]>
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. |
2e528b49 | 23 | * Copyright (c) 2013 by Delphix. All rights reserved. |
a38718a6 | 24 | * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved. |
34dc7c2f BB |
25 | */ |
26 | ||
34dc7c2f BB |
27 | #include <sys/zfs_context.h> |
28 | #include <sys/fm/fs/zfs.h> | |
29 | #include <sys/spa.h> | |
30 | #include <sys/txg.h> | |
31 | #include <sys/spa_impl.h> | |
32 | #include <sys/vdev_impl.h> | |
33 | #include <sys/zio_impl.h> | |
34 | #include <sys/zio_compress.h> | |
35 | #include <sys/zio_checksum.h> | |
428870ff BB |
36 | #include <sys/dmu_objset.h> |
37 | #include <sys/arc.h> | |
38 | #include <sys/ddt.h> | |
34dc7c2f | 39 | |
34dc7c2f BB |
40 | /* |
41 | * ========================================================================== | |
42 | * I/O type descriptions | |
43 | * ========================================================================== | |
44 | */ | |
e8b96c60 | 45 | const char *zio_type_name[ZIO_TYPES] = { |
451041db | 46 | "z_null", "z_rd", "z_wr", "z_fr", "z_cl", "z_ioctl" |
428870ff | 47 | }; |
34dc7c2f BB |
48 | |
49 | /* | |
50 | * ========================================================================== | |
51 | * I/O kmem caches | |
52 | * ========================================================================== | |
53 | */ | |
54 | kmem_cache_t *zio_cache; | |
d164b209 | 55 | kmem_cache_t *zio_link_cache; |
86dd0fd9 | 56 | kmem_cache_t *zio_vdev_cache; |
34dc7c2f BB |
57 | kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; |
58 | kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; | |
c28b2279 | 59 | int zio_bulk_flags = 0; |
a69052be | 60 | int zio_delay_max = ZIO_DELAY_MAX; |
34dc7c2f | 61 | |
6d974228 | 62 | extern int zfs_mg_alloc_failures; |
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 BB |
131 | zio_cache = kmem_cache_create("zio_cache", sizeof (zio_t), 0, |
132 | zio_cons, zio_dest, NULL, NULL, NULL, KMC_KMEM); | |
d164b209 | 133 | zio_link_cache = kmem_cache_create("zio_link_cache", |
ae6ba3db | 134 | sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, KMC_KMEM); |
d1d7e268 | 135 | zio_vdev_cache = kmem_cache_create("zio_vdev_cache", sizeof (vdev_io_t), |
86dd0fd9 | 136 | PAGESIZE, NULL, NULL, NULL, NULL, NULL, KMC_VMEM); |
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 | ||
174 | /* | |
175 | * The smallest buffers (512b) are heavily used and | |
176 | * experience a lot of churn. The slabs allocated | |
177 | * for them are also relatively small (32K). Thus | |
178 | * in over to avoid expensive calls to vmalloc() we | |
179 | * make an exception to the usual slab allocation | |
180 | * policy and force these buffers to be kmem backed. | |
181 | */ | |
182 | if (size == (1 << SPA_MINBLOCKSHIFT)) | |
183 | flags |= KMC_KMEM; | |
184 | ||
34dc7c2f BB |
185 | (void) sprintf(name, "zio_buf_%lu", (ulong_t)size); |
186 | zio_buf_cache[c] = kmem_cache_create(name, size, | |
ae6ba3db | 187 | align, NULL, NULL, NULL, NULL, NULL, flags); |
34dc7c2f BB |
188 | |
189 | (void) sprintf(name, "zio_data_buf_%lu", (ulong_t)size); | |
190 | zio_data_buf_cache[c] = kmem_cache_create(name, size, | |
ae6ba3db BB |
191 | align, NULL, NULL, NULL, NULL, |
192 | data_alloc_arena, flags); | |
34dc7c2f BB |
193 | } |
194 | } | |
195 | ||
196 | while (--c != 0) { | |
197 | ASSERT(zio_buf_cache[c] != NULL); | |
198 | if (zio_buf_cache[c - 1] == NULL) | |
199 | zio_buf_cache[c - 1] = zio_buf_cache[c]; | |
200 | ||
201 | ASSERT(zio_data_buf_cache[c] != NULL); | |
202 | if (zio_data_buf_cache[c - 1] == NULL) | |
203 | zio_data_buf_cache[c - 1] = zio_data_buf_cache[c]; | |
204 | } | |
205 | ||
6d974228 GW |
206 | /* |
207 | * The zio write taskqs have 1 thread per cpu, allow 1/2 of the taskqs | |
208 | * to fail 3 times per txg or 8 failures, whichever is greater. | |
209 | */ | |
ac72fac3 GW |
210 | if (zfs_mg_alloc_failures == 0) |
211 | zfs_mg_alloc_failures = MAX((3 * max_ncpus / 2), 8); | |
6d974228 | 212 | |
34dc7c2f | 213 | zio_inject_init(); |
9759c60f ED |
214 | |
215 | lz4_init(); | |
34dc7c2f BB |
216 | } |
217 | ||
218 | void | |
219 | zio_fini(void) | |
220 | { | |
221 | size_t c; | |
222 | kmem_cache_t *last_cache = NULL; | |
223 | kmem_cache_t *last_data_cache = NULL; | |
224 | ||
225 | for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { | |
226 | if (zio_buf_cache[c] != last_cache) { | |
227 | last_cache = zio_buf_cache[c]; | |
228 | kmem_cache_destroy(zio_buf_cache[c]); | |
229 | } | |
230 | zio_buf_cache[c] = NULL; | |
231 | ||
232 | if (zio_data_buf_cache[c] != last_data_cache) { | |
233 | last_data_cache = zio_data_buf_cache[c]; | |
234 | kmem_cache_destroy(zio_data_buf_cache[c]); | |
235 | } | |
236 | zio_data_buf_cache[c] = NULL; | |
237 | } | |
238 | ||
86dd0fd9 | 239 | kmem_cache_destroy(zio_vdev_cache); |
d164b209 | 240 | kmem_cache_destroy(zio_link_cache); |
34dc7c2f BB |
241 | kmem_cache_destroy(zio_cache); |
242 | ||
243 | zio_inject_fini(); | |
9759c60f ED |
244 | |
245 | lz4_fini(); | |
34dc7c2f BB |
246 | } |
247 | ||
248 | /* | |
249 | * ========================================================================== | |
250 | * Allocate and free I/O buffers | |
251 | * ========================================================================== | |
252 | */ | |
253 | ||
254 | /* | |
255 | * Use zio_buf_alloc to allocate ZFS metadata. This data will appear in a | |
256 | * crashdump if the kernel panics, so use it judiciously. Obviously, it's | |
257 | * useful to inspect ZFS metadata, but if possible, we should avoid keeping | |
258 | * excess / transient data in-core during a crashdump. | |
259 | */ | |
260 | void * | |
261 | zio_buf_alloc(size_t size) | |
262 | { | |
263 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
264 | ||
265 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
266 | ||
ebcfc8a5 | 267 | return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE | KM_NODEBUG)); |
34dc7c2f BB |
268 | } |
269 | ||
270 | /* | |
271 | * Use zio_data_buf_alloc to allocate data. The data will not appear in a | |
272 | * crashdump if the kernel panics. This exists so that we will limit the amount | |
273 | * of ZFS data that shows up in a kernel crashdump. (Thus reducing the amount | |
274 | * of kernel heap dumped to disk when the kernel panics) | |
275 | */ | |
276 | void * | |
277 | zio_data_buf_alloc(size_t size) | |
278 | { | |
279 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
280 | ||
281 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
282 | ||
ebcfc8a5 BB |
283 | return (kmem_cache_alloc(zio_data_buf_cache[c], |
284 | KM_PUSHPAGE | KM_NODEBUG)); | |
34dc7c2f BB |
285 | } |
286 | ||
287 | void | |
288 | zio_buf_free(void *buf, size_t size) | |
289 | { | |
290 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
291 | ||
292 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
293 | ||
294 | kmem_cache_free(zio_buf_cache[c], buf); | |
295 | } | |
296 | ||
297 | void | |
298 | zio_data_buf_free(void *buf, size_t size) | |
299 | { | |
300 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
301 | ||
302 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
303 | ||
304 | kmem_cache_free(zio_data_buf_cache[c], buf); | |
305 | } | |
306 | ||
86dd0fd9 BB |
307 | /* |
308 | * Dedicated I/O buffers to ensure that memory fragmentation never prevents | |
309 | * or significantly delays the issuing of a zio. These buffers are used | |
310 | * to aggregate I/O and could be used for raidz stripes. | |
311 | */ | |
312 | void * | |
313 | zio_vdev_alloc(void) | |
314 | { | |
315 | return (kmem_cache_alloc(zio_vdev_cache, KM_PUSHPAGE)); | |
316 | } | |
317 | ||
318 | void | |
319 | zio_vdev_free(void *buf) | |
320 | { | |
321 | kmem_cache_free(zio_vdev_cache, buf); | |
322 | ||
323 | } | |
324 | ||
34dc7c2f BB |
325 | /* |
326 | * ========================================================================== | |
327 | * Push and pop I/O transform buffers | |
328 | * ========================================================================== | |
329 | */ | |
330 | static void | |
b128c09f BB |
331 | zio_push_transform(zio_t *zio, void *data, uint64_t size, uint64_t bufsize, |
332 | zio_transform_func_t *transform) | |
34dc7c2f | 333 | { |
691f6ac4 | 334 | zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_PUSHPAGE); |
34dc7c2f | 335 | |
b128c09f BB |
336 | zt->zt_orig_data = zio->io_data; |
337 | zt->zt_orig_size = zio->io_size; | |
34dc7c2f | 338 | zt->zt_bufsize = bufsize; |
b128c09f | 339 | zt->zt_transform = transform; |
34dc7c2f BB |
340 | |
341 | zt->zt_next = zio->io_transform_stack; | |
342 | zio->io_transform_stack = zt; | |
343 | ||
344 | zio->io_data = data; | |
345 | zio->io_size = size; | |
346 | } | |
347 | ||
348 | static void | |
b128c09f | 349 | zio_pop_transforms(zio_t *zio) |
34dc7c2f | 350 | { |
b128c09f BB |
351 | zio_transform_t *zt; |
352 | ||
353 | while ((zt = zio->io_transform_stack) != NULL) { | |
354 | if (zt->zt_transform != NULL) | |
355 | zt->zt_transform(zio, | |
356 | zt->zt_orig_data, zt->zt_orig_size); | |
34dc7c2f | 357 | |
428870ff BB |
358 | if (zt->zt_bufsize != 0) |
359 | zio_buf_free(zio->io_data, zt->zt_bufsize); | |
34dc7c2f | 360 | |
b128c09f BB |
361 | zio->io_data = zt->zt_orig_data; |
362 | zio->io_size = zt->zt_orig_size; | |
363 | zio->io_transform_stack = zt->zt_next; | |
34dc7c2f | 364 | |
b128c09f | 365 | kmem_free(zt, sizeof (zio_transform_t)); |
34dc7c2f BB |
366 | } |
367 | } | |
368 | ||
b128c09f BB |
369 | /* |
370 | * ========================================================================== | |
371 | * I/O transform callbacks for subblocks and decompression | |
372 | * ========================================================================== | |
373 | */ | |
374 | static void | |
375 | zio_subblock(zio_t *zio, void *data, uint64_t size) | |
376 | { | |
377 | ASSERT(zio->io_size > size); | |
378 | ||
379 | if (zio->io_type == ZIO_TYPE_READ) | |
380 | bcopy(zio->io_data, data, size); | |
381 | } | |
382 | ||
383 | static void | |
384 | zio_decompress(zio_t *zio, void *data, uint64_t size) | |
385 | { | |
386 | if (zio->io_error == 0 && | |
387 | zio_decompress_data(BP_GET_COMPRESS(zio->io_bp), | |
428870ff | 388 | zio->io_data, data, zio->io_size, size) != 0) |
2e528b49 | 389 | zio->io_error = SET_ERROR(EIO); |
b128c09f BB |
390 | } |
391 | ||
392 | /* | |
393 | * ========================================================================== | |
394 | * I/O parent/child relationships and pipeline interlocks | |
395 | * ========================================================================== | |
396 | */ | |
d164b209 BB |
397 | /* |
398 | * NOTE - Callers to zio_walk_parents() and zio_walk_children must | |
399 | * continue calling these functions until they return NULL. | |
400 | * Otherwise, the next caller will pick up the list walk in | |
401 | * some indeterminate state. (Otherwise every caller would | |
402 | * have to pass in a cookie to keep the state represented by | |
403 | * io_walk_link, which gets annoying.) | |
404 | */ | |
405 | zio_t * | |
406 | zio_walk_parents(zio_t *cio) | |
407 | { | |
408 | zio_link_t *zl = cio->io_walk_link; | |
409 | list_t *pl = &cio->io_parent_list; | |
b128c09f | 410 | |
d164b209 BB |
411 | zl = (zl == NULL) ? list_head(pl) : list_next(pl, zl); |
412 | cio->io_walk_link = zl; | |
413 | ||
414 | if (zl == NULL) | |
415 | return (NULL); | |
416 | ||
417 | ASSERT(zl->zl_child == cio); | |
418 | return (zl->zl_parent); | |
419 | } | |
420 | ||
421 | zio_t * | |
422 | zio_walk_children(zio_t *pio) | |
423 | { | |
424 | zio_link_t *zl = pio->io_walk_link; | |
425 | list_t *cl = &pio->io_child_list; | |
426 | ||
427 | zl = (zl == NULL) ? list_head(cl) : list_next(cl, zl); | |
428 | pio->io_walk_link = zl; | |
429 | ||
430 | if (zl == NULL) | |
431 | return (NULL); | |
432 | ||
433 | ASSERT(zl->zl_parent == pio); | |
434 | return (zl->zl_child); | |
435 | } | |
436 | ||
437 | zio_t * | |
438 | zio_unique_parent(zio_t *cio) | |
439 | { | |
440 | zio_t *pio = zio_walk_parents(cio); | |
441 | ||
442 | VERIFY(zio_walk_parents(cio) == NULL); | |
443 | return (pio); | |
444 | } | |
445 | ||
446 | void | |
447 | zio_add_child(zio_t *pio, zio_t *cio) | |
b128c09f | 448 | { |
691f6ac4 | 449 | zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_PUSHPAGE); |
d6320ddb | 450 | int w; |
d164b209 BB |
451 | |
452 | /* | |
453 | * Logical I/Os can have logical, gang, or vdev children. | |
454 | * Gang I/Os can have gang or vdev children. | |
455 | * Vdev I/Os can only have vdev children. | |
456 | * The following ASSERT captures all of these constraints. | |
457 | */ | |
458 | ASSERT(cio->io_child_type <= pio->io_child_type); | |
459 | ||
460 | zl->zl_parent = pio; | |
461 | zl->zl_child = cio; | |
462 | ||
463 | mutex_enter(&cio->io_lock); | |
b128c09f | 464 | mutex_enter(&pio->io_lock); |
d164b209 BB |
465 | |
466 | ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0); | |
467 | ||
d6320ddb | 468 | for (w = 0; w < ZIO_WAIT_TYPES; w++) |
d164b209 BB |
469 | pio->io_children[cio->io_child_type][w] += !cio->io_state[w]; |
470 | ||
471 | list_insert_head(&pio->io_child_list, zl); | |
472 | list_insert_head(&cio->io_parent_list, zl); | |
473 | ||
428870ff BB |
474 | pio->io_child_count++; |
475 | cio->io_parent_count++; | |
476 | ||
b128c09f | 477 | mutex_exit(&pio->io_lock); |
d164b209 | 478 | mutex_exit(&cio->io_lock); |
b128c09f BB |
479 | } |
480 | ||
34dc7c2f | 481 | static void |
d164b209 | 482 | zio_remove_child(zio_t *pio, zio_t *cio, zio_link_t *zl) |
b128c09f | 483 | { |
d164b209 BB |
484 | ASSERT(zl->zl_parent == pio); |
485 | ASSERT(zl->zl_child == cio); | |
b128c09f | 486 | |
d164b209 | 487 | mutex_enter(&cio->io_lock); |
b128c09f | 488 | mutex_enter(&pio->io_lock); |
d164b209 BB |
489 | |
490 | list_remove(&pio->io_child_list, zl); | |
491 | list_remove(&cio->io_parent_list, zl); | |
492 | ||
428870ff BB |
493 | pio->io_child_count--; |
494 | cio->io_parent_count--; | |
495 | ||
b128c09f | 496 | mutex_exit(&pio->io_lock); |
d164b209 BB |
497 | mutex_exit(&cio->io_lock); |
498 | ||
499 | kmem_cache_free(zio_link_cache, zl); | |
b128c09f BB |
500 | } |
501 | ||
502 | static boolean_t | |
503 | zio_wait_for_children(zio_t *zio, enum zio_child child, enum zio_wait_type wait) | |
34dc7c2f | 504 | { |
b128c09f BB |
505 | uint64_t *countp = &zio->io_children[child][wait]; |
506 | boolean_t waiting = B_FALSE; | |
507 | ||
508 | mutex_enter(&zio->io_lock); | |
509 | ASSERT(zio->io_stall == NULL); | |
510 | if (*countp != 0) { | |
428870ff | 511 | zio->io_stage >>= 1; |
b128c09f BB |
512 | zio->io_stall = countp; |
513 | waiting = B_TRUE; | |
514 | } | |
515 | mutex_exit(&zio->io_lock); | |
516 | ||
517 | return (waiting); | |
518 | } | |
34dc7c2f | 519 | |
bf701a83 BB |
520 | __attribute__((always_inline)) |
521 | static inline void | |
b128c09f BB |
522 | zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait) |
523 | { | |
524 | uint64_t *countp = &pio->io_children[zio->io_child_type][wait]; | |
525 | int *errorp = &pio->io_child_error[zio->io_child_type]; | |
34dc7c2f | 526 | |
b128c09f BB |
527 | mutex_enter(&pio->io_lock); |
528 | if (zio->io_error && !(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) | |
529 | *errorp = zio_worst_error(*errorp, zio->io_error); | |
530 | pio->io_reexecute |= zio->io_reexecute; | |
531 | ASSERT3U(*countp, >, 0); | |
e8b96c60 MA |
532 | |
533 | (*countp)--; | |
534 | ||
535 | if (*countp == 0 && pio->io_stall == countp) { | |
b128c09f BB |
536 | pio->io_stall = NULL; |
537 | mutex_exit(&pio->io_lock); | |
da6b4005 | 538 | __zio_execute(pio); |
b128c09f BB |
539 | } else { |
540 | mutex_exit(&pio->io_lock); | |
34dc7c2f BB |
541 | } |
542 | } | |
543 | ||
b128c09f BB |
544 | static void |
545 | zio_inherit_child_errors(zio_t *zio, enum zio_child c) | |
546 | { | |
547 | if (zio->io_child_error[c] != 0 && zio->io_error == 0) | |
548 | zio->io_error = zio->io_child_error[c]; | |
549 | } | |
550 | ||
34dc7c2f BB |
551 | /* |
552 | * ========================================================================== | |
b128c09f | 553 | * Create the various types of I/O (read, write, free, etc) |
34dc7c2f BB |
554 | * ========================================================================== |
555 | */ | |
556 | static zio_t * | |
428870ff | 557 | zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, |
34dc7c2f | 558 | void *data, uint64_t size, zio_done_func_t *done, void *private, |
e8b96c60 | 559 | zio_type_t type, zio_priority_t priority, enum zio_flag flags, |
