xref: /illumos-gate/usr/src/grub/grub-0.97/stage2/fsys_zfs.c (revision c3d26abc9ee97b4f60233556aadeb57e0bd30bb9)

/*
 *  GRUB  --  GRand Unified Bootloader
 *  Copyright (C) 1999,2000,2001,2002,2003,2004  Free Software Foundation, Inc.
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program; if not, write to the Free Software
 *  Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
 */

/*
 * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

/*
 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
 * Copyright (c) 2014 Integros [integros.com]
 */

/*
 * The zfs plug-in routines for GRUB are:
 *
 * zfs_mount() - locates a valid uberblock of the root pool and reads
 *		in its MOS at the memory address MOS.
 *
 * zfs_open() - locates a plain file object by following the MOS
 *		and places its dnode at the memory address DNODE.
 *
 * zfs_read() - read in the data blocks pointed by the DNODE.
 *
 * ZFS_SCRATCH is used as a working area.
 *
 * (memory addr)   MOS      DNODE	ZFS_SCRATCH
 *		    |         |          |
 *	    +-------V---------V----------V---------------+
 *   memory |       | dnode   | dnode    |  scratch      |
 *	    |       | 512B    | 512B     |  area         |
 *	    +--------------------------------------------+
 */

#ifdef	FSYS_ZFS

#include "shared.h"
#include "filesys.h"
#include "fsys_zfs.h"

/* cache for a file block of the currently zfs_open()-ed file */
static void *file_buf = NULL;
static uint64_t file_start = 0;
static uint64_t file_end = 0;

/* cache for a dnode block */
static dnode_phys_t *dnode_buf = NULL;
static dnode_phys_t *dnode_mdn = NULL;
static uint64_t dnode_start = 0;
static uint64_t dnode_end = 0;

static uint64_t pool_guid = 0;
static uberblock_t current_uberblock;
static char *stackbase;

decomp_entry_t decomp_table[ZIO_COMPRESS_FUNCTIONS] =
{
	{"inherit", 0},			/* ZIO_COMPRESS_INHERIT */
	{"on", lzjb_decompress}, 	/* ZIO_COMPRESS_ON */
	{"off", 0},			/* ZIO_COMPRESS_OFF */
	{"lzjb", lzjb_decompress},	/* ZIO_COMPRESS_LZJB */
	{"empty", 0},			/* ZIO_COMPRESS_EMPTY */
	{"gzip-1", 0},			/* ZIO_COMPRESS_GZIP_1 */
	{"gzip-2", 0},			/* ZIO_COMPRESS_GZIP_2 */
	{"gzip-3", 0},			/* ZIO_COMPRESS_GZIP_3 */
	{"gzip-4", 0},			/* ZIO_COMPRESS_GZIP_4 */
	{"gzip-5", 0},			/* ZIO_COMPRESS_GZIP_5 */
	{"gzip-6", 0},			/* ZIO_COMPRESS_GZIP_6 */
	{"gzip-7", 0},			/* ZIO_COMPRESS_GZIP_7 */
	{"gzip-8", 0},			/* ZIO_COMPRESS_GZIP_8 */
	{"gzip-9", 0},			/* ZIO_COMPRESS_GZIP_9 */
	{"zle", 0},			/* ZIO_COMPRESS_ZLE */
	{"lz4", lz4_decompress}		/* ZIO_COMPRESS_LZ4 */
};

static int zio_read_data(blkptr_t *bp, void *buf, char *stack);

/*
 * Our own version of bcmp().
 */
static int
zfs_bcmp(const void *s1, const void *s2, size_t n)
{
	const uchar_t *ps1 = s1;
	const uchar_t *ps2 = s2;

	if (s1 != s2 && n != 0) {
		do {
			if (*ps1++ != *ps2++)
				return (1);
		} while (--n != 0);
	}

	return (0);
}

/*
 * Our own version of log2().  Same thing as highbit()-1.
 */
static int
zfs_log2(uint64_t num)
{
	int i = 0;

	while (num > 1) {
		i++;
		num = num >> 1;
	}

	return (i);
}

/* Checksum Functions */
static void
zio_checksum_off(const void *buf, uint64_t size, zio_cksum_t *zcp)
{
	ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
}

/* Checksum Table and Values */
zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = {
	{{NULL,			NULL},			0, 0,	"inherit"},
	{{NULL,			NULL},			0, 0,	"on"},
	{{zio_checksum_off,	zio_checksum_off},	0, 0,	"off"},
	{{zio_checksum_SHA256,	zio_checksum_SHA256},	1, 1,	"label"},
	{{zio_checksum_SHA256,	zio_checksum_SHA256},	1, 1,	"gang_header"},
	{{NULL,			NULL},			0, 0,	"zilog"},
	{{fletcher_2_native,	fletcher_2_byteswap},	0, 0,	"fletcher2"},
	{{fletcher_4_native,	fletcher_4_byteswap},	1, 0,	"fletcher4"},
	{{zio_checksum_SHA256,	zio_checksum_SHA256},	1, 0,	"SHA256"},
	{{NULL,			NULL},			0, 0,	"zilog2"},
	{{zio_checksum_off,	zio_checksum_off},	0, 0,	"noparity"},
	{{zio_checksum_SHA512,	NULL},			0, 0,	"SHA512"}
};

/*
 * zio_checksum_verify: Provides support for checksum verification.
 *
 * Fletcher2, Fletcher4, SHA-256 and SHA-512/256 are supported.
 *
 * Return:
 * 	-1 = Failure
 *	 0 = Success
 */
static int
zio_checksum_verify(blkptr_t *bp, char *data, int size)
{
	zio_cksum_t zc = bp->blk_cksum;
	uint32_t checksum = BP_GET_CHECKSUM(bp);
	int byteswap = BP_SHOULD_BYTESWAP(bp);
	zio_eck_t *zec = (zio_eck_t *)(data + size) - 1;
	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
	zio_cksum_t actual_cksum, expected_cksum;

	if (byteswap) {
		grub_printf("byteswap not supported\n");
		return (-1);
	}

	if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL) {
		grub_printf("checksum algorithm %u not supported\n", checksum);
		return (-1);
	}

	if (ci->ci_eck) {
		expected_cksum = zec->zec_cksum;
		zec->zec_cksum = zc;
		ci->ci_func[0](data, size, &actual_cksum);
		zec->zec_cksum = expected_cksum;
		zc = expected_cksum;
	} else {
		ci->ci_func[byteswap](data, size, &actual_cksum);
	}

	if ((actual_cksum.zc_word[0] - zc.zc_word[0]) |
	    (actual_cksum.zc_word[1] - zc.zc_word[1]) |
	    (actual_cksum.zc_word[2] - zc.zc_word[2]) |
	    (actual_cksum.zc_word[3] - zc.zc_word[3]))
		return (-1);

	return (0);
}

/*
 * vdev_label_start returns the physical disk offset (in bytes) of
 * label "l".
 */
static uint64_t
vdev_label_start(uint64_t psize, int l)
{
	return (l * sizeof (vdev_label_t) + (l < VDEV_LABELS / 2 ?
	    0 : psize - VDEV_LABELS * sizeof (vdev_label_t)));
}

