Home | History | Annotate | Download | only in zfs
      1    789    ahrens /*
      2    789    ahrens  * CDDL HEADER START
      3    789    ahrens  *
      4    789    ahrens  * The contents of this file are subject to the terms of the
      5   1544  eschrock  * Common Development and Distribution License (the "License").
      6   1544  eschrock  * You may not use this file except in compliance with the License.
      7    789    ahrens  *
      8    789    ahrens  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
      9    789    ahrens  * or http://www.opensolaris.org/os/licensing.
     10    789    ahrens  * See the License for the specific language governing permissions
     11    789    ahrens  * and limitations under the License.
     12    789    ahrens  *
     13    789    ahrens  * When distributing Covered Code, include this CDDL HEADER in each
     14    789    ahrens  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
     15    789    ahrens  * If applicable, add the following below this CDDL HEADER, with the
     16    789    ahrens  * fields enclosed by brackets "[]" replaced with your own identifying
     17    789    ahrens  * information: Portions Copyright [yyyy] [name of copyright owner]
     18    789    ahrens  *
     19    789    ahrens  * CDDL HEADER END
     20    789    ahrens  */
     21    789    ahrens /*
     22  10614  Jonathan  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
     23    789    ahrens  * Use is subject to license terms.
     24    789    ahrens  */
     25    789    ahrens 
     26    789    ahrens #include <sys/zfs_context.h>
     27    789    ahrens #include <sys/spa.h>
     28    789    ahrens #include <sys/zio.h>
     29    789    ahrens #include <sys/zio_checksum.h>
     30    789    ahrens 
     31    789    ahrens /*
     32    789    ahrens  * Checksum vectors.
     33    789    ahrens  *
     34    789    ahrens  * In the SPA, everything is checksummed.  We support checksum vectors
     35    789    ahrens  * for three distinct reasons:
     36    789    ahrens  *
     37    789    ahrens  *   1. Different kinds of data need different levels of protection.
     38    789    ahrens  *	For SPA metadata, we always want a very strong checksum.
     39    789    ahrens  *	For user data, we let users make the trade-off between speed
     40    789    ahrens  *	and checksum strength.
     41    789    ahrens  *
     42    789    ahrens  *   2. Cryptographic hash and MAC algorithms are an area of active research.
     43    789    ahrens  *	It is likely that in future hash functions will be at least as strong
     44    789    ahrens  *	as current best-of-breed, and may be substantially faster as well.
     45    789    ahrens  *	We want the ability to take advantage of these new hashes as soon as
     46    789    ahrens  *	they become available.
     47    789    ahrens  *
     48    789    ahrens  *   3. If someone develops hardware that can compute a strong hash quickly,
     49    789    ahrens  *	we want the ability to take advantage of that hardware.
     50    789    ahrens  *
     51    789    ahrens  * Of course, we don't want a checksum upgrade to invalidate existing
     52  10922      Jeff  * data, so we store the checksum *function* in eight bits of the bp.
     53  10922      Jeff  * This gives us room for up to 256 different checksum functions.
     54    789    ahrens  *
     55    789    ahrens  * When writing a block, we always checksum it with the latest-and-greatest
     56    789    ahrens  * checksum function of the appropriate strength.  When reading a block,
     57    789    ahrens  * we compare the expected checksum against the actual checksum, which we
     58  10922      Jeff  * compute via the checksum function specified by BP_GET_CHECKSUM(bp).
