1 // SPDX-License-Identifier: CDDL-1.0
2 /*
3 * This file and its contents are supplied under the terms of the
4 * Common Development and Distribution License ("CDDL"), version 1.0.
5 * You may only use this file in accordance with the terms of version
6 * 1.0 of the CDDL.
7 *
8 * A full copy of the text of the CDDL should have accompanied this
9 * source. A copy of the CDDL is also available via the Internet at
10 * https://opensource.org/license/CDDL-1.0.
11 */
12
13 /*
14 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
15 * Copyright 2011 Nexenta Systems, Inc. All rights reserved.
16 * Copyright (c) 2012, 2024 by Delphix. All rights reserved.
17 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
18 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
19 * Copyright 2016 Toomas Soome <tsoome@me.com>
20 * Copyright (c) 2019, Allan Jude
21 * Copyright (c) 2019, 2023, 2024, Klara Inc.
22 * Copyright (c) 2019-2020, Michael Niewöhner
23 * Copyright (c) 2024 by George Melikov. All rights reserved.
24 */
25
26 #ifndef _ZIO_H
27 #define _ZIO_H
28
29 #include <sys/zfs_context.h>
30 #include <sys/spa.h>
31 #include <sys/txg.h>
32 #include <sys/avl.h>
33 #include <sys/fs/zfs.h>
34 #include <sys/zio_impl.h>
35
36 #ifdef __cplusplus
37 extern "C" {
38 #endif
39
40 /*
41 * Embedded checksum
42 */
43 #define ZEC_MAGIC 0x210da7ab10c7a11ULL
44
45 typedef struct zio_eck {
46 uint64_t zec_magic; /* for validation, endianness */
47 zio_cksum_t zec_cksum; /* 256-bit checksum */
48 } zio_eck_t;
49
50 /*
51 * Gang block headers are self-checksumming and contain an array
52 * of block pointers. The old gang block size has enough room for 3 blkptrs,
53 * while new gang blocks can store more.
54 *
55 * Layout:
56 * +--------+--------+--------+-----+---------+-----------+
57 * | | | | | | |
58 * | blkptr | blkptr | blkptr | ... | padding | zio_eck_t |
59 * | 1 | 2 | 3 | | | |
60 * +--------+--------+--------+-----+---------+-----------+
61 * 128B 128B 128B 88B 40B
62 */
63 #define SPA_OLD_GANGBLOCKSIZE SPA_MINBLOCKSIZE
64 typedef void zio_gbh_phys_t;
65
66 static inline uint64_t
gbh_nblkptrs(uint64_t size)67 gbh_nblkptrs(uint64_t size) {
68 ASSERT(IS_P2ALIGNED(size, sizeof (blkptr_t)));
69 return ((size - sizeof (zio_eck_t)) / sizeof (blkptr_t));
70 }
71
72 static inline zio_eck_t *
gbh_eck(zio_gbh_phys_t * gbh,uint64_t size)73 gbh_eck(zio_gbh_phys_t *gbh, uint64_t size) {
74 ASSERT(IS_P2ALIGNED(size, sizeof (blkptr_t)));
75 return ((zio_eck_t *)((uintptr_t)gbh + (size_t)size -
76 sizeof (zio_eck_t)));
77 }
78
79 static inline blkptr_t *
gbh_bp(zio_gbh_phys_t * gbh,int bp)80 gbh_bp(zio_gbh_phys_t *gbh, int bp) {
81 return (&((blkptr_t *)gbh)[bp]);
82 }
83
84 enum zio_checksum {
85 ZIO_CHECKSUM_INHERIT = 0,
86 ZIO_CHECKSUM_ON,
87 ZIO_CHECKSUM_OFF,
88 ZIO_CHECKSUM_LABEL,
89 ZIO_CHECKSUM_GANG_HEADER,
90 ZIO_CHECKSUM_ZILOG,
91 ZIO_CHECKSUM_FLETCHER_2,
92 ZIO_CHECKSUM_FLETCHER_4,
93 ZIO_CHECKSUM_SHA256,
94 ZIO_CHECKSUM_ZILOG2,
95 ZIO_CHECKSUM_NOPARITY,
96 ZIO_CHECKSUM_SHA512,
97 ZIO_CHECKSUM_SKEIN,
98 ZIO_CHECKSUM_EDONR,
99 ZIO_CHECKSUM_BLAKE3,
100 ZIO_CHECKSUM_FUNCTIONS
101 };
102
103 /*
104 * The number of "legacy" compression functions which can be set on individual
105 * objects.
