xref: /openbmc/u-boot/fs/zfs/zfs.c (revision 7ca6f363)
1 /*
2  *
3  * ZFS filesystem ported to u-boot by
4  * Jorgen Lundman <lundman at lundman.net>
5  *
6  *	GRUB  --  GRand Unified Bootloader
7  *	Copyright (C) 1999,2000,2001,2002,2003,2004
8  *	Free Software Foundation, Inc.
9  *	Copyright 2004	Sun Microsystems, Inc.
10  *
11  * SPDX-License-Identifier:	GPL-2.0+
12  */
13 
14 #include <common.h>
15 #include <malloc.h>
16 #include <linux/stat.h>
17 #include <linux/time.h>
18 #include <linux/ctype.h>
19 #include <asm/byteorder.h>
20 #include "zfs_common.h"
21 #include "div64.h"
22 
23 block_dev_desc_t *zfs_dev_desc;
24 
25 /*
26  * The zfs plug-in routines for GRUB are:
27  *
28  * zfs_mount() - locates a valid uberblock of the root pool and reads
29  *		in its MOS at the memory address MOS.
30  *
31  * zfs_open() - locates a plain file object by following the MOS
32  *		and places its dnode at the memory address DNODE.
33  *
34  * zfs_read() - read in the data blocks pointed by the DNODE.
35  *
36  */
37 
38 #include <zfs/zfs.h>
39 #include <zfs/zio.h>
40 #include <zfs/dnode.h>
41 #include <zfs/uberblock_impl.h>
42 #include <zfs/vdev_impl.h>
43 #include <zfs/zio_checksum.h>
44 #include <zfs/zap_impl.h>
45 #include <zfs/zap_leaf.h>
46 #include <zfs/zfs_znode.h>
47 #include <zfs/dmu.h>
48 #include <zfs/dmu_objset.h>
49 #include <zfs/sa_impl.h>
50 #include <zfs/dsl_dir.h>
51 #include <zfs/dsl_dataset.h>
52 
53 
54 #define	ZPOOL_PROP_BOOTFS		"bootfs"
55 
56 
57 /*
58  * For nvlist manipulation. (from nvpair.h)
59  */
60 #define	NV_ENCODE_NATIVE	0
61 #define	NV_ENCODE_XDR		1
62 #define	NV_BIG_ENDIAN			0
63 #define	NV_LITTLE_ENDIAN	1
64 #define	DATA_TYPE_UINT64	8
65 #define	DATA_TYPE_STRING	9
66 #define	DATA_TYPE_NVLIST	19
67 #define	DATA_TYPE_NVLIST_ARRAY	20
68 
69 
70 /*
71  * Macros to get fields in a bp or DVA.
72  */
73 #define	P2PHASE(x, align)		((x) & ((align) - 1))
74 #define	DVA_OFFSET_TO_PHYS_SECTOR(offset)					\
75 	((offset + VDEV_LABEL_START_SIZE) >> SPA_MINBLOCKSHIFT)
76 
77 /*
78  * return x rounded down to an align boundary
79  * eg, P2ALIGN(1200, 1024) == 1024 (1*align)
80  * eg, P2ALIGN(1024, 1024) == 1024 (1*align)
81  * eg, P2ALIGN(0x1234, 0x100) == 0x1200 (0x12*align)
82  * eg, P2ALIGN(0x5600, 0x100) == 0x5600 (0x56*align)
83  */
84 #define	P2ALIGN(x, align)		((x) & -(align))
85 
86 /*
87  * FAT ZAP data structures
88  */
89 #define	ZFS_CRC64_POLY 0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
90 #define	ZAP_HASH_IDX(hash, n)	(((n) == 0) ? 0 : ((hash) >> (64 - (n))))
91 #define	CHAIN_END	0xffff	/* end of the chunk chain */
92 
93 /*
94  * The amount of space within the chunk available for the array is:
95  * chunk size - space for type (1) - space for next pointer (2)
96  */
97 #define	ZAP_LEAF_ARRAY_BYTES (ZAP_LEAF_CHUNKSIZE - 3)
98 
99 #define	ZAP_LEAF_HASH_SHIFT(bs)	(bs - 5)
100 #define	ZAP_LEAF_HASH_NUMENTRIES(bs) (1 << ZAP_LEAF_HASH_SHIFT(bs))
101 #define	LEAF_HASH(bs, h)												\
102 	((ZAP_LEAF_HASH_NUMENTRIES(bs)-1) &									\
103 	 ((h) >> (64 - ZAP_LEAF_HASH_SHIFT(bs)-l->l_hdr.lh_prefix_len)))
104 
105 /*
106  * The amount of space available for chunks is:
107  * block size shift - hash entry size (2) * number of hash
108  * entries - header space (2*chunksize)
109  */
110 #define	ZAP_LEAF_NUMCHUNKS(bs)						\
111 	(((1<<bs) - 2*ZAP_LEAF_HASH_NUMENTRIES(bs)) /	\
112 	 ZAP_LEAF_CHUNKSIZE - 2)
113 
114 /*
115  * The chunks start immediately after the hash table.  The end of the
116  * hash table is at l_hash + HASH_NUMENTRIES, which we simply cast to a
117  * chunk_t.
118  */
119 #define	ZAP_LEAF_CHUNK(l, bs, idx)										\
120 	((zap_leaf_chunk_t *)(l->l_hash + ZAP_LEAF_HASH_NUMENTRIES(bs)))[idx]
121 #define	ZAP_LEAF_ENTRY(l, bs, idx) (&ZAP_LEAF_CHUNK(l, bs, idx).l_entry)
122 
123 
124 /*
125  * Decompression Entry - lzjb
126  */
127 #ifndef	NBBY
128 #define	NBBY	8
129 #endif
130 
131 
132 
133 typedef int zfs_decomp_func_t(void *s_start, void *d_start,
134 							  uint32_t s_len, uint32_t d_len);
135 typedef struct decomp_entry {
136 	char *name;
137 	zfs_decomp_func_t *decomp_func;
138 } decomp_entry_t;
139 
140 typedef struct dnode_end {
141 	dnode_phys_t dn;
142 	zfs_endian_t endian;
143 } dnode_end_t;
144 
145 struct zfs_data {
146 	/* cache for a file block of the currently zfs_open()-ed file */
147 	char *file_buf;
148 	uint64_t file_start;
149 	uint64_t file_end;
150 
151 	/* XXX: ashift is per vdev, not per pool.  We currently only ever touch
152 	 * a single vdev, but when/if raid-z or stripes are supported, this
153 	 * may need revision.
154 	 */
155 	uint64_t vdev_ashift;
156 	uint64_t label_txg;
157 	uint64_t pool_guid;
158 
159 	/* cache for a dnode block */
160 	dnode_phys_t *dnode_buf;
161 	dnode_phys_t *dnode_mdn;
162 	uint64_t dnode_start;
163 	uint64_t dnode_end;
164 	zfs_endian_t dnode_endian;
165 
166 	uberblock_t current_uberblock;
167 
168 	dnode_end_t mos;
169 	dnode_end_t mdn;
170 	dnode_end_t dnode;
171 
172 	uint64_t vdev_phys_sector;
173 
174 	int (*userhook)(const char *, const struct zfs_dirhook_info *);
175 	struct zfs_dirhook_info *dirinfo;
176 
177 };
178 
179 
180 
181 
182 static int
183 zlib_decompress(void *s, void *d,
184 				uint32_t slen, uint32_t dlen)
185 {
186 	if (zlib_decompress(s, d, slen, dlen) < 0)
187 		return ZFS_ERR_BAD_FS;
188 	return ZFS_ERR_NONE;
189 }
190 
191 static decomp_entry_t decomp_table[ZIO_COMPRESS_FUNCTIONS] = {
192 	{"inherit", NULL},		/* ZIO_COMPRESS_INHERIT */
193 	{"on", lzjb_decompress},	/* ZIO_COMPRESS_ON */
194 	{"off", NULL},		/* ZIO_COMPRESS_OFF */
195 	{"lzjb", lzjb_decompress},	/* ZIO_COMPRESS_LZJB */
196 	{"empty", NULL},		/* ZIO_COMPRESS_EMPTY */
197 	{"gzip-1", zlib_decompress},  /* ZIO_COMPRESS_GZIP1 */
198 	{"gzip-2", zlib_decompress},  /* ZIO_COMPRESS_GZIP2 */
199 	{"gzip-3", zlib_decompress},  /* ZIO_COMPRESS_GZIP3 */
200 	{"gzip-4", zlib_decompress},  /* ZIO_COMPRESS_GZIP4 */
201 	{"gzip-5", zlib_decompress},  /* ZIO_COMPRESS_GZIP5 */
202 	{"gzip-6", zlib_decompress},  /* ZIO_COMPRESS_GZIP6 */
203 	{"gzip-7", zlib_decompress},  /* ZIO_COMPRESS_GZIP7 */
204 	{"gzip-8", zlib_decompress},  /* ZIO_COMPRESS_GZIP8 */
205 	{"gzip-9", zlib_decompress},  /* ZIO_COMPRESS_GZIP9 */
206 };
207 
208 
209 
210 static int zio_read_data(blkptr_t *bp, zfs_endian_t endian,
211 						 void *buf, struct zfs_data *data);
212 
213 static int
214 zio_read(blkptr_t *bp, zfs_endian_t endian, void **buf,
215 		 size_t *size, struct zfs_data *data);
216 
217 /*
218  * Our own version of log2().  Same thing as highbit()-1.
219  */
220 static int
221 zfs_log2(uint64_t num)
222 {
223 	int i = 0;
224 
225 	while (num > 1) {
226 		i++;
227 		num = num >> 1;
228 	}
229 
230 	return i;
231 }
232 
233 
234 /* Checksum Functions */
235 static void
236 zio_checksum_off(const void *buf __attribute__ ((unused)),
237 				 uint64_t size __attribute__ ((unused)),
238 				 zfs_endian_t endian __attribute__ ((unused)),
239 				 zio_cksum_t *zcp)
240 {
241 	ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
242 }
243 
244 /* Checksum Table and Values */
245 static zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = {
246 	{NULL, 0, 0, "inherit"},
247 	{NULL, 0, 0, "on"},
248 	{zio_checksum_off, 0, 0, "off"},
249 	{zio_checksum_SHA256, 1, 1, "label"},
250 	{zio_checksum_SHA256, 1, 1, "gang_header"},
251 	{NULL, 0, 0, "zilog"},
252 	{fletcher_2_endian, 0, 0, "fletcher2"},
253 	{fletcher_4_endian, 1, 0, "fletcher4"},
254 	{zio_checksum_SHA256, 1, 0, "SHA256"},
255 	{NULL, 0, 0, "zilog2"},
256 };
257 
258 /*
259  * zio_checksum_verify: Provides support for checksum verification.
260  *
261  * Fletcher2, Fletcher4, and SHA256 are supported.
