xref: /NextBSD/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/sys/zio.h (revision 84d351007654069f9643c8e4b4802a7f5f08ee42)
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, 2014 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/zio_priority.h>
33 #include <sys/zfs_context.h>
34 #include <sys/spa.h>
35 #include <sys/txg.h>
36 #include <sys/avl.h>
37 #include <sys/kstat.h>
38 #include <sys/fs/zfs.h>
39 #include <sys/zio_impl.h>
40 
41 #ifdef	__cplusplus
42 extern "C" {
43 #endif
44 
45 /*
46  * Embedded checksum
47  */
48 #define	ZEC_MAGIC	0x210da7ab10c7a11ULL
49 
50 typedef struct zio_eck {
51 	uint64_t	zec_magic;	/* for validation, endianness	*/
52 	zio_cksum_t	zec_cksum;	/* 256-bit checksum		*/
53 } zio_eck_t;
54 
55 /*
56  * Gang block headers are self-checksumming and contain an array
57  * of block pointers.
58  */
59 #define	SPA_GANGBLOCKSIZE	SPA_MINBLOCKSIZE
60 #define	SPA_GBH_NBLKPTRS	((SPA_GANGBLOCKSIZE - \
61 	sizeof (zio_eck_t)) / sizeof (blkptr_t))
62 #define	SPA_GBH_FILLER		((SPA_GANGBLOCKSIZE - \
63 	sizeof (zio_eck_t) - \
64 	(SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\
65 	sizeof (uint64_t))
66 
67 typedef struct zio_gbh {
68 	blkptr_t		zg_blkptr[SPA_GBH_NBLKPTRS];
69 	uint64_t		zg_filler[SPA_GBH_FILLER];
70 	zio_eck_t		zg_tail;
71 } zio_gbh_phys_t;
72 
73 enum zio_checksum {
74 	ZIO_CHECKSUM_INHERIT = 0,
75 	ZIO_CHECKSUM_ON,
76 	ZIO_CHECKSUM_OFF,
77 	ZIO_CHECKSUM_LABEL,
78 	ZIO_CHECKSUM_GANG_HEADER,
79 	ZIO_CHECKSUM_ZILOG,
80 	ZIO_CHECKSUM_FLETCHER_2,
81 	ZIO_CHECKSUM_FLETCHER_4,
82 	ZIO_CHECKSUM_SHA256,
83 	ZIO_CHECKSUM_ZILOG2,
84 	ZIO_CHECKSUM_NOPARITY,
85 #ifdef illumos
86 	ZIO_CHECKSUM_SHA512,
87 	ZIO_CHECKSUM_SKEIN,
88 	ZIO_CHECKSUM_EDONR,
89 #endif
90 	ZIO_CHECKSUM_FUNCTIONS
91 };
92 
93 /*
94  * The number of "legacy" compression functions which can be set on individual
95  * objects.
96  */
97 #define	ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2
98 
99 #define	ZIO_CHECKSUM_ON_VALUE	ZIO_CHECKSUM_FLETCHER_4
100 #define	ZIO_CHECKSUM_DEFAULT	ZIO_CHECKSUM_ON
101 
102 #define	ZIO_CHECKSUM_MASK	0xffULL
103 #define	ZIO_CHECKSUM_VERIFY	(1 << 8)
104 
105 #define	ZIO_DEDUPCHECKSUM	ZIO_CHECKSUM_SHA256
106 #define	ZIO_DEDUPDITTO_MIN	100
107 
108 enum zio_compress {
109 	ZIO_COMPRESS_INHERIT = 0,
110 	ZIO_COMPRESS_ON,
111 	ZIO_COMPRESS_OFF,
112 	ZIO_COMPRESS_LZJB,
113 	ZIO_COMPRESS_EMPTY,
114 	ZIO_COMPRESS_GZIP_1,
115 	ZIO_COMPRESS_GZIP_2,
116 	ZIO_COMPRESS_GZIP_3,
117 	ZIO_COMPRESS_GZIP_4,
118 	ZIO_COMPRESS_GZIP_5,
119 	ZIO_COMPRESS_GZIP_6,
120 	ZIO_COMPRESS_GZIP_7,
121 	ZIO_COMPRESS_GZIP_8,
122 	ZIO_COMPRESS_GZIP_9,
123 	ZIO_COMPRESS_ZLE,
124 	ZIO_COMPRESS_LZ4,
125 	ZIO_COMPRESS_FUNCTIONS
126 };
127 
128 /*
129  * The number of "legacy" compression functions which can be set on individual
130  * objects.
