zio.h revision 330238 
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 */
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
153enum 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	ZIO_FLAG_IO_ALLOCATING	= 1 << 14,
179
180#define	ZIO_FLAG_DDT_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
181#define	ZIO_FLAG_GANG_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
182
183	/*
184	 * Flags inherited by vdev children.
185	 */
186	ZIO_FLAG_IO_RETRY	= 1 << 15,	/* must be first for INHERIT */
187	ZIO_FLAG_PROBE		= 1 << 16,
188	ZIO_FLAG_TRYHARD	= 1 << 17,
189	ZIO_FLAG_OPTIONAL	= 1 << 18,
190
191#define	ZIO_FLAG_VDEV_INHERIT	(ZIO_FLAG_DONT_QUEUE - 1)
192
193	/*
194	 * Flags not inherited by any children.
195	 */
196	ZIO_FLAG_DONT_QUEUE	= 1 << 19,	/* must be first for INHERIT */
197	ZIO_FLAG_DONT_PROPAGATE	= 1 << 20,
198	ZIO_FLAG_IO_BYPASS	= 1 << 21,
199	ZIO_FLAG_IO_REWRITE	= 1 << 22,
200	ZIO_FLAG_RAW		= 1 << 23,
201	ZIO_FLAG_GANG_CHILD	= 1 << 24,
202	ZIO_FLAG_DDT_CHILD	= 1 << 25,
203	ZIO_FLAG_GODFATHER	= 1 << 26,
204	ZIO_FLAG_NOPWRITE	= 1 << 27,
205	ZIO_FLAG_REEXECUTED	= 1 << 28,
206	ZIO_FLAG_DELEGATED	= 1 << 29,
207};
208
209#define	ZIO_FLAG_MUSTSUCCEED		0
210
211#define	ZIO_DDT_CHILD_FLAGS(zio)				\
212	(((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |		\
213	ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
214
215#define	ZIO_GANG_CHILD_FLAGS(zio)				\
216	(((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |		\
217	ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
218
219#define	ZIO_VDEV_CHILD_FLAGS(zio)				\
220	(((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |		\
221	ZIO_FLAG_CANFAIL)
222
223#define	ZIO_CHILD_BIT(x)		(1 << (x))
224#define	ZIO_CHILD_BIT_IS_SET(val, x)	((val) & (1 << (x)))
225
226enum 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#define	ZIO_CHILD_VDEV_BIT		ZIO_CHILD_BIT(ZIO_CHILD_VDEV)
235#define	ZIO_CHILD_GANG_BIT		ZIO_CHILD_BIT(ZIO_CHILD_GANG)
236#define	ZIO_CHILD_DDT_BIT		ZIO_CHILD_BIT(ZIO_CHILD_DDT)
237#define	ZIO_CHILD_LOGICAL_BIT		ZIO_CHILD_BIT(ZIO_CHILD_LOGICAL)
238#define	ZIO_CHILD_ALL_BITS					\
239	(ZIO_CHILD_VDEV_BIT | ZIO_CHILD_GANG_BIT | 		\
240	ZIO_CHILD_DDT_BIT | ZIO_CHILD_LOGICAL_BIT)
241
242enum zio_wait_type {
243	ZIO_WAIT_READY = 0,
244	ZIO_WAIT_DONE,
245	ZIO_WAIT_TYPES
246};
247
248/*
249 * We'll take the number 122 and 123 to indicate checksum errors and
250 * fragmentation. Those doesn't collide with any errno values as they
251 * are greater than ELAST.
252 */
253#define	ECKSUM	122
254#define	EFRAGS	123
255
256typedef void zio_done_func_t(zio_t *zio);
257
258extern boolean_t zio_dva_throttle_enabled;
259extern const char *zio_type_name[ZIO_TYPES];
260
261/*
262 * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
263 * identifies any block in the pool.  By convention, the meta-objset (MOS)
264 * is objset 0, and the meta-dnode is object 0.  This covers all blocks
265 * except root blocks and ZIL blocks, which are defined as follows:
266 *
267 * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
268 * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
269 * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
270 * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
271 *
272 * Note: this structure is called a bookmark because its original purpose
273 * was to remember where to resume a pool-wide traverse.
274 *
275 * Note: this structure is passed between userland and the kernel, and is
276 * stored on disk (by virtue of being incorporated into other on-disk
277 * structures, e.g. dsl_scan_phys_t).
