Print this page
4045 zfs write throttle & i/o scheduler performance work
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>


   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 by Delphix. All rights reserved.
  26  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  27  */
  28 
  29 #ifndef _ZIO_H
  30 #define _ZIO_H
  31 
  32 #include <sys/zfs_context.h>
  33 #include <sys/spa.h>
  34 #include <sys/txg.h>
  35 #include <sys/avl.h>
  36 #include <sys/fs/zfs.h>
  37 #include <sys/zio_impl.h>
  38 
  39 #ifdef  __cplusplus
  40 extern "C" {
  41 #endif
  42 
  43 /*
  44  * Embedded checksum
  45  */


 109         ZIO_COMPRESS_ZLE,
 110         ZIO_COMPRESS_LZ4,
 111         ZIO_COMPRESS_FUNCTIONS
 112 };
 113 
 114 /* N.B. when altering this value, also change BOOTFS_COMPRESS_VALID below */
 115 #define ZIO_COMPRESS_ON_VALUE   ZIO_COMPRESS_LZJB
 116 #define ZIO_COMPRESS_DEFAULT    ZIO_COMPRESS_OFF
 117 
 118 #define BOOTFS_COMPRESS_VALID(compress)                 \
 119         ((compress) == ZIO_COMPRESS_LZJB ||             \
 120         (compress) == ZIO_COMPRESS_LZ4 ||               \
 121         ((compress) == ZIO_COMPRESS_ON &&               \
 122         ZIO_COMPRESS_ON_VALUE == ZIO_COMPRESS_LZJB) ||  \
 123         (compress) == ZIO_COMPRESS_OFF)
 124 
 125 #define ZIO_FAILURE_MODE_WAIT           0
 126 #define ZIO_FAILURE_MODE_CONTINUE       1
 127 #define ZIO_FAILURE_MODE_PANIC          2
 128 
 129 #define ZIO_PRIORITY_NOW                (zio_priority_table[0])
 130 #define ZIO_PRIORITY_SYNC_READ          (zio_priority_table[1])
 131 #define ZIO_PRIORITY_SYNC_WRITE         (zio_priority_table[2])
 132 #define ZIO_PRIORITY_LOG_WRITE          (zio_priority_table[3])
 133 #define ZIO_PRIORITY_CACHE_FILL         (zio_priority_table[4])
 134 #define ZIO_PRIORITY_AGG                (zio_priority_table[5])
 135 #define ZIO_PRIORITY_FREE               (zio_priority_table[6])
 136 #define ZIO_PRIORITY_ASYNC_WRITE        (zio_priority_table[7])
 137 #define ZIO_PRIORITY_ASYNC_READ         (zio_priority_table[8])
 138 #define ZIO_PRIORITY_RESILVER           (zio_priority_table[9])
 139 #define ZIO_PRIORITY_SCRUB              (zio_priority_table[10])
 140 #define ZIO_PRIORITY_DDT_PREFETCH       (zio_priority_table[11])
 141 #define ZIO_PRIORITY_TABLE_SIZE         12
 142 



 143 #define ZIO_PIPELINE_CONTINUE           0x100
 144 #define ZIO_PIPELINE_STOP               0x101
 145 
 146 enum zio_flag {
 147         /*
 148          * Flags inherited by gang, ddt, and vdev children,
 149          * and that must be equal for two zios to aggregate
 150          */
 151         ZIO_FLAG_DONT_AGGREGATE = 1 << 0,
 152         ZIO_FLAG_IO_REPAIR      = 1 << 1,
 153         ZIO_FLAG_SELF_HEAL      = 1 << 2,
 154         ZIO_FLAG_RESILVER       = 1 << 3,
 155         ZIO_FLAG_SCRUB          = 1 << 4,
 156         ZIO_FLAG_SCAN_THREAD    = 1 << 5,
 157 
 158 #define ZIO_FLAG_AGG_INHERIT    (ZIO_FLAG_CANFAIL - 1)
 159 
 160         /*
 161          * Flags inherited by ddt, gang, and vdev children.
 162          */


