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, 2016 by Delphix. All rights reserved.
26  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
27  * Copyright 2016 Toomas Soome <tsoome@me.com>
28  */
29 
30 #ifndef _ZIO_H
31 #define   _ZIO_H
32 
33 #include <sys/zio_priority.h>
34 #include <sys/zfs_context.h>
35 #include <sys/spa.h>
36 #include <sys/txg.h>
37 #include <sys/avl.h>
38 #include <sys/kstat.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 #ifndef __NetBSD__
87           ZIO_CHECKSUM_SHA512,
88           ZIO_CHECKSUM_SKEIN,
89 #endif
90 #ifdef illumos
91           ZIO_CHECKSUM_EDONR,
92 #endif
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 (1 << 8)
107 
108 #define   ZIO_DEDUPCHECKSUM   ZIO_CHECKSUM_SHA256
109 #define   ZIO_DEDUPDITTO_MIN  100
110 
111 enum zio_compress {
112           ZIO_COMPRESS_INHERIT = 0,
113           ZIO_COMPRESS_ON,
114           ZIO_COMPRESS_OFF,
115           ZIO_COMPRESS_LZJB,
116           ZIO_COMPRESS_EMPTY,
117           ZIO_COMPRESS_GZIP_1,
118           ZIO_COMPRESS_GZIP_2,
119           ZIO_COMPRESS_GZIP_3,
120           ZIO_COMPRESS_GZIP_4,
121           ZIO_COMPRESS_GZIP_5,
122           ZIO_COMPRESS_GZIP_6,
123           ZIO_COMPRESS_GZIP_7,
124           ZIO_COMPRESS_GZIP_8,
125           ZIO_COMPRESS_GZIP_9,
126           ZIO_COMPRESS_ZLE,
127           ZIO_COMPRESS_LZ4,
128           ZIO_COMPRESS_FUNCTIONS
129 };
130 
131 /*
132  * The number of "legacy" compression functions which can be set on individual
133  * objects.
134  */
135 #define   ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
136 
137 /*
138  * The meaning of "compress = on" selected by the compression features enabled
139  * on a given pool.
140  */
141 #define   ZIO_COMPRESS_LEGACY_ON_VALUE  ZIO_COMPRESS_LZJB
142 #define   ZIO_COMPRESS_LZ4_ON_VALUE     ZIO_COMPRESS_LZ4
143 
144 #define   ZIO_COMPRESS_DEFAULT                    ZIO_COMPRESS_OFF
145 
146 #define   BOOTFS_COMPRESS_VALID(compress)                             \
147           ((compress) == ZIO_COMPRESS_LZJB ||               \
148           (compress) == ZIO_COMPRESS_LZ4 ||                 \
149           (compress) == ZIO_COMPRESS_ON ||                  \
150           (compress) == ZIO_COMPRESS_OFF)
151 
152 #define   ZIO_FAILURE_MODE_WAIT                   0
153 #define   ZIO_FAILURE_MODE_CONTINUE     1
154 #define   ZIO_FAILURE_MODE_PANIC                  2
155 
156 enum zio_flag {
157           /*
158            * Flags inherited by gang, ddt, and vdev children,
159            * and that must be equal for two zios to aggregate
160            */
161           ZIO_FLAG_DONT_AGGREGATE       = 1 << 0,
162           ZIO_FLAG_IO_REPAIR  = 1 << 1,
163           ZIO_FLAG_SELF_HEAL  = 1 << 2,
164           ZIO_FLAG_RESILVER   = 1 << 3,
165           ZIO_FLAG_SCRUB                = 1 << 4,
166           ZIO_FLAG_SCAN_THREAD          = 1 << 5,
167           ZIO_FLAG_PHYSICAL   = 1 << 6,
168 
169 #define   ZIO_FLAG_AGG_INHERIT          (ZIO_FLAG_CANFAIL - 1)
170 
171           /*
172            * Flags inherited by ddt, gang, and vdev children.
173            */
174           ZIO_FLAG_CANFAIL    = 1 << 7, /* must be first for INHERIT */
175           ZIO_FLAG_SPECULATIVE          = 1 << 8,
176           ZIO_FLAG_CONFIG_WRITER        = 1 << 9,
177           ZIO_FLAG_DONT_RETRY = 1 << 10,
178           ZIO_FLAG_DONT_CACHE = 1 << 11,
179           ZIO_FLAG_NODATA               = 1 << 12,
180           ZIO_FLAG_INDUCE_DAMAGE        = 1 << 13,
181           ZIO_FLAG_IO_ALLOCATING        = 1 << 14,
182 
183 #define   ZIO_FLAG_DDT_INHERIT          (ZIO_FLAG_IO_RETRY - 1)
184 #define   ZIO_FLAG_GANG_INHERIT         (ZIO_FLAG_IO_RETRY - 1)
185 
186           /*
187            * Flags inherited by vdev children.
