1 /*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
24 * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved.
25 * Copyright (c) 2014 Integros [integros.com]
26 */
27
28 #include <sys/sysmacros.h>
29 #include <sys/zfs_context.h>
30 #include <sys/fm/fs/zfs.h>
31 #include <sys/spa.h>
32 #include <sys/txg.h>
33 #include <sys/spa_impl.h>
34 #include <sys/vdev_impl.h>
35 #include <sys/zio_impl.h>
36 #include <sys/zio_compress.h>
37 #include <sys/zio_checksum.h>
38 #include <sys/dmu_objset.h>
39 #include <sys/arc.h>
40 #include <sys/ddt.h>
41 #include <sys/trim_map.h>
42 #include <sys/blkptr.h>
43 #include <sys/zfeature.h>
44 #include <sys/metaslab_impl.h>
45
46 SYSCTL_DECL(_vfs_zfs);
47 SYSCTL_NODE(_vfs_zfs, OID_AUTO, zio, CTLFLAG_RW, 0, "ZFS ZIO");
48 #ifdef __NetBSD__
49 const int zio_use_uma = 1;
50 #else
51 #if defined(__amd64__)
52 static int zio_use_uma = 1;
53 #else
54 static int zio_use_uma = 0;
55 #endif
56 #endif
57 SYSCTL_INT(_vfs_zfs_zio, OID_AUTO, use_uma, CTLFLAG_RDTUN, &zio_use_uma, 0,
58 "Use uma(9) for ZIO allocations");
59 static int zio_exclude_metadata = 0;
60 SYSCTL_INT(_vfs_zfs_zio, OID_AUTO, exclude_metadata, CTLFLAG_RDTUN, &zio_exclude_metadata, 0,
61 "Exclude metadata buffers from dumps as well");
62
63 zio_trim_stats_t zio_trim_stats = {
64 { "bytes", KSTAT_DATA_UINT64,
65 "Number of bytes successfully TRIMmed" },
66 { "success", KSTAT_DATA_UINT64,
67 "Number of successful TRIM requests" },
68 { "unsupported", KSTAT_DATA_UINT64,
69 "Number of TRIM requests that failed because TRIM is not supported" },
70 { "failed", KSTAT_DATA_UINT64,
71 "Number of TRIM requests that failed for reasons other than not supported" },
72 };
73
74 static kstat_t *zio_trim_ksp;
75
76 /*
77 * ==========================================================================
78 * I/O type descriptions
79 * ==========================================================================
80 */
81 const char *zio_type_name[ZIO_TYPES] = {
82 "zio_null", "zio_read", "zio_write", "zio_free", "zio_claim",
83 "zio_ioctl"
84 };
85
86 boolean_t zio_dva_throttle_enabled = B_TRUE;
87 SYSCTL_INT(_vfs_zfs_zio, OID_AUTO, dva_throttle_enabled, CTLFLAG_RDTUN,
88 &zio_dva_throttle_enabled, 0, "");
89
90 /*
91 * ==========================================================================
92 * I/O kmem caches
93 * ==========================================================================
94 */
95 kmem_cache_t *zio_cache;
96 kmem_cache_t *zio_link_cache;
97 kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT];
98 kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT];
99
100 #ifdef _KERNEL
101 extern vmem_t *zio_alloc_arena;
102 #endif
103
104 #define ZIO_PIPELINE_CONTINUE 0x100
105 #define ZIO_PIPELINE_STOP 0x101
106
107 #define BP_SPANB(indblkshift, level) \
108 (((uint64_t)1) << ((level) * ((indblkshift) - SPA_BLKPTRSHIFT)))
109 #define COMPARE_META_LEVEL 0x80000000ul
110 /*
111 * The following actions directly effect the spa's sync-to-convergence logic.
112 * The values below define the sync pass when we start performing the action.
113 * Care should be taken when changing these values as they directly impact
114 * spa_sync() performance. Tuning these values may introduce subtle performance
115 * pathologies and should only be done in the context of performance analysis.
116 * These tunables will eventually be removed and replaced with #defines once
117 * enough analysis has been done to determine optimal values.
118 *
119 * The 'zfs_sync_pass_deferred_free' pass must be greater than 1 to ensure that
120 * regular blocks are not deferred.
121 */
122 int zfs_sync_pass_deferred_free = 2; /* defer frees starting in this pass */
123 SYSCTL_INT(_vfs_zfs, OID_AUTO, sync_pass_deferred_free, CTLFLAG_RDTUN,
124 &zfs_sync_pass_deferred_free, 0, "defer frees starting in this pass");
125 int zfs_sync_pass_dont_compress = 5; /* don't compress starting in this pass */
126 SYSCTL_INT(_vfs_zfs, OID_AUTO, sync_pass_dont_compress, CTLFLAG_RDTUN,
127 &zfs_sync_pass_dont_compress, 0, "don't compress starting in this pass");
128 int zfs_sync_pass_rewrite = 2; /* rewrite new bps starting in this pass */
129 SYSCTL_INT(_vfs_zfs, OID_AUTO, sync_pass_rewrite, CTLFLAG_RDTUN,
130 &zfs_sync_pass_rewrite, 0, "rewrite new bps starting in this pass");
131
132 /*
133 * An allocating zio is one that either currently has the DVA allocate
134 * stage set or will have it later in its lifetime.
135 */
136 #define IO_IS_ALLOCATING(zio) ((zio)->io_orig_pipeline & ZIO_STAGE_DVA_ALLOCATE)
137
138 boolean_t zio_requeue_io_start_cut_in_line = B_TRUE;
139
140 #ifdef illumos
141 #ifdef ZFS_DEBUG
142 int zio_buf_debug_limit = 16384;
143 #else
144 int zio_buf_debug_limit = 0;
145 #endif
146 #endif
147
148 static void zio_taskq_dispatch(zio_t *, zio_taskq_type_t, boolean_t);
149
150 void
zio_init(void)151 zio_init(void)
152 {
153 size_t c;
154 zio_cache = kmem_cache_create("zio_cache",
155 sizeof (zio_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
156 zio_link_cache = kmem_cache_create("zio_link_cache",
157 sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
158
159 if (!zio_use_uma)
160 goto out;
161
162 /*
163 * For small buffers, we want a cache for each multiple of
164 * SPA_MINBLOCKSIZE. For larger buffers, we want a cache
165 * for each quarter-power of 2.
166 */
167 for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) {
168 size_t size = (c + 1) << SPA_MINBLOCKSHIFT;
169 size_t p2 = size;
170 size_t align = 0;
171 int cflags = zio_exclude_metadata ? KMC_NODEBUG : 0;
172
173 while (!ISP2(p2))
174 p2 &= p2 - 1;
175
176 #ifdef illumos
177 #ifndef _KERNEL
178 /*
179 * If we are using watchpoints, put each buffer on its own page,
180 * to eliminate the performance overhead of trapping to the
181 * kernel when modifying a non-watched buffer that shares the
182 * page with a watched buffer.
183 */
184 if (arc_watch && !IS_P2ALIGNED(size, PAGESIZE))
185 continue;
186 #endif
187 #endif /* illumos */
188 if (size <= 4 * SPA_MINBLOCKSIZE) {
189 align = SPA_MINBLOCKSIZE;
190 } else if (IS_P2ALIGNED(size, p2 >> 2)) {
191 align = MIN(p2 >> 2, PAGESIZE);
192 }
193
194 if (align != 0) {
195 char name[36];
196 (void) sprintf(name, "zio_buf_%lu", (ulong_t)size);
197 zio_buf_cache[c] = kmem_cache_create(name, size,
198 align, NULL, NULL, NULL, NULL, NULL, cflags);
199
200 /*
201 * Since zio_data bufs do not appear in crash dumps, we
202 * pass KMC_NOTOUCH so that no allocator metadata is
203 * stored with the buffers.
204 */
205 (void) sprintf(name, "zio_data_buf_%lu", (ulong_t)size);
206 zio_data_buf_cache[c] = kmem_cache_create(name, size,
207 align, NULL, NULL, NULL, NULL, NULL,
208 cflags | KMC_NOTOUCH | KMC_NODEBUG);
209 }
210 }
211
212 while (--c != 0) {
213 ASSERT(zio_buf_cache[c] != NULL);
214 if (zio_buf_cache[c - 1] == NULL)
215 zio_buf_cache[c - 1] = zio_buf_cache[c];
216
217 ASSERT(zio_data_buf_cache[c] != NULL);
218 if (zio_data_buf_cache[c - 1] == NULL)
219 zio_data_buf_cache[c - 1] = zio_data_buf_cache[c];
220 }
221 out:
222
223 zio_inject_init();
224
225 zio_trim_ksp = kstat_create("zfs", 0, "zio_trim", "misc",
226 KSTAT_TYPE_NAMED,
227 sizeof(zio_trim_stats) / sizeof(kstat_named_t),
228 KSTAT_FLAG_VIRTUAL);
229
230 if (zio_trim_ksp != NULL) {
231 zio_trim_ksp->ks_data = &zio_trim_stats;
232 kstat_install(zio_trim_ksp);
233 }
234 }
235
236 void
zio_fini(void)237 zio_fini(void)
238 {
239 size_t c;
240 kmem_cache_t *last_cache = NULL;
241 kmem_cache_t *last_data_cache = NULL;
242
243 for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) {
244 if (zio_buf_cache[c] != last_cache) {
245 last_cache = zio_buf_cache[c];
246 kmem_cache_destroy(zio_buf_cache[c]);
247 }
248 zio_buf_cache[c] = NULL;
249
250 if (zio_data_buf_cache[c] != last_data_cache) {
251 last_data_cache = zio_data_buf_cache[c];
252 kmem_cache_destroy(zio_data_buf_cache[c]);
253 }
254 zio_data_buf_cache[c] = NULL;
255 }
256
257 kmem_cache_destroy(zio_link_cache);
258 kmem_cache_destroy(zio_cache);
259
260 zio_inject_fini();
261
262 if (zio_trim_ksp != NULL) {
263 kstat_delete(zio_trim_ksp);
264 zio_trim_ksp = NULL;
265 }
266 }
267
268 /*
269 * ==========================================================================
270 * Allocate and free I/O buffers
271 * ==========================================================================
272 */
273
274 /*
275 * Use zio_buf_alloc to allocate ZFS metadata. This data will appear in a
276 * crashdump if the kernel panics, so use it judiciously. Obviously, it's
277 * useful to inspect ZFS metadata, but if possible, we should avoid keeping
278 * excess / transient data in-core during a crashdump.
279 */
280 static void *
zio_buf_alloc_impl(size_t size,boolean_t canwait)281 zio_buf_alloc_impl(size_t size, boolean_t canwait)
282 {
283 size_t c = (size - 1) >> SPA_MINBLOCKSHIFT;
284 int flags = zio_exclude_metadata ? KM_NODEBUG : 0;
285
286 VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
287
288 if (zio_use_uma) {
289 return (kmem_cache_alloc(zio_buf_cache[c],
290 canwait ? KM_PUSHPAGE : KM_NOSLEEP));
291 } else {
292 return (kmem_alloc(size,
293 (canwait ? KM_SLEEP : KM_NOSLEEP) | flags));
294 }
295 }
296
297 void *
zio_buf_alloc(size_t size)298 zio_buf_alloc(size_t size)
299 {
300 return (zio_buf_alloc_impl(size, B_TRUE));
301 }
302
303 void *
zio_buf_alloc_nowait(size_t size)304 zio_buf_alloc_nowait(size_t size)
305 {
306 return (zio_buf_alloc_impl(size, B_FALSE));
307 }
308
309 /*
310 * Use zio_data_buf_alloc to allocate data. The data will not appear in a
311 * crashdump if the kernel panics. This exists so that we will limit the amount
312 * of ZFS data that shows up in a kernel crashdump. (Thus reducing the amount
313 * of kernel heap dumped to disk when the kernel panics)
314 */
315 void *
zio_data_buf_alloc(size_t size)316 zio_data_buf_alloc(size_t size)
317 {
318 size_t c = (size - 1) >> SPA_MINBLOCKSHIFT;
319
320 VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
321
322 if (zio_use_uma)
323 return (kmem_cache_alloc(zio_data_buf_cache[c], KM_PUSHPAGE));
324 else
325 return (kmem_alloc(size, KM_SLEEP | KM_NODEBUG));
326 }
327
328 void
zio_buf_free(void * buf,size_t size)329 zio_buf_free(void *buf, size_t size)
330 {
331 size_t c = (size - 1) >> SPA_MINBLOCKSHIFT;
332
333 VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
334
335 if (zio_use_uma)
336 kmem_cache_free(zio_buf_cache[c], buf);
337 else
338 kmem_free(buf, size);
339 }
340
341 void
zio_data_buf_free(void * buf,size_t size)342 zio_data_buf_free(void *buf, size_t size)
343 {
344 size_t c = (size - 1) >> SPA_MINBLOCKSHIFT;
345
346 VERIFY3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT);
347
348 if (zio_use_uma)
349 kmem_cache_free(zio_data_buf_cache[c], buf);
350 else
351 kmem_free(buf, size);
352 }
353
354 /*
355 * ==========================================================================
356 * Push and pop I/O transform buffers
357 * ==========================================================================
358 */
359 void
zio_push_transform(zio_t * zio,void * data,uint64_t size,uint64_t bufsize,zio_transform_func_t * transform)360 zio_push_transform(zio_t *zio, void *data, uint64_t size, uint64_t bufsize,
361 zio_transform_func_t *transform)
362 {
363 zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_SLEEP);
364
365 zt->zt_orig_data = zio->io_data;
366 zt->zt_orig_size = zio->io_size;
367 zt->zt_bufsize = bufsize;
368 zt->zt_transform = transform;
369
370 zt->zt_next = zio->io_transform_stack;
371 zio->io_transform_stack = zt;
372
373 zio->io_data = data;
374 zio->io_size = size;
375 }
376
377 void
zio_pop_transforms(zio_t * zio)378 zio_pop_transforms(zio_t *zio)
379 {
380 zio_transform_t *zt;
381
382 while ((zt = zio->io_transform_stack) != NULL) {
383 if (zt->zt_transform != NULL)
384 zt->zt_transform(zio,
385 zt->zt_orig_data, zt->zt_orig_size);
386
387 if (zt->zt_bufsize != 0)
388 zio_buf_free(zio->io_data, zt->zt_bufsize);
389
390 zio->io_data = zt->zt_orig_data;
391 zio->io_size = zt->zt_orig_size;
392 zio->io_transform_stack = zt->zt_next;
393
394 kmem_free(zt, sizeof (zio_transform_t));
395 }
396 }
397
398 /*
399 * ==========================================================================
400 * I/O transform callbacks for subblocks and decompression
401 * ==========================================================================
402 */
403 static void
zio_subblock(zio_t * zio,void * data,uint64_t size)404 zio_subblock(zio_t *zio, void *data, uint64_t size)
405 {
406 ASSERT(zio->io_size > size);
407
408 if (zio->io_type == ZIO_TYPE_READ)
409 bcopy(zio->io_data, data, size);
410 }
411
412 static void
zio_decompress(zio_t * zio,void * data,uint64_t size)413 zio_decompress(zio_t *zio, void *data, uint64_t size)
414 {
415 if (zio->io_error == 0 &&
416 zio_decompress_data(BP_GET_COMPRESS(zio->io_bp),
417 zio->io_data, data, zio->io_size, size) != 0)
418 zio->io_error = SET_ERROR(EIO);
419 }
420
421 /*
422 * ==========================================================================
423 * I/O parent/child relationships and pipeline interlocks
424 * ==========================================================================
425 */
426 zio_t *
zio_walk_parents(zio_t * cio,zio_link_t ** zl)427 zio_walk_parents(zio_t *cio, zio_link_t **zl)
428 {
429 list_t *pl = &cio->io_parent_list;
430
431 *zl = (*zl == NULL) ? list_head(pl) : list_next(pl, *zl);
432 if (*zl == NULL)
433 return (NULL);
434
435 ASSERT((*zl)->zl_child == cio);
436 return ((*zl)->zl_parent);
437 }
438
439 zio_t *
zio_walk_children(zio_t * pio,zio_link_t ** zl)440 zio_walk_children(zio_t *pio, zio_link_t **zl)
441 {
442 list_t *cl = &pio->io_child_list;
443
444 *zl = (*zl == NULL) ? list_head(cl) : list_next(cl, *zl);
445 if (*zl == NULL)
446 return (NULL);
447
448 ASSERT((*zl)->zl_parent == pio);
449 return ((*zl)->zl_child);
450 }
451
452 zio_t *
zio_unique_parent(zio_t * cio)453 zio_unique_parent(zio_t *cio)
454 {
455 zio_link_t *zl = NULL;
456 zio_t *pio = zio_walk_parents(cio, &zl);
457
458 VERIFY3P(zio_walk_parents(cio, &zl), ==, NULL);
459 return (pio);
460 }
461
462 void
zio_add_child(zio_t * pio,zio_t * cio)463 zio_add_child(zio_t *pio, zio_t *cio)
464 {
465 zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_SLEEP);
466
467 /*
468 * Logical I/Os can have logical, gang, or vdev children.
469 * Gang I/Os can have gang or vdev children.
470 * Vdev I/Os can only have vdev children.
471 * The following ASSERT captures all of these constraints.
472 */
473 ASSERT(cio->io_child_type <= pio->io_child_type);
474
475 zl->zl_parent = pio;
476 zl->zl_child = cio;
477
478 mutex_enter(&cio->io_lock);
479 mutex_enter(&pio->io_lock);
480
481 ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0);
482
483 for (int w = 0; w < ZIO_WAIT_TYPES; w++)
484 pio->io_children[cio->io_child_type][w] += !cio->io_state[w];
485
486 list_insert_head(&pio->io_child_list, zl);
487 list_insert_head(&cio->io_parent_list, zl);
488
489 pio->io_child_count++;
490 cio->io_parent_count++;
491
492 mutex_exit(&pio->io_lock);
493 mutex_exit(&cio->io_lock);
494 }
495
496 static void
zio_remove_child(zio_t * pio,zio_t * cio,zio_link_t * zl)497 zio_remove_child(zio_t *pio, zio_t *cio, zio_link_t *zl)
498 {
499 ASSERT(zl->zl_parent == pio);
500 ASSERT(zl->zl_child == cio);
501
502 mutex_enter(&cio->io_lock);
503 mutex_enter(&pio->io_lock);
504
505 list_remove(&pio->io_child_list, zl);
506 list_remove(&cio->io_parent_list, zl);
507
508 pio->io_child_count--;
509 cio->io_parent_count--;
510
511 mutex_exit(&pio->io_lock);
512 mutex_exit(&cio->io_lock);
513
514 kmem_cache_free(zio_link_cache, zl);
515 }
516
517 static boolean_t
zio_wait_for_children(zio_t * zio,enum zio_child child,enum zio_wait_type wait)518 zio_wait_for_children(zio_t *zio, enum zio_child child, enum zio_wait_type wait)
519 {
520 uint64_t *countp = &zio->io_children[child][wait];
521 boolean_t waiting = B_FALSE;
522
523 mutex_enter(&zio->io_lock);
524 ASSERT(zio->io_stall == NULL);
525 if (*countp != 0) {
526 zio->io_stage >>= 1;
527 ASSERT3U(zio->io_stage, !=, ZIO_STAGE_OPEN);
528 zio->io_stall = countp;
529 waiting = B_TRUE;
530 }
531 mutex_exit(&zio->io_lock);
532
533 return (waiting);
534 }
535
536 static void
zio_notify_parent(zio_t * pio,zio_t * zio,enum zio_wait_type wait)537 zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait)
538 {
539 uint64_t *countp = &pio->io_children[zio->io_child_type][wait];
540 int *errorp = &pio->io_child_error[zio->io_child_type];
541
542 mutex_enter(&pio->io_lock);
543 if (zio->io_error && !(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE))
544 *errorp = zio_worst_error(*errorp, zio->io_error);
545 pio->io_reexecute |= zio->io_reexecute;
546 ASSERT3U(*countp, >, 0);
547
548 (*countp)--;
549
550 if (*countp == 0 && pio->io_stall == countp) {
551 zio_taskq_type_t type =
552 pio->io_stage < ZIO_STAGE_VDEV_IO_START ? ZIO_TASKQ_ISSUE :
553 ZIO_TASKQ_INTERRUPT;
554 pio->io_stall = NULL;
555 mutex_exit(&pio->io_lock);
556 /*
557 * Dispatch the parent zio in its own taskq so that
558 * the child can continue to make progress. This also
559 * prevents overflowing the stack when we have deeply nested
560 * parent-child relationships.
