]>
Commit | Line | Data |
---|---|---|
34dc7c2f BB |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
428870ff | 22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
ebf8e3a2 | 23 | * Copyright (c) 2012 by Delphix. All rights reserved. |
a38718a6 | 24 | * Copyright (c) 2011 Nexenta Systems, Inc. All rights reserved. |
34dc7c2f BB |
25 | */ |
26 | ||
34dc7c2f BB |
27 | #include <sys/zfs_context.h> |
28 | #include <sys/fm/fs/zfs.h> | |
29 | #include <sys/spa.h> | |
30 | #include <sys/txg.h> | |
31 | #include <sys/spa_impl.h> | |
32 | #include <sys/vdev_impl.h> | |
33 | #include <sys/zio_impl.h> | |
34 | #include <sys/zio_compress.h> | |
35 | #include <sys/zio_checksum.h> | |
428870ff BB |
36 | #include <sys/dmu_objset.h> |
37 | #include <sys/arc.h> | |
38 | #include <sys/ddt.h> | |
34dc7c2f BB |
39 | |
40 | /* | |
41 | * ========================================================================== | |
42 | * I/O priority table | |
43 | * ========================================================================== | |
44 | */ | |
45 | uint8_t zio_priority_table[ZIO_PRIORITY_TABLE_SIZE] = { | |
46 | 0, /* ZIO_PRIORITY_NOW */ | |
47 | 0, /* ZIO_PRIORITY_SYNC_READ */ | |
48 | 0, /* ZIO_PRIORITY_SYNC_WRITE */ | |
34dc7c2f | 49 | 0, /* ZIO_PRIORITY_LOG_WRITE */ |
428870ff BB |
50 | 1, /* ZIO_PRIORITY_CACHE_FILL */ |
51 | 1, /* ZIO_PRIORITY_AGG */ | |
52 | 4, /* ZIO_PRIORITY_FREE */ | |
53 | 4, /* ZIO_PRIORITY_ASYNC_WRITE */ | |
54 | 6, /* ZIO_PRIORITY_ASYNC_READ */ | |
34dc7c2f BB |
55 | 10, /* ZIO_PRIORITY_RESILVER */ |
56 | 20, /* ZIO_PRIORITY_SCRUB */ | |
428870ff | 57 | 2, /* ZIO_PRIORITY_DDT_PREFETCH */ |
34dc7c2f BB |
58 | }; |
59 | ||
60 | /* | |
61 | * ========================================================================== | |
62 | * I/O type descriptions | |
63 | * ========================================================================== | |
64 | */ | |
65 | char *zio_type_name[ZIO_TYPES] = { | |
451041db | 66 | "z_null", "z_rd", "z_wr", "z_fr", "z_cl", "z_ioctl" |
428870ff | 67 | }; |
34dc7c2f BB |
68 | |
69 | /* | |
70 | * ========================================================================== | |
71 | * I/O kmem caches | |
72 | * ========================================================================== | |
73 | */ | |
74 | kmem_cache_t *zio_cache; | |
d164b209 | 75 | kmem_cache_t *zio_link_cache; |
86dd0fd9 | 76 | kmem_cache_t *zio_vdev_cache; |
34dc7c2f BB |
77 | kmem_cache_t *zio_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; |
78 | kmem_cache_t *zio_data_buf_cache[SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT]; | |
c28b2279 | 79 | int zio_bulk_flags = 0; |
a69052be | 80 | int zio_delay_max = ZIO_DELAY_MAX; |
34dc7c2f BB |
81 | |
82 | #ifdef _KERNEL | |
83 | extern vmem_t *zio_alloc_arena; | |
84 | #endif | |
6d974228 | 85 | extern int zfs_mg_alloc_failures; |
34dc7c2f | 86 | |
55d85d5a GW |
87 | /* |
88 | * The following actions directly effect the spa's sync-to-convergence logic. | |
89 | * The values below define the sync pass when we start performing the action. | |
90 | * Care should be taken when changing these values as they directly impact | |
91 | * spa_sync() performance. Tuning these values may introduce subtle performance | |
92 | * pathologies and should only be done in the context of performance analysis. | |
93 | * These tunables will eventually be removed and replaced with #defines once | |
94 | * enough analysis has been done to determine optimal values. | |
95 | * | |
96 | * The 'zfs_sync_pass_deferred_free' pass must be greater than 1 to ensure that | |
97 | * regular blocks are not deferred. | |
98 | */ | |
99 | int zfs_sync_pass_deferred_free = 2; /* defer frees starting in this pass */ | |
100 | int zfs_sync_pass_dont_compress = 5; /* don't compress starting in this pass */ | |
101 | int zfs_sync_pass_rewrite = 2; /* rewrite new bps starting in this pass */ | |
102 | ||
34dc7c2f | 103 | /* |
b128c09f BB |
104 | * An allocating zio is one that either currently has the DVA allocate |
105 | * stage set or will have it later in its lifetime. | |
34dc7c2f | 106 | */ |
428870ff BB |
107 | #define IO_IS_ALLOCATING(zio) ((zio)->io_orig_pipeline & ZIO_STAGE_DVA_ALLOCATE) |
108 | ||
c409e464 | 109 | int zio_requeue_io_start_cut_in_line = 1; |
428870ff BB |
110 | |
111 | #ifdef ZFS_DEBUG | |
112 | int zio_buf_debug_limit = 16384; | |
113 | #else | |
114 | int zio_buf_debug_limit = 0; | |
115 | #endif | |
34dc7c2f | 116 | |
da6b4005 NB |
117 | static inline void __zio_execute(zio_t *zio); |
118 | ||
49be0ccf BB |
119 | static int |
120 | zio_cons(void *arg, void *unused, int kmflag) | |
121 | { | |
122 | zio_t *zio = arg; | |
123 | ||
124 | bzero(zio, sizeof (zio_t)); | |
125 | ||
126 | mutex_init(&zio->io_lock, NULL, MUTEX_DEFAULT, NULL); | |
127 | cv_init(&zio->io_cv, NULL, CV_DEFAULT, NULL); | |
128 | ||
129 | list_create(&zio->io_parent_list, sizeof (zio_link_t), | |
130 | offsetof(zio_link_t, zl_parent_node)); | |
131 | list_create(&zio->io_child_list, sizeof (zio_link_t), | |
132 | offsetof(zio_link_t, zl_child_node)); | |
133 | ||
134 | return (0); | |
135 | } | |
136 | ||
137 | static void | |
138 | zio_dest(void *arg, void *unused) | |
139 | { | |
140 | zio_t *zio = arg; | |
141 | ||
142 | mutex_destroy(&zio->io_lock); | |
143 | cv_destroy(&zio->io_cv); | |
144 | list_destroy(&zio->io_parent_list); | |
145 | list_destroy(&zio->io_child_list); | |
146 | } | |
147 | ||
34dc7c2f BB |
148 | void |
149 | zio_init(void) | |
150 | { | |
151 | size_t c; | |
152 | vmem_t *data_alloc_arena = NULL; | |
153 | ||
154 | #ifdef _KERNEL | |
155 | data_alloc_arena = zio_alloc_arena; | |
156 | #endif | |
49be0ccf BB |
157 | zio_cache = kmem_cache_create("zio_cache", sizeof (zio_t), 0, |
158 | zio_cons, zio_dest, NULL, NULL, NULL, KMC_KMEM); | |
d164b209 | 159 | zio_link_cache = kmem_cache_create("zio_link_cache", |
ae6ba3db | 160 | sizeof (zio_link_t), 0, NULL, NULL, NULL, NULL, NULL, KMC_KMEM); |
86dd0fd9 BB |
161 | zio_vdev_cache = kmem_cache_create("zio_vdev_cache", sizeof(vdev_io_t), |
162 | PAGESIZE, NULL, NULL, NULL, NULL, NULL, KMC_VMEM); | |
34dc7c2f BB |
163 | |
164 | /* | |
165 | * For small buffers, we want a cache for each multiple of | |
166 | * SPA_MINBLOCKSIZE. For medium-size buffers, we want a cache | |
167 | * for each quarter-power of 2. For large buffers, we want | |
168 | * a cache for each multiple of PAGESIZE. | |
169 | */ | |
170 | for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { | |
171 | size_t size = (c + 1) << SPA_MINBLOCKSHIFT; | |
172 | size_t p2 = size; | |
173 | size_t align = 0; | |
174 | ||
175 | while (p2 & (p2 - 1)) | |
176 | p2 &= p2 - 1; | |
177 | ||
178 | if (size <= 4 * SPA_MINBLOCKSIZE) { | |
179 | align = SPA_MINBLOCKSIZE; | |
180 | } else if (P2PHASE(size, PAGESIZE) == 0) { | |
181 | align = PAGESIZE; | |
182 | } else if (P2PHASE(size, p2 >> 2) == 0) { | |
183 | align = p2 >> 2; | |
184 | } | |
185 | ||
186 | if (align != 0) { | |
187 | char name[36]; | |
ae6ba3db BB |
188 | int flags = zio_bulk_flags; |
189 | ||
190 | /* | |
191 | * The smallest buffers (512b) are heavily used and | |
192 | * experience a lot of churn. The slabs allocated | |
193 | * for them are also relatively small (32K). Thus | |
194 | * in over to avoid expensive calls to vmalloc() we | |
195 | * make an exception to the usual slab allocation | |
196 | * policy and force these buffers to be kmem backed. | |
197 | */ | |
198 | if (size == (1 << SPA_MINBLOCKSHIFT)) | |
199 | flags |= KMC_KMEM; | |
200 | ||
34dc7c2f BB |
201 | (void) sprintf(name, "zio_buf_%lu", (ulong_t)size); |
202 | zio_buf_cache[c] = kmem_cache_create(name, size, | |
ae6ba3db | 203 | align, NULL, NULL, NULL, NULL, NULL, flags); |
34dc7c2f BB |
204 | |
205 | (void) sprintf(name, "zio_data_buf_%lu", (ulong_t)size); | |
206 | zio_data_buf_cache[c] = kmem_cache_create(name, size, | |
ae6ba3db BB |
207 | align, NULL, NULL, NULL, NULL, |
208 | data_alloc_arena, flags); | |
34dc7c2f BB |
209 | } |
210 | } | |
211 | ||
212 | while (--c != 0) { | |
213 | ASSERT(zio_buf_cache[c] != NULL); | |
214 | if (zio_buf_cache[c - 1] == NULL) | |
215 | zio_buf_cache[c - 1] = zio_buf_cache[c]; | |
216 | ||
217 | ASSERT(zio_data_buf_cache[c] != NULL); | |
218 | if (zio_data_buf_cache[c - 1] == NULL) | |
219 | zio_data_buf_cache[c - 1] = zio_data_buf_cache[c]; | |
220 | } | |
221 | ||
6d974228 GW |
222 | /* |
223 | * The zio write taskqs have 1 thread per cpu, allow 1/2 of the taskqs | |
224 | * to fail 3 times per txg or 8 failures, whichever is greater. | |
225 | */ | |
226 | zfs_mg_alloc_failures = MAX((3 * max_ncpus / 2), 8); | |
227 | ||
34dc7c2f | 228 | zio_inject_init(); |
9759c60f ED |
229 | |
230 | lz4_init(); | |
34dc7c2f BB |
231 | } |
232 | ||
233 | void | |
234 | zio_fini(void) | |
235 | { | |
236 | size_t c; | |
237 | kmem_cache_t *last_cache = NULL; | |
238 | kmem_cache_t *last_data_cache = NULL; | |
239 | ||
240 | for (c = 0; c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT; c++) { | |
241 | if (zio_buf_cache[c] != last_cache) { | |
242 | last_cache = zio_buf_cache[c]; | |
243 | kmem_cache_destroy(zio_buf_cache[c]); | |
244 | } | |
245 | zio_buf_cache[c] = NULL; | |
246 | ||
247 | if (zio_data_buf_cache[c] != last_data_cache) { | |
248 | last_data_cache = zio_data_buf_cache[c]; | |
249 | kmem_cache_destroy(zio_data_buf_cache[c]); | |
250 | } | |
251 | zio_data_buf_cache[c] = NULL; | |
252 | } | |
253 | ||
86dd0fd9 | 254 | kmem_cache_destroy(zio_vdev_cache); |
d164b209 | 255 | kmem_cache_destroy(zio_link_cache); |
34dc7c2f BB |
256 | kmem_cache_destroy(zio_cache); |
257 | ||
258 | zio_inject_fini(); | |
9759c60f ED |
259 | |
260 | lz4_fini(); | |
34dc7c2f BB |
261 | } |
262 | ||
263 | /* | |
264 | * ========================================================================== | |
265 | * Allocate and free I/O buffers | |
266 | * ========================================================================== | |
267 | */ | |
268 | ||
269 | /* | |
270 | * Use zio_buf_alloc to allocate ZFS metadata. This data will appear in a | |
271 | * crashdump if the kernel panics, so use it judiciously. Obviously, it's | |
272 | * useful to inspect ZFS metadata, but if possible, we should avoid keeping | |
273 | * excess / transient data in-core during a crashdump. | |
274 | */ | |
275 | void * | |
276 | zio_buf_alloc(size_t size) | |
277 | { | |
278 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
279 | ||
280 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
281 | ||
ebcfc8a5 | 282 | return (kmem_cache_alloc(zio_buf_cache[c], KM_PUSHPAGE | KM_NODEBUG)); |
34dc7c2f BB |
283 | } |
284 | ||
285 | /* | |
286 | * Use zio_data_buf_alloc to allocate data. The data will not appear in a | |
287 | * crashdump if the kernel panics. This exists so that we will limit the amount | |
288 | * of ZFS data that shows up in a kernel crashdump. (Thus reducing the amount | |
289 | * of kernel heap dumped to disk when the kernel panics) | |
290 | */ | |
291 | void * | |
292 | zio_data_buf_alloc(size_t size) | |
293 | { | |
294 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
295 | ||
296 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
297 | ||
ebcfc8a5 BB |
298 | return (kmem_cache_alloc(zio_data_buf_cache[c], |
299 | KM_PUSHPAGE | KM_NODEBUG)); | |
34dc7c2f BB |
300 | } |
301 | ||
302 | void | |
303 | zio_buf_free(void *buf, size_t size) | |
304 | { | |
305 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
306 | ||
307 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
308 | ||
309 | kmem_cache_free(zio_buf_cache[c], buf); | |
310 | } | |
311 | ||
312 | void | |
313 | zio_data_buf_free(void *buf, size_t size) | |
314 | { | |
315 | size_t c = (size - 1) >> SPA_MINBLOCKSHIFT; | |
316 | ||
317 | ASSERT(c < SPA_MAXBLOCKSIZE >> SPA_MINBLOCKSHIFT); | |
318 | ||
319 | kmem_cache_free(zio_data_buf_cache[c], buf); | |
320 | } | |
321 | ||
86dd0fd9 BB |
322 | /* |
323 | * Dedicated I/O buffers to ensure that memory fragmentation never prevents | |
324 | * or significantly delays the issuing of a zio. These buffers are used | |
325 | * to aggregate I/O and could be used for raidz stripes. | |
326 | */ | |
327 | void * | |
328 | zio_vdev_alloc(void) | |
329 | { | |
330 | return (kmem_cache_alloc(zio_vdev_cache, KM_PUSHPAGE)); | |
331 | } | |
332 | ||
333 | void | |
334 | zio_vdev_free(void *buf) | |
335 | { | |
336 | kmem_cache_free(zio_vdev_cache, buf); | |
337 | ||
338 | } | |
339 | ||
34dc7c2f BB |
340 | /* |
341 | * ========================================================================== | |
342 | * Push and pop I/O transform buffers | |
343 | * ========================================================================== | |
344 | */ | |
345 | static void | |
b128c09f BB |
346 | zio_push_transform(zio_t *zio, void *data, uint64_t size, uint64_t bufsize, |
347 | zio_transform_func_t *transform) | |
34dc7c2f | 348 | { |
691f6ac4 | 349 | zio_transform_t *zt = kmem_alloc(sizeof (zio_transform_t), KM_PUSHPAGE); |
34dc7c2f | 350 | |
b128c09f BB |
351 | zt->zt_orig_data = zio->io_data; |
352 | zt->zt_orig_size = zio->io_size; | |
34dc7c2f | 353 | zt->zt_bufsize = bufsize; |
b128c09f | 354 | zt->zt_transform = transform; |
34dc7c2f BB |
355 | |
356 | zt->zt_next = zio->io_transform_stack; | |
357 | zio->io_transform_stack = zt; | |
358 | ||
359 | zio->io_data = data; | |
360 | zio->io_size = size; | |
361 | } | |
362 | ||
363 | static void | |
b128c09f | 364 | zio_pop_transforms(zio_t *zio) |
34dc7c2f | 365 | { |
b128c09f BB |
366 | zio_transform_t *zt; |
367 | ||
368 | while ((zt = zio->io_transform_stack) != NULL) { | |
369 | if (zt->zt_transform != NULL) | |
370 | zt->zt_transform(zio, | |
371 | zt->zt_orig_data, zt->zt_orig_size); | |
34dc7c2f | 372 | |
428870ff BB |
373 | if (zt->zt_bufsize != 0) |
374 | zio_buf_free(zio->io_data, zt->zt_bufsize); | |
34dc7c2f | 375 | |
b128c09f BB |
376 | zio->io_data = zt->zt_orig_data; |
377 | zio->io_size = zt->zt_orig_size; | |
378 | zio->io_transform_stack = zt->zt_next; | |
34dc7c2f | 379 | |
b128c09f | 380 | kmem_free(zt, sizeof (zio_transform_t)); |
34dc7c2f BB |
381 | } |
382 | } | |
383 | ||
b128c09f BB |
384 | /* |
385 | * ========================================================================== | |
386 | * I/O transform callbacks for subblocks and decompression | |
387 | * ========================================================================== | |
388 | */ | |
389 | static void | |
390 | zio_subblock(zio_t *zio, void *data, uint64_t size) | |
391 | { | |
392 | ASSERT(zio->io_size > size); | |
393 | ||
394 | if (zio->io_type == ZIO_TYPE_READ) | |
395 | bcopy(zio->io_data, data, size); | |
396 | } | |
397 | ||
398 | static void | |
399 | zio_decompress(zio_t *zio, void *data, uint64_t size) | |
400 | { | |
401 | if (zio->io_error == 0 && | |
402 | zio_decompress_data(BP_GET_COMPRESS(zio->io_bp), | |
428870ff | 403 | zio->io_data, data, zio->io_size, size) != 0) |
b128c09f BB |
404 | zio->io_error = EIO; |
405 | } | |
406 | ||
407 | /* | |
408 | * ========================================================================== | |
409 | * I/O parent/child relationships and pipeline interlocks | |
410 | * ========================================================================== | |
411 | */ | |
d164b209 BB |
412 | /* |
413 | * NOTE - Callers to zio_walk_parents() and zio_walk_children must | |
414 | * continue calling these functions until they return NULL. | |
415 | * Otherwise, the next caller will pick up the list walk in | |
416 | * some indeterminate state. (Otherwise every caller would | |
417 | * have to pass in a cookie to keep the state represented by | |
418 | * io_walk_link, which gets annoying.) | |
419 | */ | |
420 | zio_t * | |
421 | zio_walk_parents(zio_t *cio) | |
422 | { | |
423 | zio_link_t *zl = cio->io_walk_link; | |
424 | list_t *pl = &cio->io_parent_list; | |
b128c09f | 425 | |
d164b209 BB |
426 | zl = (zl == NULL) ? list_head(pl) : list_next(pl, zl); |
427 | cio->io_walk_link = zl; | |
428 | ||
429 | if (zl == NULL) | |
430 | return (NULL); | |
431 | ||
432 | ASSERT(zl->zl_child == cio); | |
433 | return (zl->zl_parent); | |
434 | } | |
435 | ||
436 | zio_t * | |
437 | zio_walk_children(zio_t *pio) | |
438 | { | |
439 | zio_link_t *zl = pio->io_walk_link; | |
440 | list_t *cl = &pio->io_child_list; | |
441 | ||
442 | zl = (zl == NULL) ? list_head(cl) : list_next(cl, zl); | |
443 | pio->io_walk_link = zl; | |
444 | ||
445 | if (zl == NULL) | |
446 | return (NULL); | |
447 | ||
