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) 2012, 2020 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  * Copyright (c) 2019, Allan Jude
  30  * Copyright (c) 2019, Klara Inc.
  31  * Copyright (c) 2019-2020, Michael Niewöhner
  32  */
  33
  34 #ifndef _ZIO_H
  35 #define _ZIO_H
  36
  37 #include <sys/zio_priority.h>
  38 #include <sys/zfs_context.h>
  39 #include <sys/spa.h>
  40 #include <sys/txg.h>
  41 #include <sys/avl.h>
  42 #include <sys/fs/zfs.h>
  43 #include <sys/zio_impl.h>
  44
  45 #ifdef  __cplusplus
  46 extern "C" {
  47 #endif
  48
  49 /*
  50  * Embedded checksum
  51  */
  52 #define ZEC_MAGIC       0x210da7ab10c7a11ULL
  53
  54 typedef struct zio_eck {
  55         uint64_t        zec_magic;      /* for validation, endianness   */
  56         zio_cksum_t     zec_cksum;      /* 256-bit checksum             */
  57 } zio_eck_t;
  58
  59 /*
  60  * Gang block headers are self-checksumming and contain an array
  61  * of block pointers.
  62  */
  63 #define SPA_GANGBLOCKSIZE       SPA_MINBLOCKSIZE
  64 #define SPA_GBH_NBLKPTRS        ((SPA_GANGBLOCKSIZE - \
  65         sizeof (zio_eck_t)) / sizeof (blkptr_t))
  66 #define SPA_GBH_FILLER          ((SPA_GANGBLOCKSIZE - \
  67         sizeof (zio_eck_t) - \
  68         (SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\
  69         sizeof (uint64_t))
  70
  71 typedef struct zio_gbh {
  72         blkptr_t                zg_blkptr[SPA_GBH_NBLKPTRS];
  73         uint64_t                zg_filler[SPA_GBH_FILLER];
  74         zio_eck_t               zg_tail;
  75 } zio_gbh_phys_t;
  76
  77 enum zio_checksum {
  78         ZIO_CHECKSUM_INHERIT = 0,
  79         ZIO_CHECKSUM_ON,
  80         ZIO_CHECKSUM_OFF,
  81         ZIO_CHECKSUM_LABEL,
  82         ZIO_CHECKSUM_GANG_HEADER,
  83         ZIO_CHECKSUM_ZILOG,
  84         ZIO_CHECKSUM_FLETCHER_2,
  85         ZIO_CHECKSUM_FLETCHER_4,
  86         ZIO_CHECKSUM_SHA256,
  87         ZIO_CHECKSUM_ZILOG2,
  88         ZIO_CHECKSUM_NOPARITY,
  89         ZIO_CHECKSUM_SHA512,
  90         ZIO_CHECKSUM_SKEIN,
  91 #if !defined(__FreeBSD__)
  92         ZIO_CHECKSUM_EDONR,
  93 #endif
  94         ZIO_CHECKSUM_FUNCTIONS
  95 };
  96
  97 /*
  98  * The number of "legacy" compression functions which can be set on individual
  99  * objects.
