]>
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. |
9bd274dd | 23 | * Copyright (c) 2011, 2014 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> | |
9b67f605 | 39 | #include <sys/blkptr.h> |
b0bc7a84 | 40 | #include <sys/zfeature.h> |
34dc7c2f | 41 | |
34dc7c2f BB |
42 | /* |
43 | * ========================================================================== | |
44 | * I/O type descriptions | |
45 | * ========================================================================== | |
46 | */ | |
e8b96c60 | 47 | const char *zio_type_name[ZIO_TYPES] = { |
451041db | 48 | "z_null", "z_rd", "z_wr", "z_fr", "z_cl", "z_ioctl" |
428870ff | 49 | }; |
34dc7c2f BB |
50 | |
51 | /* | |
52 | * ========================================================================== | |
53 | * I/O kmem caches | |
54 | * ========================================================================== | |
55 | */ | |
56 | kmem_cache_t *zio_cache; | |
d164b209 | 57 | kmem_cache_t *zio_link_cache; |
86dd0fd9 | 58 | kmem_cache_t *zio_vdev_cache; |
34dc7c2f BB |
59 | kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; |
60 | kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; | |
c28b2279 | 61 | int zio_bulk_flags = 0; |
a69052be | 62 | int zio_delay_max = ZIO_DELAY_MAX; |
34dc7c2f | 63 | |
55d85d5a GW |
64 | /* |
65 | * The following actions directly effect the spa's sync-to-convergence logic. | |
66 | * The values below define the sync pass when we start performing the action. | |
67 | * Care should be taken when changing these values as they directly impact | |
68 | * spa_sync() performance. Tuning these values may introduce subtle performance | |
69 | * pathologies and should only be done in the context of performance analysis. | |
70 | * These tunables will eventually be removed and replaced with #defines once | |
71 | * enough analysis has been done to determine optimal values. | |
72 | * | |
73 | * The 'zfs_sync_pass_deferred_free' pass must be greater than 1 to ensure that | |
74 | * regular blocks are not deferred. | |
75 | */ | |
76 | int zfs_sync_pass_deferred_free = 2; /* defer frees starting in this pass */ | |
77 | int zfs_sync_pass_dont_compress = 5; /* don't compress starting in this pass */ | |
78 | int zfs_sync_pass_rewrite = 2; /* rewrite new bps starting in this pass */ | |
79 | ||
34dc7c2f | 80 | /* |
b128c09f BB |
81 | * An allocating zio is one that either currently has the DVA allocate |
82 | * stage set or will have it later in its lifetime. | |
34dc7c2f | 83 | */ |
428870ff BB |
84 | #define IO_IS_ALLOCATING(zio) ((zio)->io_orig_pipeline & ZIO_STAGE_DVA_ALLOCATE) |
85 | ||
c409e464 | 86 | int zio_requeue_io_start_cut_in_line = 1; |
428870ff BB |
87 | |
88 | #ifdef ZFS_DEBUG | |
89 | int zio_buf_debug_limit = 16384; | |
90 | #else | |
91 | int zio_buf_debug_limit = 0; | |
92 | #endif | |
34dc7c2f | 93 | |
da6b4005 NB |
94 | static inline void __zio_execute(zio_t *zio); |
95 | ||
49be0ccf BB |
96 | static int |
97 | zio_cons(void *arg, void *unused, int kmflag) | |
98 | { | |
99 | zio_t *zio = arg; | |
100 | ||
101 | bzero(zio, sizeof (zio_t)); | |
102 | ||
103 | mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL); | |
104 | cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL); | |
105 | ||
106 | list_create(&zio->io_parent_list, sizeof (zio_link_t), | |
107 | offsetof(zio_link_t, zl_parent_node)); | |
108 | list_create(&zio->io_child_list, sizeof (zio_link_t), | |
109 | offsetof(zio_link_t, zl_child_node)); | |
110 | ||
111 | return (0); | |
112 | } | |
113 | ||
114 | static void | |
115 | zio_dest(void *arg, void *unused) | |
116 | { | |
117 | zio_t *zio = arg; | |
118 | ||
119 | mutex_destroy(&zio->io_lock); | |
120 | cv_destroy(&zio->io_cv); | |
121 | list_destroy(&zio->io_parent_list); | |
122 | list_destroy(&zio->io_child_list); | |
123 | } | |
124 | ||
34dc7c2f BB |
125 | void |
126 | zio_init(void) | |
127 | { | |
128 | size_t c; | |
129 | vmem_t *data_alloc_arena = NULL; | |
130 | ||
49be0ccf | 131 | zio_cache = kmem_cache_create("zio_cache", sizeof (zio_t), 0, |
6795a698 | 132 | zio_cons, zio_dest, NULL, NULL, NULL, 0); |
d164b209 | 133 | zio_link_cache = kmem_cache_create("zio_link_cache", |
6795a698 | 134 | sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, 0); |
d1d7e268 | 135 | zio_vdev_cache = kmem_cache_create("zio_vdev_cache", sizeof (vdev_io_t), |
6795a698 | 136 | PAGESIZE, NULL, NULL, NULL, NULL, NULL, 0); |
34dc7c2f BB |
137 | |
138 | /* | |
139 | * For small buffers, we want a cache for each multiple of | |
140 | * SPA_MINBLOCKSIZE. For medium-size buffers, we want a cache | |
141 | * for each quarter-power of 2. For large buffers, we want | |
142 | * a cache for each multiple of PAGESIZE. | |
143 | */ | |
144 | for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { | |
145 | size_t size = (c + 1) << SPA_MINBLOCKSHIFT; | |
146 | size_t p2 = size; | |
147 | size_t align = 0; | |
148 | ||
149 | while (p2 & (p2 - 1)) | |
150 | p2 &= p2 - 1; | |
151 | ||
498877ba MA |
152 | #ifndef _KERNEL |
153 | /* | |
154 | * If we are using watchpoints, put each buffer on its own page, | |
155 | * to eliminate the performance overhead of trapping to the | |
156 | * kernel when modifying a non-watched buffer that shares the | |
157 | * page with a watched buffer. | |
158 | */ | |
159 | if (arc_watch && !IS_P2ALIGNED(size, PAGESIZE)) | |
160 | continue; | |
161 | #endif | |
34dc7c2f BB |
162 | if (size <= 4 * SPA_MINBLOCKSIZE) { |
163 | align = SPA_MINBLOCKSIZE; | |
498877ba | 164 | } else if (IS_P2ALIGNED(size, PAGESIZE)) { |
34dc7c2f | 165 | align = PAGESIZE; |
498877ba | 166 | } else if (IS_P2ALIGNED(size, p2 >> 2)) { |
34dc7c2f BB |
167 | align = p2 >> 2; |
168 | } | |
169 | ||
170 | if (align != 0) { | |
171 | char name[36]; | |
ae6ba3db BB |
172 | int flags = zio_bulk_flags; |
173 | ||
34dc7c2f BB |
174 | (void) sprintf(name, "zio_buf_%lu", (ulong_t)size); |
175 | zio_buf_cache[c] = kmem_cache_create(name, size, | |
ae6ba3db | 176 | align, NULL, NULL, NULL, NULL, NULL, flags); |
34dc7c2f BB |
177 | |
178 | (void) sprintf(name, "zio_data_buf_%lu", (ulong_t)size); | |
179 | zio_data_buf_cache[c] = kmem_cache_create(name, size, | |
ae6ba3db BB |
180 | align, NULL, NULL, NULL, NULL, |
181 | data_alloc_arena, flags); | |
34dc7c2f BB |
182 | } |
183 | } | |
184 | ||
185 | while (--c != 0) { | |
186 | ASSERT(zio_buf_cache[c] != NULL); | |
187 | if (zio_buf_cache[c - 1] == NULL) | |
188 | zio_buf_cache[c - 1] = zio_buf_cache[c]; | |
189 | ||
190 | ASSERT(zio_data_buf_cache[c] != NULL); | |
191 | if (zio_data_buf_cache[c - 1] == NULL) | |
192 | zio_data_buf_cache[c - 1] = zio_data_buf_cache[c]; | |
193 | } | |
194 | ||
34dc7c2f | 195 | zio_inject_init(); |
9759c60f ED |
196 | |
197 | lz4_init(); | |
34dc7c2f BB |
198 | } |
199 | ||
200 | void | |
201 | zio_fini(void) | |
202 | { | |
203 | size_t c; | |
204 | kmem_cache_t *last_cache = NULL; | |
205 | kmem_cache_t *last_data_cache = NULL; | |
206 | ||
207 | for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { | |
208 | if (zio_buf_cache[c] != last_cache) { | |
209 | last_cache = zio_buf_cache[c]; | |
210 | kmem_cache_destroy(zio_buf_cache[c]); | |
211 | } | |
212 | zio_buf_cache[c] = NULL; | |
213 | ||
214 | if (zio_data_buf_cache[c] != last_data_cache) { | |
215 | last_data_cache = zio_data_buf_cache[c]; | |
216 | kmem_cache_destroy(zio_data_buf_cache[c]); | |
217 | } | |
218 | zio_data_buf_cache[c] = NULL; | |
219 | } | |
220 | ||
86dd0fd9 | 221 | kmem_cache_destroy(zio_vdev_cache); |
d164b209 | 222 | kmem_cache_destroy(zio_link_cache); |
34dc7c2f BB |
223 | kmem_cache_destroy(zio_cache); |
224 | ||
225 | zio_inject_fini(); | |
9759c60f ED |
226 | |
227 | lz4_fini(); | |
34dc7c2f BB |
228 | } |
229 | ||
230 | /* | |
231 | * ========================================================================== | |
232 | * Allocate and free I/O buffers | |
233 | * ========================================================================== | |
234 | */ | |
235 | ||
236 | /* | |
237 | * Use zio_buf_alloc to allocate ZFS metadata. This data will appear in a | |
238 | * crashdump if the kernel panics, so use it judiciously. Obviously, it's | |
239 | * useful to inspect ZFS metadata, but if possible, we should avoid keeping | |
240 | * excess / transient data in-core during a crashdump. | |
241 | */ | |
242 | void * | |
243 | zio_buf_alloc(size_t size) | |
244 | { | |
245 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
246 | ||
9b67f605 | 247 | ASSERT3U(c, <, SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); |
34dc7c2f | 248 | |
ebcfc8a5 | 249 | return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE | KM_NODEBUG)); |
34dc7c2f BB |
250 | } |
251 | ||
252 | /* | |
253 | * Use zio_data_buf_alloc to allocate data. The data will not appear in a | |
254 | * crashdump if the kernel panics. This exists so that we will limit the amount | |
255 | * of ZFS data that shows up in a kernel crashdump. (Thus reducing the amount | |
256 | * of kernel heap dumped to disk when the kernel panics) | |
257 | */ | |
258 | void * | |
259 | zio_data_buf_alloc(size_t size) | |
260 | { | |
261 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
262 | ||
263 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
264 | ||
ebcfc8a5 BB |
265 | return (kmem_cache_alloc(zio_data_buf_cache[c], |
266 | KM_PUSHPAGE | KM_NODEBUG)); | |
34dc7c2f BB |
267 | } |
268 | ||
269 | void | |
270 | zio_buf_free(void *buf, size_t size) | |
271 | { | |
272 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
273 | ||
274 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
275 | ||
276 | kmem_cache_free(zio_buf_cache[c], buf); | |
277 | } | |
278 | ||
279 | void | |
280 | zio_data_buf_free(void *buf, size_t size) | |
281 | { | |
282 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
283 | ||
284 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
285 | ||
286 | kmem_cache_free(zio_data_buf_cache[c], buf); | |
287 | } | |
288 | ||
86dd0fd9 BB |
289 | /* |
290 | * Dedicated I/O buffers to ensure that memory fragmentation never prevents | |
291 | * or significantly delays the issuing of a zio. These buffers are used | |
292 | * to aggregate I/O and could be used for raidz stripes. | |
293 | */ | |
294 | void * | |
295 | zio_vdev_alloc(void) | |
296 | { | |
297 | return (kmem_cache_alloc(zio_vdev_cache, KM_PUSHPAGE)); | |
298 | } | |
299 | ||
300 | void | |
301 | zio_vdev_free(void *buf) | |
302 | { | |
303 | kmem_cache_free(zio_vdev_cache, buf); | |
304 | ||
305 | } | |
306 | ||
34dc7c2f BB |
307 | /* |
308 | * ========================================================================== | |
309 | * Push and pop I/O transform buffers | |
310 | * ========================================================================== | |
311 | */ | |
312 | static void | |
b128c09f BB |
313 | zio_push_transform(zio_t *zio, void *data, uint64_t size, uint64_t bufsize, |
314 | zio_transform_func_t *transform) | |
34dc7c2f | 315 | { |
691f6ac4 | 316 | zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_PUSHPAGE); |
34dc7c2f | 317 | |
b128c09f BB |
318 | zt->zt_orig_data = zio->io_data; |
319 | zt->zt_orig_size = zio->io_size; | |
34dc7c2f | 320 | zt->zt_bufsize = bufsize; |
b128c09f | 321 | zt->zt_transform = transform; |
34dc7c2f BB |
322 | |
323 | zt->zt_next = zio->io_transform_stack; | |
324 | zio->io_transform_stack = zt; | |
325 | ||
326 | zio->io_data = data; | |
327 | zio->io_size = size; | |
328 | } | |
329 | ||
330 | static void | |
b128c09f | 331 | zio_pop_transforms(zio_t *zio) |
34dc7c2f | 332 | { |
b128c09f BB |
333 | zio_transform_t *zt; |
334 | ||
335 | while ((zt = zio->io_transform_stack) != NULL) { | |
336 | if (zt->zt_transform != NULL) | |
337 | zt->zt_transform(zio, | |
338 | zt->zt_orig_data, zt->zt_orig_size); | |
34dc7c2f | 339 | |
428870ff BB |
340 | if (zt->zt_bufsize != 0) |
341 | zio_buf_free(zio->io_data, zt->zt_bufsize); | |
34dc7c2f | 342 | |
b128c09f BB |
343 | zio->io_data = zt->zt_orig_data; |
344 | zio->io_size = zt->zt_orig_size; | |
345 | zio->io_transform_stack = zt->zt_next; | |
34dc7c2f | 346 | |
b128c09f | 347 | kmem_free(zt, sizeof (zio_transform_t)); |
34dc7c2f BB |
348 | } |
349 | } | |
350 | ||
b128c09f BB |
351 | /* |
352 | * ========================================================================== | |
353 | * I/O transform callbacks for subblocks and decompression | |
354 | * ========================================================================== | |
355 | */ | |
356 | static void | |
357 | zio_subblock(zio_t *zio, void *data, uint64_t size) | |
358 | { | |
359 | ASSERT(zio->io_size > size); | |
360 | ||
361 | if (zio->io_type == ZIO_TYPE_READ) | |
362 | bcopy(zio->io_data, data, size); | |
363 | } | |
364 | ||
365 | static void | |
366 | zio_decompress(zio_t *zio, void *data, uint64_t size) | |
367 | { | |
368 | if (zio->io_error == 0 && | |
369 | zio_decompress_data(BP_GET_COMPRESS(zio->io_bp), | |
428870ff | 370 | zio->io_data, data, zio->io_size, size) != 0) |
2e528b49 | 371 | zio->io_error = SET_ERROR(EIO); |
b128c09f BB |
372 | } |
373 | ||
374 | /* | |
375 | * ========================================================================== | |
376 | * I/O parent/child relationships and pipeline interlocks | |
377 | * ========================================================================== | |
378 | */ | |
d164b209 BB |
379 | /* |
380 | * NOTE - Callers to zio_walk_parents() and zio_walk_children must | |
381 | * continue calling these functions until they return NULL. | |
382 | * Otherwise, the next caller will pick up the list walk in | |
383 | * some indeterminate state. (Otherwise every caller would | |
384 | * have to pass in a cookie to keep the state represented by | |
385 | * io_walk_link, which gets annoying.) | |
386 | */ | |
387 | zio_t * | |
388 | zio_walk_parents(zio_t *cio) | |
389 | { | |
390 | zio_link_t *zl = cio->io_walk_link; | |
391 | list_t *pl = &cio->io_parent_list; | |
b128c09f | 392 | |
d164b209 BB |
393 | zl = (zl == NULL) ? list_head(pl) : list_next(pl, zl); |
394 | cio->io_walk_link = zl; | |
395 | ||
396 | if (zl == NULL) | |
397 | return (NULL); | |
398 | ||
399 | ASSERT(zl->zl_child == cio); | |
400 | return (zl->zl_parent); | |
401 | } | |
402 | ||
403 | zio_t * | |
404 | zio_walk_children(zio_t *pio) | |
405 | { | |
406 | zio_link_t *zl = pio->io_walk_link; | |
407 | list_t *cl = &pio->io_child_list; | |
408 | ||
409 | zl = (zl == NULL) ? list_head(cl) : list_next(cl, zl); | |
410 | pio->io_walk_link = zl; | |
411 | ||
412 | if (zl == NULL) | |
413 | return (NULL); | |
414 | ||
415 | ASSERT(zl->zl_parent == pio); | |
416 | return (zl->zl_child); | |
417 | } | |
418 | ||
419 | zio_t * | |
420 | zio_unique_parent(zio_t *cio) | |
421 | { | |
422 | zio_t *pio = zio_walk_parents(cio); | |
423 | ||
424 | VERIFY(zio_walk_parents(cio) == NULL); | |
425 | return (pio); | |
426 | } | |
427 | ||
428 | void | |
429 | zio_add_child(zio_t *pio, zio_t *cio) | |
b128c09f | 430 | { |
691f6ac4 | 431 | zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_PUSHPAGE); |
d6320ddb | 432 | int w; |
d164b209 BB |
433 | |
434 | /* | |
435 | * Logical I/Os can have logical, gang, or vdev children. | |
436 | * Gang I/Os can have gang or vdev children. | |
437 | * Vdev I/Os can only have vdev children. | |
438 | * The following ASSERT captures all of these constraints. | |
439 | */ | |
440 | ASSERT(cio->io_child_type <= pio->io_child_type); | |
441 | ||
442 | zl->zl_parent = pio; | |
443 | zl->zl_child = cio; | |
444 | ||
445 | mutex_enter(&cio->io_lock); | |
b128c09f | 446 | mutex_enter(&pio->io_lock); |
d164b209 BB |
447 | |
448 | ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0); | |
449 | ||
d6320ddb | 450 | for (w = 0; w < ZIO_WAIT_TYPES; w++) |
d164b209 BB |
451 | pio->io_children[cio->io_child_type][w] += !cio->io_state[w]; |
452 | ||
453 | list_insert_head(&pio->io_child_list, zl); | |
454 | list_insert_head(&cio->io_parent_list, zl); | |
455 | ||
428870ff BB |
456 | pio->io_child_count++; |
457 | cio->io_parent_count++; | |
458 | ||
b128c09f | 459 | mutex_exit(&pio->io_lock); |
d164b209 | 460 | mutex_exit(&cio->io_lock); |
b128c09f BB |
461 | } |
462 | ||
34dc7c2f | 463 | static void |
d164b209 | 464 | zio_remove_child(zio_t *pio, zio_t *cio, zio_link_t *zl) |
b128c09f | 465 | { |
d164b209 BB |
466 | ASSERT(zl->zl_parent == pio); |
467 | ASSERT(zl->zl_child == cio); | |
b128c09f | 468 | |
d164b209 | 469 | mutex_enter(&cio->io_lock); |
b128c09f | 470 | mutex_enter(&pio->io_lock); |
d164b209 BB |
471 | |
472 | list_remove(&pio->io_child_list, zl); | |
473 | list_remove(&cio->io_parent_list, zl); | |
474 | ||
428870ff BB |
475 | pio->io_child_count--; |
476 | cio->io_parent_count--; | |
477 | ||
b128c09f | 478 | mutex_exit(&pio->io_lock); |
d164b209 BB |
479 | mutex_exit(&cio->io_lock); |
480 | ||
481 | kmem_cache_free(zio_link_cache, zl); | |
b128c09f BB |
482 | } |
483 | ||
484 | static boolean_t | |
485 | zio_wait_for_children(zio_t *zio, enum zio_child child, enum zio_wait_type wait) | |
34dc7c2f | 486 | { |
b128c09f BB |
487 | uint64_t *countp = &zio->io_children[child][wait]; |
488 | boolean_t waiting = B_FALSE; | |
489 | ||
490 | mutex_enter(&zio->io_lock); | |
491 | ASSERT(zio->io_stall == NULL); | |
492 | if (*countp != 0) { | |
428870ff | 493 | zio->io_stage >>= 1; |
b128c09f BB |
494 | zio->io_stall = countp; |
495 | waiting = B_TRUE; | |
496 | } | |
497 | mutex_exit(&zio->io_lock); | |
498 | ||
499 | return (waiting); | |
500 | } | |
34dc7c2f | 501 | |
bf701a83 BB |
502 | __attribute__((always_inline)) |
503 | static inline void | |
b128c09f BB |