 177         ZIO_FLAG_IO_RETRY       = 1 << 13,        /* must be first for INHERIT */
 178         ZIO_FLAG_PROBE          = 1 << 14,
 179         ZIO_FLAG_TRYHARD        = 1 << 15,
 180         ZIO_FLAG_OPTIONAL       = 1 << 16,
 181 
 182 #define ZIO_FLAG_VDEV_INHERIT   (ZIO_FLAG_DONT_QUEUE - 1)
 183 
 184         /*
 185          * Flags not inherited by any children.
 186          */
 187         ZIO_FLAG_DONT_QUEUE     = 1 << 17,        /* must be first for INHERIT */
 188         ZIO_FLAG_DONT_PROPAGATE = 1 << 18,
 189         ZIO_FLAG_IO_BYPASS      = 1 << 19,
 190         ZIO_FLAG_IO_REWRITE     = 1 << 20,
 191         ZIO_FLAG_RAW            = 1 << 21,
 192         ZIO_FLAG_GANG_CHILD     = 1 << 22,
 193         ZIO_FLAG_DDT_CHILD      = 1 << 23,
 194         ZIO_FLAG_GODFATHER      = 1 << 24,
 195         ZIO_FLAG_NOPWRITE       = 1 << 25,
 196         ZIO_FLAG_REEXECUTED     = 1 << 26,

 197 };
 198 
 199 #define ZIO_FLAG_MUSTSUCCEED            0
 200 
 201 #define ZIO_DDT_CHILD_FLAGS(zio)                                \
 202         (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |              \
 203         ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
 204 
 205 #define ZIO_GANG_CHILD_FLAGS(zio)                               \
 206         (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |             \
 207         ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
 208 
 209 #define ZIO_VDEV_CHILD_FLAGS(zio)                               \
 210         (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |             \
 211         ZIO_FLAG_CANFAIL)
 212 
 213 enum zio_child {
 214         ZIO_CHILD_VDEV = 0,
 215         ZIO_CHILD_GANG,
 216         ZIO_CHILD_DDT,
 217         ZIO_CHILD_LOGICAL,
 218         ZIO_CHILD_TYPES
 219 };
 220 
 221 enum zio_wait_type {
 222         ZIO_WAIT_READY = 0,
 223         ZIO_WAIT_DONE,
 224         ZIO_WAIT_TYPES
 225 };
 226 
 227 /*
 228  * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent
 229  * graveyard) to indicate checksum errors and fragmentation.
 230  */
 231 #define ECKSUM  EBADE
 232 #define EFRAGS  EBADR
 233 
 234 typedef void zio_done_func_t(zio_t *zio);
 235 
 236 extern uint8_t zio_priority_table[ZIO_PRIORITY_TABLE_SIZE];
 237 extern char *zio_type_name[ZIO_TYPES];
 238 
 239 /*
 240  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
 241  * identifies any block in the pool.  By convention, the meta-objset (MOS)
 242  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
 243  * except root blocks and ZIL blocks, which are defined as follows:
 244  *
 245  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
 246  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
 247  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
 248  *
 249  * Note: this structure is called a bookmark because its original purpose
 250  * was to remember where to resume a pool-wide traverse.
 251  *
 252  * Note: this structure is passed between userland and the kernel.
 253  * Therefore it must not change size or alignment between 32/64 bit
 254  * compilation options.
 255  */
 256 typedef struct zbookmark {
 257         uint64_t        zb_objset;


 357  * the reexecute flags are protected by io_lock, modifiable by children,
 358  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
 359  */
 360 #define ZIO_REEXECUTE_NOW       0x01
 361 #define ZIO_REEXECUTE_SUSPEND   0x02
 362 
 363 typedef struct zio_link {
 364         zio_t           *zl_parent;
 365         zio_t           *zl_child;
 366         list_node_t     zl_parent_node;
 367         list_node_t     zl_child_node;
 368 } zio_link_t;
 369 
 370 struct zio {
 371         /* Core information about this I/O */
 372         zbookmark_t     io_bookmark;
 373         zio_prop_t      io_prop;
 374         zio_type_t      io_type;
 375         enum zio_child  io_child_type;
 376         int             io_cmd;
 377         uint8_t         io_priority;
 378         uint8_t         io_reexecute;
 379         uint8_t         io_state[ZIO_WAIT_TYPES];
 380         uint64_t        io_txg;
 381         spa_t           *io_spa;
 382         blkptr_t        *io_bp;
 383         blkptr_t        *io_bp_override;
 384         blkptr_t        io_bp_copy;
 385         list_t          io_parent_list;
 386         list_t          io_child_list;
 387         zio_link_t      *io_walk_link;
 388         zio_t           *io_logical;
 389         zio_transform_t *io_transform_stack;
 390 
 391         /* Callback info */
 392         zio_done_func_t *io_ready;