448 | ASSERT(zl->zl_parent == pio); | |
449 | return (zl->zl_child); | |
450 | } | |
451 | ||
452 | zio_t * | |
453 | zio_unique_parent(zio_t *cio) | |
454 | { | |
455 | zio_t *pio = zio_walk_parents(cio); | |
456 | ||
457 | VERIFY(zio_walk_parents(cio) == NULL); | |
458 | return (pio); | |
459 | } | |
460 | ||
461 | void | |
462 | zio_add_child(zio_t *pio, zio_t *cio) | |
b128c09f | 463 | { |
691f6ac4 | 464 | zio_link_t *zl = kmem_cache_alloc(zio_link_cache, KM_PUSHPAGE); |
d6320ddb | 465 | int w; |
d164b209 BB |
466 | |
467 | /* | |
468 | * Logical I/Os can have logical, gang, or vdev children. | |
469 | * Gang I/Os can have gang or vdev children. | |
470 | * Vdev I/Os can only have vdev children. | |
471 | * The following ASSERT captures all of these constraints. | |
472 | */ | |
473 | ASSERT(cio->io_child_type <= pio->io_child_type); | |
474 | ||
475 | zl->zl_parent = pio; | |
476 | zl->zl_child = cio; | |
477 | ||
478 | mutex_enter(&cio->io_lock); | |
b128c09f | 479 | mutex_enter(&pio->io_lock); |
d164b209 BB |
480 | |
481 | ASSERT(pio->io_state[ZIO_WAIT_DONE] == 0); | |
482 | ||
d6320ddb | 483 | for (w = 0; w < ZIO_WAIT_TYPES; w++) |
d164b209 BB |
484 | pio->io_children[cio->io_child_type][w] += !cio->io_state[w]; |
485 | ||
486 | list_insert_head(&pio->io_child_list, zl); | |
487 | list_insert_head(&cio->io_parent_list, zl); | |
488 | ||
428870ff BB |
489 | pio->io_child_count++; |
490 | cio->io_parent_count++; | |
491 | ||
b128c09f | 492 | mutex_exit(&pio->io_lock); |
d164b209 | 493 | mutex_exit(&cio->io_lock); |
b128c09f BB |
494 | } |
495 | ||
34dc7c2f | 496 | static void |
d164b209 | 497 | zio_remove_child(zio_t *pio, zio_t *cio, zio_link_t *zl) |
b128c09f | 498 | { |
d164b209 BB |
499 | ASSERT(zl->zl_parent == pio); |
500 | ASSERT(zl->zl_child == cio); | |
b128c09f | 501 | |
d164b209 | 502 | mutex_enter(&cio->io_lock); |
b128c09f | 503 | mutex_enter(&pio->io_lock); |
d164b209 BB |
504 | |
505 | list_remove(&pio->io_child_list, zl); | |
506 | list_remove(&cio->io_parent_list, zl); | |
507 | ||
428870ff BB |
508 | pio->io_child_count--; |
509 | cio->io_parent_count--; | |
510 | ||
b128c09f | 511 | mutex_exit(&pio->io_lock); |
d164b209 BB |
512 | mutex_exit(&cio->io_lock); |
513 | ||
514 | kmem_cache_free(zio_link_cache, zl); | |
b128c09f BB |
515 | } |
516 | ||
517 | static boolean_t | |
518 | zio_wait_for_children(zio_t *zio, enum zio_child child, enum zio_wait_type wait) | |
34dc7c2f | 519 | { |
b128c09f BB |
520 | uint64_t *countp = &zio->io_children[child][wait]; |
521 | boolean_t waiting = B_FALSE; | |
522 | ||
523 | mutex_enter(&zio->io_lock); | |
524 | ASSERT(zio->io_stall == NULL); | |
525 | if (*countp != 0) { | |
428870ff | 526 | zio->io_stage >>= 1; |
b128c09f BB |
527 | zio->io_stall = countp; |
528 | waiting = B_TRUE; | |
529 | } | |
530 | mutex_exit(&zio->io_lock); | |
531 | ||
532 | return (waiting); | |
533 | } | |
34dc7c2f | 534 | |
bf701a83 BB |
535 | __attribute__((always_inline)) |
536 | static inline void | |
b128c09f BB |
537 | zio_notify_parent(zio_t *pio, zio_t *zio, enum zio_wait_type wait) |
538 | { | |
539 | uint64_t *countp = &pio->io_children[zio->io_child_type][wait]; | |
540 | int *errorp = &pio->io_child_error[zio->io_child_type]; | |
34dc7c2f | 541 | |
b128c09f BB |
542 | mutex_enter(&pio->io_lock); |
543 | if (zio->io_error && !(zio->io_flags & ZIO_FLAG_DONT_PROPAGATE)) | |
544 | *errorp = zio_worst_error(*errorp, zio->io_error); | |
545 | pio->io_reexecute |= zio->io_reexecute; | |
546 | ASSERT3U(*countp, >, 0); | |
547 | if (--*countp == 0 && pio->io_stall == countp) { | |
548 | pio->io_stall = NULL; | |
549 | mutex_exit(&pio->io_lock); | |
da6b4005 | 550 | __zio_execute(pio); |
b128c09f BB |
551 | } else { |
552 | mutex_exit(&pio->io_lock); | |
34dc7c2f BB |
553 | } |
554 | } | |
555 | ||
b128c09f BB |
556 | static void |
557 | zio_inherit_child_errors(zio_t *zio, enum zio_child c) | |
558 | { | |
559 | if (zio->io_child_error[c] != 0 && zio->io_error == 0) | |
560 | zio->io_error = zio->io_child_error[c]; | |
561 | } | |
562 | ||
34dc7c2f BB |
563 | /* |
564 | * ========================================================================== | |
b128c09f | 565 | * Create the various types of I/O (read, write, free, etc) |
34dc7c2f BB |
566 | * ========================================================================== |
567 | */ | |
568 | static zio_t * | |
428870ff | 569 | zio_create(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, |
34dc7c2f | 570 | void *data, uint64_t size, zio_done_func_t *done, void *private, |
428870ff BB |
571 | zio_type_t type, int priority, enum zio_flag flags, |
572 | vdev_t *vd, uint64_t offset, const zbookmark_t *zb, | |
573 | enum zio_stage stage, enum zio_stage pipeline) | |
34dc7c2f BB |
574 | { |
575 | zio_t *zio; | |
576 | ||
577 | ASSERT3U(size, <=, SPA_MAXBLOCKSIZE); | |
578 | ASSERT(P2PHASE(size, SPA_MINBLOCKSIZE) == 0); | |
b128c09f BB |
579 | ASSERT(P2PHASE(offset, SPA_MINBLOCKSIZE) == 0); |
580 | ||
581 | ASSERT(!vd || spa_config_held(spa, SCL_STATE_ALL, RW_READER)); | |
582 | ASSERT(!bp || !(flags & ZIO_FLAG_CONFIG_WRITER)); | |
583 | ASSERT(vd || stage == ZIO_STAGE_OPEN); | |
34dc7c2f | 584 | |
691f6ac4 | 585 | zio = kmem_cache_alloc(zio_cache, KM_PUSHPAGE); |
d164b209 | 586 | |
b128c09f BB |
587 | if (vd != NULL) |
588 | zio->io_child_type = ZIO_CHILD_VDEV; | |
589 | else if (flags & ZIO_FLAG_GANG_CHILD) | |
590 | zio->io_child_type = ZIO_CHILD_GANG; | |
428870ff BB |
591 | else if (flags & ZIO_FLAG_DDT_CHILD) |
592 | zio->io_child_type = ZIO_CHILD_DDT; | |
b128c09f BB |
593 | else |
594 | zio->io_child_type = ZIO_CHILD_LOGICAL; | |
595 | ||
34dc7c2f | 596 | if (bp != NULL) { |
49be0ccf | 597 | zio->io_logical = NULL; |
428870ff | 598 | zio->io_bp = (blkptr_t *)bp; |
34dc7c2f BB |
599 | zio->io_bp_copy = *bp; |
600 | zio->io_bp_orig = *bp; | |
428870ff BB |
601 | if (type != ZIO_TYPE_WRITE || |
602 | zio->io_child_type == ZIO_CHILD_DDT) | |
b128c09f | 603 | zio->io_bp = &zio->io_bp_copy; /* so caller can free */ |
9babb374 | 604 | if (zio->io_child_type == ZIO_CHILD_LOGICAL) |
b128c09f | 605 | zio->io_logical = zio; |
9babb374 BB |
606 | if (zio->io_child_type > ZIO_CHILD_GANG && BP_IS_GANG(bp)) |
607 | pipeline |= ZIO_GANG_STAGES; | |
49be0ccf BB |
608 | } else { |
609 | zio->io_logical = NULL; | |
610 | zio->io_bp = NULL; | |
611 | bzero(&zio->io_bp_copy, sizeof (blkptr_t)); | |
612 | bzero(&zio->io_bp_orig, sizeof (blkptr_t)); | |
34dc7c2f | 613 | } |
b128c09f BB |
614 | |
615 | zio->io_spa = spa; | |
616 | zio->io_txg = txg; | |
49be0ccf | 617 | zio->io_ready = NULL; |
34dc7c2f BB |
618 | zio->io_done = done; |
619 | zio->io_private = private; | |
49be0ccf | 620 | zio->io_prev_space_delta = 0; |
34dc7c2f BB |
621 | zio->io_type = type; |
622 | zio->io_priority = priority; | |
b128c09f | 623 | zio->io_vd = vd; |
49be0ccf BB |
624 | zio->io_vsd = NULL; |
625 | zio->io_vsd_ops = NULL; | |
b128c09f | 626 | zio->io_offset = offset; |
49be0ccf | 627 | zio->io_deadline = 0; |
cc92e9d0 GW |
628 | zio->io_timestamp = 0; |
629 | zio->io_delta = 0; | |
630 | zio->io_delay = 0; | |
428870ff BB |
631 | zio->io_orig_data = zio->io_data = data; |
632 | zio->io_orig_size = zio->io_size = size; | |
b128c09f BB |
633 | zio->io_orig_flags = zio->io_flags = flags; |
634 | zio->io_orig_stage = zio->io_stage = stage; | |
635 | zio->io_orig_pipeline = zio->io_pipeline = pipeline; | |
49be0ccf BB |
636 | bzero(&zio->io_prop, sizeof (zio_prop_t)); |
637 | zio->io_cmd = 0; | |
638 | zio->io_reexecute = 0; | |
639 | zio->io_bp_override = NULL; | |
640 | zio->io_walk_link = NULL; | |
641 | zio->io_transform_stack = NULL; | |
49be0ccf BB |
642 | zio->io_error = 0; |
643 | zio->io_child_count = 0; | |
644 | zio->io_parent_count = 0; | |
645 | zio->io_stall = NULL; | |
646 | zio->io_gang_leader = NULL; | |
647 | zio->io_gang_tree = NULL; | |
648 | zio->io_executor = NULL; | |
649 | zio->io_waiter = NULL; | |
650 | zio->io_cksum_report = NULL; | |
651 | zio->io_ena = 0; | |
652 | bzero(zio->io_child_error, sizeof (int) * ZIO_CHILD_TYPES); | |
653 | bzero(zio->io_children, | |
654 | sizeof (uint64_t) * ZIO_CHILD_TYPES * ZIO_WAIT_TYPES); | |
655 | bzero(&zio->io_bookmark, sizeof (zbookmark_t)); | |
34dc7c2f | 656 | |
d164b209 BB |
657 | zio->io_state[ZIO_WAIT_READY] = (stage >= ZIO_STAGE_READY); |
658 | zio->io_state[ZIO_WAIT_DONE] = (stage >= ZIO_STAGE_DONE); | |
659 | ||
b128c09f BB |
660 | if (zb != NULL) |
661 | zio->io_bookmark = *zb; | |
662 | ||
663 | if (pio != NULL) { | |
b128c09f | 664 | if (zio->io_logical == NULL) |
34dc7c2f | 665 | zio->io_logical = pio->io_logical; |
9babb374 BB |
666 | if (zio->io_child_type == ZIO_CHILD_GANG) |
667 | zio->io_gang_leader = pio->io_gang_leader; | |
b128c09f | 668 | zio_add_child(pio, zio); |
34dc7c2f BB |
669 | } |
670 | ||
a38718a6 GA |
671 | taskq_init_ent(&zio->io_tqent); |
672 | ||
34dc7c2f BB |
673 | return (zio); |
674 | } | |
675 | ||
676 | static void | |
b128c09f | 677 | zio_destroy(zio_t *zio) |
34dc7c2f | 678 | { |
b128c09f | 679 | kmem_cache_free(zio_cache, zio); |
34dc7c2f BB |
680 | } |
681 | ||
682 | zio_t * | |
d164b209 | 683 | zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, zio_done_func_t *done, |
428870ff | 684 | void *private, enum zio_flag flags) |
34dc7c2f BB |
685 | { |
686 | zio_t *zio; | |
687 | ||
688 | zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, | |
d164b209 | 689 | ZIO_TYPE_NULL, ZIO_PRIORITY_NOW, flags, vd, 0, NULL, |
b128c09f | 690 | ZIO_STAGE_OPEN, ZIO_INTERLOCK_PIPELINE); |
34dc7c2f BB |
691 | |
692 | return (zio); | |
693 | } | |
694 | ||
695 | zio_t * | |
428870ff | 696 | zio_root(spa_t *spa, zio_done_func_t *done, void *private, enum zio_flag flags) |
34dc7c2f | 697 | { |
d164b209 | 698 | return (zio_null(NULL, spa, NULL, done, private, flags)); |
34dc7c2f BB |
699 | } |
700 | ||
701 | zio_t * | |
b128c09f BB |
702 | zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, |
703 | void *data, uint64_t size, zio_done_func_t *done, void *private, | |
428870ff | 704 | int priority, enum zio_flag flags, const zbookmark_t *zb) |
34dc7c2f BB |
705 | { |
706 | zio_t *zio; | |
707 | ||
428870ff | 708 | zio = zio_create(pio, spa, BP_PHYSICAL_BIRTH(bp), bp, |
b128c09f BB |
709 | data, size, done, private, |
710 | ZIO_TYPE_READ, priority, flags, NULL, 0, zb, | |
428870ff BB |
711 | ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? |
712 | ZIO_DDT_CHILD_READ_PIPELINE : ZIO_READ_PIPELINE); | |
34dc7c2f | 713 | |
b128c09f BB |
714 | return (zio); |
715 | } | |
34dc7c2f | 716 | |
34dc7c2f | 717 | zio_t * |
b128c09f | 718 | zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, |
428870ff | 719 | void *data, uint64_t size, const zio_prop_t *zp, |
b128c09f | 720 | zio_done_func_t *ready, zio_done_func_t *done, void *private, |
428870ff | 721 | int priority, enum zio_flag flags, const zbookmark_t *zb) |
34dc7c2f BB |
722 | { |
723 | zio_t *zio; | |
724 | ||
b128c09f BB |
725 | ASSERT(zp->zp_checksum >= ZIO_CHECKSUM_OFF && |
726 | zp->zp_checksum < ZIO_CHECKSUM_FUNCTIONS && | |
727 | zp->zp_compress >= ZIO_COMPRESS_OFF && | |
728 | zp->zp_compress < ZIO_COMPRESS_FUNCTIONS && | |
9ae529ec | 729 | DMU_OT_IS_VALID(zp->zp_type) && |
b128c09f | 730 | zp->zp_level < 32 && |
428870ff BB |
731 | zp->zp_copies > 0 && |
732 | zp->zp_copies <= spa_max_replication(spa) && | |
733 | zp->zp_dedup <= 1 && | |
734 | zp->zp_dedup_verify <= 1); | |
34dc7c2f BB |
735 | |
736 | zio = zio_create(pio, spa, txg, bp, data, size, done, private, | |
b128c09f | 737 | ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, |
428870ff BB |
738 | ZIO_STAGE_OPEN, (flags & ZIO_FLAG_DDT_CHILD) ? |
739 | ZIO_DDT_CHILD_WRITE_PIPELINE : ZIO_WRITE_PIPELINE); | |
34dc7c2f BB |
740 | |
741 | zio->io_ready = ready; | |
b128c09f | 742 | zio->io_prop = *zp; |
34dc7c2f BB |
743 | |
744 | return (zio); | |
745 | } | |
746 | ||
747 | zio_t * | |
b128c09f BB |
748 | zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, void *data, |
749 | uint64_t size, zio_done_func_t *done, void *private, int priority, | |
428870ff | 750 | enum zio_flag flags, zbookmark_t *zb) |
34dc7c2f BB |
751 | { |
752 | zio_t *zio; | |
753 | ||
34dc7c2f | 754 | zio = zio_create(pio, spa, txg, bp, data, size, done, private, |
b128c09f BB |
755 | ZIO_TYPE_WRITE, priority, flags, NULL, 0, zb, |
756 | ZIO_STAGE_OPEN, ZIO_REWRITE_PIPELINE); | |
34dc7c2f BB |
757 | |
758 | return (zio); | |
759 | } | |
760 | ||
428870ff BB |
761 | void |
762 | zio_write_override(zio_t *zio, blkptr_t *bp, int copies) | |
763 | { | |
764 | ASSERT(zio->io_type == ZIO_TYPE_WRITE); | |
765 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
766 | ASSERT(zio->io_stage == ZIO_STAGE_OPEN); | |
767 | ASSERT(zio->io_txg == spa_syncing_txg(zio->io_spa)); | |
768 | ||
769 | zio->io_prop.zp_copies = copies; | |
770 | zio->io_bp_override = bp; | |
771 | } | |
772 | ||
773 | void | |
774 | zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp) | |
775 | { | |
776 | bplist_append(&spa->spa_free_bplist[txg & TXG_MASK], bp); | |
777 | } | |
778 | ||
34dc7c2f | 779 | zio_t * |
428870ff BB |
780 | zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, |
781 | enum zio_flag flags) | |
34dc7c2f BB |
782 | { |
783 | zio_t *zio; | |
784 | ||
428870ff BB |
785 | dprintf_bp(bp, "freeing in txg %llu, pass %u", |
786 | (longlong_t)txg, spa->spa_sync_pass); | |
34dc7c2f | 787 | |
428870ff BB |
788 | ASSERT(!BP_IS_HOLE(bp)); |
789 | ASSERT(spa_syncing_txg(spa) == txg); | |
55d85d5a | 790 | ASSERT(spa_sync_pass(spa) < zfs_sync_pass_deferred_free); |
34dc7c2f | 791 | |
df4474f9 MA |
792 | arc_freed(spa, bp); |
793 | ||
b128c09f | 794 | zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), |
428870ff | 795 | NULL, NULL, ZIO_TYPE_FREE, ZIO_PRIORITY_FREE, flags, |
b128c09f | 796 | NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_FREE_PIPELINE); |
34dc7c2f BB |
797 | |
798 | return (zio); | |
799 | } | |
800 | ||
801 | zio_t * | |
428870ff BB |
802 | zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, const blkptr_t *bp, |
803 | zio_done_func_t *done, void *private, enum zio_flag flags) | |
34dc7c2f BB |
804 | { |
805 | zio_t *zio; | |
806 | ||
807 | /* | |
808 | * A claim is an allocation of a specific block. Claims are needed | |
809 | * to support immediate writes in the intent log. The issue is that | |
810 | * immediate writes contain committed data, but in a txg that was | |
811 | * *not* committed. Upon opening the pool after an unclean shutdown, | |
812 | * the intent log claims all blocks that contain immediate write data | |
813 | * so that the SPA knows they're in use. | |
814 | * | |
815 | * All claims *must* be resolved in the first txg -- before the SPA | |
816 | * starts allocating blocks -- so that nothing is allocated twice. | |
428870ff | 817 | * If txg == 0 we just verify that the block is claimable. |
34dc7c2f BB |
818 | */ |
819 | ASSERT3U(spa->spa_uberblock.ub_rootbp.blk_birth, <, spa_first_txg(spa)); | |
428870ff BB |
820 | ASSERT(txg == spa_first_txg(spa) || txg == 0); |
821 | ASSERT(!BP_GET_DEDUP(bp) || !spa_writeable(spa)); /* zdb(1M) */ | |
34dc7c2f | 822 | |
b128c09f BB |
823 | zio = zio_create(pio, spa, txg, bp, NULL, BP_GET_PSIZE(bp), |
824 | done, private, ZIO_TYPE_CLAIM, ZIO_PRIORITY_NOW, flags, | |
825 | NULL, 0, NULL, ZIO_STAGE_OPEN, ZIO_CLAIM_PIPELINE); | |
34dc7c2f BB |
826 | |
827 | return (zio); | |
828 | } | |
829 | ||
830 | zio_t * | |
831 | zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, | |
428870ff | 832 | zio_done_func_t *done, void *private, int priority, enum zio_flag flags) |
34dc7c2f BB |
833 | { |
834 | zio_t *zio; | |
835 | int c; | |
836 | ||
837 | if (vd->vdev_children == 0) { | |
838 | zio = zio_create(pio, spa, 0, NULL, NULL, 0, done, private, | |
b128c09f | 839 | ZIO_TYPE_IOCTL, priority, flags, vd, 0, NULL, |
34dc7c2f BB |
840 | ZIO_STAGE_OPEN, ZIO_IOCTL_PIPELINE); |
841 | ||
34dc7c2f BB |
842 | zio->io_cmd = cmd; |
843 | } else { | |
d164b209 | 844 | zio = zio_null(pio, spa, NULL, NULL, NULL, flags); |
34dc7c2f BB |
845 | |
846 | for (c = 0; c < vd->vdev_children; c++) | |
847 | zio_nowait(zio_ioctl(zio, spa, vd->vdev_child[c], cmd, | |
848 | done, private, priority, flags)); | |
849 | } | |
850 | ||
851 | return (zio); | |
852 | } | |
853 | ||
34dc7c2f BB |
854 | zio_t * |
855 | zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, | |
856 | void *data, int checksum, zio_done_func_t *done, void *private, | |
