xref: /titanic_52/usr/src/uts/common/fs/zfs/sys/zio.h (revision 9b6cddcfb35f476ab090ca2318bb30d98c21ecb2)
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  * 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 enum zio_compress {
108 	ZIO_COMPRESS_INHERIT = 0,
109 	ZIO_COMPRESS_ON,
110 	ZIO_COMPRESS_OFF,
111 	ZIO_COMPRESS_LZJB,
112 	ZIO_COMPRESS_EMPTY,
113 	ZIO_COMPRESS_GZIP_1,
114 	ZIO_COMPRESS_GZIP_2,
115 	ZIO_COMPRESS_GZIP_3,
116 	ZIO_COMPRESS_GZIP_4,
117 	ZIO_COMPRESS_GZIP_5,
118 	ZIO_COMPRESS_GZIP_6,
119 	ZIO_COMPRESS_GZIP_7,
120 	ZIO_COMPRESS_GZIP_8,
121 	ZIO_COMPRESS_GZIP_9,
122 	ZIO_COMPRESS_ZLE,
123 	ZIO_COMPRESS_LZ4,
124 	ZIO_COMPRESS_FUNCTIONS
125 };
126 
127 /*
128  * The number of "legacy" compression functions which can be set on individual
129  * objects.
130  */
131 #define	ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
132 
133 /*
134  * The meaning of "compress = on" selected by the compression features enabled
135  * on a given pool.
136  */
137 #define	ZIO_COMPRESS_LEGACY_ON_VALUE	ZIO_COMPRESS_LZJB
138 #define	ZIO_COMPRESS_LZ4_ON_VALUE	ZIO_COMPRESS_LZ4
139 
140 #define	ZIO_COMPRESS_DEFAULT		ZIO_COMPRESS_OFF
141 
142 #define	BOOTFS_COMPRESS_VALID(compress)			\
143 	((compress) == ZIO_COMPRESS_LZJB ||		\
144 	(compress) == ZIO_COMPRESS_LZ4 ||		\
145 	(compress) == ZIO_COMPRESS_GZIP_1 ||		\
146 	(compress) == ZIO_COMPRESS_GZIP_2 ||		\
147 	(compress) == ZIO_COMPRESS_GZIP_3 ||		\
148 	(compress) == ZIO_COMPRESS_GZIP_4 ||		\
149 	(compress) == ZIO_COMPRESS_GZIP_5 ||		\
150 	(compress) == ZIO_COMPRESS_GZIP_6 ||		\
151 	(compress) == ZIO_COMPRESS_GZIP_7 ||		\
152 	(compress) == ZIO_COMPRESS_GZIP_8 ||		\
153 	(compress) == ZIO_COMPRESS_GZIP_9 ||		\
154 	(compress) == ZIO_COMPRESS_ON ||		\
155 	(compress) == ZIO_COMPRESS_OFF)
156 
157 #define	ZIO_FAILURE_MODE_WAIT		0
158 #define	ZIO_FAILURE_MODE_CONTINUE	1
159 #define	ZIO_FAILURE_MODE_PANIC		2
160 
161 enum zio_flag {
162 	/*
163 	 * Flags inherited by gang, ddt, and vdev children,
164 	 * and that must be equal for two zios to aggregate
165 	 */
166 	ZIO_FLAG_DONT_AGGREGATE	= 1 << 0,
167 	ZIO_FLAG_IO_REPAIR	= 1 << 1,
168 	ZIO_FLAG_SELF_HEAL	= 1 << 2,
169 	ZIO_FLAG_RESILVER	= 1 << 3,
170 	ZIO_FLAG_SCRUB		= 1 << 4,
171 	ZIO_FLAG_SCAN_THREAD	= 1 << 5,
172 	ZIO_FLAG_PHYSICAL	= 1 << 6,
173 
174 #define	ZIO_FLAG_AGG_INHERIT	(ZIO_FLAG_CANFAIL - 1)
175 
176 	/*
177 	 * Flags inherited by ddt, gang, and vdev children.
178 	 */
179 	ZIO_FLAG_CANFAIL	= 1 << 7,	/* must be first for INHERIT */
180 	ZIO_FLAG_SPECULATIVE	= 1 << 8,
181 	ZIO_FLAG_CONFIG_WRITER	= 1 << 9,
182 	ZIO_FLAG_DONT_RETRY	= 1 << 10,
183 	ZIO_FLAG_DONT_CACHE	= 1 << 11,
184 	ZIO_FLAG_NODATA		= 1 << 12,
185 	ZIO_FLAG_INDUCE_DAMAGE	= 1 << 13,
186 
187 #define	ZIO_FLAG_DDT_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
188 #define	ZIO_FLAG_GANG_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
189 
190 	/*
191 	 * Flags inherited by vdev children.
