]>
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 | ||
22 | /* | |
428870ff | 23 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. |
a38718a6 | 24 | * Copyright 2011 Nexenta Systems, Inc. All rights reserved. |
492f64e9 | 25 | * Copyright (c) 2012, 2018 by Delphix. All rights reserved. |
9759c60f | 26 | * Copyright (c) 2013 by Saso Kiselkov. All rights reserved. |
e550644f BB |
27 | * Copyright (c) 2013, Joyent, Inc. All rights reserved. |
28 | * Copyright 2016 Toomas Soome <tsoome@me.com> | |
a38718a6 | 29 | */ |
34dc7c2f BB |
30 | |
31 | #ifndef _ZIO_H | |
32 | #define _ZIO_H | |
33 | ||
fcff0f35 | 34 | #include <sys/zio_priority.h> |
34dc7c2f BB |
35 | #include <sys/zfs_context.h> |
36 | #include <sys/spa.h> | |
37 | #include <sys/txg.h> | |
38 | #include <sys/avl.h> | |
34dc7c2f BB |
39 | #include <sys/fs/zfs.h> |
40 | #include <sys/zio_impl.h> | |
41 | ||
42 | #ifdef __cplusplus | |
43 | extern "C" { | |
44 | #endif | |
45 | ||
428870ff BB |
46 | /* |
47 | * Embedded checksum | |
48 | */ | |
49 | #define ZEC_MAGIC 0x210da7ab10c7a11ULL | |
34dc7c2f | 50 | |
428870ff BB |
51 | typedef struct zio_eck { |
52 | uint64_t zec_magic; /* for validation, endianness */ | |
53 | zio_cksum_t zec_cksum; /* 256-bit checksum */ | |
54 | } zio_eck_t; | |
34dc7c2f BB |
55 | |
56 | /* | |
57 | * Gang block headers are self-checksumming and contain an array | |
58 | * of block pointers. | |
59 | */ | |
60 | #define SPA_GANGBLOCKSIZE SPA_MINBLOCKSIZE | |
61 | #define SPA_GBH_NBLKPTRS ((SPA_GANGBLOCKSIZE - \ | |
428870ff | 62 | sizeof (zio_eck_t)) / sizeof (blkptr_t)) |
34dc7c2f | 63 | #define SPA_GBH_FILLER ((SPA_GANGBLOCKSIZE - \ |
428870ff | 64 | sizeof (zio_eck_t) - \ |
34dc7c2f BB |
65 | (SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\ |
66 | sizeof (uint64_t)) | |
67 | ||
34dc7c2f BB |
68 | typedef struct zio_gbh { |
69 | blkptr_t zg_blkptr[SPA_GBH_NBLKPTRS]; | |
70 | uint64_t zg_filler[SPA_GBH_FILLER]; | |
428870ff | 71 | zio_eck_t zg_tail; |
34dc7c2f BB |
72 | } zio_gbh_phys_t; |
73 | ||
74 | enum zio_checksum { | |
75 | ZIO_CHECKSUM_INHERIT = 0, | |
76 | ZIO_CHECKSUM_ON, | |
77 | ZIO_CHECKSUM_OFF, | |
78 | ZIO_CHECKSUM_LABEL, | |
79 | ZIO_CHECKSUM_GANG_HEADER, | |
80 | ZIO_CHECKSUM_ZILOG, | |
81 | ZIO_CHECKSUM_FLETCHER_2, | |
82 | ZIO_CHECKSUM_FLETCHER_4, | |
83 | ZIO_CHECKSUM_SHA256, | |
428870ff | 84 | ZIO_CHECKSUM_ZILOG2, |
3c67d83a TH |
85 | ZIO_CHECKSUM_NOPARITY, |
86 | ZIO_CHECKSUM_SHA512, | |
87 | ZIO_CHECKSUM_SKEIN, | |
88 | ZIO_CHECKSUM_EDONR, | |
34dc7c2f BB |
89 | ZIO_CHECKSUM_FUNCTIONS |
90 | }; | |
91 | ||
9b67f605 MA |
92 | /* |
93 | * The number of "legacy" compression functions which can be set on individual | |
94 | * objects. | |
95 | */ | |
96 | #define ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2 | |
97 | ||
9babb374 | 98 | #define ZIO_CHECKSUM_ON_VALUE ZIO_CHECKSUM_FLETCHER_4 |
34dc7c2f BB |
99 | #define ZIO_CHECKSUM_DEFAULT ZIO_CHECKSUM_ON |
100 | ||
428870ff BB |
101 | #define ZIO_CHECKSUM_MASK 0xffULL |
102 | #define ZIO_CHECKSUM_VERIFY (1 << 8) | |
103 | ||
104 | #define ZIO_DEDUPCHECKSUM ZIO_CHECKSUM_SHA256 | |
105 | #define ZIO_DEDUPDITTO_MIN 100 | |
106 | ||
b5256303 TC |
107 | /* supported encryption algorithms */ |
108 | enum zio_encrypt { | |
109 | ZIO_CRYPT_INHERIT = 0, | |
110 | ZIO_CRYPT_ON, | |
111 | ZIO_CRYPT_OFF, | |
112 | ZIO_CRYPT_AES_128_CCM, | |
113 | ZIO_CRYPT_AES_192_CCM, | |
114 | ZIO_CRYPT_AES_256_CCM, | |
115 | ZIO_CRYPT_AES_128_GCM, | |
116 | ZIO_CRYPT_AES_192_GCM, | |
117 | ZIO_CRYPT_AES_256_GCM, | |
118 | ZIO_CRYPT_FUNCTIONS | |
119 | }; | |
120 | ||
