zio.h revision 268657 
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/zfs_context.h>
33#include <sys/spa.h>
34#include <sys/txg.h>
35#include <sys/avl.h>
36#include <sys/kstat.h>
37#include <sys/fs/zfs.h>
38#include <sys/zio_impl.h>
39
40#ifdef	__cplusplus
41extern "C" {
42#endif
43
44/*
45 * Embedded checksum
46 */
47#define	ZEC_MAGIC	0x210da7ab10c7a11ULL
48
49typedef struct zio_eck {
50	uint64_t	zec_magic;	/* for validation, endianness	*/
51	zio_cksum_t	zec_cksum;	/* 256-bit checksum		*/
52} zio_eck_t;
53
54/*
55 * Gang block headers are self-checksumming and contain an array
56 * of block pointers.
57 */
58#define	SPA_GANGBLOCKSIZE	SPA_MINBLOCKSIZE
59#define	SPA_GBH_NBLKPTRS	((SPA_GANGBLOCKSIZE - \
60	sizeof (zio_eck_t)) / sizeof (blkptr_t))
61#define	SPA_GBH_FILLER		((SPA_GANGBLOCKSIZE - \
62	sizeof (zio_eck_t) - \
63	(SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\
64	sizeof (uint64_t))
65
66typedef struct zio_gbh {
67	blkptr_t		zg_blkptr[SPA_GBH_NBLKPTRS];
68	uint64_t		zg_filler[SPA_GBH_FILLER];
69	zio_eck_t		zg_tail;
70} zio_gbh_phys_t;
71
72enum zio_checksum {
73	ZIO_CHECKSUM_INHERIT = 0,
74	ZIO_CHECKSUM_ON,
75	ZIO_CHECKSUM_OFF,
76	ZIO_CHECKSUM_LABEL,
77	ZIO_CHECKSUM_GANG_HEADER,
78	ZIO_CHECKSUM_ZILOG,
79	ZIO_CHECKSUM_FLETCHER_2,
80	ZIO_CHECKSUM_FLETCHER_4,
81	ZIO_CHECKSUM_SHA256,
82	ZIO_CHECKSUM_ZILOG2,
83	ZIO_CHECKSUM_NOPARITY,
84	ZIO_CHECKSUM_FUNCTIONS
85};
86
87/*
88 * The number of "legacy" compression functions which can be set on individual
89 * objects.
90 */
91#define	ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2
92
93#define	ZIO_CHECKSUM_ON_VALUE	ZIO_CHECKSUM_FLETCHER_4
94#define	ZIO_CHECKSUM_DEFAULT	ZIO_CHECKSUM_ON
95
96#define	ZIO_CHECKSUM_MASK	0xffULL
97#define	ZIO_CHECKSUM_VERIFY	(1 << 8)
98
99#define	ZIO_DEDUPCHECKSUM	ZIO_CHECKSUM_SHA256
100#define	ZIO_DEDUPDITTO_MIN	100
101
102enum zio_compress {
103	ZIO_COMPRESS_INHERIT = 0,
104	ZIO_COMPRESS_ON,
105	ZIO_COMPRESS_OFF,
106	ZIO_COMPRESS_LZJB,
107	ZIO_COMPRESS_EMPTY,
108	ZIO_COMPRESS_GZIP_1,
109	ZIO_COMPRESS_GZIP_2,
110	ZIO_COMPRESS_GZIP_3,
111	ZIO_COMPRESS_GZIP_4,
112	ZIO_COMPRESS_GZIP_5,
113	ZIO_COMPRESS_GZIP_6,
114	ZIO_COMPRESS_GZIP_7,
115	ZIO_COMPRESS_GZIP_8,
116	ZIO_COMPRESS_GZIP_9,
117	ZIO_COMPRESS_ZLE,
118	ZIO_COMPRESS_LZ4,
119	ZIO_COMPRESS_FUNCTIONS
120};
121
122/*
123 * The number of "legacy" compression functions which can be set on individual
124 * objects.
