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  */
  27
  28 #ifndef _SYS_SPA_H
  29 #define _SYS_SPA_H
  30
  31 #include <sys/avl.h>
  32 #include <sys/zfs_context.h>
  33 #include <sys/kstat.h>
  34 #include <sys/nvpair.h>
  35 #include <sys/sysmacros.h>
  36 #include <sys/types.h>
  37 #include <sys/fs/zfs.h>
  38
  39 #ifdef  __cplusplus
  40 extern "C" {
  41 #endif
  42
  43 /*
  44  * Forward references that lots of things need.
  45  */
  46 typedef struct spa spa_t;
  47 typedef struct vdev vdev_t;
  48 typedef struct metaslab metaslab_t;
  49 typedef struct metaslab_group metaslab_group_t;
  50 typedef struct metaslab_class metaslab_class_t;
  51 typedef struct zio zio_t;
  52 typedef struct zilog zilog_t;
  53 typedef struct spa_aux_vdev spa_aux_vdev_t;
  54 typedef struct ddt ddt_t;
  55 typedef struct ddt_entry ddt_entry_t;
  56 typedef struct zbookmark_phys zbookmark_phys_t;
  57
  58 struct dsl_pool;
  59 struct dsl_dataset;
  60
  61 /*
  62  * General-purpose 32-bit and 64-bit bitfield encodings.
  63  */
  64 #define BF32_DECODE(x, low, len)        P2PHASE((x) >> (low), 1U << (len))
  65 #define BF64_DECODE(x, low, len)        P2PHASE((x) >> (low), 1ULL << (len))
  66 #define BF32_ENCODE(x, low, len)        (P2PHASE((x), 1U << (len)) << (low))
  67 #define BF64_ENCODE(x, low, len)        (P2PHASE((x), 1ULL << (len)) << (low))
  68
  69 #define BF32_GET(x, low, len)           BF32_DECODE(x, low, len)
  70 #define BF64_GET(x, low, len)           BF64_DECODE(x, low, len)
  71
  72 #define BF32_SET(x, low, len, val) do { \
  73         ASSERT3U(val, <, 1U << (len)); \
  74         ASSERT3U(low + len, <=, 32); \
  75         (x) ^= BF32_ENCODE((x >> low) ^ (val), low, len); \
  76 _NOTE(CONSTCOND) } while (0)
  77
  78 #define BF64_SET(x, low, len, val) do { \
  79         ASSERT3U(val, <, 1ULL << (len)); \
  80         ASSERT3U(low + len, <=, 64); \
  81         ((x) ^= BF64_ENCODE((x >> low) ^ (val), low, len)); \
  82 _NOTE(CONSTCOND) } while (0)
  83
  84 #define BF32_GET_SB(x, low, len, shift, bias)   \
  85         ((BF32_GET(x, low, len) + (bias)) << (shift))
  86 #define BF64_GET_SB(x, low, len, shift, bias)   \
  87         ((BF64_GET(x, low, len) + (bias)) << (shift))
  88
  89 #define BF32_SET_SB(x, low, len, shift, bias, val) do { \
  90         ASSERT(IS_P2ALIGNED(val, 1U << shift)); \
  91         ASSERT3S((val) >> (shift), >=, bias); \
  92         BF32_SET(x, low, len, ((val) >> (shift)) - (bias)); \
  93 _NOTE(CONSTCOND) } while (0)
  94 #define BF64_SET_SB(x, low, len, shift, bias, val) do { \
  95         ASSERT(IS_P2ALIGNED(val, 1ULL << shift)); \
  96         ASSERT3S((val) >> (shift), >=, bias); \
  97         BF64_SET(x, low, len, ((val) >> (shift)) - (bias)); \
  98 _NOTE(CONSTCOND) } while (0)
  99
 100 /*
 101  * We currently support block sizes from 512 bytes to 16MB.
 102  * The benefits of larger blocks, and thus larger IO, need to be weighed
 103  * against the cost of COWing a giant block to modify one byte, and the
 104  * large latency of reading or writing a large block.
 105  *
 106  * Note that although blocks up to 16MB are supported, the recordsize
 107  * property can not be set larger than zfs_max_recordsize (default 1MB).
 108  * See the comment near zfs_max_recordsize in dsl_dataset.c for details.
 109  *
 110  * Note that although the LSIZE field of the blkptr_t can store sizes up
 111  * to 32MB, the dnode's dn_datablkszsec can only store sizes up to
 112  * 32MB - 512 bytes.  Therefore, we limit SPA_MAXBLOCKSIZE to 16MB.
 113  */
 114 #define SPA_MINBLOCKSHIFT       9
 115 #define SPA_OLD_MAXBLOCKSHIFT   17
 116 #define SPA_MAXBLOCKSHIFT       24
 117 #define SPA_MINBLOCKSIZE        (1ULL << SPA_MINBLOCKSHIFT)
 118 #define SPA_OLD_MAXBLOCKSIZE    (1ULL << SPA_OLD_MAXBLOCKSHIFT)
 119 #define SPA_MAXBLOCKSIZE        (1ULL << SPA_MAXBLOCKSHIFT)
 120
 121 /*
 122  * Size of block to hold the configuration data (a packed nvlist)
 123  */
 124 #define SPA_CONFIG_BLOCKSIZE    (1ULL << 14)
 125
 126 /*
 127  * The DVA size encodings for LSIZE and PSIZE support blocks up to 32MB.
 128  * The ASIZE encoding should be at least 64 times larger (6 more bits)
 129  * to support up to 4-way RAID-Z mirror mode with worst-case gang block
 130  * overhead, three DVAs per bp, plus one more bit in case we do anything
 131  * else that expands the ASIZE.
 132  */
 133 #define SPA_LSIZEBITS           16      /* LSIZE up to 32M (2^16 * 512) */
 134 #define SPA_PSIZEBITS           16      /* PSIZE up to 32M (2^16 * 512) */
 135 #define SPA_ASIZEBITS           24      /* ASIZE up to 64 times larger  */
 136
 137 /*
 138  * All SPA data is represented by 128-bit data virtual addresses (DVAs).
 139  * The members of the dva_t should be considered opaque outside the SPA.
 140  */
 141 typedef struct dva {
 142         uint64_t        dva_word[2];
 143 } dva_t;
 144
 145 /*
 146  * Each block has a 256-bit checksum -- strong enough for cryptographic hashes.