 100  */
 101 #define ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2
 102
 103 #define ZIO_CHECKSUM_ON_VALUE   ZIO_CHECKSUM_FLETCHER_4
 104 #define ZIO_CHECKSUM_DEFAULT    ZIO_CHECKSUM_ON
 105
 106 #define ZIO_CHECKSUM_MASK       0xffULL
 107 #define ZIO_CHECKSUM_VERIFY     (1 << 8)
 108
 109 #define ZIO_DEDUPCHECKSUM       ZIO_CHECKSUM_SHA256
 110
 111 /* supported encryption algorithms */
 112 enum zio_encrypt {
 113         ZIO_CRYPT_INHERIT = 0,
 114         ZIO_CRYPT_ON,
 115         ZIO_CRYPT_OFF,
 116         ZIO_CRYPT_AES_128_CCM,
 117         ZIO_CRYPT_AES_192_CCM,
 118         ZIO_CRYPT_AES_256_CCM,
 119         ZIO_CRYPT_AES_128_GCM,
 120         ZIO_CRYPT_AES_192_GCM,
 121         ZIO_CRYPT_AES_256_GCM,
 122         ZIO_CRYPT_FUNCTIONS
 123 };
 124
 125 #define ZIO_CRYPT_ON_VALUE      ZIO_CRYPT_AES_256_GCM
 126 #define ZIO_CRYPT_DEFAULT       ZIO_CRYPT_OFF
 127
 128 /* macros defining encryption lengths */
 129 #define ZIO_OBJSET_MAC_LEN              32
 130 #define ZIO_DATA_IV_LEN                 12
 131 #define ZIO_DATA_SALT_LEN               8
 132 #define ZIO_DATA_MAC_LEN                16
 133
 134 /*
 135  * The number of "legacy" compression functions which can be set on individual
 136  * objects.
 137  */
 138 #define ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
 139
 140 /*
 141  * The meaning of "compress = on" selected by the compression features enabled
 142  * on a given pool.
 143  */
 144 #define ZIO_COMPRESS_LEGACY_ON_VALUE    ZIO_COMPRESS_LZJB
 145 #define ZIO_COMPRESS_LZ4_ON_VALUE       ZIO_COMPRESS_LZ4
 146
 147 #define ZIO_COMPRESS_DEFAULT            ZIO_COMPRESS_OFF
 148
 149 #define BOOTFS_COMPRESS_VALID(compress)                 \
 150         ((compress) == ZIO_COMPRESS_LZJB ||             \
 151         (compress) == ZIO_COMPRESS_LZ4 ||               \
 152         (compress) == ZIO_COMPRESS_GZIP_1 ||            \
 153         (compress) == ZIO_COMPRESS_GZIP_2 ||            \
 154         (compress) == ZIO_COMPRESS_GZIP_3 ||            \
 155         (compress) == ZIO_COMPRESS_GZIP_4 ||            \
 156         (compress) == ZIO_COMPRESS_GZIP_5 ||            \
 157         (compress) == ZIO_COMPRESS_GZIP_6 ||            \
 158         (compress) == ZIO_COMPRESS_GZIP_7 ||            \
 159         (compress) == ZIO_COMPRESS_GZIP_8 ||            \
 160         (compress) == ZIO_COMPRESS_GZIP_9 ||            \
 161         (compress) == ZIO_COMPRESS_ZLE ||               \
 162         (compress) == ZIO_COMPRESS_ZSTD ||              \
 163         (compress) == ZIO_COMPRESS_ON ||                \
 164         (compress) == ZIO_COMPRESS_OFF)
 165
 166
 167 #define ZIO_COMPRESS_ALGO(x)    (x & SPA_COMPRESSMASK)
 168 #define ZIO_COMPRESS_LEVEL(x)   ((x & ~SPA_COMPRESSMASK) >> SPA_COMPRESSBITS)
 169 #define ZIO_COMPRESS_RAW(type, level)   (type | ((level) << SPA_COMPRESSBITS))
 170
 171 #define ZIO_COMPLEVEL_ZSTD(level)       \
 172         ZIO_COMPRESS_RAW(ZIO_COMPRESS_ZSTD, level)
 173
 174 #define ZIO_FAILURE_MODE_WAIT           0
 175 #define ZIO_FAILURE_MODE_CONTINUE       1
 176 #define ZIO_FAILURE_MODE_PANIC          2
 177
 178 typedef enum zio_suspend_reason {
 179         ZIO_SUSPEND_NONE = 0,
 180         ZIO_SUSPEND_IOERR,
 181         ZIO_SUSPEND_MMP,
 182 } zio_suspend_reason_t;
 183
 184 enum zio_flag {
 185         /*
 186          * Flags inherited by gang, ddt, and vdev children,
 187          * and that must be equal for two zios to aggregate
 188          */
 189         ZIO_FLAG_DONT_AGGREGATE = 1 << 0,
 190         ZIO_FLAG_IO_REPAIR      = 1 << 1,
 191         ZIO_FLAG_SELF_HEAL      = 1 << 2,
 192         ZIO_FLAG_RESILVER       = 1 << 3,
 193         ZIO_FLAG_SCRUB          = 1 << 4,
 194         ZIO_FLAG_SCAN_THREAD    = 1 << 5,
 195         ZIO_FLAG_PHYSICAL       = 1 << 6,
 196
 197 #define ZIO_FLAG_AGG_INHERIT    (ZIO_FLAG_CANFAIL - 1)
 198
 199         /*
 200          * Flags inherited by ddt, gang, and vdev children.