192 	 */
193 	ZIO_FLAG_IO_RETRY	= 1 << 14,	/* must be first for INHERIT */
194 	ZIO_FLAG_PROBE		= 1 << 15,
195 	ZIO_FLAG_TRYHARD	= 1 << 16,
196 	ZIO_FLAG_OPTIONAL	= 1 << 17,
197 
198 #define	ZIO_FLAG_VDEV_INHERIT	(ZIO_FLAG_DONT_QUEUE - 1)
199 
200 	/*
201 	 * Flags not inherited by any children.
202 	 */
203 	ZIO_FLAG_DONT_QUEUE	= 1 << 18,	/* must be first for INHERIT */
204 	ZIO_FLAG_DONT_PROPAGATE	= 1 << 19,
205 	ZIO_FLAG_IO_BYPASS	= 1 << 20,
206 	ZIO_FLAG_IO_REWRITE	= 1 << 21,
207 	ZIO_FLAG_RAW		= 1 << 22,
208 	ZIO_FLAG_GANG_CHILD	= 1 << 23,
209 	ZIO_FLAG_DDT_CHILD	= 1 << 24,
210 	ZIO_FLAG_GODFATHER	= 1 << 25,
211 	ZIO_FLAG_NOPWRITE	= 1 << 26,
212 	ZIO_FLAG_REEXECUTED	= 1 << 27,
213 	ZIO_FLAG_DELEGATED	= 1 << 28,
214 };
215 
216 #define	ZIO_FLAG_MUSTSUCCEED		0
217 
218 #define	ZIO_DDT_CHILD_FLAGS(zio)				\
219 	(((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |		\
220 	ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
221 
222 #define	ZIO_GANG_CHILD_FLAGS(zio)				\
223 	(((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |		\
224 	ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
225 
226 #define	ZIO_VDEV_CHILD_FLAGS(zio)				\
227 	(((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |		\
228 	ZIO_FLAG_CANFAIL)
229 
230 #define	ZIO_CHILD_BIT(x)		(1 << (x))
231 #define	ZIO_CHILD_BIT_IS_SET(val, x)	((val) & (1 << (x)))
232 
233 enum zio_child {
234 	ZIO_CHILD_VDEV = 0,
235 	ZIO_CHILD_GANG,
236 	ZIO_CHILD_DDT,
237 	ZIO_CHILD_LOGICAL,
238 	ZIO_CHILD_TYPES
239 };
240 
241 #define	ZIO_CHILD_VDEV_BIT		ZIO_CHILD_BIT(ZIO_CHILD_VDEV)
242 #define	ZIO_CHILD_GANG_BIT		ZIO_CHILD_BIT(ZIO_CHILD_GANG)
243 #define	ZIO_CHILD_DDT_BIT		ZIO_CHILD_BIT(ZIO_CHILD_DDT)
244 #define	ZIO_CHILD_LOGICAL_BIT		ZIO_CHILD_BIT(ZIO_CHILD_LOGICAL)
245 #define	ZIO_CHILD_ALL_BITS					\
246 	(ZIO_CHILD_VDEV_BIT | ZIO_CHILD_GANG_BIT | 		\
247 	ZIO_CHILD_DDT_BIT | ZIO_CHILD_LOGICAL_BIT)
248 
249 enum zio_wait_type {
250 	ZIO_WAIT_READY = 0,
251 	ZIO_WAIT_DONE,
252 	ZIO_WAIT_TYPES
253 };
254 
255 /*
256  * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent
257  * graveyard) to indicate checksum errors and fragmentation.
258  */
259 #define	ECKSUM	EBADE
260 #define	EFRAGS	EBADR
261 
262 typedef void zio_done_func_t(zio_t *zio);
263 
264 extern const char *zio_type_name[ZIO_TYPES];
265 
266 /*
267  * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
268  * identifies any block in the pool.  By convention, the meta-objset (MOS)
269  * is objset 0, and the meta-dnode is object 0.  This covers all blocks
270  * except root blocks and ZIL blocks, which are defined as follows:
271  *
272  * Root blocks (objset_phys_t) are object 0, level -1:  <objset, 0, -1, 0>.