504 | zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait) |
505 | { | |
506 | uint64_t *countp = &pio->io_children[zio->io_child_type][wait]; | |
507 | int *errorp = &pio->io_child_error[zio->io_child_type]; | |
34dc7c2f | 508 | |
b128c09f BB |
509 | mutex_enter(&pio->io_lock); |
510 | if (zio->io_error && !(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) | |
511 | *errorp = zio_worst_error(*errorp, zio->io_error); | |
512 | pio->io_reexecute |= zio->io_reexecute; | |
513 | ASSERT3U(*countp, >, 0); | |
e8b96c60 MA |
514 | |
515 | (*countp)--; | |
516 | ||
517 | if (*countp == 0 && pio->io_stall == countp) { | |
b128c09f BB |
518 | pio->io_stall = NULL; |
519 | mutex_exit(&pio->io_lock); | |
da6b4005 | 520 | __zio_execute(pio); |
b128c09f BB |
521 | } else { |
522 | mutex_exit(&pio->io_lock); | |
34dc7c2f BB |
523 | } |
524 | } | |
525 | ||
b128c09f BB |
526 | static void |
527 | zio_inherit_child_errors(zio_t *zio, enum zio_child c) | |
528 | { | |
529 | if (zio->io_child_error[c] != 0 && zio->io_error == 0) | |
530 | zio->io_error = zio->io_child_error[c]; | |
531 | } | |
532 | ||
34dc7c2f BB |
533 | /* |
534 | * ========================================================================== | |
b128c09f | 535 | * Create the various types of I/O (read, write, free, etc) |
34dc7c2f BB |
536 | * ========================================================================== |
537 | */ | |
538 | static zio_t * | |
428870ff | 539 | zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, |
34dc7c2f | 540 | void *data, uint64_t size, zio_done_func_t *done, void *private, |
e8b96c60 | 541 | zio_type_t type, zio_priority_t priority, enum zio_flag flags, |
5dbd68a3 | 542 | vdev_t *vd, uint64_t offset, const zbookmark_phys_t *zb, |
428870ff | 543 | enum zio_stage stage, enum zio_stage pipeline) |
34dc7c2f BB |
544 | { |
545 | zio_t *zio; | |
546 | ||
547 | ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); | |
548 | ASSERT(P2PHASE(size, SPA_MINBLOCKSIZE) == 0); | |
b128c09f BB |
549 | ASSERT(P2PHASE(offset, SPA_MINBLOCKSIZE) == 0); |
550 | ||
551 | ASSERT(!vd || spa_config_held(spa, SCL_STATE_ALL, RW_READER)); | |
552 | ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER)); | |
553 | ASSERT(vd || stage == ZIO_STAGE_OPEN); | |
34dc7c2f | 554 | |
691f6ac4 | 555 | zio = kmem_cache_alloc(zio_cache, KM_PUSHPAGE); |
d164b209 | 556 | |
b128c09f BB |
557 | if (vd != NULL) |
558 | zio->io_child_type = ZIO_CHILD_VDEV; | |
559 | else if (flags & ZIO_FLAG_GANG_CHILD) | |
560 | zio->io_child_type = ZIO_CHILD_GANG; | |
428870ff BB |
561 | else if (flags & ZIO_FLAG_DDT_CHILD) |
562 | zio->io_child_type = ZIO_CHILD_DDT; | |
b128c09f BB |
563 | else |
564 | zio->io_child_type = ZIO_CHILD_LOGICAL; | |
565 | ||
34dc7c2f | 566 | if (bp != NULL) { |
49be0ccf | 567 | zio->io_logical = NULL; |
428870ff | 568 | zio->io_bp = (blkptr_t *)bp; |
34dc7c2f BB |
569 | zio->io_bp_copy = *bp; |
570 | zio->io_bp_orig = *bp; | |
428870ff BB |
571 | if (type != ZIO_TYPE_WRITE || |
572 | zio->io_child_type == ZIO_CHILD_DDT) | |
b128c09f | 573 | zio->io_bp = &zio->io_bp_copy; /* so caller can free */ |
9babb374 | 574 | if (zio->io_child_type == ZIO_CHILD_LOGICAL) |
b128c09f | 575 | zio->io_logical = zio; |
9babb374 BB |
576 | if (zio->io_child_type > ZIO_CHILD_GANG && BP_IS_GANG(bp)) |
577 | pipeline |= ZIO_GANG_STAGES; | |
49be0ccf BB |
578 | } else { |
579 | zio->io_logical = NULL; | |
580 | zio->io_bp = NULL; | |
581 | bzero(&zio->io_bp_copy, sizeof (blkptr_t)); | |
582 | bzero(&zio->io_bp_orig, sizeof (blkptr_t)); | |
34dc7c2f | 583 | } |
b128c09f BB |
584 | |
585 | zio->io_spa = spa; | |
586 | zio->io_txg = txg; | |
49be0ccf | 587 | zio->io_ready = NULL; |
e8b96c60 | 588 | zio->io_physdone = NULL; |
34dc7c2f BB |
589 | zio->io_done = done; |
590 | zio->io_private = private; | |
49be0ccf | 591 | zio->io_prev_space_delta = 0; |
34dc7c2f BB |
592 | zio->io_type = type; |
593 | zio->io_priority = priority; | |
b128c09f | 594 | zio->io_vd = vd; |
49be0ccf BB |
595 | zio->io_vsd = NULL; |
596 | zio->io_vsd_ops = NULL; | |
b128c09f | 597 | zio->io_offset = offset; |
cc92e9d0 GW |
598 | zio->io_timestamp = 0; |
599 | zio->io_delta = 0; | |
600 | zio->io_delay = 0; | |
428870ff BB |
601 | zio->io_orig_data = zio->io_data = data; |
602 | zio->io_orig_size = zio->io_size = size; | |
b128c09f BB |
603 | zio->io_orig_flags = zio->io_flags = flags; |
604 | zio->io_orig_stage = zio->io_stage = stage; | |
605 | zio->io_orig_pipeline = zio->io_pipeline = pipeline; | |
49be0ccf BB |
606 | bzero(&zio->io_prop, sizeof (zio_prop_t)); |
607 | zio->io_cmd = 0; | |
608 | zio->io_reexecute = 0; | |
609 | zio->io_bp_override = NULL; | |
610 | zio->io_walk_link = NULL; | |
611 | zio->io_transform_stack = NULL; | |
49be0ccf BB |
612 | zio->io_error = 0; |
613 | zio->io_child_count = 0; | |
e8b96c60 | 614 | zio->io_phys_children = 0; |
49be0ccf BB |
615 | zio->io_parent_count = 0; |
616 | zio->io_stall = NULL; | |
617 | zio->io_gang_leader = NULL; | |
618 | zio->io_gang_tree = NULL; | |
619 | zio->io_executor = NULL; | |
620 | zio->io_waiter = NULL; | |
621 | zio->io_cksum_report = NULL; | |
622 | zio->io_ena = 0; | |
623 | bzero(zio->io_child_error, sizeof (int) * ZIO_CHILD_TYPES); | |
624 | bzero(zio->io_children, | |
625 | sizeof (uint64_t) * ZIO_CHILD_TYPES * ZIO_WAIT_TYPES); | |
5dbd68a3 | 626 | bzero(&zio->io_bookmark, sizeof (zbookmark_phys_t)); |
34dc7c2f | 627 | |
d164b209 BB |
628 | zio->io_state[ZIO_WAIT_READY] = (stage >= ZIO_STAGE_READY); |
629 | zio->io_state[ZIO_WAIT_DONE] = (stage >= ZIO_STAGE_DONE); | |
630 | ||
b128c09f BB |
631 | if (zb != NULL) |
632 | zio->io_bookmark = *zb; | |
633 | ||
634 | if (pio != NULL) { | |
b128c09f | 635 | if (zio->io_logical == NULL) |
34dc7c2f | 636 | zio->io_logical = pio->io_logical; |
9babb374 BB |
637 | if (zio->io_child_type == ZIO_CHILD_GANG) |
638 | zio->io_gang_leader = pio->io_gang_leader; | |
b128c09f | 639 | zio_add_child(pio, zio); |
34dc7c2f BB |
640 | } |
641 | ||
a38718a6 GA |
642 | taskq_init_ent(&zio->io_tqent); |
643 | ||
34dc7c2f BB |
644 | return (zio); |
645 | } | |
646 | ||
647 | static void | |
b128c09f | 648 | zio_destroy(zio_t *zio) |
34dc7c2f | 649 | { |
b128c09f | 650 | kmem_cache_free(zio_cache, zio); |
34dc7c2f BB |
651 | } |
652 | ||
653 | zio_t * | |
d164b209 | 654 | zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done, |
428870ff | 655 | void *private, enum zio_flag flags) |
34dc7c2f BB |
656 | { |
657 | zio_t *zio; | |
658 | ||
659 | zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, | |
d164b209 | 660 | ZIO_TYPE_NULL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, |
b128c09f | 661 | ZIO_STAGE_OPEN, ZIO_INTERLOCK_PIPELINE); |
34dc7c2f BB |
662 | |
663 | return (zio); | |
664 | } | |
665 | ||
666 | zio_t * | |
428870ff | 667 | zio_root(spa_t *spa, zio_done_func_t *done, void *private, enum zio_flag flags) |
34dc7c2f | 668 | { |
d164b209 | 669 | return (zio_null(NULL, spa, NULL, done, private, flags)); |
34dc7c2f BB |
670 | } |
671 | ||
672 | zio_t * | |
b128c09f BB |
673 | zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, |
674 | void *data, uint64_t size, zio_done_func_t *done, void *private, | |
5dbd68a3 | 675 | zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb) |
34dc7c2f BB |
676 | { |
677 | zio_t *zio; | |
678 | ||
428870ff | 679 | zio = zio_create(pio, spa, BP_PHYSICAL_BIRTH(bp), bp, |
b128c09f BB |
680 | data, size, done, private, |
681 | ZIO_TYPE_READ, priority, flags, NULL, 0, zb, | |
428870ff BB |
682 | ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? |
683 | ZIO_DDT_CHILD_READ_PIPELINE : ZIO_READ_PIPELINE); | |
34dc7c2f | 684 | |
b128c09f BB |
685 | return (zio); |
686 | } | |
34dc7c2f | 687 | |
34dc7c2f | 688 | zio_t * |
b128c09f | 689 | zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, |
428870ff | 690 | void *data, uint64_t size, const zio_prop_t *zp, |
e8b96c60 MA |
691 | zio_done_func_t *ready, zio_done_func_t *physdone, zio_done_func_t *done, |
692 | void *private, | |
5dbd68a3 | 693 | zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb) |
34dc7c2f BB |
694 | { |
695 | zio_t *zio; | |
696 | ||
b128c09f BB |
697 | ASSERT(zp->zp_checksum >= ZIO_CHECKSUM_OFF && |
698 | zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS && | |
699 | zp->zp_compress >= ZIO_COMPRESS_OFF && | |
700 | zp->zp_compress < ZIO_COMPRESS_FUNCTIONS && | |
9ae529ec | 701 | DMU_OT_IS_VALID(zp->zp_type) && |
b128c09f | 702 | zp->zp_level < 32 && |
428870ff | 703 | zp->zp_copies > 0 && |
03c6040b | 704 | zp->zp_copies <= spa_max_replication(spa)); |
34dc7c2f BB |
705 | |
706 | zio = zio_create(pio, spa, txg, bp, data, size, done, private, | |
b128c09f | 707 | ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, |
428870ff BB |
708 | ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? |
709 | ZIO_DDT_CHILD_WRITE_PIPELINE : ZIO_WRITE_PIPELINE); | |
34dc7c2f BB |
710 | |
711 | zio->io_ready = ready; | |
e8b96c60 | 712 | zio->io_physdone = physdone; |
b128c09f | 713 | zio->io_prop = *zp; |
34dc7c2f | 714 | |
9b67f605 MA |
715 | /* |
716 | * Data can be NULL if we are going to call zio_write_override() to | |
717 | * provide the already-allocated BP. But we may need the data to | |
718 | * verify a dedup hit (if requested). In this case, don't try to | |
719 | * dedup (just take the already-allocated BP verbatim). | |
720 | */ | |
721 | if (data == NULL && zio->io_prop.zp_dedup_verify) { | |
722 | zio->io_prop.zp_dedup = zio->io_prop.zp_dedup_verify = B_FALSE; | |
723 | } | |
724 | ||
34dc7c2f BB |
725 | return (zio); |
726 | } | |
727 | ||
728 | zio_t * | |
b128c09f | 729 | zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data, |
e8b96c60 | 730 | uint64_t size, zio_done_func_t *done, void *private, |
5dbd68a3 | 731 | zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb) |
34dc7c2f BB |
732 | { |
733 | zio_t *zio; | |
734 | ||
34dc7c2f | 735 | zio = zio_create(pio, spa, txg, bp, data, size, done, private, |
b128c09f BB |
736 | ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, |
737 | ZIO_STAGE_OPEN, ZIO_REWRITE_PIPELINE); | |
34dc7c2f BB |
738 | |
739 | return (zio); | |
740 | } | |
741 | ||
428870ff | 742 | void |
03c6040b | 743 | zio_write_override(zio_t *zio, blkptr_t *bp, int copies, boolean_t nopwrite) |
428870ff BB |
744 | { |
745 | ASSERT(zio->io_type == ZIO_TYPE_WRITE); | |
746 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
747 | ASSERT(zio->io_stage == ZIO_STAGE_OPEN); | |
748 | ASSERT(zio->io_txg == spa_syncing_txg(zio->io_spa)); | |
749 | ||
03c6040b GW |
750 | /* |
751 | * We must reset the io_prop to match the values that existed | |
752 | * when the bp was first written by dmu_sync() keeping in mind | |
753 | * that nopwrite and dedup are mutually exclusive. | |
754 | */ | |
755 | zio->io_prop.zp_dedup = nopwrite ? B_FALSE : zio->io_prop.zp_dedup; | |
756 | zio->io_prop.zp_nopwrite = nopwrite; | |
428870ff BB |
757 | zio->io_prop.zp_copies = copies; |
758 | zio->io_bp_override = bp; | |
759 | } | |
760 | ||
761 | void | |
762 | zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp) | |
763 | { | |
9b67f605 MA |
764 | |
765 | /* | |
766 | * The check for EMBEDDED is a performance optimization. We | |
767 | * process the free here (by ignoring it) rather than | |
768 | * putting it on the list and then processing it in zio_free_sync(). | |
769 | */ | |
770 | if (BP_IS_EMBEDDED(bp)) | |
771 | return; | |
13fe0198 | 772 | metaslab_check_free(spa, bp); |
2883cad5 MA |
773 | |
774 | /* | |
775 | * Frees that are for the currently-syncing txg, are not going to be | |
776 | * deferred, and which will not need to do a read (i.e. not GANG or | |
777 | * DEDUP), can be processed immediately. Otherwise, put them on the | |
778 | * in-memory list for later processing. | |
779 | */ | |
780 | if (BP_IS_GANG(bp) || BP_GET_DEDUP(bp) || | |
781 | txg != spa->spa_syncing_txg || | |
782 | spa_sync_pass(spa) >= zfs_sync_pass_deferred_free) { | |
783 | bplist_append(&spa->spa_free_bplist[txg & TXG_MASK], bp); | |
784 | } else { | |
785 | VERIFY0(zio_wait(zio_free_sync(NULL, spa, txg, bp, 0))); | |
786 | } | |
428870ff BB |
787 | } |
788 | ||
34dc7c2f | 789 | zio_t * |
428870ff BB |
790 | zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, |
791 | enum zio_flag flags) | |
34dc7c2f BB |
792 | { |
793 | zio_t *zio; | |
2883cad5 | 794 | enum zio_stage stage = ZIO_FREE_PIPELINE; |
34dc7c2f | 795 | |
428870ff BB |
796 | ASSERT(!BP_IS_HOLE(bp)); |
797 | ASSERT(spa_syncing_txg(spa) == txg); | |
55d85d5a | 798 | ASSERT(spa_sync_pass(spa) < zfs_sync_pass_deferred_free); |
34dc7c2f | 799 | |
9b67f605 MA |
800 | if (BP_IS_EMBEDDED(bp)) |
801 | return (zio_null(pio, spa, NULL, NULL, NULL, 0)); | |
802 | ||
13fe0198 | 803 | metaslab_check_free(spa, bp); |
8c841793 | 804 | arc_freed(spa, bp); |
13fe0198 | 805 | |
2883cad5 MA |
806 | /* |
807 | * GANG and DEDUP blocks can induce a read (for the gang block header, | |
808 | * or the DDT), so issue them asynchronously so that this thread is | |
809 | * not tied up. | |
810 | */ | |
811 | if (BP_IS_GANG(bp) || BP_GET_DEDUP(bp)) | |
812 | stage |= ZIO_STAGE_ISSUE_ASYNC; | |
813 | ||
b128c09f | 814 | zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), |
2883cad5 MA |
815 | NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_NOW, flags, |
816 | NULL, 0, NULL, ZIO_STAGE_OPEN, stage); | |
817 | ||
34dc7c2f BB |
818 | return (zio); |
819 | } | |
820 | ||
821 | zio_t * | |
428870ff BB |
822 | zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, |
823 | zio_done_func_t *done, void *private, enum zio_flag flags) | |
34dc7c2f BB |
824 | { |
825 | zio_t *zio; | |
826 | ||
9b67f605 MA |
827 | dprintf_bp(bp, "claiming in txg %llu", txg); |
828 | ||
829 | if (BP_IS_EMBEDDED(bp)) | |
830 | return (zio_null(pio, spa, NULL, NULL, NULL, 0)); | |
831 | ||
34dc7c2f BB |
832 | /* |
833 | * A claim is an allocation of a specific block. Claims are needed | |
834 | * to support immediate writes in the intent log. The issue is that | |
835 | * immediate writes contain committed data, but in a txg that was | |
836 | * *not* committed. Upon opening the pool after an unclean shutdown, | |
837 | * the intent log claims all blocks that contain immediate write data | |
838 | * so that the SPA knows they're in use. | |
839 | * | |
840 | * All claims *must* be resolved in the first txg -- before the SPA | |
841 | * starts allocating blocks -- so that nothing is allocated twice. | |
428870ff | 842 | * If txg == 0 we just verify that the block is claimable. |
34dc7c2f BB |
843 | */ |
844 | ASSERT3U(spa->spa_uberblock.ub_rootbp.blk_birth, <, spa_first_txg(spa)); | |
428870ff BB |
845 | ASSERT(txg == spa_first_txg(spa) || txg == 0); |
846 | ASSERT(!BP_GET_DEDUP(bp) || !spa_writeable(spa)); /* zdb(1M) */ | |
34dc7c2f | 847 | |
b128c09f BB |
848 | zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), |
849 | done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags, | |
850 | NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE); | |
34dc7c2f BB |
851 | |
852 | return (zio); | |
853 | } | |
854 | ||
855 | zio_t * | |
856 | zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, | |
e8b96c60 | 857 | zio_done_func_t *done, void *private, enum zio_flag flags) |
34dc7c2f BB |
858 | { |
859 | zio_t *zio; | |
860 | int c; | |
861 | ||
862 | if (vd->vdev_children == 0) { | |
863 | zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, | |
e8b96c60 | 864 | ZIO_TYPE_IOCTL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, |
34dc7c2f BB |
865 | ZIO_STAGE_OPEN, ZIO_IOCTL_PIPELINE); |
866 | ||
34dc7c2f BB |
867 | zio->io_cmd = cmd; |
868 | } else { | |
d164b209 | 869 | zio = zio_null(pio, spa, NULL, NULL, NULL, flags); |
34dc7c2f BB |
870 | |
871 | for (c = 0; c < vd->vdev_children; c++) | |
872 | zio_nowait(zio_ioctl(zio, spa, vd->vdev_child[c], cmd, | |
e8b96c60 | 873 | done, private, flags)); |
34dc7c2f BB |
874 | } |
875 | ||
876 | return (zio); | |
877 | } | |
878 | ||
34dc7c2f BB |
879 | zio_t * |
880 | zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, | |
881 | void *data, int checksum, zio_done_func_t *done, void *private, | |
e8b96c60 | 882 | zio_priority_t priority, enum zio_flag flags, boolean_t labels) |
34dc7c2f BB |
883 | { |
884 | zio_t *zio; | |
34dc7c2f | 885 | |
b128c09f BB |
886 | ASSERT(vd->vdev_children == 0); |
887 | ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || | |
888 | offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); | |
889 | ASSERT3U(offset + size, <=, vd->vdev_psize); | |
34dc7c2f | 890 | |
b128c09f | 891 | zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, |
b02fe35d AR |
892 | ZIO_TYPE_READ, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset, |
893 | NULL, ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE); | |
34dc7c2f | 894 | |
b128c09f | 895 | zio->io_prop.zp_checksum = checksum; |
34dc7c2f BB |
896 | |
897 | return (zio); | |
898 | } | |
899 | ||
900 | zio_t * | |
901 | zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, | |
902 | void *data, int checksum, zio_done_func_t *done, void *private, | |
e8b96c60 | 903 | zio_priority_t priority, enum zio_flag flags, boolean_t labels) |
34dc7c2f | 904 | { |
34dc7c2f | 905 | zio_t *zio; |
34dc7c2f | 906 | |
b128c09f BB |
907 | ASSERT(vd->vdev_children == 0); |
908 | ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || | |
909 | offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); | |
910 | ASSERT3U(offset + size, <=, vd->vdev_psize); | |
34dc7c2f | 911 | |
b128c09f | 912 | zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, |
b02fe35d AR |
913 | ZIO_TYPE_WRITE, priority, flags | ZIO_FLAG_PHYSICAL, vd, offset, |
