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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
29 #include <sys/zfs_context.h>
33 #include <sys/fs/zfs.h>
34 #include <sys/zio_impl.h>
43 #define ZEC_MAGIC 0x210da7ab10c7a11ULL
45 typedef struct zio_eck
{
46 uint64_t zec_magic
; /* for validation, endianness */
47 zio_cksum_t zec_cksum
; /* 256-bit checksum */
51 * Gang block headers are self-checksumming and contain an array
54 #define SPA_GANGBLOCKSIZE SPA_MINBLOCKSIZE
55 #define SPA_GBH_NBLKPTRS ((SPA_GANGBLOCKSIZE - \
56 sizeof (zio_eck_t)) / sizeof (blkptr_t))
57 #define SPA_GBH_FILLER ((SPA_GANGBLOCKSIZE - \
58 sizeof (zio_eck_t) - \
59 (SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\
62 typedef struct zio_gbh
{
63 blkptr_t zg_blkptr
[SPA_GBH_NBLKPTRS
];
64 uint64_t zg_filler
[SPA_GBH_FILLER
];
69 ZIO_CHECKSUM_INHERIT
= 0,
73 ZIO_CHECKSUM_GANG_HEADER
,
75 ZIO_CHECKSUM_FLETCHER_2
,
76 ZIO_CHECKSUM_FLETCHER_4
,
79 ZIO_CHECKSUM_FUNCTIONS
82 #define ZIO_CHECKSUM_ON_VALUE ZIO_CHECKSUM_FLETCHER_4
83 #define ZIO_CHECKSUM_DEFAULT ZIO_CHECKSUM_ON
85 #define ZIO_CHECKSUM_MASK 0xffULL
86 #define ZIO_CHECKSUM_VERIFY (1 << 8)
88 #define ZIO_DEDUPCHECKSUM ZIO_CHECKSUM_SHA256
89 #define ZIO_DEDUPDITTO_MIN 100
92 ZIO_COMPRESS_INHERIT
= 0,
107 ZIO_COMPRESS_FUNCTIONS
110 #define ZIO_COMPRESS_ON_VALUE ZIO_COMPRESS_LZJB
111 #define ZIO_COMPRESS_DEFAULT ZIO_COMPRESS_OFF
113 #define BOOTFS_COMPRESS_VALID(compress) \
114 ((compress) == ZIO_COMPRESS_LZJB || \
115 ((compress) == ZIO_COMPRESS_ON && \
116 ZIO_COMPRESS_ON_VALUE == ZIO_COMPRESS_LZJB) || \
117 (compress) == ZIO_COMPRESS_OFF)
119 #define ZIO_FAILURE_MODE_WAIT 0
120 #define ZIO_FAILURE_MODE_CONTINUE 1
121 #define ZIO_FAILURE_MODE_PANIC 2
123 #define ZIO_PRIORITY_NOW (zio_priority_table[0])
124 #define ZIO_PRIORITY_SYNC_READ (zio_priority_table[1])
125 #define ZIO_PRIORITY_SYNC_WRITE (zio_priority_table[2])
126 #define ZIO_PRIORITY_LOG_WRITE (zio_priority_table[3])
127 #define ZIO_PRIORITY_CACHE_FILL (zio_priority_table[4])
128 #define ZIO_PRIORITY_AGG (zio_priority_table[5])
129 #define ZIO_PRIORITY_FREE (zio_priority_table[6])
130 #define ZIO_PRIORITY_ASYNC_WRITE (zio_priority_table[7])
131 #define ZIO_PRIORITY_ASYNC_READ (zio_priority_table[8])
132 #define ZIO_PRIORITY_RESILVER (zio_priority_table[9])
133 #define ZIO_PRIORITY_SCRUB (zio_priority_table[10])
134 #define ZIO_PRIORITY_DDT_PREFETCH (zio_priority_table[11])
135 #define ZIO_PRIORITY_TABLE_SIZE 12
137 #define ZIO_PIPELINE_CONTINUE 0x100
138 #define ZIO_PIPELINE_STOP 0x101
142 * Flags inherited by gang, ddt, and vdev children,
143 * and that must be equal for two zios to aggregate
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,
152 #define ZIO_FLAG_AGG_INHERIT (ZIO_FLAG_CANFAIL - 1)
155 * Flags inherited by ddt, gang, and vdev children.
