xref: /illumos-gate/usr/src/uts/common/fs/zfs/sys/zio.h (revision 2f7f7a62d7a3e8a2e75eb88b95bc65871b6b90cb)
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, 2016 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  */
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/fs/zfs.h>
39 #include <sys/zio_impl.h>
40 
41 #ifdef	__cplusplus
42 extern "C" {
43 #endif
44 
45 /*
46  * Embedded checksum
47  */
48 #define	ZEC_MAGIC	0x210da7ab10c7a11ULL
49 
50 typedef struct zio_eck {
51 	uint64_t	zec_magic;	/* for validation, endianness	*/
52 	zio_cksum_t	zec_cksum;	/* 256-bit checksum		*/
53 } zio_eck_t;
54 
55 /*
56  * Gang block headers are self-checksumming and contain an array
57  * of block pointers.
58  */
59 #define	SPA_GANGBLOCKSIZE	SPA_MINBLOCKSIZE
60 #define	SPA_GBH_NBLKPTRS	((SPA_GANGBLOCKSIZE - \
61 	sizeof (zio_eck_t)) / sizeof (blkptr_t))
62 #define	SPA_GBH_FILLER		((SPA_GANGBLOCKSIZE - \
63 	sizeof (zio_eck_t) - \
64 	(SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\
65 	sizeof (uint64_t))
66 
67 typedef struct zio_gbh {
68 	blkptr_t		zg_blkptr[SPA_GBH_NBLKPTRS];
69 	uint64_t		zg_filler[SPA_GBH_FILLER];
70 	zio_eck_t		zg_tail;
71 } zio_gbh_phys_t;
72 
73 enum zio_checksum {
74 	ZIO_CHECKSUM_INHERIT = 0,
75 	ZIO_CHECKSUM_ON,
76 	ZIO_CHECKSUM_OFF,
77 	ZIO_CHECKSUM_LABEL,
78 	ZIO_CHECKSUM_GANG_HEADER,
79 	ZIO_CHECKSUM_ZILOG,
80 	ZIO_CHECKSUM_FLETCHER_2,
81 	ZIO_CHECKSUM_FLETCHER_4,
82 	ZIO_CHECKSUM_SHA256,
83 	ZIO_CHECKSUM_ZILOG2,
84 	ZIO_CHECKSUM_NOPARITY,
85 	ZIO_CHECKSUM_SHA512,
86 	ZIO_CHECKSUM_SKEIN,
87 	ZIO_CHECKSUM_EDONR,
88 	ZIO_CHECKSUM_FUNCTIONS
89 };
90 
91 /*
92  * The number of "legacy" compression functions which can be set on individual
93  * objects.
94  */
95 #define	ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2
96 
97 #define	ZIO_CHECKSUM_ON_VALUE	ZIO_CHECKSUM_FLETCHER_4
98 #define	ZIO_CHECKSUM_DEFAULT	ZIO_CHECKSUM_ON
99 
100 #define	ZIO_CHECKSUM_MASK	0xffULL
101 #define	ZIO_CHECKSUM_VERIFY	(1 << 8)
102 
103 #define	ZIO_DEDUPCHECKSUM	ZIO_CHECKSUM_SHA256
104 #define	ZIO_DEDUPDITTO_MIN	100
105 
106 enum zio_compress {
107 	ZIO_COMPRESS_INHERIT = 0,
108 	ZIO_COMPRESS_ON,
109 	ZIO_COMPRESS_OFF,
110 	ZIO_COMPRESS_LZJB,
111 	ZIO_COMPRESS_EMPTY,
112 	ZIO_COMPRESS_GZIP_1,
113 	ZIO_COMPRESS_GZIP_2,
114 	ZIO_COMPRESS_GZIP_3,
115 	ZIO_COMPRESS_GZIP_4,
116 	ZIO_COMPRESS_GZIP_5,
117 	ZIO_COMPRESS_GZIP_6,
118 	ZIO_COMPRESS_GZIP_7,
119 	ZIO_COMPRESS_GZIP_8,
120 	ZIO_COMPRESS_GZIP_9,
121 	ZIO_COMPRESS_ZLE,
122 	ZIO_COMPRESS_LZ4,
123 	ZIO_COMPRESS_FUNCTIONS
124 };
125 
126 /*
127  * The number of "legacy" compression functions which can be set on individual
128  * objects.