     59    789    ahrens  */
     60    789    ahrens 
     61    789    ahrens /*ARGSUSED*/
     62    789    ahrens static void
     63    789    ahrens zio_checksum_off(const void *buf, uint64_t size, zio_cksum_t *zcp)
     64    789    ahrens {
     65    789    ahrens 	ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
     66    789    ahrens }
     67    789    ahrens 
     68    789    ahrens zio_checksum_info_t zio_checksum_table[ZIO_CHECKSUM_FUNCTIONS] = {
     69  10922      Jeff 	{{NULL,			NULL},			0, 0, 0, "inherit"},
     70  10922      Jeff 	{{NULL,			NULL},			0, 0, 0, "on"},
     71  10922      Jeff 	{{zio_checksum_off,	zio_checksum_off},	0, 0, 0, "off"},
     72  10922      Jeff 	{{zio_checksum_SHA256,	zio_checksum_SHA256},	1, 1, 0, "label"},
     73  10922      Jeff 	{{zio_checksum_SHA256,	zio_checksum_SHA256},	1, 1, 0, "gang_header"},
     74  10922      Jeff 	{{fletcher_2_native,	fletcher_2_byteswap},	0, 1, 0, "zilog"},
     75  10922      Jeff 	{{fletcher_2_native,	fletcher_2_byteswap},	0, 0, 0, "fletcher2"},
     76  10922      Jeff 	{{fletcher_4_native,	fletcher_4_byteswap},	1, 0, 0, "fletcher4"},
     77  10922      Jeff 	{{zio_checksum_SHA256,	zio_checksum_SHA256},	1, 0, 1, "sha256"},
     78    789    ahrens };
     79    789    ahrens 
     80  10922      Jeff enum zio_checksum
     81  10922      Jeff zio_checksum_select(enum zio_checksum child, enum zio_checksum parent)
     82    789    ahrens {
     83    789    ahrens 	ASSERT(child < ZIO_CHECKSUM_FUNCTIONS);
     84    789    ahrens 	ASSERT(parent < ZIO_CHECKSUM_FUNCTIONS);
     85    789    ahrens 	ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
     86    789    ahrens 
     87    789    ahrens 	if (child == ZIO_CHECKSUM_INHERIT)
     88    789    ahrens 		return (parent);
     89    789    ahrens 
     90    789    ahrens 	if (child == ZIO_CHECKSUM_ON)
     91    789    ahrens 		return (ZIO_CHECKSUM_ON_VALUE);
     92    789    ahrens 
     93    789    ahrens 	return (child);
     94    789    ahrens }
     95    789    ahrens 
     96  10922      Jeff enum zio_checksum
     97  10922      Jeff zio_checksum_dedup_select(spa_t *spa, enum zio_checksum child,
     98  10922      Jeff     enum zio_checksum parent)
     99  10922      Jeff {
    100  10922      Jeff 	ASSERT((child & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
    101  10922      Jeff 	ASSERT((parent & ZIO_CHECKSUM_MASK) < ZIO_CHECKSUM_FUNCTIONS);
    102  10922      Jeff 	ASSERT(parent != ZIO_CHECKSUM_INHERIT && parent != ZIO_CHECKSUM_ON);
    103  10922      Jeff 
    104  10922      Jeff 	if (child == ZIO_CHECKSUM_INHERIT)
    105  10922      Jeff 		return (parent);
    106  10922      Jeff 
    107  10922      Jeff 	if (child == ZIO_CHECKSUM_ON)
    108  10922      Jeff 		return (spa_dedup_checksum(spa));
    109  10922      Jeff 
    110  10922      Jeff 	if (child == (ZIO_CHECKSUM_ON | ZIO_CHECKSUM_VERIFY))
    111  10922      Jeff 		return (spa_dedup_checksum(spa) | ZIO_CHECKSUM_VERIFY);
    112  10922      Jeff 
    113  10922      Jeff 	ASSERT(zio_checksum_table[child & ZIO_CHECKSUM_MASK].ci_dedup ||
    114  10922      Jeff 	    (child & ZIO_CHECKSUM_VERIFY) || child == ZIO_CHECKSUM_OFF);
    115  10922      Jeff 
    116  10922      Jeff 	return (child);
    117  10922      Jeff }
    118  10922      Jeff 
    119    789    ahrens /*
    120   7754      Jeff  * Set the external verifier for a gang block based on <vdev, offset, txg>,
    121   7754      Jeff  * a tuple which is guaranteed to be unique for the life of the pool.
