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 https://opensource.org/licenses/CDDL-1.0.
  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         ZIO_CHECKSUM_EDONR,
  92         ZIO_CHECKSUM_BLAKE3,
  93         ZIO_CHECKSUM_FUNCTIONS
  94 };
  95
  96 /*
  97  * The number of "legacy" compression functions which can be set on individual
  98  * objects.
  99  */
 100 #define ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2
 101
 102 #define ZIO_CHECKSUM_ON_VALUE   ZIO_CHECKSUM_FLETCHER_4
 103 #define ZIO_CHECKSUM_DEFAULT    ZIO_CHECKSUM_ON
 104
 105 #define ZIO_CHECKSUM_MASK       0xffULL
 106 #define ZIO_CHECKSUM_VERIFY     (1U << 8)
 107
 108 #define ZIO_DEDUPCHECKSUM       ZIO_CHECKSUM_SHA256
 109
 110 /* macros defining encryption lengths */
 111 #define ZIO_OBJSET_MAC_LEN              32
 112 #define ZIO_DATA_IV_LEN                 12
 113 #define ZIO_DATA_SALT_LEN               8
 114 #define ZIO_DATA_MAC_LEN                16
 115
 116 /*
 117  * The number of "legacy" compression functions which can be set on individual
 118  * objects.
 119  */
 120 #define ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
 121
 122 /*
 123  * The meaning of "compress = on" selected by the compression features enabled
 124  * on a given pool.
 125  */
 126 #define ZIO_COMPRESS_LEGACY_ON_VALUE    ZIO_COMPRESS_LZJB
 127 #define ZIO_COMPRESS_LZ4_ON_VALUE       ZIO_COMPRESS_LZ4
 128
 129 #define ZIO_COMPRESS_DEFAULT            ZIO_COMPRESS_ON
 130
 131 #define BOOTFS_COMPRESS_VALID(compress)                 \
 132         ((compress) == ZIO_COMPRESS_LZJB ||             \
 133         (compress) == ZIO_COMPRESS_LZ4 ||               \
 134         (compress) == ZIO_COMPRESS_GZIP_1 ||            \
 135         (compress) == ZIO_COMPRESS_GZIP_2 ||            \
 136         (compress) == ZIO_COMPRESS_GZIP_3 ||            \
 137         (compress) == ZIO_COMPRESS_GZIP_4 ||            \
 138         (compress) == ZIO_COMPRESS_GZIP_5 ||            \
 139         (compress) == ZIO_COMPRESS_GZIP_6 ||            \
 140         (compress) == ZIO_COMPRESS_GZIP_7 ||            \
 141         (compress) == ZIO_COMPRESS_GZIP_8 ||            \
 142         (compress) == ZIO_COMPRESS_GZIP_9 ||            \
 143         (compress) == ZIO_COMPRESS_ZLE ||               \
 144         (compress) == ZIO_COMPRESS_ZSTD ||              \
 145         (compress) == ZIO_COMPRESS_ON ||                \
 146         (compress) == ZIO_COMPRESS_OFF)
 147
 148
 149 #define ZIO_COMPRESS_ALGO(x)    (x & SPA_COMPRESSMASK)
 150 #define ZIO_COMPRESS_LEVEL(x)   ((x & ~SPA_COMPRESSMASK) >> SPA_COMPRESSBITS)
 151 #define ZIO_COMPRESS_RAW(type, level)   (type | ((level) << SPA_COMPRESSBITS))
 152
 153 #define ZIO_COMPLEVEL_ZSTD(level)       \
 154         ZIO_COMPRESS_RAW(ZIO_COMPRESS_ZSTD, level)
 155
 156 #define ZIO_FAILURE_MODE_WAIT           0
 157 #define ZIO_FAILURE_MODE_CONTINUE       1
 158 #define ZIO_FAILURE_MODE_PANIC          2
 159
 160 typedef enum zio_suspend_reason {
 161         ZIO_SUSPEND_NONE = 0,
 162         ZIO_SUSPEND_IOERR,
 163         ZIO_SUSPEND_MMP,
 164 } zio_suspend_reason_t;
 165
 166 enum zio_flag {
 167         /*
 168          * Flags inherited by gang, ddt, and vdev children,
 169          * and that must be equal for two zios to aggregate
 170          */
 171         ZIO_FLAG_DONT_AGGREGATE = 1U << 0,
 172         ZIO_FLAG_IO_REPAIR      = 1U << 1,
 173         ZIO_FLAG_SELF_HEAL      = 1U << 2,
 174         ZIO_FLAG_RESILVER       = 1U << 3,
 175         ZIO_FLAG_SCRUB          = 1U << 4,
 176         ZIO_FLAG_SCAN_THREAD    = 1U << 5,
 177         ZIO_FLAG_PHYSICAL       = 1U << 6,
 178
 179 #define ZIO_FLAG_AGG_INHERIT    (ZIO_FLAG_CANFAIL - 1)
 180
 181         /*
 182          * Flags inherited by ddt, gang, and vdev children.
