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.
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.
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]
23 * Copyright 2008 Sun Microsystems, Inc. All rights reserved.
24 * Use is subject to license terms.
30 #pragma ident "@(#)zio.h 1.20 08/04/01 SMI"
32 #include <sys/zfs_context.h>
37 #include <sys/fs/zfs.h>
38 #include <sys/zio_impl.h>
44 #define ZBT_MAGIC 0x210da7ab10c7a11ULL /* zio data bloc tail */
46 typedef struct zio_block_tail
{
47 uint64_t zbt_magic
; /* for validation, endianness */
48 zio_cksum_t zbt_cksum
; /* 256-bit checksum */
52 * Gang block headers are self-checksumming and contain an array
55 #define SPA_GANGBLOCKSIZE SPA_MINBLOCKSIZE
56 #define SPA_GBH_NBLKPTRS ((SPA_GANGBLOCKSIZE - \
57 sizeof (zio_block_tail_t)) / sizeof (blkptr_t))
58 #define SPA_GBH_FILLER ((SPA_GANGBLOCKSIZE - \
59 sizeof (zio_block_tail_t) - \
60 (SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\
63 #define ZIO_GET_IOSIZE(zio) \
64 (BP_IS_GANG((zio)->io_bp) ? \
65 SPA_GANGBLOCKSIZE : BP_GET_PSIZE((zio)->io_bp))
67 typedef struct zio_gbh
{
68 blkptr_t zg_blkptr
[SPA_GBH_NBLKPTRS
];
69 uint64_t zg_filler
[SPA_GBH_FILLER
];
70 zio_block_tail_t zg_tail
;
74 ZIO_CHECKSUM_INHERIT
= 0,
78 ZIO_CHECKSUM_GANG_HEADER
,
80 ZIO_CHECKSUM_FLETCHER_2
,
81 ZIO_CHECKSUM_FLETCHER_4
,
83 ZIO_CHECKSUM_FUNCTIONS
86 #define ZIO_CHECKSUM_ON_VALUE ZIO_CHECKSUM_FLETCHER_2
87 #define ZIO_CHECKSUM_DEFAULT ZIO_CHECKSUM_ON
90 ZIO_COMPRESS_INHERIT
= 0,
104 ZIO_COMPRESS_FUNCTIONS
107 #define ZIO_COMPRESS_ON_VALUE ZIO_COMPRESS_LZJB
108 #define ZIO_COMPRESS_DEFAULT ZIO_COMPRESS_OFF
110 #define ZIO_FAILURE_MODE_WAIT 0
111 #define ZIO_FAILURE_MODE_CONTINUE 1
112 #define ZIO_FAILURE_MODE_PANIC 2
114 #define ZIO_PRIORITY_NOW (zio_priority_table[0])
115 #define ZIO_PRIORITY_SYNC_READ (zio_priority_table[1])
116 #define ZIO_PRIORITY_SYNC_WRITE (zio_priority_table[2])
117 #define ZIO_PRIORITY_ASYNC_READ (zio_priority_table[3])
118 #define ZIO_PRIORITY_ASYNC_WRITE (zio_priority_table[4])
119 #define ZIO_PRIORITY_FREE (zio_priority_table[5])
120 #define ZIO_PRIORITY_CACHE_FILL (zio_priority_table[6])
121 #define ZIO_PRIORITY_LOG_WRITE (zio_priority_table[7])
122 #define ZIO_PRIORITY_RESILVER (zio_priority_table[8])
123 #define ZIO_PRIORITY_SCRUB (zio_priority_table[9])
124 #define ZIO_PRIORITY_TABLE_SIZE 10
126 #define ZIO_FLAG_MUSTSUCCEED 0x00000
127 #define ZIO_FLAG_CANFAIL 0x00001
128 #define ZIO_FLAG_FAILFAST 0x00002
129 #define ZIO_FLAG_CONFIG_HELD 0x00004
130 #define ZIO_FLAG_CONFIG_GRABBED 0x00008
132 #define ZIO_FLAG_DONT_CACHE 0x00010
133 #define ZIO_FLAG_DONT_QUEUE 0x00020
134 #define ZIO_FLAG_DONT_PROPAGATE 0x00040
135 #define ZIO_FLAG_DONT_RETRY 0x00080
137 #define ZIO_FLAG_PHYSICAL 0x00100
138 #define ZIO_FLAG_IO_BYPASS 0x00200
139 #define ZIO_FLAG_IO_REPAIR 0x00400
140 #define ZIO_FLAG_SPECULATIVE 0x00800
142 #define ZIO_FLAG_RESILVER 0x01000
143 #define ZIO_FLAG_SCRUB 0x02000
144 #define ZIO_FLAG_SCRUB_THREAD 0x04000
145 #define ZIO_FLAG_SUBBLOCK 0x08000
147 #define ZIO_FLAG_NOBOOKMARK 0x10000
148 #define ZIO_FLAG_USER 0x20000
149 #define ZIO_FLAG_METADATA 0x40000
150 #define ZIO_FLAG_WRITE_RETRY 0x80000
152 #define ZIO_FLAG_GANG_INHERIT \
