]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
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 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
ebf8e3a2 | 23 | * Copyright (c) 2012 by Delphix. All rights reserved. |
a38718a6 | 24 | * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved. |
34dc7c2f BB |
25 | */ |
26 | ||
34dc7c2f BB |
27 | #include <sys/zfs_context.h> |
28 | #include <sys/fm/fs/zfs.h> | |
29 | #include <sys/spa.h> | |
30 | #include <sys/txg.h> | |
31 | #include <sys/spa_impl.h> | |
32 | #include <sys/vdev_impl.h> | |
33 | #include <sys/zio_impl.h> | |
34 | #include <sys/zio_compress.h> | |
35 | #include <sys/zio_checksum.h> | |
428870ff BB |
36 | #include <sys/dmu_objset.h> |
37 | #include <sys/arc.h> | |
38 | #include <sys/ddt.h> | |
34dc7c2f BB |
39 | |
40 | /* | |
41 | * ========================================================================== | |
42 | * I/O priority table | |
43 | * ========================================================================== | |
44 | */ | |
45 | uint8_t zio_priority_table[ZIO_PRIORITY_TABLE_SIZE] = { | |
46 | 0, /* ZIO_PRIORITY_NOW */ | |
47 | 0, /* ZIO_PRIORITY_SYNC_READ */ | |
48 | 0, /* ZIO_PRIORITY_SYNC_WRITE */ | |
34dc7c2f | 49 | 0, /* ZIO_PRIORITY_LOG_WRITE */ |
428870ff BB |
50 | 1, /* ZIO_PRIORITY_CACHE_FILL */ |
51 | 1, /* ZIO_PRIORITY_AGG */ | |
52 | 4, /* ZIO_PRIORITY_FREE */ | |
53 | 4, /* ZIO_PRIORITY_ASYNC_WRITE */ | |
54 | 6, /* ZIO_PRIORITY_ASYNC_READ */ | |
34dc7c2f BB |
55 | 10, /* ZIO_PRIORITY_RESILVER */ |
56 | 20, /* ZIO_PRIORITY_SCRUB */ | |
428870ff | 57 | 2, /* ZIO_PRIORITY_DDT_PREFETCH */ |
34dc7c2f BB |
58 | }; |
59 | ||
60 | /* | |
61 | * ========================================================================== | |
62 | * I/O type descriptions | |
63 | * ========================================================================== | |
64 | */ | |
65 | char *zio_type_name[ZIO_TYPES] = { | |
451041db | 66 | "z_null", "z_rd", "z_wr", "z_fr", "z_cl", "z_ioctl" |
428870ff | 67 | }; |
34dc7c2f BB |
68 | |
69 | /* | |
70 | * ========================================================================== | |
71 | * I/O kmem caches | |
72 | * ========================================================================== | |
73 | */ | |
74 | kmem_cache_t *zio_cache; | |
d164b209 | 75 | kmem_cache_t *zio_link_cache; |
86dd0fd9 | 76 | kmem_cache_t *zio_vdev_cache; |
34dc7c2f BB |
77 | kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; |
78 | kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; | |
c28b2279 | 79 | int zio_bulk_flags = 0; |
a69052be | 80 | int zio_delay_max = ZIO_DELAY_MAX; |
34dc7c2f BB |
81 | |
82 | #ifdef _KERNEL | |
83 | extern vmem_t *zio_alloc_arena; | |
84 | #endif | |
6d974228 | 85 | extern int zfs_mg_alloc_failures; |
34dc7c2f BB |
86 | |
87 | /* | |
b128c09f BB |
88 | * An allocating zio is one that either currently has the DVA allocate |
89 | * stage set or will have it later in its lifetime. | |
34dc7c2f | 90 | */ |
428870ff BB |
91 | #define IO_IS_ALLOCATING(zio) ((zio)->io_orig_pipeline & ZIO_STAGE_DVA_ALLOCATE) |
92 | ||
c409e464 | 93 | int zio_requeue_io_start_cut_in_line = 1; |
428870ff BB |
94 | |
95 | #ifdef ZFS_DEBUG | |
96 | int zio_buf_debug_limit = 16384; | |
97 | #else | |
98 | int zio_buf_debug_limit = 0; | |
99 | #endif | |
34dc7c2f | 100 | |
da6b4005 NB |
101 | static inline void __zio_execute(zio_t *zio); |
102 | ||
49be0ccf BB |
103 | static int |
104 | zio_cons(void *arg, void *unused, int kmflag) | |
105 | { | |
106 | zio_t *zio = arg; | |
107 | ||
108 | bzero(zio, sizeof (zio_t)); | |
109 | ||
110 | mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL); | |
111 | cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL); | |
112 | ||
113 | list_create(&zio->io_parent_list, sizeof (zio_link_t), | |
114 | offsetof(zio_link_t, zl_parent_node)); | |
115 | list_create(&zio->io_child_list, sizeof (zio_link_t), | |
116 | offsetof(zio_link_t, zl_child_node)); | |
117 | ||
118 | return (0); | |
119 | } | |
120 | ||
121 | static void | |
122 | zio_dest(void *arg, void *unused) | |
123 | { | |
124 | zio_t *zio = arg; | |
125 | ||
126 | mutex_destroy(&zio->io_lock); | |
127 | cv_destroy(&zio->io_cv); | |
128 | list_destroy(&zio->io_parent_list); | |
129 | list_destroy(&zio->io_child_list); | |
130 | } | |
131 | ||
34dc7c2f BB |
132 | void |
133 | zio_init(void) | |
134 | { | |
135 | size_t c; | |
136 | vmem_t *data_alloc_arena = NULL; | |
137 | ||
138 | #ifdef _KERNEL | |
139 | data_alloc_arena = zio_alloc_arena; | |
140 | #endif | |
49be0ccf BB |
141 | zio_cache = kmem_cache_create("zio_cache", sizeof (zio_t), 0, |
142 | zio_cons, zio_dest, NULL, NULL, NULL, KMC_KMEM); | |
d164b209 | 143 | zio_link_cache = kmem_cache_create("zio_link_cache", |
ae6ba3db | 144 | sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, KMC_KMEM); |
86dd0fd9 BB |
145 | zio_vdev_cache = kmem_cache_create("zio_vdev_cache", sizeof(vdev_io_t), |
146 | PAGESIZE, NULL, NULL, NULL, NULL, NULL, KMC_VMEM); | |
34dc7c2f BB |
147 | |
148 | /* | |
149 | * For small buffers, we want a cache for each multiple of | |
150 | * SPA_MINBLOCKSIZE. For medium-size buffers, we want a cache | |
151 | * for each quarter-power of 2. For large buffers, we want | |
152 | * a cache for each multiple of PAGESIZE. | |
153 | */ | |
154 | for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { | |
155 | size_t size = (c + 1) << SPA_MINBLOCKSHIFT; | |
156 | size_t p2 = size; | |
157 | size_t align = 0; | |
158 | ||
159 | while (p2 & (p2 - 1)) | |
160 | p2 &= p2 - 1; | |
161 | ||
162 | if (size <= 4 * SPA_MINBLOCKSIZE) { | |
163 | align = SPA_MINBLOCKSIZE; | |
164 | } else if (P2PHASE(size, PAGESIZE) == 0) { | |
165 | align = PAGESIZE; | |
166 | } else if (P2PHASE(size, p2 >> 2) == 0) { | |
167 | align = p2 >> 2; | |
168 | } | |
169 | ||
170 | if (align != 0) { | |
171 | char name[36]; | |
ae6ba3db BB |
172 | int flags = zio_bulk_flags; |
173 | ||
174 | /* | |
175 | * The smallest buffers (512b) are heavily used and | |
176 | * experience a lot of churn. The slabs allocated | |
177 | * for them are also relatively small (32K). Thus | |
178 | * in over to avoid expensive calls to vmalloc() we | |
179 | * make an exception to the usual slab allocation | |
180 | * policy and force these buffers to be kmem backed. | |
181 | */ | |
182 | if (size == (1 << SPA_MINBLOCKSHIFT)) | |
183 | flags |= KMC_KMEM; | |
184 | ||
34dc7c2f BB |
185 | (void) sprintf(name, "zio_buf_%lu", (ulong_t)size); |
186 | zio_buf_cache[c] = kmem_cache_create(name, size, | |
ae6ba3db | 187 | align, NULL, NULL, NULL, NULL, NULL, flags); |
34dc7c2f BB |
188 | |
189 | (void) sprintf(name, "zio_data_buf_%lu", (ulong_t)size); | |
190 | zio_data_buf_cache[c] = kmem_cache_create(name, size, | |
ae6ba3db BB |
191 | align, NULL, NULL, NULL, NULL, |
192 | data_alloc_arena, flags); | |
34dc7c2f BB |
193 | } |
194 | } | |
195 | ||
196 | while (--c != 0) { | |
197 | ASSERT(zio_buf_cache[c] != NULL); | |
198 | if (zio_buf_cache[c - 1] == NULL) | |
199 | zio_buf_cache[c - 1] = zio_buf_cache[c]; | |
200 | ||
201 | ASSERT(zio_data_buf_cache[c] != NULL); | |
202 | if (zio_data_buf_cache[c - 1] == NULL) | |
203 | zio_data_buf_cache[c - 1] = zio_data_buf_cache[c]; | |
204 | } | |
205 | ||
6d974228 GW |
206 | /* |
207 | * The zio write taskqs have 1 thread per cpu, allow 1/2 of the taskqs | |
208 | * to fail 3 times per txg or 8 failures, whichever is greater. | |
209 | */ | |
210 | zfs_mg_alloc_failures = MAX((3 * max_ncpus / 2), 8); | |
211 | ||
34dc7c2f BB |
212 | zio_inject_init(); |
213 | } | |
214 | ||
215 | void | |
216 | zio_fini(void) | |
217 | { | |
218 | size_t c; | |
219 | kmem_cache_t *last_cache = NULL; | |
220 | kmem_cache_t *last_data_cache = NULL; | |
221 | ||
222 | for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { | |
223 | if (zio_buf_cache[c] != last_cache) { | |
224 | last_cache = zio_buf_cache[c]; | |
225 | kmem_cache_destroy(zio_buf_cache[c]); | |
226 | } | |
227 | zio_buf_cache[c] = NULL; | |
228 | ||
229 | if (zio_data_buf_cache[c] != last_data_cache) { | |
230 | last_data_cache = zio_data_buf_cache[c]; | |
231 | kmem_cache_destroy(zio_data_buf_cache[c]); | |
232 | } | |
233 | zio_data_buf_cache[c] = NULL; | |
234 | } | |
235 | ||
86dd0fd9 | 236 | kmem_cache_destroy(zio_vdev_cache); |
d164b209 | 237 | kmem_cache_destroy(zio_link_cache); |
34dc7c2f BB |
238 | kmem_cache_destroy(zio_cache); |
239 | ||
240 | zio_inject_fini(); | |
241 | } | |
242 | ||
243 | /* | |
244 | * ========================================================================== | |
245 | * Allocate and free I/O buffers | |
246 | * ========================================================================== | |
247 | */ | |
248 | ||
249 | /* | |
250 | * Use zio_buf_alloc to allocate ZFS metadata. This data will appear in a | |
251 | * crashdump if the kernel panics, so use it judiciously. Obviously, it's | |
252 | * useful to inspect ZFS metadata, but if possible, we should avoid keeping | |
253 | * excess / transient data in-core during a crashdump. | |
254 | */ | |
255 | void * | |
256 | zio_buf_alloc(size_t size) | |
257 | { | |
258 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
259 | ||
260 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
261 | ||
262 | return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE)); | |
263 | } | |
264 | ||
265 | /* | |
266 | * Use zio_data_buf_alloc to allocate data. The data will not appear in a | |
267 | * crashdump if the kernel panics. This exists so that we will limit the amount | |
268 | * of ZFS data that shows up in a kernel crashdump. (Thus reducing the amount | |
269 | * of kernel heap dumped to disk when the kernel panics) | |
270 | */ | |
271 | void * | |
272 | zio_data_buf_alloc(size_t size) | |
273 | { | |
274 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
275 | ||
276 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
277 | ||
278 | return (kmem_cache_alloc(zio_data_buf_cache[c], KM_PUSHPAGE)); | |
279 | } | |
280 | ||
281 | void | |
282 | zio_buf_free(void *buf, size_t size) | |
283 | { | |
284 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
285 | ||
286 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
287 | ||
288 | kmem_cache_free(zio_buf_cache[c], buf); | |
289 | } | |
290 | ||
291 | void | |
292 | zio_data_buf_free(void *buf, size_t size) | |
293 | { | |
294 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
295 | ||
296 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
297 | ||
298 | kmem_cache_free(zio_data_buf_cache[c], buf); | |
299 | } | |
300 | ||
86dd0fd9 BB |
301 | /* |
302 | * Dedicated I/O buffers to ensure that memory fragmentation never prevents | |
303 | * or significantly delays the issuing of a zio. These buffers are used | |
304 | * to aggregate I/O and could be used for raidz stripes. | |
305 | */ | |
306 | void * | |
307 | zio_vdev_alloc(void) | |
308 | { | |
309 | return (kmem_cache_alloc(zio_vdev_cache, KM_PUSHPAGE)); | |
310 | } | |
311 | ||
312 | void | |
313 | zio_vdev_free(void *buf) | |
314 | { | |
315 | kmem_cache_free(zio_vdev_cache, buf); | |
316 | ||
317 | } | |
318 | ||
34dc7c2f BB |
319 | /* |
320 | * ========================================================================== | |
321 | * Push and pop I/O transform buffers | |
322 | * ========================================================================== | |
323 | */ | |
324 | static void | |
b128c09f BB |
325 | zio_push_transform(zio_t *zio, void *data, uint64_t size, uint64_t bufsize, |
326 | zio_transform_func_t *transform) | |
34dc7c2f | 327 | { |
691f6ac4 | 328 | zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_PUSHPAGE); |
34dc7c2f | 329 | |
b128c09f BB |
330 | zt->zt_orig_data = zio->io_data; |
331 | zt->zt_orig_size = zio->io_size; | |
34dc7c2f | 332 | zt->zt_bufsize = bufsize; |
b128c09f | 333 | zt->zt_transform = transform; |
34dc7c2f BB |
334 | |
335 | zt->zt_next = zio->io_transform_stack; | |
336 | zio->io_transform_stack = zt; | |
337 | ||
338 | zio->io_data = data; | |
339 | zio->io_size = size; | |
340 | } | |
341 | ||
342 | static void | |
b128c09f | 343 | zio_pop_transforms(zio_t *zio) |
34dc7c2f | 344 | { |
b128c09f BB |
345 | zio_transform_t *zt; |
346 | ||
347 | while ((zt = zio->io_transform_stack) != NULL) { | |
348 | if (zt->zt_transform != NULL) | |
349 | zt->zt_transform(zio, | |
350 | zt->zt_orig_data, zt->zt_orig_size); | |
34dc7c2f | 351 | |
428870ff BB |
352 | if (zt->zt_bufsize != 0) |
353 | zio_buf_free(zio->io_data, zt->zt_bufsize); | |
34dc7c2f | 354 | |
b128c09f BB |
355 | zio->io_data = zt->zt_orig_data; |
356 | zio->io_size = zt->zt_orig_size; | |
357 | zio->io_transform_stack = zt->zt_next; | |
34dc7c2f | 358 | |
b128c09f | 359 | kmem_free(zt, sizeof (zio_transform_t)); |
34dc7c2f BB |
360 | } |
361 | } | |
362 | ||
b128c09f BB |
363 | /* |
364 | * ========================================================================== | |
365 | * I/O transform callbacks for subblocks and decompression | |
366 | * ========================================================================== | |
367 | */ | |
368 | static void | |
369 | zio_subblock(zio_t *zio, void *data, uint64_t size) | |
370 | { | |
371 | ASSERT(zio->io_size > size); | |
372 | ||
373 | if (zio->io_type == ZIO_TYPE_READ) | |
374 | bcopy(zio->io_data, data, size); | |
375 | } | |
376 | ||
377 | static void | |
378 | zio_decompress(zio_t *zio, void *data, uint64_t size) | |
379 | { | |
380 | if (zio->io_error == 0 && | |
381 | zio_decompress_data(BP_GET_COMPRESS(zio->io_bp), | |
428870ff | 382 | zio->io_data, data, zio->io_size, size) != 0) |
b128c09f BB |
383 | zio->io_error = EIO; |
384 | } | |
385 | ||
386 | /* | |
387 | * ========================================================================== | |
388 | * I/O parent/child relationships and pipeline interlocks | |
389 | * ========================================================================== | |
390 | */ | |
d164b209 BB |
391 | /* |
392 | * NOTE - Callers to zio_walk_parents() and zio_walk_children must | |
393 | * continue calling these functions until they return NULL. | |
394 | * Otherwise, the next caller will pick up the list walk in | |
395 | * some indeterminate state. (Otherwise every caller would | |
396 | * have to pass in a cookie to keep the state represented by | |
397 | * io_walk_link, which gets annoying.) | |
398 | */ | |
399 | zio_t * | |
400 | zio_walk_parents(zio_t *cio) | |
401 | { | |
402 | zio_link_t *zl = cio->io_walk_link; | |
403 | list_t *pl = &cio->io_parent_list; | |
b128c09f | 404 | |
d164b209 BB |
405 | zl = (zl == NULL) ? list_head(pl) : list_next(pl, zl); |
406 | cio->io_walk_link = zl; | |
407 | ||
408 | if (zl == NULL) | |
409 | return (NULL); | |
410 | ||
411 | ASSERT(zl->zl_child == cio); | |
412 | return (zl->zl_parent); | |
413 | } | |
414 | ||
415 | zio_t * | |
416 | zio_walk_children(zio_t *pio) | |
417 | { | |
418 | zio_link_t *zl = pio->io_walk_link; | |
419 | list_t *cl = &pio->io_child_list; | |
420 | ||
421 | zl = (zl == NULL) ? list_head(cl) : list_next(cl, zl); | |
422 | pio->io_walk_link = zl; | |
423 | ||
424 | if (zl == NULL) | |
425 | return (NULL); | |
426 | ||
427 | ASSERT(zl->zl_parent == pio); | |
428 | return (zl->zl_child); | |
429 | } | |
430 | ||
431 | zio_t * | |
432 | zio_unique_parent(zio_t *cio) | |
433 | { | |
434 | zio_t *pio = zio_walk_parents(cio); | |
435 | ||
436 | VERIFY(zio_walk_parents(cio) == NULL); | |
437 | return (pio); | |
438 | } | |
439 | ||
440 | void | |
441 | zio_add_child(zio_t *pio, zio_t *cio) | |
b128c09f | 442 | { |
691f6ac4 | 443 | zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_PUSHPAGE); |
d6320ddb | 444 | int w; |
d164b209 BB |
445 | |
446 | /* | |
447 | * Logical I/Os can have logical, gang, or vdev children. | |
448 | * Gang I/Os can have gang or vdev children. | |
449 | * Vdev I/Os can only have vdev children. | |