428870ff BB |
560 | vdev_t *vd, uint64_t offset, const zbookmark_t *zb, |
561 | enum zio_stage stage, enum zio_stage pipeline) | |
34dc7c2f BB |
562 | { |
563 | zio_t *zio; | |
564 | ||
565 | ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); | |
566 | ASSERT(P2PHASE(size, SPA_MINBLOCKSIZE) == 0); | |
b128c09f BB |
567 | ASSERT(P2PHASE(offset, SPA_MINBLOCKSIZE) == 0); |
568 | ||
569 | ASSERT(!vd || spa_config_held(spa, SCL_STATE_ALL, RW_READER)); | |
570 | ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER)); | |
571 | ASSERT(vd || stage == ZIO_STAGE_OPEN); | |
34dc7c2f | 572 | |
691f6ac4 | 573 | zio = kmem_cache_alloc(zio_cache, KM_PUSHPAGE); |
d164b209 | 574 | |
b128c09f BB |
575 | if (vd != NULL) |
576 | zio->io_child_type = ZIO_CHILD_VDEV; | |
577 | else if (flags & ZIO_FLAG_GANG_CHILD) | |
578 | zio->io_child_type = ZIO_CHILD_GANG; | |
428870ff BB |
579 | else if (flags & ZIO_FLAG_DDT_CHILD) |
580 | zio->io_child_type = ZIO_CHILD_DDT; | |
b128c09f BB |
581 | else |
582 | zio->io_child_type = ZIO_CHILD_LOGICAL; | |
583 | ||
34dc7c2f | 584 | if (bp != NULL) { |
49be0ccf | 585 | zio->io_logical = NULL; |
428870ff | 586 | zio->io_bp = (blkptr_t *)bp; |
34dc7c2f BB |
587 | zio->io_bp_copy = *bp; |
588 | zio->io_bp_orig = *bp; | |
428870ff BB |
589 | if (type != ZIO_TYPE_WRITE || |
590 | zio->io_child_type == ZIO_CHILD_DDT) | |
b128c09f | 591 | zio->io_bp = &zio->io_bp_copy; /* so caller can free */ |
9babb374 | 592 | if (zio->io_child_type == ZIO_CHILD_LOGICAL) |
b128c09f | 593 | zio->io_logical = zio; |
9babb374 BB |
594 | if (zio->io_child_type > ZIO_CHILD_GANG && BP_IS_GANG(bp)) |
595 | pipeline |= ZIO_GANG_STAGES; | |
49be0ccf BB |
596 | } else { |
597 | zio->io_logical = NULL; | |
598 | zio->io_bp = NULL; | |
599 | bzero(&zio->io_bp_copy, sizeof (blkptr_t)); | |
600 | bzero(&zio->io_bp_orig, sizeof (blkptr_t)); | |
34dc7c2f | 601 | } |
b128c09f BB |
602 | |
603 | zio->io_spa = spa; | |
604 | zio->io_txg = txg; | |
49be0ccf | 605 | zio->io_ready = NULL; |
e8b96c60 | 606 | zio->io_physdone = NULL; |
34dc7c2f BB |
607 | zio->io_done = done; |
608 | zio->io_private = private; | |
49be0ccf | 609 | zio->io_prev_space_delta = 0; |
34dc7c2f BB |
610 | zio->io_type = type; |
611 | zio->io_priority = priority; | |
b128c09f | 612 | zio->io_vd = vd; |
49be0ccf BB |
613 | zio->io_vsd = NULL; |
614 | zio->io_vsd_ops = NULL; | |
b128c09f | 615 | zio->io_offset = offset; |
cc92e9d0 GW |
616 | zio->io_timestamp = 0; |
617 | zio->io_delta = 0; | |
618 | zio->io_delay = 0; | |
428870ff BB |
619 | zio->io_orig_data = zio->io_data = data; |
620 | zio->io_orig_size = zio->io_size = size; | |
b128c09f BB |
621 | zio->io_orig_flags = zio->io_flags = flags; |
622 | zio->io_orig_stage = zio->io_stage = stage; | |
623 | zio->io_orig_pipeline = zio->io_pipeline = pipeline; | |
49be0ccf BB |
624 | bzero(&zio->io_prop, sizeof (zio_prop_t)); |
625 | zio->io_cmd = 0; | |
626 | zio->io_reexecute = 0; | |
627 | zio->io_bp_override = NULL; | |
628 | zio->io_walk_link = NULL; | |
629 | zio->io_transform_stack = NULL; | |
49be0ccf BB |
630 | zio->io_error = 0; |
631 | zio->io_child_count = 0; | |
e8b96c60 | 632 | zio->io_phys_children = 0; |
49be0ccf BB |
633 | zio->io_parent_count = 0; |
634 | zio->io_stall = NULL; | |
635 | zio->io_gang_leader = NULL; | |
636 | zio->io_gang_tree = NULL; | |
637 | zio->io_executor = NULL; | |
638 | zio->io_waiter = NULL; | |
639 | zio->io_cksum_report = NULL; | |
640 | zio->io_ena = 0; | |
641 | bzero(zio->io_child_error, sizeof (int) * ZIO_CHILD_TYPES); | |
642 | bzero(zio->io_children, | |
643 | sizeof (uint64_t) * ZIO_CHILD_TYPES * ZIO_WAIT_TYPES); | |
644 | bzero(&zio->io_bookmark, sizeof (zbookmark_t)); | |
34dc7c2f | 645 | |
d164b209 BB |
646 | zio->io_state[ZIO_WAIT_READY] = (stage >= ZIO_STAGE_READY); |
647 | zio->io_state[ZIO_WAIT_DONE] = (stage >= ZIO_STAGE_DONE); | |
648 | ||
b128c09f BB |
649 | if (zb != NULL) |
650 | zio->io_bookmark = *zb; | |
651 | ||
652 | if (pio != NULL) { | |
b128c09f | 653 | if (zio->io_logical == NULL) |
34dc7c2f | 654 | zio->io_logical = pio->io_logical; |
9babb374 BB |
655 | if (zio->io_child_type == ZIO_CHILD_GANG) |
656 | zio->io_gang_leader = pio->io_gang_leader; | |
b128c09f | 657 | zio_add_child(pio, zio); |
34dc7c2f BB |
658 | } |
659 | ||
a38718a6 GA |
660 | taskq_init_ent(&zio->io_tqent); |
661 | ||
34dc7c2f BB |
662 | return (zio); |
663 | } | |
664 | ||
665 | static void | |
b128c09f | 666 | zio_destroy(zio_t *zio) |
34dc7c2f | 667 | { |
b128c09f | 668 | kmem_cache_free(zio_cache, zio); |
34dc7c2f BB |
669 | } |
670 | ||
671 | zio_t * | |
d164b209 | 672 | zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done, |
428870ff | 673 | void *private, enum zio_flag flags) |
34dc7c2f BB |
674 | { |
675 | zio_t *zio; | |
676 | ||
677 | zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, | |
d164b209 | 678 | ZIO_TYPE_NULL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, |
b128c09f | 679 | ZIO_STAGE_OPEN, ZIO_INTERLOCK_PIPELINE); |
34dc7c2f BB |
680 | |
681 | return (zio); | |
682 | } | |
683 | ||
684 | zio_t * | |
428870ff | 685 | zio_root(spa_t *spa, zio_done_func_t *done, void *private, enum zio_flag flags) |
34dc7c2f | 686 | { |
d164b209 | 687 | return (zio_null(NULL, spa, NULL, done, private, flags)); |
34dc7c2f BB |
688 | } |
689 | ||
690 | zio_t * | |
b128c09f BB |
691 | zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, |
692 | void *data, uint64_t size, zio_done_func_t *done, void *private, | |
e8b96c60 | 693 | zio_priority_t priority, enum zio_flag flags, const zbookmark_t *zb) |
34dc7c2f BB |
694 | { |
695 | zio_t *zio; | |
696 | ||
428870ff | 697 | zio = zio_create(pio, spa, BP_PHYSICAL_BIRTH(bp), bp, |
b128c09f BB |
698 | data, size, done, private, |
699 | ZIO_TYPE_READ, priority, flags, NULL, 0, zb, | |
428870ff BB |
700 | ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? |
701 | ZIO_DDT_CHILD_READ_PIPELINE : ZIO_READ_PIPELINE); | |
34dc7c2f | 702 | |
b128c09f BB |
703 | return (zio); |
704 | } | |
34dc7c2f | 705 | |
34dc7c2f | 706 | zio_t * |
b128c09f | 707 | zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, |
428870ff | 708 | void *data, uint64_t size, const zio_prop_t *zp, |
e8b96c60 MA |
709 | zio_done_func_t *ready, zio_done_func_t *physdone, zio_done_func_t *done, |
710 | void *private, | |
711 | zio_priority_t priority, enum zio_flag flags, const zbookmark_t *zb) | |
34dc7c2f BB |
712 | { |
713 | zio_t *zio; | |
714 | ||
b128c09f BB |
715 | ASSERT(zp->zp_checksum >= ZIO_CHECKSUM_OFF && |
716 | zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS && | |
717 | zp->zp_compress >= ZIO_COMPRESS_OFF && | |
718 | zp->zp_compress < ZIO_COMPRESS_FUNCTIONS && | |
9ae529ec | 719 | DMU_OT_IS_VALID(zp->zp_type) && |
b128c09f | 720 | zp->zp_level < 32 && |
428870ff | 721 | zp->zp_copies > 0 && |
03c6040b | 722 | zp->zp_copies <= spa_max_replication(spa)); |
34dc7c2f BB |
723 | |
724 | zio = zio_create(pio, spa, txg, bp, data, size, done, private, | |
b128c09f | 725 | ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, |
428870ff BB |
726 | ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? |
727 | ZIO_DDT_CHILD_WRITE_PIPELINE : ZIO_WRITE_PIPELINE); | |
34dc7c2f BB |
728 | |
729 | zio->io_ready = ready; | |
e8b96c60 | 730 | zio->io_physdone = physdone; |
b128c09f | 731 | zio->io_prop = *zp; |
34dc7c2f BB |
732 | |
733 | return (zio); | |
734 | } | |
735 | ||
736 | zio_t * | |
b128c09f | 737 | zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data, |
e8b96c60 MA |
738 | uint64_t size, zio_done_func_t *done, void *private, |
739 | zio_priority_t priority, enum zio_flag flags, zbookmark_t *zb) | |
34dc7c2f BB |
740 | { |
741 | zio_t *zio; | |
742 | ||
34dc7c2f | 743 | zio = zio_create(pio, spa, txg, bp, data, size, done, private, |
b128c09f BB |
744 | ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, |
745 | ZIO_STAGE_OPEN, ZIO_REWRITE_PIPELINE); | |
34dc7c2f BB |
746 | |
747 | return (zio); | |
748 | } | |
749 | ||
428870ff | 750 | void |
03c6040b | 751 | zio_write_override(zio_t *zio, blkptr_t *bp, int copies, boolean_t nopwrite) |
428870ff BB |
752 | { |
753 | ASSERT(zio->io_type == ZIO_TYPE_WRITE); | |
754 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
755 | ASSERT(zio->io_stage == ZIO_STAGE_OPEN); | |
756 | ASSERT(zio->io_txg == spa_syncing_txg(zio->io_spa)); | |
757 | ||
03c6040b GW |
758 | /* |
759 | * We must reset the io_prop to match the values that existed | |
760 | * when the bp was first written by dmu_sync() keeping in mind | |
761 | * that nopwrite and dedup are mutually exclusive. | |
762 | */ | |
763 | zio->io_prop.zp_dedup = nopwrite ? B_FALSE : zio->io_prop.zp_dedup; | |
764 | zio->io_prop.zp_nopwrite = nopwrite; | |
428870ff BB |
765 | zio->io_prop.zp_copies = copies; |
766 | zio->io_bp_override = bp; | |
767 | } | |
768 | ||
769 | void | |
770 | zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp) | |
771 | { | |
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 | dprintf_bp(bp, "freeing in txg %llu, pass %u", |
797 | (longlong_t)txg, spa->spa_sync_pass); | |
34dc7c2f | 798 | |
428870ff BB |
799 | ASSERT(!BP_IS_HOLE(bp)); |
800 | ASSERT(spa_syncing_txg(spa) == txg); | |
55d85d5a | 801 | ASSERT(spa_sync_pass(spa) < zfs_sync_pass_deferred_free); |
34dc7c2f | 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 | ||
827 | /* | |
828 | * A claim is an allocation of a specific block. Claims are needed | |
829 | * to support immediate writes in the intent log. The issue is that | |
830 | * immediate writes contain committed data, but in a txg that was | |
831 | * *not* committed. Upon opening the pool after an unclean shutdown, | |
832 | * the intent log claims all blocks that contain immediate write data | |
833 | * so that the SPA knows they're in use. | |
834 | * | |
835 | * All claims *must* be resolved in the first txg -- before the SPA | |
836 | * starts allocating blocks -- so that nothing is allocated twice. | |
428870ff | 837 | * If txg == 0 we just verify that the block is claimable. |
34dc7c2f BB |
838 | */ |
839 | ASSERT3U(spa->spa_uberblock.ub_rootbp.blk_birth, <, spa_first_txg(spa)); | |
428870ff BB |
840 | ASSERT(txg == spa_first_txg(spa) || txg == 0); |
841 | ASSERT(!BP_GET_DEDUP(bp) || !spa_writeable(spa)); /* zdb(1M) */ | |
34dc7c2f | 842 | |
b128c09f BB |
843 | zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), |
844 | done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags, | |
845 | NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE); | |
34dc7c2f BB |
846 | |
847 | return (zio); | |
848 | } | |
849 | ||
850 | zio_t * | |
851 | zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, | |
e8b96c60 | 852 | zio_done_func_t *done, void *private, enum zio_flag flags) |
34dc7c2f BB |
853 | { |
854 | zio_t *zio; | |
855 | int c; | |
856 | ||
857 | if (vd->vdev_children == 0) { | |
858 | zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, | |
e8b96c60 | 859 | ZIO_TYPE_IOCTL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, |
34dc7c2f BB |
860 | ZIO_STAGE_OPEN, ZIO_IOCTL_PIPELINE); |
861 | ||
34dc7c2f BB |
862 | zio->io_cmd = cmd; |
863 | } else { | |
d164b209 | 864 | zio = zio_null(pio, spa, NULL, NULL, NULL, flags); |
34dc7c2f BB |
865 | |
866 | for (c = 0; c < vd->vdev_children; c++) | |
867 | zio_nowait(zio_ioctl(zio, spa, vd->vdev_child[c], cmd, | |
e8b96c60 | 868 | done, private, flags)); |
34dc7c2f BB |
869 | } |
870 | ||
871 | return (zio); | |
872 | } | |
873 | ||
34dc7c2f BB |
874 | zio_t * |
875 | zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, | |
876 | void *data, int checksum, zio_done_func_t *done, void *private, | |
e8b96c60 | 877 | zio_priority_t priority, enum zio_flag flags, boolean_t labels) |
34dc7c2f BB |
878 | { |
879 | zio_t *zio; | |
34dc7c2f | 880 | |
b128c09f BB |
881 | ASSERT(vd->vdev_children == 0); |
882 | ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || | |
883 | offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); | |
884 | ASSERT3U(offset + size, <=, vd->vdev_psize); | |
34dc7c2f | 885 | |
b128c09f BB |
886 | zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, |
887 | ZIO_TYPE_READ, priority, flags, vd, offset, NULL, | |
34dc7c2f BB |
888 | ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE); |
889 | ||
b128c09f | 890 | zio->io_prop.zp_checksum = checksum; |
34dc7c2f BB |
891 | |
892 | return (zio); | |
893 | } | |
894 | ||
895 | zio_t * | |
896 | zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, | |
897 | void *data, int checksum, zio_done_func_t *done, void *private, | |
e8b96c60 | 898 | zio_priority_t priority, enum zio_flag flags, boolean_t labels) |
34dc7c2f | 899 | { |
34dc7c2f | 900 | zio_t *zio; |
34dc7c2f | 901 | |
b128c09f BB |
902 | ASSERT(vd->vdev_children == 0); |
903 | ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || | |
904 | offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); | |
905 | ASSERT3U(offset + size, <=, vd->vdev_psize); | |
34dc7c2f | 906 | |
b128c09f BB |
907 | zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, |
908 | ZIO_TYPE_WRITE, priority, flags, vd, offset, NULL, | |
34dc7c2f BB |
909 | ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE); |
910 | ||
b128c09f | 911 | zio->io_prop.zp_checksum = checksum; |
34dc7c2f | 912 | |
428870ff | 913 | if (zio_checksum_table[checksum].ci_eck) { |
34dc7c2f | 914 | /* |
428870ff | 915 | * zec checksums are necessarily destructive -- they modify |
b128c09f | 916 | * the end of the write buffer to hold the verifier/checksum. |
34dc7c2f | 917 | * Therefore, we must make a local copy in case the data is |
b128c09f | 918 | * being written to multiple places in parallel. |
34dc7c2f | 919 | */ |
b128c09f | 920 | void *wbuf = zio_buf_alloc(size); |
34dc7c2f | 921 | bcopy(data, wbuf, size); |
b128c09f | 922 | zio_push_transform(zio, wbuf, size, size, NULL); |
34dc7c2f BB |
923 | } |
924 | ||
925 | return (zio); | |
926 | } | |
927 | ||
928 | /* | |
b128c09f | 929 | * Create a child I/O to do some work for us. |
34dc7c2f BB |
930 | */ |
931 | zio_t * | |
b128c09f | 932 | zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset, |
e8b96c60 MA |
933 | void *data, uint64_t size, int type, zio_priority_t priority, |
934 | enum zio_flag flags, zio_done_func_t *done, void *private) | |
34dc7c2f | 935 | { |
428870ff | 936 | enum zio_stage pipeline = ZIO_VDEV_CHILD_PIPELINE; |
b128c09f BB |
937 | zio_t *zio; |
938 | ||
939 | ASSERT(vd->vdev_parent == | |
940 | (pio->io_vd ? pio->io_vd : pio->io_spa->spa_root_vdev)); | |
34dc7c2f BB |
941 | |
942 | if (type == ZIO_TYPE_READ && bp != NULL) { | |
943 | /* | |
944 | * If we have the bp, then the child should perform the | |
945 | * checksum and the parent need not. This pushes error | |
946 | * detection as close to the leaves as possible and | |
947 | * eliminates redundant checksums in the interior nodes. | |
948 | */ | |
428870ff BB |
949 | pipeline |= ZIO_STAGE_CHECKSUM_VERIFY; |
950 | pio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; | |
34dc7c2f BB |
951 | } |
952 | ||
b128c09f BB |
953 | if (vd->vdev_children == 0) |
954 | offset += VDEV_LABEL_START_SIZE; | |
955 | ||
428870ff BB |
956 | flags |= ZIO_VDEV_CHILD_FLAGS(pio) | ZIO_FLAG_DONT_PROPAGATE; |
957 | ||
958 | /* | |
959 | * If we've decided to do a repair, the write is not speculative -- | |
960 | * even if the original read was. | |
961 | */ | |
962 | if (flags & ZIO_FLAG_IO_REPAIR) | |
963 | flags &= ~ZIO_FLAG_SPECULATIVE; | |
964 | ||
b128c09f | 965 | zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size, |
428870ff BB |
966 | done, private, type, priority, flags, vd, offset, &pio->io_bookmark, |
967 | ZIO_STAGE_VDEV_IO_START >> 1, pipeline); | |
34dc7c2f | 968 | |
e8b96c60 MA |
969 | zio->io_physdone = pio->io_physdone; |
970 | if (vd->vdev_ops->vdev_op_leaf && zio->io_logical != NULL) | |
971 | zio->io_logical->io_phys_children++; | |
972 | ||
b128c09f | 973 | return (zio); |
34dc7c2f BB |
974 | } |
975 | ||
b128c09f BB |
976 | zio_t * |
977 | zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, void *data, uint64_t size, | |
e8b96c60 | 978 | int type, zio_priority_t priority, enum zio_flag flags, |
428870ff | 979 | zio_done_func_t *done, void *private) |
34dc7c2f | 980 | { |
b128c09f | 981 | zio_t *zio; |
34dc7c2f | 982 | |
b128c09f | 983 | ASSERT(vd->vdev_ops->vdev_op_leaf); |
34dc7c2f | 984 | |
b128c09f BB |
985 | zio = zio_create(NULL, vd->vdev_spa, 0, NULL, |
986 | data, size, done, private, type, priority, | |
e8b96c60 | 987 | flags | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY | ZIO_FLAG_DELEGATED, |