 201          */
 202         ZIO_FLAG_CANFAIL        = 1 << 7,       /* must be first for INHERIT */
 203         ZIO_FLAG_SPECULATIVE    = 1 << 8,
 204         ZIO_FLAG_CONFIG_WRITER  = 1 << 9,
 205         ZIO_FLAG_DONT_RETRY     = 1 << 10,
 206         ZIO_FLAG_DONT_CACHE     = 1 << 11,
 207         ZIO_FLAG_NODATA         = 1 << 12,
 208         ZIO_FLAG_INDUCE_DAMAGE  = 1 << 13,
 209         ZIO_FLAG_IO_ALLOCATING  = 1 << 14,
 210
 211 #define ZIO_FLAG_DDT_INHERIT    (ZIO_FLAG_IO_RETRY - 1)
 212 #define ZIO_FLAG_GANG_INHERIT   (ZIO_FLAG_IO_RETRY - 1)
 213
 214         /*
 215          * Flags inherited by vdev children.
 216          */
 217         ZIO_FLAG_IO_RETRY       = 1 << 15,      /* must be first for INHERIT */
 218         ZIO_FLAG_PROBE          = 1 << 16,
 219         ZIO_FLAG_TRYHARD        = 1 << 17,
 220         ZIO_FLAG_OPTIONAL       = 1 << 18,
 221
 222 #define ZIO_FLAG_VDEV_INHERIT   (ZIO_FLAG_DONT_QUEUE - 1)
 223
 224         /*
 225          * Flags not inherited by any children.
 226          */
 227         ZIO_FLAG_DONT_QUEUE     = 1 << 19,      /* must be first for INHERIT */
 228         ZIO_FLAG_DONT_PROPAGATE = 1 << 20,
 229         ZIO_FLAG_IO_BYPASS      = 1 << 21,
 230         ZIO_FLAG_IO_REWRITE     = 1 << 22,
 231         ZIO_FLAG_RAW_COMPRESS   = 1 << 23,
 232         ZIO_FLAG_RAW_ENCRYPT    = 1 << 24,
 233         ZIO_FLAG_GANG_CHILD     = 1 << 25,
 234         ZIO_FLAG_DDT_CHILD      = 1 << 26,
 235         ZIO_FLAG_GODFATHER      = 1 << 27,
 236         ZIO_FLAG_NOPWRITE       = 1 << 28,
 237         ZIO_FLAG_REEXECUTED     = 1 << 29,
 238         ZIO_FLAG_DELEGATED      = 1 << 30,
 239         ZIO_FLAG_FASTWRITE      = 1 << 31,
 240 };
 241
 242 #define ZIO_FLAG_MUSTSUCCEED            0
 243 #define ZIO_FLAG_RAW    (ZIO_FLAG_RAW_COMPRESS | ZIO_FLAG_RAW_ENCRYPT)
 244
 245 #define ZIO_DDT_CHILD_FLAGS(zio)                                \
 246         (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |             \
 247         ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
 248
 249 #define ZIO_GANG_CHILD_FLAGS(zio)                               \
 250         (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |            \
 251         ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
 252
 253 #define ZIO_VDEV_CHILD_FLAGS(zio)                               \
 254         (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |            \
 255         ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_CANFAIL)
 256
 257 #define ZIO_CHILD_BIT(x)                (1 << (x))
 258 #define ZIO_CHILD_BIT_IS_SET(val, x)    ((val) & (1 << (x)))
 259
 260 enum zio_child {
 261         ZIO_CHILD_VDEV = 0,
 262         ZIO_CHILD_GANG,
 263         ZIO_CHILD_DDT,
 264         ZIO_CHILD_LOGICAL,
 265         ZIO_CHILD_TYPES
 266 };
 267
 268 #define ZIO_CHILD_VDEV_BIT              ZIO_CHILD_BIT(ZIO_CHILD_VDEV)
 269 #define ZIO_CHILD_GANG_BIT              ZIO_CHILD_BIT(ZIO_CHILD_GANG)
 270 #define ZIO_CHILD_DDT_BIT               ZIO_CHILD_BIT(ZIO_CHILD_DDT)
 271 #define ZIO_CHILD_LOGICAL_BIT           ZIO_CHILD_BIT(ZIO_CHILD_LOGICAL)
 272 #define ZIO_CHILD_ALL_BITS                                      \