 393         zio_done_func_t *io_done;
 394         void            *io_private;
 395         int64_t         io_prev_space_delta;    /* DMU private */
 396         blkptr_t        io_bp_orig;
 397 
 398         /* Data represented by this I/O */
 399         void            *io_data;
 400         void            *io_orig_data;
 401         uint64_t        io_size;
 402         uint64_t        io_orig_size;
 403 
 404         /* Stuff for the vdev stack */
 405         vdev_t          *io_vd;
 406         void            *io_vsd;
 407         const zio_vsd_ops_t *io_vsd_ops;
 408 
 409         uint64_t        io_offset;
 410         uint64_t        io_deadline;
 411         hrtime_t        io_timestamp;
 412         avl_node_t      io_offset_node;
 413         avl_node_t      io_deadline_node;
 414         avl_tree_t      *io_vdev_tree;
 415 
 416         /* Internal pipeline state */
 417         enum zio_flag   io_flags;
 418         enum zio_stage  io_stage;
 419         enum zio_stage  io_pipeline;
 420         enum zio_flag   io_orig_flags;
 421         enum zio_stage  io_orig_stage;
 422         enum zio_stage  io_orig_pipeline;
 423         int             io_error;
 424         int             io_child_error[ZIO_CHILD_TYPES];
 425         uint64_t        io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
 426         uint64_t        io_child_count;

 427         uint64_t        io_parent_count;
 428         uint64_t        *io_stall;
 429         zio_t           *io_gang_leader;
 430         zio_gang_node_t *io_gang_tree;
 431         void            *io_executor;
 432         void            *io_waiter;
 433         kmutex_t        io_lock;
 434         kcondvar_t      io_cv;
 435 
 436         /* FMA state */
 437         zio_cksum_report_t *io_cksum_report;
 438         uint64_t        io_ena;
 439 
 440         /* Taskq dispatching state */
 441         taskq_ent_t     io_tqent;
 442 };
 443 
 444 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
 445     zio_done_func_t *done, void *private, enum zio_flag flags);
 446 
 447 extern zio_t *zio_root(spa_t *spa,
 448     zio_done_func_t *done, void *private, enum zio_flag flags);
 449 
 450 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
 451     uint64_t size, zio_done_func_t *done, void *private,
 452     int priority, enum zio_flag flags, const zbookmark_t *zb);
 453 
 454 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 455     void *data, uint64_t size, const zio_prop_t *zp,
 456     zio_done_func_t *ready, zio_done_func_t *done, void *private,
 457     int priority, enum zio_flag flags, const zbookmark_t *zb);

