include/sys/zio.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 /*
  23  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
  25  * Copyright (c) 2013 by Delphix. All rights reserved.
  26  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  27  */
  28
  29 #ifndef _ZIO_H
  30 #define _ZIO_H
  31
  32 #include <sys/zfs_context.h>
  33 #include <sys/spa.h>
  34 #include <sys/txg.h>
  35 #include <sys/avl.h>
  36 #include <sys/fs/zfs.h>
  37 #include <sys/zio_impl.h>
  38
  39 #ifdef  __cplusplus
  40 extern "C" {
  41 #endif
  42
  43 /*
  44  * Embedded checksum
  45  */
  46 #define ZEC_MAGIC       0x210da7ab10c7a11ULL
  47
  48 typedef struct zio_eck {
  49         uint64_t        zec_magic;      /* for validation, endianness   */
  50         zio_cksum_t     zec_cksum;      /* 256-bit checksum             */
  51 } zio_eck_t;
  52
  53 /*
  54  * Gang block headers are self-checksumming and contain an array
  55  * of block pointers.
  56  */
  57 #define SPA_GANGBLOCKSIZE       SPA_MINBLOCKSIZE
  58 #define SPA_GBH_NBLKPTRS        ((SPA_GANGBLOCKSIZE - \
  59         sizeof (zio_eck_t)) / sizeof (blkptr_t))
  60 #define SPA_GBH_FILLER          ((SPA_GANGBLOCKSIZE - \
  61         sizeof (zio_eck_t) - \
  62         (SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\
  63         sizeof (uint64_t))
  64
  65 typedef struct zio_gbh {
  66         blkptr_t                zg_blkptr[SPA_GBH_NBLKPTRS];
  67         uint64_t                zg_filler[SPA_GBH_FILLER];
  68         zio_eck_t               zg_tail;
  69 } zio_gbh_phys_t;
  70
  71 enum zio_checksum {
  72         ZIO_CHECKSUM_INHERIT = 0,
  73         ZIO_CHECKSUM_ON,
  74         ZIO_CHECKSUM_OFF,
  75         ZIO_CHECKSUM_LABEL,
  76         ZIO_CHECKSUM_GANG_HEADER,
  77         ZIO_CHECKSUM_ZILOG,
  78         ZIO_CHECKSUM_FLETCHER_2,
  79         ZIO_CHECKSUM_FLETCHER_4,
  80         ZIO_CHECKSUM_SHA256,
  81         ZIO_CHECKSUM_ZILOG2,
  82         ZIO_CHECKSUM_FUNCTIONS
  83 };
  84
  85 /*
  86  * The number of "legacy" compression functions which can be set on individual
  87  * objects.
  88  */
  89 #define ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2
  90
  91 #define ZIO_CHECKSUM_ON_VALUE   ZIO_CHECKSUM_FLETCHER_4
  92 #define ZIO_CHECKSUM_DEFAULT    ZIO_CHECKSUM_ON
  93
  94 #define ZIO_CHECKSUM_MASK       0xffULL
  95 #define ZIO_CHECKSUM_VERIFY     (1 << 8)
  96
  97 #define ZIO_DEDUPCHECKSUM       ZIO_CHECKSUM_SHA256
  98 #define ZIO_DEDUPDITTO_MIN      100
  99
 100 enum zio_compress {
 101         ZIO_COMPRESS_INHERIT = 0,
 102         ZIO_COMPRESS_ON,
 103         ZIO_COMPRESS_OFF,
 104         ZIO_COMPRESS_LZJB,
 105         ZIO_COMPRESS_EMPTY,
 106         ZIO_COMPRESS_GZIP_1,
 107         ZIO_COMPRESS_GZIP_2,
 108         ZIO_COMPRESS_GZIP_3,
 109         ZIO_COMPRESS_GZIP_4,
 110         ZIO_COMPRESS_GZIP_5,
 111         ZIO_COMPRESS_GZIP_6,
 112         ZIO_COMPRESS_GZIP_7,
 113         ZIO_COMPRESS_GZIP_8,
 114         ZIO_COMPRESS_GZIP_9,
 115         ZIO_COMPRESS_ZLE,
 116         ZIO_COMPRESS_LZ4,
 117         ZIO_COMPRESS_FUNCTIONS
 118 };
 119
 120 /*
 121  * The number of "legacy" compression functions which can be set on individual
 122  * objects.
