xref: /openbmc/u-boot/fs/zfs/zfs.c (revision af2a4093)
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  *	GRUB is free software; you can redistribute it and/or modify
12  *	it under the terms of the GNU General Public License as published by
13  *	the Free Software Foundation; either version 2 of the License, or
14  *	(at your option) any later version.
15  *
16  *	GRUB is distributed in the hope that it will be useful,
17  *	but WITHOUT ANY WARRANTY; without even the implied warranty of
18  *	MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19  *	GNU General Public License for more details.
20  *
21  *	You should have received a copy of the GNU General Public License
22  *	along with GRUB.  If not, see <http://www.gnu.org/licenses/>.
23  *
24  */
25 
26 #include <common.h>
27 #include <malloc.h>
28 #include <linux/stat.h>
29 #include <linux/time.h>
30 #include <linux/ctype.h>
31 #include <asm/byteorder.h>
32 #include "zfs_common.h"
33 
34 block_dev_desc_t *zfs_dev_desc;
35 
36 /*
37  * The zfs plug-in routines for GRUB are:
38  *
39  * zfs_mount() - locates a valid uberblock of the root pool and reads
40  *		in its MOS at the memory address MOS.
41  *
42  * zfs_open() - locates a plain file object by following the MOS
43  *		and places its dnode at the memory address DNODE.
44  *
45  * zfs_read() - read in the data blocks pointed by the DNODE.
46  *
47  */
48 
49 #include <zfs/zfs.h>
50 #include <zfs/zio.h>
51 #include <zfs/dnode.h>
52 #include <zfs/uberblock_impl.h>
53 #include <zfs/vdev_impl.h>
54 #include <zfs/zio_checksum.h>
55 #include <zfs/zap_impl.h>
56 #include <zfs/zap_leaf.h>
57 #include <zfs/zfs_znode.h>
58 #include <zfs/dmu.h>
59 #include <zfs/dmu_objset.h>
60 #include <zfs/sa_impl.h>
61 #include <zfs/dsl_dir.h>
62 #include <zfs/dsl_dataset.h>
63 
64 
65 #define	ZPOOL_PROP_BOOTFS		"bootfs"
66 
67 
68 /*
69  * For nvlist manipulation. (from nvpair.h)
70  */
71 #define	NV_ENCODE_NATIVE	0
72 #define	NV_ENCODE_XDR		1
73 #define	NV_BIG_ENDIAN			0
74 #define	NV_LITTLE_ENDIAN	1
75 #define	DATA_TYPE_UINT64	8
76 #define	DATA_TYPE_STRING	9
77 #define	DATA_TYPE_NVLIST	19
78 #define	DATA_TYPE_NVLIST_ARRAY	20
79 
80 
81 /*
82  * Macros to get fields in a bp or DVA.
83  */
84 #define	P2PHASE(x, align)		((x) & ((align) - 1))
85 #define	DVA_OFFSET_TO_PHYS_SECTOR(offset)					\
86 	((offset + VDEV_LABEL_START_SIZE) >> SPA_MINBLOCKSHIFT)
87 
88 /*
89  * return x rounded down to an align boundary
90  * eg, P2ALIGN(1200, 1024) == 1024 (1*align)
91  * eg, P2ALIGN(1024, 1024) == 1024 (1*align)
92  * eg, P2ALIGN(0x1234, 0x100) == 0x1200 (0x12*align)
93  * eg, P2ALIGN(0x5600, 0x100) == 0x5600 (0x56*align)
94  */
95 #define	P2ALIGN(x, align)		((x) & -(align))
96 
97 /*
98  * FAT ZAP data structures
99  */
100 #define	ZFS_CRC64_POLY 0xC96C5795D7870F42ULL	/* ECMA-182, reflected form */
101 #define	ZAP_HASH_IDX(hash, n)	(((n) == 0) ? 0 : ((hash) >> (64 - (n))))
102 #define	CHAIN_END	0xffff	/* end of the chunk chain */
103 
104 /*
105  * The amount of space within the chunk available for the array is:
106  * chunk size - space for type (1) - space for next pointer (2)
107  */
108 #define	ZAP_LEAF_ARRAY_BYTES (ZAP_LEAF_CHUNKSIZE - 3)
109 
110 #define	ZAP_LEAF_HASH_SHIFT(bs)	(bs - 5)
111 #define	ZAP_LEAF_HASH_NUMENTRIES(bs) (1 << ZAP_LEAF_HASH_SHIFT(bs))
112 #define	LEAF_HASH(bs, h)												\
113 	((ZAP_LEAF_HASH_NUMENTRIES(bs)-1) &									\
114 	 ((h) >> (64 - ZAP_LEAF_HASH_SHIFT(bs)-l->l_hdr.lh_prefix_len)))
115 
116 /*
117  * The amount of space available for chunks is:
118  * block size shift - hash entry size (2) * number of hash
119  * entries - header space (2*chunksize)
120  */
121 #define	ZAP_LEAF_NUMCHUNKS(bs)						\
122 	(((1<<bs) - 2*ZAP_LEAF_HASH_NUMENTRIES(bs)) /	\
123 	 ZAP_LEAF_CHUNKSIZE - 2)
124 
125 /*
126  * The chunks start immediately after the hash table.  The end of the
127  * hash table is at l_hash + HASH_NUMENTRIES, which we simply cast to a
128  * chunk_t.
129  */
130 #define	ZAP_LEAF_CHUNK(l, bs, idx)										\
131 	((zap_leaf_chunk_t *)(l->l_hash + ZAP_LEAF_HASH_NUMENTRIES(bs)))[idx]
132 #define	ZAP_LEAF_ENTRY(l, bs, idx) (&ZAP_LEAF_CHUNK(l, bs, idx).l_entry)
133 
134 
135 /*
136  * Decompression Entry - lzjb
137  */
138 #ifndef	NBBY
139 #define	NBBY	8
140 #endif
141 
142 
143 
144 typedef int zfs_decomp_func_t(void *s_start, void *d_start,
145 							  uint32_t s_len, uint32_t d_len);
146 typedef struct decomp_entry {
147 	char *name;
148 	zfs_decomp_func_t *decomp_func;
149 } decomp_entry_t;
150 
151 typedef struct dnode_end {
152 	dnode_phys_t dn;
153 	zfs_endian_t endian;
154 } dnode_end_t;
155 
156 struct zfs_data {
157 	/* cache for a file block of the currently zfs_open()-ed file */
158 	char *file_buf;
159 	uint64_t file_start;
160 	uint64_t file_end;
161 
162 	/* XXX: ashift is per vdev, not per pool.  We currently only ever touch
163 	 * a single vdev, but when/if raid-z or stripes are supported, this
164 	 * may need revision.
165 	 */
166 	uint64_t vdev_ashift;
167 	uint64_t label_txg;
168 	uint64_t pool_guid;
169 
170 	/* cache for a dnode block */
171 	dnode_phys_t *dnode_buf;
172 	dnode_phys_t *dnode_mdn;
173 	uint64_t dnode_start;
174 	uint64_t dnode_end;
175 	zfs_endian_t dnode_endian;
176 
177 	uberblock_t current_uberblock;
178 
179 	dnode_end_t mos;
180 	dnode_end_t mdn;
181 	dnode_end_t dnode;
182 
183 	uint64_t vdev_phys_sector;
184 
185 	int (*userhook)(const char *, const struct zfs_dirhook_info *);
186 	struct zfs_dirhook_info *dirinfo;
187 
188 };
189 
190 
191 
192 
193 static int
194 zlib_decompress(void *s, void *d,
195 				uint32_t slen, uint32_t dlen)
196 {
197 	if (zlib_decompress(s, d, slen, dlen) < 0)
198 		return ZFS_ERR_BAD_FS;
199 	return ZFS_ERR_NONE;
200 }
201 
202 static decomp_entry_t decomp_table[ZIO_COMPRESS_FUNCTIONS] = {
203 	{"inherit", NULL},		/* ZIO_COMPRESS_INHERIT */
204 	{"on", lzjb_decompress},	/* ZIO_COMPRESS_ON */
205 	{"off", NULL},		/* ZIO_COMPRESS_OFF */
206 	{"lzjb", lzjb_decompress},	/* ZIO_COMPRESS_LZJB */
207 	{"empty", NULL},		/* ZIO_COMPRESS_EMPTY */
208 	{"gzip-1", zlib_decompress},  /* ZIO_COMPRESS_GZIP1 */
209 	{"gzip-2", zlib_decompress},  /* ZIO_COMPRESS_GZIP2 */
210 	{"gzip-3", zlib_decompress},  /* ZIO_COMPRESS_GZIP3 */
211 	{"gzip-4", zlib_decompress},  /* ZIO_COMPRESS_GZIP4 */
212 	{"gzip-5", zlib_decompress},  /* ZIO_COMPRESS_GZIP5 */
213 	{"gzip-6", zlib_decompress},  /* ZIO_COMPRESS_GZIP6 */
214 	{"gzip-7", zlib_decompress},  /* ZIO_COMPRESS_GZIP7 */
215 	{"gzip-8", zlib_decompress},  /* ZIO_COMPRESS_GZIP8 */
216 	{"gzip-9", zlib_decompress},  /* ZIO_COMPRESS_GZIP9 */
217 };
218 
219 
220 
221 static int zio_read_data(blkptr_t *bp, zfs_endian_t endian,
222 						 void *buf, struct zfs_data *data);
223 
224 static int
225 zio_read(blkptr_t *bp, zfs_endian_t endian, void **buf,
226 		 size_t *size, struct zfs_data *data);
227 
228 /*
229  * Our own version of log2().  Same thing as highbit()-1.
230  */
231 static int
232 zfs_log2(uint64_t num)
233 {
234 	int i = 0;
235 
236 	while (num > 1) {
237 		i++;
238 		num = num >> 1;
239 	}
240 
241 	return i;
242 }
243 
244 
245 /* Checksum Functions */
246 static void
247 zio_checksum_off(const void *buf __attribute__ ((unused)),
248 				 uint64_t size __attribute__ ((unused)),
249 				 zfs_endian_t endian __attribute__ ((unused)),
250 				 zio_cksum_t *zcp)
251 {
252 	ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
253 }
254 
255 /* Checksum Table and Values */
256 static zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = {
257 	{NULL, 0, 0, "inherit"},
258 	{NULL, 0, 0, "on"},
259 	{zio_checksum_off, 0, 0, "off"},
260 	{zio_checksum_SHA256, 1, 1, "label"},
261 	{zio_checksum_SHA256, 1, 1, "gang_header"},
262 	{NULL, 0, 0, "zilog"},
263 	{fletcher_2_endian, 0, 0, "fletcher2"},
264 	{fletcher_4_endian, 1, 0, "fletcher4"},
265 	{zio_checksum_SHA256, 1, 0, "SHA256"},
266 	{NULL, 0, 0, "zilog2"},
267 };
268 
269 /*
270  * zio_checksum_verify: Provides support for checksum verification.
271  *
272  * Fletcher2, Fletcher4, and SHA256 are supported.