262  *
263  */
264 static int
265 zio_checksum_verify(zio_cksum_t zc, uint32_t checksum,
266 					zfs_endian_t endian, char *buf, int size)
267 {
268 	zio_eck_t *zec = (zio_eck_t *) (buf + size) - 1;
269 	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
270 	zio_cksum_t actual_cksum, expected_cksum;
271 
272 	if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func == NULL) {
273 		printf("zfs unknown checksum function %d\n", checksum);
274 		return ZFS_ERR_NOT_IMPLEMENTED_YET;
275 	}
276 
277 	if (ci->ci_eck) {
278 		expected_cksum = zec->zec_cksum;
279 		zec->zec_cksum = zc;
280 		ci->ci_func(buf, size, endian, &actual_cksum);
281 		zec->zec_cksum = expected_cksum;
282 		zc = expected_cksum;
283 	} else {
284 		ci->ci_func(buf, size, endian, &actual_cksum);
285 	}
286 
287 	if ((actual_cksum.zc_word[0] != zc.zc_word[0])
288 		|| (actual_cksum.zc_word[1] != zc.zc_word[1])
289 		|| (actual_cksum.zc_word[2] != zc.zc_word[2])
290 		|| (actual_cksum.zc_word[3] != zc.zc_word[3])) {
291 		return ZFS_ERR_BAD_FS;
292 	}
293 
294 	return ZFS_ERR_NONE;
295 }
296 
297 /*
298  * vdev_uberblock_compare takes two uberblock structures and returns an integer
299  * indicating the more recent of the two.
300  *	Return Value = 1 if ub2 is more recent
301  *	Return Value = -1 if ub1 is more recent
302  * The most recent uberblock is determined using its transaction number and
303  * timestamp.  The uberblock with the highest transaction number is
304  * considered "newer".	If the transaction numbers of the two blocks match, the
305  * timestamps are compared to determine the "newer" of the two.
306  */
307 static int
308 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
309 {
310 	zfs_endian_t ub1_endian, ub2_endian;
311 	if (zfs_to_cpu64(ub1->ub_magic, LITTLE_ENDIAN) == UBERBLOCK_MAGIC)
312 		ub1_endian = LITTLE_ENDIAN;
313 	else
314 		ub1_endian = BIG_ENDIAN;
315 	if (zfs_to_cpu64(ub2->ub_magic, LITTLE_ENDIAN) == UBERBLOCK_MAGIC)
316 		ub2_endian = LITTLE_ENDIAN;
317 	else
318 		ub2_endian = BIG_ENDIAN;
319 
320 	if (zfs_to_cpu64(ub1->ub_txg, ub1_endian)
321 		< zfs_to_cpu64(ub2->ub_txg, ub2_endian))
322 		return -1;
323 	if (zfs_to_cpu64(ub1->ub_txg, ub1_endian)
324 		> zfs_to_cpu64(ub2->ub_txg, ub2_endian))
325 		return 1;
326 
327 	if (zfs_to_cpu64(ub1->ub_timestamp, ub1_endian)
328 		< zfs_to_cpu64(ub2->ub_timestamp, ub2_endian))
329 		return -1;
330 	if (zfs_to_cpu64(ub1->ub_timestamp, ub1_endian)
331 		> zfs_to_cpu64(ub2->ub_timestamp, ub2_endian))
332 		return 1;
333 
334 	return 0;
335 }
336 
337 /*
338  * Three pieces of information are needed to verify an uberblock: the magic
339  * number, the version number, and the checksum.
340  *
341  * Currently Implemented: version number, magic number, label txg
342  * Need to Implement: checksum
343  *
344  */
345 static int
346 uberblock_verify(uberblock_t *uber, int offset, struct zfs_data *data)
347 {
348 	int err;
349 	zfs_endian_t endian = UNKNOWN_ENDIAN;
350 	zio_cksum_t zc;
351 
352 	if (uber->ub_txg < data->label_txg) {
353 		debug("ignoring partially written label: uber_txg < label_txg %llu %llu\n",
354 			  uber->ub_txg, data->label_txg);
355 		return ZFS_ERR_BAD_FS;
356 	}
357 
358 	if (zfs_to_cpu64(uber->ub_magic, LITTLE_ENDIAN) == UBERBLOCK_MAGIC
359 		&& zfs_to_cpu64(uber->ub_version, LITTLE_ENDIAN) > 0
360 		&& zfs_to_cpu64(uber->ub_version, LITTLE_ENDIAN) <= SPA_VERSION)
361 		endian = LITTLE_ENDIAN;
362 
363 	if (zfs_to_cpu64(uber->ub_magic, BIG_ENDIAN) == UBERBLOCK_MAGIC
364 		&& zfs_to_cpu64(uber->ub_version, BIG_ENDIAN) > 0
365 		&& zfs_to_cpu64(uber->ub_version, BIG_ENDIAN) <= SPA_VERSION)
366 		endian = BIG_ENDIAN;
367 
368 	if (endian == UNKNOWN_ENDIAN) {
369 		printf("invalid uberblock magic\n");
370 		return ZFS_ERR_BAD_FS;
371 	}
372 
373 	memset(&zc, 0, sizeof(zc));
374 	zc.zc_word[0] = cpu_to_zfs64(offset, endian);
375 	err = zio_checksum_verify(zc, ZIO_CHECKSUM_LABEL, endian,
376 							  (char *) uber, UBERBLOCK_SIZE(data->vdev_ashift));
377 
378 	if (!err) {
379 		/* Check that the data pointed by the rootbp is usable. */
380 		void *osp = NULL;
381 		size_t ospsize;
382 		err = zio_read(&uber->ub_rootbp, endian, &osp, &ospsize, data);
383 		free(osp);
384 
385 		if (!err && ospsize < OBJSET_PHYS_SIZE_V14) {
386 			printf("uberblock rootbp points to invalid data\n");
387 			return ZFS_ERR_BAD_FS;
388 		}
389 	}
390 
391 	return err;
392 }
393 
394 /*
395  * Find the best uberblock.
396  * Return:
397  *	  Success - Pointer to the best uberblock.
398  *	  Failure - NULL
399  */
400 static uberblock_t *find_bestub(char *ub_array, struct zfs_data *data)
401 {
402 	const uint64_t sector = data->vdev_phys_sector;
403 	uberblock_t *ubbest = NULL;
404 	uberblock_t *ubnext;
405 	unsigned int i, offset, pickedub = 0;
406 	int err = ZFS_ERR_NONE;
407 
408 	const unsigned int UBCOUNT = UBERBLOCK_COUNT(data->vdev_ashift);
409 	const uint64_t UBBYTES = UBERBLOCK_SIZE(data->vdev_ashift);
410 
411 	for (i = 0; i < UBCOUNT; i++) {
412 		ubnext = (uberblock_t *) (i * UBBYTES + ub_array);
413 		offset = (sector << SPA_MINBLOCKSHIFT) + VDEV_PHYS_SIZE + (i * UBBYTES);
414 
415 		err = uberblock_verify(ubnext, offset, data);
416 		if (err)
417 			continue;
418 
419 		if (ubbest == NULL || vdev_uberblock_compare(ubnext, ubbest) > 0) {
420 			ubbest = ubnext;
421 			pickedub = i;
422 		}
423 	}
424 
425 	if (ubbest)
426 		debug("zfs Found best uberblock at idx %d, txg %llu\n",
427 			  pickedub, (unsigned long long) ubbest->ub_txg);
428 
429 	return ubbest;
430 }
431 
432 static inline size_t
433 get_psize(blkptr_t *bp, zfs_endian_t endian)
434 {
435 	return (((zfs_to_cpu64((bp)->blk_prop, endian) >> 16) & 0xffff) + 1)
436 			<< SPA_MINBLOCKSHIFT;
437 }
438 
439 static uint64_t
440 dva_get_offset(dva_t *dva, zfs_endian_t endian)
441 {
442 	return zfs_to_cpu64((dva)->dva_word[1],
443 							 endian) << SPA_MINBLOCKSHIFT;
444 }
445 
446 /*
447  * Read a block of data based on the gang block address dva,
448  * and put its data in buf.
449  *
450  */
451 static int
452 zio_read_gang(blkptr_t *bp, zfs_endian_t endian, dva_t *dva, void *buf,
453 			  struct zfs_data *data)
454 {
455 	zio_gbh_phys_t *zio_gb;
456 	uint64_t offset, sector;
457 	unsigned i;
458 	int err;
459 	zio_cksum_t zc;
460 
461 	memset(&zc, 0, sizeof(zc));
462 
463 	zio_gb = malloc(SPA_GANGBLOCKSIZE);
464 	if (!zio_gb)
465 		return ZFS_ERR_OUT_OF_MEMORY;
466 
467 	offset = dva_get_offset(dva, endian);
468 	sector = DVA_OFFSET_TO_PHYS_SECTOR(offset);
469 
470 	/* read in the gang block header */
471 	err = zfs_devread(sector, 0, SPA_GANGBLOCKSIZE, (char *) zio_gb);
472 
473 	if (err) {
474 		free(zio_gb);
475 		return err;
476 	}
477 
478 	/* XXX */
479 	/* self checksuming the gang block header */
480 	ZIO_SET_CHECKSUM(&zc, DVA_GET_VDEV(dva),
481 					 dva_get_offset(dva, endian), bp->blk_birth, 0);
482 	err = zio_checksum_verify(zc, ZIO_CHECKSUM_GANG_HEADER, endian,
483 							  (char *) zio_gb, SPA_GANGBLOCKSIZE);
484 	if (err) {
485 		free(zio_gb);
486 		return err;
487 	}
488 
489 	endian = (zfs_to_cpu64(bp->blk_prop, endian) >> 63) & 1;
490 
491 	for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
492 		if (zio_gb->zg_blkptr[i].blk_birth == 0)
493 			continue;
494 
495 		err = zio_read_data(&zio_gb->zg_blkptr[i], endian, buf, data);
496 		if (err) {
497 			free(zio_gb);
498 			return err;
499 		}
500 		buf = (char *) buf + get_psize(&zio_gb->zg_blkptr[i], endian);
501 	}
502 	free(zio_gb);
503 	return ZFS_ERR_NONE;
504 }
505 
506 /*
507  * Read in a block of raw data to buf.
508  */
509 static int
510 zio_read_data(blkptr_t *bp, zfs_endian_t endian, void *buf,
511 			  struct zfs_data *data)
512 {
513 	int i, psize;
514 	int err = ZFS_ERR_NONE;
515 
516 	psize = get_psize(bp, endian);
517 
518 	/* pick a good dva from the block pointer */
519 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
520 		uint64_t offset, sector;
521 
522 		if (bp->blk_dva[i].dva_word[0] == 0 && bp->blk_dva[i].dva_word[1] == 0)
523 			continue;
524 
525 		if ((zfs_to_cpu64(bp->blk_dva[i].dva_word[1], endian)>>63) & 1) {
526 			err = zio_read_gang(bp, endian, &bp->blk_dva[i], buf, data);
527 		} else {
528 			/* read in a data block */
529 			offset = dva_get_offset(&bp->blk_dva[i], endian);
530 			sector = DVA_OFFSET_TO_PHYS_SECTOR(offset);
531 
532 			err = zfs_devread(sector, 0, psize, buf);
533 		}
534 
535 		if (!err) {
536 			/*Check the underlying checksum before we rule this DVA as "good"*/
537 			uint32_t checkalgo = (zfs_to_cpu64((bp)->blk_prop, endian) >> 40) & 0xff;
538 
539 			err = zio_checksum_verify(bp->blk_cksum, checkalgo, endian, buf, psize);
540 			if (!err)
541 				return ZFS_ERR_NONE;
542 		}
543 
544 		/* If read failed or checksum bad, reset the error.	 Hopefully we've got some more DVA's to try.*/
545 	}
546 
547 	if (!err) {
548 		printf("couldn't find a valid DVA\n");
549 		err = ZFS_ERR_BAD_FS;
550 	}
551 
552 	return err;
553 }
554 
555 /*
556  * Read in a block of data, verify its checksum, decompress if needed,
557  * and put the uncompressed data in buf.