428870ff | 857 | int priority, enum zio_flag flags, boolean_t labels) |
34dc7c2f BB |
858 | { |
859 | zio_t *zio; | |
34dc7c2f | 860 | |
b128c09f BB |
861 | ASSERT(vd->vdev_children == 0); |
862 | ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || | |
863 | offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); | |
864 | ASSERT3U(offset + size, <=, vd->vdev_psize); | |
34dc7c2f | 865 | |
b128c09f BB |
866 | zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, |
867 | ZIO_TYPE_READ, priority, flags, vd, offset, NULL, | |
34dc7c2f BB |
868 | ZIO_STAGE_OPEN, ZIO_READ_PHYS_PIPELINE); |
869 | ||
b128c09f | 870 | zio->io_prop.zp_checksum = checksum; |
34dc7c2f BB |
871 | |
872 | return (zio); | |
873 | } | |
874 | ||
875 | zio_t * | |
876 | zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size, | |
877 | void *data, int checksum, zio_done_func_t *done, void *private, | |
428870ff | 878 | int priority, enum zio_flag flags, boolean_t labels) |
34dc7c2f | 879 | { |
34dc7c2f | 880 | zio_t *zio; |
34dc7c2f | 881 | |
b128c09f BB |
882 | ASSERT(vd->vdev_children == 0); |
883 | ASSERT(!labels || offset + size <= VDEV_LABEL_START_SIZE || | |
884 | offset >= vd->vdev_psize - VDEV_LABEL_END_SIZE); | |
885 | ASSERT3U(offset + size, <=, vd->vdev_psize); | |
34dc7c2f | 886 | |
b128c09f BB |
887 | zio = zio_create(pio, vd->vdev_spa, 0, NULL, data, size, done, private, |
888 | ZIO_TYPE_WRITE, priority, flags, vd, offset, NULL, | |
34dc7c2f BB |
889 | ZIO_STAGE_OPEN, ZIO_WRITE_PHYS_PIPELINE); |
890 | ||
b128c09f | 891 | zio->io_prop.zp_checksum = checksum; |
34dc7c2f | 892 | |
428870ff | 893 | if (zio_checksum_table[checksum].ci_eck) { |
34dc7c2f | 894 | /* |
428870ff | 895 | * zec checksums are necessarily destructive -- they modify |
b128c09f | 896 | * the end of the write buffer to hold the verifier/checksum. |
34dc7c2f | 897 | * Therefore, we must make a local copy in case the data is |
b128c09f | 898 | * being written to multiple places in parallel. |
34dc7c2f | 899 | */ |
b128c09f | 900 | void *wbuf = zio_buf_alloc(size); |
34dc7c2f | 901 | bcopy(data, wbuf, size); |
b128c09f | 902 | zio_push_transform(zio, wbuf, size, size, NULL); |
34dc7c2f BB |
903 | } |
904 | ||
905 | return (zio); | |
906 | } | |
907 | ||
908 | /* | |
b128c09f | 909 | * Create a child I/O to do some work for us. |
34dc7c2f BB |
910 | */ |
911 | zio_t * | |
b128c09f | 912 | zio_vdev_child_io(zio_t *pio, blkptr_t *bp, vdev_t *vd, uint64_t offset, |
428870ff | 913 | void *data, uint64_t size, int type, int priority, enum zio_flag flags, |
34dc7c2f BB |
914 | zio_done_func_t *done, void *private) |
915 | { | |
428870ff | 916 | enum zio_stage pipeline = ZIO_VDEV_CHILD_PIPELINE; |
b128c09f BB |
917 | zio_t *zio; |
918 | ||
919 | ASSERT(vd->vdev_parent == | |
920 | (pio->io_vd ? pio->io_vd : pio->io_spa->spa_root_vdev)); | |
34dc7c2f BB |
921 | |
922 | if (type == ZIO_TYPE_READ && bp != NULL) { | |
923 | /* | |
924 | * If we have the bp, then the child should perform the | |
925 | * checksum and the parent need not. This pushes error | |
926 | * detection as close to the leaves as possible and | |
927 | * eliminates redundant checksums in the interior nodes. | |
928 | */ | |
428870ff BB |
929 | pipeline |= ZIO_STAGE_CHECKSUM_VERIFY; |
930 | pio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; | |
34dc7c2f BB |
931 | } |
932 | ||
b128c09f BB |
933 | if (vd->vdev_children == 0) |
934 | offset += VDEV_LABEL_START_SIZE; | |
935 | ||
428870ff BB |
936 | flags |= ZIO_VDEV_CHILD_FLAGS(pio) | ZIO_FLAG_DONT_PROPAGATE; |
937 | ||
938 | /* | |
939 | * If we've decided to do a repair, the write is not speculative -- | |
940 | * even if the original read was. | |
941 | */ | |
942 | if (flags & ZIO_FLAG_IO_REPAIR) | |
943 | flags &= ~ZIO_FLAG_SPECULATIVE; | |
944 | ||
b128c09f | 945 | zio = zio_create(pio, pio->io_spa, pio->io_txg, bp, data, size, |
428870ff BB |
946 | done, private, type, priority, flags, vd, offset, &pio->io_bookmark, |
947 | ZIO_STAGE_VDEV_IO_START >> 1, pipeline); | |
34dc7c2f | 948 | |
b128c09f | 949 | return (zio); |
34dc7c2f BB |
950 | } |
951 | ||
b128c09f BB |
952 | zio_t * |
953 | zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, void *data, uint64_t size, | |
428870ff BB |
954 | int type, int priority, enum zio_flag flags, |
955 | zio_done_func_t *done, void *private) | |
34dc7c2f | 956 | { |
b128c09f | 957 | zio_t *zio; |
34dc7c2f | 958 | |
b128c09f | 959 | ASSERT(vd->vdev_ops->vdev_op_leaf); |
34dc7c2f | 960 | |
b128c09f BB |
961 | zio = zio_create(NULL, vd->vdev_spa, 0, NULL, |
962 | data, size, done, private, type, priority, | |
963 | flags | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_RETRY, | |
964 | vd, offset, NULL, | |
428870ff | 965 | ZIO_STAGE_VDEV_IO_START >> 1, ZIO_VDEV_CHILD_PIPELINE); |
34dc7c2f | 966 | |
b128c09f | 967 | return (zio); |
34dc7c2f BB |
968 | } |
969 | ||
970 | void | |
b128c09f | 971 | zio_flush(zio_t *zio, vdev_t *vd) |
34dc7c2f | 972 | { |
b128c09f BB |
973 | zio_nowait(zio_ioctl(zio, zio->io_spa, vd, DKIOCFLUSHWRITECACHE, |
974 | NULL, NULL, ZIO_PRIORITY_NOW, | |
975 | ZIO_FLAG_CANFAIL | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY)); | |
34dc7c2f BB |
976 | } |
977 | ||
428870ff BB |
978 | void |
979 | zio_shrink(zio_t *zio, uint64_t size) | |
980 | { | |
981 | ASSERT(zio->io_executor == NULL); | |
982 | ASSERT(zio->io_orig_size == zio->io_size); | |
983 | ASSERT(size <= zio->io_size); | |
984 | ||
985 | /* | |
986 | * We don't shrink for raidz because of problems with the | |
987 | * reconstruction when reading back less than the block size. | |
988 | * Note, BP_IS_RAIDZ() assumes no compression. | |
989 | */ | |
990 | ASSERT(BP_GET_COMPRESS(zio->io_bp) == ZIO_COMPRESS_OFF); | |
991 | if (!BP_IS_RAIDZ(zio->io_bp)) | |
992 | zio->io_orig_size = zio->io_size = size; | |
993 | } | |
994 | ||
34dc7c2f BB |
995 | /* |
996 | * ========================================================================== | |
b128c09f | 997 | * Prepare to read and write logical blocks |
34dc7c2f BB |
998 | * ========================================================================== |
999 | */ | |
b128c09f | 1000 | |
34dc7c2f | 1001 | static int |
b128c09f | 1002 | zio_read_bp_init(zio_t *zio) |
34dc7c2f | 1003 | { |
b128c09f | 1004 | blkptr_t *bp = zio->io_bp; |
34dc7c2f | 1005 | |
fb5f0bc8 | 1006 | if (BP_GET_COMPRESS(bp) != ZIO_COMPRESS_OFF && |
9babb374 BB |
1007 | zio->io_child_type == ZIO_CHILD_LOGICAL && |
1008 | !(zio->io_flags & ZIO_FLAG_RAW)) { | |
428870ff BB |
1009 | uint64_t psize = BP_GET_PSIZE(bp); |
1010 | void *cbuf = zio_buf_alloc(psize); | |
b128c09f | 1011 | |
428870ff | 1012 | zio_push_transform(zio, cbuf, psize, psize, zio_decompress); |
34dc7c2f | 1013 | } |
34dc7c2f | 1014 | |
9ae529ec | 1015 | if (!DMU_OT_IS_METADATA(BP_GET_TYPE(bp)) && BP_GET_LEVEL(bp) == 0) |
b128c09f BB |
1016 | zio->io_flags |= ZIO_FLAG_DONT_CACHE; |
1017 | ||
428870ff BB |
1018 | if (BP_GET_TYPE(bp) == DMU_OT_DDT_ZAP) |
1019 | zio->io_flags |= ZIO_FLAG_DONT_CACHE; | |
1020 | ||
1021 | if (BP_GET_DEDUP(bp) && zio->io_child_type == ZIO_CHILD_LOGICAL) | |
1022 | zio->io_pipeline = ZIO_DDT_READ_PIPELINE; | |
1023 | ||
b128c09f | 1024 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
1025 | } |
1026 | ||
b128c09f BB |
1027 | static int |
1028 | zio_write_bp_init(zio_t *zio) | |
34dc7c2f | 1029 | { |
428870ff | 1030 | spa_t *spa = zio->io_spa; |
b128c09f | 1031 | zio_prop_t *zp = &zio->io_prop; |
428870ff | 1032 | enum zio_compress compress = zp->zp_compress; |
34dc7c2f | 1033 | blkptr_t *bp = zio->io_bp; |
b128c09f | 1034 | uint64_t lsize = zio->io_size; |
428870ff | 1035 | uint64_t psize = lsize; |
b128c09f | 1036 | int pass = 1; |
34dc7c2f | 1037 | |
b128c09f BB |
1038 | /* |
1039 | * If our children haven't all reached the ready stage, | |
1040 | * wait for them and then repeat this pipeline stage. | |
1041 | */ | |
1042 | if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || | |
1043 | zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_READY)) | |
1044 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 1045 | |
b128c09f BB |
1046 | if (!IO_IS_ALLOCATING(zio)) |
1047 | return (ZIO_PIPELINE_CONTINUE); | |
34dc7c2f | 1048 | |
428870ff BB |
1049 | ASSERT(zio->io_child_type != ZIO_CHILD_DDT); |
1050 | ||
1051 | if (zio->io_bp_override) { | |
1052 | ASSERT(bp->blk_birth != zio->io_txg); | |
1053 | ASSERT(BP_GET_DEDUP(zio->io_bp_override) == 0); | |
1054 | ||
1055 | *bp = *zio->io_bp_override; | |
1056 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
1057 | ||
1058 | if (BP_IS_HOLE(bp) || !zp->zp_dedup) | |
1059 | return (ZIO_PIPELINE_CONTINUE); | |
1060 | ||
1061 | ASSERT(zio_checksum_table[zp->zp_checksum].ci_dedup || | |
1062 | zp->zp_dedup_verify); | |
1063 | ||
1064 | if (BP_GET_CHECKSUM(bp) == zp->zp_checksum) { | |
1065 | BP_SET_DEDUP(bp, 1); | |
1066 | zio->io_pipeline |= ZIO_STAGE_DDT_WRITE; | |
1067 | return (ZIO_PIPELINE_CONTINUE); | |
1068 | } | |
1069 | zio->io_bp_override = NULL; | |
1070 | BP_ZERO(bp); | |
1071 | } | |
34dc7c2f | 1072 | |
b128c09f BB |
1073 | if (bp->blk_birth == zio->io_txg) { |
1074 | /* | |
1075 | * We're rewriting an existing block, which means we're | |
1076 | * working on behalf of spa_sync(). For spa_sync() to | |
1077 | * converge, it must eventually be the case that we don't | |
1078 | * have to allocate new blocks. But compression changes | |
1079 | * the blocksize, which forces a reallocate, and makes | |
1080 | * convergence take longer. Therefore, after the first | |
1081 | * few passes, stop compressing to ensure convergence. | |
1082 | */ | |
428870ff BB |
1083 | pass = spa_sync_pass(spa); |
1084 | ||
1085 | ASSERT(zio->io_txg == spa_syncing_txg(spa)); | |
1086 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1087 | ASSERT(!BP_GET_DEDUP(bp)); | |
34dc7c2f | 1088 | |
55d85d5a | 1089 | if (pass >= zfs_sync_pass_dont_compress) |
b128c09f | 1090 | compress = ZIO_COMPRESS_OFF; |
34dc7c2f | 1091 | |
b128c09f | 1092 | /* Make sure someone doesn't change their mind on overwrites */ |
428870ff BB |
1093 | ASSERT(MIN(zp->zp_copies + BP_IS_GANG(bp), |
1094 | spa_max_replication(spa)) == BP_GET_NDVAS(bp)); | |
b128c09f | 1095 | } |
34dc7c2f | 1096 | |
b128c09f | 1097 | if (compress != ZIO_COMPRESS_OFF) { |
428870ff BB |
1098 | void *cbuf = zio_buf_alloc(lsize); |
1099 | psize = zio_compress_data(compress, zio->io_data, cbuf, lsize); | |
1100 | if (psize == 0 || psize == lsize) { | |
b128c09f | 1101 | compress = ZIO_COMPRESS_OFF; |
428870ff BB |
1102 | zio_buf_free(cbuf, lsize); |
1103 | } else { | |
1104 | ASSERT(psize < lsize); | |
1105 | zio_push_transform(zio, cbuf, psize, lsize, NULL); | |
b128c09f BB |
1106 | } |
1107 | } | |
34dc7c2f | 1108 | |
b128c09f BB |
1109 | /* |
1110 | * The final pass of spa_sync() must be all rewrites, but the first | |
1111 | * few passes offer a trade-off: allocating blocks defers convergence, | |
1112 | * but newly allocated blocks are sequential, so they can be written | |
1113 | * to disk faster. Therefore, we allow the first few passes of | |
1114 | * spa_sync() to allocate new blocks, but force rewrites after that. | |
1115 | * There should only be a handful of blocks after pass 1 in any case. | |
1116 | */ | |
428870ff | 1117 | if (bp->blk_birth == zio->io_txg && BP_GET_PSIZE(bp) == psize && |
55d85d5a | 1118 | pass >= zfs_sync_pass_rewrite) { |
428870ff | 1119 | enum zio_stage gang_stages = zio->io_pipeline & ZIO_GANG_STAGES; |
d6320ddb | 1120 | ASSERT(psize != 0); |
b128c09f BB |
1121 | zio->io_pipeline = ZIO_REWRITE_PIPELINE | gang_stages; |
1122 | zio->io_flags |= ZIO_FLAG_IO_REWRITE; | |
1123 | } else { | |
1124 | BP_ZERO(bp); | |
1125 | zio->io_pipeline = ZIO_WRITE_PIPELINE; | |
1126 | } | |
34dc7c2f | 1127 | |
428870ff | 1128 | if (psize == 0) { |
b128c09f BB |
1129 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; |
1130 | } else { | |
1131 | ASSERT(zp->zp_checksum != ZIO_CHECKSUM_GANG_HEADER); | |
1132 | BP_SET_LSIZE(bp, lsize); | |
428870ff | 1133 | BP_SET_PSIZE(bp, psize); |
b128c09f BB |
1134 | BP_SET_COMPRESS(bp, compress); |
1135 | BP_SET_CHECKSUM(bp, zp->zp_checksum); | |
1136 | BP_SET_TYPE(bp, zp->zp_type); | |
1137 | BP_SET_LEVEL(bp, zp->zp_level); | |
428870ff | 1138 | BP_SET_DEDUP(bp, zp->zp_dedup); |
b128c09f | 1139 | BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER); |
428870ff BB |
1140 | if (zp->zp_dedup) { |
1141 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1142 | ASSERT(!(zio->io_flags & ZIO_FLAG_IO_REWRITE)); | |
1143 | zio->io_pipeline = ZIO_DDT_WRITE_PIPELINE; | |
1144 | } | |
1145 | } | |
1146 | ||
1147 | return (ZIO_PIPELINE_CONTINUE); | |
1148 | } | |
1149 | ||
1150 | static int | |
1151 | zio_free_bp_init(zio_t *zio) | |
1152 | { | |
1153 | blkptr_t *bp = zio->io_bp; | |
1154 | ||
1155 | if (zio->io_child_type == ZIO_CHILD_LOGICAL) { | |
1156 | if (BP_GET_DEDUP(bp)) | |
1157 | zio->io_pipeline = ZIO_DDT_FREE_PIPELINE; | |
b128c09f | 1158 | } |
34dc7c2f BB |
1159 | |
1160 | return (ZIO_PIPELINE_CONTINUE); | |
1161 | } | |
1162 | ||
b128c09f BB |
1163 | /* |
1164 | * ========================================================================== | |
1165 | * Execute the I/O pipeline | |
1166 | * ========================================================================== | |
1167 | */ | |
1168 | ||
1169 | static void | |
7ef5e54e | 1170 | zio_taskq_dispatch(zio_t *zio, zio_taskq_type_t q, boolean_t cutinline) |
34dc7c2f | 1171 | { |
428870ff | 1172 | spa_t *spa = zio->io_spa; |
b128c09f | 1173 | zio_type_t t = zio->io_type; |
a38718a6 | 1174 | int flags = (cutinline ? TQ_FRONT : 0); |
34dc7c2f BB |
1175 | |
1176 | /* | |
9babb374 BB |
1177 | * If we're a config writer or a probe, the normal issue and |
1178 | * interrupt threads may all be blocked waiting for the config lock. | |
1179 | * In this case, select the otherwise-unused taskq for ZIO_TYPE_NULL. | |
34dc7c2f | 1180 | */ |
9babb374 | 1181 | if (zio->io_flags & (ZIO_FLAG_CONFIG_WRITER | ZIO_FLAG_PROBE)) |
b128c09f | 1182 | t = ZIO_TYPE_NULL; |
34dc7c2f BB |
1183 | |
1184 | /* | |
b128c09f | 1185 | * A similar issue exists for the L2ARC write thread until L2ARC 2.0. |
34dc7c2f | 1186 | */ |
b128c09f BB |
1187 | if (t == ZIO_TYPE_WRITE && zio->io_vd && zio->io_vd->vdev_aux) |
1188 | t = ZIO_TYPE_NULL; | |
34dc7c2f | 1189 | |
428870ff | 1190 | /* |
7ef5e54e AL |
1191 | * If this is a high priority I/O, then use the high priority taskq if |
1192 | * available. | |
428870ff BB |
1193 | */ |
1194 | if (zio->io_priority == ZIO_PRIORITY_NOW && | |
7ef5e54e | 1195 | spa->spa_zio_taskq[t][q + 1].stqs_count != 0) |
428870ff BB |
1196 | q++; |
1197 | ||
1198 | ASSERT3U(q, <, ZIO_TASKQ_TYPES); | |
5cc556b4 | 1199 | |
a38718a6 GA |
1200 | /* |
1201 | * NB: We are assuming that the zio can only be dispatched | |
1202 | * to a single taskq at a time. It would be a grievous error | |
1203 | * to dispatch the zio to another taskq at the same time. | |
1204 | */ | |
1205 | ASSERT(taskq_empty_ent(&zio->io_tqent)); | |
7ef5e54e AL |
1206 | spa_taskq_dispatch_ent(spa, t, q, (task_func_t *)zio_execute, zio, |
1207 | flags, &zio->io_tqent); | |
b128c09f | 1208 | } |
34dc7c2f | 1209 | |
b128c09f | 1210 | static boolean_t |
7ef5e54e | 1211 | zio_taskq_member(zio_t *zio, zio_taskq_type_t q) |
b128c09f BB |
1212 | { |
1213 | kthread_t *executor = zio->io_executor; | |
1214 | spa_t *spa = zio->io_spa; | |
d6320ddb | 1215 | zio_type_t t; |
34dc7c2f | 1216 | |
7ef5e54e AL |
1217 | for (t = 0; t < ZIO_TYPES; t++) { |
1218 | spa_taskqs_t *tqs = &spa->spa_zio_taskq[t][q]; | |
1219 | uint_t i; | |
1220 | for (i = 0; i < tqs->stqs_count; i++) { | |
1221 | if (taskq_member(tqs->stqs_taskq[i], executor)) | |
1222 | return (B_TRUE); | |
1223 | } | |
1224 | } | |
34dc7c2f | 1225 | |
b128c09f BB |
1226 | return (B_FALSE); |
1227 | } | |
34dc7c2f | 1228 | |
b128c09f BB |
1229 | static int |
1230 | zio_issue_async(zio_t *zio) | |
1231 | { | |
428870ff | 1232 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); |
b128c09f BB |
1233 | |
1234 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f BB |
1235 | } |
1236 | ||
b128c09f BB |
1237 | void |
1238 | zio_interrupt(zio_t *zio) | |
34dc7c2f | 1239 | { |
428870ff | 1240 | zio_taskq_dispatch(zio, ZIO_TASKQ_INTERRUPT, B_FALSE); |
b128c09f | 1241 | } |
34dc7c2f | 1242 | |
b128c09f BB |
1243 | /* |
1244 | * Execute the I/O pipeline until one of the following occurs: | |
1245 | * (1) the I/O completes; (2) the pipeline stalls waiting for | |