121 | #define ZIO_CRYPT_ON_VALUE ZIO_CRYPT_AES_256_CCM | |
122 | #define ZIO_CRYPT_DEFAULT ZIO_CRYPT_OFF | |
123 | ||
124 | /* macros defining encryption lengths */ | |
125 | #define ZIO_OBJSET_MAC_LEN 32 | |
126 | #define ZIO_DATA_IV_LEN 12 | |
127 | #define ZIO_DATA_SALT_LEN 8 | |
128 | #define ZIO_DATA_MAC_LEN 16 | |
129 | ||
9b67f605 MA |
130 | /* |
131 | * The number of "legacy" compression functions which can be set on individual | |
132 | * objects. | |
133 | */ | |
134 | #define ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4 | |
135 | ||
99197f03 JG |
136 | /* |
137 | * The meaning of "compress = on" selected by the compression features enabled | |
138 | * on a given pool. | |
139 | */ | |
140 | #define ZIO_COMPRESS_LEGACY_ON_VALUE ZIO_COMPRESS_LZJB | |
141 | #define ZIO_COMPRESS_LZ4_ON_VALUE ZIO_COMPRESS_LZ4 | |
142 | ||
143 | #define ZIO_COMPRESS_DEFAULT ZIO_COMPRESS_OFF | |
34dc7c2f | 144 | |
428870ff BB |
145 | #define BOOTFS_COMPRESS_VALID(compress) \ |
146 | ((compress) == ZIO_COMPRESS_LZJB || \ | |
9759c60f | 147 | (compress) == ZIO_COMPRESS_LZ4 || \ |
8aab1218 TS |
148 | (compress) == ZIO_COMPRESS_GZIP_1 || \ |
149 | (compress) == ZIO_COMPRESS_GZIP_2 || \ | |
150 | (compress) == ZIO_COMPRESS_GZIP_3 || \ | |
151 | (compress) == ZIO_COMPRESS_GZIP_4 || \ | |
152 | (compress) == ZIO_COMPRESS_GZIP_5 || \ | |
153 | (compress) == ZIO_COMPRESS_GZIP_6 || \ | |
154 | (compress) == ZIO_COMPRESS_GZIP_7 || \ | |
155 | (compress) == ZIO_COMPRESS_GZIP_8 || \ | |
156 | (compress) == ZIO_COMPRESS_GZIP_9 || \ | |
157 | (compress) == ZIO_COMPRESS_ZLE || \ | |
99197f03 | 158 | (compress) == ZIO_COMPRESS_ON || \ |
428870ff BB |
159 | (compress) == ZIO_COMPRESS_OFF) |
160 | ||
a69052be BB |
161 | /* |
162 | * Default Linux timeout for a sd device. | |
163 | */ | |
d1d7e268 | 164 | #define ZIO_DELAY_MAX (30 * MILLISEC) |
a69052be | 165 | |
34dc7c2f BB |
166 | #define ZIO_FAILURE_MODE_WAIT 0 |
167 | #define ZIO_FAILURE_MODE_CONTINUE 1 | |
168 | #define ZIO_FAILURE_MODE_PANIC 2 | |
169 | ||
cec3a0a1 OF |
170 | typedef enum zio_suspend_reason { |
171 | ZIO_SUSPEND_NONE = 0, | |
172 | ZIO_SUSPEND_IOERR, | |
173 | ZIO_SUSPEND_MMP, | |
174 | } zio_suspend_reason_t; | |
175 | ||
428870ff BB |
176 | enum zio_flag { |
177 | /* | |
178 | * Flags inherited by gang, ddt, and vdev children, | |
179 | * and that must be equal for two zios to aggregate | |
180 | */ | |
181 | ZIO_FLAG_DONT_AGGREGATE = 1 << 0, | |
182 | ZIO_FLAG_IO_REPAIR = 1 << 1, | |
183 | ZIO_FLAG_SELF_HEAL = 1 << 2, | |
184 | ZIO_FLAG_RESILVER = 1 << 3, | |
185 | ZIO_FLAG_SCRUB = 1 << 4, | |
572e2857 | 186 | ZIO_FLAG_SCAN_THREAD = 1 << 5, |
b02fe35d | 187 | ZIO_FLAG_PHYSICAL = 1 << 6, |
428870ff BB |
188 | |
189 | #define ZIO_FLAG_AGG_INHERIT (ZIO_FLAG_CANFAIL - 1) | |
190 | ||
191 | /* | |
192 | * Flags inherited by ddt, gang, and vdev children. | |
193 | */ | |
b02fe35d AR |
194 | ZIO_FLAG_CANFAIL = 1 << 7, /* must be first for INHERIT */ |
195 | ZIO_FLAG_SPECULATIVE = 1 << 8, | |
196 | ZIO_FLAG_CONFIG_WRITER = 1 << 9, | |
197 | ZIO_FLAG_DONT_RETRY = 1 << 10, | |
198 | ZIO_FLAG_DONT_CACHE = 1 << 11, | |
199 | ZIO_FLAG_NODATA = 1 << 12, | |
200 | ZIO_FLAG_INDUCE_DAMAGE = 1 << 13, | |
3dfb57a3 | 201 | ZIO_FLAG_IO_ALLOCATING = 1 << 14, |
428870ff BB |
202 | |
203 | #define ZIO_FLAG_DDT_INHERIT (ZIO_FLAG_IO_RETRY - 1) | |
204 | #define ZIO_FLAG_GANG_INHERIT (ZIO_FLAG_IO_RETRY - 1) | |
205 | ||
206 | /* | |
207 | * Flags inherited by vdev children. | |
208 | */ | |
3dfb57a3 DB |
209 | ZIO_FLAG_IO_RETRY = 1 << 15, /* must be first for INHERIT */ |
210 | ZIO_FLAG_PROBE = 1 << 16, | |