558  */
559 static int
560 zio_read(blkptr_t *bp, zfs_endian_t endian, void **buf,
561 		 size_t *size, struct zfs_data *data)
562 {
563 	size_t lsize, psize;
564 	unsigned int comp;
565 	char *compbuf = NULL;
566 	int err;
567 
568 	*buf = NULL;
569 
570 	comp = (zfs_to_cpu64((bp)->blk_prop, endian)>>32) & 0xff;
571 	lsize = (BP_IS_HOLE(bp) ? 0 :
572 			 (((zfs_to_cpu64((bp)->blk_prop, endian) & 0xffff) + 1)
573 			  << SPA_MINBLOCKSHIFT));
574 	psize = get_psize(bp, endian);
575 
576 	if (size)
577 		*size = lsize;
578 
579 	if (comp >= ZIO_COMPRESS_FUNCTIONS) {
580 		printf("compression algorithm %u not supported\n", (unsigned int) comp);
581 		return ZFS_ERR_NOT_IMPLEMENTED_YET;
582 	}
583 
584 	if (comp != ZIO_COMPRESS_OFF && decomp_table[comp].decomp_func == NULL) {
585 		printf("compression algorithm %s not supported\n", decomp_table[comp].name);
586 		return ZFS_ERR_NOT_IMPLEMENTED_YET;
587 	}
588 
589 	if (comp != ZIO_COMPRESS_OFF) {
590 		compbuf = malloc(psize);
591 		if (!compbuf)
592 			return ZFS_ERR_OUT_OF_MEMORY;
593 	} else {
594 		compbuf = *buf = malloc(lsize);
595 	}
596 
597 	err = zio_read_data(bp, endian, compbuf, data);
598 	if (err) {
599 		free(compbuf);
600 		*buf = NULL;
601 		return err;
602 	}
603 
604 	if (comp != ZIO_COMPRESS_OFF) {
605 		*buf = malloc(lsize);
606 		if (!*buf) {
607 			free(compbuf);
608 			return ZFS_ERR_OUT_OF_MEMORY;
609 		}
610 
611 		err = decomp_table[comp].decomp_func(compbuf, *buf, psize, lsize);
612 		free(compbuf);
613 		if (err) {
614 			free(*buf);
615 			*buf = NULL;
616 			return err;
617 		}
618 	}
619 
620 	return ZFS_ERR_NONE;
621 }
622 
623 /*
624  * Get the block from a block id.
625  * push the block onto the stack.
626  *
627  */
628 static int
629 dmu_read(dnode_end_t *dn, uint64_t blkid, void **buf,
630 		 zfs_endian_t *endian_out, struct zfs_data *data)
631 {
632 	int idx, level;
633 	blkptr_t *bp_array = dn->dn.dn_blkptr;
634 	int epbs = dn->dn.dn_indblkshift - SPA_BLKPTRSHIFT;
635 	blkptr_t *bp;
636 	void *tmpbuf = 0;
637 	zfs_endian_t endian;
638 	int err = ZFS_ERR_NONE;
639 
640 	bp = malloc(sizeof(blkptr_t));
641 	if (!bp)
642 		return ZFS_ERR_OUT_OF_MEMORY;
643 
644 	endian = dn->endian;
645 	for (level = dn->dn.dn_nlevels - 1; level >= 0; level--) {
646 		idx = (blkid >> (epbs * level)) & ((1 << epbs) - 1);
647 		*bp = bp_array[idx];
648 		if (bp_array != dn->dn.dn_blkptr) {
649 			free(bp_array);
650 			bp_array = 0;
651 		}
652 
653 		if (BP_IS_HOLE(bp)) {
654 			size_t size = zfs_to_cpu16(dn->dn.dn_datablkszsec,
655 											dn->endian)
656 				<< SPA_MINBLOCKSHIFT;
657 			*buf = malloc(size);
658 			if (*buf) {
659 				err = ZFS_ERR_OUT_OF_MEMORY;
660 				break;
661 			}
662 			memset(*buf, 0, size);
663 			endian = (zfs_to_cpu64(bp->blk_prop, endian) >> 63) & 1;
664 			break;
665 		}
666 		if (level == 0) {
667 			err = zio_read(bp, endian, buf, 0, data);
668 			endian = (zfs_to_cpu64(bp->blk_prop, endian) >> 63) & 1;
669 			break;
670 		}
671 		err = zio_read(bp, endian, &tmpbuf, 0, data);
672 		endian = (zfs_to_cpu64(bp->blk_prop, endian) >> 63) & 1;
673 		if (err)
674 			break;
675 		bp_array = tmpbuf;
676 	}
677 	if (bp_array != dn->dn.dn_blkptr)
678 		free(bp_array);
679 	if (endian_out)
680 		*endian_out = endian;
681 
682 	free(bp);
683 	return err;
684 }
685 
686 /*
687  * mzap_lookup: Looks up property described by "name" and returns the value
688  * in "value".
689  */
690 static int
691 mzap_lookup(mzap_phys_t *zapobj, zfs_endian_t endian,
692 			int objsize, char *name, uint64_t * value)
693 {
694 	int i, chunks;
695 	mzap_ent_phys_t *mzap_ent = zapobj->mz_chunk;
696 
697 	chunks = objsize / MZAP_ENT_LEN - 1;
698 	for (i = 0; i < chunks; i++) {
699 		if (strcmp(mzap_ent[i].mze_name, name) == 0) {
700 			*value = zfs_to_cpu64(mzap_ent[i].mze_value, endian);
701 			return ZFS_ERR_NONE;
702 		}
703 	}
704 
705 	printf("couldn't find '%s'\n", name);
706 	return ZFS_ERR_FILE_NOT_FOUND;
707 }
708 
709 static int
710 mzap_iterate(mzap_phys_t *zapobj, zfs_endian_t endian, int objsize,
711 			 int (*hook)(const char *name,
712 						 uint64_t val,
713 						 struct zfs_data *data),
714 			 struct zfs_data *data)
715 {
716 	int i, chunks;
717 	mzap_ent_phys_t *mzap_ent = zapobj->mz_chunk;
718 
719 	chunks = objsize / MZAP_ENT_LEN - 1;
720 	for (i = 0; i < chunks; i++) {
721 		if (hook(mzap_ent[i].mze_name,
722 				 zfs_to_cpu64(mzap_ent[i].mze_value, endian),
723 				 data))
724 			return 1;
725 	}
726 
727 	return 0;
728 }
729 
730 static uint64_t
731 zap_hash(uint64_t salt, const char *name)
732 {
733 	static uint64_t table[256];
734 	const uint8_t *cp;
735 	uint8_t c;
736 	uint64_t crc = salt;
737 
738 	if (table[128] == 0) {
739 		uint64_t *ct;
740 		int i, j;
741 		for (i = 0; i < 256; i++) {
742 			for (ct = table + i, *ct = i, j = 8; j > 0; j--)
743 				*ct = (*ct >> 1) ^ (-(*ct & 1) & ZFS_CRC64_POLY);
744 		}
745 	}
746 
747 	for (cp = (const uint8_t *) name; (c = *cp) != '\0'; cp++)
748 		crc = (crc >> 8) ^ table[(crc ^ c) & 0xFF];
749 
750 	/*
751 	 * Only use 28 bits, since we need 4 bits in the cookie for the
752 	 * collision differentiator.  We MUST use the high bits, since
753 	 * those are the onces that we first pay attention to when
754 	 * chosing the bucket.
755 	 */
756 	crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1);
757 
758 	return crc;
759 }
760 
761 /*
762  * Only to be used on 8-bit arrays.
763  * array_len is actual len in bytes (not encoded le_value_length).
764  * buf is null-terminated.
765  */
766 /* XXX */
767 static int
768 zap_leaf_array_equal(zap_leaf_phys_t *l, zfs_endian_t endian,
769 					 int blksft, int chunk, int array_len, const char *buf)
770 {
771 	int bseen = 0;
772 
773 	while (bseen < array_len) {
774 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, blksft, chunk).l_array;
775 		int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES);
776 
777 		if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft))
778 			return 0;
779 
780 		if (memcmp(la->la_array, buf + bseen, toread) != 0)
781 			break;
782 		chunk = zfs_to_cpu16(la->la_next, endian);
783 		bseen += toread;
784 	}
785 	return (bseen == array_len);
786 }
787 
788 /* XXX */
789 static int
790 zap_leaf_array_get(zap_leaf_phys_t *l, zfs_endian_t endian, int blksft,
791 				   int chunk, int array_len, char *buf)
792 {
793 	int bseen = 0;
794 
795 	while (bseen < array_len) {
796 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, blksft, chunk).l_array;
797 		int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES);
798 
799 		if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft))
800 			/* Don't use errno because this error is to be ignored.  */
801 			return ZFS_ERR_BAD_FS;
802 
803 		memcpy(buf + bseen, la->la_array,  toread);
804 		chunk = zfs_to_cpu16(la->la_next, endian);
805 		bseen += toread;
806 	}
807 	return ZFS_ERR_NONE;
808 }
809 
810 
811 /*
812  * Given a zap_leaf_phys_t, walk thru the zap leaf chunks to get the
813  * value for the property "name".