273  *
274  */
275 static int
276 zio_checksum_verify(zio_cksum_t zc, uint32_t checksum,
277 					zfs_endian_t endian, char *buf, int size)
278 {
279 	zio_eck_t *zec = (zio_eck_t *) (buf + size) - 1;
280 	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
281 	zio_cksum_t actual_cksum, expected_cksum;
282 
283 	if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func == NULL) {
284 		printf("zfs unknown checksum function %d\n", checksum);
285 		return ZFS_ERR_NOT_IMPLEMENTED_YET;
286 	}
287 
288 	if (ci->ci_eck) {
289 		expected_cksum = zec->zec_cksum;
290 		zec->zec_cksum = zc;
291 		ci->ci_func(buf, size, endian, &actual_cksum);
292 		zec->zec_cksum = expected_cksum;
293 		zc = expected_cksum;
294 	} else {
295 		ci->ci_func(buf, size, endian, &actual_cksum);
296 	}
297 
298 	if ((actual_cksum.zc_word[0] != zc.zc_word[0])
299 		|| (actual_cksum.zc_word[1] != zc.zc_word[1])
300 		|| (actual_cksum.zc_word[2] != zc.zc_word[2])
301 		|| (actual_cksum.zc_word[3] != zc.zc_word[3])) {
302 		return ZFS_ERR_BAD_FS;
303 	}
304 
305 	return ZFS_ERR_NONE;
306 }
307 
308 /*
309  * vdev_uberblock_compare takes two uberblock structures and returns an integer
310  * indicating the more recent of the two.
311  *	Return Value = 1 if ub2 is more recent
312  *	Return Value = -1 if ub1 is more recent
313  * The most recent uberblock is determined using its transaction number and
314  * timestamp.  The uberblock with the highest transaction number is
315  * considered "newer".	If the transaction numbers of the two blocks match, the
316  * timestamps are compared to determine the "newer" of the two.
317  */
318 static int
319 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
320 {
321 	zfs_endian_t ub1_endian, ub2_endian;
322 	if (zfs_to_cpu64(ub1->ub_magic, LITTLE_ENDIAN) == UBERBLOCK_MAGIC)
323 		ub1_endian = LITTLE_ENDIAN;
324 	else
325 		ub1_endian = BIG_ENDIAN;
326 	if (zfs_to_cpu64(ub2->ub_magic, LITTLE_ENDIAN) == UBERBLOCK_MAGIC)
327 		ub2_endian = LITTLE_ENDIAN;
328 	else
329 		ub2_endian = BIG_ENDIAN;
330 
331 	if (zfs_to_cpu64(ub1->ub_txg, ub1_endian)
332 		< zfs_to_cpu64(ub2->ub_txg, ub2_endian))
333 		return -1;
334 	if (zfs_to_cpu64(ub1->ub_txg, ub1_endian)
335 		> zfs_to_cpu64(ub2->ub_txg, ub2_endian))
336 		return 1;
337 
338 	if (zfs_to_cpu64(ub1->ub_timestamp, ub1_endian)
339 		< zfs_to_cpu64(ub2->ub_timestamp, ub2_endian))
340 		return -1;
341 	if (zfs_to_cpu64(ub1->ub_timestamp, ub1_endian)
342 		> zfs_to_cpu64(ub2->ub_timestamp, ub2_endian))
343 		return 1;
344 
345 	return 0;
346 }
347 
348 /*
349  * Three pieces of information are needed to verify an uberblock: the magic
350  * number, the version number, and the checksum.
351  *
352  * Currently Implemented: version number, magic number, label txg
353  * Need to Implement: checksum
354  *
355  */
356 static int
357 uberblock_verify(uberblock_t *uber, int offset, struct zfs_data *data)
358 {
359 	int err;
360 	zfs_endian_t endian = UNKNOWN_ENDIAN;
361 	zio_cksum_t zc;
362 
363 	if (uber->ub_txg < data->label_txg) {
364 		debug("ignoring partially written label: uber_txg < label_txg %llu %llu\n",
365 			  uber->ub_txg, data->label_txg);
366 		return ZFS_ERR_BAD_FS;
367 	}
368 
369 	if (zfs_to_cpu64(uber->ub_magic, LITTLE_ENDIAN) == UBERBLOCK_MAGIC
370 		&& zfs_to_cpu64(uber->ub_version, LITTLE_ENDIAN) > 0
371 		&& zfs_to_cpu64(uber->ub_version, LITTLE_ENDIAN) <= SPA_VERSION)
372 		endian = LITTLE_ENDIAN;
373 
374 	if (zfs_to_cpu64(uber->ub_magic, BIG_ENDIAN) == UBERBLOCK_MAGIC
375 		&& zfs_to_cpu64(uber->ub_version, BIG_ENDIAN) > 0
376 		&& zfs_to_cpu64(uber->ub_version, BIG_ENDIAN) <= SPA_VERSION)
377 		endian = BIG_ENDIAN;
378 
379 	if (endian == UNKNOWN_ENDIAN) {
380 		printf("invalid uberblock magic\n");
381 		return ZFS_ERR_BAD_FS;
382 	}
383 
384 	memset(&zc, 0, sizeof(zc));
385 	zc.zc_word[0] = cpu_to_zfs64(offset, endian);
386 	err = zio_checksum_verify(zc, ZIO_CHECKSUM_LABEL, endian,
387 							  (char *) uber, UBERBLOCK_SIZE(data->vdev_ashift));
388 
389 	if (!err) {
390 		/* Check that the data pointed by the rootbp is usable. */
391 		void *osp = NULL;
392 		size_t ospsize;
393 		err = zio_read(&uber->ub_rootbp, endian, &osp, &ospsize, data);
394 		free(osp);
395 
396 		if (!err && ospsize < OBJSET_PHYS_SIZE_V14) {
397 			printf("uberblock rootbp points to invalid data\n");
398 			return ZFS_ERR_BAD_FS;
399 		}
400 	}
401 
402 	return err;
403 }
404 
405 /*
406  * Find the best uberblock.
407  * Return:
408  *	  Success - Pointer to the best uberblock.
409  *	  Failure - NULL
410  */
411 static uberblock_t *find_bestub(char *ub_array, struct zfs_data *data)
412 {
413 	const uint64_t sector = data->vdev_phys_sector;
414 	uberblock_t *ubbest = NULL;
415 	uberblock_t *ubnext;
416 	unsigned int i, offset, pickedub = 0;
417 	int err = ZFS_ERR_NONE;
418 
419 	const unsigned int UBCOUNT = UBERBLOCK_COUNT(data->vdev_ashift);
420 	const uint64_t UBBYTES = UBERBLOCK_SIZE(data->vdev_ashift);
421 
422 	for (i = 0; i < UBCOUNT; i++) {
423 		ubnext = (uberblock_t *) (i * UBBYTES + ub_array);
424 		offset = (sector << SPA_MINBLOCKSHIFT) + VDEV_PHYS_SIZE + (i * UBBYTES);
425 
426 		err = uberblock_verify(ubnext, offset, data);
427 		if (err)
428 			continue;
429 
430 		if (ubbest == NULL || vdev_uberblock_compare(ubnext, ubbest) > 0) {
431 			ubbest = ubnext;
432 			pickedub = i;
433 		}
434 	}
435 
436 	if (ubbest)
437 		debug("zfs Found best uberblock at idx %d, txg %llu\n",
438 			  pickedub, (unsigned long long) ubbest->ub_txg);
439 
440 	return ubbest;
441 }
442 
443 static inline size_t
444 get_psize(blkptr_t *bp, zfs_endian_t endian)
445 {
446 	return (((zfs_to_cpu64((bp)->blk_prop, endian) >> 16) & 0xffff) + 1)
447 			<< SPA_MINBLOCKSHIFT;
448 }
449 
450 static uint64_t
451 dva_get_offset(dva_t *dva, zfs_endian_t endian)
452 {
453 	return zfs_to_cpu64((dva)->dva_word[1],
454 							 endian) << SPA_MINBLOCKSHIFT;
455 }
456 
457 /*
458  * Read a block of data based on the gang block address dva,
459  * and put its data in buf.
460  *
461  */
462 static int
463 zio_read_gang(blkptr_t *bp, zfs_endian_t endian, dva_t *dva, void *buf,
464 			  struct zfs_data *data)
465 {
466 	zio_gbh_phys_t *zio_gb;
467 	uint64_t offset, sector;
468 	unsigned i;
469 	int err;
470 	zio_cksum_t zc;
471 
472 	memset(&zc, 0, sizeof(zc));
473 
474 	zio_gb = malloc(SPA_GANGBLOCKSIZE);
475 	if (!zio_gb)
476 		return ZFS_ERR_OUT_OF_MEMORY;
477 
478 	offset = dva_get_offset(dva, endian);
479 	sector = DVA_OFFSET_TO_PHYS_SECTOR(offset);
480 
481 	/* read in the gang block header */
482 	err = zfs_devread(sector, 0, SPA_GANGBLOCKSIZE, (char *) zio_gb);
483 
484 	if (err) {
485 		free(zio_gb);
486 		return err;
487 	}
488 
489 	/* XXX */
490 	/* self checksuming the gang block header */
491 	ZIO_SET_CHECKSUM(&zc, DVA_GET_VDEV(dva),
492 					 dva_get_offset(dva, endian), bp->blk_birth, 0);
493 	err = zio_checksum_verify(zc, ZIO_CHECKSUM_GANG_HEADER, endian,
494 							  (char *) zio_gb, SPA_GANGBLOCKSIZE);
495 	if (err) {
496 		free(zio_gb);
497 		return err;
498 	}
499 
500 	endian = (zfs_to_cpu64(bp->blk_prop, endian) >> 63) & 1;
501 
502 	for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
503 		if (zio_gb->zg_blkptr[i].blk_birth == 0)
504 			continue;
505 
506 		err = zio_read_data(&zio_gb->zg_blkptr[i], endian, buf, data);
507 		if (err) {
508 			free(zio_gb);
509 			return err;
510 		}
511 		buf = (char *) buf + get_psize(&zio_gb->zg_blkptr[i], endian);
512 	}
513 	free(zio_gb);
514 	return ZFS_ERR_NONE;
515 }
516 
517 /*
518  * Read in a block of raw data to buf.
519  */
520 static int
521 zio_read_data(blkptr_t *bp, zfs_endian_t endian, void *buf,
522 			  struct zfs_data *data)
523 {
524 	int i, psize;
525 	int err = ZFS_ERR_NONE;
526 
527 	psize = get_psize(bp, endian);
528 
529 	/* pick a good dva from the block pointer */
530 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
531 		uint64_t offset, sector;
532 
533 		if (bp->blk_dva[i].dva_word[0] == 0 && bp->blk_dva[i].dva_word[1] == 0)
534 			continue;
535 
536 		if ((zfs_to_cpu64(bp->blk_dva[i].dva_word[1], endian)>>63) & 1) {
537 			err = zio_read_gang(bp, endian, &bp->blk_dva[i], buf, data);
538 		} else {
539 			/* read in a data block */
540 			offset = dva_get_offset(&bp->blk_dva[i], endian);
541 			sector = DVA_OFFSET_TO_PHYS_SECTOR(offset);
542 
543 			err = zfs_devread(sector, 0, psize, buf);
544 		}
545 
546 		if (!err) {
547 			/*Check the underlying checksum before we rule this DVA as "good"*/
548 			uint32_t checkalgo = (zfs_to_cpu64((bp)->blk_prop, endian) >> 40) & 0xff;
549 
550 			err = zio_checksum_verify(bp->blk_cksum, checkalgo, endian, buf, psize);
551 			if (!err)
552 				return ZFS_ERR_NONE;
553 		}
554 
555 		/* If read failed or checksum bad, reset the error.	 Hopefully we've got some more DVA's to try.*/
556 	}
557 
558 	if (!err) {
559 		printf("couldn't find a valid DVA\n");
560 		err = ZFS_ERR_BAD_FS;
561 	}
562 
563 	return err;
564 }
565 
566 /*
567  * Read in a block of data, verify its checksum, decompress if needed,
568  * and put the uncompressed data in buf.