211 | ZIO_FLAG_TRYHARD = 1 << 17, | |
212 | ZIO_FLAG_OPTIONAL = 1 << 18, | |
428870ff BB |
213 | |
214 | #define ZIO_FLAG_VDEV_INHERIT (ZIO_FLAG_DONT_QUEUE - 1) | |
215 | ||
216 | /* | |
217 | * Flags not inherited by any children. | |
218 | */ | |
3dfb57a3 DB |
219 | ZIO_FLAG_DONT_QUEUE = 1 << 19, /* must be first for INHERIT */ |
220 | ZIO_FLAG_DONT_PROPAGATE = 1 << 20, | |
221 | ZIO_FLAG_IO_BYPASS = 1 << 21, | |
222 | ZIO_FLAG_IO_REWRITE = 1 << 22, | |
b5256303 TC |
223 | ZIO_FLAG_RAW_COMPRESS = 1 << 23, |
224 | ZIO_FLAG_RAW_ENCRYPT = 1 << 24, | |
225 | ZIO_FLAG_GANG_CHILD = 1 << 25, | |
226 | ZIO_FLAG_DDT_CHILD = 1 << 26, | |
227 | ZIO_FLAG_GODFATHER = 1 << 27, | |
228 | ZIO_FLAG_NOPWRITE = 1 << 28, | |
229 | ZIO_FLAG_REEXECUTED = 1 << 29, | |
230 | ZIO_FLAG_DELEGATED = 1 << 30, | |
231 | ZIO_FLAG_FASTWRITE = 1 << 31, | |
428870ff BB |
232 | }; |
233 | ||
234 | #define ZIO_FLAG_MUSTSUCCEED 0 | |
b5256303 | 235 | #define ZIO_FLAG_RAW (ZIO_FLAG_RAW_COMPRESS | ZIO_FLAG_RAW_ENCRYPT) |
428870ff BB |
236 | |
237 | #define ZIO_DDT_CHILD_FLAGS(zio) \ | |
238 | (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) | \ | |
239 | ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL) | |
240 | ||
b128c09f BB |
241 | #define ZIO_GANG_CHILD_FLAGS(zio) \ |
242 | (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) | \ | |
243 | ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL) | |
244 | ||
428870ff BB |
245 | #define ZIO_VDEV_CHILD_FLAGS(zio) \ |
246 | (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) | \ | |
a1d477c2 | 247 | ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_CANFAIL) |
428870ff | 248 | |
ddc751d5 GW |
249 | #define ZIO_CHILD_BIT(x) (1 << (x)) |
250 | #define ZIO_CHILD_BIT_IS_SET(val, x) ((val) & (1 << (x))) | |
251 | ||
b128c09f BB |
252 | enum zio_child { |
253 | ZIO_CHILD_VDEV = 0, | |
254 | ZIO_CHILD_GANG, | |
428870ff | 255 | ZIO_CHILD_DDT, |
b128c09f BB |
256 | ZIO_CHILD_LOGICAL, |
257 | ZIO_CHILD_TYPES | |
258 | }; | |
259 | ||
ddc751d5 GW |
260 | #define ZIO_CHILD_VDEV_BIT ZIO_CHILD_BIT(ZIO_CHILD_VDEV) |
261 | #define ZIO_CHILD_GANG_BIT ZIO_CHILD_BIT(ZIO_CHILD_GANG) | |
262 | #define ZIO_CHILD_DDT_BIT ZIO_CHILD_BIT(ZIO_CHILD_DDT) | |
263 | #define ZIO_CHILD_LOGICAL_BIT ZIO_CHILD_BIT(ZIO_CHILD_LOGICAL) | |
264 | #define ZIO_CHILD_ALL_BITS \ | |
62840030 | 265 | (ZIO_CHILD_VDEV_BIT | ZIO_CHILD_GANG_BIT | \ |
ddc751d5 GW |
266 | ZIO_CHILD_DDT_BIT | ZIO_CHILD_LOGICAL_BIT) |
267 | ||
b128c09f BB |
268 | enum zio_wait_type { |
269 | ZIO_WAIT_READY = 0, | |
270 | ZIO_WAIT_DONE, | |
271 | ZIO_WAIT_TYPES | |
272 | }; | |
273 | ||
34dc7c2f BB |
274 | /* |
275 | * We'll take the unused errnos, 'EBADE' and 'EBADR' (from the Convergent | |
276 | * graveyard) to indicate checksum errors and fragmentation. | |
277 | */ | |
278 | #define ECKSUM EBADE | |
279 | #define EFRAGS EBADR | |
280 | ||
a1d477c2 MA |
281 | /* Similar for ENOACTIVE */ |
282 | #define ENOTACTIVE ENOANO | |
283 | ||
34dc7c2f BB |
284 | typedef void zio_done_func_t(zio_t *zio); |
285 | ||
3dfb57a3 | 286 | extern int zio_dva_throttle_enabled; |
e8b96c60 | 287 | extern const char *zio_type_name[ZIO_TYPES]; |
34dc7c2f BB |
288 | |
289 | /* | |
290 | * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely | |
291 | * identifies any block in the pool. By convention, the meta-objset (MOS) | |
428870ff BB |
292 | * is objset 0, and the meta-dnode is object 0. This covers all blocks |
293 | * except root blocks and ZIL blocks, which are defined as follows: | |
34dc7c2f | 294 | * |
428870ff BB |
