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)
120 * Default Linux timeout for a sd device.
122 #define ZIO_DELAY_MAX (30 * MILLISEC)
124 #define ZIO_FAILURE_MODE_WAIT 0
125 #define ZIO_FAILURE_MODE_CONTINUE 1
126 #define ZIO_FAILURE_MODE_PANIC 2
128 #define ZIO_PRIORITY_NOW (zio_priority_table[0])
129 #define ZIO_PRIORITY_SYNC_READ (zio_priority_table[1])
130 #define ZIO_PRIORITY_SYNC_WRITE (zio_priority_table[2])
131 #define ZIO_PRIORITY_LOG_WRITE (zio_priority_table[3])
132 #define ZIO_PRIORITY_CACHE_FILL (zio_priority_table[4])
133 #define ZIO_PRIORITY_AGG (zio_priority_table[5])
134 #define ZIO_PRIORITY_FREE (zio_priority_table[6])
135 #define ZIO_PRIORITY_ASYNC_WRITE (zio_priority_table[7])
136 #define ZIO_PRIORITY_ASYNC_READ (zio_priority_table[8])
137 #define ZIO_PRIORITY_RESILVER (zio_priority_table[9])
138 #define ZIO_PRIORITY_SCRUB (zio_priority_table[10])
139 #define ZIO_PRIORITY_DDT_PREFETCH (zio_priority_table[11])
140 #define ZIO_PRIORITY_TABLE_SIZE 12
142 #define ZIO_PIPELINE_CONTINUE 0x100
143 #define ZIO_PIPELINE_STOP 0x101
147 * Flags inherited by gang, ddt, and vdev children,
148 * and that must be equal for two zios to aggregate
150 ZIO_FLAG_DONT_AGGREGATE = 1 << 0,
151 ZIO_FLAG_IO_REPAIR = 1 << 1,
152 ZIO_FLAG_SELF_HEAL = 1 << 2,
153 ZIO_FLAG_RESILVER = 1 << 3,
154 ZIO_FLAG_SCRUB = 1 << 4,
155 ZIO_FLAG_SCAN_THREAD = 1 << 5,
157 #define ZIO_FLAG_AGG_INHERIT (ZIO_FLAG_CANFAIL - 1)
160 * Flags inherited by ddt, gang, and vdev children.
162 ZIO_FLAG_CANFAIL = 1 << 6, /* must be first for INHERIT */
163 ZIO_FLAG_SPECULATIVE = 1 << 7,
164 ZIO_FLAG_CONFIG_WRITER = 1 << 8,
165 ZIO_FLAG_DONT_RETRY = 1 << 9,
166 ZIO_FLAG_DONT_CACHE = 1 << 10,
167 ZIO_FLAG_NODATA = 1 << 11,
168 ZIO_FLAG_INDUCE_DAMAGE = 1 << 12,
170 #define ZIO_FLAG_DDT_INHERIT (ZIO_FLAG_IO_RETRY - 1)
171 #define ZIO_FLAG_GANG_INHERIT (ZIO_FLAG_IO_RETRY - 1)
174 * Flags inherited by vdev children.
176 ZIO_FLAG_IO_RETRY = 1 << 13, /* must be first for INHERIT */
177 ZIO_FLAG_PROBE = 1 << 14,
178 ZIO_FLAG_TRYHARD = 1 << 15,
179 ZIO_FLAG_OPTIONAL = 1 << 16,
181 #define ZIO_FLAG_VDEV_INHERIT (ZIO_FLAG_DONT_QUEUE - 1)
184 * Flags not inherited by any children.
186 ZIO_FLAG_DONT_QUEUE = 1 << 17, /* must be first for INHERIT */
187 ZIO_FLAG_DONT_PROPAGATE = 1 << 18,
188 ZIO_FLAG_IO_BYPASS = 1 << 19,
189 ZIO_FLAG_IO_REWRITE = 1 << 20,
190 ZIO_FLAG_RAW = 1 << 21,
191 ZIO_FLAG_GANG_CHILD = 1 << 22,
192 ZIO_FLAG_DDT_CHILD = 1 << 23,
193 ZIO_FLAG_GODFATHER = 1 << 24
196 #define ZIO_FLAG_MUSTSUCCEED 0
198 #define ZIO_DDT_CHILD_FLAGS(zio) \
199 (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) | \
200 ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
202 #define ZIO_GANG_CHILD_FLAGS(zio) \
203 (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) | \
204 ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
206 #define ZIO_VDEV_CHILD_FLAGS(zio) \
207 (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) | \
225 * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent
226 * graveyard) to indicate checksum errors and fragmentation.