561 */
562 zio_taskq_dispatch(pio, type, B_FALSE);
563 } else {
564 mutex_exit(&pio->io_lock);
565 }
566 }
567
568 static void
zio_inherit_child_errors(zio_t * zio,enum zio_child c)569 zio_inherit_child_errors(zio_t *zio, enum zio_child c)
570 {
571 if (zio->io_child_error[c] != 0 && zio->io_error == 0)
572 zio->io_error = zio->io_child_error[c];
573 }
574
575 int
zio_timestamp_compare(const void * x1,const void * x2)576 zio_timestamp_compare(const void *x1, const void *x2)
577 {
578 const zio_t *z1 = x1;
579 const zio_t *z2 = x2;
580
581 if (z1->io_queued_timestamp < z2->io_queued_timestamp)
582 return (-1);
583 if (z1->io_queued_timestamp > z2->io_queued_timestamp)
584 return (1);
585
586 if (z1->io_bookmark.zb_objset < z2->io_bookmark.zb_objset)
587 return (-1);
588 if (z1->io_bookmark.zb_objset > z2->io_bookmark.zb_objset)
589 return (1);
590
591 if (z1->io_bookmark.zb_object < z2->io_bookmark.zb_object)
592 return (-1);
593 if (z1->io_bookmark.zb_object > z2->io_bookmark.zb_object)
594 return (1);
595
596 if (z1->io_bookmark.zb_level < z2->io_bookmark.zb_level)
597 return (-1);
598 if (z1->io_bookmark.zb_level > z2->io_bookmark.zb_level)
599 return (1);
600
601 if (z1->io_bookmark.zb_blkid < z2->io_bookmark.zb_blkid)
602 return (-1);
603 if (z1->io_bookmark.zb_blkid > z2->io_bookmark.zb_blkid)
604 return (1);
605
606 if (z1 < z2)
607 return (-1);
608 if (z1 > z2)
609 return (1);
610
611 return (0);
612 }
613
614 /*
615 * ==========================================================================
616 * Create the various types of I/O (read, write, free, etc)
617 * ==========================================================================
618 */
619 static zio_t *
zio_create(zio_t * pio,spa_t * spa,uint64_t txg,const blkptr_t * bp,void * data,uint64_t size,zio_done_func_t * done,void * private,zio_type_t type,zio_priority_t priority,enum zio_flag flags,vdev_t * vd,uint64_t offset,const zbookmark_phys_t * zb,enum zio_stage stage,enum zio_stage pipeline)620 zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
621 void *data, uint64_t size, zio_done_func_t *done, void *private,
622 zio_type_t type, zio_priority_t priority, enum zio_flag flags,
623 vdev_t *vd, uint64_t offset, const zbookmark_phys_t *zb,
624 enum zio_stage stage, enum zio_stage pipeline)
625 {
626 zio_t *zio;
627
628 ASSERT3U(type == ZIO_TYPE_FREE || size, <=, SPA_MAXBLOCKSIZE);
629 ASSERT(P2PHASE(size, SPA_MINBLOCKSIZE) == 0);
630 ASSERT(P2PHASE(offset, SPA_MINBLOCKSIZE) == 0);
631
632 ASSERT(!vd || spa_config_held(spa, SCL_STATE_ALL, RW_READER));
633 ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER));
634 ASSERT(vd || stage == ZIO_STAGE_OPEN);
635
636 zio = kmem_cache_alloc(zio_cache, KM_SLEEP);
637 bzero(zio, sizeof (zio_t));
638
639 mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL);
640 cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL);
641
642 list_create(&zio->io_parent_list, sizeof (zio_link_t),
643 offsetof(zio_link_t, zl_parent_node));
644 list_create(&zio->io_child_list, sizeof (zio_link_t),
645 offsetof(zio_link_t, zl_child_node));
646 metaslab_trace_init(&zio->io_alloc_list);
647
648 if (vd != NULL)
649 zio->io_child_type = ZIO_CHILD_VDEV;
650 else if (flags & ZIO_FLAG_GANG_CHILD)
651 zio->io_child_type = ZIO_CHILD_GANG;
652 else if (flags & ZIO_FLAG_DDT_CHILD)
653 zio->io_child_type = ZIO_CHILD_DDT;
654 else
655 zio->io_child_type = ZIO_CHILD_LOGICAL;
656
657 if (bp != NULL) {
658 zio->io_bp = (blkptr_t *)bp;
659 zio->io_bp_copy = *bp;
660 zio->io_bp_orig = *bp;
661 if (type != ZIO_TYPE_WRITE ||
662 zio->io_child_type == ZIO_CHILD_DDT)
663 zio->io_bp = &zio->io_bp_copy; /* so caller can free */
664 if (zio->io_child_type == ZIO_CHILD_LOGICAL)
665 zio->io_logical = zio;
666 if (zio->io_child_type > ZIO_CHILD_GANG && BP_IS_GANG(bp))
667 pipeline |= ZIO_GANG_STAGES;
668 }
669
670 zio->io_spa = spa;
671 zio->io_txg = txg;
672 zio->io_done = done;
673 zio->io_private = private;
674 zio->io_type = type;
675 zio->io_priority = priority;
676 zio->io_vd = vd;
677 zio->io_offset = offset;
678 zio->io_orig_data = zio->io_data = data;
679 zio->io_orig_size = zio->io_size = size;
680 zio->io_orig_flags = zio->io_flags = flags;
681 zio->io_orig_stage = zio->io_stage = stage;
682 zio->io_orig_pipeline = zio->io_pipeline = pipeline;
683 zio->io_pipeline_trace = ZIO_STAGE_OPEN;
684
685 zio->io_state[ZIO_WAIT_READY] = (stage >= ZIO_STAGE_READY);
686 zio->io_state[ZIO_WAIT_DONE] = (stage >= ZIO_STAGE_DONE);
687
688 if (zb != NULL)
689 zio->io_bookmark = *zb;
690
691 if (pio != NULL) {
692 if (zio->io_logical == NULL)
693 zio->io_logical = pio->io_logical;
694 if (zio->io_child_type == ZIO_CHILD_GANG)
695 zio->io_gang_leader = pio->io_gang_leader;
696 zio_add_child(pio, zio);
697 }
698
699 return (zio);
700 }
701
702 static void
zio_destroy(zio_t * zio)703 zio_destroy(zio_t *zio)
704 {
705 metaslab_trace_fini(&zio->io_alloc_list);
706 list_destroy(&zio->io_parent_list);
707 list_destroy(&zio->io_child_list);
708 mutex_destroy(&zio->io_lock);
709 cv_destroy(&zio->io_cv);
710 kmem_cache_free(zio_cache, zio);
711 }
712
713 zio_t *
zio_null(zio_t * pio,spa_t * spa,vdev_t * vd,zio_done_func_t * done,void * private,enum zio_flag flags)714 zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done,
715 void *private, enum zio_flag flags)
716 {
717 zio_t *zio;
718
719 zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private,
720 ZIO_TYPE_NULL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL,
721 ZIO_STAGE_OPEN, ZIO_INTERLOCK_PIPELINE);
722
723 return (zio);
724 }
725
726 zio_t *
zio_root(spa_t * spa,zio_done_func_t * done,void * private,enum zio_flag flags)727 zio_root(spa_t *spa, zio_done_func_t *done, void *private, enum zio_flag flags)
728 {
729 return (zio_null(NULL, spa, NULL, done, private, flags));
730 }
731
732 void
zfs_blkptr_verify(spa_t * spa,const blkptr_t * bp)733 zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp)
734 {
735 if (!DMU_OT_IS_VALID(BP_GET_TYPE(bp))) {
736 zfs_panic_recover("blkptr at %p has invalid TYPE %llu",
737 bp, (longlong_t)BP_GET_TYPE(bp));
738 }
739 if (BP_GET_CHECKSUM(bp) >= ZIO_CHECKSUM_FUNCTIONS ||
740 BP_GET_CHECKSUM(bp) <= ZIO_CHECKSUM_ON) {
741 zfs_panic_recover("blkptr at %p has invalid CHECKSUM %llu",
742 bp, (longlong_t)BP_GET_CHECKSUM(bp));
743 }
744 if (BP_GET_COMPRESS(bp) >= ZIO_COMPRESS_FUNCTIONS ||
745 BP_GET_COMPRESS(bp) <= ZIO_COMPRESS_ON) {
746 zfs_panic_recover("blkptr at %p has invalid COMPRESS %llu",
747 bp, (longlong_t)BP_GET_COMPRESS(bp));
748 }
749 if (BP_GET_LSIZE(bp) > SPA_MAXBLOCKSIZE) {
750 zfs_panic_recover("blkptr at %p has invalid LSIZE %llu",
751 bp, (longlong_t)BP_GET_LSIZE(bp));
752 }
753 if (BP_GET_PSIZE(bp) > SPA_MAXBLOCKSIZE) {
754 zfs_panic_recover("blkptr at %p has invalid PSIZE %llu",
755 bp, (longlong_t)BP_GET_PSIZE(bp));
756 }
757
758 if (BP_IS_EMBEDDED(bp)) {
759 if (BPE_GET_ETYPE(bp) > NUM_BP_EMBEDDED_TYPES) {
760 zfs_panic_recover("blkptr at %p has invalid ETYPE %llu",
761 bp, (longlong_t)BPE_GET_ETYPE(bp));
762 }
763 }
764
765 /*
766 * Pool-specific checks.
767 *
768 * Note: it would be nice to verify that the blk_birth and
769 * BP_PHYSICAL_BIRTH() are not too large. However, spa_freeze()
770 * allows the birth time of log blocks (and dmu_sync()-ed blocks
771 * that are in the log) to be arbitrarily large.
772 */
773 for (int i = 0; i < BP_GET_NDVAS(bp); i++) {
774 uint64_t vdevid = DVA_GET_VDEV(&bp->blk_dva[i]);
775 if (vdevid >= spa->spa_root_vdev->vdev_children) {
776 zfs_panic_recover("blkptr at %p DVA %u has invalid "
777 "VDEV %llu",
778 bp, i, (longlong_t)vdevid);
779 continue;
780 }
781 vdev_t *vd = spa->spa_root_vdev->vdev_child[vdevid];
782 if (vd == NULL) {
783 zfs_panic_recover("blkptr at %p DVA %u has invalid "
784 "VDEV %llu",
785 bp, i, (longlong_t)vdevid);
786 continue;
787 }
788 if (vd->vdev_ops == &vdev_hole_ops) {
789 zfs_panic_recover("blkptr at %p DVA %u has hole "
790 "VDEV %llu",
791 bp, i, (longlong_t)vdevid);
792 continue;
793 }
794 if (vd->vdev_ops == &vdev_missing_ops) {
795 /*
796 * "missing" vdevs are valid during import, but we
797 * don't have their detailed info (e.g. asize), so
798 * we can't perform any more checks on them.
799 */
800 continue;
801 }
802 uint64_t offset = DVA_GET_OFFSET(&bp->blk_dva[i]);
803 uint64_t asize = DVA_GET_ASIZE(&bp->blk_dva[i]);
804 if (BP_IS_GANG(bp))
805 asize = vdev_psize_to_asize(vd, SPA_GANGBLOCKSIZE);
806 if (offset + asize > vd->vdev_asize) {
807 zfs_panic_recover("blkptr at %p DVA %u has invalid "
808 "OFFSET %llu",
809 bp, i, (longlong_t)offset);
810 }
811 }
812 }
813
814 zio_t *
zio_read(zio_t * pio,spa_t * spa,const blkptr_t * bp,void * data,uint64_t size,zio_done_func_t * done,void * private,zio_priority_t priority,enum zio_flag flags,const zbookmark_phys_t * zb)815 zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
816 void *data, uint64_t size, zio_done_func_t *done, void *private,
817 zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb)
818 {
819 zio_t *zio;
820
821 zfs_blkptr_verify(spa, bp);
822
823 zio = zio_create(pio, spa, BP_PHYSICAL_BIRTH(bp), bp,
824 data, size, done, private,
825 ZIO_TYPE_READ, priority, flags, NULL, 0, zb,
826 ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ?
827 ZIO_DDT_CHILD_READ_PIPELINE : ZIO_READ_PIPELINE);
828
829 return (zio);
830 }
831
832 zio_t *
zio_write(zio_t * pio,spa_t * spa,uint64_t txg,blkptr_t * bp,void * data,uint64_t size,const zio_prop_t * zp,zio_done_func_t * ready,zio_done_func_t * children_ready,zio_done_func_t * physdone,zio_done_func_t * done,void * private,zio_priority_t priority,enum zio_flag flags,const zbookmark_phys_t * zb)833 zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
834 void *data, uint64_t size, const zio_prop_t *zp,
835 zio_done_func_t *ready, zio_done_func_t *children_ready,
836 zio_done_func_t *physdone, zio_done_func_t *done,
837 void *private, zio_priority_t priority, enum zio_flag flags,
838 const zbookmark_phys_t *zb)
839 {
840 zio_t *zio;
841
842 ASSERT(zp->zp_checksum >= ZIO_CHECKSUM_OFF &&
843 zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS &&
844 zp->zp_compress >= ZIO_COMPRESS_OFF &&
845 zp->zp_compress < ZIO_COMPRESS_FUNCTIONS &&
846 DMU_OT_IS_VALID(zp->zp_type) &&
847 zp->zp_level < 32 &&
848 zp->zp_copies > 0 &&
849 zp->zp_copies <= spa_max_replication(spa));
850
851 zio = zio_create(pio, spa, txg, bp, data, size, done, private,
852 ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb,
853 ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ?
854 ZIO_DDT_CHILD_WRITE_PIPELINE : ZIO_WRITE_PIPELINE);
855
856 zio->io_ready = ready;
857 zio->io_children_ready = children_ready;
858 zio->io_physdone = physdone;
859 zio->io_prop = *zp;
860
861 /*
862 * Data can be NULL if we are going to call zio_write_override() to
863 * provide the already-allocated BP. But we may need the data to
864 * verify a dedup hit (if requested). In this case, don't try to
865 * dedup (just take the already-allocated BP verbatim).
866 */
867 if (data == NULL && zio->io_prop.zp_dedup_verify) {
868 zio->io_prop.zp_dedup = zio->io_prop.zp_dedup_verify = B_FALSE;
869 }
870
871 return (zio);
872 }
873
874 zio_t *
zio_rewrite(zio_t * pio,spa_t * spa,uint64_t txg,blkptr_t * bp,void * data,uint64_t size,zio_done_func_t * done,void * private,zio_priority_t priority,enum zio_flag flags,zbookmark_phys_t * zb)875 zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data,
876 uint64_t size, zio_done_func_t *done, void *private,
877 zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb)
878 {
879 zio_t *zio;
880
881 zio = zio_create(pio, spa, txg, bp, data, size, done, private,
882 ZIO_TYPE_WRITE, priority, flags | ZIO_FLAG_IO_REWRITE, NULL, 0, zb,
883 ZIO_STAGE_OPEN, ZIO_REWRITE_PIPELINE);
884
885 return (zio);
886 }
887
888 void
zio_write_override(zio_t * zio,blkptr_t * bp,int copies,boolean_t nopwrite)889 zio_write_override(zio_t *zio, blkptr_t *bp, int copies, boolean_t nopwrite)
890 {
891 ASSERT(zio->io_type == ZIO_TYPE_WRITE);
892 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
893 ASSERT(zio->io_stage == ZIO_STAGE_OPEN);
894 ASSERT(zio->io_txg == spa_syncing_txg(zio->io_spa));
895
896 /*
897 * We must reset the io_prop to match the values that existed
898 * when the bp was first written by dmu_sync() keeping in mind
899 * that nopwrite and dedup are mutually exclusive.
900 */
901 zio->io_prop.zp_dedup = nopwrite ? B_FALSE : zio->io_prop.zp_dedup;
902 zio->io_prop.zp_nopwrite = nopwrite;
903 zio->io_prop.zp_copies = copies;
904 zio->io_bp_override = bp;
905 }
906
907 void
zio_free(spa_t * spa,uint64_t txg,const blkptr_t * bp)908 zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp)
909 {
910
911 /*
912 * The check for EMBEDDED is a performance optimization. We
913 * process the free here (by ignoring it) rather than
914 * putting it on the list and then processing it in zio_free_sync().
915 */
916 if (BP_IS_EMBEDDED(bp))
917 return;
918 metaslab_check_free(spa, bp);
919
920 /*
921 * Frees that are for the currently-syncing txg, are not going to be
922 * deferred, and which will not need to do a read (i.e. not GANG or
923 * DEDUP), can be processed immediately. Otherwise, put them on the
924 * in-memory list for later processing.
925 */
926 if (zfs_trim_enabled || BP_IS_GANG(bp) || BP_GET_DEDUP(bp) ||
927 txg != spa->spa_syncing_txg ||
928 spa_sync_pass(spa) >= zfs_sync_pass_deferred_free) {
929 bplist_append(&spa->spa_free_bplist[txg & TXG_MASK], bp);
930 } else {
931 VERIFY0(zio_wait(zio_free_sync(NULL, spa, txg, bp,
932 BP_GET_PSIZE(bp), 0)));
933 }
934 }
935
936 zio_t *
zio_free_sync(zio_t * pio,spa_t * spa,uint64_t txg,const blkptr_t * bp,uint64_t size,enum zio_flag flags)937 zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
938 uint64_t size, enum zio_flag flags)
939 {
940 zio_t *zio;
941 enum zio_stage stage = ZIO_FREE_PIPELINE;
942
943 ASSERT(!BP_IS_HOLE(bp));
944 ASSERT(spa_syncing_txg(spa) == txg);
945 ASSERT(spa_sync_pass(spa) < zfs_sync_pass_deferred_free);
946
947 if (BP_IS_EMBEDDED(bp))
948 return (zio_null(pio, spa, NULL, NULL, NULL, 0));
949
950 metaslab_check_free(spa, bp);
951 arc_freed(spa, bp);
952
953 if (zfs_trim_enabled)
954 stage |= ZIO_STAGE_ISSUE_ASYNC | ZIO_STAGE_VDEV_IO_START |
955 ZIO_STAGE_VDEV_IO_ASSESS;
956 /*
957 * GANG and DEDUP blocks can induce a read (for the gang block header,
958 * or the DDT), so issue them asynchronously so that this thread is
959 * not tied up.
960 */
961 else if (BP_IS_GANG(bp) || BP_GET_DEDUP(bp))
962 stage |= ZIO_STAGE_ISSUE_ASYNC;
963
964 flags |= ZIO_FLAG_DONT_QUEUE;
965
966 zio = zio_create(pio, spa, txg, bp, NULL, size,
967 NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_NOW, flags,
968 NULL, 0, NULL, ZIO_STAGE_OPEN, stage);
969
970 return (zio);
971 }
972
973 zio_t *
zio_claim(zio_t * pio,spa_t * spa,uint64_t txg,const blkptr_t * bp,zio_done_func_t * done,void * private,enum zio_flag flags)974 zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp,
975 zio_done_func_t *done, void *private, enum zio_flag flags)
976 {
977 zio_t *zio;
978
979 dprintf_bp(bp, "claiming in txg %llu", txg);
980
981 if (BP_IS_EMBEDDED(bp))
982 return (zio_null(pio, spa, NULL, NULL, NULL, 0));
983
984 /*
985 * A claim is an allocation of a specific block. Claims are needed
986 * to support immediate writes in the intent log. The issue is that
987 * immediate writes contain committed data, but in a txg that was
988 * *not* committed. Upon opening the pool after an unclean shutdown,
989 * the intent log claims all blocks that contain immediate write data
990 * so that the SPA knows they're in use.
991 *
992 * All claims *must* be resolved in the first txg -- before the SPA
993 * starts allocating blocks -- so that nothing is allocated twice.
994 * If txg == 0 we just verify that the block is claimable.
995 */
996 ASSERT3U(spa->spa_uberblock.ub_rootbp.blk_birth, <, spa_first_txg(spa));
997 ASSERT(txg == spa_first_txg(spa) || txg == 0);
998 ASSERT(!BP_GET_DEDUP(bp) || !spa_writeable(spa)); /* zdb(1M) */
999
1000 zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp),
1001 done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags,
1002 NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE);
1003 ASSERT0(zio->io_queued_timestamp);
1004
1005 return (zio);
1006 }
1007
1008 zio_t *
zio_ioctl(zio_t * pio,spa_t * spa,vdev_t * vd,int cmd,uint64_t offset,uint64_t size,zio_done_func_t * done,void * private,zio_priority_t priority,enum zio_flag flags)1009 zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, uint64_t offset,
1010 uint64_t size, zio_done_func_t *done, void *private,
1011 zio_priority_t priority, enum zio_flag flags)
1012 {
1013 zio_t *zio;
1014 int c;
1015
1016 if (vd->vdev_children == 0) {
1017 zio = zio_create(pio, spa, 0, NULL, NULL, size, done, private,
1018 ZIO_TYPE_IOCTL, priority, flags, vd, offset, NULL,
1019 ZIO_STAGE_OPEN, ZIO_IOCTL_PIPELINE);
1020
1021 zio->io_cmd = cmd;
1022 } else {
1023 zio = zio_null(pio, spa, NULL, NULL, NULL, flags);
1024
1025 for (c = 0; c < vd->vdev_children; c++)
1026 zio_nowait(zio_ioctl(zio, spa, vd->vdev_child[c], cmd,
1027 offset, size, done, private, priority, flags));
1028 }
1029
1030 return (zio);
1031 }
1032
1033 zio_t *
zio_read_phys(zio_t * pio,vdev_t * vd,uint64_t offset,uint64_t size,void * data,int checksum,zio_done_func_t * done,void * private,zio_priority_t priority,enum zio_flag flags,boolean_t labels)1034 zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
1035 void *data, int checksum, zio_done_func_t *done, void *private,
1036 zio_priority_t priority, enum zio_flag flags, boolean_t labels)
1037 {
1038 zio_t *zio;
1039
1040 ASSERT(vd->vdev_children == 0);
1041 ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE ||
1042 offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE);
1043 ASSERT3U(offset + size, <=, vd->vdev_psize);
1044
1045 zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private,
1046 ZIO_TYPE_READ, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset,
1047 NULL, ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE);
1048
1049 zio->io_prop.zp_checksum = checksum;
1050
1051 return (zio);
1052 }
1053
1054 zio_t *
zio_write_phys(zio_t * pio,vdev_t * vd,uint64_t offset,uint64_t size,void * data,int checksum,zio_done_func_t * done,void * private,zio_priority_t priority,enum zio_flag flags,boolean_t labels)1055 zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
1056 void *data, int checksum, zio_done_func_t *done, void *private,
1057 zio_priority_t priority, enum zio_flag flags, boolean_t labels)
1058 {
1059 zio_t *zio;
1060
1061 ASSERT(vd->vdev_children == 0);
1062 ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE ||
1063 offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE);
1064 ASSERT3U(offset + size, <=, vd->vdev_psize);
1065
1066 zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private,
1067 ZIO_TYPE_WRITE, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset,
1068 NULL, ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE);
1069
1070 zio->io_prop.zp_checksum = checksum;
1071
1072 if (zio_checksum_table[checksum].ci_flags & ZCHECKSUM_FLAG_EMBEDDED) {
1073 /*
1074 * zec checksums are necessarily destructive -- they modify
1075 * the end of the write buffer to hold the verifier/checksum.
