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
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 #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 
134 typedef enum zio_suspend_reason {
135 	ZIO_SUSPEND_NONE = 0,
136 	ZIO_SUSPEND_IOERR,
137 	ZIO_SUSPEND_MMP,
138 } zio_suspend_reason_t;
139 
140 enum zio_flag {
141 	/*
142 	 * Flags inherited by gang, ddt, and vdev children,
143 	 * and that must be equal for two zios to aggregate
144 	 */
145 	ZIO_FLAG_DONT_AGGREGATE	= 1 << 0,
146 	ZIO_FLAG_IO_REPAIR	= 1 << 1,
147 	ZIO_FLAG_SELF_HEAL	= 1 << 2,
148 	ZIO_FLAG_RESILVER	= 1 << 3,
149 	ZIO_FLAG_SCRUB		= 1 << 4,
150 	ZIO_FLAG_SCAN_THREAD	= 1 << 5,
151 	ZIO_FLAG_PHYSICAL	= 1 << 6,
152 
153 #define	ZIO_FLAG_AGG_INHERIT	(ZIO_FLAG_CANFAIL - 1)
154 
155 	/*
156 	 * Flags inherited by ddt, gang, and vdev children.
157 	 */
158 	ZIO_FLAG_CANFAIL	= 1 << 7,	/* must be first for INHERIT */
159 	ZIO_FLAG_SPECULATIVE	= 1 << 8,
160 	ZIO_FLAG_CONFIG_WRITER	= 1 << 9,
161 	ZIO_FLAG_DONT_RETRY	= 1 << 10,
162 	ZIO_FLAG_DONT_CACHE	= 1 << 11,
163 	ZIO_FLAG_NODATA		= 1 << 12,
164 	ZIO_FLAG_INDUCE_DAMAGE	= 1 << 13,
165 	ZIO_FLAG_IO_ALLOCATING	= 1 << 14,
166 
167 #define	ZIO_FLAG_DDT_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
168 #define	ZIO_FLAG_GANG_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
169 
170 	/*
171 	 * Flags inherited by vdev children.
172 	 */
173 	ZIO_FLAG_IO_RETRY	= 1 << 15,	/* must be first for INHERIT */
174 	ZIO_FLAG_PROBE		= 1 << 16,
175 	ZIO_FLAG_TRYHARD	= 1 << 17,
176 	ZIO_FLAG_OPTIONAL	= 1 << 18,
177 
178 #define	ZIO_FLAG_VDEV_INHERIT	(ZIO_FLAG_DONT_QUEUE - 1)
179 
180 	/*
181 	 * Flags not inherited by any children.
182 	 */
183 	ZIO_FLAG_DONT_QUEUE	= 1 << 19,	/* must be first for INHERIT */
184 	ZIO_FLAG_DONT_PROPAGATE	= 1 << 20,
185 	ZIO_FLAG_IO_BYPASS	= 1 << 21,
186 	ZIO_FLAG_IO_REWRITE	= 1 << 22,
187 	ZIO_FLAG_RAW		= 1 << 23,
188 	ZIO_FLAG_GANG_CHILD	= 1 << 24,
189 	ZIO_FLAG_DDT_CHILD	= 1 << 25,
190 	ZIO_FLAG_GODFATHER	= 1 << 26,
191 	ZIO_FLAG_NOPWRITE	= 1 << 27,
192 	ZIO_FLAG_REEXECUTED	= 1 << 28,
193 	ZIO_FLAG_DELEGATED	= 1 << 29,
194 };
195 
196 #define	ZIO_FLAG_MUSTSUCCEED		0
197 
198 #define	ZIO_DDT_CHILD_FLAGS(zio)				\
199 	(((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |		\
200 	ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
201 
202 #define	ZIO_GANG_CHILD_FLAGS(zio)				\
203 	(((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |		\
204 	ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
205 
206 #define	ZIO_VDEV_CHILD_FLAGS(zio)				\
207 	(((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |		\
208 	ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_CANFAIL)
209 
210 #define	ZIO_CHILD_BIT(x)		(1 << (x))
211 #define	ZIO_CHILD_BIT_IS_SET(val, x)	((val) & (1 << (x)))
212 
213 enum zio_child {
214 	ZIO_CHILD_VDEV = 0,
215 	ZIO_CHILD_GANG,
216 	ZIO_CHILD_DDT,
217 	ZIO_CHILD_LOGICAL,
218 	ZIO_CHILD_TYPES
219 };
220 
221 #define	ZIO_CHILD_VDEV_BIT		ZIO_CHILD_BIT(ZIO_CHILD_VDEV)
222 #define	ZIO_CHILD_GANG_BIT		ZIO_CHILD_BIT(ZIO_CHILD_GANG)
223 #define	ZIO_CHILD_DDT_BIT		ZIO_CHILD_BIT(ZIO_CHILD_DDT)
224 #define	ZIO_CHILD_LOGICAL_BIT		ZIO_CHILD_BIT(ZIO_CHILD_LOGICAL)
225 #define	ZIO_CHILD_ALL_BITS					\
226 	(ZIO_CHILD_VDEV_BIT | ZIO_CHILD_GANG_BIT |		\
227 	ZIO_CHILD_DDT_BIT | ZIO_CHILD_LOGICAL_BIT)
228 
229 enum zio_wait_type {
230 	ZIO_WAIT_READY = 0,
231 	ZIO_WAIT_DONE,
232 	ZIO_WAIT_TYPES
233 };
234 
235 /*
236  * These are bespoke errnos used in ZFS. We map them to their closest FreeBSD
237  * equivalents. This gives us more useful error messages from strerror(3).