295 | * Root blocks (objset_phys_t) are object 0, level -1: <objset, 0, -1, 0>. |
296 | * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>. | |
297 | * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>. | |
fcff0f35 | 298 | * dnode visit bookmarks are <objset, object id of dnode, -3, 0>. |
34dc7c2f | 299 | * |
428870ff BB |
300 | * Note: this structure is called a bookmark because its original purpose |
301 | * was to remember where to resume a pool-wide traverse. | |
34dc7c2f | 302 | * |
5dbd68a3 MA |
303 | * Note: this structure is passed between userland and the kernel, and is |
304 | * stored on disk (by virtue of being incorporated into other on-disk | |
305 | * structures, e.g. dsl_scan_phys_t). | |
34dc7c2f | 306 | */ |
5dbd68a3 | 307 | struct zbookmark_phys { |
34dc7c2f BB |
308 | uint64_t zb_objset; |
309 | uint64_t zb_object; | |
310 | int64_t zb_level; | |
311 | uint64_t zb_blkid; | |
1421c891 | 312 | }; |
34dc7c2f | 313 | |
428870ff BB |
314 | #define SET_BOOKMARK(zb, objset, object, level, blkid) \ |
315 | { \ | |
316 | (zb)->zb_objset = objset; \ | |
317 | (zb)->zb_object = object; \ | |
318 | (zb)->zb_level = level; \ | |
319 | (zb)->zb_blkid = blkid; \ | |
320 | } | |
321 | ||
322 | #define ZB_DESTROYED_OBJSET (-1ULL) | |
323 | ||
324 | #define ZB_ROOT_OBJECT (0ULL) | |
325 | #define ZB_ROOT_LEVEL (-1LL) | |
326 | #define ZB_ROOT_BLKID (0ULL) | |
327 | ||
328 | #define ZB_ZIL_OBJECT (0ULL) | |
329 | #define ZB_ZIL_LEVEL (-2LL) | |
330 | ||
fcff0f35 PD |
331 | #define ZB_DNODE_LEVEL (-3LL) |
332 | #define ZB_DNODE_BLKID (0ULL) | |
333 | ||
9ae529ec CS |
334 | #define ZB_IS_ZERO(zb) \ |
335 | ((zb)->zb_objset == 0 && (zb)->zb_object == 0 && \ | |
336 | (zb)->zb_level == 0 && (zb)->zb_blkid == 0) | |
337 | #define ZB_IS_ROOT(zb) \ | |
338 | ((zb)->zb_object == ZB_ROOT_OBJECT && \ | |
339 | (zb)->zb_level == ZB_ROOT_LEVEL && \ | |
340 | (zb)->zb_blkid == ZB_ROOT_BLKID) | |
341 | ||
b128c09f BB |
342 | typedef struct zio_prop { |
343 | enum zio_checksum zp_checksum; | |
344 | enum zio_compress zp_compress; | |
345 | dmu_object_type_t zp_type; | |
346 | uint8_t zp_level; | |
428870ff | 347 | uint8_t zp_copies; |
03c6040b GW |
348 | boolean_t zp_dedup; |
349 | boolean_t zp_dedup_verify; | |
350 | boolean_t zp_nopwrite; | |
b5256303 TC |
351 | boolean_t zp_encrypt; |
352 | boolean_t zp_byteorder; | |
353 | uint8_t zp_salt[ZIO_DATA_SALT_LEN]; | |
354 | uint8_t zp_iv[ZIO_DATA_IV_LEN]; | |
355 | uint8_t zp_mac[ZIO_DATA_MAC_LEN]; | |
cc99f275 | 356 | uint32_t zp_zpl_smallblk; |
b128c09f BB |
357 | } zio_prop_t; |
358 | ||
428870ff BB |
359 | typedef struct zio_cksum_report zio_cksum_report_t; |
360 | ||
361 | typedef void zio_cksum_finish_f(zio_cksum_report_t *rep, | |
84c07ada | 362 | const abd_t *good_data); |
428870ff BB |
363 | typedef void zio_cksum_free_f(void *cbdata, size_t size); |
364 | ||
365 | struct zio_bad_cksum; /* defined in zio_checksum.h */ | |
9ae529ec | 366 | struct dnode_phys; |
a6255b7f | 367 | struct abd; |
428870ff BB |
368 | |
369 | struct zio_cksum_report { | |
370 | struct zio_cksum_report *zcr_next; | |
371 | nvlist_t *zcr_ereport; | |
372 | nvlist_t *zcr_detector; | |
373 | void *zcr_cbdata; | |
374 | size_t zcr_cbinfo; /* passed to zcr_free() */ | |
375 | uint64_t zcr_align; | |
376 | uint64_t zcr_length; | |
377 | zio_cksum_finish_f *zcr_finish; | |
378 | zio_cksum_free_f *zcr_free; | |
379 | ||
380 | /* internal use only */ | |
381 | struct zio_bad_cksum *zcr_ckinfo; /* information from failure */ | |
382 | }; | |
383 | ||
384 | typedef void zio_vsd_cksum_report_f(zio_t *zio, zio_cksum_report_t *zcr, | |
