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, 2018 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
43extern "C" {
44#endif
45
46/*
47 * Embedded checksum
48 */
49#define	ZEC_MAGIC	0x210da7ab10c7a11ULL
50
51typedef 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
68typedef 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
74enum 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	ZIO_CHECKSUM_SHA512,
87	ZIO_CHECKSUM_SKEIN,
88#ifdef illumos
89	ZIO_CHECKSUM_EDONR,
90#endif
91	ZIO_CHECKSUM_FUNCTIONS
92};
93
94/*
95 * The number of "legacy" compression functions which can be set on individual
96 * objects.
97 */
98#define	ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2
99
100#define	ZIO_CHECKSUM_ON_VALUE	ZIO_CHECKSUM_FLETCHER_4
101#define	ZIO_CHECKSUM_DEFAULT	ZIO_CHECKSUM_ON
102
103#define	ZIO_CHECKSUM_MASK	0xffULL
104#define	ZIO_CHECKSUM_VERIFY	(1 << 8)
105
106#define	ZIO_DEDUPCHECKSUM	ZIO_CHECKSUM_SHA256
107#define	ZIO_DEDUPDITTO_MIN	100
108
109/*
110 * The number of "legacy" compression functions which can be set on individual
111 * objects.
112 */
113#define	ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
114
115/*
116 * The meaning of "compress = on" selected by the compression features enabled
117 * on a given pool.
118 */
119#define	ZIO_COMPRESS_LEGACY_ON_VALUE	ZIO_COMPRESS_LZJB
120#define	ZIO_COMPRESS_LZ4_ON_VALUE	ZIO_COMPRESS_LZ4
121
122#define	ZIO_COMPRESS_DEFAULT		ZIO_COMPRESS_OFF
123
124#define	BOOTFS_COMPRESS_VALID(compress)			\
125	((compress) == ZIO_COMPRESS_LZJB ||		\
126	(compress) == ZIO_COMPRESS_LZ4 ||		\
127	(compress) == ZIO_COMPRESS_ON ||		\
128	(compress) == ZIO_COMPRESS_OFF)
129
130#define	ZIO_FAILURE_MODE_WAIT		0
131#define	ZIO_FAILURE_MODE_CONTINUE	1
132#define	ZIO_FAILURE_MODE_PANIC		2
133
134enum zio_flag {
135	/*
136	 * Flags inherited by gang, ddt, and vdev children,
137	 * and that must be equal for two zios to aggregate
138	 */
139	ZIO_FLAG_DONT_AGGREGATE	= 1 << 0,
140	ZIO_FLAG_IO_REPAIR	= 1 << 1,
141	ZIO_FLAG_SELF_HEAL	= 1 << 2,
142	ZIO_FLAG_RESILVER	= 1 << 3,
143	ZIO_FLAG_SCRUB		= 1 << 4,
144	ZIO_FLAG_SCAN_THREAD	= 1 << 5,
145	ZIO_FLAG_PHYSICAL	= 1 << 6,
146
147#define	ZIO_FLAG_AGG_INHERIT	(ZIO_FLAG_CANFAIL - 1)
148
149	/*
150	 * Flags inherited by ddt, gang, and vdev children.
151	 */
152	ZIO_FLAG_CANFAIL	= 1 << 7,	/* must be first for INHERIT */
153	ZIO_FLAG_SPECULATIVE	= 1 << 8,
154	ZIO_FLAG_CONFIG_WRITER	= 1 << 9,
155	ZIO_FLAG_DONT_RETRY	= 1 << 10,
156	ZIO_FLAG_DONT_CACHE	= 1 << 11,
157	ZIO_FLAG_NODATA		= 1 << 12,
158	ZIO_FLAG_INDUCE_DAMAGE	= 1 << 13,
159	ZIO_FLAG_IO_ALLOCATING	= 1 << 14,
160
161#define	ZIO_FLAG_DDT_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
162#define	ZIO_FLAG_GANG_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
163
164	/*
165	 * Flags inherited by vdev children.
