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 /*
 167  * This was originally an enum type. However, those are 32-bit and there is no
 168  * way to make a 64-bit enum type. Since we ran out of bits for flags, we were
 169  * forced to upgrade it to a uint64_t.
 170  */
 171 typedef uint64_t zio_flag_t;
 172         /*
 173          * Flags inherited by gang, ddt, and vdev children,
 174          * and that must be equal for two zios to aggregate
 175          */
 176 #define ZIO_FLAG_DONT_AGGREGATE (1ULL << 0)
 177 #define ZIO_FLAG_IO_REPAIR      (1ULL << 1)
 178 #define ZIO_FLAG_SELF_HEAL      (1ULL << 2)
 179 #define ZIO_FLAG_RESILVER       (1ULL << 3)
 180 #define ZIO_FLAG_SCRUB          (1ULL << 4)
 181 #define ZIO_FLAG_SCAN_THREAD    (1ULL << 5)
 182 #define ZIO_FLAG_PHYSICAL       (1ULL << 6)
 183
 184 #define ZIO_FLAG_AGG_INHERIT    (ZIO_FLAG_CANFAIL - 1)
 185
 186         /*
 187          * Flags inherited by ddt, gang, and vdev children.
 188          */
 189 #define ZIO_FLAG_CANFAIL        (1ULL << 7)     /* must be first for INHERIT */
 190 #define ZIO_FLAG_SPECULATIVE    (1ULL << 8)
 191 #define ZIO_FLAG_CONFIG_WRITER  (1ULL << 9)
 192 #define ZIO_FLAG_DONT_RETRY     (1ULL << 10)
 193 #define ZIO_FLAG_NODATA         (1ULL << 12)
 194 #define ZIO_FLAG_INDUCE_DAMAGE  (1ULL << 13)
 195 #define ZIO_FLAG_IO_ALLOCATING  (1ULL << 14)
 196
 197 #define ZIO_FLAG_DDT_INHERIT    (ZIO_FLAG_IO_RETRY - 1)
 198 #define ZIO_FLAG_GANG_INHERIT   (ZIO_FLAG_IO_RETRY - 1)
 199
 200         /*
 201          * Flags inherited by vdev children.
 202          */
 203 #define ZIO_FLAG_IO_RETRY       (1ULL << 15)    /* must be first for INHERIT */
 204 #define ZIO_FLAG_PROBE          (1ULL << 16)
 205 #define ZIO_FLAG_TRYHARD        (1ULL << 17)
 206 #define ZIO_FLAG_OPTIONAL       (1ULL << 18)
 207
 208 #define ZIO_FLAG_VDEV_INHERIT   (ZIO_FLAG_DONT_QUEUE - 1)
 209
 210         /*
 211          * Flags not inherited by any children.
 212          */
 213 #define ZIO_FLAG_DONT_QUEUE     (1ULL << 19)    /* must be first for INHERIT */
 214 #define ZIO_FLAG_DONT_PROPAGATE (1ULL << 20)
 215 #define ZIO_FLAG_IO_BYPASS      (1ULL << 21)
 216 #define ZIO_FLAG_IO_REWRITE     (1ULL << 22)
 217 #define ZIO_FLAG_RAW_COMPRESS   (1ULL << 23)
 218 #define ZIO_FLAG_RAW_ENCRYPT    (1ULL << 24)
 219 #define ZIO_FLAG_GANG_CHILD     (1ULL << 25)
 220 #define ZIO_FLAG_DDT_CHILD      (1ULL << 26)
 221 #define ZIO_FLAG_GODFATHER      (1ULL << 27)
 222 #define ZIO_FLAG_NOPWRITE       (1ULL << 28)
 223 #define ZIO_FLAG_REEXECUTED     (1ULL << 29)
 224 #define ZIO_FLAG_DELEGATED      (1ULL << 30)
 225
 226 #define ZIO_FLAG_MUSTSUCCEED            0
 227 #define ZIO_FLAG_RAW    (ZIO_FLAG_RAW_COMPRESS | ZIO_FLAG_RAW_ENCRYPT)
 228
 229 #define ZIO_DDT_CHILD_FLAGS(zio)                                \
 230         (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |             \
 231         ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