131  */
132 #define	ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
133 
134 /*
135  * The meaning of "compress = on" selected by the compression features enabled
136  * on a given pool.
137  */
138 #define	ZIO_COMPRESS_LEGACY_ON_VALUE	ZIO_COMPRESS_LZJB
139 #define	ZIO_COMPRESS_LZ4_ON_VALUE	ZIO_COMPRESS_LZ4
140 
141 #define	ZIO_COMPRESS_DEFAULT		ZIO_COMPRESS_OFF
142 
143 #define	BOOTFS_COMPRESS_VALID(compress)			\
144 	((compress) == ZIO_COMPRESS_LZJB ||		\
145 	(compress) == ZIO_COMPRESS_LZ4 ||		\
146 	(compress) == ZIO_COMPRESS_ON ||		\
147 	(compress) == ZIO_COMPRESS_OFF)
148 
149 #define	ZIO_FAILURE_MODE_WAIT		0
150 #define	ZIO_FAILURE_MODE_CONTINUE	1
151 #define	ZIO_FAILURE_MODE_PANIC		2
152 
153 enum zio_flag {
154 	/*
155 	 * Flags inherited by gang, ddt, and vdev children,
156 	 * and that must be equal for two zios to aggregate
157 	 */
158 	ZIO_FLAG_DONT_AGGREGATE	= 1 << 0,
159 	ZIO_FLAG_IO_REPAIR	= 1 << 1,
160 	ZIO_FLAG_SELF_HEAL	= 1 << 2,
161 	ZIO_FLAG_RESILVER	= 1 << 3,
162 	ZIO_FLAG_SCRUB		= 1 << 4,
163 	ZIO_FLAG_SCAN_THREAD	= 1 << 5,
164 	ZIO_FLAG_PHYSICAL	= 1 << 6,
165 
166 #define	ZIO_FLAG_AGG_INHERIT	(ZIO_FLAG_CANFAIL - 1)
167 
168 	/*
169 	 * Flags inherited by ddt, gang, and vdev children.
170 	 */
171 	ZIO_FLAG_CANFAIL	= 1 << 7,	/* must be first for INHERIT */
172 	ZIO_FLAG_SPECULATIVE	= 1 << 8,
173 	ZIO_FLAG_CONFIG_WRITER	= 1 << 9,
174 	ZIO_FLAG_DONT_RETRY	= 1 << 10,
175 	ZIO_FLAG_DONT_CACHE	= 1 << 11,
176 	ZIO_FLAG_NODATA		= 1 << 12,
177 	ZIO_FLAG_INDUCE_DAMAGE	= 1 << 13,
178 
179 #define	ZIO_FLAG_DDT_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
180 #define	ZIO_FLAG_GANG_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
181 
182 	/*
183 	 * Flags inherited by vdev children.
184 	 */
185 	ZIO_FLAG_IO_RETRY	= 1 << 14,	/* must be first for INHERIT */
186 	ZIO_FLAG_PROBE		= 1 << 15,
187 	ZIO_FLAG_TRYHARD	= 1 << 16,
188 	ZIO_FLAG_OPTIONAL	= 1 << 17,
189 
190 #define	ZIO_FLAG_VDEV_INHERIT	(ZIO_FLAG_DONT_QUEUE - 1)
191 
192 	/*
193 	 * Flags not inherited by any children.
