zio.h revision 288542 
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/*
129 * The meaning of "compress = on" selected by the compression features enabled
130 * on a given pool.
131 */
132#define	ZIO_COMPRESS_LEGACY_ON_VALUE	ZIO_COMPRESS_LZJB
133#define	ZIO_COMPRESS_LZ4_ON_VALUE	ZIO_COMPRESS_LZ4
134
135#define	ZIO_COMPRESS_DEFAULT		ZIO_COMPRESS_OFF
136
137#define	BOOTFS_COMPRESS_VALID(compress)			\
138	((compress) == ZIO_COMPRESS_LZJB ||		\
139	(compress) == ZIO_COMPRESS_LZ4 ||		\
140	(compress) == ZIO_COMPRESS_ON ||		\
141	(compress) == ZIO_COMPRESS_OFF)
142
143#define	ZIO_FAILURE_MODE_WAIT		0
144#define	ZIO_FAILURE_MODE_CONTINUE	1
145#define	ZIO_FAILURE_MODE_PANIC		2
146
147typedef enum zio_priority {
148	ZIO_PRIORITY_SYNC_READ,
149	ZIO_PRIORITY_SYNC_WRITE,	/* ZIL */
150	ZIO_PRIORITY_ASYNC_READ,	/* prefetch */
151	ZIO_PRIORITY_ASYNC_WRITE,	/* spa_sync() */
152	ZIO_PRIORITY_SCRUB,		/* asynchronous scrub/resilver reads */
153	ZIO_PRIORITY_TRIM,		/* free requests used for TRIM */
154	ZIO_PRIORITY_NUM_QUEUEABLE,
155
156	ZIO_PRIORITY_NOW		/* non-queued I/Os (e.g. ioctl) */
157} zio_priority_t;
158
159#define	ZIO_PIPELINE_CONTINUE		0x100
160#define	ZIO_PIPELINE_STOP		0x101
161
162enum zio_flag {
163	/*
164	 * Flags inherited by gang, ddt, and vdev children,
165	 * and that must be equal for two zios to aggregate
166	 */
167	ZIO_FLAG_DONT_AGGREGATE	= 1 << 0,
168	ZIO_FLAG_IO_REPAIR	= 1 << 1,
169	ZIO_FLAG_SELF_HEAL	= 1 << 2,
170	ZIO_FLAG_RESILVER	= 1 << 3,
171	ZIO_FLAG_SCRUB		= 1 << 4,
172	ZIO_FLAG_SCAN_THREAD	= 1 << 5,
173	ZIO_FLAG_PHYSICAL	= 1 << 6,
174
175#define	ZIO_FLAG_AGG_INHERIT	(ZIO_FLAG_CANFAIL - 1)
176
177	/*
178	 * Flags inherited by ddt, gang, and vdev children.
179	 */
180	ZIO_FLAG_CANFAIL	= 1 << 7,	/* must be first for INHERIT */
181	ZIO_FLAG_SPECULATIVE	= 1 << 8,
182	ZIO_FLAG_CONFIG_WRITER	= 1 << 9,
183	ZIO_FLAG_DONT_RETRY	= 1 << 10,
184	ZIO_FLAG_DONT_CACHE	= 1 << 11,
185	ZIO_FLAG_NODATA		= 1 << 12,
186	ZIO_FLAG_INDUCE_DAMAGE	= 1 << 13,
187
188#define	ZIO_FLAG_DDT_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
189#define	ZIO_FLAG_GANG_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
190
191	/*
192	 * Flags inherited by vdev children.
193	 */
194	ZIO_FLAG_IO_RETRY	= 1 << 14,	/* must be first for INHERIT */
195	ZIO_FLAG_PROBE		= 1 << 15,
196	ZIO_FLAG_TRYHARD	= 1 << 16,
197	ZIO_FLAG_OPTIONAL	= 1 << 17,
198
199#define	ZIO_FLAG_VDEV_INHERIT	(ZIO_FLAG_DONT_QUEUE - 1)
200
201	/*
202	 * Flags not inherited by any children.
