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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011 by Delphix. All rights reserved.
26 #include <sys/zfs_context.h>
27 #include <sys/fm/fs/zfs.h>
30 #include <sys/spa_impl.h>
31 #include <sys/vdev_impl.h>
32 #include <sys/zio_impl.h>
33 #include <sys/zio_compress.h>
34 #include <sys/zio_checksum.h>
35 #include <sys/dmu_objset.h>
40 * ==========================================================================
42 * ==========================================================================
44 uint8_t zio_priority_table[ZIO_PRIORITY_TABLE_SIZE] = {
45 0, /* ZIO_PRIORITY_NOW */
46 0, /* ZIO_PRIORITY_SYNC_READ */
47 0, /* ZIO_PRIORITY_SYNC_WRITE */
48 0, /* ZIO_PRIORITY_LOG_WRITE */
49 1, /* ZIO_PRIORITY_CACHE_FILL */
50 1, /* ZIO_PRIORITY_AGG */
51 4, /* ZIO_PRIORITY_FREE */
52 4, /* ZIO_PRIORITY_ASYNC_WRITE */
53 6, /* ZIO_PRIORITY_ASYNC_READ */
54 10, /* ZIO_PRIORITY_RESILVER */
55 20, /* ZIO_PRIORITY_SCRUB */
56 2, /* ZIO_PRIORITY_DDT_PREFETCH */
60 * ==========================================================================
61 * I/O type descriptions
62 * ==========================================================================
64 char *zio_type_name[ZIO_TYPES] = {
65 "z_null", "z_rd", "z_wr", "z_fr", "z_cl", "z_ioctl"
69 * ==========================================================================
71 * ==========================================================================
73 kmem_cache_t *zio_cache;
74 kmem_cache_t *zio_link_cache;
75 kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT];
76 kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT];
77 int zio_bulk_flags = 0;
78 int zio_delay_max = ZIO_DELAY_MAX;
81 extern vmem_t *zio_alloc_arena;
83 extern int zfs_mg_alloc_failures;
86 * An allocating zio is one that either currently has the DVA allocate
87 * stage set or will have it later in its lifetime.
89 #define IO_IS_ALLOCATING(zio) ((zio)->io_orig_pipeline & ZIO_STAGE_DVA_ALLOCATE)
91 int zio_requeue_io_start_cut_in_line = 1;
94 int zio_buf_debug_limit = 16384;
96 int zio_buf_debug_limit = 0;
99 static inline void __zio_execute(zio_t *zio);
105 vmem_t *data_alloc_arena = NULL;
108 data_alloc_arena = zio_alloc_arena;
110 zio_cache = kmem_cache_create("zio_cache",
111 sizeof (zio_t), 0, NULL, NULL, NULL, NULL, NULL, KMC_KMEM);
112 zio_link_cache = kmem_cache_create("zio_link_cache",
113 sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, KMC_KMEM);
116 * For small buffers, we want a cache for each multiple of
117 * SPA_MINBLOCKSIZE. For medium-size buffers, we want a cache
118 * for each quarter-power of 2. For large buffers, we want
119 * a cache for each multiple of PAGESIZE.
121 for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) {
122 size_t size = (c + 1) << SPA_MINBLOCKSHIFT;
126 while (p2 & (p2 - 1))
129 if (size <= 4 * SPA_MINBLOCKSIZE) {
130 align = SPA_MINBLOCKSIZE;
131 } else if (P2PHASE(size, PAGESIZE) == 0) {
133 } else if (P2PHASE(size, p2 >> 2) == 0) {
139 int flags = zio_bulk_flags;
142 * The smallest buffers (512b) are heavily used and
143 * experience a lot of churn. The slabs allocated
144 * for them are also relatively small (32K). Thus
145 * in over to avoid expensive calls to vmalloc() we
146 * make an exception to the usual slab allocation
147 * policy and force these buffers to be kmem backed.
149 if (size == (1 << SPA_MINBLOCKSHIFT))
152 (void) sprintf(name, "zio_buf_%lu", (ulong_t)size);
153 zio_buf_cache[c] = kmem_cache_create(name, size,
154 align, NULL, NULL, NULL, NULL, NULL, flags);
156 (void) sprintf(name, "zio_data_buf_%lu", (ulong_t)size);
157 zio_data_buf_cache[c] = kmem_cache_create(name, size,
158 align, NULL, NULL, NULL, NULL,
159 data_alloc_arena, flags);
164 ASSERT(zio_buf_cache[c] != NULL);
165 if (zio_buf_cache[c - 1] == NULL)
166 zio_buf_cache[c - 1] = zio_buf_cache[c];
168 ASSERT(zio_data_buf_cache[c] != NULL);
169 if (zio_data_buf_cache[c - 1] == NULL)
170 zio_data_buf_cache[c - 1] = zio_data_buf_cache[c];
174 * The zio write taskqs have 1 thread per cpu, allow 1/2 of the taskqs
175 * to fail 3 times per txg or 8 failures, whichever is greater.
177 zfs_mg_alloc_failures = MAX((3 * max_ncpus / 2), 8);
186 kmem_cache_t *last_cache = NULL;
187 kmem_cache_t *last_data_cache = NULL;
189 for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) {
190 if (zio_buf_cache[c] != last_cache) {
191 last_cache = zio_buf_cache[c];
192 kmem_cache_destroy(zio_buf_cache[c]);
194 zio_buf_cache[c] = NULL;
196 if (zio_data_buf_cache[c] != last_data_cache) {
197 last_data_cache = zio_data_buf_cache[c];
198 kmem_cache_destroy(zio_data_buf_cache[c]);
200 zio_data_buf_cache[c] = NULL;
203 kmem_cache_destroy(zio_link_cache);
204 kmem_cache_destroy(zio_cache);
210 * ==========================================================================
211 * Allocate and free I/O buffers
212 * ==========================================================================
216 * Use zio_buf_alloc to allocate ZFS metadata. This data will appear in a
217 * crashdump if the kernel panics, so use it judiciously. Obviously, it's
218 * useful to inspect ZFS metadata, but if possible, we should avoid keeping
219 * excess / transient data in-core during a crashdump.
222 zio_buf_alloc(size_t size)
224 size_t c = (size - 1) >> SPA_MINBLOCKSHIFT;
226 ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
228 return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE));
232 * Use zio_data_buf_alloc to allocate data. The data will not appear in a
233 * crashdump if the kernel panics. This exists so that we will limit the amount
234 * of ZFS data that shows up in a kernel crashdump. (Thus reducing the amount
235 * of kernel heap dumped to disk when the kernel panics)
238 zio_data_buf_alloc(size_t size)
240 size_t c = (size - 1) >> SPA_MINBLOCKSHIFT;
242 ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
244 return (kmem_cache_alloc(zio_data_buf_cache[c], KM_PUSHPAGE));
248 zio_buf_free(void *buf, size_t size)
250 size_t c = (size - 1) >> SPA_MINBLOCKSHIFT;
252 ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
254 kmem_cache_free(zio_buf_cache[c], buf);
258 zio_data_buf_free(void *buf, size_t size)
260 size_t c = (size - 1) >> SPA_MINBLOCKSHIFT;
262 ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
264 kmem_cache_free(zio_data_buf_cache[c], buf);
268 * ==========================================================================
269 * Push and pop I/O transform buffers
270 * ==========================================================================
273 zio_push_transform(zio_t *zio, void *data, uint64_t size, uint64_t bufsize,
274 zio_transform_func_t *transform)
276 zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_PUSHPAGE);
278 zt->zt_orig_data = zio->io_data;
279 zt->zt_orig_size = zio->io_size;
280 zt->zt_bufsize = bufsize;
281 zt->zt_transform = transform;
283 zt->zt_next = zio->io_transform_stack;
284 zio->io_transform_stack = zt;
291 zio_pop_transforms(zio_t *zio)
295 while ((zt = zio->io_transform_stack) != NULL) {
296 if (zt->zt_transform != NULL)
297 zt->zt_transform(zio,
298 zt->zt_orig_data, zt->zt_orig_size);
300 if (zt->zt_bufsize != 0)
301 zio_buf_free(zio->io_data, zt->zt_bufsize);
303 zio->io_data = zt->zt_orig_data;
304 zio->io_size = zt->zt_orig_size;
305 zio->io_transform_stack = zt->zt_next;
307 kmem_free(zt, sizeof (zio_transform_t));
312 * ==========================================================================
313 * I/O transform callbacks for subblocks and decompression
314 * ==========================================================================
317 zio_subblock(zio_t *zio, void *data, uint64_t size)
319 ASSERT(zio->io_size > size);
321 if (zio->io_type == ZIO_TYPE_READ)
322 bcopy(zio->io_data, data, size);
326 zio_decompress(zio_t *zio, void *data, uint64_t size)
328 if (zio->io_error == 0 &&
329 zio_decompress_data(BP_GET_COMPRESS(zio->io_bp),
330 zio->io_data, data, zio->io_size, size) != 0)
335 * ==========================================================================
336 * I/O parent/child relationships and pipeline interlocks
337 * ==========================================================================
340 * NOTE - Callers to zio_walk_parents() and zio_walk_children must
341 * continue calling these functions until they return NULL.
342 * Otherwise, the next caller will pick up the list walk in
343 * some indeterminate state. (Otherwise every caller would
344 * have to pass in a cookie to keep the state represented by
345 * io_walk_link, which gets annoying.)
348 zio_walk_parents(zio_t *cio)
350 zio_link_t *zl = cio->io_walk_link;
351 list_t *pl = &cio->io_parent_list;
353 zl = (zl == NULL) ? list_head(pl) : list_next(pl, zl);
354 cio->io_walk_link = zl;
359 ASSERT(zl->zl_child == cio);
360 return (zl->zl_parent);
364 zio_walk_children(zio_t *pio)
366 zio_link_t *zl = pio->io_walk_link;
367 list_t *cl = &pio->io_child_list;
369 zl = (zl == NULL) ? list_head(cl) : list_next(cl, zl);
370 pio->io_walk_link = zl;
375 ASSERT(zl->zl_parent == pio);
376 return (zl->zl_child);
380 zio_unique_parent(zio_t *cio)
382 zio_t *pio = zio_walk_parents(cio);
384 VERIFY(zio_walk_parents(cio) == NULL);
389 zio_add_child(zio_t *pio, zio_t *cio)
391 zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_PUSHPAGE);
395 * Logical I/Os can have logical, gang, or vdev children.
396 * Gang I/Os can have gang or vdev children.
397 * Vdev I/Os can only have vdev children.
398 * The following ASSERT captures all of these constraints.