278 */
279typedef struct zbookmark_phys {
280	uint64_t	zb_objset;
281	uint64_t	zb_object;
282	int64_t		zb_level;
283	uint64_t	zb_blkid;
284} zbookmark_phys_t;
285
286#define	SET_BOOKMARK(zb, objset, object, level, blkid)  \
287{                                                       \
288	(zb)->zb_objset = objset;                       \
289	(zb)->zb_object = object;                       \
290	(zb)->zb_level = level;                         \
291	(zb)->zb_blkid = blkid;                         \
292}
293
294#define	ZB_DESTROYED_OBJSET	(-1ULL)
295
296#define	ZB_ROOT_OBJECT		(0ULL)
297#define	ZB_ROOT_LEVEL		(-1LL)
298#define	ZB_ROOT_BLKID		(0ULL)
299
300#define	ZB_ZIL_OBJECT		(0ULL)
301#define	ZB_ZIL_LEVEL		(-2LL)
302
303#define	ZB_DNODE_LEVEL		(-3LL)
304#define	ZB_DNODE_BLKID		(0ULL)
305
306#define	ZB_IS_ZERO(zb)						\
307	((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&	\
308	(zb)->zb_level == 0 && (zb)->zb_blkid == 0)
309#define	ZB_IS_ROOT(zb)				\
310	((zb)->zb_object == ZB_ROOT_OBJECT &&	\
311	(zb)->zb_level == ZB_ROOT_LEVEL &&	\
312	(zb)->zb_blkid == ZB_ROOT_BLKID)
313
314typedef struct zio_prop {
315	enum zio_checksum	zp_checksum;
316	enum zio_compress	zp_compress;
317	dmu_object_type_t	zp_type;
318	uint8_t			zp_level;
319	uint8_t			zp_copies;
320	boolean_t		zp_dedup;
321	boolean_t		zp_dedup_verify;
322	boolean_t		zp_nopwrite;
323} zio_prop_t;
324
325typedef struct zio_cksum_report zio_cksum_report_t;
326
327typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
328    const void *good_data);
329typedef void zio_cksum_free_f(void *cbdata, size_t size);
330
331struct zio_bad_cksum;				/* defined in zio_checksum.h */
332struct dnode_phys;
333
334struct zio_cksum_report {
335	struct zio_cksum_report *zcr_next;
336	nvlist_t		*zcr_ereport;
337	nvlist_t		*zcr_detector;
338	void			*zcr_cbdata;
339	size_t			zcr_cbinfo;	/* passed to zcr_free() */
340	uint64_t		zcr_align;
341	uint64_t		zcr_length;
342	zio_cksum_finish_f	*zcr_finish;
343	zio_cksum_free_f	*zcr_free;
344
345	/* internal use only */
346	struct zio_bad_cksum	*zcr_ckinfo;	/* information from failure */
347};
348
349typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
350    void *arg);
351
352zio_vsd_cksum_report_f	zio_vsd_default_cksum_report;
353
354typedef struct zio_vsd_ops {
355	zio_done_func_t		*vsd_free;
356	zio_vsd_cksum_report_f	*vsd_cksum_report;
357} zio_vsd_ops_t;
358
359typedef struct zio_gang_node {
360	zio_gbh_phys_t		*gn_gbh;
361	struct zio_gang_node	*gn_child[SPA_GBH_NBLKPTRS];
362} zio_gang_node_t;
363
364typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
365    zio_gang_node_t *gn, void *data);
366
367typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size);
368
369typedef struct zio_transform {
370	void			*zt_orig_data;
371	uint64_t		zt_orig_size;
372	uint64_t		zt_bufsize;
373	zio_transform_func_t	*zt_transform;
374	struct zio_transform	*zt_next;
375} zio_transform_t;
376
377typedef int zio_pipe_stage_t(zio_t *zio);
378
379/*
380 * The io_reexecute flags are distinct from io_flags because the child must
381 * be able to propagate them to the parent.  The normal io_flags are local
382 * to the zio, not protected by any lock, and not modifiable by children;
383 * the reexecute flags are protected by io_lock, modifiable by children,
384 * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
385 */
386#define	ZIO_REEXECUTE_NOW	0x01
387#define	ZIO_REEXECUTE_SUSPEND	0x02
388
389typedef struct zio_link {
390	zio_t		*zl_parent;
391	zio_t		*zl_child;
392	list_node_t	zl_parent_node;
393	list_node_t	zl_child_node;
394} zio_link_t;
395
396/*
397 * Used for TRIM kstat.
398 */
399typedef struct zio_trim_stats {
400	/*
401	 * Number of bytes successfully TRIMmed.
402	 */
403	kstat_named_t bytes;
404
405	/*
406	 * Number of successful TRIM requests.