    122   7754      Jeff  */
    123   7754      Jeff static void
    124   7754      Jeff zio_checksum_gang_verifier(zio_cksum_t *zcp, blkptr_t *bp)
    125   7754      Jeff {
    126   7754      Jeff 	dva_t *dva = BP_IDENTITY(bp);
    127  10922      Jeff 	uint64_t txg = BP_PHYSICAL_BIRTH(bp);
    128   7754      Jeff 
    129   7754      Jeff 	ASSERT(BP_IS_GANG(bp));
    130   7754      Jeff 
    131   7754      Jeff 	ZIO_SET_CHECKSUM(zcp, DVA_GET_VDEV(dva), DVA_GET_OFFSET(dva), txg, 0);
    132   7754      Jeff }
    133   7754      Jeff 
    134   7754      Jeff /*
    135   7754      Jeff  * Set the external verifier for a label block based on its offset.
    136   7754      Jeff  * The vdev is implicit, and the txg is unknowable at pool open time --
    137   7754      Jeff  * hence the logic in vdev_uberblock_load() to find the most recent copy.
    138   7754      Jeff  */
    139   7754      Jeff static void
    140   7754      Jeff zio_checksum_label_verifier(zio_cksum_t *zcp, uint64_t offset)
    141   7754      Jeff {
    142   7754      Jeff 	ZIO_SET_CHECKSUM(zcp, offset, 0, 0, 0);
    143   7754      Jeff }
    144   7754      Jeff 
    145   7754      Jeff /*
    146    789    ahrens  * Generate the checksum.
    147    789    ahrens  */
    148    789    ahrens void
    149   7754      Jeff zio_checksum_compute(zio_t *zio, enum zio_checksum checksum,
    150   7754      Jeff 	void *data, uint64_t size)
    151    789    ahrens {
    152   7754      Jeff 	blkptr_t *bp = zio->io_bp;
    153   7754      Jeff 	uint64_t offset = zio->io_offset;
    154    789    ahrens 	zio_block_tail_t *zbt = (zio_block_tail_t *)((char *)data + size) - 1;
    155    789    ahrens 	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
    156    789    ahrens 	zio_cksum_t zbt_cksum;
    157    789    ahrens 
    158   7754      Jeff 	ASSERT((uint_t)checksum < ZIO_CHECKSUM_FUNCTIONS);
    159    789    ahrens 	ASSERT(ci->ci_func[0] != NULL);
    160    789    ahrens 
    161    789    ahrens 	if (ci->ci_zbt) {
    162   7754      Jeff 		if (checksum == ZIO_CHECKSUM_GANG_HEADER)
    163   7754      Jeff 			zio_checksum_gang_verifier(&zbt->zbt_cksum, bp);
    164   7754      Jeff 		else if (checksum == ZIO_CHECKSUM_LABEL)
    165   7754      Jeff 			zio_checksum_label_verifier(&zbt->zbt_cksum, offset);
    166   7754      Jeff 		else
    167   7754      Jeff 			bp->blk_cksum = zbt->zbt_cksum;
    168    789    ahrens 		zbt->zbt_magic = ZBT_MAGIC;
    169    789    ahrens 		ci->ci_func[0](data, size, &zbt_cksum);
    170    789    ahrens 		zbt->zbt_cksum = zbt_cksum;
    171    789    ahrens 	} else {
    172   7754      Jeff 		ci->ci_func[0](data, size, &bp->blk_cksum);
    173    789    ahrens 	}
    174    789    ahrens }
    175    789    ahrens 
    176    789    ahrens int
    177  10614  Jonathan zio_checksum_error(zio_t *zio, zio_bad_cksum_t *info)
    178    789    ahrens {
    179    789    ahrens 	blkptr_t *bp = zio->io_bp;
    180   7754      Jeff 	uint_t checksum = (bp == NULL ? zio->io_prop.zp_checksum :
    181   7754      Jeff 	    (BP_IS_GANG(bp) ? ZIO_CHECKSUM_GANG_HEADER : BP_GET_CHECKSUM(bp)));
