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.
24 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2012 by Delphix. All rights reserved.
26 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
32 #include <sys/zfs_context.h>
36 #include <sys/fs/zfs.h>
37 #include <sys/zio_impl.h>
46 #define ZEC_MAGIC 0x210da7ab10c7a11ULL
48 typedef struct zio_eck {
49 uint64_t zec_magic; /* for validation, endianness */
50 zio_cksum_t zec_cksum; /* 256-bit checksum */
54 * Gang block headers are self-checksumming and contain an array
57 #define SPA_GANGBLOCKSIZE SPA_MINBLOCKSIZE
58 #define SPA_GBH_NBLKPTRS ((SPA_GANGBLOCKSIZE - \
59 sizeof (zio_eck_t)) / sizeof (blkptr_t))
60 #define SPA_GBH_FILLER ((SPA_GANGBLOCKSIZE - \
61 sizeof (zio_eck_t) - \
62 (SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\
65 typedef struct zio_gbh {
66 blkptr_t zg_blkptr[SPA_GBH_NBLKPTRS];
67 uint64_t zg_filler[SPA_GBH_FILLER];
72 ZIO_CHECKSUM_INHERIT = 0,
76 ZIO_CHECKSUM_GANG_HEADER,
78 ZIO_CHECKSUM_FLETCHER_2,
79 ZIO_CHECKSUM_FLETCHER_4,
82 ZIO_CHECKSUM_FUNCTIONS
85 #define ZIO_CHECKSUM_ON_VALUE ZIO_CHECKSUM_FLETCHER_4
86 #define ZIO_CHECKSUM_DEFAULT ZIO_CHECKSUM_ON
88 #define ZIO_CHECKSUM_MASK 0xffULL
89 #define ZIO_CHECKSUM_VERIFY (1 << 8)
91 #define ZIO_DEDUPCHECKSUM ZIO_CHECKSUM_SHA256
92 #define ZIO_DEDUPDITTO_MIN 100
95 ZIO_COMPRESS_INHERIT = 0,
111 ZIO_COMPRESS_FUNCTIONS
114 #define ZIO_COMPRESS_ON_VALUE ZIO_COMPRESS_LZJB
115 #define ZIO_COMPRESS_DEFAULT ZIO_COMPRESS_OFF
117 #define BOOTFS_COMPRESS_VALID(compress) \
118 ((compress) == ZIO_COMPRESS_LZJB || \
119 (compress) == ZIO_COMPRESS_LZ4 || \
120 ((compress) == ZIO_COMPRESS_ON && \
121 ZIO_COMPRESS_ON_VALUE == ZIO_COMPRESS_LZJB) || \
122 (compress) == ZIO_COMPRESS_OFF)
125 * Default Linux timeout for a sd device.
127 #define ZIO_DELAY_MAX (30 * MILLISEC)
129 #define ZIO_FAILURE_MODE_WAIT 0
130 #define ZIO_FAILURE_MODE_CONTINUE 1
131 #define ZIO_FAILURE_MODE_PANIC 2
133 #define ZIO_PRIORITY_NOW (zio_priority_table[0])
134 #define ZIO_PRIORITY_SYNC_READ (zio_priority_table[1])
135 #define ZIO_PRIORITY_SYNC_WRITE (zio_priority_table[2])
136 #define ZIO_PRIORITY_LOG_WRITE (zio_priority_table[3])
137 #define ZIO_PRIORITY_CACHE_FILL (zio_priority_table[4])
138 #define ZIO_PRIORITY_AGG (zio_priority_table[5])
139 #define ZIO_PRIORITY_FREE (zio_priority_table[6])
140 #define ZIO_PRIORITY_ASYNC_WRITE (zio_priority_table[7])
141 #define ZIO_PRIORITY_ASYNC_READ (zio_priority_table[8])
142 #define ZIO_PRIORITY_RESILVER (zio_priority_table[9])
143 #define ZIO_PRIORITY_SCRUB (zio_priority_table[10])
144 #define ZIO_PRIORITY_DDT_PREFETCH (zio_priority_table[11])
145 #define ZIO_PRIORITY_TABLE_SIZE 12
147 #define ZIO_PIPELINE_CONTINUE 0x100
148 #define ZIO_PIPELINE_STOP 0x101
152 * Flags inherited by gang, ddt, and vdev children,
153 * and that must be equal for two zios to aggregate
155 ZIO_FLAG_DONT_AGGREGATE = 1 << 0,
156 ZIO_FLAG_IO_REPAIR = 1 << 1,
157 ZIO_FLAG_SELF_HEAL = 1 << 2,
158 ZIO_FLAG_RESILVER = 1 << 3,
159 ZIO_FLAG_SCRUB = 1 << 4,
160 ZIO_FLAG_SCAN_THREAD = 1 << 5,
162 #define ZIO_FLAG_AGG_INHERIT (ZIO_FLAG_CANFAIL - 1)
165 * Flags inherited by ddt, gang, and vdev children.