 232
 233 #define ZIO_GANG_CHILD_FLAGS(zio)                               \
 234         (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |            \
 235         ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
 236
 237 #define ZIO_VDEV_CHILD_FLAGS(zio)                               \
 238         (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |            \
 239         ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_CANFAIL)
 240
 241 #define ZIO_CHILD_BIT(x)                (1U << (x))
 242 #define ZIO_CHILD_BIT_IS_SET(val, x)    ((val) & (1U << (x)))
 243
 244 enum zio_child {
 245         ZIO_CHILD_VDEV = 0,
 246         ZIO_CHILD_GANG,
 247         ZIO_CHILD_DDT,
 248         ZIO_CHILD_LOGICAL,
 249         ZIO_CHILD_TYPES
 250 };
 251
 252 #define ZIO_CHILD_VDEV_BIT              ZIO_CHILD_BIT(ZIO_CHILD_VDEV)
 253 #define ZIO_CHILD_GANG_BIT              ZIO_CHILD_BIT(ZIO_CHILD_GANG)
 254 #define ZIO_CHILD_DDT_BIT               ZIO_CHILD_BIT(ZIO_CHILD_DDT)
 255 #define ZIO_CHILD_LOGICAL_BIT           ZIO_CHILD_BIT(ZIO_CHILD_LOGICAL)
 256 #define ZIO_CHILD_ALL_BITS                                      \
 257         (ZIO_CHILD_VDEV_BIT | ZIO_CHILD_GANG_BIT |              \
 258         ZIO_CHILD_DDT_BIT | ZIO_CHILD_LOGICAL_BIT)
 259
 260 enum zio_wait_type {
 261         ZIO_WAIT_READY = 0,
 262         ZIO_WAIT_DONE,
 263         ZIO_WAIT_TYPES
 264 };
 265
 266 typedef void zio_done_func_t(zio_t *zio);
 267
 268 extern int zio_exclude_metadata;
 269 extern int zio_dva_throttle_enabled;
 270 extern const char *const zio_type_name[ZIO_TYPES];
 271
 272 /*
 273  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
 274  * identifies any block in the pool.  By convention, the meta-objset (MOS)
 275  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
 276  * except root blocks and ZIL blocks, which are defined as follows:
 277  *
 278  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
 279  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
 280  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
 281  * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
 282  *
 283  * Note: this structure is called a bookmark because its original purpose
 284  * was to remember where to resume a pool-wide traverse.
 285  *
 286  * Note: this structure is passed between userland and the kernel, and is
 287  * stored on disk (by virtue of being incorporated into other on-disk
 288  * structures, e.g. dsl_scan_phys_t).
 289  *
 290  * If the head_errlog feature is enabled a different on-disk format for error
 291  * logs is used. This introduces the use of an error bookmark, a four-tuple
 292  * <object, level, blkid, birth> that uniquely identifies any error block
 293  * in the pool. The birth transaction group is used to track whether the block
 294  * has been overwritten by newer data or added to a snapshot since its marking
 295  * as an error.