166	 */
167	ZIO_FLAG_IO_RETRY	= 1 << 15,	/* must be first for INHERIT */
168	ZIO_FLAG_PROBE		= 1 << 16,
169	ZIO_FLAG_TRYHARD	= 1 << 17,
170	ZIO_FLAG_OPTIONAL	= 1 << 18,
171
172#define	ZIO_FLAG_VDEV_INHERIT	(ZIO_FLAG_DONT_QUEUE - 1)
173
174	/*
175	 * Flags not inherited by any children.
176	 */
177	ZIO_FLAG_DONT_QUEUE	= 1 << 19,	/* must be first for INHERIT */
178	ZIO_FLAG_DONT_PROPAGATE	= 1 << 20,
179	ZIO_FLAG_IO_BYPASS	= 1 << 21,
180	ZIO_FLAG_IO_REWRITE	= 1 << 22,
181	ZIO_FLAG_RAW		= 1 << 23,
182	ZIO_FLAG_GANG_CHILD	= 1 << 24,
183	ZIO_FLAG_DDT_CHILD	= 1 << 25,
184	ZIO_FLAG_GODFATHER	= 1 << 26,
185	ZIO_FLAG_NOPWRITE	= 1 << 27,
186	ZIO_FLAG_REEXECUTED	= 1 << 28,
187	ZIO_FLAG_DELEGATED	= 1 << 29,
188};
189
190#define	ZIO_FLAG_MUSTSUCCEED		0
191
192#define	ZIO_DDT_CHILD_FLAGS(zio)				\
193	(((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |		\
194	ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
195
196#define	ZIO_GANG_CHILD_FLAGS(zio)				\
197	(((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |		\
198	ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
199
200#define	ZIO_VDEV_CHILD_FLAGS(zio)				\
201	(((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |		\
202	ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_CANFAIL)
203
204#define	ZIO_CHILD_BIT(x)		(1 << (x))
205#define	ZIO_CHILD_BIT_IS_SET(val, x)	((val) & (1 << (x)))
206
207enum zio_child {
208	ZIO_CHILD_VDEV = 0,
209	ZIO_CHILD_GANG,
210	ZIO_CHILD_DDT,
211	ZIO_CHILD_LOGICAL,
212	ZIO_CHILD_TYPES
213};
214
215#define	ZIO_CHILD_VDEV_BIT		ZIO_CHILD_BIT(ZIO_CHILD_VDEV)
216#define	ZIO_CHILD_GANG_BIT		ZIO_CHILD_BIT(ZIO_CHILD_GANG)
217#define	ZIO_CHILD_DDT_BIT		ZIO_CHILD_BIT(ZIO_CHILD_DDT)
218#define	ZIO_CHILD_LOGICAL_BIT		ZIO_CHILD_BIT(ZIO_CHILD_LOGICAL)
219#define	ZIO_CHILD_ALL_BITS					\
220	(ZIO_CHILD_VDEV_BIT | ZIO_CHILD_GANG_BIT |		\
221	ZIO_CHILD_DDT_BIT | ZIO_CHILD_LOGICAL_BIT)
222
223enum zio_wait_type {
224	ZIO_WAIT_READY = 0,
225	ZIO_WAIT_DONE,
226	ZIO_WAIT_TYPES
227};
228
229/*
230 * We'll take the number 122 and 123 to indicate checksum errors and
231 * fragmentation. Those doesn't collide with any errno values as they
232 * are greater than ELAST.
233 */
234#define	ECKSUM	122
235#define	EFRAGS	123
236
237typedef void zio_done_func_t(zio_t *zio);
238
239extern boolean_t zio_dva_throttle_enabled;
240extern const char *zio_type_name[ZIO_TYPES];
241
242/*
243 * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
244 * identifies any block in the pool.  By convention, the meta-objset (MOS)
245 * is objset 0, and the meta-dnode is object 0.  This covers all blocks
246 * except root blocks and ZIL blocks, which are defined as follows:
247 *
248 * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
249 * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
250 * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
251 * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
252 *
253 * Note: this structure is called a bookmark because its original purpose
254 * was to remember where to resume a pool-wide traverse.
255 *
256 * Note: this structure is passed between userland and the kernel, and is
257 * stored on disk (by virtue of being incorporated into other on-disk
258 * structures, e.g. dsl_scan_phys_t).