125 */
126#define	ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
127
128/* N.B. when altering this value, also change BOOTFS_COMPRESS_VALID below */
129#define	ZIO_COMPRESS_ON_VALUE	ZIO_COMPRESS_LZJB
130#define	ZIO_COMPRESS_DEFAULT	ZIO_COMPRESS_OFF
131
132#define	BOOTFS_COMPRESS_VALID(compress)			\
133	((compress) == ZIO_COMPRESS_LZJB ||		\
134	(compress) == ZIO_COMPRESS_LZ4 ||		\
135	((compress) == ZIO_COMPRESS_ON &&		\
136	ZIO_COMPRESS_ON_VALUE == ZIO_COMPRESS_LZJB) ||	\
137	(compress) == ZIO_COMPRESS_OFF)
138
139#define	ZIO_FAILURE_MODE_WAIT		0
140#define	ZIO_FAILURE_MODE_CONTINUE	1
141#define	ZIO_FAILURE_MODE_PANIC		2
142
143typedef enum zio_priority {
144	ZIO_PRIORITY_SYNC_READ,
145	ZIO_PRIORITY_SYNC_WRITE,	/* ZIL */
146	ZIO_PRIORITY_ASYNC_READ,	/* prefetch */
147	ZIO_PRIORITY_ASYNC_WRITE,	/* spa_sync() */
148	ZIO_PRIORITY_SCRUB,		/* asynchronous scrub/resilver reads */
149	ZIO_PRIORITY_NUM_QUEUEABLE,
150
151	ZIO_PRIORITY_NOW		/* non-queued i/os (e.g. free) */
152} zio_priority_t;
153
154#define	ZIO_PIPELINE_CONTINUE		0x100
155#define	ZIO_PIPELINE_STOP		0x101
156
157enum zio_flag {
158	/*
159	 * Flags inherited by gang, ddt, and vdev children,
160	 * and that must be equal for two zios to aggregate
161	 */
162	ZIO_FLAG_DONT_AGGREGATE	= 1 << 0,
163	ZIO_FLAG_IO_REPAIR	= 1 << 1,
164	ZIO_FLAG_SELF_HEAL	= 1 << 2,
165	ZIO_FLAG_RESILVER	= 1 << 3,
166	ZIO_FLAG_SCRUB		= 1 << 4,
167	ZIO_FLAG_SCAN_THREAD	= 1 << 5,
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 << 6,	/* must be first for INHERIT */
175	ZIO_FLAG_SPECULATIVE	= 1 << 7,
176	ZIO_FLAG_CONFIG_WRITER	= 1 << 8,
177	ZIO_FLAG_DONT_RETRY	= 1 << 9,
178	ZIO_FLAG_DONT_CACHE	= 1 << 10,
179	ZIO_FLAG_NODATA		= 1 << 11,
180	ZIO_FLAG_INDUCE_DAMAGE	= 1 << 12,
181
182#define	ZIO_FLAG_DDT_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
183#define	ZIO_FLAG_GANG_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
184
185	/*
186	 * Flags inherited by vdev children.
187	 */
188	ZIO_FLAG_IO_RETRY	= 1 << 13,	/* must be first for INHERIT */
189	ZIO_FLAG_PROBE		= 1 << 14,
190	ZIO_FLAG_TRYHARD	= 1 << 15,
191	ZIO_FLAG_OPTIONAL	= 1 << 16,
192
193#define	ZIO_FLAG_VDEV_INHERIT	(ZIO_FLAG_DONT_QUEUE - 1)
194
195	/*
196	 * Flags not inherited by any children.