450 | * The following ASSERT captures all of these constraints. | |
451 | */ | |
452 | ASSERT(cio->io_child_type <= pio->io_child_type); | |
453 | ||
454 | zl->zl_parent = pio; | |
455 | zl->zl_child = cio; | |
456 | ||
457 | mutex_enter(&cio->io_lock); | |
b128c09f | 458 | mutex_enter(&pio->io_lock); |
d164b209 BB |
459 | |
460 | ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0); | |
461 | ||
d6320ddb | 462 | for (w = 0; w < ZIO_WAIT_TYPES; w++) |
d164b209 BB |
463 | pio->io_children[cio->io_child_type][w] += !cio->io_state[w]; |
464 | ||
465 | list_insert_head(&pio->io_child_list, zl); | |
466 | list_insert_head(&cio->io_parent_list, zl); | |
467 | ||
428870ff BB |
468 | pio->io_child_count++; |
469 | cio->io_parent_count++; | |
470 | ||
b128c09f | 471 | mutex_exit(&pio->io_lock); |
d164b209 | 472 | mutex_exit(&cio->io_lock); |
b128c09f BB |
473 | } |
474 | ||
34dc7c2f | 475 | static void |
d164b209 | 476 | zio_remove_child(zio_t *pio, zio_t *cio, zio_link_t *zl) |
b128c09f | 477 | { |
d164b209 BB |
478 | ASSERT(zl->zl_parent == pio); |
479 | ASSERT(zl->zl_child == cio); | |
b128c09f | 480 | |
d164b209 | 481 | mutex_enter(&cio->io_lock); |
b128c09f | 482 | mutex_enter(&pio->io_lock); |
d164b209 BB |
483 | |
484 | list_remove(&pio->io_child_list, zl); | |
485 | list_remove(&cio->io_parent_list, zl); | |
486 | ||
428870ff BB |
487 | pio->io_child_count--; |
488 | cio->io_parent_count--; | |
489 | ||
b128c09f | 490 | mutex_exit(&pio->io_lock); |
d164b209 BB |
491 | mutex_exit(&cio->io_lock); |
492 | ||
493 | kmem_cache_free(zio_link_cache, zl); | |
b128c09f BB |
494 | } |
495 | ||
496 | static boolean_t | |
497 | zio_wait_for_children(zio_t *zio, enum zio_child child, enum zio_wait_type wait) | |
34dc7c2f | 498 | { |
b128c09f BB |
499 | uint64_t *countp = &zio->io_children[child][wait]; |
500 | boolean_t waiting = B_FALSE; | |
501 | ||
502 | mutex_enter(&zio->io_lock); | |
503 | ASSERT(zio->io_stall == NULL); | |
504 | if (*countp != 0) { | |
428870ff | 505 | zio->io_stage >>= 1; |
b128c09f BB |
506 | zio->io_stall = countp; |
507 | waiting = B_TRUE; | |
508 | } | |
509 | mutex_exit(&zio->io_lock); | |
510 | ||
511 | return (waiting); | |
512 | } | |
34dc7c2f | 513 | |
bf701a83 BB |
514 | __attribute__((always_inline)) |
515 | static inline void | |
b128c09f BB |
516 | zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait) |
517 | { | |
518 | uint64_t *countp = &pio->io_children[zio->io_child_type][wait]; | |
519 | int *errorp = &pio->io_child_error[zio->io_child_type]; | |
34dc7c2f | 520 | |
b128c09f BB |
521 | mutex_enter(&pio->io_lock); |
522 | if (zio->io_error && !(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) | |
523 | *errorp = zio_worst_error(*errorp, zio->io_error); | |
524 | pio->io_reexecute |= zio->io_reexecute; | |
525 | ASSERT3U(*countp, >, 0); | |
526 | if (--*countp == 0 && pio->io_stall == countp) { | |
527 | pio->io_stall = NULL; | |
528 | mutex_exit(&pio->io_lock); | |
da6b4005 | 529 | __zio_execute(pio); |
b128c09f BB |
530 | } else { |
531 | mutex_exit(&pio->io_lock); | |
34dc7c2f BB |
532 | } |
533 | } | |
534 | ||
b128c09f BB |
535 | static void |
536 | zio_inherit_child_errors(zio_t *zio, enum zio_child c) | |
537 | { | |
538 | if (zio->io_child_error[c] != 0 && zio->io_error == 0) | |
539 | zio->io_error = zio->io_child_error[c]; | |
540 | } | |
541 | ||
34dc7c2f BB |
542 | /* |
543 | * ========================================================================== | |
b128c09f | 544 | * Create the various types of I/O (read, write, free, etc) |
34dc7c2f BB |
545 | * ========================================================================== |
546 | */ | |
547 | static zio_t * | |
428870ff | 548 | zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, |
34dc7c2f | 549 | void *data, uint64_t size, zio_done_func_t *done, void *private, |
428870ff BB |
550 | zio_type_t type, int priority, enum zio_flag flags, |
551 | vdev_t *vd, uint64_t offset, const zbookmark_t *zb, | |
552 | enum zio_stage stage, enum zio_stage pipeline) | |
34dc7c2f BB |
553 | { |
554 | zio_t *zio; | |
555 | ||
556 | ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); | |
557 | ASSERT(P2PHASE(size, SPA_MINBLOCKSIZE) == 0); | |
b128c09f BB |
558 | ASSERT(P2PHASE(offset, SPA_MINBLOCKSIZE) == 0); |
559 | ||
560 | ASSERT(!vd || spa_config_held(spa, SCL_STATE_ALL, RW_READER)); | |
561 | ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER)); | |
562 | ASSERT(vd || stage == ZIO_STAGE_OPEN); | |
34dc7c2f | 563 | |
691f6ac4 | 564 | zio = kmem_cache_alloc(zio_cache, KM_PUSHPAGE); |
d164b209 | 565 | |
b128c09f BB |
566 | if (vd != NULL) |
567 | zio->io_child_type = ZIO_CHILD_VDEV; | |
568 | else if (flags & ZIO_FLAG_GANG_CHILD) | |
569 | zio->io_child_type = ZIO_CHILD_GANG; | |
428870ff BB |
570 | else if (flags & ZIO_FLAG_DDT_CHILD) |
571 | zio->io_child_type = ZIO_CHILD_DDT; | |
b128c09f BB |
572 | else |
573 | zio->io_child_type = ZIO_CHILD_LOGICAL; | |
574 | ||
34dc7c2f | 575 | if (bp != NULL) { |
49be0ccf | 576 | zio->io_logical = NULL; |
428870ff | 577 | zio->io_bp = (blkptr_t *)bp; |
34dc7c2f BB |
578 | zio->io_bp_copy = *bp; |
579 | zio->io_bp_orig = *bp; | |
428870ff BB |
580 | if (type != ZIO_TYPE_WRITE || |
581 | zio->io_child_type == ZIO_CHILD_DDT) | |
b128c09f | 582 | zio->io_bp = &zio->io_bp_copy; /* so caller can free */ |
9babb374 | 583 | if (zio->io_child_type == ZIO_CHILD_LOGICAL) |
b128c09f | 584 | zio->io_logical = zio; |
9babb374 BB |
585 | if (zio->io_child_type > ZIO_CHILD_GANG && BP_IS_GANG(bp)) |
586 | pipeline |= ZIO_GANG_STAGES; | |
49be0ccf BB |
587 | } else { |
588 | zio->io_logical = NULL; | |
589 | zio->io_bp = NULL; | |
590 | bzero(&zio->io_bp_copy, sizeof (blkptr_t)); | |
591 | bzero(&zio->io_bp_orig, sizeof (blkptr_t)); | |
34dc7c2f | 592 | } |
b128c09f BB |
593 | |
594 | zio->io_spa = spa; | |
595 | zio->io_txg = txg; | |
49be0ccf | 596 | zio->io_ready = NULL; |
34dc7c2f BB |
597 | zio->io_done = done; |
598 | zio->io_private = private; | |
49be0ccf | 599 | zio->io_prev_space_delta = 0; |
34dc7c2f BB |
600 | zio->io_type = type; |
601 | zio->io_priority = priority; | |
b128c09f | 602 | zio->io_vd = vd; |
49be0ccf BB |
603 | zio->io_vsd = NULL; |
604 | zio->io_vsd_ops = NULL; | |
b128c09f | 605 | zio->io_offset = offset; |
49be0ccf | 606 | zio->io_deadline = 0; |
428870ff BB |
607 | zio->io_orig_data = zio->io_data = data; |
608 | zio->io_orig_size = zio->io_size = size; | |
b128c09f BB |
609 | zio->io_orig_flags = zio->io_flags = flags; |
610 | zio->io_orig_stage = zio->io_stage = stage; | |
611 | zio->io_orig_pipeline = zio->io_pipeline = pipeline; | |
49be0ccf BB |
612 | bzero(&zio->io_prop, sizeof (zio_prop_t)); |
613 | zio->io_cmd = 0; | |
614 | zio->io_reexecute = 0; | |
615 | zio->io_bp_override = NULL; | |
616 | zio->io_walk_link = NULL; | |
617 | zio->io_transform_stack = NULL; | |
618 | zio->io_delay = 0; | |
619 | zio->io_error = 0; | |
620 | zio->io_child_count = 0; | |
621 | zio->io_parent_count = 0; | |
622 | zio->io_stall = NULL; | |
623 | zio->io_gang_leader = NULL; | |
624 | zio->io_gang_tree = NULL; | |
625 | zio->io_executor = NULL; | |
626 | zio->io_waiter = NULL; | |
627 | zio->io_cksum_report = NULL; | |
628 | zio->io_ena = 0; | |
629 | bzero(zio->io_child_error, sizeof (int) * ZIO_CHILD_TYPES); | |
630 | bzero(zio->io_children, | |
631 | sizeof (uint64_t) * ZIO_CHILD_TYPES * ZIO_WAIT_TYPES); | |
632 | bzero(&zio->io_bookmark, sizeof (zbookmark_t)); | |
34dc7c2f | 633 | |
d164b209 BB |
634 | zio->io_state[ZIO_WAIT_READY] = (stage >= ZIO_STAGE_READY); |
635 | zio->io_state[ZIO_WAIT_DONE] = (stage >= ZIO_STAGE_DONE); | |
636 | ||
b128c09f BB |
637 | if (zb != NULL) |
638 | zio->io_bookmark = *zb; | |
639 | ||
640 | if (pio != NULL) { | |
b128c09f | 641 | if (zio->io_logical == NULL) |
34dc7c2f | 642 | zio->io_logical = pio->io_logical; |
9babb374 BB |
643 | if (zio->io_child_type == ZIO_CHILD_GANG) |
644 | zio->io_gang_leader = pio->io_gang_leader; | |
b128c09f | 645 | zio_add_child(pio, zio); |
34dc7c2f BB |
646 | } |
647 | ||
a38718a6 GA |
648 | taskq_init_ent(&zio->io_tqent); |
649 | ||
34dc7c2f BB |
650 | return (zio); |
651 | } | |
652 | ||
653 | static void | |
b128c09f | 654 | zio_destroy(zio_t *zio) |
34dc7c2f | 655 | { |
b128c09f | 656 | kmem_cache_free(zio_cache, zio); |
34dc7c2f BB |
657 | } |
658 | ||
659 | zio_t * | |
d164b209 | 660 | zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done, |
428870ff | 661 | void *private, enum zio_flag flags) |
34dc7c2f BB |
662 | { |
663 | zio_t *zio; | |
664 | ||
665 | zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, | |
d164b209 | 666 | ZIO_TYPE_NULL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, |
b128c09f | 667 | ZIO_STAGE_OPEN, ZIO_INTERLOCK_PIPELINE); |
34dc7c2f BB |
668 | |
669 | return (zio); | |
670 | } | |
671 | ||
672 | zio_t * | |
428870ff | 673 | zio_root(spa_t *spa, zio_done_func_t *done, void *private, enum zio_flag flags) |
34dc7c2f | 674 | { |
d164b209 | 675 | return (zio_null(NULL, spa, NULL, done, private, flags)); |
34dc7c2f BB |
676 | } |
677 | ||
678 | zio_t * | |
b128c09f BB |
679 | zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, |
680 | void *data, uint64_t size, zio_done_func_t *done, void *private, | |
428870ff | 681 | int priority, enum zio_flag flags, const zbookmark_t *zb) |
34dc7c2f BB |
682 | { |
683 | zio_t *zio; | |
684 | ||
428870ff | 685 | zio = zio_create(pio, spa, BP_PHYSICAL_BIRTH(bp), bp, |
b128c09f BB |
686 | data, size, done, private, |
687 | ZIO_TYPE_READ, priority, flags, NULL, 0, zb, | |
428870ff BB |
688 | ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? |
689 | ZIO_DDT_CHILD_READ_PIPELINE : ZIO_READ_PIPELINE); | |
34dc7c2f | 690 | |
b128c09f BB |
691 | return (zio); |
692 | } | |
34dc7c2f | 693 | |
34dc7c2f | 694 | zio_t * |
b128c09f | 695 | zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, |
428870ff | 696 | void *data, uint64_t size, const zio_prop_t *zp, |
b128c09f | 697 | zio_done_func_t *ready, zio_done_func_t *done, void *private, |
428870ff | 698 | int priority, enum zio_flag flags, const zbookmark_t *zb) |
34dc7c2f BB |
699 | { |
700 | zio_t *zio; | |
701 | ||
b128c09f BB |
702 | ASSERT(zp->zp_checksum >= ZIO_CHECKSUM_OFF && |
703 | zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS && | |
704 | zp->zp_compress >= ZIO_COMPRESS_OFF && | |
705 | zp->zp_compress < ZIO_COMPRESS_FUNCTIONS && | |
706 | zp->zp_type < DMU_OT_NUMTYPES && | |
707 | zp->zp_level < 32 && | |
428870ff BB |
708 | zp->zp_copies > 0 && |
709 | zp->zp_copies <= spa_max_replication(spa) && | |
710 | zp->zp_dedup <= 1 && | |
711 | zp->zp_dedup_verify <= 1); | |
34dc7c2f BB |
712 | |
713 | zio = zio_create(pio, spa, txg, bp, data, size, done, private, | |
b128c09f | 714 | ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, |
428870ff BB |
715 | ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? |
716 | ZIO_DDT_CHILD_WRITE_PIPELINE : ZIO_WRITE_PIPELINE); | |
34dc7c2f BB |
717 | |
718 | zio->io_ready = ready; | |
b128c09f | 719 | zio->io_prop = *zp; |
34dc7c2f BB |
720 | |
721 | return (zio); | |
722 | } | |
723 | ||
724 | zio_t * | |
b128c09f BB |
725 | zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data, |
726 | uint64_t size, zio_done_func_t *done, void *private, int priority, | |
428870ff | 727 | enum zio_flag flags, zbookmark_t *zb) |
34dc7c2f BB |
728 | { |
729 | zio_t *zio; | |
730 | ||
34dc7c2f | 731 | zio = zio_create(pio, spa, txg, bp, data, size, done, private, |
b128c09f BB |
732 | ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, |
733 | ZIO_STAGE_OPEN, ZIO_REWRITE_PIPELINE); | |
34dc7c2f BB |
734 | |
735 | return (zio); | |
736 | } | |
737 | ||
428870ff BB |
738 | void |
739 | zio_write_override(zio_t *zio, blkptr_t *bp, int copies) | |
740 | { | |
741 | ASSERT(zio->io_type == ZIO_TYPE_WRITE); | |
742 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
743 | ASSERT(zio->io_stage == ZIO_STAGE_OPEN); | |
744 | ASSERT(zio->io_txg == spa_syncing_txg(zio->io_spa)); | |
745 | ||
746 | zio->io_prop.zp_copies = copies; | |
747 | zio->io_bp_override = bp; | |
748 | } | |
749 | ||
750 | void | |
751 | zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp) | |
752 | { | |
753 | bplist_append(&spa->spa_free_bplist[txg & TXG_MASK], bp); | |
754 | } | |
755 | ||
34dc7c2f | 756 | zio_t * |
428870ff BB |
757 | zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, |
758 | enum zio_flag flags) | |
34dc7c2f BB |
759 | { |
760 | zio_t *zio; | |
761 | ||
428870ff BB |
762 | dprintf_bp(bp, "freeing in txg %llu, pass %u", |
763 | (longlong_t)txg, spa->spa_sync_pass); | |
34dc7c2f | 764 | |
428870ff BB |
765 | ASSERT(!BP_IS_HOLE(bp)); |
766 | ASSERT(spa_syncing_txg(spa) == txg); | |
767 | ASSERT(spa_sync_pass(spa) <= SYNC_PASS_DEFERRED_FREE); | |
34dc7c2f | 768 | |
b128c09f | 769 | zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), |
428870ff | 770 | NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_FREE, flags, |
b128c09f | 771 | NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_FREE_PIPELINE); |
34dc7c2f BB |
772 | |
773 | return (zio); | |
774 | } | |
775 | ||
776 | zio_t * | |
428870ff BB |
777 | zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, |
778 | zio_done_func_t *done, void *private, enum zio_flag flags) | |
34dc7c2f BB |
779 | { |
780 | zio_t *zio; | |
781 | ||
782 | /* | |
783 | * A claim is an allocation of a specific block. Claims are needed | |
784 | * to support immediate writes in the intent log. The issue is that | |
785 | * immediate writes contain committed data, but in a txg that was | |
786 | * *not* committed. Upon opening the pool after an unclean shutdown, | |
787 | * the intent log claims all blocks that contain immediate write data | |
788 | * so that the SPA knows they're in use. | |
789 | * | |
790 | * All claims *must* be resolved in the first txg -- before the SPA | |
791 | * starts allocating blocks -- so that nothing is allocated twice. | |
428870ff | 792 | * If txg == 0 we just verify that the block is claimable. |
34dc7c2f BB |
793 | */ |
794 | ASSERT3U(spa->spa_uberblock.ub_rootbp.blk_birth, <, spa_first_txg(spa)); | |
428870ff BB |
795 | ASSERT(txg == spa_first_txg(spa) || txg == 0); |
796 | ASSERT(!BP_GET_DEDUP(bp) || !spa_writeable(spa)); /* zdb(1M) */ | |
34dc7c2f | 797 | |
b128c09f BB |
798 | zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), |
799 | done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags, | |
800 | NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE); | |
34dc7c2f BB |
801 | |
802 | return (zio); | |
803 | } | |
804 | ||
805 | zio_t * | |
806 | zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, | |
428870ff | 807 | zio_done_func_t *done, void *private, int priority, enum zio_flag flags) |
34dc7c2f BB |
808 | { |
809 | zio_t *zio; | |
810 | int c; | |
811 | ||
812 | if (vd->vdev_children == 0) { | |
813 | zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, | |
b128c09f | 814 | ZIO_TYPE_IOCTL, priority, flags, vd, 0, NULL, |
34dc7c2f BB |
815 | ZIO_STAGE_OPEN, ZIO_IOCTL_PIPELINE); |
816 | ||
34dc7c2f BB |
817 | zio->io_cmd = cmd; |
818 | } else { | |
d164b209 | 819 | zio = zio_null(pio, spa, NULL, NULL, NULL, flags); |
34dc7c2f BB |
820 | |
821 | for (c = 0; c < vd->vdev_children; c++) | |
822 | zio_nowait(zio_ioctl(zio, spa, vd->vdev_child[c], cmd, | |
823 | done, private, priority, flags)); | |
824 | } | |
825 | ||
826 | return (zio); | |
827 | } | |
828 | ||
34dc7c2f BB |
829 | zio_t * |
830 | zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, | |
831 | void *data, int checksum, zio_done_func_t *done, void *private, | |
428870ff | 832 | int priority, enum zio_flag flags, boolean_t labels) |
34dc7c2f BB |
833 | { |
834 | zio_t *zio; | |
34dc7c2f | 835 | |
b128c09f BB |
836 | ASSERT(vd->vdev_children == 0); |
837 | ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || | |
838 | offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); | |