188            */
189           ZIO_FLAG_IO_RETRY   = 1 << 15,          /* must be first for INHERIT */
190           ZIO_FLAG_PROBE                = 1 << 16,
191           ZIO_FLAG_TRYHARD    = 1 << 17,
192           ZIO_FLAG_OPTIONAL   = 1 << 18,
193 
194 #define   ZIO_FLAG_VDEV_INHERIT         (ZIO_FLAG_DONT_QUEUE - 1)
195 
196           /*
197            * Flags not inherited by any children.
198            */
199           ZIO_FLAG_DONT_QUEUE = 1 << 19,          /* must be first for INHERIT */
200           ZIO_FLAG_DONT_PROPAGATE       = 1 << 20,
201           ZIO_FLAG_IO_BYPASS  = 1 << 21,
202           ZIO_FLAG_IO_REWRITE = 1 << 22,
203           ZIO_FLAG_RAW                  = 1 << 23,
204           ZIO_FLAG_GANG_CHILD = 1 << 24,
205           ZIO_FLAG_DDT_CHILD  = 1 << 25,
206           ZIO_FLAG_GODFATHER  = 1 << 26,
207           ZIO_FLAG_NOPWRITE   = 1 << 27,
208           ZIO_FLAG_REEXECUTED = 1 << 28,
209           ZIO_FLAG_DELEGATED  = 1 << 29,
210 };
211 
212 #define   ZIO_FLAG_MUSTSUCCEED                    0
213 
214 #define   ZIO_DDT_CHILD_FLAGS(zio)                                    \
215           (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |                 \
216           ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
217 
218 #define   ZIO_GANG_CHILD_FLAGS(zio)                                   \
219           (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |                \
220           ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
221 
222 #define   ZIO_VDEV_CHILD_FLAGS(zio)                                   \
223           (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |                \
224           ZIO_FLAG_CANFAIL)
225 
226 enum zio_child {
227           ZIO_CHILD_VDEV = 0,
228           ZIO_CHILD_GANG,
229           ZIO_CHILD_DDT,
230           ZIO_CHILD_LOGICAL,
231           ZIO_CHILD_TYPES
232 };
233 
234 enum zio_wait_type {
235           ZIO_WAIT_READY = 0,
236           ZIO_WAIT_DONE,
237           ZIO_WAIT_TYPES
238 };
239 
240 /*
241  * We'll take the number 122 and 123 to indicate checksum errors and
242  * fragmentation. Those doesn't collide with any errno values as they
243  * are greater than ELAST.
244  */
245 #define   ECKSUM    122
246 #define   EFRAGS    123
247 
248 typedef void zio_done_func_t(zio_t *zio);
249 
250 extern boolean_t zio_dva_throttle_enabled;
251 extern const char *zio_type_name[ZIO_TYPES];
252 
253 /*
254  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
255  * identifies any block in the pool.  By convention, the meta-objset (MOS)
256  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
257  * except root blocks and ZIL blocks, which are defined as follows:
258  *
259  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
260  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
261  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
262  * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
263  *
264  * Note: this structure is called a bookmark because its original purpose
265  * was to remember where to resume a pool-wide traverse.
266  *
267  * Note: this structure is passed between userland and the kernel, and is
268  * stored on disk (by virtue of being incorporated into other on-disk
269  * structures, e.g. dsl_scan_phys_t).