106 */
107 #define ZIO_CHECKSUM_LEGACY_FUNCTIONS ZIO_CHECKSUM_ZILOG2
108
109 #define ZIO_CHECKSUM_ON_VALUE ZIO_CHECKSUM_FLETCHER_4
110 #define ZIO_CHECKSUM_DEFAULT ZIO_CHECKSUM_ON
111
112 #define ZIO_CHECKSUM_MASK 0xffULL
113 #define ZIO_CHECKSUM_VERIFY (1U << 8)
114
115 #define ZIO_DEDUPCHECKSUM ZIO_CHECKSUM_SHA256
116
117 /* macros defining encryption lengths */
118 #define ZIO_OBJSET_MAC_LEN 32
119 #define ZIO_DATA_IV_LEN 12
120 #define ZIO_DATA_SALT_LEN 8
121 #define ZIO_DATA_MAC_LEN 16
122
123 /*
124 * The number of "legacy" compression functions which can be set on individual
125 * objects.
126 */
127 #define ZIO_COMPRESS_LEGACY_FUNCTIONS ZIO_COMPRESS_LZ4
128
129 /*
130 * The meaning of "compress = on" selected by the compression features enabled
131 * on a given pool.
132 */
133 #define ZIO_COMPRESS_LEGACY_ON_VALUE ZIO_COMPRESS_LZJB
134 #define ZIO_COMPRESS_LZ4_ON_VALUE ZIO_COMPRESS_LZ4
135
136 #define ZIO_COMPRESS_DEFAULT ZIO_COMPRESS_ON
137
138 #define BOOTFS_COMPRESS_VALID(compress) \
139 ((compress) == ZIO_COMPRESS_LZJB || \
140 (compress) == ZIO_COMPRESS_LZ4 || \
141 (compress) == ZIO_COMPRESS_GZIP_1 || \
142 (compress) == ZIO_COMPRESS_GZIP_2 || \
143 (compress) == ZIO_COMPRESS_GZIP_3 || \
144 (compress) == ZIO_COMPRESS_GZIP_4 || \
145 (compress) == ZIO_COMPRESS_GZIP_5 || \
146 (compress) == ZIO_COMPRESS_GZIP_6 || \
147 (compress) == ZIO_COMPRESS_GZIP_7 || \
148 (compress) == ZIO_COMPRESS_GZIP_8 || \
149 (compress) == ZIO_COMPRESS_GZIP_9 || \
150 (compress) == ZIO_COMPRESS_ZLE || \
151 (compress) == ZIO_COMPRESS_ZSTD || \
152 (compress) == ZIO_COMPRESS_ON || \
153 (compress) == ZIO_COMPRESS_OFF)
154
155
156 #define ZIO_COMPRESS_ALGO(x) (x & SPA_COMPRESSMASK)
157 #define ZIO_COMPRESS_LEVEL(x) ((x & ~SPA_COMPRESSMASK) >> SPA_COMPRESSBITS)
158 #define ZIO_COMPRESS_RAW(type, level) (type | ((level) << SPA_COMPRESSBITS))
159
160 #define ZIO_COMPLEVEL_ZSTD(level) \
161 ZIO_COMPRESS_RAW(ZIO_COMPRESS_ZSTD, level)
162
163 #define ZIO_FAILURE_MODE_WAIT 0
164 #define ZIO_FAILURE_MODE_CONTINUE 1
165 #define ZIO_FAILURE_MODE_PANIC 2
166
167 typedef enum zio_suspend_reason {
168 ZIO_SUSPEND_NONE = 0,
169 ZIO_SUSPEND_IOERR,
170 ZIO_SUSPEND_MMP,
171 } zio_suspend_reason_t;
172
173 /*
174 * This was originally an enum type. However, those are 32-bit and there is no
175 * way to make a 64-bit enum type. Since we ran out of bits for flags, we were
176 * forced to upgrade it to a uint64_t.
177 *
178 * NOTE: PLEASE UPDATE THE BITFIELD STRINGS IN zfs_valstr.c IF YOU ADD ANOTHER
179 * FLAG.
180 */
181 typedef uint64_t zio_flag_t;
182 /*
183 * Flags inherited by gang, ddt, and vdev children,
184 * and that must be equal for two zios to aggregate
185 */
186 #define ZIO_FLAG_DONT_AGGREGATE (1ULL << 0)
187 #define ZIO_FLAG_IO_REPAIR (1ULL << 1)
188 #define ZIO_FLAG_SELF_HEAL (1ULL << 2)
189 #define ZIO_FLAG_RESILVER (1ULL << 3)
190 #define ZIO_FLAG_SCRUB (1ULL << 4)
191 #define ZIO_FLAG_SCAN_THREAD (1ULL << 5)
192 #define ZIO_FLAG_PHYSICAL (1ULL << 6)
193
194 #define ZIO_FLAG_AGG_INHERIT (ZIO_FLAG_CANFAIL - 1)
195
196 /*
197 * Flags inherited by ddt, gang, and vdev children.