197	 */
198	ZIO_FLAG_DONT_QUEUE	= 1 << 17,	/* must be first for INHERIT */
199	ZIO_FLAG_DONT_PROPAGATE	= 1 << 18,
200	ZIO_FLAG_IO_BYPASS	= 1 << 19,
201	ZIO_FLAG_IO_REWRITE	= 1 << 20,
202	ZIO_FLAG_RAW		= 1 << 21,
203	ZIO_FLAG_GANG_CHILD	= 1 << 22,
204	ZIO_FLAG_DDT_CHILD	= 1 << 23,
205	ZIO_FLAG_GODFATHER	= 1 << 24,
206	ZIO_FLAG_NOPWRITE	= 1 << 25,
207	ZIO_FLAG_REEXECUTED	= 1 << 26,
208	ZIO_FLAG_DELEGATED	= 1 << 27,
209};
210
211#define	ZIO_FLAG_MUSTSUCCEED		0
212
213#define	ZIO_DDT_CHILD_FLAGS(zio)				\
214	(((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |		\
215	ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
216
217#define	ZIO_GANG_CHILD_FLAGS(zio)				\
218	(((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |		\
219	ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
220
221#define	ZIO_VDEV_CHILD_FLAGS(zio)				\
222	(((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |		\
223	ZIO_FLAG_CANFAIL)
224
225enum zio_child {
226	ZIO_CHILD_VDEV = 0,
227	ZIO_CHILD_GANG,
228	ZIO_CHILD_DDT,
229	ZIO_CHILD_LOGICAL,
230	ZIO_CHILD_TYPES
231};
232
233enum zio_wait_type {
234	ZIO_WAIT_READY = 0,
235	ZIO_WAIT_DONE,
236	ZIO_WAIT_TYPES
237};
238
239/*
240 * We'll take the number 122 and 123 to indicate checksum errors and
241 * fragmentation. Those doesn't collide with any errno values as they
242 * are greater than ELAST.
243 */
244#define	ECKSUM	122
245#define	EFRAGS	123
246
247typedef void zio_done_func_t(zio_t *zio);
248
249extern const char *zio_type_name[ZIO_TYPES];
250
251/*
252 * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
253 * identifies any block in the pool.  By convention, the meta-objset (MOS)
254 * is objset 0, and the meta-dnode is object 0.  This covers all blocks
255 * except root blocks and ZIL blocks, which are defined as follows:
256 *
257 * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
258 * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
259 * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
260 *
261 * Note: this structure is called a bookmark because its original purpose
262 * was to remember where to resume a pool-wide traverse.
263 *
264 * Note: this structure is passed between userland and the kernel, and is
265 * stored on disk (by virtue of being incorporated into other on-disk
266 * structures, e.g. dsl_scan_phys_t).
267 */
268typedef struct zbookmark_phys {
269	uint64_t	zb_objset;
270	uint64_t	zb_object;
271	int64_t		zb_level;
272	uint64_t	zb_blkid;
273} zbookmark_phys_t;
274
275#define	SET_BOOKMARK(zb, objset, object, level, blkid)  \
276{                                                       \
277	(zb)->zb_objset = objset;                       \
278	(zb)->zb_object = object;                       \
279	(zb)->zb_level = level;                         \
280	(zb)->zb_blkid = blkid;                         \
281}
282
283#define	ZB_DESTROYED_OBJSET	(-1ULL)
284
285#define	ZB_ROOT_OBJECT		(0ULL)
286#define	ZB_ROOT_LEVEL		(-1LL)
287#define	ZB_ROOT_BLKID		(0ULL)
288
289#define	ZB_ZIL_OBJECT		(0ULL)
290#define	ZB_ZIL_LEVEL		(-2LL)
291
292#define	ZB_IS_ZERO(zb)						\
293	((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&	\