400 ASSERT(cio->io_child_type <= pio->io_child_type);
405 mutex_enter(&cio->io_lock);
406 mutex_enter(&pio->io_lock);
408 ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0);
410 for (w = 0; w < ZIO_WAIT_TYPES; w++)
411 pio->io_children[cio->io_child_type][w] += !cio->io_state[w];
413 list_insert_head(&pio->io_child_list, zl);
414 list_insert_head(&cio->io_parent_list, zl);
416 pio->io_child_count++;
417 cio->io_parent_count++;
419 mutex_exit(&pio->io_lock);
420 mutex_exit(&cio->io_lock);
424 zio_remove_child(zio_t *pio, zio_t *cio, zio_link_t *zl)
426 ASSERT(zl->zl_parent == pio);
427 ASSERT(zl->zl_child == cio);
429 mutex_enter(&cio->io_lock);
430 mutex_enter(&pio->io_lock);
432 list_remove(&pio->io_child_list, zl);
433 list_remove(&cio->io_parent_list, zl);
435 pio->io_child_count--;
436 cio->io_parent_count--;
438 mutex_exit(&pio->io_lock);
439 mutex_exit(&cio->io_lock);
441 kmem_cache_free(zio_link_cache, zl);
445 zio_wait_for_children(zio_t *zio, enum zio_child child, enum zio_wait_type wait)
447 uint64_t *countp = &zio->io_children[child][wait];
448 boolean_t waiting = B_FALSE;
450 mutex_enter(&zio->io_lock);
451 ASSERT(zio->io_stall == NULL);
454 zio->io_stall = countp;
457 mutex_exit(&zio->io_lock);
462 __attribute__((always_inline))
464 zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait)
466 uint64_t *countp = &pio->io_children[zio->io_child_type][wait];
467 int *errorp = &pio->io_child_error[zio->io_child_type];
469 mutex_enter(&pio->io_lock);
470 if (zio->io_error && !(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
471 *errorp = zio_worst_error(*errorp, zio->io_error);
472 pio->io_reexecute |= zio->io_reexecute;
473 ASSERT3U(*countp, >, 0);
474 if (--*countp == 0 && pio->io_stall == countp) {
475 pio->io_stall = NULL;
476 mutex_exit(&pio->io_lock);
479 mutex_exit(&pio->io_lock);
484 zio_inherit_child_errors(zio_t *zio, enum zio_child c)
486 if (zio->io_child_error[c] != 0 && zio->io_error == 0)
487 zio->io_error = zio->io_child_error[c];
491 * ==========================================================================
492 * Create the various types of I/O (read, write, free, etc)
493 * ==========================================================================
496 zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
497 void *data, uint64_t size, zio_done_func_t *done, void *private,
498 zio_type_t type, int priority, enum zio_flag flags,
499 vdev_t *vd, uint64_t offset, const zbookmark_t *zb,
500 enum zio_stage stage, enum zio_stage pipeline)
504 ASSERT3U(size, <=, SPA_MAXBLOCKSIZE);
505 ASSERT(P2PHASE(size, SPA_MINBLOCKSIZE) == 0);
506 ASSERT(P2PHASE(offset, SPA_MINBLOCKSIZE) == 0);
508 ASSERT(!vd || spa_config_held(spa, SCL_STATE_ALL, RW_READER));
509 ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER));
510 ASSERT(vd || stage == ZIO_STAGE_OPEN);
512 zio = kmem_cache_alloc(zio_cache, KM_PUSHPAGE);
513 bzero(zio, sizeof (zio_t));
515 mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL);
516 cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL);
518 list_create(&zio->io_parent_list, sizeof (zio_link_t),
519 offsetof(zio_link_t, zl_parent_node));
520 list_create(&zio->io_child_list, sizeof (zio_link_t),
521 offsetof(zio_link_t, zl_child_node));
524 zio->io_child_type = ZIO_CHILD_VDEV;
525 else if (flags & ZIO_FLAG_GANG_CHILD)
526 zio->io_child_type = ZIO_CHILD_GANG;
527 else if (flags & ZIO_FLAG_DDT_CHILD)
528 zio->io_child_type = ZIO_CHILD_DDT;
530 zio->io_child_type = ZIO_CHILD_LOGICAL;
533 zio->io_bp = (blkptr_t *)bp;
534 zio->io_bp_copy = *bp;
535 zio->io_bp_orig = *bp;
536 if (type != ZIO_TYPE_WRITE ||
537 zio->io_child_type == ZIO_CHILD_DDT)
538 zio->io_bp = &zio->io_bp_copy; /* so caller can free */
539 if (zio->io_child_type == ZIO_CHILD_LOGICAL)
540 zio->io_logical = zio;
541 if (zio->io_child_type > ZIO_CHILD_GANG && BP_IS_GANG(bp))
542 pipeline |= ZIO_GANG_STAGES;
548 zio->io_private = private;
550 zio->io_priority = priority;
552 zio->io_offset = offset;
553 zio->io_orig_data = zio->io_data = data;
554 zio->io_orig_size = zio->io_size = size;
555 zio->io_orig_flags = zio->io_flags = flags;
556 zio->io_orig_stage = zio->io_stage = stage;
557 zio->io_orig_pipeline = zio->io_pipeline = pipeline;
559 zio->io_state[ZIO_WAIT_READY] = (stage >= ZIO_STAGE_READY);
560 zio->io_state[ZIO_WAIT_DONE] = (stage >= ZIO_STAGE_DONE);
563 zio->io_bookmark = *zb;
566 if (zio->io_logical == NULL)
567 zio->io_logical = pio->io_logical;
568 if (zio->io_child_type == ZIO_CHILD_GANG)
569 zio->io_gang_leader = pio->io_gang_leader;
570 zio_add_child(pio, zio);
577 zio_destroy(zio_t *zio)
579 list_destroy(&zio->io_parent_list);
580 list_destroy(&zio->io_child_list);
581 mutex_destroy(&zio->io_lock);
582 cv_destroy(&zio->io_cv);
583 kmem_cache_free(zio_cache, zio);
587 zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done,
588 void *private, enum zio_flag flags)
592 zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private,
593 ZIO_TYPE_NULL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL,
594 ZIO_STAGE_OPEN, ZIO_INTERLOCK_PIPELINE);
600 zio_root(spa_t *spa, zio_done_func_t *done, void *private, enum zio_flag flags)
602 return (zio_null(NULL, spa, NULL, done, private, flags));
606 zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
607 void *data, uint64_t size, zio_done_func_t *done, void *private,
608 int priority, enum zio_flag flags, const zbookmark_t *zb)
612 zio = zio_create(pio, spa, BP_PHYSICAL_BIRTH(bp), bp,
613 data, size, done, private,
614 ZIO_TYPE_READ, priority, flags, NULL, 0, zb,
615 ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ?
616 ZIO_DDT_CHILD_READ_PIPELINE : ZIO_READ_PIPELINE);
622 zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
623 void *data, uint64_t size, const zio_prop_t *zp,
624 zio_done_func_t *ready, zio_done_func_t *done, void *private,
625 int priority, enum zio_flag flags, const zbookmark_t *zb)
629 ASSERT(zp->zp_checksum >= ZIO_CHECKSUM_OFF &&
630 zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS &&
631 zp->zp_compress >= ZIO_COMPRESS_OFF &&
632 zp->zp_compress < ZIO_COMPRESS_FUNCTIONS &&
633 zp->zp_type < DMU_OT_NUMTYPES &&
636 zp->zp_copies <= spa_max_replication(spa) &&
638 zp->zp_dedup_verify <= 1);
640 zio = zio_create(pio, spa, txg, bp, data, size, done, private,
641 ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb,
642 ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ?
643 ZIO_DDT_CHILD_WRITE_PIPELINE : ZIO_WRITE_PIPELINE);
645 zio->io_ready = ready;
652 zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data,
653 uint64_t size, zio_done_func_t *done, void *private, int priority,
654 enum zio_flag flags, zbookmark_t *zb)
658 zio = zio_create(pio, spa, txg, bp, data, size, done, private,
659 ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb,
660 ZIO_STAGE_OPEN, ZIO_REWRITE_PIPELINE);
666 zio_write_override(zio_t *zio, blkptr_t *bp, int copies)
668 ASSERT(zio->io_type == ZIO_TYPE_WRITE);
669 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
670 ASSERT(zio->io_stage == ZIO_STAGE_OPEN);
671 ASSERT(zio->io_txg == spa_syncing_txg(zio->io_spa));
673 zio->io_prop.zp_copies = copies;
674 zio->io_bp_override = bp;
678 zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp)
680 bplist_append(&spa->spa_free_bplist[txg & TXG_MASK], bp);
684 zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
689 dprintf_bp(bp, "freeing in txg %llu, pass %u",
690 (longlong_t)txg, spa->spa_sync_pass);
692 ASSERT(!BP_IS_HOLE(bp));
693 ASSERT(spa_syncing_txg(spa) == txg);
694 ASSERT(spa_sync_pass(spa) <= SYNC_PASS_DEFERRED_FREE);
696 zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp),
697 NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_FREE, flags,
698 NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_FREE_PIPELINE);
704 zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
705 zio_done_func_t *done, void *private, enum zio_flag flags)
710 * A claim is an allocation of a specific block. Claims are needed
711 * to support immediate writes in the intent log. The issue is that
712 * immediate writes contain committed data, but in a txg that was
713 * *not* committed. Upon opening the pool after an unclean shutdown,
714 * the intent log claims all blocks that contain immediate write data
715 * so that the SPA knows they're in use.
717 * All claims *must* be resolved in the first txg -- before the SPA
718 * starts allocating blocks -- so that nothing is allocated twice.
719 * If txg == 0 we just verify that the block is claimable.
721 ASSERT3U(spa->spa_uberblock.ub_rootbp.blk_birth, <, spa_first_txg(spa));
722 ASSERT(txg == spa_first_txg(spa) || txg == 0);
723 ASSERT(!BP_GET_DEDUP(bp) || !spa_writeable(spa)); /* zdb(1M) */
725 zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp),
726 done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags,
727 NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE);
733 zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd,
734 zio_done_func_t *done, void *private, int priority, enum zio_flag flags)
739 if (vd->vdev_children == 0) {
740 zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private,
741 ZIO_TYPE_IOCTL, priority, flags, vd, 0, NULL,
742 ZIO_STAGE_OPEN, ZIO_IOCTL_PIPELINE);
746 zio = zio_null(pio, spa, NULL, NULL, NULL, flags);
748 for (c = 0; c < vd->vdev_children; c++)
749 zio_nowait(zio_ioctl(zio, spa, vd->vdev_child[c], cmd,
750 done, private, priority, flags));
757 zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
758 void *data, int checksum, zio_done_func_t *done, void *private,
759 int priority, enum zio_flag flags, boolean_t labels)
763 ASSERT(vd->vdev_children == 0);
764 ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE ||
765 offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE);
766 ASSERT3U(offset + size, <=, vd->vdev_psize);
768 zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private,
769 ZIO_TYPE_READ, priority, flags, vd, offset, NULL,
770 ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE);
772 zio->io_prop.zp_checksum = checksum;
778 zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
779 void *data, int checksum, zio_done_func_t *done, void *private,
780 int priority, enum zio_flag flags, boolean_t labels)
784 ASSERT(vd->vdev_children == 0);
785 ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE ||
786 offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE);
787 ASSERT3U(offset + size, <=, vd->vdev_psize);
789 zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private,
790 ZIO_TYPE_WRITE, priority, flags, vd, offset, NULL,
791 ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE);
793 zio->io_prop.zp_checksum = checksum;
795 if (zio_checksum_table[checksum].ci_eck) {
797 * zec checksums are necessarily destructive -- they modify
798 * the end of the write buffer to hold the verifier/checksum.
799 * Therefore, we must make a local copy in case the data is
800 * being written to multiple places in parallel.
802 void *wbuf = zio_buf_alloc(size);
803 bcopy(data, wbuf, size);
804 zio_push_transform(zio, wbuf, size, size, NULL);
811 * Create a child I/O to do some work for us.
814 zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset,
815 void *data, uint64_t size, int type, int priority, enum zio_flag flags,
816 zio_done_func_t *done, void *private)
818 enum zio_stage pipeline = ZIO_VDEV_CHILD_PIPELINE;
821 ASSERT(vd->vdev_parent ==
822 (pio->io_vd ? pio->io_vd : pio->io_spa->spa_root_vdev));
824 if (type == ZIO_TYPE_READ && bp != NULL) {
826 * If we have the bp, then the child should perform the
827 * checksum and the parent need not. This pushes error
828 * detection as close to the leaves as possible and
829 * eliminates redundant checksums in the interior nodes.
831 pipeline |= ZIO_STAGE_CHECKSUM_VERIFY;
832 pio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY;
835 if (vd->vdev_children == 0)
836 offset += VDEV_LABEL_START_SIZE;
838 flags |= ZIO_VDEV_CHILD_FLAGS(pio) | ZIO_FLAG_DONT_PROPAGATE;
841 * If we've decided to do a repair, the write is not speculative --
842 * even if the original read was.