1246 | * dependent child I/Os; (3) the I/O issues, so we're waiting | |
1247 | * for an I/O completion interrupt; (4) the I/O is delegated by | |
1248 | * vdev-level caching or aggregation; (5) the I/O is deferred | |
1249 | * due to vdev-level queueing; (6) the I/O is handed off to | |
1250 | * another thread. In all cases, the pipeline stops whenever | |
8e07b99b | 1251 | * there's no CPU work; it never burns a thread in cv_wait_io(). |
b128c09f BB |
1252 | * |
1253 | * There's no locking on io_stage because there's no legitimate way | |
1254 | * for multiple threads to be attempting to process the same I/O. | |
1255 | */ | |
428870ff | 1256 | static zio_pipe_stage_t *zio_pipeline[]; |
34dc7c2f | 1257 | |
da6b4005 NB |
1258 | /* |
1259 | * zio_execute() is a wrapper around the static function | |
1260 | * __zio_execute() so that we can force __zio_execute() to be | |
1261 | * inlined. This reduces stack overhead which is important | |
1262 | * because __zio_execute() is called recursively in several zio | |
1263 | * code paths. zio_execute() itself cannot be inlined because | |
1264 | * it is externally visible. | |
1265 | */ | |
b128c09f BB |
1266 | void |
1267 | zio_execute(zio_t *zio) | |
da6b4005 NB |
1268 | { |
1269 | __zio_execute(zio); | |
1270 | } | |
1271 | ||
1272 | __attribute__((always_inline)) | |
1273 | static inline void | |
1274 | __zio_execute(zio_t *zio) | |
b128c09f BB |
1275 | { |
1276 | zio->io_executor = curthread; | |
34dc7c2f | 1277 | |
b128c09f | 1278 | while (zio->io_stage < ZIO_STAGE_DONE) { |
428870ff BB |
1279 | enum zio_stage pipeline = zio->io_pipeline; |
1280 | enum zio_stage stage = zio->io_stage; | |
91579709 | 1281 | dsl_pool_t *dp; |
2fac4c2a | 1282 | boolean_t cut; |
b128c09f | 1283 | int rv; |
34dc7c2f | 1284 | |
b128c09f | 1285 | ASSERT(!MUTEX_HELD(&zio->io_lock)); |
428870ff BB |
1286 | ASSERT(ISP2(stage)); |
1287 | ASSERT(zio->io_stall == NULL); | |
34dc7c2f | 1288 | |
428870ff BB |
1289 | do { |
1290 | stage <<= 1; | |
1291 | } while ((stage & pipeline) == 0); | |
b128c09f BB |
1292 | |
1293 | ASSERT(stage <= ZIO_STAGE_DONE); | |
34dc7c2f | 1294 | |
91579709 | 1295 | dp = spa_get_dsl(zio->io_spa); |
2fac4c2a BB |
1296 | cut = (stage == ZIO_STAGE_VDEV_IO_START) ? |
1297 | zio_requeue_io_start_cut_in_line : B_FALSE; | |
1298 | ||
34dc7c2f | 1299 | /* |
b128c09f BB |
1300 | * If we are in interrupt context and this pipeline stage |
1301 | * will grab a config lock that is held across I/O, | |
428870ff BB |
1302 | * or may wait for an I/O that needs an interrupt thread |
1303 | * to complete, issue async to avoid deadlock. | |
1304 | * | |
1305 | * For VDEV_IO_START, we cut in line so that the io will | |
1306 | * be sent to disk promptly. | |
34dc7c2f | 1307 | */ |
91579709 BB |
1308 | if ((stage & ZIO_BLOCKING_STAGES) && zio->io_vd == NULL && |
1309 | zio_taskq_member(zio, ZIO_TASKQ_INTERRUPT)) { | |
1310 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); | |
1311 | return; | |
1312 | } | |
1313 | ||
4cec9b2d | 1314 | #ifdef _KERNEL |
91579709 BB |
1315 | /* |
1316 | * If we executing in the context of the tx_sync_thread, | |
1317 | * or we are performing pool initialization outside of a | |
1318 | * zio_taskq[ZIO_TASKQ_ISSUE] context. Then issue the zio | |
1319 | * async to minimize stack usage for these deep call paths. | |
1320 | */ | |
1321 | if ((dp && curthread == dp->dp_tx.tx_sync_thread) || | |
1322 | (dp && spa_is_initializing(dp->dp_spa) && | |
1323 | !zio_taskq_member(zio, ZIO_TASKQ_ISSUE))) { | |
428870ff | 1324 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, cut); |
b128c09f | 1325 | return; |
34dc7c2f | 1326 | } |
4cec9b2d | 1327 | #endif |
34dc7c2f | 1328 | |
b128c09f | 1329 | zio->io_stage = stage; |
428870ff | 1330 | rv = zio_pipeline[highbit(stage) - 1](zio); |
34dc7c2f | 1331 | |
b128c09f BB |
1332 | if (rv == ZIO_PIPELINE_STOP) |
1333 | return; | |
34dc7c2f | 1334 | |
b128c09f BB |
1335 | ASSERT(rv == ZIO_PIPELINE_CONTINUE); |
1336 | } | |
34dc7c2f BB |
1337 | } |
1338 | ||
da6b4005 | 1339 | |
b128c09f BB |
1340 | /* |
1341 | * ========================================================================== | |
1342 | * Initiate I/O, either sync or async | |
1343 | * ========================================================================== | |
1344 | */ | |
1345 | int | |
1346 | zio_wait(zio_t *zio) | |
34dc7c2f | 1347 | { |
b128c09f | 1348 | int error; |
34dc7c2f | 1349 | |
b128c09f BB |
1350 | ASSERT(zio->io_stage == ZIO_STAGE_OPEN); |
1351 | ASSERT(zio->io_executor == NULL); | |
34dc7c2f | 1352 | |
b128c09f | 1353 | zio->io_waiter = curthread; |
34dc7c2f | 1354 | |
da6b4005 | 1355 | __zio_execute(zio); |
34dc7c2f | 1356 | |
b128c09f | 1357 | mutex_enter(&zio->io_lock); |
72f53c56 | 1358 | while (zio->io_executor != NULL) |
72938d69 | 1359 | cv_wait_io(&zio->io_cv, &zio->io_lock); |
b128c09f | 1360 | mutex_exit(&zio->io_lock); |
34dc7c2f | 1361 | |
b128c09f BB |
1362 | error = zio->io_error; |
1363 | zio_destroy(zio); | |
34dc7c2f | 1364 | |
b128c09f BB |
1365 | return (error); |
1366 | } | |
34dc7c2f | 1367 | |
b128c09f BB |
1368 | void |
1369 | zio_nowait(zio_t *zio) | |
1370 | { | |
1371 | ASSERT(zio->io_executor == NULL); | |
34dc7c2f | 1372 | |
d164b209 BB |
1373 | if (zio->io_child_type == ZIO_CHILD_LOGICAL && |
1374 | zio_unique_parent(zio) == NULL) { | |
34dc7c2f | 1375 | /* |
b128c09f | 1376 | * This is a logical async I/O with no parent to wait for it. |
9babb374 BB |
1377 | * We add it to the spa_async_root_zio "Godfather" I/O which |
1378 | * will ensure they complete prior to unloading the pool. | |
34dc7c2f | 1379 | */ |
b128c09f | 1380 | spa_t *spa = zio->io_spa; |
9babb374 BB |
1381 | |
1382 | zio_add_child(spa->spa_async_zio_root, zio); | |
b128c09f | 1383 | } |
34dc7c2f | 1384 | |
da6b4005 | 1385 | __zio_execute(zio); |
b128c09f | 1386 | } |
34dc7c2f | 1387 | |
b128c09f BB |
1388 | /* |
1389 | * ========================================================================== | |
1390 | * Reexecute or suspend/resume failed I/O | |
1391 | * ========================================================================== | |
1392 | */ | |
34dc7c2f | 1393 | |
b128c09f BB |
1394 | static void |
1395 | zio_reexecute(zio_t *pio) | |
1396 | { | |
d164b209 | 1397 | zio_t *cio, *cio_next; |
d6320ddb | 1398 | int c, w; |
d164b209 BB |
1399 | |
1400 | ASSERT(pio->io_child_type == ZIO_CHILD_LOGICAL); | |
1401 | ASSERT(pio->io_orig_stage == ZIO_STAGE_OPEN); | |
9babb374 BB |
1402 | ASSERT(pio->io_gang_leader == NULL); |
1403 | ASSERT(pio->io_gang_tree == NULL); | |
34dc7c2f | 1404 | |
b128c09f BB |
1405 | pio->io_flags = pio->io_orig_flags; |
1406 | pio->io_stage = pio->io_orig_stage; | |
1407 | pio->io_pipeline = pio->io_orig_pipeline; | |
1408 | pio->io_reexecute = 0; | |
1409 | pio->io_error = 0; | |
d6320ddb | 1410 | for (w = 0; w < ZIO_WAIT_TYPES; w++) |
d164b209 | 1411 | pio->io_state[w] = 0; |
d6320ddb | 1412 | for (c = 0; c < ZIO_CHILD_TYPES; c++) |
b128c09f | 1413 | pio->io_child_error[c] = 0; |
34dc7c2f | 1414 | |
428870ff BB |
1415 | if (IO_IS_ALLOCATING(pio)) |
1416 | BP_ZERO(pio->io_bp); | |
34dc7c2f | 1417 | |
b128c09f BB |
1418 | /* |
1419 | * As we reexecute pio's children, new children could be created. | |
d164b209 | 1420 | * New children go to the head of pio's io_child_list, however, |
b128c09f | 1421 | * so we will (correctly) not reexecute them. The key is that |
d164b209 BB |
1422 | * the remainder of pio's io_child_list, from 'cio_next' onward, |
1423 | * cannot be affected by any side effects of reexecuting 'cio'. | |
b128c09f | 1424 | */ |
d164b209 BB |
1425 | for (cio = zio_walk_children(pio); cio != NULL; cio = cio_next) { |
1426 | cio_next = zio_walk_children(pio); | |
b128c09f | 1427 | mutex_enter(&pio->io_lock); |
d6320ddb | 1428 | for (w = 0; w < ZIO_WAIT_TYPES; w++) |
d164b209 | 1429 | pio->io_children[cio->io_child_type][w]++; |
b128c09f | 1430 | mutex_exit(&pio->io_lock); |
d164b209 | 1431 | zio_reexecute(cio); |
34dc7c2f | 1432 | } |
34dc7c2f | 1433 | |
b128c09f BB |
1434 | /* |
1435 | * Now that all children have been reexecuted, execute the parent. | |
9babb374 BB |
1436 | * We don't reexecute "The Godfather" I/O here as it's the |
1437 | * responsibility of the caller to wait on him. | |
b128c09f | 1438 | */ |
9babb374 | 1439 | if (!(pio->io_flags & ZIO_FLAG_GODFATHER)) |
da6b4005 | 1440 | __zio_execute(pio); |
34dc7c2f BB |
1441 | } |
1442 | ||
b128c09f BB |
1443 | void |
1444 | zio_suspend(spa_t *spa, zio_t *zio) | |
34dc7c2f | 1445 | { |
b128c09f BB |
1446 | if (spa_get_failmode(spa) == ZIO_FAILURE_MODE_PANIC) |
1447 | fm_panic("Pool '%s' has encountered an uncorrectable I/O " | |
1448 | "failure and the failure mode property for this pool " | |
1449 | "is set to panic.", spa_name(spa)); | |
34dc7c2f | 1450 | |
b128c09f | 1451 | zfs_ereport_post(FM_EREPORT_ZFS_IO_FAILURE, spa, NULL, NULL, 0, 0); |
34dc7c2f | 1452 | |
b128c09f | 1453 | mutex_enter(&spa->spa_suspend_lock); |
34dc7c2f | 1454 | |
b128c09f | 1455 | if (spa->spa_suspend_zio_root == NULL) |
9babb374 BB |
1456 | spa->spa_suspend_zio_root = zio_root(spa, NULL, NULL, |
1457 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE | | |
1458 | ZIO_FLAG_GODFATHER); | |
34dc7c2f | 1459 | |
b128c09f | 1460 | spa->spa_suspended = B_TRUE; |
34dc7c2f | 1461 | |
b128c09f | 1462 | if (zio != NULL) { |
9babb374 | 1463 | ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); |
b128c09f BB |
1464 | ASSERT(zio != spa->spa_suspend_zio_root); |
1465 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
d164b209 | 1466 | ASSERT(zio_unique_parent(zio) == NULL); |
b128c09f BB |
1467 | ASSERT(zio->io_stage == ZIO_STAGE_DONE); |
1468 | zio_add_child(spa->spa_suspend_zio_root, zio); | |
1469 | } | |
34dc7c2f | 1470 | |
b128c09f BB |
1471 | mutex_exit(&spa->spa_suspend_lock); |
1472 | } | |
34dc7c2f | 1473 | |
9babb374 | 1474 | int |
b128c09f BB |
1475 | zio_resume(spa_t *spa) |
1476 | { | |
9babb374 | 1477 | zio_t *pio; |
34dc7c2f BB |
1478 | |
1479 | /* | |
b128c09f | 1480 | * Reexecute all previously suspended i/o. |
34dc7c2f | 1481 | */ |
b128c09f BB |
1482 | mutex_enter(&spa->spa_suspend_lock); |
1483 | spa->spa_suspended = B_FALSE; | |
1484 | cv_broadcast(&spa->spa_suspend_cv); | |
1485 | pio = spa->spa_suspend_zio_root; | |
1486 | spa->spa_suspend_zio_root = NULL; | |
1487 | mutex_exit(&spa->spa_suspend_lock); | |
1488 | ||
1489 | if (pio == NULL) | |
9babb374 | 1490 | return (0); |
34dc7c2f | 1491 | |
9babb374 BB |
1492 | zio_reexecute(pio); |
1493 | return (zio_wait(pio)); | |
b128c09f BB |
1494 | } |
1495 | ||
1496 | void | |
1497 | zio_resume_wait(spa_t *spa) | |
1498 | { | |
1499 | mutex_enter(&spa->spa_suspend_lock); | |
1500 | while (spa_suspended(spa)) | |
1501 | cv_wait(&spa->spa_suspend_cv, &spa->spa_suspend_lock); | |
1502 | mutex_exit(&spa->spa_suspend_lock); | |
34dc7c2f BB |
1503 | } |
1504 | ||
1505 | /* | |
1506 | * ========================================================================== | |
b128c09f BB |
1507 | * Gang blocks. |
1508 | * | |
1509 | * A gang block is a collection of small blocks that looks to the DMU | |
1510 | * like one large block. When zio_dva_allocate() cannot find a block | |
1511 | * of the requested size, due to either severe fragmentation or the pool | |
1512 | * being nearly full, it calls zio_write_gang_block() to construct the | |
1513 | * block from smaller fragments. | |
1514 | * | |
1515 | * A gang block consists of a gang header (zio_gbh_phys_t) and up to | |
1516 | * three (SPA_GBH_NBLKPTRS) gang members. The gang header is just like | |
1517 | * an indirect block: it's an array of block pointers. It consumes | |
1518 | * only one sector and hence is allocatable regardless of fragmentation. | |
1519 | * The gang header's bps point to its gang members, which hold the data. | |
1520 | * | |
1521 | * Gang blocks are self-checksumming, using the bp's <vdev, offset, txg> | |
1522 | * as the verifier to ensure uniqueness of the SHA256 checksum. | |
1523 | * Critically, the gang block bp's blk_cksum is the checksum of the data, | |
1524 | * not the gang header. This ensures that data block signatures (needed for | |
1525 | * deduplication) are independent of how the block is physically stored. | |
1526 | * | |
1527 | * Gang blocks can be nested: a gang member may itself be a gang block. | |
1528 | * Thus every gang block is a tree in which root and all interior nodes are | |
1529 | * gang headers, and the leaves are normal blocks that contain user data. | |
1530 | * The root of the gang tree is called the gang leader. | |
1531 | * | |
1532 | * To perform any operation (read, rewrite, free, claim) on a gang block, | |
1533 | * zio_gang_assemble() first assembles the gang tree (minus data leaves) | |
1534 | * in the io_gang_tree field of the original logical i/o by recursively | |
1535 | * reading the gang leader and all gang headers below it. This yields | |
1536 | * an in-core tree containing the contents of every gang header and the | |
1537 | * bps for every constituent of the gang block. | |
1538 | * | |
1539 | * With the gang tree now assembled, zio_gang_issue() just walks the gang tree | |
1540 | * and invokes a callback on each bp. To free a gang block, zio_gang_issue() | |
1541 | * calls zio_free_gang() -- a trivial wrapper around zio_free() -- for each bp. | |
1542 | * zio_claim_gang() provides a similarly trivial wrapper for zio_claim(). | |
1543 | * zio_read_gang() is a wrapper around zio_read() that omits reading gang | |
1544 | * headers, since we already have those in io_gang_tree. zio_rewrite_gang() | |
1545 | * performs a zio_rewrite() of the data or, for gang headers, a zio_rewrite() | |
1546 | * of the gang header plus zio_checksum_compute() of the data to update the | |
1547 | * gang header's blk_cksum as described above. | |
1548 | * | |
1549 | * The two-phase assemble/issue model solves the problem of partial failure -- | |
1550 | * what if you'd freed part of a gang block but then couldn't read the | |
1551 | * gang header for another part? Assembling the entire gang tree first | |
1552 | * ensures that all the necessary gang header I/O has succeeded before | |
1553 | * starting the actual work of free, claim, or write. Once the gang tree | |
1554 | * is assembled, free and claim are in-memory operations that cannot fail. | |
1555 | * | |
1556 | * In the event that a gang write fails, zio_dva_unallocate() walks the | |
1557 | * gang tree to immediately free (i.e. insert back into the space map) | |
1558 | * everything we've allocated. This ensures that we don't get ENOSPC | |
1559 | * errors during repeated suspend/resume cycles due to a flaky device. | |
1560 | * | |
1561 | * Gang rewrites only happen during sync-to-convergence. If we can't assemble | |
1562 | * the gang tree, we won't modify the block, so we can safely defer the free | |
1563 | * (knowing that the block is still intact). If we *can* assemble the gang | |
1564 | * tree, then even if some of the rewrites fail, zio_dva_unallocate() will free | |
1565 | * each constituent bp and we can allocate a new block on the next sync pass. | |
1566 | * | |
1567 | * In all cases, the gang tree allows complete recovery from partial failure. | |
34dc7c2f BB |
1568 | * ========================================================================== |
1569 | */ | |
b128c09f BB |
1570 | |
1571 | static zio_t * | |
1572 | zio_read_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
34dc7c2f | 1573 | { |
b128c09f BB |
1574 | if (gn != NULL) |
1575 | return (pio); | |
34dc7c2f | 1576 | |
b128c09f BB |
1577 | return (zio_read(pio, pio->io_spa, bp, data, BP_GET_PSIZE(bp), |
1578 | NULL, NULL, pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), | |
1579 | &pio->io_bookmark)); | |
1580 | } | |
1581 | ||
1582 | zio_t * | |
1583 | zio_rewrite_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
1584 | { | |
1585 | zio_t *zio; | |
1586 | ||
1587 | if (gn != NULL) { | |
1588 | zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, | |
1589 | gn->gn_gbh, SPA_GANGBLOCKSIZE, NULL, NULL, pio->io_priority, | |
1590 | ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); | |
34dc7c2f | 1591 | /* |
b128c09f BB |
1592 | * As we rewrite each gang header, the pipeline will compute |
1593 | * a new gang block header checksum for it; but no one will | |
1594 | * compute a new data checksum, so we do that here. The one | |
1595 | * exception is the gang leader: the pipeline already computed | |