1076 * Therefore, we must make a local copy in case the data is
1077 * being written to multiple places in parallel.
1078 */
1079 void *wbuf = zio_buf_alloc(size);
1080 bcopy(data, wbuf, size);
1081 zio_push_transform(zio, wbuf, size, size, NULL);
1082 }
1083
1084 return (zio);
1085 }
1086
1087 /*
1088 * Create a child I/O to do some work for us.
1089 */
1090 zio_t *
zio_vdev_child_io(zio_t * pio,blkptr_t * bp,vdev_t * vd,uint64_t offset,void * data,uint64_t size,int type,zio_priority_t priority,enum zio_flag flags,zio_done_func_t * done,void * private)1091 zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset,
1092 void *data, uint64_t size, int type, zio_priority_t priority,
1093 enum zio_flag flags, zio_done_func_t *done, void *private)
1094 {
1095 enum zio_stage pipeline = ZIO_VDEV_CHILD_PIPELINE;
1096 zio_t *zio;
1097
1098 ASSERT(vd->vdev_parent ==
1099 (pio->io_vd ? pio->io_vd : pio->io_spa->spa_root_vdev));
1100
1101 if (type == ZIO_TYPE_READ && bp != NULL) {
1102 /*
1103 * If we have the bp, then the child should perform the
1104 * checksum and the parent need not. This pushes error
1105 * detection as close to the leaves as possible and
1106 * eliminates redundant checksums in the interior nodes.
1107 */
1108 pipeline |= ZIO_STAGE_CHECKSUM_VERIFY;
1109 pio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY;
1110 }
1111
1112 /* Not all IO types require vdev io done stage e.g. free */
1113 if (!(pio->io_pipeline & ZIO_STAGE_VDEV_IO_DONE))
1114 pipeline &= ~ZIO_STAGE_VDEV_IO_DONE;
1115
1116 if (vd->vdev_children == 0)
1117 offset += VDEV_LABEL_START_SIZE;
1118
1119 flags |= ZIO_VDEV_CHILD_FLAGS(pio) | ZIO_FLAG_DONT_PROPAGATE;
1120
1121 /*
1122 * If we've decided to do a repair, the write is not speculative --
1123 * even if the original read was.
1124 */
1125 if (flags & ZIO_FLAG_IO_REPAIR)
1126 flags &= ~ZIO_FLAG_SPECULATIVE;
1127
1128 /*
1129 * If we're creating a child I/O that is not associated with a
1130 * top-level vdev, then the child zio is not an allocating I/O.
1131 * If this is a retried I/O then we ignore it since we will
1132 * have already processed the original allocating I/O.
1133 */
1134 if (flags & ZIO_FLAG_IO_ALLOCATING &&
1135 (vd != vd->vdev_top || (flags & ZIO_FLAG_IO_RETRY))) {
1136 metaslab_class_t *mc = spa_normal_class(pio->io_spa);
1137
1138 ASSERT(mc->mc_alloc_throttle_enabled);
1139 ASSERT(type == ZIO_TYPE_WRITE);
1140 ASSERT(priority == ZIO_PRIORITY_ASYNC_WRITE);
1141 ASSERT(!(flags & ZIO_FLAG_IO_REPAIR));
1142 ASSERT(!(pio->io_flags & ZIO_FLAG_IO_REWRITE) ||
1143 pio->io_child_type == ZIO_CHILD_GANG);
1144
1145 flags &= ~ZIO_FLAG_IO_ALLOCATING;
1146 }
1147
1148 zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size,
1149 done, private, type, priority, flags, vd, offset, &pio->io_bookmark,
1150 ZIO_STAGE_VDEV_IO_START >> 1, pipeline);
1151 ASSERT3U(zio->io_child_type, ==, ZIO_CHILD_VDEV);
1152
1153 zio->io_physdone = pio->io_physdone;
1154 if (vd->vdev_ops->vdev_op_leaf && zio->io_logical != NULL)
1155 zio->io_logical->io_phys_children++;
1156
1157 return (zio);
1158 }
1159
1160 zio_t *
zio_vdev_delegated_io(vdev_t * vd,uint64_t offset,void * data,uint64_t size,int type,zio_priority_t priority,enum zio_flag flags,zio_done_func_t * done,void * private)1161 zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, void *data, uint64_t size,
1162 int type, zio_priority_t priority, enum zio_flag flags,
1163 zio_done_func_t *done, void *private)
1164 {
1165 zio_t *zio;
1166
1167 ASSERT(vd->vdev_ops->vdev_op_leaf);
1168
1169 zio = zio_create(NULL, vd->vdev_spa, 0, NULL,
1170 data, size, done, private, type, priority,
1171 flags | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY | ZIO_FLAG_DELEGATED,
1172 vd, offset, NULL,
1173 ZIO_STAGE_VDEV_IO_START >> 1, ZIO_VDEV_CHILD_PIPELINE);
1174
1175 return (zio);
1176 }
1177
1178 void
zio_flush(zio_t * zio,vdev_t * vd)1179 zio_flush(zio_t *zio, vdev_t *vd)
1180 {
1181 zio_nowait(zio_ioctl(zio, zio->io_spa, vd, DKIOCFLUSHWRITECACHE, 0, 0,
1182 NULL, NULL, ZIO_PRIORITY_NOW,
1183 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY));
1184 }
1185
1186 zio_t *
zio_trim(zio_t * zio,spa_t * spa,vdev_t * vd,uint64_t offset,uint64_t size)1187 zio_trim(zio_t *zio, spa_t *spa, vdev_t *vd, uint64_t offset, uint64_t size)
1188 {
1189
1190 ASSERT(vd->vdev_ops->vdev_op_leaf);
1191
1192 return (zio_create(zio, spa, 0, NULL, NULL, size, NULL, NULL,
1193 ZIO_TYPE_FREE, ZIO_PRIORITY_TRIM, ZIO_FLAG_DONT_AGGREGATE |
1194 ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY,
1195 vd, offset, NULL, ZIO_STAGE_OPEN, ZIO_FREE_PHYS_PIPELINE));
1196 }
1197
1198 void
zio_shrink(zio_t * zio,uint64_t size)1199 zio_shrink(zio_t *zio, uint64_t size)
1200 {
1201 ASSERT(zio->io_executor == NULL);
1202 ASSERT(zio->io_orig_size == zio->io_size);
1203 ASSERT(size <= zio->io_size);
1204
1205 /*
1206 * We don't shrink for raidz because of problems with the
1207 * reconstruction when reading back less than the block size.
1208 * Note, BP_IS_RAIDZ() assumes no compression.
1209 */
1210 ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF);
1211 if (!BP_IS_RAIDZ(zio->io_bp))
1212 zio->io_orig_size = zio->io_size = size;
1213 }
1214
1215 /*
1216 * ==========================================================================
1217 * Prepare to read and write logical blocks
1218 * ==========================================================================
1219 */
1220
1221 static int
zio_read_bp_init(zio_t * zio)1222 zio_read_bp_init(zio_t *zio)
1223 {
1224 blkptr_t *bp = zio->io_bp;
1225
1226 if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF &&
1227 zio->io_child_type == ZIO_CHILD_LOGICAL &&
1228 !(zio->io_flags & ZIO_FLAG_RAW)) {
1229 uint64_t psize =
1230 BP_IS_EMBEDDED(bp) ? BPE_GET_PSIZE(bp) : BP_GET_PSIZE(bp);
1231 void *cbuf = zio_buf_alloc(psize);
1232
1233 zio_push_transform(zio, cbuf, psize, psize, zio_decompress);
1234 }
1235
1236 if (BP_IS_EMBEDDED(bp) && BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA) {
1237 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
1238 decode_embedded_bp_compressed(bp, zio->io_data);
1239 } else {
1240 ASSERT(!BP_IS_EMBEDDED(bp));
1241 }
1242
1243 if (!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) && BP_GET_LEVEL(bp) == 0)
1244 zio->io_flags |= ZIO_FLAG_DONT_CACHE;
1245
1246 if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP)
1247 zio->io_flags |= ZIO_FLAG_DONT_CACHE;
1248
1249 if (BP_GET_DEDUP(bp) && zio->io_child_type == ZIO_CHILD_LOGICAL)
1250 zio->io_pipeline = ZIO_DDT_READ_PIPELINE;
1251
1252 return (ZIO_PIPELINE_CONTINUE);
1253 }
1254
1255 static int
zio_write_bp_init(zio_t * zio)1256 zio_write_bp_init(zio_t *zio)
1257 {
1258 if (!IO_IS_ALLOCATING(zio))
1259 return (ZIO_PIPELINE_CONTINUE);
1260
1261 ASSERT(zio->io_child_type != ZIO_CHILD_DDT);
1262
1263 if (zio->io_bp_override) {
1264 blkptr_t *bp = zio->io_bp;
1265 zio_prop_t *zp = &zio->io_prop;
1266
1267 ASSERT(bp->blk_birth != zio->io_txg);
1268 ASSERT(BP_GET_DEDUP(zio->io_bp_override) == 0);
1269
1270 *bp = *zio->io_bp_override;
1271 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
1272
1273 if (BP_IS_EMBEDDED(bp))
1274 return (ZIO_PIPELINE_CONTINUE);
1275
1276 /*
1277 * If we've been overridden and nopwrite is set then
1278 * set the flag accordingly to indicate that a nopwrite
1279 * has already occurred.
1280 */
1281 if (!BP_IS_HOLE(bp) && zp->zp_nopwrite) {
1282 ASSERT(!zp->zp_dedup);
1283 ASSERT3U(BP_GET_CHECKSUM(bp), ==, zp->zp_checksum);
1284 zio->io_flags |= ZIO_FLAG_NOPWRITE;
1285 return (ZIO_PIPELINE_CONTINUE);
1286 }
1287
1288 ASSERT(!zp->zp_nopwrite);
1289
1290 if (BP_IS_HOLE(bp) || !zp->zp_dedup)
1291 return (ZIO_PIPELINE_CONTINUE);
1292
1293 ASSERT((zio_checksum_table[zp->zp_checksum].ci_flags &
1294 ZCHECKSUM_FLAG_DEDUP) || zp->zp_dedup_verify);
1295
1296 if (BP_GET_CHECKSUM(bp) == zp->zp_checksum) {
1297 BP_SET_DEDUP(bp, 1);
1298 zio->io_pipeline |= ZIO_STAGE_DDT_WRITE;
1299 return (ZIO_PIPELINE_CONTINUE);
1300 }
1301
1302 /*
1303 * We were unable to handle this as an override bp, treat
1304 * it as a regular write I/O.
1305 */
1306 zio->io_bp_override = NULL;
1307 *bp = zio->io_bp_orig;
1308 zio->io_pipeline = zio->io_orig_pipeline;
1309 }
1310
1311 return (ZIO_PIPELINE_CONTINUE);
1312 }
1313
1314 static int
zio_write_compress(zio_t * zio)1315 zio_write_compress(zio_t *zio)
1316 {
1317 spa_t *spa = zio->io_spa;
1318 zio_prop_t *zp = &zio->io_prop;
1319 enum zio_compress compress = zp->zp_compress;
1320 blkptr_t *bp = zio->io_bp;
1321 uint64_t lsize = zio->io_size;
1322 uint64_t psize = lsize;
1323 int pass = 1;
1324
1325 /*
1326 * If our children haven't all reached the ready stage,
1327 * wait for them and then repeat this pipeline stage.
1328 */
1329 if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) ||
1330 zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_READY))
1331 return (ZIO_PIPELINE_STOP);
1332
1333 if (!IO_IS_ALLOCATING(zio))
1334 return (ZIO_PIPELINE_CONTINUE);
1335
1336 if (zio->io_children_ready != NULL) {
1337 /*
1338 * Now that all our children are ready, run the callback
1339 * associated with this zio in case it wants to modify the
1340 * data to be written.
1341 */
1342 ASSERT3U(zp->zp_level, >, 0);
1343 zio->io_children_ready(zio);
1344 }
1345
1346 ASSERT(zio->io_child_type != ZIO_CHILD_DDT);
1347 ASSERT(zio->io_bp_override == NULL);
1348
1349 if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg) {
1350 /*
1351 * We're rewriting an existing block, which means we're
1352 * working on behalf of spa_sync(). For spa_sync() to
1353 * converge, it must eventually be the case that we don't
1354 * have to allocate new blocks. But compression changes
1355 * the blocksize, which forces a reallocate, and makes
1356 * convergence take longer. Therefore, after the first
1357 * few passes, stop compressing to ensure convergence.
1358 */
1359 pass = spa_sync_pass(spa);
1360
1361 ASSERT(zio->io_txg == spa_syncing_txg(spa));
1362 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
1363 ASSERT(!BP_GET_DEDUP(bp));
1364
1365 if (pass >= zfs_sync_pass_dont_compress)
1366 compress = ZIO_COMPRESS_OFF;
1367
1368 /* Make sure someone doesn't change their mind on overwrites */
1369 ASSERT(BP_IS_EMBEDDED(bp) || MIN(zp->zp_copies + BP_IS_GANG(bp),
1370 spa_max_replication(spa)) == BP_GET_NDVAS(bp));
1371 }
1372
1373 if (compress != ZIO_COMPRESS_OFF) {
1374 void *cbuf = zio_buf_alloc(lsize);
1375 psize = zio_compress_data(compress, zio->io_data, cbuf, lsize);
1376 if (psize == 0 || psize == lsize) {
1377 compress = ZIO_COMPRESS_OFF;
1378 zio_buf_free(cbuf, lsize);
1379 } else if (!zp->zp_dedup && psize <= BPE_PAYLOAD_SIZE &&
1380 zp->zp_level == 0 && !DMU_OT_HAS_FILL(zp->zp_type) &&
1381 spa_feature_is_enabled(spa, SPA_FEATURE_EMBEDDED_DATA)) {
1382 encode_embedded_bp_compressed(bp,
1383 cbuf, compress, lsize, psize);
1384 BPE_SET_ETYPE(bp, BP_EMBEDDED_TYPE_DATA);
1385 BP_SET_TYPE(bp, zio->io_prop.zp_type);
1386 BP_SET_LEVEL(bp, zio->io_prop.zp_level);
1387 zio_buf_free(cbuf, lsize);
1388 bp->blk_birth = zio->io_txg;
1389 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
1390 ASSERT(spa_feature_is_active(spa,
1391 SPA_FEATURE_EMBEDDED_DATA));
1392 return (ZIO_PIPELINE_CONTINUE);
1393 } else {
1394 /*
1395 * Round up compressed size up to the ashift
1396 * of the smallest-ashift device, and zero the tail.
1397 * This ensures that the compressed size of the BP
1398 * (and thus compressratio property) are correct,
1399 * in that we charge for the padding used to fill out
1400 * the last sector.
1401 */
1402 ASSERT3U(spa->spa_min_ashift, >=, SPA_MINBLOCKSHIFT);
1403 size_t rounded = (size_t)P2ROUNDUP(psize,
1404 1ULL << spa->spa_min_ashift);
1405 if (rounded >= lsize) {
1406 compress = ZIO_COMPRESS_OFF;
1407 zio_buf_free(cbuf, lsize);
1408 psize = lsize;
1409 } else {
1410 bzero((char *)cbuf + psize, rounded - psize);
1411 psize = rounded;
1412 zio_push_transform(zio, cbuf,
1413 psize, lsize, NULL);
1414 }
1415 }
1416
1417 /*
1418 * We were unable to handle this as an override bp, treat
1419 * it as a regular write I/O.
1420 */
1421 zio->io_bp_override = NULL;
1422 *bp = zio->io_bp_orig;
1423 zio->io_pipeline = zio->io_orig_pipeline;
1424 }
1425
1426 /*
1427 * The final pass of spa_sync() must be all rewrites, but the first
1428 * few passes offer a trade-off: allocating blocks defers convergence,
1429 * but newly allocated blocks are sequential, so they can be written
1430 * to disk faster. Therefore, we allow the first few passes of
1431 * spa_sync() to allocate new blocks, but force rewrites after that.
1432 * There should only be a handful of blocks after pass 1 in any case.
1433 */
1434 if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg &&
1435 BP_GET_PSIZE(bp) == psize &&
1436 pass >= zfs_sync_pass_rewrite) {
1437 ASSERT(psize != 0);
1438 enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES;
1439 zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages;
1440 zio->io_flags |= ZIO_FLAG_IO_REWRITE;
1441 } else {
1442 BP_ZERO(bp);
1443 zio->io_pipeline = ZIO_WRITE_PIPELINE;
1444 }
1445
1446 if (psize == 0) {
1447 if (zio->io_bp_orig.blk_birth != 0 &&
1448 spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) {
1449 BP_SET_LSIZE(bp, lsize);
1450 BP_SET_TYPE(bp, zp->zp_type);
1451 BP_SET_LEVEL(bp, zp->zp_level);
1452 BP_SET_BIRTH(bp, zio->io_txg, 0);
1453 }
1454 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
1455 } else {
1456 ASSERT(zp->zp_checksum != ZIO_CHECKSUM_GANG_HEADER);
1457 BP_SET_LSIZE(bp, lsize);
1458 BP_SET_TYPE(bp, zp->zp_type);
1459 BP_SET_LEVEL(bp, zp->zp_level);
1460 BP_SET_PSIZE(bp, psize);
1461 BP_SET_COMPRESS(bp, compress);
1462 BP_SET_CHECKSUM(bp, zp->zp_checksum);
1463 BP_SET_DEDUP(bp, zp->zp_dedup);
1464 BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
1465 if (zp->zp_dedup) {
1466 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
1467 ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE));
1468 zio->io_pipeline = ZIO_DDT_WRITE_PIPELINE;
1469 }
1470 if (zp->zp_nopwrite) {
1471 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
1472 ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE));
1473 zio->io_pipeline |= ZIO_STAGE_NOP_WRITE;
1474 }
1475 }
1476 return (ZIO_PIPELINE_CONTINUE);
1477 }
1478
1479 static int
zio_free_bp_init(zio_t * zio)1480 zio_free_bp_init(zio_t *zio)
1481 {
1482 blkptr_t *bp = zio->io_bp;
1483
1484 if (zio->io_child_type == ZIO_CHILD_LOGICAL) {
1485 if (BP_GET_DEDUP(bp))
1486 zio->io_pipeline = ZIO_DDT_FREE_PIPELINE;
1487 }
1488
1489 return (ZIO_PIPELINE_CONTINUE);
1490 }
1491
1492 /*
1493 * ==========================================================================
1494 * Execute the I/O pipeline
1495 * ==========================================================================
1496 */
1497
1498 static void
zio_taskq_dispatch(zio_t * zio,zio_taskq_type_t q,boolean_t cutinline)1499 zio_taskq_dispatch(zio_t *zio, zio_taskq_type_t q, boolean_t cutinline)
1500 {
1501 spa_t *spa = zio->io_spa;
1502 zio_type_t t = zio->io_type;
1503 int flags = (cutinline ? TQ_FRONT : 0);
1504
1505 ASSERT(q == ZIO_TASKQ_ISSUE || q == ZIO_TASKQ_INTERRUPT);
1506
1507 /*
1508 * If we're a config writer or a probe, the normal issue and
1509 * interrupt threads may all be blocked waiting for the config lock.
1510 * In this case, select the otherwise-unused taskq for ZIO_TYPE_NULL.
1511 */
1512 if (zio->io_flags & (ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_PROBE))
1513 t = ZIO_TYPE_NULL;
1514
1515 /*
1516 * A similar issue exists for the L2ARC write thread until L2ARC 2.0.