839 | ASSERT3U(offset + size, <=, vd->vdev_psize); | |
34dc7c2f | 840 | |
b128c09f BB |
841 | zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, |
842 | ZIO_TYPE_READ, priority, flags, vd, offset, NULL, | |
34dc7c2f BB |
843 | ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE); |
844 | ||
b128c09f | 845 | zio->io_prop.zp_checksum = checksum; |
34dc7c2f BB |
846 | |
847 | return (zio); | |
848 | } | |
849 | ||
850 | zio_t * | |
851 | zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, | |
852 | void *data, int checksum, zio_done_func_t *done, void *private, | |
428870ff | 853 | int priority, enum zio_flag flags, boolean_t labels) |
34dc7c2f | 854 | { |
34dc7c2f | 855 | zio_t *zio; |
34dc7c2f | 856 | |
b128c09f BB |
857 | ASSERT(vd->vdev_children == 0); |
858 | ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || | |
859 | offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); | |
860 | ASSERT3U(offset + size, <=, vd->vdev_psize); | |
34dc7c2f | 861 | |
b128c09f BB |
862 | zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, |
863 | ZIO_TYPE_WRITE, priority, flags, vd, offset, NULL, | |
34dc7c2f BB |
864 | ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE); |
865 | ||
b128c09f | 866 | zio->io_prop.zp_checksum = checksum; |
34dc7c2f | 867 | |
428870ff | 868 | if (zio_checksum_table[checksum].ci_eck) { |
34dc7c2f | 869 | /* |
428870ff | 870 | * zec checksums are necessarily destructive -- they modify |
b128c09f | 871 | * the end of the write buffer to hold the verifier/checksum. |
34dc7c2f | 872 | * Therefore, we must make a local copy in case the data is |
b128c09f | 873 | * being written to multiple places in parallel. |
34dc7c2f | 874 | */ |
b128c09f | 875 | void *wbuf = zio_buf_alloc(size); |
34dc7c2f | 876 | bcopy(data, wbuf, size); |
b128c09f | 877 | zio_push_transform(zio, wbuf, size, size, NULL); |
34dc7c2f BB |
878 | } |
879 | ||
880 | return (zio); | |
881 | } | |
882 | ||
883 | /* | |
b128c09f | 884 | * Create a child I/O to do some work for us. |
34dc7c2f BB |
885 | */ |
886 | zio_t * | |
b128c09f | 887 | zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset, |
428870ff | 888 | void *data, uint64_t size, int type, int priority, enum zio_flag flags, |
34dc7c2f BB |
889 | zio_done_func_t *done, void *private) |
890 | { | |
428870ff | 891 | enum zio_stage pipeline = ZIO_VDEV_CHILD_PIPELINE; |
b128c09f BB |
892 | zio_t *zio; |
893 | ||
894 | ASSERT(vd->vdev_parent == | |
895 | (pio->io_vd ? pio->io_vd : pio->io_spa->spa_root_vdev)); | |
34dc7c2f BB |
896 | |
897 | if (type == ZIO_TYPE_READ && bp != NULL) { | |
898 | /* | |
899 | * If we have the bp, then the child should perform the | |
900 | * checksum and the parent need not. This pushes error | |
901 | * detection as close to the leaves as possible and | |
902 | * eliminates redundant checksums in the interior nodes. | |
903 | */ | |
428870ff BB |
904 | pipeline |= ZIO_STAGE_CHECKSUM_VERIFY; |
905 | pio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; | |
34dc7c2f BB |
906 | } |
907 | ||
b128c09f BB |
908 | if (vd->vdev_children == 0) |
909 | offset += VDEV_LABEL_START_SIZE; | |
910 | ||
428870ff BB |
911 | flags |= ZIO_VDEV_CHILD_FLAGS(pio) | ZIO_FLAG_DONT_PROPAGATE; |
912 | ||
913 | /* | |
914 | * If we've decided to do a repair, the write is not speculative -- | |
915 | * even if the original read was. | |
916 | */ | |
917 | if (flags & ZIO_FLAG_IO_REPAIR) | |
918 | flags &= ~ZIO_FLAG_SPECULATIVE; | |
919 | ||
b128c09f | 920 | zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size, |
428870ff BB |
921 | done, private, type, priority, flags, vd, offset, &pio->io_bookmark, |
922 | ZIO_STAGE_VDEV_IO_START >> 1, pipeline); | |
34dc7c2f | 923 | |
b128c09f | 924 | return (zio); |
34dc7c2f BB |
925 | } |
926 | ||
b128c09f BB |
927 | zio_t * |
928 | zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, void *data, uint64_t size, | |
428870ff BB |
929 | int type, int priority, enum zio_flag flags, |
930 | zio_done_func_t *done, void *private) | |
34dc7c2f | 931 | { |
b128c09f | 932 | zio_t *zio; |
34dc7c2f | 933 | |
b128c09f | 934 | ASSERT(vd->vdev_ops->vdev_op_leaf); |
34dc7c2f | 935 | |
b128c09f BB |
936 | zio = zio_create(NULL, vd->vdev_spa, 0, NULL, |
937 | data, size, done, private, type, priority, | |
938 | flags | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY, | |
939 | vd, offset, NULL, | |
428870ff | 940 | ZIO_STAGE_VDEV_IO_START >> 1, ZIO_VDEV_CHILD_PIPELINE); |
34dc7c2f | 941 | |
b128c09f | 942 | return (zio); |
34dc7c2f BB |
943 | } |
944 | ||
945 | void | |
b128c09f | 946 | zio_flush(zio_t *zio, vdev_t *vd) |
34dc7c2f | 947 | { |
b128c09f BB |
948 | zio_nowait(zio_ioctl(zio, zio->io_spa, vd, DKIOCFLUSHWRITECACHE, |
949 | NULL, NULL, ZIO_PRIORITY_NOW, | |
950 | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY)); | |
34dc7c2f BB |
951 | } |
952 | ||
428870ff BB |
953 | void |
954 | zio_shrink(zio_t *zio, uint64_t size) | |
955 | { | |
956 | ASSERT(zio->io_executor == NULL); | |
957 | ASSERT(zio->io_orig_size == zio->io_size); | |
958 | ASSERT(size <= zio->io_size); | |
959 | ||
960 | /* | |
961 | * We don't shrink for raidz because of problems with the | |
962 | * reconstruction when reading back less than the block size. | |
963 | * Note, BP_IS_RAIDZ() assumes no compression. | |
964 | */ | |
965 | ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); | |
966 | if (!BP_IS_RAIDZ(zio->io_bp)) | |
967 | zio->io_orig_size = zio->io_size = size; | |
968 | } | |
969 | ||
34dc7c2f BB |
970 | /* |
971 | * ========================================================================== | |
b128c09f | 972 | * Prepare to read and write logical blocks |
34dc7c2f BB |
973 | * ========================================================================== |
974 | */ | |
b128c09f | 975 | |
34dc7c2f | 976 | static int |
b128c09f | 977 | zio_read_bp_init(zio_t *zio) |
34dc7c2f | 978 | { |
b128c09f | 979 | blkptr_t *bp = zio->io_bp; |
34dc7c2f | 980 | |
fb5f0bc8 | 981 | if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF && |
9babb374 BB |
982 | zio->io_child_type == ZIO_CHILD_LOGICAL && |
983 | !(zio->io_flags & ZIO_FLAG_RAW)) { | |
428870ff BB |
984 | uint64_t psize = BP_GET_PSIZE(bp); |
985 | void *cbuf = zio_buf_alloc(psize); | |
b128c09f | 986 | |
428870ff | 987 | zio_push_transform(zio, cbuf, psize, psize, zio_decompress); |
34dc7c2f | 988 | } |
34dc7c2f | 989 | |
b128c09f BB |
990 | if (!dmu_ot[BP_GET_TYPE(bp)].ot_metadata && BP_GET_LEVEL(bp) == 0) |
991 | zio->io_flags |= ZIO_FLAG_DONT_CACHE; | |
992 | ||
428870ff BB |
993 | if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP) |
994 | zio->io_flags |= ZIO_FLAG_DONT_CACHE; | |
995 | ||
996 | if (BP_GET_DEDUP(bp) && zio->io_child_type == ZIO_CHILD_LOGICAL) | |
997 | zio->io_pipeline = ZIO_DDT_READ_PIPELINE; | |
998 | ||
b128c09f | 999 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
1000 | } |
1001 | ||
b128c09f BB |
1002 | static int |
1003 | zio_write_bp_init(zio_t *zio) | |
34dc7c2f | 1004 | { |
428870ff | 1005 | spa_t *spa = zio->io_spa; |
b128c09f | 1006 | zio_prop_t *zp = &zio->io_prop; |
428870ff | 1007 | enum zio_compress compress = zp->zp_compress; |
34dc7c2f | 1008 | blkptr_t *bp = zio->io_bp; |
b128c09f | 1009 | uint64_t lsize = zio->io_size; |
428870ff | 1010 | uint64_t psize = lsize; |
b128c09f | 1011 | int pass = 1; |
34dc7c2f | 1012 | |
b128c09f BB |
1013 | /* |
1014 | * If our children haven't all reached the ready stage, | |
1015 | * wait for them and then repeat this pipeline stage. | |
1016 | */ | |
1017 | if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || | |
1018 | zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_READY)) | |
1019 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 1020 | |
b128c09f BB |
1021 | if (!IO_IS_ALLOCATING(zio)) |
1022 | return (ZIO_PIPELINE_CONTINUE); | |
34dc7c2f | 1023 | |
428870ff BB |
1024 | ASSERT(zio->io_child_type != ZIO_CHILD_DDT); |
1025 | ||
1026 | if (zio->io_bp_override) { | |
1027 | ASSERT(bp->blk_birth != zio->io_txg); | |
1028 | ASSERT(BP_GET_DEDUP(zio->io_bp_override) == 0); | |
1029 | ||
1030 | *bp = *zio->io_bp_override; | |
1031 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
1032 | ||
1033 | if (BP_IS_HOLE(bp) || !zp->zp_dedup) | |
1034 | return (ZIO_PIPELINE_CONTINUE); | |
1035 | ||
1036 | ASSERT(zio_checksum_table[zp->zp_checksum].ci_dedup || | |
1037 | zp->zp_dedup_verify); | |
1038 | ||
1039 | if (BP_GET_CHECKSUM(bp) == zp->zp_checksum) { | |
1040 | BP_SET_DEDUP(bp, 1); | |
1041 | zio->io_pipeline |= ZIO_STAGE_DDT_WRITE; | |
1042 | return (ZIO_PIPELINE_CONTINUE); | |
1043 | } | |
1044 | zio->io_bp_override = NULL; | |
1045 | BP_ZERO(bp); | |
1046 | } | |
34dc7c2f | 1047 | |
b128c09f BB |
1048 | if (bp->blk_birth == zio->io_txg) { |
1049 | /* | |
1050 | * We're rewriting an existing block, which means we're | |
1051 | * working on behalf of spa_sync(). For spa_sync() to | |
1052 | * converge, it must eventually be the case that we don't | |
1053 | * have to allocate new blocks. But compression changes | |
1054 | * the blocksize, which forces a reallocate, and makes | |
1055 | * convergence take longer. Therefore, after the first | |
1056 | * few passes, stop compressing to ensure convergence. | |
1057 | */ | |
428870ff BB |
1058 | pass = spa_sync_pass(spa); |
1059 | ||
1060 | ASSERT(zio->io_txg == spa_syncing_txg(spa)); | |
1061 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1062 | ASSERT(!BP_GET_DEDUP(bp)); | |
34dc7c2f | 1063 | |
b128c09f BB |
1064 | if (pass > SYNC_PASS_DONT_COMPRESS) |
1065 | compress = ZIO_COMPRESS_OFF; | |
34dc7c2f | 1066 | |
b128c09f | 1067 | /* Make sure someone doesn't change their mind on overwrites */ |
428870ff BB |
1068 | ASSERT(MIN(zp->zp_copies + BP_IS_GANG(bp), |
1069 | spa_max_replication(spa)) == BP_GET_NDVAS(bp)); | |
b128c09f | 1070 | } |
34dc7c2f | 1071 | |
b128c09f | 1072 | if (compress != ZIO_COMPRESS_OFF) { |
428870ff BB |
1073 | void *cbuf = zio_buf_alloc(lsize); |
1074 | psize = zio_compress_data(compress, zio->io_data, cbuf, lsize); | |
1075 | if (psize == 0 || psize == lsize) { | |
b128c09f | 1076 | compress = ZIO_COMPRESS_OFF; |
428870ff BB |
1077 | zio_buf_free(cbuf, lsize); |
1078 | } else { | |
1079 | ASSERT(psize < lsize); | |
1080 | zio_push_transform(zio, cbuf, psize, lsize, NULL); | |
b128c09f BB |
1081 | } |
1082 | } | |
34dc7c2f | 1083 | |
b128c09f BB |
1084 | /* |
1085 | * The final pass of spa_sync() must be all rewrites, but the first | |
1086 | * few passes offer a trade-off: allocating blocks defers convergence, | |
1087 | * but newly allocated blocks are sequential, so they can be written | |
1088 | * to disk faster. Therefore, we allow the first few passes of | |
1089 | * spa_sync() to allocate new blocks, but force rewrites after that. | |
1090 | * There should only be a handful of blocks after pass 1 in any case. | |
1091 | */ | |
428870ff | 1092 | if (bp->blk_birth == zio->io_txg && BP_GET_PSIZE(bp) == psize && |
b128c09f | 1093 | pass > SYNC_PASS_REWRITE) { |
428870ff | 1094 | enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES; |
d6320ddb | 1095 | ASSERT(psize != 0); |
b128c09f BB |
1096 | zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages; |
1097 | zio->io_flags |= ZIO_FLAG_IO_REWRITE; | |
1098 | } else { | |
1099 | BP_ZERO(bp); | |
1100 | zio->io_pipeline = ZIO_WRITE_PIPELINE; | |
1101 | } | |
34dc7c2f | 1102 | |
428870ff | 1103 | if (psize == 0) { |
b128c09f BB |
1104 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; |
1105 | } else { | |
1106 | ASSERT(zp->zp_checksum != ZIO_CHECKSUM_GANG_HEADER); | |
1107 | BP_SET_LSIZE(bp, lsize); | |
428870ff | 1108 | BP_SET_PSIZE(bp, psize); |
b128c09f BB |
1109 | BP_SET_COMPRESS(bp, compress); |
1110 | BP_SET_CHECKSUM(bp, zp->zp_checksum); | |
1111 | BP_SET_TYPE(bp, zp->zp_type); | |
1112 | BP_SET_LEVEL(bp, zp->zp_level); | |
428870ff | 1113 | BP_SET_DEDUP(bp, zp->zp_dedup); |
b128c09f | 1114 | BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); |
428870ff BB |
1115 | if (zp->zp_dedup) { |
1116 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1117 | ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); | |
1118 | zio->io_pipeline = ZIO_DDT_WRITE_PIPELINE; | |
1119 | } | |
1120 | } | |
1121 | ||
1122 | return (ZIO_PIPELINE_CONTINUE); | |
1123 | } | |
1124 | ||
1125 | static int | |
1126 | zio_free_bp_init(zio_t *zio) | |
1127 | { | |
1128 | blkptr_t *bp = zio->io_bp; | |
1129 | ||
1130 | if (zio->io_child_type == ZIO_CHILD_LOGICAL) { | |
1131 | if (BP_GET_DEDUP(bp)) | |
1132 | zio->io_pipeline = ZIO_DDT_FREE_PIPELINE; | |
b128c09f | 1133 | } |
34dc7c2f BB |
1134 | |
1135 | return (ZIO_PIPELINE_CONTINUE); | |
1136 | } | |
1137 | ||
b128c09f BB |
1138 | /* |
1139 | * ========================================================================== | |
1140 | * Execute the I/O pipeline | |
1141 | * ========================================================================== | |
1142 | */ | |
1143 | ||
1144 | static void | |
428870ff | 1145 | zio_taskq_dispatch(zio_t *zio, enum zio_taskq_type q, boolean_t cutinline) |
34dc7c2f | 1146 | { |
428870ff | 1147 | spa_t *spa = zio->io_spa; |
b128c09f | 1148 | zio_type_t t = zio->io_type; |
a38718a6 | 1149 | int flags = (cutinline ? TQ_FRONT : 0); |
34dc7c2f BB |
1150 | |
1151 | /* | |
9babb374 BB |
1152 | * If we're a config writer or a probe, the normal issue and |
1153 | * interrupt threads may all be blocked waiting for the config lock. | |
1154 | * In this case, select the otherwise-unused taskq for ZIO_TYPE_NULL. | |
34dc7c2f | 1155 | */ |
9babb374 | 1156 | if (zio->io_flags & (ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_PROBE)) |
b128c09f | 1157 | t = ZIO_TYPE_NULL; |
34dc7c2f BB |
1158 | |
1159 | /* | |
b128c09f | 1160 | * A similar issue exists for the L2ARC write thread until L2ARC 2.0. |
34dc7c2f | 1161 | */ |
b128c09f BB |
1162 | if (t == ZIO_TYPE_WRITE && zio->io_vd && zio->io_vd->vdev_aux) |
1163 | t = ZIO_TYPE_NULL; | |
34dc7c2f | 1164 | |
428870ff BB |
1165 | /* |
1166 | * If this is a high priority I/O, then use the high priority taskq. | |
1167 | */ | |
1168 | if (zio->io_priority == ZIO_PRIORITY_NOW && | |
1169 | spa->spa_zio_taskq[t][q + 1] != NULL) | |
1170 | q++; | |
1171 | ||
1172 | ASSERT3U(q, <, ZIO_TASKQ_TYPES); | |
5cc556b4 | 1173 | |
a38718a6 GA |
1174 | /* |
1175 | * NB: We are assuming that the zio can only be dispatched | |
1176 | * to a single taskq at a time. It would be a grievous error | |
1177 | * to dispatch the zio to another taskq at the same time. | |
1178 | */ | |
1179 | ASSERT(taskq_empty_ent(&zio->io_tqent)); | |
1180 | taskq_dispatch_ent(spa->spa_zio_taskq[t][q], | |
1181 | (task_func_t *)zio_execute, zio, flags, &zio->io_tqent); | |
b128c09f | 1182 | } |
34dc7c2f | 1183 | |
b128c09f BB |
1184 | static boolean_t |
1185 | zio_taskq_member(zio_t *zio, enum zio_taskq_type q) | |
1186 | { | |
1187 | kthread_t *executor = zio->io_executor; | |
1188 | spa_t *spa = zio->io_spa; | |
d6320ddb | 1189 | zio_type_t t; |
34dc7c2f | 1190 | |
d6320ddb | 1191 | for (t = 0; t < ZIO_TYPES; t++) |
b128c09f BB |
1192 | if (taskq_member(spa->spa_zio_taskq[t][q], executor)) |
1193 | return (B_TRUE); | |
34dc7c2f | 1194 | |
b128c09f BB |
1195 | return (B_FALSE); |
1196 | } | |
34dc7c2f | 1197 | |
b128c09f BB |
1198 | static int |
1199 | zio_issue_async(zio_t *zio) | |
1200 | { | |
428870ff | 1201 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); |
b128c09f BB |
1202 | |
1203 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f BB |
1204 | } |
1205 | ||
b128c09f BB |
1206 | void |
1207 | zio_interrupt(zio_t *zio) | |
34dc7c2f | 1208 | { |
428870ff | 1209 | zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT, B_FALSE); |
b128c09f | 1210 | } |
34dc7c2f | 1211 | |
b128c09f BB |
1212 | /* |
1213 | * Execute the I/O pipeline until one of the following occurs: | |
1214 | * (1) the I/O completes; (2) the pipeline stalls waiting for | |
