include/sys/spa.h

   1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 /*
  22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc.  All rights reserved.
  25  * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
  26  * Copyright 2013 Saso Kiselkov. All rights reserved.
  27  */
  28
  29 #ifndef _SYS_SPA_H
  30 #define _SYS_SPA_H
  31
  32 #include <sys/avl.h>
  33 #include <sys/zfs_context.h>
  34 #include <sys/kstat.h>
  35 #include <sys/nvpair.h>
  36 #include <sys/sysmacros.h>
  37 #include <sys/types.h>
  38 #include <sys/fs/zfs.h>
  39 #include <sys/spa_checksum.h>
  40 #include <sys/dmu.h>
  41
  42 #ifdef  __cplusplus
  43 extern "C" {
  44 #endif
  45
  46 /*
  47  * Forward references that lots of things need.
  48  */
  49 typedef struct spa spa_t;
  50 typedef struct vdev vdev_t;
  51 typedef struct metaslab metaslab_t;
  52 typedef struct metaslab_group metaslab_group_t;
  53 typedef struct metaslab_class metaslab_class_t;
  54 typedef struct zio zio_t;
  55 typedef struct zilog zilog_t;
  56 typedef struct spa_aux_vdev spa_aux_vdev_t;
  57 typedef struct ddt ddt_t;
  58 typedef struct ddt_entry ddt_entry_t;
  59 typedef struct zbookmark_phys zbookmark_phys_t;
  60
  61 struct dsl_pool;
  62 struct dsl_dataset;
  63
  64 /*
  65  * General-purpose 32-bit and 64-bit bitfield encodings.
  66  */
  67 #define BF32_DECODE(x, low, len)        P2PHASE((x) >> (low), 1U << (len))
  68 #define BF64_DECODE(x, low, len)        P2PHASE((x) >> (low), 1ULL << (len))
  69 #define BF32_ENCODE(x, low, len)        (P2PHASE((x), 1U << (len)) << (low))
  70 #define BF64_ENCODE(x, low, len)        (P2PHASE((x), 1ULL << (len)) << (low))
  71
  72 #define BF32_GET(x, low, len)           BF32_DECODE(x, low, len)
  73 #define BF64_GET(x, low, len)           BF64_DECODE(x, low, len)
  74
  75 #define BF32_SET(x, low, len, val) do { \
  76         ASSERT3U(val, <, 1U << (len)); \
  77         ASSERT3U(low + len, <=, 32); \
  78         (x) ^= BF32_ENCODE((x >> low) ^ (val), low, len); \
  79 _NOTE(CONSTCOND) } while (0)
  80
  81 #define BF64_SET(x, low, len, val) do { \
  82         ASSERT3U(val, <, 1ULL << (len)); \
  83         ASSERT3U(low + len, <=, 64); \
  84         ((x) ^= BF64_ENCODE((x >> low) ^ (val), low, len)); \
  85 _NOTE(CONSTCOND) } while (0)
  86
  87 #define BF32_GET_SB(x, low, len, shift, bias)   \
  88         ((BF32_GET(x, low, len) + (bias)) << (shift))
  89 #define BF64_GET_SB(x, low, len, shift, bias)   \
  90         ((BF64_GET(x, low, len) + (bias)) << (shift))
  91
  92 #define BF32_SET_SB(x, low, len, shift, bias, val) do { \
  93         ASSERT(IS_P2ALIGNED(val, 1U << shift)); \
  94         ASSERT3S((val) >> (shift), >=, bias); \
  95         BF32_SET(x, low, len, ((val) >> (shift)) - (bias)); \
  96 _NOTE(CONSTCOND) } while (0)
  97 #define BF64_SET_SB(x, low, len, shift, bias, val) do { \
  98         ASSERT(IS_P2ALIGNED(val, 1ULL << shift)); \
  99         ASSERT3S((val) >> (shift), >=, bias); \
 100         BF64_SET(x, low, len, ((val) >> (shift)) - (bias)); \
 101 _NOTE(CONSTCOND) } while (0)
 102
 103 /*
 104  * We currently support block sizes from 512 bytes to 16MB.
 105  * The benefits of larger blocks, and thus larger IO, need to be weighed
 106  * against the cost of COWing a giant block to modify one byte, and the
 107  * large latency of reading or writing a large block.
 108  *
 109  * Note that although blocks up to 16MB are supported, the recordsize
 110  * property can not be set larger than zfs_max_recordsize (default 1MB).
 111  * See the comment near zfs_max_recordsize in dsl_dataset.c for details.
 112  *
 113  * Note that although the LSIZE field of the blkptr_t can store sizes up
 114  * to 32MB, the dnode's dn_datablkszsec can only store sizes up to
 115  * 32MB - 512 bytes.  Therefore, we limit SPA_MAXBLOCKSIZE to 16MB.
 116  */
 117 #define SPA_MINBLOCKSHIFT       9
 118 #define SPA_OLD_MAXBLOCKSHIFT   17
 119 #define SPA_MAXBLOCKSHIFT       24
 120 #define SPA_MINBLOCKSIZE        (1ULL << SPA_MINBLOCKSHIFT)
 121 #define SPA_OLD_MAXBLOCKSIZE    (1ULL << SPA_OLD_MAXBLOCKSHIFT)
 122 #define SPA_MAXBLOCKSIZE        (1ULL << SPA_MAXBLOCKSHIFT)
 123
 124 /*
 125  * Size of block to hold the configuration data (a packed nvlist)
 126  */
 127 #define SPA_CONFIG_BLOCKSIZE    (1ULL << 14)
 128
 129 /*
 130  * The DVA size encodings for LSIZE and PSIZE support blocks up to 32MB.
 131  * The ASIZE encoding should be at least 64 times larger (6 more bits)
 132  * to support up to 4-way RAID-Z mirror mode with worst-case gang block
 133  * overhead, three DVAs per bp, plus one more bit in case we do anything
 134  * else that expands the ASIZE.