814  *
815  */
816 /* XXX */
817 static int
818 zap_leaf_lookup(zap_leaf_phys_t *l, zfs_endian_t endian,
819 				int blksft, uint64_t h,
820 				const char *name, uint64_t *value)
821 {
822 	uint16_t chunk;
823 	struct zap_leaf_entry *le;
824 
825 	/* Verify if this is a valid leaf block */
826 	if (zfs_to_cpu64(l->l_hdr.lh_block_type, endian) != ZBT_LEAF) {
827 		printf("invalid leaf type\n");
828 		return ZFS_ERR_BAD_FS;
829 	}
830 	if (zfs_to_cpu32(l->l_hdr.lh_magic, endian) != ZAP_LEAF_MAGIC) {
831 		printf("invalid leaf magic\n");
832 		return ZFS_ERR_BAD_FS;
833 	}
834 
835 	for (chunk = zfs_to_cpu16(l->l_hash[LEAF_HASH(blksft, h)], endian);
836 		 chunk != CHAIN_END; chunk = le->le_next) {
837 
838 		if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft)) {
839 			printf("invalid chunk number\n");
840 			return ZFS_ERR_BAD_FS;
841 		}
842 
843 		le = ZAP_LEAF_ENTRY(l, blksft, chunk);
844 
845 		/* Verify the chunk entry */
846 		if (le->le_type != ZAP_CHUNK_ENTRY) {
847 			printf("invalid chunk entry\n");
848 			return ZFS_ERR_BAD_FS;
849 		}
850 
851 		if (zfs_to_cpu64(le->le_hash, endian) != h)
852 			continue;
853 
854 		if (zap_leaf_array_equal(l, endian, blksft,
855 								 zfs_to_cpu16(le->le_name_chunk, endian),
856 								 zfs_to_cpu16(le->le_name_length, endian),
857 								 name)) {
858 			struct zap_leaf_array *la;
859 
860 			if (le->le_int_size != 8 || le->le_value_length != 1) {
861 				printf("invalid leaf chunk entry\n");
862 				return ZFS_ERR_BAD_FS;
863 			}
864 			/* get the uint64_t property value */
865 			la = &ZAP_LEAF_CHUNK(l, blksft, le->le_value_chunk).l_array;
866 
867 			*value = be64_to_cpu(la->la_array64);
868 
869 			return ZFS_ERR_NONE;
870 		}
871 	}
872 
873 	printf("couldn't find '%s'\n", name);
874 	return ZFS_ERR_FILE_NOT_FOUND;
875 }
876 
877 
878 /* Verify if this is a fat zap header block */
879 static int
880 zap_verify(zap_phys_t *zap)
881 {
882 	if (zap->zap_magic != (uint64_t) ZAP_MAGIC) {
883 		printf("bad ZAP magic\n");
884 		return ZFS_ERR_BAD_FS;
885 	}
886 
887 	if (zap->zap_flags != 0) {
888 		printf("bad ZAP flags\n");
889 		return ZFS_ERR_BAD_FS;
890 	}
891 
892 	if (zap->zap_salt == 0) {
893 		printf("bad ZAP salt\n");
894 		return ZFS_ERR_BAD_FS;
895 	}
896 
897 	return ZFS_ERR_NONE;
898 }
899 
900 /*
901  * Fat ZAP lookup
902  *
903  */
904 /* XXX */
905 static int
906 fzap_lookup(dnode_end_t *zap_dnode, zap_phys_t *zap,
907 			char *name, uint64_t *value, struct zfs_data *data)
908 {
909 	void *l;
910 	uint64_t hash, idx, blkid;
911 	int blksft = zfs_log2(zfs_to_cpu16(zap_dnode->dn.dn_datablkszsec,
912 											zap_dnode->endian) << DNODE_SHIFT);
913 	int err;
914 	zfs_endian_t leafendian;
915 
916 	err = zap_verify(zap);
917 	if (err)
918 		return err;
919 
920 	hash = zap_hash(zap->zap_salt, name);
921 
922 	/* get block id from index */
923 	if (zap->zap_ptrtbl.zt_numblks != 0) {
924 		printf("external pointer tables not supported\n");
925 		return ZFS_ERR_NOT_IMPLEMENTED_YET;
926 	}
927 	idx = ZAP_HASH_IDX(hash, zap->zap_ptrtbl.zt_shift);
928 	blkid = ((uint64_t *) zap)[idx + (1 << (blksft - 3 - 1))];
929 
930 	/* Get the leaf block */
931 	if ((1U << blksft) < sizeof(zap_leaf_phys_t)) {
932 		printf("ZAP leaf is too small\n");
933 		return ZFS_ERR_BAD_FS;
934 	}
935 	err = dmu_read(zap_dnode, blkid, &l, &leafendian, data);
936 	if (err)
937 		return err;
938 
939 	err = zap_leaf_lookup(l, leafendian, blksft, hash, name, value);
940 	free(l);
941 	return err;
942 }
943 
944 /* XXX */
945 static int
946 fzap_iterate(dnode_end_t *zap_dnode, zap_phys_t *zap,
947 			 int (*hook)(const char *name,
948 						 uint64_t val,
949 						 struct zfs_data *data),
950 			 struct zfs_data *data)
951 {
952 	zap_leaf_phys_t *l;
953 	void *l_in;
954 	uint64_t idx, blkid;
955 	uint16_t chunk;
956 	int blksft = zfs_log2(zfs_to_cpu16(zap_dnode->dn.dn_datablkszsec,
957 											zap_dnode->endian) << DNODE_SHIFT);
958 	int err;
959 	zfs_endian_t endian;
960 
961 	if (zap_verify(zap))
962 		return 0;
963 
964 	/* get block id from index */
965 	if (zap->zap_ptrtbl.zt_numblks != 0) {
966 		printf("external pointer tables not supported\n");
967 		return 0;
968 	}
969 	/* Get the leaf block */
970 	if ((1U << blksft) < sizeof(zap_leaf_phys_t)) {
971 		printf("ZAP leaf is too small\n");
972 		return 0;
973 	}
974 	for (idx = 0; idx < zap->zap_ptrtbl.zt_numblks; idx++) {
975 		blkid = ((uint64_t *) zap)[idx + (1 << (blksft - 3 - 1))];
976 
977 		err = dmu_read(zap_dnode, blkid, &l_in, &endian, data);
978 		l = l_in;
979 		if (err)
980 			continue;
981 
982 		/* Verify if this is a valid leaf block */
983 		if (zfs_to_cpu64(l->l_hdr.lh_block_type, endian) != ZBT_LEAF) {
984 			free(l);
985 			continue;
986 		}
987 		if (zfs_to_cpu32(l->l_hdr.lh_magic, endian) != ZAP_LEAF_MAGIC) {
988 			free(l);
989 			continue;
990 		}
991 
992 		for (chunk = 0; chunk < ZAP_LEAF_NUMCHUNKS(blksft); chunk++) {
993 			char *buf;
994 			struct zap_leaf_array *la;
995 			struct zap_leaf_entry *le;
996 			uint64_t val;
997 			le = ZAP_LEAF_ENTRY(l, blksft, chunk);
998 
999 			/* Verify the chunk entry */
1000 			if (le->le_type != ZAP_CHUNK_ENTRY)
1001 				continue;
1002 
1003 			buf = malloc(zfs_to_cpu16(le->le_name_length, endian)
1004 						 + 1);
1005 			if (zap_leaf_array_get(l, endian, blksft, le->le_name_chunk,
1006 								   le->le_name_length, buf)) {
1007 				free(buf);
1008 				continue;
1009 			}
1010 			buf[le->le_name_length] = 0;
1011 
1012 			if (le->le_int_size != 8
1013 				|| zfs_to_cpu16(le->le_value_length, endian) != 1)
1014 				continue;
1015 
1016 			/* get the uint64_t property value */
1017 			la = &ZAP_LEAF_CHUNK(l, blksft, le->le_value_chunk).l_array;
1018 			val = be64_to_cpu(la->la_array64);
1019 			if (hook(buf, val, data))
1020 				return 1;
1021 			free(buf);
1022 		}
1023 	}
1024 	return 0;
1025 }
1026 
1027 
1028 /*
1029  * Read in the data of a zap object and find the value for a matching
1030  * property name.
1031  *
1032  */
1033 static int
1034 zap_lookup(dnode_end_t *zap_dnode, char *name, uint64_t *val,
1035 		   struct zfs_data *data)
1036 {
1037 	uint64_t block_type;
1038 	int size;
1039 	void *zapbuf;
1040 	int err;
1041 	zfs_endian_t endian;
1042 
1043 	/* Read in the first block of the zap object data. */
1044 	size = zfs_to_cpu16(zap_dnode->dn.dn_datablkszsec,
1045 							 zap_dnode->endian) << SPA_MINBLOCKSHIFT;
1046 	err = dmu_read(zap_dnode, 0, &zapbuf, &endian, data);
1047 	if (err)
1048 		return err;
1049 	block_type = zfs_to_cpu64(*((uint64_t *) zapbuf), endian);
1050 
1051 	if (block_type == ZBT_MICRO) {
1052 		err = (mzap_lookup(zapbuf, endian, size, name, val));
1053 		free(zapbuf);
1054 		return err;
1055 	} else if (block_type == ZBT_HEADER) {
1056 		/* this is a fat zap */
1057 		err = (fzap_lookup(zap_dnode, zapbuf, name, val, data));
1058 		free(zapbuf);
1059 		return err;
1060 	}
1061 
1062 	printf("unknown ZAP type\n");
1063 	return ZFS_ERR_BAD_FS;
1064 }
1065 
1066 static int
1067 zap_iterate(dnode_end_t *zap_dnode,
1068 			int (*hook)(const char *name, uint64_t val,
1069 						struct zfs_data *data),
1070 			struct zfs_data *data)
1071 {
1072 	uint64_t block_type;
1073 	int size;
1074 	void *zapbuf;
1075 	int err;
1076 	int ret;
1077 	zfs_endian_t endian;
1078 
1079 	/* Read in the first block of the zap object data. */
1080 	size = zfs_to_cpu16(zap_dnode->dn.dn_datablkszsec, zap_dnode->endian) << SPA_MINBLOCKSHIFT;
1081 	err = dmu_read(zap_dnode, 0, &zapbuf, &endian, data);
1082 	if (err)
1083 		return 0;
1084 	block_type = zfs_to_cpu64(*((uint64_t *) zapbuf), endian);
1085 
1086 	if (block_type == ZBT_MICRO) {
1087 		ret = mzap_iterate(zapbuf, endian, size, hook, data);
1088 		free(zapbuf);
1089 		return ret;
1090 	} else if (block_type == ZBT_HEADER) {
1091 		/* this is a fat zap */
1092 		ret = fzap_iterate(zap_dnode, zapbuf, hook, data);
1093 		free(zapbuf);
1094 		return ret;
1095 	}
1096 	printf("unknown ZAP type\n");
1097 	return 0;
1098 }
1099 
1100 
1101 /*
1102  * Get the dnode of an object number from the metadnode of an object set.
1103  *
1104  * Input
1105  *	mdn - metadnode to get the object dnode
1106  *	objnum - object number for the object dnode
1107  *	buf - data buffer that holds the returning dnode
1108  */
1109 static int
1110 dnode_get(dnode_end_t *mdn, uint64_t objnum, uint8_t type,
1111 		  dnode_end_t *buf, struct zfs_data *data)
1112 {
1113 	uint64_t blkid, blksz;	/* the block id this object dnode is in */
1114 	int epbs;			/* shift of number of dnodes in a block */
1115 	int idx;			/* index within a block */
1116 	void *dnbuf;
1117 	int err;
1118 	zfs_endian_t endian;
1119 
1120 	blksz = zfs_to_cpu16(mdn->dn.dn_datablkszsec,
1121 							  mdn->endian) << SPA_MINBLOCKSHIFT;
1122 
1123 	epbs = zfs_log2(blksz) - DNODE_SHIFT;
1124 	blkid = objnum >> epbs;
1125 	idx = objnum & ((1 << epbs) - 1);
1126 
1127 	if (data->dnode_buf != NULL && memcmp(data->dnode_mdn, mdn,
1128 										  sizeof(*mdn)) == 0
1129 		&& objnum >= data->dnode_start && objnum < data->dnode_end) {
1130 		memmove(&(buf->dn), &(data->dnode_buf)[idx], DNODE_SIZE);
1131 		buf->endian = data->dnode_endian;
1132 		if (type && buf->dn.dn_type != type)  {
1133 			printf("incorrect dnode type: %02X != %02x\n", buf->dn.dn_type, type);
1134 			return ZFS_ERR_BAD_FS;
1135 		}
1136 		return ZFS_ERR_NONE;
1137 	}
1138 
1139 	err = dmu_read(mdn, blkid, &dnbuf, &endian, data);
1140 	if (err)
1141 		return err;
1142 
1143 	free(data->dnode_buf);
1144 	free(data->dnode_mdn);
1145 	data->dnode_mdn = malloc(sizeof(*mdn));
1146 	if (!data->dnode_mdn) {
1147 		data->dnode_buf = 0;
1148 	} else {
1149 		memcpy(data->dnode_mdn, mdn, sizeof(*mdn));
1150 		data->dnode_buf = dnbuf;
1151 		data->dnode_start = blkid << epbs;
1152 		data->dnode_end = (blkid + 1) << epbs;
1153 		data->dnode_endian = endian;
1154 	}
1155 
1156 	memmove(&(buf->dn), (dnode_phys_t *) dnbuf + idx, DNODE_SIZE);
1157 	buf->endian = endian;
1158 	if (type && buf->dn.dn_type != type) {
1159 		printf("incorrect dnode type\n");
1160 		return ZFS_ERR_BAD_FS;
1161 	}
1162 
1163 	return ZFS_ERR_NONE;
1164 }
1165 
1166 /*
1167  * Get the file dnode for a given file name where mdn is the meta dnode
1168  * for this ZFS object set. When found, place the file dnode in dn.