844 if (flags & ZIO_FLAG_IO_REPAIR)
845 flags &= ~ZIO_FLAG_SPECULATIVE;
847 zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size,
848 done, private, type, priority, flags, vd, offset, &pio->io_bookmark,
849 ZIO_STAGE_VDEV_IO_START >> 1, pipeline);
855 zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, void *data, uint64_t size,
856 int type, int priority, enum zio_flag flags,
857 zio_done_func_t *done, void *private)
861 ASSERT(vd->vdev_ops->vdev_op_leaf);
863 zio = zio_create(NULL, vd->vdev_spa, 0, NULL,
864 data, size, done, private, type, priority,
865 flags | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY,
867 ZIO_STAGE_VDEV_IO_START >> 1, ZIO_VDEV_CHILD_PIPELINE);
873 zio_flush(zio_t *zio, vdev_t *vd)
875 zio_nowait(zio_ioctl(zio, zio->io_spa, vd, DKIOCFLUSHWRITECACHE,
876 NULL, NULL, ZIO_PRIORITY_NOW,
877 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY));
881 zio_shrink(zio_t *zio, uint64_t size)
883 ASSERT(zio->io_executor == NULL);
884 ASSERT(zio->io_orig_size == zio->io_size);
885 ASSERT(size <= zio->io_size);
888 * We don't shrink for raidz because of problems with the
889 * reconstruction when reading back less than the block size.
890 * Note, BP_IS_RAIDZ() assumes no compression.
892 ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF);
893 if (!BP_IS_RAIDZ(zio->io_bp))
894 zio->io_orig_size = zio->io_size = size;
898 * ==========================================================================
899 * Prepare to read and write logical blocks
900 * ==========================================================================
904 zio_read_bp_init(zio_t *zio)
906 blkptr_t *bp = zio->io_bp;
908 if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF &&
909 zio->io_child_type == ZIO_CHILD_LOGICAL &&
910 !(zio->io_flags & ZIO_FLAG_RAW)) {
911 uint64_t psize = BP_GET_PSIZE(bp);
912 void *cbuf = zio_buf_alloc(psize);
914 zio_push_transform(zio, cbuf, psize, psize, zio_decompress);
917 if (!dmu_ot[BP_GET_TYPE(bp)].ot_metadata && BP_GET_LEVEL(bp) == 0)
918 zio->io_flags |= ZIO_FLAG_DONT_CACHE;
920 if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP)
921 zio->io_flags |= ZIO_FLAG_DONT_CACHE;
923 if (BP_GET_DEDUP(bp) && zio->io_child_type == ZIO_CHILD_LOGICAL)
924 zio->io_pipeline = ZIO_DDT_READ_PIPELINE;
926 return (ZIO_PIPELINE_CONTINUE);
930 zio_write_bp_init(zio_t *zio)
932 spa_t *spa = zio->io_spa;
933 zio_prop_t *zp = &zio->io_prop;
934 enum zio_compress compress = zp->zp_compress;
935 blkptr_t *bp = zio->io_bp;
936 uint64_t lsize = zio->io_size;
937 uint64_t psize = lsize;
941 * If our children haven't all reached the ready stage,
942 * wait for them and then repeat this pipeline stage.
944 if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) ||
945 zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_READY))
946 return (ZIO_PIPELINE_STOP);
948 if (!IO_IS_ALLOCATING(zio))
949 return (ZIO_PIPELINE_CONTINUE);
951 ASSERT(zio->io_child_type != ZIO_CHILD_DDT);
953 if (zio->io_bp_override) {
954 ASSERT(bp->blk_birth != zio->io_txg);
955 ASSERT(BP_GET_DEDUP(zio->io_bp_override) == 0);
957 *bp = *zio->io_bp_override;
958 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
960 if (BP_IS_HOLE(bp) || !zp->zp_dedup)
961 return (ZIO_PIPELINE_CONTINUE);
963 ASSERT(zio_checksum_table[zp->zp_checksum].ci_dedup ||
964 zp->zp_dedup_verify);
966 if (BP_GET_CHECKSUM(bp) == zp->zp_checksum) {
968 zio->io_pipeline |= ZIO_STAGE_DDT_WRITE;
969 return (ZIO_PIPELINE_CONTINUE);
971 zio->io_bp_override = NULL;
975 if (bp->blk_birth == zio->io_txg) {
977 * We're rewriting an existing block, which means we're
978 * working on behalf of spa_sync(). For spa_sync() to
979 * converge, it must eventually be the case that we don't
980 * have to allocate new blocks. But compression changes
981 * the blocksize, which forces a reallocate, and makes
982 * convergence take longer. Therefore, after the first
983 * few passes, stop compressing to ensure convergence.
985 pass = spa_sync_pass(spa);
987 ASSERT(zio->io_txg == spa_syncing_txg(spa));
988 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
989 ASSERT(!BP_GET_DEDUP(bp));
991 if (pass > SYNC_PASS_DONT_COMPRESS)
992 compress = ZIO_COMPRESS_OFF;
994 /* Make sure someone doesn't change their mind on overwrites */
995 ASSERT(MIN(zp->zp_copies + BP_IS_GANG(bp),
996 spa_max_replication(spa)) == BP_GET_NDVAS(bp));
999 if (compress != ZIO_COMPRESS_OFF) {
1000 void *cbuf = zio_buf_alloc(lsize);
1001 psize = zio_compress_data(compress, zio->io_data, cbuf, lsize);
1002 if (psize == 0 || psize == lsize) {
1003 compress = ZIO_COMPRESS_OFF;
1004 zio_buf_free(cbuf, lsize);
1006 ASSERT(psize < lsize);
1007 zio_push_transform(zio, cbuf, psize, lsize, NULL);
1012 * The final pass of spa_sync() must be all rewrites, but the first
1013 * few passes offer a trade-off: allocating blocks defers convergence,
1014 * but newly allocated blocks are sequential, so they can be written
1015 * to disk faster. Therefore, we allow the first few passes of
1016 * spa_sync() to allocate new blocks, but force rewrites after that.
1017 * There should only be a handful of blocks after pass 1 in any case.
1019 if (bp->blk_birth == zio->io_txg && BP_GET_PSIZE(bp) == psize &&
1020 pass > SYNC_PASS_REWRITE) {
1021 enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES;
1023 zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages;
1024 zio->io_flags |= ZIO_FLAG_IO_REWRITE;
1027 zio->io_pipeline = ZIO_WRITE_PIPELINE;
1031 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
1033 ASSERT(zp->zp_checksum != ZIO_CHECKSUM_GANG_HEADER);
1034 BP_SET_LSIZE(bp, lsize);
1035 BP_SET_PSIZE(bp, psize);
1036 BP_SET_COMPRESS(bp, compress);
1037 BP_SET_CHECKSUM(bp, zp->zp_checksum);
1038 BP_SET_TYPE(bp, zp->zp_type);
1039 BP_SET_LEVEL(bp, zp->zp_level);
1040 BP_SET_DEDUP(bp, zp->zp_dedup);
1041 BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
1043 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
1044 ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE));
1045 zio->io_pipeline = ZIO_DDT_WRITE_PIPELINE;
1049 return (ZIO_PIPELINE_CONTINUE);
1053 zio_free_bp_init(zio_t *zio)
1055 blkptr_t *bp = zio->io_bp;
1057 if (zio->io_child_type == ZIO_CHILD_LOGICAL) {
1058 if (BP_GET_DEDUP(bp))
1059 zio->io_pipeline = ZIO_DDT_FREE_PIPELINE;
1062 return (ZIO_PIPELINE_CONTINUE);
1066 * ==========================================================================
1067 * Execute the I/O pipeline
1068 * ==========================================================================
1072 zio_taskq_dispatch(zio_t *zio, enum zio_taskq_type q, boolean_t cutinline)
1074 spa_t *spa = zio->io_spa;
1075 zio_type_t t = zio->io_type;
1076 int flags = TQ_NOSLEEP | (cutinline ? TQ_FRONT : 0);
1079 * If we're a config writer or a probe, the normal issue and
1080 * interrupt threads may all be blocked waiting for the config lock.
1081 * In this case, select the otherwise-unused taskq for ZIO_TYPE_NULL.
1083 if (zio->io_flags & (ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_PROBE))
1087 * A similar issue exists for the L2ARC write thread until L2ARC 2.0.
1089 if (t == ZIO_TYPE_WRITE && zio->io_vd && zio->io_vd->vdev_aux)
1093 * If this is a high priority I/O, then use the high priority taskq.
1095 if (zio->io_priority == ZIO_PRIORITY_NOW &&
1096 spa->spa_zio_taskq[t][q + 1] != NULL)
1099 ASSERT3U(q, <, ZIO_TASKQ_TYPES);
1101 while (taskq_dispatch(spa->spa_zio_taskq[t][q],
1102 (task_func_t *)zio_execute, zio, flags) == 0); /* do nothing */
1106 zio_taskq_member(zio_t *zio, enum zio_taskq_type q)
1108 kthread_t *executor = zio->io_executor;
1109 spa_t *spa = zio->io_spa;
1112 for (t = 0; t < ZIO_TYPES; t++)
1113 if (taskq_member(spa->spa_zio_taskq[t][q], executor))
1120 zio_issue_async(zio_t *zio)
1122 zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE);
1124 return (ZIO_PIPELINE_STOP);
1128 zio_interrupt(zio_t *zio)
1130 zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT, B_FALSE);
1134 * Execute the I/O pipeline until one of the following occurs:
1135 * (1) the I/O completes; (2) the pipeline stalls waiting for
1136 * dependent child I/Os; (3) the I/O issues, so we're waiting
1137 * for an I/O completion interrupt; (4) the I/O is delegated by
1138 * vdev-level caching or aggregation; (5) the I/O is deferred
1139 * due to vdev-level queueing; (6) the I/O is handed off to
1140 * another thread. In all cases, the pipeline stops whenever
1141 * there's no CPU work; it never burns a thread in cv_wait().
1143 * There's no locking on io_stage because there's no legitimate way
1144 * for multiple threads to be attempting to process the same I/O.
1146 static zio_pipe_stage_t *zio_pipeline[];
1149 * zio_execute() is a wrapper around the static function
1150 * __zio_execute() so that we can force __zio_execute() to be
1151 * inlined. This reduces stack overhead which is important
1152 * because __zio_execute() is called recursively in several zio
1153 * code paths. zio_execute() itself cannot be inlined because
1154 * it is externally visible.
1157 zio_execute(zio_t *zio)
1162 __attribute__((always_inline))
1164 __zio_execute(zio_t *zio)
1166 zio->io_executor = curthread;
1168 while (zio->io_stage < ZIO_STAGE_DONE) {
1169 enum zio_stage pipeline = zio->io_pipeline;
1170 enum zio_stage stage = zio->io_stage;
1175 ASSERT(!MUTEX_HELD(&zio->io_lock));
1176 ASSERT(ISP2(stage));
1177 ASSERT(zio->io_stall == NULL);
1181 } while ((stage & pipeline) == 0);
1183 ASSERT(stage <= ZIO_STAGE_DONE);
1185 dsl = spa_get_dsl(zio->io_spa);
1186 cut = (stage == ZIO_STAGE_VDEV_IO_START) ?
1187 zio_requeue_io_start_cut_in_line : B_FALSE;
1190 * If we are in interrupt context and this pipeline stage
1191 * will grab a config lock that is held across I/O,
1192 * or may wait for an I/O that needs an interrupt thread
1193 * to complete, issue async to avoid deadlock.
1195 * If we are in the txg_sync_thread or being called
1196 * during pool init issue async to minimize stack depth.
1197 * Both of these call paths may be recursively called.
1199 * For VDEV_IO_START, we cut in line so that the io will
1200 * be sent to disk promptly.