1517 */
1518 if (t == ZIO_TYPE_WRITE && zio->io_vd && zio->io_vd->vdev_aux)
1519 t = ZIO_TYPE_NULL;
1520
1521 /*
1522 * If this is a high priority I/O, then use the high priority taskq if
1523 * available.
1524 */
1525 if (zio->io_priority == ZIO_PRIORITY_NOW &&
1526 spa->spa_zio_taskq[t][q + 1].stqs_count != 0)
1527 q++;
1528
1529 ASSERT3U(q, <, ZIO_TASKQ_TYPES);
1530
1531 /*
1532 * NB: We are assuming that the zio can only be dispatched
1533 * to a single taskq at a time. It would be a grievous error
1534 * to dispatch the zio to another taskq at the same time.
1535 */
1536 #if defined(illumos) || !defined(_KERNEL)
1537 ASSERT(zio->io_tqent.tqent_next == NULL);
1538 #elif defined(__NetBSD__)
1539 ASSERT(zio->io_tqent.tqent_queued == 0);
1540 #else
1541 ASSERT(zio->io_tqent.tqent_task.ta_pending == 0);
1542 #endif
1543 spa_taskq_dispatch_ent(spa, t, q, (task_func_t *)zio_execute, zio,
1544 flags, &zio->io_tqent);
1545 }
1546
1547 static boolean_t
zio_taskq_member(zio_t * zio,zio_taskq_type_t q)1548 zio_taskq_member(zio_t *zio, zio_taskq_type_t q)
1549 {
1550 kthread_t *executor = zio->io_executor;
1551 spa_t *spa = zio->io_spa;
1552
1553 for (zio_type_t t = 0; t < ZIO_TYPES; t++) {
1554 spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q];
1555 uint_t i;
1556 for (i = 0; i < tqs->stqs_count; i++) {
1557 if (taskq_member(tqs->stqs_taskq[i], executor))
1558 return (B_TRUE);
1559 }
1560 }
1561
1562 return (B_FALSE);
1563 }
1564
1565 static int
zio_issue_async(zio_t * zio)1566 zio_issue_async(zio_t *zio)
1567 {
1568 zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE);
1569
1570 return (ZIO_PIPELINE_STOP);
1571 }
1572
1573 void
zio_interrupt(zio_t * zio)1574 zio_interrupt(zio_t *zio)
1575 {
1576 zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT, B_FALSE);
1577 }
1578
1579 void
zio_delay_interrupt(zio_t * zio)1580 zio_delay_interrupt(zio_t *zio)
1581 {
1582 /*
1583 * The timeout_generic() function isn't defined in userspace, so
1584 * rather than trying to implement the function, the zio delay
1585 * functionality has been disabled for userspace builds.
1586 */
1587
1588 #ifndef __NetBSD__
1589 /* XXXNETBSD implement timeout_generic() with a callout_t in zio_t */
1590 /*
1591 * If io_target_timestamp is zero, then no delay has been registered
1592 * for this IO, thus jump to the end of this function and "skip" the
1593 * delay; issuing it directly to the zio layer.
1594 */
1595 if (zio->io_target_timestamp != 0) {
1596 hrtime_t now = gethrtime();
1597
1598 if (now >= zio->io_target_timestamp) {
1599 /*
1600 * This IO has already taken longer than the target
1601 * delay to complete, so we don't want to delay it
1602 * any longer; we "miss" the delay and issue it
1603 * directly to the zio layer. This is likely due to
1604 * the target latency being set to a value less than
1605 * the underlying hardware can satisfy (e.g. delay
1606 * set to 1ms, but the disks take 10ms to complete an
1607 * IO request).
1608 */
1609
1610 DTRACE_PROBE2(zio__delay__miss, zio_t *, zio,
1611 hrtime_t, now);
1612
1613 zio_interrupt(zio);
1614 } else {
1615 hrtime_t diff = zio->io_target_timestamp - now;
1616
1617 DTRACE_PROBE3(zio__delay__hit, zio_t *, zio,
1618 hrtime_t, now, hrtime_t, diff);
1619
1620 (void) timeout_generic(CALLOUT_NORMAL,
1621 (void (*)(void *))zio_interrupt, zio, diff, 1, 0);
1622 }
1623
1624 return;
1625 }
1626 #endif
1627
1628 DTRACE_PROBE1(zio__delay__skip, zio_t *, zio);
1629 zio_interrupt(zio);
1630 }
1631
1632 /*
1633 * Execute the I/O pipeline until one of the following occurs:
1634 *
1635 * (1) the I/O completes
1636 * (2) the pipeline stalls waiting for dependent child I/Os
1637 * (3) the I/O issues, so we're waiting for an I/O completion interrupt
1638 * (4) the I/O is delegated by vdev-level caching or aggregation
1639 * (5) the I/O is deferred due to vdev-level queueing
1640 * (6) the I/O is handed off to another thread.
1641 *
1642 * In all cases, the pipeline stops whenever there's no CPU work; it never
1643 * burns a thread in cv_wait().
1644 *
1645 * There's no locking on io_stage because there's no legitimate way
1646 * for multiple threads to be attempting to process the same I/O.
1647 */
1648 static zio_pipe_stage_t *zio_pipeline[];
1649
1650 void
zio_execute(zio_t * zio)1651 zio_execute(zio_t *zio)
1652 {
1653 zio->io_executor = curthread;
1654
1655 ASSERT3U(zio->io_queued_timestamp, >, 0);
1656
1657 while (zio->io_stage < ZIO_STAGE_DONE) {
1658 enum zio_stage pipeline = zio->io_pipeline;
1659 enum zio_stage stage = zio->io_stage;
1660 int rv;
1661
1662 ASSERT(!MUTEX_HELD(&zio->io_lock));
1663 ASSERT(ISP2(stage));
1664 ASSERT(zio->io_stall == NULL);
1665
1666 do {
1667 stage <<= 1;
1668 } while ((stage & pipeline) == 0);
1669
1670 ASSERT(stage <= ZIO_STAGE_DONE);
1671
1672 /*
1673 * If we are in interrupt context and this pipeline stage
1674 * will grab a config lock that is held across I/O,
1675 * or may wait for an I/O that needs an interrupt thread
1676 * to complete, issue async to avoid deadlock.
1677 *
1678 * For VDEV_IO_START, we cut in line so that the io will
1679 * be sent to disk promptly.
1680 */
1681 if ((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL &&
1682 zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) {
1683 boolean_t cut = (stage == ZIO_STAGE_VDEV_IO_START) ?
1684 zio_requeue_io_start_cut_in_line : B_FALSE;
1685 zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut);
1686 return;
1687 }
1688
1689 zio->io_stage = stage;
1690 zio->io_pipeline_trace |= zio->io_stage;
1691 rv = zio_pipeline[highbit64(stage) - 1](zio);
1692
1693 if (rv == ZIO_PIPELINE_STOP)
1694 return;
1695
1696 ASSERT(rv == ZIO_PIPELINE_CONTINUE);
1697 }
1698 }
1699
1700 /*
1701 * ==========================================================================
1702 * Initiate I/O, either sync or async
1703 * ==========================================================================
1704 */
1705 int
zio_wait(zio_t * zio)1706 zio_wait(zio_t *zio)
1707 {
1708 int error;
1709
1710 ASSERT(zio->io_stage == ZIO_STAGE_OPEN);
1711 ASSERT(zio->io_executor == NULL);
1712
1713 zio->io_waiter = curthread;
1714 ASSERT0(zio->io_queued_timestamp);
1715 zio->io_queued_timestamp = gethrtime();
1716
1717 zio_execute(zio);
1718
1719 mutex_enter(&zio->io_lock);
1720 while (zio->io_executor != NULL)
1721 cv_wait(&zio->io_cv, &zio->io_lock);
1722 mutex_exit(&zio->io_lock);
1723
1724 error = zio->io_error;
1725 zio_destroy(zio);
1726
1727 return (error);
1728 }
1729
1730 void
zio_nowait(zio_t * zio)1731 zio_nowait(zio_t *zio)
1732 {
1733 ASSERT(zio->io_executor == NULL);
1734
1735 if (zio->io_child_type == ZIO_CHILD_LOGICAL &&
1736 zio_unique_parent(zio) == NULL) {
1737 /*
1738 * This is a logical async I/O with no parent to wait for it.
1739 * We add it to the spa_async_root_zio "Godfather" I/O which
1740 * will ensure they complete prior to unloading the pool.
1741 */
1742 spa_t *spa = zio->io_spa;
1743
1744 zio_add_child(spa->spa_async_zio_root[CPU_SEQID], zio);
1745 }
1746
1747 ASSERT0(zio->io_queued_timestamp);
1748 zio->io_queued_timestamp = gethrtime();
1749 zio_execute(zio);
1750 }
1751
1752 /*
1753 * ==========================================================================
1754 * Reexecute or suspend/resume failed I/O
1755 * ==========================================================================
1756 */
1757
1758 static void
zio_reexecute(zio_t * pio)1759 zio_reexecute(zio_t *pio)
1760 {
1761 zio_t *cio, *cio_next;
1762
1763 ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL);
1764 ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN);
1765 ASSERT(pio->io_gang_leader == NULL);
1766 ASSERT(pio->io_gang_tree == NULL);
1767
1768 pio->io_flags = pio->io_orig_flags;
1769 pio->io_stage = pio->io_orig_stage;
1770 pio->io_pipeline = pio->io_orig_pipeline;
1771 pio->io_reexecute = 0;
1772 pio->io_flags |= ZIO_FLAG_REEXECUTED;
1773 pio->io_pipeline_trace = 0;
1774 pio->io_error = 0;
1775 for (int w = 0; w < ZIO_WAIT_TYPES; w++)
1776 pio->io_state[w] = 0;
1777 for (int c = 0; c < ZIO_CHILD_TYPES; c++)
1778 pio->io_child_error[c] = 0;
1779
1780 if (IO_IS_ALLOCATING(pio))
1781 BP_ZERO(pio->io_bp);
1782
1783 /*
1784 * As we reexecute pio's children, new children could be created.
1785 * New children go to the head of pio's io_child_list, however,
1786 * so we will (correctly) not reexecute them. The key is that
1787 * the remainder of pio's io_child_list, from 'cio_next' onward,
1788 * cannot be affected by any side effects of reexecuting 'cio'.
1789 */
1790 zio_link_t *zl = NULL;
1791 for (cio = zio_walk_children(pio, &zl); cio != NULL; cio = cio_next) {
1792 cio_next = zio_walk_children(pio, &zl);
1793 mutex_enter(&pio->io_lock);
1794 for (int w = 0; w < ZIO_WAIT_TYPES; w++)
1795 pio->io_children[cio->io_child_type][w]++;
1796 mutex_exit(&pio->io_lock);
1797 zio_reexecute(cio);
1798 }
1799
1800 /*
1801 * Now that all children have been reexecuted, execute the parent.
1802 * We don't reexecute "The Godfather" I/O here as it's the
1803 * responsibility of the caller to wait on him.
1804 */
1805 if (!(pio->io_flags & ZIO_FLAG_GODFATHER)) {
1806 pio->io_queued_timestamp = gethrtime();
1807 zio_execute(pio);
1808 }
1809 }
1810
1811 void
zio_suspend(spa_t * spa,zio_t * zio)1812 zio_suspend(spa_t *spa, zio_t *zio)
1813 {
1814 if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC)
1815 fm_panic("Pool '%s' has encountered an uncorrectable I/O "
1816 "failure and the failure mode property for this pool "
1817 "is set to panic.", spa_name(spa));
1818
1819 zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, 0, 0);
1820
1821 mutex_enter(&spa->spa_suspend_lock);
1822
1823 if (spa->spa_suspend_zio_root == NULL)
1824 spa->spa_suspend_zio_root = zio_root(spa, NULL, NULL,
1825 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
1826 ZIO_FLAG_GODFATHER);
1827
1828 spa->spa_suspended = B_TRUE;
1829
1830 if (zio != NULL) {
1831 ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER));
1832 ASSERT(zio != spa->spa_suspend_zio_root);
1833 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
1834 ASSERT(zio_unique_parent(zio) == NULL);
1835 ASSERT(zio->io_stage == ZIO_STAGE_DONE);
1836 zio_add_child(spa->spa_suspend_zio_root, zio);
1837 }
1838
1839 mutex_exit(&spa->spa_suspend_lock);
1840 }
1841
1842 int
zio_resume(spa_t * spa)1843 zio_resume(spa_t *spa)
1844 {
1845 zio_t *pio;
1846
1847 /*
1848 * Reexecute all previously suspended i/o.
1849 */
1850 mutex_enter(&spa->spa_suspend_lock);
1851 spa->spa_suspended = B_FALSE;
1852 cv_broadcast(&spa->spa_suspend_cv);
1853 pio = spa->spa_suspend_zio_root;
1854 spa->spa_suspend_zio_root = NULL;
1855 mutex_exit(&spa->spa_suspend_lock);
1856
1857 if (pio == NULL)
1858 return (0);
1859
1860 zio_reexecute(pio);
1861 return (zio_wait(pio));
1862 }
1863
1864 void
zio_resume_wait(spa_t * spa)1865 zio_resume_wait(spa_t *spa)
1866 {
1867 mutex_enter(&spa->spa_suspend_lock);
1868 while (spa_suspended(spa))
1869 cv_wait(&spa->spa_suspend_cv, &spa->spa_suspend_lock);
1870 mutex_exit(&spa->spa_suspend_lock);
1871 }
1872
1873 /*
1874 * ==========================================================================
1875 * Gang blocks.
1876 *
1877 * A gang block is a collection of small blocks that looks to the DMU
1878 * like one large block. When zio_dva_allocate() cannot find a block
1879 * of the requested size, due to either severe fragmentation or the pool
1880 * being nearly full, it calls zio_write_gang_block() to construct the
1881 * block from smaller fragments.
1882 *
1883 * A gang block consists of a gang header (zio_gbh_phys_t) and up to
1884 * three (SPA_GBH_NBLKPTRS) gang members. The gang header is just like
1885 * an indirect block: it's an array of block pointers. It consumes
1886 * only one sector and hence is allocatable regardless of fragmentation.
1887 * The gang header's bps point to its gang members, which hold the data.
1888 *
1889 * Gang blocks are self-checksumming, using the bp's <vdev, offset, txg>
1890 * as the verifier to ensure uniqueness of the SHA256 checksum.
1891 * Critically, the gang block bp's blk_cksum is the checksum of the data,
1892 * not the gang header. This ensures that data block signatures (needed for
1893 * deduplication) are independent of how the block is physically stored.
1894 *
1895 * Gang blocks can be nested: a gang member may itself be a gang block.
1896 * Thus every gang block is a tree in which root and all interior nodes are
1897 * gang headers, and the leaves are normal blocks that contain user data.
1898 * The root of the gang tree is called the gang leader.
1899 *
1900 * To perform any operation (read, rewrite, free, claim) on a gang block,
1901 * zio_gang_assemble() first assembles the gang tree (minus data leaves)
1902 * in the io_gang_tree field of the original logical i/o by recursively
1903 * reading the gang leader and all gang headers below it. This yields
1904 * an in-core tree containing the contents of every gang header and the
1905 * bps for every constituent of the gang block.
1906 *
1907 * With the gang tree now assembled, zio_gang_issue() just walks the gang tree
1908 * and invokes a callback on each bp. To free a gang block, zio_gang_issue()
1909 * calls zio_free_gang() -- a trivial wrapper around zio_free() -- for each bp.
1910 * zio_claim_gang() provides a similarly trivial wrapper for zio_claim().
1911 * zio_read_gang() is a wrapper around zio_read() that omits reading gang
1912 * headers, since we already have those in io_gang_tree. zio_rewrite_gang()
1913 * performs a zio_rewrite() of the data or, for gang headers, a zio_rewrite()
1914 * of the gang header plus zio_checksum_compute() of the data to update the
1915 * gang header's blk_cksum as described above.
1916 *
1917 * The two-phase assemble/issue model solves the problem of partial failure --
1918 * what if you'd freed part of a gang block but then couldn't read the
1919 * gang header for another part? Assembling the entire gang tree first
1920 * ensures that all the necessary gang header I/O has succeeded before
1921 * starting the actual work of free, claim, or write. Once the gang tree
1922 * is assembled, free and claim are in-memory operations that cannot fail.
1923 *
1924 * In the event that a gang write fails, zio_dva_unallocate() walks the
1925 * gang tree to immediately free (i.e. insert back into the space map)
1926 * everything we've allocated. This ensures that we don't get ENOSPC
1927 * errors during repeated suspend/resume cycles due to a flaky device.
1928 *
1929 * Gang rewrites only happen during sync-to-convergence. If we can't assemble
1930 * the gang tree, we won't modify the block, so we can safely defer the free
1931 * (knowing that the block is still intact). If we *can* assemble the gang
1932 * tree, then even if some of the rewrites fail, zio_dva_unallocate() will free
1933 * each constituent bp and we can allocate a new block on the next sync pass.
1934 *
1935 * In all cases, the gang tree allows complete recovery from partial failure.
1936 * ==========================================================================
1937 */
1938
1939 static zio_t *
zio_read_gang(zio_t * pio,blkptr_t * bp,zio_gang_node_t * gn,void * data)1940 zio_read_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data)
1941 {
1942 if (gn != NULL)
1943 return (pio);
1944
1945 return (zio_read(pio, pio->io_spa, bp, data, BP_GET_PSIZE(bp),
1946 NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio),
1947 &pio->io_bookmark));
1948 }
1949
1950 zio_t *
zio_rewrite_gang(zio_t * pio,blkptr_t * bp,zio_gang_node_t * gn,void * data)1951 zio_rewrite_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data)
1952 {
1953 zio_t *zio;
1954
1955 if (gn != NULL) {
1956 zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp,
1957 gn->gn_gbh, SPA_GANGBLOCKSIZE, NULL, NULL, pio->io_priority,
1958 ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark);
1959 /*
1960 * As we rewrite each gang header, the pipeline will compute
1961 * a new gang block header checksum for it; but no one will
1962 * compute a new data checksum, so we do that here. The one
1963 * exception is the gang leader: the pipeline already computed
1964 * its data checksum because that stage precedes gang assembly.
1965 * (Presently, nothing actually uses interior data checksums;
1966 * this is just good hygiene.)
1967 */
1968 if (gn != pio->io_gang_leader->io_gang_tree) {
1969 zio_checksum_compute(zio, BP_GET_CHECKSUM(bp),
1970 data, BP_GET_PSIZE(bp));
1971 }
1972 /*
1973 * If we are here to damage data for testing purposes,
1974 * leave the GBH alone so that we can detect the damage.