 135  */
 136 #define SPA_LSIZEBITS           16      /* LSIZE up to 32M (2^16 * 512) */
 137 #define SPA_PSIZEBITS           16      /* PSIZE up to 32M (2^16 * 512) */
 138 #define SPA_ASIZEBITS           24      /* ASIZE up to 64 times larger  */
 139
 140 #define SPA_COMPRESSBITS        7
 141
 142 /*
 143  * All SPA data is represented by 128-bit data virtual addresses (DVAs).
 144  * The members of the dva_t should be considered opaque outside the SPA.
 145  */
 146 typedef struct dva {
 147         uint64_t        dva_word[2];
 148 } dva_t;
 149
 150
 151 /*
 152  * Some checksums/hashes need a 256-bit initialization salt. This salt is kept
 153  * secret and is suitable for use in MAC algorithms as the key.
 154  */
 155 typedef struct zio_cksum_salt {
 156         uint8_t         zcs_bytes[32];
 157 } zio_cksum_salt_t;
 158
 159 /*
 160  * Each block is described by its DVAs, time of birth, checksum, etc.
 161  * The word-by-word, bit-by-bit layout of the blkptr is as follows:
 162  *
 163  *      64      56      48      40      32      24      16      8       0
 164  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 165  * 0    |               vdev1           | GRID  |         ASIZE         |
 166  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 167  * 1    |G|                      offset1                                |
 168  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 169  * 2    |               vdev2           | GRID  |         ASIZE         |
 170  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 171  * 3    |G|                      offset2                                |
 172  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 173  * 4    |               vdev3           | GRID  |         ASIZE         |
 174  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 175  * 5    |G|                      offset3                                |
 176  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 177  * 6    |BDX|lvl| type  | cksum |E| comp|    PSIZE      |     LSIZE     |
 178  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 179  * 7    |                       padding                                 |
 180  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 181  * 8    |                       padding                                 |
 182  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 183  * 9    |                       physical birth txg                      |
 184  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 185  * a    |                       logical birth txg                       |
 186  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 187  * b    |                       fill count                              |
 188  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 189  * c    |                       checksum[0]                             |
 190  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 191  * d    |                       checksum[1]                             |
 192  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 193  * e    |                       checksum[2]                             |
 194  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 195  * f    |                       checksum[3]                             |
 196  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 197  *
 198  * Legend:
 199  *
 200  * vdev         virtual device ID
 201  * offset       offset into virtual device
 202  * LSIZE        logical size
 203  * PSIZE        physical size (after compression)
 204  * ASIZE        allocated size (including RAID-Z parity and gang block headers)
 205  * GRID         RAID-Z layout information (reserved for future use)
 206  * cksum        checksum function
 207  * comp         compression function
 208  * G            gang block indicator
 209  * B            byteorder (endianness)
 210  * D            dedup
 211  * X            encryption (on version 30, which is not supported)
 212  * E            blkptr_t contains embedded data (see below)
 213  * lvl          level of indirection
 214  * type         DMU object type
 215  * phys birth   txg of block allocation; zero if same as logical birth txg
 216  * log. birth   transaction group in which the block was logically born
 217  * fill count   number of non-zero blocks under this bp
 218  * checksum[4]  256-bit checksum of the data this bp describes
 219  */
 220
 221 /*
 222  * "Embedded" blkptr_t's don't actually point to a block, instead they
 223  * have a data payload embedded in the blkptr_t itself.  See the comment
 224  * in blkptr.c for more details.
 225  *
 226  * The blkptr_t is laid out as follows:
 227  *
 228  *      64      56      48      40      32      24      16      8       0
 229  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 230  * 0    |      payload                                                  |
 231  * 1    |      payload                                                  |
 232  * 2    |      payload                                                  |
 233  * 3    |      payload                                                  |
 234  * 4    |      payload                                                  |
 235  * 5    |      payload                                                  |
 236  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 237  * 6    |BDX|lvl| type  | etype |E| comp| PSIZE|              LSIZE     |
 238  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 239  * 7    |      payload                                                  |
 240  * 8    |      payload                                                  |
 241  * 9    |      payload                                                  |
 242  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 243  * a    |                       logical birth txg                       |
 244  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 245  * b    |      payload                                                  |
 246  * c    |      payload                                                  |
 247  * d    |      payload                                                  |
 248  * e    |      payload                                                  |
 249  * f    |      payload                                                  |
 250  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 251  *
 252  * Legend:
 253  *
 254  * payload              contains the embedded data
 255  * B (byteorder)        byteorder (endianness)
 256  * D (dedup)            padding (set to zero)
 257  * X                    encryption (set to zero; see above)
 258  * E (embedded)         set to one
 259  * lvl                  indirection level
 260  * type                 DMU object type
 261  * etype                how to interpret embedded data (BP_EMBEDDED_TYPE_*)
 262  * comp                 compression function of payload
 263  * PSIZE                size of payload after compression, in bytes
 264  * LSIZE                logical size of payload, in bytes
 265  *                      note that 25 bits is enough to store the largest
 266  *                      "normal" BP's LSIZE (2^16 * 2^9) in bytes
 267  * log. birth           transaction group in which the block was logically born
 268  *
 269  * Note that LSIZE and PSIZE are stored in bytes, whereas for non-embedded
 270  * bp's they are stored in units of SPA_MINBLOCKSHIFT.
 271  * Generally, the generic BP_GET_*() macros can be used on embedded BP's.
 272  * The B, D, X, lvl, type, and comp fields are stored the same as with normal
 273  * BP's so the BP_SET_* macros can be used with them.  etype, PSIZE, LSIZE must
 274  * be set with the BPE_SET_* macros.  BP_SET_EMBEDDED() should be called before
 275  * other macros, as they assert that they are only used on BP's of the correct
 276  * "embedded-ness".