 123  */
 124 #define ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
 125
 126 #define ZIO_COMPRESS_ON_VALUE   ZIO_COMPRESS_LZJB
 127 #define ZIO_COMPRESS_DEFAULT    ZIO_COMPRESS_OFF
 128
 129 #define BOOTFS_COMPRESS_VALID(compress)                 \
 130         ((compress) == ZIO_COMPRESS_LZJB ||             \
 131         (compress) == ZIO_COMPRESS_LZ4 ||               \
 132         ((compress) == ZIO_COMPRESS_ON &&               \
 133         ZIO_COMPRESS_ON_VALUE == ZIO_COMPRESS_LZJB) ||  \
 134         (compress) == ZIO_COMPRESS_OFF)
 135
 136 /*
 137  * Default Linux timeout for a sd device.
 138  */
 139 #define ZIO_DELAY_MAX                   (30 * MILLISEC)
 140
 141 #define ZIO_FAILURE_MODE_WAIT           0
 142 #define ZIO_FAILURE_MODE_CONTINUE       1
 143 #define ZIO_FAILURE_MODE_PANIC          2
 144
 145 typedef enum zio_priority {
 146         ZIO_PRIORITY_SYNC_READ,
 147         ZIO_PRIORITY_SYNC_WRITE,        /* ZIL */
 148         ZIO_PRIORITY_ASYNC_READ,        /* prefetch */
 149         ZIO_PRIORITY_ASYNC_WRITE,       /* spa_sync() */
 150         ZIO_PRIORITY_SCRUB,             /* asynchronous scrub/resilver reads */
 151         ZIO_PRIORITY_NUM_QUEUEABLE,
 152
 153         ZIO_PRIORITY_NOW                /* non-queued i/os (e.g. free) */
 154 } zio_priority_t;
 155
 156 #define ZIO_PIPELINE_CONTINUE           0x100
 157 #define ZIO_PIPELINE_STOP               0x101
 158
 159 enum zio_flag {
 160         /*
 161          * Flags inherited by gang, ddt, and vdev children,
 162          * and that must be equal for two zios to aggregate
 163          */
 164         ZIO_FLAG_DONT_AGGREGATE = 1 << 0,
 165         ZIO_FLAG_IO_REPAIR      = 1 << 1,
 166         ZIO_FLAG_SELF_HEAL      = 1 << 2,
 167         ZIO_FLAG_RESILVER       = 1 << 3,
 168         ZIO_FLAG_SCRUB          = 1 << 4,
 169         ZIO_FLAG_SCAN_THREAD    = 1 << 5,
 170
 171 #define ZIO_FLAG_AGG_INHERIT    (ZIO_FLAG_CANFAIL - 1)
 172
 173         /*
 174          * Flags inherited by ddt, gang, and vdev children.
 175          */
 176         ZIO_FLAG_CANFAIL        = 1 << 6,       /* must be first for INHERIT */
 177         ZIO_FLAG_SPECULATIVE    = 1 << 7,
 178         ZIO_FLAG_CONFIG_WRITER  = 1 << 8,
 179         ZIO_FLAG_DONT_RETRY     = 1 << 9,
 180         ZIO_FLAG_DONT_CACHE     = 1 << 10,
 181         ZIO_FLAG_NODATA         = 1 << 11,
 182         ZIO_FLAG_INDUCE_DAMAGE  = 1 << 12,
 183
 184 #define ZIO_FLAG_DDT_INHERIT    (ZIO_FLAG_IO_RETRY - 1)
 185 #define ZIO_FLAG_GANG_INHERIT   (ZIO_FLAG_IO_RETRY - 1)
 186
 187         /*
 188          * Flags inherited by vdev children.
