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, 2014 by Delphix. All rights reserved.
26 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
32 #include <sys/zio_priority.h>
33 #include <sys/zfs_context.h>
37 #include <sys/kstat.h>
38 #include <sys/fs/zfs.h>
39 #include <sys/zio_impl.h>
48 #define ZEC_MAGIC 0x210da7ab10c7a11ULL
50 typedef struct zio_eck {
51 uint64_t zec_magic; /* for validation, endianness */
52 zio_cksum_t zec_cksum; /* 256-bit checksum */
56 * Gang block headers are self-checksumming and contain an array
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))) /\
67 typedef struct zio_gbh {
68 blkptr_t zg_blkptr[SPA_GBH_NBLKPTRS];
69 uint64_t zg_filler[SPA_GBH_FILLER];
74 ZIO_CHECKSUM_INHERIT = 0,
78 ZIO_CHECKSUM_GANG_HEADER,
80 ZIO_CHECKSUM_FLETCHER_2,
81 ZIO_CHECKSUM_FLETCHER_4,
84 ZIO_CHECKSUM_NOPARITY,
90 ZIO_CHECKSUM_FUNCTIONS
94 * The number of "legacy" compression functions which can be set on individual
97 #define ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2
99 #define ZIO_CHECKSUM_ON_VALUE ZIO_CHECKSUM_FLETCHER_4
100 #define ZIO_CHECKSUM_DEFAULT ZIO_CHECKSUM_ON
102 #define ZIO_CHECKSUM_MASK 0xffULL
103 #define ZIO_CHECKSUM_VERIFY (1 << 8)
105 #define ZIO_DEDUPCHECKSUM ZIO_CHECKSUM_SHA256
106 #define ZIO_DEDUPDITTO_MIN 100
109 ZIO_COMPRESS_INHERIT = 0,
125 ZIO_COMPRESS_FUNCTIONS
129 * The number of "legacy" compression functions which can be set on individual
132 #define ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
135 * The meaning of "compress = on" selected by the compression features enabled
138 #define ZIO_COMPRESS_LEGACY_ON_VALUE ZIO_COMPRESS_LZJB
139 #define ZIO_COMPRESS_LZ4_ON_VALUE ZIO_COMPRESS_LZ4
141 #define ZIO_COMPRESS_DEFAULT ZIO_COMPRESS_OFF
143 #define BOOTFS_COMPRESS_VALID(compress) \
144 ((compress) == ZIO_COMPRESS_LZJB || \
145 (compress) == ZIO_COMPRESS_LZ4 || \
146 (compress) == ZIO_COMPRESS_ON || \
147 (compress) == ZIO_COMPRESS_OFF)
149 #define ZIO_FAILURE_MODE_WAIT 0
150 #define ZIO_FAILURE_MODE_CONTINUE 1
151 #define ZIO_FAILURE_MODE_PANIC 2
155 * Flags inherited by gang, ddt, and vdev children,
156 * and that must be equal for two zios to aggregate
158 ZIO_FLAG_DONT_AGGREGATE = 1 << 0,
159 ZIO_FLAG_IO_REPAIR = 1 << 1,
160 ZIO_FLAG_SELF_HEAL = 1 << 2,
161 ZIO_FLAG_RESILVER = 1 << 3,
162 ZIO_FLAG_SCRUB = 1 << 4,
163 ZIO_FLAG_SCAN_THREAD = 1 << 5,
164 ZIO_FLAG_PHYSICAL = 1 << 6,
166 #define ZIO_FLAG_AGG_INHERIT (ZIO_FLAG_CANFAIL - 1)
169 * Flags inherited by ddt, gang, and vdev children.
171 ZIO_FLAG_CANFAIL = 1 << 7, /* must be first for INHERIT */
172 ZIO_FLAG_SPECULATIVE = 1 << 8,
173 ZIO_FLAG_CONFIG_WRITER = 1 << 9,
174 ZIO_FLAG_DONT_RETRY = 1 << 10,
175 ZIO_FLAG_DONT_CACHE = 1 << 11,
176 ZIO_FLAG_NODATA = 1 << 12,
177 ZIO_FLAG_INDUCE_DAMAGE = 1 << 13,
179 #define ZIO_FLAG_DDT_INHERIT (ZIO_FLAG_IO_RETRY - 1)
180 #define ZIO_FLAG_GANG_INHERIT (ZIO_FLAG_IO_RETRY - 1)
183 * Flags inherited by vdev children.