157 ZIO_FLAG_CANFAIL
= 1 << 6, /* must be first for INHERIT */
158 ZIO_FLAG_SPECULATIVE
= 1 << 7,
159 ZIO_FLAG_CONFIG_WRITER
= 1 << 8,
160 ZIO_FLAG_DONT_RETRY
= 1 << 9,
161 ZIO_FLAG_DONT_CACHE
= 1 << 10,
162 ZIO_FLAG_NODATA
= 1 << 11,
163 ZIO_FLAG_INDUCE_DAMAGE
= 1 << 12,
165 #define ZIO_FLAG_DDT_INHERIT (ZIO_FLAG_IO_RETRY - 1)
166 #define ZIO_FLAG_GANG_INHERIT (ZIO_FLAG_IO_RETRY - 1)
169 * Flags inherited by vdev children.
171 ZIO_FLAG_IO_RETRY
= 1 << 13, /* must be first for INHERIT */
172 ZIO_FLAG_PROBE
= 1 << 14,
173 ZIO_FLAG_TRYHARD
= 1 << 15,
174 ZIO_FLAG_OPTIONAL
= 1 << 16,
176 #define ZIO_FLAG_VDEV_INHERIT (ZIO_FLAG_DONT_QUEUE - 1)
179 * Flags not inherited by any children.
181 ZIO_FLAG_DONT_QUEUE
= 1 << 17, /* must be first for INHERIT */
182 ZIO_FLAG_DONT_PROPAGATE
= 1 << 18,
183 ZIO_FLAG_IO_BYPASS
= 1 << 19,
184 ZIO_FLAG_IO_REWRITE
= 1 << 20,
185 ZIO_FLAG_RAW
= 1 << 21,
186 ZIO_FLAG_GANG_CHILD
= 1 << 22,
187 ZIO_FLAG_DDT_CHILD
= 1 << 23,
188 ZIO_FLAG_GODFATHER
= 1 << 24
191 #define ZIO_FLAG_MUSTSUCCEED 0
193 #define ZIO_DDT_CHILD_FLAGS(zio) \
194 (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) | \
195 ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
197 #define ZIO_GANG_CHILD_FLAGS(zio) \
198 (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) | \
199 ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
201 #define ZIO_VDEV_CHILD_FLAGS(zio) \
202 (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) | \
220 * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent
221 * graveyard) to indicate checksum errors and fragmentation.
226 typedef void zio_done_func_t(zio_t
*zio
);
228 extern uint8_t zio_priority_table
[ZIO_PRIORITY_TABLE_SIZE
];
229 extern char *zio_type_name
[ZIO_TYPES
];
232 * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
233 * identifies any block in the pool. By convention, the meta-objset (MOS)
234 * is objset 0, and the meta-dnode is object 0. This covers all blocks
235 * except root blocks and ZIL blocks, which are defined as follows:
237 * Root blocks (objset_phys_t) are object 0, level -1: <objset, 0, -1, 0>.
238 * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
239 * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
241 * Note: this structure is called a bookmark because its original purpose
242 * was to remember where to resume a pool-wide traverse.
244 * Note: this structure is passed between userland and the kernel.
245 * Therefore it must not change size or alignment between 32/64 bit
246 * compilation options.
