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 by Delphix. All rights reserved.
26  * Copyright (c) 2013 by Delphix. All rights reserved.
27  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
28  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
29  */
30 
31 #ifndef _ZIO_H
32 #define	_ZIO_H
33 
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
43 extern "C" {
44 #endif
45 
46 /*
47  * Embedded checksum
48  */
49 #define	ZEC_MAGIC	0x210da7ab10c7a11ULL
50 
51 typedef 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 
68 typedef 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 
74 enum 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_FUNCTIONS
87 };
88 
89 #define	ZIO_CHECKSUM_ON_VALUE	ZIO_CHECKSUM_FLETCHER_4
90 #define	ZIO_CHECKSUM_DEFAULT	ZIO_CHECKSUM_ON
91 
92 #define	ZIO_CHECKSUM_MASK	0xffULL
93 #define	ZIO_CHECKSUM_VERIFY	(1 << 8)
94 
95 #define	ZIO_DEDUPCHECKSUM	ZIO_CHECKSUM_SHA256
96 #define	ZIO_DEDUPDITTO_MIN	100
97 
98 enum zio_compress {
99 	ZIO_COMPRESS_INHERIT = 0,
100 	ZIO_COMPRESS_ON,
101 	ZIO_COMPRESS_OFF,
102 	ZIO_COMPRESS_LZJB,
103 	ZIO_COMPRESS_EMPTY,
104 	ZIO_COMPRESS_GZIP_1,
105 	ZIO_COMPRESS_GZIP_2,
106 	ZIO_COMPRESS_GZIP_3,
107 	ZIO_COMPRESS_GZIP_4,
108 	ZIO_COMPRESS_GZIP_5,
109 	ZIO_COMPRESS_GZIP_6,
110 	ZIO_COMPRESS_GZIP_7,
111 	ZIO_COMPRESS_GZIP_8,
112 	ZIO_COMPRESS_GZIP_9,
113 	ZIO_COMPRESS_ZLE,
114 	ZIO_COMPRESS_LZ4,
115 	ZIO_COMPRESS_FUNCTIONS
116 };
117 
118 /* N.B. when altering this value, also change BOOTFS_COMPRESS_VALID below */
119 #define	ZIO_COMPRESS_ON_VALUE	ZIO_COMPRESS_LZJB
120 #define	ZIO_COMPRESS_DEFAULT	ZIO_COMPRESS_OFF
121 
122 #define	BOOTFS_COMPRESS_VALID(compress)			\
123 	((compress) == ZIO_COMPRESS_LZJB ||		\
124 	(compress) == ZIO_COMPRESS_LZ4 ||		\
125 	((compress) == ZIO_COMPRESS_ON &&		\
126 	ZIO_COMPRESS_ON_VALUE == ZIO_COMPRESS_LZJB) ||	\
127 	(compress) == ZIO_COMPRESS_OFF)
128 
129 #define	ZIO_FAILURE_MODE_WAIT		0
130 #define	ZIO_FAILURE_MODE_CONTINUE	1
131 #define	ZIO_FAILURE_MODE_PANIC		2
132 
133 typedef enum zio_priority {
134 	ZIO_PRIORITY_SYNC_READ,
135 	ZIO_PRIORITY_SYNC_WRITE,	/* ZIL */
136 	ZIO_PRIORITY_ASYNC_READ,	/* prefetch */
137 	ZIO_PRIORITY_ASYNC_WRITE,	/* spa_sync() */
138 	ZIO_PRIORITY_SCRUB,		/* asynchronous scrub/resilver reads */
139 	ZIO_PRIORITY_NUM_QUEUEABLE,
140 
141 	ZIO_PRIORITY_NOW		/* non-queued i/os (e.g. free) */
142 } zio_priority_t;
143 
144 #define	ZIO_PIPELINE_CONTINUE		0x100
145 #define	ZIO_PIPELINE_STOP		0x101
146 
147 enum zio_flag {
148 	/*
149 	 * Flags inherited by gang, ddt, and vdev children,
150 	 * and that must be equal for two zios to aggregate
151 	 */
152 	ZIO_FLAG_DONT_AGGREGATE	= 1 << 0,
153 	ZIO_FLAG_IO_REPAIR	= 1 << 1,
154 	ZIO_FLAG_SELF_HEAL	= 1 << 2,
155 	ZIO_FLAG_RESILVER	= 1 << 3,
156 	ZIO_FLAG_SCRUB		= 1 << 4,
157 	ZIO_FLAG_SCAN_THREAD	= 1 << 5,
158 
159 #define	ZIO_FLAG_AGG_INHERIT	(ZIO_FLAG_CANFAIL - 1)
160 
161 	/*
162 	 * Flags inherited by ddt, gang, and vdev children.
