xref: /illumos-gate/usr/src/uts/common/fs/zfs/sys/zio.h (revision c5ee46810f82e8a53d2cc5a487568a573f449039)
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, 2017 by Delphix. All rights reserved.
26  * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
27  * Copyright (c) 2013, Joyent, Inc. All rights reserved.
28  * Copyright 2016 Toomas Soome <tsoome@me.com>
29  */
30 
31 #ifndef _ZIO_H
32 #define	_ZIO_H
33 
34 #include <sys/zio_priority.h>
35 #include <sys/zfs_context.h>
36 #include <sys/spa.h>
37 #include <sys/txg.h>
38 #include <sys/avl.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_SHA512,
87 	ZIO_CHECKSUM_SKEIN,
88 	ZIO_CHECKSUM_EDONR,
89 	ZIO_CHECKSUM_FUNCTIONS
90 };
91 
92 /*
93  * The number of "legacy" compression functions which can be set on individual
94  * objects.
95  */
96 #define	ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2
97 
98 #define	ZIO_CHECKSUM_ON_VALUE	ZIO_CHECKSUM_FLETCHER_4
99 #define	ZIO_CHECKSUM_DEFAULT	ZIO_CHECKSUM_ON
100 
101 #define	ZIO_CHECKSUM_MASK	0xffULL
102 #define	ZIO_CHECKSUM_VERIFY	(1 << 8)
103 
104 #define	ZIO_DEDUPCHECKSUM	ZIO_CHECKSUM_SHA256
105 #define	ZIO_DEDUPDITTO_MIN	100
106 
107 /*
108  * The number of "legacy" compression functions which can be set on individual
109  * objects.
110  */
111 #define	ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
112 
113 /*
114  * The meaning of "compress = on" selected by the compression features enabled
115  * on a given pool.
116  */
117 #define	ZIO_COMPRESS_LEGACY_ON_VALUE	ZIO_COMPRESS_LZJB
118 #define	ZIO_COMPRESS_LZ4_ON_VALUE	ZIO_COMPRESS_LZ4
119 
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_GZIP_1 ||		\
126 	(compress) == ZIO_COMPRESS_GZIP_2 ||		\
127 	(compress) == ZIO_COMPRESS_GZIP_3 ||		\
128 	(compress) == ZIO_COMPRESS_GZIP_4 ||		\
129 	(compress) == ZIO_COMPRESS_GZIP_5 ||		\
130 	(compress) == ZIO_COMPRESS_GZIP_6 ||		\
131 	(compress) == ZIO_COMPRESS_GZIP_7 ||		\
132 	(compress) == ZIO_COMPRESS_GZIP_8 ||		\
133 	(compress) == ZIO_COMPRESS_GZIP_9 ||		\
134 	(compress) == ZIO_COMPRESS_ON ||		\
135 	(compress) == ZIO_COMPRESS_OFF)
136 
137 #define	ZIO_FAILURE_MODE_WAIT		0
138 #define	ZIO_FAILURE_MODE_CONTINUE	1
139 #define	ZIO_FAILURE_MODE_PANIC		2
140 
141 enum zio_flag {
142 	/*
143 	 * Flags inherited by gang, ddt, and vdev children,
144 	 * and that must be equal for two zios to aggregate
145 	 */
146 	ZIO_FLAG_DONT_AGGREGATE	= 1 << 0,
147 	ZIO_FLAG_IO_REPAIR	= 1 << 1,
148 	ZIO_FLAG_SELF_HEAL	= 1 << 2,
149 	ZIO_FLAG_RESILVER	= 1 << 3,
150 	ZIO_FLAG_SCRUB		= 1 << 4,
151 	ZIO_FLAG_SCAN_THREAD	= 1 << 5,
152 	ZIO_FLAG_PHYSICAL	= 1 << 6,
153 
154 #define	ZIO_FLAG_AGG_INHERIT	(ZIO_FLAG_CANFAIL - 1)
155 
156 	/*
157 	 * Flags inherited by ddt, gang, and vdev children.