914 | NULL, ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE); | |
34dc7c2f | 915 | |
b128c09f | 916 | zio->io_prop.zp_checksum = checksum; |
34dc7c2f | 917 | |
428870ff | 918 | if (zio_checksum_table[checksum].ci_eck) { |
34dc7c2f | 919 | /* |
428870ff | 920 | * zec checksums are necessarily destructive -- they modify |
b128c09f | 921 | * the end of the write buffer to hold the verifier/checksum. |
34dc7c2f | 922 | * Therefore, we must make a local copy in case the data is |
b128c09f | 923 | * being written to multiple places in parallel. |
34dc7c2f | 924 | */ |
b128c09f | 925 | void *wbuf = zio_buf_alloc(size); |
34dc7c2f | 926 | bcopy(data, wbuf, size); |
b128c09f | 927 | zio_push_transform(zio, wbuf, size, size, NULL); |
34dc7c2f BB |
928 | } |
929 | ||
930 | return (zio); | |
931 | } | |
932 | ||
933 | /* | |
b128c09f | 934 | * Create a child I/O to do some work for us. |
34dc7c2f BB |
935 | */ |
936 | zio_t * | |
b128c09f | 937 | zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset, |
e8b96c60 MA |
938 | void *data, uint64_t size, int type, zio_priority_t priority, |
939 | enum zio_flag flags, zio_done_func_t *done, void *private) | |
34dc7c2f | 940 | { |
428870ff | 941 | enum zio_stage pipeline = ZIO_VDEV_CHILD_PIPELINE; |
b128c09f BB |
942 | zio_t *zio; |
943 | ||
944 | ASSERT(vd->vdev_parent == | |
945 | (pio->io_vd ? pio->io_vd : pio->io_spa->spa_root_vdev)); | |
34dc7c2f BB |
946 | |
947 | if (type == ZIO_TYPE_READ && bp != NULL) { | |
948 | /* | |
949 | * If we have the bp, then the child should perform the | |
950 | * checksum and the parent need not. This pushes error | |
951 | * detection as close to the leaves as possible and | |
952 | * eliminates redundant checksums in the interior nodes. | |
953 | */ | |
428870ff BB |
954 | pipeline |= ZIO_STAGE_CHECKSUM_VERIFY; |
955 | pio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; | |
34dc7c2f BB |
956 | } |
957 | ||
b128c09f BB |
958 | if (vd->vdev_children == 0) |
959 | offset += VDEV_LABEL_START_SIZE; | |
960 | ||
428870ff BB |
961 | flags |= ZIO_VDEV_CHILD_FLAGS(pio) | ZIO_FLAG_DONT_PROPAGATE; |
962 | ||
963 | /* | |
964 | * If we've decided to do a repair, the write is not speculative -- | |
965 | * even if the original read was. | |
966 | */ | |
967 | if (flags & ZIO_FLAG_IO_REPAIR) | |
968 | flags &= ~ZIO_FLAG_SPECULATIVE; | |
969 | ||
b128c09f | 970 | zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size, |
428870ff BB |
971 | done, private, type, priority, flags, vd, offset, &pio->io_bookmark, |
972 | ZIO_STAGE_VDEV_IO_START >> 1, pipeline); | |
34dc7c2f | 973 | |
e8b96c60 MA |
974 | zio->io_physdone = pio->io_physdone; |
975 | if (vd->vdev_ops->vdev_op_leaf && zio->io_logical != NULL) | |
976 | zio->io_logical->io_phys_children++; | |
977 | ||
b128c09f | 978 | return (zio); |
34dc7c2f BB |
979 | } |
980 | ||
b128c09f BB |
981 | zio_t * |
982 | zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, void *data, uint64_t size, | |
e8b96c60 | 983 | int type, zio_priority_t priority, enum zio_flag flags, |
428870ff | 984 | zio_done_func_t *done, void *private) |
34dc7c2f | 985 | { |
b128c09f | 986 | zio_t *zio; |
34dc7c2f | 987 | |
b128c09f | 988 | ASSERT(vd->vdev_ops->vdev_op_leaf); |
34dc7c2f | 989 | |
b128c09f BB |
990 | zio = zio_create(NULL, vd->vdev_spa, 0, NULL, |
991 | data, size, done, private, type, priority, | |
e8b96c60 | 992 | flags | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY | ZIO_FLAG_DELEGATED, |
b128c09f | 993 | vd, offset, NULL, |
428870ff | 994 | ZIO_STAGE_VDEV_IO_START >> 1, ZIO_VDEV_CHILD_PIPELINE); |
34dc7c2f | 995 | |
b128c09f | 996 | return (zio); |
34dc7c2f BB |
997 | } |
998 | ||
999 | void | |
b128c09f | 1000 | zio_flush(zio_t *zio, vdev_t *vd) |
34dc7c2f | 1001 | { |
b128c09f | 1002 | zio_nowait(zio_ioctl(zio, zio->io_spa, vd, DKIOCFLUSHWRITECACHE, |
e8b96c60 | 1003 | NULL, NULL, |
b128c09f | 1004 | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY)); |
34dc7c2f BB |
1005 | } |
1006 | ||
428870ff BB |
1007 | void |
1008 | zio_shrink(zio_t *zio, uint64_t size) | |
1009 | { | |
1010 | ASSERT(zio->io_executor == NULL); | |
1011 | ASSERT(zio->io_orig_size == zio->io_size); | |
1012 | ASSERT(size <= zio->io_size); | |
1013 | ||
1014 | /* | |
1015 | * We don't shrink for raidz because of problems with the | |
1016 | * reconstruction when reading back less than the block size. | |
1017 | * Note, BP_IS_RAIDZ() assumes no compression. | |
1018 | */ | |
1019 | ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); | |
1020 | if (!BP_IS_RAIDZ(zio->io_bp)) | |
1021 | zio->io_orig_size = zio->io_size = size; | |
1022 | } | |
1023 | ||
34dc7c2f BB |
1024 | /* |
1025 | * ========================================================================== | |
b128c09f | 1026 | * Prepare to read and write logical blocks |
34dc7c2f BB |
1027 | * ========================================================================== |
1028 | */ | |
b128c09f | 1029 | |
34dc7c2f | 1030 | static int |
b128c09f | 1031 | zio_read_bp_init(zio_t *zio) |
34dc7c2f | 1032 | { |
b128c09f | 1033 | blkptr_t *bp = zio->io_bp; |
34dc7c2f | 1034 | |
fb5f0bc8 | 1035 | if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF && |
9babb374 BB |
1036 | zio->io_child_type == ZIO_CHILD_LOGICAL && |
1037 | !(zio->io_flags & ZIO_FLAG_RAW)) { | |
9b67f605 MA |
1038 | uint64_t psize = |
1039 | BP_IS_EMBEDDED(bp) ? BPE_GET_PSIZE(bp) : BP_GET_PSIZE(bp); | |
428870ff | 1040 | void *cbuf = zio_buf_alloc(psize); |
b128c09f | 1041 | |
428870ff | 1042 | zio_push_transform(zio, cbuf, psize, psize, zio_decompress); |
34dc7c2f | 1043 | } |
34dc7c2f | 1044 | |
9b67f605 MA |
1045 | if (BP_IS_EMBEDDED(bp) && BPE_GET_ETYPE(bp) == BP_EMBEDDED_TYPE_DATA) { |
1046 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
1047 | decode_embedded_bp_compressed(bp, zio->io_data); | |
1048 | } else { | |
1049 | ASSERT(!BP_IS_EMBEDDED(bp)); | |
1050 | } | |
1051 | ||
9ae529ec | 1052 | if (!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) && BP_GET_LEVEL(bp) == 0) |
b128c09f BB |
1053 | zio->io_flags |= ZIO_FLAG_DONT_CACHE; |
1054 | ||
428870ff BB |
1055 | if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP) |
1056 | zio->io_flags |= ZIO_FLAG_DONT_CACHE; | |
1057 | ||
1058 | if (BP_GET_DEDUP(bp) && zio->io_child_type == ZIO_CHILD_LOGICAL) | |
1059 | zio->io_pipeline = ZIO_DDT_READ_PIPELINE; | |
1060 | ||
b128c09f | 1061 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
1062 | } |
1063 | ||
b128c09f BB |
1064 | static int |
1065 | zio_write_bp_init(zio_t *zio) | |
34dc7c2f | 1066 | { |
428870ff | 1067 | spa_t *spa = zio->io_spa; |
b128c09f | 1068 | zio_prop_t *zp = &zio->io_prop; |
428870ff | 1069 | enum zio_compress compress = zp->zp_compress; |
34dc7c2f | 1070 | blkptr_t *bp = zio->io_bp; |
b128c09f | 1071 | uint64_t lsize = zio->io_size; |
428870ff | 1072 | uint64_t psize = lsize; |
b128c09f | 1073 | int pass = 1; |
34dc7c2f | 1074 | |
b128c09f BB |
1075 | /* |
1076 | * If our children haven't all reached the ready stage, | |
1077 | * wait for them and then repeat this pipeline stage. | |
1078 | */ | |
1079 | if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || | |
1080 | zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_READY)) | |
1081 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 1082 | |
b128c09f BB |
1083 | if (!IO_IS_ALLOCATING(zio)) |
1084 | return (ZIO_PIPELINE_CONTINUE); | |
34dc7c2f | 1085 | |
428870ff BB |
1086 | ASSERT(zio->io_child_type != ZIO_CHILD_DDT); |
1087 | ||
1088 | if (zio->io_bp_override) { | |
1089 | ASSERT(bp->blk_birth != zio->io_txg); | |
1090 | ASSERT(BP_GET_DEDUP(zio->io_bp_override) == 0); | |
1091 | ||
1092 | *bp = *zio->io_bp_override; | |
1093 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
1094 | ||
9b67f605 MA |
1095 | if (BP_IS_EMBEDDED(bp)) |
1096 | return (ZIO_PIPELINE_CONTINUE); | |
1097 | ||
03c6040b GW |
1098 | /* |
1099 | * If we've been overridden and nopwrite is set then | |
1100 | * set the flag accordingly to indicate that a nopwrite | |
1101 | * has already occurred. | |
1102 | */ | |
1103 | if (!BP_IS_HOLE(bp) && zp->zp_nopwrite) { | |
1104 | ASSERT(!zp->zp_dedup); | |
1105 | zio->io_flags |= ZIO_FLAG_NOPWRITE; | |
1106 | return (ZIO_PIPELINE_CONTINUE); | |
1107 | } | |
1108 | ||
1109 | ASSERT(!zp->zp_nopwrite); | |
1110 | ||
428870ff BB |
1111 | if (BP_IS_HOLE(bp) || !zp->zp_dedup) |
1112 | return (ZIO_PIPELINE_CONTINUE); | |
1113 | ||
1114 | ASSERT(zio_checksum_table[zp->zp_checksum].ci_dedup || | |
1115 | zp->zp_dedup_verify); | |
1116 | ||
1117 | if (BP_GET_CHECKSUM(bp) == zp->zp_checksum) { | |
1118 | BP_SET_DEDUP(bp, 1); | |
1119 | zio->io_pipeline |= ZIO_STAGE_DDT_WRITE; | |
1120 | return (ZIO_PIPELINE_CONTINUE); | |
1121 | } | |
1122 | zio->io_bp_override = NULL; | |
1123 | BP_ZERO(bp); | |
1124 | } | |
34dc7c2f | 1125 | |
b0bc7a84 | 1126 | if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg) { |
b128c09f BB |
1127 | /* |
1128 | * We're rewriting an existing block, which means we're | |
1129 | * working on behalf of spa_sync(). For spa_sync() to | |
1130 | * converge, it must eventually be the case that we don't | |
1131 | * have to allocate new blocks. But compression changes | |
1132 | * the blocksize, which forces a reallocate, and makes | |
1133 | * convergence take longer. Therefore, after the first | |
1134 | * few passes, stop compressing to ensure convergence. | |
1135 | */ | |
428870ff BB |
1136 | pass = spa_sync_pass(spa); |
1137 | ||
1138 | ASSERT(zio->io_txg == spa_syncing_txg(spa)); | |
1139 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1140 | ASSERT(!BP_GET_DEDUP(bp)); | |
34dc7c2f | 1141 | |
55d85d5a | 1142 | if (pass >= zfs_sync_pass_dont_compress) |
b128c09f | 1143 | compress = ZIO_COMPRESS_OFF; |
34dc7c2f | 1144 | |
b128c09f | 1145 | /* Make sure someone doesn't change their mind on overwrites */ |
9b67f605 | 1146 | ASSERT(BP_IS_EMBEDDED(bp) || MIN(zp->zp_copies + BP_IS_GANG(bp), |
428870ff | 1147 | spa_max_replication(spa)) == BP_GET_NDVAS(bp)); |
b128c09f | 1148 | } |
34dc7c2f | 1149 | |
b128c09f | 1150 | if (compress != ZIO_COMPRESS_OFF) { |
428870ff BB |
1151 | void *cbuf = zio_buf_alloc(lsize); |
1152 | psize = zio_compress_data(compress, zio->io_data, cbuf, lsize); | |
1153 | if (psize == 0 || psize == lsize) { | |
b128c09f | 1154 | compress = ZIO_COMPRESS_OFF; |
428870ff | 1155 | zio_buf_free(cbuf, lsize); |
9b67f605 MA |
1156 | } else if (!zp->zp_dedup && psize <= BPE_PAYLOAD_SIZE && |
1157 | zp->zp_level == 0 && !DMU_OT_HAS_FILL(zp->zp_type) && | |
1158 | spa_feature_is_enabled(spa, SPA_FEATURE_EMBEDDED_DATA)) { | |
1159 | encode_embedded_bp_compressed(bp, | |
1160 | cbuf, compress, lsize, psize); | |
1161 | BPE_SET_ETYPE(bp, BP_EMBEDDED_TYPE_DATA); | |
1162 | BP_SET_TYPE(bp, zio->io_prop.zp_type); | |
1163 | BP_SET_LEVEL(bp, zio->io_prop.zp_level); | |
1164 | zio_buf_free(cbuf, lsize); | |
1165 | bp->blk_birth = zio->io_txg; | |
1166 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
1167 | ASSERT(spa_feature_is_active(spa, | |
1168 | SPA_FEATURE_EMBEDDED_DATA)); | |
1169 | return (ZIO_PIPELINE_CONTINUE); | |
428870ff | 1170 | } else { |
9b67f605 MA |
1171 | /* |
1172 | * Round up compressed size to MINBLOCKSIZE and | |
1173 | * zero the tail. | |
1174 | */ | |
1175 | size_t rounded = | |
1176 | P2ROUNDUP(psize, (size_t)SPA_MINBLOCKSIZE); | |
1177 | if (rounded > psize) { | |
1178 | bzero((char *)cbuf + psize, rounded - psize); | |
1179 | psize = rounded; | |
1180 | } | |
1181 | if (psize == lsize) { | |
1182 | compress = ZIO_COMPRESS_OFF; | |
1183 | zio_buf_free(cbuf, lsize); | |
1184 | } else { | |
1185 | zio_push_transform(zio, cbuf, | |
1186 | psize, lsize, NULL); | |
1187 | } | |
b128c09f BB |
1188 | } |
1189 | } | |
34dc7c2f | 1190 | |
b128c09f BB |
1191 | /* |
1192 | * The final pass of spa_sync() must be all rewrites, but the first | |
1193 | * few passes offer a trade-off: allocating blocks defers convergence, | |
1194 | * but newly allocated blocks are sequential, so they can be written | |
1195 | * to disk faster. Therefore, we allow the first few passes of | |
1196 | * spa_sync() to allocate new blocks, but force rewrites after that. | |
1197 | * There should only be a handful of blocks after pass 1 in any case. | |
1198 | */ | |
b0bc7a84 MG |
1199 | if (!BP_IS_HOLE(bp) && bp->blk_birth == zio->io_txg && |
1200 | BP_GET_PSIZE(bp) == psize && | |
55d85d5a | 1201 | pass >= zfs_sync_pass_rewrite) { |
428870ff | 1202 | enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES; |
d6320ddb | 1203 | ASSERT(psize != 0); |
b128c09f BB |
1204 | zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages; |
1205 | zio->io_flags |= ZIO_FLAG_IO_REWRITE; | |
1206 | } else { | |
1207 | BP_ZERO(bp); | |
1208 | zio->io_pipeline = ZIO_WRITE_PIPELINE; | |
1209 | } | |
34dc7c2f | 1210 | |
428870ff | 1211 | if (psize == 0) { |
b0bc7a84 MG |
1212 | if (zio->io_bp_orig.blk_birth != 0 && |
1213 | spa_feature_is_active(spa, SPA_FEATURE_HOLE_BIRTH)) { | |
1214 | BP_SET_LSIZE(bp, lsize); | |
1215 | BP_SET_TYPE(bp, zp->zp_type); | |
1216 | BP_SET_LEVEL(bp, zp->zp_level); | |
1217 | BP_SET_BIRTH(bp, zio->io_txg, 0); | |
1218 | } | |
b128c09f BB |
1219 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; |
1220 | } else { | |
1221 | ASSERT(zp->zp_checksum != ZIO_CHECKSUM_GANG_HEADER); | |
1222 | BP_SET_LSIZE(bp, lsize); | |
b0bc7a84 MG |
1223 | BP_SET_TYPE(bp, zp->zp_type); |
1224 | BP_SET_LEVEL(bp, zp->zp_level); | |
428870ff | 1225 | BP_SET_PSIZE(bp, psize); |
b128c09f BB |
1226 | BP_SET_COMPRESS(bp, compress); |
1227 | BP_SET_CHECKSUM(bp, zp->zp_checksum); | |
428870ff | 1228 | BP_SET_DEDUP(bp, zp->zp_dedup); |
b128c09f | 1229 | BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); |
428870ff BB |
1230 | if (zp->zp_dedup) { |
1231 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1232 | ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); | |
1233 | zio->io_pipeline = ZIO_DDT_WRITE_PIPELINE; | |
1234 | } | |
03c6040b GW |
1235 | if (zp->zp_nopwrite) { |
1236 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1237 | ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); | |
1238 | zio->io_pipeline |= ZIO_STAGE_NOP_WRITE; | |
1239 | } | |
428870ff BB |
1240 | } |
1241 | ||
1242 | return (ZIO_PIPELINE_CONTINUE); | |
1243 | } | |
1244 | ||
1245 | static int | |
1246 | zio_free_bp_init(zio_t *zio) | |
1247 | { | |
1248 | blkptr_t *bp = zio->io_bp; | |
1249 | ||
1250 | if (zio->io_child_type == ZIO_CHILD_LOGICAL) { | |
1251 | if (BP_GET_DEDUP(bp)) | |
1252 | zio->io_pipeline = ZIO_DDT_FREE_PIPELINE; | |
b128c09f | 1253 | } |
34dc7c2f BB |
1254 | |
1255 | return (ZIO_PIPELINE_CONTINUE); | |
1256 | } | |
1257 | ||
b128c09f BB |
1258 | /* |
1259 | * ========================================================================== | |
1260 | * Execute the I/O pipeline | |
1261 | * ========================================================================== | |
1262 | */ | |
1263 | ||
1264 | static void | |
7ef5e54e | 1265 | zio_taskq_dispatch(zio_t *zio, zio_taskq_type_t q, boolean_t cutinline) |
34dc7c2f | 1266 | { |
428870ff | 1267 | spa_t *spa = zio->io_spa; |
b128c09f | 1268 | zio_type_t t = zio->io_type; |
a38718a6 | 1269 | int flags = (cutinline ? TQ_FRONT : 0); |
34dc7c2f BB |
1270 | |
1271 | /* | |
9babb374 BB |
1272 | * If we're a config writer or a probe, the normal issue and |
1273 | * interrupt threads may all be blocked waiting for the config lock. | |
1274 | * In this case, select the otherwise-unused taskq for ZIO_TYPE_NULL. | |
34dc7c2f | 1275 | */ |
9babb374 | 1276 | if (zio->io_flags & (ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_PROBE)) |
b128c09f | 1277 | t = ZIO_TYPE_NULL; |
34dc7c2f BB |
1278 | |
1279 | /* | |
b128c09f | 1280 | * A similar issue exists for the L2ARC write thread until L2ARC 2.0. |
34dc7c2f | 1281 | */ |
b128c09f BB |
1282 | if (t == ZIO_TYPE_WRITE && zio->io_vd && zio->io_vd->vdev_aux) |
1283 | t = ZIO_TYPE_NULL; | |
34dc7c2f | 1284 | |
428870ff | 1285 | /* |
7ef5e54e AL |
1286 | * If this is a high priority I/O, then use the high priority taskq if |
1287 | * available. | |
428870ff BB |
1288 | */ |
1289 | if (zio->io_priority == ZIO_PRIORITY_NOW && | |
7ef5e54e | 1290 | spa->spa_zio_taskq[t][q + 1].stqs_count != 0) |
428870ff BB |
1291 | q++; |
1292 | ||
1293 | ASSERT3U(q, <, ZIO_TASKQ_TYPES); | |
5cc556b4 | 1294 | |
a38718a6 GA |
1295 | /* |
1296 | * NB: We are assuming that the zio can only be dispatched | |
1297 | * to a single taskq at a time. It would be a grievous error | |
1298 | * to dispatch the zio to another taskq at the same time. | |
1299 | */ | |
1300 | ASSERT(taskq_empty_ent(&zio->io_tqent)); | |
7ef5e54e AL |
1301 | spa_taskq_dispatch_ent(spa, t, q, (task_func_t *)zio_execute, zio, |
1302 | flags, &zio->io_tqent); | |
b128c09f | 1303 | } |
34dc7c2f | 1304 | |
b128c09f | 1305 | static boolean_t |
7ef5e54e | 1306 | zio_taskq_member(zio_t *zio, zio_taskq_type_t q) |
b128c09f BB |
1307 | { |
1308 | kthread_t *executor = zio->io_executor; | |
1309 | spa_t *spa = zio->io_spa; | |
d6320ddb | 1310 | zio_type_t t; |
34dc7c2f | 1311 | |
7ef5e54e AL |
1312 | for (t = 0; t < ZIO_TYPES; t++) { |
1313 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
1314 | uint_t i; | |
1315 | for (i = 0; i < tqs->stqs_count; i++) { | |
1316 | if (taskq_member(tqs->stqs_taskq[i], executor)) | |
1317 | return (B_TRUE); | |
1318 | } | |
1319 | } | |
34dc7c2f | 1320 | |
b128c09f BB |
1321 | return (B_FALSE); |
1322 | } | |
34dc7c2f | 1323 | |
b128c09f BB |
1324 | static int |
1325 | zio_issue_async(zio_t *zio) | |
1326 | { | |