185 ZIO_FLAG_IO_RETRY = 1 << 14, /* must be first for INHERIT */
186 ZIO_FLAG_PROBE = 1 << 15,
187 ZIO_FLAG_TRYHARD = 1 << 16,
188 ZIO_FLAG_OPTIONAL = 1 << 17,
190 #define ZIO_FLAG_VDEV_INHERIT (ZIO_FLAG_DONT_QUEUE - 1)
193 * Flags not inherited by any children.
195 ZIO_FLAG_DONT_QUEUE = 1 << 18, /* must be first for INHERIT */
196 ZIO_FLAG_DONT_PROPAGATE = 1 << 19,
197 ZIO_FLAG_IO_BYPASS = 1 << 20,
198 ZIO_FLAG_IO_REWRITE = 1 << 21,
199 ZIO_FLAG_RAW = 1 << 22,
200 ZIO_FLAG_GANG_CHILD = 1 << 23,
201 ZIO_FLAG_DDT_CHILD = 1 << 24,
202 ZIO_FLAG_GODFATHER = 1 << 25,
203 ZIO_FLAG_NOPWRITE = 1 << 26,
204 ZIO_FLAG_REEXECUTED = 1 << 27,
205 ZIO_FLAG_DELEGATED = 1 << 28,
208 #define ZIO_FLAG_MUSTSUCCEED 0
210 #define ZIO_DDT_CHILD_FLAGS(zio) \
211 (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) | \
212 ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
214 #define ZIO_GANG_CHILD_FLAGS(zio) \
215 (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) | \
216 ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
218 #define ZIO_VDEV_CHILD_FLAGS(zio) \
219 (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) | \
237 * We'll take the number 122 and 123 to indicate checksum errors and
238 * fragmentation. Those doesn't collide with any errno values as they
239 * are greater than ELAST.
244 typedef void zio_done_func_t(zio_t *zio);
246 extern const char *zio_type_name[ZIO_TYPES];
249 * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
250 * identifies any block in the pool. By convention, the meta-objset (MOS)
251 * is objset 0, and the meta-dnode is object 0. This covers all blocks
252 * except root blocks and ZIL blocks, which are defined as follows:
254 * Root blocks (objset_phys_t) are object 0, level -1: <objset, 0, -1, 0>.
255 * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
256 * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
257 * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
259 * Note: this structure is called a bookmark because its original purpose
260 * was to remember where to resume a pool-wide traverse.
262 * Note: this structure is passed between userland and the kernel, and is
263 * stored on disk (by virtue of being incorporated into other on-disk
264 * structures, e.g. dsl_scan_phys_t).
266 typedef struct zbookmark_phys {
273 #define SET_BOOKMARK(zb, objset, object, level, blkid) \
275 (zb)->zb_objset = objset; \
276 (zb)->zb_object = object; \
277 (zb)->zb_level = level; \
278 (zb)->zb_blkid = blkid; \
281 #define ZB_DESTROYED_OBJSET (-1ULL)
283 #define ZB_ROOT_OBJECT (0ULL)
284 #define ZB_ROOT_LEVEL (-1LL)
285 #define ZB_ROOT_BLKID (0ULL)
287 #define ZB_ZIL_OBJECT (0ULL)
288 #define ZB_ZIL_LEVEL (-2LL)
290 #define ZB_DNODE_LEVEL (-3LL)
291 #define ZB_DNODE_BLKID (0ULL)
293 #define ZB_IS_ZERO(zb) \
294 ((zb)->zb_objset == 0 && (zb)->zb_object == 0 && \
295 (zb)->zb_level == 0 && (zb)->zb_blkid == 0)
296 #define ZB_IS_ROOT(zb) \
297 ((zb)->zb_object == ZB_ROOT_OBJECT && \
298 (zb)->zb_level == ZB_ROOT_LEVEL && \
299 (zb)->zb_blkid == ZB_ROOT_BLKID)
301 typedef struct zio_prop {
302 enum zio_checksum zp_checksum;
303 enum zio_compress zp_compress;
304 dmu_object_type_t zp_type;
308 boolean_t zp_dedup_verify;
309 boolean_t zp_nopwrite;
312 typedef struct zio_cksum_report zio_cksum_report_t;
314 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
315 const void *good_data);
316 typedef void zio_cksum_free_f(void *cbdata, size_t size);
318 struct zio_bad_cksum; /* defined in zio_checksum.h */
321 struct zio_cksum_report {
322 struct zio_cksum_report *zcr_next;
323 nvlist_t *zcr_ereport;
324 nvlist_t *zcr_detector;
326 size_t zcr_cbinfo; /* passed to zcr_free() */
329 zio_cksum_finish_f *zcr_finish;
330 zio_cksum_free_f *zcr_free;
332 /* internal use only */
333 struct zio_bad_cksum *zcr_ckinfo; /* information from failure */
336 typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr,
339 zio_vsd_cksum_report_f zio_vsd_default_cksum_report;
341 typedef struct zio_vsd_ops {
342 zio_done_func_t *vsd_free;
343 zio_vsd_cksum_report_f *vsd_cksum_report;
346 typedef struct zio_gang_node {
347 zio_gbh_phys_t *gn_gbh;
348 struct zio_gang_node *gn_child[SPA_GBH_NBLKPTRS];
351 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
352 zio_gang_node_t *gn, void *data);
