xref: /titanic_51/usr/src/grub/grub-0.97/stage2/fsys_zfs.c (revision c3d26abc9ee97b4f60233556aadeb57e0bd30bb9)
1 /*
2  *  GRUB  --  GRand Unified Bootloader
3  *  Copyright (C) 1999,2000,2001,2002,2003,2004  Free Software Foundation, Inc.
4  *
5  *  This program is free software; you can redistribute it and/or modify
6  *  it under the terms of the GNU General Public License as published by
7  *  the Free Software Foundation; either version 2 of the License, or
8  *  (at your option) any later version.
9  *
10  *  This program is distributed in the hope that it will be useful,
11  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
12  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13  *  GNU General Public License for more details.
14  *
15  *  You should have received a copy of the GNU General Public License
16  *  along with this program; if not, write to the Free Software
17  *  Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
18  */
19 
20 /*
21  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
22  * Use is subject to license terms.
23  */
24 
25 /*
26  * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
27  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
28  * Copyright (c) 2014 Integros [integros.com]
29  */
30 
31 /*
32  * The zfs plug-in routines for GRUB are:
33  *
34  * zfs_mount() - locates a valid uberblock of the root pool and reads
35  *		in its MOS at the memory address MOS.
36  *
37  * zfs_open() - locates a plain file object by following the MOS
38  *		and places its dnode at the memory address DNODE.
39  *
40  * zfs_read() - read in the data blocks pointed by the DNODE.
41  *
42  * ZFS_SCRATCH is used as a working area.
43  *
44  * (memory addr)   MOS      DNODE	ZFS_SCRATCH
45  *		    |         |          |
46  *	    +-------V---------V----------V---------------+
47  *   memory |       | dnode   | dnode    |  scratch      |
48  *	    |       | 512B    | 512B     |  area         |
49  *	    +--------------------------------------------+
50  */
51 
52 #ifdef	FSYS_ZFS
53 
54 #include "shared.h"
55 #include "filesys.h"
56 #include "fsys_zfs.h"
57 
58 /* cache for a file block of the currently zfs_open()-ed file */
59 static void *file_buf = NULL;
60 static uint64_t file_start = 0;
61 static uint64_t file_end = 0;
62 
63 /* cache for a dnode block */
64 static dnode_phys_t *dnode_buf = NULL;
65 static dnode_phys_t *dnode_mdn = NULL;
66 static uint64_t dnode_start = 0;
67 static uint64_t dnode_end = 0;
68 
69 static uint64_t pool_guid = 0;
70 static uberblock_t current_uberblock;
71 static char *stackbase;
72 
73 decomp_entry_t decomp_table[ZIO_COMPRESS_FUNCTIONS] =
74 {
75 	{"inherit", 0},			/* ZIO_COMPRESS_INHERIT */
76 	{"on", lzjb_decompress}, 	/* ZIO_COMPRESS_ON */
77 	{"off", 0},			/* ZIO_COMPRESS_OFF */
78 	{"lzjb", lzjb_decompress},	/* ZIO_COMPRESS_LZJB */
79 	{"empty", 0},			/* ZIO_COMPRESS_EMPTY */
80 	{"gzip-1", 0},			/* ZIO_COMPRESS_GZIP_1 */
81 	{"gzip-2", 0},			/* ZIO_COMPRESS_GZIP_2 */
82 	{"gzip-3", 0},			/* ZIO_COMPRESS_GZIP_3 */
83 	{"gzip-4", 0},			/* ZIO_COMPRESS_GZIP_4 */
84 	{"gzip-5", 0},			/* ZIO_COMPRESS_GZIP_5 */
85 	{"gzip-6", 0},			/* ZIO_COMPRESS_GZIP_6 */
86 	{"gzip-7", 0},			/* ZIO_COMPRESS_GZIP_7 */
87 	{"gzip-8", 0},			/* ZIO_COMPRESS_GZIP_8 */
88 	{"gzip-9", 0},			/* ZIO_COMPRESS_GZIP_9 */
89 	{"zle", 0},			/* ZIO_COMPRESS_ZLE */
90 	{"lz4", lz4_decompress}		/* ZIO_COMPRESS_LZ4 */
91 };
92 
93 static int zio_read_data(blkptr_t *bp, void *buf, char *stack);
94 
95 /*
96  * Our own version of bcmp().
97  */
98 static int
99 zfs_bcmp(const void *s1, const void *s2, size_t n)
100 {
101 	const uchar_t *ps1 = s1;
102 	const uchar_t *ps2 = s2;
103 
104 	if (s1 != s2 && n != 0) {
105 		do {
106 			if (*ps1++ != *ps2++)
107 				return (1);
108 		} while (--n != 0);
109 	}
110 
111 	return (0);
112 }
113 
114 /*
115  * Our own version of log2().  Same thing as highbit()-1.
116  */
117 static int
118 zfs_log2(uint64_t num)
119 {
120 	int i = 0;
121 
122 	while (num > 1) {
123 		i++;
124 		num = num >> 1;
125 	}
126 
127 	return (i);
128 }
129 
130 /* Checksum Functions */
131 static void
132 zio_checksum_off(const void *buf, uint64_t size, zio_cksum_t *zcp)
133 {
134 	ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
135 }
136 
137 /* Checksum Table and Values */
138 zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = {
139 	{{NULL,			NULL},			0, 0,	"inherit"},
140 	{{NULL,			NULL},			0, 0,	"on"},
141 	{{zio_checksum_off,	zio_checksum_off},	0, 0,	"off"},
142 	{{zio_checksum_SHA256,	zio_checksum_SHA256},	1, 1,	"label"},
143 	{{zio_checksum_SHA256,	zio_checksum_SHA256},	1, 1,	"gang_header"},
144 	{{NULL,			NULL},			0, 0,	"zilog"},
145 	{{fletcher_2_native,	fletcher_2_byteswap},	0, 0,	"fletcher2"},
146 	{{fletcher_4_native,	fletcher_4_byteswap},	1, 0,	"fletcher4"},
147 	{{zio_checksum_SHA256,	zio_checksum_SHA256},	1, 0,	"SHA256"},
148 	{{NULL,			NULL},			0, 0,	"zilog2"},
149 	{{zio_checksum_off,	zio_checksum_off},	0, 0,	"noparity"},
150 	{{zio_checksum_SHA512,	NULL},			0, 0,	"SHA512"}
151 };
152 
153 /*
154  * zio_checksum_verify: Provides support for checksum verification.
155  *
156  * Fletcher2, Fletcher4, SHA-256 and SHA-512/256 are supported.
157  *
158  * Return:
159  * 	-1 = Failure
160  *	 0 = Success
161  */
162 static int
163 zio_checksum_verify(blkptr_t *bp, char *data, int size)
164 {
165 	zio_cksum_t zc = bp->blk_cksum;
166 	uint32_t checksum = BP_GET_CHECKSUM(bp);
167 	int byteswap = BP_SHOULD_BYTESWAP(bp);
168 	zio_eck_t *zec = (zio_eck_t *)(data + size) - 1;
169 	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
170 	zio_cksum_t actual_cksum, expected_cksum;
171 
172 	if (byteswap) {
173 		grub_printf("byteswap not supported\n");
174 		return (-1);
175 	}
176 
177 	if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL) {
178 		grub_printf("checksum algorithm %u not supported\n", checksum);
179 		return (-1);
180 	}
181 
182 	if (ci->ci_eck) {
183 		expected_cksum = zec->zec_cksum;
184 		zec->zec_cksum = zc;
185 		ci->ci_func[0](data, size, &actual_cksum);
186 		zec->zec_cksum = expected_cksum;
187 		zc = expected_cksum;
188 	} else {
189 		ci->ci_func[byteswap](data, size, &actual_cksum);
190 	}
191 
192 	if ((actual_cksum.zc_word[0] - zc.zc_word[0]) |
193 	    (actual_cksum.zc_word[1] - zc.zc_word[1]) |
194 	    (actual_cksum.zc_word[2] - zc.zc_word[2]) |
195 	    (actual_cksum.zc_word[3] - zc.zc_word[3]))
196 		return (-1);
197 
198 	return (0);
199 }
200 
201 /*
202  * vdev_label_start returns the physical disk offset (in bytes) of
203  * label "l".
204  */
205 static uint64_t
206 vdev_label_start(uint64_t psize, int l)
207 {
208 	return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
209 	    0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
210 }
211 
212 /*
213  * vdev_uberblock_compare takes two uberblock structures and returns an integer
214  * indicating the more recent of the two.
215  * 	Return Value = 1 if ub2 is more recent
216  * 	Return Value = -1 if ub1 is more recent
217  * The most recent uberblock is determined using its transaction number and
218  * timestamp.  The uberblock with the highest transaction number is
219  * considered "newer".  If the transaction numbers of the two blocks match, the
220  * timestamps are compared to determine the "newer" of the two.