/*
 * vdev_uberblock_compare takes two uberblock structures and returns an integer
 * indicating the more recent of the two.
 * 	Return Value = 1 if ub2 is more recent
 * 	Return Value = -1 if ub1 is more recent
 * The most recent uberblock is determined using its transaction number and
 * timestamp.  The uberblock with the highest transaction number is
 * considered "newer".  If the transaction numbers of the two blocks match, the
 * timestamps are compared to determine the "newer" of the two.
 */
static int
vdev_uberblock_compare(uberblock_t *ub1, uberblock_t *ub2)
{
	if (ub1->ub_txg < ub2->ub_txg)
		return (-1);
	if (ub1->ub_txg > ub2->ub_txg)
		return (1);

	if (ub1->ub_timestamp < ub2->ub_timestamp)
		return (-1);
	if (ub1->ub_timestamp > ub2->ub_timestamp)
		return (1);

	return (0);
}

/*
 * Three pieces of information are needed to verify an uberblock: the magic
 * number, the version number, and the checksum.
 *
 * Return:
 *     0 - Success
 *    -1 - Failure
 */
static int
uberblock_verify(uberblock_t *uber, uint64_t ub_size, uint64_t offset)
{
	blkptr_t bp;

	BP_ZERO(&bp);
	BP_SET_CHECKSUM(&bp, ZIO_CHECKSUM_LABEL);
	BP_SET_BYTEORDER(&bp, ZFS_HOST_BYTEORDER);
	ZIO_SET_CHECKSUM(&bp.blk_cksum, offset, 0, 0, 0);

	if (zio_checksum_verify(&bp, (char *)uber, ub_size) != 0)
		return (-1);

	if (uber->ub_magic == UBERBLOCK_MAGIC &&
	    SPA_VERSION_IS_SUPPORTED(uber->ub_version))
		return (0);

	return (-1);
}

/*
 * Find the best uberblock.
 * Return:
 *    Success - Pointer to the best uberblock.
 *    Failure - NULL
 */
static uberblock_t *
find_bestub(char *ub_array, uint64_t ashift, uint64_t sector)
{
	uberblock_t *ubbest = NULL;
	uberblock_t *ubnext;
	uint64_t offset, ub_size;
	int i;

	ub_size = VDEV_UBERBLOCK_SIZE(ashift);

	for (i = 0; i < VDEV_UBERBLOCK_COUNT(ashift); i++) {
		ubnext = (uberblock_t *)ub_array;
		ub_array += ub_size;
		offset = (sector << SPA_MINBLOCKSHIFT) +
		    VDEV_UBERBLOCK_OFFSET(ashift, i);

		if (uberblock_verify(ubnext, ub_size, offset) != 0)
			continue;

		if (ubbest == NULL ||
		    vdev_uberblock_compare(ubnext, ubbest) > 0)
			ubbest = ubnext;
	}

	return (ubbest);
}

/*
 * Read a block of data based on the gang block address dva,
 * and put its data in buf.
 *
 * Return:
 *	0 - success
 *	1 - failure
 */
static int
zio_read_gang(blkptr_t *bp, dva_t *dva, void *buf, char *stack)
{
	zio_gbh_phys_t *zio_gb;
	uint64_t offset, sector;
	blkptr_t tmpbp;
	int i;

	zio_gb = (zio_gbh_phys_t *)stack;
	stack += SPA_GANGBLOCKSIZE;
	offset = DVA_GET_OFFSET(dva);
	sector = DVA_OFFSET_TO_PHYS_SECTOR(offset);

	/* read in the gang block header */
	if (devread(sector, 0, SPA_GANGBLOCKSIZE, (char *)zio_gb) == 0) {
		grub_printf("failed to read in a gang block header\n");
		return (1);
	}

	/* self checksuming the gang block header */
	BP_ZERO(&tmpbp);
	BP_SET_CHECKSUM(&tmpbp, ZIO_CHECKSUM_GANG_HEADER);
	BP_SET_BYTEORDER(&tmpbp, ZFS_HOST_BYTEORDER);
	ZIO_SET_CHECKSUM(&tmpbp.blk_cksum, DVA_GET_VDEV(dva),
	    DVA_GET_OFFSET(dva), bp->blk_birth, 0);
	if (zio_checksum_verify(&tmpbp, (char *)zio_gb, SPA_GANGBLOCKSIZE)) {
		grub_printf("failed to checksum a gang block header\n");
		return (1);
	}

	for (i = 0; i < SPA_GBH_NBLKPTRS; i++) {
		if (BP_IS_HOLE(&zio_gb->zg_blkptr[i]))
			continue;

		if (zio_read_data(&zio_gb->zg_blkptr[i], buf, stack))
			return (1);
		buf += BP_GET_PSIZE(&zio_gb->zg_blkptr[i]);
	}

	return (0);
}

/*
 * Read in a block of raw data to buf.
 *
 * Return:
 *	0 - success
 *	1 - failure
 */
static int
zio_read_data(blkptr_t *bp, void *buf, char *stack)
{
	int i, psize;

	psize = BP_GET_PSIZE(bp);

	/* pick a good dva from the block pointer */
	for (i = 0; i < SPA_DVAS_PER_BP; i++) {
		uint64_t offset, sector;

		if (bp->blk_dva[i].dva_word[0] == 0 &&
		    bp->blk_dva[i].dva_word[1] == 0)
			continue;

		if (DVA_GET_GANG(&bp->blk_dva[i])) {
			if (zio_read_gang(bp, &bp->blk_dva[i], buf, stack) != 0)
				continue;
		} else {
			/* read in a data block */
			offset = DVA_GET_OFFSET(&bp->blk_dva[i]);
			sector = DVA_OFFSET_TO_PHYS_SECTOR(offset);
			if (devread(sector, 0, psize, buf) == 0)
				continue;
		}

		/* verify that the checksum matches */
		if (zio_checksum_verify(bp, buf, psize) == 0) {
			return (0);
		}
	}

	grub_printf("could not read block due to EIO or ECKSUM\n");
	return (1);
}

/*
 * buf must be at least BPE_GET_PSIZE(bp) bytes long (which will never be
 * more than BPE_PAYLOAD_SIZE bytes).
 */
static void
decode_embedded_bp_compressed(const blkptr_t *bp, void *buf)
{
	int psize, i;
	uint8_t *buf8 = buf;
	uint64_t w = 0;
	const uint64_t *bp64 = (const uint64_t *)bp;

	psize = BPE_GET_PSIZE(bp);

	/*
	 * Decode the words of the block pointer into the byte array.
	 * Low bits of first word are the first byte (little endian).
	 */
	for (i = 0; i < psize; i++) {
		if (i % sizeof (w) == 0) {
			/* beginning of a word */
			w = *bp64;
			bp64++;
			if (!BPE_IS_PAYLOADWORD(bp, bp64))
				bp64++;
		}
		buf8[i] = BF64_GET(w, (i % sizeof (w)) * NBBY, NBBY);
	}
}