129  */
130 #define	ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
131 
132 /*
133  * The meaning of "compress = on" selected by the compression features enabled
134  * on a given pool.
135  */
136 #define	ZIO_COMPRESS_LEGACY_ON_VALUE	ZIO_COMPRESS_LZJB
137 #define	ZIO_COMPRESS_LZ4_ON_VALUE	ZIO_COMPRESS_LZ4
138 
139 #define	ZIO_COMPRESS_DEFAULT		ZIO_COMPRESS_OFF
140 
141 #define	BOOTFS_COMPRESS_VALID(compress)			\
142 	((compress) == ZIO_COMPRESS_LZJB ||		\
143 	(compress) == ZIO_COMPRESS_LZ4 ||		\
144 	(compress) == ZIO_COMPRESS_ON ||		\
145 	(compress) == ZIO_COMPRESS_OFF)
146 
147 #define	ZIO_FAILURE_MODE_WAIT		0
148 #define	ZIO_FAILURE_MODE_CONTINUE	1
149 #define	ZIO_FAILURE_MODE_PANIC		2
150 
151 enum zio_flag {
152 	/*
153 	 * Flags inherited by gang, ddt, and vdev children,
154 	 * and that must be equal for two zios to aggregate
155 	 */
156 	ZIO_FLAG_DONT_AGGREGATE	= 1 << 0,
157 	ZIO_FLAG_IO_REPAIR	= 1 << 1,
158 	ZIO_FLAG_SELF_HEAL	= 1 << 2,
159 	ZIO_FLAG_RESILVER	= 1 << 3,
160 	ZIO_FLAG_SCRUB		= 1 << 4,
161 	ZIO_FLAG_SCAN_THREAD	= 1 << 5,
162 	ZIO_FLAG_PHYSICAL	= 1 << 6,
163 
164 #define	ZIO_FLAG_AGG_INHERIT	(ZIO_FLAG_CANFAIL - 1)
165 
166 	/*
167 	 * Flags inherited by ddt, gang, and vdev children.
168 	 */
169 	ZIO_FLAG_CANFAIL	= 1 << 7,	/* must be first for INHERIT */
170 	ZIO_FLAG_SPECULATIVE	= 1 << 8,
171 	ZIO_FLAG_CONFIG_WRITER	= 1 << 9,
172 	ZIO_FLAG_DONT_RETRY	= 1 << 10,
173 	ZIO_FLAG_DONT_CACHE	= 1 << 11,
174 	ZIO_FLAG_NODATA		= 1 << 12,
175 	ZIO_FLAG_INDUCE_DAMAGE	= 1 << 13,
176 	ZIO_FLAG_IO_ALLOCATING	= 1 << 14,
177 
178 #define	ZIO_FLAG_DDT_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
179 #define	ZIO_FLAG_GANG_INHERIT	(ZIO_FLAG_IO_RETRY - 1)
180 
181 	/*
182 	 * Flags inherited by vdev children.
183 	 */
184 	ZIO_FLAG_IO_RETRY	= 1 << 15,	/* must be first for INHERIT */
185 	ZIO_FLAG_PROBE		= 1 << 16,
186 	ZIO_FLAG_TRYHARD	= 1 << 17,
187 	ZIO_FLAG_OPTIONAL	= 1 << 18,
188 
189 #define	ZIO_FLAG_VDEV_INHERIT	(ZIO_FLAG_DONT_QUEUE - 1)
190 
191 	/*
192 	 * Flags not inherited by any children.
