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, 2018 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 /* supported encryption algorithms */
 108 enum zio_encrypt {
 109         ZIO_CRYPT_INHERIT = 0,
 110         ZIO_CRYPT_ON,
 111         ZIO_CRYPT_OFF,
 112         ZIO_CRYPT_AES_128_CCM,
 113         ZIO_CRYPT_AES_192_CCM,
 114         ZIO_CRYPT_AES_256_CCM,
 115         ZIO_CRYPT_AES_128_GCM,
 116         ZIO_CRYPT_AES_192_GCM,
 117         ZIO_CRYPT_AES_256_GCM,
 118         ZIO_CRYPT_FUNCTIONS
 119 };
 120
 121 #define ZIO_CRYPT_ON_VALUE      ZIO_CRYPT_AES_256_CCM
 122 #define ZIO_CRYPT_DEFAULT       ZIO_CRYPT_OFF
 123
 124 /* macros defining encryption lengths */
 125 #define ZIO_OBJSET_MAC_LEN              32
 126 #define ZIO_DATA_IV_LEN                 12
 127 #define ZIO_DATA_SALT_LEN               8
 128 #define ZIO_DATA_MAC_LEN                16
 129
 130 /*
 131  * The number of "legacy" compression functions which can be set on individual
 132  * objects.
 133  */
 134 #define ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
 135
 136 /*
 137  * The meaning of "compress = on" selected by the compression features enabled
 138  * on a given pool.
 139  */
 140 #define ZIO_COMPRESS_LEGACY_ON_VALUE    ZIO_COMPRESS_LZJB
 141 #define ZIO_COMPRESS_LZ4_ON_VALUE       ZIO_COMPRESS_LZ4
 142
 143 #define ZIO_COMPRESS_DEFAULT            ZIO_COMPRESS_OFF
 144
 145 #define BOOTFS_COMPRESS_VALID(compress)                 \
 146         ((compress) == ZIO_COMPRESS_LZJB ||             \
 147         (compress) == ZIO_COMPRESS_LZ4 ||               \
 148         (compress) == ZIO_COMPRESS_GZIP_1 ||            \
 149         (compress) == ZIO_COMPRESS_GZIP_2 ||            \
 150         (compress) == ZIO_COMPRESS_GZIP_3 ||            \
 151         (compress) == ZIO_COMPRESS_GZIP_4 ||            \
 152         (compress) == ZIO_COMPRESS_GZIP_5 ||            \
 153         (compress) == ZIO_COMPRESS_GZIP_6 ||            \
 154         (compress) == ZIO_COMPRESS_GZIP_7 ||            \
 155         (compress) == ZIO_COMPRESS_GZIP_8 ||            \
 156         (compress) == ZIO_COMPRESS_GZIP_9 ||            \
 157         (compress) == ZIO_COMPRESS_ZLE ||               \
 158         (compress) == ZIO_COMPRESS_ON ||                \
 159         (compress) == ZIO_COMPRESS_OFF)
 160
 161 #define ZIO_FAILURE_MODE_WAIT           0
 162 #define ZIO_FAILURE_MODE_CONTINUE       1
 163 #define ZIO_FAILURE_MODE_PANIC          2
 164
 165 typedef enum zio_suspend_reason {
 166         ZIO_SUSPEND_NONE = 0,
 167         ZIO_SUSPEND_IOERR,
 168         ZIO_SUSPEND_MMP,
 169 } zio_suspend_reason_t;
 170
 171 enum zio_flag {
 172         /*
 173          * Flags inherited by gang, ddt, and vdev children,
 174          * and that must be equal for two zios to aggregate
 175          */
 176         ZIO_FLAG_DONT_AGGREGATE = 1 << 0,
 177         ZIO_FLAG_IO_REPAIR      = 1 << 1,
 178         ZIO_FLAG_SELF_HEAL      = 1 << 2,
 179         ZIO_FLAG_RESILVER       = 1 << 3,
 180         ZIO_FLAG_SCRUB          = 1 << 4,
 181         ZIO_FLAG_SCAN_THREAD    = 1 << 5,
 182         ZIO_FLAG_PHYSICAL       = 1 << 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         ZIO_FLAG_CANFAIL        = 1 << 7,       /* must be first for INHERIT */
 190         ZIO_FLAG_SPECULATIVE    = 1 << 8,
 191         ZIO_FLAG_CONFIG_WRITER  = 1 << 9,
 192         ZIO_FLAG_DONT_RETRY     = 1 << 10,
 193         ZIO_FLAG_DONT_CACHE     = 1 << 11,
 194         ZIO_FLAG_NODATA         = 1 << 12,
 195         ZIO_FLAG_INDUCE_DAMAGE  = 1 << 13,
 196         ZIO_FLAG_IO_ALLOCATING  = 1 << 14,
 197
 198 #define ZIO_FLAG_DDT_INHERIT    (ZIO_FLAG_IO_RETRY - 1)
 199 #define ZIO_FLAG_GANG_INHERIT   (ZIO_FLAG_IO_RETRY - 1)
 200
 201         /*
 202          * Flags inherited by vdev children.