163 	 */
164 	ZIO_FLAG_CANFAIL	= 1 << 6,	/* must be first for INHERIT */
165 	ZIO_FLAG_SPECULATIVE	= 1 << 7,
166 	ZIO_FLAG_CONFIG_WRITER	= 1 << 8,
167 	ZIO_FLAG_DONT_RETRY	= 1 << 9,
168 	ZIO_FLAG_DONT_CACHE	= 1 << 10,
169 	ZIO_FLAG_NODATA		= 1 << 11,
170 	ZIO_FLAG_INDUCE_DAMAGE	= 1 << 12,
171 
172 #define	ZIO_FLAG_DDT_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
173 #define	ZIO_FLAG_GANG_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
174 
175 	/*
176 	 * Flags inherited by vdev children.
177 	 */
178 	ZIO_FLAG_IO_RETRY	= 1 << 13,	/* must be first for INHERIT */
179 	ZIO_FLAG_PROBE		= 1 << 14,
180 	ZIO_FLAG_TRYHARD	= 1 << 15,
181 	ZIO_FLAG_OPTIONAL	= 1 << 16,
182 
183 #define	ZIO_FLAG_VDEV_INHERIT	(ZIO_FLAG_DONT_QUEUE - 1)
184 
185 	/*
186 	 * Flags not inherited by any children.
187 	 */
188 	ZIO_FLAG_DONT_QUEUE	= 1 << 17,	/* must be first for INHERIT */
189 	ZIO_FLAG_DONT_PROPAGATE	= 1 << 18,
190 	ZIO_FLAG_IO_BYPASS	= 1 << 19,
191 	ZIO_FLAG_IO_REWRITE	= 1 << 20,
192 	ZIO_FLAG_RAW		= 1 << 21,
193 	ZIO_FLAG_GANG_CHILD	= 1 << 22,
194 	ZIO_FLAG_DDT_CHILD	= 1 << 23,
195 	ZIO_FLAG_GODFATHER	= 1 << 24,
196 	ZIO_FLAG_NOPWRITE	= 1 << 25,
197 	ZIO_FLAG_REEXECUTED	= 1 << 26,
198 	ZIO_FLAG_DELEGATED	= 1 << 27,
199 };
200 
201 #define	ZIO_FLAG_MUSTSUCCEED		0
202 
203 #define	ZIO_DDT_CHILD_FLAGS(zio)				\
204 	(((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |		\
205 	ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
206 
207 #define	ZIO_GANG_CHILD_FLAGS(zio)				\
208 	(((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |		\
209 	ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
210 
211 #define	ZIO_VDEV_CHILD_FLAGS(zio)				\
212 	(((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |		\
213 	ZIO_FLAG_CANFAIL)
214 
215 enum zio_child {
216 	ZIO_CHILD_VDEV = 0,
217 	ZIO_CHILD_GANG,
218 	ZIO_CHILD_DDT,
219 	ZIO_CHILD_LOGICAL,
220 	ZIO_CHILD_TYPES
221 };
222 
223 enum 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 
237 typedef void zio_done_func_t(zio_t *zio);
238 
239 extern const char *zio_type_name[ZIO_TYPES];
240 
241 /*
242  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
243  * identifies any block in the pool.  By convention, the meta-objset (MOS)
244  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
245  * except root blocks and ZIL blocks, which are defined as follows:
246  *
247  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
248  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
249  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
250  *
251  * Note: this structure is called a bookmark because its original purpose
252  * was to remember where to resume a pool-wide traverse.
253  *
254  * Note: this structure is passed between userland and the kernel.
255  * Therefore it must not change size or alignment between 32/64 bit
256  * compilation options.