221  */
222 static int
223 vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
224 {
225 	if (ub1->ub_txg < ub2->ub_txg)
226 		return (-1);
227 	if (ub1->ub_txg > ub2->ub_txg)
228 		return (1);
229 
230 	if (ub1->ub_timestamp < ub2->ub_timestamp)
231 		return (-1);
232 	if (ub1->ub_timestamp > ub2->ub_timestamp)
233 		return (1);
234 
235 	return (0);
236 }
237 
238 /*
239  * Three pieces of information are needed to verify an uberblock: the magic
240  * number, the version number, and the checksum.
241  *
242  * Return:
243  *     0 - Success
244  *    -1 - Failure
245  */
246 static int
247 uberblock_verify(uberblock_t *uber, uint64_t ub_size, uint64_t offset)
248 {
249 	blkptr_t bp;
250 
251 	BP_ZERO(&bp);
252 	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
253 	BP_SET_BYTEORDER(&bp, ZFS_HOST_BYTEORDER);
254 	ZIO_SET_CHECKSUM(&bp.blk_cksum, offset, 0, 0, 0);
255 
256 	if (zio_checksum_verify(&bp, (char *)uber, ub_size) != 0)
257 		return (-1);
258 
259 	if (uber->ub_magic == UBERBLOCK_MAGIC &&
260 	    SPA_VERSION_IS_SUPPORTED(uber->ub_version))
261 		return (0);
262 
263 	return (-1);
264 }
265 
266 /*
267  * Find the best uberblock.
268  * Return:
269  *    Success - Pointer to the best uberblock.
270  *    Failure - NULL
271  */
272 static uberblock_t *
273 find_bestub(char *ub_array, uint64_t ashift, uint64_t sector)
274 {
275 	uberblock_t *ubbest = NULL;
276 	uberblock_t *ubnext;
277 	uint64_t offset, ub_size;
278 	int i;
279 
280 	ub_size = VDEV_UBERBLOCK_SIZE(ashift);
281 
282 	for (i = 0; i < VDEV_UBERBLOCK_COUNT(ashift); i++) {
283 		ubnext = (uberblock_t *)ub_array;
284 		ub_array += ub_size;
285 		offset = (sector << SPA_MINBLOCKSHIFT) +
286 		    VDEV_UBERBLOCK_OFFSET(ashift, i);
287 
288 		if (uberblock_verify(ubnext, ub_size, offset) != 0)
289 			continue;
290 
291 		if (ubbest == NULL ||
292 		    vdev_uberblock_compare(ubnext, ubbest) > 0)
293 			ubbest = ubnext;
294 	}
295 
296 	return (ubbest);
297 }
298 
299 /*
300  * Read a block of data based on the gang block address dva,
301  * and put its data in buf.
302  *
303  * Return:
304  *	0 - success
305  *	1 - failure
306  */
307 static int
308 zio_read_gang(blkptr_t *bp, dva_t *dva, void *buf, char *stack)
309 {
310 	zio_gbh_phys_t *zio_gb;
311 	uint64_t offset, sector;
312 	blkptr_t tmpbp;
313 	int i;
314 
315 	zio_gb = (zio_gbh_phys_t *)stack;
316 	stack += SPA_GANGBLOCKSIZE;
317 	offset = DVA_GET_OFFSET(dva);
318 	sector = DVA_OFFSET_TO_PHYS_SECTOR(offset);
319 
320 	/* read in the gang block header */
321 	if (devread(sector, 0, SPA_GANGBLOCKSIZE, (char *)zio_gb) == 0) {
322 		grub_printf("failed to read in a gang block header\n");
323 		return (1);
324 	}
325 
326 	/* self checksuming the gang block header */
327 	BP_ZERO(&tmpbp);
328 	BP_SET_CHECKSUM(&tmpbp, ZIO_CHECKSUM_GANG_HEADER);
329 	BP_SET_BYTEORDER(&tmpbp, ZFS_HOST_BYTEORDER);
330 	ZIO_SET_CHECKSUM(&tmpbp.blk_cksum, DVA_GET_VDEV(dva),
331 	    DVA_GET_OFFSET(dva), bp->blk_birth, 0);
332 	if (zio_checksum_verify(&tmpbp, (char *)zio_gb, SPA_GANGBLOCKSIZE)) {
333 		grub_printf("failed to checksum a gang block header\n");
334 		return (1);
335 	}
336 
337 	for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
338 		if (BP_IS_HOLE(&zio_gb->zg_blkptr[i]))
339 			continue;
340 
341 		if (zio_read_data(&zio_gb->zg_blkptr[i], buf, stack))
342 			return (1);
343 		buf += BP_GET_PSIZE(&zio_gb->zg_blkptr[i]);
344 	}
345 
346 	return (0);
347 }
348 
349 /*
350  * Read in a block of raw data to buf.
351  *
352  * Return:
353  *	0 - success
354  *	1 - failure
355  */
356 static int
357 zio_read_data(blkptr_t *bp, void *buf, char *stack)
358 {
359 	int i, psize;
360 
361 	psize = BP_GET_PSIZE(bp);
362 
363 	/* pick a good dva from the block pointer */
364 	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
365 		uint64_t offset, sector;
366 
367 		if (bp->blk_dva[i].dva_word[0] == 0 &&
368 		    bp->blk_dva[i].dva_word[1] == 0)
369 			continue;
370 
371 		if (DVA_GET_GANG(&bp->blk_dva[i])) {
372 			if (zio_read_gang(bp, &bp->blk_dva[i], buf, stack) != 0)
373 				continue;
374 		} else {
375 			/* read in a data block */
376 			offset = DVA_GET_OFFSET(&bp->blk_dva[i]);
377 			sector = DVA_OFFSET_TO_PHYS_SECTOR(offset);
378 			if (devread(sector, 0, psize, buf) == 0)
379 				continue;
380 		}
381 
382 		/* verify that the checksum matches */
383 		if (zio_checksum_verify(bp, buf, psize) == 0) {
384 			return (0);
385 		}
386 	}
387 
388 	grub_printf("could not read block due to EIO or ECKSUM\n");
389 	return (1);
390 }
391 
392 /*
393  * buf must be at least BPE_GET_PSIZE(bp) bytes long (which will never be
394  * more than BPE_PAYLOAD_SIZE bytes).
395  */
396 static void
397 decode_embedded_bp_compressed(const blkptr_t *bp, void *buf)
398 {
399 	int psize, i;
400 	uint8_t *buf8 = buf;
401 	uint64_t w = 0;
402 	const uint64_t *bp64 = (const uint64_t *)bp;
403 
404 	psize = BPE_GET_PSIZE(bp);
405 
406 	/*
407 	 * Decode the words of the block pointer into the byte array.
408 	 * Low bits of first word are the first byte (little endian).
409 	 */
410 	for (i = 0; i < psize; i++) {
411 		if (i % sizeof (w) == 0) {
412 			/* beginning of a word */
413 			w = *bp64;
414 			bp64++;
415 			if (!BPE_IS_PAYLOADWORD(bp, bp64))
416 				bp64++;
417 		}
418 		buf8[i] = BF64_GET(w, (i % sizeof (w)) * NBBY, NBBY);
419 	}
420 }
421 
422 /*
423  * Fill in the buffer with the (decompressed) payload of the embedded
424  * blkptr_t.  Takes into account compression and byteorder (the payload is
425  * treated as a stream of bytes).
426  * Return 0 on success, or ENOSPC if it won't fit in the buffer.
427  */
428 static int
429 decode_embedded_bp(const blkptr_t *bp, void *buf)
430 {
431 	int comp;
432 	int lsize, psize;
433 	uint8_t *dst = buf;
434 	uint64_t w = 0;
435 
436 	lsize = BPE_GET_LSIZE(bp);
437 	psize = BPE_GET_PSIZE(bp);
438 	comp = BP_GET_COMPRESS(bp);
439 
440 	if (comp != ZIO_COMPRESS_OFF) {
441 		uint8_t dstbuf[BPE_PAYLOAD_SIZE];
442 
443 		if ((unsigned int)comp >= ZIO_COMPRESS_FUNCTIONS ||
444 		    decomp_table[comp].decomp_func == NULL) {
445 			grub_printf("compression algorithm not supported\n");
446 			return (ERR_FSYS_CORRUPT);
447 		}
448 
449 		decode_embedded_bp_compressed(bp, dstbuf);
450 		decomp_table[comp].decomp_func(dstbuf, buf, psize, lsize);
451 	} else {
452 		decode_embedded_bp_compressed(bp, buf);
453 	}
454 
455 	return (0);
456 }
457 
458 /*
459  * Read in a block of data, verify its checksum, decompress if needed,
460  * and put the uncompressed data in buf.