 273         (ZIO_CHILD_VDEV_BIT | ZIO_CHILD_GANG_BIT |              \
 274         ZIO_CHILD_DDT_BIT | ZIO_CHILD_LOGICAL_BIT)
 275
 276 enum zio_wait_type {
 277         ZIO_WAIT_READY = 0,
 278         ZIO_WAIT_DONE,
 279         ZIO_WAIT_TYPES
 280 };
 281
 282 typedef void zio_done_func_t(zio_t *zio);
 283
 284 extern int zio_exclude_metadata;
 285 extern int zio_dva_throttle_enabled;
 286 extern const char *zio_type_name[ZIO_TYPES];
 287
 288 /*
 289  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
 290  * identifies any block in the pool.  By convention, the meta-objset (MOS)
 291  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
 292  * except root blocks and ZIL blocks, which are defined as follows:
 293  *
 294  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
 295  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
 296  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
 297  * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
 298  *
 299  * Note: this structure is called a bookmark because its original purpose
 300  * was to remember where to resume a pool-wide traverse.
 301  *
 302  * Note: this structure is passed between userland and the kernel, and is
 303  * stored on disk (by virtue of being incorporated into other on-disk
 304  * structures, e.g. dsl_scan_phys_t).
 305  */
 306 struct zbookmark_phys {
 307         uint64_t        zb_objset;
 308         uint64_t        zb_object;
 309         int64_t         zb_level;
 310         uint64_t        zb_blkid;
 311 };
 312
 313 #define SET_BOOKMARK(zb, objset, object, level, blkid)  \
 314 {                                                       \
 315         (zb)->zb_objset = objset;                       \
 316         (zb)->zb_object = object;                       \
 317         (zb)->zb_level = level;                         \
 318         (zb)->zb_blkid = blkid;                         \
 319 }
 320
 321 #define ZB_DESTROYED_OBJSET     (-1ULL)
 322
 323 #define ZB_ROOT_OBJECT          (0ULL)
 324 #define ZB_ROOT_LEVEL           (-1LL)
 325 #define ZB_ROOT_BLKID           (0ULL)
 326
 327 #define ZB_ZIL_OBJECT           (0ULL)
 328 #define ZB_ZIL_LEVEL            (-2LL)
 329
 330 #define ZB_DNODE_LEVEL          (-3LL)
 331 #define ZB_DNODE_BLKID          (0ULL)
 332
 333 #define ZB_IS_ZERO(zb)                                          \
 334         ((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&        \
 335         (zb)->zb_level == 0 && (zb)->zb_blkid == 0)
 336 #define ZB_IS_ROOT(zb)                          \
 337         ((zb)->zb_object == ZB_ROOT_OBJECT &&   \
 338         (zb)->zb_level == ZB_ROOT_LEVEL &&      \
 339         (zb)->zb_blkid == ZB_ROOT_BLKID)
 340
 341 typedef struct zio_prop {
 342         enum zio_checksum       zp_checksum;
 343         enum zio_compress       zp_compress;
 344         uint8_t                 zp_complevel;
 345         dmu_object_type_t       zp_type;
 346         uint8_t                 zp_level;
 347         uint8_t                 zp_copies;
 348         boolean_t               zp_dedup;
 349         boolean_t               zp_dedup_verify;
