xref: /freebsd/stand/libsa/zfs/zfsimpl.c (revision a67cc943273ba7cba2f78e33fc5897e1fbecd462)
1 /*-
2  * Copyright (c) 2007 Doug Rabson
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 /*
31  *	Stand-alone ZFS file reader.
32  */
33 
34 #include <sys/endian.h>
35 #include <sys/stat.h>
36 #include <sys/stdint.h>
37 #include <sys/list.h>
38 #include <machine/_inttypes.h>
39 
40 #include "zfsimpl.h"
41 #include "zfssubr.c"
42 
43 
44 struct zfsmount {
45 	const spa_t	*spa;
46 	objset_phys_t	objset;
47 	uint64_t	rootobj;
48 };
49 static struct zfsmount zfsmount __unused;
50 
51 /*
52  * The indirect_child_t represents the vdev that we will read from, when we
53  * need to read all copies of the data (e.g. for scrub or reconstruction).
54  * For plain (non-mirror) top-level vdevs (i.e. is_vdev is not a mirror),
55  * ic_vdev is the same as is_vdev.  However, for mirror top-level vdevs,
56  * ic_vdev is a child of the mirror.
57  */
58 typedef struct indirect_child {
59 	void *ic_data;
60 	vdev_t *ic_vdev;
61 } indirect_child_t;
62 
63 /*
64  * The indirect_split_t represents one mapped segment of an i/o to the
65  * indirect vdev. For non-split (contiguously-mapped) blocks, there will be
66  * only one indirect_split_t, with is_split_offset==0 and is_size==io_size.
67  * For split blocks, there will be several of these.
68  */
69 typedef struct indirect_split {
70 	list_node_t is_node; /* link on iv_splits */
71 
72 	/*
73 	 * is_split_offset is the offset into the i/o.
74 	 * This is the sum of the previous splits' is_size's.
75 	 */
76 	uint64_t is_split_offset;
77 
78 	vdev_t *is_vdev; /* top-level vdev */
79 	uint64_t is_target_offset; /* offset on is_vdev */
80 	uint64_t is_size;
81 	int is_children; /* number of entries in is_child[] */
82 
83 	/*
84 	 * is_good_child is the child that we are currently using to
85 	 * attempt reconstruction.
86 	 */
87 	int is_good_child;
88 
89 	indirect_child_t is_child[1]; /* variable-length */
90 } indirect_split_t;
91 
92 /*
93  * The indirect_vsd_t is associated with each i/o to the indirect vdev.
94  * It is the "Vdev-Specific Data" in the zio_t's io_vsd.
95  */
96 typedef struct indirect_vsd {
97 	boolean_t iv_split_block;
98 	boolean_t iv_reconstruct;
99 
100 	list_t iv_splits; /* list of indirect_split_t's */
101 } indirect_vsd_t;
102 
103 /*
104  * List of all vdevs, chained through v_alllink.
105  */
106 static vdev_list_t zfs_vdevs;
107 
108 /*
109  * List of ZFS features supported for read
110  */
111 static const char *features_for_read[] = {
112 	"org.illumos:lz4_compress",
113 	"com.delphix:hole_birth",
114 	"com.delphix:extensible_dataset",
115 	"com.delphix:embedded_data",
116 	"org.open-zfs:large_blocks",
117 	"org.illumos:sha512",
118 	"org.illumos:skein",
119 	"org.zfsonlinux:large_dnode",
120 	"com.joyent:multi_vdev_crash_dump",
121 	"com.delphix:spacemap_histogram",
122 	"com.delphix:zpool_checkpoint",
123 	"com.delphix:spacemap_v2",
124 	"com.datto:encryption",
125 	"org.zfsonlinux:allocation_classes",
126 	"com.datto:resilver_defer",
127 	"com.delphix:device_removal",
128 	"com.delphix:obsolete_counts",
129 	"com.intel:allocation_classes",
130 	"org.freebsd:zstd_compress",
131 	NULL
132 };
133 
134 /*
135  * List of all pools, chained through spa_link.
136  */
137 static spa_list_t zfs_pools;
138 
139 static const dnode_phys_t *dnode_cache_obj;
140 static uint64_t dnode_cache_bn;
141 static char *dnode_cache_buf;
142 
143 static int zio_read(const spa_t *spa, const blkptr_t *bp, void *buf);
144 static int zfs_get_root(const spa_t *spa, uint64_t *objid);
145 static int zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result);
146 static int zap_lookup(const spa_t *spa, const dnode_phys_t *dnode,
147     const char *name, uint64_t integer_size, uint64_t num_integers,
148     void *value);
149 static int objset_get_dnode(const spa_t *, const objset_phys_t *, uint64_t,
150     dnode_phys_t *);
151 static int dnode_read(const spa_t *, const dnode_phys_t *, off_t, void *,
152     size_t);
153 static int vdev_indirect_read(vdev_t *, const blkptr_t *, void *, off_t,
154     size_t);
155 static int vdev_mirror_read(vdev_t *, const blkptr_t *, void *, off_t, size_t);
156 vdev_indirect_mapping_t *vdev_indirect_mapping_open(spa_t *, objset_phys_t *,
157     uint64_t);
158 vdev_indirect_mapping_entry_phys_t *
159     vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *, uint64_t,
160     uint64_t, uint64_t *);
161 
162 static void
163 zfs_init(void)
164 {
165 	STAILQ_INIT(&zfs_vdevs);
166 	STAILQ_INIT(&zfs_pools);
167 
168 	dnode_cache_buf = malloc(SPA_MAXBLOCKSIZE);
169 
170 	zfs_init_crc();
171 }
172 
173 static int
174 nvlist_check_features_for_read(nvlist_t *nvl)
175 {
176 	nvlist_t *features = NULL;
177 	nvs_data_t *data;
178 	nvp_header_t *nvp;
179 	nv_string_t *nvp_name;
180 	int rc;
181 
182 	rc = nvlist_find(nvl, ZPOOL_CONFIG_FEATURES_FOR_READ,
183 	    DATA_TYPE_NVLIST, NULL, &features, NULL);
184 	if (rc != 0)
185 		return (rc);
186 
187 	data = (nvs_data_t *)features->nv_data;
188 	nvp = &data->nvl_pair;	/* first pair in nvlist */
189 
190 	while (nvp->encoded_size != 0 && nvp->decoded_size != 0) {
191 		int i, found;
192 
193 		nvp_name = (nv_string_t *)((uintptr_t)nvp + sizeof(*nvp));
194 		found = 0;
195 
196 		for (i = 0; features_for_read[i] != NULL; i++) {
197 			if (memcmp(nvp_name->nv_data, features_for_read[i],
198 			    nvp_name->nv_size) == 0) {
199 				found = 1;
200 				break;
201 			}
202 		}
203 
204 		if (!found) {
205 			printf("ZFS: unsupported feature: %.*s\n",
206 			    nvp_name->nv_size, nvp_name->nv_data);
207 			rc = EIO;
208 		}
209 		nvp = (nvp_header_t *)((uint8_t *)nvp + nvp->encoded_size);
210 	}
211 	nvlist_destroy(features);
212 
213 	return (rc);
214 }
215 
216 static int
217 vdev_read_phys(vdev_t *vdev, const blkptr_t *bp, void *buf,
218     off_t offset, size_t size)
219 {
220 	size_t psize;
221 	int rc;
222 
223 	if (!vdev->v_phys_read)
224 		return (EIO);
225 
226 	if (bp) {
227 		psize = BP_GET_PSIZE(bp);
228 	} else {
229 		psize = size;
230 	}
231 
232 	rc = vdev->v_phys_read(vdev, vdev->v_read_priv, offset, buf, psize);
233 	if (rc == 0) {
234 		if (bp != NULL)
235 			rc = zio_checksum_verify(vdev->v_spa, bp, buf);
236 	}
237 
238 	return (rc);
239 }
240 
241 typedef struct remap_segment {
242 	vdev_t *rs_vd;
243 	uint64_t rs_offset;
244 	uint64_t rs_asize;
245 	uint64_t rs_split_offset;
246 	list_node_t rs_node;
247 } remap_segment_t;
248 
249 static remap_segment_t *
250 rs_alloc(vdev_t *vd, uint64_t offset, uint64_t asize, uint64_t split_offset)
251 {
252 	remap_segment_t *rs = malloc(sizeof (remap_segment_t));
253 
254 	if (rs != NULL) {
255 		rs->rs_vd = vd;
256 		rs->rs_offset = offset;
257 		rs->rs_asize = asize;
258 		rs->rs_split_offset = split_offset;
259 	}
260 
261 	return (rs);
262 }
263 
264 vdev_indirect_mapping_t *
265 vdev_indirect_mapping_open(spa_t *spa, objset_phys_t *os,
266     uint64_t mapping_object)
267 {
268 	vdev_indirect_mapping_t *vim;
269 	vdev_indirect_mapping_phys_t *vim_phys;
270 	int rc;
271 
272 	vim = calloc(1, sizeof (*vim));
273 	if (vim == NULL)
274 		return (NULL);
275 
276 	vim->vim_dn = calloc(1, sizeof (*vim->vim_dn));
277 	if (vim->vim_dn == NULL) {
278 		free(vim);
279 		return (NULL);
280 	}
281 
282 	rc = objset_get_dnode(spa, os, mapping_object, vim->vim_dn);
283 	if (rc != 0) {
284 		free(vim->vim_dn);
285 		free(vim);
286 		return (NULL);
287 	}
288 
289 	vim->vim_spa = spa;
290 	vim->vim_phys = malloc(sizeof (*vim->vim_phys));
291 	if (vim->vim_phys == NULL) {
292 		free(vim->vim_dn);
293 		free(vim);
294 		return (NULL);
295 	}
296 
297 	vim_phys = (vdev_indirect_mapping_phys_t *)DN_BONUS(vim->vim_dn);
298 	*vim->vim_phys = *vim_phys;
299 
300 	vim->vim_objset = os;
301 	vim->vim_object = mapping_object;
302 	vim->vim_entries = NULL;
303 
304 	vim->vim_havecounts =
305 	    (vim->vim_dn->dn_bonuslen > VDEV_INDIRECT_MAPPING_SIZE_V0);
306 
307 	return (vim);
308 }
309 
310 /*
311  * Compare an offset with an indirect mapping entry; there are three
312  * possible scenarios:
313  *
314  *     1. The offset is "less than" the mapping entry; meaning the
315  *        offset is less than the source offset of the mapping entry. In
316  *        this case, there is no overlap between the offset and the
317  *        mapping entry and -1 will be returned.
318  *
319  *     2. The offset is "greater than" the mapping entry; meaning the
320  *        offset is greater than the mapping entry's source offset plus
321  *        the entry's size. In this case, there is no overlap between
322  *        the offset and the mapping entry and 1 will be returned.
323  *
324  *        NOTE: If the offset is actually equal to the entry's offset
325  *        plus size, this is considered to be "greater" than the entry,
326  *        and this case applies (i.e. 1 will be returned). Thus, the
327  *        entry's "range" can be considered to be inclusive at its
328  *        start, but exclusive at its end: e.g. [src, src + size).
329  *
330  *     3. The last case to consider is if the offset actually falls
331  *        within the mapping entry's range. If this is the case, the
332  *        offset is considered to be "equal to" the mapping entry and
333  *        0 will be returned.
334  *
335  *        NOTE: If the offset is equal to the entry's source offset,
336  *        this case applies and 0 will be returned. If the offset is
337  *        equal to the entry's source plus its size, this case does
338  *        *not* apply (see "NOTE" above for scenario 2), and 1 will be
339  *        returned.
340  */
341 static int
342 dva_mapping_overlap_compare(const void *v_key, const void *v_array_elem)
343 {
344 	const uint64_t *key = v_key;
345 	const vdev_indirect_mapping_entry_phys_t *array_elem =
346 	    v_array_elem;
347 	uint64_t src_offset = DVA_MAPPING_GET_SRC_OFFSET(array_elem);
348 
349 	if (*key < src_offset) {
350 		return (-1);
351 	} else if (*key < src_offset + DVA_GET_ASIZE(&array_elem->vimep_dst)) {
352 		return (0);
353 	} else {
354 		return (1);
355 	}
356 }
357 
358 /*
359  * Return array entry.
360  */
361 static vdev_indirect_mapping_entry_phys_t *
362 vdev_indirect_mapping_entry(vdev_indirect_mapping_t *vim, uint64_t index)
363 {
364 	uint64_t size;
365 	off_t offset = 0;
366 	int rc;
367 
368 	if (vim->vim_phys->vimp_num_entries == 0)
369 		return (NULL);
370 
371 	if (vim->vim_entries == NULL) {
372 		uint64_t bsize;
373 
374 		bsize = vim->vim_dn->dn_datablkszsec << SPA_MINBLOCKSHIFT;
375 		size = vim->vim_phys->vimp_num_entries *
376 		    sizeof (*vim->vim_entries);
377 		if (size > bsize) {
378 			size = bsize / sizeof (*vim->vim_entries);
379 			size *= sizeof (*vim->vim_entries);
380 		}
381 		vim->vim_entries = malloc(size);
382 		if (vim->vim_entries == NULL)
383 			return (NULL);
384 		vim->vim_num_entries = size / sizeof (*vim->vim_entries);
385 		offset = index * sizeof (*vim->vim_entries);
386 	}
387 
388 	/* We have data in vim_entries */
389 	if (offset == 0) {
390 		if (index >= vim->vim_entry_offset &&
391 		    index <= vim->vim_entry_offset + vim->vim_num_entries) {
392 			index -= vim->vim_entry_offset;
393 			return (&vim->vim_entries[index]);
394 		}
395 		offset = index * sizeof (*vim->vim_entries);
396 	}
397 
398 	vim->vim_entry_offset = index;
399 	size = vim->vim_num_entries * sizeof (*vim->vim_entries);
400 	rc = dnode_read(vim->vim_spa, vim->vim_dn, offset, vim->vim_entries,
401 	    size);
402 	if (rc != 0) {
403 		/* Read error, invalidate vim_entries. */
404 		free(vim->vim_entries);
405 		vim->vim_entries = NULL;
406 		return (NULL);
407 	}
408 	index -= vim->vim_entry_offset;
409 	return (&vim->vim_entries[index]);
410 }
411 
412 /*
413  * Returns the mapping entry for the given offset.
414  *
415  * It's possible that the given offset will not be in the mapping table
416  * (i.e. no mapping entries contain this offset), in which case, the
417  * return value value depends on the "next_if_missing" parameter.
418  *
419  * If the offset is not found in the table and "next_if_missing" is
420  * B_FALSE, then NULL will always be returned. The behavior is intended
421  * to allow consumers to get the entry corresponding to the offset
422  * parameter, iff the offset overlaps with an entry in the table.
423  *
424  * If the offset is not found in the table and "next_if_missing" is
425  * B_TRUE, then the entry nearest to the given offset will be returned,
426  * such that the entry's source offset is greater than the offset
427  * passed in (i.e. the "next" mapping entry in the table is returned, if
428  * the offset is missing from the table). If there are no entries whose
429  * source offset is greater than the passed in offset, NULL is returned.
430  */
431 static vdev_indirect_mapping_entry_phys_t *
432 vdev_indirect_mapping_entry_for_offset(vdev_indirect_mapping_t *vim,
433     uint64_t offset)
434 {
435 	ASSERT(vim->vim_phys->vimp_num_entries > 0);
436 
437 	vdev_indirect_mapping_entry_phys_t *entry;
438 
439 	uint64_t last = vim->vim_phys->vimp_num_entries - 1;
440 	uint64_t base = 0;
441 
442 	/*
443 	 * We don't define these inside of the while loop because we use
444 	 * their value in the case that offset isn't in the mapping.