 458 
 459 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 460     void *data, uint64_t size, zio_done_func_t *done, void *private,
 461     int priority, enum zio_flag flags, zbookmark_t *zb);
 462 
 463 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
 464     boolean_t nopwrite);
 465 
 466 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
 467 
 468 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
 469     const blkptr_t *bp,
 470     zio_done_func_t *done, void *private, enum zio_flag flags);
 471 
 472 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
 473     zio_done_func_t *done, void *private, int priority, enum zio_flag flags);
 474 
 475 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 476     uint64_t size, void *data, int checksum,
 477     zio_done_func_t *done, void *private, int priority, enum zio_flag flags,
 478     boolean_t labels);
 479 
 480 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 481     uint64_t size, void *data, int checksum,
 482     zio_done_func_t *done, void *private, int priority, enum zio_flag flags,
 483     boolean_t labels);
 484 
 485 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
 486     const blkptr_t *bp, enum zio_flag flags);
 487 
 488 extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
 489     blkptr_t *old_bp, uint64_t size, boolean_t use_slog);
 490 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
 491 extern void zio_flush(zio_t *zio, vdev_t *vd);
 492 extern void zio_shrink(zio_t *zio, uint64_t size);
 493 
 494 extern int zio_wait(zio_t *zio);
 495 extern void zio_nowait(zio_t *zio);
 496 extern void zio_execute(zio_t *zio);
 497 extern void zio_interrupt(zio_t *zio);
 498 
 499 extern zio_t *zio_walk_parents(zio_t *cio);
 500 extern zio_t *zio_walk_children(zio_t *pio);
 501 extern zio_t *zio_unique_parent(zio_t *cio);
 502 extern void zio_add_child(zio_t *pio, zio_t *cio);
 503 
 504 extern void *zio_buf_alloc(size_t size);
 505 extern void zio_buf_free(void *buf, size_t size);
 506 extern void *zio_data_buf_alloc(size_t size);
 507 extern void zio_data_buf_free(void *buf, size_t size);
 508 
 509 extern void zio_resubmit_stage_async(void *);
 510 
 511 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
 512     uint64_t offset, void *data, uint64_t size, int type, int priority,
 513     enum zio_flag flags, zio_done_func_t *done, void *private);

 514 
 515 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
 516     void *data, uint64_t size, int type, int priority,
 517     enum zio_flag flags, zio_done_func_t *done, void *private);
 518 
 519 extern void zio_vdev_io_bypass(zio_t *zio);
 520 extern void zio_vdev_io_reissue(zio_t *zio);
 521 extern void zio_vdev_io_redone(zio_t *zio);
 522 
 523 extern void zio_checksum_verified(zio_t *zio);
 524 extern int zio_worst_error(int e1, int e2);
 525 
 526 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
 527     enum zio_checksum parent);
 528 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
 529     enum zio_checksum child, enum zio_checksum parent);
 530 extern enum zio_compress zio_compress_select(enum zio_compress child,
 531     enum zio_compress parent);
 532 
 533 extern void zio_suspend(spa_t *spa, zio_t *zio);
 534 extern int zio_resume(spa_t *spa);
 535 extern void zio_resume_wait(spa_t *spa);
 536 




   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) 2013 by Delphix. All rights reserved.
  26  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
  27  */
  28 
  29 #ifndef _ZIO_H
  30 #define _ZIO_H
  31 
  32 #include <sys/zfs_context.h>
  33 #include <sys/spa.h>
  34 #include <sys/txg.h>
  35 #include <sys/avl.h>
  36 #include <sys/fs/zfs.h>
  37 #include <sys/zio_impl.h>
  38 
  39 #ifdef  __cplusplus
  40 extern "C" {
  41 #endif
  42 
  43 /*
  44  * Embedded checksum
  45  */


 109         ZIO_COMPRESS_ZLE,
 110         ZIO_COMPRESS_LZ4,
 111         ZIO_COMPRESS_FUNCTIONS
 112 };
 113 
 114 /* N.B. when altering this value, also change BOOTFS_COMPRESS_VALID below */
 115 #define ZIO_COMPRESS_ON_VALUE   ZIO_COMPRESS_LZJB
 116 #define ZIO_COMPRESS_DEFAULT    ZIO_COMPRESS_OFF
 117 
 118 #define BOOTFS_COMPRESS_VALID(compress)                 \
 119         ((compress) == ZIO_COMPRESS_LZJB ||             \
 120         (compress) == ZIO_COMPRESS_LZ4 ||               \
 121         ((compress) == ZIO_COMPRESS_ON &&               \
 122         ZIO_COMPRESS_ON_VALUE == ZIO_COMPRESS_LZJB) ||  \
 123         (compress) == ZIO_COMPRESS_OFF)
 124 
 125 #define ZIO_FAILURE_MODE_WAIT           0
 126 #define ZIO_FAILURE_MODE_CONTINUE       1
 127 #define ZIO_FAILURE_MODE_PANIC          2
 128 
 129 typedef enum zio_priority {
 130         ZIO_PRIORITY_SYNC_READ,
 131         ZIO_PRIORITY_SYNC_WRITE,        /* ZIL */
 132         ZIO_PRIORITY_ASYNC_READ,        /* prefetch */
 133         ZIO_PRIORITY_ASYNC_WRITE,       /* spa_sync() */
 134         ZIO_PRIORITY_SCRUB,             /* asynchronous scrub/resilver reads */
 135         ZIO_PRIORITY_NUM_QUEUEABLE,






