 			value = fzap_leaf_value(&zl, zc);
2524 
2525 			/* printf("%s 0x%jx\n", name, (uintmax_t)value); */
2526 			rc = callback((const char *)name, value);
2527 			if (rc != 0) {
2528 				free(zl.l_phys);
2529 				return (rc);
2530 			}
2531 		}
2532 	}
2533 
2534 	free(zl.l_phys);
2535 	return (0);
2536 }
2537 
2538 static int zfs_printf(const char *name, uint64_t value __unused)
2539 {
2540 
2541 	printf("%s\n", name);
2542 
2543 	return (0);
2544 }
2545 
2546 /*
2547  * List a zap directory.
2548  */
2549 static int
2550 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
2551 {
2552 	zap_phys_t *zap;
2553 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2554 	int rc;
2555 
2556 	zap = malloc(size);
2557 	if (zap == NULL)
2558 		return (ENOMEM);
2559 
2560 	rc = dnode_read(spa, dnode, 0, zap, size);
2561 	if (rc == 0) {
2562 		if (zap->zap_block_type == ZBT_MICRO)
2563 			rc = mzap_list((const mzap_phys_t *)zap, size,
2564 			    zfs_printf);
2565 		else
2566 			rc = fzap_list(spa, dnode, zap, zfs_printf);
2567 	}
2568 	free(zap);
2569 	return (rc);
2570 }
2571 
2572 static int
2573 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum,
2574     dnode_phys_t *dnode)
2575 {
2576 	off_t offset;
2577 
2578 	offset = objnum * sizeof(dnode_phys_t);
2579 	return dnode_read(spa, &os->os_meta_dnode, offset,
2580 		dnode, sizeof(dnode_phys_t));
2581 }
2582 
2583 /*
2584  * Lookup a name in a microzap directory.
2585  */
2586 static int
2587 mzap_rlookup(const mzap_phys_t *mz, size_t size, char *name, uint64_t value)
2588 {
2589 	const mzap_ent_phys_t *mze;
2590 	int chunks, i;
2591 
2592 	/*
2593 	 * Microzap objects use exactly one block. Read the whole
2594 	 * thing.
2595 	 */
2596 	chunks = size / MZAP_ENT_LEN - 1;
2597 	for (i = 0; i < chunks; i++) {
2598 		mze = &mz->mz_chunk[i];
2599 		if (value == mze->mze_value) {
2600 			strcpy(name, mze->mze_name);
2601 			return (0);
2602 		}
2603 	}
2604 
2605 	return (ENOENT);
2606 }
2607 
2608 static void
2609 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
2610 {
2611 	size_t namelen;
2612 	const zap_leaf_chunk_t *nc;
2613 	char *p;
2614 
2615 	namelen = zc->l_entry.le_name_numints;
2616 
2617 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2618 	p = name;
2619 	while (namelen > 0) {
2620 		size_t len;
2621 		len = namelen;
2622 		if (len > ZAP_LEAF_ARRAY_BYTES)
2623 			len = ZAP_LEAF_ARRAY_BYTES;
2624 		memcpy(p, nc->l_array.la_array, len);
2625 		p += len;
2626 		namelen -= len;
2627 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2628 	}
2629 
2630 	*p = '\0';
2631 }
2632 
2633 static int
2634 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2635     char *name, uint64_t value)
2636 {
2637 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2638 	fat_zap_t z;
2639 	uint64_t i;
2640 	int j, rc;
2641 
2642 	if (zh->zap_magic != ZAP_MAGIC)
2643 		return (EIO);
2644 
2645 	z.zap_block_shift = ilog2(bsize);
2646 	z.zap_phys = zh;
2647 
2648 	/*
2649 	 * This assumes that the leaf blocks start at block 1. The
2650 	 * documentation isn't exactly clear on this.