203	 */
204	ZIO_FLAG_DONT_QUEUE	= 1 << 18,	/* must be first for INHERIT */
205	ZIO_FLAG_DONT_PROPAGATE	= 1 << 19,
206	ZIO_FLAG_IO_BYPASS	= 1 << 20,
207	ZIO_FLAG_IO_REWRITE	= 1 << 21,
208	ZIO_FLAG_RAW		= 1 << 22,
209	ZIO_FLAG_GANG_CHILD	= 1 << 23,
210	ZIO_FLAG_DDT_CHILD	= 1 << 24,
211	ZIO_FLAG_GODFATHER	= 1 << 25,
212	ZIO_FLAG_NOPWRITE	= 1 << 26,
213	ZIO_FLAG_REEXECUTED	= 1 << 27,
214	ZIO_FLAG_DELEGATED	= 1 << 28,
215};
216
217#define	ZIO_FLAG_MUSTSUCCEED		0
218
219#define	ZIO_DDT_CHILD_FLAGS(zio)				\
220	(((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |		\
221	ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
222
223#define	ZIO_GANG_CHILD_FLAGS(zio)				\
224	(((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |		\
225	ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
226
227#define	ZIO_VDEV_CHILD_FLAGS(zio)				\
228	(((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |		\
229	ZIO_FLAG_CANFAIL)
230
231enum zio_child {
232	ZIO_CHILD_VDEV = 0,
233	ZIO_CHILD_GANG,
234	ZIO_CHILD_DDT,
235	ZIO_CHILD_LOGICAL,
236	ZIO_CHILD_TYPES
237};
238
239enum zio_wait_type {
240	ZIO_WAIT_READY = 0,
241	ZIO_WAIT_DONE,
242	ZIO_WAIT_TYPES
243};
244
245/*
246 * We'll take the number 122 and 123 to indicate checksum errors and
247 * fragmentation. Those doesn't collide with any errno values as they
248 * are greater than ELAST.
249 */
250#define	ECKSUM	122
251#define	EFRAGS	123
252
253typedef void zio_done_func_t(zio_t *zio);
254
255extern const char *zio_type_name[ZIO_TYPES];
256
257/*
258 * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
259 * identifies any block in the pool.  By convention, the meta-objset (MOS)
260 * is objset 0, and the meta-dnode is object 0.  This covers all blocks
261 * except root blocks and ZIL blocks, which are defined as follows:
262 *
263 * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
264 * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
265 * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
266 *
267 * Note: this structure is called a bookmark because its original purpose
268 * was to remember where to resume a pool-wide traverse.
269 *
270 * Note: this structure is passed between userland and the kernel, and is
271 * stored on disk (by virtue of being incorporated into other on-disk
272 * structures, e.g. dsl_scan_phys_t).