1215 | * dependent child I/Os; (3) the I/O issues, so we're waiting | |
1216 | * for an I/O completion interrupt; (4) the I/O is delegated by | |
1217 | * vdev-level caching or aggregation; (5) the I/O is deferred | |
1218 | * due to vdev-level queueing; (6) the I/O is handed off to | |
1219 | * another thread. In all cases, the pipeline stops whenever | |
1220 | * there's no CPU work; it never burns a thread in cv_wait(). | |
1221 | * | |
1222 | * There's no locking on io_stage because there's no legitimate way | |
1223 | * for multiple threads to be attempting to process the same I/O. | |
1224 | */ | |
428870ff | 1225 | static zio_pipe_stage_t *zio_pipeline[]; |
34dc7c2f | 1226 | |
da6b4005 NB |
1227 | /* |
1228 | * zio_execute() is a wrapper around the static function | |
1229 | * __zio_execute() so that we can force __zio_execute() to be | |
1230 | * inlined. This reduces stack overhead which is important | |
1231 | * because __zio_execute() is called recursively in several zio | |
1232 | * code paths. zio_execute() itself cannot be inlined because | |
1233 | * it is externally visible. | |
1234 | */ | |
b128c09f BB |
1235 | void |
1236 | zio_execute(zio_t *zio) | |
da6b4005 NB |
1237 | { |
1238 | __zio_execute(zio); | |
1239 | } | |
1240 | ||
1241 | __attribute__((always_inline)) | |
1242 | static inline void | |
1243 | __zio_execute(zio_t *zio) | |
b128c09f BB |
1244 | { |
1245 | zio->io_executor = curthread; | |
34dc7c2f | 1246 | |
b128c09f | 1247 | while (zio->io_stage < ZIO_STAGE_DONE) { |
428870ff BB |
1248 | enum zio_stage pipeline = zio->io_pipeline; |
1249 | enum zio_stage stage = zio->io_stage; | |
2fac4c2a BB |
1250 | dsl_pool_t *dsl; |
1251 | boolean_t cut; | |
b128c09f | 1252 | int rv; |
34dc7c2f | 1253 | |
b128c09f | 1254 | ASSERT(!MUTEX_HELD(&zio->io_lock)); |
428870ff BB |
1255 | ASSERT(ISP2(stage)); |
1256 | ASSERT(zio->io_stall == NULL); | |
34dc7c2f | 1257 | |
428870ff BB |
1258 | do { |
1259 | stage <<= 1; | |
1260 | } while ((stage & pipeline) == 0); | |
b128c09f BB |
1261 | |
1262 | ASSERT(stage <= ZIO_STAGE_DONE); | |
34dc7c2f | 1263 | |
2fac4c2a BB |
1264 | dsl = spa_get_dsl(zio->io_spa); |
1265 | cut = (stage == ZIO_STAGE_VDEV_IO_START) ? | |
1266 | zio_requeue_io_start_cut_in_line : B_FALSE; | |
1267 | ||
34dc7c2f | 1268 | /* |
b128c09f BB |
1269 | * If we are in interrupt context and this pipeline stage |
1270 | * will grab a config lock that is held across I/O, | |
428870ff BB |
1271 | * or may wait for an I/O that needs an interrupt thread |
1272 | * to complete, issue async to avoid deadlock. | |
1273 | * | |
2fac4c2a BB |
1274 | * If we are in the txg_sync_thread or being called |
1275 | * during pool init issue async to minimize stack depth. | |
1276 | * Both of these call paths may be recursively called. | |
1277 | * | |
428870ff BB |
1278 | * For VDEV_IO_START, we cut in line so that the io will |
1279 | * be sent to disk promptly. | |
34dc7c2f | 1280 | */ |
2fac4c2a BB |
1281 | if (((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL && |
1282 | zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) || | |
1283 | (dsl != NULL && dsl_pool_sync_context(dsl))) { | |
428870ff | 1284 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); |
b128c09f | 1285 | return; |
34dc7c2f BB |
1286 | } |
1287 | ||
b128c09f | 1288 | zio->io_stage = stage; |
428870ff | 1289 | rv = zio_pipeline[highbit(stage) - 1](zio); |
34dc7c2f | 1290 | |
b128c09f BB |
1291 | if (rv == ZIO_PIPELINE_STOP) |
1292 | return; | |
34dc7c2f | 1293 | |
b128c09f BB |
1294 | ASSERT(rv == ZIO_PIPELINE_CONTINUE); |
1295 | } | |
34dc7c2f BB |
1296 | } |
1297 | ||
da6b4005 | 1298 | |
b128c09f BB |
1299 | /* |
1300 | * ========================================================================== | |
1301 | * Initiate I/O, either sync or async | |
1302 | * ========================================================================== | |
1303 | */ | |
1304 | int | |
1305 | zio_wait(zio_t *zio) | |
34dc7c2f | 1306 | { |
b128c09f | 1307 | int error; |
34dc7c2f | 1308 | |
b128c09f BB |
1309 | ASSERT(zio->io_stage == ZIO_STAGE_OPEN); |
1310 | ASSERT(zio->io_executor == NULL); | |
34dc7c2f | 1311 | |
b128c09f | 1312 | zio->io_waiter = curthread; |
34dc7c2f | 1313 | |
da6b4005 | 1314 | __zio_execute(zio); |
34dc7c2f | 1315 | |
b128c09f BB |
1316 | mutex_enter(&zio->io_lock); |
1317 | while (zio->io_executor != NULL) | |
1318 | cv_wait(&zio->io_cv, &zio->io_lock); | |
1319 | mutex_exit(&zio->io_lock); | |
34dc7c2f | 1320 | |
b128c09f BB |
1321 | error = zio->io_error; |
1322 | zio_destroy(zio); | |
34dc7c2f | 1323 | |
b128c09f BB |
1324 | return (error); |
1325 | } | |
34dc7c2f | 1326 | |
b128c09f BB |
1327 | void |
1328 | zio_nowait(zio_t *zio) | |
1329 | { | |
1330 | ASSERT(zio->io_executor == NULL); | |
34dc7c2f | 1331 | |
d164b209 BB |
1332 | if (zio->io_child_type == ZIO_CHILD_LOGICAL && |
1333 | zio_unique_parent(zio) == NULL) { | |
34dc7c2f | 1334 | /* |
b128c09f | 1335 | * This is a logical async I/O with no parent to wait for it. |
9babb374 BB |
1336 | * We add it to the spa_async_root_zio "Godfather" I/O which |
1337 | * will ensure they complete prior to unloading the pool. | |
34dc7c2f | 1338 | */ |
b128c09f | 1339 | spa_t *spa = zio->io_spa; |
9babb374 BB |
1340 | |
1341 | zio_add_child(spa->spa_async_zio_root, zio); | |
b128c09f | 1342 | } |
34dc7c2f | 1343 | |
da6b4005 | 1344 | __zio_execute(zio); |
b128c09f | 1345 | } |
34dc7c2f | 1346 | |
b128c09f BB |
1347 | /* |
1348 | * ========================================================================== | |
1349 | * Reexecute or suspend/resume failed I/O | |
1350 | * ========================================================================== | |
1351 | */ | |
34dc7c2f | 1352 | |
b128c09f BB |
1353 | static void |
1354 | zio_reexecute(zio_t *pio) | |
1355 | { | |
d164b209 | 1356 | zio_t *cio, *cio_next; |
d6320ddb | 1357 | int c, w; |
d164b209 BB |
1358 | |
1359 | ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL); | |
1360 | ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN); | |
9babb374 BB |
1361 | ASSERT(pio->io_gang_leader == NULL); |
1362 | ASSERT(pio->io_gang_tree == NULL); | |
34dc7c2f | 1363 | |
b128c09f BB |
1364 | pio->io_flags = pio->io_orig_flags; |
1365 | pio->io_stage = pio->io_orig_stage; | |
1366 | pio->io_pipeline = pio->io_orig_pipeline; | |
1367 | pio->io_reexecute = 0; | |
1368 | pio->io_error = 0; | |
d6320ddb | 1369 | for (w = 0; w < ZIO_WAIT_TYPES; w++) |
d164b209 | 1370 | pio->io_state[w] = 0; |
d6320ddb | 1371 | for (c = 0; c < ZIO_CHILD_TYPES; c++) |
b128c09f | 1372 | pio->io_child_error[c] = 0; |
34dc7c2f | 1373 | |
428870ff BB |
1374 | if (IO_IS_ALLOCATING(pio)) |
1375 | BP_ZERO(pio->io_bp); | |
34dc7c2f | 1376 | |
b128c09f BB |
1377 | /* |
1378 | * As we reexecute pio's children, new children could be created. | |
d164b209 | 1379 | * New children go to the head of pio's io_child_list, however, |
b128c09f | 1380 | * so we will (correctly) not reexecute them. The key is that |
d164b209 BB |
1381 | * the remainder of pio's io_child_list, from 'cio_next' onward, |
1382 | * cannot be affected by any side effects of reexecuting 'cio'. | |
b128c09f | 1383 | */ |
d164b209 BB |
1384 | for (cio = zio_walk_children(pio); cio != NULL; cio = cio_next) { |
1385 | cio_next = zio_walk_children(pio); | |
b128c09f | 1386 | mutex_enter(&pio->io_lock); |
d6320ddb | 1387 | for (w = 0; w < ZIO_WAIT_TYPES; w++) |
d164b209 | 1388 | pio->io_children[cio->io_child_type][w]++; |
b128c09f | 1389 | mutex_exit(&pio->io_lock); |
d164b209 | 1390 | zio_reexecute(cio); |
34dc7c2f | 1391 | } |
34dc7c2f | 1392 | |
b128c09f BB |
1393 | /* |
1394 | * Now that all children have been reexecuted, execute the parent. | |
9babb374 BB |
1395 | * We don't reexecute "The Godfather" I/O here as it's the |
1396 | * responsibility of the caller to wait on him. | |
b128c09f | 1397 | */ |
9babb374 | 1398 | if (!(pio->io_flags & ZIO_FLAG_GODFATHER)) |
da6b4005 | 1399 | __zio_execute(pio); |
34dc7c2f BB |
1400 | } |
1401 | ||
b128c09f BB |
1402 | void |
1403 | zio_suspend(spa_t *spa, zio_t *zio) | |
34dc7c2f | 1404 | { |
b128c09f BB |
1405 | if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC) |
1406 | fm_panic("Pool '%s' has encountered an uncorrectable I/O " | |
1407 | "failure and the failure mode property for this pool " | |
1408 | "is set to panic.", spa_name(spa)); | |
34dc7c2f | 1409 | |
b128c09f | 1410 | zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, 0, 0); |
34dc7c2f | 1411 | |
b128c09f | 1412 | mutex_enter(&spa->spa_suspend_lock); |
34dc7c2f | 1413 | |
b128c09f | 1414 | if (spa->spa_suspend_zio_root == NULL) |
9babb374 BB |
1415 | spa->spa_suspend_zio_root = zio_root(spa, NULL, NULL, |
1416 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
1417 | ZIO_FLAG_GODFATHER); | |
34dc7c2f | 1418 | |
b128c09f | 1419 | spa->spa_suspended = B_TRUE; |
34dc7c2f | 1420 | |
b128c09f | 1421 | if (zio != NULL) { |
9babb374 | 1422 | ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); |
b128c09f BB |
1423 | ASSERT(zio != spa->spa_suspend_zio_root); |
1424 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
d164b209 | 1425 | ASSERT(zio_unique_parent(zio) == NULL); |
b128c09f BB |
1426 | ASSERT(zio->io_stage == ZIO_STAGE_DONE); |
1427 | zio_add_child(spa->spa_suspend_zio_root, zio); | |
1428 | } | |
34dc7c2f | 1429 | |
b128c09f BB |
1430 | mutex_exit(&spa->spa_suspend_lock); |
1431 | } | |
34dc7c2f | 1432 | |
9babb374 | 1433 | int |
b128c09f BB |
1434 | zio_resume(spa_t *spa) |
1435 | { | |
9babb374 | 1436 | zio_t *pio; |
34dc7c2f BB |
1437 | |
1438 | /* | |
b128c09f | 1439 | * Reexecute all previously suspended i/o. |
34dc7c2f | 1440 | */ |
b128c09f BB |
1441 | mutex_enter(&spa->spa_suspend_lock); |
1442 | spa->spa_suspended = B_FALSE; | |
1443 | cv_broadcast(&spa->spa_suspend_cv); | |
1444 | pio = spa->spa_suspend_zio_root; | |
1445 | spa->spa_suspend_zio_root = NULL; | |
1446 | mutex_exit(&spa->spa_suspend_lock); | |
1447 | ||
1448 | if (pio == NULL) | |
9babb374 | 1449 | return (0); |
34dc7c2f | 1450 | |
9babb374 BB |
1451 | zio_reexecute(pio); |
1452 | return (zio_wait(pio)); | |
b128c09f BB |
1453 | } |
1454 | ||
1455 | void | |
1456 | zio_resume_wait(spa_t *spa) | |
1457 | { | |
1458 | mutex_enter(&spa->spa_suspend_lock); | |
1459 | while (spa_suspended(spa)) | |
1460 | cv_wait(&spa->spa_suspend_cv, &spa->spa_suspend_lock); | |
1461 | mutex_exit(&spa->spa_suspend_lock); | |
34dc7c2f BB |
1462 | } |
1463 | ||
1464 | /* | |
1465 | * ========================================================================== | |
b128c09f BB |
1466 | * Gang blocks. |
1467 | * | |
1468 | * A gang block is a collection of small blocks that looks to the DMU | |
1469 | * like one large block. When zio_dva_allocate() cannot find a block | |
1470 | * of the requested size, due to either severe fragmentation or the pool | |
1471 | * being nearly full, it calls zio_write_gang_block() to construct the | |
1472 | * block from smaller fragments. | |
1473 | * | |
1474 | * A gang block consists of a gang header (zio_gbh_phys_t) and up to | |
1475 | * three (SPA_GBH_NBLKPTRS) gang members. The gang header is just like | |
1476 | * an indirect block: it's an array of block pointers. It consumes | |
1477 | * only one sector and hence is allocatable regardless of fragmentation. | |
1478 | * The gang header's bps point to its gang members, which hold the data. | |
1479 | * | |
1480 | * Gang blocks are self-checksumming, using the bp's <vdev, offset, txg> | |
1481 | * as the verifier to ensure uniqueness of the SHA256 checksum. | |
1482 | * Critically, the gang block bp's blk_cksum is the checksum of the data, | |
1483 | * not the gang header. This ensures that data block signatures (needed for | |
1484 | * deduplication) are independent of how the block is physically stored. | |
1485 | * | |
1486 | * Gang blocks can be nested: a gang member may itself be a gang block. | |
1487 | * Thus every gang block is a tree in which root and all interior nodes are | |
1488 | * gang headers, and the leaves are normal blocks that contain user data. | |
1489 | * The root of the gang tree is called the gang leader. | |
1490 | * | |
1491 | * To perform any operation (read, rewrite, free, claim) on a gang block, | |
1492 | * zio_gang_assemble() first assembles the gang tree (minus data leaves) | |
1493 | * in the io_gang_tree field of the original logical i/o by recursively | |
1494 | * reading the gang leader and all gang headers below it. This yields | |
1495 | * an in-core tree containing the contents of every gang header and the | |
1496 | * bps for every constituent of the gang block. | |
1497 | * | |
1498 | * With the gang tree now assembled, zio_gang_issue() just walks the gang tree | |
1499 | * and invokes a callback on each bp. To free a gang block, zio_gang_issue() | |
1500 | * calls zio_free_gang() -- a trivial wrapper around zio_free() -- for each bp. | |
1501 | * zio_claim_gang() provides a similarly trivial wrapper for zio_claim(). | |
1502 | * zio_read_gang() is a wrapper around zio_read() that omits reading gang | |
1503 | * headers, since we already have those in io_gang_tree. zio_rewrite_gang() | |
1504 | * performs a zio_rewrite() of the data or, for gang headers, a zio_rewrite() | |
1505 | * of the gang header plus zio_checksum_compute() of the data to update the | |
1506 | * gang header's blk_cksum as described above. | |
1507 | * | |
1508 | * The two-phase assemble/issue model solves the problem of partial failure -- | |
1509 | * what if you'd freed part of a gang block but then couldn't read the | |
1510 | * gang header for another part? Assembling the entire gang tree first | |
1511 | * ensures that all the necessary gang header I/O has succeeded before | |
1512 | * starting the actual work of free, claim, or write. Once the gang tree | |
1513 | * is assembled, free and claim are in-memory operations that cannot fail. | |
1514 | * | |
1515 | * In the event that a gang write fails, zio_dva_unallocate() walks the | |
1516 | * gang tree to immediately free (i.e. insert back into the space map) | |
1517 | * everything we've allocated. This ensures that we don't get ENOSPC | |
1518 | * errors during repeated suspend/resume cycles due to a flaky device. | |
1519 | * | |
1520 | * Gang rewrites only happen during sync-to-convergence. If we can't assemble | |
1521 | * the gang tree, we won't modify the block, so we can safely defer the free | |
1522 | * (knowing that the block is still intact). If we *can* assemble the gang | |
1523 | * tree, then even if some of the rewrites fail, zio_dva_unallocate() will free | |
1524 | * each constituent bp and we can allocate a new block on the next sync pass. | |
1525 | * | |
1526 | * In all cases, the gang tree allows complete recovery from partial failure. | |
34dc7c2f BB |
1527 | * ========================================================================== |
1528 | */ | |
b128c09f BB |
1529 | |
1530 | static zio_t * | |
1531 | zio_read_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
34dc7c2f | 1532 | { |
b128c09f BB |
1533 | if (gn != NULL) |
1534 | return (pio); | |
34dc7c2f | 1535 | |
b128c09f BB |
1536 | return (zio_read(pio, pio->io_spa, bp, data, BP_GET_PSIZE(bp), |
1537 | NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), | |
1538 | &pio->io_bookmark)); | |
1539 | } | |
1540 | ||
1541 | zio_t * | |
1542 | zio_rewrite_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
1543 | { | |
1544 | zio_t *zio; | |
1545 | ||
1546 | if (gn != NULL) { | |
1547 | zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, | |
1548 | gn->gn_gbh, SPA_GANGBLOCKSIZE, NULL, NULL, pio->io_priority, | |
1549 | ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); | |
34dc7c2f | 1550 | /* |
b128c09f BB |
1551 | * As we rewrite each gang header, the pipeline will compute |
1552 | * a new gang block header checksum for it; but no one will | |