2651 	 */
2652 	zap_leaf_t zl;
2653 	zl.l_bs = z.zap_block_shift;
2654 	zl.l_phys = malloc(bsize);
2655 	if (zl.l_phys == NULL)
2656 		return (ENOMEM);
2657 
2658 	for (i = 0; i < zh->zap_num_leafs; i++) {
2659 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2660 
2661 		rc = dnode_read(spa, dnode, off, zl.l_phys, bsize);
2662 		if (rc != 0)
2663 			goto done;
2664 
2665 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2666 			zap_leaf_chunk_t *zc;
2667 
2668 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2669 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2670 				continue;
2671 			if (zc->l_entry.le_value_intlen != 8 ||
2672 			    zc->l_entry.le_value_numints != 1)
2673 				continue;
2674 
2675 			if (fzap_leaf_value(&zl, zc) == value) {
2676 				fzap_name_copy(&zl, zc, name);
2677 				goto done;
2678 			}
2679 		}
2680 	}
2681 
2682 	rc = ENOENT;
2683 done:
2684 	free(zl.l_phys);
2685 	return (rc);
2686 }
2687 
2688 static int
2689 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name,
2690     uint64_t value)
2691 {
2692 	zap_phys_t *zap;
2693 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2694 	int rc;
2695 
2696 	zap = malloc(size);
2697 	if (zap == NULL)
2698 		return (ENOMEM);
2699 
2700 	rc = dnode_read(spa, dnode, 0, zap, size);
2701 	if (rc == 0) {
2702 		if (zap->zap_block_type == ZBT_MICRO)
2703 			rc = mzap_rlookup((const mzap_phys_t *)zap, size,
2704 			    name, value);
2705 		else
2706 			rc = fzap_rlookup(spa, dnode, zap, name, value);
2707 	}
2708 	free(zap);
2709 	return (rc);
2710 }
2711 
2712 static int
2713 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
2714 {
2715 	char name[256];
2716 	char component[256];
2717 	uint64_t dir_obj, parent_obj, child_dir_zapobj;
2718 	dnode_phys_t child_dir_zap, dataset, dir, parent;
2719 	dsl_dir_phys_t *dd;
2720 	dsl_dataset_phys_t *ds;
2721 	char *p;
2722 	int len;
2723 
2724 	p = &name[sizeof(name) - 1];
2725 	*p = '\0';
2726 
2727 	if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2728 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2729 		return (EIO);
2730 	}
2731 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2732 	dir_obj = ds->ds_dir_obj;
2733 
2734 	for (;;) {
2735 		if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir) != 0)
2736 			return (EIO);
2737 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2738 
2739 		/* Actual loop condition. */
2740 		parent_obj = dd->dd_parent_obj;
2741 		if (parent_obj == 0)
2742 			break;
2743 
2744 		if (objset_get_dnode(spa, &spa->spa_mos, parent_obj,
2745 		    &parent) != 0)
2746 			return (EIO);
2747 		dd = (dsl_dir_phys_t *)&parent.dn_bonus;
2748 		child_dir_zapobj = dd->dd_child_dir_zapobj;
2749 		if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj,
2750 		    &child_dir_zap) != 0)
2751 			return (EIO);
2752 		if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
2753 			return (EIO);
2754 
2755 		len = strlen(component);
2756 		p -= len;
2757 		memcpy(p, component, len);
2758 		--p;
2759 		*p = '/';
2760 
2761 		/* Actual loop iteration. */
2762 		dir_obj = parent_obj;
2763 	}
2764 
2765 	if (*p != '\0')
2766 		++p;
2767 	strcpy(result, p);
2768 
2769 	return (0);
2770 }
2771 
2772 static int
2773 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
2774 {
2775 	char element[256];
2776 	uint64_t dir_obj, child_dir_zapobj;
2777 	dnode_phys_t child_dir_zap, dir;
2778 	dsl_dir_phys_t *dd;
2779 	const char *p, *q;
2780 
2781 	if (objset_get_dnode(spa, &spa->spa_mos,
2782 	    DMU_POOL_DIRECTORY_OBJECT, &dir))
2783 		return (EIO);
2784 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
2785 	    1, &dir_obj))
2786 		return (EIO);
2787 
2788 	p = name;
2789 	for (;;) {
2790 		if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir))
2791 			return (EIO);
2792 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2793 
2794 		while (*p == '/')
2795 			p++;
2796 		/* Actual loop condition #1. */
2797 		if (*p == '\0')
2798 			break;
2799 
2800 		q = strchr(p, '/');
2801 		if (q) {
2802 			memcpy(element, p, q - p);
2803 			element[q - p] = '\0';
2804 			p = q + 1;
2805 		} else {
2806 			strcpy(element, p);
2807 			p += strlen(p);
2808 		}