385 | void *arg); | |
386 | ||
387 | zio_vsd_cksum_report_f zio_vsd_default_cksum_report; | |
388 | ||
389 | typedef struct zio_vsd_ops { | |
390 | zio_done_func_t *vsd_free; | |
391 | zio_vsd_cksum_report_f *vsd_cksum_report; | |
392 | } zio_vsd_ops_t; | |
393 | ||
b128c09f BB |
394 | typedef struct zio_gang_node { |
395 | zio_gbh_phys_t *gn_gbh; | |
396 | struct zio_gang_node *gn_child[SPA_GBH_NBLKPTRS]; | |
397 | } zio_gang_node_t; | |
398 | ||
399 | typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp, | |
a6255b7f | 400 | zio_gang_node_t *gn, struct abd *data, uint64_t offset); |
b128c09f | 401 | |
a6255b7f | 402 | typedef void zio_transform_func_t(zio_t *zio, struct abd *data, uint64_t size); |
b128c09f BB |
403 | |
404 | typedef struct zio_transform { | |
a6255b7f | 405 | struct abd *zt_orig_abd; |
b128c09f BB |
406 | uint64_t zt_orig_size; |
407 | uint64_t zt_bufsize; | |
408 | zio_transform_func_t *zt_transform; | |
409 | struct zio_transform *zt_next; | |
410 | } zio_transform_t; | |
411 | ||
62840030 | 412 | typedef zio_t *zio_pipe_stage_t(zio_t *zio); |
b128c09f BB |
413 | |
414 | /* | |
415 | * The io_reexecute flags are distinct from io_flags because the child must | |
416 | * be able to propagate them to the parent. The normal io_flags are local | |
417 | * to the zio, not protected by any lock, and not modifiable by children; | |
418 | * the reexecute flags are protected by io_lock, modifiable by children, | |
419 | * and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set. | |
420 | */ | |
421 | #define ZIO_REEXECUTE_NOW 0x01 | |
422 | #define ZIO_REEXECUTE_SUSPEND 0x02 | |
423 | ||
4e21fd06 DB |
424 | typedef struct zio_alloc_list { |
425 | list_t zal_list; | |
426 | uint64_t zal_size; | |
427 | } zio_alloc_list_t; | |
428 | ||
d164b209 BB |
429 | typedef struct zio_link { |
430 | zio_t *zl_parent; | |
431 | zio_t *zl_child; | |
432 | list_node_t zl_parent_node; | |
433 | list_node_t zl_child_node; | |
434 | } zio_link_t; | |
435 | ||
34dc7c2f BB |
436 | struct zio { |
437 | /* Core information about this I/O */ | |
5dbd68a3 | 438 | zbookmark_phys_t io_bookmark; |
b128c09f BB |
439 | zio_prop_t io_prop; |
440 | zio_type_t io_type; | |
441 | enum zio_child io_child_type; | |
442 | int io_cmd; | |
e8b96c60 | 443 | zio_priority_t io_priority; |
b128c09f | 444 | uint8_t io_reexecute; |
d164b209 | 445 | uint8_t io_state[ZIO_WAIT_TYPES]; |
34dc7c2f | 446 | uint64_t io_txg; |
b128c09f | 447 | spa_t *io_spa; |
34dc7c2f | 448 | blkptr_t *io_bp; |
428870ff | 449 | blkptr_t *io_bp_override; |
34dc7c2f | 450 | blkptr_t io_bp_copy; |
d164b209 BB |
451 | list_t io_parent_list; |
452 | list_t io_child_list; | |
34dc7c2f | 453 | zio_t *io_logical; |
b128c09f | 454 | zio_transform_t *io_transform_stack; |
34dc7c2f BB |
455 | |
456 | /* Callback info */ | |
bc77ba73 PD |
457 | zio_done_func_t *io_ready; |
458 | zio_done_func_t *io_children_ready; | |
e8b96c60 | 459 | zio_done_func_t *io_physdone; |
34dc7c2f BB |
460 | zio_done_func_t *io_done; |
461 | void *io_private; | |
428870ff | 462 | int64_t io_prev_space_delta; /* DMU private */ |
34dc7c2f | 463 | blkptr_t io_bp_orig; |
2aa34383 DK |
464 | /* io_lsize != io_orig_size iff this is a raw write */ |
465 | uint64_t io_lsize; | |
34dc7c2f BB |
466 | |
467 | /* Data represented by this I/O */ | |
a6255b7f DQ |
468 | struct abd *io_abd; |
469 | struct abd *io_orig_abd; | |
34dc7c2f | 470 | uint64_t io_size; |
428870ff | 471 | uint64_t io_orig_size; |
34dc7c2f BB |
472 | |
473 | /* Stuff for the vdev stack */ | |
474 | vdev_t *io_vd; | |
475 | void *io_vsd; | |