198 */
199 #define ZIO_FLAG_CANFAIL (1ULL << 7) /* must be first for INHERIT */
200 #define ZIO_FLAG_SPECULATIVE (1ULL << 8)
201 #define ZIO_FLAG_CONFIG_WRITER (1ULL << 9)
202 #define ZIO_FLAG_DONT_RETRY (1ULL << 10)
203 #define ZIO_FLAG_NODATA (1ULL << 12)
204 #define ZIO_FLAG_INDUCE_DAMAGE (1ULL << 13)
205 #define ZIO_FLAG_ALLOC_THROTTLED (1ULL << 14)
206
207 #define ZIO_FLAG_DDT_INHERIT (ZIO_FLAG_IO_RETRY - 1)
208 #define ZIO_FLAG_GANG_INHERIT (ZIO_FLAG_IO_RETRY - 1)
209
210 /*
211 * Flags inherited by vdev children.
212 */
213 #define ZIO_FLAG_IO_RETRY (1ULL << 15) /* must be first for INHERIT */
214 #define ZIO_FLAG_PROBE (1ULL << 16)
215 #define ZIO_FLAG_TRYHARD (1ULL << 17)
216 #define ZIO_FLAG_OPTIONAL (1ULL << 18)
217 #define ZIO_FLAG_DIO_READ (1ULL << 19)
218 #define ZIO_FLAG_BYPASSED_QUEUE (1ULL << 20)
219 #define ZIO_FLAG_VDEV_INHERIT (ZIO_FLAG_DONT_QUEUE - 1)
220
221 /*
222 * Flags not inherited by any children.
223 */
224 #define ZIO_FLAG_DONT_QUEUE (1ULL << 21) /* must be first for INHERIT */
225 #define ZIO_FLAG_DONT_PROPAGATE (1ULL << 22)
226 #define ZIO_FLAG_IO_BYPASS (1ULL << 23)
227 #define ZIO_FLAG_IO_REWRITE (1ULL << 24)
228 #define ZIO_FLAG_RAW_COMPRESS (1ULL << 25)
229 #define ZIO_FLAG_RAW_ENCRYPT (1ULL << 26)
230 #define ZIO_FLAG_GANG_CHILD (1ULL << 27)
231 #define ZIO_FLAG_DDT_CHILD (1ULL << 28)
232 #define ZIO_FLAG_GODFATHER (1ULL << 29)
233 #define ZIO_FLAG_NOPWRITE (1ULL << 30)
234 #define ZIO_FLAG_REEXECUTED (1ULL << 31)
235 #define ZIO_FLAG_DELEGATED (1ULL << 32)
236 #define ZIO_FLAG_PREALLOCATED (1ULL << 33)
237 #define ZIO_FLAG_POSTREAD (1ULL << 34)
238
239 #define ZIO_ALLOCATOR_NONE (-1)
240 #define ZIO_HAS_ALLOCATOR(zio) ((zio)->io_allocator != ZIO_ALLOCATOR_NONE)
241
242 #define ZIO_FLAG_MUSTSUCCEED 0
243 #define ZIO_FLAG_RAW (ZIO_FLAG_RAW_COMPRESS | ZIO_FLAG_RAW_ENCRYPT)
244
245 #define ZIO_DDT_CHILD_FLAGS(zio) \
246 (((zio)->io_flags & ZIO_FLAG_DDT_INHERIT) | \
247 ZIO_FLAG_DDT_CHILD | ZIO_FLAG_CANFAIL)
248
249 #define ZIO_GANG_CHILD_FLAGS(zio) \
250 (((zio)->io_flags & ZIO_FLAG_GANG_INHERIT) | \
251 ZIO_FLAG_GANG_CHILD | ZIO_FLAG_CANFAIL)
252
253 #define ZIO_VDEV_CHILD_FLAGS(zio) \
254 (((zio)->io_flags & ZIO_FLAG_VDEV_INHERIT) | \
255 ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_CANFAIL)
256
257 #define ZIO_CHILD_BIT(x) (1U << (x))
258 #define ZIO_CHILD_BIT_IS_SET(val, x) ((val) & (1U << (x)))
259
260 enum zio_child {
261 ZIO_CHILD_VDEV = 0,
262 ZIO_CHILD_GANG,
263 ZIO_CHILD_DDT,
264 ZIO_CHILD_LOGICAL,
265 ZIO_CHILD_TYPES
266 };
267
268 #define ZIO_CHILD_VDEV_BIT ZIO_CHILD_BIT(ZIO_CHILD_VDEV)
269 #define ZIO_CHILD_GANG_BIT ZIO_CHILD_BIT(ZIO_CHILD_GANG)
270 #define ZIO_CHILD_DDT_BIT ZIO_CHILD_BIT(ZIO_CHILD_DDT)
271 #define ZIO_CHILD_LOGICAL_BIT ZIO_CHILD_BIT(ZIO_CHILD_LOGICAL)
272 #define ZIO_CHILD_ALL_BITS \
273 (ZIO_CHILD_VDEV_BIT | ZIO_CHILD_GANG_BIT | \