2809 
2810 		child_dir_zapobj = dd->dd_child_dir_zapobj;
2811 		if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj,
2812 		    &child_dir_zap) != 0)
2813 			return (EIO);
2814 
2815 		/* Actual loop condition #2. */
2816 		if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
2817 		    1, &dir_obj) != 0)
2818 			return (ENOENT);
2819 	}
2820 
2821 	*objnum = dd->dd_head_dataset_obj;
2822 	return (0);
2823 }
2824 
2825 #ifndef BOOT2
2826 static int
2827 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
2828 {
2829 	uint64_t dir_obj, child_dir_zapobj;
2830 	dnode_phys_t child_dir_zap, dir, dataset;
2831 	dsl_dataset_phys_t *ds;
2832 	dsl_dir_phys_t *dd;
2833 
2834 	if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2835 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2836 		return (EIO);
2837 	}
2838 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2839 	dir_obj = ds->ds_dir_obj;
2840 
2841 	if (objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir)) {
2842 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2843 		return (EIO);
2844 	}
2845 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2846 
2847 	child_dir_zapobj = dd->dd_child_dir_zapobj;
2848 	if (objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj,
2849 	    &child_dir_zap) != 0) {
2850 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2851 		return (EIO);
2852 	}
2853 
2854 	return (zap_list(spa, &child_dir_zap) != 0);
2855 }
2856 
2857 int
2858 zfs_callback_dataset(const spa_t *spa, uint64_t objnum,
2859     int (*callback)(const char *, uint64_t))
2860 {
2861 	uint64_t dir_obj, child_dir_zapobj;
2862 	dnode_phys_t child_dir_zap, dir, dataset;
2863 	dsl_dataset_phys_t *ds;
2864 	dsl_dir_phys_t *dd;
2865 	zap_phys_t *zap;
2866 	size_t size;
2867 	int err;
2868 
2869 	err = objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset);
2870 	if (err != 0) {
2871 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2872 		return (err);
2873 	}
2874 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2875 	dir_obj = ds->ds_dir_obj;
2876 
2877 	err = objset_get_dnode(spa, &spa->spa_mos, dir_obj, &dir);
2878 	if (err != 0) {
2879 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2880 		return (err);
2881 	}
2882 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2883 
2884 	child_dir_zapobj = dd->dd_child_dir_zapobj;
2885 	err = objset_get_dnode(spa, &spa->spa_mos, child_dir_zapobj,
2886 	    &child_dir_zap);
2887 	if (err != 0) {
2888 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2889 		return (err);
2890 	}
2891 
2892 	size = child_dir_zap.dn_datablkszsec << SPA_MINBLOCKSHIFT;
2893 	zap = malloc(size);
2894 	if (zap != NULL) {
2895 		err = dnode_read(spa, &child_dir_zap, 0, zap, size);
2896 		if (err != 0)
2897 			goto done;
2898 
2899 		if (zap->zap_block_type == ZBT_MICRO)
2900 			err = mzap_list((const mzap_phys_t *)zap, size,
2901 			    callback);
2902 		else
2903 			err = fzap_list(spa, &child_dir_zap, zap, callback);
2904 	} else {
2905 		err = ENOMEM;
2906 	}
2907 done:
2908 	free(zap);
2909 	return (err);
2910 }
2911 #endif
2912 
2913 /*
2914  * Find the object set given the object number of its dataset object
2915  * and return its details in *objset
2916  */
2917 static int
2918 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
2919 {
2920 	dnode_phys_t dataset;
2921 	dsl_dataset_phys_t *ds;
2922 
2923 	if (objset_get_dnode(spa, &spa->spa_mos, objnum, &dataset)) {
2924 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2925 		return (EIO);
2926 	}
2927 
2928 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2929 	if (zio_read(spa, &ds->ds_bp, objset)) {
2930 		printf("ZFS: can't read object set for dataset %ju\n",
2931 		    (uintmax_t)objnum);
2932 		return (EIO);
2933 	}
2934 
2935 	return (0);
2936 }
2937 
2938 /*
2939  * Find the object set pointed to by the BOOTFS property or the root
2940  * dataset if there is none and return its details in *objset
2941  */
2942 static int
2943 zfs_get_root(const spa_t *spa, uint64_t *objid)
2944 {
2945 	dnode_phys_t dir, propdir;
2946 	uint64_t props, bootfs, root;
2947 
2948 	*objid = 0;
2949 
2950 	/*
2951 	 * Start with the MOS directory object.