/*
 * Fill in the buffer with the (decompressed) payload of the embedded
 * blkptr_t.  Takes into account compression and byteorder (the payload is
 * treated as a stream of bytes).
 * Return 0 on success, or ENOSPC if it won't fit in the buffer.
 */
static int
decode_embedded_bp(const blkptr_t *bp, void *buf)
{
	int comp;
	int lsize, psize;
	uint8_t *dst = buf;
	uint64_t w = 0;

	lsize = BPE_GET_LSIZE(bp);
	psize = BPE_GET_PSIZE(bp);
	comp = BP_GET_COMPRESS(bp);

	if (comp != ZIO_COMPRESS_OFF) {
		uint8_t dstbuf[BPE_PAYLOAD_SIZE];

		if ((unsigned int)comp >= ZIO_COMPRESS_FUNCTIONS ||
		    decomp_table[comp].decomp_func == NULL) {
			grub_printf("compression algorithm not supported\n");
			return (ERR_FSYS_CORRUPT);
		}

		decode_embedded_bp_compressed(bp, dstbuf);
		decomp_table[comp].decomp_func(dstbuf, buf, psize, lsize);
	} else {
		decode_embedded_bp_compressed(bp, buf);
	}

	return (0);
}

/*
 * Read in a block of data, verify its checksum, decompress if needed,
 * and put the uncompressed data in buf.
 *
 * Return:
 *	0 - success
 *	errnum - failure
 */
static int
zio_read(blkptr_t *bp, void *buf, char *stack)
{
	int lsize, psize, comp;
	char *retbuf;

	if (BP_IS_EMBEDDED(bp)) {
		if (BPE_GET_ETYPE(bp) != BP_EMBEDDED_TYPE_DATA) {
			grub_printf("unsupported embedded BP (type=%u)\n",
			    (int)BPE_GET_ETYPE(bp));
			return (ERR_FSYS_CORRUPT);
		}
		return (decode_embedded_bp(bp, buf));
	}

	comp = BP_GET_COMPRESS(bp);
	lsize = BP_GET_LSIZE(bp);
	psize = BP_GET_PSIZE(bp);

	if ((unsigned int)comp >= ZIO_COMPRESS_FUNCTIONS ||
	    (comp != ZIO_COMPRESS_OFF &&
	    decomp_table[comp].decomp_func == NULL)) {
		grub_printf("compression algorithm not supported\n");
		return (ERR_FSYS_CORRUPT);
	}

	if ((char *)buf < stack && ((char *)buf) + lsize > stack) {
		grub_printf("not enough memory to fit %u bytes on stack\n",
		    lsize);
		return (ERR_WONT_FIT);
	}

	retbuf = buf;
	if (comp != ZIO_COMPRESS_OFF) {
		buf = stack;
		stack += psize;
	}

	if (zio_read_data(bp, buf, stack) != 0) {
		grub_printf("zio_read_data failed\n");
		return (ERR_FSYS_CORRUPT);
	}

	if (comp != ZIO_COMPRESS_OFF) {
		if (decomp_table[comp].decomp_func(buf, retbuf, psize,
		    lsize) != 0) {
			grub_printf("zio_read decompression failed\n");
			return (ERR_FSYS_CORRUPT);
		}
	}

	return (0);
}

/*
 * Get the block from a block id.
 * push the block onto the stack.
 *
 * Return:
 * 	0 - success
 * 	errnum - failure
 */
static int
dmu_read(dnode_phys_t *dn, uint64_t blkid, void *buf, char *stack)
{
	int idx, level;
	blkptr_t *bp_array = dn->dn_blkptr;
	int epbs = dn->dn_indblkshift - SPA_BLKPTRSHIFT;
	blkptr_t *bp, *tmpbuf;

	bp = (blkptr_t *)stack;
	stack += sizeof (blkptr_t);

	tmpbuf = (blkptr_t *)stack;
	stack += 1<<dn->dn_indblkshift;

	for (level = dn->dn_nlevels - 1; level >= 0; level--) {
		idx = (blkid >> (epbs * level)) & ((1<<epbs)-1);
		*bp = bp_array[idx];
		if (level == 0)
			tmpbuf = buf;
		if (BP_IS_HOLE(bp)) {
			grub_memset(buf, 0,
			    dn->dn_datablkszsec << SPA_MINBLOCKSHIFT);
			break;
		} else if (errnum = zio_read(bp, tmpbuf, stack)) {
			return (errnum);
		}

		bp_array = tmpbuf;
	}

	return (0);
}

/*
 * mzap_lookup: Looks up property described by "name" and returns the value
 * in "value".
 *
 * Return:
 *	0 - success
 *	errnum - failure
 */
static int
mzap_lookup(mzap_phys_t *zapobj, int objsize, const char *name,
    uint64_t *value)
{
	int i, chunks;
	mzap_ent_phys_t *mzap_ent = zapobj->mz_chunk;

	chunks = objsize / MZAP_ENT_LEN - 1;
	for (i = 0; i < chunks; i++) {
		if (grub_strcmp(mzap_ent[i].mze_name, name) == 0) {
			*value = mzap_ent[i].mze_value;
			return (0);
		}
	}

	return (ERR_FSYS_CORRUPT);
}

static uint64_t
zap_hash(uint64_t salt, const char *name)
{
	static uint64_t table[256];
	const uint8_t *cp;
	uint8_t c;
	uint64_t crc = salt;

	if (table[128] == 0) {
		uint64_t *ct;
		int i, j;
		for (i = 0; i < 256; i++) {
			for (ct = table + i, *ct = i, j = 8; j > 0; j--)
				*ct = (*ct >> 1) ^ (-(*ct & 1) &
				    ZFS_CRC64_POLY);
		}
	}

	if (crc == 0 || table[128] != ZFS_CRC64_POLY) {
		errnum = ERR_FSYS_CORRUPT;
		return (0);
	}

	for (cp = (const uint8_t *)name; (c = *cp) != '\0'; cp++)
		crc = (crc >> 8) ^ table[(crc ^ c) & 0xFF];

	/*
	 * Only use 28 bits, since we need 4 bits in the cookie for the
	 * collision differentiator.  We MUST use the high bits, since
	 * those are the ones that we first pay attention to when
	 * choosing the bucket.
	 */
	crc &= ~((1ULL << (64 - 28)) - 1);

	return (crc);
}

/*
 * Only to be used on 8-bit arrays.
 * array_len is actual len in bytes (not encoded le_value_length).
 * buf is null-terminated.
 */
static int
zap_leaf_array_equal(zap_leaf_phys_t *l, int blksft, int chunk,
    int array_len, const char *buf)
{
	int bseen = 0;

	while (bseen < array_len) {
		struct zap_leaf_array *la =
		    &ZAP_LEAF_CHUNK(l, blksft, chunk).l_array;
		int toread = MIN(array_len - bseen, ZAP_LEAF_ARRAY_BYTES);

		if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft))
			return (0);

		if (zfs_bcmp(la->la_array, buf + bseen, toread) != 0)
			break;
		chunk = la->la_next;
		bseen += toread;
	}
	return (bseen == array_len);
}

/*
 * Given a zap_leaf_phys_t, walk thru the zap leaf chunks to get the
 * value for the property "name".
 *
 * Return:
 *	0 - success
 *	errnum - failure
 */
static int
zap_leaf_lookup(zap_leaf_phys_t *l, int blksft, uint64_t h,
    const char *name, uint64_t *value)
{
	uint16_t chunk;
	struct zap_leaf_entry *le;

	/* Verify if this is a valid leaf block */
	if (l->l_hdr.lh_block_type != ZBT_LEAF)
		return (ERR_FSYS_CORRUPT);
	if (l->l_hdr.lh_magic != ZAP_LEAF_MAGIC)
		return (ERR_FSYS_CORRUPT);

	for (chunk = l->l_hash[LEAF_HASH(blksft, h)];
	    chunk != CHAIN_END; chunk = le->le_next) {

		if (chunk >= ZAP_LEAF_NUMCHUNKS(blksft))
			return (ERR_FSYS_CORRUPT);

		le = ZAP_LEAF_ENTRY(l, blksft, chunk);

		/* Verify the chunk entry */
		if (le->le_type != ZAP_CHUNK_ENTRY)
			return (ERR_FSYS_CORRUPT);

		if (le->le_hash != h)
			continue;

		if (zap_leaf_array_equal(l, blksft, le->le_name_chunk,
		    le->le_name_length, name)) {

			struct zap_leaf_array *la;
			uint8_t *ip;

			if (le->le_int_size != 8 || le->le_value_length != 1)
				return (ERR_FSYS_CORRUPT);

