zio.h revision 290757 
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
42extern "C" {
43#endif
44
45/*
46 * Embedded checksum
47 */
48#define	ZEC_MAGIC	0x210da7ab10c7a11ULL
49
50typedef 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
67typedef 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
73enum 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
108enum 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#define	ZIO_PIPELINE_CONTINUE		0x100
154#define	ZIO_PIPELINE_STOP		0x101
155
156enum 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
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 << 14,	/* must be first for INHERIT */
189	ZIO_FLAG_PROBE		= 1 << 15,
190	ZIO_FLAG_TRYHARD	= 1 << 16,
191	ZIO_FLAG_OPTIONAL	= 1 << 17,
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 << 18,	/* must be first for INHERIT */
199	ZIO_FLAG_DONT_PROPAGATE	= 1 << 19,
200	ZIO_FLAG_IO_BYPASS	= 1 << 20,
201	ZIO_FLAG_IO_REWRITE	= 1 << 21,
202	ZIO_FLAG_RAW		= 1 << 22,
203	ZIO_FLAG_GANG_CHILD	= 1 << 23,
204	ZIO_FLAG_DDT_CHILD	= 1 << 24,
205	ZIO_FLAG_GODFATHER	= 1 << 25,
206	ZIO_FLAG_NOPWRITE	= 1 << 26,
207	ZIO_FLAG_REEXECUTED	= 1 << 27,
208	ZIO_FLAG_DELEGATED	= 1 << 28,
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 * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
261 *
262 * Note: this structure is called a bookmark because its original purpose
263 * was to remember where to resume a pool-wide traverse.
264 *
265 * Note: this structure is passed between userland and the kernel, and is
266 * stored on disk (by virtue of being incorporated into other on-disk
267 * structures, e.g. dsl_scan_phys_t).
268 */
269typedef struct zbookmark_phys {
270	uint64_t	zb_objset;
271	uint64_t	zb_object;
272	int64_t		zb_level;
273	uint64_t	zb_blkid;
274} zbookmark_phys_t;
275
276#define	SET_BOOKMARK(zb, objset, object, level, blkid)  \
277{                                                       \
278	(zb)->zb_objset = objset;                       \
279	(zb)->zb_object = object;                       \
280	(zb)->zb_level = level;                         \
281	(zb)->zb_blkid = blkid;                         \
282}
283
284#define	ZB_DESTROYED_OBJSET	(-1ULL)
285
286#define	ZB_ROOT_OBJECT		(0ULL)
287#define	ZB_ROOT_LEVEL		(-1LL)
288#define	ZB_ROOT_BLKID		(0ULL)
289
290#define	ZB_ZIL_OBJECT		(0ULL)
291#define	ZB_ZIL_LEVEL		(-2LL)
292
293#define	ZB_DNODE_LEVEL		(-3LL)
294#define	ZB_DNODE_BLKID		(0ULL)
295
296#define	ZB_IS_ZERO(zb)						\
297	((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&	\
298	(zb)->zb_level == 0 && (zb)->zb_blkid == 0)
299#define	ZB_IS_ROOT(zb)				\
300	((zb)->zb_object == ZB_ROOT_OBJECT &&	\
301	(zb)->zb_level == ZB_ROOT_LEVEL &&	\
302	(zb)->zb_blkid == ZB_ROOT_BLKID)
303
304typedef struct zio_prop {
305	enum zio_checksum	zp_checksum;
306	enum zio_compress	zp_compress;
307	dmu_object_type_t	zp_type;
308	uint8_t			zp_level;
309	uint8_t			zp_copies;
310	boolean_t		zp_dedup;
311	boolean_t		zp_dedup_verify;
312	boolean_t		zp_nopwrite;
313} zio_prop_t;
314
315typedef struct zio_cksum_report zio_cksum_report_t;
316
317typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
318    const void *good_data);
319typedef void zio_cksum_free_f(void *cbdata, size_t size);
320
321struct zio_bad_cksum;				/* defined in zio_checksum.h */
322struct dnode_phys;
323
324struct zio_cksum_report {
325	struct zio_cksum_report *zcr_next;
326	nvlist_t		*zcr_ereport;
327	nvlist_t		*zcr_detector;
328	void			*zcr_cbdata;
329	size_t			zcr_cbinfo;	/* passed to zcr_free() */
330	uint64_t		zcr_align;
331	uint64_t		zcr_length;
332	zio_cksum_finish_f	*zcr_finish;
333	zio_cksum_free_f	*zcr_free;
334
335	/* internal use only */
336	struct zio_bad_cksum	*zcr_ckinfo;	/* information from failure */
337};
338
339typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
340    void *arg);
341
342zio_vsd_cksum_report_f	zio_vsd_default_cksum_report;
343
344typedef struct zio_vsd_ops {
345	zio_done_func_t		*vsd_free;
346	zio_vsd_cksum_report_f	*vsd_cksum_report;
347} zio_vsd_ops_t;
348
349typedef struct zio_gang_node {
350	zio_gbh_phys_t		*gn_gbh;
351	struct zio_gang_node	*gn_child[SPA_GBH_NBLKPTRS];
352} zio_gang_node_t;
353
354typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
355    zio_gang_node_t *gn, void *data);
356
357typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size);
358
359typedef struct zio_transform {
360	void			*zt_orig_data;
361	uint64_t		zt_orig_size;
362	uint64_t		zt_bufsize;
363	zio_transform_func_t	*zt_transform;
364	struct zio_transform	*zt_next;
365} zio_transform_t;
366
367typedef int zio_pipe_stage_t(zio_t *zio);
368
369/*
370 * The io_reexecute flags are distinct from io_flags because the child must
371 * be able to propagate them to the parent.  The normal io_flags are local
372 * to the zio, not protected by any lock, and not modifiable by children;
373 * the reexecute flags are protected by io_lock, modifiable by children,
374 * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
375 */
376#define	ZIO_REEXECUTE_NOW	0x01
377#define	ZIO_REEXECUTE_SUSPEND	0x02
378
379typedef struct zio_link {
380	zio_t		*zl_parent;
381	zio_t		*zl_child;
382	list_node_t	zl_parent_node;
383	list_node_t	zl_child_node;
384} zio_link_t;
385
386/*
387 * Used for TRIM kstat.