 183          */
 184         ZIO_FLAG_CANFAIL        = 1U << 7,      /* must be first for INHERIT */
 185         ZIO_FLAG_SPECULATIVE    = 1U << 8,
 186         ZIO_FLAG_CONFIG_WRITER  = 1U << 9,
 187         ZIO_FLAG_DONT_RETRY     = 1U << 10,
 188         ZIO_FLAG_DONT_CACHE     = 1U << 11,
 189         ZIO_FLAG_NODATA         = 1U << 12,
 190         ZIO_FLAG_INDUCE_DAMAGE  = 1U << 13,
 191         ZIO_FLAG_IO_ALLOCATING  = 1U << 14,
 192
 193 #define ZIO_FLAG_DDT_INHERIT    (ZIO_FLAG_IO_RETRY - 1)
 194 #define ZIO_FLAG_GANG_INHERIT   (ZIO_FLAG_IO_RETRY - 1)
 195
 196         /*
 197          * Flags inherited by vdev children.
 198          */
 199         ZIO_FLAG_IO_RETRY       = 1U << 15,     /* must be first for INHERIT */
 200         ZIO_FLAG_PROBE          = 1U << 16,
 201         ZIO_FLAG_TRYHARD        = 1U << 17,
 202         ZIO_FLAG_OPTIONAL       = 1U << 18,
 203
 204 #define ZIO_FLAG_VDEV_INHERIT   (ZIO_FLAG_DONT_QUEUE - 1)
 205
 206         /*
 207          * Flags not inherited by any children.
 208          */
 209         ZIO_FLAG_DONT_QUEUE     = 1U << 19,     /* must be first for INHERIT */
 210         ZIO_FLAG_DONT_PROPAGATE = 1U << 20,
 211         ZIO_FLAG_IO_BYPASS      = 1U << 21,
 212         ZIO_FLAG_IO_REWRITE     = 1U << 22,
 213         ZIO_FLAG_RAW_COMPRESS   = 1U << 23,
 214         ZIO_FLAG_RAW_ENCRYPT    = 1U << 24,
 215         ZIO_FLAG_GANG_CHILD     = 1U << 25,
 216         ZIO_FLAG_DDT_CHILD      = 1U << 26,
 217         ZIO_FLAG_GODFATHER      = 1U << 27,
 218         ZIO_FLAG_NOPWRITE       = 1U << 28,
 219         ZIO_FLAG_REEXECUTED     = 1U << 29,
 220         ZIO_FLAG_DELEGATED      = 1U << 30,
 221         ZIO_FLAG_FASTWRITE      = 1U << 31,
 222 };
 223
 224 #define ZIO_FLAG_MUSTSUCCEED            0
 225 #define ZIO_FLAG_RAW    (ZIO_FLAG_RAW_COMPRESS | ZIO_FLAG_RAW_ENCRYPT)
 226
 227 #define ZIO_DDT_CHILD_FLAGS(zio)                                \
 228         (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |             \
 229         ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
 230
 231 #define ZIO_GANG_CHILD_FLAGS(zio)                               \
 232         (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |            \
 233         ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
 234
 235 #define ZIO_VDEV_CHILD_FLAGS(zio)                               \
 236         (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |            \
 237         ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_CANFAIL)