194 	 */
195 	ZIO_FLAG_DONT_QUEUE	= 1 << 18,	/* must be first for INHERIT */
196 	ZIO_FLAG_DONT_PROPAGATE	= 1 << 19,
197 	ZIO_FLAG_IO_BYPASS	= 1 << 20,
198 	ZIO_FLAG_IO_REWRITE	= 1 << 21,
199 	ZIO_FLAG_RAW		= 1 << 22,
200 	ZIO_FLAG_GANG_CHILD	= 1 << 23,
201 	ZIO_FLAG_DDT_CHILD	= 1 << 24,
202 	ZIO_FLAG_GODFATHER	= 1 << 25,
203 	ZIO_FLAG_NOPWRITE	= 1 << 26,
204 	ZIO_FLAG_REEXECUTED	= 1 << 27,
205 	ZIO_FLAG_DELEGATED	= 1 << 28,
206 };
207 
208 #define	ZIO_FLAG_MUSTSUCCEED		0
209 
210 #define	ZIO_DDT_CHILD_FLAGS(zio)				\
211 	(((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |		\
212 	ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
213 
214 #define	ZIO_GANG_CHILD_FLAGS(zio)				\
215 	(((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |		\
216 	ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
217 
218 #define	ZIO_VDEV_CHILD_FLAGS(zio)				\
219 	(((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |		\
220 	ZIO_FLAG_CANFAIL)
221 
222 enum zio_child {
223 	ZIO_CHILD_VDEV = 0,
224 	ZIO_CHILD_GANG,
225 	ZIO_CHILD_DDT,
226 	ZIO_CHILD_LOGICAL,
227 	ZIO_CHILD_TYPES
228 };
229 
230 enum zio_wait_type {
231 	ZIO_WAIT_READY = 0,
232 	ZIO_WAIT_DONE,
233 	ZIO_WAIT_TYPES
234 };
235 
236 /*
237  * We'll take the number 122 and 123 to indicate checksum errors and
238  * fragmentation. Those doesn't collide with any errno values as they
239  * are greater than ELAST.
240  */
241 #define	ECKSUM	122
242 #define	EFRAGS	123
243 
244 typedef void zio_done_func_t(zio_t *zio);
245 
246 extern const char *zio_type_name[ZIO_TYPES];
247 
248 /*
249  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
250  * identifies any block in the pool.  By convention, the meta-objset (MOS)
251  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
252  * except root blocks and ZIL blocks, which are defined as follows:
253  *
254  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
255  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
256  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
257  * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
258  *
259  * Note: this structure is called a bookmark because its original purpose
260  * was to remember where to resume a pool-wide traverse.
261  *
262  * Note: this structure is passed between userland and the kernel, and is
263  * stored on disk (by virtue of being incorporated into other on-disk
264  * structures, e.g. dsl_scan_phys_t).
265  */
266 typedef struct zbookmark_phys {
267 	uint64_t	zb_objset;
268 	uint64_t	zb_object;
269 	int64_t		zb_level;
270 	uint64_t	zb_blkid;
271 } zbookmark_phys_t;
272 
273 #define	SET_BOOKMARK(zb, objset, object, level, blkid)  \
274 {                                                       \
275 	(zb)->zb_objset = objset;                       \
276 	(zb)->zb_object = object;                       \
277 	(zb)->zb_level = level;                         \
278 	(zb)->zb_blkid = blkid;                         \
279 }
280 
281 #define	ZB_DESTROYED_OBJSET	(-1ULL)
282 
283 #define	ZB_ROOT_OBJECT		(0ULL)
284 #define	ZB_ROOT_LEVEL		(-1LL)
285 #define	ZB_ROOT_BLKID		(0ULL)
286 
287 #define	ZB_ZIL_OBJECT		(0ULL)
288 #define	ZB_ZIL_LEVEL		(-2LL)
289 
290 #define	ZB_DNODE_LEVEL		(-3LL)
291 #define	ZB_DNODE_BLKID		(0ULL)
292 
293 #define	ZB_IS_ZERO(zb)						\