294	(zb)->zb_level == 0 && (zb)->zb_blkid == 0)
295#define	ZB_IS_ROOT(zb)				\
296	((zb)->zb_object == ZB_ROOT_OBJECT &&	\
297	(zb)->zb_level == ZB_ROOT_LEVEL &&	\
298	(zb)->zb_blkid == ZB_ROOT_BLKID)
299
300typedef struct zio_prop {
301	enum zio_checksum	zp_checksum;
302	enum zio_compress	zp_compress;
303	dmu_object_type_t	zp_type;
304	uint8_t			zp_level;
305	uint8_t			zp_copies;
306	boolean_t		zp_dedup;
307	boolean_t		zp_dedup_verify;
308	boolean_t		zp_nopwrite;
309} zio_prop_t;
310
311typedef struct zio_cksum_report zio_cksum_report_t;
312
313typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
314    const void *good_data);
315typedef void zio_cksum_free_f(void *cbdata, size_t size);
316
317struct zio_bad_cksum;				/* defined in zio_checksum.h */
318struct dnode_phys;
319
320struct zio_cksum_report {
321	struct zio_cksum_report *zcr_next;
322	nvlist_t		*zcr_ereport;
323	nvlist_t		*zcr_detector;
324	void			*zcr_cbdata;
325	size_t			zcr_cbinfo;	/* passed to zcr_free() */
326	uint64_t		zcr_align;
327	uint64_t		zcr_length;
328	zio_cksum_finish_f	*zcr_finish;
329	zio_cksum_free_f	*zcr_free;
330
331	/* internal use only */
332	struct zio_bad_cksum	*zcr_ckinfo;	/* information from failure */
333};
334
335typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
336    void *arg);
337
338zio_vsd_cksum_report_f	zio_vsd_default_cksum_report;
339
340typedef struct zio_vsd_ops {
341	zio_done_func_t		*vsd_free;
342	zio_vsd_cksum_report_f	*vsd_cksum_report;
343} zio_vsd_ops_t;
344
345typedef struct zio_gang_node {
346	zio_gbh_phys_t		*gn_gbh;
347	struct zio_gang_node	*gn_child[SPA_GBH_NBLKPTRS];
348} zio_gang_node_t;
349
350typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
351    zio_gang_node_t *gn, void *data);
352
353typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size);
354
355typedef struct zio_transform {
356	void			*zt_orig_data;
357	uint64_t		zt_orig_size;
358	uint64_t		zt_bufsize;
359	zio_transform_func_t	*zt_transform;
360	struct zio_transform	*zt_next;
361} zio_transform_t;
362
363typedef int zio_pipe_stage_t(zio_t **ziop);
364
365/*
366 * The io_reexecute flags are distinct from io_flags because the child must
367 * be able to propagate them to the parent.  The normal io_flags are local
368 * to the zio, not protected by any lock, and not modifiable by children;
369 * the reexecute flags are protected by io_lock, modifiable by children,
370 * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
371 */
372#define	ZIO_REEXECUTE_NOW	0x01
373#define	ZIO_REEXECUTE_SUSPEND	0x02
374
375typedef struct zio_link {
376	zio_t		*zl_parent;
377	zio_t		*zl_child;
378	list_node_t	zl_parent_node;
379	list_node_t	zl_child_node;
380} zio_link_t;
381
382/*
383 * Used for TRIM kstat.
384 */
385typedef struct zio_trim_stats {
386	/*
387	 * Number of bytes successfully TRIMmed.
388	 */
389	kstat_named_t bytes;
390
391	/*
392	 * Number of successful TRIM requests.
393	 */
394	kstat_named_t success;
395
396	/*
397	 * Number of TRIM requests that failed because TRIM is not
398	 * supported.
399	 */
400	kstat_named_t unsupported;
401
402	/*
403	 * Number of TRIM requests that failed for other reasons.