231 typedef void zio_done_func_t(zio_t *zio);
233 extern uint8_t zio_priority_table[ZIO_PRIORITY_TABLE_SIZE];
234 extern char *zio_type_name[ZIO_TYPES];
237 * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
238 * identifies any block in the pool. By convention, the meta-objset (MOS)
239 * is objset 0, and the meta-dnode is object 0. This covers all blocks
240 * except root blocks and ZIL blocks, which are defined as follows:
242 * Root blocks (objset_phys_t) are object 0, level -1: <objset, 0, -1, 0>.
243 * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
244 * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
246 * Note: this structure is called a bookmark because its original purpose
247 * was to remember where to resume a pool-wide traverse.
249 * Note: this structure is passed between userland and the kernel.
250 * Therefore it must not change size or alignment between 32/64 bit
251 * compilation options.
253 typedef struct zbookmark {
260 #define SET_BOOKMARK(zb, objset, object, level, blkid) \
262 (zb)->zb_objset = objset; \
263 (zb)->zb_object = object; \
264 (zb)->zb_level = level; \
265 (zb)->zb_blkid = blkid; \
268 #define ZB_DESTROYED_OBJSET (-1ULL)
270 #define ZB_ROOT_OBJECT (0ULL)
271 #define ZB_ROOT_LEVEL (-1LL)
272 #define ZB_ROOT_BLKID (0ULL)
274 #define ZB_ZIL_OBJECT (0ULL)
275 #define ZB_ZIL_LEVEL (-2LL)
277 typedef struct zio_prop {
278 enum zio_checksum zp_checksum;
279 enum zio_compress zp_compress;
280 dmu_object_type_t zp_type;
284 uint8_t zp_dedup_verify;
287 typedef struct zio_cksum_report zio_cksum_report_t;
289 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
290 const void *good_data);
291 typedef void zio_cksum_free_f(void *cbdata, size_t size);
293 struct zio_bad_cksum; /* defined in zio_checksum.h */
295 struct zio_cksum_report {
296 struct zio_cksum_report *zcr_next;
297 nvlist_t *zcr_ereport;
298 nvlist_t *zcr_detector;
300 size_t zcr_cbinfo; /* passed to zcr_free() */
303 zio_cksum_finish_f *zcr_finish;
304 zio_cksum_free_f *zcr_free;
306 /* internal use only */
307 struct zio_bad_cksum *zcr_ckinfo; /* information from failure */
310 typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
313 zio_vsd_cksum_report_f zio_vsd_default_cksum_report;
315 typedef struct zio_vsd_ops {
316 zio_done_func_t *vsd_free;
317 zio_vsd_cksum_report_f *vsd_cksum_report;
320 typedef struct zio_gang_node {
321 zio_gbh_phys_t *gn_gbh;
322 struct zio_gang_node *gn_child[SPA_GBH_NBLKPTRS];
325 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
326 zio_gang_node_t *gn, void *data);
328 typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size);
330 typedef struct zio_transform {
332 uint64_t zt_orig_size;
334 zio_transform_func_t *zt_transform;
335 struct zio_transform *zt_next;
338 typedef int zio_pipe_stage_t(zio_t *zio);
341 * The io_reexecute flags are distinct from io_flags because the child must
342 * be able to propagate them to the parent. The normal io_flags are local
343 * to the zio, not protected by any lock, and not modifiable by children;
344 * the reexecute flags are protected by io_lock, modifiable by children,
345 * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
347 #define ZIO_REEXECUTE_NOW 0x01
348 #define ZIO_REEXECUTE_SUSPEND 0x02
350 typedef struct zio_link {
353 list_node_t zl_parent_node;
354 list_node_t zl_child_node;
358 /* Core information about this I/O */
359 zbookmark_t io_bookmark;
362 enum zio_child io_child_type;
365 uint8_t io_reexecute;
366 uint8_t io_state[ZIO_WAIT_TYPES];
370 blkptr_t *io_bp_override;
372 list_t io_parent_list;
373 list_t io_child_list;
374 zio_link_t *io_walk_link;
376 zio_transform_t *io_transform_stack;
379 zio_done_func_t *io_ready;
380 zio_done_func_t *io_done;
382 int64_t io_prev_space_delta; /* DMU private */
385 /* Data represented by this I/O */
389 uint64_t io_orig_size;
391 /* Stuff for the vdev stack */
394 const zio_vsd_ops_t *io_vsd_ops;
397 uint64_t io_deadline;
398 avl_node_t io_offset_node;
399 avl_node_t io_deadline_node;
400 avl_tree_t *io_vdev_tree;
402 /* Internal pipeline state */
403 enum zio_flag io_flags;
404 enum zio_stage io_stage;
405 enum zio_stage io_pipeline;
406 enum zio_flag io_orig_flags;
407 enum zio_stage io_orig_stage;
408 enum zio_stage io_orig_pipeline;
411 int io_child_error[ZIO_CHILD_TYPES];
412 uint64_t io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
413 uint64_t io_child_count;
414 uint64_t io_parent_count;
416 zio_t *io_gang_leader;
417 zio_gang_node_t *io_gang_tree;
424 zio_cksum_report_t *io_cksum_report;
428 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
429 zio_done_func_t *done, void *private, enum zio_flag flags);
431 extern zio_t *zio_root(spa_t *spa,
432 zio_done_func_t *done, void *private, enum zio_flag flags);
434 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