 147  */
 148 typedef struct zio_cksum {
 149         uint64_t        zc_word[4];
 150 } zio_cksum_t;
 151
 152 /*
 153  * Each block is described by its DVAs, time of birth, checksum, etc.
 154  * The word-by-word, bit-by-bit layout of the blkptr is as follows:
 155  *
 156  *      64      56      48      40      32      24      16      8       0
 157  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 158  * 0    |               vdev1           | GRID  |         ASIZE         |
 159  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 160  * 1    |G|                      offset1                                |
 161  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 162  * 2    |               vdev2           | GRID  |         ASIZE         |
 163  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 164  * 3    |G|                      offset2                                |
 165  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 166  * 4    |               vdev3           | GRID  |         ASIZE         |
 167  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 168  * 5    |G|                      offset3                                |
 169  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 170  * 6    |BDX|lvl| type  | cksum |E| comp|    PSIZE      |     LSIZE     |
 171  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 172  * 7    |                       padding                                 |
 173  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 174  * 8    |                       padding                                 |
 175  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 176  * 9    |                       physical birth txg                      |
 177  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 178  * a    |                       logical birth txg                       |
 179  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 180  * b    |                       fill count                              |
 181  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 182  * c    |                       checksum[0]                             |
 183  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 184  * d    |                       checksum[1]                             |
 185  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 186  * e    |                       checksum[2]                             |
 187  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 188  * f    |                       checksum[3]                             |
 189  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 190  *
 191  * Legend:
 192  *
 193  * vdev         virtual device ID
 194  * offset       offset into virtual device
 195  * LSIZE        logical size
 196  * PSIZE        physical size (after compression)
 197  * ASIZE        allocated size (including RAID-Z parity and gang block headers)
 198  * GRID         RAID-Z layout information (reserved for future use)
 199  * cksum        checksum function
 200  * comp         compression function
 201  * G            gang block indicator
 202  * B            byteorder (endianness)
 203  * D            dedup
 204  * X            encryption (on version 30, which is not supported)
 205  * E            blkptr_t contains embedded data (see below)
 206  * lvl          level of indirection
 207  * type         DMU object type
 208  * phys birth   txg of block allocation; zero if same as logical birth txg
 209  * log. birth   transaction group in which the block was logically born
 210  * fill count   number of non-zero blocks under this bp
 211  * checksum[4]  256-bit checksum of the data this bp describes
 212  */
 213
 214 /*
 215  * "Embedded" blkptr_t's don't actually point to a block, instead they
 216  * have a data payload embedded in the blkptr_t itself.  See the comment
 217  * in blkptr.c for more details.
 218  *
 219  * The blkptr_t is laid out as follows:
 220  *
 221  *      64      56      48      40      32      24      16      8       0
 222  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 223  * 0    |      payload                                                  |
 224  * 1    |      payload                                                  |
 225  * 2    |      payload                                                  |
 226  * 3    |      payload                                                  |
 227  * 4    |      payload                                                  |
 228  * 5    |      payload                                                  |
 229  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 230  * 6    |BDX|lvl| type  | etype |E| comp| PSIZE|              LSIZE     |
 231  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 232  * 7    |      payload                                                  |
 233  * 8    |      payload                                                  |
 234  * 9    |      payload                                                  |
 235  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 236  * a    |                       logical birth txg                       |
 237  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 238  * b    |      payload                                                  |
 239  * c    |      payload                                                  |
 240  * d    |      payload                                                  |
 241  * e    |      payload                                                  |
 242  * f    |      payload                                                  |
 243  *      +-------+-------+-------+-------+-------+-------+-------+-------+
 244  *
 245  * Legend:
 246  *
 247  * payload              contains the embedded data
 248  * B (byteorder)        byteorder (endianness)
 249  * D (dedup)            padding (set to zero)
 250  * X                    encryption (set to zero; see above)
 251  * E (embedded)         set to one
 252  * lvl                  indirection level
 253  * type                 DMU object type
 254  * etype                how to interpret embedded data (BP_EMBEDDED_TYPE_*)
 255  * comp                 compression function of payload
 256  * PSIZE                size of payload after compression, in bytes
 257  * LSIZE                logical size of payload, in bytes
 258  *                      note that 25 bits is enough to store the largest
 259  *                      "normal" BP's LSIZE (2^16 * 2^9) in bytes
 260  * log. birth           transaction group in which the block was logically born
 261  *
 262  * Note that LSIZE and PSIZE are stored in bytes, whereas for non-embedded
 263  * bp's they are stored in units of SPA_MINBLOCKSHIFT.
 264  * Generally, the generic BP_GET_*() macros can be used on embedded BP's.
 265  * The B, D, X, lvl, type, and comp fields are stored the same as with normal
 266  * BP's so the BP_SET_* macros can be used with them.  etype, PSIZE, LSIZE must
 267  * be set with the BPE_SET_* macros.  BP_SET_EMBEDDED() should be called before
 268  * other macros, as they assert that they are only used on BP's of the correct
 269  * "embedded-ness".