407	 */
408	kstat_named_t success;
409
410	/*
411	 * Number of TRIM requests that failed because TRIM is not
412	 * supported.
413	 */
414	kstat_named_t unsupported;
415
416	/*
417	 * Number of TRIM requests that failed for other reasons.
418	 */
419	kstat_named_t failed;
420} zio_trim_stats_t;
421
422extern zio_trim_stats_t zio_trim_stats;
423
424#define ZIO_TRIM_STAT_INCR(stat, val) \
425	atomic_add_64(&zio_trim_stats.stat.value.ui64, (val));
426#define ZIO_TRIM_STAT_BUMP(stat) \
427	ZIO_TRIM_STAT_INCR(stat, 1);
428
429struct zio {
430	/* Core information about this I/O */
431	zbookmark_phys_t	io_bookmark;
432	zio_prop_t	io_prop;
433	zio_type_t	io_type;
434	enum zio_child	io_child_type;
435	int		io_cmd;
436	zio_priority_t	io_priority;
437	uint8_t		io_reexecute;
438	uint8_t		io_state[ZIO_WAIT_TYPES];
439	uint64_t	io_txg;
440	spa_t		*io_spa;
441	blkptr_t	*io_bp;
442	blkptr_t	*io_bp_override;
443	blkptr_t	io_bp_copy;
444	list_t		io_parent_list;
445	list_t		io_child_list;
446	zio_t		*io_logical;
447	zio_transform_t *io_transform_stack;
448
449	/* Callback info */
450	zio_done_func_t	*io_ready;
451	zio_done_func_t	*io_children_ready;
452	zio_done_func_t	*io_physdone;
453	zio_done_func_t	*io_done;
454	void		*io_private;
455	int64_t		io_prev_space_delta;	/* DMU private */
456	blkptr_t	io_bp_orig;
457
458	/* Data represented by this I/O */
459	void		*io_data;
460	void		*io_orig_data;
461	uint64_t	io_size;
462	uint64_t	io_orig_size;
463
464	/* Stuff for the vdev stack */
465	vdev_t		*io_vd;
466	void		*io_vsd;
467	const zio_vsd_ops_t *io_vsd_ops;
468
469	uint64_t	io_offset;
470	hrtime_t	io_timestamp;
471	hrtime_t	io_queued_timestamp;
472	hrtime_t	io_target_timestamp;
473	avl_node_t	io_queue_node;
474	avl_node_t	io_offset_node;
475	avl_node_t	io_alloc_node;
476
477	/* Internal pipeline state */
478	enum zio_flag	io_flags;
479	enum zio_stage	io_stage;
480	enum zio_stage	io_pipeline;
481	enum zio_flag	io_orig_flags;
482	enum zio_stage	io_orig_stage;
483	enum zio_stage	io_orig_pipeline;
484	enum zio_stage	io_pipeline_trace;
485	int		io_error;
486	int		io_child_error[ZIO_CHILD_TYPES];
487	uint64_t	io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
488	uint64_t	io_child_count;
489	uint64_t	io_phys_children;
490	uint64_t	io_parent_count;
491	uint64_t	*io_stall;
492	zio_t		*io_gang_leader;
493	zio_gang_node_t	*io_gang_tree;
494	void		*io_executor;
495	void		*io_waiter;
496	kmutex_t	io_lock;
497	kcondvar_t	io_cv;
498
499	/* FMA state */
500	zio_cksum_report_t *io_cksum_report;
501	uint64_t	io_ena;
502
503	/* Taskq dispatching state */
504	taskq_ent_t	io_tqent;
505
506	avl_node_t	io_trim_node;
507	list_node_t	io_trim_link;
508};
509
510extern int zio_timestamp_compare(const void *, const void *);
511
512extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
513    zio_done_func_t *done, void *priv, enum zio_flag flags);
514
515extern zio_t *zio_root(spa_t *spa,
516    zio_done_func_t *done, void *priv, enum zio_flag flags);
517
518extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
519    uint64_t size, zio_done_func_t *done, void *priv,
520    zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
521
522extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
523    void *data, uint64_t size, const zio_prop_t *zp,
524    zio_done_func_t *ready, zio_done_func_t *children_ready,
525    zio_done_func_t *physdone, zio_done_func_t *done,
526    void *priv, zio_priority_t priority, enum zio_flag flags,
527    const zbookmark_phys_t *zb);
528
529extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
530    void *data, uint64_t size, zio_done_func_t *done, void *priv,
531    zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
532
533extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
534    boolean_t nopwrite);
535
536extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
537
538extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
539    const blkptr_t *bp,
540    zio_done_func_t *done, void *priv, enum zio_flag flags);
541
542extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
543    uint64_t offset, uint64_t size, zio_done_func_t *done, void *priv,
544    zio_priority_t priority, enum zio_flag flags);
545
546extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
547    uint64_t size, void *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_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
552    uint64_t size, void *data, int checksum,
553    zio_done_func_t *done, void *priv, zio_priority_t priority,
554    enum zio_flag flags, boolean_t labels);
555
556extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