			/* get the uint64_t property value */
			la = &ZAP_LEAF_CHUNK(l, blksft,
			    le->le_value_chunk).l_array;
			ip = la->la_array;

			*value = (uint64_t)ip[0] << 56 | (uint64_t)ip[1] << 48 |
			    (uint64_t)ip[2] << 40 | (uint64_t)ip[3] << 32 |
			    (uint64_t)ip[4] << 24 | (uint64_t)ip[5] << 16 |
			    (uint64_t)ip[6] << 8 | (uint64_t)ip[7];

			return (0);
		}
	}

	return (ERR_FSYS_CORRUPT);
}

/*
 * Fat ZAP lookup
 *
 * Return:
 *	0 - success
 *	errnum - failure
 */
static int
fzap_lookup(dnode_phys_t *zap_dnode, zap_phys_t *zap,
    const char *name, uint64_t *value, char *stack)
{
	zap_leaf_phys_t *l;
	uint64_t hash, idx, blkid;
	int blksft = zfs_log2(zap_dnode->dn_datablkszsec << DNODE_SHIFT);

	/* Verify if this is a fat zap header block */
	if (zap->zap_magic != (uint64_t)ZAP_MAGIC ||
	    zap->zap_flags != 0)
		return (ERR_FSYS_CORRUPT);

	hash = zap_hash(zap->zap_salt, name);
	if (errnum)
		return (errnum);

	/* get block id from index */
	if (zap->zap_ptrtbl.zt_numblks != 0) {
		/* external pointer tables not supported */
		return (ERR_FSYS_CORRUPT);
	}
	idx = ZAP_HASH_IDX(hash, zap->zap_ptrtbl.zt_shift);
	blkid = ((uint64_t *)zap)[idx + (1<<(blksft-3-1))];

	/* Get the leaf block */
	l = (zap_leaf_phys_t *)stack;
	stack += 1<<blksft;
	if ((1<<blksft) < sizeof (zap_leaf_phys_t))
		return (ERR_FSYS_CORRUPT);
	if (errnum = dmu_read(zap_dnode, blkid, l, stack))
		return (errnum);

	return (zap_leaf_lookup(l, blksft, hash, name, value));
}

/*
 * Read in the data of a zap object and find the value for a matching
 * property name.
 *
 * Return:
 *	0 - success
 *	errnum - failure
 */
static int
zap_lookup(dnode_phys_t *zap_dnode, const char *name, uint64_t *val,
    char *stack)
{
	uint64_t block_type;
	int size;
	void *zapbuf;

	/* Read in the first block of the zap object data. */
	zapbuf = stack;
	size = zap_dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
	stack += size;

	if ((errnum = dmu_read(zap_dnode, 0, zapbuf, stack)) != 0)
		return (errnum);

	block_type = *((uint64_t *)zapbuf);

	if (block_type == ZBT_MICRO) {
		return (mzap_lookup(zapbuf, size, name, val));
	} else if (block_type == ZBT_HEADER) {
		/* this is a fat zap */
		return (fzap_lookup(zap_dnode, zapbuf, name,
		    val, stack));
	}

	return (ERR_FSYS_CORRUPT);
}

typedef struct zap_attribute {
	int za_integer_length;
	uint64_t za_num_integers;
	uint64_t za_first_integer;
	char *za_name;
} zap_attribute_t;

typedef int (zap_cb_t)(zap_attribute_t *za, void *arg, char *stack);

static int
zap_iterate(dnode_phys_t *zap_dnode, zap_cb_t *cb, void *arg, char *stack)
{
	uint32_t size = zap_dnode->dn_datablkszsec << SPA_MINBLOCKSHIFT;
	zap_attribute_t za;
	int i;
	mzap_phys_t *mzp = (mzap_phys_t *)stack;
	stack += size;

	if ((errnum = dmu_read(zap_dnode, 0, mzp, stack)) != 0)
		return (errnum);

	/*
	 * Iteration over fatzap objects has not yet been implemented.
	 * If we encounter a pool in which there are more features for
	 * read than can fit inside a microzap (i.e., more than 2048
	 * features for read), we can add support for fatzap iteration.
	 * For now, fail.
	 */
	if (mzp->mz_block_type != ZBT_MICRO) {
		grub_printf("feature information stored in fatzap, pool "
		    "version not supported\n");
		return (1);
	}

	za.za_integer_length = 8;
	za.za_num_integers = 1;
	for (i = 0; i < size / MZAP_ENT_LEN - 1; i++) {
		mzap_ent_phys_t *mzep = &mzp->mz_chunk[i];
		int err;

		za.za_first_integer = mzep->mze_value;
		za.za_name = mzep->mze_name;
		err = cb(&za, arg, stack);
		if (err != 0)
			return (err);
	}

	return (0);
}

/*
 * Get the dnode of an object number from the metadnode of an object set.
 *
 * Input
 *	mdn - metadnode to get the object dnode
 *	objnum - object number for the object dnode
 *	type - if nonzero, object must be of this type
 *	buf - data buffer that holds the returning dnode
 *	stack - scratch area
 *
 * Return:
 *	0 - success
 *	errnum - failure
 */
static int
dnode_get(dnode_phys_t *mdn, uint64_t objnum, uint8_t type, dnode_phys_t *buf,
    char *stack)
{
	uint64_t blkid, blksz; /* the block id this object dnode is in */
	int epbs; /* shift of number of dnodes in a block */
	int idx; /* index within a block */
	dnode_phys_t *dnbuf;

	blksz = mdn->dn_datablkszsec << SPA_MINBLOCKSHIFT;
	epbs = zfs_log2(blksz) - DNODE_SHIFT;
	blkid = objnum >> epbs;
	idx = objnum & ((1<<epbs)-1);

	if (dnode_buf != NULL && dnode_mdn == mdn &&
	    objnum >= dnode_start && objnum < dnode_end) {
		grub_memmove(buf, &dnode_buf[idx], DNODE_SIZE);
		VERIFY_DN_TYPE(buf, type);
		return (0);
	}

	if (dnode_buf && blksz == 1<<DNODE_BLOCK_SHIFT) {
		dnbuf = dnode_buf;
		dnode_mdn = mdn;
		dnode_start = blkid << epbs;
		dnode_end = (blkid + 1) << epbs;
	} else {
		dnbuf = (dnode_phys_t *)stack;
		stack += blksz;
	}

	if (errnum = dmu_read(mdn, blkid, (char *)dnbuf, stack))
		return (errnum);

	grub_memmove(buf, &dnbuf[idx], DNODE_SIZE);
	VERIFY_DN_TYPE(buf, type);

	return (0);
}

/*
 * Check if this is a special file that resides at the top
 * dataset of the pool. Currently this is the GRUB menu,
 * boot signature and boot signature backup.
 * str starts with '/'.
 */
static int
is_top_dataset_file(char *str)
{
	char *tptr;

	if ((tptr = grub_strstr(str, "menu.lst")) &&
	    (tptr[8] == '\0' || tptr[8] == ' ') &&
	    *(tptr-1) == '/')
		return (1);

	if (grub_strncmp(str, BOOTSIGN_DIR"/",
	    grub_strlen(BOOTSIGN_DIR) + 1) == 0)
		return (1);

	if (grub_strcmp(str, BOOTSIGN_BACKUP) == 0)
		return (1);

	return (0);
}

static int
check_feature(zap_attribute_t *za, void *arg, char *stack)
{
	const char **names = arg;
	int i;

	if (za->za_first_integer == 0)
		return (0);

	for (i = 0; names[i] != NULL; i++) {
		if (grub_strcmp(za->za_name, names[i]) == 0) {
			return (0);
		}
	}
	grub_printf("missing feature for read '%s'\n", za->za_name);
	return (ERR_NEWER_VERSION);
}