2952 	 */
2953 	if (objset_get_dnode(spa, &spa->spa_mos,
2954 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
2955 		printf("ZFS: can't read MOS object directory\n");
2956 		return (EIO);
2957 	}
2958 
2959 	/*
2960 	 * Lookup the pool_props and see if we can find a bootfs.
2961 	 */
2962 	if (zap_lookup(spa, &dir, DMU_POOL_PROPS,
2963 	    sizeof(props), 1, &props) == 0 &&
2964 	    objset_get_dnode(spa, &spa->spa_mos, props, &propdir) == 0 &&
2965 	    zap_lookup(spa, &propdir, "bootfs",
2966 	    sizeof(bootfs), 1, &bootfs) == 0 && bootfs != 0) {
2967 		*objid = bootfs;
2968 		return (0);
2969 	}
2970 	/*
2971 	 * Lookup the root dataset directory
2972 	 */
2973 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET,
2974 	    sizeof(root), 1, &root) ||
2975 	    objset_get_dnode(spa, &spa->spa_mos, root, &dir)) {
2976 		printf("ZFS: can't find root dsl_dir\n");
2977 		return (EIO);
2978 	}
2979 
2980 	/*
2981 	 * Use the information from the dataset directory's bonus buffer
2982 	 * to find the dataset object and from that the object set itself.
2983 	 */
2984 	dsl_dir_phys_t *dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2985 	*objid = dd->dd_head_dataset_obj;
2986 	return (0);
2987 }
2988 
2989 static int
2990 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
2991 {
2992 
2993 	mount->spa = spa;
2994 
2995 	/*
2996 	 * Find the root object set if not explicitly provided
2997 	 */
2998 	if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
2999 		printf("ZFS: can't find root filesystem\n");
3000 		return (EIO);
3001 	}
3002 
3003 	if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
3004 		printf("ZFS: can't open root filesystem\n");
3005 		return (EIO);
3006 	}
3007 
3008 	mount->rootobj = rootobj;
3009 
3010 	return (0);
3011 }
3012 
3013 /*
3014  * callback function for feature name checks.
3015  */
3016 static int
3017 check_feature(const char *name, uint64_t value)
3018 {
3019 	int i;
3020 
3021 	if (value == 0)
3022 		return (0);
3023 	if (name[0] == '\0')
3024 		return (0);
3025 
3026 	for (i = 0; features_for_read[i] != NULL; i++) {
3027 		if (strcmp(name, features_for_read[i]) == 0)
3028 			return (0);
3029 	}
3030 	printf("ZFS: unsupported feature: %s\n", name);
3031 	return (EIO);
3032 }
3033 
3034 /*
3035  * Checks whether the MOS features that are active are supported.
3036  */
3037 static int
3038 check_mos_features(const spa_t *spa)
3039 {
3040 	dnode_phys_t dir;
3041 	zap_phys_t *zap;
3042 	uint64_t objnum;
3043 	size_t size;
3044 	int rc;
3045 
3046 	if ((rc = objset_get_dnode(spa, &spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
3047 	    &dir)) != 0)
3048 		return (rc);
3049 	if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
3050 	    sizeof (objnum), 1, &objnum)) != 0) {
3051 		/*
3052 		 * It is older pool without features. As we have already
3053 		 * tested the label, just return without raising the error.
