spa.h revision 5ad820458efd0fdb914baff9c1447c22b819fa23
fa9e4066f08beec538e775443c5be79dd423fcabahrens * CDDL HEADER START
fa9e4066f08beec538e775443c5be79dd423fcabahrens * The contents of this file are subject to the terms of the
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock * Common Development and Distribution License (the "License").
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock * You may not use this file except in compliance with the License.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
fa9e4066f08beec538e775443c5be79dd423fcabahrens * See the License for the specific language governing permissions
fa9e4066f08beec538e775443c5be79dd423fcabahrens * and limitations under the License.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * When distributing Covered Code, include this CDDL HEADER in each
fa9e4066f08beec538e775443c5be79dd423fcabahrens * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * If applicable, add the following below this CDDL HEADER, with the
fa9e4066f08beec538e775443c5be79dd423fcabahrens * fields enclosed by brackets "[]" replaced with your own identifying
fa9e4066f08beec538e775443c5be79dd423fcabahrens * information: Portions Copyright [yyyy] [name of copyright owner]
fa9e4066f08beec538e775443c5be79dd423fcabahrens * CDDL HEADER END
c67d9675bbc8392fe45f3a7dfbda1ad4daa1eb07eschrock * Copyright 2006 Sun Microsystems, Inc. All rights reserved.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Use is subject to license terms.
fa9e4066f08beec538e775443c5be79dd423fcabahrens#pragma ident "%Z%%M% %I% %E% SMI"
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern "C" {
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Forward references that lots of things need.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * General-purpose 32-bit and 64-bit bitfield encodings.
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BF32_DECODE(x, low, len) P2PHASE((x) >> (low), 1U << (len))
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BF64_DECODE(x, low, len) P2PHASE((x) >> (low), 1ULL << (len))
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BF32_ENCODE(x, low, len) (P2PHASE((x), 1U << (len)) << (low))
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BF64_ENCODE(x, low, len) (P2PHASE((x), 1ULL << (len)) << (low))
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BF32_GET(x, low, len) BF32_DECODE(x, low, len)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BF64_GET(x, low, len) BF64_DECODE(x, low, len)
fa9e4066f08beec538e775443c5be79dd423fcabahrens * We currently support nine block sizes, from 512 bytes to 128K.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * We could go higher, but the benefits are near-zero and the cost
fa9e4066f08beec538e775443c5be79dd423fcabahrens * of COWing a giant block to modify one byte would become excessive.
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define SPA_BLOCKSIZES (SPA_MAXBLOCKSHIFT - SPA_MINBLOCKSHIFT + 1)
fa9e4066f08beec538e775443c5be79dd423fcabahrens * The DVA size encodings for LSIZE and PSIZE support blocks up to 32MB.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * The ASIZE encoding should be at least 64 times larger (6 more bits)
fa9e4066f08beec538e775443c5be79dd423fcabahrens * to support up to 4-way RAID-Z mirror mode with worst-case gang block
fa9e4066f08beec538e775443c5be79dd423fcabahrens * overhead, three DVAs per bp, plus one more bit in case we do anything
fa9e4066f08beec538e775443c5be79dd423fcabahrens * else that expands the ASIZE.
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define SPA_LSIZEBITS 16 /* LSIZE up to 32M (2^16 * 512) */
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define SPA_PSIZEBITS 16 /* PSIZE up to 32M (2^16 * 512) */
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define SPA_ASIZEBITS 24 /* ASIZE up to 64 times larger */
fa9e4066f08beec538e775443c5be79dd423fcabahrens * All SPA data is represented by 128-bit data virtual addresses (DVAs).
fa9e4066f08beec538e775443c5be79dd423fcabahrens * The members of the dva_t should be considered opaque outside the SPA.
fa9e4066f08beec538e775443c5be79dd423fcabahrenstypedef struct dva {
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Each block has a 256-bit checksum -- strong enough for cryptographic hashes.