b128c09f | 988 | vd, offset, NULL, |
428870ff | 989 | ZIO_STAGE_VDEV_IO_START >> 1, ZIO_VDEV_CHILD_PIPELINE); |
34dc7c2f | 990 | |
b128c09f | 991 | return (zio); |
34dc7c2f BB |
992 | } |
993 | ||
994 | void | |
b128c09f | 995 | zio_flush(zio_t *zio, vdev_t *vd) |
34dc7c2f | 996 | { |
b128c09f | 997 | zio_nowait(zio_ioctl(zio, zio->io_spa, vd, DKIOCFLUSHWRITECACHE, |
e8b96c60 | 998 | NULL, NULL, |
b128c09f | 999 | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY)); |
34dc7c2f BB |
1000 | } |
1001 | ||
428870ff BB |
1002 | void |
1003 | zio_shrink(zio_t *zio, uint64_t size) | |
1004 | { | |
1005 | ASSERT(zio->io_executor == NULL); | |
1006 | ASSERT(zio->io_orig_size == zio->io_size); | |
1007 | ASSERT(size <= zio->io_size); | |
1008 | ||
1009 | /* | |
1010 | * We don't shrink for raidz because of problems with the | |
1011 | * reconstruction when reading back less than the block size. | |
1012 | * Note, BP_IS_RAIDZ() assumes no compression. | |
1013 | */ | |
1014 | ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); | |
1015 | if (!BP_IS_RAIDZ(zio->io_bp)) | |
1016 | zio->io_orig_size = zio->io_size = size; | |
1017 | } | |
1018 | ||
34dc7c2f BB |
1019 | /* |
1020 | * ========================================================================== | |
b128c09f | 1021 | * Prepare to read and write logical blocks |
34dc7c2f BB |
1022 | * ========================================================================== |
1023 | */ | |
b128c09f | 1024 | |
34dc7c2f | 1025 | static int |
b128c09f | 1026 | zio_read_bp_init(zio_t *zio) |
34dc7c2f | 1027 | { |
b128c09f | 1028 | blkptr_t *bp = zio->io_bp; |
34dc7c2f | 1029 | |
fb5f0bc8 | 1030 | if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF && |
9babb374 BB |
1031 | zio->io_child_type == ZIO_CHILD_LOGICAL && |
1032 | !(zio->io_flags & ZIO_FLAG_RAW)) { | |
428870ff BB |
1033 | uint64_t psize = BP_GET_PSIZE(bp); |
1034 | void *cbuf = zio_buf_alloc(psize); | |
b128c09f | 1035 | |
428870ff | 1036 | zio_push_transform(zio, cbuf, psize, psize, zio_decompress); |
34dc7c2f | 1037 | } |
34dc7c2f | 1038 | |
9ae529ec | 1039 | if (!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) && BP_GET_LEVEL(bp) == 0) |
b128c09f BB |
1040 | zio->io_flags |= ZIO_FLAG_DONT_CACHE; |
1041 | ||
428870ff BB |
1042 | if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP) |
1043 | zio->io_flags |= ZIO_FLAG_DONT_CACHE; | |
1044 | ||
1045 | if (BP_GET_DEDUP(bp) && zio->io_child_type == ZIO_CHILD_LOGICAL) | |
1046 | zio->io_pipeline = ZIO_DDT_READ_PIPELINE; | |
1047 | ||
b128c09f | 1048 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
1049 | } |
1050 | ||
b128c09f BB |
1051 | static int |
1052 | zio_write_bp_init(zio_t *zio) | |
34dc7c2f | 1053 | { |
428870ff | 1054 | spa_t *spa = zio->io_spa; |
b128c09f | 1055 | zio_prop_t *zp = &zio->io_prop; |
428870ff | 1056 | enum zio_compress compress = zp->zp_compress; |
34dc7c2f | 1057 | blkptr_t *bp = zio->io_bp; |
b128c09f | 1058 | uint64_t lsize = zio->io_size; |
428870ff | 1059 | uint64_t psize = lsize; |
b128c09f | 1060 | int pass = 1; |
34dc7c2f | 1061 | |
b128c09f BB |
1062 | /* |
1063 | * If our children haven't all reached the ready stage, | |
1064 | * wait for them and then repeat this pipeline stage. | |
1065 | */ | |
1066 | if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || | |
1067 | zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_READY)) | |
1068 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 1069 | |
b128c09f BB |
1070 | if (!IO_IS_ALLOCATING(zio)) |
1071 | return (ZIO_PIPELINE_CONTINUE); | |
34dc7c2f | 1072 | |
428870ff BB |
1073 | ASSERT(zio->io_child_type != ZIO_CHILD_DDT); |
1074 | ||
1075 | if (zio->io_bp_override) { | |
1076 | ASSERT(bp->blk_birth != zio->io_txg); | |
1077 | ASSERT(BP_GET_DEDUP(zio->io_bp_override) == 0); | |
1078 | ||
1079 | *bp = *zio->io_bp_override; | |
1080 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
1081 | ||
03c6040b GW |
1082 | /* |
1083 | * If we've been overridden and nopwrite is set then | |
1084 | * set the flag accordingly to indicate that a nopwrite | |
1085 | * has already occurred. | |
1086 | */ | |
1087 | if (!BP_IS_HOLE(bp) && zp->zp_nopwrite) { | |
1088 | ASSERT(!zp->zp_dedup); | |
1089 | zio->io_flags |= ZIO_FLAG_NOPWRITE; | |
1090 | return (ZIO_PIPELINE_CONTINUE); | |
1091 | } | |
1092 | ||
1093 | ASSERT(!zp->zp_nopwrite); | |
1094 | ||
428870ff BB |
1095 | if (BP_IS_HOLE(bp) || !zp->zp_dedup) |
1096 | return (ZIO_PIPELINE_CONTINUE); | |
1097 | ||
1098 | ASSERT(zio_checksum_table[zp->zp_checksum].ci_dedup || | |
1099 | zp->zp_dedup_verify); | |
1100 | ||
1101 | if (BP_GET_CHECKSUM(bp) == zp->zp_checksum) { | |
1102 | BP_SET_DEDUP(bp, 1); | |
1103 | zio->io_pipeline |= ZIO_STAGE_DDT_WRITE; | |
1104 | return (ZIO_PIPELINE_CONTINUE); | |
1105 | } | |
1106 | zio->io_bp_override = NULL; | |
1107 | BP_ZERO(bp); | |
1108 | } | |
34dc7c2f | 1109 | |
b128c09f BB |
1110 | if (bp->blk_birth == zio->io_txg) { |
1111 | /* | |
1112 | * We're rewriting an existing block, which means we're | |
1113 | * working on behalf of spa_sync(). For spa_sync() to | |
1114 | * converge, it must eventually be the case that we don't | |
1115 | * have to allocate new blocks. But compression changes | |
1116 | * the blocksize, which forces a reallocate, and makes | |
1117 | * convergence take longer. Therefore, after the first | |
1118 | * few passes, stop compressing to ensure convergence. | |
1119 | */ | |
428870ff BB |
1120 | pass = spa_sync_pass(spa); |
1121 | ||
1122 | ASSERT(zio->io_txg == spa_syncing_txg(spa)); | |
1123 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1124 | ASSERT(!BP_GET_DEDUP(bp)); | |
34dc7c2f | 1125 | |
55d85d5a | 1126 | if (pass >= zfs_sync_pass_dont_compress) |
b128c09f | 1127 | compress = ZIO_COMPRESS_OFF; |
34dc7c2f | 1128 | |
b128c09f | 1129 | /* Make sure someone doesn't change their mind on overwrites */ |
428870ff BB |
1130 | ASSERT(MIN(zp->zp_copies + BP_IS_GANG(bp), |
1131 | spa_max_replication(spa)) == BP_GET_NDVAS(bp)); | |
b128c09f | 1132 | } |
34dc7c2f | 1133 | |
b128c09f | 1134 | if (compress != ZIO_COMPRESS_OFF) { |
428870ff BB |
1135 | void *cbuf = zio_buf_alloc(lsize); |
1136 | psize = zio_compress_data(compress, zio->io_data, cbuf, lsize); | |
1137 | if (psize == 0 || psize == lsize) { | |
b128c09f | 1138 | compress = ZIO_COMPRESS_OFF; |
428870ff BB |
1139 | zio_buf_free(cbuf, lsize); |
1140 | } else { | |
1141 | ASSERT(psize < lsize); | |
1142 | zio_push_transform(zio, cbuf, psize, lsize, NULL); | |
b128c09f BB |
1143 | } |
1144 | } | |
34dc7c2f | 1145 | |
b128c09f BB |
1146 | /* |
1147 | * The final pass of spa_sync() must be all rewrites, but the first | |
1148 | * few passes offer a trade-off: allocating blocks defers convergence, | |
1149 | * but newly allocated blocks are sequential, so they can be written | |
1150 | * to disk faster. Therefore, we allow the first few passes of | |
1151 | * spa_sync() to allocate new blocks, but force rewrites after that. | |
1152 | * There should only be a handful of blocks after pass 1 in any case. | |
1153 | */ | |
428870ff | 1154 | if (bp->blk_birth == zio->io_txg && BP_GET_PSIZE(bp) == psize && |
55d85d5a | 1155 | pass >= zfs_sync_pass_rewrite) { |
428870ff | 1156 | enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES; |
d6320ddb | 1157 | ASSERT(psize != 0); |
b128c09f BB |
1158 | zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages; |
1159 | zio->io_flags |= ZIO_FLAG_IO_REWRITE; | |
1160 | } else { | |
1161 | BP_ZERO(bp); | |
1162 | zio->io_pipeline = ZIO_WRITE_PIPELINE; | |
1163 | } | |
34dc7c2f | 1164 | |
428870ff | 1165 | if (psize == 0) { |
b128c09f BB |
1166 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; |
1167 | } else { | |
1168 | ASSERT(zp->zp_checksum != ZIO_CHECKSUM_GANG_HEADER); | |
1169 | BP_SET_LSIZE(bp, lsize); | |
428870ff | 1170 | BP_SET_PSIZE(bp, psize); |
b128c09f BB |
1171 | BP_SET_COMPRESS(bp, compress); |
1172 | BP_SET_CHECKSUM(bp, zp->zp_checksum); | |
1173 | BP_SET_TYPE(bp, zp->zp_type); | |
1174 | BP_SET_LEVEL(bp, zp->zp_level); | |
428870ff | 1175 | BP_SET_DEDUP(bp, zp->zp_dedup); |
b128c09f | 1176 | BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); |
428870ff BB |
1177 | if (zp->zp_dedup) { |
1178 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1179 | ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); | |
1180 | zio->io_pipeline = ZIO_DDT_WRITE_PIPELINE; | |
1181 | } | |
03c6040b GW |
1182 | if (zp->zp_nopwrite) { |
1183 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1184 | ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); | |
1185 | zio->io_pipeline |= ZIO_STAGE_NOP_WRITE; | |
1186 | } | |
428870ff BB |
1187 | } |
1188 | ||
1189 | return (ZIO_PIPELINE_CONTINUE); | |
1190 | } | |
1191 | ||
1192 | static int | |
1193 | zio_free_bp_init(zio_t *zio) | |
1194 | { | |
1195 | blkptr_t *bp = zio->io_bp; | |
1196 | ||
1197 | if (zio->io_child_type == ZIO_CHILD_LOGICAL) { | |
1198 | if (BP_GET_DEDUP(bp)) | |
1199 | zio->io_pipeline = ZIO_DDT_FREE_PIPELINE; | |
b128c09f | 1200 | } |
34dc7c2f BB |
1201 | |
1202 | return (ZIO_PIPELINE_CONTINUE); | |
1203 | } | |
1204 | ||
b128c09f BB |
1205 | /* |
1206 | * ========================================================================== | |
1207 | * Execute the I/O pipeline | |
1208 | * ========================================================================== | |
1209 | */ | |
1210 | ||
1211 | static void | |
7ef5e54e | 1212 | zio_taskq_dispatch(zio_t *zio, zio_taskq_type_t q, boolean_t cutinline) |
34dc7c2f | 1213 | { |
428870ff | 1214 | spa_t *spa = zio->io_spa; |
b128c09f | 1215 | zio_type_t t = zio->io_type; |
a38718a6 | 1216 | int flags = (cutinline ? TQ_FRONT : 0); |
34dc7c2f BB |
1217 | |
1218 | /* | |
9babb374 BB |
1219 | * If we're a config writer or a probe, the normal issue and |
1220 | * interrupt threads may all be blocked waiting for the config lock. | |
1221 | * In this case, select the otherwise-unused taskq for ZIO_TYPE_NULL. | |
34dc7c2f | 1222 | */ |
9babb374 | 1223 | if (zio->io_flags & (ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_PROBE)) |
b128c09f | 1224 | t = ZIO_TYPE_NULL; |
34dc7c2f BB |
1225 | |
1226 | /* | |
b128c09f | 1227 | * A similar issue exists for the L2ARC write thread until L2ARC 2.0. |
34dc7c2f | 1228 | */ |
b128c09f BB |
1229 | if (t == ZIO_TYPE_WRITE && zio->io_vd && zio->io_vd->vdev_aux) |
1230 | t = ZIO_TYPE_NULL; | |
34dc7c2f | 1231 | |
428870ff | 1232 | /* |
7ef5e54e AL |
1233 | * If this is a high priority I/O, then use the high priority taskq if |
1234 | * available. | |
428870ff BB |
1235 | */ |
1236 | if (zio->io_priority == ZIO_PRIORITY_NOW && | |
7ef5e54e | 1237 | spa->spa_zio_taskq[t][q + 1].stqs_count != 0) |
428870ff BB |
1238 | q++; |
1239 | ||
1240 | ASSERT3U(q, <, ZIO_TASKQ_TYPES); | |
5cc556b4 | 1241 | |
a38718a6 GA |
1242 | /* |
1243 | * NB: We are assuming that the zio can only be dispatched | |
1244 | * to a single taskq at a time. It would be a grievous error | |
1245 | * to dispatch the zio to another taskq at the same time. | |
1246 | */ | |
1247 | ASSERT(taskq_empty_ent(&zio->io_tqent)); | |
7ef5e54e AL |
1248 | spa_taskq_dispatch_ent(spa, t, q, (task_func_t *)zio_execute, zio, |
1249 | flags, &zio->io_tqent); | |
b128c09f | 1250 | } |
34dc7c2f | 1251 | |
b128c09f | 1252 | static boolean_t |
7ef5e54e | 1253 | zio_taskq_member(zio_t *zio, zio_taskq_type_t q) |
b128c09f BB |
1254 | { |
1255 | kthread_t *executor = zio->io_executor; | |
1256 | spa_t *spa = zio->io_spa; | |
d6320ddb | 1257 | zio_type_t t; |
34dc7c2f | 1258 | |
7ef5e54e AL |
1259 | for (t = 0; t < ZIO_TYPES; t++) { |
1260 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
1261 | uint_t i; | |
1262 | for (i = 0; i < tqs->stqs_count; i++) { | |
1263 | if (taskq_member(tqs->stqs_taskq[i], executor)) | |
1264 | return (B_TRUE); | |
1265 | } | |
1266 | } | |
34dc7c2f | 1267 | |
b128c09f BB |
1268 | return (B_FALSE); |
1269 | } | |
34dc7c2f | 1270 | |
b128c09f BB |
1271 | static int |
1272 | zio_issue_async(zio_t *zio) | |
1273 | { | |
428870ff | 1274 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); |
b128c09f BB |
1275 | |
1276 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f BB |
1277 | } |
1278 | ||
b128c09f BB |
1279 | void |
1280 | zio_interrupt(zio_t *zio) | |
34dc7c2f | 1281 | { |
428870ff | 1282 | zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT, B_FALSE); |
b128c09f | 1283 | } |
34dc7c2f | 1284 | |
b128c09f BB |
1285 | /* |
1286 | * Execute the I/O pipeline until one of the following occurs: | |
1287 | * (1) the I/O completes; (2) the pipeline stalls waiting for | |
1288 | * dependent child I/Os; (3) the I/O issues, so we're waiting | |
1289 | * for an I/O completion interrupt; (4) the I/O is delegated by | |
1290 | * vdev-level caching or aggregation; (5) the I/O is deferred | |
1291 | * due to vdev-level queueing; (6) the I/O is handed off to | |
1292 | * another thread. In all cases, the pipeline stops whenever | |
8e07b99b | 1293 | * there's no CPU work; it never burns a thread in cv_wait_io(). |
b128c09f BB |
1294 | * |
1295 | * There's no locking on io_stage because there's no legitimate way | |
1296 | * for multiple threads to be attempting to process the same I/O. | |
1297 | */ | |
428870ff | 1298 | static zio_pipe_stage_t *zio_pipeline[]; |
34dc7c2f | 1299 | |
da6b4005 NB |
1300 | /* |
1301 | * zio_execute() is a wrapper around the static function | |
1302 | * __zio_execute() so that we can force __zio_execute() to be | |
1303 | * inlined. This reduces stack overhead which is important | |
1304 | * because __zio_execute() is called recursively in several zio | |
1305 | * code paths. zio_execute() itself cannot be inlined because | |
1306 | * it is externally visible. | |
1307 | */ | |
b128c09f BB |
1308 | void |
1309 | zio_execute(zio_t *zio) | |
da6b4005 NB |
1310 | { |
1311 | __zio_execute(zio); | |
1312 | } | |
1313 | ||
1314 | __attribute__((always_inline)) | |
1315 | static inline void | |
1316 | __zio_execute(zio_t *zio) | |
b128c09f BB |
1317 | { |
1318 | zio->io_executor = curthread; | |
34dc7c2f | 1319 | |
b128c09f | 1320 | while (zio->io_stage < ZIO_STAGE_DONE) { |
428870ff BB |
1321 | enum zio_stage pipeline = zio->io_pipeline; |
1322 | enum zio_stage stage = zio->io_stage; | |
91579709 | 1323 | dsl_pool_t *dp; |
2fac4c2a | 1324 | boolean_t cut; |
b128c09f | 1325 | int rv; |
34dc7c2f | 1326 | |
b128c09f | 1327 | ASSERT(!MUTEX_HELD(&zio->io_lock)); |
428870ff BB |
1328 | ASSERT(ISP2(stage)); |
1329 | ASSERT(zio->io_stall == NULL); | |
34dc7c2f | 1330 | |
428870ff BB |
1331 | do { |
1332 | stage <<= 1; | |
1333 | } while ((stage & pipeline) == 0); | |
b128c09f BB |
1334 | |
1335 | ASSERT(stage <= ZIO_STAGE_DONE); | |
34dc7c2f | 1336 | |
91579709 | 1337 | dp = spa_get_dsl(zio->io_spa); |
2fac4c2a BB |
1338 | cut = (stage == ZIO_STAGE_VDEV_IO_START) ? |
1339 | zio_requeue_io_start_cut_in_line : B_FALSE; | |
1340 | ||
34dc7c2f | 1341 | /* |
b128c09f BB |
1342 | * If we are in interrupt context and this pipeline stage |
1343 | * will grab a config lock that is held across I/O, | |
428870ff BB |
1344 | * or may wait for an I/O that needs an interrupt thread |
1345 | * to complete, issue async to avoid deadlock. | |
1346 | * | |
1347 | * For VDEV_IO_START, we cut in line so that the io will | |
1348 | * be sent to disk promptly. | |
34dc7c2f | 1349 | */ |
91579709 BB |
1350 | if ((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL && |
1351 | zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) { | |
1352 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); | |
1353 | return; | |
1354 | } | |
1355 | ||
4cec9b2d | 1356 | #ifdef _KERNEL |
91579709 BB |
1357 | /* |
1358 | * If we executing in the context of the tx_sync_thread, | |
1359 | * or we are performing pool initialization outside of a | |
34e14332 BB |
1360 | * zio_taskq[ZIO_TASKQ_ISSUE|ZIO_TASKQ_ISSUE_HIGH] context. |
1361 | * Then issue the zio asynchronously to minimize stack usage | |
1362 | * for these deep call paths. | |
91579709 BB |
1363 | */ |
1364 | if ((dp && curthread == dp->dp_tx.tx_sync_thread) || | |
1365 | (dp && spa_is_initializing(dp->dp_spa) && | |
34e14332 BB |
1366 | !zio_taskq_member(zio, ZIO_TASKQ_ISSUE) && |
1367 | !zio_taskq_member(zio, ZIO_TASKQ_ISSUE_HIGH))) { | |
428870ff | 1368 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); |
b128c09f | 1369 | return; |
34dc7c2f | 1370 | } |
4cec9b2d | 1371 | #endif |
34dc7c2f | 1372 | |
b128c09f | 1373 | zio->io_stage = stage; |