 189          */
 190         ZIO_FLAG_IO_RETRY       = 1 << 13,      /* must be first for INHERIT */
 191         ZIO_FLAG_PROBE          = 1 << 14,
 192         ZIO_FLAG_TRYHARD        = 1 << 15,
 193         ZIO_FLAG_OPTIONAL       = 1 << 16,
 194
 195 #define ZIO_FLAG_VDEV_INHERIT   (ZIO_FLAG_DONT_QUEUE - 1)
 196
 197         /*
 198          * Flags not inherited by any children.
 199          */
 200         ZIO_FLAG_DONT_QUEUE     = 1 << 17,      /* must be first for INHERIT */
 201         ZIO_FLAG_DONT_PROPAGATE = 1 << 18,
 202         ZIO_FLAG_IO_BYPASS      = 1 << 19,
 203         ZIO_FLAG_IO_REWRITE     = 1 << 20,
 204         ZIO_FLAG_RAW            = 1 << 21,
 205         ZIO_FLAG_GANG_CHILD     = 1 << 22,
 206         ZIO_FLAG_DDT_CHILD      = 1 << 23,
 207         ZIO_FLAG_GODFATHER      = 1 << 24,
 208         ZIO_FLAG_NOPWRITE       = 1 << 25,
 209         ZIO_FLAG_REEXECUTED     = 1 << 26,
 210         ZIO_FLAG_DELEGATED      = 1 << 27,
 211         ZIO_FLAG_FASTWRITE      = 1 << 28
 212 };
 213
 214 #define ZIO_FLAG_MUSTSUCCEED            0
 215
 216 #define ZIO_DDT_CHILD_FLAGS(zio)                                \
 217         (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |             \
 218         ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
 219
 220 #define ZIO_GANG_CHILD_FLAGS(zio)                               \
 221         (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |            \
 222         ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
 223
 224 #define ZIO_VDEV_CHILD_FLAGS(zio)                               \
 225         (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |            \
 226         ZIO_FLAG_CANFAIL)
 227
 228 enum zio_child {
 229         ZIO_CHILD_VDEV = 0,
 230         ZIO_CHILD_GANG,
 231         ZIO_CHILD_DDT,
 232         ZIO_CHILD_LOGICAL,
 233         ZIO_CHILD_TYPES
 234 };
 235
 236 enum zio_wait_type {
 237         ZIO_WAIT_READY = 0,
 238         ZIO_WAIT_DONE,
 239         ZIO_WAIT_TYPES
 240 };
 241
 242 /*
 243  * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent
 244  * graveyard) to indicate checksum errors and fragmentation.
 245  */
 246 #define ECKSUM  EBADE
 247 #define EFRAGS  EBADR
 248
 249 typedef void zio_done_func_t(zio_t *zio);
 250
 251 extern const char *zio_type_name[ZIO_TYPES];
 252
 253 /*
 254  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
 255  * identifies any block in the pool.  By convention, the meta-objset (MOS)
 256  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
 257  * except root blocks and ZIL blocks, which are defined as follows:
 258  *
 259  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
 260  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
 261  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
 262  *
 263  * Note: this structure is called a bookmark because its original purpose
 264  * was to remember where to resume a pool-wide traverse.
 265  *
 266  * Note: this structure is passed between userland and the kernel.
 267  * Therefore it must not change size or alignment between 32/64 bit
 268  * compilation options.