 277  */
 278
 279 #define BPE_GET_ETYPE(bp)       \
 280         (ASSERT(BP_IS_EMBEDDED(bp)), \
 281         BF64_GET((bp)->blk_prop, 40, 8))
 282 #define BPE_SET_ETYPE(bp, t)    do { \
 283         ASSERT(BP_IS_EMBEDDED(bp)); \
 284         BF64_SET((bp)->blk_prop, 40, 8, t); \
 285 _NOTE(CONSTCOND) } while (0)
 286
 287 #define BPE_GET_LSIZE(bp)       \
 288         (ASSERT(BP_IS_EMBEDDED(bp)), \
 289         BF64_GET_SB((bp)->blk_prop, 0, 25, 0, 1))
 290 #define BPE_SET_LSIZE(bp, x)    do { \
 291         ASSERT(BP_IS_EMBEDDED(bp)); \
 292         BF64_SET_SB((bp)->blk_prop, 0, 25, 0, 1, x); \
 293 _NOTE(CONSTCOND) } while (0)
 294
 295 #define BPE_GET_PSIZE(bp)       \
 296         (ASSERT(BP_IS_EMBEDDED(bp)), \
 297         BF64_GET_SB((bp)->blk_prop, 25, 7, 0, 1))
 298 #define BPE_SET_PSIZE(bp, x)    do { \
 299         ASSERT(BP_IS_EMBEDDED(bp)); \
 300         BF64_SET_SB((bp)->blk_prop, 25, 7, 0, 1, x); \
 301 _NOTE(CONSTCOND) } while (0)
 302
 303 typedef enum bp_embedded_type {
 304         BP_EMBEDDED_TYPE_DATA,
 305         BP_EMBEDDED_TYPE_RESERVED, /* Reserved for an unintegrated feature. */
 306         NUM_BP_EMBEDDED_TYPES = BP_EMBEDDED_TYPE_RESERVED
 307 } bp_embedded_type_t;
 308
 309 #define BPE_NUM_WORDS 14
 310 #define BPE_PAYLOAD_SIZE (BPE_NUM_WORDS * sizeof (uint64_t))
 311 #define BPE_IS_PAYLOADWORD(bp, wp) \
 312         ((wp) != &(bp)->blk_prop && (wp) != &(bp)->blk_birth)
 313
 314 #define SPA_BLKPTRSHIFT 7               /* blkptr_t is 128 bytes        */
 315 #define SPA_DVAS_PER_BP 3               /* Number of DVAs in a bp       */
 316
 317 /*
 318  * A block is a hole when it has either 1) never been written to, or
 319  * 2) is zero-filled. In both cases, ZFS can return all zeroes for all reads
 320  * without physically allocating disk space. Holes are represented in the
 321  * blkptr_t structure by zeroed blk_dva. Correct checking for holes is
 322  * done through the BP_IS_HOLE macro. For holes, the logical size, level,
 323  * DMU object type, and birth times are all also stored for holes that
 324  * were written to at some point (i.e. were punched after having been filled).
 325  */
 326 typedef struct blkptr {
 327         dva_t           blk_dva[SPA_DVAS_PER_BP]; /* Data Virtual Addresses */
 328         uint64_t        blk_prop;       /* size, compression, type, etc     */
 329         uint64_t        blk_pad[2];     /* Extra space for the future       */
 330         uint64_t        blk_phys_birth; /* txg when block was allocated     */
 331         uint64_t        blk_birth;      /* transaction group at birth       */
 332         uint64_t        blk_fill;       /* fill count                       */
 333         zio_cksum_t     blk_cksum;      /* 256-bit checksum                 */
 334 } blkptr_t;
 335
 336 /*
 337  * Macros to get and set fields in a bp or DVA.
 338  */
 339 #define DVA_GET_ASIZE(dva)      \
 340         BF64_GET_SB((dva)->dva_word[0], 0, SPA_ASIZEBITS, SPA_MINBLOCKSHIFT, 0)
 341 #define DVA_SET_ASIZE(dva, x)   \
 342         BF64_SET_SB((dva)->dva_word[0], 0, SPA_ASIZEBITS, \
 343         SPA_MINBLOCKSHIFT, 0, x)
 344
 345 #define DVA_GET_GRID(dva)       BF64_GET((dva)->dva_word[0], 24, 8)
 346 #define DVA_SET_GRID(dva, x)    BF64_SET((dva)->dva_word[0], 24, 8, x)
 347
 348 #define DVA_GET_VDEV(dva)       BF64_GET((dva)->dva_word[0], 32, 32)
 349 #define DVA_SET_VDEV(dva, x)    BF64_SET((dva)->dva_word[0], 32, 32, x)
 350
 351 #define DVA_GET_OFFSET(dva)     \
 352         BF64_GET_SB((dva)->dva_word[1], 0, 63, SPA_MINBLOCKSHIFT, 0)
 353 #define DVA_SET_OFFSET(dva, x)  \
 354         BF64_SET_SB((dva)->dva_word[1], 0, 63, SPA_MINBLOCKSHIFT, 0, x)
 355
 356 #define DVA_GET_GANG(dva)       BF64_GET((dva)->dva_word[1], 63, 1)
 357 #define DVA_SET_GANG(dva, x)    BF64_SET((dva)->dva_word[1], 63, 1, x)
 358
 359 #define BP_GET_LSIZE(bp)        \
 360         (BP_IS_EMBEDDED(bp) ?   \
 361         (BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA ? BPE_GET_LSIZE(bp) : 0): \
 362         BF64_GET_SB((bp)->blk_prop, 0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1))
 363 #define BP_SET_LSIZE(bp, x)     do { \
 364         ASSERT(!BP_IS_EMBEDDED(bp)); \
 365         BF64_SET_SB((bp)->blk_prop, \
 366             0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1, x); \
 367 _NOTE(CONSTCOND) } while (0)
 368
 369 #define BP_GET_PSIZE(bp)        \
 370         (BP_IS_EMBEDDED(bp) ? 0 : \
 371         BF64_GET_SB((bp)->blk_prop, 16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1))
 372 #define BP_SET_PSIZE(bp, x)     do { \
 373         ASSERT(!BP_IS_EMBEDDED(bp)); \
 374         BF64_SET_SB((bp)->blk_prop, \
 375             16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1, x); \
 376 _NOTE(CONSTCOND) } while (0)
 377
 378 #define BP_GET_COMPRESS(bp)             \
 379         BF64_GET((bp)->blk_prop, 32, SPA_COMPRESSBITS)
 380 #define BP_SET_COMPRESS(bp, x)          \
 381         BF64_SET((bp)->blk_prop, 32, SPA_COMPRESSBITS, x)
 382
 383 #define BP_IS_EMBEDDED(bp)              BF64_GET((bp)->blk_prop, 39, 1)