238  */
239 #define	ECKSUM	EINTEGRITY
240 #define	EFRAGS	ENOSPC
241 
242 typedef void zio_done_func_t(zio_t *zio);
243 
244 extern boolean_t zio_dva_throttle_enabled;
245 extern const char *zio_type_name[ZIO_TYPES];
246 
247 /*
248  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
249  * identifies any block in the pool.  By convention, the meta-objset (MOS)
250  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
251  * except root blocks and ZIL blocks, which are defined as follows:
252  *
253  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
254  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
255  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
256  * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
257  *
258  * Note: this structure is called a bookmark because its original purpose
259  * was to remember where to resume a pool-wide traverse.
260  *
261  * Note: this structure is passed between userland and the kernel, and is
262  * stored on disk (by virtue of being incorporated into other on-disk
263  * structures, e.g. dsl_scan_phys_t).
264  */
265 typedef struct zbookmark_phys {
266 	uint64_t	zb_objset;
267 	uint64_t	zb_object;
268 	int64_t		zb_level;
269 	uint64_t	zb_blkid;
270 } zbookmark_phys_t;
271 
272 #define	SET_BOOKMARK(zb, objset, object, level, blkid)  \
273 {                                                       \
274 	(zb)->zb_objset = objset;                       \
275 	(zb)->zb_object = object;                       \
276 	(zb)->zb_level = level;                         \
277 	(zb)->zb_blkid = blkid;                         \
278 }
279 
280 #define	ZB_DESTROYED_OBJSET	(-1ULL)
281 
282 #define	ZB_ROOT_OBJECT		(0ULL)
283 #define	ZB_ROOT_LEVEL		(-1LL)
284 #define	ZB_ROOT_BLKID		(0ULL)
285 
286 #define	ZB_ZIL_OBJECT		(0ULL)
287 #define	ZB_ZIL_LEVEL		(-2LL)
288 
289 #define	ZB_DNODE_LEVEL		(-3LL)
290 #define	ZB_DNODE_BLKID		(0ULL)
291 
292 #define	ZB_IS_ZERO(zb)						\
293 	((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&	\
294 	(zb)->zb_level == 0 && (zb)->zb_blkid == 0)
295 #define	ZB_IS_ROOT(zb)				\
296 	((zb)->zb_object == ZB_ROOT_OBJECT &&	\
297 	(zb)->zb_level == ZB_ROOT_LEVEL &&	\
298 	(zb)->zb_blkid == ZB_ROOT_BLKID)
299 
300 typedef struct zio_prop {
301 	enum zio_checksum	zp_checksum;
302 	enum zio_compress	zp_compress;
303 	dmu_object_type_t	zp_type;
304 	uint8_t			zp_level;
305 	uint8_t			zp_copies;
306 	boolean_t		zp_dedup;
307 	boolean_t		zp_dedup_verify;
308 	boolean_t		zp_nopwrite;
309 	uint32_t		zp_zpl_smallblk;
310 } zio_prop_t;
311 
312 typedef struct zio_cksum_report zio_cksum_report_t;
313 
314 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
315     const void *good_data);
316 typedef void zio_cksum_free_f(void *cbdata, size_t size);
