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) 2012, 2017 by Delphix. All rights reserved.
  26  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  27  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
  28  * Copyright 2016 Toomas Soome <tsoome@me.com>
  29  */
  30 
  31 #ifndef _ZIO_H
  32 #define _ZIO_H
  33 
  34 #include <sys/zio_priority.h>
  35 #include <sys/zfs_context.h>
  36 #include <sys/spa.h>
  37 #include <sys/txg.h>
  38 #include <sys/avl.h>
  39 #include <sys/fs/zfs.h>
  40 #include <sys/zio_impl.h>
  41 
  42 #ifdef  __cplusplus
  43 extern "C" {
  44 #endif
  45 
  46 /*
  47  * Embedded checksum
  48  */
  49 #define ZEC_MAGIC       0x210da7ab10c7a11ULL
  50 
  51 typedef struct zio_eck {
  52         uint64_t        zec_magic;      /* for validation, endianness   */
  53         zio_cksum_t     zec_cksum;      /* 256-bit checksum             */
  54 } zio_eck_t;
  55 
  56 /*
  57  * Gang block headers are self-checksumming and contain an array
  58  * of block pointers.
  59  */
  60 #define SPA_GANGBLOCKSIZE       SPA_MINBLOCKSIZE
  61 #define SPA_GBH_NBLKPTRS        ((SPA_GANGBLOCKSIZE - \
  62         sizeof (zio_eck_t)) / sizeof (blkptr_t))
  63 #define SPA_GBH_FILLER          ((SPA_GANGBLOCKSIZE - \
  64         sizeof (zio_eck_t) - \
  65         (SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\
  66         sizeof (uint64_t))
  67 
  68 typedef struct zio_gbh {
  69         blkptr_t                zg_blkptr[SPA_GBH_NBLKPTRS];
  70         uint64_t                zg_filler[SPA_GBH_FILLER];
  71         zio_eck_t               zg_tail;
  72 } zio_gbh_phys_t;
  73 
  74 enum zio_checksum {
  75         ZIO_CHECKSUM_INHERIT = 0,
  76         ZIO_CHECKSUM_ON,
  77         ZIO_CHECKSUM_OFF,
  78         ZIO_CHECKSUM_LABEL,
  79         ZIO_CHECKSUM_GANG_HEADER,
  80         ZIO_CHECKSUM_ZILOG,
  81         ZIO_CHECKSUM_FLETCHER_2,
  82         ZIO_CHECKSUM_FLETCHER_4,
  83         ZIO_CHECKSUM_SHA256,
  84         ZIO_CHECKSUM_ZILOG2,
  85         ZIO_CHECKSUM_NOPARITY,
  86         ZIO_CHECKSUM_SHA512,
  87         ZIO_CHECKSUM_SKEIN,
  88         ZIO_CHECKSUM_EDONR,
  89         ZIO_CHECKSUM_FUNCTIONS
  90 };
  91 
  92 /*
  93  * The number of "legacy" compression functions which can be set on individual
  94  * objects.
  95  */
  96 #define ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2
  97 
  98 #define ZIO_CHECKSUM_ON_VALUE   ZIO_CHECKSUM_FLETCHER_4
  99 #define ZIO_CHECKSUM_DEFAULT    ZIO_CHECKSUM_ON
 100 
 101 #define ZIO_CHECKSUM_MASK       0xffULL
 102 #define ZIO_CHECKSUM_VERIFY     (1 << 8)
 103 
 104 #define ZIO_DEDUPCHECKSUM       ZIO_CHECKSUM_SHA256
 105 #define ZIO_DEDUPDITTO_MIN      100
 106 
 107 /*
 108  * The number of "legacy" compression functions which can be set on individual
 109  * objects.
 110  */
 111 #define ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
 112 
 113 /*
 114  * The meaning of "compress = on" selected by the compression features enabled
 115  * on a given pool.