 238
 239 #define ZIO_CHILD_BIT(x)                (1U << (x))
 240 #define ZIO_CHILD_BIT_IS_SET(val, x)    ((val) & (1U << (x)))
 241
 242 enum zio_child {
 243         ZIO_CHILD_VDEV = 0,
 244         ZIO_CHILD_GANG,
 245         ZIO_CHILD_DDT,
 246         ZIO_CHILD_LOGICAL,
 247         ZIO_CHILD_TYPES
 248 };
 249
 250 #define ZIO_CHILD_VDEV_BIT              ZIO_CHILD_BIT(ZIO_CHILD_VDEV)
 251 #define ZIO_CHILD_GANG_BIT              ZIO_CHILD_BIT(ZIO_CHILD_GANG)
 252 #define ZIO_CHILD_DDT_BIT               ZIO_CHILD_BIT(ZIO_CHILD_DDT)
 253 #define ZIO_CHILD_LOGICAL_BIT           ZIO_CHILD_BIT(ZIO_CHILD_LOGICAL)
 254 #define ZIO_CHILD_ALL_BITS                                      \
 255         (ZIO_CHILD_VDEV_BIT | ZIO_CHILD_GANG_BIT |              \
 256         ZIO_CHILD_DDT_BIT | ZIO_CHILD_LOGICAL_BIT)
 257
 258 enum zio_wait_type {
 259         ZIO_WAIT_READY = 0,
 260         ZIO_WAIT_DONE,
 261         ZIO_WAIT_TYPES
 262 };
 263
 264 typedef void zio_done_func_t(zio_t *zio);
 265
 266 extern int zio_exclude_metadata;
 267 extern int zio_dva_throttle_enabled;
 268 extern const char *const zio_type_name[ZIO_TYPES];
 269
 270 /*
 271  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
 272  * identifies any block in the pool.  By convention, the meta-objset (MOS)
 273  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
 274  * except root blocks and ZIL blocks, which are defined as follows:
 275  *
 276  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
 277  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
 278  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
 279  * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
 280  *
 281  * Note: this structure is called a bookmark because its original purpose
 282  * was to remember where to resume a pool-wide traverse.
 283  *
 284  * Note: this structure is passed between userland and the kernel, and is
 285  * stored on disk (by virtue of being incorporated into other on-disk
 286  * structures, e.g. dsl_scan_phys_t).
 287  *
 288  * If the head_errlog feature is enabled a different on-disk format for error
 289  * logs is used. This introduces the use of an error bookmark, a four-tuple
 290  * <object, level, blkid, birth> that uniquely identifies any error block
 291  * in the pool. The birth transaction group is used to track whether the block
 292  * has been overwritten by newer data or added to a snapshot since its marking
 293  * as an error.