 203          */
 204         ZIO_FLAG_IO_RETRY       = 1 << 15,      /* must be first for INHERIT */
 205         ZIO_FLAG_PROBE          = 1 << 16,
 206         ZIO_FLAG_TRYHARD        = 1 << 17,
 207         ZIO_FLAG_OPTIONAL       = 1 << 18,
 208
 209 #define ZIO_FLAG_VDEV_INHERIT   (ZIO_FLAG_DONT_QUEUE - 1)
 210
 211         /*
 212          * Flags not inherited by any children.
 213          */
 214         ZIO_FLAG_DONT_QUEUE     = 1 << 19,      /* must be first for INHERIT */
 215         ZIO_FLAG_DONT_PROPAGATE = 1 << 20,
 216         ZIO_FLAG_IO_BYPASS      = 1 << 21,
 217         ZIO_FLAG_IO_REWRITE     = 1 << 22,
 218         ZIO_FLAG_RAW_COMPRESS   = 1 << 23,
 219         ZIO_FLAG_RAW_ENCRYPT    = 1 << 24,
 220         ZIO_FLAG_GANG_CHILD     = 1 << 25,
 221         ZIO_FLAG_DDT_CHILD      = 1 << 26,
 222         ZIO_FLAG_GODFATHER      = 1 << 27,
 223         ZIO_FLAG_NOPWRITE       = 1 << 28,
 224         ZIO_FLAG_REEXECUTED     = 1 << 29,
 225         ZIO_FLAG_DELEGATED      = 1 << 30,
 226         ZIO_FLAG_FASTWRITE      = 1 << 31,
 227 };
 228
 229 #define ZIO_FLAG_MUSTSUCCEED            0
 230 #define ZIO_FLAG_RAW    (ZIO_FLAG_RAW_COMPRESS | ZIO_FLAG_RAW_ENCRYPT)
 231
 232 #define ZIO_DDT_CHILD_FLAGS(zio)                                \
 233         (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |             \
 234         ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
 235
 236 #define ZIO_GANG_CHILD_FLAGS(zio)                               \
 237         (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |            \
 238         ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
 239
 240 #define ZIO_VDEV_CHILD_FLAGS(zio)                               \
 241         (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |            \
 242         ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_CANFAIL)
 243
 244 #define ZIO_CHILD_BIT(x)                (1 << (x))
 245 #define ZIO_CHILD_BIT_IS_SET(val, x)    ((val) & (1 << (x)))
 246
 247 enum zio_child {
 248         ZIO_CHILD_VDEV = 0,
 249         ZIO_CHILD_GANG,
 250         ZIO_CHILD_DDT,
 251         ZIO_CHILD_LOGICAL,
 252         ZIO_CHILD_TYPES
 253 };
 254
 255 #define ZIO_CHILD_VDEV_BIT              ZIO_CHILD_BIT(ZIO_CHILD_VDEV)
 256 #define ZIO_CHILD_GANG_BIT              ZIO_CHILD_BIT(ZIO_CHILD_GANG)
 257 #define ZIO_CHILD_DDT_BIT               ZIO_CHILD_BIT(ZIO_CHILD_DDT)
 258 #define ZIO_CHILD_LOGICAL_BIT           ZIO_CHILD_BIT(ZIO_CHILD_LOGICAL)
 259 #define ZIO_CHILD_ALL_BITS                                      \
 260         (ZIO_CHILD_VDEV_BIT | ZIO_CHILD_GANG_BIT |              \
 261         ZIO_CHILD_DDT_BIT | ZIO_CHILD_LOGICAL_BIT)
 262
 263 enum zio_wait_type {
 264         ZIO_WAIT_READY = 0,
 265         ZIO_WAIT_DONE,
 266         ZIO_WAIT_TYPES
 267 };
 268
 269 /*
 270  * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent
 271  * graveyard) to indicate checksum errors and fragmentation.
