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 /* 23 * Copyright 2007 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 #ifndef _ZIO_H 28 #define _ZIO_H 29 30 #pragma ident "%Z%%M% %I% %E% SMI" 31 32 #include <sys/zfs_context.h> 33 #include <sys/spa.h> 34 #include <sys/txg.h> 35 #include <sys/avl.h> 36 #include <sys/dkio.h> 37 #include <sys/fs/zfs.h> 38 #include <sys/zio_impl.h> 39 40 #ifdef __cplusplus 41 extern "C" { 42 #endif 43 44 #define ZBT_MAGIC 0x210da7ab10c7a11ULL /* zio data bloc tail */ 45 46 typedef struct zio_block_tail { 47 uint64_t zbt_magic; /* for validation, endianness */ 48 zio_cksum_t zbt_cksum; /* 256-bit checksum */ 49 } zio_block_tail_t; 50 51 /* 52 * Gang block headers are self-checksumming and contain an array 53 * of block pointers. 54 */ 55 #define SPA_GANGBLOCKSIZE SPA_MINBLOCKSIZE 56 #define SPA_GBH_NBLKPTRS ((SPA_GANGBLOCKSIZE - \ 57 sizeof (zio_block_tail_t)) / sizeof (blkptr_t)) 58 #define SPA_GBH_FILLER ((SPA_GANGBLOCKSIZE - \ 59 sizeof (zio_block_tail_t) - \ 60 (SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\ 61 sizeof (uint64_t)) 62 63 #define ZIO_GET_IOSIZE(zio) \ 64 (BP_IS_GANG((zio)->io_bp) ? \ 65 SPA_GANGBLOCKSIZE : BP_GET_PSIZE((zio)->io_bp)) 66 67 typedef struct zio_gbh { 68 blkptr_t zg_blkptr[SPA_GBH_NBLKPTRS]; 69 uint64_t zg_filler[SPA_GBH_FILLER]; 70 zio_block_tail_t zg_tail; 71 } zio_gbh_phys_t; 72 73 enum zio_checksum { 74 ZIO_CHECKSUM_INHERIT = 0, 75 ZIO_CHECKSUM_ON, 76 ZIO_CHECKSUM_OFF, 77 ZIO_CHECKSUM_LABEL, 78 ZIO_CHECKSUM_GANG_HEADER, 79 ZIO_CHECKSUM_ZILOG, 80 ZIO_CHECKSUM_FLETCHER_2, 81 ZIO_CHECKSUM_FLETCHER_4, 82 ZIO_CHECKSUM_SHA256, 83 ZIO_CHECKSUM_FUNCTIONS 84 }; 85 86 #define ZIO_CHECKSUM_ON_VALUE ZIO_CHECKSUM_FLETCHER_2 87 #define ZIO_CHECKSUM_DEFAULT ZIO_CHECKSUM_ON 88 89 enum zio_compress { 90 ZIO_COMPRESS_INHERIT = 0, 91 ZIO_COMPRESS_ON, 92 ZIO_COMPRESS_OFF, 93 ZIO_COMPRESS_LZJB, 94 ZIO_COMPRESS_EMPTY, 95 ZIO_COMPRESS_GZIP_1, 96 ZIO_COMPRESS_GZIP_2, 97 ZIO_COMPRESS_GZIP_3, 98 ZIO_COMPRESS_GZIP_4, 99 ZIO_COMPRESS_GZIP_5, 100 ZIO_COMPRESS_GZIP_6, 101 ZIO_COMPRESS_GZIP_7, 102 ZIO_COMPRESS_GZIP_8, 103 ZIO_COMPRESS_GZIP_9, 104 ZIO_COMPRESS_FUNCTIONS 105 }; 106 107 #define ZIO_COMPRESS_ON_VALUE ZIO_COMPRESS_LZJB 108 #define ZIO_COMPRESS_DEFAULT ZIO_COMPRESS_OFF 109 110 #define ZIO_FAILURE_MODE_WAIT 0 111 #define ZIO_FAILURE_MODE_CONTINUE 1 112 #define ZIO_FAILURE_MODE_PANIC 2 113 114 #define ZIO_PRIORITY_NOW (zio_priority_table[0]) 115 #define ZIO_PRIORITY_SYNC_READ (zio_priority_table[1]) 116 #define ZIO_PRIORITY_SYNC_WRITE (zio_priority_table[2]) 117 #define ZIO_PRIORITY_ASYNC_READ (zio_priority_table[3]) 118 #define ZIO_PRIORITY_ASYNC_WRITE (zio_priority_table[4]) 119 #define ZIO_PRIORITY_FREE (zio_priority_table[5]) 120 #define ZIO_PRIORITY_CACHE_FILL (zio_priority_table[6]) 121 #define ZIO_PRIORITY_LOG_WRITE (zio_priority_table[7]) 122 #define ZIO_PRIORITY_RESILVER (zio_priority_table[8]) 123 #define ZIO_PRIORITY_SCRUB (zio_priority_table[9]) 124 #define