428870ff | 1327 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); |
b128c09f BB |
1328 | |
1329 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f BB |
1330 | } |
1331 | ||
b128c09f BB |
1332 | void |
1333 | zio_interrupt(zio_t *zio) | |
34dc7c2f | 1334 | { |
428870ff | 1335 | zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT, B_FALSE); |
b128c09f | 1336 | } |
34dc7c2f | 1337 | |
b128c09f BB |
1338 | /* |
1339 | * Execute the I/O pipeline until one of the following occurs: | |
1340 | * (1) the I/O completes; (2) the pipeline stalls waiting for | |
1341 | * dependent child I/Os; (3) the I/O issues, so we're waiting | |
1342 | * for an I/O completion interrupt; (4) the I/O is delegated by | |
1343 | * vdev-level caching or aggregation; (5) the I/O is deferred | |
1344 | * due to vdev-level queueing; (6) the I/O is handed off to | |
1345 | * another thread. In all cases, the pipeline stops whenever | |
8e07b99b | 1346 | * there's no CPU work; it never burns a thread in cv_wait_io(). |
b128c09f BB |
1347 | * |
1348 | * There's no locking on io_stage because there's no legitimate way | |
1349 | * for multiple threads to be attempting to process the same I/O. | |
1350 | */ | |
428870ff | 1351 | static zio_pipe_stage_t *zio_pipeline[]; |
34dc7c2f | 1352 | |
da6b4005 NB |
1353 | /* |
1354 | * zio_execute() is a wrapper around the static function | |
1355 | * __zio_execute() so that we can force __zio_execute() to be | |
1356 | * inlined. This reduces stack overhead which is important | |
1357 | * because __zio_execute() is called recursively in several zio | |
1358 | * code paths. zio_execute() itself cannot be inlined because | |
1359 | * it is externally visible. | |
1360 | */ | |
b128c09f BB |
1361 | void |
1362 | zio_execute(zio_t *zio) | |
da6b4005 | 1363 | { |
92119cc2 BB |
1364 | fstrans_cookie_t cookie; |
1365 | ||
1366 | cookie = spl_fstrans_mark(); | |
da6b4005 | 1367 | __zio_execute(zio); |
92119cc2 | 1368 | spl_fstrans_unmark(cookie); |
da6b4005 NB |
1369 | } |
1370 | ||
1371 | __attribute__((always_inline)) | |
1372 | static inline void | |
1373 | __zio_execute(zio_t *zio) | |
b128c09f BB |
1374 | { |
1375 | zio->io_executor = curthread; | |
34dc7c2f | 1376 | |
b128c09f | 1377 | while (zio->io_stage < ZIO_STAGE_DONE) { |
428870ff BB |
1378 | enum zio_stage pipeline = zio->io_pipeline; |
1379 | enum zio_stage stage = zio->io_stage; | |
91579709 | 1380 | dsl_pool_t *dp; |
2fac4c2a | 1381 | boolean_t cut; |
b128c09f | 1382 | int rv; |
34dc7c2f | 1383 | |
b128c09f | 1384 | ASSERT(!MUTEX_HELD(&zio->io_lock)); |
428870ff BB |
1385 | ASSERT(ISP2(stage)); |
1386 | ASSERT(zio->io_stall == NULL); | |
34dc7c2f | 1387 | |
428870ff BB |
1388 | do { |
1389 | stage <<= 1; | |
1390 | } while ((stage & pipeline) == 0); | |
b128c09f BB |
1391 | |
1392 | ASSERT(stage <= ZIO_STAGE_DONE); | |
34dc7c2f | 1393 | |
91579709 | 1394 | dp = spa_get_dsl(zio->io_spa); |
2fac4c2a BB |
1395 | cut = (stage == ZIO_STAGE_VDEV_IO_START) ? |
1396 | zio_requeue_io_start_cut_in_line : B_FALSE; | |
1397 | ||
34dc7c2f | 1398 | /* |
b128c09f BB |
1399 | * If we are in interrupt context and this pipeline stage |
1400 | * will grab a config lock that is held across I/O, | |
428870ff BB |
1401 | * or may wait for an I/O that needs an interrupt thread |
1402 | * to complete, issue async to avoid deadlock. | |
1403 | * | |
1404 | * For VDEV_IO_START, we cut in line so that the io will | |
1405 | * be sent to disk promptly. | |
34dc7c2f | 1406 | */ |
91579709 BB |
1407 | if ((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL && |
1408 | zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) { | |
1409 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); | |
1410 | return; | |
1411 | } | |
1412 | ||
1413 | /* | |
1414 | * If we executing in the context of the tx_sync_thread, | |
1415 | * or we are performing pool initialization outside of a | |
34e14332 BB |
1416 | * zio_taskq[ZIO_TASKQ_ISSUE|ZIO_TASKQ_ISSUE_HIGH] context. |
1417 | * Then issue the zio asynchronously to minimize stack usage | |
1418 | * for these deep call paths. | |
91579709 BB |
1419 | */ |
1420 | if ((dp && curthread == dp->dp_tx.tx_sync_thread) || | |
1421 | (dp && spa_is_initializing(dp->dp_spa) && | |
34e14332 BB |
1422 | !zio_taskq_member(zio, ZIO_TASKQ_ISSUE) && |
1423 | !zio_taskq_member(zio, ZIO_TASKQ_ISSUE_HIGH))) { | |
428870ff | 1424 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); |
b128c09f | 1425 | return; |
34dc7c2f BB |
1426 | } |
1427 | ||
b128c09f | 1428 | zio->io_stage = stage; |
9bd274dd | 1429 | rv = zio_pipeline[highbit64(stage) - 1](zio); |
34dc7c2f | 1430 | |
b128c09f BB |
1431 | if (rv == ZIO_PIPELINE_STOP) |
1432 | return; | |
34dc7c2f | 1433 | |
b128c09f BB |
1434 | ASSERT(rv == ZIO_PIPELINE_CONTINUE); |
1435 | } | |
34dc7c2f BB |
1436 | } |
1437 | ||
da6b4005 | 1438 | |
b128c09f BB |
1439 | /* |
1440 | * ========================================================================== | |
1441 | * Initiate I/O, either sync or async | |
1442 | * ========================================================================== | |
1443 | */ | |
1444 | int | |
1445 | zio_wait(zio_t *zio) | |
34dc7c2f | 1446 | { |
b128c09f | 1447 | int error; |
34dc7c2f | 1448 | |
b128c09f BB |
1449 | ASSERT(zio->io_stage == ZIO_STAGE_OPEN); |
1450 | ASSERT(zio->io_executor == NULL); | |
34dc7c2f | 1451 | |
b128c09f | 1452 | zio->io_waiter = curthread; |
34dc7c2f | 1453 | |
da6b4005 | 1454 | __zio_execute(zio); |
34dc7c2f | 1455 | |
b128c09f | 1456 | mutex_enter(&zio->io_lock); |
72f53c56 | 1457 | while (zio->io_executor != NULL) |
72938d69 | 1458 | cv_wait_io(&zio->io_cv, &zio->io_lock); |
b128c09f | 1459 | mutex_exit(&zio->io_lock); |
34dc7c2f | 1460 | |
b128c09f BB |
1461 | error = zio->io_error; |
1462 | zio_destroy(zio); | |
34dc7c2f | 1463 | |
b128c09f BB |
1464 | return (error); |
1465 | } | |
34dc7c2f | 1466 | |
b128c09f BB |
1467 | void |
1468 | zio_nowait(zio_t *zio) | |
1469 | { | |
1470 | ASSERT(zio->io_executor == NULL); | |
34dc7c2f | 1471 | |
d164b209 BB |
1472 | if (zio->io_child_type == ZIO_CHILD_LOGICAL && |
1473 | zio_unique_parent(zio) == NULL) { | |
8878261f BB |
1474 | zio_t *pio; |
1475 | ||
34dc7c2f | 1476 | /* |
b128c09f | 1477 | * This is a logical async I/O with no parent to wait for it. |
9babb374 BB |
1478 | * We add it to the spa_async_root_zio "Godfather" I/O which |
1479 | * will ensure they complete prior to unloading the pool. | |
34dc7c2f | 1480 | */ |
b128c09f | 1481 | spa_t *spa = zio->io_spa; |
8878261f BB |
1482 | kpreempt_disable(); |
1483 | pio = spa->spa_async_zio_root[CPU_SEQID]; | |
1484 | kpreempt_enable(); | |
9babb374 | 1485 | |
8878261f | 1486 | zio_add_child(pio, zio); |
b128c09f | 1487 | } |
34dc7c2f | 1488 | |
da6b4005 | 1489 | __zio_execute(zio); |
b128c09f | 1490 | } |
34dc7c2f | 1491 | |
b128c09f BB |
1492 | /* |
1493 | * ========================================================================== | |
1494 | * Reexecute or suspend/resume failed I/O | |
1495 | * ========================================================================== | |
1496 | */ | |
34dc7c2f | 1497 | |
b128c09f BB |
1498 | static void |
1499 | zio_reexecute(zio_t *pio) | |
1500 | { | |
d164b209 | 1501 | zio_t *cio, *cio_next; |
d6320ddb | 1502 | int c, w; |
d164b209 BB |
1503 | |
1504 | ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL); | |
1505 | ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN); | |
9babb374 BB |
1506 | ASSERT(pio->io_gang_leader == NULL); |
1507 | ASSERT(pio->io_gang_tree == NULL); | |
34dc7c2f | 1508 | |
b128c09f BB |
1509 | pio->io_flags = pio->io_orig_flags; |
1510 | pio->io_stage = pio->io_orig_stage; | |
1511 | pio->io_pipeline = pio->io_orig_pipeline; | |
1512 | pio->io_reexecute = 0; | |
03c6040b | 1513 | pio->io_flags |= ZIO_FLAG_REEXECUTED; |
b128c09f | 1514 | pio->io_error = 0; |
d6320ddb | 1515 | for (w = 0; w < ZIO_WAIT_TYPES; w++) |
d164b209 | 1516 | pio->io_state[w] = 0; |
d6320ddb | 1517 | for (c = 0; c < ZIO_CHILD_TYPES; c++) |
b128c09f | 1518 | pio->io_child_error[c] = 0; |
34dc7c2f | 1519 | |
428870ff BB |
1520 | if (IO_IS_ALLOCATING(pio)) |
1521 | BP_ZERO(pio->io_bp); | |
34dc7c2f | 1522 | |
b128c09f BB |
1523 | /* |
1524 | * As we reexecute pio's children, new children could be created. | |
d164b209 | 1525 | * New children go to the head of pio's io_child_list, however, |
b128c09f | 1526 | * so we will (correctly) not reexecute them. The key is that |
d164b209 BB |
1527 | * the remainder of pio's io_child_list, from 'cio_next' onward, |
1528 | * cannot be affected by any side effects of reexecuting 'cio'. | |
b128c09f | 1529 | */ |
d164b209 BB |
1530 | for (cio = zio_walk_children(pio); cio != NULL; cio = cio_next) { |
1531 | cio_next = zio_walk_children(pio); | |
b128c09f | 1532 | mutex_enter(&pio->io_lock); |
d6320ddb | 1533 | for (w = 0; w < ZIO_WAIT_TYPES; w++) |
d164b209 | 1534 | pio->io_children[cio->io_child_type][w]++; |
b128c09f | 1535 | mutex_exit(&pio->io_lock); |
d164b209 | 1536 | zio_reexecute(cio); |
34dc7c2f | 1537 | } |
34dc7c2f | 1538 | |
b128c09f BB |
1539 | /* |
1540 | * Now that all children have been reexecuted, execute the parent. | |
9babb374 BB |
1541 | * We don't reexecute "The Godfather" I/O here as it's the |
1542 | * responsibility of the caller to wait on him. | |
b128c09f | 1543 | */ |
9babb374 | 1544 | if (!(pio->io_flags & ZIO_FLAG_GODFATHER)) |
da6b4005 | 1545 | __zio_execute(pio); |
34dc7c2f BB |
1546 | } |
1547 | ||
b128c09f BB |
1548 | void |
1549 | zio_suspend(spa_t *spa, zio_t *zio) | |
34dc7c2f | 1550 | { |
b128c09f BB |
1551 | if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC) |
1552 | fm_panic("Pool '%s' has encountered an uncorrectable I/O " | |
1553 | "failure and the failure mode property for this pool " | |
1554 | "is set to panic.", spa_name(spa)); | |
34dc7c2f | 1555 | |
bf89c199 BB |
1556 | cmn_err(CE_WARN, "Pool '%s' has encountered an uncorrectable I/O " |
1557 | "failure and has been suspended.\n", spa_name(spa)); | |
1558 | ||
b128c09f | 1559 | zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, 0, 0); |
34dc7c2f | 1560 | |
b128c09f | 1561 | mutex_enter(&spa->spa_suspend_lock); |
34dc7c2f | 1562 | |
b128c09f | 1563 | if (spa->spa_suspend_zio_root == NULL) |
9babb374 BB |
1564 | spa->spa_suspend_zio_root = zio_root(spa, NULL, NULL, |
1565 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
1566 | ZIO_FLAG_GODFATHER); | |
34dc7c2f | 1567 | |
b128c09f | 1568 | spa->spa_suspended = B_TRUE; |
34dc7c2f | 1569 | |
b128c09f | 1570 | if (zio != NULL) { |
9babb374 | 1571 | ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); |
b128c09f BB |
1572 | ASSERT(zio != spa->spa_suspend_zio_root); |
1573 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
d164b209 | 1574 | ASSERT(zio_unique_parent(zio) == NULL); |
b128c09f BB |
1575 | ASSERT(zio->io_stage == ZIO_STAGE_DONE); |
1576 | zio_add_child(spa->spa_suspend_zio_root, zio); | |
1577 | } | |
34dc7c2f | 1578 | |
b128c09f BB |
1579 | mutex_exit(&spa->spa_suspend_lock); |
1580 | } | |
34dc7c2f | 1581 | |
9babb374 | 1582 | int |
b128c09f BB |
1583 | zio_resume(spa_t *spa) |
1584 | { | |
9babb374 | 1585 | zio_t *pio; |
34dc7c2f BB |
1586 | |
1587 | /* | |
b128c09f | 1588 | * Reexecute all previously suspended i/o. |
34dc7c2f | 1589 | */ |
b128c09f BB |
1590 | mutex_enter(&spa->spa_suspend_lock); |
1591 | spa->spa_suspended = B_FALSE; | |
1592 | cv_broadcast(&spa->spa_suspend_cv); | |
1593 | pio = spa->spa_suspend_zio_root; | |
1594 | spa->spa_suspend_zio_root = NULL; | |
1595 | mutex_exit(&spa->spa_suspend_lock); | |
1596 | ||
1597 | if (pio == NULL) | |
9babb374 | 1598 | return (0); |
34dc7c2f | 1599 | |
9babb374 BB |
1600 | zio_reexecute(pio); |
1601 | return (zio_wait(pio)); | |
b128c09f BB |
1602 | } |
1603 | ||
1604 | void | |
1605 | zio_resume_wait(spa_t *spa) | |
1606 | { | |
1607 | mutex_enter(&spa->spa_suspend_lock); | |
1608 | while (spa_suspended(spa)) | |
1609 | cv_wait(&spa->spa_suspend_cv, &spa->spa_suspend_lock); | |
1610 | mutex_exit(&spa->spa_suspend_lock); | |
34dc7c2f BB |
1611 | } |
1612 | ||
1613 | /* | |
1614 | * ========================================================================== | |
b128c09f BB |
1615 | * Gang blocks. |
1616 | * | |
1617 | * A gang block is a collection of small blocks that looks to the DMU | |
1618 | * like one large block. When zio_dva_allocate() cannot find a block | |
1619 | * of the requested size, due to either severe fragmentation or the pool | |
1620 | * being nearly full, it calls zio_write_gang_block() to construct the | |
1621 | * block from smaller fragments. | |
1622 | * | |
1623 | * A gang block consists of a gang header (zio_gbh_phys_t) and up to | |
1624 | * three (SPA_GBH_NBLKPTRS) gang members. The gang header is just like | |
1625 | * an indirect block: it's an array of block pointers. It consumes | |
1626 | * only one sector and hence is allocatable regardless of fragmentation. | |
1627 | * The gang header's bps point to its gang members, which hold the data. | |
1628 | * | |
1629 | * Gang blocks are self-checksumming, using the bp's <vdev, offset, txg> | |
1630 | * as the verifier to ensure uniqueness of the SHA256 checksum. | |
1631 | * Critically, the gang block bp's blk_cksum is the checksum of the data, | |
1632 | * not the gang header. This ensures that data block signatures (needed for | |
1633 | * deduplication) are independent of how the block is physically stored. | |
1634 | * | |
1635 | * Gang blocks can be nested: a gang member may itself be a gang block. | |
1636 | * Thus every gang block is a tree in which root and all interior nodes are | |
1637 | * gang headers, and the leaves are normal blocks that contain user data. | |
1638 | * The root of the gang tree is called the gang leader. | |
1639 | * | |
1640 | * To perform any operation (read, rewrite, free, claim) on a gang block, | |
1641 | * zio_gang_assemble() first assembles the gang tree (minus data leaves) | |
1642 | * in the io_gang_tree field of the original logical i/o by recursively | |
1643 | * reading the gang leader and all gang headers below it. This yields | |
1644 | * an in-core tree containing the contents of every gang header and the | |
1645 | * bps for every constituent of the gang block. | |
1646 | * | |
1647 | * With the gang tree now assembled, zio_gang_issue() just walks the gang tree | |
1648 | * and invokes a callback on each bp. To free a gang block, zio_gang_issue() | |
1649 | * calls zio_free_gang() -- a trivial wrapper around zio_free() -- for each bp. | |
1650 | * zio_claim_gang() provides a similarly trivial wrapper for zio_claim(). | |
1651 | * zio_read_gang() is a wrapper around zio_read() that omits reading gang | |
1652 | * headers, since we already have those in io_gang_tree. zio_rewrite_gang() | |
1653 | * performs a zio_rewrite() of the data or, for gang headers, a zio_rewrite() | |
1654 | * of the gang header plus zio_checksum_compute() of the data to update the | |
1655 | * gang header's blk_cksum as described above. | |
1656 | * | |
1657 | * The two-phase assemble/issue model solves the problem of partial failure -- | |
1658 | * what if you'd freed part of a gang block but then couldn't read the | |
1659 | * gang header for another part? Assembling the entire gang tree first | |
1660 | * ensures that all the necessary gang header I/O has succeeded before | |
1661 | * starting the actual work of free, claim, or write. Once the gang tree | |
1662 | * is assembled, free and claim are in-memory operations that cannot fail. | |
1663 | * | |
1664 | * In the event that a gang write fails, zio_dva_unallocate() walks the | |
1665 | * gang tree to immediately free (i.e. insert back into the space map) | |
1666 | * everything we've allocated. This ensures that we don't get ENOSPC | |
1667 | * errors during repeated suspend/resume cycles due to a flaky device. | |
1668 | * | |
1669 | * Gang rewrites only happen during sync-to-convergence. If we can't assemble | |
1670 | * the gang tree, we won't modify the block, so we can safely defer the free | |
1671 | * (knowing that the block is still intact). If we *can* assemble the gang | |
1672 | * tree, then even if some of the rewrites fail, zio_dva_unallocate() will free | |
1673 | * each constituent bp and we can allocate a new block on the next sync pass. | |
1674 | * | |
1675 | * In all cases, the gang tree allows complete recovery from partial failure. | |
34dc7c2f BB |
1676 | * ========================================================================== |
1677 | */ | |
b128c09f BB |
1678 | |
1679 | static zio_t * | |
1680 | zio_read_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
34dc7c2f | 1681 | { |
b128c09f BB |
1682 | if (gn != NULL) |
1683 | return (pio); | |
34dc7c2f | 1684 | |
b128c09f BB |
1685 | return (zio_read(pio, pio->io_spa, bp, data, BP_GET_PSIZE(bp), |
1686 | NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), | |
1687 | &pio->io_bookmark)); | |
1688 | } | |
1689 | ||
1690 | zio_t * | |
1691 | zio_rewrite_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
1692 | { | |
1693 | zio_t *zio; | |
1694 | ||
1695 | if (gn != NULL) { | |
1696 | zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, | |
1697 | gn->gn_gbh, SPA_GANGBLOCKSIZE, NULL, NULL, pio->io_priority, | |
1698 | ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); | |
34dc7c2f | 1699 | /* |
b128c09f BB |
1700 | * As we rewrite each gang header, the pipeline will compute |
1701 | * a new gang block header checksum for it; but no one will | |
1702 | * compute a new data checksum, so we do that here. The one | |
1703 | * exception is the gang leader: the pipeline already computed | |