 270  */
 271
 272 #define BPE_GET_ETYPE(bp)       \
 273         (ASSERT(BP_IS_EMBEDDED(bp)), \
 274         BF64_GET((bp)->blk_prop, 40, 8))
 275 #define BPE_SET_ETYPE(bp, t)    do { \
 276         ASSERT(BP_IS_EMBEDDED(bp)); \
 277         BF64_SET((bp)->blk_prop, 40, 8, t); \
 278 _NOTE(CONSTCOND) } while (0)
 279
 280 #define BPE_GET_LSIZE(bp)       \
 281         (ASSERT(BP_IS_EMBEDDED(bp)), \
 282         BF64_GET_SB((bp)->blk_prop, 0, 25, 0, 1))
 283 #define BPE_SET_LSIZE(bp, x)    do { \
 284         ASSERT(BP_IS_EMBEDDED(bp)); \
 285         BF64_SET_SB((bp)->blk_prop, 0, 25, 0, 1, x); \
 286 _NOTE(CONSTCOND) } while (0)
 287
 288 #define BPE_GET_PSIZE(bp)       \
 289         (ASSERT(BP_IS_EMBEDDED(bp)), \
 290         BF64_GET_SB((bp)->blk_prop, 25, 7, 0, 1))
 291 #define BPE_SET_PSIZE(bp, x)    do { \
 292         ASSERT(BP_IS_EMBEDDED(bp)); \
 293         BF64_SET_SB((bp)->blk_prop, 25, 7, 0, 1, x); \
 294 _NOTE(CONSTCOND) } while (0)
 295
 296 typedef enum bp_embedded_type {
 297         BP_EMBEDDED_TYPE_DATA,
 298         BP_EMBEDDED_TYPE_RESERVED, /* Reserved for an unintegrated feature. */
 299         NUM_BP_EMBEDDED_TYPES = BP_EMBEDDED_TYPE_RESERVED
 300 } bp_embedded_type_t;
 301
 302 #define BPE_NUM_WORDS 14
 303 #define BPE_PAYLOAD_SIZE (BPE_NUM_WORDS * sizeof (uint64_t))
 304 #define BPE_IS_PAYLOADWORD(bp, wp) \
 305         ((wp) != &(bp)->blk_prop && (wp) != &(bp)->blk_birth)
 306
 307 #define SPA_BLKPTRSHIFT 7               /* blkptr_t is 128 bytes        */
 308 #define SPA_DVAS_PER_BP 3               /* Number of DVAs in a bp       */
 309
 310 /*
 311  * A block is a hole when it has either 1) never been written to, or
 312  * 2) is zero-filled. In both cases, ZFS can return all zeroes for all reads
 313  * without physically allocating disk space. Holes are represented in the
 314  * blkptr_t structure by zeroed blk_dva. Correct checking for holes is
 315  * done through the BP_IS_HOLE macro. For holes, the logical size, level,
 316  * DMU object type, and birth times are all also stored for holes that
 317  * were written to at some point (i.e. were punched after having been filled).
 318  */
 319 typedef struct blkptr {
 320         dva_t           blk_dva[SPA_DVAS_PER_BP]; /* Data Virtual Addresses */
 321         uint64_t        blk_prop;       /* size, compression, type, etc     */
 322         uint64_t        blk_pad[2];     /* Extra space for the future       */
 323         uint64_t        blk_phys_birth; /* txg when block was allocated     */
 324         uint64_t        blk_birth;      /* transaction group at birth       */
 325         uint64_t        blk_fill;       /* fill count                       */
 326         zio_cksum_t     blk_cksum;      /* 256-bit checksum                 */
 327 } blkptr_t;
 328
 329 /*
 330  * Macros to get and set fields in a bp or DVA.
 331  */
 332 #define DVA_GET_ASIZE(dva)      \
 333         BF64_GET_SB((dva)->dva_word[0], 0, SPA_ASIZEBITS, SPA_MINBLOCKSHIFT, 0)
 334 #define DVA_SET_ASIZE(dva, x)   \
 335         BF64_SET_SB((dva)->dva_word[0], 0, SPA_ASIZEBITS, \
 336         SPA_MINBLOCKSHIFT, 0, x)
 337
 338 #define DVA_GET_GRID(dva)       BF64_GET((dva)->dva_word[0], 24, 8)
 339 #define DVA_SET_GRID(dva, x)    BF64_SET((dva)->dva_word[0], 24, 8, x)
 340
 341 #define DVA_GET_VDEV(dva)       BF64_GET((dva)->dva_word[0], 32, 32)
 342 #define DVA_SET_VDEV(dva, x)    BF64_SET((dva)->dva_word[0], 32, 32, x)
 343
 344 #define DVA_GET_OFFSET(dva)     \
 345         BF64_GET_SB((dva)->dva_word[1], 0, 63, SPA_MINBLOCKSHIFT, 0)
 346 #define DVA_SET_OFFSET(dva, x)  \
 347         BF64_SET_SB((dva)->dva_word[1], 0, 63, SPA_MINBLOCKSHIFT, 0, x)
 348
 349 #define DVA_GET_GANG(dva)       BF64_GET((dva)->dva_word[1], 63, 1)
 350 #define DVA_SET_GANG(dva, x)    BF64_SET((dva)->dva_word[1], 63, 1, x)
 351
 352 #define BP_GET_LSIZE(bp)        \
 353         (BP_IS_EMBEDDED(bp) ?   \
 354         (BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA ? BPE_GET_LSIZE(bp) : 0): \
 355         BF64_GET_SB((bp)->blk_prop, 0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1))
 356 #define BP_SET_LSIZE(bp, x)     do { \
 357         ASSERT(!BP_IS_EMBEDDED(bp)); \
 358         BF64_SET_SB((bp)->blk_prop, \
 359             0, SPA_LSIZEBITS, SPA_MINBLOCKSHIFT, 1, x); \
 360 _NOTE(CONSTCOND) } while (0)
 361
 362 #define BP_GET_PSIZE(bp)        \
 363         (BP_IS_EMBEDDED(bp) ? 0 : \
 364         BF64_GET_SB((bp)->blk_prop, 16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1))
 365 #define BP_SET_PSIZE(bp, x)     do { \
 366         ASSERT(!BP_IS_EMBEDDED(bp)); \
 367         BF64_SET_SB((bp)->blk_prop, \
 368             16, SPA_PSIZEBITS, SPA_MINBLOCKSHIFT, 1, x); \
 369 _NOTE(CONSTCOND) } while (0)
 370
 371 #define BP_GET_COMPRESS(bp)             BF64_GET((bp)->blk_prop, 32, 7)
 372 #define BP_SET_COMPRESS(bp, x)          BF64_SET((bp)->blk_prop, 32, 7, x)
 373
 374 #define BP_IS_EMBEDDED(bp)              BF64_GET((bp)->blk_prop, 39, 1)
 375 #define BP_SET_EMBEDDED(bp, x)          BF64_SET((bp)->blk_prop, 39, 1, x)
 376
 377 #define BP_GET_CHECKSUM(bp)             \
 378         (BP_IS_EMBEDDED(bp) ? ZIO_CHECKSUM_OFF : \
 379         BF64_GET((bp)->blk_prop, 40, 8))
 380 #define BP_SET_CHECKSUM(bp, x)          do { \