158 	 */
159 	ZIO_FLAG_CANFAIL	= 1 << 7,	/* must be first for INHERIT */
160 	ZIO_FLAG_SPECULATIVE	= 1 << 8,
161 	ZIO_FLAG_CONFIG_WRITER	= 1 << 9,
162 	ZIO_FLAG_DONT_RETRY	= 1 << 10,
163 	ZIO_FLAG_DONT_CACHE	= 1 << 11,
164 	ZIO_FLAG_NODATA		= 1 << 12,
165 	ZIO_FLAG_INDUCE_DAMAGE	= 1 << 13,
166 	ZIO_FLAG_IO_ALLOCATING	= 1 << 14,
167 
168 #define	ZIO_FLAG_DDT_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
169 #define	ZIO_FLAG_GANG_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
170 
171 	/*
172 	 * Flags inherited by vdev children.
173 	 */
174 	ZIO_FLAG_IO_RETRY	= 1 << 15,	/* must be first for INHERIT */
175 	ZIO_FLAG_PROBE		= 1 << 16,
176 	ZIO_FLAG_TRYHARD	= 1 << 17,
177 	ZIO_FLAG_OPTIONAL	= 1 << 18,
178 
179 #define	ZIO_FLAG_VDEV_INHERIT	(ZIO_FLAG_DONT_QUEUE - 1)
180 
181 	/*
182 	 * Flags not inherited by any children.
183 	 */
184 	ZIO_FLAG_DONT_QUEUE	= 1 << 19,	/* must be first for INHERIT */
185 	ZIO_FLAG_DONT_PROPAGATE	= 1 << 20,
186 	ZIO_FLAG_IO_BYPASS	= 1 << 21,
187 	ZIO_FLAG_IO_REWRITE	= 1 << 22,
188 	ZIO_FLAG_RAW		= 1 << 23,
189 	ZIO_FLAG_GANG_CHILD	= 1 << 24,
190 	ZIO_FLAG_DDT_CHILD	= 1 << 25,
191 	ZIO_FLAG_GODFATHER	= 1 << 26,
192 	ZIO_FLAG_NOPWRITE	= 1 << 27,
193 	ZIO_FLAG_REEXECUTED	= 1 << 28,
194 	ZIO_FLAG_DELEGATED	= 1 << 29,
195 };
196 
197 #define	ZIO_FLAG_MUSTSUCCEED		0
198 
199 #define	ZIO_DDT_CHILD_FLAGS(zio)				\
200 	(((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |		\
201 	ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
202 
203 #define	ZIO_GANG_CHILD_FLAGS(zio)				\
204 	(((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |		\
205 	ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
206 
207 #define	ZIO_VDEV_CHILD_FLAGS(zio)				\
208 	(((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |		\
209 	ZIO_FLAG_CANFAIL)
210 
211 enum zio_child {
212 	ZIO_CHILD_VDEV = 0,
213 	ZIO_CHILD_GANG,
214 	ZIO_CHILD_DDT,
215 	ZIO_CHILD_LOGICAL,
216 	ZIO_CHILD_TYPES
217 };
218 
219 enum zio_wait_type {
220 	ZIO_WAIT_READY = 0,
221 	ZIO_WAIT_DONE,
222 	ZIO_WAIT_TYPES
223 };
224 
225 /*
226  * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent
227  * graveyard) to indicate checksum errors and fragmentation.
228  */
229 #define	ECKSUM	EBADE
230 #define	EFRAGS	EBADR
231 
232 typedef void zio_done_func_t(zio_t *zio);
233 
234 extern boolean_t zio_dva_throttle_enabled;
235 extern const char *zio_type_name[ZIO_TYPES];
236 
237 /*
238  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
239  * identifies any block in the pool.  By convention, the meta-objset (MOS)
240  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
241  * except root blocks and ZIL blocks, which are defined as follows:
242  *
243  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
244  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
245  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
246  * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
247  *
248  * Note: this structure is called a bookmark because its original purpose
249  * was to remember where to resume a pool-wide traverse.
250  *
251  * Note: this structure is passed between userland and the kernel, and is
252  * stored on disk (by virtue of being incorporated into other on-disk
253  * structures, e.g. dsl_scan_phys_t).