 350         boolean_t               zp_nopwrite;
 351         boolean_t               zp_encrypt;
 352         boolean_t               zp_byteorder;
 353         uint8_t                 zp_salt[ZIO_DATA_SALT_LEN];
 354         uint8_t                 zp_iv[ZIO_DATA_IV_LEN];
 355         uint8_t                 zp_mac[ZIO_DATA_MAC_LEN];
 356         uint32_t                zp_zpl_smallblk;
 357 } zio_prop_t;
 358
 359 typedef struct zio_cksum_report zio_cksum_report_t;
 360
 361 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
 362     const abd_t *good_data);
 363 typedef void zio_cksum_free_f(void *cbdata, size_t size);
 364
 365 struct zio_bad_cksum;                           /* defined in zio_checksum.h */
 366 struct dnode_phys;
 367 struct abd;
 368
 369 struct zio_cksum_report {
 370         struct zio_cksum_report *zcr_next;
 371         nvlist_t                *zcr_ereport;
 372         nvlist_t                *zcr_detector;
 373         void                    *zcr_cbdata;
 374         size_t                  zcr_cbinfo;     /* passed to zcr_free() */
 375         uint64_t                zcr_sector;
 376         uint64_t                zcr_align;
 377         uint64_t                zcr_length;
 378         zio_cksum_finish_f      *zcr_finish;
 379         zio_cksum_free_f        *zcr_free;
 380
 381         /* internal use only */
 382         struct zio_bad_cksum    *zcr_ckinfo;    /* information from failure */
 383 };
 384
 385 typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
 386     void *arg);
 387
 388 zio_vsd_cksum_report_f  zio_vsd_default_cksum_report;
 389
 390 typedef struct zio_vsd_ops {
 391         zio_done_func_t         *vsd_free;
 392         zio_vsd_cksum_report_f  *vsd_cksum_report;
 393 } zio_vsd_ops_t;
 394
 395 typedef struct zio_gang_node {
 396         zio_gbh_phys_t          *gn_gbh;
 397         struct zio_gang_node    *gn_child[SPA_GBH_NBLKPTRS];
 398 } zio_gang_node_t;
 399
 400 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
 401     zio_gang_node_t *gn, struct abd *data, uint64_t offset);
 402
 403 typedef void zio_transform_func_t(zio_t *zio, struct abd *data, uint64_t size);
 404
 405 typedef struct zio_transform {
 406         struct abd              *zt_orig_abd;
 407         uint64_t                zt_orig_size;
 408         uint64_t                zt_bufsize;
 409         zio_transform_func_t    *zt_transform;
 410         struct zio_transform    *zt_next;
 411 } zio_transform_t;
 412
 413 typedef zio_t *zio_pipe_stage_t(zio_t *zio);
 414
 415 /*
 416  * The io_reexecute flags are distinct from io_flags because the child must
 417  * be able to propagate them to the parent.  The normal io_flags are local
 418  * to the zio, not protected by any lock, and not modifiable by children;
 419  * the reexecute flags are protected by io_lock, modifiable by children,
 420  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
 421  */
 422 #define ZIO_REEXECUTE_NOW       0x01
 423 #define ZIO_REEXECUTE_SUSPEND   0x02
 424
 425 /*
 426  * The io_trim flags are used to specify the type of TRIM to perform.  They
 427  * only apply to ZIO_TYPE_TRIM zios are distinct from io_flags.