193 	 */
194 	ZIO_FLAG_DONT_QUEUE	= 1 << 19,	/* must be first for INHERIT */
195 	ZIO_FLAG_DONT_PROPAGATE	= 1 << 20,
196 	ZIO_FLAG_IO_BYPASS	= 1 << 21,
197 	ZIO_FLAG_IO_REWRITE	= 1 << 22,
198 	ZIO_FLAG_RAW		= 1 << 23,
199 	ZIO_FLAG_GANG_CHILD	= 1 << 24,
200 	ZIO_FLAG_DDT_CHILD	= 1 << 25,
201 	ZIO_FLAG_GODFATHER	= 1 << 26,
202 	ZIO_FLAG_NOPWRITE	= 1 << 27,
203 	ZIO_FLAG_REEXECUTED	= 1 << 28,
204 	ZIO_FLAG_DELEGATED	= 1 << 29,
205 };
206 
207 #define	ZIO_FLAG_MUSTSUCCEED		0
208 
209 #define	ZIO_DDT_CHILD_FLAGS(zio)				\
210 	(((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) |		\
211 	ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
212 
213 #define	ZIO_GANG_CHILD_FLAGS(zio)				\
214 	(((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) |		\
215 	ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
216 
217 #define	ZIO_VDEV_CHILD_FLAGS(zio)				\
218 	(((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) |		\
219 	ZIO_FLAG_CANFAIL)
220 
221 enum zio_child {
222 	ZIO_CHILD_VDEV = 0,
223 	ZIO_CHILD_GANG,
224 	ZIO_CHILD_DDT,
225 	ZIO_CHILD_LOGICAL,
226 	ZIO_CHILD_TYPES
227 };
228 
229 enum zio_wait_type {
230 	ZIO_WAIT_READY = 0,
231 	ZIO_WAIT_DONE,
232 	ZIO_WAIT_TYPES
233 };
234 
235 /*
236  * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent
237  * graveyard) to indicate checksum errors and fragmentation.
238  */
239 #define	ECKSUM	EBADE
240 #define	EFRAGS	EBADR
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 } zio_prop_t;
310 
311 typedef struct zio_cksum_report zio_cksum_report_t;
312 
313 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
314     const void *good_data);
315 typedef void zio_cksum_free_f(void *cbdata, size_t size);
316 
317 struct zio_bad_cksum;				/* defined in zio_checksum.h */
318 struct dnode_phys;
319 
320 struct zio_cksum_report {
321 	struct zio_cksum_report *zcr_next;
322 	nvlist_t		*zcr_ereport;
323 	nvlist_t		*zcr_detector;
324 	void			*zcr_cbdata;
325 	size_t			zcr_cbinfo;	/* passed to zcr_free() */
326 	uint64_t		zcr_align;
327 	uint64_t		zcr_length;
328 	zio_cksum_finish_f	*zcr_finish;
329 	zio_cksum_free_f	*zcr_free;
330 
331 	/* internal use only */
332 	struct zio_bad_cksum	*zcr_ckinfo;	/* information from failure */
333 };
334 
335 typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
336     void *arg);
337 
338 zio_vsd_cksum_report_f	zio_vsd_default_cksum_report;
339 
340 typedef struct zio_vsd_ops {
341 	zio_done_func_t		*vsd_free;
342 	zio_vsd_cksum_report_f	*vsd_cksum_report;
343 } zio_vsd_ops_t;
344 
345 typedef struct zio_gang_node {
346 	zio_gbh_phys_t		*gn_gbh;
347 	struct zio_gang_node	*gn_child[SPA_GBH_NBLKPTRS];
348 } zio_gang_node_t;
349 
350 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
351     zio_gang_node_t *gn, void *data);
352 
353 typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size);
354 
355 typedef struct zio_transform {
356 	void			*zt_orig_data;
357 	uint64_t		zt_orig_size;
358 	uint64_t		zt_bufsize;
359 	zio_transform_func_t	*zt_transform;
360 	struct zio_transform	*zt_next;
361 } zio_transform_t;
362 
363 typedef int zio_pipe_stage_t(zio_t *zio);
364 
365 /*
366  * The io_reexecute flags are distinct from io_flags because the child must
367  * be able to propagate them to the parent.  The normal io_flags are local
368  * to the zio, not protected by any lock, and not modifiable by children;
369  * the reexecute flags are protected by io_lock, modifiable by children,
370  * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
371  */
372 #define	ZIO_REEXECUTE_NOW	0x01
373 #define	ZIO_REEXECUTE_SUSPEND	0x02
374 
375 typedef struct zio_link {
376 	zio_t		*zl_parent;
377 	zio_t		*zl_child;
378 	list_node_t	zl_parent_node;
379 	list_node_t	zl_child_node;
380 } zio_link_t;
381 
382 struct zio {
383 	/* Core information about this I/O */
384 	zbookmark_phys_t	io_bookmark;
385 	zio_prop_t	io_prop;
386 	zio_type_t	io_type;
387 	enum zio_child	io_child_type;
388 	int		io_cmd;
389 	zio_priority_t	io_priority;
390 	uint8_t		io_reexecute;
391 	uint8_t		io_state[ZIO_WAIT_TYPES];
392 	uint64_t	io_txg;
393 	spa_t		*io_spa;
394 	blkptr_t	*io_bp;
395 	blkptr_t	*io_bp_override;