461  *
462  * Return:
463  *	0 - success
464  *	errnum - failure
465  */
466 static int
467 zio_read(blkptr_t *bp, void *buf, char *stack)
468 {
469 	int lsize, psize, comp;
470 	char *retbuf;
471 
472 	if (BP_IS_EMBEDDED(bp)) {
473 		if (BPE_GET_ETYPE(bp) != BP_EMBEDDED_TYPE_DATA) {
474 			grub_printf("unsupported embedded BP (type=%u)\n",
475 			    (int)BPE_GET_ETYPE(bp));
476 			return (ERR_FSYS_CORRUPT);
477 		}
478 		return (decode_embedded_bp(bp, buf));
479 	}
480 
481 	comp = BP_GET_COMPRESS(bp);
482 	lsize = BP_GET_LSIZE(bp);
483 	psize = BP_GET_PSIZE(bp);
484 
485 	if ((unsigned int)comp >= ZIO_COMPRESS_FUNCTIONS ||
486 	    (comp != ZIO_COMPRESS_OFF &&
487 	    decomp_table[comp].decomp_func == NULL)) {
488 		grub_printf("compression algorithm not supported\n");
489 		return (ERR_FSYS_CORRUPT);
490 	}
491 
492 	if ((char *)buf < stack && ((char *)buf) + lsize > stack) {
493 		grub_printf("not enough memory to fit %u bytes on stack\n",
494 		    lsize);
495 		return (ERR_WONT_FIT);
496 	}
497 
498 	retbuf = buf;
499 	if (comp != ZIO_COMPRESS_OFF) {
500 		buf = stack;
501 		stack += psize;
502 	}
503 
504 	if (zio_read_data(bp, buf, stack) != 0) {
505 		grub_printf("zio_read_data failed\n");
506 		return (ERR_FSYS_CORRUPT);
507 	}
508 
509 	if (comp != ZIO_COMPRESS_OFF) {
510 		if (decomp_table[comp].decomp_func(buf, retbuf, psize,
511 		    lsize) != 0) {
512 			grub_printf("zio_read decompression failed\n");
513 			return (ERR_FSYS_CORRUPT);
514 		}
515 	}
516 
517 	return (0);
518 }
519 
520 /*
521  * Get the block from a block id.
522  * push the block onto the stack.
523  *
524  * Return:
525  * 	0 - success
526  * 	errnum - failure
527  */
528 static int
529 dmu_read(dnode_phys_t *dn, uint64_t blkid, void *buf, char *stack)
530 {
531 	int idx, level;
532 	blkptr_t *bp_array = dn->dn_blkptr;
533 	int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
534 	blkptr_t *bp, *tmpbuf;
535 
536 	bp = (blkptr_t *)stack;
537 	stack += sizeof (blkptr_t);
538 
539 	tmpbuf = (blkptr_t *)stack;
540 	stack += 1<<dn->dn_indblkshift;
541 
542 	for (level = dn->dn_nlevels - 1; level >= 0; level--) {
543 		idx = (blkid >> (epbs * level)) & ((1<<epbs)-1);
544 		*bp = bp_array[idx];
545 		if (level == 0)
546 			tmpbuf = buf;
547 		if (BP_IS_HOLE(bp)) {
548 			grub_memset(buf, 0,
549 			    dn->dn_datablkszsec << SPA_MINBLOCKSHIFT);
550 			break;
551 		} else if (errnum = zio_read(bp, tmpbuf, stack)) {
552 			return (errnum);
553 		}
554 
555 		bp_array = tmpbuf;
556 	}
557 
558 	return (0);
559 }
560 
561 /*
562  * mzap_lookup: Looks up property described by "name" and returns the value
563  * in "value".
564  *
565  * Return:
566  *	0 - success
567  *	errnum - failure
568  */
569 static int
570 mzap_lookup(mzap_phys_t *zapobj, int objsize, const char *name,
571     uint64_t *value)
572 {
573 	int i, chunks;
574 	mzap_ent_phys_t *mzap_ent = zapobj->mz_chunk;
575 
576 	chunks = objsize / MZAP_ENT_LEN - 1;
577 	for (i = 0; i < chunks; i++) {
578 		if (grub_strcmp(mzap_ent[i].mze_name, name) == 0) {
579 			*value = mzap_ent[i].mze_value;
580 			return (0);
581 		}
582 	}
583 
584 	return (ERR_FSYS_CORRUPT);
585 }
586 
587 static uint64_t
588 zap_hash(uint64_t salt, const char *name)
589 {
590 	static uint64_t table[256];
591 	const uint8_t *cp;
592 	uint8_t c;
593 	uint64_t crc = salt;
594 
595 	if (table[128] == 0) {
596 		uint64_t *ct;
597 		int i, j;
598 		for (i = 0; i < 256; i++) {
599 			for (ct = table + i, *ct = i, j = 8; j > 0; j--)
600 				*ct = (*ct >> 1) ^ (-(*ct & 1) &
601 				    ZFS_CRC64_POLY);
602 		}
603 	}
604 
605 	if (crc == 0 || table[128] != ZFS_CRC64_POLY) {
606 		errnum = ERR_FSYS_CORRUPT;
607 		return (0);
608 	}
609 
610 	for (cp = (const uint8_t *)name; (c = *cp) != '\0'; cp++)
611 		crc = (crc >> 8) ^ table[(crc ^ c) & 0xFF];
612 
613 	/*
614 	 * Only use 28 bits, since we need 4 bits in the cookie for the
615 	 * collision differentiator.  We MUST use the high bits, since
616 	 * those are the ones that we first pay attention to when
617 	 * choosing the bucket.
618 	 */
619 	crc &= ~((1ULL << (64 - 28)) - 1);
620 
621 	return (crc);
622 }
623 
624 /*
625  * Only to be used on 8-bit arrays.
626  * array_len is actual len in bytes (not encoded le_value_length).
627  * buf is null-terminated.
628  */
629 static int
630 zap_leaf_array_equal(zap_leaf_phys_t *l, int blksft, int chunk,
631     int array_len, const char *buf)
632 {
633 	int bseen = 0;
634 
635 	while (bseen < array_len) {
636 		struct zap_leaf_array *la =
637 		    &ZAP_LEAF_CHUNK(l, blksft, chunk).l_array;
638 		int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES);
639 
640 		if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft))
641 			return (0);
642 
643 		if (zfs_bcmp(la->la_array, buf + bseen, toread) != 0)
644 			break;
645 		chunk = la->la_next;
646 		bseen += toread;
647 	}
648 	return (bseen == array_len);
649 }
650 
651 /*
652  * Given a zap_leaf_phys_t, walk thru the zap leaf chunks to get the
653  * value for the property "name".
654  *
655  * Return:
656  *	0 - success
657  *	errnum - failure
658  */
659 static int
660 zap_leaf_lookup(zap_leaf_phys_t *l, int blksft, uint64_t h,
661     const char *name, uint64_t *value)
662 {
663 	uint16_t chunk;
664 	struct zap_leaf_entry *le;
665 
666 	/* Verify if this is a valid leaf block */
667 	if (l->l_hdr.lh_block_type != ZBT_LEAF)
668 		return (ERR_FSYS_CORRUPT);
669 	if (l->l_hdr.lh_magic != ZAP_LEAF_MAGIC)
670 		return (ERR_FSYS_CORRUPT);
671 
672 	for (chunk = l->l_hash[LEAF_HASH(blksft, h)];
673 	    chunk != CHAIN_END; chunk = le->le_next) {
674 
675 		if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft))
676 			return (ERR_FSYS_CORRUPT);
677 
678 		le = ZAP_LEAF_ENTRY(l, blksft, chunk);
679 
680 		/* Verify the chunk entry */
681 		if (le->le_type != ZAP_CHUNK_ENTRY)
682 			return (ERR_FSYS_CORRUPT);
683 
684 		if (le->le_hash != h)
685 			continue;
686 
687 		if (zap_leaf_array_equal(l, blksft, le->le_name_chunk,
688 		    le->le_name_length, name)) {
689 
690 			struct zap_leaf_array *la;
691 			uint8_t *ip;
692 
693 			if (le->le_int_size != 8 || le->le_value_length != 1)
694 				return (ERR_FSYS_CORRUPT);
695 
696 			/* get the uint64_t property value */
697 			la = &ZAP_LEAF_CHUNK(l, blksft,
698 			    le->le_value_chunk).l_array;
699 			ip = la->la_array;
700 
701 			*value = (uint64_t)ip[0] << 56 | (uint64_t)ip[1] << 48 |
702 			    (uint64_t)ip[2] << 40 | (uint64_t)ip[3] << 32 |
703 			    (uint64_t)ip[4] << 24 | (uint64_t)ip[5] << 16 |
704 			    (uint64_t)ip[6] << 8 | (uint64_t)ip[7];
705 
706 			return (0);
707 		}
708 	}
709 
710 	return (ERR_FSYS_CORRUPT);
711 }
712 
713 /*
714  * Fat ZAP lookup
715  *
716  * Return:
717  *	0 - success
718  *	errnum - failure
719  */
720 static int
721 fzap_lookup(dnode_phys_t *zap_dnode, zap_phys_t *zap,
722     const char *name, uint64_t *value, char *stack)
723 {
724 	zap_leaf_phys_t *l;
725 	uint64_t hash, idx, blkid;
726 	int blksft = zfs_log2(zap_dnode->dn_datablkszsec << DNODE_SHIFT);
727 
728 	/* Verify if this is a fat zap header block */
729 	if (zap->zap_magic != (uint64_t)ZAP_MAGIC ||
730 	    zap->zap_flags != 0)
731 		return (ERR_FSYS_CORRUPT);
732 
733 	hash = zap_hash(zap->zap_salt, name);
734 	if (errnum)
735 		return (errnum);
736 
737 	/* get block id from index */
738 	if (zap->zap_ptrtbl.zt_numblks != 0) {
739 		/* external pointer tables not supported */
740 		return (ERR_FSYS_CORRUPT);
741 	}
742 	idx = ZAP_HASH_IDX(hash, zap->zap_ptrtbl.zt_shift);
743 	blkid = ((uint64_t *)zap)[idx + (1<<(blksft-3-1))];
744 
745 	/* Get the leaf block */
746 	l = (zap_leaf_phys_t *)stack;
747 	stack += 1<<blksft;
748 	if ((1<<blksft) < sizeof (zap_leaf_phys_t))
749 		return (ERR_FSYS_CORRUPT);
750 	if (errnum = dmu_read(zap_dnode, blkid, l, stack))
751 		return (errnum);
752 
753 	return (zap_leaf_lookup(l, blksft, hash, name, value));
754 }
755 
756 /*
757  * Read in the data of a zap object and find the value for a matching
758  * property name.