254  */
255 typedef struct zbookmark_phys {
256 	uint64_t	zb_objset;
257 	uint64_t	zb_object;
258 	int64_t		zb_level;
259 	uint64_t	zb_blkid;
260 } zbookmark_phys_t;
261 
262 #define	SET_BOOKMARK(zb, objset, object, level, blkid)  \
263 {                                                       \
264 	(zb)->zb_objset = objset;                       \
265 	(zb)->zb_object = object;                       \
266 	(zb)->zb_level = level;                         \
267 	(zb)->zb_blkid = blkid;                         \
268 }
269 
270 #define	ZB_DESTROYED_OBJSET	(-1ULL)
271 
272 #define	ZB_ROOT_OBJECT		(0ULL)
273 #define	ZB_ROOT_LEVEL		(-1LL)
274 #define	ZB_ROOT_BLKID		(0ULL)
275 
276 #define	ZB_ZIL_OBJECT		(0ULL)
277 #define	ZB_ZIL_LEVEL		(-2LL)
278 
279 #define	ZB_DNODE_LEVEL		(-3LL)
280 #define	ZB_DNODE_BLKID		(0ULL)
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 struct abd;
310 
311 struct zio_cksum_report {
312 	struct zio_cksum_report *zcr_next;
313 	nvlist_t		*zcr_ereport;
314 	nvlist_t		*zcr_detector;
315 	void			*zcr_cbdata;
316 	size_t			zcr_cbinfo;	/* passed to zcr_free() */
317 	uint64_t		zcr_align;
318 	uint64_t		zcr_length;
319 	zio_cksum_finish_f	*zcr_finish;
320 	zio_cksum_free_f	*zcr_free;
321 
322 	/* internal use only */
323 	struct zio_bad_cksum	*zcr_ckinfo;	/* information from failure */
324 };
325 
326 typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
327     void *arg);
328 
329 zio_vsd_cksum_report_f	zio_vsd_default_cksum_report;
330 
331 typedef struct zio_vsd_ops {
332 	zio_done_func_t		*vsd_free;
333 	zio_vsd_cksum_report_f	*vsd_cksum_report;
334 } zio_vsd_ops_t;
335 
336 typedef struct zio_gang_node {
337 	zio_gbh_phys_t		*gn_gbh;
338 	struct zio_gang_node	*gn_child[SPA_GBH_NBLKPTRS];
339 } zio_gang_node_t;
340 
341 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
342     zio_gang_node_t *gn, struct abd *data, uint64_t offset);
343 
344 typedef void zio_transform_func_t(zio_t *zio, struct abd *data, uint64_t size);
345 
346 typedef struct zio_transform {
347 	struct abd		*zt_orig_abd;
348 	uint64_t		zt_orig_size;
349 	uint64_t		zt_bufsize;
350 	zio_transform_func_t	*zt_transform;
351 	struct zio_transform	*zt_next;
352 } zio_transform_t;
353 
354 typedef int zio_pipe_stage_t(zio_t *zio);
355 
356 /*
357  * The io_reexecute flags are distinct from io_flags because the child must
358  * be able to propagate them to the parent.  The normal io_flags are local
359  * to the zio, not protected by any lock, and not modifiable by children;
360  * the reexecute flags are protected by io_lock, modifiable by children,
361  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
362  */
363 #define	ZIO_REEXECUTE_NOW	0x01
364 #define	ZIO_REEXECUTE_SUSPEND	0x02
365 
366 typedef struct zio_alloc_list {
367 	list_t  zal_list;
368 	uint64_t zal_size;
369 } zio_alloc_list_t;
370 
371 typedef struct zio_link {
372 	zio_t		*zl_parent;
373 	zio_t		*zl_child;
374 	list_node_t	zl_parent_node;
375 	list_node_t	zl_child_node;
376 } zio_link_t;
377 
378 struct zio {
379 	/* Core information about this I/O */
380 	zbookmark_phys_t	io_bookmark;
381 	zio_prop_t	io_prop;
382 	zio_type_t	io_type;
383 	enum zio_child	io_child_type;
384 	int		io_cmd;
385 	zio_priority_t	io_priority;
386 	uint8_t		io_reexecute;
387 	uint8_t		io_state[ZIO_WAIT_TYPES];
388 	uint64_t	io_txg;
389 	spa_t		*io_spa;
390 	blkptr_t	*io_bp;
391 	blkptr_t	*io_bp_override;
392 	blkptr_t	io_bp_copy;
393 	list_t		io_parent_list;
394 	list_t		io_child_list;
395 	zio_t		*io_logical;
396 	zio_transform_t *io_transform_stack;
397 
398 	/* Callback info */
399 	zio_done_func_t	*io_ready;
400 	zio_done_func_t	*io_children_ready;
401 	zio_done_func_t	*io_physdone;
402 	zio_done_func_t	*io_done;
403 	void		*io_private;
404 	int64_t		io_prev_space_delta;	/* DMU private */
405 	blkptr_t	io_bp_orig;
406 
407 	/* Data represented by this I/O */
408 	struct abd	*io_abd;
409 	struct abd	*io_orig_abd;
410 	uint64_t	io_size;