1704 | * its data checksum because that stage precedes gang assembly. | |
1705 | * (Presently, nothing actually uses interior data checksums; | |
1706 | * this is just good hygiene.) | |
34dc7c2f | 1707 | */ |
9babb374 | 1708 | if (gn != pio->io_gang_leader->io_gang_tree) { |
b128c09f BB |
1709 | zio_checksum_compute(zio, BP_GET_CHECKSUM(bp), |
1710 | data, BP_GET_PSIZE(bp)); | |
1711 | } | |
428870ff BB |
1712 | /* |
1713 | * If we are here to damage data for testing purposes, | |
1714 | * leave the GBH alone so that we can detect the damage. | |
1715 | */ | |
1716 | if (pio->io_gang_leader->io_flags & ZIO_FLAG_INDUCE_DAMAGE) | |
1717 | zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; | |
34dc7c2f | 1718 | } else { |
b128c09f BB |
1719 | zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, |
1720 | data, BP_GET_PSIZE(bp), NULL, NULL, pio->io_priority, | |
1721 | ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); | |
34dc7c2f BB |
1722 | } |
1723 | ||
b128c09f BB |
1724 | return (zio); |
1725 | } | |
34dc7c2f | 1726 | |
b128c09f BB |
1727 | /* ARGSUSED */ |
1728 | zio_t * | |
1729 | zio_free_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
1730 | { | |
428870ff BB |
1731 | return (zio_free_sync(pio, pio->io_spa, pio->io_txg, bp, |
1732 | ZIO_GANG_CHILD_FLAGS(pio))); | |
34dc7c2f BB |
1733 | } |
1734 | ||
b128c09f BB |
1735 | /* ARGSUSED */ |
1736 | zio_t * | |
1737 | zio_claim_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
34dc7c2f | 1738 | { |
b128c09f BB |
1739 | return (zio_claim(pio, pio->io_spa, pio->io_txg, bp, |
1740 | NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio))); | |
1741 | } | |
1742 | ||
1743 | static zio_gang_issue_func_t *zio_gang_issue_func[ZIO_TYPES] = { | |
1744 | NULL, | |
1745 | zio_read_gang, | |
1746 | zio_rewrite_gang, | |
1747 | zio_free_gang, | |
1748 | zio_claim_gang, | |
1749 | NULL | |
1750 | }; | |
34dc7c2f | 1751 | |
b128c09f | 1752 | static void zio_gang_tree_assemble_done(zio_t *zio); |
34dc7c2f | 1753 | |
b128c09f BB |
1754 | static zio_gang_node_t * |
1755 | zio_gang_node_alloc(zio_gang_node_t **gnpp) | |
1756 | { | |
1757 | zio_gang_node_t *gn; | |
34dc7c2f | 1758 | |
b128c09f | 1759 | ASSERT(*gnpp == NULL); |
34dc7c2f | 1760 | |
691f6ac4 | 1761 | gn = kmem_zalloc(sizeof (*gn), KM_PUSHPAGE); |
b128c09f BB |
1762 | gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE); |
1763 | *gnpp = gn; | |
34dc7c2f | 1764 | |
b128c09f | 1765 | return (gn); |
34dc7c2f BB |
1766 | } |
1767 | ||
34dc7c2f | 1768 | static void |
b128c09f | 1769 | zio_gang_node_free(zio_gang_node_t **gnpp) |
34dc7c2f | 1770 | { |
b128c09f | 1771 | zio_gang_node_t *gn = *gnpp; |
d6320ddb | 1772 | int g; |
34dc7c2f | 1773 | |
d6320ddb | 1774 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) |
b128c09f BB |
1775 | ASSERT(gn->gn_child[g] == NULL); |
1776 | ||
1777 | zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE); | |
1778 | kmem_free(gn, sizeof (*gn)); | |
1779 | *gnpp = NULL; | |
34dc7c2f BB |
1780 | } |
1781 | ||
b128c09f BB |
1782 | static void |
1783 | zio_gang_tree_free(zio_gang_node_t **gnpp) | |
34dc7c2f | 1784 | { |
b128c09f | 1785 | zio_gang_node_t *gn = *gnpp; |
d6320ddb | 1786 | int g; |
34dc7c2f | 1787 | |
b128c09f BB |
1788 | if (gn == NULL) |
1789 | return; | |
34dc7c2f | 1790 | |
d6320ddb | 1791 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) |
b128c09f | 1792 | zio_gang_tree_free(&gn->gn_child[g]); |
34dc7c2f | 1793 | |
b128c09f | 1794 | zio_gang_node_free(gnpp); |
34dc7c2f BB |
1795 | } |
1796 | ||
b128c09f | 1797 | static void |
9babb374 | 1798 | zio_gang_tree_assemble(zio_t *gio, blkptr_t *bp, zio_gang_node_t **gnpp) |
34dc7c2f | 1799 | { |
b128c09f BB |
1800 | zio_gang_node_t *gn = zio_gang_node_alloc(gnpp); |
1801 | ||
9babb374 | 1802 | ASSERT(gio->io_gang_leader == gio); |
b128c09f | 1803 | ASSERT(BP_IS_GANG(bp)); |
34dc7c2f | 1804 | |
9babb374 | 1805 | zio_nowait(zio_read(gio, gio->io_spa, bp, gn->gn_gbh, |
b128c09f | 1806 | SPA_GANGBLOCKSIZE, zio_gang_tree_assemble_done, gn, |
9babb374 | 1807 | gio->io_priority, ZIO_GANG_CHILD_FLAGS(gio), &gio->io_bookmark)); |
b128c09f | 1808 | } |
34dc7c2f | 1809 | |
b128c09f BB |
1810 | static void |
1811 | zio_gang_tree_assemble_done(zio_t *zio) | |
1812 | { | |
9babb374 | 1813 | zio_t *gio = zio->io_gang_leader; |
b128c09f BB |
1814 | zio_gang_node_t *gn = zio->io_private; |
1815 | blkptr_t *bp = zio->io_bp; | |
d6320ddb | 1816 | int g; |
34dc7c2f | 1817 | |
9babb374 | 1818 | ASSERT(gio == zio_unique_parent(zio)); |
428870ff | 1819 | ASSERT(zio->io_child_count == 0); |
34dc7c2f | 1820 | |
b128c09f BB |
1821 | if (zio->io_error) |
1822 | return; | |
34dc7c2f | 1823 | |
b128c09f BB |
1824 | if (BP_SHOULD_BYTESWAP(bp)) |
1825 | byteswap_uint64_array(zio->io_data, zio->io_size); | |
34dc7c2f | 1826 | |
b128c09f BB |
1827 | ASSERT(zio->io_data == gn->gn_gbh); |
1828 | ASSERT(zio->io_size == SPA_GANGBLOCKSIZE); | |
428870ff | 1829 | ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); |
34dc7c2f | 1830 | |
d6320ddb | 1831 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) { |
b128c09f BB |
1832 | blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; |
1833 | if (!BP_IS_GANG(gbp)) | |
1834 | continue; | |
9babb374 | 1835 | zio_gang_tree_assemble(gio, gbp, &gn->gn_child[g]); |
b128c09f | 1836 | } |
34dc7c2f BB |
1837 | } |
1838 | ||
b128c09f BB |
1839 | static void |
1840 | zio_gang_tree_issue(zio_t *pio, zio_gang_node_t *gn, blkptr_t *bp, void *data) | |
34dc7c2f | 1841 | { |
9babb374 | 1842 | zio_t *gio = pio->io_gang_leader; |
b128c09f | 1843 | zio_t *zio; |
d6320ddb | 1844 | int g; |
34dc7c2f | 1845 | |
b128c09f | 1846 | ASSERT(BP_IS_GANG(bp) == !!gn); |
9babb374 BB |
1847 | ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp)); |
1848 | ASSERT(BP_GET_LSIZE(bp) == BP_GET_PSIZE(bp) || gn == gio->io_gang_tree); | |
34dc7c2f | 1849 | |
b128c09f BB |
1850 | /* |
1851 | * If you're a gang header, your data is in gn->gn_gbh. | |
1852 | * If you're a gang member, your data is in 'data' and gn == NULL. | |
1853 | */ | |
9babb374 | 1854 | zio = zio_gang_issue_func[gio->io_type](pio, bp, gn, data); |
34dc7c2f | 1855 | |
b128c09f | 1856 | if (gn != NULL) { |
428870ff | 1857 | ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); |
34dc7c2f | 1858 | |
d6320ddb | 1859 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) { |
b128c09f BB |
1860 | blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; |
1861 | if (BP_IS_HOLE(gbp)) | |
1862 | continue; | |
1863 | zio_gang_tree_issue(zio, gn->gn_child[g], gbp, data); | |
1864 | data = (char *)data + BP_GET_PSIZE(gbp); | |
1865 | } | |
34dc7c2f BB |
1866 | } |
1867 | ||
9babb374 BB |
1868 | if (gn == gio->io_gang_tree) |
1869 | ASSERT3P((char *)gio->io_data + gio->io_size, ==, data); | |
34dc7c2f | 1870 | |
b128c09f BB |
1871 | if (zio != pio) |
1872 | zio_nowait(zio); | |
34dc7c2f BB |
1873 | } |
1874 | ||
1875 | static int | |
b128c09f | 1876 | zio_gang_assemble(zio_t *zio) |
34dc7c2f | 1877 | { |
b128c09f | 1878 | blkptr_t *bp = zio->io_bp; |
34dc7c2f | 1879 | |
9babb374 BB |
1880 | ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == NULL); |
1881 | ASSERT(zio->io_child_type > ZIO_CHILD_GANG); | |
1882 | ||
1883 | zio->io_gang_leader = zio; | |
34dc7c2f | 1884 | |
b128c09f | 1885 | zio_gang_tree_assemble(zio, bp, &zio->io_gang_tree); |
34dc7c2f BB |
1886 | |
1887 | return (ZIO_PIPELINE_CONTINUE); | |
1888 | } | |
1889 | ||
1890 | static int | |
b128c09f | 1891 | zio_gang_issue(zio_t *zio) |
34dc7c2f | 1892 | { |
b128c09f | 1893 | blkptr_t *bp = zio->io_bp; |
34dc7c2f | 1894 | |
b128c09f BB |
1895 | if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE)) |
1896 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 1897 | |
9babb374 BB |
1898 | ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == zio); |
1899 | ASSERT(zio->io_child_type > ZIO_CHILD_GANG); | |
34dc7c2f | 1900 | |
b128c09f | 1901 | if (zio->io_child_error[ZIO_CHILD_GANG] == 0) |
9babb374 | 1902 | zio_gang_tree_issue(zio, zio->io_gang_tree, bp, zio->io_data); |
b128c09f | 1903 | else |
9babb374 | 1904 | zio_gang_tree_free(&zio->io_gang_tree); |
34dc7c2f | 1905 | |
b128c09f | 1906 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; |
34dc7c2f BB |
1907 | |
1908 | return (ZIO_PIPELINE_CONTINUE); | |
1909 | } | |
1910 | ||
1911 | static void | |
b128c09f | 1912 | zio_write_gang_member_ready(zio_t *zio) |
34dc7c2f | 1913 | { |
d164b209 | 1914 | zio_t *pio = zio_unique_parent(zio); |
34dc7c2f BB |
1915 | dva_t *cdva = zio->io_bp->blk_dva; |
1916 | dva_t *pdva = pio->io_bp->blk_dva; | |
1917 | uint64_t asize; | |
d6320ddb | 1918 | int d; |
d1d7e268 | 1919 | ASSERTV(zio_t *gio = zio->io_gang_leader); |
34dc7c2f | 1920 | |
b128c09f BB |
1921 | if (BP_IS_HOLE(zio->io_bp)) |
1922 | return; | |
1923 | ||
1924 | ASSERT(BP_IS_HOLE(&zio->io_bp_orig)); | |
1925 | ||
1926 | ASSERT(zio->io_child_type == ZIO_CHILD_GANG); | |
428870ff BB |
1927 | ASSERT3U(zio->io_prop.zp_copies, ==, gio->io_prop.zp_copies); |
1928 | ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp)); | |
1929 | ASSERT3U(pio->io_prop.zp_copies, <=, BP_GET_NDVAS(pio->io_bp)); | |
34dc7c2f | 1930 | ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp)); |
34dc7c2f BB |
1931 | |
1932 | mutex_enter(&pio->io_lock); | |
d6320ddb | 1933 | for (d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) { |
34dc7c2f BB |
1934 | ASSERT(DVA_GET_GANG(&pdva[d])); |
1935 | asize = DVA_GET_ASIZE(&pdva[d]); | |
1936 | asize += DVA_GET_ASIZE(&cdva[d]); | |
1937 | DVA_SET_ASIZE(&pdva[d], asize); | |
1938 | } | |
1939 | mutex_exit(&pio->io_lock); | |
1940 | } | |
1941 | ||
1942 | static int | |
b128c09f | 1943 | zio_write_gang_block(zio_t *pio) |
34dc7c2f | 1944 | { |
b128c09f BB |
1945 | spa_t *spa = pio->io_spa; |
1946 | blkptr_t *bp = pio->io_bp; | |
9babb374 | 1947 | zio_t *gio = pio->io_gang_leader; |
b128c09f BB |
1948 | zio_t *zio; |
1949 | zio_gang_node_t *gn, **gnpp; | |
34dc7c2f | 1950 | zio_gbh_phys_t *gbh; |
b128c09f BB |
1951 | uint64_t txg = pio->io_txg; |
1952 | uint64_t resid = pio->io_size; | |
1953 | uint64_t lsize; | |
428870ff BB |
1954 | int copies = gio->io_prop.zp_copies; |
1955 | int gbh_copies = MIN(copies + 1, spa_max_replication(spa)); | |
b128c09f | 1956 | zio_prop_t zp; |
d6320ddb | 1957 | int g, error; |
34dc7c2f | 1958 | |
428870ff BB |
1959 | error = metaslab_alloc(spa, spa_normal_class(spa), SPA_GANGBLOCKSIZE, |
1960 | bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, | |
b128c09f | 1961 | METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER); |
34dc7c2f | 1962 | if (error) { |
b128c09f | 1963 | pio->io_error = error; |
34dc7c2f BB |
1964 | return (ZIO_PIPELINE_CONTINUE); |
1965 | } | |
1966 | ||
9babb374 BB |
1967 | if (pio == gio) { |
1968 | gnpp = &gio->io_gang_tree; | |
b128c09f BB |
1969 | } else { |
1970 | gnpp = pio->io_private; | |
1971 | ASSERT(pio->io_ready == zio_write_gang_member_ready); | |
34dc7c2f BB |
1972 | } |
1973 | ||
b128c09f BB |
1974 | gn = zio_gang_node_alloc(gnpp); |
1975 | gbh = gn->gn_gbh; | |
1976 | bzero(gbh, SPA_GANGBLOCKSIZE); | |
34dc7c2f | 1977 | |
b128c09f BB |
1978 | /* |
1979 | * Create the gang header. | |
1980 | */ | |
1981 | zio = zio_rewrite(pio, spa, txg, bp, gbh, SPA_GANGBLOCKSIZE, NULL, NULL, | |
1982 | pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); | |
34dc7c2f | 1983 | |
b128c09f BB |
1984 | /* |
1985 | * Create and nowait the gang children. | |
1986 | */ | |
d6320ddb | 1987 | for (g = 0; resid != 0; resid -= lsize, g++) { |
b128c09f BB |
1988 | lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g), |
1989 | SPA_MINBLOCKSIZE); | |
1990 | ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid); | |
1991 | ||
9babb374 | 1992 | zp.zp_checksum = gio->io_prop.zp_checksum; |
b128c09f BB |
1993 | zp.zp_compress = ZIO_COMPRESS_OFF; |
1994 | zp.zp_type = DMU_OT_NONE; | |
1995 | zp.zp_level = 0; | |
428870ff | 1996 | zp.zp_copies = gio->io_prop.zp_copies; |
03c6040b GW |
1997 | zp.zp_dedup = B_FALSE; |
1998 | zp.zp_dedup_verify = B_FALSE; | |
1999 | zp.zp_nopwrite = B_FALSE; | |
b128c09f BB |
2000 | |
2001 | zio_nowait(zio_write(zio, spa, txg, &gbh->zg_blkptr[g], | |
2002 | (char *)pio->io_data + (pio->io_size - resid), lsize, &zp, | |
e8b96c60 | 2003 | zio_write_gang_member_ready, NULL, NULL, &gn->gn_child[g], |
b128c09f BB |
2004 | pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), |
2005 | &pio->io_bookmark)); | |
2006 | } | |
34dc7c2f BB |
2007 | |
2008 | /* | |
b128c09f | 2009 | * Set pio's pipeline to just wait for zio to finish. |
34dc7c2f | 2010 | */ |
b128c09f BB |
2011 | pio->io_pipeline = ZIO_INTERLOCK_PIPELINE; |
2012 | ||
920dd524 ED |
2013 | /* |
2014 | * We didn't allocate this bp, so make sure it doesn't get unmarked. | |
2015 | */ | |
2016 | pio->io_flags &= ~ZIO_FLAG_FASTWRITE; | |
2017 | ||
b128c09f BB |
2018 | zio_nowait(zio); |
2019 | ||
2020 | return (ZIO_PIPELINE_CONTINUE); | |
34dc7c2f BB |
2021 | } |
2022 | ||
03c6040b GW |
2023 | /* |
2024 | * The zio_nop_write stage in the pipeline determines if allocating | |
2025 | * a new bp is necessary. By leveraging a cryptographically secure checksum, | |
2026 | * such as SHA256, we can compare the checksums of the new data and the old | |
2027 | * to determine if allocating a new block is required. The nopwrite | |
2028 | * feature can handle writes in either syncing or open context (i.e. zil | |
2029 | * writes) and as a result is mutually exclusive with dedup. | |
2030 | */ | |
2031 | static int | |
2032 | zio_nop_write(zio_t *zio) | |
2033 | { | |
2034 | blkptr_t *bp = zio->io_bp; | |
2035 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
2036 | zio_prop_t *zp = &zio->io_prop; | |
2037 | ||
2038 | ASSERT(BP_GET_LEVEL(bp) == 0); | |
2039 | ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); | |
2040 | ASSERT(zp->zp_nopwrite); | |
2041 | ASSERT(!zp->zp_dedup); | |
2042 | ASSERT(zio->io_bp_override == NULL); | |
2043 | ASSERT(IO_IS_ALLOCATING(zio)); | |
2044 | ||
2045 | /* | |
2046 | * Check to see if the original bp and the new bp have matching | |
2047 | * characteristics (i.e. same checksum, compression algorithms, etc). | |
2048 | * If they don't then just continue with the pipeline which will | |
2049 | * allocate a new bp. | |
2050 | */ | |
2051 | if (BP_IS_HOLE(bp_orig) || | |
2052 | !zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_dedup || | |
2053 | BP_GET_CHECKSUM(bp) != BP_GET_CHECKSUM(bp_orig) || | |
2054 | BP_GET_COMPRESS(bp) != BP_GET_COMPRESS(bp_orig) || | |
2055 | BP_GET_DEDUP(bp) != BP_GET_DEDUP(bp_orig) || | |
2056 | zp->zp_copies != BP_GET_NDVAS(bp_orig)) | |
2057 | return (ZIO_PIPELINE_CONTINUE); | |
2058 | ||
2059 | /* | |
2060 | * If the checksums match then reset the pipeline so that we | |
2061 | * avoid allocating a new bp and issuing any I/O. | |
2062 | */ | |
2063 | if (ZIO_CHECKSUM_EQUAL(bp->blk_cksum, bp_orig->blk_cksum)) { | |
2064 | ASSERT(zio_checksum_table[zp->zp_checksum].ci_dedup); | |
2065 | ASSERT3U(BP_GET_PSIZE(bp), ==, BP_GET_PSIZE(bp_orig)); | |
2066 | ASSERT3U(BP_GET_LSIZE(bp), ==, BP_GET_LSIZE(bp_orig)); | |
2067 | ASSERT(zp->zp_compress != ZIO_COMPRESS_OFF); | |
2068 | ASSERT(bcmp(&bp->blk_prop, &bp_orig->blk_prop, | |
2069 | sizeof (uint64_t)) == 0); | |
2070 | ||
2071 | *bp = *bp_orig; | |
2072 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
2073 | zio->io_flags |= ZIO_FLAG_NOPWRITE; | |
2074 | } | |
2075 | ||
2076 | return (ZIO_PIPELINE_CONTINUE); | |
2077 | } | |
2078 | ||
34dc7c2f BB |
2079 | /* |
2080 | * ========================================================================== | |
428870ff | 2081 | * Dedup |
34dc7c2f BB |
2082 | * ========================================================================== |
2083 | */ | |
428870ff BB |
2084 | static void |
2085 | zio_ddt_child_read_done(zio_t *zio) | |
2086 | { | |
2087 | blkptr_t *bp = zio->io_bp; | |
2088 | ddt_entry_t *dde = zio->io_private; | |
2089 | ddt_phys_t *ddp; | |
2090 | zio_t *pio = zio_unique_parent(zio); | |
2091 | ||
2092 | mutex_enter(&pio->io_lock); | |
2093 | ddp = ddt_phys_select(dde, bp); | |
2094 | if (zio->io_error == 0) | |
2095 | ddt_phys_clear(ddp); /* this ddp doesn't need repair */ | |
2096 | if (zio->io_error == 0 && dde->dde_repair_data == NULL) | |
2097 | dde->dde_repair_data = zio->io_data; | |
2098 | else | |
2099 | zio_buf_free(zio->io_data, zio->io_size); | |
2100 | mutex_exit(&pio->io_lock); | |
2101 | } | |
2102 | ||
2103 | static int | |
2104 | zio_ddt_read_start(zio_t *zio) | |
2105 | { | |
2106 | blkptr_t *bp = zio->io_bp; | |
d6320ddb | 2107 | int p; |
428870ff BB |
2108 | |
2109 | ASSERT(BP_GET_DEDUP(bp)); | |
2110 | ASSERT(BP_GET_PSIZE(bp) == zio->io_size); | |
2111 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
2112 | ||
2113 | if (zio->io_child_error[ZIO_CHILD_DDT]) { | |
2114 | ddt_t *ddt = ddt_select(zio->io_spa, bp); | |
2115 | ddt_entry_t *dde = ddt_repair_start(ddt, bp); | |
2116 | ddt_phys_t *ddp = dde->dde_phys; | |
2117 | ddt_phys_t *ddp_self = ddt_phys_select(dde, bp); | |
2118 | blkptr_t blk; | |
2119 | ||
2120 | ASSERT(zio->io_vsd == NULL); | |
2121 | zio->io_vsd = dde; | |
2122 | ||
2123 | if (ddp_self == NULL) | |
2124 | return (ZIO_PIPELINE_CONTINUE); | |
2125 | ||
d6320ddb | 2126 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff BB |
2127 | if (ddp->ddp_phys_birth == 0 || ddp == ddp_self) |
2128 | continue; | |
2129 | ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp, | |
2130 | &blk); | |
2131 | zio_nowait(zio_read(zio, zio->io_spa, &blk, | |
2132 | zio_buf_alloc(zio->io_size), zio->io_size, | |