759  *
760  * Return:
761  *	0 - success
762  *	errnum - failure
763  */
764 static int
765 zap_lookup(dnode_phys_t *zap_dnode, const char *name, uint64_t *val,
766     char *stack)
767 {
768 	uint64_t block_type;
769 	int size;
770 	void *zapbuf;
771 
772 	/* Read in the first block of the zap object data. */
773 	zapbuf = stack;
774 	size = zap_dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
775 	stack += size;
776 
777 	if ((errnum = dmu_read(zap_dnode, 0, zapbuf, stack)) != 0)
778 		return (errnum);
779 
780 	block_type = *((uint64_t *)zapbuf);
781 
782 	if (block_type == ZBT_MICRO) {
783 		return (mzap_lookup(zapbuf, size, name, val));
784 	} else if (block_type == ZBT_HEADER) {
785 		/* this is a fat zap */
786 		return (fzap_lookup(zap_dnode, zapbuf, name,
787 		    val, stack));
788 	}
789 
790 	return (ERR_FSYS_CORRUPT);
791 }
792 
793 typedef struct zap_attribute {
794 	int za_integer_length;
795 	uint64_t za_num_integers;
796 	uint64_t za_first_integer;
797 	char *za_name;
798 } zap_attribute_t;
799 
800 typedef int (zap_cb_t)(zap_attribute_t *za, void *arg, char *stack);
801 
802 static int
803 zap_iterate(dnode_phys_t *zap_dnode, zap_cb_t *cb, void *arg, char *stack)
804 {
805 	uint32_t size = zap_dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
806 	zap_attribute_t za;
807 	int i;
808 	mzap_phys_t *mzp = (mzap_phys_t *)stack;
809 	stack += size;
810 
811 	if ((errnum = dmu_read(zap_dnode, 0, mzp, stack)) != 0)
812 		return (errnum);
813 
814 	/*
815 	 * Iteration over fatzap objects has not yet been implemented.
816 	 * If we encounter a pool in which there are more features for
817 	 * read than can fit inside a microzap (i.e., more than 2048
818 	 * features for read), we can add support for fatzap iteration.
819 	 * For now, fail.
820 	 */
821 	if (mzp->mz_block_type != ZBT_MICRO) {
822 		grub_printf("feature information stored in fatzap, pool "
823 		    "version not supported\n");
824 		return (1);
825 	}
826 
827 	za.za_integer_length = 8;
828 	za.za_num_integers = 1;
829 	for (i = 0; i < size / MZAP_ENT_LEN - 1; i++) {
830 		mzap_ent_phys_t *mzep = &mzp->mz_chunk[i];
831 		int err;
832 
833 		za.za_first_integer = mzep->mze_value;
834 		za.za_name = mzep->mze_name;
835 		err = cb(&za, arg, stack);
836 		if (err != 0)
837 			return (err);
838 	}
839 
840 	return (0);
841 }
842 
843 /*
844  * Get the dnode of an object number from the metadnode of an object set.
845  *
846  * Input
847  *	mdn - metadnode to get the object dnode
848  *	objnum - object number for the object dnode
849  *	type - if nonzero, object must be of this type
850  *	buf - data buffer that holds the returning dnode
851  *	stack - scratch area
852  *
853  * Return:
854  *	0 - success
855  *	errnum - failure
856  */
857 static int
858 dnode_get(dnode_phys_t *mdn, uint64_t objnum, uint8_t type, dnode_phys_t *buf,
859     char *stack)
860 {
861 	uint64_t blkid, blksz; /* the block id this object dnode is in */
862 	int epbs; /* shift of number of dnodes in a block */
863 	int idx; /* index within a block */
864 	dnode_phys_t *dnbuf;
865 
866 	blksz = mdn->dn_datablkszsec << SPA_MINBLOCKSHIFT;
867 	epbs = zfs_log2(blksz) - DNODE_SHIFT;
868 	blkid = objnum >> epbs;
869 	idx = objnum & ((1<<epbs)-1);
870 
871 	if (dnode_buf != NULL && dnode_mdn == mdn &&
872 	    objnum >= dnode_start && objnum < dnode_end) {
873 		grub_memmove(buf, &dnode_buf[idx], DNODE_SIZE);
874 		VERIFY_DN_TYPE(buf, type);
875 		return (0);
876 	}
877 
878 	if (dnode_buf && blksz == 1<<DNODE_BLOCK_SHIFT) {
879 		dnbuf = dnode_buf;
880 		dnode_mdn = mdn;
881 		dnode_start = blkid << epbs;
882 		dnode_end = (blkid + 1) << epbs;
883 	} else {
884 		dnbuf = (dnode_phys_t *)stack;
885 		stack += blksz;
886 	}
887 
888 	if (errnum = dmu_read(mdn, blkid, (char *)dnbuf, stack))
889 		return (errnum);
890 
891 	grub_memmove(buf, &dnbuf[idx], DNODE_SIZE);
892 	VERIFY_DN_TYPE(buf, type);
893 
894 	return (0);
895 }
896 
897 /*
898  * Check if this is a special file that resides at the top
899  * dataset of the pool. Currently this is the GRUB menu,
900  * boot signature and boot signature backup.
901  * str starts with '/'.
902  */
903 static int
904 is_top_dataset_file(char *str)
905 {
906 	char *tptr;
907 
908 	if ((tptr = grub_strstr(str, "menu.lst")) &&
909 	    (tptr[8] == '\0' || tptr[8] == ' ') &&
910 	    *(tptr-1) == '/')
911 		return (1);
912 
913 	if (grub_strncmp(str, BOOTSIGN_DIR"/",
914 	    grub_strlen(BOOTSIGN_DIR) + 1) == 0)
915 		return (1);
916 
917 	if (grub_strcmp(str, BOOTSIGN_BACKUP) == 0)
918 		return (1);
919 
920 	return (0);
921 }
922 
923 static int
924 check_feature(zap_attribute_t *za, void *arg, char *stack)
925 {
926 	const char **names = arg;
927 	int i;
928 
929 	if (za->za_first_integer == 0)
930 		return (0);
931 
932 	for (i = 0; names[i] != NULL; i++) {
933 		if (grub_strcmp(za->za_name, names[i]) == 0) {
934 			return (0);
935 		}
936 	}
937 	grub_printf("missing feature for read '%s'\n", za->za_name);
938 	return (ERR_NEWER_VERSION);
939 }
940 
941 /*
942  * Get the file dnode for a given file name where mdn is the meta dnode
943  * for this ZFS object set. When found, place the file dnode in dn.
944  * The 'path' argument will be mangled.