 272  */
 273 #define ECKSUM  EBADE
 274 #define EFRAGS  EBADR
 275
 276 /* Similar for ENOACTIVE */
 277 #define ENOTACTIVE      ENOANO
 278
 279 typedef void zio_done_func_t(zio_t *zio);
 280
 281 extern int zio_dva_throttle_enabled;
 282 extern const char *zio_type_name[ZIO_TYPES];
 283
 284 /*
 285  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
 286  * identifies any block in the pool.  By convention, the meta-objset (MOS)
 287  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
 288  * except root blocks and ZIL blocks, which are defined as follows:
 289  *
 290  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
 291  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
 292  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
 293  * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
 294  *
 295  * Note: this structure is called a bookmark because its original purpose
 296  * was to remember where to resume a pool-wide traverse.
 297  *
 298  * Note: this structure is passed between userland and the kernel, and is
 299  * stored on disk (by virtue of being incorporated into other on-disk
 300  * structures, e.g. dsl_scan_phys_t).
 301  */
 302 struct zbookmark_phys {
 303         uint64_t        zb_objset;
 304         uint64_t        zb_object;
 305         int64_t         zb_level;
 306         uint64_t        zb_blkid;
 307 };
 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         dmu_object_type_t       zp_type;
 341         uint8_t                 zp_level;
 342         uint8_t                 zp_copies;
 343         boolean_t               zp_dedup;
 344         boolean_t               zp_dedup_verify;
 345         boolean_t               zp_nopwrite;
 346         boolean_t               zp_encrypt;
 347         boolean_t               zp_byteorder;
 348         uint8_t                 zp_salt[ZIO_DATA_SALT_LEN];
 349         uint8_t                 zp_iv[ZIO_DATA_IV_LEN];
 350         uint8_t                 zp_mac[ZIO_DATA_MAC_LEN];
 351         uint32_t                zp_zpl_smallblk;
 352 } zio_prop_t;
 353
 354 typedef struct zio_cksum_report zio_cksum_report_t;
 355
 356 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
 357     const abd_t *good_data);
 358 typedef void zio_cksum_free_f(void *cbdata, size_t size);
 359
 360 struct zio_bad_cksum;                           /* defined in zio_checksum.h */
 361 struct dnode_phys;
 362 struct abd;
 363
 364 struct zio_cksum_report {
 365         struct zio_cksum_report *zcr_next;
 366         nvlist_t                *zcr_ereport;
 367         nvlist_t                *zcr_detector;
 368         void                    *zcr_cbdata;
 369         size_t                  zcr_cbinfo;     /* passed to zcr_free() */
 370         uint64_t                zcr_align;
 371         uint64_t                zcr_length;
 372         zio_cksum_finish_f      *zcr_finish;
 373         zio_cksum_free_f        *zcr_free;
 374
 375         /* internal use only */
 376         struct zio_bad_cksum    *zcr_ckinfo;    /* information from failure */
 377 };
 378
 379 typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
 380     void *arg);
 381
 382 zio_vsd_cksum_report_f  zio_vsd_default_cksum_report;
 383
 384 typedef struct zio_vsd_ops {
 385         zio_done_func_t         *vsd_free;
 386         zio_vsd_cksum_report_f  *vsd_cksum_report;
 387 } zio_vsd_ops_t;
 388
 389 typedef struct zio_gang_node {
 390         zio_gbh_phys_t          *gn_gbh;
 391         struct zio_gang_node    *gn_child[SPA_GBH_NBLKPTRS];
 392 } zio_gang_node_t;
 393
 394 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
 395     zio_gang_node_t *gn, struct abd *data, uint64_t offset);
 396
 397 typedef void zio_transform_func_t(zio_t *zio, struct abd *data, uint64_t size);
 398
 399 typedef struct zio_transform {
 400         struct abd              *zt_orig_abd;
 401         uint64_t                zt_orig_size;
 402         uint64_t                zt_bufsize;
 403         zio_transform_func_t    *zt_transform;