273  * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
274  * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
275  * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
276  *
277  * Note: this structure is called a bookmark because its original purpose
278  * was to remember where to resume a pool-wide traverse.
279  *
280  * Note: this structure is passed between userland and the kernel, and is
281  * stored on disk (by virtue of being incorporated into other on-disk
282  * structures, e.g. dsl_scan_phys_t).
283  */
284 typedef struct zbookmark_phys {
285 	uint64_t	zb_objset;
286 	uint64_t	zb_object;
287 	int64_t		zb_level;
288 	uint64_t	zb_blkid;
289 } zbookmark_phys_t;
290 
291 #define	SET_BOOKMARK(zb, objset, object, level, blkid)  \
292 {                                                       \
293 	(zb)->zb_objset = objset;                       \
294 	(zb)->zb_object = object;                       \
295 	(zb)->zb_level = level;                         \
296 	(zb)->zb_blkid = blkid;                         \
297 }
298 
299 #define	ZB_DESTROYED_OBJSET	(-1ULL)
300 
301 #define	ZB_ROOT_OBJECT		(0ULL)
302 #define	ZB_ROOT_LEVEL		(-1LL)
303 #define	ZB_ROOT_BLKID		(0ULL)
304 
305 #define	ZB_ZIL_OBJECT		(0ULL)
306 #define	ZB_ZIL_LEVEL		(-2LL)
307 
308 #define	ZB_DNODE_LEVEL		(-3LL)
309 #define	ZB_DNODE_BLKID		(0ULL)
310 
311 #define	ZB_IS_ZERO(zb)						\
312 	((zb)->zb_objset == 0 && (zb)->zb_object == 0 &&	\
313 	(zb)->zb_level == 0 && (zb)->zb_blkid == 0)
314 #define	ZB_IS_ROOT(zb)				\
315 	((zb)->zb_object == ZB_ROOT_OBJECT &&	\
316 	(zb)->zb_level == ZB_ROOT_LEVEL &&	\
317 	(zb)->zb_blkid == ZB_ROOT_BLKID)
318 
319 typedef struct zio_prop {
320 	enum zio_checksum	zp_checksum;
321 	enum zio_compress	zp_compress;
322 	dmu_object_type_t	zp_type;
323 	uint8_t			zp_level;
324 	uint8_t			zp_copies;
325 	boolean_t		zp_dedup;
326 	boolean_t		zp_dedup_verify;
327 	boolean_t		zp_nopwrite;
328 } zio_prop_t;
329 
330 typedef struct zio_cksum_report zio_cksum_report_t;
331 
332 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
333     const void *good_data);
334 typedef void zio_cksum_free_f(void *cbdata, size_t size);
335 
336 struct zio_bad_cksum;				/* defined in zio_checksum.h */
337 struct dnode_phys;
338 
339 struct zio_cksum_report {
340 	struct zio_cksum_report *zcr_next;
341 	nvlist_t		*zcr_ereport;
342 	nvlist_t		*zcr_detector;
343 	void			*zcr_cbdata;
344 	size_t			zcr_cbinfo;	/* passed to zcr_free() */
345 	uint64_t		zcr_align;
346 	uint64_t		zcr_length;
347 	zio_cksum_finish_f	*zcr_finish;
348 	zio_cksum_free_f	*zcr_free;
349 
350 	/* internal use only */
351 	struct zio_bad_cksum	*zcr_ckinfo;	/* information from failure */
352 };
353 
354 typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
355     void *arg);
356 
357 zio_vsd_cksum_report_f	zio_vsd_default_cksum_report;
358 
359 typedef struct zio_vsd_ops {
360 	zio_done_func_t		*vsd_free;
361 	zio_vsd_cksum_report_f	*vsd_cksum_report;
362 } zio_vsd_ops_t;
363 
364 typedef struct zio_gang_node {
365 	zio_gbh_phys_t		*gn_gbh;
366 	struct zio_gang_node	*gn_child[SPA_GBH_NBLKPTRS];
367 } zio_gang_node_t;
368 
369 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
370     zio_gang_node_t *gn, void *data);
371 
372 typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size);
373 
374 typedef struct zio_transform {
375 	void			*zt_orig_data;
376 	uint64_t		zt_orig_size;
377 	uint64_t		zt_bufsize;
378 	zio_transform_func_t	*zt_transform;
379 	struct zio_transform	*zt_next;
380 } zio_transform_t;
381 
382 typedef int zio_pipe_stage_t(zio_t *zio);
383 
384 /*
385  * The io_reexecute flags are distinct from io_flags because the child must
386  * be able to propagate them to the parent.  The normal io_flags are local
387  * to the zio, not protected by any lock, and not modifiable by children;
388  * the reexecute flags are protected by io_lock, modifiable by children,
389  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
390  */
391 #define	ZIO_REEXECUTE_NOW	0x01
392 #define	ZIO_REEXECUTE_SUSPEND	0x02
393 
394 typedef struct zio_link {
395 	zio_t		*zl_parent;
396 	zio_t		*zl_child;