1553 | * compute a new data checksum, so we do that here. The one | |
1554 | * exception is the gang leader: the pipeline already computed | |
1555 | * its data checksum because that stage precedes gang assembly. | |
1556 | * (Presently, nothing actually uses interior data checksums; | |
1557 | * this is just good hygiene.) | |
34dc7c2f | 1558 | */ |
9babb374 | 1559 | if (gn != pio->io_gang_leader->io_gang_tree) { |
b128c09f BB |
1560 | zio_checksum_compute(zio, BP_GET_CHECKSUM(bp), |
1561 | data, BP_GET_PSIZE(bp)); | |
1562 | } | |
428870ff BB |
1563 | /* |
1564 | * If we are here to damage data for testing purposes, | |
1565 | * leave the GBH alone so that we can detect the damage. | |
1566 | */ | |
1567 | if (pio->io_gang_leader->io_flags & ZIO_FLAG_INDUCE_DAMAGE) | |
1568 | zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; | |
34dc7c2f | 1569 | } else { |
b128c09f BB |
1570 | zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, |
1571 | data, BP_GET_PSIZE(bp), NULL, NULL, pio->io_priority, | |
1572 | ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); | |
34dc7c2f BB |
1573 | } |
1574 | ||
b128c09f BB |
1575 | return (zio); |
1576 | } | |
34dc7c2f | 1577 | |
b128c09f BB |
1578 | /* ARGSUSED */ |
1579 | zio_t * | |
1580 | zio_free_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
1581 | { | |
428870ff BB |
1582 | return (zio_free_sync(pio, pio->io_spa, pio->io_txg, bp, |
1583 | ZIO_GANG_CHILD_FLAGS(pio))); | |
34dc7c2f BB |
1584 | } |
1585 | ||
b128c09f BB |
1586 | /* ARGSUSED */ |
1587 | zio_t * | |
1588 | zio_claim_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
34dc7c2f | 1589 | { |
b128c09f BB |
1590 | return (zio_claim(pio, pio->io_spa, pio->io_txg, bp, |
1591 | NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio))); | |
1592 | } | |
1593 | ||
1594 | static zio_gang_issue_func_t *zio_gang_issue_func[ZIO_TYPES] = { | |
1595 | NULL, | |
1596 | zio_read_gang, | |
1597 | zio_rewrite_gang, | |
1598 | zio_free_gang, | |
1599 | zio_claim_gang, | |
1600 | NULL | |
1601 | }; | |
34dc7c2f | 1602 | |
b128c09f | 1603 | static void zio_gang_tree_assemble_done(zio_t *zio); |
34dc7c2f | 1604 | |
b128c09f BB |
1605 | static zio_gang_node_t * |
1606 | zio_gang_node_alloc(zio_gang_node_t **gnpp) | |
1607 | { | |
1608 | zio_gang_node_t *gn; | |
34dc7c2f | 1609 | |
b128c09f | 1610 | ASSERT(*gnpp == NULL); |
34dc7c2f | 1611 | |
691f6ac4 | 1612 | gn = kmem_zalloc(sizeof (*gn), KM_PUSHPAGE); |
b128c09f BB |
1613 | gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE); |
1614 | *gnpp = gn; | |
34dc7c2f | 1615 | |
b128c09f | 1616 | return (gn); |
34dc7c2f BB |
1617 | } |
1618 | ||
34dc7c2f | 1619 | static void |
b128c09f | 1620 | zio_gang_node_free(zio_gang_node_t **gnpp) |
34dc7c2f | 1621 | { |
b128c09f | 1622 | zio_gang_node_t *gn = *gnpp; |
d6320ddb | 1623 | int g; |
34dc7c2f | 1624 | |
d6320ddb | 1625 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) |
b128c09f BB |
1626 | ASSERT(gn->gn_child[g] == NULL); |
1627 | ||
1628 | zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE); | |
1629 | kmem_free(gn, sizeof (*gn)); | |
1630 | *gnpp = NULL; | |
34dc7c2f BB |
1631 | } |
1632 | ||
b128c09f BB |
1633 | static void |
1634 | zio_gang_tree_free(zio_gang_node_t **gnpp) | |
34dc7c2f | 1635 | { |
b128c09f | 1636 | zio_gang_node_t *gn = *gnpp; |
d6320ddb | 1637 | int g; |
34dc7c2f | 1638 | |
b128c09f BB |
1639 | if (gn == NULL) |
1640 | return; | |
34dc7c2f | 1641 | |
d6320ddb | 1642 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) |
b128c09f | 1643 | zio_gang_tree_free(&gn->gn_child[g]); |
34dc7c2f | 1644 | |
b128c09f | 1645 | zio_gang_node_free(gnpp); |
34dc7c2f BB |
1646 | } |
1647 | ||
b128c09f | 1648 | static void |
9babb374 | 1649 | zio_gang_tree_assemble(zio_t *gio, blkptr_t *bp, zio_gang_node_t **gnpp) |
34dc7c2f | 1650 | { |
b128c09f BB |
1651 | zio_gang_node_t *gn = zio_gang_node_alloc(gnpp); |
1652 | ||
9babb374 | 1653 | ASSERT(gio->io_gang_leader == gio); |
b128c09f | 1654 | ASSERT(BP_IS_GANG(bp)); |
34dc7c2f | 1655 | |
9babb374 | 1656 | zio_nowait(zio_read(gio, gio->io_spa, bp, gn->gn_gbh, |
b128c09f | 1657 | SPA_GANGBLOCKSIZE, zio_gang_tree_assemble_done, gn, |
9babb374 | 1658 | gio->io_priority, ZIO_GANG_CHILD_FLAGS(gio), &gio->io_bookmark)); |
b128c09f | 1659 | } |
34dc7c2f | 1660 | |
b128c09f BB |
1661 | static void |
1662 | zio_gang_tree_assemble_done(zio_t *zio) | |
1663 | { | |
9babb374 | 1664 | zio_t *gio = zio->io_gang_leader; |
b128c09f BB |
1665 | zio_gang_node_t *gn = zio->io_private; |
1666 | blkptr_t *bp = zio->io_bp; | |
d6320ddb | 1667 | int g; |
34dc7c2f | 1668 | |
9babb374 | 1669 | ASSERT(gio == zio_unique_parent(zio)); |
428870ff | 1670 | ASSERT(zio->io_child_count == 0); |
34dc7c2f | 1671 | |
b128c09f BB |
1672 | if (zio->io_error) |
1673 | return; | |
34dc7c2f | 1674 | |
b128c09f BB |
1675 | if (BP_SHOULD_BYTESWAP(bp)) |
1676 | byteswap_uint64_array(zio->io_data, zio->io_size); | |
34dc7c2f | 1677 | |
b128c09f BB |
1678 | ASSERT(zio->io_data == gn->gn_gbh); |
1679 | ASSERT(zio->io_size == SPA_GANGBLOCKSIZE); | |
428870ff | 1680 | ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); |
34dc7c2f | 1681 | |
d6320ddb | 1682 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) { |
b128c09f BB |
1683 | blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; |
1684 | if (!BP_IS_GANG(gbp)) | |
1685 | continue; | |
9babb374 | 1686 | zio_gang_tree_assemble(gio, gbp, &gn->gn_child[g]); |
b128c09f | 1687 | } |
34dc7c2f BB |
1688 | } |
1689 | ||
b128c09f BB |
1690 | static void |
1691 | zio_gang_tree_issue(zio_t *pio, zio_gang_node_t *gn, blkptr_t *bp, void *data) | |
34dc7c2f | 1692 | { |
9babb374 | 1693 | zio_t *gio = pio->io_gang_leader; |
b128c09f | 1694 | zio_t *zio; |
d6320ddb | 1695 | int g; |
34dc7c2f | 1696 | |
b128c09f | 1697 | ASSERT(BP_IS_GANG(bp) == !!gn); |
9babb374 BB |
1698 | ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp)); |
1699 | ASSERT(BP_GET_LSIZE(bp) == BP_GET_PSIZE(bp) || gn == gio->io_gang_tree); | |
34dc7c2f | 1700 | |
b128c09f BB |
1701 | /* |
1702 | * If you're a gang header, your data is in gn->gn_gbh. | |
1703 | * If you're a gang member, your data is in 'data' and gn == NULL. | |
1704 | */ | |
9babb374 | 1705 | zio = zio_gang_issue_func[gio->io_type](pio, bp, gn, data); |
34dc7c2f | 1706 | |
b128c09f | 1707 | if (gn != NULL) { |
428870ff | 1708 | ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); |
34dc7c2f | 1709 | |
d6320ddb | 1710 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) { |
b128c09f BB |
1711 | blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; |
1712 | if (BP_IS_HOLE(gbp)) | |
1713 | continue; | |
1714 | zio_gang_tree_issue(zio, gn->gn_child[g], gbp, data); | |
1715 | data = (char *)data + BP_GET_PSIZE(gbp); | |
1716 | } | |
34dc7c2f BB |
1717 | } |
1718 | ||
9babb374 BB |
1719 | if (gn == gio->io_gang_tree) |
1720 | ASSERT3P((char *)gio->io_data + gio->io_size, ==, data); | |
34dc7c2f | 1721 | |
b128c09f BB |
1722 | if (zio != pio) |
1723 | zio_nowait(zio); | |
34dc7c2f BB |
1724 | } |
1725 | ||
1726 | static int | |
b128c09f | 1727 | zio_gang_assemble(zio_t *zio) |
34dc7c2f | 1728 | { |
b128c09f | 1729 | blkptr_t *bp = zio->io_bp; |
34dc7c2f | 1730 | |
9babb374 BB |
1731 | ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == NULL); |
1732 | ASSERT(zio->io_child_type > ZIO_CHILD_GANG); | |
1733 | ||
1734 | zio->io_gang_leader = zio; | |
34dc7c2f | 1735 | |
b128c09f | 1736 | zio_gang_tree_assemble(zio, bp, &zio->io_gang_tree); |
34dc7c2f BB |
1737 | |
1738 | return (ZIO_PIPELINE_CONTINUE); | |
1739 | } | |
1740 | ||
1741 | static int | |
b128c09f | 1742 | zio_gang_issue(zio_t *zio) |
34dc7c2f | 1743 | { |
b128c09f | 1744 | blkptr_t *bp = zio->io_bp; |
34dc7c2f | 1745 | |
b128c09f BB |
1746 | if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE)) |
1747 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 1748 | |
9babb374 BB |
1749 | ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == zio); |
1750 | ASSERT(zio->io_child_type > ZIO_CHILD_GANG); | |
34dc7c2f | 1751 | |
b128c09f | 1752 | if (zio->io_child_error[ZIO_CHILD_GANG] == 0) |
9babb374 | 1753 | zio_gang_tree_issue(zio, zio->io_gang_tree, bp, zio->io_data); |
b128c09f | 1754 | else |
9babb374 | 1755 | zio_gang_tree_free(&zio->io_gang_tree); |
34dc7c2f | 1756 | |
b128c09f | 1757 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; |
34dc7c2f BB |
1758 | |
1759 | return (ZIO_PIPELINE_CONTINUE); | |
1760 | } | |
1761 | ||
1762 | static void | |
b128c09f | 1763 | zio_write_gang_member_ready(zio_t *zio) |
34dc7c2f | 1764 | { |
d164b209 | 1765 | zio_t *pio = zio_unique_parent(zio); |
1fde1e37 | 1766 | ASSERTV(zio_t *gio = zio->io_gang_leader;) |
34dc7c2f BB |
1767 | dva_t *cdva = zio->io_bp->blk_dva; |
1768 | dva_t *pdva = pio->io_bp->blk_dva; | |
1769 | uint64_t asize; | |
d6320ddb | 1770 | int d; |
34dc7c2f | 1771 | |
b128c09f BB |
1772 | if (BP_IS_HOLE(zio->io_bp)) |
1773 | return; | |
1774 | ||
1775 | ASSERT(BP_IS_HOLE(&zio->io_bp_orig)); | |
1776 | ||
1777 | ASSERT(zio->io_child_type == ZIO_CHILD_GANG); | |
428870ff BB |
1778 | ASSERT3U(zio->io_prop.zp_copies, ==, gio->io_prop.zp_copies); |
1779 | ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp)); | |
1780 | ASSERT3U(pio->io_prop.zp_copies, <=, BP_GET_NDVAS(pio->io_bp)); | |
34dc7c2f | 1781 | ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp)); |
34dc7c2f BB |
1782 | |
1783 | mutex_enter(&pio->io_lock); | |
d6320ddb | 1784 | for (d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) { |
34dc7c2f BB |
1785 | ASSERT(DVA_GET_GANG(&pdva[d])); |
1786 | asize = DVA_GET_ASIZE(&pdva[d]); | |
1787 | asize += DVA_GET_ASIZE(&cdva[d]); | |
1788 | DVA_SET_ASIZE(&pdva[d], asize); | |
1789 | } | |
1790 | mutex_exit(&pio->io_lock); | |
1791 | } | |
1792 | ||
1793 | static int | |
b128c09f | 1794 | zio_write_gang_block(zio_t *pio) |
34dc7c2f | 1795 | { |
b128c09f BB |
1796 | spa_t *spa = pio->io_spa; |
1797 | blkptr_t *bp = pio->io_bp; | |
9babb374 | 1798 | zio_t *gio = pio->io_gang_leader; |
b128c09f BB |
1799 | zio_t *zio; |
1800 | zio_gang_node_t *gn, **gnpp; | |
34dc7c2f | 1801 | zio_gbh_phys_t *gbh; |
b128c09f BB |
1802 | uint64_t txg = pio->io_txg; |
1803 | uint64_t resid = pio->io_size; | |
1804 | uint64_t lsize; | |
428870ff BB |
1805 | int copies = gio->io_prop.zp_copies; |
1806 | int gbh_copies = MIN(copies + 1, spa_max_replication(spa)); | |
b128c09f | 1807 | zio_prop_t zp; |
d6320ddb | 1808 | int g, error; |
34dc7c2f | 1809 | |
428870ff BB |
1810 | error = metaslab_alloc(spa, spa_normal_class(spa), SPA_GANGBLOCKSIZE, |
1811 | bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, | |
b128c09f | 1812 | METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER); |
34dc7c2f | 1813 | if (error) { |
b128c09f | 1814 | pio->io_error = error; |
34dc7c2f BB |
1815 | return (ZIO_PIPELINE_CONTINUE); |
1816 | } | |
1817 | ||
9babb374 BB |
1818 | if (pio == gio) { |
1819 | gnpp = &gio->io_gang_tree; | |
b128c09f BB |
1820 | } else { |
1821 | gnpp = pio->io_private; | |
1822 | ASSERT(pio->io_ready == zio_write_gang_member_ready); | |
34dc7c2f BB |
1823 | } |
1824 | ||
b128c09f BB |
1825 | gn = zio_gang_node_alloc(gnpp); |
1826 | gbh = gn->gn_gbh; | |
1827 | bzero(gbh, SPA_GANGBLOCKSIZE); | |
34dc7c2f | 1828 | |
b128c09f BB |
1829 | /* |
1830 | * Create the gang header. | |
1831 | */ | |
1832 | zio = zio_rewrite(pio, spa, txg, bp, gbh, SPA_GANGBLOCKSIZE, NULL, NULL, | |
1833 | pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); | |
34dc7c2f | 1834 | |
b128c09f BB |
1835 | /* |
1836 | * Create and nowait the gang children. | |
1837 | */ | |
d6320ddb | 1838 | for (g = 0; resid != 0; resid -= lsize, g++) { |
b128c09f BB |
1839 | lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g), |
1840 | SPA_MINBLOCKSIZE); | |
1841 | ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid); | |
1842 | ||
9babb374 | 1843 | zp.zp_checksum = gio->io_prop.zp_checksum; |
b128c09f BB |
1844 | zp.zp_compress = ZIO_COMPRESS_OFF; |
1845 | zp.zp_type = DMU_OT_NONE; | |
1846 | zp.zp_level = 0; | |
428870ff BB |
1847 | zp.zp_copies = gio->io_prop.zp_copies; |
1848 | zp.zp_dedup = 0; | |
1849 | zp.zp_dedup_verify = 0; | |
b128c09f BB |
1850 | |
1851 | zio_nowait(zio_write(zio, spa, txg, &gbh->zg_blkptr[g], | |
1852 | (char *)pio->io_data + (pio->io_size - resid), lsize, &zp, | |
1853 | zio_write_gang_member_ready, NULL, &gn->gn_child[g], | |
1854 | pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), | |
1855 | &pio->io_bookmark)); | |
1856 | } | |
34dc7c2f BB |
1857 | |
1858 | /* | |
b128c09f | 1859 | * Set pio's pipeline to just wait for zio to finish. |
34dc7c2f | 1860 | */ |
b128c09f BB |
1861 | pio->io_pipeline = ZIO_INTERLOCK_PIPELINE; |
1862 | ||
1863 | zio_nowait(zio); | |
1864 | ||
1865 | return (ZIO_PIPELINE_CONTINUE); | |
34dc7c2f BB |
1866 | } |
1867 | ||
1868 | /* | |
1869 | * ========================================================================== | |
428870ff | 1870 | * Dedup |
34dc7c2f BB |
1871 | * ========================================================================== |
1872 | */ | |
428870ff BB |
1873 | static void |
1874 | zio_ddt_child_read_done(zio_t *zio) | |
1875 | { | |
1876 | blkptr_t *bp = zio->io_bp; | |
1877 | ddt_entry_t *dde = zio->io_private; | |
1878 | ddt_phys_t *ddp; | |
1879 | zio_t *pio = zio_unique_parent(zio); | |
1880 | ||
1881 | mutex_enter(&pio->io_lock); | |
1882 | ddp = ddt_phys_select(dde, bp); | |
1883 | if (zio->io_error == 0) | |
1884 | ddt_phys_clear(ddp); /* this ddp doesn't need repair */ | |
1885 | if (zio->io_error == 0 && dde->dde_repair_data == NULL) | |
1886 | dde->dde_repair_data = zio->io_data; | |
1887 | else | |
1888 | zio_buf_free(zio->io_data, zio->io_size); | |
1889 | mutex_exit(&pio->io_lock); | |
1890 | } | |
1891 | ||
1892 | static int | |
1893 | zio_ddt_read_start(zio_t *zio) | |
1894 | { | |
1895 | blkptr_t *bp = zio->io_bp; | |
d6320ddb | 1896 | int p; |
428870ff BB |
1897 | |
1898 | ASSERT(BP_GET_DEDUP(bp)); | |
1899 | ASSERT(BP_GET_PSIZE(bp) == zio->io_size); | |
1900 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1901 | ||
1902 | if (zio->io_child_error[ZIO_CHILD_DDT]) { | |
1903 | ddt_t *ddt = ddt_select(zio->io_spa, bp); | |
1904 | ddt_entry_t *dde = ddt_repair_start(ddt, bp); | |
1905 | ddt_phys_t *ddp = dde->dde_phys; | |
1906 | ddt_phys_t *ddp_self = ddt_phys_select(dde, bp); | |
1907 | blkptr_t blk; | |
1908 | ||
1909 | ASSERT(zio->io_vsd == NULL); | |
1910 | zio->io_vsd = dde; | |
1911 | ||
1912 | if (ddp_self == NULL) | |
1913 | return (ZIO_PIPELINE_CONTINUE); | |
1914 | ||
d6320ddb | 1915 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff BB |
1916 | if (ddp->ddp_phys_birth == 0 || ddp == ddp_self) |
1917 | continue; | |
1918 | ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp, | |
1919 | &blk); | |
1920 | zio_nowait(zio_read(zio, zio->io_spa, &blk, | |
1921 | zio_buf_alloc(zio->io_size), zio->io_size, | |
1922 | zio_ddt_child_read_done, dde, zio->io_priority, | |
1923 | ZIO_DDT_CHILD_FLAGS(zio) | ZIO_FLAG_DONT_PROPAGATE, | |
1924 | &zio->io_bookmark)); | |
1925 | } | |
1926 | return (ZIO_PIPELINE_CONTINUE); | |
1927 | } | |
1928 | ||
1929 | zio_nowait(zio_read(zio, zio->io_spa, bp, | |
1930 | zio->io_data, zio->io_size, NULL, NULL, zio->io_priority, | |
1931 | ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark)); | |
1932 | ||
1933 | return (ZIO_PIPELINE_CONTINUE); | |
1934 | } | |
1935 | ||
1936 | static int | |
1937 | zio_ddt_read_done(zio_t *zio) | |
1938 | { | |
1939 | blkptr_t *bp = zio->io_bp; | |
1940 | ||
1941 | if (zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE)) | |
1942 | return (ZIO_PIPELINE_STOP); | |
1943 | ||
1944 | ASSERT(BP_GET_DEDUP(bp)); | |
1945 | ASSERT(BP_GET_PSIZE(bp) == zio->io_size); | |
1946 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1947 | ||
1948 | if (zio->io_child_error[ZIO_CHILD_DDT]) { | |