257  */
258 typedef struct zbookmark {
259 	uint64_t	zb_objset;
260 	uint64_t	zb_object;
261 	int64_t		zb_level;
262 	uint64_t	zb_blkid;
263 } zbookmark_t;
264 
265 #define	SET_BOOKMARK(zb, objset, object, level, blkid)  \
266 {                                                       \
267 	(zb)->zb_objset = objset;                       \
268 	(zb)->zb_object = object;                       \
269 	(zb)->zb_level = level;                         \
270 	(zb)->zb_blkid = blkid;                         \
271 }
272 
273 #define	ZB_DESTROYED_OBJSET	(-1ULL)
274 
275 #define	ZB_ROOT_OBJECT		(0ULL)
276 #define	ZB_ROOT_LEVEL		(-1LL)
277 #define	ZB_ROOT_BLKID		(0ULL)
278 
279 #define	ZB_ZIL_OBJECT		(0ULL)
280 #define	ZB_ZIL_LEVEL		(-2LL)
281 
282 #define	ZB_IS_ZERO(zb)						\
283 	((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&	\
284 	(zb)->zb_level == 0 && (zb)->zb_blkid == 0)
285 #define	ZB_IS_ROOT(zb)				\
286 	((zb)->zb_object == ZB_ROOT_OBJECT &&	\
287 	(zb)->zb_level == ZB_ROOT_LEVEL &&	\
288 	(zb)->zb_blkid == ZB_ROOT_BLKID)
289 
290 typedef struct zio_prop {
291 	enum zio_checksum	zp_checksum;
292 	enum zio_compress	zp_compress;
293 	dmu_object_type_t	zp_type;
294 	uint8_t			zp_level;
295 	uint8_t			zp_copies;
296 	boolean_t		zp_dedup;
297 	boolean_t		zp_dedup_verify;
298 	boolean_t		zp_nopwrite;
299 } zio_prop_t;
300 
301 typedef struct zio_cksum_report zio_cksum_report_t;
302 
303 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
304     const void *good_data);
305 typedef void zio_cksum_free_f(void *cbdata, size_t size);
306 
307 struct zio_bad_cksum;				/* defined in zio_checksum.h */
308 struct dnode_phys;
309 
310 struct zio_cksum_report {
311 	struct zio_cksum_report *zcr_next;
312 	nvlist_t		*zcr_ereport;
313 	nvlist_t		*zcr_detector;
314 	void			*zcr_cbdata;
315 	size_t			zcr_cbinfo;	/* passed to zcr_free() */
316 	uint64_t		zcr_align;
317 	uint64_t		zcr_length;
318 	zio_cksum_finish_f	*zcr_finish;
319 	zio_cksum_free_f	*zcr_free;
320 
321 	/* internal use only */
322 	struct zio_bad_cksum	*zcr_ckinfo;	/* information from failure */
323 };
324 
325 typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
326     void *arg);
327 
328 zio_vsd_cksum_report_f	zio_vsd_default_cksum_report;
329 
330 typedef struct zio_vsd_ops {
331 	zio_done_func_t		*vsd_free;
332 	zio_vsd_cksum_report_f	*vsd_cksum_report;
333 } zio_vsd_ops_t;
334 
335 typedef struct zio_gang_node {
336 	zio_gbh_phys_t		*gn_gbh;
337 	struct zio_gang_node	*gn_child[SPA_GBH_NBLKPTRS];
338 } zio_gang_node_t;
339 
340 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
341     zio_gang_node_t *gn, void *data);
342 
343 typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size);
344 
345 typedef struct zio_transform {
346 	void			*zt_orig_data;
347 	uint64_t		zt_orig_size;
348 	uint64_t		zt_bufsize;
349 	zio_transform_func_t	*zt_transform;
350 	struct zio_transform	*zt_next;
351 } zio_transform_t;
352 
353 typedef int zio_pipe_stage_t(zio_t **ziop);
354 
355 /*
356  * The io_reexecute flags are distinct from io_flags because the child must
357  * be able to propagate them to the parent.  The normal io_flags are local
358  * to the zio, not protected by any lock, and not modifiable by children;
359  * the reexecute flags are protected by io_lock, modifiable by children,
360  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
361  */
362 #define	ZIO_REEXECUTE_NOW	0x01
363 #define	ZIO_REEXECUTE_SUSPEND	0x02
364 
365 typedef struct zio_link {
366 	zio_t		*zl_parent;
367 	zio_t		*zl_child;
368 	list_node_t	zl_parent_node;
369 	list_node_t	zl_child_node;
370 } zio_link_t;
371 
372 /*
373  * Used for TRIM kstat.
374  */
375 typedef struct zio_trim_stats {
376 	/*
377 	 * Number of bytes successfully TRIMmed.
378 	 */
379 	kstat_named_t bytes;
380 
381 	/*
382 	 * Number of successful TRIM requests.
383 	 */
384 	kstat_named_t success;
385 
386 	/*
387 	 * Number of TRIM requests that failed because TRIM is not
388 	 * supported.