248 typedef struct zbookmark
{
255 #define SET_BOOKMARK(zb, objset, object, level, blkid) \
257 (zb)->zb_objset = objset; \
258 (zb)->zb_object = object; \
259 (zb)->zb_level = level; \
260 (zb)->zb_blkid = blkid; \
263 #define ZB_DESTROYED_OBJSET (-1ULL)
265 #define ZB_ROOT_OBJECT (0ULL)
266 #define ZB_ROOT_LEVEL (-1LL)
267 #define ZB_ROOT_BLKID (0ULL)
269 #define ZB_ZIL_OBJECT (0ULL)
270 #define ZB_ZIL_LEVEL (-2LL)
272 typedef struct zio_prop
{
273 enum zio_checksum zp_checksum
;
274 enum zio_compress zp_compress
;
275 dmu_object_type_t zp_type
;
279 uint8_t zp_dedup_verify
;
282 typedef struct zio_cksum_report zio_cksum_report_t
;
284 typedef void zio_cksum_finish_f(zio_cksum_report_t
*rep
,
285 const void *good_data
);
286 typedef void zio_cksum_free_f(void *cbdata
, size_t size
);
288 struct zio_bad_cksum
; /* defined in zio_checksum.h */
290 struct zio_cksum_report
{
291 struct zio_cksum_report
*zcr_next
;
292 nvlist_t
*zcr_ereport
;
293 nvlist_t
*zcr_detector
;
295 size_t zcr_cbinfo
; /* passed to zcr_free() */
298 zio_cksum_finish_f
*zcr_finish
;
299 zio_cksum_free_f
*zcr_free
;
301 /* internal use only */
302 struct zio_bad_cksum
*zcr_ckinfo
; /* information from failure */
305 typedef void zio_vsd_cksum_report_f(zio_t
*zio
, zio_cksum_report_t
*zcr
,
308 zio_vsd_cksum_report_f zio_vsd_default_cksum_report
;
310 typedef struct zio_vsd_ops
{
311 zio_done_func_t
*vsd_free
;
312 zio_vsd_cksum_report_f
*vsd_cksum_report
;
315 typedef struct zio_gang_node
{
316 zio_gbh_phys_t
*gn_gbh
;
317 struct zio_gang_node
*gn_child
[SPA_GBH_NBLKPTRS
];
320 typedef zio_t
*zio_gang_issue_func_t(zio_t
*zio
, blkptr_t
*bp
,
321 zio_gang_node_t
*gn
, void *data
);
323 typedef void zio_transform_func_t(zio_t
*zio
, void *data
, uint64_t size
);
325 typedef struct zio_transform
{
327 uint64_t zt_orig_size
;
329 zio_transform_func_t
*zt_transform
;
330 struct zio_transform
*zt_next
;
333 typedef int zio_pipe_stage_t(zio_t
*zio
);
336 * The io_reexecute flags are distinct from io_flags because the child must
337 * be able to propagate them to the parent. The normal io_flags are local
338 * to the zio, not protected by any lock, and not modifiable by children;
339 * the reexecute flags are protected by io_lock, modifiable by children,
340 * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
342 #define ZIO_REEXECUTE_NOW 0x01
343 #define ZIO_REEXECUTE_SUSPEND 0x02
345 typedef struct zio_link
{
348 list_node_t zl_parent_node
;
349 list_node_t zl_child_node
;
353 /* Core information about this I/O */
354 zbookmark_t io_bookmark
;
357 enum zio_child io_child_type
;
360 uint8_t io_reexecute
;
361 uint8_t io_state
[ZIO_WAIT_TYPES
];
365 blkptr_t
*io_bp_override
;
367 list_t io_parent_list
;
368 list_t io_child_list
;
369 zio_link_t
*io_walk_link
;
371 zio_transform_t
*io_transform_stack
;
374 zio_done_func_t
*io_ready
;
375 zio_done_func_t
*io_done
;
377 int64_t io_prev_space_delta
; /* DMU private */
380 /* Data represented by this I/O */
384 uint64_t io_orig_size
;
386 /* Stuff for the vdev stack */
389 const zio_vsd_ops_t
*io_vsd_ops
;
392 uint64_t io_deadline
;
393 avl_node_t io_offset_node
;