1949 | ddt_t *ddt = ddt_select(zio->io_spa, bp); | |
1950 | ddt_entry_t *dde = zio->io_vsd; | |
1951 | if (ddt == NULL) { | |
1952 | ASSERT(spa_load_state(zio->io_spa) != SPA_LOAD_NONE); | |
1953 | return (ZIO_PIPELINE_CONTINUE); | |
1954 | } | |
1955 | if (dde == NULL) { | |
1956 | zio->io_stage = ZIO_STAGE_DDT_READ_START >> 1; | |
1957 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); | |
1958 | return (ZIO_PIPELINE_STOP); | |
1959 | } | |
1960 | if (dde->dde_repair_data != NULL) { | |
1961 | bcopy(dde->dde_repair_data, zio->io_data, zio->io_size); | |
1962 | zio->io_child_error[ZIO_CHILD_DDT] = 0; | |
1963 | } | |
1964 | ddt_repair_done(ddt, dde); | |
1965 | zio->io_vsd = NULL; | |
1966 | } | |
1967 | ||
1968 | ASSERT(zio->io_vsd == NULL); | |
1969 | ||
1970 | return (ZIO_PIPELINE_CONTINUE); | |
1971 | } | |
1972 | ||
1973 | static boolean_t | |
1974 | zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde) | |
1975 | { | |
1976 | spa_t *spa = zio->io_spa; | |
d6320ddb | 1977 | int p; |
428870ff BB |
1978 | |
1979 | /* | |
1980 | * Note: we compare the original data, not the transformed data, | |
1981 | * because when zio->io_bp is an override bp, we will not have | |
1982 | * pushed the I/O transforms. That's an important optimization | |
1983 | * because otherwise we'd compress/encrypt all dmu_sync() data twice. | |
1984 | */ | |
d6320ddb | 1985 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { |
428870ff BB |
1986 | zio_t *lio = dde->dde_lead_zio[p]; |
1987 | ||
1988 | if (lio != NULL) { | |
1989 | return (lio->io_orig_size != zio->io_orig_size || | |
1990 | bcmp(zio->io_orig_data, lio->io_orig_data, | |
1991 | zio->io_orig_size) != 0); | |
1992 | } | |
1993 | } | |
1994 | ||
d6320ddb | 1995 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { |
428870ff BB |
1996 | ddt_phys_t *ddp = &dde->dde_phys[p]; |
1997 | ||
1998 | if (ddp->ddp_phys_birth != 0) { | |
1999 | arc_buf_t *abuf = NULL; | |
2000 | uint32_t aflags = ARC_WAIT; | |
2001 | blkptr_t blk = *zio->io_bp; | |
2002 | int error; | |
2003 | ||
2004 | ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth); | |
2005 | ||
2006 | ddt_exit(ddt); | |
2007 | ||
2008 | error = arc_read_nolock(NULL, spa, &blk, | |
2009 | arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ, | |
2010 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, | |
2011 | &aflags, &zio->io_bookmark); | |
2012 | ||
2013 | if (error == 0) { | |
2014 | if (arc_buf_size(abuf) != zio->io_orig_size || | |
2015 | bcmp(abuf->b_data, zio->io_orig_data, | |
2016 | zio->io_orig_size) != 0) | |
2017 | error = EEXIST; | |
2018 | VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1); | |
2019 | } | |
2020 | ||
2021 | ddt_enter(ddt); | |
2022 | return (error != 0); | |
2023 | } | |
2024 | } | |
2025 | ||
2026 | return (B_FALSE); | |
2027 | } | |
2028 | ||
2029 | static void | |
2030 | zio_ddt_child_write_ready(zio_t *zio) | |
2031 | { | |
2032 | int p = zio->io_prop.zp_copies; | |
2033 | ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); | |
2034 | ddt_entry_t *dde = zio->io_private; | |
2035 | ddt_phys_t *ddp = &dde->dde_phys[p]; | |
2036 | zio_t *pio; | |
2037 | ||
2038 | if (zio->io_error) | |
2039 | return; | |
2040 | ||
2041 | ddt_enter(ddt); | |
2042 | ||
2043 | ASSERT(dde->dde_lead_zio[p] == zio); | |
2044 | ||
2045 | ddt_phys_fill(ddp, zio->io_bp); | |
2046 | ||
2047 | while ((pio = zio_walk_parents(zio)) != NULL) | |
2048 | ddt_bp_fill(ddp, pio->io_bp, zio->io_txg); | |
2049 | ||
2050 | ddt_exit(ddt); | |
2051 | } | |
2052 | ||
2053 | static void | |
2054 | zio_ddt_child_write_done(zio_t *zio) | |
2055 | { | |
2056 | int p = zio->io_prop.zp_copies; | |
2057 | ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); | |
2058 | ddt_entry_t *dde = zio->io_private; | |
2059 | ddt_phys_t *ddp = &dde->dde_phys[p]; | |
2060 | ||
2061 | ddt_enter(ddt); | |
2062 | ||
2063 | ASSERT(ddp->ddp_refcnt == 0); | |
2064 | ASSERT(dde->dde_lead_zio[p] == zio); | |
2065 | dde->dde_lead_zio[p] = NULL; | |
2066 | ||
2067 | if (zio->io_error == 0) { | |
2068 | while (zio_walk_parents(zio) != NULL) | |
2069 | ddt_phys_addref(ddp); | |
2070 | } else { | |
2071 | ddt_phys_clear(ddp); | |
2072 | } | |
2073 | ||
2074 | ddt_exit(ddt); | |
2075 | } | |
2076 | ||
2077 | static void | |
2078 | zio_ddt_ditto_write_done(zio_t *zio) | |
2079 | { | |
2080 | int p = DDT_PHYS_DITTO; | |
428870ff BB |
2081 | blkptr_t *bp = zio->io_bp; |
2082 | ddt_t *ddt = ddt_select(zio->io_spa, bp); | |
2083 | ddt_entry_t *dde = zio->io_private; | |
2084 | ddt_phys_t *ddp = &dde->dde_phys[p]; | |
2085 | ddt_key_t *ddk = &dde->dde_key; | |
1fde1e37 | 2086 | ASSERTV(zio_prop_t *zp = &zio->io_prop); |
428870ff BB |
2087 | |
2088 | ddt_enter(ddt); | |
2089 | ||
2090 | ASSERT(ddp->ddp_refcnt == 0); | |
2091 | ASSERT(dde->dde_lead_zio[p] == zio); | |
2092 | dde->dde_lead_zio[p] = NULL; | |
2093 | ||
2094 | if (zio->io_error == 0) { | |
2095 | ASSERT(ZIO_CHECKSUM_EQUAL(bp->blk_cksum, ddk->ddk_cksum)); | |
2096 | ASSERT(zp->zp_copies < SPA_DVAS_PER_BP); | |
2097 | ASSERT(zp->zp_copies == BP_GET_NDVAS(bp) - BP_IS_GANG(bp)); | |
2098 | if (ddp->ddp_phys_birth != 0) | |
2099 | ddt_phys_free(ddt, ddk, ddp, zio->io_txg); | |
2100 | ddt_phys_fill(ddp, bp); | |
2101 | } | |
2102 | ||
2103 | ddt_exit(ddt); | |
2104 | } | |
2105 | ||
2106 | static int | |
2107 | zio_ddt_write(zio_t *zio) | |
2108 | { | |
2109 | spa_t *spa = zio->io_spa; | |
2110 | blkptr_t *bp = zio->io_bp; | |
2111 | uint64_t txg = zio->io_txg; | |
2112 | zio_prop_t *zp = &zio->io_prop; | |
2113 | int p = zp->zp_copies; | |
2114 | int ditto_copies; | |
2115 | zio_t *cio = NULL; | |
2116 | zio_t *dio = NULL; | |
2117 | ddt_t *ddt = ddt_select(spa, bp); | |
2118 | ddt_entry_t *dde; | |
2119 | ddt_phys_t *ddp; | |
2120 | ||
2121 | ASSERT(BP_GET_DEDUP(bp)); | |
2122 | ASSERT(BP_GET_CHECKSUM(bp) == zp->zp_checksum); | |
2123 | ASSERT(BP_IS_HOLE(bp) || zio->io_bp_override); | |
2124 | ||
2125 | ddt_enter(ddt); | |
2126 | dde = ddt_lookup(ddt, bp, B_TRUE); | |
2127 | ddp = &dde->dde_phys[p]; | |
2128 | ||
2129 | if (zp->zp_dedup_verify && zio_ddt_collision(zio, ddt, dde)) { | |
2130 | /* | |
2131 | * If we're using a weak checksum, upgrade to a strong checksum | |
2132 | * and try again. If we're already using a strong checksum, | |
2133 | * we can't resolve it, so just convert to an ordinary write. | |
2134 | * (And automatically e-mail a paper to Nature?) | |
2135 | */ | |
2136 | if (!zio_checksum_table[zp->zp_checksum].ci_dedup) { | |
2137 | zp->zp_checksum = spa_dedup_checksum(spa); | |
2138 | zio_pop_transforms(zio); | |
2139 | zio->io_stage = ZIO_STAGE_OPEN; | |
2140 | BP_ZERO(bp); | |
2141 | } else { | |
2142 | zp->zp_dedup = 0; | |
2143 | } | |
2144 | zio->io_pipeline = ZIO_WRITE_PIPELINE; | |
2145 | ddt_exit(ddt); | |
2146 | return (ZIO_PIPELINE_CONTINUE); | |
2147 | } | |
2148 | ||
2149 | ditto_copies = ddt_ditto_copies_needed(ddt, dde, ddp); | |
2150 | ASSERT(ditto_copies < SPA_DVAS_PER_BP); | |
2151 | ||
2152 | if (ditto_copies > ddt_ditto_copies_present(dde) && | |
2153 | dde->dde_lead_zio[DDT_PHYS_DITTO] == NULL) { | |
2154 | zio_prop_t czp = *zp; | |
2155 | ||
2156 | czp.zp_copies = ditto_copies; | |
2157 | ||
2158 | /* | |
2159 | * If we arrived here with an override bp, we won't have run | |
2160 | * the transform stack, so we won't have the data we need to | |
2161 | * generate a child i/o. So, toss the override bp and restart. | |
2162 | * This is safe, because using the override bp is just an | |
2163 | * optimization; and it's rare, so the cost doesn't matter. | |
2164 | */ | |
2165 | if (zio->io_bp_override) { | |
2166 | zio_pop_transforms(zio); | |
2167 | zio->io_stage = ZIO_STAGE_OPEN; | |
2168 | zio->io_pipeline = ZIO_WRITE_PIPELINE; | |
2169 | zio->io_bp_override = NULL; | |
2170 | BP_ZERO(bp); | |
2171 | ddt_exit(ddt); | |
2172 | return (ZIO_PIPELINE_CONTINUE); | |
2173 | } | |
2174 | ||
2175 | dio = zio_write(zio, spa, txg, bp, zio->io_orig_data, | |
2176 | zio->io_orig_size, &czp, NULL, | |
2177 | zio_ddt_ditto_write_done, dde, zio->io_priority, | |
2178 | ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); | |
2179 | ||
2180 | zio_push_transform(dio, zio->io_data, zio->io_size, 0, NULL); | |
2181 | dde->dde_lead_zio[DDT_PHYS_DITTO] = dio; | |
2182 | } | |
2183 | ||
2184 | if (ddp->ddp_phys_birth != 0 || dde->dde_lead_zio[p] != NULL) { | |
2185 | if (ddp->ddp_phys_birth != 0) | |
2186 | ddt_bp_fill(ddp, bp, txg); | |
2187 | if (dde->dde_lead_zio[p] != NULL) | |
2188 | zio_add_child(zio, dde->dde_lead_zio[p]); | |
2189 | else | |
2190 | ddt_phys_addref(ddp); | |
2191 | } else if (zio->io_bp_override) { | |
2192 | ASSERT(bp->blk_birth == txg); | |
2193 | ASSERT(BP_EQUAL(bp, zio->io_bp_override)); | |
2194 | ddt_phys_fill(ddp, bp); | |
2195 | ddt_phys_addref(ddp); | |
2196 | } else { | |
2197 | cio = zio_write(zio, spa, txg, bp, zio->io_orig_data, | |
2198 | zio->io_orig_size, zp, zio_ddt_child_write_ready, | |
2199 | zio_ddt_child_write_done, dde, zio->io_priority, | |
2200 | ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); | |
2201 | ||
2202 | zio_push_transform(cio, zio->io_data, zio->io_size, 0, NULL); | |
2203 | dde->dde_lead_zio[p] = cio; | |
2204 | } | |
2205 | ||
2206 | ddt_exit(ddt); | |
2207 | ||
2208 | if (cio) | |
2209 | zio_nowait(cio); | |
2210 | if (dio) | |
2211 | zio_nowait(dio); | |
2212 | ||
2213 | return (ZIO_PIPELINE_CONTINUE); | |
2214 | } | |
2215 | ||
2216 | ddt_entry_t *freedde; /* for debugging */ | |
b128c09f | 2217 | |
428870ff BB |
2218 | static int |
2219 | zio_ddt_free(zio_t *zio) | |
2220 | { | |
2221 | spa_t *spa = zio->io_spa; | |
2222 | blkptr_t *bp = zio->io_bp; | |
2223 | ddt_t *ddt = ddt_select(spa, bp); | |
2224 | ddt_entry_t *dde; | |
2225 | ddt_phys_t *ddp; | |
2226 | ||
2227 | ASSERT(BP_GET_DEDUP(bp)); | |
2228 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
2229 | ||
2230 | ddt_enter(ddt); | |
2231 | freedde = dde = ddt_lookup(ddt, bp, B_TRUE); | |
2232 | ddp = ddt_phys_select(dde, bp); | |
2233 | ddt_phys_decref(ddp); | |
2234 | ddt_exit(ddt); | |
2235 | ||
2236 | return (ZIO_PIPELINE_CONTINUE); | |
2237 | } | |
2238 | ||
2239 | /* | |
2240 | * ========================================================================== | |
2241 | * Allocate and free blocks | |
2242 | * ========================================================================== | |
2243 | */ | |
34dc7c2f BB |
2244 | static int |
2245 | zio_dva_allocate(zio_t *zio) | |
2246 | { | |
2247 | spa_t *spa = zio->io_spa; | |
428870ff | 2248 | metaslab_class_t *mc = spa_normal_class(spa); |
34dc7c2f BB |
2249 | blkptr_t *bp = zio->io_bp; |
2250 | int error; | |
6d974228 | 2251 | int flags = 0; |
34dc7c2f | 2252 | |
9babb374 BB |
2253 | if (zio->io_gang_leader == NULL) { |
2254 | ASSERT(zio->io_child_type > ZIO_CHILD_GANG); | |
2255 | zio->io_gang_leader = zio; | |
2256 | } | |
2257 | ||
34dc7c2f BB |
2258 | ASSERT(BP_IS_HOLE(bp)); |
2259 | ASSERT3U(BP_GET_NDVAS(bp), ==, 0); | |
428870ff BB |
2260 | ASSERT3U(zio->io_prop.zp_copies, >, 0); |
2261 | ASSERT3U(zio->io_prop.zp_copies, <=, spa_max_replication(spa)); | |
34dc7c2f BB |
2262 | ASSERT3U(zio->io_size, ==, BP_GET_PSIZE(bp)); |
2263 | ||
6d974228 GW |
2264 | /* |
2265 | * The dump device does not support gang blocks so allocation on | |
2266 | * behalf of the dump device (i.e. ZIO_FLAG_NODATA) must avoid | |
2267 | * the "fast" gang feature. | |
2268 | */ | |
2269 | flags |= (zio->io_flags & ZIO_FLAG_NODATA) ? METASLAB_GANG_AVOID : 0; | |
2270 | flags |= (zio->io_flags & ZIO_FLAG_GANG_CHILD) ? | |
2271 | METASLAB_GANG_CHILD : 0; | |
b128c09f | 2272 | error = metaslab_alloc(spa, mc, zio->io_size, bp, |
6d974228 | 2273 | zio->io_prop.zp_copies, zio->io_txg, NULL, flags); |
34dc7c2f | 2274 | |
b128c09f | 2275 | if (error) { |
6d974228 GW |
2276 | spa_dbgmsg(spa, "%s: metaslab allocation failure: zio %p, " |
2277 | "size %llu, error %d", spa_name(spa), zio, zio->io_size, | |
2278 | error); | |
b128c09f BB |
2279 | if (error == ENOSPC && zio->io_size > SPA_MINBLOCKSIZE) |
2280 | return (zio_write_gang_block(zio)); | |
34dc7c2f BB |
2281 | zio->io_error = error; |
2282 | } | |
2283 | ||
2284 | return (ZIO_PIPELINE_CONTINUE); | |
2285 | } | |
2286 | ||
2287 | static int | |
2288 | zio_dva_free(zio_t *zio) | |
2289 | { | |
b128c09f | 2290 | metaslab_free(zio->io_spa, zio->io_bp, zio->io_txg, B_FALSE); |
34dc7c2f BB |
2291 | |
2292 | return (ZIO_PIPELINE_CONTINUE); | |
2293 | } | |
2294 | ||
2295 | static int | |
2296 | zio_dva_claim(zio_t *zio) | |
2297 | { | |
b128c09f BB |
2298 | int error; |
2299 | ||
2300 | error = metaslab_claim(zio->io_spa, zio->io_bp, zio->io_txg); | |
2301 | if (error) | |
2302 | zio->io_error = error; | |
34dc7c2f BB |
2303 | |
2304 | return (ZIO_PIPELINE_CONTINUE); | |
2305 | } | |
2306 | ||
b128c09f BB |
2307 | /* |
2308 | * Undo an allocation. This is used by zio_done() when an I/O fails | |
2309 | * and we want to give back the block we just allocated. | |
2310 | * This handles both normal blocks and gang blocks. | |
2311 | */ | |
2312 | static void | |
2313 | zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp) | |
2314 | { | |
d6320ddb BB |
2315 | int g; |
2316 | ||
b128c09f | 2317 | ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); |
428870ff | 2318 | ASSERT(zio->io_bp_override == NULL); |
b128c09f BB |
2319 | |
2320 | if (!BP_IS_HOLE(bp)) | |
428870ff | 2321 | metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE); |
b128c09f BB |
2322 | |
2323 | if (gn != NULL) { | |
d6320ddb | 2324 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) { |
b128c09f BB |
2325 | zio_dva_unallocate(zio, gn->gn_child[g], |
2326 | &gn->gn_gbh->zg_blkptr[g]); | |
2327 | } | |
2328 | } | |
2329 | } | |
2330 | ||
2331 | /* | |
2332 | * Try to allocate an intent log block. Return 0 on success, errno on failure. | |
2333 | */ | |
2334 | int | |
428870ff BB |
2335 | zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, blkptr_t *old_bp, |
2336 | uint64_t size, boolean_t use_slog) | |
b128c09f | 2337 | { |
428870ff | 2338 | int error = 1; |
b128c09f | 2339 | |
428870ff BB |
2340 | ASSERT(txg > spa_syncing_txg(spa)); |
2341 | ||
ebf8e3a2 BB |
2342 | /* |
2343 | * ZIL blocks are always contiguous (i.e. not gang blocks) so we | |
2344 | * set the METASLAB_GANG_AVOID flag so that they don't "fast gang" | |
2345 | * when allocating them. | |
2346 | */ | |
2347 | if (use_slog) { | |
428870ff | 2348 | error = metaslab_alloc(spa, spa_log_class(spa), size, |
ebf8e3a2 BB |
2349 | new_bp, 1, txg, old_bp, |
2350 | METASLAB_HINTBP_AVOID | METASLAB_GANG_AVOID); | |
2351 | } | |
b128c09f | 2352 | |
ebf8e3a2 | 2353 | if (error) { |
428870ff | 2354 | error = metaslab_alloc(spa, spa_normal_class(spa), size, |
ebf8e3a2 BB |
2355 | new_bp, 1, txg, old_bp, |
2356 | METASLAB_HINTBP_AVOID | METASLAB_GANG_AVOID); | |
2357 | } | |
b128c09f BB |
2358 | |
2359 | if (error == 0) { | |
2360 | BP_SET_LSIZE(new_bp, size); | |
2361 | BP_SET_PSIZE(new_bp, size); | |
2362 | BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF); | |
428870ff BB |
2363 | BP_SET_CHECKSUM(new_bp, |
2364 | spa_version(spa) >= SPA_VERSION_SLIM_ZIL | |
2365 | ? ZIO_CHECKSUM_ZILOG2 : ZIO_CHECKSUM_ZILOG); | |
b128c09f BB |
2366 | BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG); |
2367 | BP_SET_LEVEL(new_bp, 0); | |
428870ff | 2368 | BP_SET_DEDUP(new_bp, 0); |
b128c09f BB |
2369 | BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER); |
2370 | } | |
2371 | ||
2372 | return (error); | |
2373 | } | |
2374 | ||
2375 | /* | |
428870ff | 2376 | * Free an intent log block. |
b128c09f BB |
2377 | */ |
2378 | void | |
428870ff | 2379 | zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp) |
b128c09f | 2380 | { |
428870ff | 2381 | ASSERT(BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG); |
b128c09f BB |
2382 | ASSERT(!BP_IS_GANG(bp)); |
2383 | ||
428870ff | 2384 | zio_free(spa, txg, bp); |