404	 */
405	kstat_named_t failed;
406} zio_trim_stats_t;
407
408extern zio_trim_stats_t zio_trim_stats;
409
410#define ZIO_TRIM_STAT_INCR(stat, val) \
411	atomic_add_64(&zio_trim_stats.stat.value.ui64, (val));
412#define ZIO_TRIM_STAT_BUMP(stat) \
413	ZIO_TRIM_STAT_INCR(stat, 1);
414
415struct zio {
416	/* Core information about this I/O */
417	zbookmark_phys_t	io_bookmark;
418	zio_prop_t	io_prop;
419	zio_type_t	io_type;
420	enum zio_child	io_child_type;
421	int		io_cmd;
422	zio_priority_t	io_priority;
423	uint8_t		io_reexecute;
424	uint8_t		io_state[ZIO_WAIT_TYPES];
425	uint64_t	io_txg;
426	spa_t		*io_spa;
427	blkptr_t	*io_bp;
428	blkptr_t	*io_bp_override;
429	blkptr_t	io_bp_copy;
430	list_t		io_parent_list;
431	list_t		io_child_list;
432	zio_link_t	*io_walk_link;
433	zio_t		*io_logical;
434	zio_transform_t *io_transform_stack;
435
436	/* Callback info */
437	zio_done_func_t	*io_ready;
438	zio_done_func_t	*io_physdone;
439	zio_done_func_t	*io_done;
440	void		*io_private;
441	int64_t		io_prev_space_delta;	/* DMU private */
442	blkptr_t	io_bp_orig;
443
444	/* Data represented by this I/O */
445	void		*io_data;
446	void		*io_orig_data;
447	uint64_t	io_size;
448	uint64_t	io_orig_size;
449
450	/* Stuff for the vdev stack */
451	vdev_t		*io_vd;
452	void		*io_vsd;
453	const zio_vsd_ops_t *io_vsd_ops;
454
455	uint64_t	io_offset;
456	hrtime_t	io_timestamp;
457	avl_node_t	io_queue_node;
458
459	/* Internal pipeline state */
460	enum zio_flag	io_flags;
461	enum zio_stage	io_stage;
462	enum zio_stage	io_pipeline;
463	enum zio_flag	io_orig_flags;
464	enum zio_stage	io_orig_stage;
465	enum zio_stage	io_orig_pipeline;
466	int		io_error;
467	int		io_child_error[ZIO_CHILD_TYPES];
468	uint64_t	io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
469	uint64_t	io_child_count;
470	uint64_t	io_phys_children;
471	uint64_t	io_parent_count;
472	uint64_t	*io_stall;
473	zio_t		*io_gang_leader;
474	zio_gang_node_t	*io_gang_tree;
475	void		*io_executor;
476	void		*io_waiter;
477	kmutex_t	io_lock;
478	kcondvar_t	io_cv;
479
480	/* FMA state */
481	zio_cksum_report_t *io_cksum_report;
482	uint64_t	io_ena;
483
484	/* Taskq dispatching state */
485	taskq_ent_t	io_tqent;
486
487	avl_node_t	io_trim_node;
488	list_node_t	io_trim_link;
489};
490
491extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
492    zio_done_func_t *done, void *priv, enum zio_flag flags);
493
494extern zio_t *zio_root(spa_t *spa,
495    zio_done_func_t *done, void *priv, enum zio_flag flags);
496
497extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
498    uint64_t size, zio_done_func_t *done, void *priv,
499    zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
500
501extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
502    void *data, uint64_t size, const zio_prop_t *zp,
503    zio_done_func_t *ready, zio_done_func_t *physdone, zio_done_func_t *done,
504    void *priv,
505    zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
506
507extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
508    void *data, uint64_t size, zio_done_func_t *done, void *priv,
509    zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
510
511extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
512    boolean_t nopwrite);
513
514extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
515
516extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
517    const blkptr_t *bp,
518    zio_done_func_t *done, void *priv, enum zio_flag flags);
519
520extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
521    uint64_t offset, uint64_t size, zio_done_func_t *done, void *priv,
522    enum zio_flag flags);
523
524extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
525    uint64_t size, void *data, int checksum,
526    zio_done_func_t *done, void *priv, zio_priority_t priority,
527    enum zio_flag flags, boolean_t labels);
528
529extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
530    uint64_t size, void *data, int checksum,
531    zio_done_func_t *done, void *priv, zio_priority_t priority,
532    enum zio_flag flags, boolean_t labels);
533
534extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