274 ZIO_CHILD_DDT_BIT | ZIO_CHILD_LOGICAL_BIT)
275
276 enum zio_wait_type {
277 ZIO_WAIT_READY = 0,
278 ZIO_WAIT_DONE,
279 ZIO_WAIT_TYPES
280 };
281
282 typedef void zio_done_func_t(zio_t *zio);
283
284 extern int zio_exclude_metadata;
285 extern int zio_dva_throttle_enabled;
286 extern const char *const zio_type_name[ZIO_TYPES];
287
288 /*
289 * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely
290 * identifies any block in the pool. By convention, the meta-objset (MOS)
291 * is objset 0, and the meta-dnode is object 0. This covers all blocks
292 * except root blocks and ZIL blocks, which are defined as follows:
293 *
294 * Root blocks (objset_phys_t) are object 0, level -1: <objset, 0, -1, 0>.
295 * ZIL blocks are bookmarked <objset, 0, -2, blkid == ZIL sequence number>.
296 * dmu_sync()ed ZIL data blocks are bookmarked <objset, object, -2, blkid>.
297 * dnode visit bookmarks are <objset, object id of dnode, -3, 0>.
298 *
299 * Note: this structure is called a bookmark because its original purpose
300 * was to remember where to resume a pool-wide traverse.
301 *
302 * Note: this structure is passed between userland and the kernel, and is
303 * stored on disk (by virtue of being incorporated into other on-disk
304 * structures, e.g. dsl_scan_phys_t).
305 *
306 * If the head_errlog feature is enabled a different on-disk format for error
307 * logs is used. This introduces the use of an error bookmark, a four-tuple
308 * <object, level, blkid, birth> that uniquely identifies any error block
309 * in the pool. The birth transaction group is used to track whether the block
310 * has been overwritten by newer data or added to a snapshot since its marking
311 * as an error.
312 */
313 struct zbookmark_phys {
314 uint64_t zb_objset;
315 uint64_t zb_object;
316 int64_t zb_level;
317 uint64_t zb_blkid;
318 };
319
320 struct zbookmark_err_phys {
321 uint64_t zb_object;
322 int64_t zb_level;
323 uint64_t zb_blkid;
324 uint64_t zb_birth;
325 };
326
327 #define SET_BOOKMARK(zb, objset, object, level, blkid) \
328 { \
329 (zb)->zb_objset = objset; \
330 (zb)->zb_object = object; \
331 (zb)->zb_level = level; \
332 (zb)->zb_blkid = blkid; \
333 }
334
335 #define ZB_DESTROYED_OBJSET (-1ULL)
336
337 #define ZB_ROOT_OBJECT (0ULL)
338 #define ZB_ROOT_LEVEL (-1LL)
339 #define ZB_ROOT_BLKID (0ULL)
340
341 #define ZB_ZIL_OBJECT (0ULL)
342 #define ZB_ZIL_LEVEL (-2LL)
343
344 #define ZB_DNODE_LEVEL (-3LL)
345 #define ZB_DNODE_BLKID (0ULL)
346
347 #define ZB_IS_ZERO(zb) \
348 ((zb)->zb_objset == 0 && (zb)->zb_object == 0 && \
349 (zb)->zb_level == 0 && (zb)->zb_blkid == 0)
350 #define ZB_IS_ROOT(zb) \
351 ((zb)->zb_object == ZB_ROOT_OBJECT && \
352 (zb)->zb_level == ZB_ROOT_LEVEL && \
353 (zb)->zb_blkid == ZB_ROOT_BLKID)
354
355 typedef struct zio_prop {
356 enum zio_checksum zp_checksum:8;
357 enum zio_compress zp_compress:8;
358 uint8_t zp_complevel;