397 	list_node_t	zl_parent_node;
398 	list_node_t	zl_child_node;
399 } zio_link_t;
400 
401 struct zio {
402 	/* Core information about this I/O */
403 	zbookmark_phys_t	io_bookmark;
404 	zio_prop_t	io_prop;
405 	zio_type_t	io_type;
406 	enum zio_child	io_child_type;
407 	int		io_cmd;
408 	zio_priority_t	io_priority;
409 	uint8_t		io_reexecute;
410 	uint8_t		io_state[ZIO_WAIT_TYPES];
411 	uint64_t	io_txg;
412 	spa_t		*io_spa;
413 	blkptr_t	*io_bp;
414 	blkptr_t	*io_bp_override;
415 	blkptr_t	io_bp_copy;
416 	list_t		io_parent_list;
417 	list_t		io_child_list;
418 	zio_link_t	*io_walk_link;
419 	zio_t		*io_logical;
420 	zio_transform_t *io_transform_stack;
421 
422 	/* Callback info */
423 	zio_done_func_t	*io_ready;
424 	zio_done_func_t	*io_physdone;
425 	zio_done_func_t	*io_done;
426 	void		*io_private;
427 	int64_t		io_prev_space_delta;	/* DMU private */
428 	blkptr_t	io_bp_orig;
429 
430 	/* Data represented by this I/O */
431 	void		*io_data;
432 	void		*io_orig_data;
433 	uint64_t	io_size;
434 	uint64_t	io_orig_size;
435 
436 	/* Stuff for the vdev stack */
437 	vdev_t		*io_vd;
438 	void		*io_vsd;
439 	const zio_vsd_ops_t *io_vsd_ops;
440 
441 	uint64_t	io_offset;
442 	hrtime_t	io_timestamp;
443 	hrtime_t	io_target_timestamp;
444 	avl_node_t	io_queue_node;
445 	avl_node_t	io_offset_node;
446 
447 	/* Internal pipeline state */
448 	enum zio_flag	io_flags;
449 	enum zio_stage	io_stage;
450 	enum zio_stage	io_pipeline;
451 	enum zio_flag	io_orig_flags;
452 	enum zio_stage	io_orig_stage;
453 	enum zio_stage	io_orig_pipeline;
454 	int		io_error;
455 	int		io_child_error[ZIO_CHILD_TYPES];
456 	uint64_t	io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
457 	uint64_t	io_child_count;
458 	uint64_t	io_phys_children;
459 	uint64_t	io_parent_count;
460 	uint64_t	*io_stall;
461 	zio_t		*io_gang_leader;
462 	zio_gang_node_t	*io_gang_tree;
463 	void		*io_executor;
464 	void		*io_waiter;
465 	kmutex_t	io_lock;
466 	kcondvar_t	io_cv;
467 
468 	/* FMA state */
469 	zio_cksum_report_t *io_cksum_report;
470 	uint64_t	io_ena;
471 
472 	/* Taskq dispatching state */
473 	taskq_ent_t	io_tqent;
474 };
475 
476 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
477     zio_done_func_t *done, void *private, enum zio_flag flags);
478 
479 extern zio_t *zio_root(spa_t *spa,
480     zio_done_func_t *done, void *private, enum zio_flag flags);
481 
482 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
483     uint64_t size, zio_done_func_t *done, void *private,
484     zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
485 
486 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
487     void *data, uint64_t size, const zio_prop_t *zp,
488     zio_done_func_t *ready, zio_done_func_t *physdone, zio_done_func_t *done,
489     void *private,
490     zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
491 
492 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
493     void *data, uint64_t size, zio_done_func_t *done, void *private,
494     zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
495 
496 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
497     boolean_t nopwrite);
498 
499 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
500 
501 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
502     const blkptr_t *bp,
503     zio_done_func_t *done, void *private, enum zio_flag flags);
504 
505 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
506     zio_done_func_t *done, void *private, enum zio_flag flags);
507 
508 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
509     uint64_t size, void *data, int checksum,
510     zio_done_func_t *done, void *private, zio_priority_t priority,
511     enum zio_flag flags, boolean_t labels);
512 
513 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
514     uint64_t size, void *data, int checksum,
515     zio_done_func_t *done, void *private, zio_priority_t priority,
516     enum zio_flag flags, boolean_t labels);
517 
518 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
519     const blkptr_t *bp, enum zio_flag flags);
520 
521 extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