 428  */
 429 enum trim_flag {
 430         ZIO_TRIM_SECURE         = 1 << 0,
 431 };
 432
 433 typedef struct zio_alloc_list {
 434         list_t  zal_list;
 435         uint64_t zal_size;
 436 } zio_alloc_list_t;
 437
 438 typedef struct zio_link {
 439         zio_t           *zl_parent;
 440         zio_t           *zl_child;
 441         list_node_t     zl_parent_node;
 442         list_node_t     zl_child_node;
 443 } zio_link_t;
 444
 445 struct zio {
 446         /* Core information about this I/O */
 447         zbookmark_phys_t        io_bookmark;
 448         zio_prop_t      io_prop;
 449         zio_type_t      io_type;
 450         enum zio_child  io_child_type;
 451         enum trim_flag  io_trim_flags;
 452         int             io_cmd;
 453         zio_priority_t  io_priority;
 454         uint8_t         io_reexecute;
 455         uint8_t         io_state[ZIO_WAIT_TYPES];
 456         uint64_t        io_txg;
 457         spa_t           *io_spa;
 458         blkptr_t        *io_bp;
 459         blkptr_t        *io_bp_override;
 460         blkptr_t        io_bp_copy;
 461         list_t          io_parent_list;
 462         list_t          io_child_list;
 463         zio_t           *io_logical;
 464         zio_transform_t *io_transform_stack;
 465
 466         /* Callback info */
 467         zio_done_func_t *io_ready;
 468         zio_done_func_t *io_children_ready;
 469         zio_done_func_t *io_physdone;
 470         zio_done_func_t *io_done;
 471         void            *io_private;
 472         int64_t         io_prev_space_delta;    /* DMU private */
 473         blkptr_t        io_bp_orig;
 474         /* io_lsize != io_orig_size iff this is a raw write */
 475         uint64_t        io_lsize;
 476
 477         /* Data represented by this I/O */
 478         struct abd      *io_abd;
 479         struct abd      *io_orig_abd;
 480         uint64_t        io_size;
 481         uint64_t        io_orig_size;
 482
 483         /* Stuff for the vdev stack */
 484         vdev_t          *io_vd;
 485         void            *io_vsd;
 486         const zio_vsd_ops_t *io_vsd_ops;
 487         metaslab_class_t *io_metaslab_class;    /* dva throttle class */
 488
 489         uint64_t        io_offset;
 490         hrtime_t        io_timestamp;   /* submitted at */
 491         hrtime_t        io_queued_timestamp;
 492         hrtime_t        io_target_timestamp;
 493         hrtime_t        io_delta;       /* vdev queue service delta */
 494         hrtime_t        io_delay;       /* Device access time (disk or */
 495                                         /* file). */
 496         avl_node_t      io_queue_node;
 497         avl_node_t      io_offset_node;
 498         avl_node_t      io_alloc_node;
 499         zio_alloc_list_t        io_alloc_list;
 500
 501         /* Internal pipeline state */
 502         enum zio_flag   io_flags;
 503         enum zio_stage  io_stage;
 504         enum zio_stage  io_pipeline;
 505         enum zio_flag   io_orig_flags;
 506         enum zio_stage  io_orig_stage;
 507         enum zio_stage  io_orig_pipeline;
 508         enum zio_stage  io_pipeline_trace;
 509         int             io_error;
 510         int             io_child_error[ZIO_CHILD_TYPES];
 511         uint64_t        io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
 512         uint64_t        io_child_count;
 513         uint64_t        io_phys_children;
 514         uint64_t        io_parent_count;
 515         uint64_t        *io_stall;
 516         zio_t           *io_gang_leader;
 517         zio_gang_node_t *io_gang_tree;
 518         void            *io_executor;
 519         void            *io_waiter;
 520         void            *io_bio;
 521         kmutex_t        io_lock;
 522         kcondvar_t      io_cv;
 523         int             io_allocator;
 524
 525         /* FMA state */