1202 if (((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL &&
1203 zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) ||
1204 (dsl != NULL && dsl_pool_sync_context(dsl))) {
1205 zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut);
1209 zio->io_stage = stage;
1210 rv = zio_pipeline[highbit(stage) - 1](zio);
1212 if (rv == ZIO_PIPELINE_STOP)
1215 ASSERT(rv == ZIO_PIPELINE_CONTINUE);
1221 * ==========================================================================
1222 * Initiate I/O, either sync or async
1223 * ==========================================================================
1226 zio_wait(zio_t *zio)
1230 ASSERT(zio->io_stage == ZIO_STAGE_OPEN);
1231 ASSERT(zio->io_executor == NULL);
1233 zio->io_waiter = curthread;
1237 mutex_enter(&zio->io_lock);
1238 while (zio->io_executor != NULL)
1239 cv_wait(&zio->io_cv, &zio->io_lock);
1240 mutex_exit(&zio->io_lock);
1242 error = zio->io_error;
1249 zio_nowait(zio_t *zio)
1251 ASSERT(zio->io_executor == NULL);
1253 if (zio->io_child_type == ZIO_CHILD_LOGICAL &&
1254 zio_unique_parent(zio) == NULL) {
1256 * This is a logical async I/O with no parent to wait for it.
1257 * We add it to the spa_async_root_zio "Godfather" I/O which
1258 * will ensure they complete prior to unloading the pool.
1260 spa_t *spa = zio->io_spa;
1262 zio_add_child(spa->spa_async_zio_root, zio);
1269 * ==========================================================================
1270 * Reexecute or suspend/resume failed I/O
1271 * ==========================================================================
1275 zio_reexecute(zio_t *pio)
1277 zio_t *cio, *cio_next;
1280 ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL);
1281 ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN);
1282 ASSERT(pio->io_gang_leader == NULL);
1283 ASSERT(pio->io_gang_tree == NULL);
1285 pio->io_flags = pio->io_orig_flags;
1286 pio->io_stage = pio->io_orig_stage;
1287 pio->io_pipeline = pio->io_orig_pipeline;
1288 pio->io_reexecute = 0;
1290 for (w = 0; w < ZIO_WAIT_TYPES; w++)
1291 pio->io_state[w] = 0;
1292 for (c = 0; c < ZIO_CHILD_TYPES; c++)
1293 pio->io_child_error[c] = 0;
1295 if (IO_IS_ALLOCATING(pio))
1296 BP_ZERO(pio->io_bp);
1299 * As we reexecute pio's children, new children could be created.
1300 * New children go to the head of pio's io_child_list, however,
1301 * so we will (correctly) not reexecute them. The key is that
1302 * the remainder of pio's io_child_list, from 'cio_next' onward,
1303 * cannot be affected by any side effects of reexecuting 'cio'.
1305 for (cio = zio_walk_children(pio); cio != NULL; cio = cio_next) {
1306 cio_next = zio_walk_children(pio);
1307 mutex_enter(&pio->io_lock);
1308 for (w = 0; w < ZIO_WAIT_TYPES; w++)
1309 pio->io_children[cio->io_child_type][w]++;
1310 mutex_exit(&pio->io_lock);
1315 * Now that all children have been reexecuted, execute the parent.
1316 * We don't reexecute "The Godfather" I/O here as it's the
1317 * responsibility of the caller to wait on him.
1319 if (!(pio->io_flags & ZIO_FLAG_GODFATHER))
1324 zio_suspend(spa_t *spa, zio_t *zio)
1326 if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC)
1327 fm_panic("Pool '%s' has encountered an uncorrectable I/O "
1328 "failure and the failure mode property for this pool "
1329 "is set to panic.", spa_name(spa));
1331 zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, 0, 0);
1333 mutex_enter(&spa->spa_suspend_lock);
1335 if (spa->spa_suspend_zio_root == NULL)
1336 spa->spa_suspend_zio_root = zio_root(spa, NULL, NULL,
1337 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
1338 ZIO_FLAG_GODFATHER);
1340 spa->spa_suspended = B_TRUE;
1343 ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER));
1344 ASSERT(zio != spa->spa_suspend_zio_root);
1345 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
1346 ASSERT(zio_unique_parent(zio) == NULL);
1347 ASSERT(zio->io_stage == ZIO_STAGE_DONE);
1348 zio_add_child(spa->spa_suspend_zio_root, zio);
1351 mutex_exit(&spa->spa_suspend_lock);
1355 zio_resume(spa_t *spa)
1360 * Reexecute all previously suspended i/o.
1362 mutex_enter(&spa->spa_suspend_lock);
1363 spa->spa_suspended = B_FALSE;
1364 cv_broadcast(&spa->spa_suspend_cv);
1365 pio = spa->spa_suspend_zio_root;
1366 spa->spa_suspend_zio_root = NULL;
1367 mutex_exit(&spa->spa_suspend_lock);
1373 return (zio_wait(pio));
1377 zio_resume_wait(spa_t *spa)
1379 mutex_enter(&spa->spa_suspend_lock);
1380 while (spa_suspended(spa))
1381 cv_wait(&spa->spa_suspend_cv, &spa->spa_suspend_lock);
1382 mutex_exit(&spa->spa_suspend_lock);
1386 * ==========================================================================
1389 * A gang block is a collection of small blocks that looks to the DMU
1390 * like one large block. When zio_dva_allocate() cannot find a block
1391 * of the requested size, due to either severe fragmentation or the pool
1392 * being nearly full, it calls zio_write_gang_block() to construct the
1393 * block from smaller fragments.
1395 * A gang block consists of a gang header (zio_gbh_phys_t) and up to
1396 * three (SPA_GBH_NBLKPTRS) gang members. The gang header is just like
1397 * an indirect block: it's an array of block pointers. It consumes
1398 * only one sector and hence is allocatable regardless of fragmentation.
1399 * The gang header's bps point to its gang members, which hold the data.
1401 * Gang blocks are self-checksumming, using the bp's <vdev, offset, txg>
1402 * as the verifier to ensure uniqueness of the SHA256 checksum.
1403 * Critically, the gang block bp's blk_cksum is the checksum of the data,
1404 * not the gang header. This ensures that data block signatures (needed for
1405 * deduplication) are independent of how the block is physically stored.
1407 * Gang blocks can be nested: a gang member may itself be a gang block.
1408 * Thus every gang block is a tree in which root and all interior nodes are
1409 * gang headers, and the leaves are normal blocks that contain user data.
1410 * The root of the gang tree is called the gang leader.
1412 * To perform any operation (read, rewrite, free, claim) on a gang block,
1413 * zio_gang_assemble() first assembles the gang tree (minus data leaves)
1414 * in the io_gang_tree field of the original logical i/o by recursively
1415 * reading the gang leader and all gang headers below it. This yields
1416 * an in-core tree containing the contents of every gang header and the
1417 * bps for every constituent of the gang block.
1419 * With the gang tree now assembled, zio_gang_issue() just walks the gang tree
1420 * and invokes a callback on each bp. To free a gang block, zio_gang_issue()
1421 * calls zio_free_gang() -- a trivial wrapper around zio_free() -- for each bp.
1422 * zio_claim_gang() provides a similarly trivial wrapper for zio_claim().
1423 * zio_read_gang() is a wrapper around zio_read() that omits reading gang
1424 * headers, since we already have those in io_gang_tree. zio_rewrite_gang()
1425 * performs a zio_rewrite() of the data or, for gang headers, a zio_rewrite()
1426 * of the gang header plus zio_checksum_compute() of the data to update the
1427 * gang header's blk_cksum as described above.
1429 * The two-phase assemble/issue model solves the problem of partial failure --
1430 * what if you'd freed part of a gang block but then couldn't read the
1431 * gang header for another part? Assembling the entire gang tree first
1432 * ensures that all the necessary gang header I/O has succeeded before
1433 * starting the actual work of free, claim, or write. Once the gang tree
1434 * is assembled, free and claim are in-memory operations that cannot fail.
1436 * In the event that a gang write fails, zio_dva_unallocate() walks the
1437 * gang tree to immediately free (i.e. insert back into the space map)
1438 * everything we've allocated. This ensures that we don't get ENOSPC
1439 * errors during repeated suspend/resume cycles due to a flaky device.
1441 * Gang rewrites only happen during sync-to-convergence. If we can't assemble
1442 * the gang tree, we won't modify the block, so we can safely defer the free
1443 * (knowing that the block is still intact). If we *can* assemble the gang
1444 * tree, then even if some of the rewrites fail, zio_dva_unallocate() will free
1445 * each constituent bp and we can allocate a new block on the next sync pass.
1447 * In all cases, the gang tree allows complete recovery from partial failure.
1448 * ==========================================================================
1452 zio_read_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data)
1457 return (zio_read(pio, pio->io_spa, bp, data, BP_GET_PSIZE(bp),
1458 NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio),
1459 &pio->io_bookmark));
1463 zio_rewrite_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data)
1468 zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp,
1469 gn->gn_gbh, SPA_GANGBLOCKSIZE, NULL, NULL, pio->io_priority,
1470 ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark);
1472 * As we rewrite each gang header, the pipeline will compute
1473 * a new gang block header checksum for it; but no one will
1474 * compute a new data checksum, so we do that here. The one
1475 * exception is the gang leader: the pipeline already computed
1476 * its data checksum because that stage precedes gang assembly.
1477 * (Presently, nothing actually uses interior data checksums;
1478 * this is just good hygiene.)
1480 if (gn != pio->io_gang_leader->io_gang_tree) {
1481 zio_checksum_compute(zio, BP_GET_CHECKSUM(bp),
1482 data, BP_GET_PSIZE(bp));
1485 * If we are here to damage data for testing purposes,
1486 * leave the GBH alone so that we can detect the damage.
1488 if (pio->io_gang_leader->io_flags & ZIO_FLAG_INDUCE_DAMAGE)
1489 zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES;
1491 zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp,
1492 data, BP_GET_PSIZE(bp), NULL, NULL, pio->io_priority,
1493 ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark);
1501 zio_free_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data)
1503 return (zio_free_sync(pio, pio->io_spa, pio->io_txg, bp,
1504 ZIO_GANG_CHILD_FLAGS(pio)));
1509 zio_claim_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data)
1511 return (zio_claim(pio, pio->io_spa, pio->io_txg, bp,
1512 NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio)));
1515 static zio_gang_issue_func_t *zio_gang_issue_func[ZIO_TYPES] = {
1524 static void zio_gang_tree_assemble_done(zio_t *zio);
1526 static zio_gang_node_t *
1527 zio_gang_node_alloc(zio_gang_node_t **gnpp)
1529 zio_gang_node_t *gn;
1531 ASSERT(*gnpp == NULL);
1533 gn = kmem_zalloc(sizeof (*gn), KM_PUSHPAGE);
1534 gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE);
1541 zio_gang_node_free(zio_gang_node_t **gnpp)
1543 zio_gang_node_t *gn = *gnpp;
1546 for (g = 0; g < SPA_GBH_NBLKPTRS; g++)
1547 ASSERT(gn->gn_child[g] == NULL);
1549 zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE);
1550 kmem_free(gn, sizeof (*gn));
1555 zio_gang_tree_free(zio_gang_node_t **gnpp)
1557 zio_gang_node_t *gn = *gnpp;
1563 for (g = 0; g < SPA_GBH_NBLKPTRS; g++)
1564 zio_gang_tree_free(&gn->gn_child[g]);
1566 zio_gang_node_free(gnpp);
1570 zio_gang_tree_assemble(zio_t *gio, blkptr_t *bp, zio_gang_node_t **gnpp)
1572 zio_gang_node_t *gn = zio_gang_node_alloc(gnpp);
1574 ASSERT(gio->io_gang_leader == gio);
1575 ASSERT(BP_IS_GANG(bp));
1577 zio_nowait(zio_read(gio, gio->io_spa, bp, gn->gn_gbh,
1578 SPA_GANGBLOCKSIZE, zio_gang_tree_assemble_done, gn,
1579 gio->io_priority, ZIO_GANG_CHILD_FLAGS(gio), &gio->io_bookmark));
1583 zio_gang_tree_assemble_done(zio_t *zio)
1585 zio_t *gio = zio->io_gang_leader;
1586 zio_gang_node_t *gn = zio->io_private;
1587 blkptr_t *bp = zio->io_bp;
1590 ASSERT(gio == zio_unique_parent(zio));
1591 ASSERT(zio->io_child_count == 0);
1596 if (BP_SHOULD_BYTESWAP(bp))
1597 byteswap_uint64_array(zio->io_data, zio->io_size);
1599 ASSERT(zio->io_data == gn->gn_gbh);
1600 ASSERT(zio->io_size == SPA_GANGBLOCKSIZE);
1601 ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC);
1603 for (g = 0; g < SPA_GBH_NBLKPTRS; g++) {
1604 blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g];
1605 if (!BP_IS_GANG(gbp))
1607 zio_gang_tree_assemble(gio, gbp, &gn->gn_child[g]);
1612 zio_gang_tree_issue(zio_t *pio, zio_gang_node_t *gn, blkptr_t *bp, void *data)
1614 zio_t *gio = pio->io_gang_leader;
1618 ASSERT(BP_IS_GANG(bp) == !!gn);
1619 ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp));
1620 ASSERT(BP_GET_LSIZE(bp) == BP_GET_PSIZE(bp) || gn == gio->io_gang_tree);
1623 * If you're a gang header, your data is in gn->gn_gbh.