259 */
260typedef struct zbookmark_phys {
261	uint64_t	zb_objset;
262	uint64_t	zb_object;
263	int64_t		zb_level;
264	uint64_t	zb_blkid;
265} zbookmark_phys_t;
266
267#define	SET_BOOKMARK(zb, objset, object, level, blkid)  \
268{                                                       \
269	(zb)->zb_objset = objset;                       \
270	(zb)->zb_object = object;                       \
271	(zb)->zb_level = level;                         \
272	(zb)->zb_blkid = blkid;                         \
273}
274
275#define	ZB_DESTROYED_OBJSET	(-1ULL)
276
277#define	ZB_ROOT_OBJECT		(0ULL)
278#define	ZB_ROOT_LEVEL		(-1LL)
279#define	ZB_ROOT_BLKID		(0ULL)
280
281#define	ZB_ZIL_OBJECT		(0ULL)
282#define	ZB_ZIL_LEVEL		(-2LL)
283
284#define	ZB_DNODE_LEVEL		(-3LL)
285#define	ZB_DNODE_BLKID		(0ULL)
286
287#define	ZB_IS_ZERO(zb)						\
288	((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&	\
289	(zb)->zb_level == 0 && (zb)->zb_blkid == 0)
290#define	ZB_IS_ROOT(zb)				\
291	((zb)->zb_object == ZB_ROOT_OBJECT &&	\
292	(zb)->zb_level == ZB_ROOT_LEVEL &&	\
293	(zb)->zb_blkid == ZB_ROOT_BLKID)
294
295typedef struct zio_prop {
296	enum zio_checksum	zp_checksum;
297	enum zio_compress	zp_compress;
298	dmu_object_type_t	zp_type;
299	uint8_t			zp_level;
300	uint8_t			zp_copies;
301	boolean_t		zp_dedup;
302	boolean_t		zp_dedup_verify;
303	boolean_t		zp_nopwrite;
304} zio_prop_t;
305
306typedef struct zio_cksum_report zio_cksum_report_t;
307
308typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
309    const void *good_data);
310typedef void zio_cksum_free_f(void *cbdata, size_t size);
311
312struct zio_bad_cksum;				/* defined in zio_checksum.h */
313struct dnode_phys;
314struct abd;
315
316struct zio_cksum_report {
317	struct zio_cksum_report *zcr_next;
318	nvlist_t		*zcr_ereport;
319	nvlist_t		*zcr_detector;
320	void			*zcr_cbdata;
321	size_t			zcr_cbinfo;	/* passed to zcr_free() */
322	uint64_t		zcr_align;
323	uint64_t		zcr_length;
324	zio_cksum_finish_f	*zcr_finish;
325	zio_cksum_free_f	*zcr_free;
326
327	/* internal use only */
328	struct zio_bad_cksum	*zcr_ckinfo;	/* information from failure */
329};
330
331typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
332    void *arg);
333
334zio_vsd_cksum_report_f	zio_vsd_default_cksum_report;
335
336typedef struct zio_vsd_ops {
337	zio_done_func_t		*vsd_free;
338	zio_vsd_cksum_report_f	*vsd_cksum_report;
339} zio_vsd_ops_t;
340
341typedef struct zio_gang_node {
342	zio_gbh_phys_t		*gn_gbh;
343	struct zio_gang_node	*gn_child[SPA_GBH_NBLKPTRS];
344} zio_gang_node_t;
345
346typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
347    zio_gang_node_t *gn, struct abd *data, uint64_t offset);
348
349typedef void zio_transform_func_t(zio_t *zio, struct abd *data, uint64_t size);
350
351typedef struct zio_transform {
352	struct abd		*zt_orig_abd;
353	uint64_t		zt_orig_size;
354	uint64_t		zt_bufsize;
355	zio_transform_func_t	*zt_transform;
356	struct zio_transform	*zt_next;
357} zio_transform_t;
358
359typedef zio_t *zio_pipe_stage_t(zio_t *zio);
360
361/*
362 * The io_reexecute flags are distinct from io_flags because the child must
363 * be able to propagate them to the parent.  The normal io_flags are local
364 * to the zio, not protected by any lock, and not modifiable by children;
365 * the reexecute flags are protected by io_lock, modifiable by children,
366 * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
367 */
368#define	ZIO_REEXECUTE_NOW	0x01
369#define	ZIO_REEXECUTE_SUSPEND	0x02
370
371typedef struct zio_alloc_list {
372	list_t  zal_list;
373	uint64_t zal_size;
374} zio_alloc_list_t;
375
376typedef 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 */
386typedef 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
409extern 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
416struct 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_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_children_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	struct abd	*io_abd;
447	struct abd	*io_orig_abd;
448	uint64_t	io_size;