445 	 */
446 	uint64_t mid;
447 	int result;
448 
449 	while (last >= base) {
450 		mid = base + ((last - base) >> 1);
451 
452 		entry = vdev_indirect_mapping_entry(vim, mid);
453 		if (entry == NULL)
454 			break;
455 		result = dva_mapping_overlap_compare(&offset, entry);
456 
457 		if (result == 0) {
458 			break;
459 		} else if (result < 0) {
460 			last = mid - 1;
461 		} else {
462 			base = mid + 1;
463 		}
464 	}
465 	return (entry);
466 }
467 
468 /*
469  * Given an indirect vdev and an extent on that vdev, it duplicates the
470  * physical entries of the indirect mapping that correspond to the extent
471  * to a new array and returns a pointer to it. In addition, copied_entries
472  * is populated with the number of mapping entries that were duplicated.
473  *
474  * Finally, since we are doing an allocation, it is up to the caller to
475  * free the array allocated in this function.
476  */
477 vdev_indirect_mapping_entry_phys_t *
478 vdev_indirect_mapping_duplicate_adjacent_entries(vdev_t *vd, uint64_t offset,
479     uint64_t asize, uint64_t *copied_entries)
480 {
481 	vdev_indirect_mapping_entry_phys_t *duplicate_mappings = NULL;
482 	vdev_indirect_mapping_t *vim = vd->v_mapping;
483 	uint64_t entries = 0;
484 
485 	vdev_indirect_mapping_entry_phys_t *first_mapping =
486 	    vdev_indirect_mapping_entry_for_offset(vim, offset);
487 	ASSERT3P(first_mapping, !=, NULL);
488 
489 	vdev_indirect_mapping_entry_phys_t *m = first_mapping;
490 	while (asize > 0) {
491 		uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
492 		uint64_t inner_offset = offset - DVA_MAPPING_GET_SRC_OFFSET(m);
493 		uint64_t inner_size = MIN(asize, size - inner_offset);
494 
495 		offset += inner_size;
496 		asize -= inner_size;
497 		entries++;
498 		m++;
499 	}
500 
501 	size_t copy_length = entries * sizeof (*first_mapping);
502 	duplicate_mappings = malloc(copy_length);
503 	if (duplicate_mappings != NULL)
504 		bcopy(first_mapping, duplicate_mappings, copy_length);
505 	else
506 		entries = 0;
507 
508 	*copied_entries = entries;
509 
510 	return (duplicate_mappings);
511 }
512 
513 static vdev_t *
514 vdev_lookup_top(spa_t *spa, uint64_t vdev)
515 {
516 	vdev_t *rvd;
517 	vdev_list_t *vlist;
518 
519 	vlist = &spa->spa_root_vdev->v_children;
520 	STAILQ_FOREACH(rvd, vlist, v_childlink)
521 		if (rvd->v_id == vdev)
522 			break;
523 
524 	return (rvd);
525 }
526 
527 /*
528  * This is a callback for vdev_indirect_remap() which allocates an
529  * indirect_split_t for each split segment and adds it to iv_splits.
530  */
531 static void
532 vdev_indirect_gather_splits(uint64_t split_offset, vdev_t *vd, uint64_t offset,
533     uint64_t size, void *arg)
534 {
535 	int n = 1;
536 	zio_t *zio = arg;
537 	indirect_vsd_t *iv = zio->io_vsd;
538 
539 	if (vd->v_read == vdev_indirect_read)
540 		return;
541 
542 	if (vd->v_read == vdev_mirror_read)
543 		n = vd->v_nchildren;
544 
545 	indirect_split_t *is =
546 	    malloc(offsetof(indirect_split_t, is_child[n]));
547 	if (is == NULL) {
548 		zio->io_error = ENOMEM;
549 		return;
550 	}
551 	bzero(is, offsetof(indirect_split_t, is_child[n]));
552 
553 	is->is_children = n;
554 	is->is_size = size;
555 	is->is_split_offset = split_offset;
556 	is->is_target_offset = offset;
557 	is->is_vdev = vd;
558 
559 	/*
560 	 * Note that we only consider multiple copies of the data for
561 	 * *mirror* vdevs.  We don't for "replacing" or "spare" vdevs, even
562 	 * though they use the same ops as mirror, because there's only one
563 	 * "good" copy under the replacing/spare.
564 	 */
565 	if (vd->v_read == vdev_mirror_read) {
566 		int i = 0;
567 		vdev_t *kid;
568 
569 		STAILQ_FOREACH(kid, &vd->v_children, v_childlink) {
570 			is->is_child[i++].ic_vdev = kid;
571 		}
572 	} else {
573 		is->is_child[0].ic_vdev = vd;
574 	}
575 
576 	list_insert_tail(&iv->iv_splits, is);
577 }
578 
579 static void
580 vdev_indirect_remap(vdev_t *vd, uint64_t offset, uint64_t asize, void *arg)
581 {
582 	list_t stack;
583 	spa_t *spa = vd->v_spa;
584 	zio_t *zio = arg;
585 	remap_segment_t *rs;
586 
587 	list_create(&stack, sizeof (remap_segment_t),
588 	    offsetof(remap_segment_t, rs_node));
589 
590 	rs = rs_alloc(vd, offset, asize, 0);
591 	if (rs == NULL) {
592 		printf("vdev_indirect_remap: out of memory.\n");
593 		zio->io_error = ENOMEM;
594 	}
595 	for (; rs != NULL; rs = list_remove_head(&stack)) {
596 		vdev_t *v = rs->rs_vd;
597 		uint64_t num_entries = 0;
598 		/* vdev_indirect_mapping_t *vim = v->v_mapping; */
599 		vdev_indirect_mapping_entry_phys_t *mapping =
600 		    vdev_indirect_mapping_duplicate_adjacent_entries(v,
601 		    rs->rs_offset, rs->rs_asize, &num_entries);
602 
603 		if (num_entries == 0)
604 			zio->io_error = ENOMEM;
605 
606 		for (uint64_t i = 0; i < num_entries; i++) {
607 			vdev_indirect_mapping_entry_phys_t *m = &mapping[i];
608 			uint64_t size = DVA_GET_ASIZE(&m->vimep_dst);
609 			uint64_t dst_offset = DVA_GET_OFFSET(&m->vimep_dst);
610 			uint64_t dst_vdev = DVA_GET_VDEV(&m->vimep_dst);
611 			uint64_t inner_offset = rs->rs_offset -
612 			    DVA_MAPPING_GET_SRC_OFFSET(m);
613 			uint64_t inner_size =
614 			    MIN(rs->rs_asize, size - inner_offset);
615 			vdev_t *dst_v = vdev_lookup_top(spa, dst_vdev);
616 
617 			if (dst_v->v_read == vdev_indirect_read) {
618 				remap_segment_t *o;
619 
620 				o = rs_alloc(dst_v, dst_offset + inner_offset,
621 				    inner_size, rs->rs_split_offset);
622 				if (o == NULL) {
623 					printf("vdev_indirect_remap: "
624 					    "out of memory.\n");
625 					zio->io_error = ENOMEM;
626 					break;
627 				}
628 
629 				list_insert_head(&stack, o);
630 			}
631 			vdev_indirect_gather_splits(rs->rs_split_offset, dst_v,
632 			    dst_offset + inner_offset,
633 			    inner_size, arg);
634 
635 			/*
636 			 * vdev_indirect_gather_splits can have memory
637 			 * allocation error, we can not recover from it.
638 			 */
639 			if (zio->io_error != 0)
640 				break;
641 			rs->rs_offset += inner_size;
642 			rs->rs_asize -= inner_size;
643 			rs->rs_split_offset += inner_size;
644 		}
645 
646 		free(mapping);
647 		free(rs);
648 		if (zio->io_error != 0)
649 			break;
650 	}
651 
652 	list_destroy(&stack);
653 }
654 
655 static void
656 vdev_indirect_map_free(zio_t *zio)
657 {
658 	indirect_vsd_t *iv = zio->io_vsd;
659 	indirect_split_t *is;
660 
661 	while ((is = list_head(&iv->iv_splits)) != NULL) {
662 		for (int c = 0; c < is->is_children; c++) {
663 			indirect_child_t *ic = &is->is_child[c];
664 			free(ic->ic_data);
665 		}
666 		list_remove(&iv->iv_splits, is);
667 		free(is);
668 	}
669 	free(iv);
670 }
671 
672 static int
673 vdev_indirect_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
674     off_t offset, size_t bytes)
675 {
676 	zio_t zio;
677 	spa_t *spa = vdev->v_spa;
678 	indirect_vsd_t *iv;
679 	indirect_split_t *first;
680 	int rc = EIO;
681 
682 	iv = calloc(1, sizeof(*iv));
683 	if (iv == NULL)
684 		return (ENOMEM);
685 
686 	list_create(&iv->iv_splits,
687 	    sizeof (indirect_split_t), offsetof(indirect_split_t, is_node));
688 
689 	bzero(&zio, sizeof(zio));
690 	zio.io_spa = spa;
691 	zio.io_bp = (blkptr_t *)bp;
692 	zio.io_data = buf;
693 	zio.io_size = bytes;
694 	zio.io_offset = offset;
695 	zio.io_vd = vdev;
696 	zio.io_vsd = iv;
697 
698 	if (vdev->v_mapping == NULL) {
699 		vdev_indirect_config_t *vic;
700 
701 		vic = &vdev->vdev_indirect_config;
702 		vdev->v_mapping = vdev_indirect_mapping_open(spa,
703 		    spa->spa_mos, vic->vic_mapping_object);
704 	}
705 
706 	vdev_indirect_remap(vdev, offset, bytes, &zio);
707 	if (zio.io_error != 0)
708 		return (zio.io_error);
709 
710 	first = list_head(&iv->iv_splits);
711 	if (first->is_size == zio.io_size) {
712 		/*
713 		 * This is not a split block; we are pointing to the entire
714 		 * data, which will checksum the same as the original data.
715 		 * Pass the BP down so that the child i/o can verify the
716 		 * checksum, and try a different location if available
717 		 * (e.g. on a mirror).
718 		 *
719 		 * While this special case could be handled the same as the
720 		 * general (split block) case, doing it this way ensures
721 		 * that the vast majority of blocks on indirect vdevs
722 		 * (which are not split) are handled identically to blocks
723 		 * on non-indirect vdevs.  This allows us to be less strict
724 		 * about performance in the general (but rare) case.
725 		 */
726 		rc = first->is_vdev->v_read(first->is_vdev, zio.io_bp,
727 		    zio.io_data, first->is_target_offset, bytes);
728 	} else {
729 		iv->iv_split_block = B_TRUE;
730 		/*
731 		 * Read one copy of each split segment, from the
732 		 * top-level vdev.  Since we don't know the
733 		 * checksum of each split individually, the child
734 		 * zio can't ensure that we get the right data.
735 		 * E.g. if it's a mirror, it will just read from a
736 		 * random (healthy) leaf vdev.  We have to verify
737 		 * the checksum in vdev_indirect_io_done().
738 		 */
739 		for (indirect_split_t *is = list_head(&iv->iv_splits);
740 		    is != NULL; is = list_next(&iv->iv_splits, is)) {
741 			char *ptr = zio.io_data;
742 
743 			rc = is->is_vdev->v_read(is->is_vdev, zio.io_bp,
744 			    ptr + is->is_split_offset, is->is_target_offset,
745 			    is->is_size);
746 		}
747 		if (zio_checksum_verify(spa, zio.io_bp, zio.io_data))
748 			rc = ECKSUM;
749 		else
750 			rc = 0;
751 	}
752 
753 	vdev_indirect_map_free(&zio);
754 	if (rc == 0)
755 		rc = zio.io_error;
756 
757 	return (rc);
758 }
759 
760 static int
761 vdev_disk_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
762     off_t offset, size_t bytes)
763 {
764 
765 	return (vdev_read_phys(vdev, bp, buf,
766 	    offset + VDEV_LABEL_START_SIZE, bytes));
767 }
768 
769 static int
770 vdev_missing_read(vdev_t *vdev __unused, const blkptr_t *bp __unused,
771     void *buf __unused, off_t offset __unused, size_t bytes __unused)
772 {
773 
774 	return (ENOTSUP);
775 }
776 
777 static int
778 vdev_mirror_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
779     off_t offset, size_t bytes)
780 {
781 	vdev_t *kid;
782 	int rc;
783 
784 	rc = EIO;
785 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
786 		if (kid->v_state != VDEV_STATE_HEALTHY)
787 			continue;
788 		rc = kid->v_read(kid, bp, buf, offset, bytes);
789 		if (!rc)
790 			return (0);
791 	}
792 
793 	return (rc);
794 }
795 
796 static int
797 vdev_replacing_read(vdev_t *vdev, const blkptr_t *bp, void *buf,
798     off_t offset, size_t bytes)
799 {
800 	vdev_t *kid;
801 
802 	/*
803 	 * Here we should have two kids:
804 	 * First one which is the one we are replacing and we can trust
805 	 * only this one to have valid data, but it might not be present.
806 	 * Second one is that one we are replacing with. It is most likely
807 	 * healthy, but we can't trust it has needed data, so we won't use it.
808 	 */
809 	kid = STAILQ_FIRST(&vdev->v_children);
810 	if (kid == NULL)
811 		return (EIO);
812 	if (kid->v_state != VDEV_STATE_HEALTHY)
813 		return (EIO);
814 	return (kid->v_read(kid, bp, buf, offset, bytes));
815 }
816 
817 static vdev_t *
818 vdev_find(uint64_t guid)
819 {
820 	vdev_t *vdev;
821 
822 	STAILQ_FOREACH(vdev, &zfs_vdevs, v_alllink)
823 		if (vdev->v_guid == guid)
824 			return (vdev);
825 
826 	return (0);
827 }
828 
829 static vdev_t *
830 vdev_create(uint64_t guid, vdev_read_t *_read)
831 {
832 	vdev_t *vdev;
833 	vdev_indirect_config_t *vic;
834 
835 	vdev = calloc(1, sizeof(vdev_t));
836 	if (vdev != NULL) {
837 		STAILQ_INIT(&vdev->v_children);
838 		vdev->v_guid = guid;
839 		vdev->v_read = _read;
840 
841 		/*
842 		 * root vdev has no read function, we use this fact to
843 		 * skip setting up data we do not need for root vdev.
844 		 * We only point root vdev from spa.