    182   7754      Jeff 	int byteswap;
    183  10614  Jonathan 	int error;
    184   7754      Jeff 	uint64_t size = (bp == NULL ? zio->io_size :
    185   7754      Jeff 	    (BP_IS_GANG(bp) ? SPA_GANGBLOCKSIZE : BP_GET_PSIZE(bp)));
    186   7754      Jeff 	uint64_t offset = zio->io_offset;
    187  10614  Jonathan 	void *data = zio->io_data;
    188    789    ahrens 	zio_block_tail_t *zbt = (zio_block_tail_t *)((char *)data + size) - 1;
    189    789    ahrens 	zio_checksum_info_t *ci = &zio_checksum_table[checksum];
    190   7754      Jeff 	zio_cksum_t actual_cksum, expected_cksum, verifier;
    191    789    ahrens 
    192    789    ahrens 	if (checksum >= ZIO_CHECKSUM_FUNCTIONS || ci->ci_func[0] == NULL)
    193    789    ahrens 		return (EINVAL);
    194    789    ahrens 
    195    789    ahrens 	if (ci->ci_zbt) {
    196    789    ahrens 		if (checksum == ZIO_CHECKSUM_GANG_HEADER)
    197   7754      Jeff 			zio_checksum_gang_verifier(&verifier, bp);
    198   7754      Jeff 		else if (checksum == ZIO_CHECKSUM_LABEL)
    199   7754      Jeff 			zio_checksum_label_verifier(&verifier, offset);
    200   7754      Jeff 		else
    201   7754      Jeff 			verifier = bp->blk_cksum;
    202    789    ahrens 
    203   7754      Jeff 		byteswap = (zbt->zbt_magic == BSWAP_64(ZBT_MAGIC));
    204   7754      Jeff 
    205   7754      Jeff 		if (byteswap)
    206   7754      Jeff 			byteswap_uint64_array(&verifier, sizeof (zio_cksum_t));
    207   7754      Jeff 
    208   7754      Jeff 		expected_cksum = zbt->zbt_cksum;
    209   7754      Jeff 		zbt->zbt_cksum = verifier;
    210   7754      Jeff 		ci->ci_func[byteswap](data, size, &actual_cksum);
    211   7754      Jeff 		zbt->zbt_cksum = expected_cksum;
    212   7754      Jeff 
    213   7754      Jeff 		if (byteswap)
    214    789    ahrens 			byteswap_uint64_array(&expected_cksum,
    215    789    ahrens 			    sizeof (zio_cksum_t));
    216    789    ahrens 	} else {
    217   1775     billm 		ASSERT(!BP_IS_GANG(bp));
    218   7754      Jeff 		byteswap = BP_SHOULD_BYTESWAP(bp);
    219   7754      Jeff 		expected_cksum = bp->blk_cksum;
    220    789    ahrens 		ci->ci_func[byteswap](data, size, &actual_cksum);
    221    789    ahrens 	}
    222    789    ahrens 
    223  10614  Jonathan 	info->zbc_expected = expected_cksum;
    224  10614  Jonathan 	info->zbc_actual = actual_cksum;
    225  10614  Jonathan 	info->zbc_checksum_name = ci->ci_name;
    226  10614  Jonathan 	info->zbc_byteswapped = byteswap;
    227  10614  Jonathan 	info->zbc_injected = 0;
    228  10614  Jonathan 	info->zbc_has_cksum = 1;
    229  10614  Jonathan 
    230   7754      Jeff 	if (!ZIO_CHECKSUM_EQUAL(actual_cksum, expected_cksum))
    231    789    ahrens 		return (ECKSUM);
    232    789    ahrens 
    233  10614  Jonathan 	if (zio_injection_enabled && !zio->io_error &&
    234  10614  Jonathan 	    (error = zio_handle_fault_injection(zio, ECKSUM)) != 0) {
    235  10614  Jonathan 
    236  10614  Jonathan 		info->zbc_injected = 1;
    237  10614  Jonathan 		return (error);
    238  10614  Jonathan 	}
    239   1544  eschrock 
    240    789    ahrens 	return (0);
    241    789    ahrens }
    242