 269  */
 270 struct zbookmark {
 271         uint64_t        zb_objset;
 272         uint64_t        zb_object;
 273         int64_t         zb_level;
 274         uint64_t        zb_blkid;
 275 };
 276
 277 #define SET_BOOKMARK(zb, objset, object, level, blkid)  \
 278 {                                                       \
 279         (zb)->zb_objset = objset;                       \
 280         (zb)->zb_object = object;                       \
 281         (zb)->zb_level = level;                         \
 282         (zb)->zb_blkid = blkid;                         \
 283 }
 284
 285 #define ZB_DESTROYED_OBJSET     (-1ULL)
 286
 287 #define ZB_ROOT_OBJECT          (0ULL)
 288 #define ZB_ROOT_LEVEL           (-1LL)
 289 #define ZB_ROOT_BLKID           (0ULL)
 290
 291 #define ZB_ZIL_OBJECT           (0ULL)
 292 #define ZB_ZIL_LEVEL            (-2LL)
 293
 294 #define ZB_IS_ZERO(zb)                                          \
 295         ((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&        \
 296         (zb)->zb_level == 0 && (zb)->zb_blkid == 0)
 297 #define ZB_IS_ROOT(zb)                          \
 298         ((zb)->zb_object == ZB_ROOT_OBJECT &&   \
 299         (zb)->zb_level == ZB_ROOT_LEVEL &&      \
 300         (zb)->zb_blkid == ZB_ROOT_BLKID)
 301
 302 typedef struct zio_prop {
 303         enum zio_checksum       zp_checksum;
 304         enum zio_compress       zp_compress;
 305         dmu_object_type_t       zp_type;
 306         uint8_t                 zp_level;
 307         uint8_t                 zp_copies;
 308         boolean_t               zp_dedup;
 309         boolean_t               zp_dedup_verify;
 310         boolean_t               zp_nopwrite;
 311 } zio_prop_t;
 312
 313 typedef struct zio_cksum_report zio_cksum_report_t;
 314
 315 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
 316     const void *good_data);
 317 typedef void zio_cksum_free_f(void *cbdata, size_t size);
 318
 319 struct zio_bad_cksum;                           /* defined in zio_checksum.h */
 320 struct dnode_phys;
 321
 322 struct zio_cksum_report {
 323         struct zio_cksum_report *zcr_next;
 324         nvlist_t                *zcr_ereport;
 325         nvlist_t                *zcr_detector;
 326         void                    *zcr_cbdata;
 327         size_t                  zcr_cbinfo;     /* passed to zcr_free() */
 328         uint64_t                zcr_align;
 329         uint64_t                zcr_length;
 330         zio_cksum_finish_f      *zcr_finish;
 331         zio_cksum_free_f        *zcr_free;
 332
 333         /* internal use only */
 334         struct zio_bad_cksum    *zcr_ckinfo;    /* information from failure */
 335 };
 336
 337 typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
 338     void *arg);
 339
 340 zio_vsd_cksum_report_f  zio_vsd_default_cksum_report;
 341
 342 typedef struct zio_vsd_ops {
 343         zio_done_func_t         *vsd_free;
 344         zio_vsd_cksum_report_f  *vsd_cksum_report;
 345 } zio_vsd_ops_t;
 346
 347 typedef struct zio_gang_node {
 348         zio_gbh_phys_t          *gn_gbh;
 349         struct zio_gang_node    *gn_child[SPA_GBH_NBLKPTRS];
 350 } zio_gang_node_t;
 351
 352 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
 353     zio_gang_node_t *gn, void *data);
 354
 355 typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size);
 356
 357 typedef struct zio_transform {
 358         void                    *zt_orig_data;
 359         uint64_t                zt_orig_size;
 360         uint64_t                zt_bufsize;
 361         zio_transform_func_t    *zt_transform;
 362         struct zio_transform    *zt_next;
 363 } zio_transform_t;
 364
 365 typedef int zio_pipe_stage_t(zio_t *zio);
 366
 367 /*
 368  * The io_reexecute flags are distinct from io_flags because the child must
 369  * be able to propagate them to the parent.  The normal io_flags are local
 370  * to the zio, not protected by any lock, and not modifiable by children;
 371  * the reexecute flags are protected by io_lock, modifiable by children,
 372  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
 373  */
 374 #define ZIO_REEXECUTE_NOW       0x01
 375 #define ZIO_REEXECUTE_SUSPEND   0x02
 376
 377 typedef struct zio_link {
 378         zio_t           *zl_parent;
 379         zio_t           *zl_child;
 380         list_node_t     zl_parent_node;
 381         list_node_t     zl_child_node;
 382 } zio_link_t;
 383
 384 struct zio {
 385         /* Core information about this I/O */