2133 | zio_ddt_child_read_done, dde, zio->io_priority, | |
2134 | ZIO_DDT_CHILD_FLAGS(zio) | ZIO_FLAG_DONT_PROPAGATE, | |
2135 | &zio->io_bookmark)); | |
2136 | } | |
2137 | return (ZIO_PIPELINE_CONTINUE); | |
2138 | } | |
2139 | ||
2140 | zio_nowait(zio_read(zio, zio->io_spa, bp, | |
2141 | zio->io_data, zio->io_size, NULL, NULL, zio->io_priority, | |
2142 | ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark)); | |
2143 | ||
2144 | return (ZIO_PIPELINE_CONTINUE); | |
2145 | } | |
2146 | ||
2147 | static int | |
2148 | zio_ddt_read_done(zio_t *zio) | |
2149 | { | |
2150 | blkptr_t *bp = zio->io_bp; | |
2151 | ||
2152 | if (zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE)) | |
2153 | return (ZIO_PIPELINE_STOP); | |
2154 | ||
2155 | ASSERT(BP_GET_DEDUP(bp)); | |
2156 | ASSERT(BP_GET_PSIZE(bp) == zio->io_size); | |
2157 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
2158 | ||
2159 | if (zio->io_child_error[ZIO_CHILD_DDT]) { | |
2160 | ddt_t *ddt = ddt_select(zio->io_spa, bp); | |
2161 | ddt_entry_t *dde = zio->io_vsd; | |
2162 | if (ddt == NULL) { | |
2163 | ASSERT(spa_load_state(zio->io_spa) != SPA_LOAD_NONE); | |
2164 | return (ZIO_PIPELINE_CONTINUE); | |
2165 | } | |
2166 | if (dde == NULL) { | |
2167 | zio->io_stage = ZIO_STAGE_DDT_READ_START >> 1; | |
2168 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); | |
2169 | return (ZIO_PIPELINE_STOP); | |
2170 | } | |
2171 | if (dde->dde_repair_data != NULL) { | |
2172 | bcopy(dde->dde_repair_data, zio->io_data, zio->io_size); | |
2173 | zio->io_child_error[ZIO_CHILD_DDT] = 0; | |
2174 | } | |
2175 | ddt_repair_done(ddt, dde); | |
2176 | zio->io_vsd = NULL; | |
2177 | } | |
2178 | ||
2179 | ASSERT(zio->io_vsd == NULL); | |
2180 | ||
2181 | return (ZIO_PIPELINE_CONTINUE); | |
2182 | } | |
2183 | ||
2184 | static boolean_t | |
2185 | zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde) | |
2186 | { | |
2187 | spa_t *spa = zio->io_spa; | |
d6320ddb | 2188 | int p; |
428870ff BB |
2189 | |
2190 | /* | |
2191 | * Note: we compare the original data, not the transformed data, | |
2192 | * because when zio->io_bp is an override bp, we will not have | |
2193 | * pushed the I/O transforms. That's an important optimization | |
2194 | * because otherwise we'd compress/encrypt all dmu_sync() data twice. | |
2195 | */ | |
d6320ddb | 2196 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { |
428870ff BB |
2197 | zio_t *lio = dde->dde_lead_zio[p]; |
2198 | ||
2199 | if (lio != NULL) { | |
2200 | return (lio->io_orig_size != zio->io_orig_size || | |
2201 | bcmp(zio->io_orig_data, lio->io_orig_data, | |
2202 | zio->io_orig_size) != 0); | |
2203 | } | |
2204 | } | |
2205 | ||
d6320ddb | 2206 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { |
428870ff BB |
2207 | ddt_phys_t *ddp = &dde->dde_phys[p]; |
2208 | ||
2209 | if (ddp->ddp_phys_birth != 0) { | |
2210 | arc_buf_t *abuf = NULL; | |
2211 | uint32_t aflags = ARC_WAIT; | |
2212 | blkptr_t blk = *zio->io_bp; | |
2213 | int error; | |
2214 | ||
2215 | ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth); | |
2216 | ||
2217 | ddt_exit(ddt); | |
2218 | ||
294f6806 | 2219 | error = arc_read(NULL, spa, &blk, |
428870ff BB |
2220 | arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ, |
2221 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, | |
2222 | &aflags, &zio->io_bookmark); | |
2223 | ||
2224 | if (error == 0) { | |
2225 | if (arc_buf_size(abuf) != zio->io_orig_size || | |
2226 | bcmp(abuf->b_data, zio->io_orig_data, | |
2227 | zio->io_orig_size) != 0) | |
2e528b49 | 2228 | error = SET_ERROR(EEXIST); |
13fe0198 | 2229 | VERIFY(arc_buf_remove_ref(abuf, &abuf)); |
428870ff BB |
2230 | } |
2231 | ||
2232 | ddt_enter(ddt); | |
2233 | return (error != 0); | |
2234 | } | |
2235 | } | |
2236 | ||
2237 | return (B_FALSE); | |
2238 | } | |
2239 | ||
2240 | static void | |
2241 | zio_ddt_child_write_ready(zio_t *zio) | |
2242 | { | |
2243 | int p = zio->io_prop.zp_copies; | |
2244 | ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); | |
2245 | ddt_entry_t *dde = zio->io_private; | |
2246 | ddt_phys_t *ddp = &dde->dde_phys[p]; | |
2247 | zio_t *pio; | |
2248 | ||
2249 | if (zio->io_error) | |
2250 | return; | |
2251 | ||
2252 | ddt_enter(ddt); | |
2253 | ||
2254 | ASSERT(dde->dde_lead_zio[p] == zio); | |
2255 | ||
2256 | ddt_phys_fill(ddp, zio->io_bp); | |
2257 | ||
2258 | while ((pio = zio_walk_parents(zio)) != NULL) | |
2259 | ddt_bp_fill(ddp, pio->io_bp, zio->io_txg); | |
2260 | ||
2261 | ddt_exit(ddt); | |
2262 | } | |
2263 | ||
2264 | static void | |
2265 | zio_ddt_child_write_done(zio_t *zio) | |
2266 | { | |
2267 | int p = zio->io_prop.zp_copies; | |
2268 | ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); | |
2269 | ddt_entry_t *dde = zio->io_private; | |
2270 | ddt_phys_t *ddp = &dde->dde_phys[p]; | |
2271 | ||
2272 | ddt_enter(ddt); | |
2273 | ||
2274 | ASSERT(ddp->ddp_refcnt == 0); | |
2275 | ASSERT(dde->dde_lead_zio[p] == zio); | |
2276 | dde->dde_lead_zio[p] = NULL; | |
2277 | ||
2278 | if (zio->io_error == 0) { | |
2279 | while (zio_walk_parents(zio) != NULL) | |
2280 | ddt_phys_addref(ddp); | |
2281 | } else { | |
2282 | ddt_phys_clear(ddp); | |
2283 | } | |
2284 | ||
2285 | ddt_exit(ddt); | |
2286 | } | |
2287 | ||
2288 | static void | |
2289 | zio_ddt_ditto_write_done(zio_t *zio) | |
2290 | { | |
2291 | int p = DDT_PHYS_DITTO; | |
428870ff BB |
2292 | blkptr_t *bp = zio->io_bp; |
2293 | ddt_t *ddt = ddt_select(zio->io_spa, bp); | |
2294 | ddt_entry_t *dde = zio->io_private; | |
2295 | ddt_phys_t *ddp = &dde->dde_phys[p]; | |
2296 | ddt_key_t *ddk = &dde->dde_key; | |
1fde1e37 | 2297 | ASSERTV(zio_prop_t *zp = &zio->io_prop); |
428870ff BB |
2298 | |
2299 | ddt_enter(ddt); | |
2300 | ||
2301 | ASSERT(ddp->ddp_refcnt == 0); | |
2302 | ASSERT(dde->dde_lead_zio[p] == zio); | |
2303 | dde->dde_lead_zio[p] = NULL; | |
2304 | ||
2305 | if (zio->io_error == 0) { | |
2306 | ASSERT(ZIO_CHECKSUM_EQUAL(bp->blk_cksum, ddk->ddk_cksum)); | |
2307 | ASSERT(zp->zp_copies < SPA_DVAS_PER_BP); | |
2308 | ASSERT(zp->zp_copies == BP_GET_NDVAS(bp) - BP_IS_GANG(bp)); | |
2309 | if (ddp->ddp_phys_birth != 0) | |
2310 | ddt_phys_free(ddt, ddk, ddp, zio->io_txg); | |
2311 | ddt_phys_fill(ddp, bp); | |
2312 | } | |
2313 | ||
2314 | ddt_exit(ddt); | |
2315 | } | |
2316 | ||
2317 | static int | |
2318 | zio_ddt_write(zio_t *zio) | |
2319 | { | |
2320 | spa_t *spa = zio->io_spa; | |
2321 | blkptr_t *bp = zio->io_bp; | |
2322 | uint64_t txg = zio->io_txg; | |
2323 | zio_prop_t *zp = &zio->io_prop; | |
2324 | int p = zp->zp_copies; | |
2325 | int ditto_copies; | |
2326 | zio_t *cio = NULL; | |
2327 | zio_t *dio = NULL; | |
2328 | ddt_t *ddt = ddt_select(spa, bp); | |
2329 | ddt_entry_t *dde; | |
2330 | ddt_phys_t *ddp; | |
2331 | ||
2332 | ASSERT(BP_GET_DEDUP(bp)); | |
2333 | ASSERT(BP_GET_CHECKSUM(bp) == zp->zp_checksum); | |
2334 | ASSERT(BP_IS_HOLE(bp) || zio->io_bp_override); | |
2335 | ||
2336 | ddt_enter(ddt); | |
2337 | dde = ddt_lookup(ddt, bp, B_TRUE); | |
2338 | ddp = &dde->dde_phys[p]; | |
2339 | ||
2340 | if (zp->zp_dedup_verify && zio_ddt_collision(zio, ddt, dde)) { | |
2341 | /* | |
2342 | * If we're using a weak checksum, upgrade to a strong checksum | |
2343 | * and try again. If we're already using a strong checksum, | |
2344 | * we can't resolve it, so just convert to an ordinary write. | |
2345 | * (And automatically e-mail a paper to Nature?) | |
2346 | */ | |
2347 | if (!zio_checksum_table[zp->zp_checksum].ci_dedup) { | |
2348 | zp->zp_checksum = spa_dedup_checksum(spa); | |
2349 | zio_pop_transforms(zio); | |
2350 | zio->io_stage = ZIO_STAGE_OPEN; | |
2351 | BP_ZERO(bp); | |
2352 | } else { | |
03c6040b | 2353 | zp->zp_dedup = B_FALSE; |
428870ff BB |
2354 | } |
2355 | zio->io_pipeline = ZIO_WRITE_PIPELINE; | |
2356 | ddt_exit(ddt); | |
2357 | return (ZIO_PIPELINE_CONTINUE); | |
2358 | } | |
2359 | ||
2360 | ditto_copies = ddt_ditto_copies_needed(ddt, dde, ddp); | |
2361 | ASSERT(ditto_copies < SPA_DVAS_PER_BP); | |
2362 | ||
2363 | if (ditto_copies > ddt_ditto_copies_present(dde) && | |
2364 | dde->dde_lead_zio[DDT_PHYS_DITTO] == NULL) { | |
2365 | zio_prop_t czp = *zp; | |
2366 | ||
2367 | czp.zp_copies = ditto_copies; | |
2368 | ||
2369 | /* | |
2370 | * If we arrived here with an override bp, we won't have run | |
2371 | * the transform stack, so we won't have the data we need to | |
2372 | * generate a child i/o. So, toss the override bp and restart. | |
2373 | * This is safe, because using the override bp is just an | |
2374 | * optimization; and it's rare, so the cost doesn't matter. | |
2375 | */ | |
2376 | if (zio->io_bp_override) { | |
2377 | zio_pop_transforms(zio); | |
2378 | zio->io_stage = ZIO_STAGE_OPEN; | |
2379 | zio->io_pipeline = ZIO_WRITE_PIPELINE; | |
2380 | zio->io_bp_override = NULL; | |
2381 | BP_ZERO(bp); | |
2382 | ddt_exit(ddt); | |
2383 | return (ZIO_PIPELINE_CONTINUE); | |
2384 | } | |
2385 | ||
2386 | dio = zio_write(zio, spa, txg, bp, zio->io_orig_data, | |
e8b96c60 | 2387 | zio->io_orig_size, &czp, NULL, NULL, |
428870ff BB |
2388 | zio_ddt_ditto_write_done, dde, zio->io_priority, |
2389 | ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); | |
2390 | ||
2391 | zio_push_transform(dio, zio->io_data, zio->io_size, 0, NULL); | |
2392 | dde->dde_lead_zio[DDT_PHYS_DITTO] = dio; | |
2393 | } | |
2394 | ||
2395 | if (ddp->ddp_phys_birth != 0 || dde->dde_lead_zio[p] != NULL) { | |
2396 | if (ddp->ddp_phys_birth != 0) | |
2397 | ddt_bp_fill(ddp, bp, txg); | |
2398 | if (dde->dde_lead_zio[p] != NULL) | |
2399 | zio_add_child(zio, dde->dde_lead_zio[p]); | |
2400 | else | |
2401 | ddt_phys_addref(ddp); | |
2402 | } else if (zio->io_bp_override) { | |
2403 | ASSERT(bp->blk_birth == txg); | |
2404 | ASSERT(BP_EQUAL(bp, zio->io_bp_override)); | |
2405 | ddt_phys_fill(ddp, bp); | |
2406 | ddt_phys_addref(ddp); | |
2407 | } else { | |
2408 | cio = zio_write(zio, spa, txg, bp, zio->io_orig_data, | |
e8b96c60 | 2409 | zio->io_orig_size, zp, zio_ddt_child_write_ready, NULL, |
428870ff BB |
2410 | zio_ddt_child_write_done, dde, zio->io_priority, |
2411 | ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); | |
2412 | ||
2413 | zio_push_transform(cio, zio->io_data, zio->io_size, 0, NULL); | |
2414 | dde->dde_lead_zio[p] = cio; | |
2415 | } | |
2416 | ||
2417 | ddt_exit(ddt); | |
2418 | ||
2419 | if (cio) | |
2420 | zio_nowait(cio); | |
2421 | if (dio) | |
2422 | zio_nowait(dio); | |
2423 | ||
2424 | return (ZIO_PIPELINE_CONTINUE); | |
2425 | } | |
2426 | ||
2427 | ddt_entry_t *freedde; /* for debugging */ | |
b128c09f | 2428 | |
428870ff BB |
2429 | static int |
2430 | zio_ddt_free(zio_t *zio) | |
2431 | { | |
2432 | spa_t *spa = zio->io_spa; | |
2433 | blkptr_t *bp = zio->io_bp; | |
2434 | ddt_t *ddt = ddt_select(spa, bp); | |
2435 | ddt_entry_t *dde; | |
2436 | ddt_phys_t *ddp; | |
2437 | ||
2438 | ASSERT(BP_GET_DEDUP(bp)); | |
2439 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
2440 | ||
2441 | ddt_enter(ddt); | |
2442 | freedde = dde = ddt_lookup(ddt, bp, B_TRUE); | |
5dc6af0e BB |
2443 | if (dde) { |
2444 | ddp = ddt_phys_select(dde, bp); | |
2445 | if (ddp) | |
2446 | ddt_phys_decref(ddp); | |
2447 | } | |
428870ff BB |
2448 | ddt_exit(ddt); |
2449 | ||
2450 | return (ZIO_PIPELINE_CONTINUE); | |
2451 | } | |
2452 | ||
2453 | /* | |
2454 | * ========================================================================== | |
2455 | * Allocate and free blocks | |
2456 | * ========================================================================== | |
2457 | */ | |
34dc7c2f BB |
2458 | static int |
2459 | zio_dva_allocate(zio_t *zio) | |
2460 | { | |
2461 | spa_t *spa = zio->io_spa; | |
428870ff | 2462 | metaslab_class_t *mc = spa_normal_class(spa); |
34dc7c2f BB |
2463 | blkptr_t *bp = zio->io_bp; |
2464 | int error; | |
6d974228 | 2465 | int flags = 0; |
34dc7c2f | 2466 | |
9babb374 BB |
2467 | if (zio->io_gang_leader == NULL) { |
2468 | ASSERT(zio->io_child_type > ZIO_CHILD_GANG); | |
2469 | zio->io_gang_leader = zio; | |
2470 | } | |
2471 | ||
34dc7c2f | 2472 | ASSERT(BP_IS_HOLE(bp)); |
c99c9001 | 2473 | ASSERT0(BP_GET_NDVAS(bp)); |
428870ff BB |
2474 | ASSERT3U(zio->io_prop.zp_copies, >, 0); |
2475 | ASSERT3U(zio->io_prop.zp_copies, <=, spa_max_replication(spa)); | |
34dc7c2f BB |
2476 | ASSERT3U(zio->io_size, ==, BP_GET_PSIZE(bp)); |
2477 | ||
6d974228 GW |
2478 | /* |
2479 | * The dump device does not support gang blocks so allocation on | |
2480 | * behalf of the dump device (i.e. ZIO_FLAG_NODATA) must avoid | |
2481 | * the "fast" gang feature. | |
2482 | */ | |
2483 | flags |= (zio->io_flags & ZIO_FLAG_NODATA) ? METASLAB_GANG_AVOID : 0; | |
2484 | flags |= (zio->io_flags & ZIO_FLAG_GANG_CHILD) ? | |
2485 | METASLAB_GANG_CHILD : 0; | |
920dd524 | 2486 | flags |= (zio->io_flags & ZIO_FLAG_FASTWRITE) ? METASLAB_FASTWRITE : 0; |
b128c09f | 2487 | error = metaslab_alloc(spa, mc, zio->io_size, bp, |
6d974228 | 2488 | zio->io_prop.zp_copies, zio->io_txg, NULL, flags); |
34dc7c2f | 2489 | |
b128c09f | 2490 | if (error) { |
6d974228 GW |
2491 | spa_dbgmsg(spa, "%s: metaslab allocation failure: zio %p, " |
2492 | "size %llu, error %d", spa_name(spa), zio, zio->io_size, | |
2493 | error); | |
b128c09f BB |
2494 | if (error == ENOSPC && zio->io_size > SPA_MINBLOCKSIZE) |
2495 | return (zio_write_gang_block(zio)); | |
34dc7c2f BB |
2496 | zio->io_error = error; |
2497 | } | |
2498 | ||
2499 | return (ZIO_PIPELINE_CONTINUE); | |
2500 | } | |
2501 | ||
2502 | static int | |
2503 | zio_dva_free(zio_t *zio) | |
2504 | { | |
b128c09f | 2505 | metaslab_free(zio->io_spa, zio->io_bp, zio->io_txg, B_FALSE); |
34dc7c2f BB |
2506 | |
2507 | return (ZIO_PIPELINE_CONTINUE); | |
2508 | } | |
2509 | ||
2510 | static int | |
2511 | zio_dva_claim(zio_t *zio) | |
2512 | { | |
b128c09f BB |
2513 | int error; |
2514 | ||
2515 | error = metaslab_claim(zio->io_spa, zio->io_bp, zio->io_txg); | |
2516 | if (error) | |
2517 | zio->io_error = error; | |
34dc7c2f BB |
2518 | |
2519 | return (ZIO_PIPELINE_CONTINUE); | |
2520 | } | |
2521 | ||
b128c09f BB |
2522 | /* |
2523 | * Undo an allocation. This is used by zio_done() when an I/O fails | |
2524 | * and we want to give back the block we just allocated. | |
2525 | * This handles both normal blocks and gang blocks. | |
2526 | */ | |
2527 | static void | |
2528 | zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp) | |
2529 | { | |
d6320ddb BB |
2530 | int g; |
2531 | ||
b128c09f | 2532 | ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); |
428870ff | 2533 | ASSERT(zio->io_bp_override == NULL); |
b128c09f BB |
2534 | |
2535 | if (!BP_IS_HOLE(bp)) | |
428870ff | 2536 | metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE); |
b128c09f BB |
2537 | |
2538 | if (gn != NULL) { | |
d6320ddb | 2539 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) { |
b128c09f BB |
2540 | zio_dva_unallocate(zio, gn->gn_child[g], |
2541 | &gn->gn_gbh->zg_blkptr[g]); | |
2542 | } | |
2543 | } | |
2544 | } | |
2545 | ||
2546 | /* | |
2547 | * Try to allocate an intent log block. Return 0 on success, errno on failure. | |
2548 | */ | |
2549 | int | |
920dd524 ED |
2550 | zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, uint64_t size, |
2551 | boolean_t use_slog) | |
b128c09f | 2552 | { |
428870ff | 2553 | int error = 1; |
b128c09f | 2554 | |
428870ff BB |
2555 | ASSERT(txg > spa_syncing_txg(spa)); |
2556 | ||
ebf8e3a2 BB |
2557 | /* |
2558 | * ZIL blocks are always contiguous (i.e. not gang blocks) so we | |
2559 | * set the METASLAB_GANG_AVOID flag so that they don't "fast gang" | |
2560 | * when allocating them. | |
2561 | */ | |
2562 | if (use_slog) { | |
428870ff | 2563 | error = metaslab_alloc(spa, spa_log_class(spa), size, |
920dd524 ED |
2564 | new_bp, 1, txg, NULL, |
2565 | METASLAB_FASTWRITE | METASLAB_GANG_AVOID); | |
ebf8e3a2 | 2566 | } |
b128c09f | 2567 | |
ebf8e3a2 | 2568 | if (error) { |
428870ff | 2569 | error = metaslab_alloc(spa, spa_normal_class(spa), size, |
920dd524 | 2570 | new_bp, 1, txg, NULL, |
ac72fac3 | 2571 | METASLAB_FASTWRITE); |
ebf8e3a2 | 2572 | } |
b128c09f BB |
2573 | |
2574 | if (error == 0) { | |
2575 | BP_SET_LSIZE(new_bp, size); | |
2576 | BP_SET_PSIZE(new_bp, size); | |
2577 | BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF); | |
428870ff BB |
2578 | BP_SET_CHECKSUM(new_bp, |
2579 | spa_version(spa) >= SPA_VERSION_SLIM_ZIL | |
2580 | ? ZIO_CHECKSUM_ZILOG2 : ZIO_CHECKSUM_ZILOG); | |
b128c09f BB |
2581 | BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG); |
2582 | BP_SET_LEVEL(new_bp, 0); | |
428870ff | 2583 | BP_SET_DEDUP(new_bp, 0); |
b128c09f BB |
2584 | BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER); |
2585 | } | |
2586 | ||
2587 | return (error); | |
2588 | } | |
2589 | ||
2590 | /* | |
428870ff | 2591 | * Free an intent log block. |
b128c09f BB |
2592 | */ |
2593 | void | |
428870ff | 2594 | zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp) |
b128c09f | 2595 | { |
428870ff | 2596 | ASSERT(BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG); |
b128c09f BB |
2597 | ASSERT(!BP_IS_GANG(bp)); |
2598 | ||
428870ff | 2599 | zio_free(spa, txg, bp); |
b128c09f BB |
2600 | } |
2601 | ||
34dc7c2f BB |
2602 | /* |
2603 | * ========================================================================== | |
2604 | * Read and write to physical devices | |
2605 | * ========================================================================== | |
2606 | */ | |
34dc7c2f BB |
2607 | static int |
2608 | zio_vdev_io_start(zio_t *zio) | |
2609 | { | |
2610 | vdev_t *vd = zio->io_vd; | |
34dc7c2f BB |
2611 | uint64_t align; |