 526         zio_cksum_report_t *io_cksum_report;
 527         uint64_t        io_ena;
 528
 529         /* Taskq dispatching state */
 530         taskq_ent_t     io_tqent;
 531 };
 532
 533 enum blk_verify_flag {
 534         BLK_VERIFY_ONLY,
 535         BLK_VERIFY_LOG,
 536         BLK_VERIFY_HALT
 537 };
 538
 539 extern int zio_bookmark_compare(const void *, const void *);
 540
 541 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
 542     zio_done_func_t *done, void *priv, enum zio_flag flags);
 543
 544 extern zio_t *zio_root(spa_t *spa,
 545     zio_done_func_t *done, void *priv, enum zio_flag flags);
 546
 547 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
 548     struct abd *data, uint64_t lsize, zio_done_func_t *done, void *priv,
 549     zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
 550
 551 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 552     struct abd *data, uint64_t size, uint64_t psize, const zio_prop_t *zp,
 553     zio_done_func_t *ready, zio_done_func_t *children_ready,
 554     zio_done_func_t *physdone, zio_done_func_t *done,
 555     void *priv, zio_priority_t priority, enum zio_flag flags,
 556     const zbookmark_phys_t *zb);
 557
 558 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 559     struct abd *data, uint64_t size, zio_done_func_t *done, void *priv,
 560     zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
 561
 562 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
 563     boolean_t nopwrite);
 564
 565 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
 566
 567 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
 568     const blkptr_t *bp,
 569     zio_done_func_t *done, void *priv, enum zio_flag flags);
 570
 571 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
 572     zio_done_func_t *done, void *priv, enum zio_flag flags);
 573
 574 extern zio_t *zio_trim(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
 575     zio_done_func_t *done, void *priv, zio_priority_t priority,
 576     enum zio_flag flags, enum trim_flag trim_flags);
 577
 578 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 579     uint64_t size, struct abd *data, int checksum,
 580     zio_done_func_t *done, void *priv, zio_priority_t priority,
 581     enum zio_flag flags, boolean_t labels);
 582
 583 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 584     uint64_t size, struct abd *data, int checksum,
 585     zio_done_func_t *done, void *priv, zio_priority_t priority,
 586     enum zio_flag flags, boolean_t labels);
 587
 588 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
 589     const blkptr_t *bp, enum zio_flag flags);
 590
 591 extern int zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg,
 592     blkptr_t *new_bp, uint64_t size, boolean_t *slog);
 593 extern void zio_flush(zio_t *zio, vdev_t *vd);
 594 extern void zio_shrink(zio_t *zio, uint64_t size);
 595
 596 extern int zio_wait(zio_t *zio);
 597 extern void zio_nowait(zio_t *zio);
 598 extern void zio_execute(zio_t *zio);
 599 extern void zio_interrupt(zio_t *zio);
 600 extern void zio_delay_init(zio_t *zio);
 601 extern void zio_delay_interrupt(zio_t *zio);
 602 extern void zio_deadman(zio_t *zio, char *tag);
 603
 604 extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **);
 605 extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **);
 606 extern zio_t *zio_unique_parent(zio_t *cio);
 607 extern void zio_add_child(zio_t *pio, zio_t *cio);
 608
 609 extern void *zio_buf_alloc(size_t size);
 610 extern void zio_buf_free(void *buf, size_t size);
 611 extern void *zio_data_buf_alloc(size_t size);
 612 extern void zio_data_buf_free(void *buf, size_t size);
 613
 614 extern void zio_push_transform(zio_t *zio, struct abd *abd, uint64_t size,
 615     uint64_t bufsize, zio_transform_func_t *transform);
 616 extern void zio_pop_transforms(zio_t *zio);
 617
 618 extern void zio_resubmit_stage_async(void *);
 619