 404         struct zio_transform    *zt_next;
 405 } zio_transform_t;
 406
 407 typedef zio_t *zio_pipe_stage_t(zio_t *zio);
 408
 409 /*
 410  * The io_reexecute flags are distinct from io_flags because the child must
 411  * be able to propagate them to the parent.  The normal io_flags are local
 412  * to the zio, not protected by any lock, and not modifiable by children;
 413  * the reexecute flags are protected by io_lock, modifiable by children,
 414  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
 415  */
 416 #define ZIO_REEXECUTE_NOW       0x01
 417 #define ZIO_REEXECUTE_SUSPEND   0x02
 418
 419 /*
 420  * The io_trim flags are used to specify the type of TRIM to perform.  They
 421  * only apply to ZIO_TYPE_TRIM zios are distinct from io_flags.
 422  */
 423 enum trim_flag {
 424         ZIO_TRIM_SECURE         = 1 << 0,
 425 };
 426
 427 typedef struct zio_alloc_list {
 428         list_t  zal_list;
 429         uint64_t zal_size;
 430 } zio_alloc_list_t;
 431
 432 typedef struct zio_link {
 433         zio_t           *zl_parent;
 434         zio_t           *zl_child;
 435         list_node_t     zl_parent_node;
 436         list_node_t     zl_child_node;
 437 } zio_link_t;
 438
 439 struct zio {
 440         /* Core information about this I/O */
 441         zbookmark_phys_t        io_bookmark;
 442         zio_prop_t      io_prop;
 443         zio_type_t      io_type;
 444         enum zio_child  io_child_type;
 445         enum trim_flag  io_trim_flags;
 446         int             io_cmd;
 447         zio_priority_t  io_priority;
 448         uint8_t         io_reexecute;
 449         uint8_t         io_state[ZIO_WAIT_TYPES];
 450         uint64_t        io_txg;
 451         spa_t           *io_spa;
 452         blkptr_t        *io_bp;
 453         blkptr_t        *io_bp_override;
 454         blkptr_t        io_bp_copy;
 455         list_t          io_parent_list;
 456         list_t          io_child_list;
 457         zio_t           *io_logical;
 458         zio_transform_t *io_transform_stack;
 459
 460         /* Callback info */
 461         zio_done_func_t *io_ready;
 462         zio_done_func_t *io_children_ready;
 463         zio_done_func_t *io_physdone;
 464         zio_done_func_t *io_done;
 465         void            *io_private;
 466         int64_t         io_prev_space_delta;    /* DMU private */
 467         blkptr_t        io_bp_orig;
 468         /* io_lsize != io_orig_size iff this is a raw write */
 469         uint64_t        io_lsize;
 470
 471         /* Data represented by this I/O */
 472         struct abd      *io_abd;
 473         struct abd      *io_orig_abd;
 474         uint64_t        io_size;
 475         uint64_t        io_orig_size;
 476
 477         /* Stuff for the vdev stack */
 478         vdev_t          *io_vd;
 479         void            *io_vsd;
 480         const zio_vsd_ops_t *io_vsd_ops;
 481         metaslab_class_t *io_metaslab_class;    /* dva throttle class */
 482
 483         uint64_t        io_offset;
 484         hrtime_t        io_timestamp;   /* submitted at */
 485         hrtime_t        io_queued_timestamp;
 486         hrtime_t        io_target_timestamp;
 487         hrtime_t        io_delta;       /* vdev queue service delta */
 488         hrtime_t        io_delay;       /* Device access time (disk or */
 489                                         /* file). */
 490         avl_node_t      io_queue_node;
 491         avl_node_t      io_offset_node;
 492         avl_node_t      io_alloc_node;
 493         zio_alloc_list_t        io_alloc_list;
 494
 495         /* Internal pipeline state */
 496         enum zio_flag   io_flags;
 497         enum zio_stage  io_stage;
 498         enum zio_stage  io_pipeline;
 499         enum zio_flag   io_orig_flags;
 500         enum zio_stage  io_orig_stage;
 501         enum zio_stage  io_orig_pipeline;
 502         enum zio_stage  io_pipeline_trace;
 503         int             io_error;
 504         int             io_child_error[ZIO_CHILD_TYPES];