2612 | spa_t *spa = zio->io_spa; | |
2613 | ||
b128c09f BB |
2614 | ASSERT(zio->io_error == 0); |
2615 | ASSERT(zio->io_child_error[ZIO_CHILD_VDEV] == 0); | |
34dc7c2f | 2616 | |
b128c09f BB |
2617 | if (vd == NULL) { |
2618 | if (!(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) | |
2619 | spa_config_enter(spa, SCL_ZIO, zio, RW_READER); | |
34dc7c2f | 2620 | |
b128c09f BB |
2621 | /* |
2622 | * The mirror_ops handle multiple DVAs in a single BP. | |
2623 | */ | |
2624 | return (vdev_mirror_ops.vdev_op_io_start(zio)); | |
34dc7c2f BB |
2625 | } |
2626 | ||
572e2857 BB |
2627 | /* |
2628 | * We keep track of time-sensitive I/Os so that the scan thread | |
2629 | * can quickly react to certain workloads. In particular, we care | |
2630 | * about non-scrubbing, top-level reads and writes with the following | |
2631 | * characteristics: | |
2632 | * - synchronous writes of user data to non-slog devices | |
2633 | * - any reads of user data | |
2634 | * When these conditions are met, adjust the timestamp of spa_last_io | |
2635 | * which allows the scan thread to adjust its workload accordingly. | |
2636 | */ | |
2637 | if (!(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_bp != NULL && | |
2638 | vd == vd->vdev_top && !vd->vdev_islog && | |
2639 | zio->io_bookmark.zb_objset != DMU_META_OBJSET && | |
2640 | zio->io_txg != spa_syncing_txg(spa)) { | |
2641 | uint64_t old = spa->spa_last_io; | |
2642 | uint64_t new = ddi_get_lbolt64(); | |
2643 | if (old != new) | |
2644 | (void) atomic_cas_64(&spa->spa_last_io, old, new); | |
2645 | } | |
2646 | ||
b128c09f BB |
2647 | align = 1ULL << vd->vdev_top->vdev_ashift; |
2648 | ||
b02fe35d AR |
2649 | if (!(zio->io_flags & ZIO_FLAG_PHYSICAL) && |
2650 | P2PHASE(zio->io_size, align) != 0) { | |
2651 | /* Transform logical writes to be a full physical block size. */ | |
34dc7c2f BB |
2652 | uint64_t asize = P2ROUNDUP(zio->io_size, align); |
2653 | char *abuf = zio_buf_alloc(asize); | |
178e73b3 | 2654 | ASSERT(vd == vd->vdev_top); |
34dc7c2f BB |
2655 | if (zio->io_type == ZIO_TYPE_WRITE) { |
2656 | bcopy(zio->io_data, abuf, zio->io_size); | |
2657 | bzero(abuf + zio->io_size, asize - zio->io_size); | |
2658 | } | |
b128c09f | 2659 | zio_push_transform(zio, abuf, asize, asize, zio_subblock); |
34dc7c2f BB |
2660 | } |
2661 | ||
b02fe35d AR |
2662 | /* |
2663 | * If this is not a physical io, make sure that it is properly aligned | |
2664 | * before proceeding. | |
2665 | */ | |
2666 | if (!(zio->io_flags & ZIO_FLAG_PHYSICAL)) { | |
2667 | ASSERT0(P2PHASE(zio->io_offset, align)); | |
2668 | ASSERT0(P2PHASE(zio->io_size, align)); | |
2669 | } else { | |
2670 | /* | |
2671 | * For physical writes, we allow 512b aligned writes and assume | |
2672 | * the device will perform a read-modify-write as necessary. | |
2673 | */ | |
2674 | ASSERT0(P2PHASE(zio->io_offset, SPA_MINBLOCKSIZE)); | |
2675 | ASSERT0(P2PHASE(zio->io_size, SPA_MINBLOCKSIZE)); | |
2676 | } | |
2677 | ||
572e2857 | 2678 | VERIFY(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa)); |
fb5f0bc8 BB |
2679 | |
2680 | /* | |
2681 | * If this is a repair I/O, and there's no self-healing involved -- | |
2682 | * that is, we're just resilvering what we expect to resilver -- | |
2683 | * then don't do the I/O unless zio's txg is actually in vd's DTL. | |
2684 | * This prevents spurious resilvering with nested replication. | |
2685 | * For example, given a mirror of mirrors, (A+B)+(C+D), if only | |
2686 | * A is out of date, we'll read from C+D, then use the data to | |
2687 | * resilver A+B -- but we don't actually want to resilver B, just A. | |
2688 | * The top-level mirror has no way to know this, so instead we just | |
2689 | * discard unnecessary repairs as we work our way down the vdev tree. | |
2690 | * The same logic applies to any form of nested replication: | |
2691 | * ditto + mirror, RAID-Z + replacing, etc. This covers them all. | |
2692 | */ | |
2693 | if ((zio->io_flags & ZIO_FLAG_IO_REPAIR) && | |
2694 | !(zio->io_flags & ZIO_FLAG_SELF_HEAL) && | |
2695 | zio->io_txg != 0 && /* not a delegated i/o */ | |
2696 | !vdev_dtl_contains(vd, DTL_PARTIAL, zio->io_txg, 1)) { | |
2697 | ASSERT(zio->io_type == ZIO_TYPE_WRITE); | |
fb5f0bc8 BB |
2698 | zio_vdev_io_bypass(zio); |
2699 | return (ZIO_PIPELINE_CONTINUE); | |
2700 | } | |
34dc7c2f | 2701 | |
b128c09f BB |
2702 | if (vd->vdev_ops->vdev_op_leaf && |
2703 | (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE)) { | |
2704 | ||
b0bc7a84 | 2705 | if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio)) |
d164b209 | 2706 | return (ZIO_PIPELINE_CONTINUE); |
b128c09f BB |
2707 | |
2708 | if ((zio = vdev_queue_io(zio)) == NULL) | |
2709 | return (ZIO_PIPELINE_STOP); | |
2710 | ||
2711 | if (!vdev_accessible(vd, zio)) { | |
2e528b49 | 2712 | zio->io_error = SET_ERROR(ENXIO); |
b128c09f BB |
2713 | zio_interrupt(zio); |
2714 | return (ZIO_PIPELINE_STOP); | |
2715 | } | |
b128c09f BB |
2716 | } |
2717 | ||
34dc7c2f BB |
2718 | return (vd->vdev_ops->vdev_op_io_start(zio)); |
2719 | } | |
2720 | ||
2721 | static int | |
2722 | zio_vdev_io_done(zio_t *zio) | |
2723 | { | |
b128c09f BB |
2724 | vdev_t *vd = zio->io_vd; |
2725 | vdev_ops_t *ops = vd ? vd->vdev_ops : &vdev_mirror_ops; | |
2726 | boolean_t unexpected_error = B_FALSE; | |
34dc7c2f | 2727 | |
b128c09f BB |
2728 | if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) |
2729 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 2730 | |
b128c09f BB |
2731 | ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE); |
2732 | ||
2733 | if (vd != NULL && vd->vdev_ops->vdev_op_leaf) { | |
2734 | ||
2735 | vdev_queue_io_done(zio); | |
2736 | ||
2737 | if (zio->io_type == ZIO_TYPE_WRITE) | |
2738 | vdev_cache_write(zio); | |
2739 | ||
2740 | if (zio_injection_enabled && zio->io_error == 0) | |
9babb374 BB |
2741 | zio->io_error = zio_handle_device_injection(vd, |
2742 | zio, EIO); | |
b128c09f BB |
2743 | |
2744 | if (zio_injection_enabled && zio->io_error == 0) | |
2745 | zio->io_error = zio_handle_label_injection(zio, EIO); | |
2746 | ||
2747 | if (zio->io_error) { | |
2748 | if (!vdev_accessible(vd, zio)) { | |
2e528b49 | 2749 | zio->io_error = SET_ERROR(ENXIO); |
b128c09f BB |
2750 | } else { |
2751 | unexpected_error = B_TRUE; | |
2752 | } | |
2753 | } | |
2754 | } | |
2755 | ||
2756 | ops->vdev_op_io_done(zio); | |
34dc7c2f | 2757 | |
b128c09f | 2758 | if (unexpected_error) |
d164b209 | 2759 | VERIFY(vdev_probe(vd, zio) == NULL); |
34dc7c2f | 2760 | |
b128c09f | 2761 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
2762 | } |
2763 | ||
428870ff BB |
2764 | /* |
2765 | * For non-raidz ZIOs, we can just copy aside the bad data read from the | |
2766 | * disk, and use that to finish the checksum ereport later. | |
2767 | */ | |
2768 | static void | |
2769 | zio_vsd_default_cksum_finish(zio_cksum_report_t *zcr, | |
2770 | const void *good_buf) | |
2771 | { | |
2772 | /* no processing needed */ | |
2773 | zfs_ereport_finish_checksum(zcr, good_buf, zcr->zcr_cbdata, B_FALSE); | |
2774 | } | |
2775 | ||
2776 | /*ARGSUSED*/ | |
2777 | void | |
2778 | zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *ignored) | |
2779 | { | |
2780 | void *buf = zio_buf_alloc(zio->io_size); | |
2781 | ||
2782 | bcopy(zio->io_data, buf, zio->io_size); | |
2783 | ||
2784 | zcr->zcr_cbinfo = zio->io_size; | |
2785 | zcr->zcr_cbdata = buf; | |
2786 | zcr->zcr_finish = zio_vsd_default_cksum_finish; | |
2787 | zcr->zcr_free = zio_buf_free; | |
2788 | } | |
2789 | ||
34dc7c2f BB |
2790 | static int |
2791 | zio_vdev_io_assess(zio_t *zio) | |
2792 | { | |
2793 | vdev_t *vd = zio->io_vd; | |
b128c09f BB |
2794 | |
2795 | if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) | |
2796 | return (ZIO_PIPELINE_STOP); | |
2797 | ||
2798 | if (vd == NULL && !(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) | |
2799 | spa_config_exit(zio->io_spa, SCL_ZIO, zio); | |
2800 | ||
2801 | if (zio->io_vsd != NULL) { | |
428870ff | 2802 | zio->io_vsd_ops->vsd_free(zio); |
b128c09f | 2803 | zio->io_vsd = NULL; |
34dc7c2f BB |
2804 | } |
2805 | ||
b128c09f | 2806 | if (zio_injection_enabled && zio->io_error == 0) |
34dc7c2f BB |
2807 | zio->io_error = zio_handle_fault_injection(zio, EIO); |
2808 | ||
2809 | /* | |
2810 | * If the I/O failed, determine whether we should attempt to retry it. | |
428870ff BB |
2811 | * |
2812 | * On retry, we cut in line in the issue queue, since we don't want | |
2813 | * compression/checksumming/etc. work to prevent our (cheap) IO reissue. | |
34dc7c2f | 2814 | */ |
b128c09f BB |
2815 | if (zio->io_error && vd == NULL && |
2816 | !(zio->io_flags & (ZIO_FLAG_DONT_RETRY | ZIO_FLAG_IO_RETRY))) { | |
2817 | ASSERT(!(zio->io_flags & ZIO_FLAG_DONT_QUEUE)); /* not a leaf */ | |
2818 | ASSERT(!(zio->io_flags & ZIO_FLAG_IO_BYPASS)); /* not a leaf */ | |
34dc7c2f | 2819 | zio->io_error = 0; |
b128c09f BB |
2820 | zio->io_flags |= ZIO_FLAG_IO_RETRY | |
2821 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE; | |
428870ff BB |
2822 | zio->io_stage = ZIO_STAGE_VDEV_IO_START >> 1; |
2823 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, | |
2824 | zio_requeue_io_start_cut_in_line); | |
b128c09f | 2825 | return (ZIO_PIPELINE_STOP); |
34dc7c2f BB |
2826 | } |
2827 | ||
b128c09f BB |
2828 | /* |
2829 | * If we got an error on a leaf device, convert it to ENXIO | |
2830 | * if the device is not accessible at all. | |
2831 | */ | |
2832 | if (zio->io_error && vd != NULL && vd->vdev_ops->vdev_op_leaf && | |
2833 | !vdev_accessible(vd, zio)) | |
2e528b49 | 2834 | zio->io_error = SET_ERROR(ENXIO); |
b128c09f BB |
2835 | |
2836 | /* | |
2837 | * If we can't write to an interior vdev (mirror or RAID-Z), | |
2838 | * set vdev_cant_write so that we stop trying to allocate from it. | |
2839 | */ | |
2840 | if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE && | |
13fe0198 | 2841 | vd != NULL && !vd->vdev_ops->vdev_op_leaf) { |
b128c09f | 2842 | vd->vdev_cant_write = B_TRUE; |
13fe0198 | 2843 | } |
b128c09f BB |
2844 | |
2845 | if (zio->io_error) | |
2846 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
2847 | ||
e8b96c60 MA |
2848 | if (vd != NULL && vd->vdev_ops->vdev_op_leaf && |
2849 | zio->io_physdone != NULL) { | |
2850 | ASSERT(!(zio->io_flags & ZIO_FLAG_DELEGATED)); | |
2851 | ASSERT(zio->io_child_type == ZIO_CHILD_VDEV); | |
2852 | zio->io_physdone(zio->io_logical); | |
2853 | } | |
2854 | ||
34dc7c2f BB |
2855 | return (ZIO_PIPELINE_CONTINUE); |
2856 | } | |
2857 | ||
2858 | void | |
2859 | zio_vdev_io_reissue(zio_t *zio) | |
2860 | { | |
2861 | ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); | |
2862 | ASSERT(zio->io_error == 0); | |
2863 | ||
428870ff | 2864 | zio->io_stage >>= 1; |
34dc7c2f BB |
2865 | } |
2866 | ||
2867 | void | |
2868 | zio_vdev_io_redone(zio_t *zio) | |
2869 | { | |
2870 | ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_DONE); | |
2871 | ||
428870ff | 2872 | zio->io_stage >>= 1; |
34dc7c2f BB |
2873 | } |
2874 | ||
2875 | void | |
2876 | zio_vdev_io_bypass(zio_t *zio) | |
2877 | { | |
2878 | ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); | |
2879 | ASSERT(zio->io_error == 0); | |
2880 | ||
2881 | zio->io_flags |= ZIO_FLAG_IO_BYPASS; | |
428870ff | 2882 | zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS >> 1; |
34dc7c2f BB |
2883 | } |
2884 | ||
2885 | /* | |
2886 | * ========================================================================== | |
2887 | * Generate and verify checksums | |
2888 | * ========================================================================== | |
2889 | */ | |
2890 | static int | |
2891 | zio_checksum_generate(zio_t *zio) | |
2892 | { | |
34dc7c2f | 2893 | blkptr_t *bp = zio->io_bp; |
b128c09f | 2894 | enum zio_checksum checksum; |
34dc7c2f | 2895 | |
b128c09f BB |
2896 | if (bp == NULL) { |
2897 | /* | |
2898 | * This is zio_write_phys(). | |
2899 | * We're either generating a label checksum, or none at all. | |
2900 | */ | |
2901 | checksum = zio->io_prop.zp_checksum; | |
34dc7c2f | 2902 | |
b128c09f BB |
2903 | if (checksum == ZIO_CHECKSUM_OFF) |
2904 | return (ZIO_PIPELINE_CONTINUE); | |
2905 | ||
2906 | ASSERT(checksum == ZIO_CHECKSUM_LABEL); | |
2907 | } else { | |
2908 | if (BP_IS_GANG(bp) && zio->io_child_type == ZIO_CHILD_GANG) { | |
2909 | ASSERT(!IO_IS_ALLOCATING(zio)); | |
2910 | checksum = ZIO_CHECKSUM_GANG_HEADER; | |
2911 | } else { | |
2912 | checksum = BP_GET_CHECKSUM(bp); | |
2913 | } | |
2914 | } | |
34dc7c2f | 2915 | |
b128c09f | 2916 | zio_checksum_compute(zio, checksum, zio->io_data, zio->io_size); |
34dc7c2f BB |
2917 | |
2918 | return (ZIO_PIPELINE_CONTINUE); | |
2919 | } | |
2920 | ||
2921 | static int | |
b128c09f | 2922 | zio_checksum_verify(zio_t *zio) |
34dc7c2f | 2923 | { |
428870ff | 2924 | zio_bad_cksum_t info; |
b128c09f BB |
2925 | blkptr_t *bp = zio->io_bp; |
2926 | int error; | |
34dc7c2f | 2927 | |
428870ff BB |
2928 | ASSERT(zio->io_vd != NULL); |
2929 | ||
b128c09f BB |
2930 | if (bp == NULL) { |
2931 | /* | |
2932 | * This is zio_read_phys(). | |
2933 | * We're either verifying a label checksum, or nothing at all. | |
2934 | */ | |
2935 | if (zio->io_prop.zp_checksum == ZIO_CHECKSUM_OFF) | |
2936 | return (ZIO_PIPELINE_CONTINUE); | |
34dc7c2f | 2937 | |
b128c09f BB |
2938 | ASSERT(zio->io_prop.zp_checksum == ZIO_CHECKSUM_LABEL); |
2939 | } | |
34dc7c2f | 2940 | |
428870ff | 2941 | if ((error = zio_checksum_error(zio, &info)) != 0) { |
b128c09f BB |
2942 | zio->io_error = error; |
2943 | if (!(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { | |
428870ff BB |
2944 | zfs_ereport_start_checksum(zio->io_spa, |
2945 | zio->io_vd, zio, zio->io_offset, | |
2946 | zio->io_size, NULL, &info); | |
b128c09f | 2947 | } |
34dc7c2f BB |
2948 | } |
2949 | ||
2950 | return (ZIO_PIPELINE_CONTINUE); | |
2951 | } | |
2952 | ||
2953 | /* | |
2954 | * Called by RAID-Z to ensure we don't compute the checksum twice. | |
2955 | */ | |
2956 | void | |
2957 | zio_checksum_verified(zio_t *zio) | |
2958 | { | |
428870ff | 2959 | zio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; |
34dc7c2f BB |
2960 | } |
2961 | ||
2962 | /* | |
b128c09f BB |
2963 | * ========================================================================== |
2964 | * Error rank. Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other. | |
9b67f605 | 2965 | * An error of 0 indicates success. ENXIO indicates whole-device failure, |
b128c09f BB |
2966 | * which may be transient (e.g. unplugged) or permament. ECKSUM and EIO |
2967 | * indicate errors that are specific to one I/O, and most likely permanent. | |
2968 | * Any other error is presumed to be worse because we weren't expecting it. | |
2969 | * ========================================================================== | |
34dc7c2f | 2970 | */ |
b128c09f BB |
2971 | int |
2972 | zio_worst_error(int e1, int e2) | |
34dc7c2f | 2973 | { |
b128c09f BB |
2974 | static int zio_error_rank[] = { 0, ENXIO, ECKSUM, EIO }; |
2975 | int r1, r2; | |
2976 | ||
2977 | for (r1 = 0; r1 < sizeof (zio_error_rank) / sizeof (int); r1++) | |
2978 | if (e1 == zio_error_rank[r1]) | |
2979 | break; | |
34dc7c2f | 2980 | |
b128c09f BB |
2981 | for (r2 = 0; r2 < sizeof (zio_error_rank) / sizeof (int); r2++) |
2982 | if (e2 == zio_error_rank[r2]) | |
2983 | break; | |
2984 | ||
2985 | return (r1 > r2 ? e1 : e2); | |
34dc7c2f BB |
2986 | } |
2987 | ||
2988 | /* | |
2989 | * ========================================================================== | |
b128c09f | 2990 | * I/O completion |
34dc7c2f BB |
2991 | * ========================================================================== |
2992 | */ | |
b128c09f BB |
2993 | static int |
2994 | zio_ready(zio_t *zio) | |
34dc7c2f | 2995 | { |
b128c09f | 2996 | blkptr_t *bp = zio->io_bp; |
d164b209 | 2997 | zio_t *pio, *pio_next; |
34dc7c2f | 2998 | |
428870ff BB |
2999 | if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || |
3000 | zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_READY)) | |
9babb374 | 3001 | return (ZIO_PIPELINE_STOP); |
34dc7c2f | 3002 | |
9babb374 | 3003 | if (zio->io_ready) { |
b128c09f | 3004 | ASSERT(IO_IS_ALLOCATING(zio)); |
03c6040b GW |
3005 | ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp) || |
3006 | (zio->io_flags & ZIO_FLAG_NOPWRITE)); | |
b128c09f | 3007 | ASSERT(zio->io_children[ZIO_CHILD_GANG][ZIO_WAIT_READY] == 0); |
34dc7c2f | 3008 | |
b128c09f BB |
3009 | zio->io_ready(zio); |
3010 | } | |
34dc7c2f | 3011 | |
b128c09f BB |
3012 | if (bp != NULL && bp != &zio->io_bp_copy) |
3013 | zio->io_bp_copy = *bp; | |
34dc7c2f | 3014 | |
b128c09f BB |
3015 | if (zio->io_error) |
3016 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
34dc7c2f | 3017 | |
d164b209 BB |
3018 | mutex_enter(&zio->io_lock); |
3019 | zio->io_state[ZIO_WAIT_READY] = 1; | |
3020 | pio = zio_walk_parents(zio); | |
3021 | mutex_exit(&zio->io_lock); | |
3022 | ||
3023 | /* | |
3024 | * As we notify zio's parents, new parents could be added. | |
3025 | * New parents go to the head of zio's io_parent_list, however, | |
3026 | * so we will (correctly) not notify them. The remainder of zio's | |
3027 | * io_parent_list, from 'pio_next' onward, cannot change because | |
3028 | * all parents must wait for us to be done before they can be done. | |
3029 | */ | |
3030 | for (; pio != NULL; pio = pio_next) { | |
3031 | pio_next = zio_walk_parents(zio); | |
b128c09f | 3032 | zio_notify_parent(pio, zio, ZIO_WAIT_READY); |
d164b209 | 3033 | } |
34dc7c2f | 3034 | |
428870ff BB |
3035 | if (zio->io_flags & ZIO_FLAG_NODATA) { |
3036 | if (BP_IS_GANG(bp)) { | |
3037 | zio->io_flags &= ~ZIO_FLAG_NODATA; | |