fa9e4066f08beec538e775443c5be79dd423fcabahrenstypedef struct zio_cksum {
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Each block is described by its DVAs, time of birth, checksum, etc.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * The word-by-word, bit-by-bit layout of the blkptr is as follows:
fa9e4066f08beec538e775443c5be79dd423fcabahrens * 64 56 48 40 32 24 16 8 0
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * 0 | vdev1 | GRID | ASIZE |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * 1 |G| offset1 |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * 2 | vdev2 | GRID | ASIZE |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * 3 |G| offset2 |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * 4 | vdev3 | GRID | ASIZE |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * 5 |G| offset3 |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * 6 |E| lvl | type | cksum | comp | PSIZE | LSIZE |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * 7 | padding |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * 8 | padding |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * 9 | padding |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * a | birth txg |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * b | fill count |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * c | checksum[0] |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * d | checksum[1] |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * e | checksum[2] |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * f | checksum[3] |
fa9e4066f08beec538e775443c5be79dd423fcabahrens * +-------+-------+-------+-------+-------+-------+-------+-------+
fa9e4066f08beec538e775443c5be79dd423fcabahrens * vdev virtual device ID
fa9e4066f08beec538e775443c5be79dd423fcabahrens * offset offset into virtual device
fa9e4066f08beec538e775443c5be79dd423fcabahrens * LSIZE logical size
fa9e4066f08beec538e775443c5be79dd423fcabahrens * PSIZE physical size (after compression)
fa9e4066f08beec538e775443c5be79dd423fcabahrens * ASIZE allocated size (including RAID-Z parity and gang block headers)
fa9e4066f08beec538e775443c5be79dd423fcabahrens * GRID RAID-Z layout information (reserved for future use)
fa9e4066f08beec538e775443c5be79dd423fcabahrens * cksum checksum function
fa9e4066f08beec538e775443c5be79dd423fcabahrens * comp compression function
fa9e4066f08beec538e775443c5be79dd423fcabahrens * G gang block indicator
fa9e4066f08beec538e775443c5be79dd423fcabahrens * E endianness
fa9e4066f08beec538e775443c5be79dd423fcabahrens * type DMU object type
fa9e4066f08beec538e775443c5be79dd423fcabahrens * lvl level of indirection
fa9e4066f08beec538e775443c5be79dd423fcabahrens * birth txg transaction group in which the block was born
fa9e4066f08beec538e775443c5be79dd423fcabahrens * fill count number of non-zero blocks under this bp
fa9e4066f08beec538e775443c5be79dd423fcabahrens * checksum[4] 256-bit checksum of the data this bp describes
fa9e4066f08beec538e775443c5be79dd423fcabahrenstypedef struct blkptr {
fa9e4066f08beec538e775443c5be79dd423fcabahrens uint64_t blk_prop; /* size, compression, type, etc */
fa9e4066f08beec538e775443c5be79dd423fcabahrens uint64_t blk_pad[3]; /* Extra space for the future */
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define SPA_DVAS_PER_BP 3 /* Number of DVAs in a bp */
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Macros to get and set fields in a bp or DVA.