 116  */
 117 #define ZIO_COMPRESS_LEGACY_ON_VALUE    ZIO_COMPRESS_LZJB
 118 #define ZIO_COMPRESS_LZ4_ON_VALUE       ZIO_COMPRESS_LZ4
 119 
 120 #define ZIO_COMPRESS_DEFAULT            ZIO_COMPRESS_OFF
 121 
 122 #define BOOTFS_COMPRESS_VALID(compress)                 \
 123         ((compress) == ZIO_COMPRESS_LZJB ||             \
 124         (compress) == ZIO_COMPRESS_LZ4 ||               \
 125         (compress) == ZIO_COMPRESS_GZIP_1 ||            \
 126         (compress) == ZIO_COMPRESS_GZIP_2 ||            \
 127         (compress) == ZIO_COMPRESS_GZIP_3 ||            \
 128         (compress) == ZIO_COMPRESS_GZIP_4 ||            \
 129         (compress) == ZIO_COMPRESS_GZIP_5 ||            \
 130         (compress) == ZIO_COMPRESS_GZIP_6 ||            \
 131         (compress) == ZIO_COMPRESS_GZIP_7 ||            \
 132         (compress) == ZIO_COMPRESS_GZIP_8 ||            \
 133         (compress) == ZIO_COMPRESS_GZIP_9 ||            \
 134         (compress) == ZIO_COMPRESS_ON ||                \
 135         (compress) == ZIO_COMPRESS_OFF)
 136 
 137 #define ZIO_FAILURE_MODE_WAIT           0
 138 #define ZIO_FAILURE_MODE_CONTINUE       1
 139 #define ZIO_FAILURE_MODE_PANIC          2
 140 
 141 enum zio_flag {
 142         /*
 143          * Flags inherited by gang, ddt, and vdev children,
 144          * and that must be equal for two zios to aggregate
 145          */
 146         ZIO_FLAG_DONT_AGGREGATE = 1 << 0,
 147         ZIO_FLAG_IO_REPAIR      = 1 << 1,
 148         ZIO_FLAG_SELF_HEAL      = 1 << 2,
 149         ZIO_FLAG_RESILVER       = 1 << 3,
 150         ZIO_FLAG_SCRUB          = 1 << 4,
 151         ZIO_FLAG_SCAN_THREAD    = 1 << 5,
 152         ZIO_FLAG_PHYSICAL       = 1 << 6,
 153 
 154 #define ZIO_FLAG_AGG_INHERIT    (ZIO_FLAG_CANFAIL - 1)
 155 
 156         /*
 157          * Flags inherited by ddt, gang, and vdev children.
 158          */
 159         ZIO_FLAG_CANFAIL        = 1 << 7, /* must be first for INHERIT */
 160         ZIO_FLAG_SPECULATIVE    = 1 << 8,
 161         ZIO_FLAG_CONFIG_WRITER  = 1 << 9,
 162         ZIO_FLAG_DONT_RETRY     = 1 << 10,
 163         ZIO_FLAG_DONT_CACHE     = 1 << 11,
 164         ZIO_FLAG_NODATA         = 1 << 12,
 165         ZIO_FLAG_INDUCE_DAMAGE  = 1 << 13,
 166         ZIO_FLAG_IO_ALLOCATING  = 1 << 14,
 167 
 168 #define ZIO_FLAG_DDT_INHERIT    (ZIO_FLAG_IO_RETRY - 1)
 169 #define ZIO_FLAG_GANG_INHERIT   (ZIO_FLAG_IO_RETRY - 1)
 170 
 171         /*
 172          * Flags inherited by vdev children.
 173          */
 174         ZIO_FLAG_IO_RETRY       = 1 << 15,        /* must be first for INHERIT */
 175         ZIO_FLAG_PROBE          = 1 << 16,
 176         ZIO_FLAG_TRYHARD        = 1 << 17,
 177         ZIO_FLAG_OPTIONAL       = 1 << 18,
 178 
 179 #define ZIO_FLAG_VDEV_INHERIT   (ZIO_FLAG_DONT_QUEUE - 1)
 180 
 181         /*
 182          * Flags not inherited by any children.