153 (ZIO_FLAG_CANFAIL | \
154 ZIO_FLAG_FAILFAST | \
155 ZIO_FLAG_CONFIG_HELD | \
156 ZIO_FLAG_DONT_CACHE | \
157 ZIO_FLAG_DONT_RETRY | \
158 ZIO_FLAG_IO_REPAIR | \
159 ZIO_FLAG_SPECULATIVE | \
160 ZIO_FLAG_RESILVER | \
162 ZIO_FLAG_SCRUB_THREAD | \
166 #define ZIO_FLAG_VDEV_INHERIT \
167 (ZIO_FLAG_GANG_INHERIT | \
170 #define ZIO_FLAG_RETRY_INHERIT \
171 (ZIO_FLAG_VDEV_INHERIT | \
172 ZIO_FLAG_CONFIG_GRABBED | \
173 ZIO_FLAG_DONT_PROPAGATE | \
177 #define ZIO_PIPELINE_CONTINUE 0x100
178 #define ZIO_PIPELINE_STOP 0x101
181 * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent
182 * graveyard) to indicate checksum errors and fragmentation.
187 typedef struct zio zio_t
;
188 typedef void zio_done_func_t(zio_t
*zio
);
190 extern uint8_t zio_priority_table
[ZIO_PRIORITY_TABLE_SIZE
];
191 extern char *zio_type_name
[ZIO_TYPES
];
194 * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
195 * identifies any block in the pool. By convention, the meta-objset (MOS)
196 * is objset 0, the meta-dnode is object 0, the root block (osphys_t) is
197 * level -1 of the meta-dnode, and intent log blocks (which are chained
198 * off the root block) have blkid == sequence number. In summary:
201 * meta-dnode is object 0
202 * root block is <objset, 0, -1, 0>
203 * intent log is <objset, 0, -1, ZIL sequence number>
205 * Note: this structure is called a bookmark because its first purpose was
206 * to remember where to resume a pool-wide traverse. The absolute ordering
207 * for block visitation during traversal is defined in compare_bookmark().
209 * Note: this structure is passed between userland and the kernel.
210 * Therefore it must not change size or alignment between 32/64 bit
211 * compilation options.
213 typedef struct zbookmark
{
221 /* Core information about this I/O */
225 zbookmark_t io_bookmark
;
226 enum zio_checksum io_checksum
;
227 enum zio_compress io_compress
;
233 zio_t
*io_sibling_prev
;
234 zio_t
*io_sibling_next
;
235 zio_transform_t
*io_transform_stack
;
237 list_node_t zio_link_node
;
240 zio_done_func_t
*io_ready
;
241 zio_done_func_t
*io_done
;
245 /* Data represented by this I/O */
249 /* Stuff for the vdev stack */
253 uint64_t io_deadline
;
254 uint64_t io_timestamp
;
255 avl_node_t io_offset_node
;
256 avl_node_t io_deadline_node
;
257 avl_tree_t
*io_vdev_tree
;
258 zio_t
*io_delegate_list
;
259 zio_t
*io_delegate_next
;
261 /* Internal pipeline state */
264 enum zio_type io_type
;
265 enum zio_stage io_stage
;
266 enum zio_stage io_orig_stage
;
269 struct dk_callback io_dk_callback
;
273 uint32_t io_numerrors
;
274 uint32_t io_pipeline
;
275 uint32_t io_orig_pipeline
;
276 uint64_t io_children_notready
;
277 uint64_t io_children_notdone
;
286 extern zio_t
*zio_null(zio_t
*pio
, spa_t
*spa
,
287 zio_done_func_t
*done
, void *private, int flags
);
289 extern zio_t
*zio_root(spa_t
*spa
,
290 zio_done_func_t
*done
, void *private, int flags
);
292 extern zio_t
*zio_read(zio_t
*pio
, spa_t
*spa
, blkptr_t
*bp
, void *data
,
293 uint64_t size
, zio_done_func_t
*done
, void *private,
294 int priority
, int flags
, zbookmark_t
*zb
);
296 extern zio_t
*zio_write(zio_t
*pio
, spa_t
*spa
, int checksum
, int compress
,
297 int ncopies
, uint64_t txg
, blkptr_t
*bp
, void *data