945  *
946  * Return:
947  *	0 - success
948  *	errnum - failure
949  */
950 static int
951 dnode_get_path(dnode_phys_t *mdn, char *path, dnode_phys_t *dn,
952     char *stack)
953 {
954 	uint64_t objnum, version;
955 	char *cname, ch;
956 
957 	if (errnum = dnode_get(mdn, MASTER_NODE_OBJ, DMU_OT_MASTER_NODE,
958 	    dn, stack))
959 		return (errnum);
960 
961 	if (errnum = zap_lookup(dn, ZPL_VERSION_STR, &version, stack))
962 		return (errnum);
963 	if (version > ZPL_VERSION)
964 		return (-1);
965 
966 	if (errnum = zap_lookup(dn, ZFS_ROOT_OBJ, &objnum, stack))
967 		return (errnum);
968 
969 	if (errnum = dnode_get(mdn, objnum, DMU_OT_DIRECTORY_CONTENTS,
970 	    dn, stack))
971 		return (errnum);
972 
973 	/* skip leading slashes */
974 	while (*path == '/')
975 		path++;
976 
977 	while (*path && !grub_isspace(*path)) {
978 
979 		/* get the next component name */
980 		cname = path;
981 		while (*path && !grub_isspace(*path) && *path != '/')
982 			path++;
983 		ch = *path;
984 		*path = 0;   /* ensure null termination */
985 
986 		if (errnum = zap_lookup(dn, cname, &objnum, stack))
987 			return (errnum);
988 
989 		objnum = ZFS_DIRENT_OBJ(objnum);
990 		if (errnum = dnode_get(mdn, objnum, 0, dn, stack))
991 			return (errnum);
992 
993 		*path = ch;
994 		while (*path == '/')
995 			path++;
996 	}
997 
998 	/* We found the dnode for this file. Verify if it is a plain file. */
999 	VERIFY_DN_TYPE(dn, DMU_OT_PLAIN_FILE_CONTENTS);
1000 
1001 	return (0);
1002 }
1003 
1004 /*
1005  * Get the default 'bootfs' property value from the rootpool.
1006  *
1007  * Return:
1008  *	0 - success
1009  *	errnum -failure
1010  */
1011 static int
1012 get_default_bootfsobj(dnode_phys_t *mosmdn, uint64_t *obj, char *stack)
1013 {
1014 	uint64_t objnum = 0;
1015 	dnode_phys_t *dn = (dnode_phys_t *)stack;
1016 	stack += DNODE_SIZE;
1017 
1018 	if (errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT,
1019 	    DMU_OT_OBJECT_DIRECTORY, dn, stack))
1020 		return (errnum);
1021 
1022 	/*
1023 	 * find the object number for 'pool_props', and get the dnode
1024 	 * of the 'pool_props'.
1025 	 */
1026 	if (zap_lookup(dn, DMU_POOL_PROPS, &objnum, stack))
1027 		return (ERR_FILESYSTEM_NOT_FOUND);
1028 
1029 	if (errnum = dnode_get(mosmdn, objnum, DMU_OT_POOL_PROPS, dn, stack))
1030 		return (errnum);
1031 
1032 	if (zap_lookup(dn, ZPOOL_PROP_BOOTFS, &objnum, stack))
1033 		return (ERR_FILESYSTEM_NOT_FOUND);
1034 
1035 	if (!objnum)
1036 		return (ERR_FILESYSTEM_NOT_FOUND);
1037 
1038 	*obj = objnum;
1039 	return (0);
1040 }
1041 
1042 /*
1043  * List of pool features that the grub implementation of ZFS supports for
1044  * read. Note that features that are only required for write do not need
1045  * to be listed here since grub opens pools in read-only mode.
1046  *
1047  * When this list is updated the version number in usr/src/grub/capability
1048  * must be incremented to ensure the new grub gets installed.
1049  */
1050 static const char *spa_feature_names[] = {
1051 	"org.illumos:lz4_compress",
1052 	"com.delphix:hole_birth",
1053 	"com.delphix:extensible_dataset",
1054 	"com.delphix:embedded_data",
1055 	"org.open-zfs:large_blocks",
1056 	"org.illumos:sha512",
1057 	NULL
1058 };
1059 
1060 /*
1061  * Checks whether the MOS features that are active are supported by this
1062  * (GRUB's) implementation of ZFS.
1063  *
1064  * Return:
1065  *	0: Success.
1066  *	errnum: Failure.
1067  */
1068 static int
1069 check_mos_features(dnode_phys_t *mosmdn, char *stack)
1070 {
1071 	uint64_t objnum;
1072 	dnode_phys_t *dn;
1073 	uint8_t error = 0;
1074 
1075 	dn = (dnode_phys_t *)stack;
1076 	stack += DNODE_SIZE;
1077 
1078 	if ((errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT,
1079 	    DMU_OT_OBJECT_DIRECTORY, dn, stack)) != 0)
1080 		return (errnum);
1081 
1082 	/*
1083 	 * Find the object number for 'features_for_read' and retrieve its
1084 	 * corresponding dnode. Note that we don't check features_for_write
1085 	 * because GRUB is not opening the pool for write.
1086 	 */
1087 	if ((errnum = zap_lookup(dn, DMU_POOL_FEATURES_FOR_READ, &objnum,
1088 	    stack)) != 0)
1089 		return (errnum);
1090 
1091 	if ((errnum = dnode_get(mosmdn, objnum, DMU_OTN_ZAP_METADATA,
1092 	    dn, stack)) != 0)
1093 		return (errnum);
1094 
1095 	return (zap_iterate(dn, check_feature, spa_feature_names, stack));
1096 }
1097 
1098 /*
1099  * Given a MOS metadnode, get the metadnode of a given filesystem name (fsname),
1100  * e.g. pool/rootfs, or a given object number (obj), e.g. the object number
1101  * of pool/rootfs.
1102  *
1103  * If no fsname and no obj are given, return the DSL_DIR metadnode.
1104  * If fsname is given, return its metadnode and its matching object number.
1105  * If only obj is given, return the metadnode for this object number.
1106  *
1107  * Return:
1108  *	0 - success
1109  *	errnum - failure
1110  */
1111 static int
1112 get_objset_mdn(dnode_phys_t *mosmdn, char *fsname, uint64_t *obj,
1113     dnode_phys_t *mdn, char *stack)
1114 {
1115 	uint64_t objnum, headobj;
1116 	char *cname, ch;
1117 	blkptr_t *bp;
1118 	objset_phys_t *osp;
1119 	int issnapshot = 0;
1120 	char *snapname;
1121 
1122 	if (fsname == NULL && obj) {
1123 		headobj = *obj;
1124 		goto skip;
1125 	}
1126 
1127 	if (errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT,
1128 	    DMU_OT_OBJECT_DIRECTORY, mdn, stack))
1129 		return (errnum);
1130 
1131 	if (errnum = zap_lookup(mdn, DMU_POOL_ROOT_DATASET, &objnum,
1132 	    stack))
1133 		return (errnum);
1134 
1135 	if (errnum = dnode_get(mosmdn, objnum, 0, mdn, stack))
1136 		return (errnum);
1137 
1138 	if (fsname == NULL) {
1139 		headobj =
1140 		    ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj;
1141 		goto skip;
1142 	}
1143 
1144 	/* take out the pool name */
1145 	while (*fsname && !grub_isspace(*fsname) && *fsname != '/')
1146 		fsname++;
1147 
1148 	while (*fsname && !grub_isspace(*fsname)) {
1149 		uint64_t childobj;
1150 
1151 		while (*fsname == '/')
1152 			fsname++;
1153 
1154 		cname = fsname;
1155 		while (*fsname && !grub_isspace(*fsname) && *fsname != '/')
1156 			fsname++;
1157 		ch = *fsname;
1158 		*fsname = 0;
1159 
1160 		snapname = cname;
1161 		while (*snapname && !grub_isspace(*snapname) && *snapname !=
1162 		    '@')
1163 			snapname++;
1164 		if (*snapname == '@') {
1165 			issnapshot = 1;
1166 			*snapname = 0;
1167 		}
1168 		childobj =
1169 		    ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_child_dir_zapobj;
1170 		if (errnum = dnode_get(mosmdn, childobj,
1171 		    DMU_OT_DSL_DIR_CHILD_MAP, mdn, stack))
1172 			return (errnum);
1173 
1174 		if (zap_lookup(mdn, cname, &objnum, stack))
1175 			return (ERR_FILESYSTEM_NOT_FOUND);
1176 
1177 		if (errnum = dnode_get(mosmdn, objnum, 0,
1178 		    mdn, stack))
1179 			return (errnum);
1180 
1181 		*fsname = ch;
1182 		if (issnapshot)
1183 			*snapname = '@';
1184 	}
1185 	headobj = ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj;
1186 	if (obj)
1187 		*obj = headobj;
1188 
1189 skip:
1190 	if (errnum = dnode_get(mosmdn, headobj, 0, mdn, stack))
1191 		return (errnum);
1192 	if (issnapshot) {
1193 		uint64_t snapobj;
1194 
1195 		snapobj = ((dsl_dataset_phys_t *)DN_BONUS(mdn))->
1196 		    ds_snapnames_zapobj;
1197 
1198 		if (errnum = dnode_get(mosmdn, snapobj,
1199 		    DMU_OT_DSL_DS_SNAP_MAP, mdn, stack))
1200 			return (errnum);
1201 		if (zap_lookup(mdn, snapname + 1, &headobj, stack))
1202 			return (ERR_FILESYSTEM_NOT_FOUND);
1203 		if (errnum = dnode_get(mosmdn, headobj, 0, mdn, stack))
1204 			return (errnum);
1205 		if (obj)
1206 			*obj = headobj;
1207 	}
1208 
1209 	bp = &((dsl_dataset_phys_t *)DN_BONUS(mdn))->ds_bp;
1210 	osp = (objset_phys_t *)stack;
1211 	stack += sizeof (objset_phys_t);
1212 	if (errnum = zio_read(bp, osp, stack))
1213 		return (errnum);
1214 
1215 	grub_memmove((char *)mdn, (char *)&osp->os_meta_dnode, DNODE_SIZE);
1216 
1217 	return (0);
1218 }
1219 
1220 /*
1221  * For a given XDR packed nvlist, verify the first 4 bytes and move on.