1624 * If you're a gang member, your data is in 'data' and gn == NULL.
1626 zio = zio_gang_issue_func[gio->io_type](pio, bp, gn, data);
1629 ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC);
1631 for (g = 0; g < SPA_GBH_NBLKPTRS; g++) {
1632 blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g];
1633 if (BP_IS_HOLE(gbp))
1635 zio_gang_tree_issue(zio, gn->gn_child[g], gbp, data);
1636 data = (char *)data + BP_GET_PSIZE(gbp);
1640 if (gn == gio->io_gang_tree)
1641 ASSERT3P((char *)gio->io_data + gio->io_size, ==, data);
1648 zio_gang_assemble(zio_t *zio)
1650 blkptr_t *bp = zio->io_bp;
1652 ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == NULL);
1653 ASSERT(zio->io_child_type > ZIO_CHILD_GANG);
1655 zio->io_gang_leader = zio;
1657 zio_gang_tree_assemble(zio, bp, &zio->io_gang_tree);
1659 return (ZIO_PIPELINE_CONTINUE);
1663 zio_gang_issue(zio_t *zio)
1665 blkptr_t *bp = zio->io_bp;
1667 if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE))
1668 return (ZIO_PIPELINE_STOP);
1670 ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == zio);
1671 ASSERT(zio->io_child_type > ZIO_CHILD_GANG);
1673 if (zio->io_child_error[ZIO_CHILD_GANG] == 0)
1674 zio_gang_tree_issue(zio, zio->io_gang_tree, bp, zio->io_data);
1676 zio_gang_tree_free(&zio->io_gang_tree);
1678 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
1680 return (ZIO_PIPELINE_CONTINUE);
1684 zio_write_gang_member_ready(zio_t *zio)
1686 zio_t *pio = zio_unique_parent(zio);
1687 ASSERTV(zio_t *gio = zio->io_gang_leader;)
1688 dva_t *cdva = zio->io_bp->blk_dva;
1689 dva_t *pdva = pio->io_bp->blk_dva;
1693 if (BP_IS_HOLE(zio->io_bp))
1696 ASSERT(BP_IS_HOLE(&zio->io_bp_orig));
1698 ASSERT(zio->io_child_type == ZIO_CHILD_GANG);
1699 ASSERT3U(zio->io_prop.zp_copies, ==, gio->io_prop.zp_copies);
1700 ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp));
1701 ASSERT3U(pio->io_prop.zp_copies, <=, BP_GET_NDVAS(pio->io_bp));
1702 ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp));
1704 mutex_enter(&pio->io_lock);
1705 for (d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) {
1706 ASSERT(DVA_GET_GANG(&pdva[d]));
1707 asize = DVA_GET_ASIZE(&pdva[d]);
1708 asize += DVA_GET_ASIZE(&cdva[d]);
1709 DVA_SET_ASIZE(&pdva[d], asize);
1711 mutex_exit(&pio->io_lock);
1715 zio_write_gang_block(zio_t *pio)
1717 spa_t *spa = pio->io_spa;
1718 blkptr_t *bp = pio->io_bp;
1719 zio_t *gio = pio->io_gang_leader;
1721 zio_gang_node_t *gn, **gnpp;
1722 zio_gbh_phys_t *gbh;
1723 uint64_t txg = pio->io_txg;
1724 uint64_t resid = pio->io_size;
1726 int copies = gio->io_prop.zp_copies;
1727 int gbh_copies = MIN(copies + 1, spa_max_replication(spa));
1731 error = metaslab_alloc(spa, spa_normal_class(spa), SPA_GANGBLOCKSIZE,
1732 bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp,
1733 METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER);
1735 pio->io_error = error;
1736 return (ZIO_PIPELINE_CONTINUE);
1740 gnpp = &gio->io_gang_tree;
1742 gnpp = pio->io_private;
1743 ASSERT(pio->io_ready == zio_write_gang_member_ready);
1746 gn = zio_gang_node_alloc(gnpp);
1748 bzero(gbh, SPA_GANGBLOCKSIZE);
1751 * Create the gang header.
1753 zio = zio_rewrite(pio, spa, txg, bp, gbh, SPA_GANGBLOCKSIZE, NULL, NULL,
1754 pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark);
1757 * Create and nowait the gang children.
1759 for (g = 0; resid != 0; resid -= lsize, g++) {
1760 lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g),
1762 ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid);
1764 zp.zp_checksum = gio->io_prop.zp_checksum;
1765 zp.zp_compress = ZIO_COMPRESS_OFF;
1766 zp.zp_type = DMU_OT_NONE;
1768 zp.zp_copies = gio->io_prop.zp_copies;
1770 zp.zp_dedup_verify = 0;
1772 zio_nowait(zio_write(zio, spa, txg, &gbh->zg_blkptr[g],
1773 (char *)pio->io_data + (pio->io_size - resid), lsize, &zp,
1774 zio_write_gang_member_ready, NULL, &gn->gn_child[g],
1775 pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio),
1776 &pio->io_bookmark));
1780 * Set pio's pipeline to just wait for zio to finish.
1782 pio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
1786 return (ZIO_PIPELINE_CONTINUE);
1790 * ==========================================================================
1792 * ==========================================================================
1795 zio_ddt_child_read_done(zio_t *zio)
1797 blkptr_t *bp = zio->io_bp;
1798 ddt_entry_t *dde = zio->io_private;
1800 zio_t *pio = zio_unique_parent(zio);
1802 mutex_enter(&pio->io_lock);
1803 ddp = ddt_phys_select(dde, bp);
1804 if (zio->io_error == 0)
1805 ddt_phys_clear(ddp); /* this ddp doesn't need repair */
1806 if (zio->io_error == 0 && dde->dde_repair_data == NULL)
1807 dde->dde_repair_data = zio->io_data;
1809 zio_buf_free(zio->io_data, zio->io_size);
1810 mutex_exit(&pio->io_lock);
1814 zio_ddt_read_start(zio_t *zio)
1816 blkptr_t *bp = zio->io_bp;
1819 ASSERT(BP_GET_DEDUP(bp));
1820 ASSERT(BP_GET_PSIZE(bp) == zio->io_size);
1821 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
1823 if (zio->io_child_error[ZIO_CHILD_DDT]) {
1824 ddt_t *ddt = ddt_select(zio->io_spa, bp);
1825 ddt_entry_t *dde = ddt_repair_start(ddt, bp);
1826 ddt_phys_t *ddp = dde->dde_phys;
1827 ddt_phys_t *ddp_self = ddt_phys_select(dde, bp);
1830 ASSERT(zio->io_vsd == NULL);
1833 if (ddp_self == NULL)
1834 return (ZIO_PIPELINE_CONTINUE);
1836 for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
1837 if (ddp->ddp_phys_birth == 0 || ddp == ddp_self)
1839 ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp,
1841 zio_nowait(zio_read(zio, zio->io_spa, &blk,
1842 zio_buf_alloc(zio->io_size), zio->io_size,
1843 zio_ddt_child_read_done, dde, zio->io_priority,
1844 ZIO_DDT_CHILD_FLAGS(zio) | ZIO_FLAG_DONT_PROPAGATE,
1845 &zio->io_bookmark));
1847 return (ZIO_PIPELINE_CONTINUE);
1850 zio_nowait(zio_read(zio, zio->io_spa, bp,
1851 zio->io_data, zio->io_size, NULL, NULL, zio->io_priority,
1852 ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark));
1854 return (ZIO_PIPELINE_CONTINUE);
1858 zio_ddt_read_done(zio_t *zio)
1860 blkptr_t *bp = zio->io_bp;
1862 if (zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE))
1863 return (ZIO_PIPELINE_STOP);
1865 ASSERT(BP_GET_DEDUP(bp));
1866 ASSERT(BP_GET_PSIZE(bp) == zio->io_size);
1867 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
1869 if (zio->io_child_error[ZIO_CHILD_DDT]) {
1870 ddt_t *ddt = ddt_select(zio->io_spa, bp);
1871 ddt_entry_t *dde = zio->io_vsd;
1873 ASSERT(spa_load_state(zio->io_spa) != SPA_LOAD_NONE);
1874 return (ZIO_PIPELINE_CONTINUE);
1877 zio->io_stage = ZIO_STAGE_DDT_READ_START >> 1;
1878 zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE);
1879 return (ZIO_PIPELINE_STOP);
1881 if (dde->dde_repair_data != NULL) {
1882 bcopy(dde->dde_repair_data, zio->io_data, zio->io_size);
1883 zio->io_child_error[ZIO_CHILD_DDT] = 0;
1885 ddt_repair_done(ddt, dde);
1889 ASSERT(zio->io_vsd == NULL);
1891 return (ZIO_PIPELINE_CONTINUE);
1895 zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde)
1897 spa_t *spa = zio->io_spa;
1901 * Note: we compare the original data, not the transformed data,
1902 * because when zio->io_bp is an override bp, we will not have
1903 * pushed the I/O transforms. That's an important optimization
1904 * because otherwise we'd compress/encrypt all dmu_sync() data twice.