 384 #define BP_SET_EMBEDDED(bp, x)          BF64_SET((bp)->blk_prop, 39, 1, x)
 385
 386 #define BP_GET_CHECKSUM(bp)             \
 387         (BP_IS_EMBEDDED(bp) ? ZIO_CHECKSUM_OFF : \
 388         BF64_GET((bp)->blk_prop, 40, 8))
 389 #define BP_SET_CHECKSUM(bp, x)          do { \
 390         ASSERT(!BP_IS_EMBEDDED(bp)); \
 391         BF64_SET((bp)->blk_prop, 40, 8, x); \
 392 _NOTE(CONSTCOND) } while (0)
 393
 394 #define BP_GET_TYPE(bp)                 BF64_GET((bp)->blk_prop, 48, 8)
 395 #define BP_SET_TYPE(bp, x)              BF64_SET((bp)->blk_prop, 48, 8, x)
 396
 397 #define BP_GET_LEVEL(bp)                BF64_GET((bp)->blk_prop, 56, 5)
 398 #define BP_SET_LEVEL(bp, x)             BF64_SET((bp)->blk_prop, 56, 5, x)
 399
 400 #define BP_GET_DEDUP(bp)                BF64_GET((bp)->blk_prop, 62, 1)
 401 #define BP_SET_DEDUP(bp, x)             BF64_SET((bp)->blk_prop, 62, 1, x)
 402
 403 #define BP_GET_BYTEORDER(bp)            BF64_GET((bp)->blk_prop, 63, 1)
 404 #define BP_SET_BYTEORDER(bp, x)         BF64_SET((bp)->blk_prop, 63, 1, x)
 405
 406 #define BP_PHYSICAL_BIRTH(bp)           \
 407         (BP_IS_EMBEDDED(bp) ? 0 : \
 408         (bp)->blk_phys_birth ? (bp)->blk_phys_birth : (bp)->blk_birth)
 409
 410 #define BP_SET_BIRTH(bp, logical, physical)     \
 411 {                                               \
 412         ASSERT(!BP_IS_EMBEDDED(bp));            \
 413         (bp)->blk_birth = (logical);            \
 414         (bp)->blk_phys_birth = ((logical) == (physical) ? 0 : (physical)); \
 415 }
 416
 417 #define BP_GET_FILL(bp) (BP_IS_EMBEDDED(bp) ? 1 : (bp)->blk_fill)
 418
 419 #define BP_IS_METADATA(bp)      \
 420         (BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp)))
 421
 422 #define BP_GET_ASIZE(bp)        \
 423         (BP_IS_EMBEDDED(bp) ? 0 : \
 424         DVA_GET_ASIZE(&(bp)->blk_dva[0]) + \
 425         DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \
 426         DVA_GET_ASIZE(&(bp)->blk_dva[2]))
 427
 428 #define BP_GET_UCSIZE(bp)       \
 429         (BP_IS_METADATA(bp) ? BP_GET_PSIZE(bp) : BP_GET_LSIZE(bp))
 430
 431 #define BP_GET_NDVAS(bp)        \
 432         (BP_IS_EMBEDDED(bp) ? 0 : \
 433         !!DVA_GET_ASIZE(&(bp)->blk_dva[0]) + \
 434         !!DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \
 435         !!DVA_GET_ASIZE(&(bp)->blk_dva[2]))
 436
 437 #define BP_COUNT_GANG(bp)       \
 438         (BP_IS_EMBEDDED(bp) ? 0 : \
 439         (DVA_GET_GANG(&(bp)->blk_dva[0]) + \
 440         DVA_GET_GANG(&(bp)->blk_dva[1]) + \
 441         DVA_GET_GANG(&(bp)->blk_dva[2])))
 442
 443 #define DVA_EQUAL(dva1, dva2)   \
 444         ((dva1)->dva_word[1] == (dva2)->dva_word[1] && \
 445         (dva1)->dva_word[0] == (dva2)->dva_word[0])
 446
 447 #define BP_EQUAL(bp1, bp2)      \
 448         (BP_PHYSICAL_BIRTH(bp1) == BP_PHYSICAL_BIRTH(bp2) &&    \
 449         (bp1)->blk_birth == (bp2)->blk_birth &&                 \
 450         DVA_EQUAL(&(bp1)->blk_dva[0], &(bp2)->blk_dva[0]) &&    \
 451         DVA_EQUAL(&(bp1)->blk_dva[1], &(bp2)->blk_dva[1]) &&    \
 452         DVA_EQUAL(&(bp1)->blk_dva[2], &(bp2)->blk_dva[2]))
 453
 454
 455 #define DVA_IS_VALID(dva)       (DVA_GET_ASIZE(dva) != 0)
 456
 457 #define BP_IDENTITY(bp)         (ASSERT(!BP_IS_EMBEDDED(bp)), &(bp)->blk_dva[0])
 458 #define BP_IS_GANG(bp)          \
 459         (BP_IS_EMBEDDED(bp) ? B_FALSE : DVA_GET_GANG(BP_IDENTITY(bp)))
 460 #define DVA_IS_EMPTY(dva)       ((dva)->dva_word[0] == 0ULL &&  \
 461                                 (dva)->dva_word[1] == 0ULL)
 462 #define BP_IS_HOLE(bp) \
 463         (!BP_IS_EMBEDDED(bp) && DVA_IS_EMPTY(BP_IDENTITY(bp)))
 464
 465 /* BP_IS_RAIDZ(bp) assumes no block compression */
 466 #define BP_IS_RAIDZ(bp)         (DVA_GET_ASIZE(&(bp)->blk_dva[0]) > \
 467                                 BP_GET_PSIZE(bp))
 468
 469 #define BP_ZERO(bp)                             \
 470 {                                               \
 471         (bp)->blk_dva[0].dva_word[0] = 0;       \
 472         (bp)->blk_dva[0].dva_word[1] = 0;       \
 473         (bp)->blk_dva[1].dva_word[0] = 0;       \
 474         (bp)->blk_dva[1].dva_word[1] = 0;       \
 475         (bp)->blk_dva[2].dva_word[0] = 0;       \
 476         (bp)->blk_dva[2].dva_word[1] = 0;       \
 477         (bp)->blk_prop = 0;                     \
 478         (bp)->blk_pad[0] = 0;                   \
 479         (bp)->blk_pad[1] = 0;                   \
 480         (bp)->blk_phys_birth = 0;               \
 481         (bp)->blk_birth = 0;                    \
 482         (bp)->blk_fill = 0;                     \
 483         ZIO_SET_CHECKSUM(&(bp)->blk_cksum, 0, 0, 0, 0); \
 484 }
 485
 486 #ifdef _BIG_ENDIAN
 487 #define ZFS_HOST_BYTEORDER      (0ULL)
 488 #else
 489 #define ZFS_HOST_BYTEORDER      (1ULL)
 490 #endif
 491
 492 #define BP_SHOULD_BYTESWAP(bp)  (BP_GET_BYTEORDER(bp) != ZFS_HOST_BYTEORDER)
 493
 494 #define BP_SPRINTF_LEN  320
 495
 496 /*
 497  * This macro allows code sharing between zfs, libzpool, and mdb.