1975 */
1976 if (pio->io_gang_leader->io_flags & ZIO_FLAG_INDUCE_DAMAGE)
1977 zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES;
1978 } else {
1979 zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp,
1980 data, BP_GET_PSIZE(bp), NULL, NULL, pio->io_priority,
1981 ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark);
1982 }
1983
1984 return (zio);
1985 }
1986
1987 /* ARGSUSED */
1988 zio_t *
zio_free_gang(zio_t * pio,blkptr_t * bp,zio_gang_node_t * gn,void * data)1989 zio_free_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data)
1990 {
1991 return (zio_free_sync(pio, pio->io_spa, pio->io_txg, bp,
1992 BP_IS_GANG(bp) ? SPA_GANGBLOCKSIZE : BP_GET_PSIZE(bp),
1993 ZIO_GANG_CHILD_FLAGS(pio)));
1994 }
1995
1996 /* ARGSUSED */
1997 zio_t *
zio_claim_gang(zio_t * pio,blkptr_t * bp,zio_gang_node_t * gn,void * data)1998 zio_claim_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data)
1999 {
2000 return (zio_claim(pio, pio->io_spa, pio->io_txg, bp,
2001 NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio)));
2002 }
2003
2004 static zio_gang_issue_func_t *zio_gang_issue_func[ZIO_TYPES] = {
2005 NULL,
2006 zio_read_gang,
2007 zio_rewrite_gang,
2008 zio_free_gang,
2009 zio_claim_gang,
2010 NULL
2011 };
2012
2013 static void zio_gang_tree_assemble_done(zio_t *zio);
2014
2015 static zio_gang_node_t *
zio_gang_node_alloc(zio_gang_node_t ** gnpp)2016 zio_gang_node_alloc(zio_gang_node_t **gnpp)
2017 {
2018 zio_gang_node_t *gn;
2019
2020 ASSERT(*gnpp == NULL);
2021
2022 gn = kmem_zalloc(sizeof (*gn), KM_SLEEP);
2023 gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE);
2024 *gnpp = gn;
2025
2026 return (gn);
2027 }
2028
2029 static void
zio_gang_node_free(zio_gang_node_t ** gnpp)2030 zio_gang_node_free(zio_gang_node_t **gnpp)
2031 {
2032 zio_gang_node_t *gn = *gnpp;
2033
2034 for (int g = 0; g < SPA_GBH_NBLKPTRS; g++)
2035 ASSERT(gn->gn_child[g] == NULL);
2036
2037 zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE);
2038 kmem_free(gn, sizeof (*gn));
2039 *gnpp = NULL;
2040 }
2041
2042 static void
zio_gang_tree_free(zio_gang_node_t ** gnpp)2043 zio_gang_tree_free(zio_gang_node_t **gnpp)
2044 {
2045 zio_gang_node_t *gn = *gnpp;
2046
2047 if (gn == NULL)
2048 return;
2049
2050 for (int g = 0; g < SPA_GBH_NBLKPTRS; g++)
2051 zio_gang_tree_free(&gn->gn_child[g]);
2052
2053 zio_gang_node_free(gnpp);
2054 }
2055
2056 static void
zio_gang_tree_assemble(zio_t * gio,blkptr_t * bp,zio_gang_node_t ** gnpp)2057 zio_gang_tree_assemble(zio_t *gio, blkptr_t *bp, zio_gang_node_t **gnpp)
2058 {
2059 zio_gang_node_t *gn = zio_gang_node_alloc(gnpp);
2060
2061 ASSERT(gio->io_gang_leader == gio);
2062 ASSERT(BP_IS_GANG(bp));
2063
2064 zio_nowait(zio_read(gio, gio->io_spa, bp, gn->gn_gbh,
2065 SPA_GANGBLOCKSIZE, zio_gang_tree_assemble_done, gn,
2066 gio->io_priority, ZIO_GANG_CHILD_FLAGS(gio), &gio->io_bookmark));
2067 }
2068
2069 static void
zio_gang_tree_assemble_done(zio_t * zio)2070 zio_gang_tree_assemble_done(zio_t *zio)
2071 {
2072 zio_t *gio = zio->io_gang_leader;
2073 zio_gang_node_t *gn = zio->io_private;
2074 blkptr_t *bp = zio->io_bp;
2075
2076 ASSERT(gio == zio_unique_parent(zio));
2077 ASSERT(zio->io_child_count == 0);
2078
2079 if (zio->io_error)
2080 return;
2081
2082 if (BP_SHOULD_BYTESWAP(bp))
2083 byteswap_uint64_array(zio->io_data, zio->io_size);
2084
2085 ASSERT(zio->io_data == gn->gn_gbh);
2086 ASSERT(zio->io_size == SPA_GANGBLOCKSIZE);
2087 ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC);
2088
2089 for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) {
2090 blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g];
2091 if (!BP_IS_GANG(gbp))
2092 continue;
2093 zio_gang_tree_assemble(gio, gbp, &gn->gn_child[g]);
2094 }
2095 }
2096
2097 static void
zio_gang_tree_issue(zio_t * pio,zio_gang_node_t * gn,blkptr_t * bp,void * data)2098 zio_gang_tree_issue(zio_t *pio, zio_gang_node_t *gn, blkptr_t *bp, void *data)
2099 {
2100 zio_t *gio = pio->io_gang_leader;
2101 zio_t *zio;
2102
2103 ASSERT(BP_IS_GANG(bp) == !!gn);
2104 ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp));
2105 ASSERT(BP_GET_LSIZE(bp) == BP_GET_PSIZE(bp) || gn == gio->io_gang_tree);
2106
2107 /*
2108 * If you're a gang header, your data is in gn->gn_gbh.
2109 * If you're a gang member, your data is in 'data' and gn == NULL.
2110 */
2111 zio = zio_gang_issue_func[gio->io_type](pio, bp, gn, data);
2112
2113 if (gn != NULL) {
2114 ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC);
2115
2116 for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) {
2117 blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g];
2118 if (BP_IS_HOLE(gbp))
2119 continue;
2120 zio_gang_tree_issue(zio, gn->gn_child[g], gbp, data);
2121 data = (char *)data + BP_GET_PSIZE(gbp);
2122 }
2123 }
2124
2125 if (gn == gio->io_gang_tree && gio->io_data != NULL)
2126 ASSERT3P((char *)gio->io_data + gio->io_size, ==, data);
2127
2128 if (zio != pio)
2129 zio_nowait(zio);
2130 }
2131
2132 static int
zio_gang_assemble(zio_t * zio)2133 zio_gang_assemble(zio_t *zio)
2134 {
2135 blkptr_t *bp = zio->io_bp;
2136
2137 ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == NULL);
2138 ASSERT(zio->io_child_type > ZIO_CHILD_GANG);
2139
2140 zio->io_gang_leader = zio;
2141
2142 zio_gang_tree_assemble(zio, bp, &zio->io_gang_tree);
2143
2144 return (ZIO_PIPELINE_CONTINUE);
2145 }
2146
2147 static int
zio_gang_issue(zio_t * zio)2148 zio_gang_issue(zio_t *zio)
2149 {
2150 blkptr_t *bp = zio->io_bp;
2151
2152 if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE))
2153 return (ZIO_PIPELINE_STOP);
2154
2155 ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == zio);
2156 ASSERT(zio->io_child_type > ZIO_CHILD_GANG);
2157
2158 if (zio->io_child_error[ZIO_CHILD_GANG] == 0)
2159 zio_gang_tree_issue(zio, zio->io_gang_tree, bp, zio->io_data);
2160 else
2161 zio_gang_tree_free(&zio->io_gang_tree);
2162
2163 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
2164
2165 return (ZIO_PIPELINE_CONTINUE);
2166 }
2167
2168 static void
zio_write_gang_member_ready(zio_t * zio)2169 zio_write_gang_member_ready(zio_t *zio)
2170 {
2171 zio_t *pio = zio_unique_parent(zio);
2172 zio_t *gio = zio->io_gang_leader;
2173 dva_t *cdva = zio->io_bp->blk_dva;
2174 dva_t *pdva = pio->io_bp->blk_dva;
2175 uint64_t asize;
2176
2177 if (BP_IS_HOLE(zio->io_bp))
2178 return;
2179
2180 ASSERT(BP_IS_HOLE(&zio->io_bp_orig));
2181
2182 ASSERT(zio->io_child_type == ZIO_CHILD_GANG);
2183 ASSERT3U(zio->io_prop.zp_copies, ==, gio->io_prop.zp_copies);
2184 ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp));
2185 ASSERT3U(pio->io_prop.zp_copies, <=, BP_GET_NDVAS(pio->io_bp));
2186 ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp));
2187
2188 mutex_enter(&pio->io_lock);
2189 for (int d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) {
2190 ASSERT(DVA_GET_GANG(&pdva[d]));
2191 asize = DVA_GET_ASIZE(&pdva[d]);
2192 asize += DVA_GET_ASIZE(&cdva[d]);
2193 DVA_SET_ASIZE(&pdva[d], asize);
2194 }
2195 mutex_exit(&pio->io_lock);
2196 }
2197
2198 static int
zio_write_gang_block(zio_t * pio)2199 zio_write_gang_block(zio_t *pio)
2200 {
2201 spa_t *spa = pio->io_spa;
2202 metaslab_class_t *mc = spa_normal_class(spa);
2203 blkptr_t *bp = pio->io_bp;
2204 zio_t *gio = pio->io_gang_leader;
2205 zio_t *zio;
2206 zio_gang_node_t *gn, **gnpp;
2207 zio_gbh_phys_t *gbh;
2208 uint64_t txg = pio->io_txg;
2209 uint64_t resid = pio->io_size;
2210 uint64_t lsize;
2211 int copies = gio->io_prop.zp_copies;
2212 int gbh_copies = MIN(copies + 1, spa_max_replication(spa));
2213 zio_prop_t zp;
2214 int error;
2215
2216 int flags = METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER;
2217 if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) {
2218 ASSERT(pio->io_priority == ZIO_PRIORITY_ASYNC_WRITE);
2219 ASSERT(!(pio->io_flags & ZIO_FLAG_NODATA));
2220
2221 flags |= METASLAB_ASYNC_ALLOC;
2222 VERIFY(refcount_held(&mc->mc_alloc_slots, pio));
2223
2224 /*
2225 * The logical zio has already placed a reservation for
2226 * 'copies' allocation slots but gang blocks may require
2227 * additional copies. These additional copies
2228 * (i.e. gbh_copies - copies) are guaranteed to succeed
2229 * since metaslab_class_throttle_reserve() always allows
2230 * additional reservations for gang blocks.
2231 */
2232 VERIFY(metaslab_class_throttle_reserve(mc, gbh_copies - copies,
2233 pio, flags));
2234 }
2235
2236 error = metaslab_alloc(spa, mc, SPA_GANGBLOCKSIZE,
2237 bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, flags,
2238 &pio->io_alloc_list, pio);
2239 if (error) {
2240 if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) {
2241 ASSERT(pio->io_priority == ZIO_PRIORITY_ASYNC_WRITE);
2242 ASSERT(!(pio->io_flags & ZIO_FLAG_NODATA));
2243
2244 /*
2245 * If we failed to allocate the gang block header then
2246 * we remove any additional allocation reservations that
2247 * we placed here. The original reservation will
2248 * be removed when the logical I/O goes to the ready
2249 * stage.
2250 */
2251 metaslab_class_throttle_unreserve(mc,
2252 gbh_copies - copies, pio);
2253 }
2254 pio->io_error = error;
2255 return (ZIO_PIPELINE_CONTINUE);
2256 }
2257
2258 if (pio == gio) {
2259 gnpp = &gio->io_gang_tree;
2260 } else {
2261 gnpp = pio->io_private;
2262 ASSERT(pio->io_ready == zio_write_gang_member_ready);
2263 }
2264
2265 gn = zio_gang_node_alloc(gnpp);
2266 gbh = gn->gn_gbh;
2267 bzero(gbh, SPA_GANGBLOCKSIZE);
2268
2269 /*
2270 * Create the gang header.
2271 */
2272 zio = zio_rewrite(pio, spa, txg, bp, gbh, SPA_GANGBLOCKSIZE, NULL, NULL,
2273 pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark);
2274
2275 /*
2276 * Create and nowait the gang children.
2277 */
2278 for (int g = 0; resid != 0; resid -= lsize, g++) {
2279 lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g),
2280 SPA_MINBLOCKSIZE);
2281 ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid);
2282
2283 zp.zp_checksum = gio->io_prop.zp_checksum;
2284 zp.zp_compress = ZIO_COMPRESS_OFF;
2285 zp.zp_type = DMU_OT_NONE;
2286 zp.zp_level = 0;
2287 zp.zp_copies = gio->io_prop.zp_copies;
2288 zp.zp_dedup = B_FALSE;
2289 zp.zp_dedup_verify = B_FALSE;
2290 zp.zp_nopwrite = B_FALSE;
2291
2292 zio_t *cio = zio_write(zio, spa, txg, &gbh->zg_blkptr[g],
2293 (char *)pio->io_data + (pio->io_size - resid), lsize, &zp,
2294 zio_write_gang_member_ready, NULL, NULL, NULL,
2295 &gn->gn_child[g], pio->io_priority,
2296 ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark);
2297
2298 if (pio->io_flags & ZIO_FLAG_IO_ALLOCATING) {
2299 ASSERT(pio->io_priority == ZIO_PRIORITY_ASYNC_WRITE);
2300 ASSERT(!(pio->io_flags & ZIO_FLAG_NODATA));
2301
2302 /*
2303 * Gang children won't throttle but we should
2304 * account for their work, so reserve an allocation
2305 * slot for them here.
2306 */
2307 VERIFY(metaslab_class_throttle_reserve(mc,
2308 zp.zp_copies, cio, flags));
2309 }
2310 zio_nowait(cio);
2311 }
2312
2313 /*
2314 * Set pio's pipeline to just wait for zio to finish.
2315 */
2316 pio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
2317
2318 zio_nowait(zio);
2319
2320 return (ZIO_PIPELINE_CONTINUE);
2321 }
2322
2323 /*
2324 * The zio_nop_write stage in the pipeline determines if allocating a
2325 * new bp is necessary. The nopwrite feature can handle writes in
2326 * either syncing or open context (i.e. zil writes) and as a result is
2327 * mutually exclusive with dedup.
2328 *
2329 * By leveraging a cryptographically secure checksum, such as SHA256, we
2330 * can compare the checksums of the new data and the old to determine if
2331 * allocating a new block is required. Note that our requirements for
2332 * cryptographic strength are fairly weak: there can't be any accidental
2333 * hash collisions, but we don't need to be secure against intentional
2334 * (malicious) collisions. To trigger a nopwrite, you have to be able
2335 * to write the file to begin with, and triggering an incorrect (hash
2336 * collision) nopwrite is no worse than simply writing to the file.
2337 * That said, there are no known attacks against the checksum algorithms
2338 * used for nopwrite, assuming that the salt and the checksums
2339 * themselves remain secret.
2340 */
2341 static int
zio_nop_write(zio_t * zio)2342 zio_nop_write(zio_t *zio)
2343 {
2344 blkptr_t *bp = zio->io_bp;
2345 blkptr_t *bp_orig = &zio->io_bp_orig;
2346 zio_prop_t *zp = &zio->io_prop;
2347
2348 ASSERT(BP_GET_LEVEL(bp) == 0);
2349 ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE));
2350 ASSERT(zp->zp_nopwrite);
2351 ASSERT(!zp->zp_dedup);
2352 ASSERT(zio->io_bp_override == NULL);
2353 ASSERT(IO_IS_ALLOCATING(zio));
2354
2355 /*
2356 * Check to see if the original bp and the new bp have matching
2357 * characteristics (i.e. same checksum, compression algorithms, etc).
2358 * If they don't then just continue with the pipeline which will
2359 * allocate a new bp.
2360 */
2361 if (BP_IS_HOLE(bp_orig) ||
2362 !(zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_flags &
2363 ZCHECKSUM_FLAG_NOPWRITE) ||
2364 BP_GET_CHECKSUM(bp) != BP_GET_CHECKSUM(bp_orig) ||
2365 BP_GET_COMPRESS(bp) != BP_GET_COMPRESS(bp_orig) ||
2366 BP_GET_DEDUP(bp) != BP_GET_DEDUP(bp_orig) ||
2367 zp->zp_copies != BP_GET_NDVAS(bp_orig))
2368 return (ZIO_PIPELINE_CONTINUE);
2369
2370 /*
2371 * If the checksums match then reset the pipeline so that we
2372 * avoid allocating a new bp and issuing any I/O.
2373 */
2374 if (ZIO_CHECKSUM_EQUAL(bp->blk_cksum, bp_orig->blk_cksum)) {
2375 ASSERT(zio_checksum_table[zp->zp_checksum].ci_flags &
2376 ZCHECKSUM_FLAG_NOPWRITE);
2377 ASSERT3U(BP_GET_PSIZE(bp), ==, BP_GET_PSIZE(bp_orig));
2378 ASSERT3U(BP_GET_LSIZE(bp), ==, BP_GET_LSIZE(bp_orig));
2379 ASSERT(zp->zp_compress != ZIO_COMPRESS_OFF);
2380 ASSERT(bcmp(&bp->blk_prop, &bp_orig->blk_prop,
2381 sizeof (uint64_t)) == 0);
2382
2383 *bp = *bp_orig;
2384 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
2385 zio->io_flags |= ZIO_FLAG_NOPWRITE;
2386 }
2387
2388 return (ZIO_PIPELINE_CONTINUE);
2389 }
2390
2391 /*
2392 * ==========================================================================
2393 * Dedup
2394 * ==========================================================================
2395 */
2396 static void
zio_ddt_child_read_done(zio_t * zio)2397 zio_ddt_child_read_done(zio_t *zio)
2398 {
2399 blkptr_t *bp = zio->io_bp;
2400 ddt_entry_t *dde = zio->io_private;
2401 ddt_phys_t *ddp;
2402 zio_t *pio = zio_unique_parent(zio);
2403
2404 mutex_enter(&pio->io_lock);
2405 ddp = ddt_phys_select(dde, bp);
2406 if (zio->io_error == 0)
2407 ddt_phys_clear(ddp); /* this ddp doesn't need repair */
2408 if (zio->io_error == 0 && dde->dde_repair_data == NULL)
2409 dde->dde_repair_data = zio->io_data;
2410 else
2411 zio_buf_free(zio->io_data, zio->io_size);
2412 mutex_exit(&pio->io_lock);
2413 }
2414
2415 static int
zio_ddt_read_start(zio_t * zio)2416 zio_ddt_read_start(zio_t *zio)
2417 {
2418 blkptr_t *bp = zio->io_bp;
2419
2420 ASSERT(BP_GET_DEDUP(bp));
2421 ASSERT(BP_GET_PSIZE(bp) == zio->io_size);
2422 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
2423
2424 if (zio->io_child_error[ZIO_CHILD_DDT]) {
2425 ddt_t *ddt = ddt_select(zio->io_spa, bp);
2426 ddt_entry_t *dde = ddt_repair_start(ddt, bp);
2427 ddt_phys_t *ddp = dde->dde_phys;
2428 ddt_phys_t *ddp_self = ddt_phys_select(dde, bp);
2429 blkptr_t blk;
2430
2431 ASSERT(zio->io_vsd == NULL);
2432 zio->io_vsd = dde;
2433
2434 if (ddp_self == NULL)
2435 return (ZIO_PIPELINE_CONTINUE);
2436
2437 for (int p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
2438 if (ddp->ddp_phys_birth == 0 || ddp == ddp_self)
2439 continue;
2440 ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp,
2441 &blk);
2442 zio_nowait(zio_read(zio, zio->io_spa, &blk,
2443 zio_buf_alloc(zio->io_size), zio->io_size,
2444 zio_ddt_child_read_done, dde, zio->io_priority,
2445 ZIO_DDT_CHILD_FLAGS(zio) | ZIO_FLAG_DONT_PROPAGATE,
2446 &zio->io_bookmark));
2447 }
2448 return (ZIO_PIPELINE_CONTINUE);
2449 }
2450
2451 zio_nowait(zio_read(zio, zio->io_spa, bp,
2452 zio->io_data, zio->io_size, NULL, NULL, zio->io_priority,
2453 ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark));
2454
2455 return (ZIO_PIPELINE_CONTINUE);
2456 }
2457
2458 static int
zio_ddt_read_done(zio_t * zio)2459 zio_ddt_read_done(zio_t *zio)
2460 {
2461 blkptr_t *bp = zio->io_bp;
2462
2463 if (zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE))
2464 return (ZIO_PIPELINE_STOP);
2465
2466 ASSERT(BP_GET_DEDUP(bp));
2467 ASSERT(BP_GET_PSIZE(bp) == zio->io_size);
2468 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
2469
2470 if (zio->io_child_error[ZIO_CHILD_DDT]) {
2471 ddt_t *ddt = ddt_select(zio->io_spa, bp);
2472 ddt_entry_t *dde = zio->io_vsd;
2473 if (ddt == NULL) {
2474 ASSERT(spa_load_state(zio->io_spa) != SPA_LOAD_NONE);
2475 return (ZIO_PIPELINE_CONTINUE);
2476 }
2477 if (dde == NULL) {
2478 zio->io_stage = ZIO_STAGE_DDT_READ_START >> 1;
2479 zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE);
2480 return (ZIO_PIPELINE_STOP);
2481 }
2482 if (dde->dde_repair_data != NULL) {
2483 bcopy(dde->dde_repair_data, zio->io_data, zio->io_size);
2484 zio->io_child_error[ZIO_CHILD_DDT] = 0;
2485 }
2486 ddt_repair_done(ddt, dde);
2487 zio->io_vsd = NULL;
2488 }
2489
2490 ASSERT(zio->io_vsd == NULL);
2491
2492 return (ZIO_PIPELINE_CONTINUE);
2493 }
2494
2495 static boolean_t
zio_ddt_collision(zio_t * zio,ddt_t * ddt,ddt_entry_t * dde)2496 zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde)
2497 {
2498 spa_t *spa = zio->io_spa;
2499
2500 /*
2501 * Note: we compare the original data, not the transformed data,
2502 * because when zio->io_bp is an override bp, we will not have
2503 * pushed the I/O transforms. That's an important optimization
2504 * because otherwise we'd compress/encrypt all dmu_sync() data twice.