1596 | * its data checksum because that stage precedes gang assembly. | |
1597 | * (Presently, nothing actually uses interior data checksums; | |
1598 | * this is just good hygiene.) | |
34dc7c2f | 1599 | */ |
9babb374 | 1600 | if (gn != pio->io_gang_leader->io_gang_tree) { |
b128c09f BB |
1601 | zio_checksum_compute(zio, BP_GET_CHECKSUM(bp), |
1602 | data, BP_GET_PSIZE(bp)); | |
1603 | } | |
428870ff BB |
1604 | /* |
1605 | * If we are here to damage data for testing purposes, | |
1606 | * leave the GBH alone so that we can detect the damage. | |
1607 | */ | |
1608 | if (pio->io_gang_leader->io_flags & ZIO_FLAG_INDUCE_DAMAGE) | |
1609 | zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; | |
34dc7c2f | 1610 | } else { |
b128c09f BB |
1611 | zio = zio_rewrite(pio, pio->io_spa, pio->io_txg, bp, |
1612 | data, BP_GET_PSIZE(bp), NULL, NULL, pio->io_priority, | |
1613 | ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); | |
34dc7c2f BB |
1614 | } |
1615 | ||
b128c09f BB |
1616 | return (zio); |
1617 | } | |
34dc7c2f | 1618 | |
b128c09f BB |
1619 | /* ARGSUSED */ |
1620 | zio_t * | |
1621 | zio_free_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
1622 | { | |
428870ff BB |
1623 | return (zio_free_sync(pio, pio->io_spa, pio->io_txg, bp, |
1624 | ZIO_GANG_CHILD_FLAGS(pio))); | |
34dc7c2f BB |
1625 | } |
1626 | ||
b128c09f BB |
1627 | /* ARGSUSED */ |
1628 | zio_t * | |
1629 | zio_claim_gang(zio_t *pio, blkptr_t *bp, zio_gang_node_t *gn, void *data) | |
34dc7c2f | 1630 | { |
b128c09f BB |
1631 | return (zio_claim(pio, pio->io_spa, pio->io_txg, bp, |
1632 | NULL, NULL, ZIO_GANG_CHILD_FLAGS(pio))); | |
1633 | } | |
1634 | ||
1635 | static zio_gang_issue_func_t *zio_gang_issue_func[ZIO_TYPES] = { | |
1636 | NULL, | |
1637 | zio_read_gang, | |
1638 | zio_rewrite_gang, | |
1639 | zio_free_gang, | |
1640 | zio_claim_gang, | |
1641 | NULL | |
1642 | }; | |
34dc7c2f | 1643 | |
b128c09f | 1644 | static void zio_gang_tree_assemble_done(zio_t *zio); |
34dc7c2f | 1645 | |
b128c09f BB |
1646 | static zio_gang_node_t * |
1647 | zio_gang_node_alloc(zio_gang_node_t **gnpp) | |
1648 | { | |
1649 | zio_gang_node_t *gn; | |
34dc7c2f | 1650 | |
b128c09f | 1651 | ASSERT(*gnpp == NULL); |
34dc7c2f | 1652 | |
691f6ac4 | 1653 | gn = kmem_zalloc(sizeof (*gn), KM_PUSHPAGE); |
b128c09f BB |
1654 | gn->gn_gbh = zio_buf_alloc(SPA_GANGBLOCKSIZE); |
1655 | *gnpp = gn; | |
34dc7c2f | 1656 | |
b128c09f | 1657 | return (gn); |
34dc7c2f BB |
1658 | } |
1659 | ||
34dc7c2f | 1660 | static void |
b128c09f | 1661 | zio_gang_node_free(zio_gang_node_t **gnpp) |
34dc7c2f | 1662 | { |
b128c09f | 1663 | zio_gang_node_t *gn = *gnpp; |
d6320ddb | 1664 | int g; |
34dc7c2f | 1665 | |
d6320ddb | 1666 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) |
b128c09f BB |
1667 | ASSERT(gn->gn_child[g] == NULL); |
1668 | ||
1669 | zio_buf_free(gn->gn_gbh, SPA_GANGBLOCKSIZE); | |
1670 | kmem_free(gn, sizeof (*gn)); | |
1671 | *gnpp = NULL; | |
34dc7c2f BB |
1672 | } |
1673 | ||
b128c09f BB |
1674 | static void |
1675 | zio_gang_tree_free(zio_gang_node_t **gnpp) | |
34dc7c2f | 1676 | { |
b128c09f | 1677 | zio_gang_node_t *gn = *gnpp; |
d6320ddb | 1678 | int g; |
34dc7c2f | 1679 | |
b128c09f BB |
1680 | if (gn == NULL) |
1681 | return; | |
34dc7c2f | 1682 | |
d6320ddb | 1683 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) |
b128c09f | 1684 | zio_gang_tree_free(&gn->gn_child[g]); |
34dc7c2f | 1685 | |
b128c09f | 1686 | zio_gang_node_free(gnpp); |
34dc7c2f BB |
1687 | } |
1688 | ||
b128c09f | 1689 | static void |
9babb374 | 1690 | zio_gang_tree_assemble(zio_t *gio, blkptr_t *bp, zio_gang_node_t **gnpp) |
34dc7c2f | 1691 | { |
b128c09f BB |
1692 | zio_gang_node_t *gn = zio_gang_node_alloc(gnpp); |
1693 | ||
9babb374 | 1694 | ASSERT(gio->io_gang_leader == gio); |
b128c09f | 1695 | ASSERT(BP_IS_GANG(bp)); |
34dc7c2f | 1696 | |
9babb374 | 1697 | zio_nowait(zio_read(gio, gio->io_spa, bp, gn->gn_gbh, |
b128c09f | 1698 | SPA_GANGBLOCKSIZE, zio_gang_tree_assemble_done, gn, |
9babb374 | 1699 | gio->io_priority, ZIO_GANG_CHILD_FLAGS(gio), &gio->io_bookmark)); |
b128c09f | 1700 | } |
34dc7c2f | 1701 | |
b128c09f BB |
1702 | static void |
1703 | zio_gang_tree_assemble_done(zio_t *zio) | |
1704 | { | |
9babb374 | 1705 | zio_t *gio = zio->io_gang_leader; |
b128c09f BB |
1706 | zio_gang_node_t *gn = zio->io_private; |
1707 | blkptr_t *bp = zio->io_bp; | |
d6320ddb | 1708 | int g; |
34dc7c2f | 1709 | |
9babb374 | 1710 | ASSERT(gio == zio_unique_parent(zio)); |
428870ff | 1711 | ASSERT(zio->io_child_count == 0); |
34dc7c2f | 1712 | |
b128c09f BB |
1713 | if (zio->io_error) |
1714 | return; | |
34dc7c2f | 1715 | |
b128c09f BB |
1716 | if (BP_SHOULD_BYTESWAP(bp)) |
1717 | byteswap_uint64_array(zio->io_data, zio->io_size); | |
34dc7c2f | 1718 | |
b128c09f BB |
1719 | ASSERT(zio->io_data == gn->gn_gbh); |
1720 | ASSERT(zio->io_size == SPA_GANGBLOCKSIZE); | |
428870ff | 1721 | ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); |
34dc7c2f | 1722 | |
d6320ddb | 1723 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) { |
b128c09f BB |
1724 | blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; |
1725 | if (!BP_IS_GANG(gbp)) | |
1726 | continue; | |
9babb374 | 1727 | zio_gang_tree_assemble(gio, gbp, &gn->gn_child[g]); |
b128c09f | 1728 | } |
34dc7c2f BB |
1729 | } |
1730 | ||
b128c09f BB |
1731 | static void |
1732 | zio_gang_tree_issue(zio_t *pio, zio_gang_node_t *gn, blkptr_t *bp, void *data) | |
34dc7c2f | 1733 | { |
9babb374 | 1734 | zio_t *gio = pio->io_gang_leader; |
b128c09f | 1735 | zio_t *zio; |
d6320ddb | 1736 | int g; |
34dc7c2f | 1737 | |
b128c09f | 1738 | ASSERT(BP_IS_GANG(bp) == !!gn); |
9babb374 BB |
1739 | ASSERT(BP_GET_CHECKSUM(bp) == BP_GET_CHECKSUM(gio->io_bp)); |
1740 | ASSERT(BP_GET_LSIZE(bp) == BP_GET_PSIZE(bp) || gn == gio->io_gang_tree); | |
34dc7c2f | 1741 | |
b128c09f BB |
1742 | /* |
1743 | * If you're a gang header, your data is in gn->gn_gbh. | |
1744 | * If you're a gang member, your data is in 'data' and gn == NULL. | |
1745 | */ | |
9babb374 | 1746 | zio = zio_gang_issue_func[gio->io_type](pio, bp, gn, data); |
34dc7c2f | 1747 | |
b128c09f | 1748 | if (gn != NULL) { |
428870ff | 1749 | ASSERT(gn->gn_gbh->zg_tail.zec_magic == ZEC_MAGIC); |
34dc7c2f | 1750 | |
d6320ddb | 1751 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) { |
b128c09f BB |
1752 | blkptr_t *gbp = &gn->gn_gbh->zg_blkptr[g]; |
1753 | if (BP_IS_HOLE(gbp)) | |
1754 | continue; | |
1755 | zio_gang_tree_issue(zio, gn->gn_child[g], gbp, data); | |
1756 | data = (char *)data + BP_GET_PSIZE(gbp); | |
1757 | } | |
34dc7c2f BB |
1758 | } |
1759 | ||
9babb374 BB |
1760 | if (gn == gio->io_gang_tree) |
1761 | ASSERT3P((char *)gio->io_data + gio->io_size, ==, data); | |
34dc7c2f | 1762 | |
b128c09f BB |
1763 | if (zio != pio) |
1764 | zio_nowait(zio); | |
34dc7c2f BB |
1765 | } |
1766 | ||
1767 | static int | |
b128c09f | 1768 | zio_gang_assemble(zio_t *zio) |
34dc7c2f | 1769 | { |
b128c09f | 1770 | blkptr_t *bp = zio->io_bp; |
34dc7c2f | 1771 | |
9babb374 BB |
1772 | ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == NULL); |
1773 | ASSERT(zio->io_child_type > ZIO_CHILD_GANG); | |
1774 | ||
1775 | zio->io_gang_leader = zio; | |
34dc7c2f | 1776 | |
b128c09f | 1777 | zio_gang_tree_assemble(zio, bp, &zio->io_gang_tree); |
34dc7c2f BB |
1778 | |
1779 | return (ZIO_PIPELINE_CONTINUE); | |
1780 | } | |
1781 | ||
1782 | static int | |
b128c09f | 1783 | zio_gang_issue(zio_t *zio) |
34dc7c2f | 1784 | { |
b128c09f | 1785 | blkptr_t *bp = zio->io_bp; |
34dc7c2f | 1786 | |
b128c09f BB |
1787 | if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE)) |
1788 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 1789 | |
9babb374 BB |
1790 | ASSERT(BP_IS_GANG(bp) && zio->io_gang_leader == zio); |
1791 | ASSERT(zio->io_child_type > ZIO_CHILD_GANG); | |
34dc7c2f | 1792 | |
b128c09f | 1793 | if (zio->io_child_error[ZIO_CHILD_GANG] == 0) |
9babb374 | 1794 | zio_gang_tree_issue(zio, zio->io_gang_tree, bp, zio->io_data); |
b128c09f | 1795 | else |
9babb374 | 1796 | zio_gang_tree_free(&zio->io_gang_tree); |
34dc7c2f | 1797 | |
b128c09f | 1798 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; |
34dc7c2f BB |
1799 | |
1800 | return (ZIO_PIPELINE_CONTINUE); | |
1801 | } | |
1802 | ||
1803 | static void | |
b128c09f | 1804 | zio_write_gang_member_ready(zio_t *zio) |
34dc7c2f | 1805 | { |
d164b209 | 1806 | zio_t *pio = zio_unique_parent(zio); |
1fde1e37 | 1807 | ASSERTV(zio_t *gio = zio->io_gang_leader;) |
34dc7c2f BB |
1808 | dva_t *cdva = zio->io_bp->blk_dva; |
1809 | dva_t *pdva = pio->io_bp->blk_dva; | |
1810 | uint64_t asize; | |
d6320ddb | 1811 | int d; |
34dc7c2f | 1812 | |
b128c09f BB |
1813 | if (BP_IS_HOLE(zio->io_bp)) |
1814 | return; | |
1815 | ||
1816 | ASSERT(BP_IS_HOLE(&zio->io_bp_orig)); | |
1817 | ||
1818 | ASSERT(zio->io_child_type == ZIO_CHILD_GANG); | |
428870ff BB |
1819 | ASSERT3U(zio->io_prop.zp_copies, ==, gio->io_prop.zp_copies); |
1820 | ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp)); | |
1821 | ASSERT3U(pio->io_prop.zp_copies, <=, BP_GET_NDVAS(pio->io_bp)); | |
34dc7c2f | 1822 | ASSERT3U(BP_GET_NDVAS(zio->io_bp), <=, BP_GET_NDVAS(pio->io_bp)); |
34dc7c2f BB |
1823 | |
1824 | mutex_enter(&pio->io_lock); | |
d6320ddb | 1825 | for (d = 0; d < BP_GET_NDVAS(zio->io_bp); d++) { |
34dc7c2f BB |
1826 | ASSERT(DVA_GET_GANG(&pdva[d])); |
1827 | asize = DVA_GET_ASIZE(&pdva[d]); | |
1828 | asize += DVA_GET_ASIZE(&cdva[d]); | |
1829 | DVA_SET_ASIZE(&pdva[d], asize); | |
1830 | } | |
1831 | mutex_exit(&pio->io_lock); | |
1832 | } | |
1833 | ||
1834 | static int | |
b128c09f | 1835 | zio_write_gang_block(zio_t *pio) |
34dc7c2f | 1836 | { |
b128c09f BB |
1837 | spa_t *spa = pio->io_spa; |
1838 | blkptr_t *bp = pio->io_bp; | |
9babb374 | 1839 | zio_t *gio = pio->io_gang_leader; |
b128c09f BB |
1840 | zio_t *zio; |
1841 | zio_gang_node_t *gn, **gnpp; | |
34dc7c2f | 1842 | zio_gbh_phys_t *gbh; |
b128c09f BB |
1843 | uint64_t txg = pio->io_txg; |
1844 | uint64_t resid = pio->io_size; | |
1845 | uint64_t lsize; | |
428870ff BB |
1846 | int copies = gio->io_prop.zp_copies; |
1847 | int gbh_copies = MIN(copies + 1, spa_max_replication(spa)); | |
b128c09f | 1848 | zio_prop_t zp; |
d6320ddb | 1849 | int g, error; |
34dc7c2f | 1850 | |
428870ff BB |
1851 | error = metaslab_alloc(spa, spa_normal_class(spa), SPA_GANGBLOCKSIZE, |
1852 | bp, gbh_copies, txg, pio == gio ? NULL : gio->io_bp, | |
b128c09f | 1853 | METASLAB_HINTBP_FAVOR | METASLAB_GANG_HEADER); |
34dc7c2f | 1854 | if (error) { |
b128c09f | 1855 | pio->io_error = error; |
34dc7c2f BB |
1856 | return (ZIO_PIPELINE_CONTINUE); |
1857 | } | |
1858 | ||
9babb374 BB |
1859 | if (pio == gio) { |
1860 | gnpp = &gio->io_gang_tree; | |
b128c09f BB |
1861 | } else { |
1862 | gnpp = pio->io_private; | |
1863 | ASSERT(pio->io_ready == zio_write_gang_member_ready); | |
34dc7c2f BB |
1864 | } |
1865 | ||
b128c09f BB |
1866 | gn = zio_gang_node_alloc(gnpp); |
1867 | gbh = gn->gn_gbh; | |
1868 | bzero(gbh, SPA_GANGBLOCKSIZE); | |
34dc7c2f | 1869 | |
b128c09f BB |
1870 | /* |
1871 | * Create the gang header. | |
1872 | */ | |
1873 | zio = zio_rewrite(pio, spa, txg, bp, gbh, SPA_GANGBLOCKSIZE, NULL, NULL, | |
1874 | pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), &pio->io_bookmark); | |
34dc7c2f | 1875 | |
b128c09f BB |
1876 | /* |
1877 | * Create and nowait the gang children. | |
1878 | */ | |
d6320ddb | 1879 | for (g = 0; resid != 0; resid -= lsize, g++) { |
b128c09f BB |
1880 | lsize = P2ROUNDUP(resid / (SPA_GBH_NBLKPTRS - g), |
1881 | SPA_MINBLOCKSIZE); | |
1882 | ASSERT(lsize >= SPA_MINBLOCKSIZE && lsize <= resid); | |
1883 | ||
9babb374 | 1884 | zp.zp_checksum = gio->io_prop.zp_checksum; |
b128c09f BB |
1885 | zp.zp_compress = ZIO_COMPRESS_OFF; |
1886 | zp.zp_type = DMU_OT_NONE; | |
1887 | zp.zp_level = 0; | |
428870ff BB |
1888 | zp.zp_copies = gio->io_prop.zp_copies; |
1889 | zp.zp_dedup = 0; | |
1890 | zp.zp_dedup_verify = 0; | |
b128c09f BB |
1891 | |
1892 | zio_nowait(zio_write(zio, spa, txg, &gbh->zg_blkptr[g], | |
1893 | (char *)pio->io_data + (pio->io_size - resid), lsize, &zp, | |
1894 | zio_write_gang_member_ready, NULL, &gn->gn_child[g], | |
1895 | pio->io_priority, ZIO_GANG_CHILD_FLAGS(pio), | |
1896 | &pio->io_bookmark)); | |
1897 | } | |
34dc7c2f BB |
1898 | |
1899 | /* | |
b128c09f | 1900 | * Set pio's pipeline to just wait for zio to finish. |
34dc7c2f | 1901 | */ |
b128c09f BB |
1902 | pio->io_pipeline = ZIO_INTERLOCK_PIPELINE; |
1903 | ||
920dd524 ED |
1904 | /* |
1905 | * We didn't allocate this bp, so make sure it doesn't get unmarked. | |
1906 | */ | |
1907 | pio->io_flags &= ~ZIO_FLAG_FASTWRITE; | |
1908 | ||
b128c09f BB |
1909 | zio_nowait(zio); |
1910 | ||
1911 | return (ZIO_PIPELINE_CONTINUE); | |
34dc7c2f BB |
1912 | } |
1913 | ||
1914 | /* | |
1915 | * ========================================================================== | |
428870ff | 1916 | * Dedup |
34dc7c2f BB |
1917 | * ========================================================================== |
1918 | */ | |
428870ff BB |
1919 | static void |
1920 | zio_ddt_child_read_done(zio_t *zio) | |
1921 | { | |
1922 | blkptr_t *bp = zio->io_bp; | |
1923 | ddt_entry_t *dde = zio->io_private; | |
1924 | ddt_phys_t *ddp; | |
1925 | zio_t *pio = zio_unique_parent(zio); | |
1926 | ||
1927 | mutex_enter(&pio->io_lock); | |
1928 | ddp = ddt_phys_select(dde, bp); | |
1929 | if (zio->io_error == 0) | |
1930 | ddt_phys_clear(ddp); /* this ddp doesn't need repair */ | |
1931 | if (zio->io_error == 0 && dde->dde_repair_data == NULL) | |
1932 | dde->dde_repair_data = zio->io_data; | |
1933 | else | |
1934 | zio_buf_free(zio->io_data, zio->io_size); | |
1935 | mutex_exit(&pio->io_lock); | |
1936 | } | |
1937 | ||
1938 | static int | |
1939 | zio_ddt_read_start(zio_t *zio) | |
1940 | { | |
1941 | blkptr_t *bp = zio->io_bp; | |
d6320ddb | 1942 | int p; |
428870ff BB |
1943 | |
1944 | ASSERT(BP_GET_DEDUP(bp)); | |
1945 | ASSERT(BP_GET_PSIZE(bp) == zio->io_size); | |
1946 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1947 | ||
1948 | if (zio->io_child_error[ZIO_CHILD_DDT]) { | |
1949 | ddt_t *ddt = ddt_select(zio->io_spa, bp); | |
1950 | ddt_entry_t *dde = ddt_repair_start(ddt, bp); | |
1951 | ddt_phys_t *ddp = dde->dde_phys; | |
1952 | ddt_phys_t *ddp_self = ddt_phys_select(dde, bp); | |
1953 | blkptr_t blk; | |
1954 | ||
1955 | ASSERT(zio->io_vsd == NULL); | |
1956 | zio->io_vsd = dde; | |
1957 | ||
1958 | if (ddp_self == NULL) | |
1959 | return (ZIO_PIPELINE_CONTINUE); | |
1960 | ||
d6320ddb | 1961 | for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) { |
428870ff BB |
1962 | if (ddp->ddp_phys_birth == 0 || ddp == ddp_self) |
1963 | continue; | |
1964 | ddt_bp_create(ddt->ddt_checksum, &dde->dde_key, ddp, | |
1965 | &blk); | |
1966 | zio_nowait(zio_read(zio, zio->io_spa, &blk, | |
1967 | zio_buf_alloc(zio->io_size), zio->io_size, | |
1968 | zio_ddt_child_read_done, dde, zio->io_priority, | |
1969 | ZIO_DDT_CHILD_FLAGS(zio) | ZIO_FLAG_DONT_PROPAGATE, | |
1970 | &zio->io_bookmark)); | |
1971 | } | |
1972 | return (ZIO_PIPELINE_CONTINUE); | |
1973 | } | |
1974 | ||
1975 | zio_nowait(zio_read(zio, zio->io_spa, bp, | |
1976 | zio->io_data, zio->io_size, NULL, NULL, zio->io_priority, | |
1977 | ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark)); | |
1978 | ||
1979 | return (ZIO_PIPELINE_CONTINUE); | |
1980 | } | |
1981 | ||
1982 | static int | |
1983 | zio_ddt_read_done(zio_t *zio) | |
1984 | { | |
1985 | blkptr_t *bp = zio->io_bp; | |
1986 | ||
1987 | if (zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE)) | |
1988 | return (ZIO_PIPELINE_STOP); | |
1989 | ||
1990 | ASSERT(BP_GET_DEDUP(bp)); | |
1991 | ASSERT(BP_GET_PSIZE(bp) == zio->io_size); | |
1992 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
1993 | ||
1994 | if (zio->io_child_error[ZIO_CHILD_DDT]) { | |
1995 | ddt_t *ddt = ddt_select(zio->io_spa, bp); | |
1996 | ddt_entry_t *dde = zio->io_vsd; | |
1997 | if (ddt == NULL) { | |
1998 | ASSERT(spa_load_state(zio->io_spa) != SPA_LOAD_NONE); | |
1999 | return (ZIO_PIPELINE_CONTINUE); | |
2000 | } | |
2001 | if (dde == NULL) { | |
2002 | zio->io_stage = ZIO_STAGE_DDT_READ_START >> 1; | |
2003 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, B_FALSE); | |
2004 | return (ZIO_PIPELINE_STOP); | |
2005 | } | |
2006 | if (dde->dde_repair_data != NULL) { | |
2007 | bcopy(dde->dde_repair_data, zio->io_data, zio->io_size); | |
2008 | zio->io_child_error[ZIO_CHILD_DDT] = 0; | |
2009 | } | |