 498  * 'func' is either snprintf() or mdb_snprintf().
 499  * 'ws' (whitespace) can be ' ' for single-line format, '\n' for multi-line.
 500  */
 501 #define SNPRINTF_BLKPTR(func, ws, buf, size, bp, type, checksum, compress) \
 502 {                                                                       \
 503         static const char *copyname[] =                                 \
 504             { "zero", "single", "double", "triple" };                   \
 505         int len = 0;                                                    \
 506         int copies = 0;                                                 \
 507         int d;                                                          \
 508                                                                         \
 509         if (bp == NULL) {                                               \
 510                 len += func(buf + len, size - len, "<NULL>");           \
 511         } else if (BP_IS_HOLE(bp)) {                                    \
 512                 len += func(buf + len, size - len,                      \
 513                     "HOLE [L%llu %s] "                                  \
 514                     "size=%llxL birth=%lluL",                           \
 515                     (u_longlong_t)BP_GET_LEVEL(bp),                     \
 516                     type,                                               \
 517                     (u_longlong_t)BP_GET_LSIZE(bp),                     \
 518                     (u_longlong_t)bp->blk_birth);                       \
 519         } else if (BP_IS_EMBEDDED(bp)) {                                \
 520                 len = func(buf + len, size - len,                       \
 521                     "EMBEDDED [L%llu %s] et=%u %s "                     \
 522                     "size=%llxL/%llxP birth=%lluL",                     \
 523                     (u_longlong_t)BP_GET_LEVEL(bp),                     \
 524                     type,                                               \
 525                     (int)BPE_GET_ETYPE(bp),                             \
 526                     compress,                                           \
 527                     (u_longlong_t)BPE_GET_LSIZE(bp),                    \
 528                     (u_longlong_t)BPE_GET_PSIZE(bp),                    \
 529                     (u_longlong_t)bp->blk_birth);                       \
 530         } else {                                                        \
 531                 for (d = 0; d < BP_GET_NDVAS(bp); d++) {                \
 532                         const dva_t *dva = &bp->blk_dva[d];             \
 533                         if (DVA_IS_VALID(dva))                          \
 534                                 copies++;                               \
 535                         len += func(buf + len, size - len,              \
 536                             "DVA[%d]=<%llu:%llx:%llx>%c", d,            \
 537                             (u_longlong_t)DVA_GET_VDEV(dva),            \
 538                             (u_longlong_t)DVA_GET_OFFSET(dva),          \
 539                             (u_longlong_t)DVA_GET_ASIZE(dva),           \
 540                             ws);                                        \
 541                 }                                                       \
 542                 if (BP_IS_GANG(bp) &&                                   \
 543                     DVA_GET_ASIZE(&bp->blk_dva[2]) <=                   \
 544                     DVA_GET_ASIZE(&bp->blk_dva[1]) / 2)                 \
 545                         copies--;                                       \
 546                 len += func(buf + len, size - len,                      \
 547                     "[L%llu %s] %s %s %s %s %s %s%c"                    \
 548                     "size=%llxL/%llxP birth=%lluL/%lluP fill=%llu%c"    \
 549                     "cksum=%llx:%llx:%llx:%llx",                        \
 550                     (u_longlong_t)BP_GET_LEVEL(bp),                     \
 551                     type,                                               \
 552                     checksum,                                           \
 553                     compress,                                           \
 554                     BP_GET_BYTEORDER(bp) == 0 ? "BE" : "LE",            \
 555                     BP_IS_GANG(bp) ? "gang" : "contiguous",             \
 556                     BP_GET_DEDUP(bp) ? "dedup" : "unique",              \
 557                     copyname[copies],                                   \
 558                     ws,                                                 \
 559                     (u_longlong_t)BP_GET_LSIZE(bp),                     \
 560                     (u_longlong_t)BP_GET_PSIZE(bp),                     \
 561                     (u_longlong_t)bp->blk_birth,                        \
 562                     (u_longlong_t)BP_PHYSICAL_BIRTH(bp),                \
 563                     (u_longlong_t)BP_GET_FILL(bp),                      \
 564                     ws,                                                 \
 565                     (u_longlong_t)bp->blk_cksum.zc_word[0],             \
 566                     (u_longlong_t)bp->blk_cksum.zc_word[1],             \