270  */
271 typedef struct zbookmark_phys {
272           uint64_t  zb_objset;
273           uint64_t  zb_object;
274           int64_t             zb_level;
275           uint64_t  zb_blkid;
276 } zbookmark_phys_t;
277 
278 #define   SET_BOOKMARK(zb, objset, object, level, blkid)  \
279 {                                                       \
280           (zb)->zb_objset = objset;                       \
281           (zb)->zb_object = object;                       \
282           (zb)->zb_level = level;                         \
283           (zb)->zb_blkid = blkid;                         \
284 }
285 
286 #define   ZB_DESTROYED_OBJSET (-1ULL)
287 
288 #define   ZB_ROOT_OBJECT                (0ULL)
289 #define   ZB_ROOT_LEVEL                 (-1LL)
290 #define   ZB_ROOT_BLKID                 (0ULL)
291 
292 #define   ZB_ZIL_OBJECT                 (0ULL)
293 #define   ZB_ZIL_LEVEL                  (-2LL)
294 
295 #define   ZB_DNODE_LEVEL                (-3LL)
296 #define   ZB_DNODE_BLKID                (0ULL)
297 
298 #define   ZB_IS_ZERO(zb)                                                        \
299           ((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&  \
300           (zb)->zb_level == 0 && (zb)->zb_blkid == 0)
301 #define   ZB_IS_ROOT(zb)                                    \
302           ((zb)->zb_object == ZB_ROOT_OBJECT &&   \
303           (zb)->zb_level == ZB_ROOT_LEVEL &&      \
304           (zb)->zb_blkid == ZB_ROOT_BLKID)
305 
306 typedef struct zio_prop {
307           enum zio_checksum   zp_checksum;
308           enum zio_compress   zp_compress;
309           dmu_object_type_t   zp_type;
310           uint8_t                       zp_level;
311           uint8_t                       zp_copies;
312           boolean_t           zp_dedup;
313           boolean_t           zp_dedup_verify;
314           boolean_t           zp_nopwrite;
315 } zio_prop_t;
316 
317 typedef struct zio_cksum_report zio_cksum_report_t;
318 
319 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
320     const void *good_data);
321 typedef void zio_cksum_free_f(void *cbdata, size_t size);
322 
323 struct zio_bad_cksum;                                       /* defined in zio_checksum.h */
324 struct dnode_phys;
325 
326 struct zio_cksum_report {
327           struct zio_cksum_report *zcr_next;
328           nvlist_t            *zcr_ereport;
329           nvlist_t            *zcr_detector;
330           void                          *zcr_cbdata;
331           size_t                        zcr_cbinfo;         /* passed to zcr_free() */
332           uint64_t            zcr_align;
333           uint64_t            zcr_length;
334           zio_cksum_finish_f  *zcr_finish;
335           zio_cksum_free_f    *zcr_free;
336 
337           /* internal use only */
338           struct zio_bad_cksum          *zcr_ckinfo;        /* information from failure */
339 };
340 
341 typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
342     void *arg);
343 
344 zio_vsd_cksum_report_f        zio_vsd_default_cksum_report;
345 
346 typedef struct zio_vsd_ops {
347           zio_done_func_t               *vsd_free;
348           zio_vsd_cksum_report_f        *vsd_cksum_report;
349 } zio_vsd_ops_t;
350 
351 typedef struct zio_gang_node {
352           zio_gbh_phys_t                *gn_gbh;
353           struct zio_gang_node          *gn_child[SPA_GBH_NBLKPTRS];
354 } zio_gang_node_t;
355 
356 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
357     zio_gang_node_t *gn, void *data);
358 
359 typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size);
360 
361 typedef struct zio_transform {
362           void                          *zt_orig_data;
363           uint64_t            zt_orig_size;
364           uint64_t            zt_bufsize;
365           zio_transform_func_t          *zt_transform;
366           struct zio_transform          *zt_next;
367 } zio_transform_t;
368 
369 typedef int zio_pipe_stage_t(zio_t *zio);
370 
371 /*
372  * The io_reexecute flags are distinct from io_flags because the child must
373  * be able to propagate them to the parent.  The normal io_flags are local
374  * to the zio, not protected by any lock, and not modifiable by children;
375  * the reexecute flags are protected by io_lock, modifiable by children,
376  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
377  */
378 #define   ZIO_REEXECUTE_NOW   0x01
379 #define   ZIO_REEXECUTE_SUSPEND         0x02
380 
381 typedef struct zio_alloc_list {
382           list_t  zal_list;
383           uint64_t zal_size;
384 } zio_alloc_list_t;
385 
386 typedef struct zio_link {
387           zio_t               *zl_parent;
388           zio_t               *zl_child;
389           list_node_t         zl_parent_node;
390           list_node_t         zl_child_node;
391 } zio_link_t;
392 
393 /*
394  * Used for TRIM kstat.
395  */
396 typedef struct zio_trim_stats {
397           /*
398            * Number of bytes successfully TRIMmed.
399            */
400           kstat_named_t bytes;
401 
402           /*
403            * Number of successful TRIM requests.
404            */
405           kstat_named_t success;
406 
407           /*
408            * Number of TRIM requests that failed because TRIM is not
409            * supported.
410            */
411           kstat_named_t unsupported;
412 
413           /*
414            * Number of TRIM requests that failed for other reasons.