294 	((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&	\
295 	(zb)->zb_level == 0 && (zb)->zb_blkid == 0)
296 #define	ZB_IS_ROOT(zb)				\
297 	((zb)->zb_object == ZB_ROOT_OBJECT &&	\
298 	(zb)->zb_level == ZB_ROOT_LEVEL &&	\
299 	(zb)->zb_blkid == ZB_ROOT_BLKID)
300 
301 typedef struct zio_prop {
302 	enum zio_checksum	zp_checksum;
303 	enum zio_compress	zp_compress;
304 	dmu_object_type_t	zp_type;
305 	uint8_t			zp_level;
306 	uint8_t			zp_copies;
307 	boolean_t		zp_dedup;
308 	boolean_t		zp_dedup_verify;
309 	boolean_t		zp_nopwrite;
310 } zio_prop_t;
311 
312 typedef struct zio_cksum_report zio_cksum_report_t;
313 
314 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
315     const void *good_data);
316 typedef void zio_cksum_free_f(void *cbdata, size_t size);
317 
318 struct zio_bad_cksum;				/* defined in zio_checksum.h */
319 struct dnode_phys;
320 
321 struct zio_cksum_report {
322 	struct zio_cksum_report *zcr_next;
323 	nvlist_t		*zcr_ereport;
324 	nvlist_t		*zcr_detector;
325 	void			*zcr_cbdata;
326 	size_t			zcr_cbinfo;	/* passed to zcr_free() */
327 	uint64_t		zcr_align;
328 	uint64_t		zcr_length;
329 	zio_cksum_finish_f	*zcr_finish;
330 	zio_cksum_free_f	*zcr_free;
331 
332 	/* internal use only */
333 	struct zio_bad_cksum	*zcr_ckinfo;	/* information from failure */
334 };
335 
336 typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
337     void *arg);
338 
339 zio_vsd_cksum_report_f	zio_vsd_default_cksum_report;
340 
341 typedef struct zio_vsd_ops {
342 	zio_done_func_t		*vsd_free;
343 	zio_vsd_cksum_report_f	*vsd_cksum_report;
344 } zio_vsd_ops_t;
345 
346 typedef struct zio_gang_node {
347 	zio_gbh_phys_t		*gn_gbh;
348 	struct zio_gang_node	*gn_child[SPA_GBH_NBLKPTRS];
349 } zio_gang_node_t;
350 
351 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
352     zio_gang_node_t *gn, void *data);
353 
354 typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size);
355 
356 typedef struct zio_transform {
357 	void			*zt_orig_data;
358 	uint64_t		zt_orig_size;
359 	uint64_t		zt_bufsize;
360 	zio_transform_func_t	*zt_transform;
361 	struct zio_transform	*zt_next;
362 } zio_transform_t;
363 
364 typedef int zio_pipe_stage_t(zio_t *zio);
365 
366 /*
367  * The io_reexecute flags are distinct from io_flags because the child must
368  * be able to propagate them to the parent.  The normal io_flags are local
369  * to the zio, not protected by any lock, and not modifiable by children;
370  * the reexecute flags are protected by io_lock, modifiable by children,
371  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
372  */
373 #define	ZIO_REEXECUTE_NOW	0x01
374 #define	ZIO_REEXECUTE_SUSPEND	0x02
375 
376 typedef struct zio_link {
377 	zio_t		*zl_parent;
378 	zio_t		*zl_child;
379 	list_node_t	zl_parent_node;
380 	list_node_t	zl_child_node;
381 } zio_link_t;
382 
383 /*
384  * Used for TRIM kstat.
385  */
386 typedef struct zio_trim_stats {
387 	/*
388 	 * Number of bytes successfully TRIMmed.
389 	 */
390 	kstat_named_t bytes;
391 
392 	/*
393 	 * Number of successful TRIM requests.
394 	 */
395 	kstat_named_t success;
396 
397 	/*
398 	 * Number of TRIM requests that failed because TRIM is not
399 	 * supported.
400 	 */
401 	kstat_named_t unsupported;
402 
403 	/*
404 	 * Number of TRIM requests that failed for other reasons.