 567                     (u_longlong_t)bp->blk_cksum.zc_word[2],             \
 568                     (u_longlong_t)bp->blk_cksum.zc_word[3]);            \
 569         }                                                               \
 570         ASSERT(len < size);                                             \
 571 }
 572
 573 #define BP_GET_BUFC_TYPE(bp)                                            \
 574         (BP_IS_METADATA(bp) ? ARC_BUFC_METADATA : ARC_BUFC_DATA)
 575
 576 typedef enum spa_import_type {
 577         SPA_IMPORT_EXISTING,
 578         SPA_IMPORT_ASSEMBLE
 579 } spa_import_type_t;
 580
 581 /* state manipulation functions */
 582 extern int spa_open(const char *pool, spa_t **, void *tag);
 583 extern int spa_open_rewind(const char *pool, spa_t **, void *tag,
 584     nvlist_t *policy, nvlist_t **config);
 585 extern int spa_get_stats(const char *pool, nvlist_t **config, char *altroot,
 586     size_t buflen);
 587 extern int spa_create(const char *pool, nvlist_t *config, nvlist_t *props,
 588     nvlist_t *zplprops);
 589 extern int spa_import(char *pool, nvlist_t *config, nvlist_t *props,
 590     uint64_t flags);
 591 extern nvlist_t *spa_tryimport(nvlist_t *tryconfig);
 592 extern int spa_destroy(char *pool);
 593 extern int spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
 594     boolean_t hardforce);
 595 extern int spa_reset(char *pool);
 596 extern void spa_async_request(spa_t *spa, int flag);
 597 extern void spa_async_unrequest(spa_t *spa, int flag);
 598 extern void spa_async_suspend(spa_t *spa);
 599 extern void spa_async_resume(spa_t *spa);
 600 extern spa_t *spa_inject_addref(char *pool);
 601 extern void spa_inject_delref(spa_t *spa);
 602 extern void spa_scan_stat_init(spa_t *spa);
 603 extern int spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps);
 604
 605 #define SPA_ASYNC_CONFIG_UPDATE 0x01
 606 #define SPA_ASYNC_REMOVE        0x02
 607 #define SPA_ASYNC_PROBE         0x04
 608 #define SPA_ASYNC_RESILVER_DONE 0x08
 609 #define SPA_ASYNC_RESILVER      0x10
 610 #define SPA_ASYNC_AUTOEXPAND    0x20
 611 #define SPA_ASYNC_REMOVE_DONE   0x40
 612 #define SPA_ASYNC_REMOVE_STOP   0x80
 613
 614 /*
 615  * Controls the behavior of spa_vdev_remove().
 616  */
 617 #define SPA_REMOVE_UNSPARE      0x01
 618 #define SPA_REMOVE_DONE         0x02
 619
 620 /* device manipulation */
 621 extern int spa_vdev_add(spa_t *spa, nvlist_t *nvroot);
 622 extern int spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot,
 623     int replacing);
 624 extern int spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid,
 625     int replace_done);
 626 extern int spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare);
 627 extern boolean_t spa_vdev_remove_active(spa_t *spa);
 628 extern int spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath);
 629 extern int spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru);
 630 extern int spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
 631     nvlist_t *props, boolean_t exp);
 632
 633 /* spare state (which is global across all pools) */
 634 extern void spa_spare_add(vdev_t *vd);
 635 extern void spa_spare_remove(vdev_t *vd);
 636 extern boolean_t spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt);
 637 extern void spa_spare_activate(vdev_t *vd);
 638
 639 /* L2ARC state (which is global across all pools) */
 640 extern void spa_l2cache_add(vdev_t *vd);
 641 extern void spa_l2cache_remove(vdev_t *vd);
 642 extern boolean_t spa_l2cache_exists(uint64_t guid, uint64_t *pool);
 643 extern void spa_l2cache_activate(vdev_t *vd);
 644 extern void spa_l2cache_drop(spa_t *spa);
 645
 646 /* scanning */
 647 extern int spa_scan(spa_t *spa, pool_scan_func_t func);
 648 extern int spa_scan_stop(spa_t *spa);
 649
 650 /* spa syncing */
 651 extern void spa_sync(spa_t *spa, uint64_t txg); /* only for DMU use */
 652 extern void spa_sync_allpools(void);
 653
 654 extern int zfs_sync_pass_deferred_free;
 655
 656 /* spa namespace global mutex */
 657 extern kmutex_t spa_namespace_lock;
 658
 659 /*
 660  * SPA configuration functions in spa_config.c
 661  */
 662
 663 #define SPA_CONFIG_UPDATE_POOL  0
 664 #define SPA_CONFIG_UPDATE_VDEVS 1
 665
 666 extern void spa_config_sync(spa_t *, boolean_t, boolean_t);
 667 extern void spa_config_load(void);
 668 extern nvlist_t *spa_all_configs(uint64_t *);
 669 extern void spa_config_set(spa_t *spa, nvlist_t *config);
 670 extern nvlist_t *spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg,
 671     int getstats);
 672 extern void spa_config_update(spa_t *spa, int what);
 673
 674 /*
 675  * Miscellaneous SPA routines in spa_misc.c
 676  */
 677
 678 /* Namespace manipulation */
 679 extern spa_t *spa_lookup(const char *name);
 680 extern spa_t *spa_add(const char *name, nvlist_t *config, const char *altroot);
 681 extern void spa_remove(spa_t *spa);
 682 extern spa_t *spa_next(spa_t *prev);
 683
 684 /* Refcount functions */
 685 extern void spa_open_ref(spa_t *spa, void *tag);
 686 extern void spa_close(spa_t *spa, void *tag);
 687 extern void spa_async_close(spa_t *spa, void *tag);
 688 extern boolean_t spa_refcount_zero(spa_t *spa);
 689
 690 #define SCL_NONE        0x00
 691 #define SCL_CONFIG      0x01
 692 #define SCL_STATE       0x02
 693 #define SCL_L2ARC       0x04            /* hack until L2ARC 2.0 */