415            */
416           kstat_named_t failed;
417 } zio_trim_stats_t;
418 
419 extern zio_trim_stats_t zio_trim_stats;
420 
421 #define ZIO_TRIM_STAT_INCR(stat, val) \
422           atomic_add_64(&zio_trim_stats.stat.value.ui64, (val));
423 #define ZIO_TRIM_STAT_BUMP(stat) \
424           ZIO_TRIM_STAT_INCR(stat, 1);
425 
426 struct zio {
427           /* Core information about this I/O */
428           zbookmark_phys_t    io_bookmark;
429           zio_prop_t          io_prop;
430           zio_type_t          io_type;
431           enum zio_child      io_child_type;
432           int                 io_cmd;
433           zio_priority_t      io_priority;
434           uint8_t             io_reexecute;
435           uint8_t             io_state[ZIO_WAIT_TYPES];
436           uint64_t  io_txg;
437           spa_t               *io_spa;
438           blkptr_t  *io_bp;
439           blkptr_t  *io_bp_override;
440           blkptr_t  io_bp_copy;
441           list_t              io_parent_list;
442           list_t              io_child_list;
443           zio_t               *io_logical;
444           zio_transform_t *io_transform_stack;
445 
446           /* Callback info */
447           zio_done_func_t     *io_ready;
448           zio_done_func_t     *io_children_ready;
449           zio_done_func_t     *io_physdone;
450           zio_done_func_t     *io_done;
451           void                *io_private;
452           int64_t             io_prev_space_delta;          /* DMU private */
453           blkptr_t  io_bp_orig;
454 
455           /* Data represented by this I/O */
456           void                *io_data;
457           void                *io_orig_data;
458           uint64_t  io_size;
459           uint64_t  io_orig_size;
460 
461           /* Stuff for the vdev stack */
462           vdev_t              *io_vd;
463           void                *io_vsd;
464           const zio_vsd_ops_t *io_vsd_ops;
465 
466           uint64_t  io_offset;
467           hrtime_t  io_timestamp;
468           hrtime_t  io_queued_timestamp;
469           hrtime_t  io_target_timestamp;
470           avl_node_t          io_queue_node;
471           avl_node_t          io_offset_node;
472           avl_node_t          io_alloc_node;
473           zio_alloc_list_t    io_alloc_list;
474 
475           /* Internal pipeline state */
476           enum zio_flag       io_flags;
477           enum zio_stage      io_stage;
478           enum zio_stage      io_pipeline;
479           enum zio_flag       io_orig_flags;
480           enum zio_stage      io_orig_stage;
481           enum zio_stage      io_orig_pipeline;
482           enum zio_stage      io_pipeline_trace;
483           int                 io_error;
484           int                 io_child_error[ZIO_CHILD_TYPES];
485           uint64_t  io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
486           uint64_t  io_child_count;
487           uint64_t  io_phys_children;
488           uint64_t  io_parent_count;
489           uint64_t  *io_stall;
490           zio_t               *io_gang_leader;
491           zio_gang_node_t     *io_gang_tree;
492           void                *io_executor;
493           void                *io_waiter;
494           kmutex_t  io_lock;
495           kcondvar_t          io_cv;
496 
497           /* FMA state */
498           zio_cksum_report_t *io_cksum_report;
499           uint64_t  io_ena;
500 
501           /* Taskq dispatching state */
502           taskq_ent_t         io_tqent;
503 
504           avl_node_t          io_trim_node;
505           list_node_t         io_trim_link;
506 };
507 
508 extern int zio_timestamp_compare(const void *, const void *);
509 
510 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
511     zio_done_func_t *done, void *priv, enum zio_flag flags);
512 
513 extern zio_t *zio_root(spa_t *spa,
514     zio_done_func_t *done, void *priv, enum zio_flag flags);
515 
516 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
517     uint64_t size, zio_done_func_t *done, void *priv,
518     zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
519 
520 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
521     void *data, uint64_t size, const zio_prop_t *zp,
522     zio_done_func_t *ready, zio_done_func_t *children_ready,
523     zio_done_func_t *physdone, zio_done_func_t *done,
524     void *priv, zio_priority_t priority, enum zio_flag flags,
525     const zbookmark_phys_t *zb);
526 
527 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
528     void *data, uint64_t size, zio_done_func_t *done, void *priv,
529     zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
530 
531 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
532     boolean_t nopwrite);
533 
534 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
535 
536 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
537     const blkptr_t *bp,
538     zio_done_func_t *done, void *priv, enum zio_flag flags);
539 
540 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
541     uint64_t offset, uint64_t size, zio_done_func_t *done, void *priv,
542     zio_priority_t priority, enum zio_flag flags);