 381         ASSERT(!BP_IS_EMBEDDED(bp)); \
 382         BF64_SET((bp)->blk_prop, 40, 8, x); \
 383 _NOTE(CONSTCOND) } while (0)
 384
 385 #define BP_GET_TYPE(bp)                 BF64_GET((bp)->blk_prop, 48, 8)
 386 #define BP_SET_TYPE(bp, x)              BF64_SET((bp)->blk_prop, 48, 8, x)
 387
 388 #define BP_GET_LEVEL(bp)                BF64_GET((bp)->blk_prop, 56, 5)
 389 #define BP_SET_LEVEL(bp, x)             BF64_SET((bp)->blk_prop, 56, 5, x)
 390
 391 #define BP_GET_DEDUP(bp)                BF64_GET((bp)->blk_prop, 62, 1)
 392 #define BP_SET_DEDUP(bp, x)             BF64_SET((bp)->blk_prop, 62, 1, x)
 393
 394 #define BP_GET_BYTEORDER(bp)            BF64_GET((bp)->blk_prop, 63, 1)
 395 #define BP_SET_BYTEORDER(bp, x)         BF64_SET((bp)->blk_prop, 63, 1, x)
 396
 397 #define BP_PHYSICAL_BIRTH(bp)           \
 398         (BP_IS_EMBEDDED(bp) ? 0 : \
 399         (bp)->blk_phys_birth ? (bp)->blk_phys_birth : (bp)->blk_birth)
 400
 401 #define BP_SET_BIRTH(bp, logical, physical)     \
 402 {                                               \
 403         ASSERT(!BP_IS_EMBEDDED(bp));            \
 404         (bp)->blk_birth = (logical);            \
 405         (bp)->blk_phys_birth = ((logical) == (physical) ? 0 : (physical)); \
 406 }
 407
 408 #define BP_GET_FILL(bp) (BP_IS_EMBEDDED(bp) ? 1 : (bp)->blk_fill)
 409
 410 #define BP_GET_ASIZE(bp)        \
 411         (BP_IS_EMBEDDED(bp) ? 0 : \
 412         DVA_GET_ASIZE(&(bp)->blk_dva[0]) + \
 413         DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \
 414         DVA_GET_ASIZE(&(bp)->blk_dva[2]))
 415
 416 #define BP_GET_UCSIZE(bp) \
 417         ((BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp))) ? \
 418         BP_GET_PSIZE(bp) : BP_GET_LSIZE(bp))
 419
 420 #define BP_GET_NDVAS(bp)        \
 421         (BP_IS_EMBEDDED(bp) ? 0 : \
 422         !!DVA_GET_ASIZE(&(bp)->blk_dva[0]) + \
 423         !!DVA_GET_ASIZE(&(bp)->blk_dva[1]) + \
 424         !!DVA_GET_ASIZE(&(bp)->blk_dva[2]))
 425
 426 #define BP_COUNT_GANG(bp)       \
 427         (BP_IS_EMBEDDED(bp) ? 0 : \
 428         (DVA_GET_GANG(&(bp)->blk_dva[0]) + \
 429         DVA_GET_GANG(&(bp)->blk_dva[1]) + \
 430         DVA_GET_GANG(&(bp)->blk_dva[2])))
 431
 432 #define DVA_EQUAL(dva1, dva2)   \
 433         ((dva1)->dva_word[1] == (dva2)->dva_word[1] && \
 434         (dva1)->dva_word[0] == (dva2)->dva_word[0])
 435
 436 #define BP_EQUAL(bp1, bp2)      \
 437         (BP_PHYSICAL_BIRTH(bp1) == BP_PHYSICAL_BIRTH(bp2) &&    \
 438         (bp1)->blk_birth == (bp2)->blk_birth &&                 \
 439         DVA_EQUAL(&(bp1)->blk_dva[0], &(bp2)->blk_dva[0]) &&    \
 440         DVA_EQUAL(&(bp1)->blk_dva[1], &(bp2)->blk_dva[1]) &&    \
 441         DVA_EQUAL(&(bp1)->blk_dva[2], &(bp2)->blk_dva[2]))
 442
 443 #define ZIO_CHECKSUM_EQUAL(zc1, zc2) \
 444         (0 == (((zc1).zc_word[0] - (zc2).zc_word[0]) | \
 445         ((zc1).zc_word[1] - (zc2).zc_word[1]) | \
 446         ((zc1).zc_word[2] - (zc2).zc_word[2]) | \
 447         ((zc1).zc_word[3] - (zc2).zc_word[3])))
 448
 449 #define DVA_IS_VALID(dva)       (DVA_GET_ASIZE(dva) != 0)
 450
 451 #define ZIO_SET_CHECKSUM(zcp, w0, w1, w2, w3)   \
 452 {                                               \
 453         (zcp)->zc_word[0] = w0;                 \
 454         (zcp)->zc_word[1] = w1;                 \
 455         (zcp)->zc_word[2] = w2;                 \
 456         (zcp)->zc_word[3] = w3;                 \
 457 }
 458
 459 #define BP_IDENTITY(bp)         (ASSERT(!BP_IS_EMBEDDED(bp)), &(bp)->blk_dva[0])
 460 #define BP_IS_GANG(bp)          \
 461         (BP_IS_EMBEDDED(bp) ? B_FALSE : DVA_GET_GANG(BP_IDENTITY(bp)))
 462 #define DVA_IS_EMPTY(dva)       ((dva)->dva_word[0] == 0ULL &&  \
 463                                 (dva)->dva_word[1] == 0ULL)
 464 #define BP_IS_HOLE(bp) \
 465         (!BP_IS_EMBEDDED(bp) && DVA_IS_EMPTY(BP_IDENTITY(bp)))
 466
 467 /* BP_IS_RAIDZ(bp) assumes no block compression */
 468 #define BP_IS_RAIDZ(bp)         (DVA_GET_ASIZE(&(bp)->blk_dva[0]) > \
 469                                 BP_GET_PSIZE(bp))
 470
 471 #define BP_ZERO(bp)                             \
 472 {                                               \
 473         (bp)->blk_dva[0].dva_word[0] = 0;       \
 474         (bp)->blk_dva[0].dva_word[1] = 0;       \
 475         (bp)->blk_dva[1].dva_word[0] = 0;       \
 476         (bp)->blk_dva[1].dva_word[1] = 0;       \
 477         (bp)->blk_dva[2].dva_word[0] = 0;       \
 478         (bp)->blk_dva[2].dva_word[1] = 0;       \
 479         (bp)->blk_prop = 0;                     \
 480         (bp)->blk_pad[0] = 0;                   \
 481         (bp)->blk_pad[1] = 0;                   \
 482         (bp)->blk_phys_birth = 0;               \
 483         (bp)->blk_birth = 0;                    \
 484         (bp)->blk_fill = 0;                     \
 485         ZIO_SET_CHECKSUM(&(bp)->blk_cksum, 0, 0, 0, 0); \
 486 }
 487
 488 #ifdef _BIG_ENDIAN
 489 #define ZFS_HOST_BYTEORDER      (0ULL)
 490 #else
 491 #define ZFS_HOST_BYTEORDER      (1ULL)
 492 #endif
 493
 494 #define BP_SHOULD_BYTESWAP(bp)  (BP_GET_BYTEORDER(bp) != ZFS_HOST_BYTEORDER)
 495
 496 #define BP_SPRINTF_LEN  320
 497
 498 /*
 499  * This macro allows code sharing between zfs, libzpool, and mdb.