3038 | } else { | |
3039 | ASSERT((uintptr_t)zio->io_data < SPA_MAXBLOCKSIZE); | |
3040 | zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; | |
3041 | } | |
3042 | } | |
3043 | ||
3044 | if (zio_injection_enabled && | |
3045 | zio->io_spa->spa_syncing_txg == zio->io_txg) | |
3046 | zio_handle_ignored_writes(zio); | |
3047 | ||
b128c09f | 3048 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
3049 | } |
3050 | ||
b128c09f BB |
3051 | static int |
3052 | zio_done(zio_t *zio) | |
34dc7c2f | 3053 | { |
d164b209 | 3054 | zio_t *pio, *pio_next; |
d6320ddb | 3055 | int c, w; |
34dc7c2f | 3056 | |
b128c09f | 3057 | /* |
9babb374 | 3058 | * If our children haven't all completed, |
b128c09f BB |
3059 | * wait for them and then repeat this pipeline stage. |
3060 | */ | |
3061 | if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE) || | |
3062 | zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE) || | |
428870ff | 3063 | zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE) || |
b128c09f BB |
3064 | zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_DONE)) |
3065 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 3066 | |
d6320ddb BB |
3067 | for (c = 0; c < ZIO_CHILD_TYPES; c++) |
3068 | for (w = 0; w < ZIO_WAIT_TYPES; w++) | |
b128c09f BB |
3069 | ASSERT(zio->io_children[c][w] == 0); |
3070 | ||
9b67f605 | 3071 | if (zio->io_bp != NULL && !BP_IS_EMBEDDED(zio->io_bp)) { |
c776b317 BB |
3072 | ASSERT(zio->io_bp->blk_pad[0] == 0); |
3073 | ASSERT(zio->io_bp->blk_pad[1] == 0); | |
d1d7e268 MK |
3074 | ASSERT(bcmp(zio->io_bp, &zio->io_bp_copy, |
3075 | sizeof (blkptr_t)) == 0 || | |
c776b317 BB |
3076 | (zio->io_bp == zio_unique_parent(zio)->io_bp)); |
3077 | if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(zio->io_bp) && | |
428870ff | 3078 | zio->io_bp_override == NULL && |
b128c09f | 3079 | !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) { |
c776b317 | 3080 | ASSERT(!BP_SHOULD_BYTESWAP(zio->io_bp)); |
d1d7e268 MK |
3081 | ASSERT3U(zio->io_prop.zp_copies, <=, |
3082 | BP_GET_NDVAS(zio->io_bp)); | |
c776b317 | 3083 | ASSERT(BP_COUNT_GANG(zio->io_bp) == 0 || |
d1d7e268 MK |
3084 | (BP_COUNT_GANG(zio->io_bp) == |
3085 | BP_GET_NDVAS(zio->io_bp))); | |
b128c09f | 3086 | } |
03c6040b GW |
3087 | if (zio->io_flags & ZIO_FLAG_NOPWRITE) |
3088 | VERIFY(BP_EQUAL(zio->io_bp, &zio->io_bp_orig)); | |
b128c09f BB |
3089 | } |
3090 | ||
3091 | /* | |
428870ff | 3092 | * If there were child vdev/gang/ddt errors, they apply to us now. |
b128c09f BB |
3093 | */ |
3094 | zio_inherit_child_errors(zio, ZIO_CHILD_VDEV); | |
3095 | zio_inherit_child_errors(zio, ZIO_CHILD_GANG); | |
428870ff BB |
3096 | zio_inherit_child_errors(zio, ZIO_CHILD_DDT); |
3097 | ||
3098 | /* | |
3099 | * If the I/O on the transformed data was successful, generate any | |
3100 | * checksum reports now while we still have the transformed data. | |
3101 | */ | |
3102 | if (zio->io_error == 0) { | |
3103 | while (zio->io_cksum_report != NULL) { | |
3104 | zio_cksum_report_t *zcr = zio->io_cksum_report; | |
3105 | uint64_t align = zcr->zcr_align; | |
c776b317 | 3106 | uint64_t asize = P2ROUNDUP(zio->io_size, align); |
428870ff BB |
3107 | char *abuf = zio->io_data; |
3108 | ||
c776b317 | 3109 | if (asize != zio->io_size) { |
428870ff | 3110 | abuf = zio_buf_alloc(asize); |
c776b317 | 3111 | bcopy(zio->io_data, abuf, zio->io_size); |
d1d7e268 | 3112 | bzero(abuf+zio->io_size, asize-zio->io_size); |
428870ff BB |
3113 | } |
3114 | ||
3115 | zio->io_cksum_report = zcr->zcr_next; | |
3116 | zcr->zcr_next = NULL; | |
3117 | zcr->zcr_finish(zcr, abuf); | |
3118 | zfs_ereport_free_checksum(zcr); | |
3119 | ||
c776b317 | 3120 | if (asize != zio->io_size) |
428870ff BB |
3121 | zio_buf_free(abuf, asize); |
3122 | } | |
3123 | } | |
b128c09f BB |
3124 | |
3125 | zio_pop_transforms(zio); /* note: may set zio->io_error */ | |
3126 | ||
c776b317 | 3127 | vdev_stat_update(zio, zio->io_size); |
b128c09f | 3128 | |
a69052be | 3129 | /* |
cc92e9d0 | 3130 | * If this I/O is attached to a particular vdev is slow, exceeding |
72f53c56 MJ |
3131 | * 30 seconds to complete, post an error described the I/O delay. |
3132 | * We ignore these errors if the device is currently unavailable. | |
a69052be | 3133 | */ |
cc92e9d0 | 3134 | if (zio->io_delay >= MSEC_TO_TICK(zio_delay_max)) { |
72f53c56 MJ |
3135 | if (zio->io_vd != NULL && !vdev_is_dead(zio->io_vd)) |
3136 | zfs_ereport_post(FM_EREPORT_ZFS_DELAY, zio->io_spa, | |
d1d7e268 | 3137 | zio->io_vd, zio, 0, 0); |
72f53c56 | 3138 | } |
a69052be | 3139 | |
b128c09f BB |
3140 | if (zio->io_error) { |
3141 | /* | |
3142 | * If this I/O is attached to a particular vdev, | |
3143 | * generate an error message describing the I/O failure | |
3144 | * at the block level. We ignore these errors if the | |
3145 | * device is currently unavailable. | |
3146 | */ | |
c776b317 BB |
3147 | if (zio->io_error != ECKSUM && zio->io_vd != NULL && |
3148 | !vdev_is_dead(zio->io_vd)) | |
3149 | zfs_ereport_post(FM_EREPORT_ZFS_IO, zio->io_spa, | |
3150 | zio->io_vd, zio, 0, 0); | |
34dc7c2f | 3151 | |
428870ff BB |
3152 | if ((zio->io_error == EIO || !(zio->io_flags & |
3153 | (ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) && | |
c776b317 | 3154 | zio == zio->io_logical) { |
b128c09f BB |
3155 | /* |
3156 | * For logical I/O requests, tell the SPA to log the | |
3157 | * error and generate a logical data ereport. | |
3158 | */ | |
c776b317 | 3159 | spa_log_error(zio->io_spa, zio); |
d1d7e268 MK |
3160 | zfs_ereport_post(FM_EREPORT_ZFS_DATA, zio->io_spa, |
3161 | NULL, zio, 0, 0); | |
b128c09f BB |
3162 | } |
3163 | } | |
34dc7c2f | 3164 | |
c776b317 | 3165 | if (zio->io_error && zio == zio->io_logical) { |
b128c09f BB |
3166 | /* |
3167 | * Determine whether zio should be reexecuted. This will | |
3168 | * propagate all the way to the root via zio_notify_parent(). | |
3169 | */ | |
c776b317 | 3170 | ASSERT(zio->io_vd == NULL && zio->io_bp != NULL); |
428870ff | 3171 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); |
b128c09f | 3172 | |
428870ff BB |
3173 | if (IO_IS_ALLOCATING(zio) && |
3174 | !(zio->io_flags & ZIO_FLAG_CANFAIL)) { | |
b128c09f BB |
3175 | if (zio->io_error != ENOSPC) |
3176 | zio->io_reexecute |= ZIO_REEXECUTE_NOW; | |
3177 | else | |
3178 | zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; | |
428870ff | 3179 | } |
b128c09f BB |
3180 | |
3181 | if ((zio->io_type == ZIO_TYPE_READ || | |
3182 | zio->io_type == ZIO_TYPE_FREE) && | |
572e2857 | 3183 | !(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && |
b128c09f | 3184 | zio->io_error == ENXIO && |
c776b317 BB |
3185 | spa_load_state(zio->io_spa) == SPA_LOAD_NONE && |
3186 | spa_get_failmode(zio->io_spa) != ZIO_FAILURE_MODE_CONTINUE) | |
b128c09f BB |
3187 | zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; |
3188 | ||
3189 | if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute) | |
3190 | zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; | |
428870ff BB |
3191 | |
3192 | /* | |
3193 | * Here is a possibly good place to attempt to do | |
3194 | * either combinatorial reconstruction or error correction | |
3195 | * based on checksums. It also might be a good place | |
3196 | * to send out preliminary ereports before we suspend | |
3197 | * processing. | |
3198 | */ | |
34dc7c2f BB |
3199 | } |
3200 | ||
3201 | /* | |
b128c09f BB |
3202 | * If there were logical child errors, they apply to us now. |
3203 | * We defer this until now to avoid conflating logical child | |
3204 | * errors with errors that happened to the zio itself when | |
3205 | * updating vdev stats and reporting FMA events above. | |
34dc7c2f | 3206 | */ |
b128c09f | 3207 | zio_inherit_child_errors(zio, ZIO_CHILD_LOGICAL); |
34dc7c2f | 3208 | |
428870ff BB |
3209 | if ((zio->io_error || zio->io_reexecute) && |
3210 | IO_IS_ALLOCATING(zio) && zio->io_gang_leader == zio && | |
03c6040b | 3211 | !(zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE))) |
c776b317 | 3212 | zio_dva_unallocate(zio, zio->io_gang_tree, zio->io_bp); |
9babb374 BB |
3213 | |
3214 | zio_gang_tree_free(&zio->io_gang_tree); | |
3215 | ||
3216 | /* | |
3217 | * Godfather I/Os should never suspend. | |
3218 | */ | |
3219 | if ((zio->io_flags & ZIO_FLAG_GODFATHER) && | |
3220 | (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) | |
3221 | zio->io_reexecute = 0; | |
3222 | ||
b128c09f BB |
3223 | if (zio->io_reexecute) { |
3224 | /* | |
3225 | * This is a logical I/O that wants to reexecute. | |
3226 | * | |
3227 | * Reexecute is top-down. When an i/o fails, if it's not | |
3228 | * the root, it simply notifies its parent and sticks around. | |
3229 | * The parent, seeing that it still has children in zio_done(), | |
3230 | * does the same. This percolates all the way up to the root. | |
3231 | * The root i/o will reexecute or suspend the entire tree. | |
3232 | * | |
3233 | * This approach ensures that zio_reexecute() honors | |
3234 | * all the original i/o dependency relationships, e.g. | |
3235 | * parents not executing until children are ready. | |
3236 | */ | |
3237 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
34dc7c2f | 3238 | |
9babb374 | 3239 | zio->io_gang_leader = NULL; |
b128c09f | 3240 | |
d164b209 BB |
3241 | mutex_enter(&zio->io_lock); |
3242 | zio->io_state[ZIO_WAIT_DONE] = 1; | |
3243 | mutex_exit(&zio->io_lock); | |
3244 | ||
9babb374 BB |
3245 | /* |
3246 | * "The Godfather" I/O monitors its children but is | |
3247 | * not a true parent to them. It will track them through | |
3248 | * the pipeline but severs its ties whenever they get into | |
3249 | * trouble (e.g. suspended). This allows "The Godfather" | |
3250 | * I/O to return status without blocking. | |
3251 | */ | |
3252 | for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { | |
3253 | zio_link_t *zl = zio->io_walk_link; | |
3254 | pio_next = zio_walk_parents(zio); | |
3255 | ||
3256 | if ((pio->io_flags & ZIO_FLAG_GODFATHER) && | |
3257 | (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) { | |
3258 | zio_remove_child(pio, zio, zl); | |
3259 | zio_notify_parent(pio, zio, ZIO_WAIT_DONE); | |
3260 | } | |
3261 | } | |
3262 | ||
d164b209 | 3263 | if ((pio = zio_unique_parent(zio)) != NULL) { |
b128c09f BB |
3264 | /* |
3265 | * We're not a root i/o, so there's nothing to do | |
3266 | * but notify our parent. Don't propagate errors | |
3267 | * upward since we haven't permanently failed yet. | |
3268 | */ | |
9babb374 | 3269 | ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); |
b128c09f BB |
3270 | zio->io_flags |= ZIO_FLAG_DONT_PROPAGATE; |
3271 | zio_notify_parent(pio, zio, ZIO_WAIT_DONE); | |
3272 | } else if (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND) { | |
3273 | /* | |
3274 | * We'd fail again if we reexecuted now, so suspend | |
3275 | * until conditions improve (e.g. device comes online). | |
3276 | */ | |
c776b317 | 3277 | zio_suspend(zio->io_spa, zio); |
b128c09f BB |
3278 | } else { |
3279 | /* | |
3280 | * Reexecution is potentially a huge amount of work. | |
3281 | * Hand it off to the otherwise-unused claim taskq. | |
3282 | */ | |
a38718a6 | 3283 | ASSERT(taskq_empty_ent(&zio->io_tqent)); |
7ef5e54e AL |
3284 | spa_taskq_dispatch_ent(zio->io_spa, |
3285 | ZIO_TYPE_CLAIM, ZIO_TASKQ_ISSUE, | |
a38718a6 GA |
3286 | (task_func_t *)zio_reexecute, zio, 0, |
3287 | &zio->io_tqent); | |
b128c09f BB |
3288 | } |
3289 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f BB |
3290 | } |
3291 | ||
428870ff | 3292 | ASSERT(zio->io_child_count == 0); |
b128c09f BB |
3293 | ASSERT(zio->io_reexecute == 0); |
3294 | ASSERT(zio->io_error == 0 || (zio->io_flags & ZIO_FLAG_CANFAIL)); | |
34dc7c2f | 3295 | |
428870ff BB |
3296 | /* |
3297 | * Report any checksum errors, since the I/O is complete. | |
3298 | */ | |
3299 | while (zio->io_cksum_report != NULL) { | |
3300 | zio_cksum_report_t *zcr = zio->io_cksum_report; | |
3301 | zio->io_cksum_report = zcr->zcr_next; | |
3302 | zcr->zcr_next = NULL; | |
3303 | zcr->zcr_finish(zcr, NULL); | |
3304 | zfs_ereport_free_checksum(zcr); | |
3305 | } | |
3306 | ||
920dd524 | 3307 | if (zio->io_flags & ZIO_FLAG_FASTWRITE && zio->io_bp && |
9b67f605 MA |
3308 | !BP_IS_HOLE(zio->io_bp) && !BP_IS_EMBEDDED(zio->io_bp) && |
3309 | !(zio->io_flags & ZIO_FLAG_NOPWRITE)) { | |
920dd524 ED |
3310 | metaslab_fastwrite_unmark(zio->io_spa, zio->io_bp); |
3311 | } | |
3312 | ||
d164b209 BB |
3313 | /* |
3314 | * It is the responsibility of the done callback to ensure that this | |
3315 | * particular zio is no longer discoverable for adoption, and as | |
3316 | * such, cannot acquire any new parents. | |
3317 | */ | |
b128c09f BB |
3318 | if (zio->io_done) |
3319 | zio->io_done(zio); | |
34dc7c2f | 3320 | |
d164b209 BB |
3321 | mutex_enter(&zio->io_lock); |
3322 | zio->io_state[ZIO_WAIT_DONE] = 1; | |
3323 | mutex_exit(&zio->io_lock); | |
34dc7c2f | 3324 | |
d164b209 BB |
3325 | for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { |
3326 | zio_link_t *zl = zio->io_walk_link; | |
3327 | pio_next = zio_walk_parents(zio); | |
3328 | zio_remove_child(pio, zio, zl); | |
b128c09f BB |
3329 | zio_notify_parent(pio, zio, ZIO_WAIT_DONE); |
3330 | } | |
34dc7c2f | 3331 | |
b128c09f BB |
3332 | if (zio->io_waiter != NULL) { |
3333 | mutex_enter(&zio->io_lock); | |
3334 | zio->io_executor = NULL; | |
3335 | cv_broadcast(&zio->io_cv); | |
3336 | mutex_exit(&zio->io_lock); | |
3337 | } else { | |
3338 | zio_destroy(zio); | |
3339 | } | |
34dc7c2f | 3340 | |
b128c09f | 3341 | return (ZIO_PIPELINE_STOP); |
34dc7c2f BB |
3342 | } |
3343 | ||
3344 | /* | |
b128c09f BB |
3345 | * ========================================================================== |
3346 | * I/O pipeline definition | |
3347 | * ========================================================================== | |
34dc7c2f | 3348 | */ |
428870ff | 3349 | static zio_pipe_stage_t *zio_pipeline[] = { |
b128c09f | 3350 | NULL, |
b128c09f | 3351 | zio_read_bp_init, |
428870ff BB |
3352 | zio_free_bp_init, |
3353 | zio_issue_async, | |
b128c09f BB |
3354 | zio_write_bp_init, |
3355 | zio_checksum_generate, | |
03c6040b | 3356 | zio_nop_write, |
428870ff BB |
3357 | zio_ddt_read_start, |
3358 | zio_ddt_read_done, | |
3359 | zio_ddt_write, | |
3360 | zio_ddt_free, | |
b128c09f BB |
3361 | zio_gang_assemble, |
3362 | zio_gang_issue, | |
3363 | zio_dva_allocate, | |
3364 | zio_dva_free, | |
3365 | zio_dva_claim, | |
3366 | zio_ready, | |
3367 | zio_vdev_io_start, | |
3368 | zio_vdev_io_done, | |
3369 | zio_vdev_io_assess, | |
3370 | zio_checksum_verify, | |
3371 | zio_done | |
3372 | }; | |
c28b2279 | 3373 | |
9ae529ec CS |
3374 | /* dnp is the dnode for zb1->zb_object */ |
3375 | boolean_t | |
5dbd68a3 MA |
3376 | zbookmark_is_before(const dnode_phys_t *dnp, const zbookmark_phys_t *zb1, |
3377 | const zbookmark_phys_t *zb2) | |
9ae529ec CS |
3378 | { |
3379 | uint64_t zb1nextL0, zb2thisobj; | |
3380 | ||
3381 | ASSERT(zb1->zb_objset == zb2->zb_objset); | |
3382 | ASSERT(zb2->zb_level == 0); | |
3383 | ||
9ae529ec CS |
3384 | /* The objset_phys_t isn't before anything. */ |
3385 | if (dnp == NULL) | |
3386 | return (B_FALSE); | |
3387 | ||
3388 | zb1nextL0 = (zb1->zb_blkid + 1) << | |
3389 | ((zb1->zb_level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)); | |
3390 | ||
3391 | zb2thisobj = zb2->zb_object ? zb2->zb_object : | |
3392 | zb2->zb_blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT); | |
3393 | ||
3394 | if (zb1->zb_object == DMU_META_DNODE_OBJECT) { | |
3395 | uint64_t nextobj = zb1nextL0 * | |
3396 | (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT) >> DNODE_SHIFT; | |
3397 | return (nextobj <= zb2thisobj); | |
3398 | } | |
3399 | ||
3400 | if (zb1->zb_object < zb2thisobj) | |
3401 | return (B_TRUE); | |
3402 | if (zb1->zb_object > zb2thisobj) | |
3403 | return (B_FALSE); | |
3404 | if (zb2->zb_object == DMU_META_DNODE_OBJECT) | |
3405 | return (B_FALSE); | |
3406 | return (zb1nextL0 <= zb2->zb_blkid); | |
3407 | } | |
3408 | ||
c28b2279 BB |
3409 | #if defined(_KERNEL) && defined(HAVE_SPL) |
3410 | /* Fault injection */ | |
3411 | EXPORT_SYMBOL(zio_injection_enabled); | |
3412 | EXPORT_SYMBOL(zio_inject_fault); | |
3413 | EXPORT_SYMBOL(zio_inject_list_next); | |
3414 | EXPORT_SYMBOL(zio_clear_fault); | |
3415 | EXPORT_SYMBOL(zio_handle_fault_injection); | |
3416 | EXPORT_SYMBOL(zio_handle_device_injection); | |
3417 | EXPORT_SYMBOL(zio_handle_label_injection); | |
c28b2279 BB |
3418 | EXPORT_SYMBOL(zio_type_name); |
3419 | ||
3420 | module_param(zio_bulk_flags, int, 0644); | |
3421 | MODULE_PARM_DESC(zio_bulk_flags, "Additional flags to pass to bulk buffers"); | |
a69052be BB |
3422 | |
3423 | module_param(zio_delay_max, int, 0644); | |
c409e464 BB |
3424 | MODULE_PARM_DESC(zio_delay_max, "Max zio millisec delay before posting event"); |
3425 | ||
3426 | module_param(zio_requeue_io_start_cut_in_line, int, 0644); | |
3427 | MODULE_PARM_DESC(zio_requeue_io_start_cut_in_line, "Prioritize requeued I/O"); | |
29dee3ee CP |
3428 | |
3429 | module_param(zfs_sync_pass_deferred_free, int, 0644); | |
3430 | MODULE_PARM_DESC(zfs_sync_pass_deferred_free, | |
d1d7e268 | 3431 | "Defer frees starting in this pass"); |
29dee3ee CP |
3432 | |
3433 | module_param(zfs_sync_pass_dont_compress, int, 0644); | |
3434 | MODULE_PARM_DESC(zfs_sync_pass_dont_compress, | |
d1d7e268 | 3435 | "Don't compress starting in this pass"); |
29dee3ee CP |
3436 | |
3437 | module_param(zfs_sync_pass_rewrite, int, 0644); | |
3438 | MODULE_PARM_DESC(zfs_sync_pass_rewrite, | |
d1d7e268 | 3439 | "Rewrite new bps starting in this pass"); |
c28b2279 | 3440 | #endif |