fa9e4066f08beec538e775443c5be79dd423fcabahrens BF64_GET_SB((dva)->dva_word[0], 0, 24, SPA_MINBLOCKSHIFT, 0)
fa9e4066f08beec538e775443c5be79dd423fcabahrens BF64_SET_SB((dva)->dva_word[0], 0, 24, SPA_MINBLOCKSHIFT, 0, x)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define DVA_GET_GRID(dva) BF64_GET((dva)->dva_word[0], 24, 8)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define DVA_SET_GRID(dva, x) BF64_SET((dva)->dva_word[0], 24, 8, x)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define DVA_GET_VDEV(dva) BF64_GET((dva)->dva_word[0], 32, 32)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define DVA_SET_VDEV(dva, x) BF64_SET((dva)->dva_word[0], 32, 32, x)
fa9e4066f08beec538e775443c5be79dd423fcabahrens BF64_GET_SB((dva)->dva_word[1], 0, 63, SPA_MINBLOCKSHIFT, 0)
fa9e4066f08beec538e775443c5be79dd423fcabahrens BF64_SET_SB((dva)->dva_word[1], 0, 63, SPA_MINBLOCKSHIFT, 0, x)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define DVA_GET_GANG(dva) BF64_GET((dva)->dva_word[1], 63, 1)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define DVA_SET_GANG(dva, x) BF64_SET((dva)->dva_word[1], 63, 1, x)
fa9e4066f08beec538e775443c5be79dd423fcabahrens BF64_GET_SB((bp)->blk_prop, 0, 16, SPA_MINBLOCKSHIFT, 1))
fa9e4066f08beec538e775443c5be79dd423fcabahrens BF64_SET_SB((bp)->blk_prop, 0, 16, SPA_MINBLOCKSHIFT, 1, x)
fa9e4066f08beec538e775443c5be79dd423fcabahrens BF64_GET_SB((bp)->blk_prop, 16, 16, SPA_MINBLOCKSHIFT, 1)
fa9e4066f08beec538e775443c5be79dd423fcabahrens BF64_SET_SB((bp)->blk_prop, 16, 16, SPA_MINBLOCKSHIFT, 1, x)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BP_GET_COMPRESS(bp) BF64_GET((bp)->blk_prop, 32, 8)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BP_SET_COMPRESS(bp, x) BF64_SET((bp)->blk_prop, 32, 8, x)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BP_GET_CHECKSUM(bp) BF64_GET((bp)->blk_prop, 40, 8)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BP_SET_CHECKSUM(bp, x) BF64_SET((bp)->blk_prop, 40, 8, x)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BP_GET_TYPE(bp) BF64_GET((bp)->blk_prop, 48, 8)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BP_SET_TYPE(bp, x) BF64_SET((bp)->blk_prop, 48, 8, x)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BP_GET_LEVEL(bp) BF64_GET((bp)->blk_prop, 56, 5)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BP_SET_LEVEL(bp, x) BF64_SET((bp)->blk_prop, 56, 5, x)
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BP_GET_BYTEORDER(bp) (0 - BF64_GET((bp)->blk_prop, 63, 1))
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BP_SET_BYTEORDER(bp, x) BF64_SET((bp)->blk_prop, 63, 1, x)
fa9e4066f08beec538e775443c5be79dd423fcabahrens (DVA_GET_ASIZE(&(bp)->blk_dva[0]) + DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \
99653d4ee642c6528e88224f12409a5f23060994eschrock ((BP_GET_LEVEL(bp) > 0 || dmu_ot[BP_GET_TYPE(bp)].ot_metadata) ? \
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Note: the byteorder is either 0 or -1, both of which are palindromes.
fa9e4066f08beec538e775443c5be79dd423fcabahrens * This simplifies the endianness handling a bit.
fa9e4066f08beec538e775443c5be79dd423fcabahrens#define BP_SHOULD_BYTESWAP(bp) (BP_GET_BYTEORDER(bp) != ZFS_HOST_BYTEORDER)
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Routines found in spa.c
fa9e4066f08beec538e775443c5be79dd423fcabahrens/* state manipulation functions */
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern int spa_open(const char *pool, spa_t **, void *tag);
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrockextern int spa_get_stats(const char *pool, nvlist_t **config,
0373e76b3c3643df49ef3483e0f293fdea61d8c6bonwickextern int spa_create(const char *pool, nvlist_t *config, const char *altroot);
0373e76b3c3643df49ef3483e0f293fdea61d8c6bonwickextern int spa_import(const char *pool, nvlist_t *config, const char *altroot);
44cd46cadd9aab751dae6a4023c1cb5bf316d274billmextern int spa_export(char *pool, nvlist_t **oldconfig);
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrockextern void spa_async_request(spa_t *spa, int flag);
fa9e4066f08beec538e775443c5be79dd423fcabahrens/* device manipulation */
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern int spa_vdev_add(spa_t *spa, nvlist_t *nvroot);
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrockextern int spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot,
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrockextern int spa_vdev_detach(spa_t *spa, uint64_t guid, int replace_done);
99653d4ee642c6528e88224f12409a5f23060994eschrockextern int spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare);
c67d9675bbc8392fe45f3a7dfbda1ad4daa1eb07eschrockextern int spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath);