411 	uint64_t	io_orig_size;
412 	/* io_lsize != io_orig_size iff this is a raw write */
413 	uint64_t	io_lsize;
414 
415 	/* Stuff for the vdev stack */
416 	vdev_t		*io_vd;
417 	void		*io_vsd;
418 	const zio_vsd_ops_t *io_vsd_ops;
419 
420 	uint64_t	io_offset;
421 	hrtime_t	io_timestamp;
422 	hrtime_t	io_queued_timestamp;
423 	hrtime_t	io_target_timestamp;
424 	avl_node_t	io_queue_node;
425 	avl_node_t	io_offset_node;
426 	avl_node_t	io_alloc_node;
427 	zio_alloc_list_t 	io_alloc_list;
428 
429 	/* Internal pipeline state */
430 	enum zio_flag	io_flags;
431 	enum zio_stage	io_stage;
432 	enum zio_stage	io_pipeline;
433 	enum zio_flag	io_orig_flags;
434 	enum zio_stage	io_orig_stage;
435 	enum zio_stage	io_orig_pipeline;
436 	enum zio_stage	io_pipeline_trace;
437 	int		io_error;
438 	int		io_child_error[ZIO_CHILD_TYPES];
439 	uint64_t	io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
440 	uint64_t	io_child_count;
441 	uint64_t	io_phys_children;
442 	uint64_t	io_parent_count;
443 	uint64_t	*io_stall;
444 	zio_t		*io_gang_leader;
445 	zio_gang_node_t	*io_gang_tree;
446 	void		*io_executor;
447 	void		*io_waiter;
448 	kmutex_t	io_lock;
449 	kcondvar_t	io_cv;
450 
451 	/* FMA state */
452 	zio_cksum_report_t *io_cksum_report;
453 	uint64_t	io_ena;
454 
455 	/* Taskq dispatching state */
456 	taskq_ent_t	io_tqent;
457 };
458 
459 extern int zio_bookmark_compare(const void *, const void *);
460 
461 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
462     zio_done_func_t *done, void *private, enum zio_flag flags);
463 
464 extern zio_t *zio_root(spa_t *spa,
465     zio_done_func_t *done, void *private, enum zio_flag flags);
466 
467 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
468     struct abd *data, uint64_t lsize, zio_done_func_t *done, void *private,
469     zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
470 
471 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
472     struct abd *data, uint64_t size, uint64_t psize, const zio_prop_t *zp,
473     zio_done_func_t *ready, zio_done_func_t *children_ready,
474     zio_done_func_t *physdone, zio_done_func_t *done,
475     void *private, zio_priority_t priority, enum zio_flag flags,
476     const zbookmark_phys_t *zb);
477 
478 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
479     struct abd *data, uint64_t size, zio_done_func_t *done, void *private,
480     zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
481 
482 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
483     boolean_t nopwrite);
484 
485 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
486 
487 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
488     const blkptr_t *bp,
489     zio_done_func_t *done, void *private, enum zio_flag flags);
490 
491 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
492     zio_done_func_t *done, void *private, enum zio_flag flags);
493 
494 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
495     uint64_t size, struct abd *data, int checksum,
496     zio_done_func_t *done, void *private, zio_priority_t priority,
497     enum zio_flag flags, boolean_t labels);
498 
499 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
500     uint64_t size, struct abd *data, int checksum,
501     zio_done_func_t *done, void *private, zio_priority_t priority,
502     enum zio_flag flags, boolean_t labels);
503 
504 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
505     const blkptr_t *bp, enum zio_flag flags);
506 
507 extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
508     blkptr_t *old_bp, uint64_t size, boolean_t *slog);
509 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
510 extern void zio_flush(zio_t *zio, vdev_t *vd);
511 extern void zio_shrink(zio_t *zio, uint64_t size);
512 
513 extern int zio_wait(zio_t *zio);
514 extern void zio_nowait(zio_t *zio);
515 extern void zio_execute(zio_t *zio);
516 extern void zio_interrupt(zio_t *zio);
517 extern void zio_delay_init(zio_t *zio);