 386         zbookmark_t     io_bookmark;
 387         zio_prop_t      io_prop;
 388         zio_type_t      io_type;
 389         enum zio_child  io_child_type;
 390         int             io_cmd;
 391         zio_priority_t  io_priority;
 392         uint8_t         io_reexecute;
 393         uint8_t         io_state[ZIO_WAIT_TYPES];
 394         uint64_t        io_txg;
 395         spa_t           *io_spa;
 396         blkptr_t        *io_bp;
 397         blkptr_t        *io_bp_override;
 398         blkptr_t        io_bp_copy;
 399         list_t          io_parent_list;
 400         list_t          io_child_list;
 401         zio_link_t      *io_walk_link;
 402         zio_t           *io_logical;
 403         zio_transform_t *io_transform_stack;
 404
 405         /* Callback info */
 406         zio_done_func_t *io_ready;
 407         zio_done_func_t *io_physdone;
 408         zio_done_func_t *io_done;
 409         void            *io_private;
 410         int64_t         io_prev_space_delta;    /* DMU private */
 411         blkptr_t        io_bp_orig;
 412
 413         /* Data represented by this I/O */
 414         void            *io_data;
 415         void            *io_orig_data;
 416         uint64_t        io_size;
 417         uint64_t        io_orig_size;
 418
 419         /* Stuff for the vdev stack */
 420         vdev_t          *io_vd;
 421         void            *io_vsd;
 422         const zio_vsd_ops_t *io_vsd_ops;
 423
 424         uint64_t        io_offset;
 425         hrtime_t        io_timestamp;   /* submitted at */
 426         hrtime_t        io_delta;       /* vdev queue service delta */
 427         uint64_t        io_delay;       /* vdev disk service delta (ticks) */
 428         avl_node_t      io_queue_node;
 429
 430         /* Internal pipeline state */
 431         enum zio_flag   io_flags;
 432         enum zio_stage  io_stage;
 433         enum zio_stage  io_pipeline;
 434         enum zio_flag   io_orig_flags;
 435         enum zio_stage  io_orig_stage;
 436         enum zio_stage  io_orig_pipeline;
 437         int             io_error;
 438         int             io_child_error[ZIO_CHILD_TYPES];
 439         uint64_t        io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
 440         uint64_t        io_child_count;
 441         uint64_t        io_phys_children;
 442         uint64_t        io_parent_count;
 443         uint64_t        *io_stall;
 444         zio_t           *io_gang_leader;
 445         zio_gang_node_t *io_gang_tree;
 446         void            *io_executor;
 447         void            *io_waiter;
 448         kmutex_t        io_lock;
 449         kcondvar_t      io_cv;
 450
 451         /* FMA state */
 452         zio_cksum_report_t *io_cksum_report;
 453         uint64_t        io_ena;
 454
 455         /* Taskq dispatching state */
 456         taskq_ent_t     io_tqent;
 457 };
 458
 459 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
 460     zio_done_func_t *done, void *private, enum zio_flag flags);
 461
 462 extern zio_t *zio_root(spa_t *spa,
 463     zio_done_func_t *done, void *private, enum zio_flag flags);
 464
 465 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
 466     uint64_t size, zio_done_func_t *done, void *private,
 467     zio_priority_t priority, enum zio_flag flags, const zbookmark_t *zb);
 468
 469 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 470     void *data, uint64_t size, const zio_prop_t *zp,
 471     zio_done_func_t *ready, zio_done_func_t *physdone, zio_done_func_t *done,
 472     void *private,
 473     zio_priority_t priority, enum zio_flag flags, const zbookmark_t *zb);
 474
 475 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 476     void *data, uint64_t size, zio_done_func_t *done, void *private,
 477     zio_priority_t priority, enum zio_flag flags, zbookmark_t *zb);
 478
 479 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
 480     boolean_t nopwrite);
 481
 482 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
 483
 484 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
 485     const blkptr_t *bp,
 486     zio_done_func_t *done, void *private, enum zio_flag flags);
 487
 488 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
 489     zio_done_func_t *done, void *private, enum zio_flag flags);
 490
 491 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 492     uint64_t size, void *data, int checksum,
 493     zio_done_func_t *done, void *private, zio_priority_t priority,