273 */
274typedef struct zbookmark_phys {
275	uint64_t	zb_objset;
276	uint64_t	zb_object;
277	int64_t		zb_level;
278	uint64_t	zb_blkid;
279} zbookmark_phys_t;
280
281#define	SET_BOOKMARK(zb, objset, object, level, blkid)  \
282{                                                       \
283	(zb)->zb_objset = objset;                       \
284	(zb)->zb_object = object;                       \
285	(zb)->zb_level = level;                         \
286	(zb)->zb_blkid = blkid;                         \
287}
288
289#define	ZB_DESTROYED_OBJSET	(-1ULL)
290
291#define	ZB_ROOT_OBJECT		(0ULL)
292#define	ZB_ROOT_LEVEL		(-1LL)
293#define	ZB_ROOT_BLKID		(0ULL)
294
295#define	ZB_ZIL_OBJECT		(0ULL)
296#define	ZB_ZIL_LEVEL		(-2LL)
297
298#define	ZB_IS_ZERO(zb)						\
299	((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&	\
300	(zb)->zb_level == 0 && (zb)->zb_blkid == 0)
301#define	ZB_IS_ROOT(zb)				\
302	((zb)->zb_object == ZB_ROOT_OBJECT &&	\
303	(zb)->zb_level == ZB_ROOT_LEVEL &&	\
304	(zb)->zb_blkid == ZB_ROOT_BLKID)
305
306typedef struct zio_prop {
307	enum zio_checksum	zp_checksum;
308	enum zio_compress	zp_compress;
309	dmu_object_type_t	zp_type;
310	uint8_t			zp_level;
311	uint8_t			zp_copies;
312	boolean_t		zp_dedup;
313	boolean_t		zp_dedup_verify;
314	boolean_t		zp_nopwrite;
315} zio_prop_t;
316
317typedef struct zio_cksum_report zio_cksum_report_t;
318
319typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
320    const void *good_data);
321typedef void zio_cksum_free_f(void *cbdata, size_t size);
322
323struct zio_bad_cksum;				/* defined in zio_checksum.h */
324struct dnode_phys;
325
326struct zio_cksum_report {
327	struct zio_cksum_report *zcr_next;
328	nvlist_t		*zcr_ereport;
329	nvlist_t		*zcr_detector;
330	void			*zcr_cbdata;
331	size_t			zcr_cbinfo;	/* passed to zcr_free() */
332	uint64_t		zcr_align;
333	uint64_t		zcr_length;
334	zio_cksum_finish_f	*zcr_finish;
335	zio_cksum_free_f	*zcr_free;
336
337	/* internal use only */
338	struct zio_bad_cksum	*zcr_ckinfo;	/* information from failure */
339};
340
341typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
342    void *arg);
343
344zio_vsd_cksum_report_f	zio_vsd_default_cksum_report;
345
346typedef struct zio_vsd_ops {
347	zio_done_func_t		*vsd_free;
348	zio_vsd_cksum_report_f	*vsd_cksum_report;
349} zio_vsd_ops_t;
350
351typedef struct zio_gang_node {
352	zio_gbh_phys_t		*gn_gbh;
353	struct zio_gang_node	*gn_child[SPA_GBH_NBLKPTRS];
354} zio_gang_node_t;
355
356typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
357    zio_gang_node_t *gn, void *data);
358
359typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size);
360
361typedef struct zio_transform {
362	void			*zt_orig_data;
363	uint64_t		zt_orig_size;
364	uint64_t		zt_bufsize;
365	zio_transform_func_t	*zt_transform;
366	struct zio_transform	*zt_next;
367} zio_transform_t;
368
369typedef int zio_pipe_stage_t(zio_t *zio);
370
371/*
372 * The io_reexecute flags are distinct from io_flags because the child must
373 * be able to propagate them to the parent.  The normal io_flags are local
374 * to the zio, not protected by any lock, and not modifiable by children;
375 * the reexecute flags are protected by io_lock, modifiable by children,
376 * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
377 */
378#define	ZIO_REEXECUTE_NOW	0x01
379#define	ZIO_REEXECUTE_SUSPEND	0x02
380
381typedef struct zio_link {
382	zio_t		*zl_parent;
383	zio_t		*zl_child;
384	list_node_t	zl_parent_node;
385	list_node_t	zl_child_node;
386} zio_link_t;
387
388/*
389 * Used for TRIM kstat.
390 */
391typedef struct zio_trim_stats {
392	/*
393	 * Number of bytes successfully TRIMmed.
394	 */
395	kstat_named_t bytes;
396
397	/*
398	 * Number of successful TRIM requests.
399	 */
400	kstat_named_t success;
401
402	/*
403	 * Number of TRIM requests that failed because TRIM is not
404	 * supported.
405	 */
406	kstat_named_t unsupported;
407
408	/*
409	 * Number of TRIM requests that failed for other reasons.