99653d4ee642c6528e88224f12409a5f23060994eschrock/* spare state (which is global across all pools) */
fa9e4066f08beec538e775443c5be79dd423fcabahrens/* scrubbing */
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern int spa_scrub(spa_t *spa, pool_scrub_type_t type, boolean_t force);
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern void spa_scrub_restart(spa_t *spa, uint64_t txg);
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrockextern void spa_scrub_throttle(spa_t *spa, int direction);
fa9e4066f08beec538e775443c5be79dd423fcabahrens/* spa syncing */
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern void spa_sync(spa_t *spa, uint64_t txg); /* only for DMU use */
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern void spa_sync_allpools(void);
fa9e4066f08beec538e775443c5be79dd423fcabahrens * SPA configuration functions in spa_config.c
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern void spa_config_sync(void);
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern void spa_config_load(void);
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern void spa_config_set(spa_t *spa, nvlist_t *config);
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern nvlist_t *spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg,
fa9e4066f08beec538e775443c5be79dd423fcabahrens * Miscellaneous SPA routines in spa_misc.c
fa9e4066f08beec538e775443c5be79dd423fcabahrens/* Namespace manipulation */
0373e76b3c3643df49ef3483e0f293fdea61d8c6bonwickextern spa_t *spa_add(const char *name, const char *altroot);
fa9e4066f08beec538e775443c5be79dd423fcabahrens/* Refcount functions */
fa9e4066f08beec538e775443c5be79dd423fcabahrens/* Pool configuration lock */
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrockextern void spa_config_enter(spa_t *spa, krw_t rw, void *tag);
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern boolean_t spa_config_held(spa_t *spa, krw_t rw);
fa9e4066f08beec538e775443c5be79dd423fcabahrens/* Pool vdev add/remove lock */
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern int spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error);
fa9e4066f08beec538e775443c5be79dd423fcabahrens/* Accessor functions */
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern void spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp);
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern struct metaslab_class *spa_metaslab_class_select(spa_t *spa);
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern uint64_t spa_get_asize(spa_t *spa, uint64_t lsize);
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern int spa_busy(void);
fa9e4066f08beec538e775443c5be79dd423fcabahrens/* Miscellaneous support routines */
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern int spa_rename(const char *oldname, const char *newname);
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern boolean_t spa_guid_exists(uint64_t pool_guid, uint64_t device_guid);
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern char *spa_strdup(const char *);
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern void spa_strfree(char *);
d80c45e0f58fa434ba37259ea2e2b12e0380c19abonwickextern void sprintf_blkptr(char *buf, int len, const blkptr_t *bp);
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern void spa_evict_all(void);
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrockextern vdev_t *spa_lookup_by_guid(spa_t *spa, uint64_t guid);
99653d4ee642c6528e88224f12409a5f23060994eschrockextern boolean_t spa_has_spare(spa_t *, uint64_t guid);
99653d4ee642c6528e88224f12409a5f23060994eschrockextern uint64_t bp_get_dasize(spa_t *spa, const blkptr_t *bp);
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock/* error handling */
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrockextern void spa_log_error(spa_t *spa, struct zio *zio);
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrockextern void zfs_ereport_post(const char *class, spa_t *spa, vdev_t *vd,
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrock struct zio *zio, uint64_t stateoroffset, uint64_t length);
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrockextern int spa_get_errlog(spa_t *spa, void *uaddr, size_t *count);
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrockextern void spa_errlog_sync(spa_t *spa, uint64_t txg);
e9dbad6f263d5570ed7ff5443ec5b958af8c24d7eschrockextern int spa_bookmark_name(spa_t *spa, struct zbookmark *zb,
ea8dc4b6d2251b437950c0056bc626b311c73c27eschrockextern void spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub);
fa9e4066f08beec538e775443c5be79dd423fcabahrens/* Initialization and termination */
fa9e4066f08beec538e775443c5be79dd423fcabahrensextern void spa_fini(void);
fa9e4066f08beec538e775443c5be79dd423fcabahrens#endif /* _SYS_SPA_H */