 183          */
 184         ZIO_FLAG_DONT_QUEUE     = 1 << 19,        /* must be first for INHERIT */
 185         ZIO_FLAG_DONT_PROPAGATE = 1 << 20,
 186         ZIO_FLAG_IO_BYPASS      = 1 << 21,
 187         ZIO_FLAG_IO_REWRITE     = 1 << 22,
 188         ZIO_FLAG_RAW            = 1 << 23,
 189         ZIO_FLAG_GANG_CHILD     = 1 << 24,
 190         ZIO_FLAG_DDT_CHILD      = 1 << 25,
 191         ZIO_FLAG_GODFATHER      = 1 << 26,
 192         ZIO_FLAG_NOPWRITE       = 1 << 27,
 193         ZIO_FLAG_REEXECUTED     = 1 << 28,
 194         ZIO_FLAG_DELEGATED      = 1 << 29,
 195 };
 196 
 197 #define ZIO_FLAG_MUSTSUCCEED            0
 198 
 199 #define ZIO_DDT_CHILD_FLAGS(zio)                                \
 200         (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |              \
 201         ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
 202 
 203 #define ZIO_GANG_CHILD_FLAGS(zio)                               \
 204         (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |             \
 205         ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
 206 
 207 #define ZIO_VDEV_CHILD_FLAGS(zio)                               \
 208         (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |             \
 209         ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_CANFAIL)
 210 
 211 #define ZIO_CHILD_BIT(x)                (1 << (x))
 212 #define ZIO_CHILD_BIT_IS_SET(val, x)    ((val) & (1 << (x)))
 213 
 214 enum zio_child {
 215         ZIO_CHILD_VDEV = 0,
 216         ZIO_CHILD_GANG,
 217         ZIO_CHILD_DDT,
 218         ZIO_CHILD_LOGICAL,
 219         ZIO_CHILD_TYPES
 220 };
 221 
 222 #define ZIO_CHILD_VDEV_BIT              ZIO_CHILD_BIT(ZIO_CHILD_VDEV)
 223 #define ZIO_CHILD_GANG_BIT              ZIO_CHILD_BIT(ZIO_CHILD_GANG)
 224 #define ZIO_CHILD_DDT_BIT               ZIO_CHILD_BIT(ZIO_CHILD_DDT)
 225 #define ZIO_CHILD_LOGICAL_BIT           ZIO_CHILD_BIT(ZIO_CHILD_LOGICAL)
 226 #define ZIO_CHILD_ALL_BITS                                      \
 227         (ZIO_CHILD_VDEV_BIT | ZIO_CHILD_GANG_BIT |              \
 228         ZIO_CHILD_DDT_BIT | ZIO_CHILD_LOGICAL_BIT)
 229 
 230 enum zio_wait_type {
 231         ZIO_WAIT_READY = 0,
 232         ZIO_WAIT_DONE,
 233         ZIO_WAIT_TYPES
 234 };
 235 
 236 /*
 237  * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent
 238  * graveyard) to indicate checksum errors and fragmentation.
 239  */
 240 #define ECKSUM  EBADE
 241 #define EFRAGS  EBADR
 242 
 243 typedef void zio_done_func_t(zio_t *zio);
 244 
 245 extern boolean_t zio_dva_throttle_enabled;
 246 extern const char *zio_type_name[ZIO_TYPES];
 247 
 248 /*
 249  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
 250  * identifies any block in the pool.  By convention, the meta-objset (MOS)
 251  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
 252  * except root blocks and ZIL blocks, which are defined as follows:
 253  *
 254  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
 255  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
 256  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
 257  * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
 258  *
 259  * Note: this structure is called a bookmark because its original purpose
 260  * was to remember where to resume a pool-wide traverse.
 261  *
 262  * Note: this structure is passed between userland and the kernel, and is
 263  * stored on disk (by virtue of being incorporated into other on-disk
 264  * structures, e.g. dsl_scan_phys_t).