522     blkptr_t *old_bp, uint64_t size, boolean_t use_slog);
523 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
524 extern void zio_flush(zio_t *zio, vdev_t *vd);
525 extern void zio_shrink(zio_t *zio, uint64_t size);
526 
527 extern int zio_wait(zio_t *zio);
528 extern void zio_nowait(zio_t *zio);
529 extern void zio_execute(zio_t *zio);
530 extern void zio_interrupt(zio_t *zio);
531 extern void zio_delay_init(zio_t *zio);
532 extern void zio_delay_interrupt(zio_t *zio);
533 
534 extern zio_t *zio_walk_parents(zio_t *cio);
535 extern zio_t *zio_walk_children(zio_t *pio);
536 extern zio_t *zio_unique_parent(zio_t *cio);
537 extern void zio_add_child(zio_t *pio, zio_t *cio);
538 
539 extern void *zio_buf_alloc(size_t size);
540 extern void zio_buf_free(void *buf, size_t size);
541 extern void *zio_data_buf_alloc(size_t size);
542 extern void zio_data_buf_free(void *buf, size_t size);
543 
544 extern void zio_resubmit_stage_async(void *);
545 
546 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
547     uint64_t offset, void *data, uint64_t size, int type,
548     zio_priority_t priority, enum zio_flag flags,
549     zio_done_func_t *done, void *private);
550 
551 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
552     void *data, uint64_t size, int type, zio_priority_t priority,
553     enum zio_flag flags, zio_done_func_t *done, void *private);
554 
555 extern void zio_vdev_io_bypass(zio_t *zio);
556 extern void zio_vdev_io_reissue(zio_t *zio);
557 extern void zio_vdev_io_redone(zio_t *zio);
558 
559 extern void zio_checksum_verified(zio_t *zio);
560 extern int zio_worst_error(int e1, int e2);
561 
562 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
563     enum zio_checksum parent);
564 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
565     enum zio_checksum child, enum zio_checksum parent);
566 extern enum zio_compress zio_compress_select(spa_t *spa,
567     enum zio_compress child, enum zio_compress parent);
568 
569 extern void zio_suspend(spa_t *spa, zio_t *zio);
570 extern int zio_resume(spa_t *spa);
571 extern void zio_resume_wait(spa_t *spa);
572 
573 /*
574  * Initial setup and teardown.
575  */
576 extern void zio_init(void);
577 extern void zio_fini(void);
578 
579 /*
580  * Fault injection
581  */
582 struct zinject_record;
583 extern uint32_t zio_injection_enabled;
584 extern int zio_inject_fault(char *name, int flags, int *id,
585     struct zinject_record *record);
586 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
587     struct zinject_record *record);
588 extern int zio_clear_fault(int id);
589 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
590 extern int zio_handle_fault_injection(zio_t *zio, int error);
591 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
592 extern int zio_handle_label_injection(zio_t *zio, int error);
593 extern void zio_handle_ignored_writes(zio_t *zio);
594 extern hrtime_t zio_handle_io_delay(zio_t *zio);
595 
596 /*
597  * Checksum ereport functions
598  */
599 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
600     uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
601 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
602     const void *good_data, const void *bad_data, boolean_t drop_if_identical);
603 
604 extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
605 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
606 
607 /* If we have the good data in hand, this function can be used */
608 extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
609     struct zio *zio, uint64_t offset, uint64_t length,
610     const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
611 
612 /* Called from spa_sync(), but primarily an injection handler */
613 extern void spa_handle_ignored_writes(spa_t *spa);
614 
615 /* zbookmark_phys functions */
616 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
617     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
618 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
619     uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
620 
621 #ifdef	__cplusplus
622 }
623 #endif
624 
625 #endif	/* _ZIO_H */
626