518 extern void zio_delay_interrupt(zio_t *zio);
519 
520 extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **);
521 extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **);
522 extern zio_t *zio_unique_parent(zio_t *cio);
523 extern void zio_add_child(zio_t *pio, zio_t *cio);
524 
525 extern void *zio_buf_alloc(size_t size);
526 extern void zio_buf_free(void *buf, size_t size);
527 extern void *zio_data_buf_alloc(size_t size);
528 extern void zio_data_buf_free(void *buf, size_t size);
529 
530 extern void zio_push_transform(zio_t *zio, struct abd *abd, uint64_t size,
531     uint64_t bufsize, zio_transform_func_t *transform);
532 extern void zio_pop_transforms(zio_t *zio);
533 
534 extern void zio_resubmit_stage_async(void *);
535 
536 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
537     uint64_t offset, struct abd *data, uint64_t size, int type,
538     zio_priority_t priority, enum zio_flag flags,
539     zio_done_func_t *done, void *private);
540 
541 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
542     struct abd *data, uint64_t size, int type, zio_priority_t priority,
543     enum zio_flag flags, zio_done_func_t *done, void *private);
544 
545 extern void zio_vdev_io_bypass(zio_t *zio);
546 extern void zio_vdev_io_reissue(zio_t *zio);
547 extern void zio_vdev_io_redone(zio_t *zio);
548 
549 extern void zio_checksum_verified(zio_t *zio);
550 extern int zio_worst_error(int e1, int e2);
551 
552 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
553     enum zio_checksum parent);
554 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
555     enum zio_checksum child, enum zio_checksum parent);
556 extern enum zio_compress zio_compress_select(spa_t *spa,
557     enum zio_compress child, enum zio_compress parent);
558 
559 extern void zio_suspend(spa_t *spa, zio_t *zio);
560 extern int zio_resume(spa_t *spa);
561 extern void zio_resume_wait(spa_t *spa);
562 
563 /*
564  * Initial setup and teardown.
565  */
566 extern void zio_init(void);
567 extern void zio_fini(void);
568 
569 /*
570  * Fault injection
571  */
572 struct zinject_record;
573 extern uint32_t zio_injection_enabled;
574 extern int zio_inject_fault(char *name, int flags, int *id,
575     struct zinject_record *record);
576 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
577     struct zinject_record *record);
578 extern int zio_clear_fault(int id);
579 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
580 extern int zio_handle_fault_injection(zio_t *zio, int error);
581 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
582 extern int zio_handle_label_injection(zio_t *zio, int error);
583 extern void zio_handle_ignored_writes(zio_t *zio);
584 extern hrtime_t zio_handle_io_delay(zio_t *zio);
585 
586 /*
587  * Checksum ereport functions
588  */
589 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
590     uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
591 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
592     const void *good_data, const void *bad_data, boolean_t drop_if_identical);
593 
594 extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
595 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
596 
597 /* If we have the good data in hand, this function can be used */
598 extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
599     struct zio *zio, uint64_t offset, uint64_t length,
600     const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
601 
602 /* Called from spa_sync(), but primarily an injection handler */
603 extern void spa_handle_ignored_writes(spa_t *spa);
604 
605 /* zbookmark_phys functions */
606 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
607     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
608 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
609     uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
610 
611 #ifdef	__cplusplus
612 }
613 #endif
614 
615 #endif	/* _ZIO_H */
616