845 		 */
846 		if (_read != NULL) {
847 			vic = &vdev->vdev_indirect_config;
848 			vic->vic_prev_indirect_vdev = UINT64_MAX;
849 			STAILQ_INSERT_TAIL(&zfs_vdevs, vdev, v_alllink);
850 		}
851 	}
852 
853 	return (vdev);
854 }
855 
856 static void
857 vdev_set_initial_state(vdev_t *vdev, const nvlist_t *nvlist)
858 {
859 	uint64_t is_offline, is_faulted, is_degraded, is_removed, isnt_present;
860 	uint64_t is_log;
861 
862 	is_offline = is_removed = is_faulted = is_degraded = isnt_present = 0;
863 	is_log = 0;
864 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_OFFLINE, DATA_TYPE_UINT64, NULL,
865 	    &is_offline, NULL);
866 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_REMOVED, DATA_TYPE_UINT64, NULL,
867 	    &is_removed, NULL);
868 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_FAULTED, DATA_TYPE_UINT64, NULL,
869 	    &is_faulted, NULL);
870 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_DEGRADED, DATA_TYPE_UINT64,
871 	    NULL, &is_degraded, NULL);
872 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_NOT_PRESENT, DATA_TYPE_UINT64,
873 	    NULL, &isnt_present, NULL);
874 	(void) nvlist_find(nvlist, ZPOOL_CONFIG_IS_LOG, DATA_TYPE_UINT64, NULL,
875 	    &is_log, NULL);
876 
877 	if (is_offline != 0)
878 		vdev->v_state = VDEV_STATE_OFFLINE;
879 	else if (is_removed != 0)
880 		vdev->v_state = VDEV_STATE_REMOVED;
881 	else if (is_faulted != 0)
882 		vdev->v_state = VDEV_STATE_FAULTED;
883 	else if (is_degraded != 0)
884 		vdev->v_state = VDEV_STATE_DEGRADED;
885 	else if (isnt_present != 0)
886 		vdev->v_state = VDEV_STATE_CANT_OPEN;
887 
888 	vdev->v_islog = is_log != 0;
889 }
890 
891 static int
892 vdev_init(uint64_t guid, const nvlist_t *nvlist, vdev_t **vdevp)
893 {
894 	uint64_t id, ashift, asize, nparity;
895 	const char *path;
896 	const char *type;
897 	int len, pathlen;
898 	char *name;
899 	vdev_t *vdev;
900 
901 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ID, DATA_TYPE_UINT64, NULL, &id,
902 	    NULL) ||
903 	    nvlist_find(nvlist, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING, NULL,
904 	    &type, &len)) {
905 		return (ENOENT);
906 	}
907 
908 	if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 &&
909 	    memcmp(type, VDEV_TYPE_DISK, len) != 0 &&
910 #ifdef ZFS_TEST
911 	    memcmp(type, VDEV_TYPE_FILE, len) != 0 &&
912 #endif
913 	    memcmp(type, VDEV_TYPE_RAIDZ, len) != 0 &&
914 	    memcmp(type, VDEV_TYPE_INDIRECT, len) != 0 &&
915 	    memcmp(type, VDEV_TYPE_REPLACING, len) != 0 &&
916 	    memcmp(type, VDEV_TYPE_HOLE, len) != 0) {
917 		printf("ZFS: can only boot from disk, mirror, raidz1, "
918 		    "raidz2 and raidz3 vdevs, got: %.*s\n", len, type);
919 		return (EIO);
920 	}
921 
922 	if (memcmp(type, VDEV_TYPE_MIRROR, len) == 0)
923 		vdev = vdev_create(guid, vdev_mirror_read);
924 	else if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0)
925 		vdev = vdev_create(guid, vdev_raidz_read);
926 	else if (memcmp(type, VDEV_TYPE_REPLACING, len) == 0)
927 		vdev = vdev_create(guid, vdev_replacing_read);
928 	else if (memcmp(type, VDEV_TYPE_INDIRECT, len) == 0) {
929 		vdev_indirect_config_t *vic;
930 
931 		vdev = vdev_create(guid, vdev_indirect_read);
932 		if (vdev != NULL) {
933 			vdev->v_state = VDEV_STATE_HEALTHY;
934 			vic = &vdev->vdev_indirect_config;
935 
936 			nvlist_find(nvlist,
937 			    ZPOOL_CONFIG_INDIRECT_OBJECT,
938 			    DATA_TYPE_UINT64,
939 			    NULL, &vic->vic_mapping_object, NULL);
940 			nvlist_find(nvlist,
941 			    ZPOOL_CONFIG_INDIRECT_BIRTHS,
942 			    DATA_TYPE_UINT64,
943 			    NULL, &vic->vic_births_object, NULL);
944 			nvlist_find(nvlist,
945 			    ZPOOL_CONFIG_PREV_INDIRECT_VDEV,
946 			    DATA_TYPE_UINT64,
947 			    NULL, &vic->vic_prev_indirect_vdev, NULL);
948 		}
949 	} else if (memcmp(type, VDEV_TYPE_HOLE, len) == 0) {
950 		vdev = vdev_create(guid, vdev_missing_read);
951 	} else {
952 		vdev = vdev_create(guid, vdev_disk_read);
953 	}
954 
955 	if (vdev == NULL)
956 		return (ENOMEM);
957 
958 	vdev_set_initial_state(vdev, nvlist);
959 	vdev->v_id = id;
960 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASHIFT,
961 	    DATA_TYPE_UINT64, NULL, &ashift, NULL) == 0)
962 		vdev->v_ashift = ashift;
963 
964 	if (nvlist_find(nvlist, ZPOOL_CONFIG_ASIZE,
965 	    DATA_TYPE_UINT64, NULL, &asize, NULL) == 0) {
966 		vdev->v_psize = asize +
967 		    VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
968 	}
969 
970 	if (nvlist_find(nvlist, ZPOOL_CONFIG_NPARITY,
971 	    DATA_TYPE_UINT64, NULL, &nparity, NULL) == 0)
972 		vdev->v_nparity = nparity;
973 
974 	if (nvlist_find(nvlist, ZPOOL_CONFIG_PATH,
975 	    DATA_TYPE_STRING, NULL, &path, &pathlen) == 0) {
976 		char prefix[] = "/dev/";
977 
978 		len = strlen(prefix);
979 		if (len < pathlen && memcmp(path, prefix, len) == 0) {
980 			path += len;
981 			pathlen -= len;
982 		}
983 		name = malloc(pathlen + 1);
984 		bcopy(path, name, pathlen);
985 		name[pathlen] = '\0';
986 		vdev->v_name = name;
987 	} else {
988 		name = NULL;
989 		if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) {
990 			if (vdev->v_nparity < 1 ||
991 			    vdev->v_nparity > 3) {
992 				printf("ZFS: invalid raidz parity: %d\n",
993 				    vdev->v_nparity);
994 				return (EIO);
995 			}
996 			(void) asprintf(&name, "%.*s%d-%" PRIu64, len, type,
997 			    vdev->v_nparity, id);
998 		} else {
999 			(void) asprintf(&name, "%.*s-%" PRIu64, len, type, id);
1000 		}
1001 		vdev->v_name = name;
1002 	}
1003 	*vdevp = vdev;
1004 	return (0);
1005 }
1006 
1007 /*
1008  * Find slot for vdev. We return either NULL to signal to use
1009  * STAILQ_INSERT_HEAD, or we return link element to be used with
1010  * STAILQ_INSERT_AFTER.
1011  */
1012 static vdev_t *
1013 vdev_find_previous(vdev_t *top_vdev, vdev_t *vdev)
1014 {
1015 	vdev_t *v, *previous;
1016 
1017 	if (STAILQ_EMPTY(&top_vdev->v_children))
1018 		return (NULL);
1019 
1020 	previous = NULL;
1021 	STAILQ_FOREACH(v, &top_vdev->v_children, v_childlink) {
1022 		if (v->v_id > vdev->v_id)
1023 			return (previous);
1024 
1025 		if (v->v_id == vdev->v_id)
1026 			return (v);
1027 
1028 		if (v->v_id < vdev->v_id)
1029 			previous = v;
1030 	}
1031 	return (previous);
1032 }
1033 
1034 static size_t
1035 vdev_child_count(vdev_t *vdev)
1036 {
1037 	vdev_t *v;
1038 	size_t count;
1039 
1040 	count = 0;
1041 	STAILQ_FOREACH(v, &vdev->v_children, v_childlink) {
1042 		count++;
1043 	}
1044 	return (count);
1045 }
1046 
1047 /*
1048  * Insert vdev into top_vdev children list. List is ordered by v_id.
1049  */
1050 static void
1051 vdev_insert(vdev_t *top_vdev, vdev_t *vdev)
1052 {
1053 	vdev_t *previous;
1054 	size_t count;
1055 
1056 	/*
1057 	 * The top level vdev can appear in random order, depending how
1058 	 * the firmware is presenting the disk devices.
1059 	 * However, we will insert vdev to create list ordered by v_id,
1060 	 * so we can use either STAILQ_INSERT_HEAD or STAILQ_INSERT_AFTER
1061 	 * as STAILQ does not have insert before.
1062 	 */
1063 	previous = vdev_find_previous(top_vdev, vdev);
1064 
1065 	if (previous == NULL) {
1066 		STAILQ_INSERT_HEAD(&top_vdev->v_children, vdev, v_childlink);
1067 	} else if (previous->v_id == vdev->v_id) {
1068 		/*
1069 		 * This vdev was configured from label config,
1070 		 * do not insert duplicate.
1071 		 */
1072 		return;
1073 	} else {
1074 		STAILQ_INSERT_AFTER(&top_vdev->v_children, previous, vdev,
1075 		    v_childlink);
1076 	}
1077 
1078 	count = vdev_child_count(top_vdev);
1079 	if (top_vdev->v_nchildren < count)
1080 		top_vdev->v_nchildren = count;
1081 }
1082 
1083 static int
1084 vdev_from_nvlist(spa_t *spa, uint64_t top_guid, const nvlist_t *nvlist)
1085 {
1086 	vdev_t *top_vdev, *vdev;
1087 	nvlist_t *kids = NULL;
1088 	int rc, nkids;
1089 
1090 	/* Get top vdev. */
1091 	top_vdev = vdev_find(top_guid);
1092 	if (top_vdev == NULL) {
1093 		rc = vdev_init(top_guid, nvlist, &top_vdev);
1094 		if (rc != 0)
1095 			return (rc);
1096 		top_vdev->v_spa = spa;
1097 		top_vdev->v_top = top_vdev;
1098 		vdev_insert(spa->spa_root_vdev, top_vdev);
1099 	}
1100 
1101 	/* Add children if there are any. */
1102 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1103 	    &nkids, &kids, NULL);
1104 	if (rc == 0) {
1105 		for (int i = 0; i < nkids; i++) {
1106 			uint64_t guid;
1107 
1108 			rc = nvlist_find(kids, ZPOOL_CONFIG_GUID,
1109 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1110 			if (rc != 0) {
1111 				nvlist_destroy(kids);
1112 				return (rc);
1113 			}
1114 			rc = vdev_init(guid, kids, &vdev);
1115 			if (rc != 0) {
1116 				nvlist_destroy(kids);
1117 				return (rc);
1118 			}
1119 
1120 			vdev->v_spa = spa;
1121 			vdev->v_top = top_vdev;
1122 			vdev_insert(top_vdev, vdev);
1123 
1124 			rc = nvlist_next(kids);
1125 			if (rc != 0) {
1126 				nvlist_destroy(kids);
1127 				return (rc);
1128 			}
1129 		}
1130 	} else {
1131 		/*
1132 		 * When there are no children, nvlist_find() does return
1133 		 * error, reset it because leaf devices have no children.
1134 		 */
1135 		rc = 0;
1136 	}
1137 	nvlist_destroy(kids);
1138 
1139 	return (rc);
1140 }
1141 
1142 static int
1143 vdev_init_from_label(spa_t *spa, const nvlist_t *nvlist)
1144 {
1145 	uint64_t pool_guid, top_guid;
1146 	nvlist_t *vdevs;
1147 	int rc;
1148 
1149 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1150 	    NULL, &pool_guid, NULL) ||
1151 	    nvlist_find(nvlist, ZPOOL_CONFIG_TOP_GUID, DATA_TYPE_UINT64,
1152 	    NULL, &top_guid, NULL) ||
1153 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1154 	    NULL, &vdevs, NULL)) {
1155 		printf("ZFS: can't find vdev details\n");
1156 		return (ENOENT);
1157 	}
1158 
1159 	rc = vdev_from_nvlist(spa, top_guid, vdevs);
1160 	nvlist_destroy(vdevs);
1161 	return (rc);
1162 }
1163 
1164 static void
1165 vdev_set_state(vdev_t *vdev)
1166 {
1167 	vdev_t *kid;
1168 	int good_kids;
1169 	int bad_kids;
1170 
1171 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1172 		vdev_set_state(kid);
1173 	}
1174 
1175 	/*
1176 	 * A mirror or raidz is healthy if all its kids are healthy. A
1177 	 * mirror is degraded if any of its kids is healthy; a raidz
1178 	 * is degraded if at most nparity kids are offline.
1179 	 */
1180 	if (STAILQ_FIRST(&vdev->v_children)) {
1181 		good_kids = 0;
1182 		bad_kids = 0;
1183 		STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1184 			if (kid->v_state == VDEV_STATE_HEALTHY)
1185 				good_kids++;
1186 			else
1187 				bad_kids++;
1188 		}
1189 		if (bad_kids == 0) {
1190 			vdev->v_state = VDEV_STATE_HEALTHY;
1191 		} else {
1192 			if (vdev->v_read == vdev_mirror_read) {
1193 				if (good_kids) {
1194 					vdev->v_state = VDEV_STATE_DEGRADED;
1195 				} else {
1196 					vdev->v_state = VDEV_STATE_OFFLINE;
1197 				}
1198 			} else if (vdev->v_read == vdev_raidz_read) {
1199 				if (bad_kids > vdev->v_nparity) {
1200 					vdev->v_state = VDEV_STATE_OFFLINE;
1201 				} else {
1202 					vdev->v_state = VDEV_STATE_DEGRADED;
1203 				}
1204 			}
1205 		}
1206 	}
1207 }
1208 
1209 static int
1210 vdev_update_from_nvlist(uint64_t top_guid, const nvlist_t *nvlist)
1211 {
1212 	vdev_t *vdev;
1213 	nvlist_t *kids = NULL;
1214 	int rc, nkids;
1215 
1216 	/* Update top vdev. */
1217 	vdev = vdev_find(top_guid);
1218 	if (vdev != NULL)
1219 		vdev_set_initial_state(vdev, nvlist);
1220 
1221 	/* Update children if there are any. */
1222 	rc = nvlist_find(nvlist, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1223 	    &nkids, &kids, NULL);
1224 	if (rc == 0) {
1225 		for (int i = 0; i < nkids; i++) {
1226 			uint64_t guid;
1227 
1228 			rc = nvlist_find(kids, ZPOOL_CONFIG_GUID,
1229 			    DATA_TYPE_UINT64, NULL, &guid, NULL);
1230 			if (rc != 0)
1231 				break;
1232 
1233 			vdev = vdev_find(guid);
1234 			if (vdev != NULL)
1235 				vdev_set_initial_state(vdev, kids);
1236 
1237 			rc = nvlist_next(kids);
1238 			if (rc != 0)
1239 				break;
1240 		}
1241 	} else {
1242 		rc = 0;
1243 	}
1244 	nvlist_destroy(kids);
1245 
1246 	return (rc);
1247 }
1248 
1249 static int
1250 vdev_init_from_nvlist(spa_t *spa, const nvlist_t *nvlist)
1251 {
1252 	uint64_t pool_guid, vdev_children;
1253 	nvlist_t *vdevs = NULL, *kids = NULL;
1254 	int rc, nkids;
1255 
1256 	if (nvlist_find(nvlist, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1257 	    NULL, &pool_guid, NULL) ||
1258 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_CHILDREN, DATA_TYPE_UINT64,
1259 	    NULL, &vdev_children, NULL) ||
1260 	    nvlist_find(nvlist, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1261 	    NULL, &vdevs, NULL)) {
1262 		printf("ZFS: can't find vdev details\n");
1263 		return (ENOENT);
1264 	}
1265 
1266 	/* Wrong guid?! */
1267 	if (spa->spa_guid != pool_guid) {
1268 		nvlist_destroy(vdevs);
1269 		return (EINVAL);
1270 	}
1271 
1272 	spa->spa_root_vdev->v_nchildren = vdev_children;
1273 
1274 	rc = nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN, DATA_TYPE_NVLIST_ARRAY,
1275 	    &nkids, &kids, NULL);
1276 	nvlist_destroy(vdevs);
1277 
1278 	/*
1279 	 * MOS config has at least one child for root vdev.
1280 	 */
1281 	if (rc != 0)
1282 		return (rc);
1283 
1284 	for (int i = 0; i < nkids; i++) {
1285 		uint64_t guid;
1286 		vdev_t *vdev;
1287 
1288 		rc = nvlist_find(kids, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1289 		    NULL, &guid, NULL);
1290 		if (rc != 0)
1291 			break;
1292 		vdev = vdev_find(guid);
1293 		/*
1294 		 * Top level vdev is missing, create it.
1295 		 */
1296 		if (vdev == NULL)
1297 			rc = vdev_from_nvlist(spa, guid, kids);
1298 		else
1299 			rc = vdev_update_from_nvlist(guid, kids);
1300 		if (rc != 0)
1301 			break;
1302 		rc = nvlist_next(kids);
1303 		if (rc != 0)
1304 			break;
1305 	}
1306 	nvlist_destroy(kids);
1307 
1308 	/*
1309 	 * Re-evaluate top-level vdev state.