388 */
389typedef struct zio_trim_stats {
390	/*
391	 * Number of bytes successfully TRIMmed.
392	 */
393	kstat_named_t bytes;
394
395	/*
396	 * Number of successful TRIM requests.
397	 */
398	kstat_named_t success;
399
400	/*
401	 * Number of TRIM requests that failed because TRIM is not
402	 * supported.
403	 */
404	kstat_named_t unsupported;
405
406	/*
407	 * Number of TRIM requests that failed for other reasons.
408	 */
409	kstat_named_t failed;
410} zio_trim_stats_t;
411
412extern zio_trim_stats_t zio_trim_stats;
413
414#define ZIO_TRIM_STAT_INCR(stat, val) \
415	atomic_add_64(&zio_trim_stats.stat.value.ui64, (val));
416#define ZIO_TRIM_STAT_BUMP(stat) \
417	ZIO_TRIM_STAT_INCR(stat, 1);
418
419struct zio {
420	/* Core information about this I/O */
421	zbookmark_phys_t	io_bookmark;
422	zio_prop_t	io_prop;
423	zio_type_t	io_type;
424	enum zio_child	io_child_type;
425	int		io_cmd;
426	zio_priority_t	io_priority;
427	uint8_t		io_reexecute;
428	uint8_t		io_state[ZIO_WAIT_TYPES];
429	uint64_t	io_txg;
430	spa_t		*io_spa;
431	blkptr_t	*io_bp;
432	blkptr_t	*io_bp_override;
433	blkptr_t	io_bp_copy;
434	list_t		io_parent_list;
435	list_t		io_child_list;
436	zio_link_t	*io_walk_link;
437	zio_t		*io_logical;
438	zio_transform_t *io_transform_stack;
439
440	/* Callback info */
441	zio_done_func_t	*io_ready;
442	zio_done_func_t	*io_physdone;
443	zio_done_func_t	*io_done;
444	void		*io_private;
445	int64_t		io_prev_space_delta;	/* DMU private */
446	blkptr_t	io_bp_orig;
447
448	/* Data represented by this I/O */
449	void		*io_data;
450	void		*io_orig_data;
451	uint64_t	io_size;
452	uint64_t	io_orig_size;
453
454	/* Stuff for the vdev stack */
455	vdev_t		*io_vd;
456	void		*io_vsd;
457	const zio_vsd_ops_t *io_vsd_ops;
458
459	uint64_t	io_offset;
460	hrtime_t	io_timestamp;
461	avl_node_t	io_queue_node;
462	avl_node_t	io_offset_node;
463
464	/* Internal pipeline state */
465	enum zio_flag	io_flags;
466	enum zio_stage	io_stage;
467	enum zio_stage	io_pipeline;
468	enum zio_flag	io_orig_flags;
469	enum zio_stage	io_orig_stage;
470	enum zio_stage	io_orig_pipeline;
471	int		io_error;
472	int		io_child_error[ZIO_CHILD_TYPES];
473	uint64_t	io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
474	uint64_t	io_child_count;
475	uint64_t	io_phys_children;
476	uint64_t	io_parent_count;
477	uint64_t	*io_stall;
478	zio_t		*io_gang_leader;
479	zio_gang_node_t	*io_gang_tree;
480	void		*io_executor;
481	void		*io_waiter;
482	kmutex_t	io_lock;
483	kcondvar_t	io_cv;
484
485	/* FMA state */
486	zio_cksum_report_t *io_cksum_report;
487	uint64_t	io_ena;
488
489	/* Taskq dispatching state */
490	taskq_ent_t	io_tqent;
491
492	avl_node_t	io_trim_node;
493	list_node_t	io_trim_link;
494};
495
496extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
497    zio_done_func_t *done, void *priv, enum zio_flag flags);
498
499extern zio_t *zio_root(spa_t *spa,
500    zio_done_func_t *done, void *priv, enum zio_flag flags);
501
502extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
503    uint64_t size, zio_done_func_t *done, void *priv,
504    zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
505
506extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
507    void *data, uint64_t size, const zio_prop_t *zp,
508    zio_done_func_t *ready, zio_done_func_t *physdone, zio_done_func_t *done,
509    void *priv,
510    zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
511
512extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
513    void *data, uint64_t size, zio_done_func_t *done, void *priv,
514    zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
515
516extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
517    boolean_t nopwrite);
518
519extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
520
521extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
522    const blkptr_t *bp,
523    zio_done_func_t *done, void *priv, enum zio_flag flags);
524
525extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
526    uint64_t offset, uint64_t size, zio_done_func_t *done, void *priv,
527    zio_priority_t priority, enum zio_flag flags);
528
529extern zio_t *zio_read_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_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
535    uint64_t size, void *data, int checksum,
536    zio_done_func_t *done, void *priv, zio_priority_t priority,