1222  *
1223  * An XDR packed nvlist is encoded as (comments from nvs_xdr_create) :
1224  *
1225  *      encoding method/host endian     (4 bytes)
1226  *      nvl_version                     (4 bytes)
1227  *      nvl_nvflag                      (4 bytes)
1228  *	encoded nvpairs:
1229  *		encoded size of the nvpair      (4 bytes)
1230  *		decoded size of the nvpair      (4 bytes)
1231  *		name string size                (4 bytes)
1232  *		name string data                (sizeof(NV_ALIGN4(string))
1233  *		data type                       (4 bytes)
1234  *		# of elements in the nvpair     (4 bytes)
1235  *		data
1236  *      2 zero's for the last nvpair
1237  *		(end of the entire list)	(8 bytes)
1238  *
1239  * Return:
1240  *	0 - success
1241  *	1 - failure
1242  */
1243 static int
1244 nvlist_unpack(char *nvlist, char **out)
1245 {
1246 	/* Verify if the 1st and 2nd byte in the nvlist are valid. */
1247 	if (nvlist[0] != NV_ENCODE_XDR || nvlist[1] != HOST_ENDIAN)
1248 		return (1);
1249 
1250 	*out = nvlist + 4;
1251 	return (0);
1252 }
1253 
1254 static char *
1255 nvlist_array(char *nvlist, int index)
1256 {
1257 	int i, encode_size;
1258 
1259 	for (i = 0; i < index; i++) {
1260 		/* skip the header, nvl_version, and nvl_nvflag */
1261 		nvlist = nvlist + 4 * 2;
1262 
1263 		while (encode_size = BSWAP_32(*(uint32_t *)nvlist))
1264 			nvlist += encode_size; /* goto the next nvpair */
1265 
1266 		nvlist = nvlist + 4 * 2; /* skip the ending 2 zeros - 8 bytes */
1267 	}
1268 
1269 	return (nvlist);
1270 }
1271 
1272 /*
1273  * The nvlist_next_nvpair() function returns a handle to the next nvpair in the
1274  * list following nvpair. If nvpair is NULL, the first pair is returned. If
1275  * nvpair is the last pair in the nvlist, NULL is returned.
1276  */
1277 static char *
1278 nvlist_next_nvpair(char *nvl, char *nvpair)
1279 {
1280 	char *cur, *prev;
1281 	int encode_size;
1282 
1283 	if (nvl == NULL)
1284 		return (NULL);
1285 
1286 	if (nvpair == NULL) {
1287 		/* skip over nvl_version and nvl_nvflag */
1288 		nvpair = nvl + 4 * 2;
1289 	} else {
1290 		/* skip to the next nvpair */
1291 		encode_size = BSWAP_32(*(uint32_t *)nvpair);
1292 		nvpair += encode_size;
1293 	}
1294 
1295 	/* 8 bytes of 0 marks the end of the list */
1296 	if (*(uint64_t *)nvpair == 0)
1297 		return (NULL);
1298 
1299 	return (nvpair);
1300 }
1301 
1302 /*
1303  * This function returns 0 on success and 1 on failure. On success, a string
1304  * containing the name of nvpair is saved in buf.
1305  */
1306 static int
1307 nvpair_name(char *nvp, char *buf, int buflen)
1308 {
1309 	int len;
1310 
1311 	/* skip over encode/decode size */
1312 	nvp += 4 * 2;
1313 
1314 	len = BSWAP_32(*(uint32_t *)nvp);
1315 	if (buflen < len + 1)
1316 		return (1);
1317 
1318 	grub_memmove(buf, nvp + 4, len);
1319 	buf[len] = '\0';
1320 
1321 	return (0);
1322 }
1323 
1324 /*
1325  * This function retrieves the value of the nvpair in the form of enumerated
1326  * type data_type_t. This is used to determine the appropriate type to pass to
1327  * nvpair_value().
1328  */
1329 static int
1330 nvpair_type(char *nvp)
1331 {
1332 	int name_len, type;
1333 
1334 	/* skip over encode/decode size */
1335 	nvp += 4 * 2;
1336 
1337 	/* skip over name_len */
1338 	name_len = BSWAP_32(*(uint32_t *)nvp);
1339 	nvp += 4;
1340 
1341 	/* skip over name */
1342 	nvp = nvp + ((name_len + 3) & ~3); /* align */
1343 
1344 	type = BSWAP_32(*(uint32_t *)nvp);
1345 
1346 	return (type);
1347 }
1348 
1349 static int
1350 nvpair_value(char *nvp, void *val, int valtype, int *nelmp)
1351 {
1352 	int name_len, type, slen;
1353 	char *strval = val;
1354 	uint64_t *intval = val;
1355 
1356 	/* skip over encode/decode size */
1357 	nvp += 4 * 2;
1358 
1359 	/* skip over name_len */
1360 	name_len = BSWAP_32(*(uint32_t *)nvp);
1361 	nvp += 4;
1362 
1363 	/* skip over name */
1364 	nvp = nvp + ((name_len + 3) & ~3); /* align */
1365 
1366 	/* skip over type */
1367 	type = BSWAP_32(*(uint32_t *)nvp);
1368 	nvp += 4;
1369 
1370 	if (type == valtype) {
1371 		int nelm;
1372 
1373 		nelm = BSWAP_32(*(uint32_t *)nvp);
1374 		if (valtype != DATA_TYPE_BOOLEAN && nelm < 1)
1375 			return (1);
1376 		nvp += 4;
1377 
1378 		switch (valtype) {
1379 		case DATA_TYPE_BOOLEAN:
1380 			return (0);
1381 
1382 		case DATA_TYPE_STRING:
1383 			slen = BSWAP_32(*(uint32_t *)nvp);
1384 			nvp += 4;
1385 			grub_memmove(strval, nvp, slen);
1386 			strval[slen] = '\0';
1387 			return (0);
1388 
1389 		case DATA_TYPE_UINT64:
1390 			*intval = BSWAP_64(*(uint64_t *)nvp);
1391 			return (0);
1392 
1393 		case DATA_TYPE_NVLIST:
1394 			*(void **)val = (void *)nvp;
1395 			return (0);
1396 
1397 		case DATA_TYPE_NVLIST_ARRAY:
1398 			*(void **)val = (void *)nvp;
1399 			if (nelmp)
1400 				*nelmp = nelm;
1401 			return (0);
1402 		}
1403 	}
1404 
1405 	return (1);
1406 }
1407 
1408 static int
1409 nvlist_lookup_value(char *nvlist, char *name, void *val, int valtype,
1410     int *nelmp)
1411 {
1412 	char *nvpair;
1413 
1414 	for (nvpair = nvlist_next_nvpair(nvlist, NULL);
1415 	    nvpair != NULL;
1416 	    nvpair = nvlist_next_nvpair(nvlist, nvpair)) {
1417 		int name_len = BSWAP_32(*(uint32_t *)(nvpair + 4 * 2));
1418 		char *nvp_name = nvpair + 4 * 3;
1419 
1420 		if ((grub_strncmp(nvp_name, name, name_len) == 0) &&
1421 		    nvpair_type(nvpair) == valtype) {
1422 			return (nvpair_value(nvpair, val, valtype, nelmp));
1423 		}
1424 	}
1425 	return (1);
1426 }
1427 
1428 /*
1429  * Check if this vdev is online and is in a good state.
1430  */
1431 static int
1432 vdev_validate(char *nv)
1433 {
1434 	uint64_t ival;
1435 
1436 	if (nvlist_lookup_value(nv, ZPOOL_CONFIG_OFFLINE, &ival,
1437 	    DATA_TYPE_UINT64, NULL) == 0 ||
1438 	    nvlist_lookup_value(nv, ZPOOL_CONFIG_FAULTED, &ival,
1439 	    DATA_TYPE_UINT64, NULL) == 0 ||
1440 	    nvlist_lookup_value(nv, ZPOOL_CONFIG_REMOVED, &ival,
1441 	    DATA_TYPE_UINT64, NULL) == 0)
1442 		return (ERR_DEV_VALUES);
1443 
1444 	return (0);
1445 }
1446 
1447 /*
1448  * Get a valid vdev pathname/devid from the boot device.
1449  * The caller should already allocate MAXPATHLEN memory for bootpath and devid.