1310 	 */
1311 	vdev_set_state(spa->spa_root_vdev);
1312 
1313 	return (rc);
1314 }
1315 
1316 static spa_t *
1317 spa_find_by_guid(uint64_t guid)
1318 {
1319 	spa_t *spa;
1320 
1321 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1322 		if (spa->spa_guid == guid)
1323 			return (spa);
1324 
1325 	return (NULL);
1326 }
1327 
1328 static spa_t *
1329 spa_find_by_name(const char *name)
1330 {
1331 	spa_t *spa;
1332 
1333 	STAILQ_FOREACH(spa, &zfs_pools, spa_link)
1334 		if (strcmp(spa->spa_name, name) == 0)
1335 			return (spa);
1336 
1337 	return (NULL);
1338 }
1339 
1340 static spa_t *
1341 spa_create(uint64_t guid, const char *name)
1342 {
1343 	spa_t *spa;
1344 
1345 	if ((spa = calloc(1, sizeof(spa_t))) == NULL)
1346 		return (NULL);
1347 	if ((spa->spa_name = strdup(name)) == NULL) {
1348 		free(spa);
1349 		return (NULL);
1350 	}
1351 	spa->spa_uberblock = &spa->spa_uberblock_master;
1352 	spa->spa_mos = &spa->spa_mos_master;
1353 	spa->spa_guid = guid;
1354 	spa->spa_root_vdev = vdev_create(guid, NULL);
1355 	if (spa->spa_root_vdev == NULL) {
1356 		free(spa->spa_name);
1357 		free(spa);
1358 		return (NULL);
1359 	}
1360 	spa->spa_root_vdev->v_name = strdup("root");
1361 	STAILQ_INSERT_TAIL(&zfs_pools, spa, spa_link);
1362 
1363 	return (spa);
1364 }
1365 
1366 static const char *
1367 state_name(vdev_state_t state)
1368 {
1369 	static const char *names[] = {
1370 		"UNKNOWN",
1371 		"CLOSED",
1372 		"OFFLINE",
1373 		"REMOVED",
1374 		"CANT_OPEN",
1375 		"FAULTED",
1376 		"DEGRADED",
1377 		"ONLINE"
1378 	};
1379 	return (names[state]);
1380 }
1381 
1382 #ifdef BOOT2
1383 
1384 #define pager_printf printf
1385 
1386 #else
1387 
1388 static int
1389 pager_printf(const char *fmt, ...)
1390 {
1391 	char line[80];
1392 	va_list args;
1393 
1394 	va_start(args, fmt);
1395 	vsnprintf(line, sizeof(line), fmt, args);
1396 	va_end(args);
1397 	return (pager_output(line));
1398 }
1399 
1400 #endif
1401 
1402 #define	STATUS_FORMAT	"        %s %s\n"
1403 
1404 static int
1405 print_state(int indent, const char *name, vdev_state_t state)
1406 {
1407 	int i;
1408 	char buf[512];
1409 
1410 	buf[0] = 0;
1411 	for (i = 0; i < indent; i++)
1412 		strcat(buf, "  ");
1413 	strcat(buf, name);
1414 	return (pager_printf(STATUS_FORMAT, buf, state_name(state)));
1415 }
1416 
1417 static int
1418 vdev_status(vdev_t *vdev, int indent)
1419 {
1420 	vdev_t *kid;
1421 	int ret;
1422 
1423 	if (vdev->v_islog) {
1424 		(void) pager_output("        logs\n");
1425 		indent++;
1426 	}
1427 
1428 	ret = print_state(indent, vdev->v_name, vdev->v_state);
1429 	if (ret != 0)
1430 		return (ret);
1431 
1432 	STAILQ_FOREACH(kid, &vdev->v_children, v_childlink) {
1433 		ret = vdev_status(kid, indent + 1);
1434 		if (ret != 0)
1435 			return (ret);
1436 	}
1437 	return (ret);
1438 }
1439 
1440 static int
1441 spa_status(spa_t *spa)
1442 {
1443 	static char bootfs[ZFS_MAXNAMELEN];
1444 	uint64_t rootid;
1445 	vdev_list_t *vlist;
1446 	vdev_t *vdev;
1447 	int good_kids, bad_kids, degraded_kids, ret;
1448 	vdev_state_t state;
1449 
1450 	ret = pager_printf("  pool: %s\n", spa->spa_name);
1451 	if (ret != 0)
1452 		return (ret);
1453 
1454 	if (zfs_get_root(spa, &rootid) == 0 &&
1455 	    zfs_rlookup(spa, rootid, bootfs) == 0) {
1456 		if (bootfs[0] == '\0')
1457 			ret = pager_printf("bootfs: %s\n", spa->spa_name);
1458 		else
1459 			ret = pager_printf("bootfs: %s/%s\n", spa->spa_name,
1460 			    bootfs);
1461 		if (ret != 0)
1462 			return (ret);
1463 	}
1464 	ret = pager_printf("config:\n\n");
1465 	if (ret != 0)
1466 		return (ret);
1467 	ret = pager_printf(STATUS_FORMAT, "NAME", "STATE");
1468 	if (ret != 0)
1469 		return (ret);
1470 
1471 	good_kids = 0;
1472 	degraded_kids = 0;
1473 	bad_kids = 0;
1474 	vlist = &spa->spa_root_vdev->v_children;
1475 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1476 		if (vdev->v_state == VDEV_STATE_HEALTHY)
1477 			good_kids++;
1478 		else if (vdev->v_state == VDEV_STATE_DEGRADED)
1479 			degraded_kids++;
1480 		else
1481 			bad_kids++;
1482 	}
1483 
1484 	state = VDEV_STATE_CLOSED;
1485 	if (good_kids > 0 && (degraded_kids + bad_kids) == 0)
1486 		state = VDEV_STATE_HEALTHY;
1487 	else if ((good_kids + degraded_kids) > 0)
1488 		state = VDEV_STATE_DEGRADED;
1489 
1490 	ret = print_state(0, spa->spa_name, state);
1491 	if (ret != 0)
1492 		return (ret);
1493 
1494 	STAILQ_FOREACH(vdev, vlist, v_childlink) {
1495 		ret = vdev_status(vdev, 1);
1496 		if (ret != 0)
1497 			return (ret);
1498 	}
1499 	return (ret);
1500 }
1501 
1502 static int
1503 spa_all_status(void)
1504 {
1505 	spa_t *spa;
1506 	int first = 1, ret = 0;
1507 
1508 	STAILQ_FOREACH(spa, &zfs_pools, spa_link) {
1509 		if (!first) {
1510 			ret = pager_printf("\n");
1511 			if (ret != 0)
1512 				return (ret);
1513 		}
1514 		first = 0;
1515 		ret = spa_status(spa);
1516 		if (ret != 0)
1517 			return (ret);
1518 	}
1519 	return (ret);
1520 }
1521 
1522 static uint64_t
1523 vdev_label_offset(uint64_t psize, int l, uint64_t offset)
1524 {
1525 	uint64_t label_offset;
1526 
1527 	if (l < VDEV_LABELS / 2)
1528 		label_offset = 0;
1529 	else
1530 		label_offset = psize - VDEV_LABELS * sizeof (vdev_label_t);
1531 
1532 	return (offset + l * sizeof (vdev_label_t) + label_offset);
1533 }
1534 
1535 static int
1536 vdev_uberblock_compare(const uberblock_t *ub1, const uberblock_t *ub2)
1537 {
1538 	unsigned int seq1 = 0;
1539 	unsigned int seq2 = 0;
1540 	int cmp = AVL_CMP(ub1->ub_txg, ub2->ub_txg);
1541 
1542 	if (cmp != 0)
1543 		return (cmp);
1544 
1545 	cmp = AVL_CMP(ub1->ub_timestamp, ub2->ub_timestamp);
1546 	if (cmp != 0)
1547 		return (cmp);
1548 
1549 	if (MMP_VALID(ub1) && MMP_SEQ_VALID(ub1))
1550 		seq1 = MMP_SEQ(ub1);
1551 
1552 	if (MMP_VALID(ub2) && MMP_SEQ_VALID(ub2))
1553 		seq2 = MMP_SEQ(ub2);
1554 
1555 	return (AVL_CMP(seq1, seq2));
1556 }
1557 
1558 static int
1559 uberblock_verify(uberblock_t *ub)
1560 {
1561 	if (ub->ub_magic == BSWAP_64((uint64_t)UBERBLOCK_MAGIC)) {
1562 		byteswap_uint64_array(ub, sizeof (uberblock_t));
1563 	}
1564 
1565 	if (ub->ub_magic != UBERBLOCK_MAGIC ||
1566 	    !SPA_VERSION_IS_SUPPORTED(ub->ub_version))
1567 		return (EINVAL);
1568 
1569 	return (0);
1570 }
1571 
1572 static int
1573 vdev_label_read(vdev_t *vd, int l, void *buf, uint64_t offset,
1574     size_t size)
1575 {
1576 	blkptr_t bp;
1577 	off_t off;
1578 
1579 	off = vdev_label_offset(vd->v_psize, l, offset);
1580 
1581 	BP_ZERO(&bp);
1582 	BP_SET_LSIZE(&bp, size);
1583 	BP_SET_PSIZE(&bp, size);
1584 	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
1585 	BP_SET_COMPRESS(&bp, ZIO_COMPRESS_OFF);
1586 	DVA_SET_OFFSET(BP_IDENTITY(&bp), off);
1587 	ZIO_SET_CHECKSUM(&bp.blk_cksum, off, 0, 0, 0);
1588 
1589 	return (vdev_read_phys(vd, &bp, buf, off, size));
1590 }
1591 
1592 static uint64_t
1593 vdev_get_label_asize(nvlist_t *nvl)
1594 {
1595 	nvlist_t *vdevs;
1596 	uint64_t asize;
1597 	const char *type;
1598 	int len;
1599 
1600 	asize = 0;
1601 	/* Get vdev tree */
1602 	if (nvlist_find(nvl, ZPOOL_CONFIG_VDEV_TREE, DATA_TYPE_NVLIST,
1603 	    NULL, &vdevs, NULL) != 0)
1604 		return (asize);
1605 
1606 	/*
1607 	 * Get vdev type. We will calculate asize for raidz, mirror and disk.
1608 	 * For raidz, the asize is raw size of all children.
1609 	 */
1610 	if (nvlist_find(vdevs, ZPOOL_CONFIG_TYPE, DATA_TYPE_STRING,
1611 	    NULL, &type, &len) != 0)
1612 		goto done;
1613 
1614 	if (memcmp(type, VDEV_TYPE_MIRROR, len) != 0 &&
1615 	    memcmp(type, VDEV_TYPE_DISK, len) != 0 &&
1616 	    memcmp(type, VDEV_TYPE_RAIDZ, len) != 0)
1617 		goto done;
1618 
1619 	if (nvlist_find(vdevs, ZPOOL_CONFIG_ASIZE, DATA_TYPE_UINT64,
1620 	    NULL, &asize, NULL) != 0)
1621 		goto done;
1622 
1623 	if (memcmp(type, VDEV_TYPE_RAIDZ, len) == 0) {
1624 		nvlist_t *kids;
1625 		int nkids;
1626 
1627 		if (nvlist_find(vdevs, ZPOOL_CONFIG_CHILDREN,
1628 		    DATA_TYPE_NVLIST_ARRAY, &nkids, &kids, NULL) != 0) {
1629 			asize = 0;
1630 			goto done;
1631 		}
1632 
1633 		asize /= nkids;
1634 		nvlist_destroy(kids);
1635 	}
1636 
1637 	asize += VDEV_LABEL_START_SIZE + VDEV_LABEL_END_SIZE;
1638 done:
1639 	nvlist_destroy(vdevs);
1640 	return (asize);
1641 }
1642 
1643 static nvlist_t *
1644 vdev_label_read_config(vdev_t *vd, uint64_t txg)
1645 {
1646 	vdev_phys_t *label;
1647 	uint64_t best_txg = 0;
1648 	uint64_t label_txg = 0;
1649 	uint64_t asize;
1650 	nvlist_t *nvl = NULL, *tmp;
1651 	int error;
1652 
1653 	label = malloc(sizeof (vdev_phys_t));
1654 	if (label == NULL)
1655 		return (NULL);
1656 
1657 	for (int l = 0; l < VDEV_LABELS; l++) {
1658 		const unsigned char *nvlist;
1659 
1660 		if (vdev_label_read(vd, l, label,
1661 		    offsetof(vdev_label_t, vl_vdev_phys),
1662 		    sizeof (vdev_phys_t)))
1663 			continue;
1664 
1665 		nvlist = (const unsigned char *) label->vp_nvlist;
1666 		tmp = nvlist_import(nvlist + 4, nvlist[0], nvlist[1]);
1667 		if (tmp == NULL)
1668 			continue;
1669 
1670 		error = nvlist_find(tmp, ZPOOL_CONFIG_POOL_TXG,
1671 		    DATA_TYPE_UINT64, NULL, &label_txg, NULL);
1672 		if (error != 0 || label_txg == 0) {
1673 			nvlist_destroy(nvl);
1674 			nvl = tmp;
1675 			goto done;
1676 		}
1677 
1678 		if (label_txg <= txg && label_txg > best_txg) {
1679 			best_txg = label_txg;
1680 			nvlist_destroy(nvl);
1681 			nvl = tmp;
1682 			tmp = NULL;
1683 
1684 			/*
1685 			 * Use asize from pool config. We need this
1686 			 * because we can get bad value from BIOS.
1687 			 */
1688 			asize = vdev_get_label_asize(nvl);
1689 			if (asize != 0) {
1690 				vd->v_psize = asize;
1691 			}
1692 		}
1693 		nvlist_destroy(tmp);
1694 	}
1695 
1696 	if (best_txg == 0) {
1697 		nvlist_destroy(nvl);
1698 		nvl = NULL;
1699 	}
1700 done:
1701 	free(label);
1702 	return (nvl);
1703 }
1704 
1705 static void
1706 vdev_uberblock_load(vdev_t *vd, uberblock_t *ub)
1707 {
1708 	uberblock_t *buf;
1709 
1710 	buf = malloc(VDEV_UBERBLOCK_SIZE(vd));
1711 	if (buf == NULL)
1712 		return;
1713 
1714 	for (int l = 0; l < VDEV_LABELS; l++) {
1715 		for (int n = 0; n < VDEV_UBERBLOCK_COUNT(vd); n++) {
1716 			if (vdev_label_read(vd, l, buf,
1717 			    VDEV_UBERBLOCK_OFFSET(vd, n),
1718 			    VDEV_UBERBLOCK_SIZE(vd)))
1719 				continue;
1720 			if (uberblock_verify(buf) != 0)
1721 				continue;
1722 
1723 			if (vdev_uberblock_compare(buf, ub) > 0)
1724 				*ub = *buf;
1725 		}
1726 	}
1727 	free(buf);
1728 }
1729 
1730 static int
1731 vdev_probe(vdev_phys_read_t *_read, void *read_priv, spa_t **spap)
1732 {
1733 	vdev_t vtmp;
1734 	spa_t *spa;
1735 	vdev_t *vdev;
1736 	nvlist_t *nvl;
1737 	uint64_t val;
1738 	uint64_t guid, vdev_children;
1739 	uint64_t pool_txg, pool_guid;
1740 	const char *pool_name;
1741 	int rc, namelen;
1742 
1743 	/*
1744 	 * Load the vdev label and figure out which
1745 	 * uberblock is most current.