435 uint64_t size, zio_done_func_t *done, void *private,
436 int priority, enum zio_flag flags, const zbookmark_t *zb);
438 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
439 void *data, uint64_t size, const zio_prop_t *zp,
440 zio_done_func_t *ready, zio_done_func_t *done, void *private,
441 int priority, enum zio_flag flags, const zbookmark_t *zb);
443 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
444 void *data, uint64_t size, zio_done_func_t *done, void *private,
445 int priority, enum zio_flag flags, zbookmark_t *zb);
447 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies);
449 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
451 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
453 zio_done_func_t *done, void *private, enum zio_flag flags);
455 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
456 zio_done_func_t *done, void *private, int priority, enum zio_flag flags);
458 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
459 uint64_t size, void *data, int checksum,
460 zio_done_func_t *done, void *private, int priority, enum zio_flag flags,
463 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
464 uint64_t size, void *data, int checksum,
465 zio_done_func_t *done, void *private, int priority, enum zio_flag flags,
468 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
469 const blkptr_t *bp, enum zio_flag flags);
471 extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
472 blkptr_t *old_bp, uint64_t size, boolean_t use_slog);
473 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
474 extern void zio_flush(zio_t *zio, vdev_t *vd);
475 extern void zio_shrink(zio_t *zio, uint64_t size);
477 extern int zio_wait(zio_t *zio);
478 extern void zio_nowait(zio_t *zio);
479 extern void zio_execute(zio_t *zio);
480 extern void zio_interrupt(zio_t *zio);
482 extern zio_t *zio_walk_parents(zio_t *cio);
483 extern zio_t *zio_walk_children(zio_t *pio);
484 extern zio_t *zio_unique_parent(zio_t *cio);
485 extern void zio_add_child(zio_t *pio, zio_t *cio);
487 extern void *zio_buf_alloc(size_t size);
488 extern void zio_buf_free(void *buf, size_t size);
489 extern void *zio_data_buf_alloc(size_t size);
490 extern void zio_data_buf_free(void *buf, size_t size);
492 extern void zio_resubmit_stage_async(void *);
494 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
495 uint64_t offset, void *data, uint64_t size, int type, int priority,
496 enum zio_flag flags, zio_done_func_t *done, void *private);
498 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
499 void *data, uint64_t size, int type, int priority,
500 enum zio_flag flags, zio_done_func_t *done, void *private);
502 extern void zio_vdev_io_bypass(zio_t *zio);
503 extern void zio_vdev_io_reissue(zio_t *zio);
504 extern void zio_vdev_io_redone(zio_t *zio);
506 extern void zio_checksum_verified(zio_t *zio);
507 extern int zio_worst_error(int e1, int e2);
509 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
510 enum zio_checksum parent);
511 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
512 enum zio_checksum child, enum zio_checksum parent);
513 extern enum zio_compress zio_compress_select(enum zio_compress child,
514 enum zio_compress parent);
516 extern void zio_suspend(spa_t *spa, zio_t *zio);
517 extern int zio_resume(spa_t *spa);
518 extern void zio_resume_wait(spa_t *spa);
521 * Initial setup and teardown.
523 extern void zio_init(void);
524 extern void zio_fini(void);
529 struct zinject_record;
530 extern uint32_t zio_injection_enabled;
531 extern int zio_inject_fault(char *name, int flags, int *id,
532 struct zinject_record *record);
533 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
534 struct zinject_record *record);
535 extern int zio_clear_fault(int id);
536 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
537 extern int zio_handle_fault_injection(zio_t *zio, int error);
538 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
539 extern int zio_handle_label_injection(zio_t *zio, int error);
540 extern void zio_handle_ignored_writes(zio_t *zio);
543 * Checksum ereport functions
545 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
546 uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
547 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
548 const void *good_data, const void *bad_data, boolean_t drop_if_identical);
550 extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
551 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
553 /* If we have the good data in hand, this function can be used */
554 extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
555 struct zio *zio, uint64_t offset, uint64_t length,
556 const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
558 /* Called from spa_sync(), but primarily an injection handler */
559 extern void spa_handle_ignored_writes(spa_t *spa);