569  */
570 static int
571 zio_read(blkptr_t *bp, zfs_endian_t endian, void **buf,
572 		 size_t *size, struct zfs_data *data)
573 {
574 	size_t lsize, psize;
575 	unsigned int comp;
576 	char *compbuf = NULL;
577 	int err;
578 
579 	*buf = NULL;
580 
581 	comp = (zfs_to_cpu64((bp)->blk_prop, endian)>>32) & 0xff;
582 	lsize = (BP_IS_HOLE(bp) ? 0 :
583 			 (((zfs_to_cpu64((bp)->blk_prop, endian) & 0xffff) + 1)
584 			  << SPA_MINBLOCKSHIFT));
585 	psize = get_psize(bp, endian);
586 
587 	if (size)
588 		*size = lsize;
589 
590 	if (comp >= ZIO_COMPRESS_FUNCTIONS) {
591 		printf("compression algorithm %u not supported\n", (unsigned int) comp);
592 		return ZFS_ERR_NOT_IMPLEMENTED_YET;
593 	}
594 
595 	if (comp != ZIO_COMPRESS_OFF && decomp_table[comp].decomp_func == NULL) {
596 		printf("compression algorithm %s not supported\n", decomp_table[comp].name);
597 		return ZFS_ERR_NOT_IMPLEMENTED_YET;
598 	}
599 
600 	if (comp != ZIO_COMPRESS_OFF) {
601 		compbuf = malloc(psize);
602 		if (!compbuf)
603 			return ZFS_ERR_OUT_OF_MEMORY;
604 	} else {
605 		compbuf = *buf = malloc(lsize);
606 	}
607 
608 	err = zio_read_data(bp, endian, compbuf, data);
609 	if (err) {
610 		free(compbuf);
611 		*buf = NULL;
612 		return err;
613 	}
614 
615 	if (comp != ZIO_COMPRESS_OFF) {
616 		*buf = malloc(lsize);
617 		if (!*buf) {
618 			free(compbuf);
619 			return ZFS_ERR_OUT_OF_MEMORY;
620 		}
621 
622 		err = decomp_table[comp].decomp_func(compbuf, *buf, psize, lsize);
623 		free(compbuf);
624 		if (err) {
625 			free(*buf);
626 			*buf = NULL;
627 			return err;
628 		}
629 	}
630 
631 	return ZFS_ERR_NONE;
632 }
633 
634 /*
635  * Get the block from a block id.
636  * push the block onto the stack.
637  *
638  */
639 static int
640 dmu_read(dnode_end_t *dn, uint64_t blkid, void **buf,
641 		 zfs_endian_t *endian_out, struct zfs_data *data)
642 {
643 	int idx, level;
644 	blkptr_t *bp_array = dn->dn.dn_blkptr;
645 	int epbs = dn->dn.dn_indblkshift - SPA_BLKPTRSHIFT;
646 	blkptr_t *bp;
647 	void *tmpbuf = 0;
648 	zfs_endian_t endian;
649 	int err = ZFS_ERR_NONE;
650 
651 	bp = malloc(sizeof(blkptr_t));
652 	if (!bp)
653 		return ZFS_ERR_OUT_OF_MEMORY;
654 
655 	endian = dn->endian;
656 	for (level = dn->dn.dn_nlevels - 1; level >= 0; level--) {
657 		idx = (blkid >> (epbs * level)) & ((1 << epbs) - 1);
658 		*bp = bp_array[idx];
659 		if (bp_array != dn->dn.dn_blkptr) {
660 			free(bp_array);
661 			bp_array = 0;
662 		}
663 
664 		if (BP_IS_HOLE(bp)) {
665 			size_t size = zfs_to_cpu16(dn->dn.dn_datablkszsec,
666 											dn->endian)
667 				<< SPA_MINBLOCKSHIFT;
668 			*buf = malloc(size);
669 			if (*buf) {
670 				err = ZFS_ERR_OUT_OF_MEMORY;
671 				break;
672 			}
673 			memset(*buf, 0, size);
674 			endian = (zfs_to_cpu64(bp->blk_prop, endian) >> 63) & 1;
675 			break;
676 		}
677 		if (level == 0) {
678 			err = zio_read(bp, endian, buf, 0, data);
679 			endian = (zfs_to_cpu64(bp->blk_prop, endian) >> 63) & 1;
680 			break;
681 		}
682 		err = zio_read(bp, endian, &tmpbuf, 0, data);
683 		endian = (zfs_to_cpu64(bp->blk_prop, endian) >> 63) & 1;
684 		if (err)
685 			break;
686 		bp_array = tmpbuf;
687 	}
688 	if (bp_array != dn->dn.dn_blkptr)
689 		free(bp_array);
690 	if (endian_out)
691 		*endian_out = endian;
692 
693 	free(bp);
694 	return err;
695 }
696 
697 /*
698  * mzap_lookup: Looks up property described by "name" and returns the value
699  * in "value".
700  */
701 static int
702 mzap_lookup(mzap_phys_t *zapobj, zfs_endian_t endian,
703 			int objsize, char *name, uint64_t * value)
704 {
705 	int i, chunks;
706 	mzap_ent_phys_t *mzap_ent = zapobj->mz_chunk;
707 
708 	chunks = objsize / MZAP_ENT_LEN - 1;
709 	for (i = 0; i < chunks; i++) {
710 		if (strcmp(mzap_ent[i].mze_name, name) == 0) {
711 			*value = zfs_to_cpu64(mzap_ent[i].mze_value, endian);
712 			return ZFS_ERR_NONE;
713 		}
714 	}
715 
716 	printf("couldn't find '%s'\n", name);
717 	return ZFS_ERR_FILE_NOT_FOUND;
718 }
719 
720 static int
721 mzap_iterate(mzap_phys_t *zapobj, zfs_endian_t endian, int objsize,
722 			 int (*hook)(const char *name,
723 						 uint64_t val,
724 						 struct zfs_data *data),
725 			 struct zfs_data *data)
726 {
727 	int i, chunks;
728 	mzap_ent_phys_t *mzap_ent = zapobj->mz_chunk;
729 
730 	chunks = objsize / MZAP_ENT_LEN - 1;
731 	for (i = 0; i < chunks; i++) {
732 		if (hook(mzap_ent[i].mze_name,
733 				 zfs_to_cpu64(mzap_ent[i].mze_value, endian),
734 				 data))
735 			return 1;
736 	}
737 
738 	return 0;
739 }
740 
741 static uint64_t
742 zap_hash(uint64_t salt, const char *name)
743 {
744 	static uint64_t table[256];
745 	const uint8_t *cp;
746 	uint8_t c;
747 	uint64_t crc = salt;
748 
749 	if (table[128] == 0) {
750 		uint64_t *ct;
751 		int i, j;
752 		for (i = 0; i < 256; i++) {
753 			for (ct = table + i, *ct = i, j = 8; j > 0; j--)
754 				*ct = (*ct >> 1) ^ (-(*ct & 1) & ZFS_CRC64_POLY);
755 		}
756 	}
757 
758 	for (cp = (const uint8_t *) name; (c = *cp) != '\0'; cp++)
759 		crc = (crc >> 8) ^ table[(crc ^ c) & 0xFF];
760 
761 	/*
762 	 * Only use 28 bits, since we need 4 bits in the cookie for the
763 	 * collision differentiator.  We MUST use the high bits, since
764 	 * those are the onces that we first pay attention to when
765 	 * chosing the bucket.
766 	 */
767 	crc &= ~((1ULL << (64 - ZAP_HASHBITS)) - 1);
768 
769 	return crc;
770 }
771 
772 /*
773  * Only to be used on 8-bit arrays.
774  * array_len is actual len in bytes (not encoded le_value_length).
775  * buf is null-terminated.
776  */
777 /* XXX */
778 static int
779 zap_leaf_array_equal(zap_leaf_phys_t *l, zfs_endian_t endian,
780 					 int blksft, int chunk, int array_len, const char *buf)
781 {
782 	int bseen = 0;
783 
784 	while (bseen < array_len) {
785 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, blksft, chunk).l_array;
786 		int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES);
787 
788 		if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft))
789 			return 0;
790 
791 		if (memcmp(la->la_array, buf + bseen, toread) != 0)
792 			break;
793 		chunk = zfs_to_cpu16(la->la_next, endian);
794 		bseen += toread;
795 	}
796 	return (bseen == array_len);
797 }
798 
799 /* XXX */
800 static int
801 zap_leaf_array_get(zap_leaf_phys_t *l, zfs_endian_t endian, int blksft,
802 				   int chunk, int array_len, char *buf)
803 {
804 	int bseen = 0;
805 
806 	while (bseen < array_len) {
807 		struct zap_leaf_array *la = &ZAP_LEAF_CHUNK(l, blksft, chunk).l_array;
808 		int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES);
809 
810 		if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft))
811 			/* Don't use errno because this error is to be ignored.  */
812 			return ZFS_ERR_BAD_FS;
813 
814 		memcpy(buf + bseen, la->la_array,  toread);
815 		chunk = zfs_to_cpu16(la->la_next, endian);
816 		bseen += toread;
817 	}
818 	return ZFS_ERR_NONE;
819 }
820 
821 
822 /*
823  * Given a zap_leaf_phys_t, walk thru the zap leaf chunks to get the
824  * value for the property "name".