396 	blkptr_t	io_bp_copy;
397 	list_t		io_parent_list;
398 	list_t		io_child_list;
399 	zio_t		*io_logical;
400 	zio_transform_t *io_transform_stack;
401 
402 	/* Callback info */
403 	zio_done_func_t	*io_ready;
404 	zio_done_func_t	*io_children_ready;
405 	zio_done_func_t	*io_physdone;
406 	zio_done_func_t	*io_done;
407 	void		*io_private;
408 	int64_t		io_prev_space_delta;	/* DMU private */
409 	blkptr_t	io_bp_orig;
410 
411 	/* Data represented by this I/O */
412 	void		*io_data;
413 	void		*io_orig_data;
414 	uint64_t	io_size;
415 	uint64_t	io_orig_size;
416 
417 	/* Stuff for the vdev stack */
418 	vdev_t		*io_vd;
419 	void		*io_vsd;
420 	const zio_vsd_ops_t *io_vsd_ops;
421 
422 	uint64_t	io_offset;
423 	hrtime_t	io_timestamp;
424 	hrtime_t	io_queued_timestamp;
425 	hrtime_t	io_target_timestamp;
426 	avl_node_t	io_queue_node;
427 	avl_node_t	io_offset_node;
428 	avl_node_t	io_alloc_node;
429 
430 	/* Internal pipeline state */
431 	enum zio_flag	io_flags;
432 	enum zio_stage	io_stage;
433 	enum zio_stage	io_pipeline;
434 	enum zio_flag	io_orig_flags;
435 	enum zio_stage	io_orig_stage;
436 	enum zio_stage	io_orig_pipeline;
437 	enum zio_stage	io_pipeline_trace;
438 	int		io_error;
439 	int		io_child_error[ZIO_CHILD_TYPES];
440 	uint64_t	io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
441 	uint64_t	io_child_count;
442 	uint64_t	io_phys_children;
443 	uint64_t	io_parent_count;
444 	uint64_t	*io_stall;
445 	zio_t		*io_gang_leader;
446 	zio_gang_node_t	*io_gang_tree;
447 	void		*io_executor;
448 	void		*io_waiter;
449 	kmutex_t	io_lock;
450 	kcondvar_t	io_cv;
451 
452 	/* FMA state */
453 	zio_cksum_report_t *io_cksum_report;
454 	uint64_t	io_ena;
455 
456 	/* Taskq dispatching state */
457 	taskq_ent_t	io_tqent;
458 };
459 
460 extern int zio_timestamp_compare(const void *, const void *);
461 
462 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
463     zio_done_func_t *done, void *private, enum zio_flag flags);
464 
465 extern zio_t *zio_root(spa_t *spa,
466     zio_done_func_t *done, void *private, enum zio_flag flags);
467 
468 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
469     uint64_t size, zio_done_func_t *done, void *private,
470     zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
471 
472 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
473     void *data, uint64_t size, const zio_prop_t *zp,
474     zio_done_func_t *ready, zio_done_func_t *children_ready,
475     zio_done_func_t *physdone, zio_done_func_t *done,
476     void *private, zio_priority_t priority, enum zio_flag flags,
477     const zbookmark_phys_t *zb);
478 
479 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
480     void *data, uint64_t size, zio_done_func_t *done, void *private,
481     zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
482 
483 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
484     boolean_t nopwrite);
485 
486 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
487 
488 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
489     const blkptr_t *bp,
490     zio_done_func_t *done, void *private, enum zio_flag flags);
491 
492 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
493     zio_done_func_t *done, void *private, enum zio_flag flags);
494 
495 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
496     uint64_t size, void *data, int checksum,
497     zio_done_func_t *done, void *private, zio_priority_t priority,
498     enum zio_flag flags, boolean_t labels);
499 
500 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
501     uint64_t size, void *data, int checksum,
502     zio_done_func_t *done, void *private, zio_priority_t priority,
503     enum zio_flag flags, boolean_t labels);
504 
505 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
506     const blkptr_t *bp, enum zio_flag flags);
507 
508 extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
509     blkptr_t *old_bp, uint64_t size, boolean_t use_slog);
510 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
511 extern void zio_flush(zio_t *zio, vdev_t *vd);
512 extern void zio_shrink(zio_t *zio, uint64_t size);