1746 	 */
1747 	memset(&vtmp, 0, sizeof(vtmp));
1748 	vtmp.v_phys_read = _read;
1749 	vtmp.v_read_priv = read_priv;
1750 	vtmp.v_psize = P2ALIGN(ldi_get_size(read_priv),
1751 	    (uint64_t)sizeof (vdev_label_t));
1752 
1753 	/* Test for minimum device size. */
1754 	if (vtmp.v_psize < SPA_MINDEVSIZE)
1755 		return (EIO);
1756 
1757 	nvl = vdev_label_read_config(&vtmp, UINT64_MAX);
1758 	if (nvl == NULL)
1759 		return (EIO);
1760 
1761 	if (nvlist_find(nvl, ZPOOL_CONFIG_VERSION, DATA_TYPE_UINT64,
1762 	    NULL, &val, NULL) != 0) {
1763 		nvlist_destroy(nvl);
1764 		return (EIO);
1765 	}
1766 
1767 	if (!SPA_VERSION_IS_SUPPORTED(val)) {
1768 		printf("ZFS: unsupported ZFS version %u (should be %u)\n",
1769 		    (unsigned)val, (unsigned)SPA_VERSION);
1770 		nvlist_destroy(nvl);
1771 		return (EIO);
1772 	}
1773 
1774 	/* Check ZFS features for read */
1775 	rc = nvlist_check_features_for_read(nvl);
1776 	if (rc != 0) {
1777 		nvlist_destroy(nvl);
1778 		return (EIO);
1779 	}
1780 
1781 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_STATE, DATA_TYPE_UINT64,
1782 	    NULL, &val, NULL) != 0) {
1783 		nvlist_destroy(nvl);
1784 		return (EIO);
1785 	}
1786 
1787 	if (val == POOL_STATE_DESTROYED) {
1788 		/* We don't boot only from destroyed pools. */
1789 		nvlist_destroy(nvl);
1790 		return (EIO);
1791 	}
1792 
1793 	if (nvlist_find(nvl, ZPOOL_CONFIG_POOL_TXG, DATA_TYPE_UINT64,
1794 	    NULL, &pool_txg, NULL) != 0 ||
1795 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_GUID, DATA_TYPE_UINT64,
1796 	    NULL, &pool_guid, NULL) != 0 ||
1797 	    nvlist_find(nvl, ZPOOL_CONFIG_POOL_NAME, DATA_TYPE_STRING,
1798 	    NULL, &pool_name, &namelen) != 0) {
1799 		/*
1800 		 * Cache and spare devices end up here - just ignore
1801 		 * them.
1802 		 */
1803 		nvlist_destroy(nvl);
1804 		return (EIO);
1805 	}
1806 
1807 	/*
1808 	 * Create the pool if this is the first time we've seen it.
1809 	 */
1810 	spa = spa_find_by_guid(pool_guid);
1811 	if (spa == NULL) {
1812 		char *name;
1813 
1814 		nvlist_find(nvl, ZPOOL_CONFIG_VDEV_CHILDREN,
1815 		    DATA_TYPE_UINT64, NULL, &vdev_children, NULL);
1816 		name = malloc(namelen + 1);
1817 		if (name == NULL) {
1818 			nvlist_destroy(nvl);
1819 			return (ENOMEM);
1820 		}
1821 		bcopy(pool_name, name, namelen);
1822 		name[namelen] = '\0';
1823 		spa = spa_create(pool_guid, name);
1824 		free(name);
1825 		if (spa == NULL) {
1826 			nvlist_destroy(nvl);
1827 			return (ENOMEM);
1828 		}
1829 		spa->spa_root_vdev->v_nchildren = vdev_children;
1830 	}
1831 	if (pool_txg > spa->spa_txg)
1832 		spa->spa_txg = pool_txg;
1833 
1834 	/*
1835 	 * Get the vdev tree and create our in-core copy of it.
1836 	 * If we already have a vdev with this guid, this must
1837 	 * be some kind of alias (overlapping slices, dangerously dedicated
1838 	 * disks etc).
1839 	 */
1840 	if (nvlist_find(nvl, ZPOOL_CONFIG_GUID, DATA_TYPE_UINT64,
1841 	    NULL, &guid, NULL) != 0) {
1842 		nvlist_destroy(nvl);
1843 		return (EIO);
1844 	}
1845 	vdev = vdev_find(guid);
1846 	/* Has this vdev already been inited? */
1847 	if (vdev && vdev->v_phys_read) {
1848 		nvlist_destroy(nvl);
1849 		return (EIO);
1850 	}
1851 
1852 	rc = vdev_init_from_label(spa, nvl);
1853 	nvlist_destroy(nvl);
1854 	if (rc != 0)
1855 		return (rc);
1856 
1857 	/*
1858 	 * We should already have created an incomplete vdev for this
1859 	 * vdev. Find it and initialise it with our read proc.
1860 	 */
1861 	vdev = vdev_find(guid);
1862 	if (vdev != NULL) {
1863 		vdev->v_phys_read = _read;
1864 		vdev->v_read_priv = read_priv;
1865 		vdev->v_psize = vtmp.v_psize;
1866 		/*
1867 		 * If no other state is set, mark vdev healthy.
1868 		 */
1869 		if (vdev->v_state == VDEV_STATE_UNKNOWN)
1870 			vdev->v_state = VDEV_STATE_HEALTHY;
1871 	} else {
1872 		printf("ZFS: inconsistent nvlist contents\n");
1873 		return (EIO);
1874 	}
1875 
1876 	if (vdev->v_islog)
1877 		spa->spa_with_log = vdev->v_islog;
1878 
1879 	/*
1880 	 * Re-evaluate top-level vdev state.
1881 	 */
1882 	vdev_set_state(vdev->v_top);
1883 
1884 	/*
1885 	 * Ok, we are happy with the pool so far. Lets find
1886 	 * the best uberblock and then we can actually access
1887 	 * the contents of the pool.
1888 	 */
1889 	vdev_uberblock_load(vdev, spa->spa_uberblock);
1890 
1891 	if (spap != NULL)
1892 		*spap = spa;
1893 	return (0);
1894 }
1895 
1896 static int
1897 ilog2(int n)
1898 {
1899 	int v;
1900 
1901 	for (v = 0; v < 32; v++)
1902 		if (n == (1 << v))
1903 			return (v);
1904 	return (-1);
1905 }
1906 
1907 static int
1908 zio_read_gang(const spa_t *spa, const blkptr_t *bp, void *buf)
1909 {
1910 	blkptr_t gbh_bp;
1911 	zio_gbh_phys_t zio_gb;
1912 	char *pbuf;
1913 	int i;
1914 
1915 	/* Artificial BP for gang block header. */
1916 	gbh_bp = *bp;
1917 	BP_SET_PSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1918 	BP_SET_LSIZE(&gbh_bp, SPA_GANGBLOCKSIZE);
1919 	BP_SET_CHECKSUM(&gbh_bp, ZIO_CHECKSUM_GANG_HEADER);
1920 	BP_SET_COMPRESS(&gbh_bp, ZIO_COMPRESS_OFF);
1921 	for (i = 0; i < SPA_DVAS_PER_BP; i++)
1922 		DVA_SET_GANG(&gbh_bp.blk_dva[i], 0);
1923 
1924 	/* Read gang header block using the artificial BP. */
1925 	if (zio_read(spa, &gbh_bp, &zio_gb))
1926 		return (EIO);
1927 
1928 	pbuf = buf;
1929 	for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
1930 		blkptr_t *gbp = &zio_gb.zg_blkptr[i];
1931 
1932 		if (BP_IS_HOLE(gbp))
1933 			continue;
1934 		if (zio_read(spa, gbp, pbuf))
1935 			return (EIO);
1936 		pbuf += BP_GET_PSIZE(gbp);
1937 	}
1938 
1939 	if (zio_checksum_verify(spa, bp, buf))
1940 		return (EIO);
1941 	return (0);
1942 }
1943 
1944 static int
1945 zio_read(const spa_t *spa, const blkptr_t *bp, void *buf)
1946 {
1947 	int cpfunc = BP_GET_COMPRESS(bp);
1948 	uint64_t align, size;
1949 	void *pbuf;
1950 	int i, error;
1951 
1952 	/*
1953 	 * Process data embedded in block pointer
1954 	 */
1955 	if (BP_IS_EMBEDDED(bp)) {
1956 		ASSERT(BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA);
1957 
1958 		size = BPE_GET_PSIZE(bp);
1959 		ASSERT(size <= BPE_PAYLOAD_SIZE);
1960 
1961 		if (cpfunc != ZIO_COMPRESS_OFF)
1962 			pbuf = malloc(size);
1963 		else
1964 			pbuf = buf;
1965 
1966 		if (pbuf == NULL)
1967 			return (ENOMEM);
1968 
1969 		decode_embedded_bp_compressed(bp, pbuf);
1970 		error = 0;
1971 
1972 		if (cpfunc != ZIO_COMPRESS_OFF) {
1973 			error = zio_decompress_data(cpfunc, pbuf,
1974 			    size, buf, BP_GET_LSIZE(bp));
1975 			free(pbuf);
1976 		}
1977 		if (error != 0)
1978 			printf("ZFS: i/o error - unable to decompress "
1979 			    "block pointer data, error %d\n", error);
1980 		return (error);
1981 	}
1982 
1983 	error = EIO;
1984 
1985 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
1986 		const dva_t *dva = &bp->blk_dva[i];
1987 		vdev_t *vdev;
1988 		vdev_list_t *vlist;
1989 		uint64_t vdevid;
1990 		off_t offset;
1991 
1992 		if (!dva->dva_word[0] && !dva->dva_word[1])
1993 			continue;
1994 
1995 		vdevid = DVA_GET_VDEV(dva);
1996 		offset = DVA_GET_OFFSET(dva);
1997 		vlist = &spa->spa_root_vdev->v_children;
1998 		STAILQ_FOREACH(vdev, vlist, v_childlink) {
1999 			if (vdev->v_id == vdevid)
2000 				break;
2001 		}
2002 		if (!vdev || !vdev->v_read)
2003 			continue;
2004 
2005 		size = BP_GET_PSIZE(bp);
2006 		if (vdev->v_read == vdev_raidz_read) {
2007 			align = 1ULL << vdev->v_ashift;
2008 			if (P2PHASE(size, align) != 0)
2009 				size = P2ROUNDUP(size, align);
2010 		}
2011 		if (size != BP_GET_PSIZE(bp) || cpfunc != ZIO_COMPRESS_OFF)
2012 			pbuf = malloc(size);
2013 		else
2014 			pbuf = buf;
2015 
2016 		if (pbuf == NULL) {
2017 			error = ENOMEM;
2018 			break;
2019 		}
2020 
2021 		if (DVA_GET_GANG(dva))
2022 			error = zio_read_gang(spa, bp, pbuf);
2023 		else
2024 			error = vdev->v_read(vdev, bp, pbuf, offset, size);
2025 		if (error == 0) {
2026 			if (cpfunc != ZIO_COMPRESS_OFF)
2027 				error = zio_decompress_data(cpfunc, pbuf,
2028 				    BP_GET_PSIZE(bp), buf, BP_GET_LSIZE(bp));
2029 			else if (size != BP_GET_PSIZE(bp))
2030 				bcopy(pbuf, buf, BP_GET_PSIZE(bp));
2031 		} else {
2032 			printf("zio_read error: %d\n", error);
2033 		}
2034 		if (buf != pbuf)
2035 			free(pbuf);
2036 		if (error == 0)
2037 			break;
2038 	}
2039 	if (error != 0)
2040 		printf("ZFS: i/o error - all block copies unavailable\n");
2041 
2042 	return (error);
2043 }
2044 
2045 static int
2046 dnode_read(const spa_t *spa, const dnode_phys_t *dnode, off_t offset,
2047     void *buf, size_t buflen)
2048 {
2049 	int ibshift = dnode->dn_indblkshift - SPA_BLKPTRSHIFT;
2050 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2051 	int nlevels = dnode->dn_nlevels;
2052 	int i, rc;
2053 
2054 	if (bsize > SPA_MAXBLOCKSIZE) {
2055 		printf("ZFS: I/O error - blocks larger than %llu are not "
2056 		    "supported\n", SPA_MAXBLOCKSIZE);
2057 		return (EIO);
2058 	}
2059 
2060 	/*
2061 	 * Note: bsize may not be a power of two here so we need to do an
2062 	 * actual divide rather than a bitshift.
2063 	 */
2064 	while (buflen > 0) {
2065 		uint64_t bn = offset / bsize;
2066 		int boff = offset % bsize;
2067 		int ibn;
2068 		const blkptr_t *indbp;
2069 		blkptr_t bp;
2070 
2071 		if (bn > dnode->dn_maxblkid)
2072 			return (EIO);
2073 
2074 		if (dnode == dnode_cache_obj && bn == dnode_cache_bn)
2075 			goto cached;
2076 
2077 		indbp = dnode->dn_blkptr;
2078 		for (i = 0; i < nlevels; i++) {
2079 			/*
2080 			 * Copy the bp from the indirect array so that
2081 			 * we can re-use the scratch buffer for multi-level
2082 			 * objects.
2083 			 */
2084 			ibn = bn >> ((nlevels - i - 1) * ibshift);
2085 			ibn &= ((1 << ibshift) - 1);
2086 			bp = indbp[ibn];
2087 			if (BP_IS_HOLE(&bp)) {
2088 				memset(dnode_cache_buf, 0, bsize);
2089 				break;
2090 			}
2091 			rc = zio_read(spa, &bp, dnode_cache_buf);
2092 			if (rc)
2093 				return (rc);
2094 			indbp = (const blkptr_t *) dnode_cache_buf;
2095 		}
2096 		dnode_cache_obj = dnode;
2097 		dnode_cache_bn = bn;
2098 	cached:
2099 
2100 		/*
2101 		 * The buffer contains our data block. Copy what we
2102 		 * need from it and loop.
2103 		 */
2104 		i = bsize - boff;
2105 		if (i > buflen) i = buflen;
2106 		memcpy(buf, &dnode_cache_buf[boff], i);
2107 		buf = ((char *)buf) + i;
2108 		offset += i;
2109 		buflen -= i;
2110 	}
2111 
2112 	return (0);
2113 }
2114 
2115 /*
2116  * Lookup a value in a microzap directory.
2117  */
2118 static int
2119 mzap_lookup(const mzap_phys_t *mz, size_t size, const char *name,
2120     uint64_t *value)
2121 {
2122 	const mzap_ent_phys_t *mze;
2123 	int chunks, i;
2124 
2125 	/*
2126 	 * Microzap objects use exactly one block. Read the whole
2127 	 * thing.
2128 	 */
2129 	chunks = size / MZAP_ENT_LEN - 1;
2130 	for (i = 0; i < chunks; i++) {
2131 		mze = &mz->mz_chunk[i];
2132 		if (strcmp(mze->mze_name, name) == 0) {
2133 			*value = mze->mze_value;
2134 			return (0);
2135 		}
2136 	}
2137 
2138 	return (ENOENT);
2139 }
2140 
2141 /*
2142  * Compare a name with a zap leaf entry. Return non-zero if the name
2143  * matches.
2144  */
2145 static int
2146 fzap_name_equal(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2147     const char *name)
2148 {
2149 	size_t namelen;
2150 	const zap_leaf_chunk_t *nc;
2151 	const char *p;
2152 
2153 	namelen = zc->l_entry.le_name_numints;
2154 
2155 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2156 	p = name;
2157 	while (namelen > 0) {
2158 		size_t len;
2159 
2160 		len = namelen;
2161 		if (len > ZAP_LEAF_ARRAY_BYTES)
2162 			len = ZAP_LEAF_ARRAY_BYTES;
2163 		if (memcmp(p, nc->l_array.la_array, len))
2164 			return (0);
2165 		p += len;
2166 		namelen -= len;
2167 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2168 	}
2169 
2170 	return (1);
2171 }
2172 
2173 /*
2174  * Extract a uint64_t value from a zap leaf entry.