 136 
 137         ZIO_PRIORITY_NOW                /* non-queued i/os (e.g. free) */
 138 } zio_priority_t;
 139 
 140 #define ZIO_PIPELINE_CONTINUE           0x100
 141 #define ZIO_PIPELINE_STOP               0x101
 142 
 143 enum zio_flag {
 144         /*
 145          * Flags inherited by gang, ddt, and vdev children,
 146          * and that must be equal for two zios to aggregate
 147          */
 148         ZIO_FLAG_DONT_AGGREGATE = 1 << 0,
 149         ZIO_FLAG_IO_REPAIR      = 1 << 1,
 150         ZIO_FLAG_SELF_HEAL      = 1 << 2,
 151         ZIO_FLAG_RESILVER       = 1 << 3,
 152         ZIO_FLAG_SCRUB          = 1 << 4,
 153         ZIO_FLAG_SCAN_THREAD    = 1 << 5,
 154 
 155 #define ZIO_FLAG_AGG_INHERIT    (ZIO_FLAG_CANFAIL - 1)
 156 
 157         /*
 158          * Flags inherited by ddt, gang, and vdev children.
 159          */


 174         ZIO_FLAG_IO_RETRY       = 1 << 13,        /* must be first for INHERIT */
 175         ZIO_FLAG_PROBE          = 1 << 14,
 176         ZIO_FLAG_TRYHARD        = 1 << 15,
 177         ZIO_FLAG_OPTIONAL       = 1 << 16,
 178 
 179 #define ZIO_FLAG_VDEV_INHERIT   (ZIO_FLAG_DONT_QUEUE - 1)
 180 
 181         /*
 182          * Flags not inherited by any children.
 183          */
 184         ZIO_FLAG_DONT_QUEUE     = 1 << 17,        /* must be first for INHERIT */
 185         ZIO_FLAG_DONT_PROPAGATE = 1 << 18,
 186         ZIO_FLAG_IO_BYPASS      = 1 << 19,
 187         ZIO_FLAG_IO_REWRITE     = 1 << 20,
 188         ZIO_FLAG_RAW            = 1 << 21,
 189         ZIO_FLAG_GANG_CHILD     = 1 << 22,
 190         ZIO_FLAG_DDT_CHILD      = 1 << 23,
 191         ZIO_FLAG_GODFATHER      = 1 << 24,
 192         ZIO_FLAG_NOPWRITE       = 1 << 25,
 193         ZIO_FLAG_REEXECUTED     = 1 << 26,
 194         ZIO_FLAG_DELEGATED      = 1 << 27,
 195 };
 196 
 197 #define ZIO_FLAG_MUSTSUCCEED            0
 198 
 199 #define ZIO_DDT_CHILD_FLAGS(zio)                                \
 200         (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |              \
 201         ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
 202 
 203 #define ZIO_GANG_CHILD_FLAGS(zio)                               \
 204         (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |             \
 205         ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
 206 
 207 #define ZIO_VDEV_CHILD_FLAGS(zio)                               \
 208         (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |             \
 209         ZIO_FLAG_CANFAIL)
 210 
 211 enum zio_child {
 212         ZIO_CHILD_VDEV = 0,
 213         ZIO_CHILD_GANG,
 214         ZIO_CHILD_DDT,
 215         ZIO_CHILD_LOGICAL,
 216         ZIO_CHILD_TYPES
 217 };
 218 
 219 enum zio_wait_type {
 220         ZIO_WAIT_READY = 0,
 221         ZIO_WAIT_DONE,
 222         ZIO_WAIT_TYPES
 223 };
 224 
 225 /*
 226  * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent
 227  * graveyard) to indicate checksum errors and fragmentation.
 228  */
 229 #define ECKSUM  EBADE
 230 #define EFRAGS  EBADR
 231 
 232 typedef void zio_done_func_t(zio_t *zio);
 233 
 234 extern const char *zio_type_name[ZIO_TYPES];