1169  * The 'path' argument will be mangled.
1170  *
1171  */
1172 static int
1173 dnode_get_path(dnode_end_t *mdn, const char *path_in, dnode_end_t *dn,
1174 			   struct zfs_data *data)
1175 {
1176 	uint64_t objnum, version;
1177 	char *cname, ch;
1178 	int err = ZFS_ERR_NONE;
1179 	char *path, *path_buf;
1180 	struct dnode_chain {
1181 		struct dnode_chain *next;
1182 		dnode_end_t dn;
1183 	};
1184 	struct dnode_chain *dnode_path = 0, *dn_new, *root;
1185 
1186 	dn_new = malloc(sizeof(*dn_new));
1187 	if (!dn_new)
1188 		return ZFS_ERR_OUT_OF_MEMORY;
1189 	dn_new->next = 0;
1190 	dnode_path = root = dn_new;
1191 
1192 	err = dnode_get(mdn, MASTER_NODE_OBJ, DMU_OT_MASTER_NODE,
1193 					&(dnode_path->dn), data);
1194 	if (err) {
1195 		free(dn_new);
1196 		return err;
1197 	}
1198 
1199 	err = zap_lookup(&(dnode_path->dn), ZPL_VERSION_STR, &version, data);
1200 	if (err) {
1201 		free(dn_new);
1202 		return err;
1203 	}
1204 	if (version > ZPL_VERSION) {
1205 		free(dn_new);
1206 		printf("too new ZPL version\n");
1207 		return ZFS_ERR_NOT_IMPLEMENTED_YET;
1208 	}
1209 
1210 	err = zap_lookup(&(dnode_path->dn), ZFS_ROOT_OBJ, &objnum, data);
1211 	if (err) {
1212 		free(dn_new);
1213 		return err;
1214 	}
1215 
1216 	err = dnode_get(mdn, objnum, 0, &(dnode_path->dn), data);
1217 	if (err) {
1218 		free(dn_new);
1219 		return err;
1220 	}
1221 
1222 	path = path_buf = strdup(path_in);
1223 	if (!path_buf) {
1224 		free(dn_new);
1225 		return ZFS_ERR_OUT_OF_MEMORY;
1226 	}
1227 
1228 	while (1) {
1229 		/* skip leading slashes */
1230 		while (*path == '/')
1231 			path++;
1232 		if (!*path)
1233 			break;
1234 		/* get the next component name */
1235 		cname = path;
1236 		while (*path && *path != '/')
1237 			path++;
1238 		/* Skip dot.  */
1239 		if (cname + 1 == path && cname[0] == '.')
1240 			continue;
1241 		/* Handle double dot.  */
1242 		if (cname + 2 == path && cname[0] == '.' && cname[1] == '.')  {
1243 			if (dn_new->next) {
1244 				dn_new = dnode_path;
1245 				dnode_path = dn_new->next;
1246 				free(dn_new);
1247 			} else {
1248 				printf("can't resolve ..\n");
1249 				err = ZFS_ERR_FILE_NOT_FOUND;
1250 				break;
1251 			}
1252 			continue;
1253 		}
1254 
1255 		ch = *path;
1256 		*path = 0;		/* ensure null termination */
1257 
1258 		if (dnode_path->dn.dn.dn_type != DMU_OT_DIRECTORY_CONTENTS) {
1259 			free(path_buf);
1260 			printf("not a directory\n");
1261 			return ZFS_ERR_BAD_FILE_TYPE;
1262 		}
1263 		err = zap_lookup(&(dnode_path->dn), cname, &objnum, data);
1264 		if (err)
1265 			break;
1266 
1267 		dn_new = malloc(sizeof(*dn_new));
1268 		if (!dn_new) {
1269 			err = ZFS_ERR_OUT_OF_MEMORY;
1270 			break;
1271 		}
1272 		dn_new->next = dnode_path;
1273 		dnode_path = dn_new;
1274 
1275 		objnum = ZFS_DIRENT_OBJ(objnum);
1276 		err = dnode_get(mdn, objnum, 0, &(dnode_path->dn), data);
1277 		if (err)
1278 			break;
1279 
1280 		*path = ch;
1281 	}
1282 
1283 	if (!err)
1284 		memcpy(dn, &(dnode_path->dn), sizeof(*dn));
1285 
1286 	while (dnode_path) {
1287 		dn_new = dnode_path->next;
1288 		free(dnode_path);
1289 		dnode_path = dn_new;
1290 	}
1291 	free(path_buf);
1292 	return err;
1293 }
1294 
1295 
1296 /*
1297  * Given a MOS metadnode, get the metadnode of a given filesystem name (fsname),
1298  * e.g. pool/rootfs, or a given object number (obj), e.g. the object number
1299  * of pool/rootfs.
1300  *
1301  * If no fsname and no obj are given, return the DSL_DIR metadnode.
1302  * If fsname is given, return its metadnode and its matching object number.
1303  * If only obj is given, return the metadnode for this object number.
1304  *
1305  */
1306 static int
1307 get_filesystem_dnode(dnode_end_t *mosmdn, char *fsname,
1308 					 dnode_end_t *mdn, struct zfs_data *data)
1309 {
1310 	uint64_t objnum;
1311 	int err;
1312 
1313 	err = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT,
1314 					DMU_OT_OBJECT_DIRECTORY, mdn, data);
1315 	if (err)
1316 		return err;
1317 
1318 	err = zap_lookup(mdn, DMU_POOL_ROOT_DATASET, &objnum, data);
1319 	if (err)
1320 		return err;
1321 
1322 	err = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, mdn, data);
1323 	if (err)
1324 		return err;
1325 
1326 	while (*fsname) {
1327 		uint64_t childobj;
1328 		char *cname, ch;
1329 
1330 		while (*fsname == '/')
1331 			fsname++;
1332 
1333 		if (!*fsname || *fsname == '@')
1334 			break;
1335 
1336 		cname = fsname;
1337 		while (*fsname && !isspace(*fsname) && *fsname != '/')
1338 			fsname++;
1339 		ch = *fsname;
1340 		*fsname = 0;
1341 
1342 		childobj = zfs_to_cpu64((((dsl_dir_phys_t *) DN_BONUS(&mdn->dn)))->dd_child_dir_zapobj, mdn->endian);
1343 		err = dnode_get(mosmdn, childobj,
1344 						DMU_OT_DSL_DIR_CHILD_MAP, mdn, data);
1345 		if (err)
1346 			return err;
1347 
1348 		err = zap_lookup(mdn, cname, &objnum, data);
1349 		if (err)
1350 			return err;
1351 
1352 		err = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, mdn, data);
1353 		if (err)
1354 			return err;
1355 
1356 		*fsname = ch;
1357 	}
1358 	return ZFS_ERR_NONE;
1359 }
1360 
1361 static int
1362 make_mdn(dnode_end_t *mdn, struct zfs_data *data)
1363 {
1364 	void *osp;
1365 	blkptr_t *bp;
1366 	size_t ospsize;
1367 	int err;
1368 
1369 	bp = &(((dsl_dataset_phys_t *) DN_BONUS(&mdn->dn))->ds_bp);
1370 	err = zio_read(bp, mdn->endian, &osp, &ospsize, data);
1371 	if (err)
1372 		return err;
1373 	if (ospsize < OBJSET_PHYS_SIZE_V14) {
1374 		free(osp);
1375 		printf("too small osp\n");
1376 		return ZFS_ERR_BAD_FS;
1377 	}
1378 
1379 	mdn->endian = (zfs_to_cpu64(bp->blk_prop, mdn->endian)>>63) & 1;
1380 	memmove((char *) &(mdn->dn),
1381 			(char *) &((objset_phys_t *) osp)->os_meta_dnode, DNODE_SIZE);
1382 	free(osp);
1383 	return ZFS_ERR_NONE;
1384 }
1385 
1386 static int
1387 dnode_get_fullpath(const char *fullpath, dnode_end_t *mdn,
1388 				   uint64_t *mdnobj, dnode_end_t *dn, int *isfs,
1389 				   struct zfs_data *data)
1390 {
1391 	char *fsname, *snapname;
1392 	const char *ptr_at, *filename;
1393 	uint64_t headobj;
1394 	int err;
1395 
1396 	ptr_at = strchr(fullpath, '@');
1397 	if (!ptr_at) {
1398 		*isfs = 1;
1399 		filename = 0;
1400 		snapname = 0;
1401 		fsname = strdup(fullpath);
1402 	} else {
1403 		const char *ptr_slash = strchr(ptr_at, '/');
1404 
1405 		*isfs = 0;
1406 		fsname = malloc(ptr_at - fullpath + 1);
1407 		if (!fsname)
1408 			return ZFS_ERR_OUT_OF_MEMORY;
1409 		memcpy(fsname, fullpath, ptr_at - fullpath);
1410 		fsname[ptr_at - fullpath] = 0;
1411 		if (ptr_at[1] && ptr_at[1] != '/') {
1412 			snapname = malloc(ptr_slash - ptr_at);
1413 			if (!snapname) {
1414 				free(fsname);
1415 				return ZFS_ERR_OUT_OF_MEMORY;
1416 			}
1417 			memcpy(snapname, ptr_at + 1, ptr_slash - ptr_at - 1);
1418 			snapname[ptr_slash - ptr_at - 1] = 0;
1419 		} else {
1420 			snapname = 0;
1421 		}
1422 		if (ptr_slash)
1423 			filename = ptr_slash;
1424 		else
1425 			filename = "/";
1426 		printf("zfs fsname = '%s' snapname='%s' filename = '%s'\n",
1427 			   fsname, snapname, filename);
1428 	}
1429 
1430 
1431 	err = get_filesystem_dnode(&(data->mos), fsname, dn, data);
1432 
1433 	if (err) {
1434 		free(fsname);
1435 		free(snapname);
1436 		return err;
1437 	}
1438 
1439 	headobj = zfs_to_cpu64(((dsl_dir_phys_t *) DN_BONUS(&dn->dn))->dd_head_dataset_obj, dn->endian);
1440 
1441 	err = dnode_get(&(data->mos), headobj, DMU_OT_DSL_DATASET, mdn, data);
1442 	if (err) {
1443 		free(fsname);
1444 		free(snapname);
1445 		return err;
1446 	}
1447 
1448 	if (snapname) {
1449 		uint64_t snapobj;
1450 
1451 		snapobj = zfs_to_cpu64(((dsl_dataset_phys_t *) DN_BONUS(&mdn->dn))->ds_snapnames_zapobj, mdn->endian);
1452 
1453 		err = dnode_get(&(data->mos), snapobj,
1454 						DMU_OT_DSL_DS_SNAP_MAP, mdn, data);
1455 		if (!err)
1456 			err = zap_lookup(mdn, snapname, &headobj, data);
1457 		if (!err)
1458 			err = dnode_get(&(data->mos), headobj, DMU_OT_DSL_DATASET, mdn, data);
1459 		if (err) {
1460 			free(fsname);
1461 			free(snapname);
1462 			return err;
1463 		}
1464 	}
1465 
1466 	if (mdnobj)
1467 		*mdnobj = headobj;
1468 
1469 	make_mdn(mdn, data);
1470 
1471 	if (*isfs) {
1472 		free(fsname);
1473 		free(snapname);
1474 		return ZFS_ERR_NONE;
1475 	}
1476 	err = dnode_get_path(mdn, filename, dn, data);
1477 	free(fsname);
1478 	free(snapname);
1479 	return err;
1480 }
1481 
1482 /*
1483  * For a given XDR packed nvlist, verify the first 4 bytes and move on.