 296  */
 297 struct zbookmark_phys {
 298         uint64_t        zb_objset;
 299         uint64_t        zb_object;
 300         int64_t         zb_level;
 301         uint64_t        zb_blkid;
 302 };
 303
 304 struct zbookmark_err_phys {
 305         uint64_t        zb_object;
 306         int64_t         zb_level;
 307         uint64_t        zb_blkid;
 308         uint64_t        zb_birth;
 309 };
 310
 311 #define SET_BOOKMARK(zb, objset, object, level, blkid)  \
 312 {                                                       \
 313         (zb)->zb_objset = objset;                       \
 314         (zb)->zb_object = object;                       \
 315         (zb)->zb_level = level;                         \
 316         (zb)->zb_blkid = blkid;                         \
 317 }
 318
 319 #define ZB_DESTROYED_OBJSET     (-1ULL)
 320
 321 #define ZB_ROOT_OBJECT          (0ULL)
 322 #define ZB_ROOT_LEVEL           (-1LL)
 323 #define ZB_ROOT_BLKID           (0ULL)
 324
 325 #define ZB_ZIL_OBJECT           (0ULL)
 326 #define ZB_ZIL_LEVEL            (-2LL)
 327
 328 #define ZB_DNODE_LEVEL          (-3LL)
 329 #define ZB_DNODE_BLKID          (0ULL)
 330
 331 #define ZB_IS_ZERO(zb)                                          \
 332         ((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&        \
 333         (zb)->zb_level == 0 && (zb)->zb_blkid == 0)
 334 #define ZB_IS_ROOT(zb)                          \
 335         ((zb)->zb_object == ZB_ROOT_OBJECT &&   \
 336         (zb)->zb_level == ZB_ROOT_LEVEL &&      \
 337         (zb)->zb_blkid == ZB_ROOT_BLKID)
 338
 339 typedef struct zio_prop {
 340         enum zio_checksum       zp_checksum;
 341         enum zio_compress       zp_compress;
 342         uint8_t                 zp_complevel;
 343         uint8_t                 zp_level;
 344         uint8_t                 zp_copies;
 345         dmu_object_type_t       zp_type;
 346         boolean_t               zp_dedup;
 347         boolean_t               zp_dedup_verify;
 348         boolean_t               zp_nopwrite;
 349         boolean_t               zp_brtwrite;
 350         boolean_t               zp_encrypt;
 351         boolean_t               zp_byteorder;
 352         uint8_t                 zp_salt[ZIO_DATA_SALT_LEN];
 353         uint8_t                 zp_iv[ZIO_DATA_IV_LEN];
 354         uint8_t                 zp_mac[ZIO_DATA_MAC_LEN];
 355         uint32_t                zp_zpl_smallblk;
 356 } zio_prop_t;
 357
 358 typedef struct zio_cksum_report zio_cksum_report_t;
 359
 360 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
 361     const abd_t *good_data);
 362 typedef void zio_cksum_free_f(void *cbdata, size_t size);
 363
 364 struct zio_bad_cksum;                           /* defined in zio_checksum.h */
 365 struct dnode_phys;
 366 struct abd;
 367
 368 struct zio_cksum_report {
 369         struct zio_cksum_report *zcr_next;
 370         nvlist_t                *zcr_ereport;
 371         nvlist_t                *zcr_detector;
 372         void                    *zcr_cbdata;
 373         size_t                  zcr_cbinfo;     /* passed to zcr_free() */
 374         uint64_t                zcr_sector;
 375         uint64_t                zcr_align;
 376         uint64_t                zcr_length;
 377         zio_cksum_finish_f      *zcr_finish;
 378         zio_cksum_free_f        *zcr_free;
 379
 380         /* internal use only */
 381         struct zio_bad_cksum    *zcr_ckinfo;    /* information from failure */
 382 };
 383
 384 typedef struct zio_vsd_ops {
 385         zio_done_func_t         *vsd_free;
 386 } zio_vsd_ops_t;
 387
 388 typedef struct zio_gang_node {
 389         zio_gbh_phys_t          *gn_gbh;
 390         struct zio_gang_node    *gn_child[SPA_GBH_NBLKPTRS];
 391 } zio_gang_node_t;
 392
 393 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
 394     zio_gang_node_t *gn, struct abd *data, uint64_t offset);
 395
 396 typedef void zio_transform_func_t(zio_t *zio, struct abd *data, uint64_t size);
 397
 398 typedef struct zio_transform {
 399         struct abd              *zt_orig_abd;
 400         uint64_t                zt_orig_size;
 401         uint64_t                zt_bufsize;
 402         zio_transform_func_t    *zt_transform;
 403         struct zio_transform    *zt_next;
 404 } zio_transform_t;
 405
 406 typedef zio_t *zio_pipe_stage_t(zio_t *zio);
 407
 408 /*
 409  * The io_reexecute flags are distinct from io_flags because the child must
 410  * be able to propagate them to the parent.  The normal io_flags are local
 411  * to the zio, not protected by any lock, and not modifiable by children;
 412  * the reexecute flags are protected by io_lock, modifiable by children,
 413  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
 414  */
 415 #define ZIO_REEXECUTE_NOW       0x01
 416 #define ZIO_REEXECUTE_SUSPEND   0x02
 417
 418 /*
 419  * The io_trim flags are used to specify the type of TRIM to perform.  They
 420  * only apply to ZIO_TYPE_TRIM zios are distinct from io_flags.