 235 
 236 /*
 237  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
 238  * identifies any block in the pool.  By convention, the meta-objset (MOS)
 239  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
 240  * except root blocks and ZIL blocks, which are defined as follows:
 241  *
 242  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
 243  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
 244  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
 245  *
 246  * Note: this structure is called a bookmark because its original purpose
 247  * was to remember where to resume a pool-wide traverse.
 248  *
 249  * Note: this structure is passed between userland and the kernel.
 250  * Therefore it must not change size or alignment between 32/64 bit
 251  * compilation options.
 252  */
 253 typedef struct zbookmark {
 254         uint64_t        zb_objset;


 354  * the reexecute flags are protected by io_lock, modifiable by children,
 355  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
 356  */
 357 #define ZIO_REEXECUTE_NOW       0x01
 358 #define ZIO_REEXECUTE_SUSPEND   0x02
 359 
 360 typedef struct zio_link {
 361         zio_t           *zl_parent;
 362         zio_t           *zl_child;
 363         list_node_t     zl_parent_node;
 364         list_node_t     zl_child_node;
 365 } zio_link_t;
 366 
 367 struct zio {
 368         /* Core information about this I/O */
 369         zbookmark_t     io_bookmark;
 370         zio_prop_t      io_prop;
 371         zio_type_t      io_type;
 372         enum zio_child  io_child_type;
 373         int             io_cmd;
 374         zio_priority_t  io_priority;
 375         uint8_t         io_reexecute;
 376         uint8_t         io_state[ZIO_WAIT_TYPES];
 377         uint64_t        io_txg;
 378         spa_t           *io_spa;
 379         blkptr_t        *io_bp;
 380         blkptr_t        *io_bp_override;
 381         blkptr_t        io_bp_copy;
 382         list_t          io_parent_list;
 383         list_t          io_child_list;
 384         zio_link_t      *io_walk_link;
 385         zio_t           *io_logical;
 386         zio_transform_t *io_transform_stack;
 387 
 388         /* Callback info */
 389         zio_done_func_t *io_ready;
 390         zio_done_func_t *io_physdone;
 391         zio_done_func_t *io_done;
 392         void            *io_private;
 393         int64_t         io_prev_space_delta;    /* DMU private */
 394         blkptr_t        io_bp_orig;
 395 
 396         /* Data represented by this I/O */
 397         void            *io_data;
 398         void            *io_orig_data;
 399         uint64_t        io_size;
 400         uint64_t        io_orig_size;
 401 
 402         /* Stuff for the vdev stack */
 403         vdev_t          *io_vd;
 404         void            *io_vsd;
 405         const zio_vsd_ops_t *io_vsd_ops;
 406 
 407         uint64_t        io_offset;

 408         hrtime_t        io_timestamp;
 409         avl_node_t      io_queue_node;