1906 for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) {
1907 zio_t *lio = dde->dde_lead_zio[p];
1910 return (lio->io_orig_size != zio->io_orig_size ||
1911 bcmp(zio->io_orig_data, lio->io_orig_data,
1912 zio->io_orig_size) != 0);
1916 for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) {
1917 ddt_phys_t *ddp = &dde->dde_phys[p];
1919 if (ddp->ddp_phys_birth != 0) {
1920 arc_buf_t *abuf = NULL;
1921 uint32_t aflags = ARC_WAIT;
1922 blkptr_t blk = *zio->io_bp;
1925 ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth);
1929 error = arc_read_nolock(NULL, spa, &blk,
1930 arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ,
1931 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
1932 &aflags, &zio->io_bookmark);
1935 if (arc_buf_size(abuf) != zio->io_orig_size ||
1936 bcmp(abuf->b_data, zio->io_orig_data,
1937 zio->io_orig_size) != 0)
1939 VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1);
1943 return (error != 0);
1951 zio_ddt_child_write_ready(zio_t *zio)
1953 int p = zio->io_prop.zp_copies;
1954 ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp);
1955 ddt_entry_t *dde = zio->io_private;
1956 ddt_phys_t *ddp = &dde->dde_phys[p];
1964 ASSERT(dde->dde_lead_zio[p] == zio);
1966 ddt_phys_fill(ddp, zio->io_bp);
1968 while ((pio = zio_walk_parents(zio)) != NULL)
1969 ddt_bp_fill(ddp, pio->io_bp, zio->io_txg);
1975 zio_ddt_child_write_done(zio_t *zio)
1977 int p = zio->io_prop.zp_copies;
1978 ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp);
1979 ddt_entry_t *dde = zio->io_private;
1980 ddt_phys_t *ddp = &dde->dde_phys[p];
1984 ASSERT(ddp->ddp_refcnt == 0);
1985 ASSERT(dde->dde_lead_zio[p] == zio);
1986 dde->dde_lead_zio[p] = NULL;
1988 if (zio->io_error == 0) {
1989 while (zio_walk_parents(zio) != NULL)
1990 ddt_phys_addref(ddp);
1992 ddt_phys_clear(ddp);
1999 zio_ddt_ditto_write_done(zio_t *zio)
2001 int p = DDT_PHYS_DITTO;
2002 blkptr_t *bp = zio->io_bp;
2003 ddt_t *ddt = ddt_select(zio->io_spa, bp);
2004 ddt_entry_t *dde = zio->io_private;
2005 ddt_phys_t *ddp = &dde->dde_phys[p];
2006 ddt_key_t *ddk = &dde->dde_key;
2007 ASSERTV(zio_prop_t *zp = &zio->io_prop);
2011 ASSERT(ddp->ddp_refcnt == 0);
2012 ASSERT(dde->dde_lead_zio[p] == zio);
2013 dde->dde_lead_zio[p] = NULL;
2015 if (zio->io_error == 0) {
2016 ASSERT(ZIO_CHECKSUM_EQUAL(bp->blk_cksum, ddk->ddk_cksum));
2017 ASSERT(zp->zp_copies < SPA_DVAS_PER_BP);
2018 ASSERT(zp->zp_copies == BP_GET_NDVAS(bp) - BP_IS_GANG(bp));
2019 if (ddp->ddp_phys_birth != 0)
2020 ddt_phys_free(ddt, ddk, ddp, zio->io_txg);
2021 ddt_phys_fill(ddp, bp);
2028 zio_ddt_write(zio_t *zio)
2030 spa_t *spa = zio->io_spa;
2031 blkptr_t *bp = zio->io_bp;
2032 uint64_t txg = zio->io_txg;
2033 zio_prop_t *zp = &zio->io_prop;
2034 int p = zp->zp_copies;
2038 ddt_t *ddt = ddt_select(spa, bp);
2042 ASSERT(BP_GET_DEDUP(bp));
2043 ASSERT(BP_GET_CHECKSUM(bp) == zp->zp_checksum);
2044 ASSERT(BP_IS_HOLE(bp) || zio->io_bp_override);
2047 dde = ddt_lookup(ddt, bp, B_TRUE);
2048 ddp = &dde->dde_phys[p];
2050 if (zp->zp_dedup_verify && zio_ddt_collision(zio, ddt, dde)) {
2052 * If we're using a weak checksum, upgrade to a strong checksum
2053 * and try again. If we're already using a strong checksum,
2054 * we can't resolve it, so just convert to an ordinary write.
2055 * (And automatically e-mail a paper to Nature?)
2057 if (!zio_checksum_table[zp->zp_checksum].ci_dedup) {
2058 zp->zp_checksum = spa_dedup_checksum(spa);
2059 zio_pop_transforms(zio);
2060 zio->io_stage = ZIO_STAGE_OPEN;
2065 zio->io_pipeline = ZIO_WRITE_PIPELINE;
2067 return (ZIO_PIPELINE_CONTINUE);
2070 ditto_copies = ddt_ditto_copies_needed(ddt, dde, ddp);
2071 ASSERT(ditto_copies < SPA_DVAS_PER_BP);
2073 if (ditto_copies > ddt_ditto_copies_present(dde) &&
2074 dde->dde_lead_zio[DDT_PHYS_DITTO] == NULL) {
2075 zio_prop_t czp = *zp;
2077 czp.zp_copies = ditto_copies;
2080 * If we arrived here with an override bp, we won't have run
2081 * the transform stack, so we won't have the data we need to
2082 * generate a child i/o. So, toss the override bp and restart.
2083 * This is safe, because using the override bp is just an
2084 * optimization; and it's rare, so the cost doesn't matter.
2086 if (zio->io_bp_override) {
2087 zio_pop_transforms(zio);
2088 zio->io_stage = ZIO_STAGE_OPEN;
2089 zio->io_pipeline = ZIO_WRITE_PIPELINE;
2090 zio->io_bp_override = NULL;
2093 return (ZIO_PIPELINE_CONTINUE);
2096 dio = zio_write(zio, spa, txg, bp, zio->io_orig_data,
2097 zio->io_orig_size, &czp, NULL,
2098 zio_ddt_ditto_write_done, dde, zio->io_priority,
2099 ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark);
2101 zio_push_transform(dio, zio->io_data, zio->io_size, 0, NULL);
2102 dde->dde_lead_zio[DDT_PHYS_DITTO] = dio;
2105 if (ddp->ddp_phys_birth != 0 || dde->dde_lead_zio[p] != NULL) {
2106 if (ddp->ddp_phys_birth != 0)
2107 ddt_bp_fill(ddp, bp, txg);
2108 if (dde->dde_lead_zio[p] != NULL)
2109 zio_add_child(zio, dde->dde_lead_zio[p]);
2111 ddt_phys_addref(ddp);
2112 } else if (zio->io_bp_override) {
2113 ASSERT(bp->blk_birth == txg);
2114 ASSERT(BP_EQUAL(bp, zio->io_bp_override));
2115 ddt_phys_fill(ddp, bp);
2116 ddt_phys_addref(ddp);
2118 cio = zio_write(zio, spa, txg, bp, zio->io_orig_data,
2119 zio->io_orig_size, zp, zio_ddt_child_write_ready,
2120 zio_ddt_child_write_done, dde, zio->io_priority,
2121 ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark);
2123 zio_push_transform(cio, zio->io_data, zio->io_size, 0, NULL);
2124 dde->dde_lead_zio[p] = cio;
2134 return (ZIO_PIPELINE_CONTINUE);
2137 ddt_entry_t *freedde; /* for debugging */
2140 zio_ddt_free(zio_t *zio)
2142 spa_t *spa = zio->io_spa;
2143 blkptr_t *bp = zio->io_bp;
2144 ddt_t *ddt = ddt_select(spa, bp);
2148 ASSERT(BP_GET_DEDUP(bp));
2149 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
2152 freedde = dde = ddt_lookup(ddt, bp, B_TRUE);
2153 ddp = ddt_phys_select(dde, bp);
2154 ddt_phys_decref(ddp);
2157 return (ZIO_PIPELINE_CONTINUE);
2161 * ==========================================================================
2162 * Allocate and free blocks
2163 * ==========================================================================
2166 zio_dva_allocate(zio_t *zio)
2168 spa_t *spa = zio->io_spa;
2169 metaslab_class_t *mc = spa_normal_class(spa);
2170 blkptr_t *bp = zio->io_bp;
2174 if (zio->io_gang_leader == NULL) {
2175 ASSERT(zio->io_child_type > ZIO_CHILD_GANG);
2176 zio->io_gang_leader = zio;
2179 ASSERT(BP_IS_HOLE(bp));
2180 ASSERT3U(BP_GET_NDVAS(bp), ==, 0);
2181 ASSERT3U(zio->io_prop.zp_copies, >, 0);
2182 ASSERT3U(zio->io_prop.zp_copies, <=, spa_max_replication(spa));
2183 ASSERT3U(zio->io_size, ==, BP_GET_PSIZE(bp));
2186 * The dump device does not support gang blocks so allocation on
2187 * behalf of the dump device (i.e. ZIO_FLAG_NODATA) must avoid
2188 * the "fast" gang feature.
2190 flags |= (zio->io_flags & ZIO_FLAG_NODATA) ? METASLAB_GANG_AVOID : 0;
2191 flags |= (zio->io_flags & ZIO_FLAG_GANG_CHILD) ?
2192 METASLAB_GANG_CHILD : 0;
2193 error = metaslab_alloc(spa, mc, zio->io_size, bp,
2194 zio->io_prop.zp_copies, zio->io_txg, NULL, flags);
2197 spa_dbgmsg(spa, "%s: metaslab allocation failure: zio %p, "
2198 "size %llu, error %d", spa_name(spa), zio, zio->io_size,
2200 if (error == ENOSPC && zio->io_size > SPA_MINBLOCKSIZE)
2201 return (zio_write_gang_block(zio));
2202 zio->io_error = error;
2205 return (ZIO_PIPELINE_CONTINUE);
2209 zio_dva_free(zio_t *zio)
2211 metaslab_free(zio->io_spa, zio->io_bp, zio->io_txg, B_FALSE);
2213 return (ZIO_PIPELINE_CONTINUE);
2217 zio_dva_claim(zio_t *zio)
2221 error = metaslab_claim(zio->io_spa, zio->io_bp, zio->io_txg);
2223 zio->io_error = error;
2225 return (ZIO_PIPELINE_CONTINUE);
2229 * Undo an allocation. This is used by zio_done() when an I/O fails
2230 * and we want to give back the block we just allocated.
2231 * This handles both normal blocks and gang blocks.
2234 zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp)
2238 ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp));
2239 ASSERT(zio->io_bp_override == NULL);
2241 if (!BP_IS_HOLE(bp))
2242 metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE);
2245 for (g = 0; g < SPA_GBH_NBLKPTRS; g++) {
2246 zio_dva_unallocate(zio, gn->gn_child[g],
2247 &gn->gn_gbh->zg_blkptr[g]);
2253 * Try to allocate an intent log block. Return 0 on success, errno on failure.
2256 zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, blkptr_t *old_bp,
2257 uint64_t size, boolean_t use_slog)
2261 ASSERT(txg > spa_syncing_txg(spa));
2264 error = metaslab_alloc(spa, spa_log_class(spa), size,
2265 new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID);
2268 error = metaslab_alloc(spa, spa_normal_class(spa), size,
2269 new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID);
2272 BP_SET_LSIZE(new_bp, size);
2273 BP_SET_PSIZE(new_bp, size);
2274 BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF);
2275 BP_SET_CHECKSUM(new_bp,
2276 spa_version(spa) >= SPA_VERSION_SLIM_ZIL
2277 ? ZIO_CHECKSUM_ZILOG2 : ZIO_CHECKSUM_ZILOG);
2278 BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG);
2279 BP_SET_LEVEL(new_bp, 0);
2280 BP_SET_DEDUP(new_bp, 0);
2281 BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER);
2288 * Free an intent log block.
2291 zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp)
2293 ASSERT(BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG);
2294 ASSERT(!BP_IS_GANG(bp));
2296 zio_free(spa, txg, bp);
2300 * ==========================================================================
2301 * Read and write to physical devices
2302 * ==========================================================================
2305 zio_vdev_io_start(zio_t *zio)
2307 vdev_t *vd = zio->io_vd;
2309 spa_t *spa = zio->io_spa;
2311 ASSERT(zio->io_error == 0);
2312 ASSERT(zio->io_child_error[ZIO_CHILD_VDEV] == 0);
2315 if (!(zio->io_flags & ZIO_FLAG_CONFIG_WRITER))
2316 spa_config_enter(spa, SCL_ZIO, zio, RW_READER);
2319 * The mirror_ops handle multiple DVAs in a single BP.
2321 return (vdev_mirror_ops.vdev_op_io_start(zio));
2325 * We keep track of time-sensitive I/Os so that the scan thread
2326 * can quickly react to certain workloads. In particular, we care
2327 * about non-scrubbing, top-level reads and writes with the following
2329 * - synchronous writes of user data to non-slog devices
2330 * - any reads of user data
2331 * When these conditions are met, adjust the timestamp of spa_last_io
2332 * which allows the scan thread to adjust its workload accordingly.