/*
 * Get the file dnode for a given file name where mdn is the meta dnode
 * for this ZFS object set. When found, place the file dnode in dn.
 * The 'path' argument will be mangled.
 *
 * Return:
 *	0 - success
 *	errnum - failure
 */
static int
dnode_get_path(dnode_phys_t *mdn, char *path, dnode_phys_t *dn,
    char *stack)
{
	uint64_t objnum, version;
	char *cname, ch;

	if (errnum = dnode_get(mdn, MASTER_NODE_OBJ, DMU_OT_MASTER_NODE,
	    dn, stack))
		return (errnum);

	if (errnum = zap_lookup(dn, ZPL_VERSION_STR, &version, stack))
		return (errnum);
	if (version > ZPL_VERSION)
		return (-1);

	if (errnum = zap_lookup(dn, ZFS_ROOT_OBJ, &objnum, stack))
		return (errnum);

	if (errnum = dnode_get(mdn, objnum, DMU_OT_DIRECTORY_CONTENTS,
	    dn, stack))
		return (errnum);

	/* skip leading slashes */
	while (*path == '/')
		path++;

	while (*path && !grub_isspace(*path)) {

		/* get the next component name */
		cname = path;
		while (*path && !grub_isspace(*path) && *path != '/')
			path++;
		ch = *path;
		*path = 0;   /* ensure null termination */

		if (errnum = zap_lookup(dn, cname, &objnum, stack))
			return (errnum);

		objnum = ZFS_DIRENT_OBJ(objnum);
		if (errnum = dnode_get(mdn, objnum, 0, dn, stack))
			return (errnum);

		*path = ch;
		while (*path == '/')
			path++;
	}

	/* We found the dnode for this file. Verify if it is a plain file. */
	VERIFY_DN_TYPE(dn, DMU_OT_PLAIN_FILE_CONTENTS);

	return (0);
}

/*
 * Get the default 'bootfs' property value from the rootpool.
 *
 * Return:
 *	0 - success
 *	errnum -failure
 */
static int
get_default_bootfsobj(dnode_phys_t *mosmdn, uint64_t *obj, char *stack)
{
	uint64_t objnum = 0;
	dnode_phys_t *dn = (dnode_phys_t *)stack;
	stack += DNODE_SIZE;

	if (errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT,
	    DMU_OT_OBJECT_DIRECTORY, dn, stack))
		return (errnum);

	/*
	 * find the object number for 'pool_props', and get the dnode
	 * of the 'pool_props'.
	 */
	if (zap_lookup(dn, DMU_POOL_PROPS, &objnum, stack))
		return (ERR_FILESYSTEM_NOT_FOUND);

	if (errnum = dnode_get(mosmdn, objnum, DMU_OT_POOL_PROPS, dn, stack))
		return (errnum);

	if (zap_lookup(dn, ZPOOL_PROP_BOOTFS, &objnum, stack))
		return (ERR_FILESYSTEM_NOT_FOUND);

	if (!objnum)
		return (ERR_FILESYSTEM_NOT_FOUND);

	*obj = objnum;
	return (0);
}

/*
 * List of pool features that the grub implementation of ZFS supports for
 * read. Note that features that are only required for write do not need
 * to be listed here since grub opens pools in read-only mode.
 *
 * When this list is updated the version number in usr/src/grub/capability
 * must be incremented to ensure the new grub gets installed.
 */
static const char *spa_feature_names[] = {
	"org.illumos:lz4_compress",
	"com.delphix:hole_birth",
	"com.delphix:extensible_dataset",
	"com.delphix:embedded_data",
	"org.open-zfs:large_blocks",
	"org.illumos:sha512",
	NULL
};

/*
 * Checks whether the MOS features that are active are supported by this
 * (GRUB's) implementation of ZFS.
 *
 * Return:
 *	0: Success.
 *	errnum: Failure.
 */
static int
check_mos_features(dnode_phys_t *mosmdn, char *stack)
{
	uint64_t objnum;
	dnode_phys_t *dn;
	uint8_t error = 0;

	dn = (dnode_phys_t *)stack;
	stack += DNODE_SIZE;

	if ((errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT,
	    DMU_OT_OBJECT_DIRECTORY, dn, stack)) != 0)
		return (errnum);

	/*
	 * Find the object number for 'features_for_read' and retrieve its
	 * corresponding dnode. Note that we don't check features_for_write
	 * because GRUB is not opening the pool for write.
	 */
	if ((errnum = zap_lookup(dn, DMU_POOL_FEATURES_FOR_READ, &objnum,
	    stack)) != 0)
		return (errnum);

	if ((errnum = dnode_get(mosmdn, objnum, DMU_OTN_ZAP_METADATA,
	    dn, stack)) != 0)
		return (errnum);

	return (zap_iterate(dn, check_feature, spa_feature_names, stack));
}

/*
 * Given a MOS metadnode, get the metadnode of a given filesystem name (fsname),
 * e.g. pool/rootfs, or a given object number (obj), e.g. the object number
 * of pool/rootfs.
 *
 * If no fsname and no obj are given, return the DSL_DIR metadnode.
 * If fsname is given, return its metadnode and its matching object number.
 * If only obj is given, return the metadnode for this object number.
 *
 * Return:
 *	0 - success
 *	errnum - failure
 */
static int
get_objset_mdn(dnode_phys_t *mosmdn, char *fsname, uint64_t *obj,
    dnode_phys_t *mdn, char *stack)
{
	uint64_t objnum, headobj;
	char *cname, ch;
	blkptr_t *bp;
	objset_phys_t *osp;
	int issnapshot = 0;
	char *snapname;

	if (fsname == NULL && obj) {
		headobj = *obj;
		goto skip;
	}

	if (errnum = dnode_get(mosmdn, DMU_POOL_DIRECTORY_OBJECT,
	    DMU_OT_OBJECT_DIRECTORY, mdn, stack))
		return (errnum);

	if (errnum = zap_lookup(mdn, DMU_POOL_ROOT_DATASET, &objnum,
	    stack))
		return (errnum);

	if (errnum = dnode_get(mosmdn, objnum, 0, mdn, stack))
		return (errnum);

	if (fsname == NULL) {
		headobj =
		    ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj;
		goto skip;
	}

	/* take out the pool name */
	while (*fsname && !grub_isspace(*fsname) && *fsname != '/')
		fsname++;

	while (*fsname && !grub_isspace(*fsname)) {
		uint64_t childobj;

		while (*fsname == '/')
			fsname++;

		cname = fsname;
		while (*fsname && !grub_isspace(*fsname) && *fsname != '/')
			fsname++;
		ch = *fsname;
		*fsname = 0;

		snapname = cname;
		while (*snapname && !grub_isspace(*snapname) && *snapname !=
		    '@')
			snapname++;
		if (*snapname == '@') {
			issnapshot = 1;
			*snapname = 0;
		}
		childobj =
		    ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_child_dir_zapobj;
		if (errnum = dnode_get(mosmdn, childobj,
		    DMU_OT_DSL_DIR_CHILD_MAP, mdn, stack))
			return (errnum);

		if (zap_lookup(mdn, cname, &objnum, stack))
			return (ERR_FILESYSTEM_NOT_FOUND);

		if (errnum = dnode_get(mosmdn, objnum, 0,
		    mdn, stack))
			return (errnum);