825  *
826  */
827 /* XXX */
828 static int
829 zap_leaf_lookup(zap_leaf_phys_t *l, zfs_endian_t endian,
830 				int blksft, uint64_t h,
831 				const char *name, uint64_t *value)
832 {
833 	uint16_t chunk;
834 	struct zap_leaf_entry *le;
835 
836 	/* Verify if this is a valid leaf block */
837 	if (zfs_to_cpu64(l->l_hdr.lh_block_type, endian) != ZBT_LEAF) {
838 		printf("invalid leaf type\n");
839 		return ZFS_ERR_BAD_FS;
840 	}
841 	if (zfs_to_cpu32(l->l_hdr.lh_magic, endian) != ZAP_LEAF_MAGIC) {
842 		printf("invalid leaf magic\n");
843 		return ZFS_ERR_BAD_FS;
844 	}
845 
846 	for (chunk = zfs_to_cpu16(l->l_hash[LEAF_HASH(blksft, h)], endian);
847 		 chunk != CHAIN_END; chunk = le->le_next) {
848 
849 		if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft)) {
850 			printf("invalid chunk number\n");
851 			return ZFS_ERR_BAD_FS;
852 		}
853 
854 		le = ZAP_LEAF_ENTRY(l, blksft, chunk);
855 
856 		/* Verify the chunk entry */
857 		if (le->le_type != ZAP_CHUNK_ENTRY) {
858 			printf("invalid chunk entry\n");
859 			return ZFS_ERR_BAD_FS;
860 		}
861 
862 		if (zfs_to_cpu64(le->le_hash, endian) != h)
863 			continue;
864 
865 		if (zap_leaf_array_equal(l, endian, blksft,
866 								 zfs_to_cpu16(le->le_name_chunk, endian),
867 								 zfs_to_cpu16(le->le_name_length, endian),
868 								 name)) {
869 			struct zap_leaf_array *la;
870 
871 			if (le->le_int_size != 8 || le->le_value_length != 1) {
872 				printf("invalid leaf chunk entry\n");
873 				return ZFS_ERR_BAD_FS;
874 			}
875 			/* get the uint64_t property value */
876 			la = &ZAP_LEAF_CHUNK(l, blksft, le->le_value_chunk).l_array;
877 
878 			*value = be64_to_cpu(la->la_array64);
879 
880 			return ZFS_ERR_NONE;
881 		}
882 	}
883 
884 	printf("couldn't find '%s'\n", name);
885 	return ZFS_ERR_FILE_NOT_FOUND;
886 }
887 
888 
889 /* Verify if this is a fat zap header block */
890 static int
891 zap_verify(zap_phys_t *zap)
892 {
893 	if (zap->zap_magic != (uint64_t) ZAP_MAGIC) {
894 		printf("bad ZAP magic\n");
895 		return ZFS_ERR_BAD_FS;
896 	}
897 
898 	if (zap->zap_flags != 0) {
899 		printf("bad ZAP flags\n");
900 		return ZFS_ERR_BAD_FS;
901 	}
902 
903 	if (zap->zap_salt == 0) {
904 		printf("bad ZAP salt\n");
905 		return ZFS_ERR_BAD_FS;
906 	}
907 
908 	return ZFS_ERR_NONE;
909 }
910 
911 /*
912  * Fat ZAP lookup
913  *
914  */
915 /* XXX */
916 static int
917 fzap_lookup(dnode_end_t *zap_dnode, zap_phys_t *zap,
918 			char *name, uint64_t *value, struct zfs_data *data)
919 {
920 	void *l;
921 	uint64_t hash, idx, blkid;
922 	int blksft = zfs_log2(zfs_to_cpu16(zap_dnode->dn.dn_datablkszsec,
923 											zap_dnode->endian) << DNODE_SHIFT);
924 	int err;
925 	zfs_endian_t leafendian;
926 
927 	err = zap_verify(zap);
928 	if (err)
929 		return err;
930 
931 	hash = zap_hash(zap->zap_salt, name);
932 
933 	/* get block id from index */
934 	if (zap->zap_ptrtbl.zt_numblks != 0) {
935 		printf("external pointer tables not supported\n");
936 		return ZFS_ERR_NOT_IMPLEMENTED_YET;
937 	}
938 	idx = ZAP_HASH_IDX(hash, zap->zap_ptrtbl.zt_shift);
939 	blkid = ((uint64_t *) zap)[idx + (1 << (blksft - 3 - 1))];
940 
941 	/* Get the leaf block */
942 	if ((1U << blksft) < sizeof(zap_leaf_phys_t)) {
943 		printf("ZAP leaf is too small\n");
944 		return ZFS_ERR_BAD_FS;
945 	}
946 	err = dmu_read(zap_dnode, blkid, &l, &leafendian, data);
947 	if (err)
948 		return err;
949 
950 	err = zap_leaf_lookup(l, leafendian, blksft, hash, name, value);
951 	free(l);
952 	return err;
953 }
954 
955 /* XXX */
956 static int
957 fzap_iterate(dnode_end_t *zap_dnode, zap_phys_t *zap,
958 			 int (*hook)(const char *name,
959 						 uint64_t val,
960 						 struct zfs_data *data),
961 			 struct zfs_data *data)
962 {
963 	zap_leaf_phys_t *l;
964 	void *l_in;
965 	uint64_t idx, blkid;
966 	uint16_t chunk;
967 	int blksft = zfs_log2(zfs_to_cpu16(zap_dnode->dn.dn_datablkszsec,
968 											zap_dnode->endian) << DNODE_SHIFT);
969 	int err;
970 	zfs_endian_t endian;
971 
972 	if (zap_verify(zap))
973 		return 0;
974 
975 	/* get block id from index */
976 	if (zap->zap_ptrtbl.zt_numblks != 0) {
977 		printf("external pointer tables not supported\n");
978 		return 0;
979 	}
980 	/* Get the leaf block */
981 	if ((1U << blksft) < sizeof(zap_leaf_phys_t)) {
982 		printf("ZAP leaf is too small\n");
983 		return 0;
984 	}
985 	for (idx = 0; idx < zap->zap_ptrtbl.zt_numblks; idx++) {
986 		blkid = ((uint64_t *) zap)[idx + (1 << (blksft - 3 - 1))];
987 
988 		err = dmu_read(zap_dnode, blkid, &l_in, &endian, data);
989 		l = l_in;
990 		if (err)
991 			continue;
992 
993 		/* Verify if this is a valid leaf block */
994 		if (zfs_to_cpu64(l->l_hdr.lh_block_type, endian) != ZBT_LEAF) {
995 			free(l);
996 			continue;
997 		}
998 		if (zfs_to_cpu32(l->l_hdr.lh_magic, endian) != ZAP_LEAF_MAGIC) {
999 			free(l);
1000 			continue;
1001 		}
1002 
1003 		for (chunk = 0; chunk < ZAP_LEAF_NUMCHUNKS(blksft); chunk++) {
1004 			char *buf;
1005 			struct zap_leaf_array *la;
1006 			struct zap_leaf_entry *le;
1007 			uint64_t val;
1008 			le = ZAP_LEAF_ENTRY(l, blksft, chunk);
1009 
1010 			/* Verify the chunk entry */
1011 			if (le->le_type != ZAP_CHUNK_ENTRY)
1012 				continue;
1013 
1014 			buf = malloc(zfs_to_cpu16(le->le_name_length, endian)
1015 						 + 1);
1016 			if (zap_leaf_array_get(l, endian, blksft, le->le_name_chunk,
1017 								   le->le_name_length, buf)) {
1018 				free(buf);
1019 				continue;
1020 			}
1021 			buf[le->le_name_length] = 0;
1022 
1023 			if (le->le_int_size != 8
1024 				|| zfs_to_cpu16(le->le_value_length, endian) != 1)
1025 				continue;
1026 
1027 			/* get the uint64_t property value */
1028 			la = &ZAP_LEAF_CHUNK(l, blksft, le->le_value_chunk).l_array;
1029 			val = be64_to_cpu(la->la_array64);
1030 			if (hook(buf, val, data))
1031 				return 1;
1032 			free(buf);
1033 		}
1034 	}
1035 	return 0;
1036 }
1037 
1038 
1039 /*
1040  * Read in the data of a zap object and find the value for a matching
1041  * property name.
1042  *
1043  */
1044 static int
1045 zap_lookup(dnode_end_t *zap_dnode, char *name, uint64_t *val,
1046 		   struct zfs_data *data)
1047 {
1048 	uint64_t block_type;
1049 	int size;
1050 	void *zapbuf;
1051 	int err;
1052 	zfs_endian_t endian;
1053 
1054 	/* Read in the first block of the zap object data. */
1055 	size = zfs_to_cpu16(zap_dnode->dn.dn_datablkszsec,
1056 							 zap_dnode->endian) << SPA_MINBLOCKSHIFT;
1057 	err = dmu_read(zap_dnode, 0, &zapbuf, &endian, data);
1058 	if (err)
1059 		return err;
1060 	block_type = zfs_to_cpu64(*((uint64_t *) zapbuf), endian);
1061 
1062 	if (block_type == ZBT_MICRO) {
1063 		err = (mzap_lookup(zapbuf, endian, size, name, val));
1064 		free(zapbuf);
1065 		return err;
1066 	} else if (block_type == ZBT_HEADER) {
1067 		/* this is a fat zap */
1068 		err = (fzap_lookup(zap_dnode, zapbuf, name, val, data));
1069 		free(zapbuf);
1070 		return err;
1071 	}
1072 
1073 	printf("unknown ZAP type\n");
1074 	return ZFS_ERR_BAD_FS;
1075 }
1076 
1077 static int
1078 zap_iterate(dnode_end_t *zap_dnode,
1079 			int (*hook)(const char *name, uint64_t val,
1080 						struct zfs_data *data),
1081 			struct zfs_data *data)
1082 {
1083 	uint64_t block_type;
1084 	int size;
1085 	void *zapbuf;
1086 	int err;
1087 	int ret;
1088 	zfs_endian_t endian;
1089 
1090 	/* Read in the first block of the zap object data. */
1091 	size = zfs_to_cpu16(zap_dnode->dn.dn_datablkszsec, zap_dnode->endian) << SPA_MINBLOCKSHIFT;
1092 	err = dmu_read(zap_dnode, 0, &zapbuf, &endian, data);
1093 	if (err)
1094 		return 0;
1095 	block_type = zfs_to_cpu64(*((uint64_t *) zapbuf), endian);
1096 
1097 	if (block_type == ZBT_MICRO) {
1098 		ret = mzap_iterate(zapbuf, endian, size, hook, data);
1099 		free(zapbuf);
1100 		return ret;
1101 	} else if (block_type == ZBT_HEADER) {
1102 		/* this is a fat zap */
1103 		ret = fzap_iterate(zap_dnode, zapbuf, hook, data);
1104 		free(zapbuf);
1105 		return ret;
1106 	}
1107 	printf("unknown ZAP type\n");
1108 	return 0;
1109 }
1110 
1111 
1112 /*
1113  * Get the dnode of an object number from the metadnode of an object set.
1114  *
1115  * Input
1116  *	mdn - metadnode to get the object dnode
1117  *	objnum - object number for the object dnode
1118  *	buf - data buffer that holds the returning dnode
1119  */
1120 static int
1121 dnode_get(dnode_end_t *mdn, uint64_t objnum, uint8_t type,
1122 		  dnode_end_t *buf, struct zfs_data *data)
1123 {
1124 	uint64_t blkid, blksz;	/* the block id this object dnode is in */
1125 	int epbs;			/* shift of number of dnodes in a block */
1126 	int idx;			/* index within a block */
1127 	void *dnbuf;
1128 	int err;
1129 	zfs_endian_t endian;
1130 
1131 	blksz = zfs_to_cpu16(mdn->dn.dn_datablkszsec,
1132 							  mdn->endian) << SPA_MINBLOCKSHIFT;
1133 
1134 	epbs = zfs_log2(blksz) - DNODE_SHIFT;
1135 	blkid = objnum >> epbs;
1136 	idx = objnum & ((1 << epbs) - 1);
1137 
1138 	if (data->dnode_buf != NULL && memcmp(data->dnode_mdn, mdn,
1139 										  sizeof(*mdn)) == 0
1140 		&& objnum >= data->dnode_start && objnum < data->dnode_end) {
1141 		memmove(&(buf->dn), &(data->dnode_buf)[idx], DNODE_SIZE);
1142 		buf->endian = data->dnode_endian;
1143 		if (type && buf->dn.dn_type != type)  {
1144 			printf("incorrect dnode type: %02X != %02x\n", buf->dn.dn_type, type);
1145 			return ZFS_ERR_BAD_FS;
1146 		}
1147 		return ZFS_ERR_NONE;
1148 	}
1149 
1150 	err = dmu_read(mdn, blkid, &dnbuf, &endian, data);
1151 	if (err)
1152 		return err;
1153 
1154 	free(data->dnode_buf);
1155 	free(data->dnode_mdn);
1156 	data->dnode_mdn = malloc(sizeof(*mdn));
1157 	if (!data->dnode_mdn) {
1158 		data->dnode_buf = 0;
1159 	} else {
1160 		memcpy(data->dnode_mdn, mdn, sizeof(*mdn));
1161 		data->dnode_buf = dnbuf;
1162 		data->dnode_start = blkid << epbs;
1163 		data->dnode_end = (blkid + 1) << epbs;
1164 		data->dnode_endian = endian;
1165 	}
1166 
1167 	memmove(&(buf->dn), (dnode_phys_t *) dnbuf + idx, DNODE_SIZE);
1168 	buf->endian = endian;
1169 	if (type && buf->dn.dn_type != type) {
1170 		printf("incorrect dnode type\n");
1171 		return ZFS_ERR_BAD_FS;
1172 	}
1173 
1174 	return ZFS_ERR_NONE;
1175 }
1176 
1177 /*
1178  * Get the file dnode for a given file name where mdn is the meta dnode
1179  * for this ZFS object set. When found, place the file dnode in dn.