2334 if (!(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_bp != NULL &&
2335 vd == vd->vdev_top && !vd->vdev_islog &&
2336 zio->io_bookmark.zb_objset != DMU_META_OBJSET &&
2337 zio->io_txg != spa_syncing_txg(spa)) {
2338 uint64_t old = spa->spa_last_io;
2339 uint64_t new = ddi_get_lbolt64();
2341 (void) atomic_cas_64(&spa->spa_last_io, old, new);
2344 align = 1ULL << vd->vdev_top->vdev_ashift;
2346 if (P2PHASE(zio->io_size, align) != 0) {
2347 uint64_t asize = P2ROUNDUP(zio->io_size, align);
2348 char *abuf = zio_buf_alloc(asize);
2349 ASSERT(vd == vd->vdev_top);
2350 if (zio->io_type == ZIO_TYPE_WRITE) {
2351 bcopy(zio->io_data, abuf, zio->io_size);
2352 bzero(abuf + zio->io_size, asize - zio->io_size);
2354 zio_push_transform(zio, abuf, asize, asize, zio_subblock);
2357 ASSERT(P2PHASE(zio->io_offset, align) == 0);
2358 ASSERT(P2PHASE(zio->io_size, align) == 0);
2359 VERIFY(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa));
2362 * If this is a repair I/O, and there's no self-healing involved --
2363 * that is, we're just resilvering what we expect to resilver --
2364 * then don't do the I/O unless zio's txg is actually in vd's DTL.
2365 * This prevents spurious resilvering with nested replication.
2366 * For example, given a mirror of mirrors, (A+B)+(C+D), if only
2367 * A is out of date, we'll read from C+D, then use the data to
2368 * resilver A+B -- but we don't actually want to resilver B, just A.
2369 * The top-level mirror has no way to know this, so instead we just
2370 * discard unnecessary repairs as we work our way down the vdev tree.
2371 * The same logic applies to any form of nested replication:
2372 * ditto + mirror, RAID-Z + replacing, etc. This covers them all.
2374 if ((zio->io_flags & ZIO_FLAG_IO_REPAIR) &&
2375 !(zio->io_flags & ZIO_FLAG_SELF_HEAL) &&
2376 zio->io_txg != 0 && /* not a delegated i/o */
2377 !vdev_dtl_contains(vd, DTL_PARTIAL, zio->io_txg, 1)) {
2378 ASSERT(zio->io_type == ZIO_TYPE_WRITE);
2379 zio_vdev_io_bypass(zio);
2380 return (ZIO_PIPELINE_CONTINUE);
2383 if (vd->vdev_ops->vdev_op_leaf &&
2384 (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE)) {
2386 if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0)
2387 return (ZIO_PIPELINE_CONTINUE);
2389 if ((zio = vdev_queue_io(zio)) == NULL)
2390 return (ZIO_PIPELINE_STOP);
2392 if (!vdev_accessible(vd, zio)) {
2393 zio->io_error = ENXIO;
2395 return (ZIO_PIPELINE_STOP);
2399 return (vd->vdev_ops->vdev_op_io_start(zio));
2403 zio_vdev_io_done(zio_t *zio)
2405 vdev_t *vd = zio->io_vd;
2406 vdev_ops_t *ops = vd ? vd->vdev_ops : &vdev_mirror_ops;
2407 boolean_t unexpected_error = B_FALSE;
2409 if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE))
2410 return (ZIO_PIPELINE_STOP);
2412 ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE);
2414 if (vd != NULL && vd->vdev_ops->vdev_op_leaf) {
2416 vdev_queue_io_done(zio);
2418 if (zio->io_type == ZIO_TYPE_WRITE)
2419 vdev_cache_write(zio);
2421 if (zio_injection_enabled && zio->io_error == 0)
2422 zio->io_error = zio_handle_device_injection(vd,
2425 if (zio_injection_enabled && zio->io_error == 0)
2426 zio->io_error = zio_handle_label_injection(zio, EIO);
2428 if (zio->io_error) {
2429 if (!vdev_accessible(vd, zio)) {
2430 zio->io_error = ENXIO;
2432 unexpected_error = B_TRUE;
2437 ops->vdev_op_io_done(zio);
2439 if (unexpected_error)
2440 VERIFY(vdev_probe(vd, zio) == NULL);
2442 return (ZIO_PIPELINE_CONTINUE);
2446 * For non-raidz ZIOs, we can just copy aside the bad data read from the
2447 * disk, and use that to finish the checksum ereport later.
2450 zio_vsd_default_cksum_finish(zio_cksum_report_t *zcr,
2451 const void *good_buf)
2453 /* no processing needed */
2454 zfs_ereport_finish_checksum(zcr, good_buf, zcr->zcr_cbdata, B_FALSE);
2459 zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *ignored)
2461 void *buf = zio_buf_alloc(zio->io_size);
2463 bcopy(zio->io_data, buf, zio->io_size);
2465 zcr->zcr_cbinfo = zio->io_size;
2466 zcr->zcr_cbdata = buf;
2467 zcr->zcr_finish = zio_vsd_default_cksum_finish;
2468 zcr->zcr_free = zio_buf_free;
2472 zio_vdev_io_assess(zio_t *zio)
2474 vdev_t *vd = zio->io_vd;
2476 if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE))
2477 return (ZIO_PIPELINE_STOP);
2479 if (vd == NULL && !(zio->io_flags & ZIO_FLAG_CONFIG_WRITER))
2480 spa_config_exit(zio->io_spa, SCL_ZIO, zio);
2482 if (zio->io_vsd != NULL) {
2483 zio->io_vsd_ops->vsd_free(zio);
2487 if (zio_injection_enabled && zio->io_error == 0)
2488 zio->io_error = zio_handle_fault_injection(zio, EIO);
2491 * If the I/O failed, determine whether we should attempt to retry it.
2493 * On retry, we cut in line in the issue queue, since we don't want
2494 * compression/checksumming/etc. work to prevent our (cheap) IO reissue.
2496 if (zio->io_error && vd == NULL &&
2497 !(zio->io_flags & (ZIO_FLAG_DONT_RETRY | ZIO_FLAG_IO_RETRY))) {
2498 ASSERT(!(zio->io_flags & ZIO_FLAG_DONT_QUEUE)); /* not a leaf */
2499 ASSERT(!(zio->io_flags & ZIO_FLAG_IO_BYPASS)); /* not a leaf */
2501 zio->io_flags |= ZIO_FLAG_IO_RETRY |
2502 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE;
2503 zio->io_stage = ZIO_STAGE_VDEV_IO_START >> 1;
2504 zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE,
2505 zio_requeue_io_start_cut_in_line);
2506 return (ZIO_PIPELINE_STOP);
2510 * If we got an error on a leaf device, convert it to ENXIO
2511 * if the device is not accessible at all.
2513 if (zio->io_error && vd != NULL && vd->vdev_ops->vdev_op_leaf &&
2514 !vdev_accessible(vd, zio))
2515 zio->io_error = ENXIO;
2518 * If we can't write to an interior vdev (mirror or RAID-Z),
2519 * set vdev_cant_write so that we stop trying to allocate from it.
2521 if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE &&
2522 vd != NULL && !vd->vdev_ops->vdev_op_leaf)
2523 vd->vdev_cant_write = B_TRUE;
2526 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
2528 return (ZIO_PIPELINE_CONTINUE);
2532 zio_vdev_io_reissue(zio_t *zio)
2534 ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START);
2535 ASSERT(zio->io_error == 0);
2537 zio->io_stage >>= 1;
2541 zio_vdev_io_redone(zio_t *zio)
2543 ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_DONE);
2545 zio->io_stage >>= 1;
2549 zio_vdev_io_bypass(zio_t *zio)
2551 ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START);
2552 ASSERT(zio->io_error == 0);
2554 zio->io_flags |= ZIO_FLAG_IO_BYPASS;
2555 zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS >> 1;
2559 * ==========================================================================
2560 * Generate and verify checksums
2561 * ==========================================================================
2564 zio_checksum_generate(zio_t *zio)
2566 blkptr_t *bp = zio->io_bp;
2567 enum zio_checksum checksum;
2571 * This is zio_write_phys().
2572 * We're either generating a label checksum, or none at all.
2574 checksum = zio->io_prop.zp_checksum;
2576 if (checksum == ZIO_CHECKSUM_OFF)
2577 return (ZIO_PIPELINE_CONTINUE);
2579 ASSERT(checksum == ZIO_CHECKSUM_LABEL);
2581 if (BP_IS_GANG(bp) && zio->io_child_type == ZIO_CHILD_GANG) {
2582 ASSERT(!IO_IS_ALLOCATING(zio));
2583 checksum = ZIO_CHECKSUM_GANG_HEADER;
2585 checksum = BP_GET_CHECKSUM(bp);
2589 zio_checksum_compute(zio, checksum, zio->io_data, zio->io_size);
2591 return (ZIO_PIPELINE_CONTINUE);
2595 zio_checksum_verify(zio_t *zio)
2597 zio_bad_cksum_t info;
2598 blkptr_t *bp = zio->io_bp;
2601 ASSERT(zio->io_vd != NULL);
2605 * This is zio_read_phys().
2606 * We're either verifying a label checksum, or nothing at all.
2608 if (zio->io_prop.zp_checksum == ZIO_CHECKSUM_OFF)
2609 return (ZIO_PIPELINE_CONTINUE);
2611 ASSERT(zio->io_prop.zp_checksum == ZIO_CHECKSUM_LABEL);
2614 if ((error = zio_checksum_error(zio, &info)) != 0) {
2615 zio->io_error = error;
2616 if (!(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
2617 zfs_ereport_start_checksum(zio->io_spa,
2618 zio->io_vd, zio, zio->io_offset,
2619 zio->io_size, NULL, &info);
2623 return (ZIO_PIPELINE_CONTINUE);
2627 * Called by RAID-Z to ensure we don't compute the checksum twice.
2630 zio_checksum_verified(zio_t *zio)
2632 zio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY;
2636 * ==========================================================================
2637 * Error rank. Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other.
2638 * An error of 0 indictes success. ENXIO indicates whole-device failure,
2639 * which may be transient (e.g. unplugged) or permament. ECKSUM and EIO
2640 * indicate errors that are specific to one I/O, and most likely permanent.
2641 * Any other error is presumed to be worse because we weren't expecting it.
2642 * ==========================================================================
2645 zio_worst_error(int e1, int e2)
2647 static int zio_error_rank[] = { 0, ENXIO, ECKSUM, EIO };
2650 for (r1 = 0; r1 < sizeof (zio_error_rank) / sizeof (int); r1++)
2651 if (e1 == zio_error_rank[r1])
2654 for (r2 = 0; r2 < sizeof (zio_error_rank) / sizeof (int); r2++)
2655 if (e2 == zio_error_rank[r2])
2658 return (r1 > r2 ? e1 : e2);
2662 * ==========================================================================
2664 * ==========================================================================
2667 zio_ready(zio_t *zio)
2669 blkptr_t *bp = zio->io_bp;
2670 zio_t *pio, *pio_next;
2672 if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) ||
2673 zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_READY))
2674 return (ZIO_PIPELINE_STOP);
2676 if (zio->io_ready) {
2677 ASSERT(IO_IS_ALLOCATING(zio));
2678 ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp));
2679 ASSERT(zio->io_children[ZIO_CHILD_GANG][ZIO_WAIT_READY] == 0);
2684 if (bp != NULL && bp != &zio->io_bp_copy)
2685 zio->io_bp_copy = *bp;
2688 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
2690 mutex_enter(&zio->io_lock);
2691 zio->io_state[ZIO_WAIT_READY] = 1;
2692 pio = zio_walk_parents(zio);
2693 mutex_exit(&zio->io_lock);
2696 * As we notify zio's parents, new parents could be added.
2697 * New parents go to the head of zio's io_parent_list, however,
2698 * so we will (correctly) not notify them. The remainder of zio's
2699 * io_parent_list, from 'pio_next' onward, cannot change because
2700 * all parents must wait for us to be done before they can be done.