2010 | ddt_repair_done(ddt, dde); | |
2011 | zio->io_vsd = NULL; | |
2012 | } | |
2013 | ||
2014 | ASSERT(zio->io_vsd == NULL); | |
2015 | ||
2016 | return (ZIO_PIPELINE_CONTINUE); | |
2017 | } | |
2018 | ||
2019 | static boolean_t | |
2020 | zio_ddt_collision(zio_t *zio, ddt_t *ddt, ddt_entry_t *dde) | |
2021 | { | |
2022 | spa_t *spa = zio->io_spa; | |
d6320ddb | 2023 | int p; |
428870ff BB |
2024 | |
2025 | /* | |
2026 | * Note: we compare the original data, not the transformed data, | |
2027 | * because when zio->io_bp is an override bp, we will not have | |
2028 | * pushed the I/O transforms. That's an important optimization | |
2029 | * because otherwise we'd compress/encrypt all dmu_sync() data twice. | |
2030 | */ | |
d6320ddb | 2031 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { |
428870ff BB |
2032 | zio_t *lio = dde->dde_lead_zio[p]; |
2033 | ||
2034 | if (lio != NULL) { | |
2035 | return (lio->io_orig_size != zio->io_orig_size || | |
2036 | bcmp(zio->io_orig_data, lio->io_orig_data, | |
2037 | zio->io_orig_size) != 0); | |
2038 | } | |
2039 | } | |
2040 | ||
d6320ddb | 2041 | for (p = DDT_PHYS_SINGLE; p <= DDT_PHYS_TRIPLE; p++) { |
428870ff BB |
2042 | ddt_phys_t *ddp = &dde->dde_phys[p]; |
2043 | ||
2044 | if (ddp->ddp_phys_birth != 0) { | |
2045 | arc_buf_t *abuf = NULL; | |
2046 | uint32_t aflags = ARC_WAIT; | |
2047 | blkptr_t blk = *zio->io_bp; | |
2048 | int error; | |
2049 | ||
2050 | ddt_bp_fill(ddp, &blk, ddp->ddp_phys_birth); | |
2051 | ||
2052 | ddt_exit(ddt); | |
2053 | ||
294f6806 | 2054 | error = arc_read(NULL, spa, &blk, |
428870ff BB |
2055 | arc_getbuf_func, &abuf, ZIO_PRIORITY_SYNC_READ, |
2056 | ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE, | |
2057 | &aflags, &zio->io_bookmark); | |
2058 | ||
2059 | if (error == 0) { | |
2060 | if (arc_buf_size(abuf) != zio->io_orig_size || | |
2061 | bcmp(abuf->b_data, zio->io_orig_data, | |
2062 | zio->io_orig_size) != 0) | |
2063 | error = EEXIST; | |
2064 | VERIFY(arc_buf_remove_ref(abuf, &abuf) == 1); | |
2065 | } | |
2066 | ||
2067 | ddt_enter(ddt); | |
2068 | return (error != 0); | |
2069 | } | |
2070 | } | |
2071 | ||
2072 | return (B_FALSE); | |
2073 | } | |
2074 | ||
2075 | static void | |
2076 | zio_ddt_child_write_ready(zio_t *zio) | |
2077 | { | |
2078 | int p = zio->io_prop.zp_copies; | |
2079 | ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); | |
2080 | ddt_entry_t *dde = zio->io_private; | |
2081 | ddt_phys_t *ddp = &dde->dde_phys[p]; | |
2082 | zio_t *pio; | |
2083 | ||
2084 | if (zio->io_error) | |
2085 | return; | |
2086 | ||
2087 | ddt_enter(ddt); | |
2088 | ||
2089 | ASSERT(dde->dde_lead_zio[p] == zio); | |
2090 | ||
2091 | ddt_phys_fill(ddp, zio->io_bp); | |
2092 | ||
2093 | while ((pio = zio_walk_parents(zio)) != NULL) | |
2094 | ddt_bp_fill(ddp, pio->io_bp, zio->io_txg); | |
2095 | ||
2096 | ddt_exit(ddt); | |
2097 | } | |
2098 | ||
2099 | static void | |
2100 | zio_ddt_child_write_done(zio_t *zio) | |
2101 | { | |
2102 | int p = zio->io_prop.zp_copies; | |
2103 | ddt_t *ddt = ddt_select(zio->io_spa, zio->io_bp); | |
2104 | ddt_entry_t *dde = zio->io_private; | |
2105 | ddt_phys_t *ddp = &dde->dde_phys[p]; | |
2106 | ||
2107 | ddt_enter(ddt); | |
2108 | ||
2109 | ASSERT(ddp->ddp_refcnt == 0); | |
2110 | ASSERT(dde->dde_lead_zio[p] == zio); | |
2111 | dde->dde_lead_zio[p] = NULL; | |
2112 | ||
2113 | if (zio->io_error == 0) { | |
2114 | while (zio_walk_parents(zio) != NULL) | |
2115 | ddt_phys_addref(ddp); | |
2116 | } else { | |
2117 | ddt_phys_clear(ddp); | |
2118 | } | |
2119 | ||
2120 | ddt_exit(ddt); | |
2121 | } | |
2122 | ||
2123 | static void | |
2124 | zio_ddt_ditto_write_done(zio_t *zio) | |
2125 | { | |
2126 | int p = DDT_PHYS_DITTO; | |
428870ff BB |
2127 | blkptr_t *bp = zio->io_bp; |
2128 | ddt_t *ddt = ddt_select(zio->io_spa, bp); | |
2129 | ddt_entry_t *dde = zio->io_private; | |
2130 | ddt_phys_t *ddp = &dde->dde_phys[p]; | |
2131 | ddt_key_t *ddk = &dde->dde_key; | |
1fde1e37 | 2132 | ASSERTV(zio_prop_t *zp = &zio->io_prop); |
428870ff BB |
2133 | |
2134 | ddt_enter(ddt); | |
2135 | ||
2136 | ASSERT(ddp->ddp_refcnt == 0); | |
2137 | ASSERT(dde->dde_lead_zio[p] == zio); | |
2138 | dde->dde_lead_zio[p] = NULL; | |
2139 | ||
2140 | if (zio->io_error == 0) { | |
2141 | ASSERT(ZIO_CHECKSUM_EQUAL(bp->blk_cksum, ddk->ddk_cksum)); | |
2142 | ASSERT(zp->zp_copies < SPA_DVAS_PER_BP); | |
2143 | ASSERT(zp->zp_copies == BP_GET_NDVAS(bp) - BP_IS_GANG(bp)); | |
2144 | if (ddp->ddp_phys_birth != 0) | |
2145 | ddt_phys_free(ddt, ddk, ddp, zio->io_txg); | |
2146 | ddt_phys_fill(ddp, bp); | |
2147 | } | |
2148 | ||
2149 | ddt_exit(ddt); | |
2150 | } | |
2151 | ||
2152 | static int | |
2153 | zio_ddt_write(zio_t *zio) | |
2154 | { | |
2155 | spa_t *spa = zio->io_spa; | |
2156 | blkptr_t *bp = zio->io_bp; | |
2157 | uint64_t txg = zio->io_txg; | |
2158 | zio_prop_t *zp = &zio->io_prop; | |
2159 | int p = zp->zp_copies; | |
2160 | int ditto_copies; | |
2161 | zio_t *cio = NULL; | |
2162 | zio_t *dio = NULL; | |
2163 | ddt_t *ddt = ddt_select(spa, bp); | |
2164 | ddt_entry_t *dde; | |
2165 | ddt_phys_t *ddp; | |
2166 | ||
2167 | ASSERT(BP_GET_DEDUP(bp)); | |
2168 | ASSERT(BP_GET_CHECKSUM(bp) == zp->zp_checksum); | |
2169 | ASSERT(BP_IS_HOLE(bp) || zio->io_bp_override); | |
2170 | ||
2171 | ddt_enter(ddt); | |
2172 | dde = ddt_lookup(ddt, bp, B_TRUE); | |
2173 | ddp = &dde->dde_phys[p]; | |
2174 | ||
2175 | if (zp->zp_dedup_verify && zio_ddt_collision(zio, ddt, dde)) { | |
2176 | /* | |
2177 | * If we're using a weak checksum, upgrade to a strong checksum | |
2178 | * and try again. If we're already using a strong checksum, | |
2179 | * we can't resolve it, so just convert to an ordinary write. | |
2180 | * (And automatically e-mail a paper to Nature?) | |
2181 | */ | |
2182 | if (!zio_checksum_table[zp->zp_checksum].ci_dedup) { | |
2183 | zp->zp_checksum = spa_dedup_checksum(spa); | |
2184 | zio_pop_transforms(zio); | |
2185 | zio->io_stage = ZIO_STAGE_OPEN; | |
2186 | BP_ZERO(bp); | |
2187 | } else { | |
2188 | zp->zp_dedup = 0; | |
2189 | } | |
2190 | zio->io_pipeline = ZIO_WRITE_PIPELINE; | |
2191 | ddt_exit(ddt); | |
2192 | return (ZIO_PIPELINE_CONTINUE); | |
2193 | } | |
2194 | ||
2195 | ditto_copies = ddt_ditto_copies_needed(ddt, dde, ddp); | |
2196 | ASSERT(ditto_copies < SPA_DVAS_PER_BP); | |
2197 | ||
2198 | if (ditto_copies > ddt_ditto_copies_present(dde) && | |
2199 | dde->dde_lead_zio[DDT_PHYS_DITTO] == NULL) { | |
2200 | zio_prop_t czp = *zp; | |
2201 | ||
2202 | czp.zp_copies = ditto_copies; | |
2203 | ||
2204 | /* | |
2205 | * If we arrived here with an override bp, we won't have run | |
2206 | * the transform stack, so we won't have the data we need to | |
2207 | * generate a child i/o. So, toss the override bp and restart. | |
2208 | * This is safe, because using the override bp is just an | |
2209 | * optimization; and it's rare, so the cost doesn't matter. | |
2210 | */ | |
2211 | if (zio->io_bp_override) { | |
2212 | zio_pop_transforms(zio); | |
2213 | zio->io_stage = ZIO_STAGE_OPEN; | |
2214 | zio->io_pipeline = ZIO_WRITE_PIPELINE; | |
2215 | zio->io_bp_override = NULL; | |
2216 | BP_ZERO(bp); | |
2217 | ddt_exit(ddt); | |
2218 | return (ZIO_PIPELINE_CONTINUE); | |
2219 | } | |
2220 | ||
2221 | dio = zio_write(zio, spa, txg, bp, zio->io_orig_data, | |
2222 | zio->io_orig_size, &czp, NULL, | |
2223 | zio_ddt_ditto_write_done, dde, zio->io_priority, | |
2224 | ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); | |
2225 | ||
2226 | zio_push_transform(dio, zio->io_data, zio->io_size, 0, NULL); | |
2227 | dde->dde_lead_zio[DDT_PHYS_DITTO] = dio; | |
2228 | } | |
2229 | ||
2230 | if (ddp->ddp_phys_birth != 0 || dde->dde_lead_zio[p] != NULL) { | |
2231 | if (ddp->ddp_phys_birth != 0) | |
2232 | ddt_bp_fill(ddp, bp, txg); | |
2233 | if (dde->dde_lead_zio[p] != NULL) | |
2234 | zio_add_child(zio, dde->dde_lead_zio[p]); | |
2235 | else | |
2236 | ddt_phys_addref(ddp); | |
2237 | } else if (zio->io_bp_override) { | |
2238 | ASSERT(bp->blk_birth == txg); | |
2239 | ASSERT(BP_EQUAL(bp, zio->io_bp_override)); | |
2240 | ddt_phys_fill(ddp, bp); | |
2241 | ddt_phys_addref(ddp); | |
2242 | } else { | |
2243 | cio = zio_write(zio, spa, txg, bp, zio->io_orig_data, | |
2244 | zio->io_orig_size, zp, zio_ddt_child_write_ready, | |
2245 | zio_ddt_child_write_done, dde, zio->io_priority, | |
2246 | ZIO_DDT_CHILD_FLAGS(zio), &zio->io_bookmark); | |
2247 | ||
2248 | zio_push_transform(cio, zio->io_data, zio->io_size, 0, NULL); | |
2249 | dde->dde_lead_zio[p] = cio; | |
2250 | } | |
2251 | ||
2252 | ddt_exit(ddt); | |
2253 | ||
2254 | if (cio) | |
2255 | zio_nowait(cio); | |
2256 | if (dio) | |
2257 | zio_nowait(dio); | |
2258 | ||
2259 | return (ZIO_PIPELINE_CONTINUE); | |
2260 | } | |
2261 | ||
2262 | ddt_entry_t *freedde; /* for debugging */ | |
b128c09f | 2263 | |
428870ff BB |
2264 | static int |
2265 | zio_ddt_free(zio_t *zio) | |
2266 | { | |
2267 | spa_t *spa = zio->io_spa; | |
2268 | blkptr_t *bp = zio->io_bp; | |
2269 | ddt_t *ddt = ddt_select(spa, bp); | |
2270 | ddt_entry_t *dde; | |
2271 | ddt_phys_t *ddp; | |
2272 | ||
2273 | ASSERT(BP_GET_DEDUP(bp)); | |
2274 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
2275 | ||
2276 | ddt_enter(ddt); | |
2277 | freedde = dde = ddt_lookup(ddt, bp, B_TRUE); | |
5dc6af0e BB |
2278 | if (dde) { |
2279 | ddp = ddt_phys_select(dde, bp); | |
2280 | if (ddp) | |
2281 | ddt_phys_decref(ddp); | |
2282 | } | |
428870ff BB |
2283 | ddt_exit(ddt); |
2284 | ||
2285 | return (ZIO_PIPELINE_CONTINUE); | |
2286 | } | |
2287 | ||
2288 | /* | |
2289 | * ========================================================================== | |
2290 | * Allocate and free blocks | |
2291 | * ========================================================================== | |
2292 | */ | |
34dc7c2f BB |
2293 | static int |
2294 | zio_dva_allocate(zio_t *zio) | |
2295 | { | |
2296 | spa_t *spa = zio->io_spa; | |
428870ff | 2297 | metaslab_class_t *mc = spa_normal_class(spa); |
34dc7c2f BB |
2298 | blkptr_t *bp = zio->io_bp; |
2299 | int error; | |
6d974228 | 2300 | int flags = 0; |
34dc7c2f | 2301 | |
9babb374 BB |
2302 | if (zio->io_gang_leader == NULL) { |
2303 | ASSERT(zio->io_child_type > ZIO_CHILD_GANG); | |
2304 | zio->io_gang_leader = zio; | |
2305 | } | |
2306 | ||
34dc7c2f | 2307 | ASSERT(BP_IS_HOLE(bp)); |
c99c9001 | 2308 | ASSERT0(BP_GET_NDVAS(bp)); |
428870ff BB |
2309 | ASSERT3U(zio->io_prop.zp_copies, >, 0); |
2310 | ASSERT3U(zio->io_prop.zp_copies, <=, spa_max_replication(spa)); | |
34dc7c2f BB |
2311 | ASSERT3U(zio->io_size, ==, BP_GET_PSIZE(bp)); |
2312 | ||
6d974228 GW |
2313 | /* |
2314 | * The dump device does not support gang blocks so allocation on | |
2315 | * behalf of the dump device (i.e. ZIO_FLAG_NODATA) must avoid | |
2316 | * the "fast" gang feature. | |
2317 | */ | |
2318 | flags |= (zio->io_flags & ZIO_FLAG_NODATA) ? METASLAB_GANG_AVOID : 0; | |
2319 | flags |= (zio->io_flags & ZIO_FLAG_GANG_CHILD) ? | |
2320 | METASLAB_GANG_CHILD : 0; | |
920dd524 | 2321 | flags |= (zio->io_flags & ZIO_FLAG_FASTWRITE) ? METASLAB_FASTWRITE : 0; |
b128c09f | 2322 | error = metaslab_alloc(spa, mc, zio->io_size, bp, |
6d974228 | 2323 | zio->io_prop.zp_copies, zio->io_txg, NULL, flags); |
34dc7c2f | 2324 | |
b128c09f | 2325 | if (error) { |
6d974228 GW |
2326 | spa_dbgmsg(spa, "%s: metaslab allocation failure: zio %p, " |
2327 | "size %llu, error %d", spa_name(spa), zio, zio->io_size, | |
2328 | error); | |
b128c09f BB |
2329 | if (error == ENOSPC && zio->io_size > SPA_MINBLOCKSIZE) |
2330 | return (zio_write_gang_block(zio)); | |
34dc7c2f BB |
2331 | zio->io_error = error; |
2332 | } | |
2333 | ||
2334 | return (ZIO_PIPELINE_CONTINUE); | |
2335 | } | |
2336 | ||
2337 | static int | |
2338 | zio_dva_free(zio_t *zio) | |
2339 | { | |
b128c09f | 2340 | metaslab_free(zio->io_spa, zio->io_bp, zio->io_txg, B_FALSE); |
34dc7c2f BB |
2341 | |
2342 | return (ZIO_PIPELINE_CONTINUE); | |
2343 | } | |
2344 | ||
2345 | static int | |
2346 | zio_dva_claim(zio_t *zio) | |
2347 | { | |
b128c09f BB |
2348 | int error; |
2349 | ||
2350 | error = metaslab_claim(zio->io_spa, zio->io_bp, zio->io_txg); | |
2351 | if (error) | |
2352 | zio->io_error = error; | |
34dc7c2f BB |
2353 | |
2354 | return (ZIO_PIPELINE_CONTINUE); | |
2355 | } | |
2356 | ||
b128c09f BB |
2357 | /* |
2358 | * Undo an allocation. This is used by zio_done() when an I/O fails | |
2359 | * and we want to give back the block we just allocated. | |
2360 | * This handles both normal blocks and gang blocks. | |
2361 | */ | |
2362 | static void | |
2363 | zio_dva_unallocate(zio_t *zio, zio_gang_node_t *gn, blkptr_t *bp) | |
2364 | { | |
d6320ddb BB |
2365 | int g; |
2366 | ||
b128c09f | 2367 | ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); |
428870ff | 2368 | ASSERT(zio->io_bp_override == NULL); |
b128c09f BB |
2369 | |
2370 | if (!BP_IS_HOLE(bp)) | |
428870ff | 2371 | metaslab_free(zio->io_spa, bp, bp->blk_birth, B_TRUE); |
b128c09f BB |
2372 | |
2373 | if (gn != NULL) { | |
d6320ddb | 2374 | for (g = 0; g < SPA_GBH_NBLKPTRS; g++) { |
b128c09f BB |
2375 | zio_dva_unallocate(zio, gn->gn_child[g], |
2376 | &gn->gn_gbh->zg_blkptr[g]); | |
2377 | } | |
2378 | } | |
2379 | } | |
2380 | ||
2381 | /* | |
2382 | * Try to allocate an intent log block. Return 0 on success, errno on failure. | |
2383 | */ | |
2384 | int | |
920dd524 ED |
2385 | zio_alloc_zil(spa_t *spa, uint64_t txg, blkptr_t *new_bp, uint64_t size, |
2386 | boolean_t use_slog) | |
b128c09f | 2387 | { |
428870ff | 2388 | int error = 1; |
b128c09f | 2389 | |
428870ff BB |
2390 | ASSERT(txg > spa_syncing_txg(spa)); |
2391 | ||
ebf8e3a2 BB |
2392 | /* |
2393 | * ZIL blocks are always contiguous (i.e. not gang blocks) so we | |
2394 | * set the METASLAB_GANG_AVOID flag so that they don't "fast gang" | |
2395 | * when allocating them. | |
2396 | */ | |
2397 | if (use_slog) { | |
428870ff | 2398 | error = metaslab_alloc(spa, spa_log_class(spa), size, |
920dd524 ED |
2399 | new_bp, 1, txg, NULL, |
2400 | METASLAB_FASTWRITE | METASLAB_GANG_AVOID); | |
ebf8e3a2 | 2401 | } |
b128c09f | 2402 | |
ebf8e3a2 | 2403 | if (error) { |
428870ff | 2404 | error = metaslab_alloc(spa, spa_normal_class(spa), size, |
920dd524 ED |
2405 | new_bp, 1, txg, NULL, |
2406 | METASLAB_FASTWRITE | METASLAB_GANG_AVOID); | |
ebf8e3a2 | 2407 | } |
b128c09f BB |
2408 | |
2409 | if (error == 0) { | |
2410 | BP_SET_LSIZE(new_bp, size); | |
2411 | BP_SET_PSIZE(new_bp, size); | |
2412 | BP_SET_COMPRESS(new_bp, ZIO_COMPRESS_OFF); | |
428870ff BB |
2413 | BP_SET_CHECKSUM(new_bp, |
2414 | spa_version(spa) >= SPA_VERSION_SLIM_ZIL | |
2415 | ? ZIO_CHECKSUM_ZILOG2 : ZIO_CHECKSUM_ZILOG); | |
b128c09f BB |
2416 | BP_SET_TYPE(new_bp, DMU_OT_INTENT_LOG); |
2417 | BP_SET_LEVEL(new_bp, 0); | |
428870ff | 2418 | BP_SET_DEDUP(new_bp, 0); |
b128c09f BB |
2419 | BP_SET_BYTEORDER(new_bp, ZFS_HOST_BYTEORDER); |
2420 | } | |
2421 | ||
2422 | return (error); | |
2423 | } | |
2424 | ||
2425 | /* | |
428870ff | 2426 | * Free an intent log block. |
b128c09f BB |
2427 | */ |
2428 | void | |
428870ff | 2429 | zio_free_zil(spa_t *spa, uint64_t txg, blkptr_t *bp) |
b128c09f | 2430 | { |
428870ff | 2431 | ASSERT(BP_GET_TYPE(bp) == DMU_OT_INTENT_LOG); |
b128c09f BB |
2432 | ASSERT(!BP_IS_GANG(bp)); |
2433 | ||
428870ff | 2434 | zio_free(spa, txg, bp); |
b128c09f BB |
2435 | } |
2436 | ||
34dc7c2f BB |
2437 | /* |
2438 | * ========================================================================== | |
2439 | * Read and write to physical devices | |
2440 | * ========================================================================== | |
2441 | */ | |
34dc7c2f BB |
2442 | static int |
2443 | zio_vdev_io_start(zio_t *zio) | |
2444 | { | |
2445 | vdev_t *vd = zio->io_vd; | |
34dc7c2f BB |
2446 | uint64_t align; |
2447 | spa_t *spa = zio->io_spa; | |
2448 | ||
b128c09f BB |
2449 | ASSERT(zio->io_error == 0); |
2450 | ASSERT(zio->io_child_error[ZIO_CHILD_VDEV] == 0); | |
34dc7c2f | 2451 | |
b128c09f BB |
2452 | if (vd == NULL) { |
2453 | if (!(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) | |
2454 | spa_config_enter(spa, SCL_ZIO, zio, RW_READER); | |
34dc7c2f | 2455 | |
b128c09f BB |
2456 | /* |
2457 | * The mirror_ops handle multiple DVAs in a single BP. | |
2458 | */ | |
2459 | return (vdev_mirror_ops.vdev_op_io_start(zio)); | |
34dc7c2f BB |
2460 | } |
2461 | ||
572e2857 BB |
2462 | /* |
2463 | * We keep track of time-sensitive I/Os so that the scan thread | |