3054 		 */
3055 		return (0);
3056 	}
3057 
3058 	if ((rc = objset_get_dnode(spa, &spa->spa_mos, objnum, &dir)) != 0)
3059 		return (rc);
3060 
3061 	if (dir.dn_type != DMU_OTN_ZAP_METADATA)
3062 		return (EIO);
3063 
3064 	size = dir.dn_datablkszsec << SPA_MINBLOCKSHIFT;
3065 	zap = malloc(size);
3066 	if (zap == NULL)
3067 		return (ENOMEM);
3068 
3069 	if (dnode_read(spa, &dir, 0, zap, size)) {
3070 		free(zap);
3071 		return (EIO);
3072 	}
3073 
3074 	if (zap->zap_block_type == ZBT_MICRO)
3075 		rc = mzap_list((const mzap_phys_t *)zap, size, check_feature);
3076 	else
3077 		rc = fzap_list(spa, &dir, zap, check_feature);
3078 
3079 	free(zap);
3080 	return (rc);
3081 }
3082 
3083 static int
3084 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
3085 {
3086 	dnode_phys_t dir;
3087 	size_t size;
3088 	int rc;
3089 	unsigned char *nv;
3090 
3091 	*value = NULL;
3092 	if ((rc = objset_get_dnode(spa, &spa->spa_mos, obj, &dir)) != 0)
3093 		return (rc);
3094 	if (dir.dn_type != DMU_OT_PACKED_NVLIST &&
3095 	    dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) {
3096 		return (EIO);
3097 	}
3098 
3099 	if (dir.dn_bonuslen != sizeof (uint64_t))
3100 		return (EIO);
3101 
3102 	size = *(uint64_t *)DN_BONUS(&dir);
3103 	nv = malloc(size);
3104 	if (nv == NULL)
3105 		return (ENOMEM);
3106 
3107 	rc = dnode_read(spa, &dir, 0, nv, size);
3108 	if (rc != 0) {
3109 		free(nv);
3110 		nv = NULL;
3111 		return (rc);
3112 	}
3113 	*value = nvlist_import(nv + 4, nv[0], nv[1]);
3114 	free(nv);
3115 	return (rc);
3116 }
3117 
3118 static int
3119 zfs_spa_init(spa_t *spa)
3120 {
3121 	dnode_phys_t dir;
3122 	uint64_t config_object;
3123 	nvlist_t *nvlist;
3124 	int rc;
3125 
3126 	if (zio_read(spa, &spa->spa_uberblock.ub_rootbp, &spa->spa_mos)) {
3127 		printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
3128 		return (EIO);
3129 	}
3130 	if (spa->spa_mos.os_type != DMU_OST_META) {
3131 		printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
3132 		return (EIO);
3133 	}
3134 
3135 	if (objset_get_dnode(spa, &spa->spa_mos, DMU_POOL_DIRECTORY_OBJECT,
3136 	    &dir)) {
3137 		printf("ZFS: failed to read pool %s directory object\n",
3138 		    spa->spa_name);
3139 		return (EIO);
3140 	}
3141 	/* this is allowed to fail, older pools do not have salt */
3142 	rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
3143 	    sizeof (spa->spa_cksum_salt.zcs_bytes),
3144 	    spa->spa_cksum_salt.zcs_bytes);
3145 
3146 	rc = check_mos_features(spa);
3147 	if (rc != 0) {
3148 		printf("ZFS: pool %s is not supported\n", spa->spa_name);
3149 		return (rc);
3150 	}
3151 
3152 	rc = zap_lookup(spa, &dir, DMU_POOL_CONFIG,
3153 	    sizeof (config_object), 1, &config_object);
3154 	if (rc != 0) {
3155 		printf("ZFS: can not read MOS %s\n", DMU_POOL_CONFIG);
3156 		return (EIO);
3157 	}
3158 	rc = load_nvlist(spa, config_object, &nvlist);
3159 	if (rc != 0)
3160 		return (rc);
3161 	/*
3162 	 * Update vdevs from MOS config. Note, we do skip encoding bytes
3163 	 * here. See also vdev_label_read_config().