 500  * 'func' is either snprintf() or mdb_snprintf().
 501  * 'ws' (whitespace) can be ' ' for single-line format, '\n' for multi-line.
 502  */
 503 #define SNPRINTF_BLKPTR(func, ws, buf, size, bp, type, checksum, compress) \
 504 {                                                                       \
 505         static const char *copyname[] =                                 \
 506             { "zero", "single", "double", "triple" };                   \
 507         int len = 0;                                                    \
 508         int copies = 0;                                                 \
 509         int d;                                                          \
 510                                                                         \
 511         if (bp == NULL) {                                               \
 512                 len += func(buf + len, size - len, "<NULL>");           \
 513         } else if (BP_IS_HOLE(bp)) {                                    \
 514                 len += func(buf + len, size - len,                      \
 515                     "HOLE [L%llu %s] "                                  \
 516                     "size=%llxL birth=%lluL",                           \
 517                     (u_longlong_t)BP_GET_LEVEL(bp),                     \
 518                     type,                                               \
 519                     (u_longlong_t)BP_GET_LSIZE(bp),                     \
 520                     (u_longlong_t)bp->blk_birth);                       \
 521         } else if (BP_IS_EMBEDDED(bp)) {                                \
 522                 len = func(buf + len, size - len,                       \
 523                     "EMBEDDED [L%llu %s] et=%u %s "                     \
 524                     "size=%llxL/%llxP birth=%lluL",                     \
 525                     (u_longlong_t)BP_GET_LEVEL(bp),                     \
 526                     type,                                               \
 527                     (int)BPE_GET_ETYPE(bp),                             \
 528                     compress,                                           \
 529                     (u_longlong_t)BPE_GET_LSIZE(bp),                    \
 530                     (u_longlong_t)BPE_GET_PSIZE(bp),                    \
 531                     (u_longlong_t)bp->blk_birth);                       \
 532         } else {                                                        \
 533                 for (d = 0; d < BP_GET_NDVAS(bp); d++) {                \
 534                         const dva_t *dva = &bp->blk_dva[d];             \
 535                         if (DVA_IS_VALID(dva))                          \
 536                                 copies++;                               \
 537                         len += func(buf + len, size - len,              \
 538                             "DVA[%d]=<%llu:%llx:%llx>%c", d,            \
 539                             (u_longlong_t)DVA_GET_VDEV(dva),            \
 540                             (u_longlong_t)DVA_GET_OFFSET(dva),          \
 541                             (u_longlong_t)DVA_GET_ASIZE(dva),           \
 542                             ws);                                        \
 543                 }                                                       \
 544                 if (BP_IS_GANG(bp) &&                                   \
 545                     DVA_GET_ASIZE(&bp->blk_dva[2]) <=                   \
 546                     DVA_GET_ASIZE(&bp->blk_dva[1]) / 2)                 \
 547                         copies--;                                       \
 548                 len += func(buf + len, size - len,                      \
 549                     "[L%llu %s] %s %s %s %s %s %s%c"                    \
 550                     "size=%llxL/%llxP birth=%lluL/%lluP fill=%llu%c"    \
 551                     "cksum=%llx:%llx:%llx:%llx",                        \
 552                     (u_longlong_t)BP_GET_LEVEL(bp),                     \
 553                     type,                                               \
 554                     checksum,                                           \
 555                     compress,                                           \
 556                     BP_GET_BYTEORDER(bp) == 0 ? "BE" : "LE",            \
 557                     BP_IS_GANG(bp) ? "gang" : "contiguous",             \
 558                     BP_GET_DEDUP(bp) ? "dedup" : "unique",              \
 559                     copyname[copies],                                   \
 560                     ws,                                                 \
 561                     (u_longlong_t)BP_GET_LSIZE(bp),                     \
 562                     (u_longlong_t)BP_GET_PSIZE(bp),                     \
 563                     (u_longlong_t)bp->blk_birth,                        \
 564                     (u_longlong_t)BP_PHYSICAL_BIRTH(bp),                \
 565                     (u_longlong_t)BP_GET_FILL(bp),                      \
 566                     ws,                                                 \
 567                     (u_longlong_t)bp->blk_cksum.zc_word[0],             \