535    const blkptr_t *bp, uint64_t size, enum zio_flag flags);
536
537extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
538    blkptr_t *old_bp, uint64_t size, boolean_t use_slog);
539extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
540extern void zio_flush(zio_t *zio, vdev_t *vd);
541extern zio_t *zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset,
542    uint64_t size);
543extern void zio_shrink(zio_t *zio, uint64_t size);
544
545extern int zio_wait(zio_t *zio);
546extern void zio_nowait(zio_t *zio);
547extern void zio_execute(zio_t *zio);
548extern void zio_interrupt(zio_t *zio);
549
550extern zio_t *zio_walk_parents(zio_t *cio);
551extern zio_t *zio_walk_children(zio_t *pio);
552extern zio_t *zio_unique_parent(zio_t *cio);
553extern void zio_add_child(zio_t *pio, zio_t *cio);
554
555extern void *zio_buf_alloc(size_t size);
556extern void zio_buf_free(void *buf, size_t size);
557extern void *zio_data_buf_alloc(size_t size);
558extern void zio_data_buf_free(void *buf, size_t size);
559
560extern void zio_resubmit_stage_async(void *);
561
562extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
563    uint64_t offset, void *data, uint64_t size, int type,
564    zio_priority_t priority, enum zio_flag flags,
565    zio_done_func_t *done, void *priv);
566
567extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
568    void *data, uint64_t size, int type, zio_priority_t priority,
569    enum zio_flag flags, zio_done_func_t *done, void *priv);
570
571extern void zio_vdev_io_bypass(zio_t *zio);
572extern void zio_vdev_io_reissue(zio_t *zio);
573extern void zio_vdev_io_redone(zio_t *zio);
574
575extern void zio_checksum_verified(zio_t *zio);
576extern int zio_worst_error(int e1, int e2);
577
578extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
579    enum zio_checksum parent);
580extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
581    enum zio_checksum child, enum zio_checksum parent);
582extern enum zio_compress zio_compress_select(enum zio_compress child,
583    enum zio_compress parent);
584
585extern void zio_suspend(spa_t *spa, zio_t *zio);
586extern int zio_resume(spa_t *spa);
587extern void zio_resume_wait(spa_t *spa);
588
589/*
590 * Initial setup and teardown.
591 */
592extern void zio_init(void);
593extern void zio_fini(void);
594
595/*
596 * Fault injection
597 */
598struct zinject_record;
599extern uint32_t zio_injection_enabled;
600extern int zio_inject_fault(char *name, int flags, int *id,
601    struct zinject_record *record);
602extern int zio_inject_list_next(int *id, char *name, size_t buflen,
603    struct zinject_record *record);
604extern int zio_clear_fault(int id);
605extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
606extern int zio_handle_fault_injection(zio_t *zio, int error);
607extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
608extern int zio_handle_label_injection(zio_t *zio, int error);
609extern void zio_handle_ignored_writes(zio_t *zio);
610extern uint64_t zio_handle_io_delay(zio_t *zio);
611
612/*
613 * Checksum ereport functions
614 */
615extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
616    uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
617extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
618    const void *good_data, const void *bad_data, boolean_t drop_if_identical);
619
620extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
621extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
622
623/* If we have the good data in hand, this function can be used */
624extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
625    struct zio *zio, uint64_t offset, uint64_t length,
626    const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
627
628/* Called from spa_sync(), but primarily an injection handler */
629extern void spa_handle_ignored_writes(spa_t *spa);
630
631/* zbookmark_phys functions */
632boolean_t zbookmark_is_before(const struct dnode_phys *dnp,
633    const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
634
635#ifdef	__cplusplus
636}
637#endif
638
639#endif	/* _ZIO_H */
640