354 typedef void zio_transform_func_t(zio_t *zio, void *data, uint64_t size);
356 typedef struct zio_transform {
358 uint64_t zt_orig_size;
360 zio_transform_func_t *zt_transform;
361 struct zio_transform *zt_next;
364 typedef int zio_pipe_stage_t(zio_t *zio);
367 * The io_reexecute flags are distinct from io_flags because the child must
368 * be able to propagate them to the parent. The normal io_flags are local
369 * to the zio, not protected by any lock, and not modifiable by children;
370 * the reexecute flags are protected by io_lock, modifiable by children,
371 * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
373 #define ZIO_REEXECUTE_NOW 0x01
374 #define ZIO_REEXECUTE_SUSPEND 0x02
376 typedef struct zio_link {
379 list_node_t zl_parent_node;
380 list_node_t zl_child_node;
384 * Used for TRIM kstat.
386 typedef struct zio_trim_stats {
388 * Number of bytes successfully TRIMmed.
393 * Number of successful TRIM requests.
395 kstat_named_t success;
398 * Number of TRIM requests that failed because TRIM is not
401 kstat_named_t unsupported;
404 * Number of TRIM requests that failed for other reasons.
406 kstat_named_t failed;
409 extern zio_trim_stats_t zio_trim_stats;
411 #define ZIO_TRIM_STAT_INCR(stat, val) \
412 atomic_add_64(&zio_trim_stats.stat.value.ui64, (val));
413 #define ZIO_TRIM_STAT_BUMP(stat) \
414 ZIO_TRIM_STAT_INCR(stat, 1);
417 /* Core information about this I/O */
418 zbookmark_phys_t io_bookmark;
421 enum zio_child io_child_type;
423 zio_priority_t io_priority;
424 uint8_t io_reexecute;
425 uint8_t io_state[ZIO_WAIT_TYPES];
429 blkptr_t *io_bp_override;
431 list_t io_parent_list;
432 list_t io_child_list;
433 zio_link_t *io_walk_link;
435 zio_transform_t *io_transform_stack;
438 zio_done_func_t *io_ready;
439 zio_done_func_t *io_physdone;
440 zio_done_func_t *io_done;
442 int64_t io_prev_space_delta; /* DMU private */
445 /* Data represented by this I/O */
449 uint64_t io_orig_size;
451 /* Stuff for the vdev stack */
454 const zio_vsd_ops_t *io_vsd_ops;
457 hrtime_t io_timestamp;
458 hrtime_t io_target_timestamp;
459 avl_node_t io_queue_node;
460 avl_node_t io_offset_node;
462 /* Internal pipeline state */
463 enum zio_flag io_flags;
464 enum zio_stage io_stage;
465 enum zio_stage io_pipeline;
466 enum zio_flag io_orig_flags;
467 enum zio_stage io_orig_stage;
468 enum zio_stage io_orig_pipeline;
470 int io_child_error[ZIO_CHILD_TYPES];
471 uint64_t io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
472 uint64_t io_child_count;
473 uint64_t io_phys_children;
474 uint64_t io_parent_count;
476 zio_t *io_gang_leader;
477 zio_gang_node_t *io_gang_tree;
484 zio_cksum_report_t *io_cksum_report;
487 /* Taskq dispatching state */
488 taskq_ent_t io_tqent;
490 avl_node_t io_trim_node;
491 list_node_t io_trim_link;
494 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
495 zio_done_func_t *done, void *priv, enum zio_flag flags);
497 extern zio_t *zio_root(spa_t *spa,
498 zio_done_func_t *done, void *priv, enum zio_flag flags);
500 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, void *data,
501 uint64_t size, zio_done_func_t *done, void *priv,
502 zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
504 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
505 void *data, uint64_t size, const zio_prop_t *zp,
506 zio_done_func_t *ready, zio_done_func_t *physdone, zio_done_func_t *done,
508 zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb);
510 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
511 void *data, uint64_t size, zio_done_func_t *done, void *priv,
512 zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb);
514 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
517 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
519 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
521 zio_done_func_t *done, void *priv, enum zio_flag flags);
523 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
524 uint64_t offset, uint64_t size, zio_done_func_t *done, void *priv,
525 zio_priority_t priority, enum zio_flag flags);
527 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
528 uint64_t size, void *data, int checksum,
529 zio_done_func_t *done, void *priv, zio_priority_t priority,
530 enum zio_flag flags, boolean_t labels);
532 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
533 uint64_t size, void *data, int checksum,