2175  */
2176 static uint64_t
2177 fzap_leaf_value(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc)
2178 {
2179 	const zap_leaf_chunk_t *vc;
2180 	int i;
2181 	uint64_t value;
2182 	const uint8_t *p;
2183 
2184 	vc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_value_chunk);
2185 	for (i = 0, value = 0, p = vc->l_array.la_array; i < 8; i++) {
2186 		value = (value << 8) | p[i];
2187 	}
2188 
2189 	return (value);
2190 }
2191 
2192 static void
2193 stv(int len, void *addr, uint64_t value)
2194 {
2195 	switch (len) {
2196 	case 1:
2197 		*(uint8_t *)addr = value;
2198 		return;
2199 	case 2:
2200 		*(uint16_t *)addr = value;
2201 		return;
2202 	case 4:
2203 		*(uint32_t *)addr = value;
2204 		return;
2205 	case 8:
2206 		*(uint64_t *)addr = value;
2207 		return;
2208 	}
2209 }
2210 
2211 /*
2212  * Extract a array from a zap leaf entry.
2213  */
2214 static void
2215 fzap_leaf_array(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc,
2216     uint64_t integer_size, uint64_t num_integers, void *buf)
2217 {
2218 	uint64_t array_int_len = zc->l_entry.le_value_intlen;
2219 	uint64_t value = 0;
2220 	uint64_t *u64 = buf;
2221 	char *p = buf;
2222 	int len = MIN(zc->l_entry.le_value_numints, num_integers);
2223 	int chunk = zc->l_entry.le_value_chunk;
2224 	int byten = 0;
2225 
2226 	if (integer_size == 8 && len == 1) {
2227 		*u64 = fzap_leaf_value(zl, zc);
2228 		return;
2229 	}
2230 
2231 	while (len > 0) {
2232 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(zl, chunk).l_array;
2233 		int i;
2234 
2235 		ASSERT3U(chunk, <, ZAP_LEAF_NUMCHUNKS(zl));
2236 		for (i = 0; i < ZAP_LEAF_ARRAY_BYTES && len > 0; i++) {
2237 			value = (value << 8) | la->la_array[i];
2238 			byten++;
2239 			if (byten == array_int_len) {
2240 				stv(integer_size, p, value);
2241 				byten = 0;
2242 				len--;
2243 				if (len == 0)
2244 					return;
2245 				p += integer_size;
2246 			}
2247 		}
2248 		chunk = la->la_next;
2249 	}
2250 }
2251 
2252 static int
2253 fzap_check_size(uint64_t integer_size, uint64_t num_integers)
2254 {
2255 
2256 	switch (integer_size) {
2257 	case 1:
2258 	case 2:
2259 	case 4:
2260 	case 8:
2261 		break;
2262 	default:
2263 		return (EINVAL);
2264 	}
2265 
2266 	if (integer_size * num_integers > ZAP_MAXVALUELEN)
2267 		return (E2BIG);
2268 
2269 	return (0);
2270 }
2271 
2272 static void
2273 zap_leaf_free(zap_leaf_t *leaf)
2274 {
2275 	free(leaf->l_phys);
2276 	free(leaf);
2277 }
2278 
2279 static int
2280 zap_get_leaf_byblk(fat_zap_t *zap, uint64_t blk, zap_leaf_t **lp)
2281 {
2282 	int bs = FZAP_BLOCK_SHIFT(zap);
2283 	int err;
2284 
2285 	*lp = malloc(sizeof(**lp));
2286 	if (*lp == NULL)
2287 		return (ENOMEM);
2288 
2289 	(*lp)->l_bs = bs;
2290 	(*lp)->l_phys = malloc(1 << bs);
2291 
2292 	if ((*lp)->l_phys == NULL) {
2293 		free(*lp);
2294 		return (ENOMEM);
2295 	}
2296 	err = dnode_read(zap->zap_spa, zap->zap_dnode, blk << bs, (*lp)->l_phys,
2297 	    1 << bs);
2298 	if (err != 0) {
2299 		zap_leaf_free(*lp);
2300 	}
2301 	return (err);
2302 }
2303 
2304 static int
2305 zap_table_load(fat_zap_t *zap, zap_table_phys_t *tbl, uint64_t idx,
2306     uint64_t *valp)
2307 {
2308 	int bs = FZAP_BLOCK_SHIFT(zap);
2309 	uint64_t blk = idx >> (bs - 3);
2310 	uint64_t off = idx & ((1 << (bs - 3)) - 1);
2311 	uint64_t *buf;
2312 	int rc;
2313 
2314 	buf = malloc(1 << zap->zap_block_shift);
2315 	if (buf == NULL)
2316 		return (ENOMEM);
2317 	rc = dnode_read(zap->zap_spa, zap->zap_dnode, (tbl->zt_blk + blk) << bs,
2318 	    buf, 1 << zap->zap_block_shift);
2319 	if (rc == 0)
2320 		*valp = buf[off];
2321 	free(buf);
2322 	return (rc);
2323 }
2324 
2325 static int
2326 zap_idx_to_blk(fat_zap_t *zap, uint64_t idx, uint64_t *valp)
2327 {
2328 	if (zap->zap_phys->zap_ptrtbl.zt_numblks == 0) {
2329 		*valp = ZAP_EMBEDDED_PTRTBL_ENT(zap, idx);
2330 		return (0);
2331 	} else {
2332 		return (zap_table_load(zap, &zap->zap_phys->zap_ptrtbl,
2333 		    idx, valp));
2334 	}
2335 }
2336 
2337 #define	ZAP_HASH_IDX(hash, n)	(((n) == 0) ? 0 : ((hash) >> (64 - (n))))
2338 static int
2339 zap_deref_leaf(fat_zap_t *zap, uint64_t h, zap_leaf_t **lp)
2340 {
2341 	uint64_t idx, blk;
2342 	int err;
2343 
2344 	idx = ZAP_HASH_IDX(h, zap->zap_phys->zap_ptrtbl.zt_shift);
2345 	err = zap_idx_to_blk(zap, idx, &blk);
2346 	if (err != 0)
2347 		return (err);
2348 	return (zap_get_leaf_byblk(zap, blk, lp));
2349 }
2350 
2351 #define	CHAIN_END	0xffff	/* end of the chunk chain */
2352 #define	LEAF_HASH(l, h) \
2353 	((ZAP_LEAF_HASH_NUMENTRIES(l)-1) & \
2354 	((h) >> \
2355 	(64 - ZAP_LEAF_HASH_SHIFT(l) - (l)->l_phys->l_hdr.lh_prefix_len)))
2356 #define	LEAF_HASH_ENTPTR(l, h)	(&(l)->l_phys->l_hash[LEAF_HASH(l, h)])
2357 
2358 static int
2359 zap_leaf_lookup(zap_leaf_t *zl, uint64_t hash, const char *name,
2360     uint64_t integer_size, uint64_t num_integers, void *value)
2361 {
2362 	int rc;
2363 	uint16_t *chunkp;
2364 	struct zap_leaf_entry *le;
2365 
2366 	/*
2367 	 * Make sure this chunk matches our hash.
2368 	 */
2369 	if (zl->l_phys->l_hdr.lh_prefix_len > 0 &&
2370 	    zl->l_phys->l_hdr.lh_prefix !=
2371 	    hash >> (64 - zl->l_phys->l_hdr.lh_prefix_len))
2372 		return (EIO);
2373 
2374 	rc = ENOENT;
2375 	for (chunkp = LEAF_HASH_ENTPTR(zl, hash);
2376 	    *chunkp != CHAIN_END; chunkp = &le->le_next) {
2377 		zap_leaf_chunk_t *zc;
2378 		uint16_t chunk = *chunkp;
2379 
2380 		le = ZAP_LEAF_ENTRY(zl, chunk);
2381 		if (le->le_hash != hash)
2382 			continue;
2383 		zc = &ZAP_LEAF_CHUNK(zl, chunk);
2384 		if (fzap_name_equal(zl, zc, name)) {
2385 			if (zc->l_entry.le_value_intlen > integer_size) {
2386 				rc = EINVAL;
2387 			} else {
2388 				fzap_leaf_array(zl, zc, integer_size,
2389 				    num_integers, value);
2390 				rc = 0;
2391 			}
2392 			break;
2393 		}
2394 	}
2395 	return (rc);
2396 }
2397 
2398 /*
2399  * Lookup a value in a fatzap directory.
2400  */
2401 static int
2402 fzap_lookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2403     const char *name, uint64_t integer_size, uint64_t num_integers,
2404     void *value)
2405 {
2406 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2407 	fat_zap_t z;
2408 	zap_leaf_t *zl;
2409 	uint64_t hash;
2410 	int rc;
2411 
2412 	if (zh->zap_magic != ZAP_MAGIC)
2413 		return (EIO);
2414 
2415 	if ((rc = fzap_check_size(integer_size, num_integers)) != 0) {
2416 		return (rc);
2417 	}
2418 
2419 	z.zap_block_shift = ilog2(bsize);
2420 	z.zap_phys = zh;
2421 	z.zap_spa = spa;
2422 	z.zap_dnode = dnode;
2423 
2424 	hash = zap_hash(zh->zap_salt, name);
2425 	rc = zap_deref_leaf(&z, hash, &zl);
2426 	if (rc != 0)
2427 		return (rc);
2428 
2429 	rc = zap_leaf_lookup(zl, hash, name, integer_size, num_integers, value);
2430 
2431 	zap_leaf_free(zl);
2432 	return (rc);
2433 }
2434 
2435 /*
2436  * Lookup a name in a zap object and return its value as a uint64_t.
2437  */
2438 static int
2439 zap_lookup(const spa_t *spa, const dnode_phys_t *dnode, const char *name,
2440     uint64_t integer_size, uint64_t num_integers, void *value)
2441 {
2442 	int rc;
2443 	zap_phys_t *zap;
2444 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2445 
2446 	zap = malloc(size);
2447 	if (zap == NULL)
2448 		return (ENOMEM);
2449 
2450 	rc = dnode_read(spa, dnode, 0, zap, size);
2451 	if (rc)
2452 		goto done;
2453 
2454 	switch (zap->zap_block_type) {
2455 	case ZBT_MICRO:
2456 		rc = mzap_lookup((const mzap_phys_t *)zap, size, name, value);
2457 		break;
2458 	case ZBT_HEADER:
2459 		rc = fzap_lookup(spa, dnode, zap, name, integer_size,
2460 		    num_integers, value);
2461 		break;
2462 	default:
2463 		printf("ZFS: invalid zap_type=%" PRIx64 "\n",
2464 		    zap->zap_block_type);
2465 		rc = EIO;
2466 	}
2467 done:
2468 	free(zap);
2469 	return (rc);
2470 }
2471 
2472 /*
2473  * List a microzap directory.
2474  */
2475 static int
2476 mzap_list(const mzap_phys_t *mz, size_t size,
2477     int (*callback)(const char *, uint64_t))
2478 {
2479 	const mzap_ent_phys_t *mze;
2480 	int chunks, i, rc;
2481 
2482 	/*
2483 	 * Microzap objects use exactly one block. Read the whole
2484 	 * thing.
2485 	 */
2486 	rc = 0;
2487 	chunks = size / MZAP_ENT_LEN - 1;
2488 	for (i = 0; i < chunks; i++) {
2489 		mze = &mz->mz_chunk[i];
2490 		if (mze->mze_name[0]) {
2491 			rc = callback(mze->mze_name, mze->mze_value);
2492 			if (rc != 0)
2493 				break;
2494 		}
2495 	}
2496 
2497 	return (rc);
2498 }
2499 
2500 /*
2501  * List a fatzap directory.
2502  */
2503 static int
2504 fzap_list(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2505     int (*callback)(const char *, uint64_t))
2506 {
2507 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2508 	fat_zap_t z;
2509 	uint64_t i;
2510 	int j, rc;
2511 
2512 	if (zh->zap_magic != ZAP_MAGIC)
2513 		return (EIO);
2514 
2515 	z.zap_block_shift = ilog2(bsize);
2516 	z.zap_phys = zh;
2517 
2518 	/*
2519 	 * This assumes that the leaf blocks start at block 1. The
2520 	 * documentation isn't exactly clear on this.
2521 	 */
2522 	zap_leaf_t zl;
2523 	zl.l_bs = z.zap_block_shift;
2524 	zl.l_phys = malloc(bsize);
2525 	if (zl.l_phys == NULL)
2526 		return (ENOMEM);
2527 
2528 	for (i = 0; i < zh->zap_num_leafs; i++) {
2529 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2530 		char name[256], *p;
2531 		uint64_t value;
2532 
2533 		if (dnode_read(spa, dnode, off, zl.l_phys, bsize)) {
2534 			free(zl.l_phys);
2535 			return (EIO);
2536 		}
2537 
2538 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2539 			zap_leaf_chunk_t *zc, *nc;
2540 			int namelen;
2541 
2542 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2543 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2544 				continue;
2545 			namelen = zc->l_entry.le_name_numints;
2546 			if (namelen > sizeof(name))
2547 				namelen = sizeof(name);
2548 
2549 			/*
2550 			 * Paste the name back together.
2551 			 */
2552 			nc = &ZAP_LEAF_CHUNK(&zl, zc->l_entry.le_name_chunk);
2553 			p = name;
2554 			while (namelen > 0) {
2555 				int len;
2556 				len = namelen;
2557 				if (len > ZAP_LEAF_ARRAY_BYTES)
2558 					len = ZAP_LEAF_ARRAY_BYTES;
2559 				memcpy(p, nc->l_array.la_array, len);
2560 				p += len;
2561 				namelen -= len;
2562 				nc = &ZAP_LEAF_CHUNK(&zl, nc->l_array.la_next);
2563 			}
2564 
2565 			/*
2566 			 * Assume the first eight bytes of the value are
2567 			 * a uint64_t.
2568 			 */
2569 			value = fzap_leaf_value(&zl, zc);
2570 
2571 			/* printf("%s 0x%jx\n", name, (uintmax_t)value); */
2572 			rc = callback((const char *)name, value);
2573 			if (rc != 0) {
2574 				free(zl.l_phys);
2575 				return (rc);
2576 			}
2577 		}
2578 	}
2579 
2580 	free(zl.l_phys);
2581 	return (0);
2582 }
2583 
2584 static int zfs_printf(const char *name, uint64_t value __unused)
2585 {
2586 
2587 	printf("%s\n", name);
2588 
2589 	return (0);
2590 }
2591 
2592 /*
2593  * List a zap directory.
2594  */
2595 static int
2596 zap_list(const spa_t *spa, const dnode_phys_t *dnode)
2597 {
2598 	zap_phys_t *zap;
2599 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2600 	int rc;
2601 
2602 	zap = malloc(size);
2603 	if (zap == NULL)
2604 		return (ENOMEM);
2605 
2606 	rc = dnode_read(spa, dnode, 0, zap, size);
2607 	if (rc == 0) {
2608 		if (zap->zap_block_type == ZBT_MICRO)
2609 			rc = mzap_list((const mzap_phys_t *)zap, size,
2610 			    zfs_printf);
2611 		else
2612 			rc = fzap_list(spa, dnode, zap, zfs_printf);
2613 	}
2614 	free(zap);
2615 	return (rc);
2616 }
2617 
2618 static int
2619 objset_get_dnode(const spa_t *spa, const objset_phys_t *os, uint64_t objnum,
2620     dnode_phys_t *dnode)
2621 {
2622 	off_t offset;
2623 
2624 	offset = objnum * sizeof(dnode_phys_t);
2625 	return dnode_read(spa, &os->os_meta_dnode, offset,
2626 		dnode, sizeof(dnode_phys_t));
2627 }
2628 
2629 /*
2630  * Lookup a name in a microzap directory.
2631  */
2632 static int
2633 mzap_rlookup(const mzap_phys_t *mz, size_t size, char *name, uint64_t value)
2634 {
2635 	const mzap_ent_phys_t *mze;
2636 	int chunks, i;
2637 
2638 	/*
2639 	 * Microzap objects use exactly one block. Read the whole
2640 	 * thing.