428870ff | 1374 | rv = zio_pipeline[highbit(stage) - 1](zio); |
34dc7c2f | 1375 | |
b128c09f BB |
1376 | if (rv == ZIO_PIPELINE_STOP) |
1377 | return; | |
34dc7c2f | 1378 | |
b128c09f BB |
1379 | ASSERT(rv == ZIO_PIPELINE_CONTINUE); |
1380 | } | |
34dc7c2f BB |
1381 | } |
1382 | ||
da6b4005 | 1383 | |
b128c09f BB |
1384 | /* |
1385 | * ========================================================================== | |
1386 | * Initiate I/O, either sync or async | |
1387 | * ========================================================================== | |
1388 | */ | |
1389 | int | |
1390 | zio_wait(zio_t *zio) | |
34dc7c2f | 1391 | { |
b128c09f | 1392 | int error; |
34dc7c2f | 1393 | |
b128c09f BB |
1394 | ASSERT(zio->io_stage == ZIO_STAGE_OPEN); |
1395 | ASSERT(zio->io_executor == NULL); | |
34dc7c2f | 1396 | |
b128c09f | 1397 | zio->io_waiter = curthread; |
34dc7c2f | 1398 | |
da6b4005 | 1399 | __zio_execute(zio); |
34dc7c2f | 1400 | |
b128c09f | 1401 | mutex_enter(&zio->io_lock); |
72f53c56 | 1402 | while (zio->io_executor != NULL) |
72938d69 | 1403 | cv_wait_io(&zio->io_cv, &zio->io_lock); |
b128c09f | 1404 | mutex_exit(&zio->io_lock); |
34dc7c2f | 1405 | |
b128c09f BB |
1406 | error = zio->io_error; |
1407 | zio_destroy(zio); | |
34dc7c2f | 1408 | |
b128c09f BB |
1409 | return (error); |
1410 | } | |
34dc7c2f | 1411 | |
b128c09f BB |
1412 | void |
1413 | zio_nowait(zio_t *zio) | |
1414 | { | |
1415 | ASSERT(zio->io_executor == NULL); | |
34dc7c2f | 1416 | |
d164b209 BB |
1417 | if (zio->io_child_type == ZIO_CHILD_LOGICAL && |
1418 | zio_unique_parent(zio) == NULL) { | |
34dc7c2f | 1419 | /* |
b128c09f | 1420 | * This is a logical async I/O with no parent to wait for it. |
9babb374 BB |
1421 | * We add it to the spa_async_root_zio "Godfather" I/O which |
1422 | * will ensure they complete prior to unloading the pool. | |
34dc7c2f | 1423 | */ |
b128c09f | 1424 | spa_t *spa = zio->io_spa; |
9babb374 BB |
1425 | |
1426 | zio_add_child(spa->spa_async_zio_root, zio); | |
b128c09f | 1427 | } |
34dc7c2f | 1428 | |
da6b4005 | 1429 | __zio_execute(zio); |
b128c09f | 1430 | } |
34dc7c2f | 1431 | |
b128c09f BB |
1432 | /* |
1433 | * ========================================================================== | |
1434 | * Reexecute or suspend/resume failed I/O | |
1435 | * ========================================================================== | |
1436 | */ | |
34dc7c2f | 1437 | |
b128c09f BB |
1438 | static void |
1439 | zio_reexecute(zio_t *pio) | |
1440 | { | |
d164b209 | 1441 | zio_t *cio, *cio_next; |
d6320ddb | 1442 | int c, w; |
d164b209 BB |
1443 | |
1444 | ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL); | |
1445 | ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN); | |
9babb374 BB |
1446 | ASSERT(pio->io_gang_leader == NULL); |
1447 | ASSERT(pio->io_gang_tree == NULL); | |
34dc7c2f | 1448 | |
b128c09f BB |
1449 | pio->io_flags = pio->io_orig_flags; |
1450 | pio->io_stage = pio->io_orig_stage; | |
1451 | pio->io_pipeline = pio->io_orig_pipeline; | |
1452 | pio->io_reexecute = 0; | |
03c6040b | 1453 | pio->io_flags |= ZIO_FLAG_REEXECUTED; |
b128c09f | 1454 | pio->io_error = 0; |
d6320ddb | 1455 | for (w = 0; w < ZIO_WAIT_TYPES; w++) |
d164b209 | 1456 | pio->io_state[w] = 0; |
d6320ddb | 1457 | for (c = 0; c < ZIO_CHILD_TYPES; c++) |
b128c09f | 1458 | pio->io_child_error[c] = 0; |
34dc7c2f | 1459 | |
428870ff BB |
1460 | if (IO_IS_ALLOCATING(pio)) |
1461 | BP_ZERO(pio->io_bp); | |
34dc7c2f | 1462 | |
b128c09f BB |
1463 | /* |
1464 | * As we reexecute pio's children, new children could be created. | |
d164b209 | 1465 | * New children go to the head of pio's io_child_list, however, |
b128c09f | 1466 | * so we will (correctly) not reexecute them. The key is that |
d164b209 BB |
1467 | * the remainder of pio's io_child_list, from 'cio_next' onward, |
1468 | * cannot be affected by any side effects of reexecuting 'cio'. | |
b128c09f | 1469 | */ |
d164b209 BB |
1470 | for (cio = zio_walk_children(pio); cio != NULL; cio = cio_next) { |
1471 | cio_next = zio_walk_children(pio); | |
b128c09f | 1472 | mutex_enter(&pio->io_lock); |
d6320ddb | 1473 | for (w = 0; w < ZIO_WAIT_TYPES; w++) |
d164b209 | 1474 | pio->io_children[cio->io_child_type][w]++; |
b128c09f | 1475 | mutex_exit(&pio->io_lock); |
d164b209 | 1476 | zio_reexecute(cio); |
34dc7c2f | 1477 | } |
34dc7c2f | 1478 | |
b128c09f BB |
1479 | /* |
1480 | * Now that all children have been reexecuted, execute the parent. | |
9babb374 BB |
1481 | * We don't reexecute "The Godfather" I/O here as it's the |
1482 | * responsibility of the caller to wait on him. | |
b128c09f | 1483 | */ |
9babb374 | 1484 | if (!(pio->io_flags & ZIO_FLAG_GODFATHER)) |
da6b4005 | 1485 | __zio_execute(pio); |
34dc7c2f BB |
1486 | } |
1487 | ||
b128c09f BB |
1488 | void |
1489 | zio_suspend(spa_t *spa, zio_t *zio) | |
34dc7c2f | 1490 | { |
b128c09f BB |
1491 | if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC) |
1492 | fm_panic("Pool '%s' has encountered an uncorrectable I/O " | |
1493 | "failure and the failure mode property for this pool " | |
1494 | "is set to panic.", spa_name(spa)); | |
34dc7c2f | 1495 | |
bf89c199 BB |
1496 | cmn_err(CE_WARN, "Pool '%s' has encountered an uncorrectable I/O " |
1497 | "failure and has been suspended.\n", spa_name(spa)); | |
1498 | ||
b128c09f | 1499 | zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, 0, 0); |
34dc7c2f | 1500 | |
b128c09f | 1501 | mutex_enter(&spa->spa_suspend_lock); |
34dc7c2f | 1502 | |
b128c09f | 1503 | if (spa->spa_suspend_zio_root == NULL) |
9babb374 BB |
1504 | spa->spa_suspend_zio_root = zio_root(spa, NULL, NULL, |
1505 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
1506 | ZIO_FLAG_GODFATHER); | |
34dc7c2f | 1507 | |
b128c09f | 1508 | spa->spa_suspended = B_TRUE; |
34dc7c2f | 1509 | |
b128c09f | 1510 | if (zio != NULL) { |
9babb374 | 1511 | ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); |
b128c09f BB |
1512 | ASSERT(zio != spa->spa_suspend_zio_root); |
1513 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
d164b209 | 1514 | ASSERT(zio_unique_parent(zio) == NULL); |
b128c09f BB |
1515 | ASSERT(zio->io_stage == ZIO_STAGE_DONE); |
1516 | zio_add_child(spa->spa_suspend_zio_root, zio); | |
1517 | } | |
34dc7c2f | 1518 | |
b128c09f BB |
1519 | mutex_exit(&spa->spa_suspend_lock); |
1520 | } | |
34dc7c2f | 1521 | |
9babb374 | 1522 | int |
b128c09f BB |
1523 | zio_resume(spa_t *spa) |
1524 | { | |
9babb374 | 1525 | zio_t *pio; |
34dc7c2f BB |
1526 | |
1527 | /* | |
b128c09f | 1528 | * Reexecute all previously suspended i/o. |
34dc7c2f | 1529 | */ |
b128c09f BB |
1530 | mutex_enter(&spa->spa_suspend_lock); |
1531 | spa->spa_suspended = B_FALSE; | |
1532 | cv_broadcast(&spa->spa_suspend_cv); | |
1533 | pio = spa->spa_suspend_zio_root; | |
1534 | spa->spa_suspend_zio_root = NULL; | |
1535 | mutex_exit(&spa->spa_suspend_lock); | |
1536 | ||
1537 | if (pio == NULL) | |
9babb374 | 1538 | return (0); |
34dc7c2f | 1539 | |
9babb374 BB |
1540 | zio_reexecute(pio); |
1541 | return (zio_wait(pio)); | |
b128c09f BB |
1542 | } |
1543 | ||
1544 | void | |
1545 | zio_resume_wait(spa_t *spa) | |
1546 | { | |
1547 | mutex_enter(&spa->spa_suspend_lock); | |
1548 | while (spa_suspended(spa)) | |
1549 | cv_wait(&spa->spa_suspend_cv, &spa->spa_suspend_lock); | |
1550 | mutex_exit(&spa->spa_suspend_lock); | |
34dc7c2f BB |
1551 | } |
1552 | ||
1553 | /* | |
1554 | * ========================================================================== | |
b128c09f BB |
1555 | * Gang blocks. |
1556 | * | |
1557 | * A gang block is a collection of small blocks that looks to the DMU | |
1558 | * like one large block. When zio_dva_allocate() cannot find a block | |
1559 | * of the requested size, due to either severe fragmentation or the pool | |
1560 | * being nearly full, it calls zio_write_gang_block() to construct the | |
1561 | * block from smaller fragments. | |
1562 | * | |
1563 | * A gang block consists of a gang header (zio_gbh_phys_t) and up to | |
1564 | * three (SPA_GBH_NBLKPTRS) gang members. The gang header is just like | |
1565 | * an indirect block: it's an array of block pointers. It consumes | |
1566 | * only one sector and hence is allocatable regardless of fragmentation. | |
1567 | * The gang header's bps point to its gang members, which hold the data. | |
1568 | * | |
1569 | * Gang blocks are self-checksumming, using the bp's <vdev, offset, txg> | |
1570 | * as the verifier to ensure uniqueness of the SHA256 checksum. | |
1571 | * Critically, the gang block bp's blk_cksum is the checksum of the data, | |
1572 | * not the gang header. This ensures that data block signatures (needed for | |
1573 | * deduplication) are independent of how the block is physically stored. | |
1574 | * | |
1575 | * Gang blocks can be nested: a gang member may itself be a gang block. | |
1576 | * Thus every gang block is a tree in which root and all interior nodes are | |
1577 | * gang headers, and the leaves are normal blocks that contain user data. | |
1578 | * The root of the gang tree is called the gang leader. | |
1579 | * | |
1580 | * To perform any operation (read, rewrite, free, claim) on a gang block, | |
1581 | * zio_gang_assemble() first assembles the gang tree (minus data leaves) | |
1582 | * in the io_gang_tree field of the original logical i/o by recursively | |
1583 | * reading the gang leader and all gang headers below it. This yields | |
1584 | * an in-core tree containing the contents of every gang header and the | |
1585 | * bps for every constituent of the gang block. | |
1586 | * | |
1587 | * With the gang tree now assembled, zio_gang_issue() just walks the gang tree | |
1588 | * and invokes a callback on each bp. To free a gang block, zio_gang_issue() | |
1589 | * calls zio_free_gang() -- a trivial wrapper around zio_free() -- for each bp. | |
1590 | * zio_claim_gang() provides a similarly trivial wrapper for zio_claim(). | |
1591 | * zio_read_gang() is a wrapper around zio_read() that omits reading gang | |
1592 | * headers, since we already have those in io_gang_tree. zio_rewrite_gang() | |
1593 | * performs a zio_rewrite() of the data or, for gang headers, a zio_rewrite() | |
1594 | * of the gang header plus zio_checksum_compute() of the data to update the | |
1595 | * gang header's blk_cksum as described above. | |
1596 | * | |
1597 | * The two-phase assemble/issue model solves the problem of partial failure -- | |
1598 | * what if you'd freed part of a gang block but then couldn't read the | |
1599 | * gang header for another part? Assembling the entire gang tree first | |
1600 | * ensures that all the necessary gang header I/O has succeeded before | |
1601 | * starting the actual work of free, claim, or write. Once the gang tree | |
1602 | * is assembled, free and claim are in-memory operations that cannot fail. | |
1603 | * | |
1604 | * In the event that a gang write fails, zio_dva_unallocate() walks the | |
1605 | * gang tree to immediately free (i.e. insert back into the space map) | |
1606 | * everything we've allocated. This ensures that we don't get ENOSPC | |
1607 | * errors during repeated suspend/resume cycles due to a flaky device. | |
1608 | * | |
1609 | * Gang rewrites only happen during sync-to-convergence. If we can't assemble | |
1610 | * the gang tree, we won't modify the block, so we can safely defer the free | |
1611 | * (knowing that the block is still intact). If we *can* assemble the gang | |
1612 | * tree, then even if some of the rewrites fail, zio_dva_unallocate() will free | |
1613 | * each constituent bp and we can allocate a new block on the next sync pass. | |
1614 | * | |
1615 | * In all cases, the gang tree allows complete recovery from partial failure. | |
34dc7c2f BB |
1616 | * ========================================================================== |
1617 | */ | |
b128c09f BB |
1618 | |
1619 | static zio_t * | |
1620 | zio_read_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
34dc7c2f | 1621 | { |
b128c09f BB |
1622 | if (gn != NULL) |
1623 | return (pio); | |
34dc7c2f | 1624 | |
b128c09f BB |
1625 | return (zio_read(pio, pio->io_spa, bp, data, BP_GET_PSIZE(bp), |
1626 | NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), | |
1627 | &pio->io_bookmark)); | |
1628 | } | |
1629 | ||
1630 | zio_t * | |
1631 | zio_rewrite_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
1632 | { | |
1633 | zio_t *zio; | |
1634 | ||
1635 | if (gn != NULL) { | |
1636 | zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, | |
1637 | gn->gn_gbh, SPA_GANGBLOCKSIZE, NULL, NULL, pio->io_priority, | |
1638 | ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); | |
34dc7c2f | 1639 | /* |
b128c09f BB |
1640 | * As we rewrite each gang header, the pipeline will compute |
1641 | * a new gang block header checksum for it; but no one will | |
1642 | * compute a new data checksum, so we do that here. The one | |
1643 | * exception is the gang leader: the pipeline already computed | |
1644 | * its data checksum because that stage precedes gang assembly. | |
1645 | * (Presently, nothing actually uses interior data checksums; | |
1646 | * this is just good hygiene.) | |
34dc7c2f | 1647 | */ |
9babb374 | 1648 | if (gn != pio->io_gang_leader->io_gang_tree) { |
b128c09f BB |
1649 | zio_checksum_compute(zio, BP_GET_CHECKSUM(bp), |
1650 | data, BP_GET_PSIZE(bp)); | |
1651 | } | |
428870ff BB |
1652 | /* |
1653 | * If we are here to damage data for testing purposes, | |
1654 | * leave the GBH alone so that we can detect the damage. | |
1655 | */ | |
1656 | if (pio->io_gang_leader->io_flags & ZIO_FLAG_INDUCE_DAMAGE) | |
1657 | zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; | |
34dc7c2f | 1658 | } else { |
b128c09f BB |
1659 | zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, |
1660 | data, BP_GET_PSIZE(bp), NULL, NULL, pio->io_priority, | |
1661 | ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); | |
34dc7c2f BB |
1662 | } |
1663 | ||
b128c09f BB |
1664 | return (zio); |
1665 | } | |
34dc7c2f | 1666 | |
b128c09f BB |
1667 | /* ARGSUSED */ |
1668 | zio_t * | |
1669 | zio_free_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
1670 | { | |
428870ff BB |
1671 | return (zio_free_sync(pio, pio->io_spa, pio->io_txg, bp, |
1672 | ZIO_GANG_CHILD_FLAGS(pio))); | |
34dc7c2f BB |
1673 | } |
1674 | ||
b128c09f BB |
1675 | /* ARGSUSED */ |
1676 | zio_t * | |
1677 | zio_claim_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
34dc7c2f | 1678 | { |
b128c09f BB |
1679 | return (zio_claim(pio, pio->io_spa, pio->io_txg, bp, |
1680 | NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio))); | |
1681 | } | |
1682 | ||
1683 | static zio_gang_issue_func_t *zio_gang_issue_func[ZIO_TYPES] = { | |
1684 | NULL, | |
1685 | zio_read_gang, | |
1686 | zio_rewrite_gang, | |
1687 | zio_free_gang, | |
1688 | zio_claim_gang, | |
1689 | NULL | |
1690 | }; | |
34dc7c2f | 1691 | |
b128c09f | 1692 | static void zio_gang_tree_assemble_done(zio_t *zio); |
34dc7c2f | 1693 | |
b128c09f BB |
1694 | static zio_gang_node_t * |
1695 | zio_gang_node_alloc(zio_gang_node_t **gnpp) | |
1696 | { | |
1697 | zio_gang_node_t *gn; | |
34dc7c2f | 1698 | |
b128c09f | 1699 | ASSERT(*gnpp == NULL); |
34dc7c2f | 1700 | |
691f6ac4 | 1701 | gn = kmem_zalloc(sizeof (*gn), KM_PUSHPAGE); |
b128c09f BB |
1702 | gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE); |
1703 | *gnpp = gn; | |
34dc7c2f | 1704 | |
b128c09f | 1705 | return (gn); |
34dc7c2f BB |
1706 | } |
1707 | ||
34dc7c2f | 1708 | static void |
b128c09f | 1709 | zio_gang_node_free(zio_gang_node_t **gnpp) |
34dc7c2f | 1710 | { |
b128c09f | 1711 | zio_gang_node_t *gn = *gnpp; |
d6320ddb | 1712 | int g; |
34dc7c2f | 1713 | |
d6320ddb | 1714 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) |
b128c09f BB |
1715 | ASSERT(gn->gn_child[g] == NULL); |
1716 | ||
1717 | zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE); | |
1718 | kmem_free(gn, sizeof (*gn)); | |
1719 | *gnpp = NULL; | |
34dc7c2f BB |
1720 | } |
1721 | ||
b128c09f BB |
1722 | static void |
1723 | zio_gang_tree_free(zio_gang_node_t **gnpp) | |
34dc7c2f | 1724 | { |
b128c09f | 1725 | zio_gang_node_t *gn = *gnpp; |
d6320ddb | 1726 | int g; |
34dc7c2f | 1727 | |
b128c09f BB |
1728 | if (gn == NULL) |
1729 | return; | |
34dc7c2f | 1730 | |
d6320ddb | 1731 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) |
b128c09f | 1732 | zio_gang_tree_free(&gn->gn_child[g]); |
34dc7c2f | 1733 | |
b128c09f | 1734 | zio_gang_node_free(gnpp); |
34dc7c2f BB |
1735 | } |
1736 | ||
b128c09f | 1737 | static void |