 421  */
 422 enum trim_flag {
 423         ZIO_TRIM_SECURE         = 1U << 0,
 424 };
 425
 426 typedef struct zio_alloc_list {
 427         list_t  zal_list;
 428         uint64_t zal_size;
 429 } zio_alloc_list_t;
 430
 431 typedef struct zio_link {
 432         zio_t           *zl_parent;
 433         zio_t           *zl_child;
 434         list_node_t     zl_parent_node;
 435         list_node_t     zl_child_node;
 436 } zio_link_t;
 437
 438 enum zio_qstate {
 439         ZIO_QS_NONE = 0,
 440         ZIO_QS_QUEUED,
 441         ZIO_QS_ACTIVE,
 442 };
 443
 444 struct zio {
 445         /* Core information about this I/O */
 446         zbookmark_phys_t        io_bookmark;
 447         zio_prop_t      io_prop;
 448         zio_type_t      io_type;
 449         enum zio_child  io_child_type;
 450         enum trim_flag  io_trim_flags;
 451         int             io_cmd;
 452         zio_priority_t  io_priority;
 453         uint8_t         io_reexecute;
 454         uint8_t         io_state[ZIO_WAIT_TYPES];
 455         uint64_t        io_txg;
 456         spa_t           *io_spa;
 457         blkptr_t        *io_bp;
 458         blkptr_t        *io_bp_override;
 459         blkptr_t        io_bp_copy;
 460         list_t          io_parent_list;
 461         list_t          io_child_list;
 462         zio_t           *io_logical;
 463         zio_transform_t *io_transform_stack;
 464
 465         /* Callback info */
 466         zio_done_func_t *io_ready;
 467         zio_done_func_t *io_children_ready;
 468         zio_done_func_t *io_done;
 469         void            *io_private;
 470         int64_t         io_prev_space_delta;    /* DMU private */
 471         blkptr_t        io_bp_orig;
 472         /* io_lsize != io_orig_size iff this is a raw write */
 473         uint64_t        io_lsize;
 474
 475         /* Data represented by this I/O */
 476         struct abd      *io_abd;
 477         struct abd      *io_orig_abd;
 478         uint64_t        io_size;
 479         uint64_t        io_orig_size;
 480
 481         /* Stuff for the vdev stack */
 482         vdev_t          *io_vd;
 483         void            *io_vsd;
 484         const zio_vsd_ops_t *io_vsd_ops;
 485         metaslab_class_t *io_metaslab_class;    /* dva throttle class */
 486
 487         enum zio_qstate io_queue_state; /* vdev queue state */
 488         union {
 489                 list_node_t l;
 490                 avl_node_t a;
 491         } io_queue_node ____cacheline_aligned;  /* allocator and vdev queues */
 492         avl_node_t      io_offset_node; /* vdev offset queues */
 493         uint64_t        io_offset;
 494         hrtime_t        io_timestamp;   /* submitted at */
 495         hrtime_t        io_queued_timestamp;
 496         hrtime_t        io_target_timestamp;
 497         hrtime_t        io_delta;       /* vdev queue service delta */
 498         hrtime_t        io_delay;       /* Device access time (disk or */
 499                                         /* file). */
 500         zio_alloc_list_t        io_alloc_list;
 501
 502         /* Internal pipeline state */
 503         zio_flag_t      io_flags;
 504         enum zio_stage  io_stage;
 505         enum zio_stage  io_pipeline;
 506         zio_flag_t      io_orig_flags;
 507         enum zio_stage  io_orig_stage;
 508         enum zio_stage  io_orig_pipeline;
 509         enum zio_stage  io_pipeline_trace;
 510         int             io_error;
 511         int             io_child_error[ZIO_CHILD_TYPES];
 512         uint64_t        io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
 513         uint64_t        *io_stall;
 514         zio_t           *io_gang_leader;
 515         zio_gang_node_t *io_gang_tree;
 516         void            *io_executor;
 517         void            *io_waiter;
 518         void            *io_bio;
 519         kmutex_t        io_lock;
 520         kcondvar_t      io_cv;
 521         int             io_allocator;
 522
 523         /* FMA state */
 524         zio_cksum_report_t *io_cksum_report;
 525         uint64_t        io_ena;
 526
 527         /* Taskq dispatching state */