 294  */
 295 struct zbookmark_phys {
 296         uint64_t        zb_objset;
 297         uint64_t        zb_object;
 298         int64_t         zb_level;
 299         uint64_t        zb_blkid;
 300 };
 301
 302 typedef struct zbookmark_err_phys {
 303         uint64_t        zb_object;
 304         int64_t         zb_level;
 305         uint64_t        zb_blkid;
 306         uint64_t        zb_birth;
 307 } zbookmark_err_phys_t;
 308
 309 #define SET_BOOKMARK(zb, objset, object, level, blkid)  \
 310 {                                                       \
 311         (zb)->zb_objset = objset;                       \
 312         (zb)->zb_object = object;                       \
 313         (zb)->zb_level = level;                         \
 314         (zb)->zb_blkid = blkid;                         \
 315 }
 316
 317 #define ZB_DESTROYED_OBJSET     (-1ULL)
 318
 319 #define ZB_ROOT_OBJECT          (0ULL)
 320 #define ZB_ROOT_LEVEL           (-1LL)
 321 #define ZB_ROOT_BLKID           (0ULL)
 322
 323 #define ZB_ZIL_OBJECT           (0ULL)
 324 #define ZB_ZIL_LEVEL            (-2LL)
 325
 326 #define ZB_DNODE_LEVEL          (-3LL)
 327 #define ZB_DNODE_BLKID          (0ULL)
 328
 329 #define ZB_IS_ZERO(zb)                                          \
 330         ((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&        \
 331         (zb)->zb_level == 0 && (zb)->zb_blkid == 0)
 332 #define ZB_IS_ROOT(zb)                          \
 333         ((zb)->zb_object == ZB_ROOT_OBJECT &&   \
 334         (zb)->zb_level == ZB_ROOT_LEVEL &&      \
 335         (zb)->zb_blkid == ZB_ROOT_BLKID)
 336
 337 typedef struct zio_prop {
 338         enum zio_checksum       zp_checksum;
 339         enum zio_compress       zp_compress;
 340         uint8_t                 zp_complevel;
 341         dmu_object_type_t       zp_type;
 342         uint8_t                 zp_level;
 343         uint8_t                 zp_copies;
 344         boolean_t               zp_dedup;
 345         boolean_t               zp_dedup_verify;
 346         boolean_t               zp_nopwrite;
 347         boolean_t               zp_encrypt;
 348         boolean_t               zp_byteorder;
 349         uint8_t                 zp_salt[ZIO_DATA_SALT_LEN];
 350         uint8_t                 zp_iv[ZIO_DATA_IV_LEN];
 351         uint8_t                 zp_mac[ZIO_DATA_MAC_LEN];
 352         uint32_t                zp_zpl_smallblk;
 353 } zio_prop_t;
 354
 355 typedef struct zio_cksum_report zio_cksum_report_t;
 356
 357 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
 358     const abd_t *good_data);
 359 typedef void zio_cksum_free_f(void *cbdata, size_t size);
 360
 361 struct zio_bad_cksum;                           /* defined in zio_checksum.h */
 362 struct dnode_phys;
 363 struct abd;
 364
 365 struct zio_cksum_report {
 366         struct zio_cksum_report *zcr_next;
 367         nvlist_t                *zcr_ereport;
 368         nvlist_t                *zcr_detector;
 369         void                    *zcr_cbdata;
 370         size_t                  zcr_cbinfo;     /* passed to zcr_free() */
 371         uint64_t                zcr_sector;
 372         uint64_t                zcr_align;
 373         uint64_t                zcr_length;
 374         zio_cksum_finish_f      *zcr_finish;
 375         zio_cksum_free_f        *zcr_free;
 376
 377         /* internal use only */
 378         struct zio_bad_cksum    *zcr_ckinfo;    /* information from failure */
 379 };
 380
 381 typedef struct zio_vsd_ops {
 382         zio_done_func_t         *vsd_free;
 383 } zio_vsd_ops_t;
 384
 385 typedef struct zio_gang_node {
 386         zio_gbh_phys_t          *gn_gbh;
 387         struct zio_gang_node    *gn_child[SPA_GBH_NBLKPTRS];
 388 } zio_gang_node_t;
 389
 390 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
 391     zio_gang_node_t *gn, struct abd *data, uint64_t offset);
 392
 393 typedef void zio_transform_func_t(zio_t *zio, struct abd *data, uint64_t size);
 394
 395 typedef struct zio_transform {
 396         struct abd              *zt_orig_abd;
 397         uint64_t                zt_orig_size;
 398         uint64_t                zt_bufsize;
 399         zio_transform_func_t    *zt_transform;
 400         struct zio_transform    *zt_next;
 401 } zio_transform_t;
 402
 403 typedef zio_t *zio_pipe_stage_t(zio_t *zio);
 404
 405 /*
 406  * The io_reexecute flags are distinct from io_flags because the child must
 407  * be able to propagate them to the parent.  The normal io_flags are local
 408  * to the zio, not protected by any lock, and not modifiable by children;
 409  * the reexecute flags are protected by io_lock, modifiable by children,
 410  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
 411  */
 412 #define ZIO_REEXECUTE_NOW       0x01
 413 #define ZIO_REEXECUTE_SUSPEND   0x02
 414
 415 /*
 416  * The io_trim flags are used to specify the type of TRIM to perform.  They
 417  * only apply to ZIO_TYPE_TRIM zios are distinct from io_flags.