3164 	 */
3165 	rc = vdev_init_from_nvlist(spa, nvlist);
3166 	nvlist_destroy(nvlist);
3167 	return (rc);
3168 }
3169 
3170 static int
3171 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
3172 {
3173 
3174 	if (dn->dn_bonustype != DMU_OT_SA) {
3175 		znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
3176 
3177 		sb->st_mode = zp->zp_mode;
3178 		sb->st_uid = zp->zp_uid;
3179 		sb->st_gid = zp->zp_gid;
3180 		sb->st_size = zp->zp_size;
3181 	} else {
3182 		sa_hdr_phys_t *sahdrp;
3183 		int hdrsize;
3184 		size_t size = 0;
3185 		void *buf = NULL;
3186 
3187 		if (dn->dn_bonuslen != 0)
3188 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3189 		else {
3190 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
3191 				blkptr_t *bp = DN_SPILL_BLKPTR(dn);
3192 				int error;
3193 
3194 				size = BP_GET_LSIZE(bp);
3195 				buf = malloc(size);
3196 				if (buf == NULL)
3197 					error = ENOMEM;
3198 				else
3199 					error = zio_read(spa, bp, buf);
3200 
3201 				if (error != 0) {
3202 					free(buf);
3203 					return (error);
3204 				}
3205 				sahdrp = buf;
3206 			} else {
3207 				return (EIO);
3208 			}
3209 		}
3210 		hdrsize = SA_HDR_SIZE(sahdrp);
3211 		sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
3212 		    SA_MODE_OFFSET);
3213 		sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
3214 		    SA_UID_OFFSET);
3215 		sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
3216 		    SA_GID_OFFSET);
3217 		sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
3218 		    SA_SIZE_OFFSET);
3219 		free(buf);
3220 	}
3221 
3222 	return (0);
3223 }
3224 
3225 static int
3226 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
3227 {
3228 	int rc = 0;
3229 
3230 	if (dn->dn_bonustype == DMU_OT_SA) {
3231 		sa_hdr_phys_t *sahdrp = NULL;
3232 		size_t size = 0;
3233 		void *buf = NULL;
3234 		int hdrsize;
3235 		char *p;
3236 
3237 		if (dn->dn_bonuslen != 0) {
3238 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3239 		} else {
3240 			blkptr_t *bp;
3241 
3242 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
3243 				return (EIO);
3244 			bp = DN_SPILL_BLKPTR(dn);
3245 
3246 			size = BP_GET_LSIZE(bp);
3247 			buf = malloc(size);
3248 			if (buf == NULL)
3249 				rc = ENOMEM;
3250 			else
3251 				rc = zio_read(spa, bp, buf);
3252 			if (rc != 0) {
3253 				free(buf);
3254 				return (rc);
3255 			}
3256 			sahdrp = buf;
3257 		}
3258 		hdrsize = SA_HDR_SIZE(sahdrp);
3259 		p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
3260 		memcpy(path, p, psize);
3261 		free(buf);
3262 		return (0);
3263 	}
3264 	/*
3265 	 * Second test is purely to silence bogus compiler
3266 	 * warning about accessing past the end of dn_bonus.
3267 	 */
3268 	if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
3269 	    sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
3270 		memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
3271 	} else {
3272 		rc = dnode_read(spa, dn, 0, path, psize);
3273 	}
3274 	return (rc);
3275 }
3276 
3277 struct obj_list {
3278 	uint64_t		objnum;
3279 	STAILQ_ENTRY(obj_list)	entry;
3280 };
3281 
3282 /*
3283  * Lookup a file and return its dnode.
3284  */
3285 static int
3286 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
3287 {
3288 	int rc;
3289 	uint64_t objnum;
3290 	const spa_t *spa;
3291 	dnode_phys_t dn;
3292 	const char *p, *q;
3293 	char element[256];
3294 	char path[1024];
3295 	int symlinks_followed = 0;
3296 	struct stat sb;
3297 	struct obj_list *entry, *tentry;
3298 	STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
3299 
3300 	spa = mount->spa;
3301 	if (mount->objset.os_type != DMU_OST_ZFS) {
3302 		printf("ZFS: unexpected object set type %ju\n",
3303 		    (uintmax_t)mount->objset.os_type);
3304 		return (EIO);
3305 	}
3306 
3307 	if ((entry = malloc(sizeof(struct obj_list))) == NULL)
3308 		return (ENOMEM);
3309 
3310 	/*
3311 	 * Get the root directory dnode.