 528         taskq_ent_t     io_tqent;
 529 };
 530
 531 enum blk_verify_flag {
 532         BLK_VERIFY_ONLY,
 533         BLK_VERIFY_LOG,
 534         BLK_VERIFY_HALT
 535 };
 536
 537 enum blk_config_flag {
 538         BLK_CONFIG_HELD,   // SCL_VDEV held for writer
 539         BLK_CONFIG_NEEDED, // SCL_VDEV should be obtained for reader
 540         BLK_CONFIG_SKIP,   // skip checks which require SCL_VDEV
 541 };
 542
 543 extern int zio_bookmark_compare(const void *, const void *);
 544
 545 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
 546     zio_done_func_t *done, void *priv, zio_flag_t flags);
 547
 548 extern zio_t *zio_root(spa_t *spa,
 549     zio_done_func_t *done, void *priv, zio_flag_t flags);
 550
 551 extern void zio_destroy(zio_t *zio);
 552
 553 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
 554     struct abd *data, uint64_t lsize, zio_done_func_t *done, void *priv,
 555     zio_priority_t priority, zio_flag_t flags, const zbookmark_phys_t *zb);
 556
 557 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 558     struct abd *data, uint64_t size, uint64_t psize, const zio_prop_t *zp,
 559     zio_done_func_t *ready, zio_done_func_t *children_ready,
 560     zio_done_func_t *done, void *priv, zio_priority_t priority,
 561     zio_flag_t flags, const zbookmark_phys_t *zb);
 562
 563 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 564     struct abd *data, uint64_t size, zio_done_func_t *done, void *priv,
 565     zio_priority_t priority, zio_flag_t flags, zbookmark_phys_t *zb);
 566
 567 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
 568     boolean_t nopwrite, boolean_t brtwrite);
 569
 570 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
 571
 572 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
 573     const blkptr_t *bp,
 574     zio_done_func_t *done, void *priv, zio_flag_t flags);
 575
 576 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
 577     zio_done_func_t *done, void *priv, zio_flag_t flags);
 578
 579 extern zio_t *zio_trim(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
 580     zio_done_func_t *done, void *priv, zio_priority_t priority,
 581     zio_flag_t flags, enum trim_flag trim_flags);
 582
 583 extern zio_t *zio_read_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     zio_flag_t flags, boolean_t labels);
 587
 588 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 589     uint64_t size, struct abd *data, int checksum,
 590     zio_done_func_t *done, void *priv, zio_priority_t priority,
 591     zio_flag_t flags, boolean_t labels);
 592
 593 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
 594     const blkptr_t *bp, zio_flag_t flags);
 595
 596 extern int zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg,
 597     blkptr_t *new_bp, uint64_t size, boolean_t *slog);
 598 extern void zio_flush(zio_t *zio, vdev_t *vd);
 599 extern void zio_shrink(zio_t *zio, uint64_t size);
 600
 601 extern int zio_wait(zio_t *zio);
 602 extern void zio_nowait(zio_t *zio);
 603 extern void zio_execute(void *zio);
 604 extern void zio_interrupt(void *zio);
 605 extern void zio_delay_init(zio_t *zio);
 606 extern void zio_delay_interrupt(zio_t *zio);
 607 extern void zio_deadman(zio_t *zio, const char *tag);
 608
 609 extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **);
 610 extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **);
 611 extern zio_t *zio_unique_parent(zio_t *cio);
 612 extern void zio_add_child(zio_t *pio, zio_t *cio);
 613 extern void zio_add_child_first(zio_t *pio, zio_t *cio);
 614
 615 extern void *zio_buf_alloc(size_t size);
 616 extern void zio_buf_free(void *buf, size_t size);
 617 extern void *zio_data_buf_alloc(size_t size);
 618 extern void zio_data_buf_free(void *buf, size_t size);
 619
 620 extern void zio_push_transform(zio_t *zio, struct abd *abd, uint64_t size,
 621     uint64_t bufsize, zio_transform_func_t *transform);
 622 extern void zio_pop_transforms(zio_t *zio);
 623
 624 extern void zio_resubmit_stage_async(void *);
 625
 626 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
 627     uint64_t offset, struct abd *data, uint64_t size, int type,