 568                     (u_longlong_t)bp->blk_cksum.zc_word[1],             \
 569                     (u_longlong_t)bp->blk_cksum.zc_word[2],             \
 570                     (u_longlong_t)bp->blk_cksum.zc_word[3]);            \
 571         }                                                               \
 572         ASSERT(len < size);                                             \
 573 }
 574
 575 #include <sys/dmu.h>
 576
 577 #define BP_GET_BUFC_TYPE(bp)                                            \
 578         (((BP_GET_LEVEL(bp) > 0) || (DMU_OT_IS_METADATA(BP_GET_TYPE(bp)))) ? \
 579         ARC_BUFC_METADATA : ARC_BUFC_DATA)
 580
 581 typedef enum spa_import_type {
 582         SPA_IMPORT_EXISTING,
 583         SPA_IMPORT_ASSEMBLE
 584 } spa_import_type_t;
 585
 586 /* state manipulation functions */
 587 extern int spa_open(const char *pool, spa_t **, void *tag);
 588 extern int spa_open_rewind(const char *pool, spa_t **, void *tag,
 589     nvlist_t *policy, nvlist_t **config);
 590 extern int spa_get_stats(const char *pool, nvlist_t **config, char *altroot,
 591     size_t buflen);
 592 extern int spa_create(const char *pool, nvlist_t *config, nvlist_t *props,
 593     nvlist_t *zplprops);
 594 extern int spa_import_rootpool(char *devpath, char *devid);
 595 extern int spa_import(char *pool, nvlist_t *config, nvlist_t *props,
 596     uint64_t flags);
 597 extern nvlist_t *spa_tryimport(nvlist_t *tryconfig);
 598 extern int spa_destroy(char *pool);
 599 extern int spa_export(char *pool, nvlist_t **oldconfig, boolean_t force,
 600     boolean_t hardforce);
 601 extern int spa_reset(char *pool);
 602 extern void spa_async_request(spa_t *spa, int flag);
 603 extern void spa_async_unrequest(spa_t *spa, int flag);
 604 extern void spa_async_suspend(spa_t *spa);
 605 extern void spa_async_resume(spa_t *spa);
 606 extern spa_t *spa_inject_addref(char *pool);
 607 extern void spa_inject_delref(spa_t *spa);
 608 extern void spa_scan_stat_init(spa_t *spa);
 609 extern int spa_scan_get_stats(spa_t *spa, pool_scan_stat_t *ps);
 610
 611 #define SPA_ASYNC_CONFIG_UPDATE 0x01
 612 #define SPA_ASYNC_REMOVE        0x02
 613 #define SPA_ASYNC_PROBE         0x04
 614 #define SPA_ASYNC_RESILVER_DONE 0x08
 615 #define SPA_ASYNC_RESILVER      0x10
 616 #define SPA_ASYNC_AUTOEXPAND    0x20
 617 #define SPA_ASYNC_REMOVE_DONE   0x40
 618 #define SPA_ASYNC_REMOVE_STOP   0x80
 619
 620 /*
 621  * Controls the behavior of spa_vdev_remove().
 622  */
 623 #define SPA_REMOVE_UNSPARE      0x01
 624 #define SPA_REMOVE_DONE         0x02
 625
 626 /* device manipulation */
 627 extern int spa_vdev_add(spa_t *spa, nvlist_t *nvroot);
 628 extern int spa_vdev_attach(spa_t *spa, uint64_t guid, nvlist_t *nvroot,
 629     int replacing);
 630 extern int spa_vdev_detach(spa_t *spa, uint64_t guid, uint64_t pguid,
 631     int replace_done);
 632 extern int spa_vdev_remove(spa_t *spa, uint64_t guid, boolean_t unspare);
 633 extern boolean_t spa_vdev_remove_active(spa_t *spa);
 634 extern int spa_vdev_setpath(spa_t *spa, uint64_t guid, const char *newpath);
 635 extern int spa_vdev_setfru(spa_t *spa, uint64_t guid, const char *newfru);
 636 extern int spa_vdev_split_mirror(spa_t *spa, char *newname, nvlist_t *config,
 637     nvlist_t *props, boolean_t exp);
 638
 639 /* spare state (which is global across all pools) */
 640 extern void spa_spare_add(vdev_t *vd);
 641 extern void spa_spare_remove(vdev_t *vd);
 642 extern boolean_t spa_spare_exists(uint64_t guid, uint64_t *pool, int *refcnt);
 643 extern void spa_spare_activate(vdev_t *vd);
 644
 645 /* L2ARC state (which is global across all pools) */
 646 extern void spa_l2cache_add(vdev_t *vd);
 647 extern void spa_l2cache_remove(vdev_t *vd);
 648 extern boolean_t spa_l2cache_exists(uint64_t guid, uint64_t *pool);
 649 extern void spa_l2cache_activate(vdev_t *vd);
 650 extern void spa_l2cache_drop(spa_t *spa);
 651
 652 /* scanning */
 653 extern int spa_scan(spa_t *spa, pool_scan_func_t func);
 654 extern int spa_scan_stop(spa_t *spa);
 655
 656 /* spa syncing */
 657 extern void spa_sync(spa_t *spa, uint64_t txg); /* only for DMU use */
 658 extern void spa_sync_allpools(void);
 659
 660 extern int zfs_sync_pass_deferred_free;
 661
 662 /* spa namespace global mutex */
 663 extern kmutex_t spa_namespace_lock;
 664
 665 /*
 666  * SPA configuration functions in spa_config.c
 667  */
 668
 669 #define SPA_CONFIG_UPDATE_POOL  0
 670 #define SPA_CONFIG_UPDATE_VDEVS 1
 671
 672 extern void spa_config_sync(spa_t *, boolean_t, boolean_t);
 673 extern void spa_config_load(void);
 674 extern nvlist_t *spa_all_configs(uint64_t *);
 675 extern void spa_config_set(spa_t *spa, nvlist_t *config);
 676 extern nvlist_t *spa_config_generate(spa_t *spa, vdev_t *vd, uint64_t txg,
 677     int getstats);
 678 extern void spa_config_update(spa_t *spa, int what);
 679
 680 /*
 681  * Miscellaneous SPA routines in spa_misc.c
 682  */
 683
 684 /* Namespace manipulation */
 685 extern spa_t *spa_lookup(const char *name);
 686 extern spa_t *spa_add(const char *name, nvlist_t *config, const char *altroot);
 687 extern void spa_remove(spa_t *spa);
 688 extern spa_t *spa_next(spa_t *prev);
 689
 690 /* Refcount functions */
 691 extern void spa_open_ref(spa_t *spa, void *tag);
 692 extern void spa_close(spa_t *spa, void *tag);