1180  * The 'path' argument will be mangled.
1181  *
1182  */
1183 static int
1184 dnode_get_path(dnode_end_t *mdn, const char *path_in, dnode_end_t *dn,
1185 			   struct zfs_data *data)
1186 {
1187 	uint64_t objnum, version;
1188 	char *cname, ch;
1189 	int err = ZFS_ERR_NONE;
1190 	char *path, *path_buf;
1191 	struct dnode_chain {
1192 		struct dnode_chain *next;
1193 		dnode_end_t dn;
1194 	};
1195 	struct dnode_chain *dnode_path = 0, *dn_new, *root;
1196 
1197 	dn_new = malloc(sizeof(*dn_new));
1198 	if (!dn_new)
1199 		return ZFS_ERR_OUT_OF_MEMORY;
1200 	dn_new->next = 0;
1201 	dnode_path = root = dn_new;
1202 
1203 	err = dnode_get(mdn, MASTER_NODE_OBJ, DMU_OT_MASTER_NODE,
1204 					&(dnode_path->dn), data);
1205 	if (err) {
1206 		free(dn_new);
1207 		return err;
1208 	}
1209 
1210 	err = zap_lookup(&(dnode_path->dn), ZPL_VERSION_STR, &version, data);
1211 	if (err) {
1212 		free(dn_new);
1213 		return err;
1214 	}
1215 	if (version > ZPL_VERSION) {
1216 		free(dn_new);
1217 		printf("too new ZPL version\n");
1218 		return ZFS_ERR_NOT_IMPLEMENTED_YET;
1219 	}
1220 
1221 	err = zap_lookup(&(dnode_path->dn), ZFS_ROOT_OBJ, &objnum, data);
1222 	if (err) {
1223 		free(dn_new);
1224 		return err;
1225 	}
1226 
1227 	err = dnode_get(mdn, objnum, 0, &(dnode_path->dn), data);
1228 	if (err) {
1229 		free(dn_new);
1230 		return err;
1231 	}
1232 
1233 	path = path_buf = strdup(path_in);
1234 	if (!path_buf) {
1235 		free(dn_new);
1236 		return ZFS_ERR_OUT_OF_MEMORY;
1237 	}
1238 
1239 	while (1) {
1240 		/* skip leading slashes */
1241 		while (*path == '/')
1242 			path++;
1243 		if (!*path)
1244 			break;
1245 		/* get the next component name */
1246 		cname = path;
1247 		while (*path && *path != '/')
1248 			path++;
1249 		/* Skip dot.  */
1250 		if (cname + 1 == path && cname[0] == '.')
1251 			continue;
1252 		/* Handle double dot.  */
1253 		if (cname + 2 == path && cname[0] == '.' && cname[1] == '.')  {
1254 			if (dn_new->next) {
1255 				dn_new = dnode_path;
1256 				dnode_path = dn_new->next;
1257 				free(dn_new);
1258 			} else {
1259 				printf("can't resolve ..\n");
1260 				err = ZFS_ERR_FILE_NOT_FOUND;
1261 				break;
1262 			}
1263 			continue;
1264 		}
1265 
1266 		ch = *path;
1267 		*path = 0;		/* ensure null termination */
1268 
1269 		if (dnode_path->dn.dn.dn_type != DMU_OT_DIRECTORY_CONTENTS) {
1270 			free(path_buf);
1271 			printf("not a directory\n");
1272 			return ZFS_ERR_BAD_FILE_TYPE;
1273 		}
1274 		err = zap_lookup(&(dnode_path->dn), cname, &objnum, data);
1275 		if (err)
1276 			break;
1277 
1278 		dn_new = malloc(sizeof(*dn_new));
1279 		if (!dn_new) {
1280 			err = ZFS_ERR_OUT_OF_MEMORY;
1281 			break;
1282 		}
1283 		dn_new->next = dnode_path;
1284 		dnode_path = dn_new;
1285 
1286 		objnum = ZFS_DIRENT_OBJ(objnum);
1287 		err = dnode_get(mdn, objnum, 0, &(dnode_path->dn), data);
1288 		if (err)
1289 			break;
1290 
1291 		*path = ch;
1292 	}
1293 
1294 	if (!err)
1295 		memcpy(dn, &(dnode_path->dn), sizeof(*dn));
1296 
1297 	while (dnode_path) {
1298 		dn_new = dnode_path->next;
1299 		free(dnode_path);
1300 		dnode_path = dn_new;
1301 	}
1302 	free(path_buf);
1303 	return err;
1304 }
1305 
1306 
1307 /*
1308  * Given a MOS metadnode, get the metadnode of a given filesystem name (fsname),
1309  * e.g. pool/rootfs, or a given object number (obj), e.g. the object number
1310  * of pool/rootfs.
1311  *
1312  * If no fsname and no obj are given, return the DSL_DIR metadnode.
1313  * If fsname is given, return its metadnode and its matching object number.
1314  * If only obj is given, return the metadnode for this object number.
1315  *
1316  */
1317 static int
1318 get_filesystem_dnode(dnode_end_t *mosmdn, char *fsname,
1319 					 dnode_end_t *mdn, struct zfs_data *data)
1320 {
1321 	uint64_t objnum;
1322 	int err;
1323 
1324 	err = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT,
1325 					DMU_OT_OBJECT_DIRECTORY, mdn, data);
1326 	if (err)
1327 		return err;
1328 
1329 	err = zap_lookup(mdn, DMU_POOL_ROOT_DATASET, &objnum, data);
1330 	if (err)
1331 		return err;
1332 
1333 	err = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, mdn, data);
1334 	if (err)
1335 		return err;
1336 
1337 	while (*fsname) {
1338 		uint64_t childobj;
1339 		char *cname, ch;
1340 
1341 		while (*fsname == '/')
1342 			fsname++;
1343 
1344 		if (!*fsname || *fsname == '@')
1345 			break;
1346 
1347 		cname = fsname;
1348 		while (*fsname && !isspace(*fsname) && *fsname != '/')
1349 			fsname++;
1350 		ch = *fsname;
1351 		*fsname = 0;
1352 
1353 		childobj = zfs_to_cpu64((((dsl_dir_phys_t *) DN_BONUS(&mdn->dn)))->dd_child_dir_zapobj, mdn->endian);
1354 		err = dnode_get(mosmdn, childobj,
1355 						DMU_OT_DSL_DIR_CHILD_MAP, mdn, data);
1356 		if (err)
1357 			return err;
1358 
1359 		err = zap_lookup(mdn, cname, &objnum, data);
1360 		if (err)
1361 			return err;
1362 
1363 		err = dnode_get(mosmdn, objnum, DMU_OT_DSL_DIR, mdn, data);
1364 		if (err)
1365 			return err;
1366 
1367 		*fsname = ch;
1368 	}
1369 	return ZFS_ERR_NONE;
1370 }
1371 
1372 static int
1373 make_mdn(dnode_end_t *mdn, struct zfs_data *data)
1374 {
1375 	void *osp;
1376 	blkptr_t *bp;
1377 	size_t ospsize;
1378 	int err;
1379 
1380 	bp = &(((dsl_dataset_phys_t *) DN_BONUS(&mdn->dn))->ds_bp);
1381 	err = zio_read(bp, mdn->endian, &osp, &ospsize, data);
1382 	if (err)
1383 		return err;
1384 	if (ospsize < OBJSET_PHYS_SIZE_V14) {
1385 		free(osp);
1386 		printf("too small osp\n");
1387 		return ZFS_ERR_BAD_FS;
1388 	}
1389 
1390 	mdn->endian = (zfs_to_cpu64(bp->blk_prop, mdn->endian)>>63) & 1;
1391 	memmove((char *) &(mdn->dn),
1392 			(char *) &((objset_phys_t *) osp)->os_meta_dnode, DNODE_SIZE);
1393 	free(osp);
1394 	return ZFS_ERR_NONE;
1395 }
1396 
1397 static int
1398 dnode_get_fullpath(const char *fullpath, dnode_end_t *mdn,
1399 				   uint64_t *mdnobj, dnode_end_t *dn, int *isfs,
1400 				   struct zfs_data *data)
1401 {
1402 	char *fsname, *snapname;
1403 	const char *ptr_at, *filename;
1404 	uint64_t headobj;
1405 	int err;
1406 
1407 	ptr_at = strchr(fullpath, '@');
1408 	if (!ptr_at) {
1409 		*isfs = 1;
1410 		filename = 0;
1411 		snapname = 0;
1412 		fsname = strdup(fullpath);
1413 	} else {
1414 		const char *ptr_slash = strchr(ptr_at, '/');
1415 
1416 		*isfs = 0;
1417 		fsname = malloc(ptr_at - fullpath + 1);
1418 		if (!fsname)
1419 			return ZFS_ERR_OUT_OF_MEMORY;
1420 		memcpy(fsname, fullpath, ptr_at - fullpath);
1421 		fsname[ptr_at - fullpath] = 0;
1422 		if (ptr_at[1] && ptr_at[1] != '/') {
1423 			snapname = malloc(ptr_slash - ptr_at);
1424 			if (!snapname) {
1425 				free(fsname);
1426 				return ZFS_ERR_OUT_OF_MEMORY;
1427 			}
1428 			memcpy(snapname, ptr_at + 1, ptr_slash - ptr_at - 1);
1429 			snapname[ptr_slash - ptr_at - 1] = 0;
1430 		} else {
1431 			snapname = 0;
1432 		}
1433 		if (ptr_slash)
1434 			filename = ptr_slash;
1435 		else
1436 			filename = "/";
1437 		printf("zfs fsname = '%s' snapname='%s' filename = '%s'\n",
1438 			   fsname, snapname, filename);
1439 	}
1440 
1441 
1442 	err = get_filesystem_dnode(&(data->mos), fsname, dn, data);
1443 
1444 	if (err) {
1445 		free(fsname);
1446 		free(snapname);
1447 		return err;
1448 	}
1449 
1450 	headobj = zfs_to_cpu64(((dsl_dir_phys_t *) DN_BONUS(&dn->dn))->dd_head_dataset_obj, dn->endian);
1451 
1452 	err = dnode_get(&(data->mos), headobj, DMU_OT_DSL_DATASET, mdn, data);
1453 	if (err) {
1454 		free(fsname);
1455 		free(snapname);
1456 		return err;
1457 	}
1458 
1459 	if (snapname) {
1460 		uint64_t snapobj;
1461 
1462 		snapobj = zfs_to_cpu64(((dsl_dataset_phys_t *) DN_BONUS(&mdn->dn))->ds_snapnames_zapobj, mdn->endian);
1463 
1464 		err = dnode_get(&(data->mos), snapobj,
1465 						DMU_OT_DSL_DS_SNAP_MAP, mdn, data);
1466 		if (!err)
1467 			err = zap_lookup(mdn, snapname, &headobj, data);
1468 		if (!err)
1469 			err = dnode_get(&(data->mos), headobj, DMU_OT_DSL_DATASET, mdn, data);
1470 		if (err) {
1471 			free(fsname);
1472 			free(snapname);
1473 			return err;
1474 		}
1475 	}
1476 
1477 	if (mdnobj)
1478 		*mdnobj = headobj;
1479 
1480 	make_mdn(mdn, data);
1481 
1482 	if (*isfs) {
1483 		free(fsname);
1484 		free(snapname);
1485 		return ZFS_ERR_NONE;
1486 	}
1487 	err = dnode_get_path(mdn, filename, dn, data);
1488 	free(fsname);
1489 	free(snapname);
1490 	return err;
1491 }
1492 
1493 /*
1494  * For a given XDR packed nvlist, verify the first 4 bytes and move on.