 265  */
 266 typedef struct zbookmark_phys {
 267         uint64_t        zb_objset;
 268         uint64_t        zb_object;
 269         int64_t         zb_level;
 270         uint64_t        zb_blkid;
 271 } zbookmark_phys_t;
 272 
 273 #define SET_BOOKMARK(zb, objset, object, level, blkid)  \
 274 {                                                       \
 275         (zb)->zb_objset = objset;                       \
 276         (zb)->zb_object = object;                       \
 277         (zb)->zb_level = level;                         \
 278         (zb)->zb_blkid = blkid;                         \
 279 }
 280 
 281 #define ZB_DESTROYED_OBJSET     (-1ULL)
 282 
 283 #define ZB_ROOT_OBJECT          (0ULL)
 284 #define ZB_ROOT_LEVEL           (-1LL)
 285 #define ZB_ROOT_BLKID           (0ULL)
 286 
 287 #define ZB_ZIL_OBJECT           (0ULL)
 288 #define ZB_ZIL_LEVEL            (-2LL)
 289 
 290 #define ZB_DNODE_LEVEL          (-3LL)
 291 #define ZB_DNODE_BLKID          (0ULL)
 292 
 293 #define ZB_IS_ZERO(zb)                                          \
 294         ((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&  \
 295         (zb)->zb_level == 0 && (zb)->zb_blkid == 0)
 296 #define ZB_IS_ROOT(zb)                          \
 297         ((zb)->zb_object == ZB_ROOT_OBJECT &&        \
 298         (zb)->zb_level == ZB_ROOT_LEVEL &&   \
 299         (zb)->zb_blkid == ZB_ROOT_BLKID)
 300 
 301 typedef struct zio_prop {
 302         enum zio_checksum       zp_checksum;
 303         enum zio_compress       zp_compress;
 304         dmu_object_type_t       zp_type;
 305         uint8_t                 zp_level;
 306         uint8_t                 zp_copies;
 307         boolean_t               zp_dedup;
 308         boolean_t               zp_dedup_verify;
 309         boolean_t               zp_nopwrite;
 310 } zio_prop_t;
 311 
 312 typedef struct zio_cksum_report zio_cksum_report_t;
 313 
 314 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
 315     const void *good_data);
 316 typedef void zio_cksum_free_f(void *cbdata, size_t size);
 317 
 318 struct zio_bad_cksum;                           /* defined in zio_checksum.h */
 319 struct dnode_phys;
 320 struct abd;
 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, struct abd *data, uint64_t offset);
 354 
 355 typedef void zio_transform_func_t(zio_t *zio, struct abd *data, uint64_t size);
 356 
 357 typedef struct zio_transform {
 358         struct abd              *zt_orig_abd;
 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_alloc_list {
 378         list_t  zal_list;
 379         uint64_t zal_size;
 380 } zio_alloc_list_t;
 381 
 382 typedef struct zio_link {
 383         zio_t           *zl_parent;
 384         zio_t           *zl_child;
 385         list_node_t     zl_parent_node;
 386         list_node_t     zl_child_node;
 387 } zio_link_t;
 388 
 389 struct zio {
 390         /* Core information about this I/O */
 391         zbookmark_phys_t        io_bookmark;
 392         zio_prop_t      io_prop;
 393         zio_type_t      io_type;
 394         enum zio_child  io_child_type;
 395         int             io_cmd;
 396         zio_priority_t  io_priority;
 397         uint8_t         io_reexecute;
 398         uint8_t         io_state[ZIO_WAIT_TYPES];
 399         uint64_t        io_txg;
 400         spa_t           *io_spa;
 401         blkptr_t        *io_bp;
 402         blkptr_t        *io_bp_override;
 403         blkptr_t        io_bp_copy;
 404         list_t          io_parent_list;
 405         list_t          io_child_list;
 406         zio_t           *io_logical;
 407         zio_transform_t *io_transform_stack;
 408 
 409         /* Callback info */
 410         zio_done_func_t *io_ready;
 411         zio_done_func_t *io_children_ready;
 412         zio_done_func_t *io_physdone;
 413         zio_done_func_t *io_done;
 414         void            *io_private;
 415         int64_t         io_prev_space_delta;    /* DMU private */
 416         blkptr_t        io_bp_orig;
 417 
 418         /* Data represented by this I/O */
 419         struct abd      *io_abd;
 420         struct abd      *io_orig_abd;
 421         uint64_t        io_size;
 422         uint64_t        io_orig_size;