		*fsname = ch;
		if (issnapshot)
			*snapname = '@';
	}
	headobj = ((dsl_dir_phys_t *)DN_BONUS(mdn))->dd_head_dataset_obj;
	if (obj)
		*obj = headobj;

skip:
	if (errnum = dnode_get(mosmdn, headobj, 0, mdn, stack))
		return (errnum);
	if (issnapshot) {
		uint64_t snapobj;

		snapobj = ((dsl_dataset_phys_t *)DN_BONUS(mdn))->
		    ds_snapnames_zapobj;

		if (errnum = dnode_get(mosmdn, snapobj,
		    DMU_OT_DSL_DS_SNAP_MAP, mdn, stack))
			return (errnum);
		if (zap_lookup(mdn, snapname + 1, &headobj, stack))
			return (ERR_FILESYSTEM_NOT_FOUND);
		if (errnum = dnode_get(mosmdn, headobj, 0, mdn, stack))
			return (errnum);
		if (obj)
			*obj = headobj;
	}

	bp = &((dsl_dataset_phys_t *)DN_BONUS(mdn))->ds_bp;
	osp = (objset_phys_t *)stack;
	stack += sizeof (objset_phys_t);
	if (errnum = zio_read(bp, osp, stack))
		return (errnum);

	grub_memmove((char *)mdn, (char *)&osp->os_meta_dnode, DNODE_SIZE);

	return (0);
}

/*
 * For a given XDR packed nvlist, verify the first 4 bytes and move on.
 *
 * An XDR packed nvlist is encoded as (comments from nvs_xdr_create) :
 *
 *      encoding method/host endian     (4 bytes)
 *      nvl_version                     (4 bytes)
 *      nvl_nvflag                      (4 bytes)
 *	encoded nvpairs:
 *		encoded size of the nvpair      (4 bytes)
 *		decoded size of the nvpair      (4 bytes)
 *		name string size                (4 bytes)
 *		name string data                (sizeof(NV_ALIGN4(string))
 *		data type                       (4 bytes)
 *		# of elements in the nvpair     (4 bytes)
 *		data
 *      2 zero's for the last nvpair
 *		(end of the entire list)	(8 bytes)
 *
 * Return:
 *	0 - success
 *	1 - failure
 */
static int
nvlist_unpack(char *nvlist, char **out)
{
	/* Verify if the 1st and 2nd byte in the nvlist are valid. */
	if (nvlist[0] != NV_ENCODE_XDR || nvlist[1] != HOST_ENDIAN)
		return (1);

	*out = nvlist + 4;
	return (0);
}

static char *
nvlist_array(char *nvlist, int index)
{
	int i, encode_size;

	for (i = 0; i < index; i++) {
		/* skip the header, nvl_version, and nvl_nvflag */
		nvlist = nvlist + 4 * 2;

		while (encode_size = BSWAP_32(*(uint32_t *)nvlist))
			nvlist += encode_size; /* goto the next nvpair */

		nvlist = nvlist + 4 * 2; /* skip the ending 2 zeros - 8 bytes */
	}

	return (nvlist);
}

/*
 * The nvlist_next_nvpair() function returns a handle to the next nvpair in the
 * list following nvpair. If nvpair is NULL, the first pair is returned. If
 * nvpair is the last pair in the nvlist, NULL is returned.
 */
static char *
nvlist_next_nvpair(char *nvl, char *nvpair)
{
	char *cur, *prev;
	int encode_size;

	if (nvl == NULL)
		return (NULL);

	if (nvpair == NULL) {
		/* skip over nvl_version and nvl_nvflag */
		nvpair = nvl + 4 * 2;
	} else {
		/* skip to the next nvpair */
		encode_size = BSWAP_32(*(uint32_t *)nvpair);
		nvpair += encode_size;
	}

	/* 8 bytes of 0 marks the end of the list */
	if (*(uint64_t *)nvpair == 0)
		return (NULL);

	return (nvpair);
}

/*
 * This function returns 0 on success and 1 on failure. On success, a string
 * containing the name of nvpair is saved in buf.
 */
static int
nvpair_name(char *nvp, char *buf, int buflen)
{
	int len;

	/* skip over encode/decode size */
	nvp += 4 * 2;

	len = BSWAP_32(*(uint32_t *)nvp);
	if (buflen < len + 1)
		return (1);

	grub_memmove(buf, nvp + 4, len);
	buf[len] = '\0';

	return (0);
}

/*
 * This function retrieves the value of the nvpair in the form of enumerated
 * type data_type_t. This is used to determine the appropriate type to pass to
 * nvpair_value().
 */
static int
nvpair_type(char *nvp)
{
	int name_len, type;

	/* skip over encode/decode size */
	nvp += 4 * 2;

	/* skip over name_len */
	name_len = BSWAP_32(*(uint32_t *)nvp);
	nvp += 4;

	/* skip over name */
	nvp = nvp + ((name_len + 3) & ~3); /* align */

	type = BSWAP_32(*(uint32_t *)nvp);

	return (type);
}

static int
nvpair_value(char *nvp, void *val, int valtype, int *nelmp)
{
	int name_len, type, slen;
	char *strval = val;
	uint64_t *intval = val;

	/* skip over encode/decode size */
	nvp += 4 * 2;

	/* skip over name_len */
	name_len = BSWAP_32(*(uint32_t *)nvp);
	nvp += 4;

	/* skip over name */
	nvp = nvp + ((name_len + 3) & ~3); /* align */

	/* skip over type */
	type = BSWAP_32(*(uint32_t *)nvp);
	nvp += 4;

	if (type == valtype) {
		int nelm;

		nelm = BSWAP_32(*(uint32_t *)nvp);
		if (valtype != DATA_TYPE_BOOLEAN && nelm < 1)
			return (1);
		nvp += 4;

		switch (valtype) {
		case DATA_TYPE_BOOLEAN:
			return (0);

		case DATA_TYPE_STRING:
			slen = BSWAP_32(*(uint32_t *)nvp);
			nvp += 4;
			grub_memmove(strval, nvp, slen);
			strval[slen] = '\0';
			return (0);

		case DATA_TYPE_UINT64:
			*intval = BSWAP_64(*(uint64_t *)nvp);
			return (0);

		case DATA_TYPE_NVLIST:
			*(void **)val = (void *)nvp;
			return (0);

		case DATA_TYPE_NVLIST_ARRAY:
			*(void **)val = (void *)nvp;
			if (nelmp)
				*nelmp = nelm;
			return (0);
		}
	}

	return (1);
}

static int
nvlist_lookup_value(char *nvlist, char *name, void *val, int valtype,
    int *nelmp)
{
	char *nvpair;

	for (nvpair = nvlist_next_nvpair(nvlist, NULL);
	    nvpair != NULL;
	    nvpair = nvlist_next_nvpair(nvlist, nvpair)) {
		int name_len = BSWAP_32(*(uint32_t *)(nvpair + 4 * 2));
		char *nvp_name = nvpair + 4 * 3;

		if ((grub_strncmp(nvp_name, name, name_len) == 0) &&
		    nvpair_type(nvpair) == valtype) {
			return (nvpair_value(nvpair, val, valtype, nelmp));
		}
	}
	return (1);
}