1484  *
1485  * An XDR packed nvlist is encoded as (comments from nvs_xdr_create) :
1486  *
1487  *		encoding method/host endian		(4 bytes)
1488  *		nvl_version						(4 bytes)
1489  *		nvl_nvflag						(4 bytes)
1490  *	encoded nvpairs:
1491  *		encoded size of the nvpair		(4 bytes)
1492  *		decoded size of the nvpair		(4 bytes)
1493  *		name string size				(4 bytes)
1494  *		name string data				(sizeof(NV_ALIGN4(string))
1495  *		data type						(4 bytes)
1496  *		# of elements in the nvpair		(4 bytes)
1497  *		data
1498  *		2 zero's for the last nvpair
1499  *		(end of the entire list)	(8 bytes)
1500  *
1501  */
1502 
1503 static int
1504 nvlist_find_value(char *nvlist, char *name, int valtype, char **val,
1505 				  size_t *size_out, size_t *nelm_out)
1506 {
1507 	int name_len, type, encode_size;
1508 	char *nvpair, *nvp_name;
1509 
1510 	/* Verify if the 1st and 2nd byte in the nvlist are valid. */
1511 	/* NOTE: independently of what endianness header announces all
1512 	   subsequent values are big-endian.  */
1513 	if (nvlist[0] != NV_ENCODE_XDR || (nvlist[1] != NV_LITTLE_ENDIAN
1514 									   && nvlist[1] != NV_BIG_ENDIAN)) {
1515 		printf("zfs incorrect nvlist header\n");
1516 		return ZFS_ERR_BAD_FS;
1517 	}
1518 
1519 	/* skip the header, nvl_version, and nvl_nvflag */
1520 	nvlist = nvlist + 4 * 3;
1521 	/*
1522 	 * Loop thru the nvpair list
1523 	 * The XDR representation of an integer is in big-endian byte order.
1524 	 */
1525 	while ((encode_size = be32_to_cpu(*(uint32_t *) nvlist))) {
1526 		int nelm;
1527 
1528 		nvpair = nvlist + 4 * 2;	/* skip the encode/decode size */
1529 
1530 		name_len = be32_to_cpu(*(uint32_t *) nvpair);
1531 		nvpair += 4;
1532 
1533 		nvp_name = nvpair;
1534 		nvpair = nvpair + ((name_len + 3) & ~3);	/* align */
1535 
1536 		type = be32_to_cpu(*(uint32_t *) nvpair);
1537 		nvpair += 4;
1538 
1539 		nelm = be32_to_cpu(*(uint32_t *) nvpair);
1540 		if (nelm < 1) {
1541 			printf("empty nvpair\n");
1542 			return ZFS_ERR_BAD_FS;
1543 		}
1544 
1545 		nvpair += 4;
1546 
1547 		if ((strncmp(nvp_name, name, name_len) == 0) && type == valtype) {
1548 			*val = nvpair;
1549 			*size_out = encode_size;
1550 			if (nelm_out)
1551 				*nelm_out = nelm;
1552 			return 1;
1553 		}
1554 
1555 		nvlist += encode_size;	/* goto the next nvpair */
1556 	}
1557 	return 0;
1558 }
1559 
1560 int
1561 zfs_nvlist_lookup_uint64(char *nvlist, char *name, uint64_t *out)
1562 {
1563 	char *nvpair;
1564 	size_t size;
1565 	int found;
1566 
1567 	found = nvlist_find_value(nvlist, name, DATA_TYPE_UINT64, &nvpair, &size, 0);
1568 	if (!found)
1569 		return 0;
1570 	if (size < sizeof(uint64_t)) {
1571 		printf("invalid uint64\n");
1572 		return ZFS_ERR_BAD_FS;
1573 	}
1574 
1575 	*out = be64_to_cpu(*(uint64_t *) nvpair);
1576 	return 1;
1577 }
1578 
1579 char *
1580 zfs_nvlist_lookup_string(char *nvlist, char *name)
1581 {
1582 	char *nvpair;
1583 	char *ret;
1584 	size_t slen;
1585 	size_t size;
1586 	int found;
1587 
1588 	found = nvlist_find_value(nvlist, name, DATA_TYPE_STRING, &nvpair, &size, 0);
1589 	if (!found)
1590 		return 0;
1591 	if (size < 4) {
1592 		printf("invalid string\n");
1593 		return 0;
1594 	}
1595 	slen = be32_to_cpu(*(uint32_t *) nvpair);
1596 	if (slen > size - 4)
1597 		slen = size - 4;
1598 	ret = malloc(slen + 1);
1599 	if (!ret)
1600 		return 0;
1601 	memcpy(ret, nvpair + 4, slen);
1602 	ret[slen] = 0;
1603 	return ret;
1604 }
1605 
1606 char *
1607 zfs_nvlist_lookup_nvlist(char *nvlist, char *name)
1608 {
1609 	char *nvpair;
1610 	char *ret;
1611 	size_t size;
1612 	int found;
1613 
1614 	found = nvlist_find_value(nvlist, name, DATA_TYPE_NVLIST, &nvpair,
1615 							  &size, 0);
1616 	if (!found)
1617 		return 0;
1618 	ret = calloc(1, size + 3 * sizeof(uint32_t));
1619 	if (!ret)
1620 		return 0;
1621 	memcpy(ret, nvlist, sizeof(uint32_t));
1622 
1623 	memcpy(ret + sizeof(uint32_t), nvpair, size);
1624 	return ret;
1625 }
1626 
1627 int
1628 zfs_nvlist_lookup_nvlist_array_get_nelm(char *nvlist, char *name)
1629 {
1630 	char *nvpair;
1631 	size_t nelm, size;
1632 	int found;
1633 
1634 	found = nvlist_find_value(nvlist, name, DATA_TYPE_NVLIST, &nvpair,
1635 							  &size, &nelm);
1636 	if (!found)
1637 		return -1;
1638 	return nelm;
1639 }
1640 
1641 char *
1642 zfs_nvlist_lookup_nvlist_array(char *nvlist, char *name,
1643 									size_t index)
1644 {
1645 	char *nvpair, *nvpairptr;
1646 	int found;
1647 	char *ret;
1648 	size_t size;
1649 	unsigned i;
1650 	size_t nelm;
1651 
1652 	found = nvlist_find_value(nvlist, name, DATA_TYPE_NVLIST, &nvpair,
1653 							  &size, &nelm);
1654 	if (!found)
1655 		return 0;
1656 	if (index >= nelm) {
1657 		printf("trying to lookup past nvlist array\n");
1658 		return 0;
1659 	}
1660 
1661 	nvpairptr = nvpair;
1662 
1663 	for (i = 0; i < index; i++) {
1664 		uint32_t encode_size;
1665 
1666 		/* skip the header, nvl_version, and nvl_nvflag */
1667 		nvpairptr = nvpairptr + 4 * 2;
1668 
1669 		while (nvpairptr < nvpair + size
1670 			   && (encode_size = be32_to_cpu(*(uint32_t *) nvpairptr)))
1671 			nvlist += encode_size;	/* goto the next nvpair */
1672 
1673 		nvlist = nvlist + 4 * 2;	/* skip the ending 2 zeros - 8 bytes */
1674 	}
1675 
1676 	if (nvpairptr >= nvpair + size
1677 		|| nvpairptr + be32_to_cpu(*(uint32_t *) (nvpairptr + 4 * 2))
1678 		>= nvpair + size) {
1679 		printf("incorrect nvlist array\n");
1680 		return 0;
1681 	}
1682 
1683 	ret = calloc(1, be32_to_cpu(*(uint32_t *) (nvpairptr + 4 * 2))
1684 				 + 3 * sizeof(uint32_t));
1685 	if (!ret)
1686 		return 0;
1687 	memcpy(ret, nvlist, sizeof(uint32_t));
1688 
1689 	memcpy(ret + sizeof(uint32_t), nvpairptr, size);
1690 	return ret;
1691 }
1692 
1693 static int
1694 int_zfs_fetch_nvlist(struct zfs_data *data, char **nvlist)
1695 {
1696 	int err;
1697 
1698 	*nvlist = malloc(VDEV_PHYS_SIZE);
1699 	/* Read in the vdev name-value pair list (112K). */
1700 	err = zfs_devread(data->vdev_phys_sector, 0, VDEV_PHYS_SIZE, *nvlist);
1701 	if (err) {
1702 		free(*nvlist);
1703 		*nvlist = 0;
1704 		return err;
1705 	}
1706 	return ZFS_ERR_NONE;
1707 }
1708 
1709 /*
1710  * Check the disk label information and retrieve needed vdev name-value pairs.