9babb374 | 1738 | zio_gang_tree_assemble(zio_t *gio, blkptr_t *bp, zio_gang_node_t **gnpp) |
34dc7c2f | 1739 | { |
b128c09f BB |
1740 | zio_gang_node_t *gn = zio_gang_node_alloc(gnpp); |
1741 | ||
9babb374 | 1742 | ASSERT(gio->io_gang_leader == gio); |
b128c09f | 1743 | ASSERT(BP_IS_GANG(bp)); |
34dc7c2f | 1744 | |
9babb374 | 1745 | zio_nowait(zio_read(gio, gio->io_spa, bp, gn->gn_gbh, |
b128c09f | 1746 | SPA_GANGBLOCKSIZE, zio_gang_tree_assemble_done, gn, |
9babb374 | 1747 | gio->io_priority, ZIO_GANG_CHILD_FLAGS(gio), &gio->io_bookmark)); |
b128c09f | 1748 | } |
34dc7c2f | 1749 | |
b128c09f BB |
1750 | static void |
1751 | zio_gang_tree_assemble_done(zio_t *zio) | |
1752 | { | |
9babb374 | 1753 | zio_t *gio = zio->io_gang_leader; |
b128c09f BB |
1754 | zio_gang_node_t *gn = zio->io_private; |
1755 | blkptr_t *bp = zio->io_bp; | |
d6320ddb | 1756 | int g; |
34dc7c2f | 1757 | |
9babb374 | 1758 | ASSERT(gio == zio_unique_parent(zio)); |
428870ff | 1759 | ASSERT(zio->io_child_count == 0); |
34dc7c2f | 1760 | |
b128c09f BB |
1761 | if (zio->io_error) |
1762 | return; | |
34dc7c2f | 1763 | |
b128c09f BB |
1764 | if (BP_SHOULD_BYTESWAP(bp)) |
1765 | byteswap_uint64_array(zio->io_data, zio->io_size); | |
34dc7c2f | 1766 | |
b128c09f BB |
1767 | ASSERT(zio->io_data == gn->gn_gbh); |
1768 | ASSERT(zio->io_size == SPA_GANGBLOCKSIZE); | |
428870ff | 1769 | ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); |
34dc7c2f | 1770 | |
d6320ddb | 1771 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) { |
b128c09f BB |
1772 | blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; |
1773 | if (!BP_IS_GANG(gbp)) | |
1774 | continue; | |
9babb374 | 1775 | zio_gang_tree_assemble(gio, gbp, &gn->gn_child[g]); |
b128c09f | 1776 | } |
34dc7c2f BB |
1777 | } |
1778 | ||
b128c09f BB |
1779 | static void |
1780 | zio_gang_tree_issue(zio_t *pio, zio_gang_node_t *gn, blkptr_t *bp, void *data) | |
34dc7c2f | 1781 | { |
9babb374 | 1782 | zio_t *gio = pio->io_gang_leader; |
b128c09f | 1783 | zio_t *zio; |
d6320ddb | 1784 | int g; |
34dc7c2f | 1785 | |
b128c09f | 1786 | ASSERT(BP_IS_GANG(bp) == !!gn); |
9babb374 BB |
1787 | ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp)); |
1788 | ASSERT(BP_GET_LSIZE(bp) == BP_GET_PSIZE(bp) || gn == gio->io_gang_tree); | |
34dc7c2f | 1789 | |
b128c09f BB |
1790 | /* |
1791 | * If you're a gang header, your data is in gn->gn_gbh. | |
1792 | * If you're a gang member, your data is in 'data' and gn == NULL. | |
1793 | */ | |
9babb374 | 1794 | zio = zio_gang_issue_func[gio->io_type](pio, bp, gn, data); |
34dc7c2f | 1795 | |
b128c09f | 1796 | if (gn != NULL) { |
428870ff | 1797 | ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); |
34dc7c2f | 1798 | |
d6320ddb | 1799 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) { |
b128c09f BB |
1800 | blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; |
1801 | if (BP_IS_HOLE(gbp)) | |
1802 | continue; | |
1803 | zio_gang_tree_issue(zio, gn->gn_child[g], gbp, data); | |
1804 | data = (char *)data + BP_GET_PSIZE(gbp); | |
1805 | } | |
34dc7c2f BB |
1806 | } |
1807 | ||
9babb374 BB |
1808 | if (gn == gio->io_gang_tree) |
1809 | ASSERT3P((char *)gio->io_data + gio->io_size, ==, data); | |
34dc7c2f | 1810 | |
b128c09f BB |
1811 | if (zio != pio) |
1812 | zio_nowait(zio); | |
34dc7c2f BB |
1813 | } |
1814 | ||
1815 | static int | |
b128c09f | 1816 | zio_gang_assemble(zio_t *zio) |
34dc7c2f | 1817 | { |
b128c09f | 1818 | blkptr_t *bp = zio->io_bp; |
34dc7c2f | 1819 | |
9babb374 BB |
1820 | ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == NULL); |
1821 | ASSERT(zio->io_child_type > ZIO_CHILD_GANG); | |
1822 | ||
1823 | zio->io_gang_leader = zio; | |
34dc7c2f | 1824 | |
b128c09f | 1825 | zio_gang_tree_assemble(zio, bp, &zio->io_gang_tree); |
34dc7c2f BB |
1826 | |
1827 | return (ZIO_PIPELINE_CONTINUE); | |
1828 | } | |
1829 | ||
1830 | static int | |
b128c09f | 1831 | zio_gang_issue(zio_t *zio) |
34dc7c2f | 1832 | { |
b128c09f | 1833 | blkptr_t *bp = zio->io_bp; |
34dc7c2f | 1834 | |
b128c09f BB |
1835 | if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE)) |
1836 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 1837 | |
9babb374 BB |
1838 | ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == zio); |
1839 | ASSERT(zio->io_child_type > ZIO_CHILD_GANG); | |
34dc7c2f | 1840 | |
b128c09f | 1841 | if (zio->io_child_error[ZIO_CHILD_GANG] == 0) |
9babb374 | 1842 | zio_gang_tree_issue(zio, zio->io_gang_tree, bp, zio->io_data); |
b128c09f | 1843 | else |
9babb374 | 1844 | zio_gang_tree_free(&zio->io_gang_tree); |
34dc7c2f | 1845 | |
b128c09f | 1846 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; |
34dc7c2f BB |
1847 | |
1848 | return (ZIO_PIPELINE_CONTINUE); | |
1849 | } | |
1850 | ||
1851 | static void | |
b128c09f | 1852 | zio_write_gang_member_ready(zio_t *zio) |
34dc7c2f | 1853 | { |
d164b209 | 1854 | zio_t *pio = zio_unique_parent(zio); |
34dc7c2f BB |
1855 | dva_t *cdva = zio->io_bp->blk_dva; |
1856 | dva_t *pdva = pio->io_bp->blk_dva; | |
1857 | uint64_t asize; | |
d6320ddb | 1858 | int d; |
d1d7e268 | 1859 | ASSERTV(zio_t *gio = zio->io_gang_leader); |
34dc7c2f | 1860 | |
b128c09f BB |
1861 | if (BP_IS_HOLE(zio->io_bp)) |
1862 | return; | |
1863 | ||
1864 | ASSERT(BP_IS_HOLE(&zio->io_bp_orig)); | |
1865 | ||
1866 | ASSERT(zio->io_child_type == ZIO_CHILD_GANG); | |
428870ff BB |
1867 | ASSERT3U(zio->io_prop.zp_copies, ==, gio->io_prop.zp_copies); |
1868 | ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp)); | |
1869 | ASSERT3U(pio->io_prop.zp_copies, <=, BP_GET_NDVAS(pio->io_bp)); | |
34dc7c2f | 1870 | ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp)); |
34dc7c2f BB |
1871 | |
1872 | mutex_enter(&pio->io_lock); | |
d6320ddb | 1873 | for (d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) { |
34dc7c2f BB |
1874 | ASSERT(DVA_GET_GANG(&pdva[d])); |
1875 | asize = DVA_GET_ASIZE(&pdva[d]); | |
1876 | asize += DVA_GET_ASIZE(&cdva[d]); | |
1877 | DVA_SET_ASIZE(&pdva[d], asize); | |
1878 | } | |
1879 | mutex_exit(&pio->io_lock); | |
1880 | } | |
1881 | ||
1882 | static int | |
b128c09f | 1883 | zio_write_gang_block(zio_t *pio) |
34dc7c2f | 1884 | { |
b128c09f BB |
1885 | spa_t *spa = pio->io_spa; |
1886 | blkptr_t *bp = pio->io_bp; | |
9babb374 | 1887 | zio_t *gio = pio->io_gang_leader; |
b128c09f BB |
1888 | zio_t *zio; |
1889 | zio_gang_node_t *gn, **gnpp; | |
34dc7c2f | 1890 | zio_gbh_phys_t *gbh; |
b128c09f BB |
1891 | uint64_t txg = pio->io_txg; |
1892 | uint64_t resid = pio->io_size; | |
1893 | uint64_t lsize; | |
428870ff BB |
1894 | int copies = gio->io_prop.zp_copies; |
1895 | int gbh_copies = MIN(copies + 1, spa_max_replication(spa)); | |
b128c09f | 1896 | zio_prop_t zp; |
d6320ddb | 1897 | int g, error; |
34dc7c2f | 1898 | |
428870ff BB |
1899 | error = metaslab_alloc(spa, spa_normal_class(spa), SPA_GANGBLOCKSIZE, |
1900 | bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, | |
b128c09f | 1901 | METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER); |
34dc7c2f | 1902 | if (error) { |
b128c09f | 1903 | pio->io_error = error; |
34dc7c2f BB |
1904 | return (ZIO_PIPELINE_CONTINUE); |
1905 | } | |
1906 | ||
9babb374 BB |
1907 | if (pio == gio) { |
1908 | gnpp = &gio->io_gang_tree; | |
b128c09f BB |
1909 | } else { |
1910 | gnpp = pio->io_private; | |
1911 | ASSERT(pio->io_ready == zio_write_gang_member_ready); | |
34dc7c2f BB |
1912 | } |
1913 | ||
b128c09f BB |
1914 | gn = zio_gang_node_alloc(gnpp); |
1915 | gbh = gn->gn_gbh; | |
1916 | bzero(gbh, SPA_GANGBLOCKSIZE); | |
34dc7c2f | 1917 | |
b128c09f BB |
1918 | /* |
1919 | * Create the gang header. | |
1920 | */ | |
1921 | zio = zio_rewrite(pio, spa, txg, bp, gbh, SPA_GANGBLOCKSIZE, NULL, NULL, | |
1922 | pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); | |
34dc7c2f | 1923 | |
b128c09f BB |
1924 | /* |
1925 | * Create and nowait the gang children. | |
1926 | */ | |
d6320ddb | 1927 | for (g = 0; resid != 0; resid -= lsize, g++) { |
b128c09f BB |
1928 | lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g), |
1929 | SPA_MINBLOCKSIZE); | |
1930 | ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid); | |
1931 | ||
9babb374 | 1932 | zp.zp_checksum = gio->io_prop.zp_checksum; |
b128c09f BB |
1933 | zp.zp_compress = ZIO_COMPRESS_OFF; |
1934 | zp.zp_type = DMU_OT_NONE; | |
1935 | zp.zp_level = 0; | |
428870ff | 1936 | zp.zp_copies = gio->io_prop.zp_copies; |
03c6040b GW |
1937 | zp.zp_dedup = B_FALSE; |
1938 | zp.zp_dedup_verify = B_FALSE; | |
1939 | zp.zp_nopwrite = B_FALSE; | |
b128c09f BB |
1940 | |
1941 | zio_nowait(zio_write(zio, spa, txg, &gbh->zg_blkptr[g], | |
1942 | (char *)pio->io_data + (pio->io_size - resid), lsize, &zp, | |
e8b96c60 | 1943 | zio_write_gang_member_ready, NULL, NULL, &gn->gn_child[g], |
b128c09f BB |
1944 | pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), |
1945 | &pio->io_bookmark)); | |
1946 | } | |
34dc7c2f BB |
1947 | |
1948 | /* | |
b128c09f | 1949 | * Set pio's pipeline to just wait for zio to finish. |
34dc7c2f | 1950 | */ |
b128c09f BB |
1951 | pio->io_pipeline = ZIO_INTERLOCK_PIPELINE; |
1952 | ||
920dd524 ED |
1953 | /* |
1954 | * We didn't allocate this bp, so make sure it doesn't get unmarked. | |
1955 | */ | |
1956 | pio->io_flags &= ~ZIO_FLAG_FASTWRITE; | |
1957 | ||
b128c09f BB |
1958 | zio_nowait(zio); |
1959 | ||
1960 | return (ZIO_PIPELINE_CONTINUE); | |
34dc7c2f BB |
1961 | } |
1962 | ||
03c6040b GW |
1963 | /* |
1964 | * The zio_nop_write stage in the pipeline determines if allocating | |
1965 | * a new bp is necessary. By leveraging a cryptographically secure checksum, | |
1966 | * such as SHA256, we can compare the checksums of the new data and the old | |
1967 | * to determine if allocating a new block is required. The nopwrite | |
1968 | * feature can handle writes in either syncing or open context (i.e. zil | |
1969 | * writes) and as a result is mutually exclusive with dedup. | |
1970 | */ | |
1971 | static int | |
1972 | zio_nop_write(zio_t *zio) | |
1973 | { | |
1974 | blkptr_t *bp = zio->io_bp; | |
1975 | blkptr_t *bp_orig = &zio->io_bp_orig; | |
1976 | zio_prop_t *zp = &zio->io_prop; | |
1977 | ||
1978 | ASSERT(BP_GET_LEVEL(bp) == 0); | |
1979 | ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); | |
1980 | ASSERT(zp->zp_nopwrite); | |
1981 | ASSERT(!zp->zp_dedup); | |
1982 | ASSERT(zio->io_bp_override == NULL); | |
1983 | ASSERT(IO_IS_ALLOCATING(zio)); | |
1984 | ||
1985 | /* | |
1986 | * Check to see if the original bp and the new bp have matching | |
1987 | * characteristics (i.e. same checksum, compression algorithms, etc). | |
1988 | * If they don't then just continue with the pipeline which will | |
1989 | * allocate a new bp. | |
1990 | */ | |
1991 | if (BP_IS_HOLE(bp_orig) || | |
1992 | !zio_checksum_table[BP_GET_CHECKSUM(bp)].ci_dedup || | |
1993 | BP_GET_CHECKSUM(bp) != BP_GET_CHECKSUM(bp_orig) || | |
1994 | BP_GET_COMPRESS(bp) != BP_GET_COMPRESS(bp_orig) || | |
1995 | BP_GET_DEDUP(bp) != BP_GET_DEDUP(bp_orig) || | |
1996 | zp->zp_copies != BP_GET_NDVAS(bp_orig)) | |
1997 | return (ZIO_PIPELINE_CONTINUE); | |
1998 | ||
1999 | /* | |
2000 | * If the checksums match then reset the pipeline so that we | |
2001 | * avoid allocating a new bp and issuing any I/O. | |
2002 | */ | |
2003 | if (ZIO_CHECKSUM_EQUAL(bp->blk_cksum, bp_orig->blk_cksum)) { | |
2004 | ASSERT(zio_checksum_table[zp->zp_checksum].ci_dedup); | |
2005 | ASSERT3U(BP_GET_PSIZE(bp), ==, BP_GET_PSIZE(bp_orig)); | |
2006 | ASSERT3U(BP_GET_LSIZE(bp), ==, BP_GET_LSIZE(bp_orig)); | |
2007 | ASSERT(zp->zp_compress != ZIO_COMPRESS_OFF); | |
2008 | ASSERT(bcmp(&bp->blk_prop, &bp_orig->blk_prop, | |
2009 | sizeof (uint64_t)) == 0); | |
2010 | ||
2011 | *bp = *bp_orig; | |
2012 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
2013 | zio->io_flags |= ZIO_FLAG_NOPWRITE; | |
2014 | } | |
2015 | ||
2016 | return (ZIO_PIPELINE_CONTINUE); | |
2017 | } | |
2018 | ||
34dc7c2f BB |
2019 | /* |
2020 | * ========================================================================== | |
428870ff | 2021 | * Dedup |
34dc7c2f BB |
2022 | * ========================================================================== |
2023 | */ | |
428870ff BB |
2024 | static void |
2025 | zio_ddt_child_read_done(zio_t *zio) | |
2026 | { | |
2027 | blkptr_t *bp = zio->io_bp; | |
2028 | ddt_entry_t *dde = zio->io_private; | |
2029 | ddt_phys_t *ddp; | |
2030 | zio_t *pio = zio_unique_parent(zio); | |
2031 | ||
2032 | mutex_enter(&pio->io_lock); | |
2033 | ddp = ddt_phys_select(dde, bp); | |
2034 | if (zio->io_error == 0) | |
2035 | ddt_phys_clear(ddp); /* this ddp doesn't need repair */ | |
2036 | if (zio->io_error == 0 && dde->dde_repair_data == NULL) | |
2037 | dde->dde_repair_data = zio->io_data; | |
2038 | else | |
2039 | zio_buf_free(zio->io_data, zio->io_size); | |
2040 | mutex_exit(&pio->io_lock); | |
2041 | } | |
2042 | ||
2043 | static int | |
2044 | zio_ddt_read_start(zio_t *zio) | |
2045 | { | |
2046 | blkptr_t *bp = zio->io_bp; | |
d6320ddb | 2047 | int p; |
428870ff BB |
2048 | |
2049 | ASSERT(BP_GET_DEDUP(bp)); | |
2050 | ASSERT(BP_GET_PSIZE(bp) == zio->io_size); | |
2051 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
2052 | ||
2053 | if (zio->io_child_error[ZIO_CHILD_DDT]) { | |
2054 | ddt_t *ddt = ddt_select(zio->io_spa, bp); | |
2055 | ddt_entry_t *dde = ddt_repair_start(ddt, bp); | |
2056 | ddt_phys_t *ddp = dde->dde_phys; | |
2057 | ddt_phys_t *ddp_self = ddt_phys_select(dde, bp); | |
2058 | blkptr_t blk; | |
2059 | ||
2060 | ASSERT(zio->io_vsd == NULL); | |
2061 | zio->io_vsd = dde; | |
2062 | ||
2063 | if (ddp_self == NULL) | |
2064 | return (ZIO_PIPELINE_CONTINUE); | |
2065 | ||
d6320ddb | 2066 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff BB |
2067 | if (ddp->ddp_phys_birth == 0 || ddp == ddp_self) |
2068 | continue; | |
2069 | ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp, | |
2070 | &blk); | |
2071 | zio_nowait(zio_read(zio, zio->io_spa, &blk, | |
2072 | zio_buf_alloc(zio->io_size), zio->io_size, | |
2073 | zio_ddt_child_read_done, dde, zio->io_priority, | |
2074 | ZIO_DDT_CHILD_FLAGS(zio) | ZIO_FLAG_DONT_PROPAGATE, | |
2075 | &zio->io_bookmark)); | |
2076 | } | |
2077 | return (ZIO_PIPELINE_CONTINUE); | |
2078 | } | |
2079 | ||
2080 | zio_nowait(zio_read(zio, zio->io_spa, bp, | |
2081 | zio->io_data, zio->io_size, NULL, NULL, zio->io_priority, | |
2082 | ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark)); | |
2083 | ||
2084 | return (ZIO_PIPELINE_CONTINUE); | |
2085 | } | |
2086 | ||
2087 | static int | |
2088 | zio_ddt_read_done(zio_t *zio) | |
2089 | { | |
2090 | blkptr_t *bp = zio->io_bp; | |
2091 | ||
2092 | if (zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE)) | |
2093 | return (ZIO_PIPELINE_STOP); | |
2094 | ||
2095 | ASSERT(BP_GET_DEDUP(bp)); | |
2096 | ASSERT(BP_GET_PSIZE(bp) == zio->io_size); | |
2097 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
2098 | ||
2099 | if (zio->io_child_error[ZIO_CHILD_DDT]) { | |
2100 | ddt_t *ddt = ddt_select(zio->io_spa, bp); | |
2101 | ddt_entry_t *dde = zio->io_vsd; | |
2102 | if (ddt == NULL) { | |
2103 | ASSERT(spa_load_state(zio->io_spa) != SPA_LOAD_NONE); | |
2104 | return (ZIO_PIPELINE_CONTINUE); | |
2105 | } | |
2106 | if (dde == NULL) { | |
2107 | zio->io_stage = ZIO_STAGE_DDT_READ_START >> 1; | |
2108 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); | |
2109 | return (ZIO_PIPELINE_STOP); | |
2110 | } | |
2111 | if (dde->dde_repair_data != NULL) { | |
2112 | bcopy(dde->dde_repair_data, zio->io_data, zio->io_size); | |
2113 | zio->io_child_error[ZIO_CHILD_DDT] = 0; | |
2114 | } | |
2115 | ddt_repair_done(ddt, dde); | |
2116 | zio->io_vsd = NULL; | |
2117 | } | |
2118 | ||
2119 | ASSERT(zio->io_vsd == NULL); | |
2120 | ||
2121 | return (ZIO_PIPELINE_CONTINUE); | |
2122 | } | |
2123 | ||
2124 | static boolean_t | |
2125 | zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde) | |
2126 | { | |
2127 | spa_t *spa = zio->io_spa; | |
d6320ddb | 2128 | int p; |
428870ff BB |
2129 | |
2130 | /* | |
2131 | * Note: we compare the original data, not the transformed data, | |
2132 | * because when zio->io_bp is an override bp, we will not have | |
2133 | * pushed the I/O transforms. That's an important optimization | |
2134 | * because otherwise we'd compress/encrypt all dmu_sync() data twice. | |
2135 | */ | |
d6320ddb | 2136 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { |
428870ff BB |
2137 | zio_t *lio = dde->dde_lead_zio[p]; |
2138 | ||
2139 | if (lio != NULL) { | |
2140 | return (lio->io_orig_size != zio->io_orig_size || | |
2141 | bcmp(zio->io_orig_data, lio->io_orig_data, | |