428870ff | 476 | const zio_vsd_ops_t *io_vsd_ops; |
cc99f275 | 477 | metaslab_class_t *io_metaslab_class; /* dva throttle class */ |
428870ff | 478 | |
34dc7c2f | 479 | uint64_t io_offset; |
cb682a17 | 480 | hrtime_t io_timestamp; /* submitted at */ |
3dfb57a3 | 481 | hrtime_t io_queued_timestamp; |
26ef0cc7 | 482 | hrtime_t io_target_timestamp; |
cb682a17 | 483 | hrtime_t io_delta; /* vdev queue service delta */ |
193a37cb TH |
484 | hrtime_t io_delay; /* Device access time (disk or */ |
485 | /* file). */ | |
e8b96c60 | 486 | avl_node_t io_queue_node; |
ec8501ee | 487 | avl_node_t io_offset_node; |
3dfb57a3 | 488 | avl_node_t io_alloc_node; |
4e21fd06 | 489 | zio_alloc_list_t io_alloc_list; |
34dc7c2f BB |
490 | |
491 | /* Internal pipeline state */ | |
428870ff BB |
492 | enum zio_flag io_flags; |
493 | enum zio_stage io_stage; | |
494 | enum zio_stage io_pipeline; | |
495 | enum zio_flag io_orig_flags; | |
496 | enum zio_stage io_orig_stage; | |
497 | enum zio_stage io_orig_pipeline; | |
3dfb57a3 | 498 | enum zio_stage io_pipeline_trace; |
b128c09f BB |
499 | int io_error; |
500 | int io_child_error[ZIO_CHILD_TYPES]; | |
501 | uint64_t io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES]; | |
428870ff | 502 | uint64_t io_child_count; |
e8b96c60 | 503 | uint64_t io_phys_children; |
428870ff | 504 | uint64_t io_parent_count; |
b128c09f | 505 | uint64_t *io_stall; |
9babb374 | 506 | zio_t *io_gang_leader; |
b128c09f BB |
507 | zio_gang_node_t *io_gang_tree; |
508 | void *io_executor; | |
34dc7c2f BB |
509 | void *io_waiter; |
510 | kmutex_t io_lock; | |
511 | kcondvar_t io_cv; | |
492f64e9 | 512 | int io_allocator; |
34dc7c2f BB |
513 | |
514 | /* FMA state */ | |
428870ff | 515 | zio_cksum_report_t *io_cksum_report; |
34dc7c2f | 516 | uint64_t io_ena; |
a38718a6 GA |
517 | |
518 | /* Taskq dispatching state */ | |
519 | taskq_ent_t io_tqent; | |
34dc7c2f BB |
520 | }; |
521 | ||
64fc7762 | 522 | extern int zio_bookmark_compare(const void *, const void *); |
3dfb57a3 | 523 | |
d164b209 | 524 | extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd, |
428870ff | 525 | zio_done_func_t *done, void *private, enum zio_flag flags); |
34dc7c2f BB |
526 | |
527 | extern zio_t *zio_root(spa_t *spa, | |
428870ff | 528 | zio_done_func_t *done, void *private, enum zio_flag flags); |
34dc7c2f | 529 | |
a6255b7f DQ |
530 | extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp, |
531 | struct abd *data, uint64_t lsize, zio_done_func_t *done, void *private, | |
5dbd68a3 | 532 | zio_priority_t priority, enum zio_flag flags, const zbookmark_phys_t *zb); |
34dc7c2f | 533 | |
b128c09f | 534 | extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, |
a6255b7f | 535 | struct abd *data, uint64_t size, uint64_t psize, const zio_prop_t *zp, |
bc77ba73 PD |
536 | zio_done_func_t *ready, zio_done_func_t *children_ready, |
537 | zio_done_func_t *physdone, zio_done_func_t *done, | |
538 | void *private, zio_priority_t priority, enum zio_flag flags, | |
539 | const zbookmark_phys_t *zb); | |
34dc7c2f | 540 | |
b128c09f | 541 | extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, |
a6255b7f | 542 | struct abd *data, uint64_t size, zio_done_func_t *done, void *private, |
5dbd68a3 | 543 | zio_priority_t priority, enum zio_flag flags, zbookmark_phys_t *zb); |
b128c09f | 544 | |
03c6040b GW |
545 | extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies, |
546 | boolean_t nopwrite); | |
34dc7c2f | 547 | |
428870ff | 548 | extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp); |
34dc7c2f | 549 | |
428870ff BB |