405 	 */
406 	kstat_named_t failed;
407 } zio_trim_stats_t;
408 
409 extern zio_trim_stats_t zio_trim_stats;
410 
411 #define ZIO_TRIM_STAT_INCR(stat, val) \
412 	atomic_add_64(&zio_trim_stats.stat.value.ui64, (val));
413 #define ZIO_TRIM_STAT_BUMP(stat) \
414 	ZIO_TRIM_STAT_INCR(stat, 1);
415 
416 struct zio {
417 	/* Core information about this I/O */
418 	zbookmark_phys_t	io_bookmark;
419 	zio_prop_t	io_prop;
420 	zio_type_t	io_type;
421 	enum zio_child	io_child_type;
422 	int		io_cmd;
423 	zio_priority_t	io_priority;
424 	uint8_t		io_reexecute;
425 	uint8_t		io_state[ZIO_WAIT_TYPES];
426 	uint64_t	io_txg;
427 	spa_t		*io_spa;
428 	blkptr_t	*io_bp;
429 	blkptr_t	*io_bp_override;
430 	blkptr_t	io_bp_copy;
431 	list_t		io_parent_list;
432 	list_t		io_child_list;
433 	zio_link_t	*io_walk_link;
434 	zio_t		*io_logical;
435 	zio_transform_t *io_transform_stack;
436 
437 	/* Callback info */
438 	zio_done_func_t	*io_ready;
439 	zio_done_func_t	*io_physdone;
440 	zio_done_func_t	*io_done;
441 	void		*io_private;
442 	int64_t		io_prev_space_delta;	/* DMU private */
443 	blkptr_t	io_bp_orig;
444 
445 	/* Data represented by this I/O */
446 	void		*io_data;
447 	void		*io_orig_data;
448 	uint64_t	io_size;
449 	uint64_t	io_orig_size;
450 
451 	/* Stuff for the vdev stack */
452 	vdev_t		*io_vd;
453 	void		*io_vsd;
454 	const zio_vsd_ops_t *io_vsd_ops;
455 
456 	uint64_t	io_offset;
457 	hrtime_t	io_timestamp;
458 	avl_node_t	io_queue_node;
459 	avl_node_t	io_offset_node;
460 
461 	/* Internal pipeline state */
462 	enum zio_flag	io_flags;
463 	enum zio_stage	io_stage;
464 	enum zio_stage	io_pipeline;
465 	enum zio_flag	io_orig_flags;
466 	enum zio_stage	io_orig_stage;
467 	enum zio_stage	io_orig_pipeline;
468 	int		io_error;
469 	int		io_child_error[ZIO_CHILD_TYPES];
470 	uint64_t	io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
471 	uint64_t	io_child_count;
472 	uint64_t	io_phys_children;
473 	uint64_t	io_parent_count;
474 	uint64_t	*io_stall;
475 	zio_t		*io_gang_leader;
476 	zio_gang_node_t	*io_gang_tree;
477 	void		*io_executor;
478 	void		*io_waiter;
479 	kmutex_t	io_lock;
480 	kcondvar_t	io_cv;
481 
482 	/* FMA state */
483 	zio_cksum_report_t *io_cksum_report;
484 	uint64_t	io_ena;
485 
486 	/* Taskq dispatching state */
487 	taskq_ent_t	io_tqent;
488 
489 	avl_node_t	io_trim_node;
490 	list_node_t	io_trim_link;
491 };
492 
493 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
494     zio_done_func_t *done, void *priv, enum zio_flag flags);
495 
496 extern zio_t *zio_root(spa_t *spa,
497     zio_done_func_t *done, void *priv, enum zio_flag flags);
498 
499 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
500     uint64_t size, zio_done_func_t *done, void *priv,
501     zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
502 
503 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
504     void *data, uint64_t size, const zio_prop_t *zp,
505     zio_done_func_t *ready, zio_done_func_t *physdone, zio_done_func_t *done,
506     void *priv,
507     zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
508 
509 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
510     void *data, uint64_t size, zio_done_func_t *done, void *priv,
511     zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
512 
513 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
514     boolean_t nopwrite);
515 
516 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
517 
518 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
519     const blkptr_t *bp,
520     zio_done_func_t *done, void *priv, enum zio_flag flags);
521 
522 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
523     uint64_t offset, uint64_t size, zio_done_func_t *done, void *priv,
524     zio_priority_t priority, enum zio_flag flags);
525 
526 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
527     uint64_t size, void *data, int checksum,
528     zio_done_func_t *done, void *priv, zio_priority_t priority,
529     enum zio_flag flags, boolean_t labels);
530 
531 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
532     uint64_t size, void *data, int checksum,
533     zio_done_func_t *done, void *priv, zio_priority_t priority,
534     enum zio_flag flags, boolean_t labels);
535 
536 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