2142 | zio->io_orig_size) != 0); | |
2143 | } | |
2144 | } | |
2145 | ||
d6320ddb | 2146 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { |
428870ff BB |
2147 | ddt_phys_t *ddp = &dde->dde_phys[p]; |
2148 | ||
2149 | if (ddp->ddp_phys_birth != 0) { | |
2150 | arc_buf_t *abuf = NULL; | |
2151 | uint32_t aflags = ARC_WAIT; | |
2152 | blkptr_t blk = *zio->io_bp; | |
2153 | int error; | |
2154 | ||
2155 | ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth); | |
2156 | ||
2157 | ddt_exit(ddt); | |
2158 | ||
294f6806 | 2159 | error = arc_read(NULL, spa, &blk, |
428870ff BB |
2160 | arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ, |
2161 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, | |
2162 | &aflags, &zio->io_bookmark); | |
2163 | ||
2164 | if (error == 0) { | |
2165 | if (arc_buf_size(abuf) != zio->io_orig_size || | |
2166 | bcmp(abuf->b_data, zio->io_orig_data, | |
2167 | zio->io_orig_size) != 0) | |
2e528b49 | 2168 | error = SET_ERROR(EEXIST); |
13fe0198 | 2169 | VERIFY(arc_buf_remove_ref(abuf, &abuf)); |
428870ff BB |
2170 | } |
2171 | ||
2172 | ddt_enter(ddt); | |
2173 | return (error != 0); | |
2174 | } | |
2175 | } | |
2176 | ||
2177 | return (B_FALSE); | |
2178 | } | |
2179 | ||
2180 | static void | |
2181 | zio_ddt_child_write_ready(zio_t *zio) | |
2182 | { | |
2183 | int p = zio->io_prop.zp_copies; | |
2184 | ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); | |
2185 | ddt_entry_t *dde = zio->io_private; | |
2186 | ddt_phys_t *ddp = &dde->dde_phys[p]; | |
2187 | zio_t *pio; | |
2188 | ||
2189 | if (zio->io_error) | |
2190 | return; | |
2191 | ||
2192 | ddt_enter(ddt); | |
2193 | ||
2194 | ASSERT(dde->dde_lead_zio[p] == zio); | |
2195 | ||
2196 | ddt_phys_fill(ddp, zio->io_bp); | |
2197 | ||
2198 | while ((pio = zio_walk_parents(zio)) != NULL) | |
2199 | ddt_bp_fill(ddp, pio->io_bp, zio->io_txg); | |
2200 | ||
2201 | ddt_exit(ddt); | |
2202 | } | |
2203 | ||
2204 | static void | |
2205 | zio_ddt_child_write_done(zio_t *zio) | |
2206 | { | |
2207 | int p = zio->io_prop.zp_copies; | |
2208 | ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); | |
2209 | ddt_entry_t *dde = zio->io_private; | |
2210 | ddt_phys_t *ddp = &dde->dde_phys[p]; | |
2211 | ||
2212 | ddt_enter(ddt); | |
2213 | ||
2214 | ASSERT(ddp->ddp_refcnt == 0); | |
2215 | ASSERT(dde->dde_lead_zio[p] == zio); | |
2216 | dde->dde_lead_zio[p] = NULL; | |
2217 | ||
2218 | if (zio->io_error == 0) { | |
2219 | while (zio_walk_parents(zio) != NULL) | |
2220 | ddt_phys_addref(ddp); | |
2221 | } else { | |
2222 | ddt_phys_clear(ddp); | |
2223 | } | |
2224 | ||
2225 | ddt_exit(ddt); | |
2226 | } | |
2227 | ||
2228 | static void | |
2229 | zio_ddt_ditto_write_done(zio_t *zio) | |
2230 | { | |
2231 | int p = DDT_PHYS_DITTO; | |
428870ff BB |
2232 | blkptr_t *bp = zio->io_bp; |
2233 | ddt_t *ddt = ddt_select(zio->io_spa, bp); | |
2234 | ddt_entry_t *dde = zio->io_private; | |
2235 | ddt_phys_t *ddp = &dde->dde_phys[p]; | |
2236 | ddt_key_t *ddk = &dde->dde_key; | |
1fde1e37 | 2237 | ASSERTV(zio_prop_t *zp = &zio->io_prop); |
428870ff BB |
2238 | |
2239 | ddt_enter(ddt); | |
2240 | ||
2241 | ASSERT(ddp->ddp_refcnt == 0); | |
2242 | ASSERT(dde->dde_lead_zio[p] == zio); | |
2243 | dde->dde_lead_zio[p] = NULL; | |
2244 | ||
2245 | if (zio->io_error == 0) { | |
2246 | ASSERT(ZIO_CHECKSUM_EQUAL(bp->blk_cksum, ddk->ddk_cksum)); | |
2247 | ASSERT(zp->zp_copies < SPA_DVAS_PER_BP); | |
2248 | ASSERT(zp->zp_copies == BP_GET_NDVAS(bp) - BP_IS_GANG(bp)); | |
2249 | if (ddp->ddp_phys_birth != 0) | |
2250 | ddt_phys_free(ddt, ddk, ddp, zio->io_txg); | |
2251 | ddt_phys_fill(ddp, bp); | |
2252 | } | |
2253 | ||
2254 | ddt_exit(ddt); | |
2255 | } | |
2256 | ||
2257 | static int | |
2258 | zio_ddt_write(zio_t *zio) | |
2259 | { | |
2260 | spa_t *spa = zio->io_spa; | |
2261 | blkptr_t *bp = zio->io_bp; | |
2262 | uint64_t txg = zio->io_txg; | |
2263 | zio_prop_t *zp = &zio->io_prop; | |
2264 | int p = zp->zp_copies; | |
2265 | int ditto_copies; | |
2266 | zio_t *cio = NULL; | |
2267 | zio_t *dio = NULL; | |
2268 | ddt_t *ddt = ddt_select(spa, bp); | |
2269 | ddt_entry_t *dde; | |
2270 | ddt_phys_t *ddp; | |
2271 | ||
2272 | ASSERT(BP_GET_DEDUP(bp)); | |
2273 | ASSERT(BP_GET_CHECKSUM(bp) == zp->zp_checksum); | |
2274 | ASSERT(BP_IS_HOLE(bp) || zio->io_bp_override); | |
2275 | ||
2276 | ddt_enter(ddt); | |
2277 | dde = ddt_lookup(ddt, bp, B_TRUE); | |
2278 | ddp = &dde->dde_phys[p]; | |
2279 | ||
2280 | if (zp->zp_dedup_verify && zio_ddt_collision(zio, ddt, dde)) { | |
2281 | /* | |
2282 | * If we're using a weak checksum, upgrade to a strong checksum | |
2283 | * and try again. If we're already using a strong checksum, | |
2284 | * we can't resolve it, so just convert to an ordinary write. | |
2285 | * (And automatically e-mail a paper to Nature?) | |
2286 | */ | |
2287 | if (!zio_checksum_table[zp->zp_checksum].ci_dedup) { | |
2288 | zp->zp_checksum = spa_dedup_checksum(spa); | |
2289 | zio_pop_transforms(zio); | |
2290 | zio->io_stage = ZIO_STAGE_OPEN; | |
2291 | BP_ZERO(bp); | |
2292 | } else { | |
03c6040b | 2293 | zp->zp_dedup = B_FALSE; |
428870ff BB |
2294 | } |
2295 | zio->io_pipeline = ZIO_WRITE_PIPELINE; | |
2296 | ddt_exit(ddt); | |
2297 | return (ZIO_PIPELINE_CONTINUE); | |
2298 | } | |
2299 | ||
2300 | ditto_copies = ddt_ditto_copies_needed(ddt, dde, ddp); | |
2301 | ASSERT(ditto_copies < SPA_DVAS_PER_BP); | |
2302 | ||
2303 | if (ditto_copies > ddt_ditto_copies_present(dde) && | |
2304 | dde->dde_lead_zio[DDT_PHYS_DITTO] == NULL) { | |
2305 | zio_prop_t czp = *zp; | |
2306 | ||
2307 | czp.zp_copies = ditto_copies; | |
2308 | ||
2309 | /* | |
2310 | * If we arrived here with an override bp, we won't have run | |
2311 | * the transform stack, so we won't have the data we need to | |
2312 | * generate a child i/o. So, toss the override bp and restart. | |
2313 | * This is safe, because using the override bp is just an | |
2314 | * optimization; and it's rare, so the cost doesn't matter. | |
2315 | */ | |
2316 | if (zio->io_bp_override) { | |
2317 | zio_pop_transforms(zio); | |
2318 | zio->io_stage = ZIO_STAGE_OPEN; | |
2319 | zio->io_pipeline = ZIO_WRITE_PIPELINE; | |
2320 | zio->io_bp_override = NULL; | |
2321 | BP_ZERO(bp); | |
2322 | ddt_exit(ddt); | |
2323 | return (ZIO_PIPELINE_CONTINUE); | |
2324 | } | |
2325 | ||
2326 | dio = zio_write(zio, spa, txg, bp, zio->io_orig_data, | |
e8b96c60 | 2327 | zio->io_orig_size, &czp, NULL, NULL, |
428870ff BB |
2328 | zio_ddt_ditto_write_done, dde, zio->io_priority, |
2329 | ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); | |
2330 | ||
2331 | zio_push_transform(dio, zio->io_data, zio->io_size, 0, NULL); | |
2332 | dde->dde_lead_zio[DDT_PHYS_DITTO] = dio; | |
2333 | } | |
2334 | ||
2335 | if (ddp->ddp_phys_birth != 0 || dde->dde_lead_zio[p] != NULL) { | |
2336 | if (ddp->ddp_phys_birth != 0) | |
2337 | ddt_bp_fill(ddp, bp, txg); | |
2338 | if (dde->dde_lead_zio[p] != NULL) | |
2339 | zio_add_child(zio, dde->dde_lead_zio[p]); | |
2340 | else | |
2341 | ddt_phys_addref(ddp); | |
2342 | } else if (zio->io_bp_override) { | |
2343 | ASSERT(bp->blk_birth == txg); | |
2344 | ASSERT(BP_EQUAL(bp, zio->io_bp_override)); | |
2345 | ddt_phys_fill(ddp, bp); | |
2346 | ddt_phys_addref(ddp); | |
2347 | } else { | |
2348 | cio = zio_write(zio, spa, txg, bp, zio->io_orig_data, | |
e8b96c60 | 2349 | zio->io_orig_size, zp, zio_ddt_child_write_ready, NULL, |
428870ff BB |
2350 | zio_ddt_child_write_done, dde, zio->io_priority, |
2351 | ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); | |
2352 | ||
2353 | zio_push_transform(cio, zio->io_data, zio->io_size, 0, NULL); | |
2354 | dde->dde_lead_zio[p] = cio; | |
2355 | } | |
2356 | ||
2357 | ddt_exit(ddt); | |
2358 | ||
2359 | if (cio) | |
2360 | zio_nowait(cio); | |
2361 | if (dio) | |
2362 | zio_nowait(dio); | |
2363 | ||
2364 | return (ZIO_PIPELINE_CONTINUE); | |
2365 | } | |
2366 | ||
2367 | ddt_entry_t *freedde; /* for debugging */ | |
b128c09f | 2368 | |
428870ff BB |
2369 | static int |
2370 | zio_ddt_free(zio_t *zio) | |
2371 | { | |
2372 | spa_t *spa = zio->io_spa; | |
2373 | blkptr_t *bp = zio->io_bp; | |
2374 | ddt_t *ddt = ddt_select(spa, bp); | |
2375 | ddt_entry_t *dde; | |
2376 | ddt_phys_t *ddp; | |
2377 | ||
2378 | ASSERT(BP_GET_DEDUP(bp)); | |
2379 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
2380 | ||
2381 | ddt_enter(ddt); | |
2382 | freedde = dde = ddt_lookup(ddt, bp, B_TRUE); | |
5dc6af0e BB |
2383 | if (dde) { |
2384 | ddp = ddt_phys_select(dde, bp); | |
2385 | if (ddp) | |
2386 | ddt_phys_decref(ddp); | |
2387 | } | |
428870ff BB |
2388 | ddt_exit(ddt); |
2389 | ||
2390 | return (ZIO_PIPELINE_CONTINUE); | |
2391 | } | |
2392 | ||
2393 | /* | |
2394 | * ========================================================================== | |
2395 | * Allocate and free blocks | |
2396 | * ========================================================================== | |
2397 | */ | |
34dc7c2f BB |
2398 | static int |
2399 | zio_dva_allocate(zio_t *zio) | |
2400 | { | |
2401 | spa_t *spa = zio->io_spa; | |
428870ff | 2402 | metaslab_class_t *mc = spa_normal_class(spa); |
34dc7c2f BB |
2403 | blkptr_t *bp = zio->io_bp; |
2404 | int error; | |
6d974228 | 2405 | int flags = 0; |
34dc7c2f | 2406 | |
9babb374 BB |
2407 | if (zio->io_gang_leader == NULL) { |
2408 | ASSERT(zio->io_child_type > ZIO_CHILD_GANG); | |
2409 | zio->io_gang_leader = zio; | |
2410 | } | |
2411 | ||
34dc7c2f | 2412 | ASSERT(BP_IS_HOLE(bp)); |
c99c9001 | 2413 | ASSERT0(BP_GET_NDVAS(bp)); |
428870ff BB |
2414 | ASSERT3U(zio->io_prop.zp_copies, >, 0); |
2415 | ASSERT3U(zio->io_prop.zp_copies, <=, spa_max_replication(spa)); | |
34dc7c2f BB |
2416 | ASSERT3U(zio->io_size, ==, BP_GET_PSIZE(bp)); |
2417 | ||
6d974228 GW |
2418 | /* |
2419 | * The dump device does not support gang blocks so allocation on | |
2420 | * behalf of the dump device (i.e. ZIO_FLAG_NODATA) must avoid | |
2421 | * the "fast" gang feature. | |
2422 | */ | |
2423 | flags |= (zio->io_flags & ZIO_FLAG_NODATA) ? METASLAB_GANG_AVOID : 0; | |
2424 | flags |= (zio->io_flags & ZIO_FLAG_GANG_CHILD) ? | |
2425 | METASLAB_GANG_CHILD : 0; | |
920dd524 | 2426 | flags |= (zio->io_flags & ZIO_FLAG_FASTWRITE) ? METASLAB_FASTWRITE : 0; |
b128c09f | 2427 | error = metaslab_alloc(spa, mc, zio->io_size, bp, |
6d974228 | 2428 | zio->io_prop.zp_copies, zio->io_txg, NULL, flags); |
34dc7c2f | 2429 | |
b128c09f | 2430 | if (error) { |
6d974228 GW |
2431 | spa_dbgmsg(spa, "%s: metaslab allocation failure: zio %p, " |
2432 | "size %llu, error %d", spa_name(spa), zio, zio->io_size, | |
2433 | error); | |
b128c09f BB |
2434 | if (error == ENOSPC && zio->io_size > SPA_MINBLOCKSIZE) |
2435 | return (zio_write_gang_block(zio)); | |
34dc7c2f BB |
2436 | zio->io_error = error; |
2437 | } | |
2438 | ||
2439 | return (ZIO_PIPELINE_CONTINUE); | |
2440 | } | |
2441 | ||
2442 | static int | |
2443 | zio_dva_free(zio_t *zio) | |
2444 | { | |
b128c09f | 2445 | metaslab_free(zio->io_spa, zio->io_bp, zio->io_txg, B_FALSE); |
34dc7c2f BB |
2446 | |
2447 | return (ZIO_PIPELINE_CONTINUE); | |
2448 | } | |
2449 | ||
2450 | static int | |
2451 | zio_dva_claim(zio_t *zio) | |
2452 | { | |
b128c09f BB |
2453 | int error; |
2454 | ||
2455 | error = metaslab_claim(zio->io_spa, zio->io_bp, zio->io_txg); | |
2456 | if (error) | |
2457 | zio->io_error = error; | |
34dc7c2f BB |
2458 | |
2459 | return (ZIO_PIPELINE_CONTINUE); | |
2460 | } | |
2461 | ||
b128c09f BB |
2462 | /* |
2463 | * Undo an allocation. This is used by zio_done() when an I/O fails | |
2464 | * and we want to give back the block we just allocated. | |
2465 | * This handles both normal blocks and gang blocks. | |
2466 | */ | |
2467 | static void | |
2468 | zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp) | |
2469 | { | |
d6320ddb BB |
2470 | int g; |
2471 | ||
b128c09f | 2472 | ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); |
428870ff | 2473 | ASSERT(zio->io_bp_override == NULL); |
b128c09f BB |
2474 | |
2475 | if (!BP_IS_HOLE(bp)) | |
428870ff | 2476 | metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE); |
b128c09f BB |
2477 | |
2478 | if (gn != NULL) { | |
d6320ddb | 2479 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) { |
b128c09f BB |
2480 | zio_dva_unallocate(zio, gn->gn_child[g], |
2481 | &gn->gn_gbh->zg_blkptr[g]); | |
2482 | } | |
2483 | } | |
2484 | } | |
2485 | ||
2486 | /* | |
2487 | * Try to allocate an intent log block. Return 0 on success, errno on failure. | |
2488 | */ | |
2489 | int | |
920dd524 ED |
2490 | zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, uint64_t size, |
2491 | boolean_t use_slog) | |
b128c09f | 2492 | { |
428870ff | 2493 | int error = 1; |
b128c09f | 2494 | |
428870ff BB |
2495 | ASSERT(txg > spa_syncing_txg(spa)); |
2496 | ||
ebf8e3a2 BB |
2497 | /* |
2498 | * ZIL blocks are always contiguous (i.e. not gang blocks) so we | |
2499 | * set the METASLAB_GANG_AVOID flag so that they don't "fast gang" | |
2500 | * when allocating them. | |
2501 | */ | |
2502 | if (use_slog) { | |
428870ff | 2503 | error = metaslab_alloc(spa, spa_log_class(spa), size, |
920dd524 ED |
2504 | new_bp, 1, txg, NULL, |
2505 | METASLAB_FASTWRITE | METASLAB_GANG_AVOID); | |
ebf8e3a2 | 2506 | } |
b128c09f | 2507 | |
ebf8e3a2 | 2508 | if (error) { |
428870ff | 2509 | error = metaslab_alloc(spa, spa_normal_class(spa), size, |
920dd524 | 2510 | new_bp, 1, txg, NULL, |
ac72fac3 | 2511 | METASLAB_FASTWRITE); |
ebf8e3a2 | 2512 | } |
b128c09f BB |
2513 | |
2514 | if (error == 0) { | |
2515 | BP_SET_LSIZE(new_bp, size); | |
2516 | BP_SET_PSIZE(new_bp, size); | |
2517 | BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF); | |
428870ff BB |
2518 | BP_SET_CHECKSUM(new_bp, |
2519 | spa_version(spa) >= SPA_VERSION_SLIM_ZIL | |
2520 | ? ZIO_CHECKSUM_ZILOG2 : ZIO_CHECKSUM_ZILOG); | |
b128c09f BB |
2521 | BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG); |
2522 | BP_SET_LEVEL(new_bp, 0); | |
428870ff | 2523 | BP_SET_DEDUP(new_bp, 0); |
b128c09f BB |
2524 | BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER); |
2525 | } | |
2526 | ||
2527 | return (error); | |
2528 | } | |
2529 | ||
2530 | /* | |
428870ff | 2531 | * Free an intent log block. |
b128c09f BB |
2532 | */ |
2533 | void | |
428870ff | 2534 | zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp) |
b128c09f | 2535 | { |
428870ff | 2536 | ASSERT(BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG); |
b128c09f BB |
2537 | ASSERT(!BP_IS_GANG(bp)); |
2538 | ||
428870ff | 2539 | zio_free(spa, txg, bp); |
b128c09f BB |
2540 | } |
2541 | ||
34dc7c2f BB |
2542 | /* |
2543 | * ========================================================================== | |
2544 | * Read and write to physical devices | |
2545 | * ========================================================================== | |
2546 | */ | |
34dc7c2f BB |
2547 | static int |
2548 | zio_vdev_io_start(zio_t *zio) | |
2549 | { | |
2550 | vdev_t *vd = zio->io_vd; | |
34dc7c2f BB |
2551 | uint64_t align; |
2552 | spa_t *spa = zio->io_spa; | |
2553 | ||
b128c09f BB |
2554 | ASSERT(zio->io_error == 0); |
2555 | ASSERT(zio->io_child_error[ZIO_CHILD_VDEV] == 0); | |
34dc7c2f | 2556 | |
b128c09f BB |
2557 | if (vd == NULL) { |
2558 | if (!(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) | |
2559 | spa_config_enter(spa, SCL_ZIO, zio, RW_READER); | |
34dc7c2f | 2560 | |
b128c09f BB |
2561 | /* |
2562 | * The mirror_ops handle multiple DVAs in a single BP. | |
2563 | */ | |
2564 | return (vdev_mirror_ops.vdev_op_io_start(zio)); | |
34dc7c2f BB |
2565 | } |
2566 | ||
572e2857 BB |
2567 | /* |
2568 | * We keep track of time-sensitive I/Os so that the scan thread | |
2569 | * can quickly react to certain workloads. In particular, we care | |
2570 | * about non-scrubbing, top-level reads and writes with the following | |
2571 | * characteristics: | |
2572 | * - synchronous writes of user data to non-slog devices | |
2573 | * - any reads of user data | |
2574 | * When these conditions are met, adjust the timestamp of spa_last_io | |
2575 | * which allows the scan thread to adjust its workload accordingly. | |
2576 | */ | |
2577 | if (!(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_bp != NULL && | |
2578 | vd == vd->vdev_top && !vd->vdev_islog && | |
2579 | zio->io_bookmark.zb_objset != DMU_META_OBJSET && | |
2580 | zio->io_txg != spa_syncing_txg(spa)) { | |
2581 | uint64_t old = spa->spa_last_io; | |
2582 | uint64_t new = ddi_get_lbolt64(); | |
2583 | if (old != new) | |
2584 | (void) atomic_cas_64(&spa->spa_last_io, old, new); | |