b128c09f BB |
2385 | } |
2386 | ||
34dc7c2f BB |
2387 | /* |
2388 | * ========================================================================== | |
2389 | * Read and write to physical devices | |
2390 | * ========================================================================== | |
2391 | */ | |
34dc7c2f BB |
2392 | static int |
2393 | zio_vdev_io_start(zio_t *zio) | |
2394 | { | |
2395 | vdev_t *vd = zio->io_vd; | |
34dc7c2f BB |
2396 | uint64_t align; |
2397 | spa_t *spa = zio->io_spa; | |
2398 | ||
b128c09f BB |
2399 | ASSERT(zio->io_error == 0); |
2400 | ASSERT(zio->io_child_error[ZIO_CHILD_VDEV] == 0); | |
34dc7c2f | 2401 | |
b128c09f BB |
2402 | if (vd == NULL) { |
2403 | if (!(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) | |
2404 | spa_config_enter(spa, SCL_ZIO, zio, RW_READER); | |
34dc7c2f | 2405 | |
b128c09f BB |
2406 | /* |
2407 | * The mirror_ops handle multiple DVAs in a single BP. | |
2408 | */ | |
2409 | return (vdev_mirror_ops.vdev_op_io_start(zio)); | |
34dc7c2f BB |
2410 | } |
2411 | ||
572e2857 BB |
2412 | /* |
2413 | * We keep track of time-sensitive I/Os so that the scan thread | |
2414 | * can quickly react to certain workloads. In particular, we care | |
2415 | * about non-scrubbing, top-level reads and writes with the following | |
2416 | * characteristics: | |
2417 | * - synchronous writes of user data to non-slog devices | |
2418 | * - any reads of user data | |
2419 | * When these conditions are met, adjust the timestamp of spa_last_io | |
2420 | * which allows the scan thread to adjust its workload accordingly. | |
2421 | */ | |
2422 | if (!(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_bp != NULL && | |
2423 | vd == vd->vdev_top && !vd->vdev_islog && | |
2424 | zio->io_bookmark.zb_objset != DMU_META_OBJSET && | |
2425 | zio->io_txg != spa_syncing_txg(spa)) { | |
2426 | uint64_t old = spa->spa_last_io; | |
2427 | uint64_t new = ddi_get_lbolt64(); | |
2428 | if (old != new) | |
2429 | (void) atomic_cas_64(&spa->spa_last_io, old, new); | |
2430 | } | |
2431 | ||
b128c09f BB |
2432 | align = 1ULL << vd->vdev_top->vdev_ashift; |
2433 | ||
34dc7c2f BB |
2434 | if (P2PHASE(zio->io_size, align) != 0) { |
2435 | uint64_t asize = P2ROUNDUP(zio->io_size, align); | |
2436 | char *abuf = zio_buf_alloc(asize); | |
b128c09f | 2437 | ASSERT(vd == vd->vdev_top); |
34dc7c2f BB |
2438 | if (zio->io_type == ZIO_TYPE_WRITE) { |
2439 | bcopy(zio->io_data, abuf, zio->io_size); | |
2440 | bzero(abuf + zio->io_size, asize - zio->io_size); | |
2441 | } | |
b128c09f | 2442 | zio_push_transform(zio, abuf, asize, asize, zio_subblock); |
34dc7c2f BB |
2443 | } |
2444 | ||
2445 | ASSERT(P2PHASE(zio->io_offset, align) == 0); | |
2446 | ASSERT(P2PHASE(zio->io_size, align) == 0); | |
572e2857 | 2447 | VERIFY(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa)); |
fb5f0bc8 BB |
2448 | |
2449 | /* | |
2450 | * If this is a repair I/O, and there's no self-healing involved -- | |
2451 | * that is, we're just resilvering what we expect to resilver -- | |
2452 | * then don't do the I/O unless zio's txg is actually in vd's DTL. | |
2453 | * This prevents spurious resilvering with nested replication. | |
2454 | * For example, given a mirror of mirrors, (A+B)+(C+D), if only | |
2455 | * A is out of date, we'll read from C+D, then use the data to | |
2456 | * resilver A+B -- but we don't actually want to resilver B, just A. | |
2457 | * The top-level mirror has no way to know this, so instead we just | |
2458 | * discard unnecessary repairs as we work our way down the vdev tree. | |
2459 | * The same logic applies to any form of nested replication: | |
2460 | * ditto + mirror, RAID-Z + replacing, etc. This covers them all. | |
2461 | */ | |
2462 | if ((zio->io_flags & ZIO_FLAG_IO_REPAIR) && | |
2463 | !(zio->io_flags & ZIO_FLAG_SELF_HEAL) && | |
2464 | zio->io_txg != 0 && /* not a delegated i/o */ | |
2465 | !vdev_dtl_contains(vd, DTL_PARTIAL, zio->io_txg, 1)) { | |
2466 | ASSERT(zio->io_type == ZIO_TYPE_WRITE); | |
fb5f0bc8 BB |
2467 | zio_vdev_io_bypass(zio); |
2468 | return (ZIO_PIPELINE_CONTINUE); | |
2469 | } | |
34dc7c2f | 2470 | |
b128c09f BB |
2471 | if (vd->vdev_ops->vdev_op_leaf && |
2472 | (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE)) { | |
2473 | ||
2474 | if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0) | |
d164b209 | 2475 | return (ZIO_PIPELINE_CONTINUE); |
b128c09f BB |
2476 | |
2477 | if ((zio = vdev_queue_io(zio)) == NULL) | |
2478 | return (ZIO_PIPELINE_STOP); | |
2479 | ||
2480 | if (!vdev_accessible(vd, zio)) { | |
2481 | zio->io_error = ENXIO; | |
2482 | zio_interrupt(zio); | |
2483 | return (ZIO_PIPELINE_STOP); | |
2484 | } | |
b128c09f BB |
2485 | } |
2486 | ||
34dc7c2f BB |
2487 | return (vd->vdev_ops->vdev_op_io_start(zio)); |
2488 | } | |
2489 | ||
2490 | static int | |
2491 | zio_vdev_io_done(zio_t *zio) | |
2492 | { | |
b128c09f BB |
2493 | vdev_t *vd = zio->io_vd; |
2494 | vdev_ops_t *ops = vd ? vd->vdev_ops : &vdev_mirror_ops; | |
2495 | boolean_t unexpected_error = B_FALSE; | |
34dc7c2f | 2496 | |
b128c09f BB |
2497 | if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) |
2498 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 2499 | |
b128c09f BB |
2500 | ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE); |
2501 | ||
2502 | if (vd != NULL && vd->vdev_ops->vdev_op_leaf) { | |
2503 | ||
2504 | vdev_queue_io_done(zio); | |
2505 | ||
2506 | if (zio->io_type == ZIO_TYPE_WRITE) | |
2507 | vdev_cache_write(zio); | |
2508 | ||
2509 | if (zio_injection_enabled && zio->io_error == 0) | |
9babb374 BB |
2510 | zio->io_error = zio_handle_device_injection(vd, |
2511 | zio, EIO); | |
b128c09f BB |
2512 | |
2513 | if (zio_injection_enabled && zio->io_error == 0) | |
2514 | zio->io_error = zio_handle_label_injection(zio, EIO); | |
2515 | ||
2516 | if (zio->io_error) { | |
2517 | if (!vdev_accessible(vd, zio)) { | |
2518 | zio->io_error = ENXIO; | |
2519 | } else { | |
2520 | unexpected_error = B_TRUE; | |
2521 | } | |
2522 | } | |
2523 | } | |
2524 | ||
2525 | ops->vdev_op_io_done(zio); | |
34dc7c2f | 2526 | |
b128c09f | 2527 | if (unexpected_error) |
d164b209 | 2528 | VERIFY(vdev_probe(vd, zio) == NULL); |
34dc7c2f | 2529 | |
b128c09f | 2530 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
2531 | } |
2532 | ||
428870ff BB |
2533 | /* |
2534 | * For non-raidz ZIOs, we can just copy aside the bad data read from the | |
2535 | * disk, and use that to finish the checksum ereport later. | |
2536 | */ | |
2537 | static void | |
2538 | zio_vsd_default_cksum_finish(zio_cksum_report_t *zcr, | |
2539 | const void *good_buf) | |
2540 | { | |
2541 | /* no processing needed */ | |
2542 | zfs_ereport_finish_checksum(zcr, good_buf, zcr->zcr_cbdata, B_FALSE); | |
2543 | } | |
2544 | ||
2545 | /*ARGSUSED*/ | |
2546 | void | |
2547 | zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *ignored) | |
2548 | { | |
2549 | void *buf = zio_buf_alloc(zio->io_size); | |
2550 | ||
2551 | bcopy(zio->io_data, buf, zio->io_size); | |
2552 | ||
2553 | zcr->zcr_cbinfo = zio->io_size; | |
2554 | zcr->zcr_cbdata = buf; | |
2555 | zcr->zcr_finish = zio_vsd_default_cksum_finish; | |
2556 | zcr->zcr_free = zio_buf_free; | |
2557 | } | |
2558 | ||
34dc7c2f BB |
2559 | static int |
2560 | zio_vdev_io_assess(zio_t *zio) | |
2561 | { | |
2562 | vdev_t *vd = zio->io_vd; | |
b128c09f BB |
2563 | |
2564 | if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) | |
2565 | return (ZIO_PIPELINE_STOP); | |
2566 | ||
2567 | if (vd == NULL && !(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) | |
2568 | spa_config_exit(zio->io_spa, SCL_ZIO, zio); | |
2569 | ||
2570 | if (zio->io_vsd != NULL) { | |
428870ff | 2571 | zio->io_vsd_ops->vsd_free(zio); |
b128c09f | 2572 | zio->io_vsd = NULL; |
34dc7c2f BB |
2573 | } |
2574 | ||
b128c09f | 2575 | if (zio_injection_enabled && zio->io_error == 0) |
34dc7c2f BB |
2576 | zio->io_error = zio_handle_fault_injection(zio, EIO); |
2577 | ||
2578 | /* | |
2579 | * If the I/O failed, determine whether we should attempt to retry it. | |
428870ff BB |
2580 | * |
2581 | * On retry, we cut in line in the issue queue, since we don't want | |
2582 | * compression/checksumming/etc. work to prevent our (cheap) IO reissue. | |
34dc7c2f | 2583 | */ |
b128c09f BB |
2584 | if (zio->io_error && vd == NULL && |
2585 | !(zio->io_flags & (ZIO_FLAG_DONT_RETRY | ZIO_FLAG_IO_RETRY))) { | |
2586 | ASSERT(!(zio->io_flags & ZIO_FLAG_DONT_QUEUE)); /* not a leaf */ | |
2587 | ASSERT(!(zio->io_flags & ZIO_FLAG_IO_BYPASS)); /* not a leaf */ | |
34dc7c2f | 2588 | zio->io_error = 0; |
b128c09f BB |
2589 | zio->io_flags |= ZIO_FLAG_IO_RETRY | |
2590 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE; | |
428870ff BB |
2591 | zio->io_stage = ZIO_STAGE_VDEV_IO_START >> 1; |
2592 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, | |
2593 | zio_requeue_io_start_cut_in_line); | |
b128c09f | 2594 | return (ZIO_PIPELINE_STOP); |
34dc7c2f BB |
2595 | } |
2596 | ||
b128c09f BB |
2597 | /* |
2598 | * If we got an error on a leaf device, convert it to ENXIO | |
2599 | * if the device is not accessible at all. | |
2600 | */ | |
2601 | if (zio->io_error && vd != NULL && vd->vdev_ops->vdev_op_leaf && | |
2602 | !vdev_accessible(vd, zio)) | |
2603 | zio->io_error = ENXIO; | |
2604 | ||
2605 | /* | |
2606 | * If we can't write to an interior vdev (mirror or RAID-Z), | |
2607 | * set vdev_cant_write so that we stop trying to allocate from it. | |
2608 | */ | |
2609 | if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE && | |
2610 | vd != NULL && !vd->vdev_ops->vdev_op_leaf) | |
2611 | vd->vdev_cant_write = B_TRUE; | |
2612 | ||
2613 | if (zio->io_error) | |
2614 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
2615 | ||
34dc7c2f BB |
2616 | return (ZIO_PIPELINE_CONTINUE); |
2617 | } | |
2618 | ||
2619 | void | |
2620 | zio_vdev_io_reissue(zio_t *zio) | |
2621 | { | |
2622 | ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); | |
2623 | ASSERT(zio->io_error == 0); | |
2624 | ||
428870ff | 2625 | zio->io_stage >>= 1; |
34dc7c2f BB |
2626 | } |
2627 | ||
2628 | void | |
2629 | zio_vdev_io_redone(zio_t *zio) | |
2630 | { | |
2631 | ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_DONE); | |
2632 | ||
428870ff | 2633 | zio->io_stage >>= 1; |
34dc7c2f BB |
2634 | } |
2635 | ||
2636 | void | |
2637 | zio_vdev_io_bypass(zio_t *zio) | |
2638 | { | |
2639 | ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); | |
2640 | ASSERT(zio->io_error == 0); | |
2641 | ||
2642 | zio->io_flags |= ZIO_FLAG_IO_BYPASS; | |
428870ff | 2643 | zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS >> 1; |
34dc7c2f BB |
2644 | } |
2645 | ||
2646 | /* | |
2647 | * ========================================================================== | |
2648 | * Generate and verify checksums | |
2649 | * ========================================================================== | |
2650 | */ | |
2651 | static int | |
2652 | zio_checksum_generate(zio_t *zio) | |
2653 | { | |
34dc7c2f | 2654 | blkptr_t *bp = zio->io_bp; |
b128c09f | 2655 | enum zio_checksum checksum; |
34dc7c2f | 2656 | |
b128c09f BB |
2657 | if (bp == NULL) { |
2658 | /* | |
2659 | * This is zio_write_phys(). | |
2660 | * We're either generating a label checksum, or none at all. | |
2661 | */ | |
2662 | checksum = zio->io_prop.zp_checksum; | |
34dc7c2f | 2663 | |
b128c09f BB |
2664 | if (checksum == ZIO_CHECKSUM_OFF) |
2665 | return (ZIO_PIPELINE_CONTINUE); | |
2666 | ||
2667 | ASSERT(checksum == ZIO_CHECKSUM_LABEL); | |
2668 | } else { | |
2669 | if (BP_IS_GANG(bp) && zio->io_child_type == ZIO_CHILD_GANG) { | |
2670 | ASSERT(!IO_IS_ALLOCATING(zio)); | |
2671 | checksum = ZIO_CHECKSUM_GANG_HEADER; | |
2672 | } else { | |
2673 | checksum = BP_GET_CHECKSUM(bp); | |
2674 | } | |
2675 | } | |
34dc7c2f | 2676 | |
b128c09f | 2677 | zio_checksum_compute(zio, checksum, zio->io_data, zio->io_size); |
34dc7c2f BB |
2678 | |
2679 | return (ZIO_PIPELINE_CONTINUE); | |
2680 | } | |
2681 | ||
2682 | static int | |
b128c09f | 2683 | zio_checksum_verify(zio_t *zio) |
34dc7c2f | 2684 | { |
428870ff | 2685 | zio_bad_cksum_t info; |
b128c09f BB |
2686 | blkptr_t *bp = zio->io_bp; |
2687 | int error; | |
34dc7c2f | 2688 | |
428870ff BB |
2689 | ASSERT(zio->io_vd != NULL); |
2690 | ||
b128c09f BB |
2691 | if (bp == NULL) { |
2692 | /* | |
2693 | * This is zio_read_phys(). | |
2694 | * We're either verifying a label checksum, or nothing at all. | |
2695 | */ | |
2696 | if (zio->io_prop.zp_checksum == ZIO_CHECKSUM_OFF) | |
2697 | return (ZIO_PIPELINE_CONTINUE); | |
34dc7c2f | 2698 | |
b128c09f BB |
2699 | ASSERT(zio->io_prop.zp_checksum == ZIO_CHECKSUM_LABEL); |
2700 | } | |
34dc7c2f | 2701 | |
428870ff | 2702 | if ((error = zio_checksum_error(zio, &info)) != 0) { |
b128c09f BB |
2703 | zio->io_error = error; |
2704 | if (!(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { | |
428870ff BB |
2705 | zfs_ereport_start_checksum(zio->io_spa, |
2706 | zio->io_vd, zio, zio->io_offset, | |
2707 | zio->io_size, NULL, &info); | |
b128c09f | 2708 | } |
34dc7c2f BB |
2709 | } |
2710 | ||
2711 | return (ZIO_PIPELINE_CONTINUE); | |
2712 | } | |
2713 | ||
2714 | /* | |
2715 | * Called by RAID-Z to ensure we don't compute the checksum twice. | |
2716 | */ | |
2717 | void | |
2718 | zio_checksum_verified(zio_t *zio) | |
2719 | { | |
428870ff | 2720 | zio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; |
34dc7c2f BB |
2721 | } |
2722 | ||
2723 | /* | |
b128c09f BB |
2724 | * ========================================================================== |
2725 | * Error rank. Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other. | |
2726 | * An error of 0 indictes success. ENXIO indicates whole-device failure, | |
2727 | * which may be transient (e.g. unplugged) or permament. ECKSUM and EIO | |
2728 | * indicate errors that are specific to one I/O, and most likely permanent. | |
2729 | * Any other error is presumed to be worse because we weren't expecting it. | |
2730 | * ========================================================================== | |
34dc7c2f | 2731 | */ |
b128c09f BB |
2732 | int |
2733 | zio_worst_error(int e1, int e2) | |
34dc7c2f | 2734 | { |
b128c09f BB |
2735 | static int zio_error_rank[] = { 0, ENXIO, ECKSUM, EIO }; |
2736 | int r1, r2; | |
2737 | ||
2738 | for (r1 = 0; r1 < sizeof (zio_error_rank) / sizeof (int); r1++) | |
2739 | if (e1 == zio_error_rank[r1]) | |
2740 | break; | |
34dc7c2f | 2741 | |
b128c09f BB |
2742 | for (r2 = 0; r2 < sizeof (zio_error_rank) / sizeof (int); r2++) |
2743 | if (e2 == zio_error_rank[r2]) | |
2744 | break; | |
2745 | ||
2746 | return (r1 > r2 ? e1 : e2); | |
34dc7c2f BB |
2747 | } |
2748 | ||
2749 | /* | |
2750 | * ========================================================================== | |
b128c09f | 2751 | * I/O completion |
34dc7c2f BB |
2752 | * ========================================================================== |
2753 | */ | |
b128c09f BB |
2754 | static int |
2755 | zio_ready(zio_t *zio) | |
34dc7c2f | 2756 | { |
b128c09f | 2757 | blkptr_t *bp = zio->io_bp; |
d164b209 | 2758 | zio_t *pio, *pio_next; |
34dc7c2f | 2759 | |
428870ff BB |
2760 | if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || |
2761 | zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_READY)) | |
9babb374 | 2762 | return (ZIO_PIPELINE_STOP); |
34dc7c2f | 2763 | |
9babb374 | 2764 | if (zio->io_ready) { |
b128c09f BB |
2765 | ASSERT(IO_IS_ALLOCATING(zio)); |
2766 | ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); | |
2767 | ASSERT(zio->io_children[ZIO_CHILD_GANG][ZIO_WAIT_READY] == 0); | |
34dc7c2f | 2768 | |
b128c09f BB |
2769 | zio->io_ready(zio); |
2770 | } | |
34dc7c2f | 2771 | |
b128c09f BB |
2772 | if (bp != NULL && bp != &zio->io_bp_copy) |
2773 | zio->io_bp_copy = *bp; | |
34dc7c2f | 2774 | |
b128c09f BB |
2775 | if (zio->io_error) |
2776 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