2641 	 */
2642 	chunks = size / MZAP_ENT_LEN - 1;
2643 	for (i = 0; i < chunks; i++) {
2644 		mze = &mz->mz_chunk[i];
2645 		if (value == mze->mze_value) {
2646 			strcpy(name, mze->mze_name);
2647 			return (0);
2648 		}
2649 	}
2650 
2651 	return (ENOENT);
2652 }
2653 
2654 static void
2655 fzap_name_copy(const zap_leaf_t *zl, const zap_leaf_chunk_t *zc, char *name)
2656 {
2657 	size_t namelen;
2658 	const zap_leaf_chunk_t *nc;
2659 	char *p;
2660 
2661 	namelen = zc->l_entry.le_name_numints;
2662 
2663 	nc = &ZAP_LEAF_CHUNK(zl, zc->l_entry.le_name_chunk);
2664 	p = name;
2665 	while (namelen > 0) {
2666 		size_t len;
2667 		len = namelen;
2668 		if (len > ZAP_LEAF_ARRAY_BYTES)
2669 			len = ZAP_LEAF_ARRAY_BYTES;
2670 		memcpy(p, nc->l_array.la_array, len);
2671 		p += len;
2672 		namelen -= len;
2673 		nc = &ZAP_LEAF_CHUNK(zl, nc->l_array.la_next);
2674 	}
2675 
2676 	*p = '\0';
2677 }
2678 
2679 static int
2680 fzap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, zap_phys_t *zh,
2681     char *name, uint64_t value)
2682 {
2683 	int bsize = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2684 	fat_zap_t z;
2685 	uint64_t i;
2686 	int j, rc;
2687 
2688 	if (zh->zap_magic != ZAP_MAGIC)
2689 		return (EIO);
2690 
2691 	z.zap_block_shift = ilog2(bsize);
2692 	z.zap_phys = zh;
2693 
2694 	/*
2695 	 * This assumes that the leaf blocks start at block 1. The
2696 	 * documentation isn't exactly clear on this.
2697 	 */
2698 	zap_leaf_t zl;
2699 	zl.l_bs = z.zap_block_shift;
2700 	zl.l_phys = malloc(bsize);
2701 	if (zl.l_phys == NULL)
2702 		return (ENOMEM);
2703 
2704 	for (i = 0; i < zh->zap_num_leafs; i++) {
2705 		off_t off = ((off_t)(i + 1)) << zl.l_bs;
2706 
2707 		rc = dnode_read(spa, dnode, off, zl.l_phys, bsize);
2708 		if (rc != 0)
2709 			goto done;
2710 
2711 		for (j = 0; j < ZAP_LEAF_NUMCHUNKS(&zl); j++) {
2712 			zap_leaf_chunk_t *zc;
2713 
2714 			zc = &ZAP_LEAF_CHUNK(&zl, j);
2715 			if (zc->l_entry.le_type != ZAP_CHUNK_ENTRY)
2716 				continue;
2717 			if (zc->l_entry.le_value_intlen != 8 ||
2718 			    zc->l_entry.le_value_numints != 1)
2719 				continue;
2720 
2721 			if (fzap_leaf_value(&zl, zc) == value) {
2722 				fzap_name_copy(&zl, zc, name);
2723 				goto done;
2724 			}
2725 		}
2726 	}
2727 
2728 	rc = ENOENT;
2729 done:
2730 	free(zl.l_phys);
2731 	return (rc);
2732 }
2733 
2734 static int
2735 zap_rlookup(const spa_t *spa, const dnode_phys_t *dnode, char *name,
2736     uint64_t value)
2737 {
2738 	zap_phys_t *zap;
2739 	size_t size = dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
2740 	int rc;
2741 
2742 	zap = malloc(size);
2743 	if (zap == NULL)
2744 		return (ENOMEM);
2745 
2746 	rc = dnode_read(spa, dnode, 0, zap, size);
2747 	if (rc == 0) {
2748 		if (zap->zap_block_type == ZBT_MICRO)
2749 			rc = mzap_rlookup((const mzap_phys_t *)zap, size,
2750 			    name, value);
2751 		else
2752 			rc = fzap_rlookup(spa, dnode, zap, name, value);
2753 	}
2754 	free(zap);
2755 	return (rc);
2756 }
2757 
2758 static int
2759 zfs_rlookup(const spa_t *spa, uint64_t objnum, char *result)
2760 {
2761 	char name[256];
2762 	char component[256];
2763 	uint64_t dir_obj, parent_obj, child_dir_zapobj;
2764 	dnode_phys_t child_dir_zap, dataset, dir, parent;
2765 	dsl_dir_phys_t *dd;
2766 	dsl_dataset_phys_t *ds;
2767 	char *p;
2768 	int len;
2769 
2770 	p = &name[sizeof(name) - 1];
2771 	*p = '\0';
2772 
2773 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
2774 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2775 		return (EIO);
2776 	}
2777 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2778 	dir_obj = ds->ds_dir_obj;
2779 
2780 	for (;;) {
2781 		if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir) != 0)
2782 			return (EIO);
2783 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2784 
2785 		/* Actual loop condition. */
2786 		parent_obj = dd->dd_parent_obj;
2787 		if (parent_obj == 0)
2788 			break;
2789 
2790 		if (objset_get_dnode(spa, spa->spa_mos, parent_obj,
2791 		    &parent) != 0)
2792 			return (EIO);
2793 		dd = (dsl_dir_phys_t *)&parent.dn_bonus;
2794 		child_dir_zapobj = dd->dd_child_dir_zapobj;
2795 		if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
2796 		    &child_dir_zap) != 0)
2797 			return (EIO);
2798 		if (zap_rlookup(spa, &child_dir_zap, component, dir_obj) != 0)
2799 			return (EIO);
2800 
2801 		len = strlen(component);
2802 		p -= len;
2803 		memcpy(p, component, len);
2804 		--p;
2805 		*p = '/';
2806 
2807 		/* Actual loop iteration. */
2808 		dir_obj = parent_obj;
2809 	}
2810 
2811 	if (*p != '\0')
2812 		++p;
2813 	strcpy(result, p);
2814 
2815 	return (0);
2816 }
2817 
2818 static int
2819 zfs_lookup_dataset(const spa_t *spa, const char *name, uint64_t *objnum)
2820 {
2821 	char element[256];
2822 	uint64_t dir_obj, child_dir_zapobj;
2823 	dnode_phys_t child_dir_zap, dir;
2824 	dsl_dir_phys_t *dd;
2825 	const char *p, *q;
2826 
2827 	if (objset_get_dnode(spa, spa->spa_mos,
2828 	    DMU_POOL_DIRECTORY_OBJECT, &dir))
2829 		return (EIO);
2830 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET, sizeof (dir_obj),
2831 	    1, &dir_obj))
2832 		return (EIO);
2833 
2834 	p = name;
2835 	for (;;) {
2836 		if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir))
2837 			return (EIO);
2838 		dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2839 
2840 		while (*p == '/')
2841 			p++;
2842 		/* Actual loop condition #1. */
2843 		if (*p == '\0')
2844 			break;
2845 
2846 		q = strchr(p, '/');
2847 		if (q) {
2848 			memcpy(element, p, q - p);
2849 			element[q - p] = '\0';
2850 			p = q + 1;
2851 		} else {
2852 			strcpy(element, p);
2853 			p += strlen(p);
2854 		}
2855 
2856 		child_dir_zapobj = dd->dd_child_dir_zapobj;
2857 		if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
2858 		    &child_dir_zap) != 0)
2859 			return (EIO);
2860 
2861 		/* Actual loop condition #2. */
2862 		if (zap_lookup(spa, &child_dir_zap, element, sizeof (dir_obj),
2863 		    1, &dir_obj) != 0)
2864 			return (ENOENT);
2865 	}
2866 
2867 	*objnum = dd->dd_head_dataset_obj;
2868 	return (0);
2869 }
2870 
2871 #ifndef BOOT2
2872 static int
2873 zfs_list_dataset(const spa_t *spa, uint64_t objnum/*, int pos, char *entry*/)
2874 {
2875 	uint64_t dir_obj, child_dir_zapobj;
2876 	dnode_phys_t child_dir_zap, dir, dataset;
2877 	dsl_dataset_phys_t *ds;
2878 	dsl_dir_phys_t *dd;
2879 
2880 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
2881 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2882 		return (EIO);
2883 	}
2884 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2885 	dir_obj = ds->ds_dir_obj;
2886 
2887 	if (objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir)) {
2888 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2889 		return (EIO);
2890 	}
2891 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2892 
2893 	child_dir_zapobj = dd->dd_child_dir_zapobj;
2894 	if (objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
2895 	    &child_dir_zap) != 0) {
2896 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2897 		return (EIO);
2898 	}
2899 
2900 	return (zap_list(spa, &child_dir_zap) != 0);
2901 }
2902 
2903 int
2904 zfs_callback_dataset(const spa_t *spa, uint64_t objnum,
2905     int (*callback)(const char *, uint64_t))
2906 {
2907 	uint64_t dir_obj, child_dir_zapobj;
2908 	dnode_phys_t child_dir_zap, dir, dataset;
2909 	dsl_dataset_phys_t *ds;
2910 	dsl_dir_phys_t *dd;
2911 	zap_phys_t *zap;
2912 	size_t size;
2913 	int err;
2914 
2915 	err = objset_get_dnode(spa, spa->spa_mos, objnum, &dataset);
2916 	if (err != 0) {
2917 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2918 		return (err);
2919 	}
2920 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2921 	dir_obj = ds->ds_dir_obj;
2922 
2923 	err = objset_get_dnode(spa, spa->spa_mos, dir_obj, &dir);
2924 	if (err != 0) {
2925 		printf("ZFS: can't find dirobj %ju\n", (uintmax_t)dir_obj);
2926 		return (err);
2927 	}
2928 	dd = (dsl_dir_phys_t *)&dir.dn_bonus;
2929 
2930 	child_dir_zapobj = dd->dd_child_dir_zapobj;
2931 	err = objset_get_dnode(spa, spa->spa_mos, child_dir_zapobj,
2932 	    &child_dir_zap);
2933 	if (err != 0) {
2934 		printf("ZFS: can't find child zap %ju\n", (uintmax_t)dir_obj);
2935 		return (err);
2936 	}
2937 
2938 	size = child_dir_zap.dn_datablkszsec << SPA_MINBLOCKSHIFT;
2939 	zap = malloc(size);
2940 	if (zap != NULL) {
2941 		err = dnode_read(spa, &child_dir_zap, 0, zap, size);
2942 		if (err != 0)
2943 			goto done;
2944 
2945 		if (zap->zap_block_type == ZBT_MICRO)
2946 			err = mzap_list((const mzap_phys_t *)zap, size,
2947 			    callback);
2948 		else
2949 			err = fzap_list(spa, &child_dir_zap, zap, callback);
2950 	} else {
2951 		err = ENOMEM;
2952 	}
2953 done:
2954 	free(zap);
2955 	return (err);
2956 }
2957 #endif
2958 
2959 /*
2960  * Find the object set given the object number of its dataset object
2961  * and return its details in *objset
2962  */
2963 static int
2964 zfs_mount_dataset(const spa_t *spa, uint64_t objnum, objset_phys_t *objset)
2965 {
2966 	dnode_phys_t dataset;
2967 	dsl_dataset_phys_t *ds;
2968 
2969 	if (objset_get_dnode(spa, spa->spa_mos, objnum, &dataset)) {
2970 		printf("ZFS: can't find dataset %ju\n", (uintmax_t)objnum);
2971 		return (EIO);
2972 	}
2973 
2974 	ds = (dsl_dataset_phys_t *)&dataset.dn_bonus;
2975 	if (zio_read(spa, &ds->ds_bp, objset)) {
2976 		printf("ZFS: can't read object set for dataset %ju\n",
2977 		    (uintmax_t)objnum);
2978 		return (EIO);
2979 	}
2980 
2981 	return (0);
2982 }
2983 
2984 /*
2985  * Find the object set pointed to by the BOOTFS property or the root
2986  * dataset if there is none and return its details in *objset
2987  */
2988 static int
2989 zfs_get_root(const spa_t *spa, uint64_t *objid)
2990 {
2991 	dnode_phys_t dir, propdir;
2992 	uint64_t props, bootfs, root;
2993 
2994 	*objid = 0;
2995 
2996 	/*
2997 	 * Start with the MOS directory object.
2998 	 */
2999 	if (objset_get_dnode(spa, spa->spa_mos,
3000 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3001 		printf("ZFS: can't read MOS object directory\n");
3002 		return (EIO);
3003 	}
3004 
3005 	/*
3006 	 * Lookup the pool_props and see if we can find a bootfs.
3007 	 */
3008 	if (zap_lookup(spa, &dir, DMU_POOL_PROPS,
3009 	    sizeof(props), 1, &props) == 0 &&
3010 	    objset_get_dnode(spa, spa->spa_mos, props, &propdir) == 0 &&
3011 	    zap_lookup(spa, &propdir, "bootfs",
3012 	    sizeof(bootfs), 1, &bootfs) == 0 && bootfs != 0) {
3013 		*objid = bootfs;
3014 		return (0);
3015 	}
3016 	/*
3017 	 * Lookup the root dataset directory
3018 	 */
3019 	if (zap_lookup(spa, &dir, DMU_POOL_ROOT_DATASET,
3020 	    sizeof(root), 1, &root) ||
3021 	    objset_get_dnode(spa, spa->spa_mos, root, &dir)) {
3022 		printf("ZFS: can't find root dsl_dir\n");
3023 		return (EIO);
3024 	}
3025 
3026 	/*
3027 	 * Use the information from the dataset directory's bonus buffer
3028 	 * to find the dataset object and from that the object set itself.
3029 	 */
3030 	dsl_dir_phys_t *dd = (dsl_dir_phys_t *)&dir.dn_bonus;
3031 	*objid = dd->dd_head_dataset_obj;
3032 	return (0);
3033 }
3034 
3035 static int
3036 zfs_mount(const spa_t *spa, uint64_t rootobj, struct zfsmount *mount)
3037 {
3038 
3039 	mount->spa = spa;
3040 
3041 	/*
3042 	 * Find the root object set if not explicitly provided
3043 	 */
3044 	if (rootobj == 0 && zfs_get_root(spa, &rootobj)) {
3045 		printf("ZFS: can't find root filesystem\n");
3046 		return (EIO);
3047 	}
3048 
3049 	if (zfs_mount_dataset(spa, rootobj, &mount->objset)) {
3050 		printf("ZFS: can't open root filesystem\n");
3051 		return (EIO);
3052 	}
3053 
3054 	mount->rootobj = rootobj;
3055 
3056 	return (0);
3057 }
3058 
3059 /*
3060  * callback function for feature name checks.
3061  */
3062 static int
3063 check_feature(const char *name, uint64_t value)
3064 {
3065 	int i;
3066 
3067 	if (value == 0)
3068 		return (0);
3069 	if (name[0] == '\0')
3070 		return (0);
3071 
3072 	for (i = 0; features_for_read[i] != NULL; i++) {
3073 		if (strcmp(name, features_for_read[i]) == 0)
3074 			return (0);
3075 	}
3076 	printf("ZFS: unsupported feature: %s\n", name);
3077 	return (EIO);
3078 }
3079 
3080 /*
3081  * Checks whether the MOS features that are active are supported.
3082  */
3083 static int
3084 check_mos_features(const spa_t *spa)
3085 {
3086 	dnode_phys_t dir;
3087 	zap_phys_t *zap;
3088 	uint64_t objnum;
3089 	size_t size;
3090 	int rc;
3091 
3092 	if ((rc = objset_get_dnode(spa, spa->spa_mos, DMU_OT_OBJECT_DIRECTORY,
3093 	    &dir)) != 0)
3094 		return (rc);
3095 	if ((rc = zap_lookup(spa, &dir, DMU_POOL_FEATURES_FOR_READ,
3096 	    sizeof (objnum), 1, &objnum)) != 0) {
3097 		/*
3098 		 * It is older pool without features. As we have already
3099 		 * tested the label, just return without raising the error.