2585 | } | |
2586 | ||
b128c09f BB |
2587 | align = 1ULL << vd->vdev_top->vdev_ashift; |
2588 | ||
178e73b3 | 2589 | if (P2PHASE(zio->io_size, align) != 0) { |
34dc7c2f BB |
2590 | uint64_t asize = P2ROUNDUP(zio->io_size, align); |
2591 | char *abuf = zio_buf_alloc(asize); | |
178e73b3 | 2592 | ASSERT(vd == vd->vdev_top); |
34dc7c2f BB |
2593 | if (zio->io_type == ZIO_TYPE_WRITE) { |
2594 | bcopy(zio->io_data, abuf, zio->io_size); | |
2595 | bzero(abuf + zio->io_size, asize - zio->io_size); | |
2596 | } | |
b128c09f | 2597 | zio_push_transform(zio, abuf, asize, asize, zio_subblock); |
34dc7c2f BB |
2598 | } |
2599 | ||
2600 | ASSERT(P2PHASE(zio->io_offset, align) == 0); | |
178e73b3 | 2601 | ASSERT(P2PHASE(zio->io_size, align) == 0); |
572e2857 | 2602 | VERIFY(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa)); |
fb5f0bc8 BB |
2603 | |
2604 | /* | |
2605 | * If this is a repair I/O, and there's no self-healing involved -- | |
2606 | * that is, we're just resilvering what we expect to resilver -- | |
2607 | * then don't do the I/O unless zio's txg is actually in vd's DTL. | |
2608 | * This prevents spurious resilvering with nested replication. | |
2609 | * For example, given a mirror of mirrors, (A+B)+(C+D), if only | |
2610 | * A is out of date, we'll read from C+D, then use the data to | |
2611 | * resilver A+B -- but we don't actually want to resilver B, just A. | |
2612 | * The top-level mirror has no way to know this, so instead we just | |
2613 | * discard unnecessary repairs as we work our way down the vdev tree. | |
2614 | * The same logic applies to any form of nested replication: | |
2615 | * ditto + mirror, RAID-Z + replacing, etc. This covers them all. | |
2616 | */ | |
2617 | if ((zio->io_flags & ZIO_FLAG_IO_REPAIR) && | |
2618 | !(zio->io_flags & ZIO_FLAG_SELF_HEAL) && | |
2619 | zio->io_txg != 0 && /* not a delegated i/o */ | |
2620 | !vdev_dtl_contains(vd, DTL_PARTIAL, zio->io_txg, 1)) { | |
2621 | ASSERT(zio->io_type == ZIO_TYPE_WRITE); | |
fb5f0bc8 BB |
2622 | zio_vdev_io_bypass(zio); |
2623 | return (ZIO_PIPELINE_CONTINUE); | |
2624 | } | |
34dc7c2f | 2625 | |
b128c09f BB |
2626 | if (vd->vdev_ops->vdev_op_leaf && |
2627 | (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE)) { | |
2628 | ||
2629 | if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0) | |
d164b209 | 2630 | return (ZIO_PIPELINE_CONTINUE); |
b128c09f BB |
2631 | |
2632 | if ((zio = vdev_queue_io(zio)) == NULL) | |
2633 | return (ZIO_PIPELINE_STOP); | |
2634 | ||
2635 | if (!vdev_accessible(vd, zio)) { | |
2e528b49 | 2636 | zio->io_error = SET_ERROR(ENXIO); |
b128c09f BB |
2637 | zio_interrupt(zio); |
2638 | return (ZIO_PIPELINE_STOP); | |
2639 | } | |
b128c09f BB |
2640 | } |
2641 | ||
34dc7c2f BB |
2642 | return (vd->vdev_ops->vdev_op_io_start(zio)); |
2643 | } | |
2644 | ||
2645 | static int | |
2646 | zio_vdev_io_done(zio_t *zio) | |
2647 | { | |
b128c09f BB |
2648 | vdev_t *vd = zio->io_vd; |
2649 | vdev_ops_t *ops = vd ? vd->vdev_ops : &vdev_mirror_ops; | |
2650 | boolean_t unexpected_error = B_FALSE; | |
34dc7c2f | 2651 | |
b128c09f BB |
2652 | if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) |
2653 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 2654 | |
b128c09f BB |
2655 | ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE); |
2656 | ||
2657 | if (vd != NULL && vd->vdev_ops->vdev_op_leaf) { | |
2658 | ||
2659 | vdev_queue_io_done(zio); | |
2660 | ||
2661 | if (zio->io_type == ZIO_TYPE_WRITE) | |
2662 | vdev_cache_write(zio); | |
2663 | ||
2664 | if (zio_injection_enabled && zio->io_error == 0) | |
9babb374 BB |
2665 | zio->io_error = zio_handle_device_injection(vd, |
2666 | zio, EIO); | |
b128c09f BB |
2667 | |
2668 | if (zio_injection_enabled && zio->io_error == 0) | |
2669 | zio->io_error = zio_handle_label_injection(zio, EIO); | |
2670 | ||
2671 | if (zio->io_error) { | |
2672 | if (!vdev_accessible(vd, zio)) { | |
2e528b49 | 2673 | zio->io_error = SET_ERROR(ENXIO); |
b128c09f BB |
2674 | } else { |
2675 | unexpected_error = B_TRUE; | |
2676 | } | |
2677 | } | |
2678 | } | |
2679 | ||
2680 | ops->vdev_op_io_done(zio); | |
34dc7c2f | 2681 | |
b128c09f | 2682 | if (unexpected_error) |
d164b209 | 2683 | VERIFY(vdev_probe(vd, zio) == NULL); |
34dc7c2f | 2684 | |
b128c09f | 2685 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
2686 | } |
2687 | ||
428870ff BB |
2688 | /* |
2689 | * For non-raidz ZIOs, we can just copy aside the bad data read from the | |
2690 | * disk, and use that to finish the checksum ereport later. | |
2691 | */ | |
2692 | static void | |
2693 | zio_vsd_default_cksum_finish(zio_cksum_report_t *zcr, | |
2694 | const void *good_buf) | |
2695 | { | |
2696 | /* no processing needed */ | |
2697 | zfs_ereport_finish_checksum(zcr, good_buf, zcr->zcr_cbdata, B_FALSE); | |
2698 | } | |
2699 | ||
2700 | /*ARGSUSED*/ | |
2701 | void | |
2702 | zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *ignored) | |
2703 | { | |
2704 | void *buf = zio_buf_alloc(zio->io_size); | |
2705 | ||
2706 | bcopy(zio->io_data, buf, zio->io_size); | |
2707 | ||
2708 | zcr->zcr_cbinfo = zio->io_size; | |
2709 | zcr->zcr_cbdata = buf; | |
2710 | zcr->zcr_finish = zio_vsd_default_cksum_finish; | |
2711 | zcr->zcr_free = zio_buf_free; | |
2712 | } | |
2713 | ||
34dc7c2f BB |
2714 | static int |
2715 | zio_vdev_io_assess(zio_t *zio) | |
2716 | { | |
2717 | vdev_t *vd = zio->io_vd; | |
b128c09f BB |
2718 | |
2719 | if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) | |
2720 | return (ZIO_PIPELINE_STOP); | |
2721 | ||
2722 | if (vd == NULL && !(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) | |
2723 | spa_config_exit(zio->io_spa, SCL_ZIO, zio); | |
2724 | ||
2725 | if (zio->io_vsd != NULL) { | |
428870ff | 2726 | zio->io_vsd_ops->vsd_free(zio); |
b128c09f | 2727 | zio->io_vsd = NULL; |
34dc7c2f BB |
2728 | } |
2729 | ||
b128c09f | 2730 | if (zio_injection_enabled && zio->io_error == 0) |
34dc7c2f BB |
2731 | zio->io_error = zio_handle_fault_injection(zio, EIO); |
2732 | ||
2733 | /* | |
2734 | * If the I/O failed, determine whether we should attempt to retry it. | |
428870ff BB |
2735 | * |
2736 | * On retry, we cut in line in the issue queue, since we don't want | |
2737 | * compression/checksumming/etc. work to prevent our (cheap) IO reissue. | |
34dc7c2f | 2738 | */ |
b128c09f BB |
2739 | if (zio->io_error && vd == NULL && |
2740 | !(zio->io_flags & (ZIO_FLAG_DONT_RETRY | ZIO_FLAG_IO_RETRY))) { | |
2741 | ASSERT(!(zio->io_flags & ZIO_FLAG_DONT_QUEUE)); /* not a leaf */ | |
2742 | ASSERT(!(zio->io_flags & ZIO_FLAG_IO_BYPASS)); /* not a leaf */ | |
34dc7c2f | 2743 | zio->io_error = 0; |
b128c09f BB |
2744 | zio->io_flags |= ZIO_FLAG_IO_RETRY | |
2745 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE; | |
428870ff BB |
2746 | zio->io_stage = ZIO_STAGE_VDEV_IO_START >> 1; |
2747 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, | |
2748 | zio_requeue_io_start_cut_in_line); | |
b128c09f | 2749 | return (ZIO_PIPELINE_STOP); |
34dc7c2f BB |
2750 | } |
2751 | ||
b128c09f BB |
2752 | /* |
2753 | * If we got an error on a leaf device, convert it to ENXIO | |
2754 | * if the device is not accessible at all. | |
2755 | */ | |
2756 | if (zio->io_error && vd != NULL && vd->vdev_ops->vdev_op_leaf && | |
2757 | !vdev_accessible(vd, zio)) | |
2e528b49 | 2758 | zio->io_error = SET_ERROR(ENXIO); |
b128c09f BB |
2759 | |
2760 | /* | |
2761 | * If we can't write to an interior vdev (mirror or RAID-Z), | |
2762 | * set vdev_cant_write so that we stop trying to allocate from it. | |
2763 | */ | |
2764 | if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE && | |
13fe0198 | 2765 | vd != NULL && !vd->vdev_ops->vdev_op_leaf) { |
b128c09f | 2766 | vd->vdev_cant_write = B_TRUE; |
13fe0198 | 2767 | } |
b128c09f BB |
2768 | |
2769 | if (zio->io_error) | |
2770 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
2771 | ||
e8b96c60 MA |
2772 | if (vd != NULL && vd->vdev_ops->vdev_op_leaf && |
2773 | zio->io_physdone != NULL) { | |
2774 | ASSERT(!(zio->io_flags & ZIO_FLAG_DELEGATED)); | |
2775 | ASSERT(zio->io_child_type == ZIO_CHILD_VDEV); | |
2776 | zio->io_physdone(zio->io_logical); | |
2777 | } | |
2778 | ||
34dc7c2f BB |
2779 | return (ZIO_PIPELINE_CONTINUE); |
2780 | } | |
2781 | ||
2782 | void | |
2783 | zio_vdev_io_reissue(zio_t *zio) | |
2784 | { | |
2785 | ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); | |
2786 | ASSERT(zio->io_error == 0); | |
2787 | ||
428870ff | 2788 | zio->io_stage >>= 1; |
34dc7c2f BB |
2789 | } |
2790 | ||
2791 | void | |
2792 | zio_vdev_io_redone(zio_t *zio) | |
2793 | { | |
2794 | ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_DONE); | |
2795 | ||
428870ff | 2796 | zio->io_stage >>= 1; |
34dc7c2f BB |
2797 | } |
2798 | ||
2799 | void | |
2800 | zio_vdev_io_bypass(zio_t *zio) | |
2801 | { | |
2802 | ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); | |
2803 | ASSERT(zio->io_error == 0); | |
2804 | ||
2805 | zio->io_flags |= ZIO_FLAG_IO_BYPASS; | |
428870ff | 2806 | zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS >> 1; |
34dc7c2f BB |
2807 | } |
2808 | ||
2809 | /* | |
2810 | * ========================================================================== | |
2811 | * Generate and verify checksums | |
2812 | * ========================================================================== | |
2813 | */ | |
2814 | static int | |
2815 | zio_checksum_generate(zio_t *zio) | |
2816 | { | |
34dc7c2f | 2817 | blkptr_t *bp = zio->io_bp; |
b128c09f | 2818 | enum zio_checksum checksum; |
34dc7c2f | 2819 | |
b128c09f BB |
2820 | if (bp == NULL) { |
2821 | /* | |
2822 | * This is zio_write_phys(). | |
2823 | * We're either generating a label checksum, or none at all. | |
2824 | */ | |
2825 | checksum = zio->io_prop.zp_checksum; | |
34dc7c2f | 2826 | |
b128c09f BB |
2827 | if (checksum == ZIO_CHECKSUM_OFF) |
2828 | return (ZIO_PIPELINE_CONTINUE); | |
2829 | ||
2830 | ASSERT(checksum == ZIO_CHECKSUM_LABEL); | |
2831 | } else { | |
2832 | if (BP_IS_GANG(bp) && zio->io_child_type == ZIO_CHILD_GANG) { | |
2833 | ASSERT(!IO_IS_ALLOCATING(zio)); | |
2834 | checksum = ZIO_CHECKSUM_GANG_HEADER; | |
2835 | } else { | |
2836 | checksum = BP_GET_CHECKSUM(bp); | |
2837 | } | |
2838 | } | |
34dc7c2f | 2839 | |
b128c09f | 2840 | zio_checksum_compute(zio, checksum, zio->io_data, zio->io_size); |
34dc7c2f BB |
2841 | |
2842 | return (ZIO_PIPELINE_CONTINUE); | |
2843 | } | |
2844 | ||
2845 | static int | |
b128c09f | 2846 | zio_checksum_verify(zio_t *zio) |
34dc7c2f | 2847 | { |
428870ff | 2848 | zio_bad_cksum_t info; |
b128c09f BB |
2849 | blkptr_t *bp = zio->io_bp; |
2850 | int error; | |
34dc7c2f | 2851 | |
428870ff BB |
2852 | ASSERT(zio->io_vd != NULL); |
2853 | ||
b128c09f BB |
2854 | if (bp == NULL) { |
2855 | /* | |
2856 | * This is zio_read_phys(). | |
2857 | * We're either verifying a label checksum, or nothing at all. | |
2858 | */ | |
2859 | if (zio->io_prop.zp_checksum == ZIO_CHECKSUM_OFF) | |
2860 | return (ZIO_PIPELINE_CONTINUE); | |
34dc7c2f | 2861 | |
b128c09f BB |
2862 | ASSERT(zio->io_prop.zp_checksum == ZIO_CHECKSUM_LABEL); |
2863 | } | |
34dc7c2f | 2864 | |
428870ff | 2865 | if ((error = zio_checksum_error(zio, &info)) != 0) { |
b128c09f BB |
2866 | zio->io_error = error; |
2867 | if (!(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { | |
428870ff BB |
2868 | zfs_ereport_start_checksum(zio->io_spa, |
2869 | zio->io_vd, zio, zio->io_offset, | |
2870 | zio->io_size, NULL, &info); | |
b128c09f | 2871 | } |
34dc7c2f BB |
2872 | } |
2873 | ||
2874 | return (ZIO_PIPELINE_CONTINUE); | |
2875 | } | |
2876 | ||
2877 | /* | |
2878 | * Called by RAID-Z to ensure we don't compute the checksum twice. | |
2879 | */ | |
2880 | void | |
2881 | zio_checksum_verified(zio_t *zio) | |
2882 | { | |
428870ff | 2883 | zio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; |
34dc7c2f BB |
2884 | } |
2885 | ||
2886 | /* | |
b128c09f BB |
2887 | * ========================================================================== |
2888 | * Error rank. Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other. | |
2889 | * An error of 0 indictes success. ENXIO indicates whole-device failure, | |
2890 | * which may be transient (e.g. unplugged) or permament. ECKSUM and EIO | |
2891 | * indicate errors that are specific to one I/O, and most likely permanent. | |
2892 | * Any other error is presumed to be worse because we weren't expecting it. | |
2893 | * ========================================================================== | |
34dc7c2f | 2894 | */ |
b128c09f BB |
2895 | int |
2896 | zio_worst_error(int e1, int e2) | |
34dc7c2f | 2897 | { |
b128c09f BB |
2898 | static int zio_error_rank[] = { 0, ENXIO, ECKSUM, EIO }; |
2899 | int r1, r2; | |
2900 | ||
2901 | for (r1 = 0; r1 < sizeof (zio_error_rank) / sizeof (int); r1++) | |
2902 | if (e1 == zio_error_rank[r1]) | |
2903 | break; | |
34dc7c2f | 2904 | |
b128c09f BB |
2905 | for (r2 = 0; r2 < sizeof (zio_error_rank) / sizeof (int); r2++) |
2906 | if (e2 == zio_error_rank[r2]) | |
2907 | break; | |
2908 | ||
2909 | return (r1 > r2 ? e1 : e2); | |
34dc7c2f BB |
2910 | } |
2911 | ||
2912 | /* | |
2913 | * ========================================================================== | |
b128c09f | 2914 | * I/O completion |
34dc7c2f BB |
2915 | * ========================================================================== |
2916 | */ | |
b128c09f BB |
2917 | static int |
2918 | zio_ready(zio_t *zio) | |
34dc7c2f | 2919 | { |
b128c09f | 2920 | blkptr_t *bp = zio->io_bp; |
d164b209 | 2921 | zio_t *pio, *pio_next; |
34dc7c2f | 2922 | |
428870ff BB |
2923 | if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || |
2924 | zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_READY)) | |
9babb374 | 2925 | return (ZIO_PIPELINE_STOP); |
34dc7c2f | 2926 | |
9babb374 | 2927 | if (zio->io_ready) { |
b128c09f | 2928 | ASSERT(IO_IS_ALLOCATING(zio)); |
03c6040b GW |
2929 | ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp) || |
2930 | (zio->io_flags & ZIO_FLAG_NOPWRITE)); | |
b128c09f | 2931 | ASSERT(zio->io_children[ZIO_CHILD_GANG][ZIO_WAIT_READY] == 0); |
34dc7c2f | 2932 | |
b128c09f BB |
2933 | zio->io_ready(zio); |
2934 | } | |
34dc7c2f | 2935 | |
b128c09f BB |
2936 | if (bp != NULL && bp != &zio->io_bp_copy) |
2937 | zio->io_bp_copy = *bp; | |
34dc7c2f | 2938 | |
b128c09f BB |
2939 | if (zio->io_error) |
2940 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
34dc7c2f | 2941 | |
d164b209 BB |
2942 | mutex_enter(&zio->io_lock); |
2943 | zio->io_state[ZIO_WAIT_READY] = 1; | |
2944 | pio = zio_walk_parents(zio); | |
2945 | mutex_exit(&zio->io_lock); | |
2946 | ||
2947 | /* | |
2948 | * As we notify zio's parents, new parents could be added. | |
2949 | * New parents go to the head of zio's io_parent_list, however, | |
2950 | * so we will (correctly) not notify them. The remainder of zio's | |
2951 | * io_parent_list, from 'pio_next' onward, cannot change because | |
2952 | * all parents must wait for us to be done before they can be done. | |
2953 | */ | |
2954 | for (; pio != NULL; pio = pio_next) { | |
2955 | pio_next = zio_walk_parents(zio); | |
b128c09f | 2956 | zio_notify_parent(pio, zio, ZIO_WAIT_READY); |
d164b209 | 2957 | } |
34dc7c2f | 2958 | |
428870ff BB |
2959 | if (zio->io_flags & ZIO_FLAG_NODATA) { |
2960 | if (BP_IS_GANG(bp)) { | |
2961 | zio->io_flags &= ~ZIO_FLAG_NODATA; | |
2962 | } else { | |
2963 | ASSERT((uintptr_t)zio->io_data < SPA_MAXBLOCKSIZE); | |
2964 | zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; | |
2965 | } | |
2966 | } | |
2967 | ||
2968 | if (zio_injection_enabled && | |
2969 | zio->io_spa->spa_syncing_txg == zio->io_txg) | |
2970 | zio_handle_ignored_writes(zio); | |
2971 | ||
b128c09f | 2972 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
2973 | } |
2974 | ||
b128c09f BB |
2975 | static int |
2976 | zio_done(zio_t *zio) | |
34dc7c2f | 2977 | { |
d164b209 | 2978 | zio_t *pio, *pio_next; |
d6320ddb | 2979 | int c, w; |
34dc7c2f | 2980 | |
b128c09f | 2981 | /* |
9babb374 | 2982 | * If our children haven't all completed, |
b128c09f BB |
2983 | * wait for them and then repeat this pipeline stage. |
2984 | */ | |
2985 | if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE) || | |
2986 | zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE) || | |
428870ff | 2987 | zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE) || |
b128c09f BB |
2988 | zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_DONE)) |
2989 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 2990 | |