 694 #define SCL_ALLOC       0x08
 695 #define SCL_ZIO         0x10
 696 #define SCL_FREE        0x20
 697 #define SCL_VDEV        0x40
 698 #define SCL_LOCKS       7
 699 #define SCL_ALL         ((1 << SCL_LOCKS) - 1)
 700 #define SCL_STATE_ALL   (SCL_STATE | SCL_L2ARC | SCL_ZIO)
 701
 702 /* Historical pool statistics */
 703 typedef struct spa_stats_history {
 704         kmutex_t                lock;
 705         uint64_t                count;
 706         uint64_t                size;
 707         kstat_t                 *kstat;
 708         void                    *private;
 709         list_t                  list;
 710 } spa_stats_history_t;
 711
 712 typedef struct spa_stats {
 713         spa_stats_history_t     read_history;
 714         spa_stats_history_t     txg_history;
 715         spa_stats_history_t     tx_assign_histogram;
 716         spa_stats_history_t     io_history;
 717 } spa_stats_t;
 718
 719 typedef enum txg_state {
 720         TXG_STATE_BIRTH         = 0,
 721         TXG_STATE_OPEN          = 1,
 722         TXG_STATE_QUIESCED      = 2,
 723         TXG_STATE_WAIT_FOR_SYNC = 3,
 724         TXG_STATE_SYNCED        = 4,
 725         TXG_STATE_COMMITTED     = 5,
 726 } txg_state_t;
 727
 728 typedef struct txg_stat {
 729         vdev_stat_t             vs1;
 730         vdev_stat_t             vs2;
 731         uint64_t                txg;
 732         uint64_t                ndirty;
 733 } txg_stat_t;
 734
 735 extern void spa_stats_init(spa_t *spa);
 736 extern void spa_stats_destroy(spa_t *spa);
 737 extern void spa_read_history_add(spa_t *spa, const zbookmark_phys_t *zb,
 738     uint32_t aflags);
 739 extern void spa_txg_history_add(spa_t *spa, uint64_t txg, hrtime_t birth_time);
 740 extern int spa_txg_history_set(spa_t *spa,  uint64_t txg,
 741     txg_state_t completed_state, hrtime_t completed_time);
 742 extern txg_stat_t *spa_txg_history_init_io(spa_t *, uint64_t,
 743     struct dsl_pool *);
 744 extern void spa_txg_history_fini_io(spa_t *, txg_stat_t *);
 745 extern void spa_tx_assign_add_nsecs(spa_t *spa, uint64_t nsecs);
 746
 747 /* Pool configuration locks */
 748 extern int spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw);
 749 extern void spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw);
 750 extern void spa_config_exit(spa_t *spa, int locks, void *tag);
 751 extern int spa_config_held(spa_t *spa, int locks, krw_t rw);
 752
 753 /* Pool vdev add/remove lock */
 754 extern uint64_t spa_vdev_enter(spa_t *spa);
 755 extern uint64_t spa_vdev_config_enter(spa_t *spa);
 756 extern void spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg,
 757     int error, char *tag);
 758 extern int spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error);
 759
 760 /* Pool vdev state change lock */
 761 extern void spa_vdev_state_enter(spa_t *spa, int oplock);
 762 extern int spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error);
 763
 764 /* Log state */
 765 typedef enum spa_log_state {
 766         SPA_LOG_UNKNOWN = 0,    /* unknown log state */
 767         SPA_LOG_MISSING,        /* missing log(s) */
 768         SPA_LOG_CLEAR,          /* clear the log(s) */
 769         SPA_LOG_GOOD,           /* log(s) are good */
 770 } spa_log_state_t;
 771
 772 extern spa_log_state_t spa_get_log_state(spa_t *spa);
 773 extern void spa_set_log_state(spa_t *spa, spa_log_state_t state);
 774 extern int spa_offline_log(spa_t *spa);
 775
 776 /* Log claim callback */
 777 extern void spa_claim_notify(zio_t *zio);
 778 extern void spa_deadman(void *);
 779
 780 /* Accessor functions */
 781 extern boolean_t spa_shutting_down(spa_t *spa);
 782 extern struct dsl_pool *spa_get_dsl(spa_t *spa);
 783 extern boolean_t spa_is_initializing(spa_t *spa);
 784 extern blkptr_t *spa_get_rootblkptr(spa_t *spa);
 785 extern void spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp);
 786 extern void spa_altroot(spa_t *, char *, size_t);
 787 extern int spa_sync_pass(spa_t *spa);
 788 extern char *spa_name(spa_t *spa);
 789 extern uint64_t spa_guid(spa_t *spa);
 790 extern uint64_t spa_load_guid(spa_t *spa);
 791 extern uint64_t spa_last_synced_txg(spa_t *spa);
 792 extern uint64_t spa_first_txg(spa_t *spa);
 793 extern uint64_t spa_syncing_txg(spa_t *spa);
 794 extern uint64_t spa_version(spa_t *spa);
 795 extern pool_state_t spa_state(spa_t *spa);
 796 extern spa_load_state_t spa_load_state(spa_t *spa);
 797 extern uint64_t spa_freeze_txg(spa_t *spa);
 798 extern uint64_t spa_get_worst_case_asize(spa_t *spa, uint64_t lsize);
 799 extern uint64_t spa_get_dspace(spa_t *spa);
 800 extern uint64_t spa_get_slop_space(spa_t *spa);
 801 extern void spa_update_dspace(spa_t *spa);
 802 extern uint64_t spa_version(spa_t *spa);
 803 extern boolean_t spa_deflate(spa_t *spa);
 804 extern metaslab_class_t *spa_normal_class(spa_t *spa);
 805 extern metaslab_class_t *spa_log_class(spa_t *spa);
 806 extern void spa_evicting_os_register(spa_t *, objset_t *os);
 807 extern void spa_evicting_os_deregister(spa_t *, objset_t *os);
 808 extern void spa_evicting_os_wait(spa_t *spa);
 809 extern int spa_max_replication(spa_t *spa);
 810 extern int spa_prev_software_version(spa_t *spa);
 811 extern uint8_t spa_get_failmode(spa_t *spa);
 812 extern boolean_t spa_suspended(spa_t *spa);
 813 extern uint64_t spa_bootfs(spa_t *spa);