167 ZIO_FLAG_CANFAIL = 1 << 6, /* must be first for INHERIT */
168 ZIO_FLAG_SPECULATIVE = 1 << 7,
169 ZIO_FLAG_CONFIG_WRITER = 1 << 8,
170 ZIO_FLAG_DONT_RETRY = 1 << 9,
171 ZIO_FLAG_DONT_CACHE = 1 << 10,
172 ZIO_FLAG_NODATA = 1 << 11,
173 ZIO_FLAG_INDUCE_DAMAGE = 1 << 12,
175 #define ZIO_FLAG_DDT_INHERIT (ZIO_FLAG_IO_RETRY - 1)
176 #define ZIO_FLAG_GANG_INHERIT (ZIO_FLAG_IO_RETRY - 1)
179 * Flags inherited by vdev children.
181 ZIO_FLAG_IO_RETRY = 1 << 13, /* must be first for INHERIT */
182 ZIO_FLAG_PROBE = 1 << 14,
183 ZIO_FLAG_TRYHARD = 1 << 15,
184 ZIO_FLAG_OPTIONAL = 1 << 16,
186 #define ZIO_FLAG_VDEV_INHERIT (ZIO_FLAG_DONT_QUEUE - 1)
189 * Flags not inherited by any children.
191 ZIO_FLAG_DONT_QUEUE = 1 << 17, /* must be first for INHERIT */
192 ZIO_FLAG_DONT_PROPAGATE = 1 << 18,
193 ZIO_FLAG_IO_BYPASS = 1 << 19,
194 ZIO_FLAG_IO_REWRITE = 1 << 20,
195 ZIO_FLAG_RAW = 1 << 21,
196 ZIO_FLAG_GANG_CHILD = 1 << 22,
197 ZIO_FLAG_DDT_CHILD = 1 << 23,
198 ZIO_FLAG_GODFATHER = 1 << 24,
199 ZIO_FLAG_FASTWRITE = 1 << 25
202 #define ZIO_FLAG_MUSTSUCCEED 0
204 #define ZIO_DDT_CHILD_FLAGS(zio) \
205 (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) | \
206 ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
208 #define ZIO_GANG_CHILD_FLAGS(zio) \
209 (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) | \
210 ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
212 #define ZIO_VDEV_CHILD_FLAGS(zio) \
213 (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) | \
231 * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent
232 * graveyard) to indicate checksum errors and fragmentation.
237 typedef void zio_done_func_t(zio_t *zio);
239 extern uint8_t zio_priority_table[ZIO_PRIORITY_TABLE_SIZE];
240 extern char *zio_type_name[ZIO_TYPES];
243 * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
244 * identifies any block in the pool. By convention, the meta-objset (MOS)
245 * is objset 0, and the meta-dnode is object 0. This covers all blocks
246 * except root blocks and ZIL blocks, which are defined as follows:
248 * Root blocks (objset_phys_t) are object 0, level -1: <objset, 0, -1, 0>.
249 * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
250 * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
252 * Note: this structure is called a bookmark because its original purpose
253 * was to remember where to resume a pool-wide traverse.
255 * Note: this structure is passed between userland and the kernel.
256 * Therefore it must not change size or alignment between 32/64 bit
257 * compilation options.