537     const blkptr_t *bp, uint64_t size, enum zio_flag flags);
538 
539 extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
540     blkptr_t *old_bp, uint64_t size, boolean_t use_slog);
541 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
542 extern void zio_flush(zio_t *zio, vdev_t *vd);
543 extern zio_t *zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset,
544     uint64_t size);
545 extern void zio_shrink(zio_t *zio, uint64_t size);
546 
547 extern int zio_wait(zio_t *zio);
548 extern void zio_nowait(zio_t *zio);
549 extern void zio_execute(zio_t *zio);
550 extern void zio_interrupt(zio_t *zio);
551 
552 extern zio_t *zio_walk_parents(zio_t *cio);
553 extern zio_t *zio_walk_children(zio_t *pio);
554 extern zio_t *zio_unique_parent(zio_t *cio);
555 extern void zio_add_child(zio_t *pio, zio_t *cio);
556 
557 extern void *zio_buf_alloc(size_t size);
558 extern void zio_buf_free(void *buf, size_t size);
559 extern void *zio_data_buf_alloc(size_t size);
560 extern void zio_data_buf_free(void *buf, size_t size);
561 
562 extern void zio_resubmit_stage_async(void *);
563 
564 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
565     uint64_t offset, void *data, uint64_t size, int type,
566     zio_priority_t priority, enum zio_flag flags,
567     zio_done_func_t *done, void *priv);
568 
569 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
570     void *data, uint64_t size, int type, zio_priority_t priority,
571     enum zio_flag flags, zio_done_func_t *done, void *priv);
572 
573 extern void zio_vdev_io_bypass(zio_t *zio);
574 extern void zio_vdev_io_reissue(zio_t *zio);
575 extern void zio_vdev_io_redone(zio_t *zio);
576 
577 extern void zio_checksum_verified(zio_t *zio);
578 extern int zio_worst_error(int e1, int e2);
579 
580 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
581     enum zio_checksum parent);
582 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
583     enum zio_checksum child, enum zio_checksum parent);
584 extern enum zio_compress zio_compress_select(spa_t *spa,
585     enum zio_compress child, enum zio_compress parent);
586 
587 extern void zio_suspend(spa_t *spa, zio_t *zio);
588 extern int zio_resume(spa_t *spa);
589 extern void zio_resume_wait(spa_t *spa);
590 
591 /*
592  * Initial setup and teardown.
593  */
594 extern void zio_init(void);
595 extern void zio_fini(void);
596 
597 /*
598  * Fault injection
599  */
600 struct zinject_record;
601 extern uint32_t zio_injection_enabled;
602 extern int zio_inject_fault(char *name, int flags, int *id,
603     struct zinject_record *record);
604 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
605     struct zinject_record *record);
606 extern int zio_clear_fault(int id);
607 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
608 extern int zio_handle_fault_injection(zio_t *zio, int error);
609 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
610 extern int zio_handle_label_injection(zio_t *zio, int error);
611 extern void zio_handle_ignored_writes(zio_t *zio);
612 extern uint64_t zio_handle_io_delay(zio_t *zio);
613 
614 /*
615  * Checksum ereport functions
616  */
617 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
618     uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
619 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
620     const void *good_data, const void *bad_data, boolean_t drop_if_identical);
621 
622 extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
623 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
624 
625 /* If we have the good data in hand, this function can be used */
626 extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
627     struct zio *zio, uint64_t offset, uint64_t length,
628     const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
629 
630 /* Called from spa_sync(), but primarily an injection handler */
631 extern void spa_handle_ignored_writes(spa_t *spa);
632 
633 /* zbookmark_phys functions */
634 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
635     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
636 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
637     uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
638 
639 #ifdef	__cplusplus
640 }
641 #endif
642 
643 #endif	/* _ZIO_H */
644