1450  */
1451 static int
1452 vdev_get_bootpath(char *nv, uint64_t inguid, char *devid, char *bootpath,
1453     int is_spare)
1454 {
1455 	char type[16];
1456 
1457 	if (nvlist_lookup_value(nv, ZPOOL_CONFIG_TYPE, &type, DATA_TYPE_STRING,
1458 	    NULL))
1459 		return (ERR_FSYS_CORRUPT);
1460 
1461 	if (grub_strcmp(type, VDEV_TYPE_DISK) == 0) {
1462 		uint64_t guid;
1463 
1464 		if (vdev_validate(nv) != 0)
1465 			return (ERR_NO_BOOTPATH);
1466 
1467 		if (nvlist_lookup_value(nv, ZPOOL_CONFIG_GUID,
1468 		    &guid, DATA_TYPE_UINT64, NULL) != 0)
1469 			return (ERR_NO_BOOTPATH);
1470 
1471 		if (guid != inguid)
1472 			return (ERR_NO_BOOTPATH);
1473 
1474 		/* for a spare vdev, pick the disk labeled with "is_spare" */
1475 		if (is_spare) {
1476 			uint64_t spare = 0;
1477 			(void) nvlist_lookup_value(nv, ZPOOL_CONFIG_IS_SPARE,
1478 			    &spare, DATA_TYPE_UINT64, NULL);
1479 			if (!spare)
1480 				return (ERR_NO_BOOTPATH);
1481 		}
1482 
1483 		if (nvlist_lookup_value(nv, ZPOOL_CONFIG_PHYS_PATH,
1484 		    bootpath, DATA_TYPE_STRING, NULL) != 0)
1485 			bootpath[0] = '\0';
1486 
1487 		if (nvlist_lookup_value(nv, ZPOOL_CONFIG_DEVID,
1488 		    devid, DATA_TYPE_STRING, NULL) != 0)
1489 			devid[0] = '\0';
1490 
1491 		if (grub_strlen(bootpath) >= MAXPATHLEN ||
1492 		    grub_strlen(devid) >= MAXPATHLEN)
1493 			return (ERR_WONT_FIT);
1494 
1495 		return (0);
1496 
1497 	} else if (grub_strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
1498 	    grub_strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
1499 	    (is_spare = (grub_strcmp(type, VDEV_TYPE_SPARE) == 0))) {
1500 		int nelm, i;
1501 		char *child;
1502 
1503 		if (nvlist_lookup_value(nv, ZPOOL_CONFIG_CHILDREN, &child,
1504 		    DATA_TYPE_NVLIST_ARRAY, &nelm))
1505 			return (ERR_FSYS_CORRUPT);
1506 
1507 		for (i = 0; i < nelm; i++) {
1508 			char *child_i;
1509 
1510 			child_i = nvlist_array(child, i);
1511 			if (vdev_get_bootpath(child_i, inguid, devid,
1512 			    bootpath, is_spare) == 0)
1513 				return (0);
1514 		}
1515 	}
1516 
1517 	return (ERR_NO_BOOTPATH);
1518 }
1519 
1520 /*
1521  * Check the disk label information and retrieve needed vdev name-value pairs.
1522  *
1523  * Return:
1524  *	0 - success
1525  *	ERR_* - failure
1526  */
1527 static int
1528 check_pool_label(uint64_t sector, char *stack, char *outdevid,
1529     char *outpath, uint64_t *outguid, uint64_t *outashift, uint64_t *outversion)
1530 {
1531 	vdev_phys_t *vdev;
1532 	uint64_t pool_state, txg = 0;
1533 	char *nvlist, *nv, *features;
1534 	uint64_t diskguid;
1535 
1536 	sector += (VDEV_SKIP_SIZE >> SPA_MINBLOCKSHIFT);
1537 
1538 	/* Read in the vdev name-value pair list (112K). */
1539 	if (devread(sector, 0, VDEV_PHYS_SIZE, stack) == 0)
1540 		return (ERR_READ);
1541 
1542 	vdev = (vdev_phys_t *)stack;
1543 	stack += sizeof (vdev_phys_t);
1544 
1545 	if (nvlist_unpack(vdev->vp_nvlist, &nvlist))
1546 		return (ERR_FSYS_CORRUPT);
1547 
1548 	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_STATE, &pool_state,
1549 	    DATA_TYPE_UINT64, NULL))
1550 		return (ERR_FSYS_CORRUPT);
1551 
1552 	if (pool_state == POOL_STATE_DESTROYED)
1553 		return (ERR_FILESYSTEM_NOT_FOUND);
1554 
1555 	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_NAME,
1556 	    current_rootpool, DATA_TYPE_STRING, NULL))
1557 		return (ERR_FSYS_CORRUPT);
1558 
1559 	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_TXG, &txg,
1560 	    DATA_TYPE_UINT64, NULL))
1561 		return (ERR_FSYS_CORRUPT);
1562 
1563 	/* not an active device */
1564 	if (txg == 0)
1565 		return (ERR_NO_BOOTPATH);
1566 
1567 	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_VERSION, outversion,
1568 	    DATA_TYPE_UINT64, NULL))
1569 		return (ERR_FSYS_CORRUPT);
1570 	if (!SPA_VERSION_IS_SUPPORTED(*outversion))
1571 		return (ERR_NEWER_VERSION);
1572 	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_VDEV_TREE, &nv,
1573 	    DATA_TYPE_NVLIST, NULL))
1574 		return (ERR_FSYS_CORRUPT);
1575 	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_GUID, &diskguid,
1576 	    DATA_TYPE_UINT64, NULL))
1577 		return (ERR_FSYS_CORRUPT);
1578 	if (nvlist_lookup_value(nv, ZPOOL_CONFIG_ASHIFT, outashift,
1579 	    DATA_TYPE_UINT64, NULL) != 0)
1580 		return (ERR_FSYS_CORRUPT);
1581 	if (vdev_get_bootpath(nv, diskguid, outdevid, outpath, 0))
1582 		return (ERR_NO_BOOTPATH);
1583 	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_GUID, outguid,
1584 	    DATA_TYPE_UINT64, NULL))
1585 		return (ERR_FSYS_CORRUPT);
1586 
1587 	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_FEATURES_FOR_READ,
1588 	    &features, DATA_TYPE_NVLIST, NULL) == 0) {
1589 		char *nvp;
1590 		char *name = stack;
1591 		stack += MAXNAMELEN;
1592 
1593 		for (nvp = nvlist_next_nvpair(features, NULL);
1594 		    nvp != NULL;
1595 		    nvp = nvlist_next_nvpair(features, nvp)) {
1596 			zap_attribute_t za;
1597 
1598 			if (nvpair_name(nvp, name, MAXNAMELEN) != 0)
1599 				return (ERR_FSYS_CORRUPT);
1600 
1601 			za.za_integer_length = 8;
1602 			za.za_num_integers = 1;
1603 			za.za_first_integer = 1;
1604 			za.za_name = name;
1605 			if (check_feature(&za, spa_feature_names, stack) != 0)
1606 				return (ERR_NEWER_VERSION);
1607 		}
1608 	}
1609 
1610 	return (0);
1611 }
1612 
1613 /*
1614  * zfs_mount() locates a valid uberblock of the root pool and read in its MOS
1615  * to the memory address MOS.
1616  *
1617  * Return:
1618  *	1 - success
1619  *	0 - failure
1620  */
1621 int
1622 zfs_mount(void)
1623 {
1624 	char *stack, *ub_array;
1625 	int label = 0;
1626 	uberblock_t *ubbest;
1627 	objset_phys_t *osp;
1628 	char tmp_bootpath[MAXNAMELEN];
1629 	char tmp_devid[MAXNAMELEN];
1630 	uint64_t tmp_guid, ashift, version;
1631 	uint64_t adjpl = (uint64_t)part_length << SPA_MINBLOCKSHIFT;
1632 	int err = errnum; /* preserve previous errnum state */
1633 
1634 	/* if it's our first time here, zero the best uberblock out */
1635 	if (best_drive == 0 && best_part == 0 && find_best_root) {
1636 		grub_memset(&current_uberblock, 0, sizeof (uberblock_t));
1637 		pool_guid = 0;
1638 	}
1639 
1640 	stackbase = ZFS_SCRATCH;
1641 	stack = stackbase;
1642 	ub_array = stack;
1643 	stack += VDEV_UBERBLOCK_RING;
1644 
1645 	osp = (objset_phys_t *)stack;
1646 	stack += sizeof (objset_phys_t);
1647 	adjpl = P2ALIGN(adjpl, (uint64_t)sizeof (vdev_label_t));
1648 
1649 	for (label = 0; label < VDEV_LABELS; label++) {
1650 
1651 		/*
1652 		 * some eltorito stacks don't give us a size and
1653 		 * we end up setting the size to MAXUINT, further
1654 		 * some of these devices stop working once a single
1655 		 * read past the end has been issued. Checking
1656 		 * for a maximum part_length and skipping the backup
1657 		 * labels at the end of the slice/partition/device
1658 		 * avoids breaking down on such devices.