 418  */
 419 enum trim_flag {
 420         ZIO_TRIM_SECURE         = 1U << 0,
 421 };
 422
 423 typedef struct zio_alloc_list {
 424         list_t  zal_list;
 425         uint64_t zal_size;
 426 } zio_alloc_list_t;
 427
 428 typedef struct zio_link {
 429         zio_t           *zl_parent;
 430         zio_t           *zl_child;
 431         list_node_t     zl_parent_node;
 432         list_node_t     zl_child_node;
 433 } zio_link_t;
 434
 435 struct zio {
 436         /* Core information about this I/O */
 437         zbookmark_phys_t        io_bookmark;
 438         zio_prop_t      io_prop;
 439         zio_type_t      io_type;
 440         enum zio_child  io_child_type;
 441         enum trim_flag  io_trim_flags;
 442         int             io_cmd;
 443         zio_priority_t  io_priority;
 444         uint8_t         io_reexecute;
 445         uint8_t         io_state[ZIO_WAIT_TYPES];
 446         uint64_t        io_txg;
 447         spa_t           *io_spa;
 448         blkptr_t        *io_bp;
 449         blkptr_t        *io_bp_override;
 450         blkptr_t        io_bp_copy;
 451         list_t          io_parent_list;
 452         list_t          io_child_list;
 453         zio_t           *io_logical;
 454         zio_transform_t *io_transform_stack;
 455
 456         /* Callback info */
 457         zio_done_func_t *io_ready;
 458         zio_done_func_t *io_children_ready;
 459         zio_done_func_t *io_physdone;
 460         zio_done_func_t *io_done;
 461         void            *io_private;
 462         int64_t         io_prev_space_delta;    /* DMU private */
 463         blkptr_t        io_bp_orig;
 464         /* io_lsize != io_orig_size iff this is a raw write */
 465         uint64_t        io_lsize;
 466
 467         /* Data represented by this I/O */
 468         struct abd      *io_abd;
 469         struct abd      *io_orig_abd;
 470         uint64_t        io_size;
 471         uint64_t        io_orig_size;
 472
 473         /* Stuff for the vdev stack */
 474         vdev_t          *io_vd;
 475         void            *io_vsd;
 476         const zio_vsd_ops_t *io_vsd_ops;
 477         metaslab_class_t *io_metaslab_class;    /* dva throttle class */
 478
 479         uint64_t        io_offset;
 480         hrtime_t        io_timestamp;   /* submitted at */
 481         hrtime_t        io_queued_timestamp;
 482         hrtime_t        io_target_timestamp;
 483         hrtime_t        io_delta;       /* vdev queue service delta */
 484         hrtime_t        io_delay;       /* Device access time (disk or */
 485                                         /* file). */
 486         avl_node_t      io_queue_node;
 487         avl_node_t      io_offset_node;
 488         avl_node_t      io_alloc_node;
 489         zio_alloc_list_t        io_alloc_list;
 490
 491         /* Internal pipeline state */
 492         enum zio_flag   io_flags;
 493         enum zio_stage  io_stage;
 494         enum zio_stage  io_pipeline;
 495         enum zio_flag   io_orig_flags;
 496         enum zio_stage  io_orig_stage;
 497         enum zio_stage  io_orig_pipeline;
 498         enum zio_stage  io_pipeline_trace;
 499         int             io_error;
 500         int             io_child_error[ZIO_CHILD_TYPES];
 501         uint64_t        io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
 502         uint64_t        io_child_count;
 503         uint64_t        io_phys_children;
 504         uint64_t        io_parent_count;
 505         uint64_t        *io_stall;
 506         zio_t           *io_gang_leader;
 507         zio_gang_node_t *io_gang_tree;
 508         void            *io_executor;
 509         void            *io_waiter;
 510         void            *io_bio;
 511         kmutex_t        io_lock;
 512         kcondvar_t      io_cv;
 513         int             io_allocator;
 514
 515         /* FMA state */
 516         zio_cksum_report_t *io_cksum_report;
 517         uint64_t        io_ena;
 518
 519         /* Taskq dispatching state */