 505         uint64_t        io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
 506         uint64_t        io_child_count;
 507         uint64_t        io_phys_children;
 508         uint64_t        io_parent_count;
 509         uint64_t        *io_stall;
 510         zio_t           *io_gang_leader;
 511         zio_gang_node_t *io_gang_tree;
 512         void            *io_executor;
 513         void            *io_waiter;
 514         kmutex_t        io_lock;
 515         kcondvar_t      io_cv;
 516         int             io_allocator;
 517
 518         /* FMA state */
 519         zio_cksum_report_t *io_cksum_report;
 520         uint64_t        io_ena;
 521
 522         /* Taskq dispatching state */
 523         taskq_ent_t     io_tqent;
 524 };
 525
 526 extern int zio_bookmark_compare(const void *, const void *);
 527
 528 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
 529     zio_done_func_t *done, void *private, enum zio_flag flags);
 530
 531 extern zio_t *zio_root(spa_t *spa,
 532     zio_done_func_t *done, void *private, enum zio_flag flags);
 533
 534 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
 535     struct abd *data, uint64_t lsize, zio_done_func_t *done, void *private,
 536     zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
 537
 538 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 539     struct abd *data, uint64_t size, uint64_t psize, const zio_prop_t *zp,
 540     zio_done_func_t *ready, zio_done_func_t *children_ready,
 541     zio_done_func_t *physdone, zio_done_func_t *done,
 542     void *private, zio_priority_t priority, enum zio_flag flags,
 543     const zbookmark_phys_t *zb);
 544
 545 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 546     struct abd *data, uint64_t size, zio_done_func_t *done, void *private,
 547     zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
 548
 549 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
 550     boolean_t nopwrite);
 551
 552 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
 553
 554 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
 555     const blkptr_t *bp,
 556     zio_done_func_t *done, void *private, enum zio_flag flags);
 557
 558 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
 559     zio_done_func_t *done, void *private, enum zio_flag flags);
 560
 561 extern zio_t *zio_trim(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
 562     zio_done_func_t *done, void *private, zio_priority_t priority,
 563     enum zio_flag flags, enum trim_flag trim_flags);
 564
 565 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 566     uint64_t size, struct abd *data, int checksum,
 567     zio_done_func_t *done, void *private, zio_priority_t priority,
 568     enum zio_flag flags, boolean_t labels);
 569
 570 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 571     uint64_t size, struct abd *data, int checksum,
 572     zio_done_func_t *done, void *private, zio_priority_t priority,
 573     enum zio_flag flags, boolean_t labels);
 574
 575 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
 576     const blkptr_t *bp, enum zio_flag flags);
 577
 578 extern int zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg,
 579     blkptr_t *new_bp, uint64_t size, boolean_t *slog);
 580 extern void zio_flush(zio_t *zio, vdev_t *vd);
 581 extern void zio_shrink(zio_t *zio, uint64_t size);
 582
 583 extern int zio_wait(zio_t *zio);
 584 extern void zio_nowait(zio_t *zio);
 585 extern void zio_execute(zio_t *zio);
 586 extern void zio_interrupt(zio_t *zio);
 587 extern void zio_delay_init(zio_t *zio);
 588 extern void zio_delay_interrupt(zio_t *zio);
 589 extern void zio_deadman(zio_t *zio, char *tag);
 590
 591 extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **);
 592 extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **);
 593 extern zio_t *zio_unique_parent(zio_t *cio);
 594 extern void zio_add_child(zio_t *pio, zio_t *cio);
 595
 596 extern void *zio_buf_alloc(size_t size);
 597 extern void zio_buf_free(void *buf, size_t size);