 628     zio_priority_t priority, zio_flag_t flags,
 629     zio_done_func_t *done, void *priv);
 630
 631 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
 632     struct abd *data, uint64_t size, zio_type_t type, zio_priority_t priority,
 633     zio_flag_t flags, zio_done_func_t *done, void *priv);
 634
 635 extern void zio_vdev_io_bypass(zio_t *zio);
 636 extern void zio_vdev_io_reissue(zio_t *zio);
 637 extern void zio_vdev_io_redone(zio_t *zio);
 638
 639 extern void zio_change_priority(zio_t *pio, zio_priority_t priority);
 640
 641 extern void zio_checksum_verified(zio_t *zio);
 642 extern int zio_worst_error(int e1, int e2);
 643
 644 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
 645     enum zio_checksum parent);
 646 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
 647     enum zio_checksum child, enum zio_checksum parent);
 648 extern enum zio_compress zio_compress_select(spa_t *spa,
 649     enum zio_compress child, enum zio_compress parent);
 650 extern uint8_t zio_complevel_select(spa_t *spa, enum zio_compress compress,
 651     uint8_t child, uint8_t parent);
 652
 653 extern void zio_suspend(spa_t *spa, zio_t *zio, zio_suspend_reason_t);
 654 extern int zio_resume(spa_t *spa);
 655 extern void zio_resume_wait(spa_t *spa);
 656
 657 extern boolean_t zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp,
 658     enum blk_config_flag blk_config, enum blk_verify_flag blk_verify);
 659
 660 /*
 661  * Initial setup and teardown.
 662  */
 663 extern void zio_init(void);
 664 extern void zio_fini(void);
 665
 666 /*
 667  * Fault injection
 668  */
 669 struct zinject_record;
 670 extern uint32_t zio_injection_enabled;
 671 extern int zio_inject_fault(char *name, int flags, int *id,
 672     struct zinject_record *record);
 673 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
 674     struct zinject_record *record);
 675 extern int zio_clear_fault(int id);
 676 extern void zio_handle_panic_injection(spa_t *spa, const char *tag,
 677     uint64_t type);
 678 extern int zio_handle_decrypt_injection(spa_t *spa, const zbookmark_phys_t *zb,
 679     uint64_t type, int error);
 680 extern int zio_handle_fault_injection(zio_t *zio, int error);
 681 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
 682 extern int zio_handle_device_injections(vdev_t *vd, zio_t *zio, int err1,
 683     int err2);
 684 extern int zio_handle_label_injection(zio_t *zio, int error);
 685 extern void zio_handle_ignored_writes(zio_t *zio);
 686 extern hrtime_t zio_handle_io_delay(zio_t *zio);
 687
 688 /*
 689  * Checksum ereport functions
 690  */
 691 extern int zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd,
 692     const zbookmark_phys_t *zb, struct zio *zio, uint64_t offset,
 693     uint64_t length, struct zio_bad_cksum *info);
 694 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
 695     const abd_t *good_data, const abd_t *bad_data, boolean_t drop_if_identical);
 696
 697 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
 698
 699 /* If we have the good data in hand, this function can be used */
 700 extern int zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
 701     const zbookmark_phys_t *zb, struct zio *zio, uint64_t offset,
 702     uint64_t length, const abd_t *good_data, const abd_t *bad_data,
 703     struct zio_bad_cksum *info);
 704
 705 void zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr);
 706 extern void zfs_ereport_snapshot_post(const char *subclass, spa_t *spa,
 707     const char *name);
 708
 709 /* Called from spa_sync(), but primarily an injection handler */
 710 extern void spa_handle_ignored_writes(spa_t *spa);
 711
 712 /* zbookmark_phys functions */
 713 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
 714     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
 715 boolean_t zbookmark_subtree_tbd(const struct dnode_phys *dnp,
 716     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
 717 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
 718     uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
 719
 720 #ifdef  __cplusplus
 721 }
 722 #endif
 723
 724 #endif  /* _ZIO_H */