550 | extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, |
551 | const blkptr_t *bp, | |
552 | zio_done_func_t *done, void *private, enum zio_flag flags); | |
34dc7c2f BB |
553 | |
554 | extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, | |
e8b96c60 | 555 | zio_done_func_t *done, void *private, enum zio_flag flags); |
34dc7c2f BB |
556 | |
557 | extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, | |
a6255b7f | 558 | uint64_t size, struct abd *data, int checksum, |
e8b96c60 MA |
559 | zio_done_func_t *done, void *private, zio_priority_t priority, |
560 | enum zio_flag flags, boolean_t labels); | |
34dc7c2f BB |
561 | |
562 | extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, | |
a6255b7f | 563 | uint64_t size, struct abd *data, int checksum, |
e8b96c60 MA |
564 | zio_done_func_t *done, void *private, zio_priority_t priority, |
565 | enum zio_flag flags, boolean_t labels); | |
34dc7c2f | 566 | |
428870ff BB |
567 | extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg, |
568 | const blkptr_t *bp, enum zio_flag flags); | |
569 | ||
b5256303 TC |
570 | extern int zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg, |
571 | blkptr_t *new_bp, uint64_t size, boolean_t *slog); | |
34dc7c2f | 572 | extern void zio_flush(zio_t *zio, vdev_t *vd); |
428870ff | 573 | extern void zio_shrink(zio_t *zio, uint64_t size); |
34dc7c2f BB |
574 | |
575 | extern int zio_wait(zio_t *zio); | |
576 | extern void zio_nowait(zio_t *zio); | |
577 | extern void zio_execute(zio_t *zio); | |
578 | extern void zio_interrupt(zio_t *zio); | |
26ef0cc7 TH |
579 | extern void zio_delay_init(zio_t *zio); |
580 | extern void zio_delay_interrupt(zio_t *zio); | |
8fb1ede1 | 581 | extern void zio_deadman(zio_t *zio, char *tag); |
34dc7c2f | 582 | |
3dfb57a3 DB |
583 | extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **); |
584 | extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **); | |
d164b209 BB |
585 | extern zio_t *zio_unique_parent(zio_t *cio); |
586 | extern void zio_add_child(zio_t *pio, zio_t *cio); | |
587 | ||
34dc7c2f BB |
588 | extern void *zio_buf_alloc(size_t size); |
589 | extern void zio_buf_free(void *buf, size_t size); | |
590 | extern void *zio_data_buf_alloc(size_t size); | |
591 | extern void zio_data_buf_free(void *buf, size_t size); | |
592 | ||
a6255b7f | 593 | extern void zio_push_transform(zio_t *zio, struct abd *abd, uint64_t size, |
d3c2ae1c GW |
594 | uint64_t bufsize, zio_transform_func_t *transform); |
595 | extern void zio_pop_transforms(zio_t *zio); | |
596 | ||
34dc7c2f BB |
597 | extern void zio_resubmit_stage_async(void *); |
598 | ||
34dc7c2f | 599 | extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd, |
a6255b7f | 600 | uint64_t offset, struct abd *data, uint64_t size, int type, |
e8b96c60 MA |
601 | zio_priority_t priority, enum zio_flag flags, |
602 | zio_done_func_t *done, void *private); | |
34dc7c2f | 603 | |
b128c09f | 604 | extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset, |
9e052db4 | 605 | struct abd *data, uint64_t size, zio_type_t type, zio_priority_t priority, |
428870ff | 606 | enum zio_flag flags, zio_done_func_t *done, void *private); |
b128c09f | 607 | |
34dc7c2f BB |
608 | extern void zio_vdev_io_bypass(zio_t *zio); |
609 | extern void zio_vdev_io_reissue(zio_t *zio); | |
610 | extern void zio_vdev_io_redone(zio_t *zio); | |
611 | ||
a8b2e306 TC |
612 | extern void zio_change_priority(zio_t *pio, zio_priority_t priority); |
613 | ||
34dc7c2f | 614 | extern void zio_checksum_verified(zio_t *zio); |
b128c09f | 615 | extern int zio_worst_error(int e1, int e2); |