359 uint8_t zp_level;
360 uint8_t zp_copies;
361 uint8_t zp_gang_copies;
362 dmu_object_type_t zp_type:8;
363 dmu_object_type_t zp_storage_type:8;
364 boolean_t zp_dedup:1;
365 boolean_t zp_dedup_verify:1;
366 boolean_t zp_nopwrite:1;
367 boolean_t zp_brtwrite:1;
368 boolean_t zp_encrypt:1;
369 boolean_t zp_byteorder:1;
370 boolean_t zp_direct_write:1;
371 boolean_t zp_rewrite:1;
372 uint32_t zp_zpl_smallblk;
373 uint8_t zp_salt[ZIO_DATA_SALT_LEN];
374 uint8_t zp_iv[ZIO_DATA_IV_LEN];
375 uint8_t zp_mac[ZIO_DATA_MAC_LEN];
376 } zio_prop_t;
377
378 typedef struct zio_cksum_report zio_cksum_report_t;
379
380 typedef void zio_cksum_finish_f(zio_cksum_report_t *rep,
381 const abd_t *good_data);
382 typedef void zio_cksum_free_f(void *cbdata, size_t size);
383
384 struct zio_bad_cksum; /* defined in zio_checksum.h */
385 struct dnode_phys;
386 struct abd;
387
388 struct zio_cksum_report {
389 struct zio_cksum_report *zcr_next;
390 nvlist_t *zcr_ereport;
391 nvlist_t *zcr_detector;
392 void *zcr_cbdata;
393 size_t zcr_cbinfo; /* passed to zcr_free() */
394 uint64_t zcr_sector;
395 uint64_t zcr_align;
396 uint64_t zcr_length;
397 zio_cksum_finish_f *zcr_finish;
398 zio_cksum_free_f *zcr_free;
399
400 /* internal use only */
401 struct zio_bad_cksum *zcr_ckinfo; /* information from failure */
402 };
403
404 typedef struct zio_vsd_ops {
405 zio_done_func_t *vsd_free;
406 } zio_vsd_ops_t;
407
408 typedef struct zio_gang_node {
409 zio_gbh_phys_t *gn_gbh;
410 uint64_t gn_gangblocksize;
411 uint64_t gn_allocsize;
412 struct zio_gang_node *gn_child[];
413 } zio_gang_node_t;
414
415 typedef zio_t *zio_gang_issue_func_t(zio_t *zio, blkptr_t *bp,
416 zio_gang_node_t *gn, struct abd *data, uint64_t offset);
417
418 typedef void zio_transform_func_t(zio_t *zio, struct abd *data, uint64_t size);
419
420 typedef struct zio_transform {
421 struct abd *zt_orig_abd;
422 uint64_t zt_orig_size;
423 uint64_t zt_bufsize;
424 zio_transform_func_t *zt_transform;
425 struct zio_transform *zt_next;
426 } zio_transform_t;
427
428 typedef zio_t *zio_pipe_stage_t(zio_t *zio);
429
430 /*
431 * The io_post flags describe additional actions that a parent IO should
432 * consider or perform on behalf of a child. They are distinct from io_flags
433 * because the child must be able to propagate them to the parent. The normal
434 * io_flags are local to the zio, not protected by any lock, and not modifiable
435 * by children; the reexecute flags are protected by io_lock, modifiable by
436 * children, and always propagated -- even when ZIO_FLAG_DONT_PROPAGATE is set.
437 */
438 #define ZIO_POST_REEXECUTE (1 << 0)
439 #define ZIO_POST_SUSPEND (1 << 1)
440 #define ZIO_POST_DIO_CHKSUM_ERR (1 << 2)
441
442 /*
443 * The io_trim flags are used to specify the type of TRIM to perform. They
444 * only apply to ZIO_TYPE_TRIM zios are distinct from io_flags.