 814 extern uint64_t spa_delegation(spa_t *spa);
 815 extern objset_t *spa_meta_objset(spa_t *spa);
 816 extern uint64_t spa_deadman_synctime(spa_t *spa);
 817
 818 /* Miscellaneous support routines */
 819 extern void spa_activate_mos_feature(spa_t *spa, const char *feature,
 820     dmu_tx_t *tx);
 821 extern void spa_deactivate_mos_feature(spa_t *spa, const char *feature);
 822 extern int spa_rename(const char *oldname, const char *newname);
 823 extern spa_t *spa_by_guid(uint64_t pool_guid, uint64_t device_guid);
 824 extern boolean_t spa_guid_exists(uint64_t pool_guid, uint64_t device_guid);
 825 extern char *spa_strdup(const char *);
 826 extern void spa_strfree(char *);
 827 extern uint64_t spa_get_random(uint64_t range);
 828 extern uint64_t spa_generate_guid(spa_t *spa);
 829 extern void snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp);
 830 extern void spa_freeze(spa_t *spa);
 831 extern int spa_change_guid(spa_t *spa);
 832 extern void spa_upgrade(spa_t *spa, uint64_t version);
 833 extern void spa_evict_all(void);
 834 extern vdev_t *spa_lookup_by_guid(spa_t *spa, uint64_t guid,
 835     boolean_t l2cache);
 836 extern boolean_t spa_has_spare(spa_t *, uint64_t guid);
 837 extern uint64_t dva_get_dsize_sync(spa_t *spa, const dva_t *dva);
 838 extern uint64_t bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp);
 839 extern uint64_t bp_get_dsize(spa_t *spa, const blkptr_t *bp);
 840 extern boolean_t spa_has_slogs(spa_t *spa);
 841 extern boolean_t spa_is_root(spa_t *spa);
 842 extern boolean_t spa_writeable(spa_t *spa);
 843 extern boolean_t spa_has_pending_synctask(spa_t *spa);
 844 extern int spa_maxblocksize(spa_t *spa);
 845 extern int spa_maxdnodesize(spa_t *spa);
 846 extern void zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp);
 847
 848 extern int spa_mode(spa_t *spa);
 849 extern uint64_t strtonum(const char *str, char **nptr);
 850
 851 extern char *spa_his_ievent_table[];
 852
 853 extern void spa_history_create_obj(spa_t *spa, dmu_tx_t *tx);
 854 extern int spa_history_get(spa_t *spa, uint64_t *offset, uint64_t *len_read,
 855     char *his_buf);
 856 extern int spa_history_log(spa_t *spa, const char *his_buf);
 857 extern int spa_history_log_nvl(spa_t *spa, nvlist_t *nvl);
 858 extern void spa_history_log_version(spa_t *spa, const char *operation);
 859 extern void spa_history_log_internal(spa_t *spa, const char *operation,
 860     dmu_tx_t *tx, const char *fmt, ...);
 861 extern void spa_history_log_internal_ds(struct dsl_dataset *ds, const char *op,
 862     dmu_tx_t *tx, const char *fmt, ...);
 863 extern void spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation,
 864     dmu_tx_t *tx, const char *fmt, ...);
 865
 866 /* error handling */
 867 struct zbookmark_phys;
 868 extern void spa_log_error(spa_t *spa, zio_t *zio);
 869 extern void zfs_ereport_post(const char *class, spa_t *spa, vdev_t *vd,
 870     zio_t *zio, uint64_t stateoroffset, uint64_t length);
 871 extern void zfs_post_remove(spa_t *spa, vdev_t *vd);
 872 extern void zfs_post_state_change(spa_t *spa, vdev_t *vd, uint64_t laststate);
 873 extern void zfs_post_autoreplace(spa_t *spa, vdev_t *vd);
 874 extern void zfs_post_sysevent(spa_t *spa, vdev_t *vd, const char *name);
 875 extern uint64_t spa_get_errlog_size(spa_t *spa);
 876 extern int spa_get_errlog(spa_t *spa, void *uaddr, size_t *count);
 877 extern void spa_errlog_rotate(spa_t *spa);
 878 extern void spa_errlog_drain(spa_t *spa);
 879 extern void spa_errlog_sync(spa_t *spa, uint64_t txg);
 880 extern void spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub);
 881
 882 /* vdev cache */
 883 extern void vdev_cache_stat_init(void);
 884 extern void vdev_cache_stat_fini(void);
 885
 886 /* Initialization and termination */
 887 extern void spa_init(int flags);
 888 extern void spa_fini(void);
 889 extern void spa_boot_init(void);
 890
 891 /* properties */
 892 extern int spa_prop_set(spa_t *spa, nvlist_t *nvp);
 893 extern int spa_prop_get(spa_t *spa, nvlist_t **nvp);
 894 extern void spa_prop_clear_bootfs(spa_t *spa, uint64_t obj, dmu_tx_t *tx);
 895 extern void spa_configfile_set(spa_t *, nvlist_t *, boolean_t);
 896
 897 /* asynchronous event notification */
 898 extern void spa_event_notify(spa_t *spa, vdev_t *vdev, const char *name);
 899
 900 #ifdef ZFS_DEBUG
 901 #define dprintf_bp(bp, fmt, ...) do {                           \
 902         if (zfs_flags & ZFS_DEBUG_DPRINTF) {                    \
 903         char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP);  \
 904         snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, (bp));        \
 905         dprintf(fmt " %s\n", __VA_ARGS__, __blkbuf);            \
 906         kmem_free(__blkbuf, BP_SPRINTF_LEN);                    \
 907         } \
 908 _NOTE(CONSTCOND) } while (0)
 909 #else
 910 #define dprintf_bp(bp, fmt, ...)
 911 #endif
 912
 913 extern boolean_t spa_debug_enabled(spa_t *spa);
 914 #define spa_dbgmsg(spa, ...)                    \
 915 {                                               \
 916         if (spa_debug_enabled(spa))             \
 917                 zfs_dbgmsg(__VA_ARGS__);        \
 918 }
 919
 920 extern int spa_mode_global;                     /* mode, e.g. FREAD | FWRITE */
 921
 922 #ifdef  __cplusplus
 923 }
 924 #endif
 925
 926 #endif  /* _SYS_SPA_H */