2702 for (; pio != NULL; pio = pio_next) {
2703 pio_next = zio_walk_parents(zio);
2704 zio_notify_parent(pio, zio, ZIO_WAIT_READY);
2707 if (zio->io_flags & ZIO_FLAG_NODATA) {
2708 if (BP_IS_GANG(bp)) {
2709 zio->io_flags &= ~ZIO_FLAG_NODATA;
2711 ASSERT((uintptr_t)zio->io_data < SPA_MAXBLOCKSIZE);
2712 zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES;
2716 if (zio_injection_enabled &&
2717 zio->io_spa->spa_syncing_txg == zio->io_txg)
2718 zio_handle_ignored_writes(zio);
2720 return (ZIO_PIPELINE_CONTINUE);
2724 zio_done(zio_t *zio)
2726 zio_t *pio, *pio_next;
2730 * If our children haven't all completed,
2731 * wait for them and then repeat this pipeline stage.
2733 if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE) ||
2734 zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE) ||
2735 zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE) ||
2736 zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_DONE))
2737 return (ZIO_PIPELINE_STOP);
2739 for (c = 0; c < ZIO_CHILD_TYPES; c++)
2740 for (w = 0; w < ZIO_WAIT_TYPES; w++)
2741 ASSERT(zio->io_children[c][w] == 0);
2743 if (zio->io_bp != NULL) {
2744 ASSERT(zio->io_bp->blk_pad[0] == 0);
2745 ASSERT(zio->io_bp->blk_pad[1] == 0);
2746 ASSERT(bcmp(zio->io_bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 ||
2747 (zio->io_bp == zio_unique_parent(zio)->io_bp));
2748 if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(zio->io_bp) &&
2749 zio->io_bp_override == NULL &&
2750 !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) {
2751 ASSERT(!BP_SHOULD_BYTESWAP(zio->io_bp));
2752 ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp));
2753 ASSERT(BP_COUNT_GANG(zio->io_bp) == 0 ||
2754 (BP_COUNT_GANG(zio->io_bp) == BP_GET_NDVAS(zio->io_bp)));
2759 * If there were child vdev/gang/ddt errors, they apply to us now.
2761 zio_inherit_child_errors(zio, ZIO_CHILD_VDEV);
2762 zio_inherit_child_errors(zio, ZIO_CHILD_GANG);
2763 zio_inherit_child_errors(zio, ZIO_CHILD_DDT);
2766 * If the I/O on the transformed data was successful, generate any
2767 * checksum reports now while we still have the transformed data.
2769 if (zio->io_error == 0) {
2770 while (zio->io_cksum_report != NULL) {
2771 zio_cksum_report_t *zcr = zio->io_cksum_report;
2772 uint64_t align = zcr->zcr_align;
2773 uint64_t asize = P2ROUNDUP(zio->io_size, align);
2774 char *abuf = zio->io_data;
2776 if (asize != zio->io_size) {
2777 abuf = zio_buf_alloc(asize);
2778 bcopy(zio->io_data, abuf, zio->io_size);
2779 bzero(abuf + zio->io_size, asize - zio->io_size);
2782 zio->io_cksum_report = zcr->zcr_next;
2783 zcr->zcr_next = NULL;
2784 zcr->zcr_finish(zcr, abuf);
2785 zfs_ereport_free_checksum(zcr);
2787 if (asize != zio->io_size)
2788 zio_buf_free(abuf, asize);
2792 zio_pop_transforms(zio); /* note: may set zio->io_error */
2794 vdev_stat_update(zio, zio->io_size);
2797 * If this I/O is attached to a particular vdev is slow, exeeding
2798 * 30 seconds to complete, post an error described the I/O delay.
2799 * We ignore these errors if the device is currently unavailable.
2801 if (zio->io_delay >= zio_delay_max) {
2802 if (zio->io_vd != NULL && !vdev_is_dead(zio->io_vd))
2803 zfs_ereport_post(FM_EREPORT_ZFS_DELAY, zio->io_spa,
2804 zio->io_vd, zio, 0, 0);
2807 if (zio->io_error) {
2809 * If this I/O is attached to a particular vdev,
2810 * generate an error message describing the I/O failure
2811 * at the block level. We ignore these errors if the
2812 * device is currently unavailable.
2814 if (zio->io_error != ECKSUM && zio->io_vd != NULL &&
2815 !vdev_is_dead(zio->io_vd))
2816 zfs_ereport_post(FM_EREPORT_ZFS_IO, zio->io_spa,
2817 zio->io_vd, zio, 0, 0);
2819 if ((zio->io_error == EIO || !(zio->io_flags &
2820 (ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) &&
2821 zio == zio->io_logical) {
2823 * For logical I/O requests, tell the SPA to log the
2824 * error and generate a logical data ereport.
2826 spa_log_error(zio->io_spa, zio);
2827 zfs_ereport_post(FM_EREPORT_ZFS_DATA, zio->io_spa, NULL, zio,
2832 if (zio->io_error && zio == zio->io_logical) {
2834 * Determine whether zio should be reexecuted. This will
2835 * propagate all the way to the root via zio_notify_parent().
2837 ASSERT(zio->io_vd == NULL && zio->io_bp != NULL);
2838 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
2840 if (IO_IS_ALLOCATING(zio) &&
2841 !(zio->io_flags & ZIO_FLAG_CANFAIL)) {
2842 if (zio->io_error != ENOSPC)
2843 zio->io_reexecute |= ZIO_REEXECUTE_NOW;
2845 zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND;
2848 if ((zio->io_type == ZIO_TYPE_READ ||
2849 zio->io_type == ZIO_TYPE_FREE) &&
2850 !(zio->io_flags & ZIO_FLAG_SCAN_THREAD) &&
2851 zio->io_error == ENXIO &&
2852 spa_load_state(zio->io_spa) == SPA_LOAD_NONE &&
2853 spa_get_failmode(zio->io_spa) != ZIO_FAILURE_MODE_CONTINUE)
2854 zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND;
2856 if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute)
2857 zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND;
2860 * Here is a possibly good place to attempt to do
2861 * either combinatorial reconstruction or error correction
2862 * based on checksums. It also might be a good place
2863 * to send out preliminary ereports before we suspend
2869 * If there were logical child errors, they apply to us now.
2870 * We defer this until now to avoid conflating logical child
2871 * errors with errors that happened to the zio itself when
2872 * updating vdev stats and reporting FMA events above.
2874 zio_inherit_child_errors(zio, ZIO_CHILD_LOGICAL);
2876 if ((zio->io_error || zio->io_reexecute) &&
2877 IO_IS_ALLOCATING(zio) && zio->io_gang_leader == zio &&
2878 !(zio->io_flags & ZIO_FLAG_IO_REWRITE))
2879 zio_dva_unallocate(zio, zio->io_gang_tree, zio->io_bp);
2881 zio_gang_tree_free(&zio->io_gang_tree);
2884 * Godfather I/Os should never suspend.
2886 if ((zio->io_flags & ZIO_FLAG_GODFATHER) &&
2887 (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND))
2888 zio->io_reexecute = 0;
2890 if (zio->io_reexecute) {
2892 * This is a logical I/O that wants to reexecute.
2894 * Reexecute is top-down. When an i/o fails, if it's not
2895 * the root, it simply notifies its parent and sticks around.
2896 * The parent, seeing that it still has children in zio_done(),
2897 * does the same. This percolates all the way up to the root.
2898 * The root i/o will reexecute or suspend the entire tree.
2900 * This approach ensures that zio_reexecute() honors
2901 * all the original i/o dependency relationships, e.g.
2902 * parents not executing until children are ready.
2904 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
2906 zio->io_gang_leader = NULL;
2908 mutex_enter(&zio->io_lock);
2909 zio->io_state[ZIO_WAIT_DONE] = 1;
2910 mutex_exit(&zio->io_lock);
2913 * "The Godfather" I/O monitors its children but is
2914 * not a true parent to them. It will track them through
2915 * the pipeline but severs its ties whenever they get into
2916 * trouble (e.g. suspended). This allows "The Godfather"
2917 * I/O to return status without blocking.
2919 for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) {
2920 zio_link_t *zl = zio->io_walk_link;
2921 pio_next = zio_walk_parents(zio);
2923 if ((pio->io_flags & ZIO_FLAG_GODFATHER) &&
2924 (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) {
2925 zio_remove_child(pio, zio, zl);
2926 zio_notify_parent(pio, zio, ZIO_WAIT_DONE);
2930 if ((pio = zio_unique_parent(zio)) != NULL) {
2932 * We're not a root i/o, so there's nothing to do
2933 * but notify our parent. Don't propagate errors
2934 * upward since we haven't permanently failed yet.
2936 ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER));
2937 zio->io_flags |= ZIO_FLAG_DONT_PROPAGATE;
2938 zio_notify_parent(pio, zio, ZIO_WAIT_DONE);
2939 } else if (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND) {
2941 * We'd fail again if we reexecuted now, so suspend
2942 * until conditions improve (e.g. device comes online).
2944 zio_suspend(zio->io_spa, zio);
2947 * Reexecution is potentially a huge amount of work.
2948 * Hand it off to the otherwise-unused claim taskq.
2950 (void) taskq_dispatch(
2951 zio->io_spa->spa_zio_taskq[ZIO_TYPE_CLAIM][ZIO_TASKQ_ISSUE],
2952 (task_func_t *)zio_reexecute, zio, TQ_SLEEP);
2954 return (ZIO_PIPELINE_STOP);
2957 ASSERT(zio->io_child_count == 0);
2958 ASSERT(zio->io_reexecute == 0);
2959 ASSERT(zio->io_error == 0 || (zio->io_flags & ZIO_FLAG_CANFAIL));
2962 * Report any checksum errors, since the I/O is complete.
2964 while (zio->io_cksum_report != NULL) {
2965 zio_cksum_report_t *zcr = zio->io_cksum_report;
2966 zio->io_cksum_report = zcr->zcr_next;
2967 zcr->zcr_next = NULL;
2968 zcr->zcr_finish(zcr, NULL);
2969 zfs_ereport_free_checksum(zcr);
2973 * It is the responsibility of the done callback to ensure that this
2974 * particular zio is no longer discoverable for adoption, and as
2975 * such, cannot acquire any new parents.
2980 mutex_enter(&zio->io_lock);
2981 zio->io_state[ZIO_WAIT_DONE] = 1;
2982 mutex_exit(&zio->io_lock);
2984 for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) {
2985 zio_link_t *zl = zio->io_walk_link;
2986 pio_next = zio_walk_parents(zio);
2987 zio_remove_child(pio, zio, zl);
2988 zio_notify_parent(pio, zio, ZIO_WAIT_DONE);
2991 if (zio->io_waiter != NULL) {
2992 mutex_enter(&zio->io_lock);
2993 zio->io_executor = NULL;
2994 cv_broadcast(&zio->io_cv);
2995 mutex_exit(&zio->io_lock);
3000 return (ZIO_PIPELINE_STOP);
3004 * ==========================================================================
3005 * I/O pipeline definition
3006 * ==========================================================================
3008 static zio_pipe_stage_t *zio_pipeline[] = {
3014 zio_checksum_generate,
3028 zio_checksum_verify,
3032 #if defined(_KERNEL) && defined(HAVE_SPL)
3033 /* Fault injection */
3034 EXPORT_SYMBOL(zio_injection_enabled);
3035 EXPORT_SYMBOL(zio_inject_fault);
3036 EXPORT_SYMBOL(zio_inject_list_next);
3037 EXPORT_SYMBOL(zio_clear_fault);
3038 EXPORT_SYMBOL(zio_handle_fault_injection);
3039 EXPORT_SYMBOL(zio_handle_device_injection);
3040 EXPORT_SYMBOL(zio_handle_label_injection);
3041 EXPORT_SYMBOL(zio_priority_table);
3042 EXPORT_SYMBOL(zio_type_name);
3044 module_param(zio_bulk_flags, int, 0644);
3045 MODULE_PARM_DESC(zio_bulk_flags, "Additional flags to pass to bulk buffers");
3047 module_param(zio_delay_max, int, 0644);
3048 MODULE_PARM_DESC(zio_delay_max, "Max zio millisec delay before posting event");
3050 module_param(zio_requeue_io_start_cut_in_line, int, 0644);
3051 MODULE_PARM_DESC(zio_requeue_io_start_cut_in_line, "Prioritize requeued I/O");