1711  *
1712  */
1713 static int
1714 check_pool_label(struct zfs_data *data)
1715 {
1716 	uint64_t pool_state;
1717 	char *nvlist;			/* for the pool */
1718 	char *vdevnvlist;		/* for the vdev */
1719 	uint64_t diskguid;
1720 	uint64_t version;
1721 	int found;
1722 	int err;
1723 
1724 	err = int_zfs_fetch_nvlist(data, &nvlist);
1725 	if (err)
1726 		return err;
1727 
1728 	found = zfs_nvlist_lookup_uint64(nvlist, ZPOOL_CONFIG_POOL_STATE,
1729 										  &pool_state);
1730 	if (!found) {
1731 		free(nvlist);
1732 		printf("zfs pool state not found\n");
1733 		return ZFS_ERR_BAD_FS;
1734 	}
1735 
1736 	if (pool_state == POOL_STATE_DESTROYED) {
1737 		free(nvlist);
1738 		printf("zpool is marked as destroyed\n");
1739 		return ZFS_ERR_BAD_FS;
1740 	}
1741 
1742 	data->label_txg = 0;
1743 	found = zfs_nvlist_lookup_uint64(nvlist, ZPOOL_CONFIG_POOL_TXG,
1744 										  &data->label_txg);
1745 	if (!found) {
1746 		free(nvlist);
1747 		printf("zfs pool txg not found\n");
1748 		return ZFS_ERR_BAD_FS;
1749 	}
1750 
1751 	/* not an active device */
1752 	if (data->label_txg == 0) {
1753 		free(nvlist);
1754 		printf("zpool is not active\n");
1755 		return ZFS_ERR_BAD_FS;
1756 	}
1757 
1758 	found = zfs_nvlist_lookup_uint64(nvlist, ZPOOL_CONFIG_VERSION,
1759 										  &version);
1760 	if (!found) {
1761 		free(nvlist);
1762 		printf("zpool config version not found\n");
1763 		return ZFS_ERR_BAD_FS;
1764 	}
1765 
1766 	if (version > SPA_VERSION) {
1767 		free(nvlist);
1768 		printf("SPA version too new %llu > %llu\n",
1769 			   (unsigned long long) version,
1770 			   (unsigned long long) SPA_VERSION);
1771 		return ZFS_ERR_NOT_IMPLEMENTED_YET;
1772 	}
1773 
1774 	vdevnvlist = zfs_nvlist_lookup_nvlist(nvlist, ZPOOL_CONFIG_VDEV_TREE);
1775 	if (!vdevnvlist) {
1776 		free(nvlist);
1777 		printf("ZFS config vdev tree not found\n");
1778 		return ZFS_ERR_BAD_FS;
1779 	}
1780 
1781 	found = zfs_nvlist_lookup_uint64(vdevnvlist, ZPOOL_CONFIG_ASHIFT,
1782 										  &data->vdev_ashift);
1783 	free(vdevnvlist);
1784 	if (!found) {
1785 		free(nvlist);
1786 		printf("ZPOOL config ashift not found\n");
1787 		return ZFS_ERR_BAD_FS;
1788 	}
1789 
1790 	found = zfs_nvlist_lookup_uint64(nvlist, ZPOOL_CONFIG_GUID, &diskguid);
1791 	if (!found) {
1792 		free(nvlist);
1793 		printf("ZPOOL config guid not found\n");
1794 		return ZFS_ERR_BAD_FS;
1795 	}
1796 
1797 	found = zfs_nvlist_lookup_uint64(nvlist, ZPOOL_CONFIG_POOL_GUID, &data->pool_guid);
1798 	if (!found) {
1799 		free(nvlist);
1800 		printf("ZPOOL config pool guid not found\n");
1801 		return ZFS_ERR_BAD_FS;
1802 	}
1803 
1804 	free(nvlist);
1805 
1806 	printf("ZFS Pool GUID: %llu (%016llx) Label: GUID: %llu (%016llx), txg: %llu, SPA v%llu, ashift: %llu\n",
1807 		   (unsigned long long) data->pool_guid,
1808 		   (unsigned long long) data->pool_guid,
1809 		   (unsigned long long) diskguid,
1810 		   (unsigned long long) diskguid,
1811 		   (unsigned long long) data->label_txg,
1812 		   (unsigned long long) version,
1813 		   (unsigned long long) data->vdev_ashift);
1814 
1815 	return ZFS_ERR_NONE;
1816 }
1817 
1818 /*
1819  * vdev_label_start returns the physical disk offset (in bytes) of
1820  * label "l".
1821  */
1822 static uint64_t vdev_label_start(uint64_t psize, int l)
1823 {
1824 	return (l * sizeof(vdev_label_t) + (l < VDEV_LABELS / 2 ?
1825 										0 : psize -
1826 										VDEV_LABELS * sizeof(vdev_label_t)));
1827 }
1828 
1829 void
1830 zfs_unmount(struct zfs_data *data)
1831 {
1832 	free(data->dnode_buf);
1833 	free(data->dnode_mdn);
1834 	free(data->file_buf);
1835 	free(data);
1836 }
1837 
1838 /*
1839  * zfs_mount() locates a valid uberblock of the root pool and read in its MOS
1840  * to the memory address MOS.
1841  *
1842  */
1843 struct zfs_data *
1844 zfs_mount(device_t dev)
1845 {
1846 	struct zfs_data *data = 0;
1847 	int label = 0, bestlabel = -1;
1848 	char *ub_array;
1849 	uberblock_t *ubbest;
1850 	uberblock_t *ubcur = NULL;
1851 	void *osp = 0;
1852 	size_t ospsize;
1853 	int err;
1854 
1855 	data = malloc(sizeof(*data));
1856 	if (!data)
1857 		return 0;
1858 	memset(data, 0, sizeof(*data));
1859 
1860 	ub_array = malloc(VDEV_UBERBLOCK_RING);
1861 	if (!ub_array) {
1862 		zfs_unmount(data);
1863 		return 0;
1864 	}
1865 
1866 	ubbest = malloc(sizeof(*ubbest));
1867 	if (!ubbest) {
1868 		zfs_unmount(data);
1869 		return 0;
1870 	}
1871 	memset(ubbest, 0, sizeof(*ubbest));
1872 
1873 	/*
1874 	 * some eltorito stacks don't give us a size and
1875 	 * we end up setting the size to MAXUINT, further
1876 	 * some of these devices stop working once a single
1877 	 * read past the end has been issued. Checking
1878 	 * for a maximum part_length and skipping the backup
1879 	 * labels at the end of the slice/partition/device
1880 	 * avoids breaking down on such devices.
1881 	 */
1882 	const int vdevnum =
1883 		dev->part_length == 0 ?
1884 		VDEV_LABELS / 2 : VDEV_LABELS;
1885 
1886 	/* Size in bytes of the device (disk or partition) aligned to label size*/
1887 	uint64_t device_size =
1888 		dev->part_length << SECTOR_BITS;
1889 
1890 	const uint64_t alignedbytes =
1891 		P2ALIGN(device_size, (uint64_t) sizeof(vdev_label_t));
1892 
1893 	for (label = 0; label < vdevnum; label++) {
1894 		uint64_t labelstartbytes = vdev_label_start(alignedbytes, label);
1895 		uint64_t labelstart = labelstartbytes >> SECTOR_BITS;
1896 
1897 		debug("zfs reading label %d at sector %llu (byte %llu)\n",
1898 			  label, (unsigned long long) labelstart,
1899 			  (unsigned long long) labelstartbytes);
1900 
1901 		data->vdev_phys_sector = labelstart +
1902 			((VDEV_SKIP_SIZE + VDEV_BOOT_HEADER_SIZE) >> SECTOR_BITS);
1903 
1904 		err = check_pool_label(data);
1905 		if (err) {
1906 			printf("zfs error checking label %d\n", label);
1907 			continue;
1908 		}
1909 
1910 		/* Read in the uberblock ring (128K). */
1911 		err = zfs_devread(data->vdev_phys_sector  +
1912 						  (VDEV_PHYS_SIZE >> SECTOR_BITS),
1913 						  0, VDEV_UBERBLOCK_RING, ub_array);
1914 		if (err) {
1915 			printf("zfs error reading uberblock ring for label %d\n", label);
1916 			continue;
1917 		}
1918 
1919 		ubcur = find_bestub(ub_array, data);
1920 		if (!ubcur) {
1921 			printf("zfs No good uberblocks found in label %d\n", label);
1922 			continue;
1923 		}
1924 
1925 		if (vdev_uberblock_compare(ubcur, ubbest) > 0) {
1926 			/* Looks like the block is good, so use it.*/
1927 			memcpy(ubbest, ubcur, sizeof(*ubbest));
1928 			bestlabel = label;
1929 			debug("zfs Current best uberblock found in label %d\n", label);
1930 		}
1931 	}
1932 	free(ub_array);
1933 
1934 	/* We zero'd the structure to begin with.  If we never assigned to it,
1935 	   magic will still be zero. */
1936 	if (!ubbest->ub_magic) {
1937 		printf("couldn't find a valid ZFS label\n");
1938 		zfs_unmount(data);
1939 		free(ubbest);
1940 		return 0;
1941 	}
1942 
1943 	debug("zfs ubbest %p in label %d\n", ubbest, bestlabel);
1944 
1945 	zfs_endian_t ub_endian =
1946 		zfs_to_cpu64(ubbest->ub_magic, LITTLE_ENDIAN) == UBERBLOCK_MAGIC
1947 		? LITTLE_ENDIAN : BIG_ENDIAN;
1948 
1949 	debug("zfs endian set to %s\n", !ub_endian ? "big" : "little");
1950 
1951 	err = zio_read(&ubbest->ub_rootbp, ub_endian, &osp, &ospsize, data);
1952 
1953 	if (err) {
1954 		printf("couldn't zio_read object directory\n");
1955 		zfs_unmount(data);
1956 		free(ubbest);
1957 		return 0;
1958 	}
1959 
1960 	if (ospsize < OBJSET_PHYS_SIZE_V14) {
1961 		printf("osp too small\n");
1962 		zfs_unmount(data);
1963 		free(osp);
1964 		free(ubbest);
1965 		return 0;
1966 	}
1967 
1968 	/* Got the MOS. Save it at the memory addr MOS. */
1969 	memmove(&(data->mos.dn), &((objset_phys_t *) osp)->os_meta_dnode, DNODE_SIZE);
1970 	data->mos.endian =
1971 		(zfs_to_cpu64(ubbest->ub_rootbp.blk_prop, ub_endian) >> 63) & 1;
1972 	memmove(&(data->current_uberblock), ubbest, sizeof(uberblock_t));
1973 
1974 	free(osp);
1975 	free(ubbest);
1976 
1977 	return data;
1978 }
1979 
1980 int
1981 zfs_fetch_nvlist(device_t dev, char **nvlist)
1982 {
1983 	struct zfs_data *zfs;
1984 	int err;
1985 
1986 	zfs = zfs_mount(dev);
1987 	if (!zfs)
1988 		return ZFS_ERR_BAD_FS;
1989 	err = int_zfs_fetch_nvlist(zfs, nvlist);
1990 	zfs_unmount(zfs);
1991 	return err;
1992 }
1993 
1994 /*
1995  * zfs_open() locates a file in the rootpool by following the
1996  * MOS and places the dnode of the file in the memory address DNODE.
1997  */
1998 int
1999 zfs_open(struct zfs_file *file, const char *fsfilename)
2000 {
2001 	struct zfs_data *data;
2002 	int err;
2003 	int isfs;
2004 
2005 	data = zfs_mount(file->device);
2006 	if (!data)
2007 		return ZFS_ERR_BAD_FS;
2008 
2009 	err = dnode_get_fullpath(fsfilename, &(data->mdn), 0,
2010 							 &(data->dnode), &isfs, data);
2011 	if (err) {
2012 		zfs_unmount(data);
2013 		return err;
2014 	}
2015 
2016 	if (isfs) {
2017 		zfs_unmount(data);
2018 		printf("Missing @ or / separator\n");
2019 		return ZFS_ERR_FILE_NOT_FOUND;
2020 	}
2021 
2022 	/* We found the dnode for this file. Verify if it is a plain file. */
2023 	if (data->dnode.dn.dn_type != DMU_OT_PLAIN_FILE_CONTENTS) {
2024 		zfs_unmount(data);
2025 		printf("not a file\n");
2026 		return ZFS_ERR_BAD_FILE_TYPE;
2027 	}
2028 
2029 	/* get the file size and set the file position to 0 */
2030 
2031 	/*
2032 	 * For DMU_OT_SA we will need to locate the SIZE attribute
2033 	 * attribute, which could be either in the bonus buffer
2034 	 * or the "spill" block.