543 
544 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
545     uint64_t size, void *data, int checksum,
546     zio_done_func_t *done, void *priv, zio_priority_t priority,
547     enum zio_flag flags, boolean_t labels);
548 
549 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
550     uint64_t size, void *data, int checksum,
551     zio_done_func_t *done, void *priv, zio_priority_t priority,
552     enum zio_flag flags, boolean_t labels);
553 
554 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
555     const blkptr_t *bp, uint64_t size, enum zio_flag flags);
556 
557 extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
558     blkptr_t *old_bp, uint64_t size, boolean_t *slog);
559 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
560 extern void zio_flush(zio_t *zio, vdev_t *vd);
561 extern zio_t *zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset,
562     uint64_t size);
563 extern void zio_shrink(zio_t *zio, uint64_t size);
564 
565 extern int zio_wait(zio_t *zio);
566 extern void zio_nowait(zio_t *zio);
567 extern void zio_execute(zio_t *zio);
568 extern void zio_interrupt(zio_t *zio);
569 extern void zio_delay_init(zio_t *zio);
570 extern void zio_delay_interrupt(zio_t *zio);
571 
572 extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **);
573 extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **);
574 extern zio_t *zio_unique_parent(zio_t *cio);
575 extern void zio_add_child(zio_t *pio, zio_t *cio);
576 
577 extern void *zio_buf_alloc(size_t size);
578 extern void *zio_buf_alloc_nowait(size_t size);
579 extern void zio_buf_free(void *buf, size_t size);
580 extern void *zio_data_buf_alloc(size_t size);
581 extern void zio_data_buf_free(void *buf, size_t size);
582 
583 extern void zio_push_transform(zio_t *zio, void *data, uint64_t size,
584     uint64_t bufsize, zio_transform_func_t *transform);
585 extern void zio_pop_transforms(zio_t *zio);
586 
587 extern void zio_resubmit_stage_async(void *);
588 
589 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
590     uint64_t offset, void *data, uint64_t size, int type,
591     zio_priority_t priority, enum zio_flag flags,
592     zio_done_func_t *done, void *priv);
593 
594 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
595     void *data, uint64_t size, int type, zio_priority_t priority,
596     enum zio_flag flags, zio_done_func_t *done, void *priv);
597 
598 extern void zio_vdev_io_bypass(zio_t *zio);
599 extern void zio_vdev_io_reissue(zio_t *zio);
600 extern void zio_vdev_io_redone(zio_t *zio);
601 
602 extern void zio_checksum_verified(zio_t *zio);
603 extern int zio_worst_error(int e1, int e2);
604 
605 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
606     enum zio_checksum parent);
607 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
608     enum zio_checksum child, enum zio_checksum parent);
609 extern enum zio_compress zio_compress_select(spa_t *spa,
610     enum zio_compress child, enum zio_compress parent);
611 
612 extern void zio_suspend(spa_t *spa, zio_t *zio);
613 extern int zio_resume(spa_t *spa);
614 extern void zio_resume_wait(spa_t *spa);
615 
616 /*
617  * Initial setup and teardown.
618  */
619 extern void zio_init(void);
620 extern void zio_fini(void);
621 
622 /*
623  * Fault injection
624  */
625 struct zinject_record;
626 extern uint32_t zio_injection_enabled;
627 extern int zio_inject_fault(char *name, int flags, int *id,
628     struct zinject_record *record);
629 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
630     struct zinject_record *record);
631 extern int zio_clear_fault(int id);
632 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
633 extern int zio_handle_fault_injection(zio_t *zio, int error);
634 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
635 extern int zio_handle_label_injection(zio_t *zio, int error);
636 extern void zio_handle_ignored_writes(zio_t *zio);
637 extern hrtime_t zio_handle_io_delay(zio_t *zio);
638 
639 /*
640  * Checksum ereport functions
641  */
642 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
643     uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
644 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
645     const void *good_data, const void *bad_data, boolean_t drop_if_identical);
646 
647 extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
648 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
649 
650 /* If we have the good data in hand, this function can be used */
651 extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
652     struct zio *zio, uint64_t offset, uint64_t length,
653     const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
654 
655 /* Called from spa_sync(), but primarily an injection handler */
656 extern void spa_handle_ignored_writes(spa_t *spa);
657 
658 /* zbookmark_phys functions */
659 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
660     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
661 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
662     uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
663 
664 #ifdef    __cplusplus
665 }
666 #endif
667 
668 #endif    /* _ZIO_H */
669