 620 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
 621     uint64_t offset, struct abd *data, uint64_t size, int type,
 622     zio_priority_t priority, enum zio_flag flags,
 623     zio_done_func_t *done, void *priv);
 624
 625 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
 626     struct abd *data, uint64_t size, zio_type_t type, zio_priority_t priority,
 627     enum zio_flag flags, zio_done_func_t *done, void *priv);
 628
 629 extern void zio_vdev_io_bypass(zio_t *zio);
 630 extern void zio_vdev_io_reissue(zio_t *zio);
 631 extern void zio_vdev_io_redone(zio_t *zio);
 632
 633 extern void zio_change_priority(zio_t *pio, zio_priority_t priority);
 634
 635 extern void zio_checksum_verified(zio_t *zio);
 636 extern int zio_worst_error(int e1, int e2);
 637
 638 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
 639     enum zio_checksum parent);
 640 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
 641     enum zio_checksum child, enum zio_checksum parent);
 642 extern enum zio_compress zio_compress_select(spa_t *spa,
 643     enum zio_compress child, enum zio_compress parent);
 644 extern uint8_t zio_complevel_select(spa_t *spa, enum zio_compress compress,
 645     uint8_t child, uint8_t parent);
 646
 647 extern void zio_suspend(spa_t *spa, zio_t *zio, zio_suspend_reason_t);
 648 extern int zio_resume(spa_t *spa);
 649 extern void zio_resume_wait(spa_t *spa);
 650
 651 extern boolean_t zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp,
 652     boolean_t config_held, enum blk_verify_flag blk_verify);
 653
 654 /*
 655  * Initial setup and teardown.
 656  */
 657 extern void zio_init(void);
 658 extern void zio_fini(void);
 659
 660 /*
 661  * Fault injection
 662  */
 663 struct zinject_record;
 664 extern uint32_t zio_injection_enabled;
 665 extern int zio_inject_fault(char *name, int flags, int *id,
 666     struct zinject_record *record);
 667 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
 668     struct zinject_record *record);
 669 extern int zio_clear_fault(int id);
 670 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
 671 extern int zio_handle_decrypt_injection(spa_t *spa, const zbookmark_phys_t *zb,
 672     uint64_t type, int error);
 673 extern int zio_handle_fault_injection(zio_t *zio, int error);
 674 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
 675 extern int zio_handle_device_injections(vdev_t *vd, zio_t *zio, int err1,
 676     int err2);
 677 extern int zio_handle_label_injection(zio_t *zio, int error);
 678 extern void zio_handle_ignored_writes(zio_t *zio);
 679 extern hrtime_t zio_handle_io_delay(zio_t *zio);
 680
 681 /*
 682  * Checksum ereport functions
 683  */
 684 extern int zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd,
 685     const zbookmark_phys_t *zb, struct zio *zio, uint64_t offset,
 686     uint64_t length, void *arg, struct zio_bad_cksum *info);
 687 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
 688     const abd_t *good_data, const abd_t *bad_data, boolean_t drop_if_identical);
 689
 690 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
 691
 692 /* If we have the good data in hand, this function can be used */
 693 extern int zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
 694     const zbookmark_phys_t *zb, struct zio *zio, uint64_t offset,
 695     uint64_t length, const abd_t *good_data, const abd_t *bad_data,
 696     struct zio_bad_cksum *info);
 697
 698 /* Called from spa_sync(), but primarily an injection handler */
 699 extern void spa_handle_ignored_writes(spa_t *spa);
 700
 701 /* zbookmark_phys functions */
 702 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
 703     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
 704 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
 705     uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
 706
 707 #ifdef  __cplusplus
 708 }
 709 #endif
 710
 711 #endif  /* _ZIO_H */