317 
318 struct zio_bad_cksum;				/* defined in zio_checksum.h */
319 struct dnode_phys;
320 struct abd;
321 
322 struct zio_cksum_report {
323 	struct zio_cksum_report *zcr_next;
324 	nvlist_t		*zcr_ereport;
325 	nvlist_t		*zcr_detector;
326 	void			*zcr_cbdata;
327 	size_t			zcr_cbinfo;	/* passed to zcr_free() */
328 	uint64_t		zcr_align;
329 	uint64_t		zcr_length;
330 	zio_cksum_finish_f	*zcr_finish;
331 	zio_cksum_free_f	*zcr_free;
332 
333 	/* internal use only */
334 	struct zio_bad_cksum	*zcr_ckinfo;	/* information from failure */
335 };
336 
337 typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
338     void *arg);
339 
340 zio_vsd_cksum_report_f	zio_vsd_default_cksum_report;
341 
342 typedef struct zio_vsd_ops {
343 	zio_done_func_t		*vsd_free;
344 	zio_vsd_cksum_report_f	*vsd_cksum_report;
345 } zio_vsd_ops_t;
346 
347 typedef struct zio_gang_node {
348 	zio_gbh_phys_t		*gn_gbh;
349 	struct zio_gang_node	*gn_child[SPA_GBH_NBLKPTRS];
350 } zio_gang_node_t;
351 
352 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
353     zio_gang_node_t *gn, struct abd *data, uint64_t offset);
354 
355 typedef void zio_transform_func_t(zio_t *zio, struct abd *data, uint64_t size);
356 
357 typedef struct zio_transform {
358 	struct abd		*zt_orig_abd;
359 	uint64_t		zt_orig_size;
360 	uint64_t		zt_bufsize;
361 	zio_transform_func_t	*zt_transform;
362 	struct zio_transform	*zt_next;
363 } zio_transform_t;
364 
365 typedef zio_t *zio_pipe_stage_t(zio_t *zio);
366 
367 /*
368  * The io_reexecute flags are distinct from io_flags because the child must
369  * be able to propagate them to the parent.  The normal io_flags are local
370  * to the zio, not protected by any lock, and not modifiable by children;
371  * the reexecute flags are protected by io_lock, modifiable by children,
372  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
373  */
374 #define	ZIO_REEXECUTE_NOW	0x01
375 #define	ZIO_REEXECUTE_SUSPEND	0x02
376 
377 typedef struct zio_alloc_list {
378 	list_t  zal_list;
379 	uint64_t zal_size;
380 } zio_alloc_list_t;
381 
382 typedef struct zio_link {
383 	zio_t		*zl_parent;
384 	zio_t		*zl_child;
385 	list_node_t	zl_parent_node;
386 	list_node_t	zl_child_node;
387 } zio_link_t;
388 
389 /*
390  * Used for TRIM kstat.
391  */
392 typedef struct zio_trim_stats {
393 	/*
394 	 * Number of bytes successfully TRIMmed.
395 	 */
396 	kstat_named_t bytes;
397 
398 	/*
399 	 * Number of successful TRIM requests.
400 	 */
401 	kstat_named_t success;
402 
403 	/*
404 	 * Number of TRIM requests that failed because TRIM is not
405 	 * supported.
406 	 */
407 	kstat_named_t unsupported;
408 
409 	/*
410 	 * Number of TRIM requests that failed for other reasons.