259 typedef struct zbookmark {
266 #define SET_BOOKMARK(zb, objset, object, level, blkid) \
268 (zb)->zb_objset = objset; \
269 (zb)->zb_object = object; \
270 (zb)->zb_level = level; \
271 (zb)->zb_blkid = blkid; \
274 #define ZB_DESTROYED_OBJSET (-1ULL)
276 #define ZB_ROOT_OBJECT (0ULL)
277 #define ZB_ROOT_LEVEL (-1LL)
278 #define ZB_ROOT_BLKID (0ULL)
280 #define ZB_ZIL_OBJECT (0ULL)
281 #define ZB_ZIL_LEVEL (-2LL)
283 #define ZB_IS_ZERO(zb) \
284 ((zb)->zb_objset == 0 && (zb)->zb_object == 0 && \
285 (zb)->zb_level == 0 && (zb)->zb_blkid == 0)
286 #define ZB_IS_ROOT(zb) \
287 ((zb)->zb_object == ZB_ROOT_OBJECT && \
288 (zb)->zb_level == ZB_ROOT_LEVEL && \
289 (zb)->zb_blkid == ZB_ROOT_BLKID)
291 typedef struct zio_prop {
292 enum zio_checksum zp_checksum;
293 enum zio_compress zp_compress;
294 dmu_object_type_t zp_type;
298 uint8_t zp_dedup_verify;
301 typedef struct zio_cksum_report zio_cksum_report_t;
303 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
304 const void *good_data);
305 typedef void zio_cksum_free_f(void *cbdata, size_t size);
307 struct zio_bad_cksum; /* defined in zio_checksum.h */
310 struct zio_cksum_report {
311 struct zio_cksum_report *zcr_next;
312 nvlist_t *zcr_ereport;
313 nvlist_t *zcr_detector;
315 size_t zcr_cbinfo; /* passed to zcr_free() */
318 zio_cksum_finish_f *zcr_finish;
319 zio_cksum_free_f *zcr_free;
321 /* internal use only */
322 struct zio_bad_cksum *zcr_ckinfo; /* information from failure */
325 typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
328 zio_vsd_cksum_report_f zio_vsd_default_cksum_report;
330 typedef struct zio_vsd_ops {
331 zio_done_func_t *vsd_free;
332 zio_vsd_cksum_report_f *vsd_cksum_report;
335 typedef struct zio_gang_node {
336 zio_gbh_phys_t *gn_gbh;
337 struct zio_gang_node *gn_child[SPA_GBH_NBLKPTRS];
340 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
341 zio_gang_node_t *gn, void *data);
343 typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size);
345 typedef struct zio_transform {
347 uint64_t zt_orig_size;
349 zio_transform_func_t *zt_transform;
350 struct zio_transform *zt_next;
353 typedef int zio_pipe_stage_t(zio_t *zio);
356 * The io_reexecute flags are distinct from io_flags because the child must
357 * be able to propagate them to the parent. The normal io_flags are local
358 * to the zio, not protected by any lock, and not modifiable by children;
359 * the reexecute flags are protected by io_lock, modifiable by children,
360 * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
362 #define ZIO_REEXECUTE_NOW 0x01
363 #define ZIO_REEXECUTE_SUSPEND 0x02
365 typedef struct zio_link {
368 list_node_t zl_parent_node;
369 list_node_t zl_child_node;
373 /* Core information about this I/O */
374 zbookmark_t io_bookmark;
377 enum zio_child io_child_type;
380 uint8_t io_reexecute;
381 uint8_t io_state[ZIO_WAIT_TYPES];
385 blkptr_t *io_bp_override;
387 list_t io_parent_list;
388 list_t io_child_list;
389 zio_link_t *io_walk_link;
391 zio_transform_t *io_transform_stack;
394 zio_done_func_t *io_ready;
395 zio_done_func_t *io_done;
397 int64_t io_prev_space_delta; /* DMU private */
400 /* Data represented by this I/O */
404 uint64_t io_orig_size;
406 /* Stuff for the vdev stack */
409 const zio_vsd_ops_t *io_vsd_ops;
412 uint64_t io_deadline; /* expires at timestamp + deadline */
413 uint64_t io_timestamp; /* submitted at (ticks) */
414 uint64_t io_delta; /* vdev queue service delta (ticks) */
415 uint64_t io_delay; /* vdev disk service delta (ticks) */
416 avl_node_t io_offset_node;
417 avl_node_t io_deadline_node;
418 avl_tree_t *io_vdev_tree;
420 /* Internal pipeline state */
421 enum zio_flag io_flags;
422 enum zio_stage io_stage;
423 enum zio_stage io_pipeline;
424 enum zio_flag io_orig_flags;
425 enum zio_stage io_orig_stage;
426 enum zio_stage io_orig_pipeline;
428 int io_child_error[ZIO_CHILD_TYPES];
429 uint64_t io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
430 uint64_t io_child_count;
431 uint64_t io_parent_count;
433 zio_t *io_gang_leader;
434 zio_gang_node_t *io_gang_tree;
441 zio_cksum_report_t *io_cksum_report;
444 /* Taskq dispatching state */
445 taskq_ent_t io_tqent;
448 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
449 zio_done_func_t *done, void *private, enum zio_flag flags);
451 extern zio_t *zio_root(spa_t *spa,
452 zio_done_func_t *done, void *private, enum zio_flag flags);
454 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
455 uint64_t size, zio_done_func_t *done, void *private,
456 int priority, enum zio_flag flags, const zbookmark_t *zb);
458 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
459 void *data, uint64_t size, const zio_prop_t *zp,
460 zio_done_func_t *ready, zio_done_func_t *done, void *private,