1495  *
1496  * An XDR packed nvlist is encoded as (comments from nvs_xdr_create) :
1497  *
1498  *		encoding method/host endian		(4 bytes)
1499  *		nvl_version						(4 bytes)
1500  *		nvl_nvflag						(4 bytes)
1501  *	encoded nvpairs:
1502  *		encoded size of the nvpair		(4 bytes)
1503  *		decoded size of the nvpair		(4 bytes)
1504  *		name string size				(4 bytes)
1505  *		name string data				(sizeof(NV_ALIGN4(string))
1506  *		data type						(4 bytes)
1507  *		# of elements in the nvpair		(4 bytes)
1508  *		data
1509  *		2 zero's for the last nvpair
1510  *		(end of the entire list)	(8 bytes)
1511  *
1512  */
1513 
1514 static int
1515 nvlist_find_value(char *nvlist, char *name, int valtype, char **val,
1516 				  size_t *size_out, size_t *nelm_out)
1517 {
1518 	int name_len, type, encode_size;
1519 	char *nvpair, *nvp_name;
1520 
1521 	/* Verify if the 1st and 2nd byte in the nvlist are valid. */
1522 	/* NOTE: independently of what endianness header announces all
1523 	   subsequent values are big-endian.  */
1524 	if (nvlist[0] != NV_ENCODE_XDR || (nvlist[1] != NV_LITTLE_ENDIAN
1525 									   && nvlist[1] != NV_BIG_ENDIAN)) {
1526 		printf("zfs incorrect nvlist header\n");
1527 		return ZFS_ERR_BAD_FS;
1528 	}
1529 
1530 	/* skip the header, nvl_version, and nvl_nvflag */
1531 	nvlist = nvlist + 4 * 3;
1532 	/*
1533 	 * Loop thru the nvpair list
1534 	 * The XDR representation of an integer is in big-endian byte order.
1535 	 */
1536 	while ((encode_size = be32_to_cpu(*(uint32_t *) nvlist))) {
1537 		int nelm;
1538 
1539 		nvpair = nvlist + 4 * 2;	/* skip the encode/decode size */
1540 
1541 		name_len = be32_to_cpu(*(uint32_t *) nvpair);
1542 		nvpair += 4;
1543 
1544 		nvp_name = nvpair;
1545 		nvpair = nvpair + ((name_len + 3) & ~3);	/* align */
1546 
1547 		type = be32_to_cpu(*(uint32_t *) nvpair);
1548 		nvpair += 4;
1549 
1550 		nelm = be32_to_cpu(*(uint32_t *) nvpair);
1551 		if (nelm < 1) {
1552 			printf("empty nvpair\n");
1553 			return ZFS_ERR_BAD_FS;
1554 		}
1555 
1556 		nvpair += 4;
1557 
1558 		if ((strncmp(nvp_name, name, name_len) == 0) && type == valtype) {
1559 			*val = nvpair;
1560 			*size_out = encode_size;
1561 			if (nelm_out)
1562 				*nelm_out = nelm;
1563 			return 1;
1564 		}
1565 
1566 		nvlist += encode_size;	/* goto the next nvpair */
1567 	}
1568 	return 0;
1569 }
1570 
1571 int
1572 zfs_nvlist_lookup_uint64(char *nvlist, char *name, uint64_t *out)
1573 {
1574 	char *nvpair;
1575 	size_t size;
1576 	int found;
1577 
1578 	found = nvlist_find_value(nvlist, name, DATA_TYPE_UINT64, &nvpair, &size, 0);
1579 	if (!found)
1580 		return 0;
1581 	if (size < sizeof(uint64_t)) {
1582 		printf("invalid uint64\n");
1583 		return ZFS_ERR_BAD_FS;
1584 	}
1585 
1586 	*out = be64_to_cpu(*(uint64_t *) nvpair);
1587 	return 1;
1588 }
1589 
1590 char *
1591 zfs_nvlist_lookup_string(char *nvlist, char *name)
1592 {
1593 	char *nvpair;
1594 	char *ret;
1595 	size_t slen;
1596 	size_t size;
1597 	int found;
1598 
1599 	found = nvlist_find_value(nvlist, name, DATA_TYPE_STRING, &nvpair, &size, 0);
1600 	if (!found)
1601 		return 0;
1602 	if (size < 4) {
1603 		printf("invalid string\n");
1604 		return 0;
1605 	}
1606 	slen = be32_to_cpu(*(uint32_t *) nvpair);
1607 	if (slen > size - 4)
1608 		slen = size - 4;
1609 	ret = malloc(slen + 1);
1610 	if (!ret)
1611 		return 0;
1612 	memcpy(ret, nvpair + 4, slen);
1613 	ret[slen] = 0;
1614 	return ret;
1615 }
1616 
1617 char *
1618 zfs_nvlist_lookup_nvlist(char *nvlist, char *name)
1619 {
1620 	char *nvpair;
1621 	char *ret;
1622 	size_t size;
1623 	int found;
1624 
1625 	found = nvlist_find_value(nvlist, name, DATA_TYPE_NVLIST, &nvpair,
1626 							  &size, 0);
1627 	if (!found)
1628 		return 0;
1629 	ret = calloc(1, size + 3 * sizeof(uint32_t));
1630 	if (!ret)
1631 		return 0;
1632 	memcpy(ret, nvlist, sizeof(uint32_t));
1633 
1634 	memcpy(ret + sizeof(uint32_t), nvpair, size);
1635 	return ret;
1636 }
1637 
1638 int
1639 zfs_nvlist_lookup_nvlist_array_get_nelm(char *nvlist, char *name)
1640 {
1641 	char *nvpair;
1642 	size_t nelm, size;
1643 	int found;
1644 
1645 	found = nvlist_find_value(nvlist, name, DATA_TYPE_NVLIST, &nvpair,
1646 							  &size, &nelm);
1647 	if (!found)
1648 		return -1;
1649 	return nelm;
1650 }
1651 
1652 char *
1653 zfs_nvlist_lookup_nvlist_array(char *nvlist, char *name,
1654 									size_t index)
1655 {
1656 	char *nvpair, *nvpairptr;
1657 	int found;
1658 	char *ret;
1659 	size_t size;
1660 	unsigned i;
1661 	size_t nelm;
1662 
1663 	found = nvlist_find_value(nvlist, name, DATA_TYPE_NVLIST, &nvpair,
1664 							  &size, &nelm);
1665 	if (!found)
1666 		return 0;
1667 	if (index >= nelm) {
1668 		printf("trying to lookup past nvlist array\n");
1669 		return 0;
1670 	}
1671 
1672 	nvpairptr = nvpair;
1673 
1674 	for (i = 0; i < index; i++) {
1675 		uint32_t encode_size;
1676 
1677 		/* skip the header, nvl_version, and nvl_nvflag */
1678 		nvpairptr = nvpairptr + 4 * 2;
1679 
1680 		while (nvpairptr < nvpair + size
1681 			   && (encode_size = be32_to_cpu(*(uint32_t *) nvpairptr)))
1682 			nvlist += encode_size;	/* goto the next nvpair */
1683 
1684 		nvlist = nvlist + 4 * 2;	/* skip the ending 2 zeros - 8 bytes */
1685 	}
1686 
1687 	if (nvpairptr >= nvpair + size
1688 		|| nvpairptr + be32_to_cpu(*(uint32_t *) (nvpairptr + 4 * 2))
1689 		>= nvpair + size) {
1690 		printf("incorrect nvlist array\n");
1691 		return 0;
1692 	}
1693 
1694 	ret = calloc(1, be32_to_cpu(*(uint32_t *) (nvpairptr + 4 * 2))
1695 				 + 3 * sizeof(uint32_t));
1696 	if (!ret)
1697 		return 0;
1698 	memcpy(ret, nvlist, sizeof(uint32_t));
1699 
1700 	memcpy(ret + sizeof(uint32_t), nvpairptr, size);
1701 	return ret;
1702 }
1703 
1704 static int
1705 int_zfs_fetch_nvlist(struct zfs_data *data, char **nvlist)
1706 {
1707 	int err;
1708 
1709 	*nvlist = malloc(VDEV_PHYS_SIZE);
1710 	/* Read in the vdev name-value pair list (112K). */
1711 	err = zfs_devread(data->vdev_phys_sector, 0, VDEV_PHYS_SIZE, *nvlist);
1712 	if (err) {
1713 		free(*nvlist);
1714 		*nvlist = 0;
1715 		return err;
1716 	}
1717 	return ZFS_ERR_NONE;
1718 }
1719 
1720 /*
1721  * Check the disk label information and retrieve needed vdev name-value pairs.
1722  *
1723  */
1724 static int
1725 check_pool_label(struct zfs_data *data)
1726 {
1727 	uint64_t pool_state;
1728 	char *nvlist;			/* for the pool */
1729 	char *vdevnvlist;		/* for the vdev */
1730 	uint64_t diskguid;
1731 	uint64_t version;
1732 	int found;
1733 	int err;
1734 
1735 	err = int_zfs_fetch_nvlist(data, &nvlist);
1736 	if (err)
1737 		return err;
1738 
1739 	found = zfs_nvlist_lookup_uint64(nvlist, ZPOOL_CONFIG_POOL_STATE,
1740 										  &pool_state);
1741 	if (!found) {
1742 		free(nvlist);
1743 		printf("zfs pool state not found\n");
1744 		return ZFS_ERR_BAD_FS;
1745 	}
1746 
1747 	if (pool_state == POOL_STATE_DESTROYED) {
1748 		free(nvlist);
1749 		printf("zpool is marked as destroyed\n");
1750 		return ZFS_ERR_BAD_FS;
1751 	}
1752 
1753 	data->label_txg = 0;
1754 	found = zfs_nvlist_lookup_uint64(nvlist, ZPOOL_CONFIG_POOL_TXG,
1755 										  &data->label_txg);
1756 	if (!found) {
1757 		free(nvlist);
1758 		printf("zfs pool txg not found\n");
1759 		return ZFS_ERR_BAD_FS;
1760 	}
1761 
1762 	/* not an active device */
1763 	if (data->label_txg == 0) {
1764 		free(nvlist);
1765 		printf("zpool is not active\n");
1766 		return ZFS_ERR_BAD_FS;
1767 	}
1768 
1769 	found = zfs_nvlist_lookup_uint64(nvlist, ZPOOL_CONFIG_VERSION,
1770 										  &version);
1771 	if (!found) {
1772 		free(nvlist);
1773 		printf("zpool config version not found\n");
1774 		return ZFS_ERR_BAD_FS;
1775 	}
1776 
1777 	if (version > SPA_VERSION) {
1778 		free(nvlist);
1779 		printf("SPA version too new %llu > %llu\n",
1780 			   (unsigned long long) version,
1781 			   (unsigned long long) SPA_VERSION);
1782 		return ZFS_ERR_NOT_IMPLEMENTED_YET;
1783 	}
1784 
1785 	vdevnvlist = zfs_nvlist_lookup_nvlist(nvlist, ZPOOL_CONFIG_VDEV_TREE);
1786 	if (!vdevnvlist) {
1787 		free(nvlist);
1788 		printf("ZFS config vdev tree not found\n");
1789 		return ZFS_ERR_BAD_FS;
1790 	}
1791 
1792 	found = zfs_nvlist_lookup_uint64(vdevnvlist, ZPOOL_CONFIG_ASHIFT,
1793 										  &data->vdev_ashift);
1794 	free(vdevnvlist);
1795 	if (!found) {
1796 		free(nvlist);
1797 		printf("ZPOOL config ashift not found\n");
1798 		return ZFS_ERR_BAD_FS;
1799 	}
1800 
1801 	found = zfs_nvlist_lookup_uint64(nvlist, ZPOOL_CONFIG_GUID, &diskguid);
1802 	if (!found) {
1803 		free(nvlist);
1804 		printf("ZPOOL config guid not found\n");
1805 		return ZFS_ERR_BAD_FS;
1806 	}
1807 
1808 	found = zfs_nvlist_lookup_uint64(nvlist, ZPOOL_CONFIG_POOL_GUID, &data->pool_guid);
1809 	if (!found) {
1810 		free(nvlist);
1811 		printf("ZPOOL config pool guid not found\n");
1812 		return ZFS_ERR_BAD_FS;
1813 	}
1814 
1815 	free(nvlist);
1816 
1817 	printf("ZFS Pool GUID: %llu (%016llx) Label: GUID: %llu (%016llx), txg: %llu, SPA v%llu, ashift: %llu\n",
1818 		   (unsigned long long) data->pool_guid,
1819 		   (unsigned long long) data->pool_guid,
1820 		   (unsigned long long) diskguid,
1821 		   (unsigned long long) diskguid,
1822 		   (unsigned long long) data->label_txg,
1823 		   (unsigned long long) version,
1824 		   (unsigned long long) data->vdev_ashift);
1825 
1826 	return ZFS_ERR_NONE;
1827 }
1828 
1829 /*
1830  * vdev_label_start returns the physical disk offset (in bytes) of
1831  * label "l".