, uint64_t size
,
298 zio_done_func_t
*ready
, zio_done_func_t
*done
, void *private, int priority
,
299 int flags
, zbookmark_t
*zb
);
301 extern zio_t
*zio_rewrite(zio_t
*pio
, spa_t
*spa
, int checksum
,
302 uint64_t txg
, blkptr_t
*bp
, void *data
, uint64_t size
,
303 zio_done_func_t
*done
, void *private, int priority
, int flags
,
306 extern zio_t
*zio_free(zio_t
*pio
, spa_t
*spa
, uint64_t txg
, blkptr_t
*bp
,
307 zio_done_func_t
*done
, void *private);
309 extern zio_t
*zio_claim(zio_t
*pio
, spa_t
*spa
, uint64_t txg
, blkptr_t
*bp
,
310 zio_done_func_t
*done
, void *private);
312 extern zio_t
*zio_ioctl(zio_t
*pio
, spa_t
*spa
, vdev_t
*vd
, int cmd
,
313 zio_done_func_t
*done
, void *private, int priority
, int flags
);
315 extern zio_t
*zio_read_phys(zio_t
*pio
, vdev_t
*vd
, uint64_t offset
,
316 uint64_t size
, void *data
, int checksum
,
317 zio_done_func_t
*done
, void *private, int priority
, int flags
,
320 extern zio_t
*zio_write_phys(zio_t
*pio
, vdev_t
*vd
, uint64_t offset
,
321 uint64_t size
, void *data
, int checksum
,
322 zio_done_func_t
*done
, void *private, int priority
, int flags
,
325 extern int zio_alloc_blk(spa_t
*spa
, uint64_t size
, blkptr_t
*new_bp
,
326 blkptr_t
*old_bp
, uint64_t txg
);
327 extern void zio_free_blk(spa_t
*spa
, blkptr_t
*bp
, uint64_t txg
);
328 extern void zio_flush(zio_t
*zio
, vdev_t
*vd
);
330 extern int zio_wait(zio_t
*zio
);
331 extern void zio_nowait(zio_t
*zio
);
332 extern void zio_execute(zio_t
*zio
);
333 extern void zio_interrupt(zio_t
*zio
);
335 extern int zio_wait_for_children_ready(zio_t
*zio
);
336 extern int zio_wait_for_children_done(zio_t
*zio
);
338 extern void *zio_buf_alloc(size_t size
);
339 extern void zio_buf_free(void *buf
, size_t size
);
340 extern void *zio_data_buf_alloc(size_t size
);
341 extern void zio_data_buf_free(void *buf
, size_t size
);
343 extern void zio_resubmit_stage_async(void *);
346 * Delegate I/O to a child vdev.
348 extern zio_t
*zio_vdev_child_io(zio_t
*zio
, blkptr_t
*bp
, vdev_t
*vd
,
349 uint64_t offset
, void *data
, uint64_t size
, int type
, int priority
,
350 int flags
, zio_done_func_t
*done
, void *private);
352 extern void zio_vdev_io_bypass(zio_t
*zio
);
353 extern void zio_vdev_io_reissue(zio_t
*zio
);
354 extern void zio_vdev_io_redone(zio_t
*zio
);
356 extern void zio_checksum_verified(zio_t
*zio
);
357 extern void zio_set_gang_verifier(zio_t
*zio
, zio_cksum_t
*zcp
);
359 extern uint8_t zio_checksum_select(uint8_t child
, uint8_t parent
);
360 extern uint8_t zio_compress_select(uint8_t child
, uint8_t parent
);
362 extern boolean_t
zio_should_retry(zio_t
*zio
);
363 extern int zio_vdev_resume_io(spa_t
*);
366 * Initial setup and teardown.
368 extern void zio_init(void);
369 extern void zio_fini(void);
374 struct zinject_record
;
375 extern uint32_t zio_injection_enabled
;
376 extern int zio_inject_fault(char *name
, int flags
, int *id
,
377 struct zinject_record
*record
);
378 extern int zio_inject_list_next(int *id
, char *name
, size_t buflen
,
379 struct zinject_record
*record
);
380 extern int zio_clear_fault(int id
);
381 extern int zio_handle_fault_injection(zio_t
*zio
, int error
);
382 extern int zio_handle_device_injection(vdev_t
*vd
, int error
);