34dc7c2f | 2777 | |
d164b209 BB |
2778 | mutex_enter(&zio->io_lock); |
2779 | zio->io_state[ZIO_WAIT_READY] = 1; | |
2780 | pio = zio_walk_parents(zio); | |
2781 | mutex_exit(&zio->io_lock); | |
2782 | ||
2783 | /* | |
2784 | * As we notify zio's parents, new parents could be added. | |
2785 | * New parents go to the head of zio's io_parent_list, however, | |
2786 | * so we will (correctly) not notify them. The remainder of zio's | |
2787 | * io_parent_list, from 'pio_next' onward, cannot change because | |
2788 | * all parents must wait for us to be done before they can be done. | |
2789 | */ | |
2790 | for (; pio != NULL; pio = pio_next) { | |
2791 | pio_next = zio_walk_parents(zio); | |
b128c09f | 2792 | zio_notify_parent(pio, zio, ZIO_WAIT_READY); |
d164b209 | 2793 | } |
34dc7c2f | 2794 | |
428870ff BB |
2795 | if (zio->io_flags & ZIO_FLAG_NODATA) { |
2796 | if (BP_IS_GANG(bp)) { | |
2797 | zio->io_flags &= ~ZIO_FLAG_NODATA; | |
2798 | } else { | |
2799 | ASSERT((uintptr_t)zio->io_data < SPA_MAXBLOCKSIZE); | |
2800 | zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; | |
2801 | } | |
2802 | } | |
2803 | ||
2804 | if (zio_injection_enabled && | |
2805 | zio->io_spa->spa_syncing_txg == zio->io_txg) | |
2806 | zio_handle_ignored_writes(zio); | |
2807 | ||
b128c09f | 2808 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
2809 | } |
2810 | ||
b128c09f BB |
2811 | static int |
2812 | zio_done(zio_t *zio) | |
34dc7c2f | 2813 | { |
d164b209 | 2814 | zio_t *pio, *pio_next; |
d6320ddb | 2815 | int c, w; |
34dc7c2f | 2816 | |
b128c09f | 2817 | /* |
9babb374 | 2818 | * If our children haven't all completed, |
b128c09f BB |
2819 | * wait for them and then repeat this pipeline stage. |
2820 | */ | |
2821 | if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE) || | |
2822 | zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE) || | |
428870ff | 2823 | zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE) || |
b128c09f BB |
2824 | zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_DONE)) |
2825 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 2826 | |
d6320ddb BB |
2827 | for (c = 0; c < ZIO_CHILD_TYPES; c++) |
2828 | for (w = 0; w < ZIO_WAIT_TYPES; w++) | |
b128c09f BB |
2829 | ASSERT(zio->io_children[c][w] == 0); |
2830 | ||
c776b317 BB |
2831 | if (zio->io_bp != NULL) { |
2832 | ASSERT(zio->io_bp->blk_pad[0] == 0); | |
2833 | ASSERT(zio->io_bp->blk_pad[1] == 0); | |
2834 | ASSERT(bcmp(zio->io_bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 || | |
2835 | (zio->io_bp == zio_unique_parent(zio)->io_bp)); | |
2836 | if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(zio->io_bp) && | |
428870ff | 2837 | zio->io_bp_override == NULL && |
b128c09f | 2838 | !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) { |
c776b317 BB |
2839 | ASSERT(!BP_SHOULD_BYTESWAP(zio->io_bp)); |
2840 | ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp)); | |
2841 | ASSERT(BP_COUNT_GANG(zio->io_bp) == 0 || | |
2842 | (BP_COUNT_GANG(zio->io_bp) == BP_GET_NDVAS(zio->io_bp))); | |
b128c09f BB |
2843 | } |
2844 | } | |
2845 | ||
2846 | /* | |
428870ff | 2847 | * If there were child vdev/gang/ddt errors, they apply to us now. |
b128c09f BB |
2848 | */ |
2849 | zio_inherit_child_errors(zio, ZIO_CHILD_VDEV); | |
2850 | zio_inherit_child_errors(zio, ZIO_CHILD_GANG); | |
428870ff BB |
2851 | zio_inherit_child_errors(zio, ZIO_CHILD_DDT); |
2852 | ||
2853 | /* | |
2854 | * If the I/O on the transformed data was successful, generate any | |
2855 | * checksum reports now while we still have the transformed data. | |
2856 | */ | |
2857 | if (zio->io_error == 0) { | |
2858 | while (zio->io_cksum_report != NULL) { | |
2859 | zio_cksum_report_t *zcr = zio->io_cksum_report; | |
2860 | uint64_t align = zcr->zcr_align; | |
c776b317 | 2861 | uint64_t asize = P2ROUNDUP(zio->io_size, align); |
428870ff BB |
2862 | char *abuf = zio->io_data; |
2863 | ||
c776b317 | 2864 | if (asize != zio->io_size) { |
428870ff | 2865 | abuf = zio_buf_alloc(asize); |
c776b317 BB |
2866 | bcopy(zio->io_data, abuf, zio->io_size); |
2867 | bzero(abuf + zio->io_size, asize - zio->io_size); | |
428870ff BB |
2868 | } |
2869 | ||
2870 | zio->io_cksum_report = zcr->zcr_next; | |
2871 | zcr->zcr_next = NULL; | |
2872 | zcr->zcr_finish(zcr, abuf); | |
2873 | zfs_ereport_free_checksum(zcr); | |
2874 | ||
c776b317 | 2875 | if (asize != zio->io_size) |
428870ff BB |
2876 | zio_buf_free(abuf, asize); |
2877 | } | |
2878 | } | |
b128c09f BB |
2879 | |
2880 | zio_pop_transforms(zio); /* note: may set zio->io_error */ | |
2881 | ||
c776b317 | 2882 | vdev_stat_update(zio, zio->io_size); |
b128c09f | 2883 | |
a69052be BB |
2884 | /* |
2885 | * If this I/O is attached to a particular vdev is slow, exeeding | |
2886 | * 30 seconds to complete, post an error described the I/O delay. | |
2887 | * We ignore these errors if the device is currently unavailable. | |
2888 | */ | |
2889 | if (zio->io_delay >= zio_delay_max) { | |
2890 | if (zio->io_vd != NULL && !vdev_is_dead(zio->io_vd)) | |
2891 | zfs_ereport_post(FM_EREPORT_ZFS_DELAY, zio->io_spa, | |
2892 | zio->io_vd, zio, 0, 0); | |
2893 | } | |
2894 | ||
b128c09f BB |
2895 | if (zio->io_error) { |
2896 | /* | |
2897 | * If this I/O is attached to a particular vdev, | |
2898 | * generate an error message describing the I/O failure | |
2899 | * at the block level. We ignore these errors if the | |
2900 | * device is currently unavailable. | |
2901 | */ | |
c776b317 BB |
2902 | if (zio->io_error != ECKSUM && zio->io_vd != NULL && |
2903 | !vdev_is_dead(zio->io_vd)) | |
2904 | zfs_ereport_post(FM_EREPORT_ZFS_IO, zio->io_spa, | |
2905 | zio->io_vd, zio, 0, 0); | |
34dc7c2f | 2906 | |
428870ff BB |
2907 | if ((zio->io_error == EIO || !(zio->io_flags & |
2908 | (ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) && | |
c776b317 | 2909 | zio == zio->io_logical) { |
b128c09f BB |
2910 | /* |
2911 | * For logical I/O requests, tell the SPA to log the | |
2912 | * error and generate a logical data ereport. | |
2913 | */ | |
c776b317 BB |
2914 | spa_log_error(zio->io_spa, zio); |
2915 | zfs_ereport_post(FM_EREPORT_ZFS_DATA, zio->io_spa, NULL, zio, | |
b128c09f BB |
2916 | 0, 0); |
2917 | } | |
2918 | } | |
34dc7c2f | 2919 | |
c776b317 | 2920 | if (zio->io_error && zio == zio->io_logical) { |
b128c09f BB |
2921 | /* |
2922 | * Determine whether zio should be reexecuted. This will | |
2923 | * propagate all the way to the root via zio_notify_parent(). | |
2924 | */ | |
c776b317 | 2925 | ASSERT(zio->io_vd == NULL && zio->io_bp != NULL); |
428870ff | 2926 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); |
b128c09f | 2927 | |
428870ff BB |
2928 | if (IO_IS_ALLOCATING(zio) && |
2929 | !(zio->io_flags & ZIO_FLAG_CANFAIL)) { | |
b128c09f BB |
2930 | if (zio->io_error != ENOSPC) |
2931 | zio->io_reexecute |= ZIO_REEXECUTE_NOW; | |
2932 | else | |
2933 | zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; | |
428870ff | 2934 | } |
b128c09f BB |
2935 | |
2936 | if ((zio->io_type == ZIO_TYPE_READ || | |
2937 | zio->io_type == ZIO_TYPE_FREE) && | |
572e2857 | 2938 | !(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && |
b128c09f | 2939 | zio->io_error == ENXIO && |
c776b317 BB |
2940 | spa_load_state(zio->io_spa) == SPA_LOAD_NONE && |
2941 | spa_get_failmode(zio->io_spa) != ZIO_FAILURE_MODE_CONTINUE) | |
b128c09f BB |
2942 | zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; |
2943 | ||
2944 | if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute) | |
2945 | zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; | |
428870ff BB |
2946 | |
2947 | /* | |
2948 | * Here is a possibly good place to attempt to do | |
2949 | * either combinatorial reconstruction or error correction | |
2950 | * based on checksums. It also might be a good place | |
2951 | * to send out preliminary ereports before we suspend | |
2952 | * processing. | |
2953 | */ | |
34dc7c2f BB |
2954 | } |
2955 | ||
2956 | /* | |
b128c09f BB |
2957 | * If there were logical child errors, they apply to us now. |
2958 | * We defer this until now to avoid conflating logical child | |
2959 | * errors with errors that happened to the zio itself when | |
2960 | * updating vdev stats and reporting FMA events above. | |
34dc7c2f | 2961 | */ |
b128c09f | 2962 | zio_inherit_child_errors(zio, ZIO_CHILD_LOGICAL); |
34dc7c2f | 2963 | |
428870ff BB |
2964 | if ((zio->io_error || zio->io_reexecute) && |
2965 | IO_IS_ALLOCATING(zio) && zio->io_gang_leader == zio && | |
2966 | !(zio->io_flags & ZIO_FLAG_IO_REWRITE)) | |
c776b317 | 2967 | zio_dva_unallocate(zio, zio->io_gang_tree, zio->io_bp); |
9babb374 BB |
2968 | |
2969 | zio_gang_tree_free(&zio->io_gang_tree); | |
2970 | ||
2971 | /* | |
2972 | * Godfather I/Os should never suspend. | |
2973 | */ | |
2974 | if ((zio->io_flags & ZIO_FLAG_GODFATHER) && | |
2975 | (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) | |
2976 | zio->io_reexecute = 0; | |
2977 | ||
b128c09f BB |
2978 | if (zio->io_reexecute) { |
2979 | /* | |
2980 | * This is a logical I/O that wants to reexecute. | |
2981 | * | |
2982 | * Reexecute is top-down. When an i/o fails, if it's not | |
2983 | * the root, it simply notifies its parent and sticks around. | |
2984 | * The parent, seeing that it still has children in zio_done(), | |
2985 | * does the same. This percolates all the way up to the root. | |
2986 | * The root i/o will reexecute or suspend the entire tree. | |
2987 | * | |
2988 | * This approach ensures that zio_reexecute() honors | |
2989 | * all the original i/o dependency relationships, e.g. | |
2990 | * parents not executing until children are ready. | |
2991 | */ | |
2992 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
34dc7c2f | 2993 | |
9babb374 | 2994 | zio->io_gang_leader = NULL; |
b128c09f | 2995 | |
d164b209 BB |
2996 | mutex_enter(&zio->io_lock); |
2997 | zio->io_state[ZIO_WAIT_DONE] = 1; | |
2998 | mutex_exit(&zio->io_lock); | |
2999 | ||
9babb374 BB |
3000 | /* |
3001 | * "The Godfather" I/O monitors its children but is | |
3002 | * not a true parent to them. It will track them through | |
3003 | * the pipeline but severs its ties whenever they get into | |
3004 | * trouble (e.g. suspended). This allows "The Godfather" | |
3005 | * I/O to return status without blocking. | |
3006 | */ | |
3007 | for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { | |
3008 | zio_link_t *zl = zio->io_walk_link; | |
3009 | pio_next = zio_walk_parents(zio); | |
3010 | ||
3011 | if ((pio->io_flags & ZIO_FLAG_GODFATHER) && | |
3012 | (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) { | |
3013 | zio_remove_child(pio, zio, zl); | |
3014 | zio_notify_parent(pio, zio, ZIO_WAIT_DONE); | |
3015 | } | |
3016 | } | |
3017 | ||
d164b209 | 3018 | if ((pio = zio_unique_parent(zio)) != NULL) { |
b128c09f BB |
3019 | /* |
3020 | * We're not a root i/o, so there's nothing to do | |
3021 | * but notify our parent. Don't propagate errors | |
3022 | * upward since we haven't permanently failed yet. | |
3023 | */ | |
9babb374 | 3024 | ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); |
b128c09f BB |
3025 | zio->io_flags |= ZIO_FLAG_DONT_PROPAGATE; |
3026 | zio_notify_parent(pio, zio, ZIO_WAIT_DONE); | |
3027 | } else if (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND) { | |
3028 | /* | |
3029 | * We'd fail again if we reexecuted now, so suspend | |
3030 | * until conditions improve (e.g. device comes online). | |
3031 | */ | |
c776b317 | 3032 | zio_suspend(zio->io_spa, zio); |
b128c09f BB |
3033 | } else { |
3034 | /* | |
3035 | * Reexecution is potentially a huge amount of work. | |
3036 | * Hand it off to the otherwise-unused claim taskq. | |
3037 | */ | |
a38718a6 GA |
3038 | ASSERT(taskq_empty_ent(&zio->io_tqent)); |
3039 | (void) taskq_dispatch_ent( | |
c776b317 | 3040 | zio->io_spa->spa_zio_taskq[ZIO_TYPE_CLAIM][ZIO_TASKQ_ISSUE], |
a38718a6 GA |
3041 | (task_func_t *)zio_reexecute, zio, 0, |
3042 | &zio->io_tqent); | |
b128c09f BB |
3043 | } |
3044 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f BB |
3045 | } |
3046 | ||
428870ff | 3047 | ASSERT(zio->io_child_count == 0); |
b128c09f BB |
3048 | ASSERT(zio->io_reexecute == 0); |
3049 | ASSERT(zio->io_error == 0 || (zio->io_flags & ZIO_FLAG_CANFAIL)); | |
34dc7c2f | 3050 | |
428870ff BB |
3051 | /* |
3052 | * Report any checksum errors, since the I/O is complete. | |
3053 | */ | |
3054 | while (zio->io_cksum_report != NULL) { | |
3055 | zio_cksum_report_t *zcr = zio->io_cksum_report; | |
3056 | zio->io_cksum_report = zcr->zcr_next; | |
3057 | zcr->zcr_next = NULL; | |
3058 | zcr->zcr_finish(zcr, NULL); | |
3059 | zfs_ereport_free_checksum(zcr); | |
3060 | } | |
3061 | ||
d164b209 BB |
3062 | /* |
3063 | * It is the responsibility of the done callback to ensure that this | |
3064 | * particular zio is no longer discoverable for adoption, and as | |
3065 | * such, cannot acquire any new parents. | |
3066 | */ | |
b128c09f BB |
3067 | if (zio->io_done) |
3068 | zio->io_done(zio); | |
34dc7c2f | 3069 | |
d164b209 BB |
3070 | mutex_enter(&zio->io_lock); |
3071 | zio->io_state[ZIO_WAIT_DONE] = 1; | |
3072 | mutex_exit(&zio->io_lock); | |
34dc7c2f | 3073 | |
d164b209 BB |
3074 | for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { |
3075 | zio_link_t *zl = zio->io_walk_link; | |
3076 | pio_next = zio_walk_parents(zio); | |
3077 | zio_remove_child(pio, zio, zl); | |
b128c09f BB |
3078 | zio_notify_parent(pio, zio, ZIO_WAIT_DONE); |
3079 | } | |
34dc7c2f | 3080 | |
b128c09f BB |
3081 | if (zio->io_waiter != NULL) { |
3082 | mutex_enter(&zio->io_lock); | |
3083 | zio->io_executor = NULL; | |
3084 | cv_broadcast(&zio->io_cv); | |
3085 | mutex_exit(&zio->io_lock); | |
3086 | } else { | |
3087 | zio_destroy(zio); | |
3088 | } | |
34dc7c2f | 3089 | |
b128c09f | 3090 | return (ZIO_PIPELINE_STOP); |
34dc7c2f BB |
3091 | } |
3092 | ||
3093 | /* | |
b128c09f BB |
3094 | * ========================================================================== |
3095 | * I/O pipeline definition | |
3096 | * ========================================================================== | |
34dc7c2f | 3097 | */ |
428870ff | 3098 | static zio_pipe_stage_t *zio_pipeline[] = { |
b128c09f | 3099 | NULL, |
b128c09f | 3100 | zio_read_bp_init, |
428870ff BB |
3101 | zio_free_bp_init, |
3102 | zio_issue_async, | |
b128c09f BB |
3103 | zio_write_bp_init, |
3104 | zio_checksum_generate, | |
428870ff BB |
3105 | zio_ddt_read_start, |
3106 | zio_ddt_read_done, | |
3107 | zio_ddt_write, | |
3108 | zio_ddt_free, | |
b128c09f BB |
3109 | zio_gang_assemble, |
3110 | zio_gang_issue, | |
3111 | zio_dva_allocate, | |
3112 | zio_dva_free, | |
3113 | zio_dva_claim, | |
3114 | zio_ready, | |
3115 | zio_vdev_io_start, | |
3116 | zio_vdev_io_done, | |
3117 | zio_vdev_io_assess, | |
3118 | zio_checksum_verify, | |
3119 | zio_done | |
3120 | }; | |
c28b2279 BB |
3121 | |
3122 | #if defined(_KERNEL) && defined(HAVE_SPL) | |
3123 | /* Fault injection */ | |
3124 | EXPORT_SYMBOL(zio_injection_enabled); | |
3125 | EXPORT_SYMBOL(zio_inject_fault); | |
3126 | EXPORT_SYMBOL(zio_inject_list_next); | |
3127 | EXPORT_SYMBOL(zio_clear_fault); | |
3128 | EXPORT_SYMBOL(zio_handle_fault_injection); | |
3129 | EXPORT_SYMBOL(zio_handle_device_injection); | |
3130 | EXPORT_SYMBOL(zio_handle_label_injection); | |
3131 | EXPORT_SYMBOL(zio_priority_table); | |
3132 | EXPORT_SYMBOL(zio_type_name); | |
3133 | ||
3134 | module_param(zio_bulk_flags, int, 0644); | |
3135 | MODULE_PARM_DESC(zio_bulk_flags, "Additional flags to pass to bulk buffers"); | |
a69052be BB |
3136 | |
3137 | module_param(zio_delay_max, int, 0644); | |
c409e464 BB |
3138 | MODULE_PARM_DESC(zio_delay_max, "Max zio millisec delay before posting event"); |
3139 | ||
3140 | module_param(zio_requeue_io_start_cut_in_line, int, 0644); | |
3141 | MODULE_PARM_DESC(zio_requeue_io_start_cut_in_line, "Prioritize requeued I/O"); | |
c28b2279 | 3142 | #endif |