1832  */
1833 static uint64_t vdev_label_start(uint64_t psize, int l)
1834 {
1835 	return (l * sizeof(vdev_label_t) + (l < VDEV_LABELS / 2 ?
1836 										0 : psize -
1837 										VDEV_LABELS * sizeof(vdev_label_t)));
1838 }
1839 
1840 void
1841 zfs_unmount(struct zfs_data *data)
1842 {
1843 	free(data->dnode_buf);
1844 	free(data->dnode_mdn);
1845 	free(data->file_buf);
1846 	free(data);
1847 }
1848 
1849 /*
1850  * zfs_mount() locates a valid uberblock of the root pool and read in its MOS
1851  * to the memory address MOS.
1852  *
1853  */
1854 struct zfs_data *
1855 zfs_mount(device_t dev)
1856 {
1857 	struct zfs_data *data = 0;
1858 	int label = 0, bestlabel = -1;
1859 	char *ub_array;
1860 	uberblock_t *ubbest;
1861 	uberblock_t *ubcur = NULL;
1862 	void *osp = 0;
1863 	size_t ospsize;
1864 	int err;
1865 
1866 	data = malloc(sizeof(*data));
1867 	if (!data)
1868 		return 0;
1869 	memset(data, 0, sizeof(*data));
1870 
1871 	ub_array = malloc(VDEV_UBERBLOCK_RING);
1872 	if (!ub_array) {
1873 		zfs_unmount(data);
1874 		return 0;
1875 	}
1876 
1877 	ubbest = malloc(sizeof(*ubbest));
1878 	if (!ubbest) {
1879 		zfs_unmount(data);
1880 		return 0;
1881 	}
1882 	memset(ubbest, 0, sizeof(*ubbest));
1883 
1884 	/*
1885 	 * some eltorito stacks don't give us a size and
1886 	 * we end up setting the size to MAXUINT, further
1887 	 * some of these devices stop working once a single
1888 	 * read past the end has been issued. Checking
1889 	 * for a maximum part_length and skipping the backup
1890 	 * labels at the end of the slice/partition/device
1891 	 * avoids breaking down on such devices.
1892 	 */
1893 	const int vdevnum =
1894 		dev->part_length == 0 ?
1895 		VDEV_LABELS / 2 : VDEV_LABELS;
1896 
1897 	/* Size in bytes of the device (disk or partition) aligned to label size*/
1898 	uint64_t device_size =
1899 		dev->part_length << SECTOR_BITS;
1900 
1901 	const uint64_t alignedbytes =
1902 		P2ALIGN(device_size, (uint64_t) sizeof(vdev_label_t));
1903 
1904 	for (label = 0; label < vdevnum; label++) {
1905 		uint64_t labelstartbytes = vdev_label_start(alignedbytes, label);
1906 		uint64_t labelstart = labelstartbytes >> SECTOR_BITS;
1907 
1908 		debug("zfs reading label %d at sector %llu (byte %llu)\n",
1909 			  label, (unsigned long long) labelstart,
1910 			  (unsigned long long) labelstartbytes);
1911 
1912 		data->vdev_phys_sector = labelstart +
1913 			((VDEV_SKIP_SIZE + VDEV_BOOT_HEADER_SIZE) >> SECTOR_BITS);
1914 
1915 		err = check_pool_label(data);
1916 		if (err) {
1917 			printf("zfs error checking label %d\n", label);
1918 			continue;
1919 		}
1920 
1921 		/* Read in the uberblock ring (128K). */
1922 		err = zfs_devread(data->vdev_phys_sector  +
1923 						  (VDEV_PHYS_SIZE >> SECTOR_BITS),
1924 						  0, VDEV_UBERBLOCK_RING, ub_array);
1925 		if (err) {
1926 			printf("zfs error reading uberblock ring for label %d\n", label);
1927 			continue;
1928 		}
1929 
1930 		ubcur = find_bestub(ub_array, data);
1931 		if (!ubcur) {
1932 			printf("zfs No good uberblocks found in label %d\n", label);
1933 			continue;
1934 		}
1935 
1936 		if (vdev_uberblock_compare(ubcur, ubbest) > 0) {
1937 			/* Looks like the block is good, so use it.*/
1938 			memcpy(ubbest, ubcur, sizeof(*ubbest));
1939 			bestlabel = label;
1940 			debug("zfs Current best uberblock found in label %d\n", label);
1941 		}
1942 	}
1943 	free(ub_array);
1944 
1945 	/* We zero'd the structure to begin with.  If we never assigned to it,
1946 	   magic will still be zero. */
1947 	if (!ubbest->ub_magic) {
1948 		printf("couldn't find a valid ZFS label\n");
1949 		zfs_unmount(data);
1950 		free(ubbest);
1951 		return 0;
1952 	}
1953 
1954 	debug("zfs ubbest %p in label %d\n", ubbest, bestlabel);
1955 
1956 	zfs_endian_t ub_endian =
1957 		zfs_to_cpu64(ubbest->ub_magic, LITTLE_ENDIAN) == UBERBLOCK_MAGIC
1958 		? LITTLE_ENDIAN : BIG_ENDIAN;
1959 
1960 	debug("zfs endian set to %s\n", !ub_endian ? "big" : "little");
1961 
1962 	err = zio_read(&ubbest->ub_rootbp, ub_endian, &osp, &ospsize, data);
1963 
1964 	if (err) {
1965 		printf("couldn't zio_read object directory\n");
1966 		zfs_unmount(data);
1967 		free(ubbest);
1968 		return 0;
1969 	}
1970 
1971 	if (ospsize < OBJSET_PHYS_SIZE_V14) {
1972 		printf("osp too small\n");
1973 		zfs_unmount(data);
1974 		free(osp);
1975 		free(ubbest);
1976 		return 0;
1977 	}
1978 
1979 	/* Got the MOS. Save it at the memory addr MOS. */
1980 	memmove(&(data->mos.dn), &((objset_phys_t *) osp)->os_meta_dnode, DNODE_SIZE);
1981 	data->mos.endian =
1982 		(zfs_to_cpu64(ubbest->ub_rootbp.blk_prop, ub_endian) >> 63) & 1;
1983 	memmove(&(data->current_uberblock), ubbest, sizeof(uberblock_t));
1984 
1985 	free(osp);
1986 	free(ubbest);
1987 
1988 	return data;
1989 }
1990 
1991 int
1992 zfs_fetch_nvlist(device_t dev, char **nvlist)
1993 {
1994 	struct zfs_data *zfs;
1995 	int err;
1996 
1997 	zfs = zfs_mount(dev);
1998 	if (!zfs)
1999 		return ZFS_ERR_BAD_FS;
2000 	err = int_zfs_fetch_nvlist(zfs, nvlist);
2001 	zfs_unmount(zfs);
2002 	return err;
2003 }
2004 
2005 static int
2006 zfs_label(device_t device, char **label)
2007 {
2008 	char *nvlist;
2009 	int err;
2010 	struct zfs_data *data;
2011 
2012 	data = zfs_mount(device);
2013 	if (!data)
2014 		return ZFS_ERR_BAD_FS;
2015 
2016 	err = int_zfs_fetch_nvlist(data, &nvlist);
2017 	if (err) {
2018 		zfs_unmount(data);
2019 		return err;
2020 	}
2021 
2022 	*label = zfs_nvlist_lookup_string(nvlist, ZPOOL_CONFIG_POOL_NAME);
2023 	free(nvlist);
2024 	zfs_unmount(data);
2025 	return ZFS_ERR_NONE;
2026 }
2027 
2028 static int
2029 zfs_uuid(device_t device, char **uuid)
2030 {
2031 	struct zfs_data *data;
2032 
2033 	data = zfs_mount(device);
2034 	if (!data)
2035 		return ZFS_ERR_BAD_FS;
2036 
2037 	*uuid = malloc(17); /* %016llx + nil */
2038 	if (!*uuid)
2039 		return ZFS_ERR_OUT_OF_MEMORY;
2040 
2041 	/* *uuid = xasprintf ("%016llx", (long long unsigned) data->pool_guid);*/
2042 	snprintf(*uuid, 17, "%016llx", (long long unsigned) data->pool_guid);
2043 	zfs_unmount(data);
2044 
2045 	return ZFS_ERR_NONE;
2046 }
2047 
2048 /*
2049  * zfs_open() locates a file in the rootpool by following the
2050  * MOS and places the dnode of the file in the memory address DNODE.
2051  */
2052 int
2053 zfs_open(struct zfs_file *file, const char *fsfilename)
2054 {
2055 	struct zfs_data *data;
2056 	int err;
2057 	int isfs;
2058 
2059 	data = zfs_mount(file->device);
2060 	if (!data)
2061 		return ZFS_ERR_BAD_FS;
2062 
2063 	err = dnode_get_fullpath(fsfilename, &(data->mdn), 0,
2064 							 &(data->dnode), &isfs, data);
2065 	if (err) {
2066 		zfs_unmount(data);
2067 		return err;
2068 	}
2069 
2070 	if (isfs) {
2071 		zfs_unmount(data);
2072 		printf("Missing @ or / separator\n");
2073 		return ZFS_ERR_FILE_NOT_FOUND;
2074 	}
2075 
2076 	/* We found the dnode for this file. Verify if it is a plain file. */
2077 	if (data->dnode.dn.dn_type != DMU_OT_PLAIN_FILE_CONTENTS) {
2078 		zfs_unmount(data);
2079 		printf("not a file\n");
2080 		return ZFS_ERR_BAD_FILE_TYPE;
2081 	}
2082 
2083 	/* get the file size and set the file position to 0 */
2084 
2085 	/*
2086 	 * For DMU_OT_SA we will need to locate the SIZE attribute
2087 	 * attribute, which could be either in the bonus buffer
2088 	 * or the "spill" block.