394 avl_node_t io_deadline_node
;
395 avl_tree_t
*io_vdev_tree
;
397 /* Internal pipeline state */
398 enum zio_flag io_flags
;
399 enum zio_stage io_stage
;
400 enum zio_stage io_pipeline
;
401 enum zio_flag io_orig_flags
;
402 enum zio_stage io_orig_stage
;
403 enum zio_stage io_orig_pipeline
;
405 int io_child_error
[ZIO_CHILD_TYPES
];
406 uint64_t io_children
[ZIO_CHILD_TYPES
][ZIO_WAIT_TYPES
];
407 uint64_t io_child_count
;
408 uint64_t io_parent_count
;
410 zio_t
*io_gang_leader
;
411 zio_gang_node_t
*io_gang_tree
;
418 zio_cksum_report_t
*io_cksum_report
;
422 extern zio_t
*zio_null(zio_t
*pio
, spa_t
*spa
, vdev_t
*vd
,
423 zio_done_func_t
*done
, void *private, enum zio_flag flags
);
425 extern zio_t
*zio_root(spa_t
*spa
,
426 zio_done_func_t
*done
, void *private, enum zio_flag flags
);
428 extern zio_t
*zio_read(zio_t
*pio
, spa_t
*spa
, const blkptr_t
*bp
, void *data
,
429 uint64_t size
, zio_done_func_t
*done
, void *private,
430 int priority
, enum zio_flag flags
, const zbookmark_t
*zb
);
432 extern zio_t
*zio_write(zio_t
*pio
, spa_t
*spa
, uint64_t txg
, blkptr_t
*bp
,
433 void *data
, uint64_t size
, const zio_prop_t
*zp
,
434 zio_done_func_t
*ready
, zio_done_func_t
*done
, void *private,
435 int priority
, enum zio_flag flags
, const zbookmark_t
*zb
);
437 extern zio_t
*zio_rewrite(zio_t
*pio
, spa_t
*spa
, uint64_t txg
, blkptr_t
*bp
,
438 void *data
, uint64_t size
, zio_done_func_t
*done
, void *private,
439 int priority
, enum zio_flag flags
, zbookmark_t
*zb
);
441 extern void zio_write_override(zio_t
*zio
, blkptr_t
*bp
, int copies
);
443 extern void zio_free(spa_t
*spa
, uint64_t txg
, const blkptr_t
*bp
);
445 extern zio_t
*zio_claim(zio_t
*pio
, spa_t
*spa
, uint64_t txg
,
447 zio_done_func_t
*done
, void *private, enum zio_flag flags
);
449 extern zio_t
*zio_ioctl(zio_t
*pio
, spa_t
*spa
, vdev_t
*vd
, int cmd
,
450 zio_done_func_t
*done
, void *private, int priority
, enum zio_flag flags
);
452 extern zio_t
*zio_read_phys(zio_t
*pio
, vdev_t
*vd
, uint64_t offset
,
453 uint64_t size
, void *data
, int checksum
,
454 zio_done_func_t
*done
, void *private, int priority
, enum zio_flag flags
,
457 extern zio_t
*zio_write_phys(zio_t
*pio
, vdev_t
*vd
, uint64_t offset
,
458 uint64_t size
, void *data
, int checksum
,
459 zio_done_func_t
*done
, void *private, int priority
, enum zio_flag flags
,
462 extern zio_t
*zio_free_sync(zio_t
*pio
, spa_t
*spa
, uint64_t txg
,
463 const blkptr_t
*bp
, enum zio_flag flags
);
465 extern int zio_alloc_zil(spa_t
*spa
, uint64_t txg
, blkptr_t
*new_bp
,
466 blkptr_t
*old_bp
, uint64_t size
, boolean_t use_slog
);
467 extern void zio_free_zil(spa_t
*spa
, uint64_t txg
, blkptr_t
*bp
);
468 extern void zio_flush(zio_t
*zio
, vdev_t
*vd
);
469 extern void zio_shrink(zio_t
*zio
, uint64_t size
);
471 extern int zio_wait(zio_t
*zio
);
472 extern void zio_nowait(zio_t
*zio
);
473 extern void zio_execute(zio_t
*zio
);
474 extern void zio_interrupt(zio_t
*zio
);
476 extern zio_t
*zio_walk_parents(zio_t
*cio
);
477 extern zio_t
*zio_walk_children(zio_t
*pio
);
478 extern zio_t
*zio_unique_parent(zio_t
*cio
);
479 extern void zio_add_child(zio_t