 520         taskq_ent_t     io_tqent;
 521 };
 522
 523 enum blk_verify_flag {
 524         BLK_VERIFY_ONLY,
 525         BLK_VERIFY_LOG,
 526         BLK_VERIFY_HALT
 527 };
 528
 529 extern int zio_bookmark_compare(const void *, const void *);
 530
 531 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
 532     zio_done_func_t *done, void *priv, enum zio_flag flags);
 533
 534 extern zio_t *zio_root(spa_t *spa,
 535     zio_done_func_t *done, void *priv, enum zio_flag flags);
 536
 537 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
 538     struct abd *data, uint64_t lsize, zio_done_func_t *done, void *priv,
 539     zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
 540
 541 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 542     struct abd *data, uint64_t size, uint64_t psize, const zio_prop_t *zp,
 543     zio_done_func_t *ready, zio_done_func_t *children_ready,
 544     zio_done_func_t *physdone, zio_done_func_t *done,
 545     void *priv, zio_priority_t priority, enum zio_flag flags,
 546     const zbookmark_phys_t *zb);
 547
 548 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 549     struct abd *data, uint64_t size, zio_done_func_t *done, void *priv,
 550     zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
 551
 552 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
 553     boolean_t nopwrite);
 554
 555 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
 556
 557 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
 558     const blkptr_t *bp,
 559     zio_done_func_t *done, void *priv, enum zio_flag flags);
 560
 561 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
 562     zio_done_func_t *done, void *priv, enum zio_flag flags);
 563
 564 extern zio_t *zio_trim(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
 565     zio_done_func_t *done, void *priv, zio_priority_t priority,
 566     enum zio_flag flags, enum trim_flag trim_flags);
 567
 568 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 569     uint64_t size, struct abd *data, int checksum,
 570     zio_done_func_t *done, void *priv, zio_priority_t priority,
 571     enum zio_flag flags, boolean_t labels);
 572
 573 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 574     uint64_t size, struct abd *data, int checksum,
 575     zio_done_func_t *done, void *priv, zio_priority_t priority,
 576     enum zio_flag flags, boolean_t labels);
 577
 578 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
 579     const blkptr_t *bp, enum zio_flag flags);
 580
 581 extern int zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg,
 582     blkptr_t *new_bp, uint64_t size, boolean_t *slog);
 583 extern void zio_flush(zio_t *zio, vdev_t *vd);
 584 extern void zio_shrink(zio_t *zio, uint64_t size);
 585
 586 extern int zio_wait(zio_t *zio);
 587 extern void zio_nowait(zio_t *zio);
 588 extern void zio_execute(void *zio);
 589 extern void zio_interrupt(void *zio);
 590 extern void zio_delay_init(zio_t *zio);
 591 extern void zio_delay_interrupt(zio_t *zio);
 592 extern void zio_deadman(zio_t *zio, const char *tag);
 593
 594 extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **);
 595 extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **);
 596 extern zio_t *zio_unique_parent(zio_t *cio);
 597 extern void zio_add_child(zio_t *pio, zio_t *cio);
 598
 599 extern void *zio_buf_alloc(size_t size);
 600 extern void zio_buf_free(void *buf, size_t size);
 601 extern void *zio_data_buf_alloc(size_t size);
 602 extern void zio_data_buf_free(void *buf, size_t size);
 603
 604 extern void zio_push_transform(zio_t *zio, struct abd *abd, uint64_t size,
 605     uint64_t bufsize, zio_transform_func_t *transform);
 606 extern void zio_pop_transforms(zio_t *zio);
 607
 608 extern void zio_resubmit_stage_async(void *);
 609
 610 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
 611     uint64_t offset, struct abd *data, uint64_t size, int type,
 612     zio_priority_t priority, enum zio_flag flags,
 613     zio_done_func_t *done, void *priv);
 614