/*
 * Check if this vdev is online and is in a good state.
 */
static int
vdev_validate(char *nv)
{
	uint64_t ival;

	if (nvlist_lookup_value(nv, ZPOOL_CONFIG_OFFLINE, &ival,
	    DATA_TYPE_UINT64, NULL) == 0 ||
	    nvlist_lookup_value(nv, ZPOOL_CONFIG_FAULTED, &ival,
	    DATA_TYPE_UINT64, NULL) == 0 ||
	    nvlist_lookup_value(nv, ZPOOL_CONFIG_REMOVED, &ival,
	    DATA_TYPE_UINT64, NULL) == 0)
		return (ERR_DEV_VALUES);

	return (0);
}

/*
 * Get a valid vdev pathname/devid from the boot device.
 * The caller should already allocate MAXPATHLEN memory for bootpath and devid.
 */
static int
vdev_get_bootpath(char *nv, uint64_t inguid, char *devid, char *bootpath,
    int is_spare)
{
	char type[16];

	if (nvlist_lookup_value(nv, ZPOOL_CONFIG_TYPE, &type, DATA_TYPE_STRING,
	    NULL))
		return (ERR_FSYS_CORRUPT);

	if (grub_strcmp(type, VDEV_TYPE_DISK) == 0) {
		uint64_t guid;

		if (vdev_validate(nv) != 0)
			return (ERR_NO_BOOTPATH);

		if (nvlist_lookup_value(nv, ZPOOL_CONFIG_GUID,
		    &guid, DATA_TYPE_UINT64, NULL) != 0)
			return (ERR_NO_BOOTPATH);

		if (guid != inguid)
			return (ERR_NO_BOOTPATH);

		/* for a spare vdev, pick the disk labeled with "is_spare" */
		if (is_spare) {
			uint64_t spare = 0;
			(void) nvlist_lookup_value(nv, ZPOOL_CONFIG_IS_SPARE,
			    &spare, DATA_TYPE_UINT64, NULL);
			if (!spare)
				return (ERR_NO_BOOTPATH);
		}

		if (nvlist_lookup_value(nv, ZPOOL_CONFIG_PHYS_PATH,
		    bootpath, DATA_TYPE_STRING, NULL) != 0)
			bootpath[0] = '\0';

		if (nvlist_lookup_value(nv, ZPOOL_CONFIG_DEVID,
		    devid, DATA_TYPE_STRING, NULL) != 0)
			devid[0] = '\0';

		if (grub_strlen(bootpath) >= MAXPATHLEN ||
		    grub_strlen(devid) >= MAXPATHLEN)
			return (ERR_WONT_FIT);

		return (0);

	} else if (grub_strcmp(type, VDEV_TYPE_MIRROR) == 0 ||
	    grub_strcmp(type, VDEV_TYPE_REPLACING) == 0 ||
	    (is_spare = (grub_strcmp(type, VDEV_TYPE_SPARE) == 0))) {
		int nelm, i;
		char *child;

		if (nvlist_lookup_value(nv, ZPOOL_CONFIG_CHILDREN, &child,
		    DATA_TYPE_NVLIST_ARRAY, &nelm))
			return (ERR_FSYS_CORRUPT);

		for (i = 0; i < nelm; i++) {
			char *child_i;

			child_i = nvlist_array(child, i);
			if (vdev_get_bootpath(child_i, inguid, devid,
			    bootpath, is_spare) == 0)
				return (0);
		}
	}

	return (ERR_NO_BOOTPATH);
}

/*
 * Check the disk label information and retrieve needed vdev name-value pairs.
 *
 * Return:
 *	0 - success
 *	ERR_* - failure
 */
static int
check_pool_label(uint64_t sector, char *stack, char *outdevid,
    char *outpath, uint64_t *outguid, uint64_t *outashift, uint64_t *outversion)
{
	vdev_phys_t *vdev;
	uint64_t pool_state, txg = 0;
	char *nvlist, *nv, *features;
	uint64_t diskguid;

	sector += (VDEV_SKIP_SIZE >> SPA_MINBLOCKSHIFT);

	/* Read in the vdev name-value pair list (112K). */
	if (devread(sector, 0, VDEV_PHYS_SIZE, stack) == 0)
		return (ERR_READ);

	vdev = (vdev_phys_t *)stack;
	stack += sizeof (vdev_phys_t);

	if (nvlist_unpack(vdev->vp_nvlist, &nvlist))
		return (ERR_FSYS_CORRUPT);

	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_STATE, &pool_state,
	    DATA_TYPE_UINT64, NULL))
		return (ERR_FSYS_CORRUPT);

	if (pool_state == POOL_STATE_DESTROYED)
		return (ERR_FILESYSTEM_NOT_FOUND);

	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_NAME,
	    current_rootpool, DATA_TYPE_STRING, NULL))
		return (ERR_FSYS_CORRUPT);

	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_TXG, &txg,
	    DATA_TYPE_UINT64, NULL))
		return (ERR_FSYS_CORRUPT);

	/* not an active device */
	if (txg == 0)
		return (ERR_NO_BOOTPATH);

	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_VERSION, outversion,
	    DATA_TYPE_UINT64, NULL))
		return (ERR_FSYS_CORRUPT);
	if (!SPA_VERSION_IS_SUPPORTED(*outversion))
		return (ERR_NEWER_VERSION);
	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_VDEV_TREE, &nv,
	    DATA_TYPE_NVLIST, NULL))
		return (ERR_FSYS_CORRUPT);
	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_GUID, &diskguid,
	    DATA_TYPE_UINT64, NULL))
		return (ERR_FSYS_CORRUPT);
	if (nvlist_lookup_value(nv, ZPOOL_CONFIG_ASHIFT, outashift,
	    DATA_TYPE_UINT64, NULL) != 0)
		return (ERR_FSYS_CORRUPT);
	if (vdev_get_bootpath(nv, diskguid, outdevid, outpath, 0))
		return (ERR_NO_BOOTPATH);
	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_POOL_GUID, outguid,
	    DATA_TYPE_UINT64, NULL))
		return (ERR_FSYS_CORRUPT);

	if (nvlist_lookup_value(nvlist, ZPOOL_CONFIG_FEATURES_FOR_READ,
	    &features, DATA_TYPE_NVLIST, NULL) == 0) {
		char *nvp;
		char *name = stack;
		stack += MAXNAMELEN;

		for (nvp = nvlist_next_nvpair(features, NULL);
		    nvp != NULL;
		    nvp = nvlist_next_nvpair(features, nvp)) {
			zap_attribute_t za;

			if (nvpair_name(nvp, name, MAXNAMELEN) != 0)
				return (ERR_FSYS_CORRUPT);

			za.za_integer_length = 8;
			za.za_num_integers = 1;
			za.za_first_integer = 1;
			za.za_name = name;
			if (check_feature(&za, spa_feature_names, stack) != 0)
				return (ERR_NEWER_VERSION);
		}
	}

	return (0);
}

/*
 * zfs_mount() locates a valid uberblock of the root pool and read in its MOS
 * to the memory address MOS.
 *
 * Return:
 *	1 - success
 *	0 - failure
 */
int
zfs_mount(void)
{
	char *stack, *ub_array;
	int label = 0;
	uberblock_t *ubbest;
	objset_phys_t *osp;
	char tmp_bootpath[MAXNAMELEN];
	char tmp_devid[MAXNAMELEN];
	uint64_t tmp_guid, ashift, version;
	uint64_t adjpl = (uint64_t)part_length << SPA_MINBLOCKSHIFT;
	int err = errnum; /* preserve previous errnum state */