445 */
446 enum trim_flag {
447 ZIO_TRIM_SECURE = 1U << 0,
448 };
449
450 #ifdef METASLAB_TRACE
451 typedef struct zio_alloc_list {
452 list_t zal_list;
453 uint64_t zal_size;
454 } zio_alloc_list_t;
455 #define ZIO_ALLOC_LIST(zio) (&(zio)->io_alloc_list)
456 #else
457 typedef struct zio_alloc_list {
458 uint8_t zal_pad;
459 } zio_alloc_list_t;
460 #define ZIO_ALLOC_LIST(zio) NULL
461 #endif
462
463 typedef struct zio_link {
464 zio_t *zl_parent;
465 zio_t *zl_child;
466 list_node_t zl_parent_node;
467 list_node_t zl_child_node;
468 } zio_link_t;
469
470 enum zio_qstate {
471 ZIO_QS_NONE = 0,
472 ZIO_QS_QUEUED,
473 ZIO_QS_ACTIVE,
474 };
475
476 struct zio {
477 /* Core information about this I/O */
478 zbookmark_phys_t io_bookmark;
479 zio_prop_t io_prop;
480 zio_type_t io_type;
481 enum zio_child io_child_type;
482 enum trim_flag io_trim_flags;
483 zio_priority_t io_priority;
484 uint8_t io_post;
485 uint8_t io_state[ZIO_WAIT_TYPES];
486 uint64_t io_txg;
487 spa_t *io_spa;
488 blkptr_t *io_bp;
489 blkptr_t *io_bp_override;
490 blkptr_t io_bp_copy;
491 list_t io_parent_list;
492 list_t io_child_list;
493 zio_t *io_logical;
494 zio_transform_t *io_transform_stack;
495
496 /* Callback info */
497 zio_done_func_t *io_ready;
498 zio_done_func_t *io_children_ready;
499 zio_done_func_t *io_done;
500 void *io_private;
501 int64_t io_prev_space_delta; /* DMU private */
502 blkptr_t io_bp_orig;
503 /* io_lsize != io_orig_size iff this is a raw write */
504 uint64_t io_lsize;
505
506 /* Data represented by this I/O */
507 struct abd *io_abd;
508 struct abd *io_orig_abd;
509 uint64_t io_size;
510 uint64_t io_orig_size;
511
512 /* Stuff for the vdev stack */
513 vdev_t *io_vd;
514 void *io_vsd;
515 const zio_vsd_ops_t *io_vsd_ops;
516 metaslab_class_t *io_metaslab_class; /* dva throttle class */
517
518 enum zio_qstate io_queue_state; /* vdev queue state */
519 union {
520 list_node_t l;
521 avl_node_t a;
522 } io_queue_node ____cacheline_aligned; /* allocator and vdev queues */
523 avl_node_t io_offset_node; /* vdev offset queues */
524 uint64_t io_offset;
525 hrtime_t io_timestamp; /* submitted at */
526 hrtime_t io_queued_timestamp;
527 hrtime_t io_target_timestamp;
528 hrtime_t io_delta; /* vdev queue service delta */
529 hrtime_t io_delay; /* Device access time (disk or */
530 /* file). */
531 #ifdef METASLAB_TRACE
532 zio_alloc_list_t io_alloc_list;
533 #endif
534
535 /* Internal pipeline state */
536 zio_flag_t io_flags;
537 enum zio_stage io_stage;
538 enum zio_stage io_pipeline;
539 zio_flag_t io_orig_flags;
540 enum zio_stage io_orig_stage;
541 enum zio_stage io_orig_pipeline;
542 enum zio_stage io_pipeline_trace;
543 int io_error;
544 int io_child_error[ZIO_CHILD_TYPES];
545 uint64_t io_children[ZIO_CHILD_TYPES][ZIO_WAIT_TYPES];
546 uint64_t *io_stall;
547 zio_t *io_gang_leader;
548 zio_gang_node_t *io_gang_tree;
549 void *io_executor;
550 void *io_waiter;
551 void *io_bio;
552 kmutex_t io_lock;
553 kcondvar_t io_cv;
554 int io_allocator;
555
556 /* FMA state */
557 zio_cksum_report_t *io_cksum_report;
558 uint64_t io_ena;
559
560 /* Taskq dispatching state */
561 taskq_ent_t io_tqent;
562 };
563
564 enum blk_verify_flag {
565 BLK_VERIFY_ONLY,
566 BLK_VERIFY_LOG,
567 BLK_VERIFY_HALT
568 };
569
570 enum blk_config_flag {
571 BLK_CONFIG_HELD, // SCL_VDEV held for writer
572 BLK_CONFIG_NEEDED, // SCL_VDEV should be obtained for reader
573 BLK_CONFIG_NEEDED_TRY, // Try with SCL_VDEV for reader
574 BLK_CONFIG_SKIP, // skip checks which require SCL_VDEV
575 };
576
577 extern int zio_bookmark_compare(const void *, const void *);
578
579 extern zio_t *zio_null(zio_t *pio, spa_t *spa, vdev_t *vd,
580 zio_done_func_t *done, void *priv, zio_flag_t flags);
581
582 extern zio_t *zio_root(spa_t *spa,
583 zio_done_func_t *done, void *priv, zio_flag_t flags);
584
585 extern void zio_destroy(zio_t *zio);
586
587 extern zio_t *zio_read(zio_t *pio, spa_t *spa, const blkptr_t *bp,
588 struct abd *data, uint64_t lsize, zio_done_func_t *done, void *priv,
589 zio_priority_t priority, zio_flag_t flags, const zbookmark_phys_t *zb);
590
591 extern zio_t *zio_write(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
592 struct abd *data, uint64_t size, uint64_t psize, const zio_prop_t *zp,
593 zio_done_func_t *ready, zio_done_func_t *children_ready,
594 zio_done_func_t *done, void *priv, zio_priority_t priority,
595 zio_flag_t flags, const zbookmark_phys_t *zb);