389 	 */
390 	kstat_named_t unsupported;
391 
392 	/*
393 	 * Number of TRIM requests that failed for other reasons.
394 	 */
395 	kstat_named_t failed;
396 } zio_trim_stats_t;
397 
398 extern zio_trim_stats_t zio_trim_stats;
399 
400 #define ZIO_TRIM_STAT_INCR(stat, val) \
401 	atomic_add_64(&zio_trim_stats.stat.value.ui64, (val));
402 #define ZIO_TRIM_STAT_BUMP(stat) \
403 	ZIO_TRIM_STAT_INCR(stat, 1);
404 
405 struct zio {
406 	/* Core information about this I/O */
407 	zbookmark_t	io_bookmark;
408 	zio_prop_t	io_prop;
409 	zio_type_t	io_type;
410 	enum zio_child	io_child_type;
411 	int		io_cmd;
412 	zio_priority_t	io_priority;
413 	uint8_t		io_reexecute;
414 	uint8_t		io_state[ZIO_WAIT_TYPES];
415 	uint64_t	io_txg;
416 	spa_t		*io_spa;
417 	blkptr_t	*io_bp;
418 	blkptr_t	*io_bp_override;
419 	blkptr_t	io_bp_copy;
420 	list_t		io_parent_list;
421 	list_t		io_child_list;
422 	zio_link_t	*io_walk_link;
423 	zio_t		*io_logical;
424 	zio_transform_t *io_transform_stack;
425 
426 	/* Callback info */
427 	zio_done_func_t	*io_ready;
428 	zio_done_func_t	*io_physdone;
429 	zio_done_func_t	*io_done;
430 	void		*io_private;
431 	int64_t		io_prev_space_delta;	/* DMU private */
432 	blkptr_t	io_bp_orig;
433 
434 	/* Data represented by this I/O */
435 	void		*io_data;
436 	void		*io_orig_data;
437 	uint64_t	io_size;
438 	uint64_t	io_orig_size;
439 
440 	/* Stuff for the vdev stack */
441 	vdev_t		*io_vd;
442 	void		*io_vsd;
443 	const zio_vsd_ops_t *io_vsd_ops;
444 
445 	uint64_t	io_offset;
446 	hrtime_t	io_timestamp;
447 	avl_node_t	io_queue_node;
448 
449 	/* Internal pipeline state */
450 	enum zio_flag	io_flags;
451 	enum zio_stage	io_stage;
452 	enum zio_stage	io_pipeline;
453 	enum zio_flag	io_orig_flags;
454 	enum zio_stage	io_orig_stage;
455 	enum zio_stage	io_orig_pipeline;
456 	int		io_error;
457 	int		io_child_error[ZIO_CHILD_TYPES];
458 	uint64_t	io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
459 	uint64_t	io_child_count;
460 	uint64_t	io_phys_children;
461 	uint64_t	io_parent_count;
462 	uint64_t	*io_stall;
463 	zio_t		*io_gang_leader;
464 	zio_gang_node_t	*io_gang_tree;
465 	void		*io_executor;
466 	void		*io_waiter;
467 	kmutex_t	io_lock;
468 	kcondvar_t	io_cv;
469 
470 	/* FMA state */
471 	zio_cksum_report_t *io_cksum_report;
472 	uint64_t	io_ena;
473 
474 	/* Taskq dispatching state */
475 	taskq_ent_t	io_tqent;
476 
477 	avl_node_t	io_trim_node;
478 	list_node_t	io_trim_link;
479 };
480 
481 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
482     zio_done_func_t *done, void *priv, enum zio_flag flags);
483 
484 extern zio_t *zio_root(spa_t *spa,
485     zio_done_func_t *done, void *priv, enum zio_flag flags);
486 
487 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
488     uint64_t size, zio_done_func_t *done, void *priv,
489     zio_priority_t priority, enum zio_flag flags, const zbookmark_t *zb);
490 
491 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
492     void *data, uint64_t size, const zio_prop_t *zp,
493     zio_done_func_t *ready, zio_done_func_t *physdone, zio_done_func_t *done,
494     void *priv,
495     zio_priority_t priority, enum zio_flag flags, const zbookmark_t *zb);
496 
497 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
498     void *data, uint64_t size, zio_done_func_t *done, void *priv,
499     zio_priority_t priority, enum zio_flag flags, zbookmark_t *zb);
500 
501 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
502     boolean_t nopwrite);
503 
504 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
505 
506 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
507     const blkptr_t *bp,
508     zio_done_func_t *done, void *priv, enum zio_flag flags);
509 
510 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
511     uint64_t offset, uint64_t size, zio_done_func_t *done, void *priv,
512     enum zio_flag flags);
513 
514 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
515     uint64_t size, void *data, int checksum,
516     zio_done_func_t *done, void *priv, zio_priority_t priority,
517     enum zio_flag flags, boolean_t labels);
518 
519 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
520     uint64_t size, void *data, int checksum,
521     zio_done_func_t *done, void *priv, zio_priority_t priority,