34dc7c2f | 616 | |
428870ff BB |
617 | extern enum zio_checksum zio_checksum_select(enum zio_checksum child, |
618 | enum zio_checksum parent); | |
619 | extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa, | |
620 | enum zio_checksum child, enum zio_checksum parent); | |
99197f03 JG |
621 | extern enum zio_compress zio_compress_select(spa_t *spa, |
622 | enum zio_compress child, enum zio_compress parent); | |
34dc7c2f | 623 | |
cec3a0a1 | 624 | extern void zio_suspend(spa_t *spa, zio_t *zio, zio_suspend_reason_t); |
9babb374 | 625 | extern int zio_resume(spa_t *spa); |
b128c09f | 626 | extern void zio_resume_wait(spa_t *spa); |
34dc7c2f BB |
627 | |
628 | /* | |
629 | * Initial setup and teardown. | |
630 | */ | |
631 | extern void zio_init(void); | |
632 | extern void zio_fini(void); | |
633 | ||
634 | /* | |
635 | * Fault injection | |
636 | */ | |
637 | struct zinject_record; | |
638 | extern uint32_t zio_injection_enabled; | |
639 | extern int zio_inject_fault(char *name, int flags, int *id, | |
640 | struct zinject_record *record); | |
641 | extern int zio_inject_list_next(int *id, char *name, size_t buflen, | |
642 | struct zinject_record *record); | |
643 | extern int zio_clear_fault(int id); | |
428870ff | 644 | extern void zio_handle_panic_injection(spa_t *spa, char *tag, uint64_t type); |
be9a5c35 TC |
645 | extern int zio_handle_decrypt_injection(spa_t *spa, const zbookmark_phys_t *zb, |
646 | uint64_t type, int error); | |
34dc7c2f | 647 | extern int zio_handle_fault_injection(zio_t *zio, int error); |
9babb374 | 648 | extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error); |
d977122d DB |
649 | extern int zio_handle_device_injections(vdev_t *vd, zio_t *zio, int err1, |
650 | int err2); | |
b128c09f | 651 | extern int zio_handle_label_injection(zio_t *zio, int error); |
428870ff | 652 | extern void zio_handle_ignored_writes(zio_t *zio); |
26ef0cc7 | 653 | extern hrtime_t zio_handle_io_delay(zio_t *zio); |
428870ff BB |
654 | |
655 | /* | |
656 | * Checksum ereport functions | |
657 | */ | |
b5256303 | 658 | extern void zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd, |
a2c2ed1b TC |
659 | const zbookmark_phys_t *zb, struct zio *zio, uint64_t offset, |
660 | uint64_t length, void *arg, struct zio_bad_cksum *info); | |
428870ff | 661 | extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report, |
84c07ada | 662 | const abd_t *good_data, const abd_t *bad_data, boolean_t drop_if_identical); |
428870ff | 663 | |
428870ff BB |
664 | extern void zfs_ereport_free_checksum(zio_cksum_report_t *report); |
665 | ||
666 | /* If we have the good data in hand, this function can be used */ | |
667 | extern void zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd, | |
a2c2ed1b TC |
668 | const zbookmark_phys_t *zb, struct zio *zio, uint64_t offset, |
669 | uint64_t length, const abd_t *good_data, const abd_t *bad_data, | |
670 | struct zio_bad_cksum *info); | |
428870ff BB |
671 | |
672 | /* Called from spa_sync(), but primarily an injection handler */ | |
673 | extern void spa_handle_ignored_writes(spa_t *spa); | |
34dc7c2f | 674 | |
5dbd68a3 | 675 | /* zbookmark_phys functions */ |
fcff0f35 PD |
676 | boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp, |
677 | const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block); | |
678 | int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2, | |
679 | uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2); | |
9ae529ec | 680 | |
34dc7c2f BB |
681 | #ifdef __cplusplus |
682 | } | |
683 | #endif | |
684 | ||
685 | #endif /* _ZIO_H */ |