449	uint64_t	io_orig_size;
450	/* io_lsize != io_orig_size iff this is a raw write */
451	uint64_t	io_lsize;
452
453	/* Stuff for the vdev stack */
454	vdev_t		*io_vd;
455	void		*io_vsd;
456	const zio_vsd_ops_t *io_vsd_ops;
457
458	uint64_t	io_offset;
459	hrtime_t	io_timestamp;
460	hrtime_t	io_queued_timestamp;
461	hrtime_t	io_target_timestamp;
462	avl_node_t	io_queue_node;
463	avl_node_t	io_offset_node;
464	avl_node_t	io_alloc_node;
465	zio_alloc_list_t 	io_alloc_list;
466
467#ifdef __FreeBSD__
468	struct bio	*io_bio;
469#endif
470
471	/* Internal pipeline state */
472	enum zio_flag	io_flags;
473	enum zio_stage	io_stage;
474	enum zio_stage	io_pipeline;
475	enum zio_flag	io_orig_flags;
476	enum zio_stage	io_orig_stage;
477	enum zio_stage	io_orig_pipeline;
478	enum zio_stage	io_pipeline_trace;
479	int		io_error;
480	int		io_child_error[ZIO_CHILD_TYPES];
481	uint64_t	io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
482	uint64_t	io_child_count;
483	uint64_t	io_phys_children;
484	uint64_t	io_parent_count;
485	uint64_t	*io_stall;
486	zio_t		*io_gang_leader;
487	zio_gang_node_t	*io_gang_tree;
488	void		*io_executor;
489	void		*io_waiter;
490	kmutex_t	io_lock;
491	kcondvar_t	io_cv;
492	int		io_allocator;
493
494	/* FMA state */
495	zio_cksum_report_t *io_cksum_report;
496	uint64_t	io_ena;
497
498	/* Taskq dispatching state */
499	taskq_ent_t	io_tqent;
500
501	avl_node_t	io_trim_node;
502	list_node_t	io_trim_link;
503};
504
505extern int zio_bookmark_compare(const void *, const void *);
506
507extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
508    zio_done_func_t *done, void *priv, enum zio_flag flags);
509
510extern zio_t *zio_root(spa_t *spa,
511    zio_done_func_t *done, void *priv, enum zio_flag flags);
512
513extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
514    struct abd *data, uint64_t lsize, zio_done_func_t *done, void *priv,
515    zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
516
517extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
518    struct abd *data, uint64_t size, uint64_t psize, const zio_prop_t *zp,
519    zio_done_func_t *ready, zio_done_func_t *children_ready,
520    zio_done_func_t *physdone, zio_done_func_t *done,
521    void *priv, zio_priority_t priority, enum zio_flag flags,
522    const zbookmark_phys_t *zb);
523
524extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
525    struct abd *data, uint64_t size, zio_done_func_t *done, void *priv,
526    zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
527
528extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
529    boolean_t nopwrite);
530
531extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
532
533extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
534    const blkptr_t *bp,
535    zio_done_func_t *done, void *priv, enum zio_flag flags);
536
537extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
538    uint64_t offset, uint64_t size, zio_done_func_t *done, void *priv,
539    zio_priority_t priority, enum zio_flag flags);
540
541extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
542    uint64_t size, struct abd *data, int checksum,
543    zio_done_func_t *done, void *priv, zio_priority_t priority,
544    enum zio_flag flags, boolean_t labels);
545
546extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
547    uint64_t size, struct abd *data, int checksum,
548    zio_done_func_t *done, void *priv, zio_priority_t priority,
549    enum zio_flag flags, boolean_t labels);
550
551extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
552    const blkptr_t *bp, uint64_t size, enum zio_flag flags);
553
554extern int zio_alloc_zil(spa_t *spa, uint64_t objset, uint64_t txg,
555    blkptr_t *new_bp, blkptr_t *old_bp, uint64_t size, boolean_t *slog);
556extern void zio_flush(zio_t *zio, vdev_t *vd);
557extern zio_t *zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset,
558    uint64_t size);
559extern void zio_shrink(zio_t *zio, uint64_t size);
560
561extern int zio_wait(zio_t *zio);
562extern void zio_nowait(zio_t *zio);
563extern void zio_execute(zio_t *zio);