*pio
, zio_t
*cio
);
481 extern void *zio_buf_alloc(size_t size
);
482 extern void zio_buf_free(void *buf
, size_t size
);
483 extern void *zio_data_buf_alloc(size_t size
);
484 extern void zio_data_buf_free(void *buf
, size_t size
);
486 extern void zio_resubmit_stage_async(void *);
488 extern zio_t
*zio_vdev_child_io(zio_t
*zio
, blkptr_t
*bp
, vdev_t
*vd
,
489 uint64_t offset
, void *data
, uint64_t size
, int type
, int priority
,
490 enum zio_flag flags
, zio_done_func_t
*done
, void *private);
492 extern zio_t
*zio_vdev_delegated_io(vdev_t
*vd
, uint64_t offset
,
493 void *data
, uint64_t size
, int type
, int priority
,
494 enum zio_flag flags
, zio_done_func_t
*done
, void *private);
496 extern void zio_vdev_io_bypass(zio_t
*zio
);
497 extern void zio_vdev_io_reissue(zio_t
*zio
);
498 extern void zio_vdev_io_redone(zio_t
*zio
);
500 extern void zio_checksum_verified(zio_t
*zio
);
501 extern int zio_worst_error(int e1
, int e2
);
503 extern enum zio_checksum
zio_checksum_select(enum zio_checksum child
,
504 enum zio_checksum parent
);
505 extern enum zio_checksum
zio_checksum_dedup_select(spa_t
*spa
,
506 enum zio_checksum child
, enum zio_checksum parent
);
507 extern enum zio_compress
zio_compress_select(enum zio_compress child
,
508 enum zio_compress parent
);
510 extern void zio_suspend(spa_t
*spa
, zio_t
*zio
);
511 extern int zio_resume(spa_t
*spa
);
512 extern void zio_resume_wait(spa_t
*spa
);
515 * Initial setup and teardown.
517 extern void zio_init(void);
518 extern void zio_fini(void);
523 struct zinject_record
;
524 extern uint32_t zio_injection_enabled
;
525 extern int zio_inject_fault(char *name
, int flags
, int *id
,
526 struct zinject_record
*record
);
527 extern int zio_inject_list_next(int *id
, char *name
, size_t buflen
,
528 struct zinject_record
*record
);
529 extern int zio_clear_fault(int id
);
530 extern void zio_handle_panic_injection(spa_t
*spa
, char *tag
, uint64_t type
);
531 extern int zio_handle_fault_injection(zio_t
*zio
, int error
);
532 extern int zio_handle_device_injection(vdev_t
*vd
, zio_t
*zio
, int error
);
533 extern int zio_handle_label_injection(zio_t
*zio
, int error
);
534 extern void zio_handle_ignored_writes(zio_t
*zio
);
537 * Checksum ereport functions
539 extern void zfs_ereport_start_checksum(spa_t
*spa
, vdev_t
*vd
, struct zio
*zio
,
540 uint64_t offset
, uint64_t length
, void *arg
, struct zio_bad_cksum
*info
);
541 extern void zfs_ereport_finish_checksum(zio_cksum_report_t
*report
,
542 const void *good_data
, const void *bad_data
, boolean_t drop_if_identical
);
544 extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t
*report
);
545 extern void zfs_ereport_free_checksum(zio_cksum_report_t
*report
);
547 /* If we have the good data in hand, this function can be used */
548 extern void zfs_ereport_post_checksum(spa_t
*spa
, vdev_t
*vd
,
549 struct zio
*zio
, uint64_t offset
, uint64_t length
,
550 const void *good_data
, const void *bad_data
, struct zio_bad_cksum
*info
);
552 /* Called from spa_sync(), but primarily an injection handler */
553 extern void spa_handle_ignored_writes(spa_t
*spa
);