2505 */
2506 for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) {
2507 zio_t *lio = dde->dde_lead_zio[p];
2508
2509 if (lio != NULL) {
2510 return (lio->io_orig_size != zio->io_orig_size ||
2511 bcmp(zio->io_orig_data, lio->io_orig_data,
2512 zio->io_orig_size) != 0);
2513 }
2514 }
2515
2516 for (int p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) {
2517 ddt_phys_t *ddp = &dde->dde_phys[p];
2518
2519 if (ddp->ddp_phys_birth != 0) {
2520 arc_buf_t *abuf = NULL;
2521 arc_flags_t aflags = ARC_FLAG_WAIT;
2522 blkptr_t blk = *zio->io_bp;
2523 int error;
2524
2525 ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth);
2526
2527 ddt_exit(ddt);
2528
2529 error = arc_read(NULL, spa, &blk,
2530 arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ,
2531 ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE,
2532 &aflags, &zio->io_bookmark);
2533
2534 if (error == 0) {
2535 if (arc_buf_size(abuf) != zio->io_orig_size ||
2536 bcmp(abuf->b_data, zio->io_orig_data,
2537 zio->io_orig_size) != 0)
2538 error = SET_ERROR(EEXIST);
2539 arc_buf_destroy(abuf, &abuf);
2540 }
2541
2542 ddt_enter(ddt);
2543 return (error != 0);
2544 }
2545 }
2546
2547 return (B_FALSE);
2548 }
2549
2550 static void
zio_ddt_child_write_ready(zio_t * zio)2551 zio_ddt_child_write_ready(zio_t *zio)
2552 {
2553 int p = zio->io_prop.zp_copies;
2554 ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp);
2555 ddt_entry_t *dde = zio->io_private;
2556 ddt_phys_t *ddp = &dde->dde_phys[p];
2557 zio_t *pio;
2558
2559 if (zio->io_error)
2560 return;
2561
2562 ddt_enter(ddt);
2563
2564 ASSERT(dde->dde_lead_zio[p] == zio);
2565
2566 ddt_phys_fill(ddp, zio->io_bp);
2567
2568 zio_link_t *zl = NULL;
2569 while ((pio = zio_walk_parents(zio, &zl)) != NULL)
2570 ddt_bp_fill(ddp, pio->io_bp, zio->io_txg);
2571
2572 ddt_exit(ddt);
2573 }
2574
2575 static void
zio_ddt_child_write_done(zio_t * zio)2576 zio_ddt_child_write_done(zio_t *zio)
2577 {
2578 int p = zio->io_prop.zp_copies;
2579 ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp);
2580 ddt_entry_t *dde = zio->io_private;
2581 ddt_phys_t *ddp = &dde->dde_phys[p];
2582
2583 ddt_enter(ddt);
2584
2585 ASSERT(ddp->ddp_refcnt == 0);
2586 ASSERT(dde->dde_lead_zio[p] == zio);
2587 dde->dde_lead_zio[p] = NULL;
2588
2589 if (zio->io_error == 0) {
2590 zio_link_t *zl = NULL;
2591 while (zio_walk_parents(zio, &zl) != NULL)
2592 ddt_phys_addref(ddp);
2593 } else {
2594 ddt_phys_clear(ddp);
2595 }
2596
2597 ddt_exit(ddt);
2598 }
2599
2600 static void
zio_ddt_ditto_write_done(zio_t * zio)2601 zio_ddt_ditto_write_done(zio_t *zio)
2602 {
2603 int p = DDT_PHYS_DITTO;
2604 zio_prop_t *zp = &zio->io_prop;
2605 blkptr_t *bp = zio->io_bp;
2606 ddt_t *ddt = ddt_select(zio->io_spa, bp);
2607 ddt_entry_t *dde = zio->io_private;
2608 ddt_phys_t *ddp = &dde->dde_phys[p];
2609 ddt_key_t *ddk = &dde->dde_key;
2610
2611 ddt_enter(ddt);
2612
2613 ASSERT(ddp->ddp_refcnt == 0);
2614 ASSERT(dde->dde_lead_zio[p] == zio);
2615 dde->dde_lead_zio[p] = NULL;
2616
2617 if (zio->io_error == 0) {
2618 ASSERT(ZIO_CHECKSUM_EQUAL(bp->blk_cksum, ddk->ddk_cksum));
2619 ASSERT(zp->zp_copies < SPA_DVAS_PER_BP);
2620 ASSERT(zp->zp_copies == BP_GET_NDVAS(bp) - BP_IS_GANG(bp));
2621 if (ddp->ddp_phys_birth != 0)
2622 ddt_phys_free(ddt, ddk, ddp, zio->io_txg);
2623 ddt_phys_fill(ddp, bp);
2624 }
2625
2626 ddt_exit(ddt);
2627 }
2628
2629 static int
zio_ddt_write(zio_t * zio)2630 zio_ddt_write(zio_t *zio)
2631 {
2632 spa_t *spa = zio->io_spa;
2633 blkptr_t *bp = zio->io_bp;
2634 uint64_t txg = zio->io_txg;
2635 zio_prop_t *zp = &zio->io_prop;
2636 int p = zp->zp_copies;
2637 int ditto_copies;
2638 zio_t *cio = NULL;
2639 zio_t *dio = NULL;
2640 ddt_t *ddt = ddt_select(spa, bp);
2641 ddt_entry_t *dde;
2642 ddt_phys_t *ddp;
2643
2644 ASSERT(BP_GET_DEDUP(bp));
2645 ASSERT(BP_GET_CHECKSUM(bp) == zp->zp_checksum);
2646 ASSERT(BP_IS_HOLE(bp) || zio->io_bp_override);
2647
2648 ddt_enter(ddt);
2649 dde = ddt_lookup(ddt, bp, B_TRUE);
2650 ddp = &dde->dde_phys[p];
2651
2652 if (zp->zp_dedup_verify && zio_ddt_collision(zio, ddt, dde)) {
2653 /*
2654 * If we're using a weak checksum, upgrade to a strong checksum
2655 * and try again. If we're already using a strong checksum,
2656 * we can't resolve it, so just convert to an ordinary write.
2657 * (And automatically e-mail a paper to Nature?)
2658 */
2659 if (!(zio_checksum_table[zp->zp_checksum].ci_flags &
2660 ZCHECKSUM_FLAG_DEDUP)) {
2661 zp->zp_checksum = spa_dedup_checksum(spa);
2662 zio_pop_transforms(zio);
2663 zio->io_stage = ZIO_STAGE_OPEN;
2664 BP_ZERO(bp);
2665 } else {
2666 zp->zp_dedup = B_FALSE;
2667 }
2668 zio->io_pipeline = ZIO_WRITE_PIPELINE;
2669 ddt_exit(ddt);
2670 return (ZIO_PIPELINE_CONTINUE);
2671 }
2672
2673 ditto_copies = ddt_ditto_copies_needed(ddt, dde, ddp);
2674 ASSERT(ditto_copies < SPA_DVAS_PER_BP);
2675
2676 if (ditto_copies > ddt_ditto_copies_present(dde) &&
2677 dde->dde_lead_zio[DDT_PHYS_DITTO] == NULL) {
2678 zio_prop_t czp = *zp;
2679
2680 czp.zp_copies = ditto_copies;
2681
2682 /*
2683 * If we arrived here with an override bp, we won't have run
2684 * the transform stack, so we won't have the data we need to
2685 * generate a child i/o. So, toss the override bp and restart.
2686 * This is safe, because using the override bp is just an
2687 * optimization; and it's rare, so the cost doesn't matter.
2688 */
2689 if (zio->io_bp_override) {
2690 zio_pop_transforms(zio);
2691 zio->io_stage = ZIO_STAGE_OPEN;
2692 zio->io_pipeline = ZIO_WRITE_PIPELINE;
2693 zio->io_bp_override = NULL;
2694 BP_ZERO(bp);
2695 ddt_exit(ddt);
2696 return (ZIO_PIPELINE_CONTINUE);
2697 }
2698
2699 dio = zio_write(zio, spa, txg, bp, zio->io_orig_data,
2700 zio->io_orig_size, &czp, NULL, NULL,
2701 NULL, zio_ddt_ditto_write_done, dde, zio->io_priority,
2702 ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark);
2703
2704 zio_push_transform(dio, zio->io_data, zio->io_size, 0, NULL);
2705 dde->dde_lead_zio[DDT_PHYS_DITTO] = dio;
2706 }
2707
2708 if (ddp->ddp_phys_birth != 0 || dde->dde_lead_zio[p] != NULL) {
2709 if (ddp->ddp_phys_birth != 0)
2710 ddt_bp_fill(ddp, bp, txg);
2711 if (dde->dde_lead_zio[p] != NULL)
2712 zio_add_child(zio, dde->dde_lead_zio[p]);
2713 else
2714 ddt_phys_addref(ddp);
2715 } else if (zio->io_bp_override) {
2716 ASSERT(bp->blk_birth == txg);
2717 ASSERT(BP_EQUAL(bp, zio->io_bp_override));
2718 ddt_phys_fill(ddp, bp);
2719 ddt_phys_addref(ddp);
2720 } else {
2721 cio = zio_write(zio, spa, txg, bp, zio->io_orig_data,
2722 zio->io_orig_size, zp,
2723 zio_ddt_child_write_ready, NULL, NULL,
2724 zio_ddt_child_write_done, dde, zio->io_priority,
2725 ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark);
2726
2727 zio_push_transform(cio, zio->io_data, zio->io_size, 0, NULL);
2728 dde->dde_lead_zio[p] = cio;
2729 }
2730
2731 ddt_exit(ddt);
2732
2733 if (cio)
2734 zio_nowait(cio);
2735 if (dio)
2736 zio_nowait(dio);
2737
2738 return (ZIO_PIPELINE_CONTINUE);
2739 }
2740
2741 ddt_entry_t *freedde; /* for debugging */
2742
2743 static int
zio_ddt_free(zio_t * zio)2744 zio_ddt_free(zio_t *zio)
2745 {
2746 spa_t *spa = zio->io_spa;
2747 blkptr_t *bp = zio->io_bp;
2748 ddt_t *ddt = ddt_select(spa, bp);
2749 ddt_entry_t *dde;
2750 ddt_phys_t *ddp;
2751
2752 ASSERT(BP_GET_DEDUP(bp));
2753 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
2754
2755 ddt_enter(ddt);
2756 freedde = dde = ddt_lookup(ddt, bp, B_TRUE);
2757 ddp = ddt_phys_select(dde, bp);
2758 ddt_phys_decref(ddp);
2759 ddt_exit(ddt);
2760
2761 return (ZIO_PIPELINE_CONTINUE);
2762 }
2763
2764 /*
2765 * ==========================================================================
2766 * Allocate and free blocks
2767 * ==========================================================================
2768 */
2769
2770 static zio_t *
zio_io_to_allocate(spa_t * spa)2771 zio_io_to_allocate(spa_t *spa)
2772 {
2773 zio_t *zio;
2774
2775 ASSERT(MUTEX_HELD(&spa->spa_alloc_lock));
2776
2777 zio = avl_first(&spa->spa_alloc_tree);
2778 if (zio == NULL)
2779 return (NULL);
2780
2781 ASSERT(IO_IS_ALLOCATING(zio));
2782
2783 /*
2784 * Try to place a reservation for this zio. If we're unable to
2785 * reserve then we throttle.
2786 */
2787 if (!metaslab_class_throttle_reserve(spa_normal_class(spa),
2788 zio->io_prop.zp_copies, zio, 0)) {
2789 return (NULL);
2790 }
2791
2792 avl_remove(&spa->spa_alloc_tree, zio);
2793 ASSERT3U(zio->io_stage, <, ZIO_STAGE_DVA_ALLOCATE);
2794
2795 return (zio);
2796 }
2797
2798 static int
zio_dva_throttle(zio_t * zio)2799 zio_dva_throttle(zio_t *zio)
2800 {
2801 spa_t *spa = zio->io_spa;
2802 zio_t *nio;
2803
2804 if (zio->io_priority == ZIO_PRIORITY_SYNC_WRITE ||
2805 !spa_normal_class(zio->io_spa)->mc_alloc_throttle_enabled ||
2806 zio->io_child_type == ZIO_CHILD_GANG ||
2807 zio->io_flags & ZIO_FLAG_NODATA) {
2808 return (ZIO_PIPELINE_CONTINUE);
2809 }
2810
2811 ASSERT(zio->io_child_type > ZIO_CHILD_GANG);
2812
2813 ASSERT3U(zio->io_queued_timestamp, >, 0);
2814 ASSERT(zio->io_stage == ZIO_STAGE_DVA_THROTTLE);
2815
2816 mutex_enter(&spa->spa_alloc_lock);
2817
2818 ASSERT(zio->io_type == ZIO_TYPE_WRITE);
2819 avl_add(&spa->spa_alloc_tree, zio);
2820
2821 nio = zio_io_to_allocate(zio->io_spa);
2822 mutex_exit(&spa->spa_alloc_lock);
2823
2824 if (nio == zio)
2825 return (ZIO_PIPELINE_CONTINUE);
2826
2827 if (nio != NULL) {
2828 ASSERT3U(nio->io_queued_timestamp, <=,
2829 zio->io_queued_timestamp);
2830 ASSERT(nio->io_stage == ZIO_STAGE_DVA_THROTTLE);
2831 /*
2832 * We are passing control to a new zio so make sure that
2833 * it is processed by a different thread. We do this to
2834 * avoid stack overflows that can occur when parents are
2835 * throttled and children are making progress. We allow
2836 * it to go to the head of the taskq since it's already
2837 * been waiting.
2838 */
2839 zio_taskq_dispatch(nio, ZIO_TASKQ_ISSUE, B_TRUE);
2840 }
2841 return (ZIO_PIPELINE_STOP);
2842 }
2843
2844 void
zio_allocate_dispatch(spa_t * spa)2845 zio_allocate_dispatch(spa_t *spa)
2846 {
2847 zio_t *zio;
2848
2849 mutex_enter(&spa->spa_alloc_lock);
2850 zio = zio_io_to_allocate(spa);
2851 mutex_exit(&spa->spa_alloc_lock);
2852 if (zio == NULL)
2853 return;
2854
2855 ASSERT3U(zio->io_stage, ==, ZIO_STAGE_DVA_THROTTLE);
2856 ASSERT0(zio->io_error);
2857 zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_TRUE);
2858 }
2859
2860 static int
zio_dva_allocate(zio_t * zio)2861 zio_dva_allocate(zio_t *zio)
2862 {
2863 spa_t *spa = zio->io_spa;
2864 metaslab_class_t *mc = spa_normal_class(spa);
2865 blkptr_t *bp = zio->io_bp;
2866 int error;
2867 int flags = 0;
2868
2869 if (zio->io_gang_leader == NULL) {
2870 ASSERT(zio->io_child_type > ZIO_CHILD_GANG);
2871 zio->io_gang_leader = zio;
2872 }
2873
2874 ASSERT(BP_IS_HOLE(bp));
2875 ASSERT0(BP_GET_NDVAS(bp));
2876 ASSERT3U(zio->io_prop.zp_copies, >, 0);
2877 ASSERT3U(zio->io_prop.zp_copies, <=, spa_max_replication(spa));
2878 ASSERT3U(zio->io_size, ==, BP_GET_PSIZE(bp));
2879
2880 if (zio->io_flags & ZIO_FLAG_NODATA) {
2881 flags |= METASLAB_DONT_THROTTLE;
2882 }
2883 if (zio->io_flags & ZIO_FLAG_GANG_CHILD) {
2884 flags |= METASLAB_GANG_CHILD;
2885 }
2886 if (zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE) {
2887 flags |= METASLAB_ASYNC_ALLOC;
2888 }
2889
2890 error = metaslab_alloc(spa, mc, zio->io_size, bp,
2891 zio->io_prop.zp_copies, zio->io_txg, NULL, flags,
2892 &zio->io_alloc_list, zio);
2893
2894 if (error != 0) {
2895 spa_dbgmsg(spa, "%s: metaslab allocation failure: zio %p, "
2896 "size %llu, error %d", spa_name(spa), zio, zio->io_size,
2897 error);
2898 if (error == ENOSPC && zio->io_size > SPA_MINBLOCKSIZE)
2899 return (zio_write_gang_block(zio));
2900 zio->io_error = error;
2901 }
2902
2903 return (ZIO_PIPELINE_CONTINUE);
2904 }
2905
2906 static int
zio_dva_free(zio_t * zio)2907 zio_dva_free(zio_t *zio)
2908 {
2909 metaslab_free(zio->io_spa, zio->io_bp, zio->io_txg, B_FALSE);
2910
2911 return (ZIO_PIPELINE_CONTINUE);
2912 }
2913
2914 static int
zio_dva_claim(zio_t * zio)2915 zio_dva_claim(zio_t *zio)
2916 {
2917 int error;
2918
2919 error = metaslab_claim(zio->io_spa, zio->io_bp, zio->io_txg);
2920 if (error)
2921 zio->io_error = error;
2922
2923 return (ZIO_PIPELINE_CONTINUE);
2924 }
2925
2926 /*
2927 * Undo an allocation. This is used by zio_done() when an I/O fails
2928 * and we want to give back the block we just allocated.
2929 * This handles both normal blocks and gang blocks.
2930 */
2931 static void
zio_dva_unallocate(zio_t * zio,zio_gang_node_t * gn,blkptr_t * bp)2932 zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp)
2933 {
2934 ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp));
2935 ASSERT(zio->io_bp_override == NULL);
2936
2937 if (!BP_IS_HOLE(bp))
2938 metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE);
2939
2940 if (gn != NULL) {
2941 for (int g = 0; g < SPA_GBH_NBLKPTRS; g++) {
2942 zio_dva_unallocate(zio, gn->gn_child[g],
2943 &gn->gn_gbh->zg_blkptr[g]);
2944 }
2945 }
2946 }
2947
2948 /*
2949 * Try to allocate an intent log block. Return 0 on success, errno on failure.
2950 */
2951 int
zio_alloc_zil(spa_t * spa,uint64_t txg,blkptr_t * new_bp,blkptr_t * old_bp,uint64_t size,boolean_t * slog)2952 zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, blkptr_t *old_bp,
2953 uint64_t size, boolean_t *slog)
2954 {
2955 int error = 1;
2956 zio_alloc_list_t io_alloc_list;
2957
2958 ASSERT(txg > spa_syncing_txg(spa));
2959
2960 metaslab_trace_init(&io_alloc_list);
2961 error = metaslab_alloc(spa, spa_log_class(spa), size, new_bp, 1,
2962 txg, old_bp, METASLAB_HINTBP_AVOID, &io_alloc_list, NULL);
2963 if (error == 0) {
2964 *slog = TRUE;
2965 } else {
2966 error = metaslab_alloc(spa, spa_normal_class(spa), size,
2967 new_bp, 1, txg, old_bp, METASLAB_HINTBP_AVOID,
2968 &io_alloc_list, NULL);
2969 if (error == 0)
2970 *slog = FALSE;
2971 }
2972 metaslab_trace_fini(&io_alloc_list);
2973
2974 if (error == 0) {
2975 BP_SET_LSIZE(new_bp, size);
2976 BP_SET_PSIZE(new_bp, size);
2977 BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF);
2978 BP_SET_CHECKSUM(new_bp,
2979 spa_version(spa) >= SPA_VERSION_SLIM_ZIL
2980 ? ZIO_CHECKSUM_ZILOG2 : ZIO_CHECKSUM_ZILOG);
2981 BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG);
2982 BP_SET_LEVEL(new_bp, 0);
2983 BP_SET_DEDUP(new_bp, 0);
2984 BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER);
2985 }
2986
2987 return (error);
2988 }
2989
2990 /*
2991 * Free an intent log block.
2992 */
2993 void
zio_free_zil(spa_t * spa,uint64_t txg,blkptr_t * bp)2994 zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp)
2995 {
2996 ASSERT(BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG);
2997 ASSERT(!BP_IS_GANG(bp));
2998
2999 zio_free(spa, txg, bp);
3000 }
3001
3002 /*
3003 * ==========================================================================
3004 * Read, write and delete to physical devices
3005 * ==========================================================================
3006 */
3007
3008
3009 /*
3010 * Issue an I/O to the underlying vdev. Typically the issue pipeline
3011 * stops after this stage and will resume upon I/O completion.
3012 * However, there are instances where the vdev layer may need to
3013 * continue the pipeline when an I/O was not issued. Since the I/O
3014 * that was sent to the vdev layer might be different than the one
3015 * currently active in the pipeline (see vdev_queue_io()), we explicitly
3016 * force the underlying vdev layers to call either zio_execute() or
3017 * zio_interrupt() to ensure that the pipeline continues with the correct I/O.
3018 */
3019 static int
zio_vdev_io_start(zio_t * zio)3020 zio_vdev_io_start(zio_t *zio)
3021 {
3022 vdev_t *vd = zio->io_vd;
3023 uint64_t align;
3024 spa_t *spa = zio->io_spa;
3025 int ret;
3026
3027 ASSERT(zio->io_error == 0);
3028 ASSERT(zio->io_child_error[ZIO_CHILD_VDEV] == 0);
3029
3030 if (vd == NULL) {
3031 if (!(zio->io_flags & ZIO_FLAG_CONFIG_WRITER))
3032 spa_config_enter(spa, SCL_ZIO, zio, RW_READER);
3033
3034 /*
3035 * The mirror_ops handle multiple DVAs in a single BP.
3036 */
3037 vdev_mirror_ops.vdev_op_io_start(zio);
3038 return (ZIO_PIPELINE_STOP);
3039 }
3040
3041 if (vd->vdev_ops->vdev_op_leaf && zio->io_type == ZIO_TYPE_FREE &&
3042 zio->io_priority == ZIO_PRIORITY_NOW) {
3043 trim_map_free(vd, zio->io_offset, zio->io_size, zio->io_txg);
3044 return (ZIO_PIPELINE_CONTINUE);
3045 }
3046
3047 ASSERT3P(zio->io_logical, !=, zio);
3048
3049 /*
3050 * We keep track of time-sensitive I/Os so that the scan thread
3051 * can quickly react to certain workloads. In particular, we care
3052 * about non-scrubbing, top-level reads and writes with the following
3053 * characteristics:
3054 * - synchronous writes of user data to non-slog devices
3055 * - any reads of user data
3056 * When these conditions are met, adjust the timestamp of spa_last_io
3057 * which allows the scan thread to adjust its workload accordingly.