2035 	 */
2036 	if (data->dnode.dn.dn_bonustype == DMU_OT_SA) {
2037 		void *sahdrp;
2038 		int hdrsize;
2039 
2040 		if (data->dnode.dn.dn_bonuslen != 0) {
2041 			sahdrp = (sa_hdr_phys_t *) DN_BONUS(&data->dnode.dn);
2042 		} else if (data->dnode.dn.dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
2043 			blkptr_t *bp = &data->dnode.dn.dn_spill;
2044 
2045 			err = zio_read(bp, data->dnode.endian, &sahdrp, NULL, data);
2046 			if (err)
2047 				return err;
2048 		} else {
2049 			printf("filesystem is corrupt :(\n");
2050 			return ZFS_ERR_BAD_FS;
2051 		}
2052 
2053 		hdrsize = SA_HDR_SIZE(((sa_hdr_phys_t *) sahdrp));
2054 		file->size = *(uint64_t *) ((char *) sahdrp + hdrsize + SA_SIZE_OFFSET);
2055 	} else {
2056 		file->size = zfs_to_cpu64(((znode_phys_t *) DN_BONUS(&data->dnode.dn))->zp_size, data->dnode.endian);
2057 	}
2058 
2059 	file->data = data;
2060 	file->offset = 0;
2061 
2062 	return ZFS_ERR_NONE;
2063 }
2064 
2065 uint64_t
2066 zfs_read(zfs_file_t file, char *buf, uint64_t len)
2067 {
2068 	struct zfs_data *data = (struct zfs_data *) file->data;
2069 	int blksz, movesize;
2070 	uint64_t length;
2071 	int64_t red;
2072 	int err;
2073 
2074 	if (data->file_buf == NULL) {
2075 		data->file_buf = malloc(SPA_MAXBLOCKSIZE);
2076 		if (!data->file_buf)
2077 			return -1;
2078 		data->file_start = data->file_end = 0;
2079 	}
2080 
2081 	/*
2082 	 * If offset is in memory, move it into the buffer provided and return.
2083 	 */
2084 	if (file->offset >= data->file_start
2085 		&& file->offset + len <= data->file_end) {
2086 		memmove(buf, data->file_buf + file->offset - data->file_start,
2087 				len);
2088 		return len;
2089 	}
2090 
2091 	blksz = zfs_to_cpu16(data->dnode.dn.dn_datablkszsec,
2092 							  data->dnode.endian) << SPA_MINBLOCKSHIFT;
2093 
2094 	/*
2095 	 * Entire Dnode is too big to fit into the space available.	 We
2096 	 * will need to read it in chunks.	This could be optimized to
2097 	 * read in as large a chunk as there is space available, but for
2098 	 * now, this only reads in one data block at a time.
2099 	 */
2100 	length = len;
2101 	red = 0;
2102 	while (length) {
2103 		void *t;
2104 		/*
2105 		 * Find requested blkid and the offset within that block.
2106 		 */
2107 		uint64_t blkid = file->offset + red;
2108 		blkid = do_div(blkid, blksz);
2109 		free(data->file_buf);
2110 		data->file_buf = 0;
2111 
2112 		err = dmu_read(&(data->dnode), blkid, &t,
2113 					   0, data);
2114 		data->file_buf = t;
2115 		if (err)
2116 			return -1;
2117 
2118 		data->file_start = blkid * blksz;
2119 		data->file_end = data->file_start + blksz;
2120 
2121 		movesize = MIN(length, data->file_end - (int) file->offset - red);
2122 
2123 		memmove(buf, data->file_buf + file->offset + red
2124 				- data->file_start, movesize);
2125 		buf += movesize;
2126 		length -= movesize;
2127 		red += movesize;
2128 	}
2129 
2130 	return len;
2131 }
2132 
2133 int
2134 zfs_close(zfs_file_t file)
2135 {
2136 	zfs_unmount((struct zfs_data *) file->data);
2137 	return ZFS_ERR_NONE;
2138 }
2139 
2140 int
2141 zfs_getmdnobj(device_t dev, const char *fsfilename,
2142 				   uint64_t *mdnobj)
2143 {
2144 	struct zfs_data *data;
2145 	int err;
2146 	int isfs;
2147 
2148 	data = zfs_mount(dev);
2149 	if (!data)
2150 		return ZFS_ERR_BAD_FS;
2151 
2152 	err = dnode_get_fullpath(fsfilename, &(data->mdn), mdnobj,
2153 							 &(data->dnode), &isfs, data);
2154 	zfs_unmount(data);
2155 	return err;
2156 }
2157 
2158 static void
2159 fill_fs_info(struct zfs_dirhook_info *info,
2160 			 dnode_end_t mdn, struct zfs_data *data)
2161 {
2162 	int err;
2163 	dnode_end_t dn;
2164 	uint64_t objnum;
2165 	uint64_t headobj;
2166 
2167 	memset(info, 0, sizeof(*info));
2168 
2169 	info->dir = 1;
2170 
2171 	if (mdn.dn.dn_type == DMU_OT_DSL_DIR) {
2172 		headobj = zfs_to_cpu64(((dsl_dir_phys_t *) DN_BONUS(&mdn.dn))->dd_head_dataset_obj, mdn.endian);
2173 
2174 		err = dnode_get(&(data->mos), headobj, DMU_OT_DSL_DATASET, &mdn, data);
2175 		if (err) {
2176 			printf("zfs failed here 1\n");
2177 			return;
2178 		}
2179 	}
2180 	make_mdn(&mdn, data);
2181 	err = dnode_get(&mdn, MASTER_NODE_OBJ, DMU_OT_MASTER_NODE,
2182 					&dn, data);
2183 	if (err) {
2184 		printf("zfs failed here 2\n");
2185 		return;
2186 	}
2187 
2188 	err = zap_lookup(&dn, ZFS_ROOT_OBJ, &objnum, data);
2189 	if (err) {
2190 		printf("zfs failed here 3\n");
2191 		return;
2192 	}
2193 
2194 	err = dnode_get(&mdn, objnum, 0, &dn, data);
2195 	if (err) {
2196 		printf("zfs failed here 4\n");
2197 		return;
2198 	}
2199 
2200 	info->mtimeset = 1;
2201 	info->mtime = zfs_to_cpu64(((znode_phys_t *) DN_BONUS(&dn.dn))->zp_mtime[0], dn.endian);
2202 
2203 	return;
2204 }
2205 
2206 static int iterate_zap(const char *name, uint64_t val, struct zfs_data *data)
2207 {
2208 	struct zfs_dirhook_info info;
2209 	dnode_end_t dn;
2210 
2211 	memset(&info, 0, sizeof(info));
2212 
2213 	dnode_get(&(data->mdn), val, 0, &dn, data);
2214 	info.mtimeset = 1;
2215 	info.mtime = zfs_to_cpu64(((znode_phys_t *) DN_BONUS(&dn.dn))->zp_mtime[0], dn.endian);
2216 	info.dir = (dn.dn.dn_type == DMU_OT_DIRECTORY_CONTENTS);
2217 	debug("zfs type=%d, name=%s\n",
2218 		  (int)dn.dn.dn_type, (char *)name);
2219 	if (!data->userhook)
2220 		return 0;
2221 	return data->userhook(name, &info);
2222 }
2223 
2224 static int iterate_zap_fs(const char *name, uint64_t val, struct zfs_data *data)
2225 {
2226 	struct zfs_dirhook_info info;
2227 	dnode_end_t mdn;
2228 	int err;
2229 	err = dnode_get(&(data->mos), val, 0, &mdn, data);
2230 	if (err)
2231 		return 0;
2232 	if (mdn.dn.dn_type != DMU_OT_DSL_DIR)
2233 		return 0;
2234 
2235 	fill_fs_info(&info, mdn, data);
2236 
2237 	if (!data->userhook)
2238 		return 0;
2239 	return data->userhook(name, &info);
2240 }
2241 
2242 static int iterate_zap_snap(const char *name, uint64_t val, struct zfs_data *data)
2243 {
2244 	struct zfs_dirhook_info info;
2245 	char *name2;
2246 	int ret = 0;
2247 	dnode_end_t mdn;
2248 	int err;
2249 
2250 	err = dnode_get(&(data->mos), val, 0, &mdn, data);
2251 	if (err)
2252 		return 0;
2253 
2254 	if (mdn.dn.dn_type != DMU_OT_DSL_DATASET)
2255 		return 0;
2256 
2257 	fill_fs_info(&info, mdn, data);
2258 
2259 	name2 = malloc(strlen(name) + 2);
2260 	name2[0] = '@';
2261 	memcpy(name2 + 1, name, strlen(name) + 1);
2262 	if (data->userhook)
2263 		ret = data->userhook(name2, &info);
2264 	free(name2);
2265 	return ret;
2266 }
2267 
2268 int
2269 zfs_ls(device_t device, const char *path,
2270 	   int (*hook)(const char *, const struct zfs_dirhook_info *))
2271 {
2272 	struct zfs_data *data;
2273 	int err;
2274 	int isfs;
2275 
2276 	data = zfs_mount(device);
2277 	if (!data)
2278 		return ZFS_ERR_BAD_FS;
2279 
2280 	data->userhook = hook;
2281 
2282 	err = dnode_get_fullpath(path, &(data->mdn), 0, &(data->dnode), &isfs, data);
2283 	if (err) {
2284 		zfs_unmount(data);
2285 		return err;
2286 	}
2287 	if (isfs) {
2288 		uint64_t childobj, headobj;
2289 		uint64_t snapobj;
2290 		dnode_end_t dn;
2291 		struct zfs_dirhook_info info;
2292 
2293 		fill_fs_info(&info, data->dnode, data);
2294 		hook("@", &info);
2295 
2296 		childobj = zfs_to_cpu64(((dsl_dir_phys_t *) DN_BONUS(&data->dnode.dn))->dd_child_dir_zapobj, data->dnode.endian);
2297 		headobj = zfs_to_cpu64(((dsl_dir_phys_t *) DN_BONUS(&data->dnode.dn))->dd_head_dataset_obj, data->dnode.endian);
2298 		err = dnode_get(&(data->mos), childobj,
2299 						DMU_OT_DSL_DIR_CHILD_MAP, &dn, data);
2300 		if (err) {
2301 			zfs_unmount(data);
2302 			return err;
2303 		}
2304 
2305 
2306 		zap_iterate(&dn, iterate_zap_fs, data);
2307 
2308 		err = dnode_get(&(data->mos), headobj, DMU_OT_DSL_DATASET, &dn, data);
2309 		if (err) {
2310 			zfs_unmount(data);
2311 			return err;
2312 		}
2313 
2314 		snapobj = zfs_to_cpu64(((dsl_dataset_phys_t *) DN_BONUS(&dn.dn))->ds_snapnames_zapobj, dn.endian);
2315 
2316 		err = dnode_get(&(data->mos), snapobj,
2317 						DMU_OT_DSL_DS_SNAP_MAP, &dn, data);
2318 		if (err) {
2319 			zfs_unmount(data);
2320 			return err;
2321 		}
2322 
2323 		zap_iterate(&dn, iterate_zap_snap, data);
2324 	} else {
2325 		if (data->dnode.dn.dn_type != DMU_OT_DIRECTORY_CONTENTS) {
2326 			zfs_unmount(data);
2327 			printf("not a directory\n");
2328 			return ZFS_ERR_BAD_FILE_TYPE;
2329 		}
2330 		zap_iterate(&(data->dnode), iterate_zap, data);
2331 	}
2332 	zfs_unmount(data);
2333 	return ZFS_ERR_NONE;
2334 }
2335