2464 | * can quickly react to certain workloads. In particular, we care | |
2465 | * about non-scrubbing, top-level reads and writes with the following | |
2466 | * characteristics: | |
2467 | * - synchronous writes of user data to non-slog devices | |
2468 | * - any reads of user data | |
2469 | * When these conditions are met, adjust the timestamp of spa_last_io | |
2470 | * which allows the scan thread to adjust its workload accordingly. | |
2471 | */ | |
2472 | if (!(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && zio->io_bp != NULL && | |
2473 | vd == vd->vdev_top && !vd->vdev_islog && | |
2474 | zio->io_bookmark.zb_objset != DMU_META_OBJSET && | |
2475 | zio->io_txg != spa_syncing_txg(spa)) { | |
2476 | uint64_t old = spa->spa_last_io; | |
2477 | uint64_t new = ddi_get_lbolt64(); | |
2478 | if (old != new) | |
2479 | (void) atomic_cas_64(&spa->spa_last_io, old, new); | |
2480 | } | |
2481 | ||
b128c09f BB |
2482 | align = 1ULL << vd->vdev_top->vdev_ashift; |
2483 | ||
178e73b3 | 2484 | if (P2PHASE(zio->io_size, align) != 0) { |
34dc7c2f BB |
2485 | uint64_t asize = P2ROUNDUP(zio->io_size, align); |
2486 | char *abuf = zio_buf_alloc(asize); | |
178e73b3 | 2487 | ASSERT(vd == vd->vdev_top); |
34dc7c2f BB |
2488 | if (zio->io_type == ZIO_TYPE_WRITE) { |
2489 | bcopy(zio->io_data, abuf, zio->io_size); | |
2490 | bzero(abuf + zio->io_size, asize - zio->io_size); | |
2491 | } | |
b128c09f | 2492 | zio_push_transform(zio, abuf, asize, asize, zio_subblock); |
34dc7c2f BB |
2493 | } |
2494 | ||
2495 | ASSERT(P2PHASE(zio->io_offset, align) == 0); | |
178e73b3 | 2496 | ASSERT(P2PHASE(zio->io_size, align) == 0); |
572e2857 | 2497 | VERIFY(zio->io_type != ZIO_TYPE_WRITE || spa_writeable(spa)); |
fb5f0bc8 BB |
2498 | |
2499 | /* | |
2500 | * If this is a repair I/O, and there's no self-healing involved -- | |
2501 | * that is, we're just resilvering what we expect to resilver -- | |
2502 | * then don't do the I/O unless zio's txg is actually in vd's DTL. | |
2503 | * This prevents spurious resilvering with nested replication. | |
2504 | * For example, given a mirror of mirrors, (A+B)+(C+D), if only | |
2505 | * A is out of date, we'll read from C+D, then use the data to | |
2506 | * resilver A+B -- but we don't actually want to resilver B, just A. | |
2507 | * The top-level mirror has no way to know this, so instead we just | |
2508 | * discard unnecessary repairs as we work our way down the vdev tree. | |
2509 | * The same logic applies to any form of nested replication: | |
2510 | * ditto + mirror, RAID-Z + replacing, etc. This covers them all. | |
2511 | */ | |
2512 | if ((zio->io_flags & ZIO_FLAG_IO_REPAIR) && | |
2513 | !(zio->io_flags & ZIO_FLAG_SELF_HEAL) && | |
2514 | zio->io_txg != 0 && /* not a delegated i/o */ | |
2515 | !vdev_dtl_contains(vd, DTL_PARTIAL, zio->io_txg, 1)) { | |
2516 | ASSERT(zio->io_type == ZIO_TYPE_WRITE); | |
fb5f0bc8 BB |
2517 | zio_vdev_io_bypass(zio); |
2518 | return (ZIO_PIPELINE_CONTINUE); | |
2519 | } | |
34dc7c2f | 2520 | |
b128c09f BB |
2521 | if (vd->vdev_ops->vdev_op_leaf && |
2522 | (zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE)) { | |
2523 | ||
2524 | if (zio->io_type == ZIO_TYPE_READ && vdev_cache_read(zio) == 0) | |
d164b209 | 2525 | return (ZIO_PIPELINE_CONTINUE); |
b128c09f BB |
2526 | |
2527 | if ((zio = vdev_queue_io(zio)) == NULL) | |
2528 | return (ZIO_PIPELINE_STOP); | |
2529 | ||
2530 | if (!vdev_accessible(vd, zio)) { | |
2531 | zio->io_error = ENXIO; | |
2532 | zio_interrupt(zio); | |
2533 | return (ZIO_PIPELINE_STOP); | |
2534 | } | |
b128c09f BB |
2535 | } |
2536 | ||
34dc7c2f BB |
2537 | return (vd->vdev_ops->vdev_op_io_start(zio)); |
2538 | } | |
2539 | ||
2540 | static int | |
2541 | zio_vdev_io_done(zio_t *zio) | |
2542 | { | |
b128c09f BB |
2543 | vdev_t *vd = zio->io_vd; |
2544 | vdev_ops_t *ops = vd ? vd->vdev_ops : &vdev_mirror_ops; | |
2545 | boolean_t unexpected_error = B_FALSE; | |
34dc7c2f | 2546 | |
b128c09f BB |
2547 | if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) |
2548 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 2549 | |
b128c09f BB |
2550 | ASSERT(zio->io_type == ZIO_TYPE_READ || zio->io_type == ZIO_TYPE_WRITE); |
2551 | ||
2552 | if (vd != NULL && vd->vdev_ops->vdev_op_leaf) { | |
2553 | ||
2554 | vdev_queue_io_done(zio); | |
2555 | ||
2556 | if (zio->io_type == ZIO_TYPE_WRITE) | |
2557 | vdev_cache_write(zio); | |
2558 | ||
2559 | if (zio_injection_enabled && zio->io_error == 0) | |
9babb374 BB |
2560 | zio->io_error = zio_handle_device_injection(vd, |
2561 | zio, EIO); | |
b128c09f BB |
2562 | |
2563 | if (zio_injection_enabled && zio->io_error == 0) | |
2564 | zio->io_error = zio_handle_label_injection(zio, EIO); | |
2565 | ||
2566 | if (zio->io_error) { | |
2567 | if (!vdev_accessible(vd, zio)) { | |
2568 | zio->io_error = ENXIO; | |
2569 | } else { | |
2570 | unexpected_error = B_TRUE; | |
2571 | } | |
2572 | } | |
2573 | } | |
2574 | ||
2575 | ops->vdev_op_io_done(zio); | |
34dc7c2f | 2576 | |
b128c09f | 2577 | if (unexpected_error) |
d164b209 | 2578 | VERIFY(vdev_probe(vd, zio) == NULL); |
34dc7c2f | 2579 | |
b128c09f | 2580 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
2581 | } |
2582 | ||
428870ff BB |
2583 | /* |
2584 | * For non-raidz ZIOs, we can just copy aside the bad data read from the | |
2585 | * disk, and use that to finish the checksum ereport later. | |
2586 | */ | |
2587 | static void | |
2588 | zio_vsd_default_cksum_finish(zio_cksum_report_t *zcr, | |
2589 | const void *good_buf) | |
2590 | { | |
2591 | /* no processing needed */ | |
2592 | zfs_ereport_finish_checksum(zcr, good_buf, zcr->zcr_cbdata, B_FALSE); | |
2593 | } | |
2594 | ||
2595 | /*ARGSUSED*/ | |
2596 | void | |
2597 | zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr, void *ignored) | |
2598 | { | |
2599 | void *buf = zio_buf_alloc(zio->io_size); | |
2600 | ||
2601 | bcopy(zio->io_data, buf, zio->io_size); | |
2602 | ||
2603 | zcr->zcr_cbinfo = zio->io_size; | |
2604 | zcr->zcr_cbdata = buf; | |
2605 | zcr->zcr_finish = zio_vsd_default_cksum_finish; | |
2606 | zcr->zcr_free = zio_buf_free; | |
2607 | } | |
2608 | ||
34dc7c2f BB |
2609 | static int |
2610 | zio_vdev_io_assess(zio_t *zio) | |
2611 | { | |
2612 | vdev_t *vd = zio->io_vd; | |
b128c09f BB |
2613 | |
2614 | if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE)) | |
2615 | return (ZIO_PIPELINE_STOP); | |
2616 | ||
2617 | if (vd == NULL && !(zio->io_flags & ZIO_FLAG_CONFIG_WRITER)) | |
2618 | spa_config_exit(zio->io_spa, SCL_ZIO, zio); | |
2619 | ||
2620 | if (zio->io_vsd != NULL) { | |
428870ff | 2621 | zio->io_vsd_ops->vsd_free(zio); |
b128c09f | 2622 | zio->io_vsd = NULL; |
34dc7c2f BB |
2623 | } |
2624 | ||
b128c09f | 2625 | if (zio_injection_enabled && zio->io_error == 0) |
34dc7c2f BB |
2626 | zio->io_error = zio_handle_fault_injection(zio, EIO); |
2627 | ||
2628 | /* | |
2629 | * If the I/O failed, determine whether we should attempt to retry it. | |
428870ff BB |
2630 | * |
2631 | * On retry, we cut in line in the issue queue, since we don't want | |
2632 | * compression/checksumming/etc. work to prevent our (cheap) IO reissue. | |
34dc7c2f | 2633 | */ |
b128c09f BB |
2634 | if (zio->io_error && vd == NULL && |
2635 | !(zio->io_flags & (ZIO_FLAG_DONT_RETRY | ZIO_FLAG_IO_RETRY))) { | |
2636 | ASSERT(!(zio->io_flags & ZIO_FLAG_DONT_QUEUE)); /* not a leaf */ | |
2637 | ASSERT(!(zio->io_flags & ZIO_FLAG_IO_BYPASS)); /* not a leaf */ | |
34dc7c2f | 2638 | zio->io_error = 0; |
b128c09f BB |
2639 | zio->io_flags |= ZIO_FLAG_IO_RETRY | |
2640 | ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_AGGREGATE; | |
428870ff BB |
2641 | zio->io_stage = ZIO_STAGE_VDEV_IO_START >> 1; |
2642 | zio_taskq_dispatch(zio, ZIO_TASKQ_ISSUE, | |
2643 | zio_requeue_io_start_cut_in_line); | |
b128c09f | 2644 | return (ZIO_PIPELINE_STOP); |
34dc7c2f BB |
2645 | } |
2646 | ||
b128c09f BB |
2647 | /* |
2648 | * If we got an error on a leaf device, convert it to ENXIO | |
2649 | * if the device is not accessible at all. | |
2650 | */ | |
2651 | if (zio->io_error && vd != NULL && vd->vdev_ops->vdev_op_leaf && | |
2652 | !vdev_accessible(vd, zio)) | |
2653 | zio->io_error = ENXIO; | |
2654 | ||
2655 | /* | |
2656 | * If we can't write to an interior vdev (mirror or RAID-Z), | |
2657 | * set vdev_cant_write so that we stop trying to allocate from it. | |
2658 | */ | |
2659 | if (zio->io_error == ENXIO && zio->io_type == ZIO_TYPE_WRITE && | |
2660 | vd != NULL && !vd->vdev_ops->vdev_op_leaf) | |
2661 | vd->vdev_cant_write = B_TRUE; | |
2662 | ||
2663 | if (zio->io_error) | |
2664 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
2665 | ||
34dc7c2f BB |
2666 | return (ZIO_PIPELINE_CONTINUE); |
2667 | } | |
2668 | ||
2669 | void | |
2670 | zio_vdev_io_reissue(zio_t *zio) | |
2671 | { | |
2672 | ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); | |
2673 | ASSERT(zio->io_error == 0); | |
2674 | ||
428870ff | 2675 | zio->io_stage >>= 1; |
34dc7c2f BB |
2676 | } |
2677 | ||
2678 | void | |
2679 | zio_vdev_io_redone(zio_t *zio) | |
2680 | { | |
2681 | ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_DONE); | |
2682 | ||
428870ff | 2683 | zio->io_stage >>= 1; |
34dc7c2f BB |
2684 | } |
2685 | ||
2686 | void | |
2687 | zio_vdev_io_bypass(zio_t *zio) | |
2688 | { | |
2689 | ASSERT(zio->io_stage == ZIO_STAGE_VDEV_IO_START); | |
2690 | ASSERT(zio->io_error == 0); | |
2691 | ||
2692 | zio->io_flags |= ZIO_FLAG_IO_BYPASS; | |
428870ff | 2693 | zio->io_stage = ZIO_STAGE_VDEV_IO_ASSESS >> 1; |
34dc7c2f BB |
2694 | } |
2695 | ||
2696 | /* | |
2697 | * ========================================================================== | |
2698 | * Generate and verify checksums | |
2699 | * ========================================================================== | |
2700 | */ | |
2701 | static int | |
2702 | zio_checksum_generate(zio_t *zio) | |
2703 | { | |
34dc7c2f | 2704 | blkptr_t *bp = zio->io_bp; |
b128c09f | 2705 | enum zio_checksum checksum; |
34dc7c2f | 2706 | |
b128c09f BB |
2707 | if (bp == NULL) { |
2708 | /* | |
2709 | * This is zio_write_phys(). | |
2710 | * We're either generating a label checksum, or none at all. | |
2711 | */ | |
2712 | checksum = zio->io_prop.zp_checksum; | |
34dc7c2f | 2713 | |
b128c09f BB |
2714 | if (checksum == ZIO_CHECKSUM_OFF) |
2715 | return (ZIO_PIPELINE_CONTINUE); | |
2716 | ||
2717 | ASSERT(checksum == ZIO_CHECKSUM_LABEL); | |
2718 | } else { | |
2719 | if (BP_IS_GANG(bp) && zio->io_child_type == ZIO_CHILD_GANG) { | |
2720 | ASSERT(!IO_IS_ALLOCATING(zio)); | |
2721 | checksum = ZIO_CHECKSUM_GANG_HEADER; | |
2722 | } else { | |
2723 | checksum = BP_GET_CHECKSUM(bp); | |
2724 | } | |
2725 | } | |
34dc7c2f | 2726 | |
b128c09f | 2727 | zio_checksum_compute(zio, checksum, zio->io_data, zio->io_size); |
34dc7c2f BB |
2728 | |
2729 | return (ZIO_PIPELINE_CONTINUE); | |
2730 | } | |
2731 | ||
2732 | static int | |
b128c09f | 2733 | zio_checksum_verify(zio_t *zio) |
34dc7c2f | 2734 | { |
428870ff | 2735 | zio_bad_cksum_t info; |
b128c09f BB |
2736 | blkptr_t *bp = zio->io_bp; |
2737 | int error; | |
34dc7c2f | 2738 | |
428870ff BB |
2739 | ASSERT(zio->io_vd != NULL); |
2740 | ||
b128c09f BB |
2741 | if (bp == NULL) { |
2742 | /* | |
2743 | * This is zio_read_phys(). | |
2744 | * We're either verifying a label checksum, or nothing at all. | |
2745 | */ | |
2746 | if (zio->io_prop.zp_checksum == ZIO_CHECKSUM_OFF) | |
2747 | return (ZIO_PIPELINE_CONTINUE); | |
34dc7c2f | 2748 | |
b128c09f BB |
2749 | ASSERT(zio->io_prop.zp_checksum == ZIO_CHECKSUM_LABEL); |
2750 | } | |
34dc7c2f | 2751 | |
428870ff | 2752 | if ((error = zio_checksum_error(zio, &info)) != 0) { |
b128c09f BB |
2753 | zio->io_error = error; |
2754 | if (!(zio->io_flags & ZIO_FLAG_SPECULATIVE)) { | |
428870ff BB |
2755 | zfs_ereport_start_checksum(zio->io_spa, |
2756 | zio->io_vd, zio, zio->io_offset, | |
2757 | zio->io_size, NULL, &info); | |
b128c09f | 2758 | } |
34dc7c2f BB |
2759 | } |
2760 | ||
2761 | return (ZIO_PIPELINE_CONTINUE); | |
2762 | } | |
2763 | ||
2764 | /* | |
2765 | * Called by RAID-Z to ensure we don't compute the checksum twice. | |
2766 | */ | |
2767 | void | |
2768 | zio_checksum_verified(zio_t *zio) | |
2769 | { | |
428870ff | 2770 | zio->io_pipeline &= ~ZIO_STAGE_CHECKSUM_VERIFY; |
34dc7c2f BB |
2771 | } |
2772 | ||
2773 | /* | |
b128c09f BB |
2774 | * ========================================================================== |
2775 | * Error rank. Error are ranked in the order 0, ENXIO, ECKSUM, EIO, other. | |
2776 | * An error of 0 indictes success. ENXIO indicates whole-device failure, | |
2777 | * which may be transient (e.g. unplugged) or permament. ECKSUM and EIO | |
2778 | * indicate errors that are specific to one I/O, and most likely permanent. | |
2779 | * Any other error is presumed to be worse because we weren't expecting it. | |
2780 | * ========================================================================== | |
34dc7c2f | 2781 | */ |
b128c09f BB |
2782 | int |
2783 | zio_worst_error(int e1, int e2) | |
34dc7c2f | 2784 | { |
b128c09f BB |
2785 | static int zio_error_rank[] = { 0, ENXIO, ECKSUM, EIO }; |
2786 | int r1, r2; | |
2787 | ||
2788 | for (r1 = 0; r1 < sizeof (zio_error_rank) / sizeof (int); r1++) | |
2789 | if (e1 == zio_error_rank[r1]) | |
2790 | break; | |
34dc7c2f | 2791 | |
b128c09f BB |
2792 | for (r2 = 0; r2 < sizeof (zio_error_rank) / sizeof (int); r2++) |
2793 | if (e2 == zio_error_rank[r2]) | |
2794 | break; | |
2795 | ||
2796 | return (r1 > r2 ? e1 : e2); | |
34dc7c2f BB |
2797 | } |
2798 | ||
2799 | /* | |
2800 | * ========================================================================== | |
b128c09f | 2801 | * I/O completion |
34dc7c2f BB |
2802 | * ========================================================================== |
2803 | */ | |
b128c09f BB |
2804 | static int |
2805 | zio_ready(zio_t *zio) | |
34dc7c2f | 2806 | { |
b128c09f | 2807 | blkptr_t *bp = zio->io_bp; |
d164b209 | 2808 | zio_t *pio, *pio_next; |
34dc7c2f | 2809 | |
428870ff BB |
2810 | if (zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_READY) || |
2811 | zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_READY)) | |
9babb374 | 2812 | return (ZIO_PIPELINE_STOP); |
34dc7c2f | 2813 | |
9babb374 | 2814 | if (zio->io_ready) { |
b128c09f BB |
2815 | ASSERT(IO_IS_ALLOCATING(zio)); |
2816 | ASSERT(bp->blk_birth == zio->io_txg || BP_IS_HOLE(bp)); | |
2817 | ASSERT(zio->io_children[ZIO_CHILD_GANG][ZIO_WAIT_READY] == 0); | |
34dc7c2f | 2818 | |
b128c09f BB |
2819 | zio->io_ready(zio); |
2820 | } | |
34dc7c2f | 2821 | |
b128c09f BB |
2822 | if (bp != NULL && bp != &zio->io_bp_copy) |
2823 | zio->io_bp_copy = *bp; | |
34dc7c2f | 2824 | |
b128c09f BB |
2825 | if (zio->io_error) |
2826 | zio->io_pipeline = ZIO_INTERLOCK_PIPELINE; | |
34dc7c2f | 2827 | |
d164b209 BB |
2828 | mutex_enter(&zio->io_lock); |
2829 | zio->io_state[ZIO_WAIT_READY] = 1; | |
2830 | pio = zio_walk_parents(zio); | |
2831 | mutex_exit(&zio->io_lock); | |
2832 | ||
2833 | /* | |
2834 | * As we notify zio's parents, new parents could be added. | |
2835 | * New parents go to the head of zio's io_parent_list, however, | |
2836 | * so we will (correctly) not notify them. The remainder of zio's | |
2837 | * io_parent_list, from 'pio_next' onward, cannot change because | |
2838 | * all parents must wait for us to be done before they can be done. | |
2839 | */ | |
2840 | for (; pio != NULL; pio = pio_next) { | |
2841 | pio_next = zio_walk_parents(zio); | |
b128c09f | 2842 | zio_notify_parent(pio, zio, ZIO_WAIT_READY); |
d164b209 | 2843 | } |
34dc7c2f | 2844 | |
428870ff BB |
2845 | if (zio->io_flags & ZIO_FLAG_NODATA) { |
2846 | if (BP_IS_GANG(bp)) { | |
2847 | zio->io_flags &= ~ZIO_FLAG_NODATA; | |
2848 | } else { | |
2849 | ASSERT((uintptr_t)zio->io_data < SPA_MAXBLOCKSIZE); | |
2850 | zio->io_pipeline &= ~ZIO_VDEV_IO_STAGES; | |
2851 | } | |
2852 | } | |
2853 | ||
2854 | if (zio_injection_enabled && | |
2855 | zio->io_spa->spa_syncing_txg == zio->io_txg) | |
2856 | zio_handle_ignored_writes(zio); | |
2857 | ||
b128c09f | 2858 | return (ZIO_PIPELINE_CONTINUE); |
34dc7c2f BB |
2859 | } |
2860 | ||
b128c09f BB |
2861 | static int |
2862 | zio_done(zio_t *zio) | |
34dc7c2f | 2863 | { |
d164b209 | 2864 | zio_t *pio, *pio_next; |
d6320ddb | 2865 | int c, w; |
34dc7c2f | 2866 | |
b128c09f | 2867 | /* |
9babb374 | 2868 | * If our children haven't all completed, |
b128c09f BB |
2869 | * wait for them and then repeat this pipeline stage. |