596
597 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp,
598 struct abd *data, uint64_t size, zio_done_func_t *done, void *priv,
599 zio_priority_t priority, zio_flag_t flags, zbookmark_phys_t *zb);
600
601 extern void zio_write_override(zio_t *zio, blkptr_t *bp, int copies,
602 int gang_copies, boolean_t nopwrite, boolean_t brtwrite);
603
604 extern void zio_free(spa_t *spa, uint64_t txg, const blkptr_t *bp);
605
606 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg,
607 const blkptr_t *bp,
608 zio_done_func_t *done, void *priv, zio_flag_t flags);
609
610 extern zio_t *zio_trim(zio_t *pio, vdev_t *vd, uint64_t offset, uint64_t size,
611 zio_done_func_t *done, void *priv, zio_priority_t priority,
612 zio_flag_t flags, enum trim_flag trim_flags);
613
614 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
615 uint64_t size, struct abd *data, int checksum,
616 zio_done_func_t *done, void *priv, zio_priority_t priority,
617 zio_flag_t flags, boolean_t labels);
618
619 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset,
620 uint64_t size, struct abd *data, int checksum,
621 zio_done_func_t *done, void *priv, zio_priority_t priority,
622 zio_flag_t flags, boolean_t labels);
623
624 extern zio_t *zio_free_sync(zio_t *pio, spa_t *spa, uint64_t txg,
625 const blkptr_t *bp, zio_flag_t flags);
626
627 extern int zio_alloc_zil(spa_t *spa, objset_t *os, uint64_t txg,
628 blkptr_t *new_bp, uint64_t min_size, uint64_t max_size, boolean_t *slog,
629 boolean_t allow_larger);
630 extern void zio_flush(zio_t *zio, vdev_t *vd);
631 extern void zio_shrink(zio_t *zio, uint64_t size);
632
633 extern size_t zio_get_compression_max_size(enum zio_compress compress,
634 uint64_t gcd_alloc, uint64_t min_alloc, size_t s_len);
635 extern int zio_wait(zio_t *zio);
636 extern void zio_nowait(zio_t *zio);
637 extern void zio_execute(void *zio);
638 extern void zio_interrupt(void *zio);
639 extern void zio_delay_init(zio_t *zio);
640 extern void zio_delay_interrupt(zio_t *zio);
641 extern void zio_deadman(zio_t *zio, const char *tag);
642
643 extern zio_t *zio_walk_parents(zio_t *cio, zio_link_t **);
644 extern zio_t *zio_walk_children(zio_t *pio, zio_link_t **);
645 extern zio_t *zio_unique_parent(zio_t *cio);
646 extern void zio_add_child(zio_t *pio, zio_t *cio);
647
648 extern void *zio_buf_alloc(size_t size);
649 extern void zio_buf_free(void *buf, size_t size);
650 extern void *zio_data_buf_alloc(size_t size);
651 extern void zio_data_buf_free(void *buf, size_t size);
652
653 extern void zio_push_transform(zio_t *zio, struct abd *abd, uint64_t size,
654 uint64_t bufsize, zio_transform_func_t *transform);
655 extern void zio_pop_transforms(zio_t *zio);
656
657 extern void zio_resubmit_stage_async(void *);
658
659 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd,
660 uint64_t offset, struct abd *data, uint64_t size, int type,
661 zio_priority_t priority, zio_flag_t flags,
662 zio_done_func_t *done, void *priv);
663
664 extern zio_t *zio_vdev_delegated_io(vdev_t *vd, uint64_t offset,
665 struct abd *data, uint64_t size, zio_type_t type, zio_priority_t priority,
666 zio_flag_t flags, zio_done_func_t *done, void *priv);
667
668 extern void zio_vdev_io_bypass(zio_t *zio);
669 extern void zio_vdev_io_reissue(zio_t *zio);
670 extern void zio_vdev_io_redone(zio_t *zio);
671
672 extern void zio_change_priority(zio_t *pio, zio_priority_t priority);
673
674 extern void zio_checksum_verified(zio_t *zio);
675 extern void zio_dio_chksum_verify_error_report(zio_t *zio);
676 extern int zio_worst_error(int e1, int e2);
677
678 extern enum zio_checksum zio_checksum_select(enum zio_checksum child,
679 enum zio_checksum parent);
680 extern enum zio_checksum zio_checksum_dedup_select(spa_t *spa,
681 enum zio_checksum child, enum zio_checksum parent);
682 extern enum zio_compress zio_compress_select(spa_t *spa,
683 enum zio_compress child, enum zio_compress parent);
684 extern uint8_t zio_complevel_select(spa_t *spa, enum zio_compress compress,
685 uint8_t child, uint8_t parent);
686
687 extern void zio_suspend(spa_t *spa, zio_t *zio, zio_suspend_reason_t);
688 extern int zio_resume(spa_t *spa);
689 extern void zio_resume_wait(spa_t *spa);
690
691 extern int zfs_blkptr_verify(spa_t *spa, const blkptr_t *bp,
692 enum blk_config_flag blk_config, enum blk_verify_flag blk_verify);
693
694 /*
695 * Initial setup and teardown.