3312 	 */
3313 	rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
3314 	if (rc) {
3315 		free(entry);
3316 		return (rc);
3317 	}
3318 
3319 	rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof(objnum), 1, &objnum);
3320 	if (rc) {
3321 		free(entry);
3322 		return (rc);
3323 	}
3324 	entry->objnum = objnum;
3325 	STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3326 
3327 	rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3328 	if (rc != 0)
3329 		goto done;
3330 
3331 	p = upath;
3332 	while (p && *p) {
3333 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3334 		if (rc != 0)
3335 			goto done;
3336 
3337 		while (*p == '/')
3338 			p++;
3339 		if (*p == '\0')
3340 			break;
3341 		q = p;
3342 		while (*q != '\0' && *q != '/')
3343 			q++;
3344 
3345 		/* skip dot */
3346 		if (p + 1 == q && p[0] == '.') {
3347 			p++;
3348 			continue;
3349 		}
3350 		/* double dot */
3351 		if (p + 2 == q && p[0] == '.' && p[1] == '.') {
3352 			p += 2;
3353 			if (STAILQ_FIRST(&on_cache) ==
3354 			    STAILQ_LAST(&on_cache, obj_list, entry)) {
3355 				rc = ENOENT;
3356 				goto done;
3357 			}
3358 			entry = STAILQ_FIRST(&on_cache);
3359 			STAILQ_REMOVE_HEAD(&on_cache, entry);
3360 			free(entry);
3361 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3362 			continue;
3363 		}
3364 		if (q - p + 1 > sizeof(element)) {
3365 			rc = ENAMETOOLONG;
3366 			goto done;
3367 		}
3368 		memcpy(element, p, q - p);
3369 		element[q - p] = 0;
3370 		p = q;
3371 
3372 		if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
3373 			goto done;
3374 		if (!S_ISDIR(sb.st_mode)) {
3375 			rc = ENOTDIR;
3376 			goto done;
3377 		}
3378 
3379 		rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum);
3380 		if (rc)
3381 			goto done;
3382 		objnum = ZFS_DIRENT_OBJ(objnum);
3383 
3384 		if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
3385 			rc = ENOMEM;
3386 			goto done;
3387 		}
3388 		entry->objnum = objnum;
3389 		STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3390 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3391 		if (rc)
3392 			goto done;
3393 
3394 		/*
3395 		 * Check for symlink.
3396 		 */
3397 		rc = zfs_dnode_stat(spa, &dn, &sb);
3398 		if (rc)
3399 			goto done;
3400 		if (S_ISLNK(sb.st_mode)) {
3401 			if (symlinks_followed > 10) {
3402 				rc = EMLINK;
3403 				goto done;
3404 			}
3405 			symlinks_followed++;
3406 
3407 			/*
3408 			 * Read the link value and copy the tail of our
3409 			 * current path onto the end.
3410 			 */
3411 			if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
3412 				rc = ENAMETOOLONG;
3413 				goto done;
3414 			}
3415 			strcpy(&path[sb.st_size], p);
3416 
3417 			rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
3418 			if (rc != 0)
3419 				goto done;
3420 
3421 			/*
3422 			 * Restart with the new path, starting either at
3423 			 * the root or at the parent depending whether or
3424 			 * not the link is relative.
3425 			 */
3426 			p = path;
3427 			if (*p == '/') {
3428 				while (STAILQ_FIRST(&on_cache) !=
3429 				    STAILQ_LAST(&on_cache, obj_list, entry)) {
3430 					entry = STAILQ_FIRST(&on_cache);
3431 					STAILQ_REMOVE_HEAD(&on_cache, entry);
3432 					free(entry);
3433 				}
3434 			} else {
3435 				entry = STAILQ_FIRST(&on_cache);
3436 				STAILQ_REMOVE_HEAD(&on_cache, entry);
3437 				free(entry);
3438 			}
3439 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3440 		}
3441 	}
3442 
3443 	*dnode = dn;
3444 done:
3445 	STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)
3446 		free(entry);
3447 	return (rc);
3448 }
3449