 615 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
 616     struct abd *data, uint64_t size, zio_type_t type, zio_priority_t priority,
 617     enum zio_flag flags, zio_done_func_t *done, void *priv);
 618
 619 extern void zio_vdev_io_bypass(zio_t *zio);
 620 extern void zio_vdev_io_reissue(zio_t *zio);
 621 extern void zio_vdev_io_redone(zio_t *zio);
 622
 623 extern void zio_change_priority(zio_t *pio, zio_priority_t priority);
 624
 625 extern void zio_checksum_verified(zio_t *zio);
 626 extern int zio_worst_error(int e1, int e2);
 627
 628 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
 629     enum zio_checksum parent);
 630 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
 631     enum zio_checksum child, enum zio_checksum parent);
 632 extern enum zio_compress zio_compress_select(spa_t *spa,
 633     enum zio_compress child, enum zio_compress parent);
 634 extern uint8_t zio_complevel_select(spa_t *spa, enum zio_compress compress,
 635     uint8_t child, uint8_t parent);
 636
 637 extern void zio_suspend(spa_t *spa, zio_t *zio, zio_suspend_reason_t);
 638 extern int zio_resume(spa_t *spa);
 639 extern void zio_resume_wait(spa_t *spa);
 640
 641 extern boolean_t zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp,
 642     boolean_t config_held, enum blk_verify_flag blk_verify);
 643
 644 /*
 645  * Initial setup and teardown.
 646  */
 647 extern void zio_init(void);
 648 extern void zio_fini(void);
 649
 650 /*
 651  * Fault injection
 652  */
 653 struct zinject_record;
 654 extern uint32_t zio_injection_enabled;
 655 extern int zio_inject_fault(char *name, int flags, int *id,
 656     struct zinject_record *record);
 657 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
 658     struct zinject_record *record);
 659 extern int zio_clear_fault(int id);
 660 extern void zio_handle_panic_injection(spa_t *spa, const char *tag,
 661     uint64_t type);
 662 extern int zio_handle_decrypt_injection(spa_t *spa, const zbookmark_phys_t *zb,
 663     uint64_t type, int error);
 664 extern int zio_handle_fault_injection(zio_t *zio, int error);
 665 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
 666 extern int zio_handle_device_injections(vdev_t *vd, zio_t *zio, int err1,
 667     int err2);
 668 extern int zio_handle_label_injection(zio_t *zio, int error);
 669 extern void zio_handle_ignored_writes(zio_t *zio);
 670 extern hrtime_t zio_handle_io_delay(zio_t *zio);
 671
 672 /*
 673  * Checksum ereport functions
 674  */
 675 extern int zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd,
 676     const zbookmark_phys_t *zb, struct zio *zio, uint64_t offset,
 677     uint64_t length, struct zio_bad_cksum *info);
 678 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
 679     const abd_t *good_data, const abd_t *bad_data, boolean_t drop_if_identical);
 680
 681 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
 682
 683 /* If we have the good data in hand, this function can be used */
 684 extern int zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
 685     const zbookmark_phys_t *zb, struct zio *zio, uint64_t offset,
 686     uint64_t length, const abd_t *good_data, const abd_t *bad_data,
 687     struct zio_bad_cksum *info);
 688
 689 void zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr);
 690 extern void zfs_ereport_snapshot_post(const char *subclass, spa_t *spa,
 691     const char *name);
 692
 693 /* Called from spa_sync(), but primarily an injection handler */
 694 extern void spa_handle_ignored_writes(spa_t *spa);
 695
 696 /* zbookmark_phys functions */
 697 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
 698     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
 699 boolean_t zbookmark_subtree_tbd(const struct dnode_phys *dnp,
 700     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
 701 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
 702     uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
 703
 704 #ifdef  __cplusplus
 705 }
 706 #endif
 707
 708 #endif  /* _ZIO_H */