3058 */
3059 if (!(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_bp != NULL &&
3060 vd == vd->vdev_top && !vd->vdev_islog &&
3061 zio->io_bookmark.zb_objset != DMU_META_OBJSET &&
3062 zio->io_txg != spa_syncing_txg(spa)) {
3063 uint64_t old = spa->spa_last_io;
3064 uint64_t new = ddi_get_lbolt64();
3065 if (old != new)
3066 (void) atomic_cas_64(&spa->spa_last_io, old, new);
3067 }
3068
3069 align = 1ULL << vd->vdev_top->vdev_ashift;
3070
3071 if (!(zio->io_flags & ZIO_FLAG_PHYSICAL) &&
3072 P2PHASE(zio->io_size, align) != 0) {
3073 /* Transform logical writes to be a full physical block size. */
3074 uint64_t asize = P2ROUNDUP(zio->io_size, align);
3075 char *abuf = NULL;
3076 if (zio->io_type == ZIO_TYPE_READ ||
3077 zio->io_type == ZIO_TYPE_WRITE)
3078 abuf = zio_buf_alloc(asize);
3079 ASSERT(vd == vd->vdev_top);
3080 if (zio->io_type == ZIO_TYPE_WRITE) {
3081 bcopy(zio->io_data, abuf, zio->io_size);
3082 bzero(abuf + zio->io_size, asize - zio->io_size);
3083 }
3084 zio_push_transform(zio, abuf, asize, abuf ? asize : 0,
3085 zio_subblock);
3086 }
3087
3088 /*
3089 * If this is not a physical io, make sure that it is properly aligned
3090 * before proceeding.
3091 */
3092 if (!(zio->io_flags & ZIO_FLAG_PHYSICAL)) {
3093 ASSERT0(P2PHASE(zio->io_offset, align));
3094 ASSERT0(P2PHASE(zio->io_size, align));
3095 } else {
3096 /*
3097 * For the physical io we allow alignment
3098 * to a logical block size.
3099 */
3100 uint64_t log_align =
3101 1ULL << vd->vdev_top->vdev_logical_ashift;
3102 ASSERT0(P2PHASE(zio->io_offset, log_align));
3103 ASSERT0(P2PHASE(zio->io_size, log_align));
3104 }
3105
3106 VERIFY(zio->io_type == ZIO_TYPE_READ || spa_writeable(spa));
3107
3108 /*
3109 * If this is a repair I/O, and there's no self-healing involved --
3110 * that is, we're just resilvering what we expect to resilver --
3111 * then don't do the I/O unless zio's txg is actually in vd's DTL.
3112 * This prevents spurious resilvering with nested replication.
3113 * For example, given a mirror of mirrors, (A+B)+(C+D), if only
3114 * A is out of date, we'll read from C+D, then use the data to
3115 * resilver A+B -- but we don't actually want to resilver B, just A.
3116 * The top-level mirror has no way to know this, so instead we just
3117 * discard unnecessary repairs as we work our way down the vdev tree.
3118 * The same logic applies to any form of nested replication:
3119 * ditto + mirror, RAID-Z + replacing, etc. This covers them all.
3120 */
3121 if ((zio->io_flags & ZIO_FLAG_IO_REPAIR) &&
3122 !(zio->io_flags & ZIO_FLAG_SELF_HEAL) &&
3123 zio->io_txg != 0 && /* not a delegated i/o */
3124 !vdev_dtl_contains(vd, DTL_PARTIAL, zio->io_txg, 1)) {
3125 ASSERT(zio->io_type == ZIO_TYPE_WRITE);
3126 zio_vdev_io_bypass(zio);
3127 return (ZIO_PIPELINE_CONTINUE);
3128 }
3129
3130 if (vd->vdev_ops->vdev_op_leaf) {
3131 switch (zio->io_type) {
3132 case ZIO_TYPE_READ:
3133 if (vdev_cache_read(zio))
3134 return (ZIO_PIPELINE_CONTINUE);
3135 /* FALLTHROUGH */
3136 case ZIO_TYPE_WRITE:
3137 case ZIO_TYPE_FREE:
3138 if ((zio = vdev_queue_io(zio)) == NULL)
3139 return (ZIO_PIPELINE_STOP);
3140
3141 if (!vdev_accessible(vd, zio)) {
3142 zio->io_error = SET_ERROR(ENXIO);
3143 zio_interrupt(zio);
3144 return (ZIO_PIPELINE_STOP);
3145 }
3146 break;
3147 }
3148 /*
3149 * Note that we ignore repair writes for TRIM because they can
3150 * conflict with normal writes. This isn't an issue because, by
3151 * definition, we only repair blocks that aren't freed.
3152 */
3153 if (zio->io_type == ZIO_TYPE_WRITE &&
3154 !(zio->io_flags & ZIO_FLAG_IO_REPAIR) &&
3155 !trim_map_write_start(zio))
3156 return (ZIO_PIPELINE_STOP);
3157 }
3158
3159 vd->vdev_ops->vdev_op_io_start(zio);
3160 return (ZIO_PIPELINE_STOP);
3161 }
3162
3163 static int
zio_vdev_io_done(zio_t * zio)3164 zio_vdev_io_done(zio_t *zio)
3165 {
3166 vdev_t *vd = zio->io_vd;
3167 vdev_ops_t *ops = vd ? vd->vdev_ops : &vdev_mirror_ops;
3168 boolean_t unexpected_error = B_FALSE;
3169
3170 if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE))
3171 return (ZIO_PIPELINE_STOP);
3172
3173 ASSERT(zio->io_type == ZIO_TYPE_READ ||
3174 zio->io_type == ZIO_TYPE_WRITE || zio->io_type == ZIO_TYPE_FREE);
3175
3176 if (vd != NULL && vd->vdev_ops->vdev_op_leaf &&
3177 (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE ||
3178 zio->io_type == ZIO_TYPE_FREE)) {
3179
3180 if (zio->io_type == ZIO_TYPE_WRITE &&
3181 !(zio->io_flags & ZIO_FLAG_IO_REPAIR))
3182 trim_map_write_done(zio);
3183
3184 vdev_queue_io_done(zio);
3185
3186 if (zio->io_type == ZIO_TYPE_WRITE)
3187 vdev_cache_write(zio);
3188
3189 if (zio_injection_enabled && zio->io_error == 0)
3190 zio->io_error = zio_handle_device_injection(vd,
3191 zio, EIO);
3192
3193 if (zio_injection_enabled && zio->io_error == 0)
3194 zio->io_error = zio_handle_label_injection(zio, EIO);
3195
3196 if (zio->io_error) {
3197 if (zio->io_error == ENOTSUP &&
3198 zio->io_type == ZIO_TYPE_FREE) {
3199 /* Not all devices support TRIM. */
3200 } else if (!vdev_accessible(vd, zio)) {
3201 zio->io_error = SET_ERROR(ENXIO);
3202 } else {
3203 unexpected_error = B_TRUE;
3204 }
3205 }
3206 }
3207
3208 ops->vdev_op_io_done(zio);
3209
3210 if (unexpected_error)
3211 VERIFY(vdev_probe(vd, zio) == NULL);
3212
3213 return (ZIO_PIPELINE_CONTINUE);
3214 }
3215
3216 /*
3217 * For non-raidz ZIOs, we can just copy aside the bad data read from the
3218 * disk, and use that to finish the checksum ereport later.
3219 */
3220 static void
zio_vsd_default_cksum_finish(zio_cksum_report_t * zcr,const void * good_buf)3221 zio_vsd_default_cksum_finish(zio_cksum_report_t *zcr,
3222 const void *good_buf)
3223 {
3224 /* no processing needed */
3225 zfs_ereport_finish_checksum(zcr, good_buf, zcr->zcr_cbdata, B_FALSE);
3226 }
3227
3228 /*ARGSUSED*/
3229 void
zio_vsd_default_cksum_report(zio_t * zio,zio_cksum_report_t * zcr,void * ignored)3230 zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *ignored)
3231 {
3232 void *buf = zio_buf_alloc(zio->io_size);
3233
3234 bcopy(zio->io_data, buf, zio->io_size);
3235
3236 zcr->zcr_cbinfo = zio->io_size;
3237 zcr->zcr_cbdata = buf;
3238 zcr->zcr_finish = zio_vsd_default_cksum_finish;
3239 zcr->zcr_free = zio_buf_free;
3240 }
3241
3242 static int
zio_vdev_io_assess(zio_t * zio)3243 zio_vdev_io_assess(zio_t *zio)
3244 {
3245 vdev_t *vd = zio->io_vd;
3246
3247 if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE))
3248 return (ZIO_PIPELINE_STOP);
3249
3250 if (vd == NULL && !(zio->io_flags & ZIO_FLAG_CONFIG_WRITER))
3251 spa_config_exit(zio->io_spa, SCL_ZIO, zio);
3252
3253 if (zio->io_vsd != NULL) {
3254 zio->io_vsd_ops->vsd_free(zio);
3255 zio->io_vsd = NULL;
3256 }
3257
3258 if (zio_injection_enabled && zio->io_error == 0)
3259 zio->io_error = zio_handle_fault_injection(zio, EIO);
3260
3261 if (zio->io_type == ZIO_TYPE_FREE &&
3262 zio->io_priority != ZIO_PRIORITY_NOW) {
3263 switch (zio->io_error) {
3264 case 0:
3265 ZIO_TRIM_STAT_INCR(bytes, zio->io_size);
3266 ZIO_TRIM_STAT_BUMP(success);
3267 break;
3268 case EOPNOTSUPP:
3269 ZIO_TRIM_STAT_BUMP(unsupported);
3270 break;
3271 default:
3272 ZIO_TRIM_STAT_BUMP(failed);
3273 break;
3274 }
3275 }
3276
3277 /*
3278 * If the I/O failed, determine whether we should attempt to retry it.
3279 *
3280 * On retry, we cut in line in the issue queue, since we don't want
3281 * compression/checksumming/etc. work to prevent our (cheap) IO reissue.
3282 */
3283 if (zio->io_error && vd == NULL &&
3284 !(zio->io_flags & (ZIO_FLAG_DONT_RETRY | ZIO_FLAG_IO_RETRY))) {
3285 ASSERT(!(zio->io_flags & ZIO_FLAG_DONT_QUEUE)); /* not a leaf */
3286 ASSERT(!(zio->io_flags & ZIO_FLAG_IO_BYPASS)); /* not a leaf */
3287 zio->io_error = 0;
3288 zio->io_flags |= ZIO_FLAG_IO_RETRY |
3289 ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE;
3290 zio->io_stage = ZIO_STAGE_VDEV_IO_START >> 1;
3291 zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE,
3292 zio_requeue_io_start_cut_in_line);
3293 return (ZIO_PIPELINE_STOP);
3294 }
3295
3296 /*
3297 * If we got an error on a leaf device, convert it to ENXIO
3298 * if the device is not accessible at all.
3299 */
3300 if (zio->io_error && vd != NULL && vd->vdev_ops->vdev_op_leaf &&
3301 !vdev_accessible(vd, zio))
3302 zio->io_error = SET_ERROR(ENXIO);
3303
3304 /*
3305 * If we can't write to an interior vdev (mirror or RAID-Z),
3306 * set vdev_cant_write so that we stop trying to allocate from it.
3307 */
3308 if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE &&
3309 vd != NULL && !vd->vdev_ops->vdev_op_leaf) {
3310 vd->vdev_cant_write = B_TRUE;
3311 }
3312
3313 if (zio->io_error)
3314 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
3315
3316 if (vd != NULL && vd->vdev_ops->vdev_op_leaf &&
3317 zio->io_physdone != NULL) {
3318 ASSERT(!(zio->io_flags & ZIO_FLAG_DELEGATED));
3319 ASSERT(zio->io_child_type == ZIO_CHILD_VDEV);
3320 zio->io_physdone(zio->io_logical);
3321 }
3322
3323 return (ZIO_PIPELINE_CONTINUE);
3324 }
3325
3326 void
zio_vdev_io_reissue(zio_t * zio)3327 zio_vdev_io_reissue(zio_t *zio)
3328 {
3329 ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START);
3330 ASSERT(zio->io_error == 0);
3331
3332 zio->io_stage >>= 1;
3333 }
3334
3335 void
zio_vdev_io_redone(zio_t * zio)3336 zio_vdev_io_redone(zio_t *zio)
3337 {
3338 ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_DONE);
3339
3340 zio->io_stage >>= 1;
3341 }
3342
3343 void
zio_vdev_io_bypass(zio_t * zio)3344 zio_vdev_io_bypass(zio_t *zio)
3345 {
3346 ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START);
3347 ASSERT(zio->io_error == 0);
3348
3349 zio->io_flags |= ZIO_FLAG_IO_BYPASS;
3350 zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS >> 1;
3351 }
3352
3353 /*
3354 * ==========================================================================
3355 * Generate and verify checksums
3356 * ==========================================================================
3357 */
3358 static int
zio_checksum_generate(zio_t * zio)3359 zio_checksum_generate(zio_t *zio)
3360 {
3361 blkptr_t *bp = zio->io_bp;
3362 enum zio_checksum checksum;
3363
3364 if (bp == NULL) {
3365 /*
3366 * This is zio_write_phys().
3367 * We're either generating a label checksum, or none at all.
3368 */
3369 checksum = zio->io_prop.zp_checksum;
3370
3371 if (checksum == ZIO_CHECKSUM_OFF)
3372 return (ZIO_PIPELINE_CONTINUE);
3373
3374 ASSERT(checksum == ZIO_CHECKSUM_LABEL);
3375 } else {
3376 if (BP_IS_GANG(bp) && zio->io_child_type == ZIO_CHILD_GANG) {
3377 ASSERT(!IO_IS_ALLOCATING(zio));
3378 checksum = ZIO_CHECKSUM_GANG_HEADER;
3379 } else {
3380 checksum = BP_GET_CHECKSUM(bp);
3381 }
3382 }
3383
3384 zio_checksum_compute(zio, checksum, zio->io_data, zio->io_size);
3385
3386 return (ZIO_PIPELINE_CONTINUE);
3387 }
3388
3389 static int
zio_checksum_verify(zio_t * zio)3390 zio_checksum_verify(zio_t *zio)
3391 {
3392 zio_bad_cksum_t info;
3393 blkptr_t *bp = zio->io_bp;
3394 int error;
3395
3396 ASSERT(zio->io_vd != NULL);
3397
3398 if (bp == NULL) {
3399 /*
3400 * This is zio_read_phys().
3401 * We're either verifying a label checksum, or nothing at all.
3402 */
3403 if (zio->io_prop.zp_checksum == ZIO_CHECKSUM_OFF)
3404 return (ZIO_PIPELINE_CONTINUE);
3405
3406 ASSERT(zio->io_prop.zp_checksum == ZIO_CHECKSUM_LABEL);
3407 }
3408
3409 if ((error = zio_checksum_error(zio, &info)) != 0) {
3410 zio->io_error = error;
3411 if (error == ECKSUM &&
3412 !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
3413 zfs_ereport_start_checksum(zio->io_spa,
3414 zio->io_vd, zio, zio->io_offset,
3415 zio->io_size, NULL, &info);
3416 }
3417 }
3418
3419 return (ZIO_PIPELINE_CONTINUE);
3420 }
3421
3422 /*
3423 * Called by RAID-Z to ensure we don't compute the checksum twice.
3424 */
3425 void
zio_checksum_verified(zio_t * zio)3426 zio_checksum_verified(zio_t *zio)
3427 {
3428 zio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY;
3429 }
3430
3431 /*
3432 * ==========================================================================
3433 * Error rank. Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other.
3434 * An error of 0 indicates success. ENXIO indicates whole-device failure,
3435 * which may be transient (e.g. unplugged) or permament. ECKSUM and EIO
3436 * indicate errors that are specific to one I/O, and most likely permanent.
3437 * Any other error is presumed to be worse because we weren't expecting it.
3438 * ==========================================================================
3439 */
3440 int
zio_worst_error(int e1,int e2)3441 zio_worst_error(int e1, int e2)
3442 {
3443 static int zio_error_rank[] = { 0, ENXIO, ECKSUM, EIO };
3444 int r1, r2;
3445
3446 for (r1 = 0; r1 < sizeof (zio_error_rank) / sizeof (int); r1++)
3447 if (e1 == zio_error_rank[r1])
3448 break;
3449
3450 for (r2 = 0; r2 < sizeof (zio_error_rank) / sizeof (int); r2++)
3451 if (e2 == zio_error_rank[r2])
3452 break;
3453
3454 return (r1 > r2 ? e1 : e2);
3455 }
3456
3457 /*
3458 * ==========================================================================
3459 * I/O completion
3460 * ==========================================================================
3461 */
3462 static int
zio_ready(zio_t * zio)3463 zio_ready(zio_t *zio)
3464 {
3465 blkptr_t *bp = zio->io_bp;
3466 zio_t *pio, *pio_next;
3467 zio_link_t *zl = NULL;
3468
3469 if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) ||
3470 zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_READY))
3471 return (ZIO_PIPELINE_STOP);
3472
3473 if (zio->io_ready) {
3474 ASSERT(IO_IS_ALLOCATING(zio));
3475 ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp) ||
3476 (zio->io_flags & ZIO_FLAG_NOPWRITE));
3477 ASSERT(zio->io_children[ZIO_CHILD_GANG][ZIO_WAIT_READY] == 0);
3478
3479 zio->io_ready(zio);
3480 }
3481
3482 if (bp != NULL && bp != &zio->io_bp_copy)
3483 zio->io_bp_copy = *bp;
3484
3485 if (zio->io_error != 0) {
3486 zio->io_pipeline = ZIO_INTERLOCK_PIPELINE;
3487
3488 if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING) {
3489 ASSERT(IO_IS_ALLOCATING(zio));
3490 ASSERT(zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE);
3491 /*
3492 * We were unable to allocate anything, unreserve and
3493 * issue the next I/O to allocate.
3494 */
3495 metaslab_class_throttle_unreserve(
3496 spa_normal_class(zio->io_spa),
3497 zio->io_prop.zp_copies, zio);
3498 zio_allocate_dispatch(zio->io_spa);
3499 }
3500 }
3501
3502 mutex_enter(&zio->io_lock);
3503 zio->io_state[ZIO_WAIT_READY] = 1;
3504 pio = zio_walk_parents(zio, &zl);
3505 mutex_exit(&zio->io_lock);
3506
3507 /*
3508 * As we notify zio's parents, new parents could be added.
3509 * New parents go to the head of zio's io_parent_list, however,
3510 * so we will (correctly) not notify them. The remainder of zio's
3511 * io_parent_list, from 'pio_next' onward, cannot change because
3512 * all parents must wait for us to be done before they can be done.
3513 */
3514 for (; pio != NULL; pio = pio_next) {
3515 pio_next = zio_walk_parents(zio, &zl);
3516 zio_notify_parent(pio, zio, ZIO_WAIT_READY);
3517 }
3518
3519 if (zio->io_flags & ZIO_FLAG_NODATA) {
3520 if (BP_IS_GANG(bp)) {
3521 zio->io_flags &= ~ZIO_FLAG_NODATA;
3522 } else {
3523 ASSERT((uintptr_t)zio->io_data < SPA_MAXBLOCKSIZE);
3524 zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES;
3525 }
3526 }
3527
3528 if (zio_injection_enabled &&
3529 zio->io_spa->spa_syncing_txg == zio->io_txg)
3530 zio_handle_ignored_writes(zio);
3531
3532 return (ZIO_PIPELINE_CONTINUE);
3533 }
3534
3535 /*
3536 * Update the allocation throttle accounting.
3537 */
3538 static void
zio_dva_throttle_done(zio_t * zio)3539 zio_dva_throttle_done(zio_t *zio)
3540 {
3541 zio_t *lio = zio->io_logical;
3542 zio_t *pio = zio_unique_parent(zio);
3543 vdev_t *vd = zio->io_vd;
3544 int flags = METASLAB_ASYNC_ALLOC;
3545
3546 ASSERT3P(zio->io_bp, !=, NULL);
3547 ASSERT3U(zio->io_type, ==, ZIO_TYPE_WRITE);
3548 ASSERT3U(zio->io_priority, ==, ZIO_PRIORITY_ASYNC_WRITE);
3549 ASSERT3U(zio->io_child_type, ==, ZIO_CHILD_VDEV);
3550 ASSERT(vd != NULL);
3551 ASSERT3P(vd, ==, vd->vdev_top);
3552 ASSERT(!(zio->io_flags & (ZIO_FLAG_IO_REPAIR | ZIO_FLAG_IO_RETRY)));
3553 ASSERT(zio->io_flags & ZIO_FLAG_IO_ALLOCATING);
3554 ASSERT(!(lio->io_flags & ZIO_FLAG_IO_REWRITE));
3555 ASSERT(!(lio->io_orig_flags & ZIO_FLAG_NODATA));
3556
3557 /*
3558 * Parents of gang children can have two flavors -- ones that
3559 * allocated the gang header (will have ZIO_FLAG_IO_REWRITE set)
3560 * and ones that allocated the constituent blocks. The allocation
3561 * throttle needs to know the allocating parent zio so we must find
3562 * it here.