2870 | */ | |
2871 | if (zio_wait_for_children(zio, ZIO_CHILD_VDEV, ZIO_WAIT_DONE) || | |
2872 | zio_wait_for_children(zio, ZIO_CHILD_GANG, ZIO_WAIT_DONE) || | |
428870ff | 2873 | zio_wait_for_children(zio, ZIO_CHILD_DDT, ZIO_WAIT_DONE) || |
b128c09f BB |
2874 | zio_wait_for_children(zio, ZIO_CHILD_LOGICAL, ZIO_WAIT_DONE)) |
2875 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f | 2876 | |
d6320ddb BB |
2877 | for (c = 0; c < ZIO_CHILD_TYPES; c++) |
2878 | for (w = 0; w < ZIO_WAIT_TYPES; w++) | |
b128c09f BB |
2879 | ASSERT(zio->io_children[c][w] == 0); |
2880 | ||
c776b317 BB |
2881 | if (zio->io_bp != NULL) { |
2882 | ASSERT(zio->io_bp->blk_pad[0] == 0); | |
2883 | ASSERT(zio->io_bp->blk_pad[1] == 0); | |
2884 | ASSERT(bcmp(zio->io_bp, &zio->io_bp_copy, sizeof (blkptr_t)) == 0 || | |
2885 | (zio->io_bp == zio_unique_parent(zio)->io_bp)); | |
2886 | if (zio->io_type == ZIO_TYPE_WRITE && !BP_IS_HOLE(zio->io_bp) && | |
428870ff | 2887 | zio->io_bp_override == NULL && |
b128c09f | 2888 | !(zio->io_flags & ZIO_FLAG_IO_REPAIR)) { |
c776b317 BB |
2889 | ASSERT(!BP_SHOULD_BYTESWAP(zio->io_bp)); |
2890 | ASSERT3U(zio->io_prop.zp_copies, <=, BP_GET_NDVAS(zio->io_bp)); | |
2891 | ASSERT(BP_COUNT_GANG(zio->io_bp) == 0 || | |
2892 | (BP_COUNT_GANG(zio->io_bp) == BP_GET_NDVAS(zio->io_bp))); | |
b128c09f BB |
2893 | } |
2894 | } | |
2895 | ||
2896 | /* | |
428870ff | 2897 | * If there were child vdev/gang/ddt errors, they apply to us now. |
b128c09f BB |
2898 | */ |
2899 | zio_inherit_child_errors(zio, ZIO_CHILD_VDEV); | |
2900 | zio_inherit_child_errors(zio, ZIO_CHILD_GANG); | |
428870ff BB |
2901 | zio_inherit_child_errors(zio, ZIO_CHILD_DDT); |
2902 | ||
2903 | /* | |
2904 | * If the I/O on the transformed data was successful, generate any | |
2905 | * checksum reports now while we still have the transformed data. | |
2906 | */ | |
2907 | if (zio->io_error == 0) { | |
2908 | while (zio->io_cksum_report != NULL) { | |
2909 | zio_cksum_report_t *zcr = zio->io_cksum_report; | |
2910 | uint64_t align = zcr->zcr_align; | |
c776b317 | 2911 | uint64_t asize = P2ROUNDUP(zio->io_size, align); |
428870ff BB |
2912 | char *abuf = zio->io_data; |
2913 | ||
c776b317 | 2914 | if (asize != zio->io_size) { |
428870ff | 2915 | abuf = zio_buf_alloc(asize); |
c776b317 BB |
2916 | bcopy(zio->io_data, abuf, zio->io_size); |
2917 | bzero(abuf + zio->io_size, asize - zio->io_size); | |
428870ff BB |
2918 | } |
2919 | ||
2920 | zio->io_cksum_report = zcr->zcr_next; | |
2921 | zcr->zcr_next = NULL; | |
2922 | zcr->zcr_finish(zcr, abuf); | |
2923 | zfs_ereport_free_checksum(zcr); | |
2924 | ||
c776b317 | 2925 | if (asize != zio->io_size) |
428870ff BB |
2926 | zio_buf_free(abuf, asize); |
2927 | } | |
2928 | } | |
b128c09f BB |
2929 | |
2930 | zio_pop_transforms(zio); /* note: may set zio->io_error */ | |
2931 | ||
c776b317 | 2932 | vdev_stat_update(zio, zio->io_size); |
b128c09f | 2933 | |
a69052be | 2934 | /* |
cc92e9d0 | 2935 | * If this I/O is attached to a particular vdev is slow, exceeding |
72f53c56 MJ |
2936 | * 30 seconds to complete, post an error described the I/O delay. |
2937 | * We ignore these errors if the device is currently unavailable. | |
a69052be | 2938 | */ |
cc92e9d0 | 2939 | if (zio->io_delay >= MSEC_TO_TICK(zio_delay_max)) { |
72f53c56 MJ |
2940 | if (zio->io_vd != NULL && !vdev_is_dead(zio->io_vd)) |
2941 | zfs_ereport_post(FM_EREPORT_ZFS_DELAY, zio->io_spa, | |
2942 | zio->io_vd, zio, 0, 0); | |
2943 | } | |
a69052be | 2944 | |
b128c09f BB |
2945 | if (zio->io_error) { |
2946 | /* | |
2947 | * If this I/O is attached to a particular vdev, | |
2948 | * generate an error message describing the I/O failure | |
2949 | * at the block level. We ignore these errors if the | |
2950 | * device is currently unavailable. | |
2951 | */ | |
c776b317 BB |
2952 | if (zio->io_error != ECKSUM && zio->io_vd != NULL && |
2953 | !vdev_is_dead(zio->io_vd)) | |
2954 | zfs_ereport_post(FM_EREPORT_ZFS_IO, zio->io_spa, | |
2955 | zio->io_vd, zio, 0, 0); | |
34dc7c2f | 2956 | |
428870ff BB |
2957 | if ((zio->io_error == EIO || !(zio->io_flags & |
2958 | (ZIO_FLAG_SPECULATIVE | ZIO_FLAG_DONT_PROPAGATE))) && | |
c776b317 | 2959 | zio == zio->io_logical) { |
b128c09f BB |
2960 | /* |
2961 | * For logical I/O requests, tell the SPA to log the | |
2962 | * error and generate a logical data ereport. | |
2963 | */ | |
c776b317 BB |
2964 | spa_log_error(zio->io_spa, zio); |
2965 | zfs_ereport_post(FM_EREPORT_ZFS_DATA, zio->io_spa, NULL, zio, | |
b128c09f BB |
2966 | 0, 0); |
2967 | } | |
2968 | } | |
34dc7c2f | 2969 | |
c776b317 | 2970 | if (zio->io_error && zio == zio->io_logical) { |
b128c09f BB |
2971 | /* |
2972 | * Determine whether zio should be reexecuted. This will | |
2973 | * propagate all the way to the root via zio_notify_parent(). | |
2974 | */ | |
c776b317 | 2975 | ASSERT(zio->io_vd == NULL && zio->io_bp != NULL); |
428870ff | 2976 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); |
b128c09f | 2977 | |
428870ff BB |
2978 | if (IO_IS_ALLOCATING(zio) && |
2979 | !(zio->io_flags & ZIO_FLAG_CANFAIL)) { | |
b128c09f BB |
2980 | if (zio->io_error != ENOSPC) |
2981 | zio->io_reexecute |= ZIO_REEXECUTE_NOW; | |
2982 | else | |
2983 | zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; | |
428870ff | 2984 | } |
b128c09f BB |
2985 | |
2986 | if ((zio->io_type == ZIO_TYPE_READ || | |
2987 | zio->io_type == ZIO_TYPE_FREE) && | |
572e2857 | 2988 | !(zio->io_flags & ZIO_FLAG_SCAN_THREAD) && |
b128c09f | 2989 | zio->io_error == ENXIO && |
c776b317 BB |
2990 | spa_load_state(zio->io_spa) == SPA_LOAD_NONE && |
2991 | spa_get_failmode(zio->io_spa) != ZIO_FAILURE_MODE_CONTINUE) | |
b128c09f BB |
2992 | zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; |
2993 | ||
2994 | if (!(zio->io_flags & ZIO_FLAG_CANFAIL) && !zio->io_reexecute) | |
2995 | zio->io_reexecute |= ZIO_REEXECUTE_SUSPEND; | |
428870ff BB |
2996 | |
2997 | /* | |
2998 | * Here is a possibly good place to attempt to do | |
2999 | * either combinatorial reconstruction or error correction | |
3000 | * based on checksums. It also might be a good place | |
3001 | * to send out preliminary ereports before we suspend | |
3002 | * processing. | |
3003 | */ | |
34dc7c2f BB |
3004 | } |
3005 | ||
3006 | /* | |
b128c09f BB |
3007 | * If there were logical child errors, they apply to us now. |
3008 | * We defer this until now to avoid conflating logical child | |
3009 | * errors with errors that happened to the zio itself when | |
3010 | * updating vdev stats and reporting FMA events above. | |
34dc7c2f | 3011 | */ |
b128c09f | 3012 | zio_inherit_child_errors(zio, ZIO_CHILD_LOGICAL); |
34dc7c2f | 3013 | |
428870ff BB |
3014 | if ((zio->io_error || zio->io_reexecute) && |
3015 | IO_IS_ALLOCATING(zio) && zio->io_gang_leader == zio && | |
3016 | !(zio->io_flags & ZIO_FLAG_IO_REWRITE)) | |
c776b317 | 3017 | zio_dva_unallocate(zio, zio->io_gang_tree, zio->io_bp); |
9babb374 BB |
3018 | |
3019 | zio_gang_tree_free(&zio->io_gang_tree); | |
3020 | ||
3021 | /* | |
3022 | * Godfather I/Os should never suspend. | |
3023 | */ | |
3024 | if ((zio->io_flags & ZIO_FLAG_GODFATHER) && | |
3025 | (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) | |
3026 | zio->io_reexecute = 0; | |
3027 | ||
b128c09f BB |
3028 | if (zio->io_reexecute) { |
3029 | /* | |
3030 | * This is a logical I/O that wants to reexecute. | |
3031 | * | |
3032 | * Reexecute is top-down. When an i/o fails, if it's not | |
3033 | * the root, it simply notifies its parent and sticks around. | |
3034 | * The parent, seeing that it still has children in zio_done(), | |
3035 | * does the same. This percolates all the way up to the root. | |
3036 | * The root i/o will reexecute or suspend the entire tree. | |
3037 | * | |
3038 | * This approach ensures that zio_reexecute() honors | |
3039 | * all the original i/o dependency relationships, e.g. | |
3040 | * parents not executing until children are ready. | |
3041 | */ | |
3042 | ASSERT(zio->io_child_type == ZIO_CHILD_LOGICAL); | |
34dc7c2f | 3043 | |
9babb374 | 3044 | zio->io_gang_leader = NULL; |
b128c09f | 3045 | |
d164b209 BB |
3046 | mutex_enter(&zio->io_lock); |
3047 | zio->io_state[ZIO_WAIT_DONE] = 1; | |
3048 | mutex_exit(&zio->io_lock); | |
3049 | ||
9babb374 BB |
3050 | /* |
3051 | * "The Godfather" I/O monitors its children but is | |
3052 | * not a true parent to them. It will track them through | |
3053 | * the pipeline but severs its ties whenever they get into | |
3054 | * trouble (e.g. suspended). This allows "The Godfather" | |
3055 | * I/O to return status without blocking. | |
3056 | */ | |
3057 | for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { | |
3058 | zio_link_t *zl = zio->io_walk_link; | |
3059 | pio_next = zio_walk_parents(zio); | |
3060 | ||
3061 | if ((pio->io_flags & ZIO_FLAG_GODFATHER) && | |
3062 | (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND)) { | |
3063 | zio_remove_child(pio, zio, zl); | |
3064 | zio_notify_parent(pio, zio, ZIO_WAIT_DONE); | |
3065 | } | |
3066 | } | |
3067 | ||
d164b209 | 3068 | if ((pio = zio_unique_parent(zio)) != NULL) { |
b128c09f BB |
3069 | /* |
3070 | * We're not a root i/o, so there's nothing to do | |
3071 | * but notify our parent. Don't propagate errors | |
3072 | * upward since we haven't permanently failed yet. | |
3073 | */ | |
9babb374 | 3074 | ASSERT(!(zio->io_flags & ZIO_FLAG_GODFATHER)); |
b128c09f BB |
3075 | zio->io_flags |= ZIO_FLAG_DONT_PROPAGATE; |
3076 | zio_notify_parent(pio, zio, ZIO_WAIT_DONE); | |
3077 | } else if (zio->io_reexecute & ZIO_REEXECUTE_SUSPEND) { | |
3078 | /* | |
3079 | * We'd fail again if we reexecuted now, so suspend | |
3080 | * until conditions improve (e.g. device comes online). | |
3081 | */ | |
c776b317 | 3082 | zio_suspend(zio->io_spa, zio); |
b128c09f BB |
3083 | } else { |
3084 | /* | |
3085 | * Reexecution is potentially a huge amount of work. | |
3086 | * Hand it off to the otherwise-unused claim taskq. | |
3087 | */ | |
a38718a6 | 3088 | ASSERT(taskq_empty_ent(&zio->io_tqent)); |
7ef5e54e AL |
3089 | spa_taskq_dispatch_ent(zio->io_spa, |
3090 | ZIO_TYPE_CLAIM, ZIO_TASKQ_ISSUE, | |
a38718a6 GA |
3091 | (task_func_t *)zio_reexecute, zio, 0, |
3092 | &zio->io_tqent); | |
b128c09f BB |
3093 | } |
3094 | return (ZIO_PIPELINE_STOP); | |
34dc7c2f BB |
3095 | } |
3096 | ||
428870ff | 3097 | ASSERT(zio->io_child_count == 0); |
b128c09f BB |
3098 | ASSERT(zio->io_reexecute == 0); |
3099 | ASSERT(zio->io_error == 0 || (zio->io_flags & ZIO_FLAG_CANFAIL)); | |
34dc7c2f | 3100 | |
428870ff BB |
3101 | /* |
3102 | * Report any checksum errors, since the I/O is complete. | |
3103 | */ | |
3104 | while (zio->io_cksum_report != NULL) { | |
3105 | zio_cksum_report_t *zcr = zio->io_cksum_report; | |
3106 | zio->io_cksum_report = zcr->zcr_next; | |
3107 | zcr->zcr_next = NULL; | |
3108 | zcr->zcr_finish(zcr, NULL); | |
3109 | zfs_ereport_free_checksum(zcr); | |
3110 | } | |
3111 | ||
920dd524 ED |
3112 | if (zio->io_flags & ZIO_FLAG_FASTWRITE && zio->io_bp && |
3113 | !BP_IS_HOLE(zio->io_bp)) { | |
3114 | metaslab_fastwrite_unmark(zio->io_spa, zio->io_bp); | |
3115 | } | |
3116 | ||
d164b209 BB |
3117 | /* |
3118 | * It is the responsibility of the done callback to ensure that this | |
3119 | * particular zio is no longer discoverable for adoption, and as | |
3120 | * such, cannot acquire any new parents. | |
3121 | */ | |
b128c09f BB |
3122 | if (zio->io_done) |
3123 | zio->io_done(zio); | |
34dc7c2f | 3124 | |
d164b209 BB |
3125 | mutex_enter(&zio->io_lock); |
3126 | zio->io_state[ZIO_WAIT_DONE] = 1; | |
3127 | mutex_exit(&zio->io_lock); | |
34dc7c2f | 3128 | |
d164b209 BB |
3129 | for (pio = zio_walk_parents(zio); pio != NULL; pio = pio_next) { |
3130 | zio_link_t *zl = zio->io_walk_link; | |
3131 | pio_next = zio_walk_parents(zio); | |
3132 | zio_remove_child(pio, zio, zl); | |
b128c09f BB |
3133 | zio_notify_parent(pio, zio, ZIO_WAIT_DONE); |
3134 | } | |
34dc7c2f | 3135 | |
b128c09f BB |
3136 | if (zio->io_waiter != NULL) { |
3137 | mutex_enter(&zio->io_lock); | |
3138 | zio->io_executor = NULL; | |
3139 | cv_broadcast(&zio->io_cv); | |
3140 | mutex_exit(&zio->io_lock); | |
3141 | } else { | |
3142 | zio_destroy(zio); | |
3143 | } | |
34dc7c2f | 3144 | |
b128c09f | 3145 | return (ZIO_PIPELINE_STOP); |
34dc7c2f BB |
3146 | } |
3147 | ||
3148 | /* | |
b128c09f BB |
3149 | * ========================================================================== |
3150 | * I/O pipeline definition | |
3151 | * ========================================================================== | |
34dc7c2f | 3152 | */ |
428870ff | 3153 | static zio_pipe_stage_t *zio_pipeline[] = { |
b128c09f | 3154 | NULL, |
b128c09f | 3155 | zio_read_bp_init, |
428870ff BB |
3156 | zio_free_bp_init, |
3157 | zio_issue_async, | |
b128c09f BB |
3158 | zio_write_bp_init, |
3159 | zio_checksum_generate, | |
428870ff BB |
3160 | zio_ddt_read_start, |
3161 | zio_ddt_read_done, | |
3162 | zio_ddt_write, | |
3163 | zio_ddt_free, | |
b128c09f BB |
3164 | zio_gang_assemble, |
3165 | zio_gang_issue, | |
3166 | zio_dva_allocate, | |
3167 | zio_dva_free, | |
3168 | zio_dva_claim, | |
3169 | zio_ready, | |
3170 | zio_vdev_io_start, | |
3171 | zio_vdev_io_done, | |
3172 | zio_vdev_io_assess, | |
3173 | zio_checksum_verify, | |
3174 | zio_done | |
3175 | }; | |
c28b2279 | 3176 | |
9ae529ec CS |
3177 | /* dnp is the dnode for zb1->zb_object */ |
3178 | boolean_t | |
3179 | zbookmark_is_before(const dnode_phys_t *dnp, const zbookmark_t *zb1, | |
3180 | const zbookmark_t *zb2) | |
3181 | { | |
3182 | uint64_t zb1nextL0, zb2thisobj; | |
3183 | ||
3184 | ASSERT(zb1->zb_objset == zb2->zb_objset); | |
3185 | ASSERT(zb2->zb_level == 0); | |
3186 | ||
3187 | /* | |
3188 | * A bookmark in the deadlist is considered to be after | |
3189 | * everything else. | |
3190 | */ | |
3191 | if (zb2->zb_object == DMU_DEADLIST_OBJECT) | |
3192 | return (B_TRUE); | |
3193 | ||
3194 | /* The objset_phys_t isn't before anything. */ | |
3195 | if (dnp == NULL) | |
3196 | return (B_FALSE); | |
3197 | ||
3198 | zb1nextL0 = (zb1->zb_blkid + 1) << | |
3199 | ((zb1->zb_level) * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT)); | |
3200 | ||
3201 | zb2thisobj = zb2->zb_object ? zb2->zb_object : | |
3202 | zb2->zb_blkid << (DNODE_BLOCK_SHIFT - DNODE_SHIFT); | |
3203 | ||
3204 | if (zb1->zb_object == DMU_META_DNODE_OBJECT) { | |
3205 | uint64_t nextobj = zb1nextL0 * | |
3206 | (dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT) >> DNODE_SHIFT; | |
3207 | return (nextobj <= zb2thisobj); | |
3208 | } | |
3209 | ||
3210 | if (zb1->zb_object < zb2thisobj) | |
3211 | return (B_TRUE); | |
3212 | if (zb1->zb_object > zb2thisobj) | |
3213 | return (B_FALSE); | |
3214 | if (zb2->zb_object == DMU_META_DNODE_OBJECT) | |
3215 | return (B_FALSE); | |
3216 | return (zb1nextL0 <= zb2->zb_blkid); | |
3217 | } | |
3218 | ||
c28b2279 BB |
3219 | #if defined(_KERNEL) && defined(HAVE_SPL) |
3220 | /* Fault injection */ | |
3221 | EXPORT_SYMBOL(zio_injection_enabled); | |
3222 | EXPORT_SYMBOL(zio_inject_fault); | |
3223 | EXPORT_SYMBOL(zio_inject_list_next); | |
3224 | EXPORT_SYMBOL(zio_clear_fault); | |
3225 | EXPORT_SYMBOL(zio_handle_fault_injection); | |
3226 | EXPORT_SYMBOL(zio_handle_device_injection); | |
3227 | EXPORT_SYMBOL(zio_handle_label_injection); | |
3228 | EXPORT_SYMBOL(zio_priority_table); | |
3229 | EXPORT_SYMBOL(zio_type_name); | |
3230 | ||
3231 | module_param(zio_bulk_flags, int, 0644); | |
3232 | MODULE_PARM_DESC(zio_bulk_flags, "Additional flags to pass to bulk buffers"); | |
a69052be BB |
3233 | |
3234 | module_param(zio_delay_max, int, 0644); | |
c409e464 BB |
3235 | MODULE_PARM_DESC(zio_delay_max, "Max zio millisec delay before posting event"); |
3236 | ||
3237 | module_param(zio_requeue_io_start_cut_in_line, int, 0644); | |
3238 | MODULE_PARM_DESC(zio_requeue_io_start_cut_in_line, "Prioritize requeued I/O"); | |
29dee3ee CP |
3239 | |
3240 | module_param(zfs_sync_pass_deferred_free, int, 0644); | |
3241 | MODULE_PARM_DESC(zfs_sync_pass_deferred_free, | |
3242 | "defer frees starting in this pass"); | |
3243 | ||
3244 | module_param(zfs_sync_pass_dont_compress, int, 0644); | |
3245 | MODULE_PARM_DESC(zfs_sync_pass_dont_compress, | |
3246 | "don't compress starting in this pass"); | |
3247 | ||
3248 | module_param(zfs_sync_pass_rewrite, int, 0644); | |
3249 | MODULE_PARM_DESC(zfs_sync_pass_rewrite, | |
3250 | "rewrite new bps starting in this pass"); | |
c28b2279 | 3251 | #endif |