 598 extern void *zio_data_buf_alloc(size_t size);
 599 extern void zio_data_buf_free(void *buf, size_t size);
 600
 601 extern void zio_push_transform(zio_t *zio, struct abd *abd, uint64_t size,
 602     uint64_t bufsize, zio_transform_func_t *transform);
 603 extern void zio_pop_transforms(zio_t *zio);
 604
 605 extern void zio_resubmit_stage_async(void *);
 606
 607 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
 608     uint64_t offset, struct abd *data, uint64_t size, int type,
 609     zio_priority_t priority, enum zio_flag flags,
 610     zio_done_func_t *done, void *private);
 611
 612 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
 613     struct abd *data, uint64_t size, zio_type_t type, zio_priority_t priority,
 614     enum zio_flag flags, zio_done_func_t *done, void *private);
 615
 616 extern void zio_vdev_io_bypass(zio_t *zio);
 617 extern void zio_vdev_io_reissue(zio_t *zio);
 618 extern void zio_vdev_io_redone(zio_t *zio);
 619
 620 extern void zio_change_priority(zio_t *pio, zio_priority_t priority);
 621
 622 extern void zio_checksum_verified(zio_t *zio);
 623 extern int zio_worst_error(int e1, int e2);
 624
 625 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
 626     enum zio_checksum parent);
 627 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
 628     enum zio_checksum child, enum zio_checksum parent);
 629 extern enum zio_compress zio_compress_select(spa_t *spa,
 630     enum zio_compress child, enum zio_compress parent);
 631
 632 extern void zio_suspend(spa_t *spa, zio_t *zio, zio_suspend_reason_t);
 633 extern int zio_resume(spa_t *spa);
 634 extern void zio_resume_wait(spa_t *spa);
 635
 636 /*
 637  * Initial setup and teardown.
 638  */
 639 extern void zio_init(void);
 640 extern void zio_fini(void);
 641
 642 /*
 643  * Fault injection
 644  */
 645 struct zinject_record;
 646 extern uint32_t zio_injection_enabled;
 647 extern int zio_inject_fault(char *name, int flags, int *id,
 648     struct zinject_record *record);
 649 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
 650     struct zinject_record *record);
 651 extern int zio_clear_fault(int id);
 652 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
 653 extern int zio_handle_decrypt_injection(spa_t *spa, const zbookmark_phys_t *zb,
 654     uint64_t type, int error);
 655 extern int zio_handle_fault_injection(zio_t *zio, int error);
 656 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
 657 extern int zio_handle_device_injections(vdev_t *vd, zio_t *zio, int err1,
 658     int err2);
 659 extern int zio_handle_label_injection(zio_t *zio, int error);
 660 extern void zio_handle_ignored_writes(zio_t *zio);
 661 extern hrtime_t zio_handle_io_delay(zio_t *zio);
 662
 663 /*
 664  * Checksum ereport functions
 665  */
 666 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd,
 667     const zbookmark_phys_t *zb, struct zio *zio, uint64_t offset,
 668     uint64_t length, void *arg, struct zio_bad_cksum *info);
 669 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
 670     const abd_t *good_data, const abd_t *bad_data, boolean_t drop_if_identical);
 671
 672 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
 673
 674 /* If we have the good data in hand, this function can be used */
 675 extern int zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
 676     const zbookmark_phys_t *zb, struct zio *zio, uint64_t offset,
 677     uint64_t length, const abd_t *good_data, const abd_t *bad_data,
 678     struct zio_bad_cksum *info);
 679
 680 /* Called from spa_sync(), but primarily an injection handler */
 681 extern void spa_handle_ignored_writes(spa_t *spa);
 682
 683 /* zbookmark_phys functions */
 684 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
 685     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
 686 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
 687     uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
 688
 689 #ifdef  __cplusplus
 690 }
 691 #endif
 692
 693 #endif  /* _ZIO_H */