2089 	 */
2090 	if (data->dnode.dn.dn_bonustype == DMU_OT_SA) {
2091 		void *sahdrp;
2092 		int hdrsize;
2093 
2094 		if (data->dnode.dn.dn_bonuslen != 0) {
2095 			sahdrp = (sa_hdr_phys_t *) DN_BONUS(&data->dnode.dn);
2096 		} else if (data->dnode.dn.dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
2097 			blkptr_t *bp = &data->dnode.dn.dn_spill;
2098 
2099 			err = zio_read(bp, data->dnode.endian, &sahdrp, NULL, data);
2100 			if (err)
2101 				return err;
2102 		} else {
2103 			printf("filesystem is corrupt :(\n");
2104 			return ZFS_ERR_BAD_FS;
2105 		}
2106 
2107 		hdrsize = SA_HDR_SIZE(((sa_hdr_phys_t *) sahdrp));
2108 		file->size = *(uint64_t *) ((char *) sahdrp + hdrsize + SA_SIZE_OFFSET);
2109 	} else {
2110 		file->size = zfs_to_cpu64(((znode_phys_t *) DN_BONUS(&data->dnode.dn))->zp_size, data->dnode.endian);
2111 	}
2112 
2113 	file->data = data;
2114 	file->offset = 0;
2115 
2116 	return ZFS_ERR_NONE;
2117 }
2118 
2119 uint64_t
2120 zfs_read(zfs_file_t file, char *buf, uint64_t len)
2121 {
2122 	struct zfs_data *data = (struct zfs_data *) file->data;
2123 	int blksz, movesize;
2124 	uint64_t length;
2125 	int64_t red;
2126 	int err;
2127 
2128 	if (data->file_buf == NULL) {
2129 		data->file_buf = malloc(SPA_MAXBLOCKSIZE);
2130 		if (!data->file_buf)
2131 			return -1;
2132 		data->file_start = data->file_end = 0;
2133 	}
2134 
2135 	/*
2136 	 * If offset is in memory, move it into the buffer provided and return.
2137 	 */
2138 	if (file->offset >= data->file_start
2139 		&& file->offset + len <= data->file_end) {
2140 		memmove(buf, data->file_buf + file->offset - data->file_start,
2141 				len);
2142 		return len;
2143 	}
2144 
2145 	blksz = zfs_to_cpu16(data->dnode.dn.dn_datablkszsec,
2146 							  data->dnode.endian) << SPA_MINBLOCKSHIFT;
2147 
2148 	/*
2149 	 * Entire Dnode is too big to fit into the space available.	 We
2150 	 * will need to read it in chunks.	This could be optimized to
2151 	 * read in as large a chunk as there is space available, but for
2152 	 * now, this only reads in one data block at a time.
2153 	 */
2154 	length = len;
2155 	red = 0;
2156 	while (length) {
2157 		void *t;
2158 		/*
2159 		 * Find requested blkid and the offset within that block.
2160 		 */
2161 		uint64_t blkid = (file->offset + red) /	 blksz;
2162 		free(data->file_buf);
2163 		data->file_buf = 0;
2164 
2165 		err = dmu_read(&(data->dnode), blkid, &t,
2166 					   0, data);
2167 		data->file_buf = t;
2168 		if (err)
2169 			return -1;
2170 
2171 		data->file_start = blkid * blksz;
2172 		data->file_end = data->file_start + blksz;
2173 
2174 		movesize = MIN(length, data->file_end - (int) file->offset - red);
2175 
2176 		memmove(buf, data->file_buf + file->offset + red
2177 				- data->file_start, movesize);
2178 		buf += movesize;
2179 		length -= movesize;
2180 		red += movesize;
2181 	}
2182 
2183 	return len;
2184 }
2185 
2186 int
2187 zfs_close(zfs_file_t file)
2188 {
2189 	zfs_unmount((struct zfs_data *) file->data);
2190 	return ZFS_ERR_NONE;
2191 }
2192 
2193 int
2194 zfs_getmdnobj(device_t dev, const char *fsfilename,
2195 				   uint64_t *mdnobj)
2196 {
2197 	struct zfs_data *data;
2198 	int err;
2199 	int isfs;
2200 
2201 	data = zfs_mount(dev);
2202 	if (!data)
2203 		return ZFS_ERR_BAD_FS;
2204 
2205 	err = dnode_get_fullpath(fsfilename, &(data->mdn), mdnobj,
2206 							 &(data->dnode), &isfs, data);
2207 	zfs_unmount(data);
2208 	return err;
2209 }
2210 
2211 static void
2212 fill_fs_info(struct zfs_dirhook_info *info,
2213 			 dnode_end_t mdn, struct zfs_data *data)
2214 {
2215 	int err;
2216 	dnode_end_t dn;
2217 	uint64_t objnum;
2218 	uint64_t headobj;
2219 
2220 	memset(info, 0, sizeof(*info));
2221 
2222 	info->dir = 1;
2223 
2224 	if (mdn.dn.dn_type == DMU_OT_DSL_DIR) {
2225 		headobj = zfs_to_cpu64(((dsl_dir_phys_t *) DN_BONUS(&mdn.dn))->dd_head_dataset_obj, mdn.endian);
2226 
2227 		err = dnode_get(&(data->mos), headobj, DMU_OT_DSL_DATASET, &mdn, data);
2228 		if (err) {
2229 			printf("zfs failed here 1\n");
2230 			return;
2231 		}
2232 	}
2233 	make_mdn(&mdn, data);
2234 	err = dnode_get(&mdn, MASTER_NODE_OBJ, DMU_OT_MASTER_NODE,
2235 					&dn, data);
2236 	if (err) {
2237 		printf("zfs failed here 2\n");
2238 		return;
2239 	}
2240 
2241 	err = zap_lookup(&dn, ZFS_ROOT_OBJ, &objnum, data);
2242 	if (err) {
2243 		printf("zfs failed here 3\n");
2244 		return;
2245 	}
2246 
2247 	err = dnode_get(&mdn, objnum, 0, &dn, data);
2248 	if (err) {
2249 		printf("zfs failed here 4\n");
2250 		return;
2251 	}
2252 
2253 	info->mtimeset = 1;
2254 	info->mtime = zfs_to_cpu64(((znode_phys_t *) DN_BONUS(&dn.dn))->zp_mtime[0], dn.endian);
2255 
2256 	return;
2257 }
2258 
2259 static int iterate_zap(const char *name, uint64_t val, struct zfs_data *data)
2260 {
2261 	struct zfs_dirhook_info info;
2262 	dnode_end_t dn;
2263 
2264 	memset(&info, 0, sizeof(info));
2265 
2266 	dnode_get(&(data->mdn), val, 0, &dn, data);
2267 	info.mtimeset = 1;
2268 	info.mtime = zfs_to_cpu64(((znode_phys_t *) DN_BONUS(&dn.dn))->zp_mtime[0], dn.endian);
2269 	info.dir = (dn.dn.dn_type == DMU_OT_DIRECTORY_CONTENTS);
2270 	debug("zfs type=%d, name=%s\n",
2271 		  (int)dn.dn.dn_type, (char *)name);
2272 	if (!data->userhook)
2273 		return 0;
2274 	return data->userhook(name, &info);
2275 }
2276 
2277 static int iterate_zap_fs(const char *name, uint64_t val, struct zfs_data *data)
2278 {
2279 	struct zfs_dirhook_info info;
2280 	dnode_end_t mdn;
2281 	int err;
2282 	err = dnode_get(&(data->mos), val, 0, &mdn, data);
2283 	if (err)
2284 		return 0;
2285 	if (mdn.dn.dn_type != DMU_OT_DSL_DIR)
2286 		return 0;
2287 
2288 	fill_fs_info(&info, mdn, data);
2289 
2290 	if (!data->userhook)
2291 		return 0;
2292 	return data->userhook(name, &info);
2293 }
2294 
2295 static int iterate_zap_snap(const char *name, uint64_t val, struct zfs_data *data)
2296 {
2297 	struct zfs_dirhook_info info;
2298 	char *name2;
2299 	int ret = 0;
2300 	dnode_end_t mdn;
2301 	int err;
2302 
2303 	err = dnode_get(&(data->mos), val, 0, &mdn, data);
2304 	if (err)
2305 		return 0;
2306 
2307 	if (mdn.dn.dn_type != DMU_OT_DSL_DATASET)
2308 		return 0;
2309 
2310 	fill_fs_info(&info, mdn, data);
2311 
2312 	name2 = malloc(strlen(name) + 2);
2313 	name2[0] = '@';
2314 	memcpy(name2 + 1, name, strlen(name) + 1);
2315 	if (data->userhook)
2316 		ret = data->userhook(name2, &info);
2317 	free(name2);
2318 	return ret;
2319 }
2320 
2321 int
2322 zfs_ls(device_t device, const char *path,
2323 	   int (*hook)(const char *, const struct zfs_dirhook_info *))
2324 {
2325 	struct zfs_data *data;
2326 	int err;
2327 	int isfs;
2328 #if 0
2329 	char *label = NULL;
2330 
2331 	zfs_label(device, &label);
2332 	if (label)
2333 		printf("ZPOOL label '%s'\n",
2334 			   label);
2335 #endif
2336 
2337 	data = zfs_mount(device);
2338 	if (!data)
2339 		return ZFS_ERR_BAD_FS;
2340 
2341 	data->userhook = hook;
2342 
2343 	err = dnode_get_fullpath(path, &(data->mdn), 0, &(data->dnode), &isfs, data);
2344 	if (err) {
2345 		zfs_unmount(data);
2346 		return err;
2347 	}
2348 	if (isfs) {
2349 		uint64_t childobj, headobj;
2350 		uint64_t snapobj;
2351 		dnode_end_t dn;
2352 		struct zfs_dirhook_info info;
2353 
2354 		fill_fs_info(&info, data->dnode, data);
2355 		hook("@", &info);
2356 
2357 		childobj = zfs_to_cpu64(((dsl_dir_phys_t *) DN_BONUS(&data->dnode.dn))->dd_child_dir_zapobj, data->dnode.endian);
2358 		headobj = zfs_to_cpu64(((dsl_dir_phys_t *) DN_BONUS(&data->dnode.dn))->dd_head_dataset_obj, data->dnode.endian);
2359 		err = dnode_get(&(data->mos), childobj,
2360 						DMU_OT_DSL_DIR_CHILD_MAP, &dn, data);
2361 		if (err) {
2362 			zfs_unmount(data);
2363 			return err;
2364 		}
2365 
2366 
2367 		zap_iterate(&dn, iterate_zap_fs, data);
2368 
2369 		err = dnode_get(&(data->mos), headobj, DMU_OT_DSL_DATASET, &dn, data);
2370 		if (err) {
2371 			zfs_unmount(data);
2372 			return err;
2373 		}
2374 
2375 		snapobj = zfs_to_cpu64(((dsl_dataset_phys_t *) DN_BONUS(&dn.dn))->ds_snapnames_zapobj, dn.endian);
2376 
2377 		err = dnode_get(&(data->mos), snapobj,
2378 						DMU_OT_DSL_DS_SNAP_MAP, &dn, data);
2379 		if (err) {
2380 			zfs_unmount(data);
2381 			return err;
2382 		}
2383 
2384 		zap_iterate(&dn, iterate_zap_snap, data);
2385 	} else {
2386 		if (data->dnode.dn.dn_type != DMU_OT_DIRECTORY_CONTENTS) {
2387 			zfs_unmount(data);
2388 			printf("not a directory\n");
2389 			return ZFS_ERR_BAD_FILE_TYPE;
2390 		}
2391 		zap_iterate(&(data->dnode), iterate_zap, data);
2392 	}
2393 	zfs_unmount(data);
2394 	return ZFS_ERR_NONE;
2395 }
2396