513 
514 extern int zio_wait(zio_t *zio);
515 extern void zio_nowait(zio_t *zio);
516 extern void zio_execute(zio_t *zio);
517 extern void zio_interrupt(zio_t *zio);
518 extern void zio_delay_init(zio_t *zio);
519 extern void zio_delay_interrupt(zio_t *zio);
520 
521 extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **);
522 extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **);
523 extern zio_t *zio_unique_parent(zio_t *cio);
524 extern void zio_add_child(zio_t *pio, zio_t *cio);
525 
526 extern void *zio_buf_alloc(size_t size);
527 extern void zio_buf_free(void *buf, size_t size);
528 extern void *zio_data_buf_alloc(size_t size);
529 extern void zio_data_buf_free(void *buf, size_t size);
530 
531 extern void zio_push_transform(zio_t *zio, void *data, uint64_t size,
532     uint64_t bufsize, zio_transform_func_t *transform);
533 extern void zio_pop_transforms(zio_t *zio);
534 
535 extern void zio_resubmit_stage_async(void *);
536 
537 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
538     uint64_t offset, void *data, uint64_t size, int type,
539     zio_priority_t priority, enum zio_flag flags,
540     zio_done_func_t *done, void *private);
541 
542 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
543     void *data, uint64_t size, int type, zio_priority_t priority,
544     enum zio_flag flags, zio_done_func_t *done, void *private);
545 
546 extern void zio_vdev_io_bypass(zio_t *zio);
547 extern void zio_vdev_io_reissue(zio_t *zio);
548 extern void zio_vdev_io_redone(zio_t *zio);
549 
550 extern void zio_checksum_verified(zio_t *zio);
551 extern int zio_worst_error(int e1, int e2);
552 
553 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
554     enum zio_checksum parent);
555 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
556     enum zio_checksum child, enum zio_checksum parent);
557 extern enum zio_compress zio_compress_select(spa_t *spa,
558     enum zio_compress child, enum zio_compress parent);
559 
560 extern void zio_suspend(spa_t *spa, zio_t *zio);
561 extern int zio_resume(spa_t *spa);
562 extern void zio_resume_wait(spa_t *spa);
563 
564 /*
565  * Initial setup and teardown.
566  */
567 extern void zio_init(void);
568 extern void zio_fini(void);
569 
570 /*
571  * Fault injection
572  */
573 struct zinject_record;
574 extern uint32_t zio_injection_enabled;
575 extern int zio_inject_fault(char *name, int flags, int *id,
576     struct zinject_record *record);
577 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
578     struct zinject_record *record);
579 extern int zio_clear_fault(int id);
580 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
581 extern int zio_handle_fault_injection(zio_t *zio, int error);
582 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
583 extern int zio_handle_label_injection(zio_t *zio, int error);
584 extern void zio_handle_ignored_writes(zio_t *zio);
585 extern hrtime_t zio_handle_io_delay(zio_t *zio);
586 
587 /*
588  * Checksum ereport functions
589  */
590 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
591     uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
592 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
593     const void *good_data, const void *bad_data, boolean_t drop_if_identical);
594 
595 extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
596 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
597 
598 /* If we have the good data in hand, this function can be used */
599 extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
600     struct zio *zio, uint64_t offset, uint64_t length,
601     const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
602 
603 /* Called from spa_sync(), but primarily an injection handler */
604 extern void spa_handle_ignored_writes(spa_t *spa);
605 
606 /* zbookmark_phys functions */
607 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
608     const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
609 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
610     uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
611 
612 #ifdef	__cplusplus
613 }
614 #endif
615 
616 #endif	/* _ZIO_H */
617