 410 
 411         /* Internal pipeline state */
 412         enum zio_flag   io_flags;
 413         enum zio_stage  io_stage;
 414         enum zio_stage  io_pipeline;
 415         enum zio_flag   io_orig_flags;
 416         enum zio_stage  io_orig_stage;
 417         enum zio_stage  io_orig_pipeline;
 418         int             io_error;
 419         int             io_child_error[ZIO_CHILD_TYPES];
 420         uint64_t        io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
 421         uint64_t        io_child_count;
 422         uint64_t        io_phys_children;
 423         uint64_t        io_parent_count;
 424         uint64_t        *io_stall;
 425         zio_t           *io_gang_leader;
 426         zio_gang_node_t *io_gang_tree;
 427         void            *io_executor;
 428         void            *io_waiter;
 429         kmutex_t        io_lock;
 430         kcondvar_t      io_cv;
 431 
 432         /* FMA state */
 433         zio_cksum_report_t *io_cksum_report;
 434         uint64_t        io_ena;
 435 
 436         /* Taskq dispatching state */
 437         taskq_ent_t     io_tqent;
 438 };
 439 
 440 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
 441     zio_done_func_t *done, void *private, enum zio_flag flags);
 442 
 443 extern zio_t *zio_root(spa_t *spa,
 444     zio_done_func_t *done, void *private, enum zio_flag flags);
 445 
 446 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
 447     uint64_t size, zio_done_func_t *done, void *private,
 448     zio_priority_t priority, enum zio_flag flags, const zbookmark_t *zb);
 449 
 450 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 451     void *data, uint64_t size, const zio_prop_t *zp,
 452     zio_done_func_t *ready, zio_done_func_t *physdone, zio_done_func_t *done,
 453     void *private,
 454     zio_priority_t priority, enum zio_flag flags, const zbookmark_t *zb);
 455 
 456 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
 457     void *data, uint64_t size, zio_done_func_t *done, void *private,
 458     zio_priority_t priority, enum zio_flag flags, zbookmark_t *zb);
 459 
 460 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
 461     boolean_t nopwrite);
 462 
 463 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
 464 
 465 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
 466     const blkptr_t *bp,
 467     zio_done_func_t *done, void *private, enum zio_flag flags);
 468 
 469 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
 470     zio_done_func_t *done, void *private, enum zio_flag flags);
 471 
 472 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 473     uint64_t size, void *data, int checksum,
 474     zio_done_func_t *done, void *private, zio_priority_t priority,
 475     enum zio_flag flags, boolean_t labels);
 476 
 477 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
 478     uint64_t size, void *data, int checksum,
 479     zio_done_func_t *done, void *private, zio_priority_t priority,
 480     enum zio_flag flags, boolean_t labels);
 481 
 482 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
 483     const blkptr_t *bp, enum zio_flag flags);
 484 
 485 extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
 486     blkptr_t *old_bp, uint64_t size, boolean_t use_slog);
 487 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
 488 extern void zio_flush(zio_t *zio, vdev_t *vd);
 489 extern void zio_shrink(zio_t *zio, uint64_t size);
 490 
 491 extern int zio_wait(zio_t *zio);
 492 extern void zio_nowait(zio_t *zio);
 493 extern void zio_execute(zio_t *zio);
 494 extern void zio_interrupt(zio_t *zio);
 495 
 496 extern zio_t *zio_walk_parents(zio_t *cio);
 497 extern zio_t *zio_walk_children(zio_t *pio);
 498 extern zio_t *zio_unique_parent(zio_t *cio);
 499 extern void zio_add_child(zio_t *pio, zio_t *cio);
 500 
 501 extern void *zio_buf_alloc(size_t size);
 502 extern void zio_buf_free(void *buf, size_t size);
 503 extern void *zio_data_buf_alloc(size_t size);
 504 extern void zio_data_buf_free(void *buf, size_t size);
 505 
 506 extern void zio_resubmit_stage_async(void *);
 507 
 508 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
 509     uint64_t offset, void *data, uint64_t size, int type,
 510     zio_priority_t priority, enum zio_flag flags,
 511     zio_done_func_t *done, void *private);
 512 
 513 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
 514     void *data, uint64_t size, int type, zio_priority_t priority,
 515     enum zio_flag flags, zio_done_func_t *done, void *private);
 516 
 517 extern void zio_vdev_io_bypass(zio_t *zio);
 518 extern void zio_vdev_io_reissue(zio_t *zio);
 519 extern void zio_vdev_io_redone(zio_t *zio);
 520 
 521 extern void zio_checksum_verified(zio_t *zio);
 522 extern int zio_worst_error(int e1, int e2);
 523 
 524 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
 525     enum zio_checksum parent);
 526 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
 527     enum zio_checksum child, enum zio_checksum parent);
 528 extern enum zio_compress zio_compress_select(enum zio_compress child,
 529     enum zio_compress parent);
 530 
 531 extern void zio_suspend(spa_t *spa, zio_t *zio);
 532 extern int zio_resume(spa_t *spa);
 533 extern void zio_resume_wait(spa_t *spa);
 534