411 	 */
412 	kstat_named_t failed;
413 } zio_trim_stats_t;
414 
415 extern zio_trim_stats_t zio_trim_stats;
416 
417 #define ZIO_TRIM_STAT_INCR(stat, val) \
418 	atomic_add_64(&zio_trim_stats.stat.value.ui64, (val));
419 #define ZIO_TRIM_STAT_BUMP(stat) \
420 	ZIO_TRIM_STAT_INCR(stat, 1);
421 
422 struct zio {
423 	/* Core information about this I/O */
424 	zbookmark_phys_t	io_bookmark;
425 	zio_prop_t	io_prop;
426 	zio_type_t	io_type;
427 	enum zio_child	io_child_type;
428 	int		io_cmd;
429 	zio_priority_t	io_priority;
430 	uint8_t		io_reexecute;
431 	uint8_t		io_state[ZIO_WAIT_TYPES];
432 	uint64_t	io_txg;
433 	spa_t		*io_spa;
434 	blkptr_t	*io_bp;
435 	blkptr_t	*io_bp_override;
436 	blkptr_t	io_bp_copy;
437 	list_t		io_parent_list;
438 	list_t		io_child_list;
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_children_ready;
445 	zio_done_func_t	*io_physdone;
446 	zio_done_func_t	*io_done;
447 	void		*io_private;
448 	int64_t		io_prev_space_delta;	/* DMU private */
449 	blkptr_t	io_bp_orig;
450 
451 	/* Data represented by this I/O */
452 	struct abd	*io_abd;
453 	struct abd	*io_orig_abd;
454 	uint64_t	io_size;
455 	uint64_t	io_orig_size;
456 	/* io_lsize != io_orig_size iff this is a raw write */
457 	uint64_t	io_lsize;
458 
459 	/* Stuff for the vdev stack */
460 	vdev_t		*io_vd;
461 	void		*io_vsd;
462 	const zio_vsd_ops_t *io_vsd_ops;
463 	metaslab_class_t *io_metaslab_class;	/* dva throttle class */
464 
465 	uint64_t	io_offset;
466 	hrtime_t	io_timestamp;
467 	hrtime_t	io_queued_timestamp;
468 	hrtime_t	io_target_timestamp;
469 	avl_node_t	io_queue_node;
470 	avl_node_t	io_offset_node;
471 	avl_node_t	io_alloc_node;
472 	zio_alloc_list_t	io_alloc_list;
473 
474 #ifdef __FreeBSD__
475 	struct bio	*io_bio;
476 #endif
477 
478 	/* Internal pipeline state */
479 	enum zio_flag	io_flags;
480 	enum zio_stage	io_stage;
481 	enum zio_stage	io_pipeline;
482 	enum zio_flag	io_orig_flags;
483 	enum zio_stage	io_orig_stage;
484 	enum zio_stage	io_orig_pipeline;
485 	enum zio_stage	io_pipeline_trace;
486 	int		io_error;
487 	int		io_child_error[ZIO_CHILD_TYPES];
488 	uint64_t	io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
489 	uint64_t	io_child_count;
490 	uint64_t	io_phys_children;
491 	uint64_t	io_parent_count;
492 	uint64_t	*io_stall;
493 	zio_t		*io_gang_leader;
494 	zio_gang_node_t	*io_gang_tree;
495 	void		*io_executor;
496 	void		*io_waiter;
497 	kmutex_t	io_lock;
498 	kcondvar_t	io_cv;
499 	int		io_allocator;
500 
501 	/* FMA state */
502 	zio_cksum_report_t *io_cksum_report;
503 	uint64_t	io_ena;
504 
505 	/* Taskq dispatching state */
506 	taskq_ent_t	io_tqent;
507 
508 	avl_node_t	io_trim_node;
509 	list_node_t	io_trim_link;
510 };
511 
512 extern int zio_bookmark_compare(const void *, const void *);
513 
514 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
515     zio_done_func_t *done, void *priv, enum zio_flag flags);
516 
517 extern zio_t *zio_root(spa_t *spa,
518     zio_done_func_t *done, void *priv, enum zio_flag flags);
519 
520 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
521     struct abd *data, uint64_t lsize, zio_done_func_t *done, void *priv,
522     zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
523 
524 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
525     struct abd *data, uint64_t size, uint64_t psize, const zio_prop_t *zp,
526     zio_done_func_t *ready, zio_done_func_t *children_ready,
527     zio_done_func_t *physdone, zio_done_func_t *done,
528     void *priv, zio_priority_t priority, enum zio_flag flags,
529     const zbookmark_phys_t *zb);
530 
531 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
532     struct abd *data, uint64_t size, zio_done_func_t *done, void *priv,
533     zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
534 
535 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
536     boolean_t nopwrite);
537 
538 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
539 
540 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
541     const blkptr_t *bp,
542     zio_done_func_t *done, void *priv, enum zio_flag flags);
543 
544 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
545     uint64_t offset, uint64_t size, zio_done_func_t *done, void *priv,
546     zio_priority_t priority, enum zio_flag flags);
547 
548 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
549     uint64_t size, struct abd *data, int checksum,
550     zio_done_func_t *done, void *priv, zio_priority_t priority,
551     enum zio_flag flags, boolean_t labels);
552 
553 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
554     uint64_t size, struct abd *data, int checksum,
555     zio_done_func_t *done, void *priv, zio_priority_t priority,
556     enum zio_flag flags, boolean_t labels);