3100 		 */
3101 		return (0);
3102 	}
3103 
3104 	if ((rc = objset_get_dnode(spa, spa->spa_mos, objnum, &dir)) != 0)
3105 		return (rc);
3106 
3107 	if (dir.dn_type != DMU_OTN_ZAP_METADATA)
3108 		return (EIO);
3109 
3110 	size = dir.dn_datablkszsec << SPA_MINBLOCKSHIFT;
3111 	zap = malloc(size);
3112 	if (zap == NULL)
3113 		return (ENOMEM);
3114 
3115 	if (dnode_read(spa, &dir, 0, zap, size)) {
3116 		free(zap);
3117 		return (EIO);
3118 	}
3119 
3120 	if (zap->zap_block_type == ZBT_MICRO)
3121 		rc = mzap_list((const mzap_phys_t *)zap, size, check_feature);
3122 	else
3123 		rc = fzap_list(spa, &dir, zap, check_feature);
3124 
3125 	free(zap);
3126 	return (rc);
3127 }
3128 
3129 static int
3130 load_nvlist(spa_t *spa, uint64_t obj, nvlist_t **value)
3131 {
3132 	dnode_phys_t dir;
3133 	size_t size;
3134 	int rc;
3135 	unsigned char *nv;
3136 
3137 	*value = NULL;
3138 	if ((rc = objset_get_dnode(spa, spa->spa_mos, obj, &dir)) != 0)
3139 		return (rc);
3140 	if (dir.dn_type != DMU_OT_PACKED_NVLIST &&
3141 	    dir.dn_bonustype != DMU_OT_PACKED_NVLIST_SIZE) {
3142 		return (EIO);
3143 	}
3144 
3145 	if (dir.dn_bonuslen != sizeof (uint64_t))
3146 		return (EIO);
3147 
3148 	size = *(uint64_t *)DN_BONUS(&dir);
3149 	nv = malloc(size);
3150 	if (nv == NULL)
3151 		return (ENOMEM);
3152 
3153 	rc = dnode_read(spa, &dir, 0, nv, size);
3154 	if (rc != 0) {
3155 		free(nv);
3156 		nv = NULL;
3157 		return (rc);
3158 	}
3159 	*value = nvlist_import(nv + 4, nv[0], nv[1]);
3160 	free(nv);
3161 	return (rc);
3162 }
3163 
3164 static int
3165 zfs_spa_init(spa_t *spa)
3166 {
3167 	struct uberblock checkpoint;
3168 	dnode_phys_t dir;
3169 	uint64_t config_object;
3170 	nvlist_t *nvlist;
3171 	int rc;
3172 
3173 	if (zio_read(spa, &spa->spa_uberblock->ub_rootbp, spa->spa_mos)) {
3174 		printf("ZFS: can't read MOS of pool %s\n", spa->spa_name);
3175 		return (EIO);
3176 	}
3177 	if (spa->spa_mos->os_type != DMU_OST_META) {
3178 		printf("ZFS: corrupted MOS of pool %s\n", spa->spa_name);
3179 		return (EIO);
3180 	}
3181 
3182 	if (objset_get_dnode(spa, &spa->spa_mos_master,
3183 	    DMU_POOL_DIRECTORY_OBJECT, &dir)) {
3184 		printf("ZFS: failed to read pool %s directory object\n",
3185 		    spa->spa_name);
3186 		return (EIO);
3187 	}
3188 	/* this is allowed to fail, older pools do not have salt */
3189 	rc = zap_lookup(spa, &dir, DMU_POOL_CHECKSUM_SALT, 1,
3190 	    sizeof (spa->spa_cksum_salt.zcs_bytes),
3191 	    spa->spa_cksum_salt.zcs_bytes);
3192 
3193 	rc = check_mos_features(spa);
3194 	if (rc != 0) {
3195 		printf("ZFS: pool %s is not supported\n", spa->spa_name);
3196 		return (rc);
3197 	}
3198 
3199 	rc = zap_lookup(spa, &dir, DMU_POOL_CONFIG,
3200 	    sizeof (config_object), 1, &config_object);
3201 	if (rc != 0) {
3202 		printf("ZFS: can not read MOS %s\n", DMU_POOL_CONFIG);
3203 		return (EIO);
3204 	}
3205 	rc = load_nvlist(spa, config_object, &nvlist);
3206 	if (rc != 0)
3207 		return (rc);
3208 
3209 	rc = zap_lookup(spa, &dir, DMU_POOL_ZPOOL_CHECKPOINT,
3210 	    sizeof(uint64_t), sizeof(checkpoint) / sizeof(uint64_t),
3211 	    &checkpoint);
3212 	if (rc == 0 && checkpoint.ub_checkpoint_txg != 0) {
3213 		memcpy(&spa->spa_uberblock_checkpoint, &checkpoint,
3214 		    sizeof(checkpoint));
3215 		if (zio_read(spa, &spa->spa_uberblock_checkpoint.ub_rootbp,
3216 		    &spa->spa_mos_checkpoint)) {
3217 			printf("ZFS: can not read checkpoint data.\n");
3218 			return (EIO);
3219 		}
3220 	}
3221 
3222 	/*
3223 	 * Update vdevs from MOS config. Note, we do skip encoding bytes
3224 	 * here. See also vdev_label_read_config().
3225 	 */
3226 	rc = vdev_init_from_nvlist(spa, nvlist);
3227 	nvlist_destroy(nvlist);
3228 	return (rc);
3229 }
3230 
3231 static int
3232 zfs_dnode_stat(const spa_t *spa, dnode_phys_t *dn, struct stat *sb)
3233 {
3234 
3235 	if (dn->dn_bonustype != DMU_OT_SA) {
3236 		znode_phys_t *zp = (znode_phys_t *)dn->dn_bonus;
3237 
3238 		sb->st_mode = zp->zp_mode;
3239 		sb->st_uid = zp->zp_uid;
3240 		sb->st_gid = zp->zp_gid;
3241 		sb->st_size = zp->zp_size;
3242 	} else {
3243 		sa_hdr_phys_t *sahdrp;
3244 		int hdrsize;
3245 		size_t size = 0;
3246 		void *buf = NULL;
3247 
3248 		if (dn->dn_bonuslen != 0)
3249 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3250 		else {
3251 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) != 0) {
3252 				blkptr_t *bp = DN_SPILL_BLKPTR(dn);
3253 				int error;
3254 
3255 				size = BP_GET_LSIZE(bp);
3256 				buf = malloc(size);
3257 				if (buf == NULL)
3258 					error = ENOMEM;
3259 				else
3260 					error = zio_read(spa, bp, buf);
3261 
3262 				if (error != 0) {
3263 					free(buf);
3264 					return (error);
3265 				}
3266 				sahdrp = buf;
3267 			} else {
3268 				return (EIO);
3269 			}
3270 		}
3271 		hdrsize = SA_HDR_SIZE(sahdrp);
3272 		sb->st_mode = *(uint64_t *)((char *)sahdrp + hdrsize +
3273 		    SA_MODE_OFFSET);
3274 		sb->st_uid = *(uint64_t *)((char *)sahdrp + hdrsize +
3275 		    SA_UID_OFFSET);
3276 		sb->st_gid = *(uint64_t *)((char *)sahdrp + hdrsize +
3277 		    SA_GID_OFFSET);
3278 		sb->st_size = *(uint64_t *)((char *)sahdrp + hdrsize +
3279 		    SA_SIZE_OFFSET);
3280 		free(buf);
3281 	}
3282 
3283 	return (0);
3284 }
3285 
3286 static int
3287 zfs_dnode_readlink(const spa_t *spa, dnode_phys_t *dn, char *path, size_t psize)
3288 {
3289 	int rc = 0;
3290 
3291 	if (dn->dn_bonustype == DMU_OT_SA) {
3292 		sa_hdr_phys_t *sahdrp = NULL;
3293 		size_t size = 0;
3294 		void *buf = NULL;
3295 		int hdrsize;
3296 		char *p;
3297 
3298 		if (dn->dn_bonuslen != 0) {
3299 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(dn);
3300 		} else {
3301 			blkptr_t *bp;
3302 
3303 			if ((dn->dn_flags & DNODE_FLAG_SPILL_BLKPTR) == 0)
3304 				return (EIO);
3305 			bp = DN_SPILL_BLKPTR(dn);
3306 
3307 			size = BP_GET_LSIZE(bp);
3308 			buf = malloc(size);
3309 			if (buf == NULL)
3310 				rc = ENOMEM;
3311 			else
3312 				rc = zio_read(spa, bp, buf);
3313 			if (rc != 0) {
3314 				free(buf);
3315 				return (rc);
3316 			}
3317 			sahdrp = buf;
3318 		}
3319 		hdrsize = SA_HDR_SIZE(sahdrp);
3320 		p = (char *)((uintptr_t)sahdrp + hdrsize + SA_SYMLINK_OFFSET);
3321 		memcpy(path, p, psize);
3322 		free(buf);
3323 		return (0);
3324 	}
3325 	/*
3326 	 * Second test is purely to silence bogus compiler
3327 	 * warning about accessing past the end of dn_bonus.
3328 	 */
3329 	if (psize + sizeof(znode_phys_t) <= dn->dn_bonuslen &&
3330 	    sizeof(znode_phys_t) <= sizeof(dn->dn_bonus)) {
3331 		memcpy(path, &dn->dn_bonus[sizeof(znode_phys_t)], psize);
3332 	} else {
3333 		rc = dnode_read(spa, dn, 0, path, psize);
3334 	}
3335 	return (rc);
3336 }
3337 
3338 struct obj_list {
3339 	uint64_t		objnum;
3340 	STAILQ_ENTRY(obj_list)	entry;
3341 };
3342 
3343 /*
3344  * Lookup a file and return its dnode.
3345  */
3346 static int
3347 zfs_lookup(const struct zfsmount *mount, const char *upath, dnode_phys_t *dnode)
3348 {
3349 	int rc;
3350 	uint64_t objnum;
3351 	const spa_t *spa;
3352 	dnode_phys_t dn;
3353 	const char *p, *q;
3354 	char element[256];
3355 	char path[1024];
3356 	int symlinks_followed = 0;
3357 	struct stat sb;
3358 	struct obj_list *entry, *tentry;
3359 	STAILQ_HEAD(, obj_list) on_cache = STAILQ_HEAD_INITIALIZER(on_cache);
3360 
3361 	spa = mount->spa;
3362 	if (mount->objset.os_type != DMU_OST_ZFS) {
3363 		printf("ZFS: unexpected object set type %ju\n",
3364 		    (uintmax_t)mount->objset.os_type);
3365 		return (EIO);
3366 	}
3367 
3368 	if ((entry = malloc(sizeof(struct obj_list))) == NULL)
3369 		return (ENOMEM);
3370 
3371 	/*
3372 	 * Get the root directory dnode.
3373 	 */
3374 	rc = objset_get_dnode(spa, &mount->objset, MASTER_NODE_OBJ, &dn);
3375 	if (rc) {
3376 		free(entry);
3377 		return (rc);
3378 	}
3379 
3380 	rc = zap_lookup(spa, &dn, ZFS_ROOT_OBJ, sizeof(objnum), 1, &objnum);
3381 	if (rc) {
3382 		free(entry);
3383 		return (rc);
3384 	}
3385 	entry->objnum = objnum;
3386 	STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3387 
3388 	rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3389 	if (rc != 0)
3390 		goto done;
3391 
3392 	p = upath;
3393 	while (p && *p) {
3394 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3395 		if (rc != 0)
3396 			goto done;
3397 
3398 		while (*p == '/')
3399 			p++;
3400 		if (*p == '\0')
3401 			break;
3402 		q = p;
3403 		while (*q != '\0' && *q != '/')
3404 			q++;
3405 
3406 		/* skip dot */
3407 		if (p + 1 == q && p[0] == '.') {
3408 			p++;
3409 			continue;
3410 		}
3411 		/* double dot */
3412 		if (p + 2 == q && p[0] == '.' && p[1] == '.') {
3413 			p += 2;
3414 			if (STAILQ_FIRST(&on_cache) ==
3415 			    STAILQ_LAST(&on_cache, obj_list, entry)) {
3416 				rc = ENOENT;
3417 				goto done;
3418 			}
3419 			entry = STAILQ_FIRST(&on_cache);
3420 			STAILQ_REMOVE_HEAD(&on_cache, entry);
3421 			free(entry);
3422 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3423 			continue;
3424 		}
3425 		if (q - p + 1 > sizeof(element)) {
3426 			rc = ENAMETOOLONG;
3427 			goto done;
3428 		}
3429 		memcpy(element, p, q - p);
3430 		element[q - p] = 0;
3431 		p = q;
3432 
3433 		if ((rc = zfs_dnode_stat(spa, &dn, &sb)) != 0)
3434 			goto done;
3435 		if (!S_ISDIR(sb.st_mode)) {
3436 			rc = ENOTDIR;
3437 			goto done;
3438 		}
3439 
3440 		rc = zap_lookup(spa, &dn, element, sizeof (objnum), 1, &objnum);
3441 		if (rc)
3442 			goto done;
3443 		objnum = ZFS_DIRENT_OBJ(objnum);
3444 
3445 		if ((entry = malloc(sizeof(struct obj_list))) == NULL) {
3446 			rc = ENOMEM;
3447 			goto done;
3448 		}
3449 		entry->objnum = objnum;
3450 		STAILQ_INSERT_HEAD(&on_cache, entry, entry);
3451 		rc = objset_get_dnode(spa, &mount->objset, objnum, &dn);
3452 		if (rc)
3453 			goto done;
3454 
3455 		/*
3456 		 * Check for symlink.
3457 		 */
3458 		rc = zfs_dnode_stat(spa, &dn, &sb);
3459 		if (rc)
3460 			goto done;
3461 		if (S_ISLNK(sb.st_mode)) {
3462 			if (symlinks_followed > 10) {
3463 				rc = EMLINK;
3464 				goto done;
3465 			}
3466 			symlinks_followed++;
3467 
3468 			/*
3469 			 * Read the link value and copy the tail of our
3470 			 * current path onto the end.
3471 			 */
3472 			if (sb.st_size + strlen(p) + 1 > sizeof(path)) {
3473 				rc = ENAMETOOLONG;
3474 				goto done;
3475 			}
3476 			strcpy(&path[sb.st_size], p);
3477 
3478 			rc = zfs_dnode_readlink(spa, &dn, path, sb.st_size);
3479 			if (rc != 0)
3480 				goto done;
3481 
3482 			/*
3483 			 * Restart with the new path, starting either at
3484 			 * the root or at the parent depending whether or
3485 			 * not the link is relative.
3486 			 */
3487 			p = path;
3488 			if (*p == '/') {
3489 				while (STAILQ_FIRST(&on_cache) !=
3490 				    STAILQ_LAST(&on_cache, obj_list, entry)) {
3491 					entry = STAILQ_FIRST(&on_cache);
3492 					STAILQ_REMOVE_HEAD(&on_cache, entry);
3493 					free(entry);
3494 				}
3495 			} else {
3496 				entry = STAILQ_FIRST(&on_cache);
3497 				STAILQ_REMOVE_HEAD(&on_cache, entry);
3498 				free(entry);
3499 			}
3500 			objnum = (STAILQ_FIRST(&on_cache))->objnum;
3501 		}
3502 	}
3503 
3504 	*dnode = dn;
3505 done:
3506 	STAILQ_FOREACH_SAFE(entry, &on_cache, entry, tentry)
3507 		free(entry);
3508 	return (rc);
3509 }
3510