696 */
697 extern void zio_init(void);
698 extern void zio_fini(void);
699
700 /*
701 * Fault injection
702 */
703 struct zinject_record;
704 extern uint32_t zio_injection_enabled;
705 extern int zio_inject_fault(char *name, int flags, int *id,
706 struct zinject_record *record);
707 extern int zio_inject_list_next(int *id, char *name, size_t buflen,
708 struct zinject_record *record);
709 extern int zio_clear_fault(int id);
710 extern void zio_handle_panic_injection(spa_t *spa, const char *tag,
711 uint64_t type);
712 extern int zio_handle_decrypt_injection(spa_t *spa, const zbookmark_phys_t *zb,
713 uint64_t type, int error);
714 extern int zio_handle_fault_injection(zio_t *zio, int error);
715 extern int zio_handle_device_injection(vdev_t *vd, zio_t *zio, int error);
716 extern int zio_handle_device_injections(vdev_t *vd, zio_t *zio, int err1,
717 int err2);
718 extern int zio_handle_label_injection(zio_t *zio, int error);
719 extern void zio_handle_ignored_writes(zio_t *zio);
720 extern hrtime_t zio_handle_io_delay(zio_t *zio);
721 extern void zio_handle_import_delay(spa_t *spa, hrtime_t elapsed);
722 extern void zio_handle_export_delay(spa_t *spa, hrtime_t elapsed);
723 extern hrtime_t zio_handle_ready_delay(zio_t *zio);
724
725 /*
726 * Checksum ereport functions
727 */
728 extern int zfs_ereport_start_checksum(spa_t *spa, vdev_t *vd,
729 const zbookmark_phys_t *zb, struct zio *zio, uint64_t offset,
730 uint64_t length, struct zio_bad_cksum *info);
731 extern void zfs_ereport_finish_checksum(zio_cksum_report_t *report,
732 const abd_t *good_data, const abd_t *bad_data, boolean_t drop_if_identical);
733
734 extern void zfs_ereport_free_checksum(zio_cksum_report_t *report);
735
736 /* If we have the good data in hand, this function can be used */
737 extern int zfs_ereport_post_checksum(spa_t *spa, vdev_t *vd,
738 const zbookmark_phys_t *zb, struct zio *zio, uint64_t offset,
739 uint64_t length, const abd_t *good_data, const abd_t *bad_data,
740 struct zio_bad_cksum *info);
741
742 void zio_vsd_default_cksum_report(zio_t *zio, zio_cksum_report_t *zcr);
743 extern void zfs_ereport_snapshot_post(const char *subclass, spa_t *spa,
744 const char *name);
745
746 /* Called from spa_sync(), but primarily an injection handler */
747 extern void spa_handle_ignored_writes(spa_t *spa);
748
749 /* zbookmark_phys functions */
750 boolean_t zbookmark_subtree_completed(const struct dnode_phys *dnp,
751 const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
752 boolean_t zbookmark_subtree_tbd(const struct dnode_phys *dnp,
753 const zbookmark_phys_t *subtree_root, const zbookmark_phys_t *last_block);
754 int zbookmark_compare(uint16_t dbss1, uint8_t ibs1, uint16_t dbss2,
755 uint8_t ibs2, const zbookmark_phys_t *zb1, const zbookmark_phys_t *zb2);
756
757 #ifdef __cplusplus
758 }
759 #endif
760
761 #endif /* _ZIO_H */
762