1659 		 */
1660 		if (part_length == MAXUINT && label == 2)
1661 			break;
1662 
1663 		uint64_t sector = vdev_label_start(adjpl,
1664 		    label) >> SPA_MINBLOCKSHIFT;
1665 
1666 		/* Read in the uberblock ring (128K). */
1667 		if (devread(sector  +
1668 		    ((VDEV_SKIP_SIZE + VDEV_PHYS_SIZE) >> SPA_MINBLOCKSHIFT),
1669 		    0, VDEV_UBERBLOCK_RING, ub_array) == 0)
1670 			continue;
1671 
1672 		if (check_pool_label(sector, stack, tmp_devid,
1673 		    tmp_bootpath, &tmp_guid, &ashift, &version))
1674 			continue;
1675 
1676 		if (pool_guid == 0)
1677 			pool_guid = tmp_guid;
1678 
1679 		if ((ubbest = find_bestub(ub_array, ashift, sector)) == NULL ||
1680 		    zio_read(&ubbest->ub_rootbp, osp, stack) != 0)
1681 			continue;
1682 
1683 		VERIFY_OS_TYPE(osp, DMU_OST_META);
1684 
1685 		if (version >= SPA_VERSION_FEATURES &&
1686 		    check_mos_features(&osp->os_meta_dnode, stack) != 0)
1687 			continue;
1688 
1689 		if (find_best_root && ((pool_guid != tmp_guid) ||
1690 		    vdev_uberblock_compare(ubbest, &(current_uberblock)) <= 0))
1691 			continue;
1692 
1693 		/* Got the MOS. Save it at the memory addr MOS. */
1694 		grub_memmove(MOS, &osp->os_meta_dnode, DNODE_SIZE);
1695 		grub_memmove(&current_uberblock, ubbest, sizeof (uberblock_t));
1696 		grub_memmove(current_bootpath, tmp_bootpath, MAXNAMELEN);
1697 		grub_memmove(current_devid, tmp_devid, grub_strlen(tmp_devid));
1698 		is_zfs_mount = 1;
1699 		return (1);
1700 	}
1701 
1702 	/*
1703 	 * While some fs impls. (tftp) rely on setting and keeping
1704 	 * global errnums set, others won't reset it and will break
1705 	 * when issuing rawreads. The goal here is to simply not
1706 	 * have zfs mount attempts impact the previous state.
1707 	 */
1708 	errnum = err;
1709 	return (0);
1710 }
1711 
1712 /*
1713  * zfs_open() locates a file in the rootpool by following the
1714  * MOS and places the dnode of the file in the memory address DNODE.
1715  *
1716  * Return:
1717  *	1 - success
1718  *	0 - failure
1719  */
1720 int
1721 zfs_open(char *filename)
1722 {
1723 	char *stack;
1724 	dnode_phys_t *mdn;
1725 
1726 	file_buf = NULL;
1727 	stackbase = ZFS_SCRATCH;
1728 	stack = stackbase;
1729 
1730 	mdn = (dnode_phys_t *)stack;
1731 	stack += sizeof (dnode_phys_t);
1732 
1733 	dnode_mdn = NULL;
1734 	dnode_buf = (dnode_phys_t *)stack;
1735 	stack += 1<<DNODE_BLOCK_SHIFT;
1736 
1737 	/*
1738 	 * menu.lst is placed at the root pool filesystem level,
1739 	 * do not goto 'current_bootfs'.
1740 	 */
1741 	if (is_top_dataset_file(filename)) {
1742 		if (errnum = get_objset_mdn(MOS, NULL, NULL, mdn, stack))
1743 			return (0);
1744 
1745 		current_bootfs_obj = 0;
1746 	} else {
1747 		if (current_bootfs[0] == '\0') {
1748 			/* Get the default root filesystem object number */
1749 			if (errnum = get_default_bootfsobj(MOS,
1750 			    &current_bootfs_obj, stack))
1751 				return (0);
1752 
1753 			if (errnum = get_objset_mdn(MOS, NULL,
1754 			    &current_bootfs_obj, mdn, stack))
1755 				return (0);
1756 		} else {
1757 			if (errnum = get_objset_mdn(MOS, current_bootfs,
1758 			    &current_bootfs_obj, mdn, stack)) {
1759 				grub_memset(current_bootfs, 0, MAXNAMELEN);
1760 				return (0);
1761 			}
1762 		}
1763 	}
1764 
1765 	if (dnode_get_path(mdn, filename, DNODE, stack)) {
1766 		errnum = ERR_FILE_NOT_FOUND;
1767 		return (0);
1768 	}
1769 
1770 	/* get the file size and set the file position to 0 */
1771 
1772 	/*
1773 	 * For DMU_OT_SA we will need to locate the SIZE attribute
1774 	 * attribute, which could be either in the bonus buffer
1775 	 * or the "spill" block.
1776 	 */
1777 	if (DNODE->dn_bonustype == DMU_OT_SA) {
1778 		sa_hdr_phys_t *sahdrp;
1779 		int hdrsize;
1780 
1781 		if (DNODE->dn_bonuslen != 0) {
1782 			sahdrp = (sa_hdr_phys_t *)DN_BONUS(DNODE);
1783 		} else {
1784 			if (DNODE->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
1785 				blkptr_t *bp = &DNODE->dn_spill;
1786 				void *buf;
1787 
1788 				buf = (void *)stack;
1789 				stack += BP_GET_LSIZE(bp);
1790 
1791 				/* reset errnum to rawread() failure */
1792 				errnum = 0;
1793 				if (zio_read(bp, buf, stack) != 0) {
1794 					return (0);
1795 				}
1796 				sahdrp = buf;
1797 			} else {
1798 				errnum = ERR_FSYS_CORRUPT;
1799 				return (0);
1800 			}
1801 		}
1802 		hdrsize = SA_HDR_SIZE(sahdrp);
1803 		filemax = *(uint64_t *)((char *)sahdrp + hdrsize +
1804 		    SA_SIZE_OFFSET);
1805 	} else {
1806 		filemax = ((znode_phys_t *)DN_BONUS(DNODE))->zp_size;
1807 	}
1808 	filepos = 0;
1809 
1810 	dnode_buf = NULL;
1811 	return (1);
1812 }
1813 
1814 /*
1815  * zfs_read reads in the data blocks pointed by the DNODE.
1816  *
1817  * Return:
1818  *	len - the length successfully read in to the buffer
1819  *	0   - failure
1820  */
1821 int
1822 zfs_read(char *buf, int len)
1823 {
1824 	char *stack;
1825 	int blksz, length, movesize;
1826 
1827 	if (file_buf == NULL) {
1828 		file_buf = stackbase;
1829 		stackbase += SPA_MAXBLOCKSIZE;
1830 		file_start = file_end = 0;
1831 	}
1832 	stack = stackbase;
1833 
1834 	/*
1835 	 * If offset is in memory, move it into the buffer provided and return.
1836 	 */
1837 	if (filepos >= file_start && filepos+len <= file_end) {
1838 		grub_memmove(buf, file_buf + filepos - file_start, len);
1839 		filepos += len;
1840 		return (len);
1841 	}
1842 
1843 	blksz = DNODE->dn_datablkszsec << SPA_MINBLOCKSHIFT;
1844 
1845 	/*
1846 	 * Note: for GRUB, SPA_MAXBLOCKSIZE is 128KB.  There is not enough
1847 	 * memory to allocate the new max blocksize (16MB), so while
1848 	 * GRUB understands the large_blocks on-disk feature, it can't
1849 	 * actually read large blocks.
1850 	 */
1851 	if (blksz > SPA_MAXBLOCKSIZE) {
1852 		grub_printf("blocks larger than 128K are not supported\n");
1853 		return (0);
1854 	}
1855 
1856 	/*
1857 	 * Entire Dnode is too big to fit into the space available.  We
1858 	 * will need to read it in chunks.  This could be optimized to
1859 	 * read in as large a chunk as there is space available, but for
1860 	 * now, this only reads in one data block at a time.
1861 	 */
1862 	length = len;
1863 	while (length) {
1864 		/*
1865 		 * Find requested blkid and the offset within that block.
1866 		 */
1867 		uint64_t blkid = filepos / blksz;
1868 
1869 		if (errnum = dmu_read(DNODE, blkid, file_buf, stack))
1870 			return (0);
1871 
1872 		file_start = blkid * blksz;
1873 		file_end = file_start + blksz;
1874 
1875 		movesize = MIN(length, file_end - filepos);
1876 
1877 		grub_memmove(buf, file_buf + filepos - file_start,
1878 		    movesize);
1879 		buf += movesize;
1880 		length -= movesize;
1881 		filepos += movesize;
1882 	}
1883 
1884 	return (len);
1885 }
1886 
1887 /*
1888  * No-Op
1889  */
1890 int
1891 zfs_embed(int *start_sector, int needed_sectors)
1892 {
1893 	return (1);
1894 }
1895 
1896 #endif /* FSYS_ZFS */
1897