 423         /* io_lsize != io_orig_size iff this is a raw write */
 424         uint64_t        io_lsize;
 425 
 426         /* Stuff for the vdev stack */
 427         vdev_t          *io_vd;
 428         void            *io_vsd;
 429         const zio_vsd_ops_t *io_vsd_ops;
 430 
 431         uint64_t        io_offset;
 432         hrtime_t        io_timestamp;
 433         hrtime_t        io_queued_timestamp;
 434         hrtime_t        io_target_timestamp;
 435         avl_node_t      io_queue_node;
 436         avl_node_t      io_offset_node;
 437         avl_node_t      io_alloc_node;
 438         zio_alloc_list_t        io_alloc_list;
 439 
 440         /* Internal pipeline state */
 441         enum zio_flag   io_flags;
 442         enum zio_stage  io_stage;
 443         enum zio_stage  io_pipeline;
 444         enum zio_flag   io_orig_flags;
 445         enum zio_stage  io_orig_stage;
 446         enum zio_stage  io_orig_pipeline;
 447         enum zio_stage  io_pipeline_trace;
 448         int             io_error;
 449         int             io_child_error[ZIO_CHILD_TYPES];
 450         uint64_t        io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
 451         uint64_t        io_child_count;
 452         uint64_t        io_phys_children;
 453         uint64_t        io_parent_count;
 454         uint64_t        *io_stall;
 455         zio_t           *io_gang_leader;
 456         zio_gang_node_t *io_gang_tree;
 457         void            *io_executor;
 458         void            *io_waiter;
 459         kmutex_t        io_lock;
 460         kcondvar_t      io_cv;
 461 
 462         /* FMA state */
 463         zio_cksum_report_t *io_cksum_report;
 464         uint64_t        io_ena;
 465 
 466         /* Taskq dispatching state */
 467         taskq_ent_t     io_tqent;
 468 };
 469 
 470 extern int zio_bookmark_compare(const void *, const void *);
 471 
 472 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
 473     zio_done_func_t *done, void *private, enum zio_flag flags);
 474 
 475 extern zio_t *zio_root(spa_t *spa,
 476     zio_done_func_t *done, void *private, enum zio_flag flags);
 477 
 478 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
 479     struct abd *data, uint64_t lsize, zio_done_func_t *done, void *private,
 480     zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
 481 
 482 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 483     struct abd *data, uint64_t size, uint64_t psize, const zio_prop_t *zp,
 484     zio_done_func_t *ready, zio_done_func_t *children_ready,
 485     zio_done_func_t *physdone, zio_done_func_t *done,
 486     void *private, zio_priority_t priority, enum zio_flag flags,
 487     const zbookmark_phys_t *zb);
 488 
 489 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 490     struct abd *data, uint64_t size, zio_done_func_t *done, void *private,
 491     zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
 492 
 493 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
 494     boolean_t nopwrite);
 495 
 496 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
 497 
 498 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
 499     const blkptr_t *bp,
 500     zio_done_func_t *done, void *private, enum zio_flag flags);
 501 
 502 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
 503     zio_done_func_t *done, void *private, enum zio_flag flags);
 504 
 505 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 506     uint64_t size, struct abd *data, int checksum,
 507     zio_done_func_t *done, void *private, zio_priority_t priority,
 508     enum zio_flag flags, boolean_t labels);
 509 
 510 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 511     uint64_t size, struct abd *data, int checksum,
 512     zio_done_func_t *done, void *private, zio_priority_t priority,
 513     enum zio_flag flags, boolean_t labels);
 514 
 515 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
 516     const blkptr_t *bp, enum zio_flag flags);
 517 
 518 extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
 519     blkptr_t *old_bp, uint64_t size, boolean_t *slog);
 520 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
 521 extern void zio_flush(zio_t *zio, vdev_t *vd);
 522 extern void zio_shrink(zio_t *zio, uint64_t size);