534 zio_done_func_t *done, void *priv, zio_priority_t priority,
535 enum zio_flag flags, boolean_t labels);
537 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
538 const blkptr_t *bp, uint64_t size, enum zio_flag flags);
540 extern int zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp,
541 blkptr_t *old_bp, uint64_t size, boolean_t use_slog);
542 extern void zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp);
543 extern void zio_flush(zio_t *zio, vdev_t *vd);
544 extern zio_t *zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset,
546 extern void zio_shrink(zio_t *zio, uint64_t size);
548 extern int zio_wait(zio_t *zio);
549 extern void zio_nowait(zio_t *zio);
550 extern void zio_execute(zio_t *zio);
551 extern void zio_interrupt(zio_t *zio);
552 extern void zio_delay_init(zio_t *zio);
553 extern void zio_delay_interrupt(zio_t *zio);
555 extern zio_t *zio_walk_parents(zio_t *cio);
556 extern zio_t *zio_walk_children(zio_t *pio);
557 extern zio_t *zio_unique_parent(zio_t *cio);
558 extern void zio_add_child(zio_t *pio, zio_t *cio);
560 extern void *zio_buf_alloc(size_t size);
561 extern void zio_buf_free(void *buf, size_t size);
562 extern void *zio_data_buf_alloc(size_t size);
563 extern void zio_data_buf_free(void *buf, size_t size);
565 extern void zio_resubmit_stage_async(void *);
567 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
568 uint64_t offset, void *data, uint64_t size, int type,
569 zio_priority_t priority, enum zio_flag flags,
570 zio_done_func_t *done, void *priv);
572 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
573 void *data, uint64_t size, int type, zio_priority_t priority,
574 enum zio_flag flags, zio_done_func_t *done, void *priv);
576 extern void zio_vdev_io_bypass(zio_t *zio);
577 extern void zio_vdev_io_reissue(zio_t *zio);
578 extern void zio_vdev_io_redone(zio_t *zio);
580 extern void zio_checksum_verified(zio_t *zio);
581 extern int zio_worst_error(int e1, int e2);
583 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
584 enum zio_checksum parent);
585 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
586 enum zio_checksum child, enum zio_checksum parent);
587 extern enum zio_compress zio_compress_select(spa_t *spa,
588 enum zio_compress child, enum zio_compress parent);
590 extern void zio_suspend(spa_t *spa, zio_t *zio);
591 extern int zio_resume(spa_t *spa);
592 extern void zio_resume_wait(spa_t *spa);
595 * Initial setup and teardown.
597 extern void zio_init(void);
598 extern void zio_fini(void);
603 struct zinject_record;
604 extern uint32_t zio_injection_enabled;
605 extern int zio_inject_fault(char *name, int flags, int *id,
606 struct zinject_record *record);
607 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
608 struct zinject_record *record);
609 extern int zio_clear_fault(int id);
610 extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type);
611 extern int zio_handle_fault_injection(zio_t *zio, int error);
612 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
613 extern int zio_handle_label_injection(zio_t *zio, int error);
614 extern void zio_handle_ignored_writes(zio_t *zio);
615 extern hrtime_t zio_handle_io_delay(zio_t *zio);
618 * Checksum ereport functions
620 extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, struct zio *zio,
621 uint64_t offset, uint64_t length, void *arg, struct zio_bad_cksum *info);
622 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
623 const void *good_data, const void *bad_data, boolean_t drop_if_identical);
625 extern void zfs_ereport_send_interim_checksum(zio_cksum_report_t *report);
626 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
628 /* If we have the good data in hand, this function can be used */
629 extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
630 struct zio *zio, uint64_t offset, uint64_t length,
631 const void *good_data, const void *bad_data, struct zio_bad_cksum *info);
633 /* Called from spa_sync(), but primarily an injection handler */
634 extern void spa_handle_ignored_writes(spa_t *spa);
636 /* zbookmark_phys functions */
637 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
638 const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
639 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
640 uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);