410	 */
411	kstat_named_t failed;
412} zio_trim_stats_t;
413
414extern zio_trim_stats_t zio_trim_stats;
415
416#define ZIO_TRIM_STAT_INCR(stat, val) \
417	atomic_add_64(&zio_trim_stats.stat.value.ui64, (val));
418#define ZIO_TRIM_STAT_BUMP(stat) \
419	ZIO_TRIM_STAT_INCR(stat, 1);
420
421struct zio {
422	/* Core information about this I/O */
423	zbookmark_phys_t	io_bookmark;
424	zio_prop_t	io_prop;
425	zio_type_t	io_type;
426	enum zio_child	io_child_type;
427	int		io_cmd;
428	zio_priority_t	io_priority;
429	uint8_t		io_reexecute;
430	uint8_t		io_state[ZIO_WAIT_TYPES];
431	uint64_t	io_txg;
432	spa_t		*io_spa;
433	blkptr_t	*io_bp;
434	blkptr_t	*io_bp_override;
435	blkptr_t	io_bp_copy;
436	list_t		io_parent_list;
437	list_t		io_child_list;
438	zio_link_t	*io_walk_link;
439	zio_t		*io_logical;
440	zio_transform_t *io_transform_stack;
441
442	/* Callback info */
443	zio_done_func_t	*io_ready;
444	zio_done_func_t	*io_physdone;
445	zio_done_func_t	*io_done;
446	void		*io_private;
447	int64_t		io_prev_space_delta;	/* DMU private */
448	blkptr_t	io_bp_orig;
449
450	/* Data represented by this I/O */
451	void		*io_data;
452	void		*io_orig_data;
453	uint64_t	io_size;
454	uint64_t	io_orig_size;
455
456	/* Stuff for the vdev stack */
457	vdev_t		*io_vd;
458	void		*io_vsd;
459	const zio_vsd_ops_t *io_vsd_ops;
460
461	uint64_t	io_offset;
462	hrtime_t	io_timestamp;
463	avl_node_t	io_queue_node;
464
465	/* Internal pipeline state */
466	enum zio_flag	io_flags;
467	enum zio_stage	io_stage;
468	enum zio_stage	io_pipeline;
469	enum zio_flag	io_orig_flags;
470	enum zio_stage	io_orig_stage;
471	enum zio_stage	io_orig_pipeline;
472	int		io_error;
473	int		io_child_error[ZIO_CHILD_TYPES];
474	uint64_t	io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
475	uint64_t	io_child_count;
476	uint64_t	io_phys_children;
477	uint64_t	io_parent_count;
478	uint64_t	*io_stall;
479	zio_t		*io_gang_leader;
480	zio_gang_node_t	*io_gang_tree;
481	void		*io_executor;
482	void		*io_waiter;
483	kmutex_t	io_lock;
484	kcondvar_t	io_cv;
485
486	/* FMA state */
487	zio_cksum_report_t *io_cksum_report;
488	uint64_t	io_ena;
489
490	/* Taskq dispatching state */
491	taskq_ent_t	io_tqent;
492
493	avl_node_t	io_trim_node;
494	list_node_t	io_trim_link;
495};
496
497extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
498    zio_done_func_t *done, void *priv, enum zio_flag flags);
499
500extern zio_t *zio_root(spa_t *spa,
501    zio_done_func_t *done, void *priv, enum zio_flag flags);
502
503extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
504    uint64_t size, zio_done_func_t *done, void *priv,
505    zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
506
507extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
508    void *data, uint64_t size, const zio_prop_t *zp,
509    zio_done_func_t *ready, zio_done_func_t *physdone, zio_done_func_t *done,
510    void *priv,
511    zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
512
513extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
514    void *data, uint64_t size, zio_done_func_t *done, void *priv,
515    zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
516
517extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
518    boolean_t nopwrite);
519
520extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
521
522extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
523    const blkptr_t *bp,
524    zio_done_func_t *done, void *priv, enum zio_flag flags);
525
526extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
527    uint64_t offset, uint64_t size, zio_done_func_t *done, void *priv,
528    zio_priority_t priority, enum zio_flag flags);
529
530extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
531    uint64_t size, void *data, int checksum,
532    zio_done_func_t *done, void *priv, zio_priority_t priority,
533    enum zio_flag flags, boolean_t labels);
534
535extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
536    uint64_t size, void *data, int checksum,
537    zio_done_func_t *done, void *priv, zio_priority_t priority,
538    enum zio_flag flags, boolean_t labels);
539
540extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
541    const blkptr_t *bp, uint64_t size, enum zio_flag flags);
542
543extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
544    blkptr_t *old_bp, uint64_t size, boolean_t use_slog);
545extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
546extern void zio_flush(zio_t *zio, vdev_t *vd);
547extern zio_t *zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset,
548    uint64_t size);
549extern void zio_shrink(zio_t *zio, uint64_t size);
550
551extern int zio_wait(zio_t *zio);
552extern void zio_nowait(zio_t *zio);
553extern void zio_execute(zio_t *zio);
554extern void zio_interrupt(zio_t *zio);
555
556extern zio_t *zio_walk_parents(zio_t *cio);
557extern zio_t *zio_walk_children(zio_t *pio);
558extern zio_t *zio_unique_parent(zio_t *cio);
559extern void zio_add_child(zio_t *pio, zio_t *cio);
560
561extern void *zio_buf_alloc(size_t size);
562extern void zio_buf_free(void *buf, size_t size);
563extern void *zio_data_buf_alloc(size_t size);
564extern void zio_data_buf_free(void *buf, size_t size);
565
566extern void zio_resubmit_stage_async(void *);
567
568extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
569    uint64_t offset, void *data, uint64_t size, int type,
570    zio_priority_t priority, enum zio_flag flags,
571    zio_done_func_t *done, void *priv);
572
573extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
574    void *data, uint64_t size, int type, zio_priority_t priority,
575    enum zio_flag flags, zio_done_func_t *done, void *priv);
576
577extern void zio_vdev_io_bypass(zio_t *zio);
578extern void zio_vdev_io_reissue(zio_t *zio);
579extern void zio_vdev_io_redone(zio_t *zio);
580
581extern void zio_checksum_verified(zio_t *zio);
582extern int zio_worst_error(int e1, int e2);
583
584extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
585    enum zio_checksum parent);
586extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
587    enum zio_checksum child, enum zio_checksum parent);
588extern enum zio_compress zio_compress_select(spa_t *spa,
589    enum zio_compress child, enum zio_compress parent);
590
591extern void zio_suspend(spa_t *spa, zio_t *zio);
592extern int zio_resume(spa_t *spa);
593extern void zio_resume_wait(spa_t *spa);
594
595/*
596 * Initial setup and teardown.
597 */
598extern void zio_init(void);
599extern void zio_fini(void);
600
601/*
602 * Fault injection
603 */
604struct zinject_record;
605extern uint32_t zio_injection_enabled;
606extern int zio_inject_fault(char *name, int flags, int *id,
607    struct zinject_record *record);
608extern int zio_inject_list_next(int *id, char *name, size_t buflen,
609    struct zinject_record *record);
610extern int zio_clear_fault(int id);
611extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
612extern int zio_handle_fault_injection(zio_t *zio, int error);
613extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
614extern int zio_handle_label_injection(zio_t *zio, int error);
615extern void zio_handle_ignored_writes(zio_t *zio);
616extern uint64_t zio_handle_io_delay(zio_t *zio);
617
618/*
619 * Checksum ereport functions
620 */
621extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
622    uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
623extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
624    const void *good_data, const void *bad_data, boolean_t drop_if_identical);
625
626extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
627extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
628
629/* If we have the good data in hand, this function can be used */
630extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
631    struct zio *zio, uint64_t offset, uint64_t length,
632    const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
633
634/* Called from spa_sync(), but primarily an injection handler */
635extern void spa_handle_ignored_writes(spa_t *spa);
636
637/* zbookmark_phys functions */
638boolean_t zbookmark_is_before(const struct dnode_phys *dnp,
639    const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
640
641#ifdef	__cplusplus
642}
643#endif
644
645#endif	/* _ZIO_H */
646