 693 extern void spa_async_close(spa_t *spa, void *tag);
 694 extern boolean_t spa_refcount_zero(spa_t *spa);
 695
 696 #define SCL_NONE        0x00
 697 #define SCL_CONFIG      0x01
 698 #define SCL_STATE       0x02
 699 #define SCL_L2ARC       0x04            /* hack until L2ARC 2.0 */
 700 #define SCL_ALLOC       0x08
 701 #define SCL_ZIO         0x10
 702 #define SCL_FREE        0x20
 703 #define SCL_VDEV        0x40
 704 #define SCL_LOCKS       7
 705 #define SCL_ALL         ((1 << SCL_LOCKS) - 1)
 706 #define SCL_STATE_ALL   (SCL_STATE | SCL_L2ARC | SCL_ZIO)
 707
 708 /* Historical pool statistics */
 709 typedef struct spa_stats_history {
 710         kmutex_t                lock;
 711         uint64_t                count;
 712         uint64_t                size;
 713         kstat_t                 *kstat;
 714         void                    *private;
 715         list_t                  list;
 716 } spa_stats_history_t;
 717
 718 typedef struct spa_stats {
 719         spa_stats_history_t     read_history;
 720         spa_stats_history_t     txg_history;
 721         spa_stats_history_t     tx_assign_histogram;
 722         spa_stats_history_t     io_history;
 723 } spa_stats_t;
 724
 725 typedef enum txg_state {
 726         TXG_STATE_BIRTH         = 0,
 727         TXG_STATE_OPEN          = 1,
 728         TXG_STATE_QUIESCED      = 2,
 729         TXG_STATE_WAIT_FOR_SYNC = 3,
 730         TXG_STATE_SYNCED        = 4,
 731         TXG_STATE_COMMITTED     = 5,
 732 } txg_state_t;
 733
 734 extern void spa_stats_init(spa_t *spa);
 735 extern void spa_stats_destroy(spa_t *spa);
 736 extern void spa_read_history_add(spa_t *spa, const zbookmark_phys_t *zb,
 737     uint32_t aflags);
 738 extern void spa_txg_history_add(spa_t *spa, uint64_t txg, hrtime_t birth_time);
 739 extern int spa_txg_history_set(spa_t *spa,  uint64_t txg,
 740     txg_state_t completed_state, hrtime_t completed_time);
 741 extern int spa_txg_history_set_io(spa_t *spa,  uint64_t txg, uint64_t nread,
 742     uint64_t nwritten, uint64_t reads, uint64_t writes, uint64_t ndirty);
 743 extern void spa_tx_assign_add_nsecs(spa_t *spa, uint64_t nsecs);
 744
 745 /* Pool configuration locks */
 746 extern int spa_config_tryenter(spa_t *spa, int locks, void *tag, krw_t rw);
 747 extern void spa_config_enter(spa_t *spa, int locks, void *tag, krw_t rw);
 748 extern void spa_config_exit(spa_t *spa, int locks, void *tag);
 749 extern int spa_config_held(spa_t *spa, int locks, krw_t rw);
 750
 751 /* Pool vdev add/remove lock */
 752 extern uint64_t spa_vdev_enter(spa_t *spa);
 753 extern uint64_t spa_vdev_config_enter(spa_t *spa);
 754 extern void spa_vdev_config_exit(spa_t *spa, vdev_t *vd, uint64_t txg,
 755     int error, char *tag);
 756 extern int spa_vdev_exit(spa_t *spa, vdev_t *vd, uint64_t txg, int error);
 757
 758 /* Pool vdev state change lock */
 759 extern void spa_vdev_state_enter(spa_t *spa, int oplock);
 760 extern int spa_vdev_state_exit(spa_t *spa, vdev_t *vd, int error);
 761
 762 /* Log state */
 763 typedef enum spa_log_state {
 764         SPA_LOG_UNKNOWN = 0,    /* unknown log state */
 765         SPA_LOG_MISSING,        /* missing log(s) */
 766         SPA_LOG_CLEAR,          /* clear the log(s) */
 767         SPA_LOG_GOOD,           /* log(s) are good */
 768 } spa_log_state_t;
 769
 770 extern spa_log_state_t spa_get_log_state(spa_t *spa);
 771 extern void spa_set_log_state(spa_t *spa, spa_log_state_t state);
 772 extern int spa_offline_log(spa_t *spa);
 773
 774 /* Log claim callback */
 775 extern void spa_claim_notify(zio_t *zio);
 776 extern void spa_deadman(void *);
 777
 778 /* Accessor functions */
 779 extern boolean_t spa_shutting_down(spa_t *spa);
 780 extern struct dsl_pool *spa_get_dsl(spa_t *spa);
 781 extern boolean_t spa_is_initializing(spa_t *spa);
 782 extern blkptr_t *spa_get_rootblkptr(spa_t *spa);
 783 extern void spa_set_rootblkptr(spa_t *spa, const blkptr_t *bp);
 784 extern void spa_altroot(spa_t *, char *, size_t);
 785 extern int spa_sync_pass(spa_t *spa);
 786 extern char *spa_name(spa_t *spa);
 787 extern uint64_t spa_guid(spa_t *spa);
 788 extern uint64_t spa_load_guid(spa_t *spa);
 789 extern uint64_t spa_last_synced_txg(spa_t *spa);
 790 extern uint64_t spa_first_txg(spa_t *spa);
 791 extern uint64_t spa_syncing_txg(spa_t *spa);
 792 extern uint64_t spa_version(spa_t *spa);
 793 extern pool_state_t spa_state(spa_t *spa);
 794 extern spa_load_state_t spa_load_state(spa_t *spa);
 795 extern uint64_t spa_freeze_txg(spa_t *spa);
 796 extern uint64_t spa_get_asize(spa_t *spa, uint64_t lsize);
 797 extern uint64_t spa_get_dspace(spa_t *spa);
 798 extern uint64_t spa_get_slop_space(spa_t *spa);
 799 extern void spa_update_dspace(spa_t *spa);
 800 extern uint64_t spa_version(spa_t *spa);
 801 extern boolean_t spa_deflate(spa_t *spa);
 802 extern metaslab_class_t *spa_normal_class(spa_t *spa);
 803 extern metaslab_class_t *spa_log_class(spa_t *spa);
 804 extern void spa_evicting_os_register(spa_t *, objset_t *os);
 805 extern void spa_evicting_os_deregister(spa_t *, objset_t *os);
 806 extern void spa_evicting_os_wait(spa_t *spa);
 807 extern int spa_max_replication(spa_t *spa);
 808 extern int spa_prev_software_version(spa_t *spa);
 809 extern uint8_t spa_get_failmode(spa_t *spa);