522     enum zio_flag flags, boolean_t labels);
523 
524 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
525     const blkptr_t *bp, uint64_t size, enum zio_flag flags);
526 
527 extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
528     blkptr_t *old_bp, uint64_t size, boolean_t use_slog);
529 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
530 extern void zio_flush(zio_t *zio, vdev_t *vd);
531 extern zio_t *zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset,
532     uint64_t size);
533 extern void zio_shrink(zio_t *zio, uint64_t size);
534 
535 extern int zio_wait(zio_t *zio);
536 extern void zio_nowait(zio_t *zio);
537 extern void zio_execute(zio_t *zio);
538 extern void zio_interrupt(zio_t *zio);
539 
540 extern zio_t *zio_walk_parents(zio_t *cio);
541 extern zio_t *zio_walk_children(zio_t *pio);
542 extern zio_t *zio_unique_parent(zio_t *cio);
543 extern void zio_add_child(zio_t *pio, zio_t *cio);
544 
545 extern void *zio_buf_alloc(size_t size);
546 extern void zio_buf_free(void *buf, size_t size);
547 extern void *zio_data_buf_alloc(size_t size);
548 extern void zio_data_buf_free(void *buf, size_t size);
549 
550 extern void zio_resubmit_stage_async(void *);
551 
552 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
553     uint64_t offset, void *data, uint64_t size, int type,
554     zio_priority_t priority, enum zio_flag flags,
555     zio_done_func_t *done, void *priv);
556 
557 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
558     void *data, uint64_t size, int type, zio_priority_t priority,
559     enum zio_flag flags, zio_done_func_t *done, void *priv);
560 
561 extern void zio_vdev_io_bypass(zio_t *zio);
562 extern void zio_vdev_io_reissue(zio_t *zio);
563 extern void zio_vdev_io_redone(zio_t *zio);
564 
565 extern void zio_checksum_verified(zio_t *zio);
566 extern int zio_worst_error(int e1, int e2);
567 
568 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
569     enum zio_checksum parent);
570 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
571     enum zio_checksum child, enum zio_checksum parent);
572 extern enum zio_compress zio_compress_select(enum zio_compress child,
573     enum zio_compress parent);
574 
575 extern void zio_suspend(spa_t *spa, zio_t *zio);
576 extern int zio_resume(spa_t *spa);
577 extern void zio_resume_wait(spa_t *spa);
578 
579 /*
580  * Initial setup and teardown.
581  */
582 extern void zio_init(void);
583 extern void zio_fini(void);
584 
585 /*
586  * Fault injection
587  */
588 struct zinject_record;
589 extern uint32_t zio_injection_enabled;
590 extern int zio_inject_fault(char *name, int flags, int *id,
591     struct zinject_record *record);
592 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
593     struct zinject_record *record);
594 extern int zio_clear_fault(int id);
595 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
596 extern int zio_handle_fault_injection(zio_t *zio, int error);
597 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
598 extern int zio_handle_label_injection(zio_t *zio, int error);
599 extern void zio_handle_ignored_writes(zio_t *zio);
600 extern uint64_t zio_handle_io_delay(zio_t *zio);
601 
602 /*
603  * Checksum ereport functions
604  */
605 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
606     uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
607 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
608     const void *good_data, const void *bad_data, boolean_t drop_if_identical);
609 
610 extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
611 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
612 
613 /* If we have the good data in hand, this function can be used */
614 extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
615     struct zio *zio, uint64_t offset, uint64_t length,
616     const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
617 
618 /* Called from spa_sync(), but primarily an injection handler */
619 extern void spa_handle_ignored_writes(spa_t *spa);
620 
621 /* zbookmark functions */
622 boolean_t zbookmark_is_before(const struct dnode_phys *dnp,
623     const zbookmark_t *zb1, const zbookmark_t *zb2);
624 
625 #ifdef	__cplusplus
626 }
627 #endif
628 
629 #endif	/* _ZIO_H */
630