d6320ddb BB |
2991 | for (c = 0; c < ZIO_CHILD_TYPES; c++) |
2992 | for (w = 0; w < ZIO_WAIT_TYPES; w++) | |
b128c09f BB |
2993 | ASSERT(zio->io_children[c][w] == 0); |
2994 | ||
c776b317 BB |
2995 | if (zio->io_bp != NULL) { |
2996 | ASSERT(zio->io_bp->blk_pad[0] == 0); | |
2997 | ASSERT(zio->io_bp->blk_pad[1] == 0); | |
d1d7e268 MK |
2998 | ASSERT(bcmp(zio->io_bp, &zio->io_bp_copy, |
2999 | sizeof (blkptr_t)) == 0 || | |
c776b317 BB |
3000 | (zio->io_bp == zio_unique_parent(zio)->io_bp)); |
3001 | if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(zio->io_bp) && | |
428870ff | 3002 | zio->io_bp_override == NULL && |
b128c09f | 3003 | !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) { |
c776b317 | 3004 | ASSERT(!BP_SHOULD_BYTESWAP(zio->io_bp)); |
d1d7e268 MK |
3005 | ASSERT3U(zio->io_prop.zp_copies, <=, |
3006 | BP_GET_NDVAS(zio->io_bp)); | |
c776b317 | 3007 | ASSERT(BP_COUNT_GANG(zio->io_bp) == 0 || |
d1d7e268 MK |
3008 | (BP_COUNT_GANG(zio->io_bp) == |
3009 | BP_GET_NDVAS(zio->io_bp))); | |
b128c09f | 3010 | } |
03c6040b GW |
3011 | if (zio->io_flags & ZIO_FLAG_NOPWRITE) |
3012 | VERIFY(BP_EQUAL(zio->io_bp, &zio->io_bp_orig)); | |
b128c09f BB |
3013 | } |
3014 | ||
3015 | /* | |
428870ff | 3016 | * If there were child vdev/gang/ddt errors, they apply to us now. |
b128c09f BB |
3017 | */ |
3018 | zio_inherit_child_errors(zio, ZIO_CHILD_VDEV); | |
3019 | zio_inherit_child_errors(zio, ZIO_CHILD_GANG); | |
428870ff BB |
3020 | zio_inherit_child_errors(zio, ZIO_CHILD_DDT); |
3021 | ||
3022 | /* | |
3023 | * If the I/O on the transformed data was successful, generate any | |
3024 | * checksum reports now while we still have the transformed data. | |
3025 | */ | |
3026 | if (zio->io_error == 0) { | |
3027 | while (zio->io_cksum_report != NULL) { | |
3028 | zio_cksum_report_t *zcr = zio->io_cksum_report; | |
3029 | uint64_t align = zcr->zcr_align; | |
c776b317 | 3030 | uint64_t asize = P2ROUNDUP(zio->io_size, align); |
428870ff BB |
3031 | char *abuf = zio->io_data; |
3032 | ||
c776b317 | 3033 | if (asize != zio->io_size) { |
428870ff | 3034 | abuf = zio_buf_alloc(asize); |
c776b317 | 3035 | bcopy(zio->io_data, abuf, zio->io_size); |
d1d7e268 | 3036 | bzero(abuf+zio->io_size, asize-zio->io_size); |
428870ff BB |
3037 | } |
3038 | ||
3039 | zio->io_cksum_report = zcr->zcr_next; | |
3040 | zcr->zcr_next = NULL; | |
3041 | zcr->zcr_finish(zcr, abuf); | |
3042 | zfs_ereport_free_checksum(zcr); | |
3043 | ||
c776b317 | 3044 | if (asize != zio->io_size) |
428870ff BB |
3045 | zio_buf_free(abuf, asize); |
3046 | } | |
3047 | } | |
b128c09f BB |
3048 | |
3049 | zio_pop_transforms(zio); /* note: may set zio->io_error */ | |
3050 | ||
c776b317 | 3051 | vdev_stat_update(zio, zio->io_size); |
b128c09f | 3052 | |
a69052be | 3053 | /* |
cc92e9d0 | 3054 | * If this I/O is attached to a particular vdev is slow, exceeding |
72f53c56 MJ |
3055 | * 30 seconds to complete, post an error described the I/O delay. |
3056 | * We ignore these errors if the device is currently unavailable. | |
a69052be | 3057 | */ |
cc92e9d0 | 3058 | if (zio->io_delay >= MSEC_TO_TICK(zio_delay_max)) { |
72f53c56 MJ |
3059 | if (zio->io_vd != NULL && !vdev_is_dead(zio->io_vd)) |
3060 | zfs_ereport_post(FM_EREPORT_ZFS_DELAY, zio->io_spa, | |
d1d7e268 | 3061 | zio->io_vd, zio, 0, 0); |
72f53c56 | 3062 | } |
a69052be | 3063 | |
b128c09f BB |
3064 | if (zio->io_error) { |
3065 | /* | |
3066 | * If this I/O is attached to a particular vdev, | |
3067 | * generate an error message describing the I/O failure | |
3068 | * at the block level. We ignore these errors if the | |
3069 | * device is currently unavailable. | |
3070 | */ | |
c776b317 BB |
3071 | if (zio->io_error != ECKSUM && zio->io_vd != NULL && |
3072 | !vdev_is_dead(zio->io_vd)) | |
3073 | zfs_ereport_post(FM_EREPORT_ZFS_IO, zio->io_spa, | |
3074 | zio->io_vd, zio, 0, 0); | |
34dc7c2f | 3075 | |
428870ff BB |
3076 | if ((zio->io_error == EIO || !(zio->io_flags & |
3077 | (ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) && | |
c776b317 | 3078 | zio == zio->io_logical) { |
b128c09f BB |
3079 | /* |
3080 | * For logical I/O requests, tell the SPA to log the | |
3081 | * error and generate a logical data ereport. | |
3082 | */ | |
c776b317 | 3083 | spa_log_error(zio->io_spa, zio); |
d1d7e268 MK |
3084 | zfs_ereport_post(FM_EREPORT_ZFS_DATA, zio->io_spa, |
3085 | NULL, zio, 0, 0); | |
b128c09f BB |
3086 | } |
3087 | } | |
34dc7c2f | 3088 | |
c776b317 | 3089 | if (zio->io_error && zio == zio->io_logical) { |
b128c09f BB |
3090 | /* |
3091 | * Determine whether zio should be reexecuted. This will | |
3092 | * propagate all the way to the root via zio_notify_parent(). | |
3093 | */ | |
c776b317 | 3094 | ASSERT(zio->io_vd == NULL && zio->io_bp != NULL); |
428870ff | 3095 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); |
b128c09f | 3096 | |
428870ff BB |
3097 | if (IO_IS_ALLOCATING(zio) && |
3098 | !(zio->io_flags & ZIO_FLAG_CANFAIL)) { | |
b128c09f BB |
3099 | if (zio->io_error != ENOSPC) |
3100 | zio->io_reexecute |= ZIO_REEXECUTE_NOW; | |
3101 | else | |
3102 | zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; | |
428870ff | 3103 | } |
b128c09f BB |
3104 | |
3105 | if ((zio->io_type == ZIO_TYPE_READ || | |
3106 | zio->io_type == ZIO_TYPE_FREE) && | |
572e2857 | 3107 | !(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && |
b128c09f | 3108 | zio->io_error == ENXIO && |
c776b317 BB |
3109 | spa_load_state(zio->io_spa) == SPA_LOAD_NONE && |
3110 | spa_get_failmode(zio->io_spa) != ZIO_FAILURE_MODE_CONTINUE) | |
b128c09f BB |
3111 | zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; |
3112 | ||
3113 | if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute) | |
3114 | zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; | |
428870ff BB |
3115 | |
3116 | /* | |
3117 | * Here is a possibly good place to attempt to do | |
3118 | * either combinatorial reconstruction or error correction | |
3119 | * based on checksums. It also might be a good place | |
3120 | * to send out preliminary ereports before we suspend | |
3121 | * processing. | |
3122 | */ | |
34dc7c2f BB |
3123 | } |
3124 | ||
3125 | /* | |
b128c09f BB |
3126 | * If there were logical child errors, they apply to us now. |
3127 | * We defer this until now to avoid conflating logical child | |
3128 | * errors with errors that happened to the zio itself when | |
3129 | * updating vdev stats and reporting FMA events above. | |
34dc7c2f | 3130 | */ |
b128c09f | 3131 | zio_inherit_child_errors(zio, ZIO_CHILD_LOGICAL); |
34dc7c2f | 3132 | |
428870ff BB |
3133 | if ((zio->io_error || zio->io_reexecute) && |
3134 | IO_IS_ALLOCATING(zio) && zio->io_gang_leader == zio && | |
03c6040b | 3135 | !(zio->io_flags & (ZIO_FLAG_IO_REWRITE | ZIO_FLAG_NOPWRITE))) |
c776b317 | 3136 | zio_dva_unallocate(zio, zio->io_gang_tree, zio->io_bp); |
9babb374 BB |
3137 | |
3138 | zio_gang_tree_free(&zio->io_gang_tree); | |
3139 | ||
3140 | /* | |
3141 | * Godfather I/Os should never suspend. | |
3142 | */ | |
3143 | if ((zio->io_flags & ZIO_FLAG_GODFATHER) && | |
3144 | (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) | |
3145 | zio->io_reexecute = 0; | |
3146 | ||
b128c09f BB |
3147 | if (zio->io_reexecute) { |
3148 | /* | |
3149 | * This is a logical I/O that wants to reexecute. | |
3150 | * | |
3151 | * Reexecute is top-down. When an i/o fails, if it's not | |
3152 | * the root, it simply notifies its parent and sticks around. | |
3153 | * The parent, seeing that it still has children in zio_done(), | |
3154 | * does the same. This percolates all the way up to the root. | |
3155 | * The root i/o will reexecute or suspend the entire tree. | |
3156 | * | |
3157 | * This approach ensures that zio_reexecute() honors | |
3158 | * all the original i/o dependency relationships, e.g. | |
3159 | * parents not executing until children are ready. | |
3160 | */ | |
3161 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
34dc7c2f | 3162 | |
9babb374 | 3163 | zio->io_gang_leader = NULL; |
b128c09f | 3164 | |
d164b209 BB |
3165 | mutex_enter(&zio->io_lock); |
3166 | zio->io_state[ZIO_WAIT_DONE] = 1; | |
3167 | mutex_exit(&zio->io_lock); | |
3168 | ||
9babb374 BB |
3169 | /* |
3170 | * "The Godfather" I/O monitors its children but is | |
3171 | * not a true parent to them. It will track them through | |
3172 | * the pipeline but severs its ties whenever they get into | |
3173 | * trouble (e.g. suspended). This allows "The Godfather" | |
3174 | * I/O to return status without blocking. | |
3175 | */ | |
3176 | for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { | |
3177 | zio_link_t *zl = zio->io_walk_link; | |
3178 | pio_next = zio_walk_parents(zio); | |
3179 | ||
3180 | if ((pio->io_flags & ZIO_FLAG_GODFATHER) && | |
3181 | (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) { | |
3182 | zio_remove_child(pio, zio, zl); | |
3183 | zio_notify_parent(pio, zio, ZIO_WAIT_DONE); | |
3184 | } | |
3185 | } | |
3186 | ||
d164b209 | 3187 | if ((pio = zio_unique_parent(zio)) != NULL) { |
b128c09f BB |
3188 | /* |
3189 | * We're not a root i/o, so there's nothing to do | |
3190 | * but notify our parent. Don't propagate errors | |
3191 | * upward since we haven't permanently failed yet. | |
3192 | */ | |
9babb374 | 3193 | ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); |
b128c09f BB |
3194 | zio->io_flags |= ZIO_FLAG_DONT_PROPAGATE; |
3195 | zio_notify_parent(pio, zio, ZIO_WAIT_DONE); | |
3196 | } else if (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND) { | |
3197 | /* | |
3198 | * We'd fail again if we reexecuted now, so suspend | |
3199 | * until conditions improve (e.g. device comes online). | |
3200 | */ | |
c776b317 | 3201 | zio_suspend(zio->io_spa, zio); |
b128c09f BB |
3202 | } else { |
3203 | /* | |
3204 | * Reexecution is potentially a huge amount of work. | |
3205 | * Hand it off to the otherwise-unused claim taskq. | |
3206 | */ | |
a38718a6 | 3207 | ASSERT(taskq_empty_ent(&zio->io_tqent)); |
7ef5e54e AL |
3208 | spa_taskq_dispatch_ent(zio->io_spa, |
3209 | ZIO_TYPE_CLAIM, ZIO_TASKQ_ISSUE, | |
a38718a6 GA |
3210 | (task_func_t *)zio_reexecute, zio, 0, |
3211 | &zio->io_tqent); | |
b128c09f BB |
3212 | } |
3213 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f BB |
3214 | } |
3215 | ||
428870ff | 3216 | ASSERT(zio->io_child_count == 0); |
b128c09f BB |
3217 | ASSERT(zio->io_reexecute == 0); |
3218 | ASSERT(zio->io_error == 0 || (zio->io_flags & ZIO_FLAG_CANFAIL)); | |
34dc7c2f | 3219 | |
428870ff BB |
3220 | /* |
3221 | * Report any checksum errors, since the I/O is complete. | |
3222 | */ | |
3223 | while (zio->io_cksum_report != NULL) { | |
3224 | zio_cksum_report_t *zcr = zio->io_cksum_report; | |
3225 | zio->io_cksum_report = zcr->zcr_next; | |
3226 | zcr->zcr_next = NULL; | |
3227 | zcr->zcr_finish(zcr, NULL); | |
3228 | zfs_ereport_free_checksum(zcr); | |
3229 | } | |
3230 | ||
920dd524 | 3231 | if (zio->io_flags & ZIO_FLAG_FASTWRITE && zio->io_bp && |
03c6040b | 3232 | !BP_IS_HOLE(zio->io_bp) && !(zio->io_flags & ZIO_FLAG_NOPWRITE)) { |
920dd524 ED |
3233 | metaslab_fastwrite_unmark(zio->io_spa, zio->io_bp); |
3234 | } | |
3235 | ||
d164b209 BB |
3236 | /* |
3237 | * It is the responsibility of the done callback to ensure that this | |
3238 | * particular zio is no longer discoverable for adoption, and as | |
3239 | * such, cannot acquire any new parents. | |
3240 | */ | |
b128c09f BB |
3241 | if (zio->io_done) |
3242 | zio->io_done(zio); | |
34dc7c2f | 3243 | |
d164b209 BB |
3244 | mutex_enter(&zio->io_lock); |
3245 | zio->io_state[ZIO_WAIT_DONE] = 1; | |
3246 | mutex_exit(&zio->io_lock); | |
34dc7c2f | 3247 | |
d164b209 BB |
3248 | for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { |
3249 | zio_link_t *zl = zio->io_walk_link; | |
3250 | pio_next = zio_walk_parents(zio); | |
3251 | zio_remove_child(pio, zio, zl); | |
b128c09f BB |
3252 | zio_notify_parent(pio, zio, ZIO_WAIT_DONE); |
3253 | } | |
34dc7c2f | 3254 | |
b128c09f BB |
3255 | if (zio->io_waiter != NULL) { |
3256 | mutex_enter(&zio->io_lock); | |
3257 | zio->io_executor = NULL; | |
3258 | cv_broadcast(&zio->io_cv); | |
3259 | mutex_exit(&zio->io_lock); | |
3260 | } else { | |
3261 | zio_destroy(zio); | |
3262 | } | |
34dc7c2f | 3263 | |
b128c09f | 3264 | return (ZIO_PIPELINE_STOP); |
34dc7c2f BB |
3265 | } |
3266 | ||
3267 | /* | |
b128c09f BB |
3268 | * ========================================================================== |
3269 | * I/O pipeline definition | |
3270 | * ========================================================================== | |
34dc7c2f | 3271 | */ |
428870ff | 3272 | static zio_pipe_stage_t *zio_pipeline[] = { |
b128c09f | 3273 | NULL, |
b128c09f | 3274 | zio_read_bp_init, |
428870ff BB |
3275 | zio_free_bp_init, |
3276 | zio_issue_async, | |
b128c09f BB |
3277 | zio_write_bp_init, |
3278 | zio_checksum_generate, | |
03c6040b | 3279 | zio_nop_write, |
428870ff BB |
3280 | zio_ddt_read_start, |
3281 | zio_ddt_read_done, | |
3282 | zio_ddt_write, | |
3283 | zio_ddt_free, | |
b128c09f BB |
3284 | zio_gang_assemble, |
3285 | zio_gang_issue, | |
3286 | zio_dva_allocate, | |
3287 | zio_dva_free, | |
3288 | zio_dva_claim, | |
3289 | zio_ready, | |
3290 | zio_vdev_io_start, | |
3291 | zio_vdev_io_done, | |
3292 | zio_vdev_io_assess, | |
3293 | zio_checksum_verify, | |
3294 | zio_done | |
3295 | }; | |
c28b2279 | 3296 | |
9ae529ec CS |
3297 | /* dnp is the dnode for zb1->zb_object */ |
3298 | boolean_t | |
3299 | zbookmark_is_before(const dnode_phys_t *dnp, const zbookmark_t *zb1, | |
3300 | const zbookmark_t *zb2) | |
3301 | { | |
3302 | uint64_t zb1nextL0, zb2thisobj; | |
3303 | ||
3304 | ASSERT(zb1->zb_objset == zb2->zb_objset); | |
3305 | ASSERT(zb2->zb_level == 0); | |
3306 | ||
3307 | /* | |
3308 | * A bookmark in the deadlist is considered to be after | |
3309 | * everything else. | |
3310 | */ | |
3311 | if (zb2->zb_object == DMU_DEADLIST_OBJECT) | |
3312 | return (B_TRUE); | |
3313 | ||
3314 | /* The objset_phys_t isn't before anything. */ | |
3315 | if (dnp == NULL) | |
3316 | return (B_FALSE); | |
3317 | ||
3318 | zb1nextL0 = (zb1->zb_blkid + 1) << | |
3319 | ((zb1->zb_level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)); | |
3320 | ||
3321 | zb2thisobj = zb2->zb_object ? zb2->zb_object : | |
3322 | zb2->zb_blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT); | |
3323 | ||
3324 | if (zb1->zb_object == DMU_META_DNODE_OBJECT) { | |
3325 | uint64_t nextobj = zb1nextL0 * | |
3326 | (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT) >> DNODE_SHIFT; | |
3327 | return (nextobj <= zb2thisobj); | |
3328 | } | |
3329 | ||
3330 | if (zb1->zb_object < zb2thisobj) | |
3331 | return (B_TRUE); | |
3332 | if (zb1->zb_object > zb2thisobj) | |
3333 | return (B_FALSE); | |
3334 | if (zb2->zb_object == DMU_META_DNODE_OBJECT) | |
3335 | return (B_FALSE); | |
3336 | return (zb1nextL0 <= zb2->zb_blkid); | |
3337 | } | |
3338 | ||
c28b2279 BB |
3339 | #if defined(_KERNEL) && defined(HAVE_SPL) |
3340 | /* Fault injection */ | |
3341 | EXPORT_SYMBOL(zio_injection_enabled); | |
3342 | EXPORT_SYMBOL(zio_inject_fault); | |
3343 | EXPORT_SYMBOL(zio_inject_list_next); | |
3344 | EXPORT_SYMBOL(zio_clear_fault); | |
3345 | EXPORT_SYMBOL(zio_handle_fault_injection); | |
3346 | EXPORT_SYMBOL(zio_handle_device_injection); | |
3347 | EXPORT_SYMBOL(zio_handle_label_injection); | |
c28b2279 BB |
3348 | EXPORT_SYMBOL(zio_type_name); |
3349 | ||
3350 | module_param(zio_bulk_flags, int, 0644); | |
3351 | MODULE_PARM_DESC(zio_bulk_flags, "Additional flags to pass to bulk buffers"); | |
a69052be BB |
3352 | |
3353 | module_param(zio_delay_max, int, 0644); | |
c409e464 BB |
3354 | MODULE_PARM_DESC(zio_delay_max, "Max zio millisec delay before posting event"); |
3355 | ||
3356 | module_param(zio_requeue_io_start_cut_in_line, int, 0644); | |
3357 | MODULE_PARM_DESC(zio_requeue_io_start_cut_in_line, "Prioritize requeued I/O"); | |
29dee3ee CP |
3358 | |
3359 | module_param(zfs_sync_pass_deferred_free, int, 0644); | |
3360 | MODULE_PARM_DESC(zfs_sync_pass_deferred_free, | |
d1d7e268 | 3361 | "Defer frees starting in this pass"); |
29dee3ee CP |
3362 | |
3363 | module_param(zfs_sync_pass_dont_compress, int, 0644); | |
3364 | MODULE_PARM_DESC(zfs_sync_pass_dont_compress, | |
d1d7e268 | 3365 | "Don't compress starting in this pass"); |
29dee3ee CP |
3366 | |
3367 | module_param(zfs_sync_pass_rewrite, int, 0644); | |
3368 | MODULE_PARM_DESC(zfs_sync_pass_rewrite, | |
d1d7e268 | 3369 | "Rewrite new bps starting in this pass"); |
c28b2279 | 3370 | #endif |