 494     enum zio_flag flags, boolean_t labels);
 495
 496 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 497     uint64_t size, void *data, int checksum,
 498     zio_done_func_t *done, void *private, zio_priority_t priority,
 499     enum zio_flag flags, boolean_t labels);
 500
 501 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
 502     const blkptr_t *bp, enum zio_flag flags);
 503
 504 extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
 505     uint64_t size, boolean_t use_slog);
 506 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
 507 extern void zio_flush(zio_t *zio, vdev_t *vd);
 508 extern void zio_shrink(zio_t *zio, uint64_t size);
 509
 510 extern int zio_wait(zio_t *zio);
 511 extern void zio_nowait(zio_t *zio);
 512 extern void zio_execute(zio_t *zio);
 513 extern void zio_interrupt(zio_t *zio);
 514
 515 extern zio_t *zio_walk_parents(zio_t *cio);
 516 extern zio_t *zio_walk_children(zio_t *pio);
 517 extern zio_t *zio_unique_parent(zio_t *cio);
 518 extern void zio_add_child(zio_t *pio, zio_t *cio);
 519
 520 extern void *zio_buf_alloc(size_t size);
 521 extern void zio_buf_free(void *buf, size_t size);
 522 extern void *zio_data_buf_alloc(size_t size);
 523 extern void zio_data_buf_free(void *buf, size_t size);
 524 extern void *zio_vdev_alloc(void);
 525 extern void zio_vdev_free(void *buf);
 526
 527 extern void zio_resubmit_stage_async(void *);
 528
 529 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
 530     uint64_t offset, void *data, uint64_t size, int type,
 531     zio_priority_t priority, enum zio_flag flags,
 532     zio_done_func_t *done, void *private);
 533
 534 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
 535     void *data, uint64_t size, int type, zio_priority_t priority,
 536     enum zio_flag flags, zio_done_func_t *done, void *private);
 537
 538 extern void zio_vdev_io_bypass(zio_t *zio);
 539 extern void zio_vdev_io_reissue(zio_t *zio);
 540 extern void zio_vdev_io_redone(zio_t *zio);
 541
 542 extern void zio_checksum_verified(zio_t *zio);
 543 extern int zio_worst_error(int e1, int e2);
 544
 545 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
 546     enum zio_checksum parent);
 547 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
 548     enum zio_checksum child, enum zio_checksum parent);
 549 extern enum zio_compress zio_compress_select(enum zio_compress child,
 550     enum zio_compress parent);
 551
 552 extern void zio_suspend(spa_t *spa, zio_t *zio);
 553 extern int zio_resume(spa_t *spa);
 554 extern void zio_resume_wait(spa_t *spa);
 555
 556 /*
 557  * Initial setup and teardown.
 558  */
 559 extern void zio_init(void);
 560 extern void zio_fini(void);
 561
 562 /*
 563  * Fault injection
 564  */
 565 struct zinject_record;
 566 extern uint32_t zio_injection_enabled;
 567 extern int zio_inject_fault(char *name, int flags, int *id,
 568     struct zinject_record *record);
 569 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
 570     struct zinject_record *record);
 571 extern int zio_clear_fault(int id);
 572 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
 573 extern int zio_handle_fault_injection(zio_t *zio, int error);
 574 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
 575 extern int zio_handle_label_injection(zio_t *zio, int error);
 576 extern void zio_handle_ignored_writes(zio_t *zio);
 577 extern uint64_t zio_handle_io_delay(zio_t *zio);
 578
 579 /*
 580  * Checksum ereport functions
 581  */
 582 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
 583     uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
 584 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
 585     const void *good_data, const void *bad_data, boolean_t drop_if_identical);
 586
 587 extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
 588 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
 589
 590 /* If we have the good data in hand, this function can be used */
 591 extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
 592     struct zio *zio, uint64_t offset, uint64_t length,
 593     const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
 594
 595 /* Called from spa_sync(), but primarily an injection handler */
 596 extern void spa_handle_ignored_writes(spa_t *spa);
 597
 598 /* zbookmark functions */
 599 boolean_t zbookmark_is_before(const struct dnode_phys *dnp,
 600     const zbookmark_t *zb1, const zbookmark_t *zb2);
 601
 602 #ifdef  __cplusplus
 603 }
 604 #endif
 605
 606 #endif  /* _ZIO_H */