 523 
 524 extern int zio_wait(zio_t *zio);
 525 extern void zio_nowait(zio_t *zio);
 526 extern void zio_execute(zio_t *zio);
 527 extern void zio_interrupt(zio_t *zio);
 528 extern void zio_delay_init(zio_t *zio);
 529 extern void zio_delay_interrupt(zio_t *zio);
 530 
 531 extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **);
 532 extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **);
 533 extern zio_t *zio_unique_parent(zio_t *cio);
 534 extern void zio_add_child(zio_t *pio, zio_t *cio);
 535 
 536 extern void *zio_buf_alloc(size_t size);
 537 extern void zio_buf_free(void *buf, size_t size);
 538 extern void *zio_data_buf_alloc(size_t size);
 539 extern void zio_data_buf_free(void *buf, size_t size);
 540 
 541 extern void zio_push_transform(zio_t *zio, struct abd *abd, uint64_t size,
 542     uint64_t bufsize, zio_transform_func_t *transform);
 543 extern void zio_pop_transforms(zio_t *zio);
 544 
 545 extern void zio_resubmit_stage_async(void *);
 546 
 547 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
 548     uint64_t offset, struct abd *data, uint64_t size, int type,
 549     zio_priority_t priority, enum zio_flag flags,
 550     zio_done_func_t *done, void *private);
 551 
 552 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
 553     struct abd *data, uint64_t size, int type, zio_priority_t priority,
 554     enum zio_flag flags, zio_done_func_t *done, void *private);
 555 
 556 extern void zio_vdev_io_bypass(zio_t *zio);
 557 extern void zio_vdev_io_reissue(zio_t *zio);
 558 extern void zio_vdev_io_redone(zio_t *zio);
 559 
 560 extern void zio_checksum_verified(zio_t *zio);
 561 extern int zio_worst_error(int e1, int e2);
 562 
 563 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
 564     enum zio_checksum parent);
 565 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
 566     enum zio_checksum child, enum zio_checksum parent);
 567 extern enum zio_compress zio_compress_select(spa_t *spa,
 568     enum zio_compress child, enum zio_compress parent);
 569 
 570 extern void zio_suspend(spa_t *spa, zio_t *zio);
 571 extern int zio_resume(spa_t *spa);
 572 extern void zio_resume_wait(spa_t *spa);
 573 
 574 /*
 575  * Initial setup and teardown.
 576  */
 577 extern void zio_init(void);
 578 extern void zio_fini(void);
 579 
 580 /*
 581  * Fault injection
 582  */
 583 struct zinject_record;
 584 extern uint32_t zio_injection_enabled;
 585 extern int zio_inject_fault(char *name, int flags, int *id,
 586     struct zinject_record *record);
 587 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
 588     struct zinject_record *record);
 589 extern int zio_clear_fault(int id);
 590 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
 591 extern int zio_handle_fault_injection(zio_t *zio, int error);
 592 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
 593 extern int zio_handle_label_injection(zio_t *zio, int error);
 594 extern void zio_handle_ignored_writes(zio_t *zio);
 595 extern hrtime_t zio_handle_io_delay(zio_t *zio);
 596 
 597 /*
 598  * Checksum ereport functions
 599  */
 600 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
 601     uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
 602 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
 603     const void *good_data, const void *bad_data, boolean_t drop_if_identical);
 604 
 605 extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
 606 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
 607 
 608 /* If we have the good data in hand, this function can be used */
 609 extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
 610     struct zio *zio, uint64_t offset, uint64_t length,
 611     const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
 612 
 613 /* Called from spa_sync(), but primarily an injection handler */
 614 extern void spa_handle_ignored_writes(spa_t *spa);
 615 
 616 /* zbookmark_phys functions */
 617 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
 618     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
 619 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
 620     uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
 621 
 622 #ifdef  __cplusplus
 623 }
 624 #endif
 625 
 626 #endif  /* _ZIO_H */