3563 */
3564 if (pio->io_child_type == ZIO_CHILD_GANG) {
3565 /*
3566 * If our parent is a rewrite gang child then our grandparent
3567 * would have been the one that performed the allocation.
3568 */
3569 if (pio->io_flags & ZIO_FLAG_IO_REWRITE)
3570 pio = zio_unique_parent(pio);
3571 flags |= METASLAB_GANG_CHILD;
3572 }
3573
3574 ASSERT(IO_IS_ALLOCATING(pio));
3575 ASSERT3P(zio, !=, zio->io_logical);
3576 ASSERT(zio->io_logical != NULL);
3577 ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REPAIR));
3578 ASSERT0(zio->io_flags & ZIO_FLAG_NOPWRITE);
3579
3580 mutex_enter(&pio->io_lock);
3581 metaslab_group_alloc_decrement(zio->io_spa, vd->vdev_id, pio, flags);
3582 mutex_exit(&pio->io_lock);
3583
3584 metaslab_class_throttle_unreserve(spa_normal_class(zio->io_spa),
3585 1, pio);
3586
3587 /*
3588 * Call into the pipeline to see if there is more work that
3589 * needs to be done. If there is work to be done it will be
3590 * dispatched to another taskq thread.
3591 */
3592 zio_allocate_dispatch(zio->io_spa);
3593 }
3594
3595 static int
zio_done(zio_t * zio)3596 zio_done(zio_t *zio)
3597 {
3598 spa_t *spa = zio->io_spa;
3599 zio_t *lio = zio->io_logical;
3600 blkptr_t *bp = zio->io_bp;
3601 vdev_t *vd = zio->io_vd;
3602 uint64_t psize = zio->io_size;
3603 zio_t *pio, *pio_next;
3604 metaslab_class_t *mc = spa_normal_class(spa);
3605 zio_link_t *zl = NULL;
3606
3607 /*
3608 * If our children haven't all completed,
3609 * wait for them and then repeat this pipeline stage.
3610 */
3611 if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE) ||
3612 zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE) ||
3613 zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE) ||
3614 zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_DONE))
3615 return (ZIO_PIPELINE_STOP);
3616
3617 /*
3618 * If the allocation throttle is enabled, then update the accounting.
3619 * We only track child I/Os that are part of an allocating async
3620 * write. We must do this since the allocation is performed
3621 * by the logical I/O but the actual write is done by child I/Os.
3622 */
3623 if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING &&
3624 zio->io_child_type == ZIO_CHILD_VDEV) {
3625 ASSERT(mc->mc_alloc_throttle_enabled);
3626 zio_dva_throttle_done(zio);
3627 }
3628
3629 /*
3630 * If the allocation throttle is enabled, verify that
3631 * we have decremented the refcounts for every I/O that was throttled.
3632 */
3633 if (zio->io_flags & ZIO_FLAG_IO_ALLOCATING) {
3634 ASSERT(zio->io_type == ZIO_TYPE_WRITE);
3635 ASSERT(zio->io_priority == ZIO_PRIORITY_ASYNC_WRITE);
3636 ASSERT(bp != NULL);
3637 metaslab_group_alloc_verify(spa, zio->io_bp, zio);
3638 VERIFY(refcount_not_held(&mc->mc_alloc_slots, zio));
3639 }
3640
3641 for (int c = 0; c < ZIO_CHILD_TYPES; c++)
3642 for (int w = 0; w < ZIO_WAIT_TYPES; w++)
3643 ASSERT(zio->io_children[c][w] == 0);
3644
3645 if (bp != NULL && !BP_IS_EMBEDDED(bp)) {
3646 ASSERT(bp->blk_pad[0] == 0);
3647 ASSERT(bp->blk_pad[1] == 0);
3648 ASSERT(bcmp(bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 ||
3649 (bp == zio_unique_parent(zio)->io_bp));
3650 if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(bp) &&
3651 zio->io_bp_override == NULL &&
3652 !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) {
3653 ASSERT(!BP_SHOULD_BYTESWAP(bp));
3654 ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(bp));
3655 ASSERT(BP_COUNT_GANG(bp) == 0 ||
3656 (BP_COUNT_GANG(bp) == BP_GET_NDVAS(bp)));
3657 }
3658 if (zio->io_flags & ZIO_FLAG_NOPWRITE)
3659 VERIFY(BP_EQUAL(bp, &zio->io_bp_orig));
3660 }
3661
3662 /*
3663 * If there were child vdev/gang/ddt errors, they apply to us now.
3664 */
3665 zio_inherit_child_errors(zio, ZIO_CHILD_VDEV);
3666 zio_inherit_child_errors(zio, ZIO_CHILD_GANG);
3667 zio_inherit_child_errors(zio, ZIO_CHILD_DDT);
3668
3669 /*
3670 * If the I/O on the transformed data was successful, generate any
3671 * checksum reports now while we still have the transformed data.
3672 */
3673 if (zio->io_error == 0) {
3674 while (zio->io_cksum_report != NULL) {
3675 zio_cksum_report_t *zcr = zio->io_cksum_report;
3676 uint64_t align = zcr->zcr_align;
3677 uint64_t asize = P2ROUNDUP(psize, align);
3678 char *abuf = zio->io_data;
3679
3680 if (asize != psize) {
3681 abuf = zio_buf_alloc(asize);
3682 bcopy(zio->io_data, abuf, psize);
3683 bzero(abuf + psize, asize - psize);
3684 }
3685
3686 zio->io_cksum_report = zcr->zcr_next;
3687 zcr->zcr_next = NULL;
3688 zcr->zcr_finish(zcr, abuf);
3689 zfs_ereport_free_checksum(zcr);
3690
3691 if (asize != psize)
3692 zio_buf_free(abuf, asize);
3693 }
3694 }
3695
3696 zio_pop_transforms(zio); /* note: may set zio->io_error */
3697
3698 vdev_stat_update(zio, psize);
3699
3700 if (zio->io_error) {
3701 /*
3702 * If this I/O is attached to a particular vdev,
3703 * generate an error message describing the I/O failure
3704 * at the block level. We ignore these errors if the
3705 * device is currently unavailable.
3706 */
3707 if (zio->io_error != ECKSUM && vd != NULL && !vdev_is_dead(vd))
3708 zfs_ereport_post(FM_EREPORT_ZFS_IO, spa, vd, zio, 0, 0);
3709
3710 if ((zio->io_error == EIO || !(zio->io_flags &
3711 (ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) &&
3712 zio == lio) {
3713 /*
3714 * For logical I/O requests, tell the SPA to log the
3715 * error and generate a logical data ereport.
3716 */
3717 spa_log_error(spa, zio);
3718 zfs_ereport_post(FM_EREPORT_ZFS_DATA, spa, NULL, zio,
3719 0, 0);
3720 }
3721 }
3722
3723 if (zio->io_error && zio == lio) {
3724 /*
3725 * Determine whether zio should be reexecuted. This will
3726 * propagate all the way to the root via zio_notify_parent().
3727 */
3728 ASSERT(vd == NULL && bp != NULL);
3729 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
3730
3731 if (IO_IS_ALLOCATING(zio) &&
3732 !(zio->io_flags & ZIO_FLAG_CANFAIL)) {
3733 if (zio->io_error != ENOSPC)
3734 zio->io_reexecute |= ZIO_REEXECUTE_NOW;
3735 else
3736 zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND;
3737 }
3738
3739 if ((zio->io_type == ZIO_TYPE_READ ||
3740 zio->io_type == ZIO_TYPE_FREE) &&
3741 !(zio->io_flags & ZIO_FLAG_SCAN_THREAD) &&
3742 zio->io_error == ENXIO &&
3743 spa_load_state(spa) == SPA_LOAD_NONE &&
3744 spa_get_failmode(spa) != ZIO_FAILURE_MODE_CONTINUE)
3745 zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND;
3746
3747 if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute)
3748 zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND;
3749
3750 /*
3751 * Here is a possibly good place to attempt to do
3752 * either combinatorial reconstruction or error correction
3753 * based on checksums. It also might be a good place
3754 * to send out preliminary ereports before we suspend
3755 * processing.
3756 */
3757 }
3758
3759 /*
3760 * If there were logical child errors, they apply to us now.
3761 * We defer this until now to avoid conflating logical child
3762 * errors with errors that happened to the zio itself when
3763 * updating vdev stats and reporting FMA events above.
3764 */
3765 zio_inherit_child_errors(zio, ZIO_CHILD_LOGICAL);
3766
3767 if ((zio->io_error || zio->io_reexecute) &&
3768 IO_IS_ALLOCATING(zio) && zio->io_gang_leader == zio &&
3769 !(zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE)))
3770 zio_dva_unallocate(zio, zio->io_gang_tree, bp);
3771
3772 zio_gang_tree_free(&zio->io_gang_tree);
3773
3774 /*
3775 * Godfather I/Os should never suspend.
3776 */
3777 if ((zio->io_flags & ZIO_FLAG_GODFATHER) &&
3778 (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND))
3779 zio->io_reexecute = 0;
3780
3781 if (zio->io_reexecute) {
3782 /*
3783 * This is a logical I/O that wants to reexecute.
3784 *
3785 * Reexecute is top-down. When an i/o fails, if it's not
3786 * the root, it simply notifies its parent and sticks around.
3787 * The parent, seeing that it still has children in zio_done(),
3788 * does the same. This percolates all the way up to the root.
3789 * The root i/o will reexecute or suspend the entire tree.
3790 *
3791 * This approach ensures that zio_reexecute() honors
3792 * all the original i/o dependency relationships, e.g.
3793 * parents not executing until children are ready.
3794 */
3795 ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL);
3796
3797 zio->io_gang_leader = NULL;
3798
3799 mutex_enter(&zio->io_lock);
3800 zio->io_state[ZIO_WAIT_DONE] = 1;
3801 mutex_exit(&zio->io_lock);
3802
3803 /*
3804 * "The Godfather" I/O monitors its children but is
3805 * not a true parent to them. It will track them through
3806 * the pipeline but severs its ties whenever they get into
3807 * trouble (e.g. suspended). This allows "The Godfather"
3808 * I/O to return status without blocking.
3809 */
3810 zl = NULL;
3811 for (pio = zio_walk_parents(zio, &zl); pio != NULL;
3812 pio = pio_next) {
3813 zio_link_t *remove_zl = zl;
3814 pio_next = zio_walk_parents(zio, &zl);
3815
3816 if ((pio->io_flags & ZIO_FLAG_GODFATHER) &&
3817 (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) {
3818 zio_remove_child(pio, zio, remove_zl);
3819 zio_notify_parent(pio, zio, ZIO_WAIT_DONE);
3820 }
3821 }
3822
3823 if ((pio = zio_unique_parent(zio)) != NULL) {
3824 /*
3825 * We're not a root i/o, so there's nothing to do
3826 * but notify our parent. Don't propagate errors
3827 * upward since we haven't permanently failed yet.
3828 */
3829 ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER));
3830 zio->io_flags |= ZIO_FLAG_DONT_PROPAGATE;
3831 zio_notify_parent(pio, zio, ZIO_WAIT_DONE);
3832 } else if (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND) {
3833 /*
3834 * We'd fail again if we reexecuted now, so suspend
3835 * until conditions improve (e.g. device comes online).
3836 */
3837 zio_suspend(spa, zio);
3838 } else {
3839 /*
3840 * Reexecution is potentially a huge amount of work.
3841 * Hand it off to the otherwise-unused claim taskq.
3842 */
3843 #if defined(illumos) || !defined(_KERNEL)
3844 ASSERT(zio->io_tqent.tqent_next == NULL);
3845 #elif defined(__NetBSD__)
3846 ASSERT(zio->io_tqent.tqent_queued == 0);
3847 #else
3848 ASSERT(zio->io_tqent.tqent_task.ta_pending == 0);
3849 #endif
3850 spa_taskq_dispatch_ent(spa, ZIO_TYPE_CLAIM,
3851 ZIO_TASKQ_ISSUE, (task_func_t *)zio_reexecute, zio,
3852 0, &zio->io_tqent);
3853 }
3854 return (ZIO_PIPELINE_STOP);
3855 }
3856
3857 ASSERT(zio->io_child_count == 0);
3858 ASSERT(zio->io_reexecute == 0);
3859 ASSERT(zio->io_error == 0 || (zio->io_flags & ZIO_FLAG_CANFAIL));
3860
3861 /*
3862 * Report any checksum errors, since the I/O is complete.
3863 */
3864 while (zio->io_cksum_report != NULL) {
3865 zio_cksum_report_t *zcr = zio->io_cksum_report;
3866 zio->io_cksum_report = zcr->zcr_next;
3867 zcr->zcr_next = NULL;
3868 zcr->zcr_finish(zcr, NULL);
3869 zfs_ereport_free_checksum(zcr);
3870 }
3871
3872 /*
3873 * It is the responsibility of the done callback to ensure that this
3874 * particular zio is no longer discoverable for adoption, and as
3875 * such, cannot acquire any new parents.
3876 */
3877 if (zio->io_done)
3878 zio->io_done(zio);
3879
3880 mutex_enter(&zio->io_lock);
3881 zio->io_state[ZIO_WAIT_DONE] = 1;
3882 mutex_exit(&zio->io_lock);
3883
3884 zl = NULL;
3885 for (pio = zio_walk_parents(zio, &zl); pio != NULL; pio = pio_next) {
3886 zio_link_t *remove_zl = zl;
3887 pio_next = zio_walk_parents(zio, &zl);
3888 zio_remove_child(pio, zio, remove_zl);
3889 zio_notify_parent(pio, zio, ZIO_WAIT_DONE);
3890 }
3891
3892 if (zio->io_waiter != NULL) {
3893 mutex_enter(&zio->io_lock);
3894 zio->io_executor = NULL;
3895 cv_broadcast(&zio->io_cv);
3896 mutex_exit(&zio->io_lock);
3897 } else {
3898 zio_destroy(zio);
3899 }
3900
3901 return (ZIO_PIPELINE_STOP);
3902 }
3903
3904 /*
3905 * ==========================================================================
3906 * I/O pipeline definition
3907 * ==========================================================================
3908 */
3909 static zio_pipe_stage_t *zio_pipeline[] = {
3910 NULL,
3911 zio_read_bp_init,
3912 zio_write_bp_init,
3913 zio_free_bp_init,
3914 zio_issue_async,
3915 zio_write_compress,
3916 zio_checksum_generate,
3917 zio_nop_write,
3918 zio_ddt_read_start,
3919 zio_ddt_read_done,
3920 zio_ddt_write,
3921 zio_ddt_free,
3922 zio_gang_assemble,
3923 zio_gang_issue,
3924 zio_dva_throttle,
3925 zio_dva_allocate,
3926 zio_dva_free,
3927 zio_dva_claim,
3928 zio_ready,
3929 zio_vdev_io_start,
3930 zio_vdev_io_done,
3931 zio_vdev_io_assess,
3932 zio_checksum_verify,
3933 zio_done
3934 };
3935
3936
3937
3938
3939 /*
3940 * Compare two zbookmark_phys_t's to see which we would reach first in a
3941 * pre-order traversal of the object tree.
3942 *
3943 * This is simple in every case aside from the meta-dnode object. For all other
3944 * objects, we traverse them in order (object 1 before object 2, and so on).
3945 * However, all of these objects are traversed while traversing object 0, since
3946 * the data it points to is the list of objects. Thus, we need to convert to a
3947 * canonical representation so we can compare meta-dnode bookmarks to
3948 * non-meta-dnode bookmarks.
3949 *
3950 * We do this by calculating "equivalents" for each field of the zbookmark.
3951 * zbookmarks outside of the meta-dnode use their own object and level, and
3952 * calculate the level 0 equivalent (the first L0 blkid that is contained in the
3953 * blocks this bookmark refers to) by multiplying their blkid by their span
3954 * (the number of L0 blocks contained within one block at their level).
3955 * zbookmarks inside the meta-dnode calculate their object equivalent
3956 * (which is L0equiv * dnodes per data block), use 0 for their L0equiv, and use
3957 * level + 1<<31 (any value larger than a level could ever be) for their level.
3958 * This causes them to always compare before a bookmark in their object
3959 * equivalent, compare appropriately to bookmarks in other objects, and to
3960 * compare appropriately to other bookmarks in the meta-dnode.
3961 */
3962 int
zbookmark_compare(uint16_t dbss1,uint8_t ibs1,uint16_t dbss2,uint8_t ibs2,const zbookmark_phys_t * zb1,const zbookmark_phys_t * zb2)3963 zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2, uint8_t ibs2,
3964 const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2)
3965 {
3966 /*
3967 * These variables represent the "equivalent" values for the zbookmark,
3968 * after converting zbookmarks inside the meta dnode to their
3969 * normal-object equivalents.
3970 */
3971 uint64_t zb1obj, zb2obj;
3972 uint64_t zb1L0, zb2L0;
3973 uint64_t zb1level, zb2level;
3974
3975 if (zb1->zb_object == zb2->zb_object &&
3976 zb1->zb_level == zb2->zb_level &&
3977 zb1->zb_blkid == zb2->zb_blkid)
3978 return (0);
3979
3980 /*
3981 * BP_SPANB calculates the span in blocks.
3982 */
3983 zb1L0 = (zb1->zb_blkid) * BP_SPANB(ibs1, zb1->zb_level);
3984 zb2L0 = (zb2->zb_blkid) * BP_SPANB(ibs2, zb2->zb_level);
3985
3986 if (zb1->zb_object == DMU_META_DNODE_OBJECT) {
3987 zb1obj = zb1L0 * (dbss1 << (SPA_MINBLOCKSHIFT - DNODE_SHIFT));
3988 zb1L0 = 0;
3989 zb1level = zb1->zb_level + COMPARE_META_LEVEL;
3990 } else {
3991 zb1obj = zb1->zb_object;
3992 zb1level = zb1->zb_level;
3993 }
3994
3995 if (zb2->zb_object == DMU_META_DNODE_OBJECT) {
3996 zb2obj = zb2L0 * (dbss2 << (SPA_MINBLOCKSHIFT - DNODE_SHIFT));
3997 zb2L0 = 0;
3998 zb2level = zb2->zb_level + COMPARE_META_LEVEL;
3999 } else {
4000 zb2obj = zb2->zb_object;
4001 zb2level = zb2->zb_level;
4002 }
4003
4004 /* Now that we have a canonical representation, do the comparison. */
4005 if (zb1obj != zb2obj)
4006 return (zb1obj < zb2obj ? -1 : 1);
4007 else if (zb1L0 != zb2L0)
4008 return (zb1L0 < zb2L0 ? -1 : 1);
4009 else if (zb1level != zb2level)
4010 return (zb1level > zb2level ? -1 : 1);
4011 /*
4012 * This can (theoretically) happen if the bookmarks have the same object
4013 * and level, but different blkids, if the block sizes are not the same.
4014 * There is presently no way to change the indirect block sizes
4015 */
4016 return (0);
4017 }
4018
4019 /*
4020 * This function checks the following: given that last_block is the place that
4021 * our traversal stopped last time, does that guarantee that we've visited
4022 * every node under subtree_root? Therefore, we can't just use the raw output
4023 * of zbookmark_compare. We have to pass in a modified version of
4024 * subtree_root; by incrementing the block id, and then checking whether
4025 * last_block is before or equal to that, we can tell whether or not having
4026 * visited last_block implies that all of subtree_root's children have been
4027 * visited.
4028 */
4029 boolean_t
zbookmark_subtree_completed(const dnode_phys_t * dnp,const zbookmark_phys_t * subtree_root,const zbookmark_phys_t * last_block)4030 zbookmark_subtree_completed(const dnode_phys_t *dnp,
4031 const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block)
4032 {
4033 zbookmark_phys_t mod_zb = *subtree_root;
4034 mod_zb.zb_blkid++;
4035 ASSERT(last_block->zb_level == 0);
4036
4037 /* The objset_phys_t isn't before anything. */
4038 if (dnp == NULL)
4039 return (B_FALSE);
4040
4041 /*
4042 * We pass in 1ULL << (DNODE_BLOCK_SHIFT - SPA_MINBLOCKSHIFT) for the
4043 * data block size in sectors, because that variable is only used if
4044 * the bookmark refers to a block in the meta-dnode. Since we don't
4045 * know without examining it what object it refers to, and there's no
4046 * harm in passing in this value in other cases, we always pass it in.
4047 *
4048 * We pass in 0 for the indirect block size shift because zb2 must be
4049 * level 0. The indirect block size is only used to calculate the span
4050 * of the bookmark, but since the bookmark must be level 0, the span is
4051 * always 1, so the math works out.
4052 *
4053 * If you make changes to how the zbookmark_compare code works, be sure
4054 * to make sure that this code still works afterwards.
4055 */
4056 return (zbookmark_compare(dnp->dn_datablkszsec, dnp->dn_indblkshift,
4057 1ULL << (DNODE_BLOCK_SHIFT - SPA_MINBLOCKSHIFT), 0, &mod_zb,
4058 last_block) <= 0);
4059 }
4060