537    enum zio_flag flags, boolean_t labels);
538
539extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
540    const blkptr_t *bp, uint64_t size, enum zio_flag flags);
541
542extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
543    blkptr_t *old_bp, uint64_t size, boolean_t use_slog);
544extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
545extern void zio_flush(zio_t *zio, vdev_t *vd);
546extern zio_t *zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset,
547    uint64_t size);
548extern void zio_shrink(zio_t *zio, uint64_t size);
549
550extern int zio_wait(zio_t *zio);
551extern void zio_nowait(zio_t *zio);
552extern void zio_execute(zio_t *zio);
553extern void zio_interrupt(zio_t *zio);
554
555extern zio_t *zio_walk_parents(zio_t *cio);
556extern zio_t *zio_walk_children(zio_t *pio);
557extern zio_t *zio_unique_parent(zio_t *cio);
558extern void zio_add_child(zio_t *pio, zio_t *cio);
559
560extern void *zio_buf_alloc(size_t size);
561extern void zio_buf_free(void *buf, size_t size);
562extern void *zio_data_buf_alloc(size_t size);
563extern void zio_data_buf_free(void *buf, size_t size);
564
565extern void zio_resubmit_stage_async(void *);
566
567extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
568    uint64_t offset, void *data, uint64_t size, int type,
569    zio_priority_t priority, enum zio_flag flags,
570    zio_done_func_t *done, void *priv);
571
572extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
573    void *data, uint64_t size, int type, zio_priority_t priority,
574    enum zio_flag flags, zio_done_func_t *done, void *priv);
575
576extern void zio_vdev_io_bypass(zio_t *zio);
577extern void zio_vdev_io_reissue(zio_t *zio);
578extern void zio_vdev_io_redone(zio_t *zio);
579
580extern void zio_checksum_verified(zio_t *zio);
581extern int zio_worst_error(int e1, int e2);
582
583extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
584    enum zio_checksum parent);
585extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
586    enum zio_checksum child, enum zio_checksum parent);
587extern enum zio_compress zio_compress_select(spa_t *spa,
588    enum zio_compress child, enum zio_compress parent);
589
590extern void zio_suspend(spa_t *spa, zio_t *zio);
591extern int zio_resume(spa_t *spa);
592extern void zio_resume_wait(spa_t *spa);
593
594/*
595 * Initial setup and teardown.
596 */
597extern void zio_init(void);
598extern void zio_fini(void);
599
600/*
601 * Fault injection
602 */
603struct zinject_record;
604extern uint32_t zio_injection_enabled;
605extern int zio_inject_fault(char *name, int flags, int *id,
606    struct zinject_record *record);
607extern int zio_inject_list_next(int *id, char *name, size_t buflen,
608    struct zinject_record *record);
609extern int zio_clear_fault(int id);
610extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
611extern int zio_handle_fault_injection(zio_t *zio, int error);
612extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
613extern int zio_handle_label_injection(zio_t *zio, int error);
614extern void zio_handle_ignored_writes(zio_t *zio);
615extern uint64_t zio_handle_io_delay(zio_t *zio);
616
617/*
618 * Checksum ereport functions
619 */
620extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
621    uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
622extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
623    const void *good_data, const void *bad_data, boolean_t drop_if_identical);
624
625extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
626extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
627
628/* If we have the good data in hand, this function can be used */
629extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
630    struct zio *zio, uint64_t offset, uint64_t length,
631    const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
632
633/* Called from spa_sync(), but primarily an injection handler */
634extern void spa_handle_ignored_writes(spa_t *spa);
635
636/* zbookmark_phys functions */
637boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
638    const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
639int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
640    uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
641
642#ifdef	__cplusplus
643}
644#endif
645
646#endif	/* _ZIO_H */
647