461 int priority, enum zio_flag flags, const zbookmark_t *zb);
463 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
464 void *data, uint64_t size, zio_done_func_t *done, void *private,
465 int priority, enum zio_flag flags, zbookmark_t *zb);
467 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies);
469 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
471 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
473 zio_done_func_t *done, void *private, enum zio_flag flags);
475 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
476 zio_done_func_t *done, void *private, int priority, enum zio_flag flags);
478 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
479 uint64_t size, void *data, int checksum,
480 zio_done_func_t *done, void *private, int priority, enum zio_flag flags,
483 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
484 uint64_t size, void *data, int checksum,
485 zio_done_func_t *done, void *private, int priority, enum zio_flag flags,
488 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
489 const blkptr_t *bp, enum zio_flag flags);
491 extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
492 uint64_t size, boolean_t use_slog);
493 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
494 extern void zio_flush(zio_t *zio, vdev_t *vd);
495 extern void zio_shrink(zio_t *zio, uint64_t size);
497 extern int zio_wait(zio_t *zio);
498 extern void zio_nowait(zio_t *zio);
499 extern void zio_execute(zio_t *zio);
500 extern void zio_interrupt(zio_t *zio);
502 extern zio_t *zio_walk_parents(zio_t *cio);
503 extern zio_t *zio_walk_children(zio_t *pio);
504 extern zio_t *zio_unique_parent(zio_t *cio);
505 extern void zio_add_child(zio_t *pio, zio_t *cio);
507 extern void *zio_buf_alloc(size_t size);
508 extern void zio_buf_free(void *buf, size_t size);
509 extern void *zio_data_buf_alloc(size_t size);
510 extern void zio_data_buf_free(void *buf, size_t size);
511 extern void *zio_vdev_alloc(void);
512 extern void zio_vdev_free(void *buf);
514 extern void zio_resubmit_stage_async(void *);
516 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
517 uint64_t offset, void *data, uint64_t size, int type, int priority,
518 enum zio_flag flags, zio_done_func_t *done, void *private);
520 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
521 void *data, uint64_t size, int type, int priority,
522 enum zio_flag flags, zio_done_func_t *done, void *private);
524 extern void zio_vdev_io_bypass(zio_t *zio);
525 extern void zio_vdev_io_reissue(zio_t *zio);
526 extern void zio_vdev_io_redone(zio_t *zio);
528 extern void zio_checksum_verified(zio_t *zio);
529 extern int zio_worst_error(int e1, int e2);
531 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
532 enum zio_checksum parent);
533 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
534 enum zio_checksum child, enum zio_checksum parent);
535 extern enum zio_compress zio_compress_select(enum zio_compress child,
536 enum zio_compress parent);
538 extern void zio_suspend(spa_t *spa, zio_t *zio);
539 extern int zio_resume(spa_t *spa);
540 extern void zio_resume_wait(spa_t *spa);
543 * Initial setup and teardown.
545 extern void zio_init(void);
546 extern void zio_fini(void);
551 struct zinject_record;
552 extern uint32_t zio_injection_enabled;
553 extern int zio_inject_fault(char *name, int flags, int *id,
554 struct zinject_record *record);
555 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
556 struct zinject_record *record);
557 extern int zio_clear_fault(int id);
558 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
559 extern int zio_handle_fault_injection(zio_t *zio, int error);
560 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
561 extern int zio_handle_label_injection(zio_t *zio, int error);
562 extern void zio_handle_ignored_writes(zio_t *zio);
563 extern uint64_t zio_handle_io_delay(zio_t *zio);
566 * Checksum ereport functions
568 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
569 uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
570 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
571 const void *good_data, const void *bad_data, boolean_t drop_if_identical);
573 extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
574 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
576 /* If we have the good data in hand, this function can be used */
577 extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
578 struct zio *zio, uint64_t offset, uint64_t length,
579 const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
581 /* Called from spa_sync(), but primarily an injection handler */
582 extern void spa_handle_ignored_writes(spa_t *spa);
584 /* zbookmark functions */
585 boolean_t zbookmark_is_before(const struct dnode_phys *dnp,
586 const zbookmark_t *zb1, const zbookmark_t *zb2);