564extern void zio_interrupt(zio_t *zio);
565extern void zio_delay_init(zio_t *zio);
566extern void zio_delay_interrupt(zio_t *zio);
567
568extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **);
569extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **);
570extern zio_t *zio_unique_parent(zio_t *cio);
571extern void zio_add_child(zio_t *pio, zio_t *cio);
572
573extern void *zio_buf_alloc(size_t size);
574extern void zio_buf_free(void *buf, size_t size);
575extern void *zio_data_buf_alloc(size_t size);
576extern void zio_data_buf_free(void *buf, size_t size);
577
578extern void zio_push_transform(zio_t *zio, struct abd *abd, uint64_t size,
579    uint64_t bufsize, zio_transform_func_t *transform);
580extern void zio_pop_transforms(zio_t *zio);
581
582extern void zio_resubmit_stage_async(void *);
583
584extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
585    uint64_t offset, struct abd *data, uint64_t size, int type,
586    zio_priority_t priority, enum zio_flag flags,
587    zio_done_func_t *done, void *priv);
588
589extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
590    struct abd *data, uint64_t size, zio_type_t type, zio_priority_t priority,
591    enum zio_flag flags, zio_done_func_t *done, void *priv);
592
593extern void zio_vdev_io_bypass(zio_t *zio);
594extern void zio_vdev_io_reissue(zio_t *zio);
595extern void zio_vdev_io_redone(zio_t *zio);
596
597extern void zio_change_priority(zio_t *pio, zio_priority_t priority);
598
599extern void zio_checksum_verified(zio_t *zio);
600extern int zio_worst_error(int e1, int e2);
601
602extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
603    enum zio_checksum parent);
604extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
605    enum zio_checksum child, enum zio_checksum parent);
606extern enum zio_compress zio_compress_select(spa_t *spa,
607    enum zio_compress child, enum zio_compress parent);
608
609extern void zio_suspend(spa_t *spa, zio_t *zio);
610extern int zio_resume(spa_t *spa);
611extern void zio_resume_wait(spa_t *spa);
612
613/*
614 * Initial setup and teardown.
615 */
616extern void zio_init(void);
617extern void zio_fini(void);
618
619/*
620 * Fault injection
621 */
622struct zinject_record;
623extern uint32_t zio_injection_enabled;
624extern int zio_inject_fault(char *name, int flags, int *id,
625    struct zinject_record *record);
626extern int zio_inject_list_next(int *id, char *name, size_t buflen,
627    struct zinject_record *record);
628extern int zio_clear_fault(int id);
629extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
630extern int zio_handle_fault_injection(zio_t *zio, int error);
631extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
632extern int zio_handle_label_injection(zio_t *zio, int error);
633extern void zio_handle_ignored_writes(zio_t *zio);
634extern hrtime_t zio_handle_io_delay(zio_t *zio);
635
636/*
637 * Checksum ereport functions
638 */
639extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
640    uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
641extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
642    const void *good_data, const void *bad_data, boolean_t drop_if_identical);
643
644extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
645extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
646
647/* If we have the good data in hand, this function can be used */
648extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
649    struct zio *zio, uint64_t offset, uint64_t length,
650    const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
651
652/* Called from spa_sync(), but primarily an injection handler */
653extern void spa_handle_ignored_writes(spa_t *spa);
654
655/* zbookmark_phys functions */
656boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
657    const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
658int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
659    uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
660
661#ifdef	__cplusplus
662}
663#endif
664
665#endif	/* _ZIO_H */
666