	/* if it's our first time here, zero the best uberblock out */
	if (best_drive == 0 && best_part == 0 && find_best_root) {
		grub_memset(&current_uberblock, 0, sizeof (uberblock_t));
		pool_guid = 0;
	}

	stackbase = ZFS_SCRATCH;
	stack = stackbase;
	ub_array = stack;
	stack += VDEV_UBERBLOCK_RING;

	osp = (objset_phys_t *)stack;
	stack += sizeof (objset_phys_t);
	adjpl = P2ALIGN(adjpl, (uint64_t)sizeof (vdev_label_t));

	for (label = 0; label < VDEV_LABELS; label++) {

		/*
		 * some eltorito stacks don't give us a size and
		 * we end up setting the size to MAXUINT, further
		 * some of these devices stop working once a single
		 * read past the end has been issued. Checking
		 * for a maximum part_length and skipping the backup
		 * labels at the end of the slice/partition/device
		 * avoids breaking down on such devices.
		 */
		if (part_length == MAXUINT && label == 2)
			break;

		uint64_t sector = vdev_label_start(adjpl,
		    label) >> SPA_MINBLOCKSHIFT;

		/* Read in the uberblock ring (128K). */
		if (devread(sector  +
		    ((VDEV_SKIP_SIZE + VDEV_PHYS_SIZE) >> SPA_MINBLOCKSHIFT),
		    0, VDEV_UBERBLOCK_RING, ub_array) == 0)
			continue;

		if (check_pool_label(sector, stack, tmp_devid,
		    tmp_bootpath, &tmp_guid, &ashift, &version))
			continue;

		if (pool_guid == 0)
			pool_guid = tmp_guid;

		if ((ubbest = find_bestub(ub_array, ashift, sector)) == NULL ||
		    zio_read(&ubbest->ub_rootbp, osp, stack) != 0)
			continue;

		VERIFY_OS_TYPE(osp, DMU_OST_META);

		if (version >= SPA_VERSION_FEATURES &&
		    check_mos_features(&osp->os_meta_dnode, stack) != 0)
			continue;

		if (find_best_root && ((pool_guid != tmp_guid) ||
		    vdev_uberblock_compare(ubbest, &(current_uberblock)) <= 0))
			continue;

		/* Got the MOS. Save it at the memory addr MOS. */
		grub_memmove(MOS, &osp->os_meta_dnode, DNODE_SIZE);
		grub_memmove(&current_uberblock, ubbest, sizeof (uberblock_t));
		grub_memmove(current_bootpath, tmp_bootpath, MAXNAMELEN);
		grub_memmove(current_devid, tmp_devid, grub_strlen(tmp_devid));
		is_zfs_mount = 1;
		return (1);
	}

	/*
	 * While some fs impls. (tftp) rely on setting and keeping
	 * global errnums set, others won't reset it and will break
	 * when issuing rawreads. The goal here is to simply not
	 * have zfs mount attempts impact the previous state.
	 */
	errnum = err;
	return (0);
}

/*
 * zfs_open() locates a file in the rootpool by following the
 * MOS and places the dnode of the file in the memory address DNODE.
 *
 * Return:
 *	1 - success
 *	0 - failure
 */
int
zfs_open(char *filename)
{
	char *stack;
	dnode_phys_t *mdn;

	file_buf = NULL;
	stackbase = ZFS_SCRATCH;
	stack = stackbase;

	mdn = (dnode_phys_t *)stack;
	stack += sizeof (dnode_phys_t);

	dnode_mdn = NULL;
	dnode_buf = (dnode_phys_t *)stack;
	stack += 1<<DNODE_BLOCK_SHIFT;

	/*
	 * menu.lst is placed at the root pool filesystem level,
	 * do not goto 'current_bootfs'.
	 */
	if (is_top_dataset_file(filename)) {
		if (errnum = get_objset_mdn(MOS, NULL, NULL, mdn, stack))
			return (0);

		current_bootfs_obj = 0;
	} else {
		if (current_bootfs[0] == '\0') {
			/* Get the default root filesystem object number */
			if (errnum = get_default_bootfsobj(MOS,
			    &current_bootfs_obj, stack))
				return (0);

			if (errnum = get_objset_mdn(MOS, NULL,
			    &current_bootfs_obj, mdn, stack))
				return (0);
		} else {
			if (errnum = get_objset_mdn(MOS, current_bootfs,
			    &current_bootfs_obj, mdn, stack)) {
				grub_memset(current_bootfs, 0, MAXNAMELEN);
				return (0);
			}
		}
	}

	if (dnode_get_path(mdn, filename, DNODE, stack)) {
		errnum = ERR_FILE_NOT_FOUND;
		return (0);
	}

	/* get the file size and set the file position to 0 */

	/*
	 * For DMU_OT_SA we will need to locate the SIZE attribute
	 * attribute, which could be either in the bonus buffer
	 * or the "spill" block.
	 */
	if (DNODE->dn_bonustype == DMU_OT_SA) {
		sa_hdr_phys_t *sahdrp;
		int hdrsize;

		if (DNODE->dn_bonuslen != 0) {
			sahdrp = (sa_hdr_phys_t *)DN_BONUS(DNODE);
		} else {
			if (DNODE->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
				blkptr_t *bp = &DNODE->dn_spill;
				void *buf;

				buf = (void *)stack;
				stack += BP_GET_LSIZE(bp);

				/* reset errnum to rawread() failure */
				errnum = 0;
				if (zio_read(bp, buf, stack) != 0) {
					return (0);
				}
				sahdrp = buf;
			} else {
				errnum = ERR_FSYS_CORRUPT;
				return (0);
			}
		}
		hdrsize = SA_HDR_SIZE(sahdrp);
		filemax = *(uint64_t *)((char *)sahdrp + hdrsize +
		    SA_SIZE_OFFSET);
	} else {
		filemax = ((znode_phys_t *)DN_BONUS(DNODE))->zp_size;
	}
	filepos = 0;

	dnode_buf = NULL;
	return (1);
}

/*
 * zfs_read reads in the data blocks pointed by the DNODE.
 *
 * Return:
 *	len - the length successfully read in to the buffer
 *	0   - failure
 */
int
zfs_read(char *buf, int len)
{
	char *stack;
	int blksz, length, movesize;

	if (file_buf == NULL) {
		file_buf = stackbase;
		stackbase += SPA_MAXBLOCKSIZE;
		file_start = file_end = 0;
	}
	stack = stackbase;

	/*
	 * If offset is in memory, move it into the buffer provided and return.
	 */
	if (filepos >= file_start && filepos+len <= file_end) {
		grub_memmove(buf, file_buf + filepos - file_start, len);
		filepos += len;
		return (len);
	}

	blksz = DNODE->dn_datablkszsec << SPA_MINBLOCKSHIFT;

	/*
	 * Note: for GRUB, SPA_MAXBLOCKSIZE is 128KB.  There is not enough
	 * memory to allocate the new max blocksize (16MB), so while
	 * GRUB understands the large_blocks on-disk feature, it can't
	 * actually read large blocks.
	 */
	if (blksz > SPA_MAXBLOCKSIZE) {
		grub_printf("blocks larger than 128K are not supported\n");
		return (0);
	}

	/*
	 * Entire Dnode is too big to fit into the space available.  We
	 * will need to read it in chunks.  This could be optimized to
	 * read in as large a chunk as there is space available, but for
	 * now, this only reads in one data block at a time.
	 */
	length = len;
	while (length) {
		/*
		 * Find requested blkid and the offset within that block.
		 */
		uint64_t blkid = filepos / blksz;

		if (errnum = dmu_read(DNODE, blkid, file_buf, stack))
			return (0);

		file_start = blkid * blksz;
		file_end = file_start + blksz;

		movesize = MIN(length, file_end - filepos);

		grub_memmove(buf, file_buf + filepos - file_start,
		    movesize);
		buf += movesize;
		length -= movesize;
		filepos += movesize;
	}

	return (len);
}

/*
 * No-Op
 */
int
zfs_embed(int *start_sector, int needed_sectors)
{
	return (1);
}

#endif /* FSYS_ZFS */