557 
558 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
559     const blkptr_t *bp, uint64_t size, enum zio_flag flags);
560 
561 extern int zio_alloc_zil(spa_t *spa, uint64_t objset, uint64_t txg,
562     blkptr_t *new_bp, blkptr_t *old_bp, uint64_t size, boolean_t *slog);
563 extern void zio_flush(zio_t *zio, vdev_t *vd);
564 extern zio_t *zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset,
565     uint64_t size);
566 extern void zio_shrink(zio_t *zio, uint64_t size);
567 
568 extern int zio_wait(zio_t *zio);
569 extern void zio_nowait(zio_t *zio);
570 extern void zio_execute(zio_t *zio);
571 extern void zio_interrupt(zio_t *zio);
572 extern void zio_delay_init(zio_t *zio);
573 extern void zio_delay_interrupt(zio_t *zio);
574 
575 extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **);
576 extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **);
577 extern zio_t *zio_unique_parent(zio_t *cio);
578 extern void zio_add_child(zio_t *pio, zio_t *cio);
579 
580 extern void *zio_buf_alloc(size_t size);
581 extern void zio_buf_free(void *buf, size_t size);
582 extern void *zio_data_buf_alloc(size_t size);
583 extern void zio_data_buf_free(void *buf, size_t size);
584 
585 extern void zio_push_transform(zio_t *zio, struct abd *abd, uint64_t size,
586     uint64_t bufsize, zio_transform_func_t *transform);
587 extern void zio_pop_transforms(zio_t *zio);
588 
589 extern void zio_resubmit_stage_async(void *);
590 
591 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
592     uint64_t offset, struct abd *data, uint64_t size, int type,
593     zio_priority_t priority, enum zio_flag flags,
594     zio_done_func_t *done, void *priv);
595 
596 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
597     struct abd *data, uint64_t size, zio_type_t type, zio_priority_t priority,
598     enum zio_flag flags, zio_done_func_t *done, void *priv);
599 
600 extern void zio_vdev_io_bypass(zio_t *zio);
601 extern void zio_vdev_io_reissue(zio_t *zio);
602 extern void zio_vdev_io_redone(zio_t *zio);
603 
604 extern void zio_change_priority(zio_t *pio, zio_priority_t priority);
605 
606 extern void zio_checksum_verified(zio_t *zio);
607 extern int zio_worst_error(int e1, int e2);
608 
609 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
610     enum zio_checksum parent);
611 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
612     enum zio_checksum child, enum zio_checksum parent);
613 extern enum zio_compress zio_compress_select(spa_t *spa,
614     enum zio_compress child, enum zio_compress parent);
615 
616 extern void zio_suspend(spa_t *spa, zio_t *zio, zio_suspend_reason_t);
617 extern int zio_resume(spa_t *spa);
618 extern void zio_resume_wait(spa_t *spa);
619 
620 /*
621  * Initial setup and teardown.
622  */
623 extern void zio_init(void);
624 extern void zio_fini(void);
625 
626 /*
627  * Fault injection
628  */
629 struct zinject_record;
630 extern uint32_t zio_injection_enabled;
631 extern int zio_inject_fault(char *name, int flags, int *id,
632     struct zinject_record *record);
633 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
634     struct zinject_record *record);
635 extern int zio_clear_fault(int id);
636 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
637 extern int zio_handle_fault_injection(zio_t *zio, int error);
638 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
639 extern int zio_handle_label_injection(zio_t *zio, int error);
640 extern void zio_handle_ignored_writes(zio_t *zio);
641 extern hrtime_t zio_handle_io_delay(zio_t *zio);
642 
643 /*
644  * Checksum ereport functions
645  */
646 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
647     uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
648 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
649     const void *good_data, const void *bad_data, boolean_t drop_if_identical);
650 
651 extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
652 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
653 
654 /* If we have the good data in hand, this function can be used */
655 extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
656     struct zio *zio, uint64_t offset, uint64_t length,
657     const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
658 
659 /* Called from spa_sync(), but primarily an injection handler */
660 extern void spa_handle_ignored_writes(spa_t *spa);
661 
662 /* zbookmark_phys functions */
663 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
664     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
665 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
666     uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
667 
668 #ifdef	__cplusplus
669 }
670 #endif
671 
672 #endif	/* _ZIO_H */
673