 810 extern boolean_t spa_suspended(spa_t *spa);
 811 extern uint64_t spa_bootfs(spa_t *spa);
 812 extern uint64_t spa_delegation(spa_t *spa);
 813 extern objset_t *spa_meta_objset(spa_t *spa);
 814 extern uint64_t spa_deadman_synctime(spa_t *spa);
 815
 816 /* Miscellaneous support routines */
 817 extern void spa_activate_mos_feature(spa_t *spa, const char *feature,
 818     dmu_tx_t *tx);
 819 extern void spa_deactivate_mos_feature(spa_t *spa, const char *feature);
 820 extern int spa_rename(const char *oldname, const char *newname);
 821 extern spa_t *spa_by_guid(uint64_t pool_guid, uint64_t device_guid);
 822 extern boolean_t spa_guid_exists(uint64_t pool_guid, uint64_t device_guid);
 823 extern char *spa_strdup(const char *);
 824 extern void spa_strfree(char *);
 825 extern uint64_t spa_get_random(uint64_t range);
 826 extern uint64_t spa_generate_guid(spa_t *spa);
 827 extern void snprintf_blkptr(char *buf, size_t buflen, const blkptr_t *bp);
 828 extern void spa_freeze(spa_t *spa);
 829 extern int spa_change_guid(spa_t *spa);
 830 extern void spa_upgrade(spa_t *spa, uint64_t version);
 831 extern void spa_evict_all(void);
 832 extern vdev_t *spa_lookup_by_guid(spa_t *spa, uint64_t guid,
 833     boolean_t l2cache);
 834 extern boolean_t spa_has_spare(spa_t *, uint64_t guid);
 835 extern uint64_t dva_get_dsize_sync(spa_t *spa, const dva_t *dva);
 836 extern uint64_t bp_get_dsize_sync(spa_t *spa, const blkptr_t *bp);
 837 extern uint64_t bp_get_dsize(spa_t *spa, const blkptr_t *bp);
 838 extern boolean_t spa_has_slogs(spa_t *spa);
 839 extern boolean_t spa_is_root(spa_t *spa);
 840 extern boolean_t spa_writeable(spa_t *spa);
 841 extern boolean_t spa_has_pending_synctask(spa_t *spa);
 842 extern int spa_maxblocksize(spa_t *spa);
 843 extern void zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp);
 844
 845 extern int spa_mode(spa_t *spa);
 846 extern uint64_t strtonum(const char *str, char **nptr);
 847
 848 extern char *spa_his_ievent_table[];
 849
 850 extern void spa_history_create_obj(spa_t *spa, dmu_tx_t *tx);
 851 extern int spa_history_get(spa_t *spa, uint64_t *offset, uint64_t *len_read,
 852     char *his_buf);
 853 extern int spa_history_log(spa_t *spa, const char *his_buf);
 854 extern int spa_history_log_nvl(spa_t *spa, nvlist_t *nvl);
 855 extern void spa_history_log_version(spa_t *spa, const char *operation);
 856 extern void spa_history_log_internal(spa_t *spa, const char *operation,
 857     dmu_tx_t *tx, const char *fmt, ...);
 858 extern void spa_history_log_internal_ds(struct dsl_dataset *ds, const char *op,
 859     dmu_tx_t *tx, const char *fmt, ...);
 860 extern void spa_history_log_internal_dd(dsl_dir_t *dd, const char *operation,
 861     dmu_tx_t *tx, const char *fmt, ...);
 862
 863 /* error handling */
 864 struct zbookmark_phys;
 865 extern void spa_log_error(spa_t *spa, zio_t *zio);
 866 extern void zfs_ereport_post(const char *class, spa_t *spa, vdev_t *vd,
 867     zio_t *zio, uint64_t stateoroffset, uint64_t length);
 868 extern void zfs_post_remove(spa_t *spa, vdev_t *vd);
 869 extern void zfs_post_state_change(spa_t *spa, vdev_t *vd);
 870 extern void zfs_post_autoreplace(spa_t *spa, vdev_t *vd);
 871 extern uint64_t spa_get_errlog_size(spa_t *spa);
 872 extern int spa_get_errlog(spa_t *spa, void *uaddr, size_t *count);
 873 extern void spa_errlog_rotate(spa_t *spa);
 874 extern void spa_errlog_drain(spa_t *spa);
 875 extern void spa_errlog_sync(spa_t *spa, uint64_t txg);
 876 extern void spa_get_errlists(spa_t *spa, avl_tree_t *last, avl_tree_t *scrub);
 877
 878 /* vdev cache */
 879 extern void vdev_cache_stat_init(void);
 880 extern void vdev_cache_stat_fini(void);
 881
 882 /* Initialization and termination */
 883 extern void spa_init(int flags);
 884 extern void spa_fini(void);
 885 extern void spa_boot_init(void);
 886
 887 /* properties */
 888 extern int spa_prop_set(spa_t *spa, nvlist_t *nvp);
 889 extern int spa_prop_get(spa_t *spa, nvlist_t **nvp);
 890 extern void spa_prop_clear_bootfs(spa_t *spa, uint64_t obj, dmu_tx_t *tx);
 891 extern void spa_configfile_set(spa_t *, nvlist_t *, boolean_t);
 892
 893 /* asynchronous event notification */
 894 extern void spa_event_notify(spa_t *spa, vdev_t *vdev, const char *name);
 895
 896 #ifdef ZFS_DEBUG
 897 #define dprintf_bp(bp, fmt, ...) do {                           \
 898         if (zfs_flags & ZFS_DEBUG_DPRINTF) {                    \
 899         char *__blkbuf = kmem_alloc(BP_SPRINTF_LEN, KM_SLEEP);  \
 900         snprintf_blkptr(__blkbuf, BP_SPRINTF_LEN, (bp));        \
 901         dprintf(fmt " %s\n", __VA_ARGS__, __blkbuf);            \
 902         kmem_free(__blkbuf, BP_SPRINTF_LEN);                    \
 903         } \
 904 _NOTE(CONSTCOND) } while (0)
 905 #else
 906 #define dprintf_bp(bp, fmt, ...)
 907 #endif
 908
 909 extern boolean_t spa_debug_enabled(spa_t *spa);
 910 #define spa_dbgmsg(spa, ...)                    \
 911 {                                               \
 912         if (spa_debug_enabled(spa))             \
 913                 zfs_dbgmsg(__VA_ARGS__);        \
 914 }
 915
 916 extern int spa_mode_global;                     /* mode, e.g. FREAD | FWRITE */
 917
 918 #ifdef  __cplusplus
 919 }
 920 #endif
 921
 922 #endif  /* _SYS_SPA_H */