557    const blkptr_t *bp, uint64_t size, enum zio_flag flags);
558
559extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
560    blkptr_t *old_bp, uint64_t size, boolean_t *slog);
561extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
562extern void zio_flush(zio_t *zio, vdev_t *vd);
563extern zio_t *zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset,
564    uint64_t size);
565extern void zio_shrink(zio_t *zio, uint64_t size);
566
567extern int zio_wait(zio_t *zio);
568extern void zio_nowait(zio_t *zio);
569extern void zio_execute(zio_t *zio);
570extern void zio_interrupt(zio_t *zio);
571extern void zio_delay_init(zio_t *zio);
572extern void zio_delay_interrupt(zio_t *zio);
573
574extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **);
575extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **);
576extern zio_t *zio_unique_parent(zio_t *cio);
577extern void zio_add_child(zio_t *pio, zio_t *cio);
578
579extern void *zio_buf_alloc(size_t size);
580extern void *zio_buf_alloc_nowait(size_t size);
581extern void zio_buf_free(void *buf, size_t size);
582extern void *zio_data_buf_alloc(size_t size);
583extern void zio_data_buf_free(void *buf, size_t size);
584
585extern void zio_push_transform(zio_t *zio, void *data, uint64_t size,
586    uint64_t bufsize, zio_transform_func_t *transform);
587extern void zio_pop_transforms(zio_t *zio);
588
589extern void zio_resubmit_stage_async(void *);
590
591extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
592    uint64_t offset, void *data, uint64_t size, int type,
593    zio_priority_t priority, enum zio_flag flags,
594    zio_done_func_t *done, void *priv);
595
596extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
597    void *data, uint64_t size, int type, zio_priority_t priority,
598    enum zio_flag flags, zio_done_func_t *done, void *priv);
599
600extern void zio_vdev_io_bypass(zio_t *zio);
601extern void zio_vdev_io_reissue(zio_t *zio);
602extern void zio_vdev_io_redone(zio_t *zio);
603
604extern void zio_checksum_verified(zio_t *zio);
605extern int zio_worst_error(int e1, int e2);
606
607extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
608    enum zio_checksum parent);
609extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
610    enum zio_checksum child, enum zio_checksum parent);
611extern enum zio_compress zio_compress_select(spa_t *spa,
612    enum zio_compress child, enum zio_compress parent);
613
614extern void zio_suspend(spa_t *spa, zio_t *zio);
615extern int zio_resume(spa_t *spa);
616extern void zio_resume_wait(spa_t *spa);
617
618/*
619 * Initial setup and teardown.
620 */
621extern void zio_init(void);
622extern void zio_fini(void);
623
624/*
625 * Fault injection
626 */
627struct zinject_record;
628extern uint32_t zio_injection_enabled;
629extern int zio_inject_fault(char *name, int flags, int *id,
630    struct zinject_record *record);
631extern int zio_inject_list_next(int *id, char *name, size_t buflen,
632    struct zinject_record *record);
633extern int zio_clear_fault(int id);
634extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
635extern int zio_handle_fault_injection(zio_t *zio, int error);
636extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
637extern int zio_handle_label_injection(zio_t *zio, int error);
638extern void zio_handle_ignored_writes(zio_t *zio);
639extern hrtime_t zio_handle_io_delay(zio_t *zio);
640
641/*
642 * Checksum ereport functions
643 */
644extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
645    uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
646extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
647    const void *good_data, const void *bad_data, boolean_t drop_if_identical);
648
649extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
650extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
651
652/* If we have the good data in hand, this function can be used */
653extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
654    struct zio *zio, uint64_t offset, uint64_t length,
655    const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
656
657/* Called from spa_sync(), but primarily an injection handler */
658extern void spa_handle_ignored_writes(spa_t *spa);
659
660/* zbookmark_phys functions */
661boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
662    const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
663int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
664    uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
665
666#ifdef	__cplusplus
667}
668#endif
669
670#endif	/* _ZIO_H */
671