ZIO_PRIORITY_TABLE_SIZE 10 125 126 #define ZIO_FLAG_MUSTSUCCEED 0x00000 127 #define ZIO_FLAG_CANFAIL 0x00001 128 #define ZIO_FLAG_FAILFAST 0x00002 129 #define ZIO_FLAG_CONFIG_HELD 0x00004 130 #define ZIO_FLAG_CONFIG_GRABBED 0x00008 131 132 #define ZIO_FLAG_DONT_CACHE 0x00010 133 #define ZIO_FLAG_DONT_QUEUE 0x00020 134 #define ZIO_FLAG_DONT_PROPAGATE 0x00040 135 #define ZIO_FLAG_DONT_RETRY 0x00080 136 137 #define ZIO_FLAG_PHYSICAL 0x00100 138 #define ZIO_FLAG_IO_BYPASS 0x00200 139 #define ZIO_FLAG_IO_REPAIR 0x00400 140 #define ZIO_FLAG_SPECULATIVE 0x00800 141 142 #define ZIO_FLAG_RESILVER 0x01000 143 #define ZIO_FLAG_SCRUB 0x02000 144 #define ZIO_FLAG_SCRUB_THREAD 0x04000 145 #define ZIO_FLAG_SUBBLOCK 0x08000 146 147 #define ZIO_FLAG_NOBOOKMARK 0x10000 148 #define ZIO_FLAG_USER 0x20000 149 #define ZIO_FLAG_METADATA 0x40000 150 #define ZIO_FLAG_WRITE_RETRY 0x80000 151 152 #define ZIO_FLAG_GANG_INHERIT \ 153 (ZIO_FLAG_CANFAIL | \ 154 ZIO_FLAG_FAILFAST | \ 155 ZIO_FLAG_CONFIG_HELD | \ 156 ZIO_FLAG_DONT_RETRY | \ 157 ZIO_FLAG_IO_REPAIR | \ 158 ZIO_FLAG_SPECULATIVE | \ 159 ZIO_FLAG_RESILVER | \ 160 ZIO_FLAG_SCRUB | \ 161 ZIO_FLAG_SCRUB_THREAD | \ 162 ZIO_FLAG_USER | \ 163 ZIO_FLAG_METADATA) 164 165 #define ZIO_FLAG_VDEV_INHERIT \ 166 (ZIO_FLAG_GANG_INHERIT | \ 167 ZIO_FLAG_DONT_CACHE | \ 168 ZIO_FLAG_PHYSICAL) 169 170 /* 171 * We'll take the unused errno 'EBADE' (from the Convergent graveyard) 172 * to indicate checksum errors. 173 */ 174 #define ECKSUM EBADE 175 176 typedef struct zio zio_t; 177 typedef void zio_done_func_t(zio_t *zio); 178 179 extern uint8_t zio_priority_table[ZIO_PRIORITY_TABLE_SIZE]; 180 extern char *zio_type_name[ZIO_TYPES]; 181 182 /* 183 * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely 184 * identifies any block in the pool. By convention, the meta-objset (MOS) 185 * is objset 0, the meta-dnode is object 0, the root block (osphys_t) is 186 * level -1 of the meta-dnode, and intent log blocks (which are chained 187 * off the root block) have blkid == sequence number. In summary: 188 * 189 * mos is objset 0 190 * meta-dnode is object 0 191 * root block is <objset, 0, -1, 0> 192 * intent log is <objset, 0, -1, ZIL sequence number> 193 * 194 * Note: this structure is called a bookmark because its first purpose was 195 * to remember where to resume a pool-wide traverse. The absolute ordering 196 * for block visitation during traversal is defined in compare_bookmark(). 197 * 198 * Note: this structure is passed between userland and the kernel. 199 * Therefore it must not change size or alignment between 32/64 bit 200 * compilation options. 201 */ 202 typedef struct zbookmark { 203 uint64_t zb_objset; 204 uint64_t zb_object; 205 int64_t zb_level; 206 uint64_t zb_blkid; 207 } zbookmark_t; 208 209 struct zio { 210 /* Core information about this I/O */ 211 zio_t *io_parent; 212 zio_t *io_root; 213 spa_t *io_spa; 214 zbookmark_t io_bookmark; 215 enum zio_checksum io_checksum; 216 enum zio_compress io_compress; 217 int io_ndvas; 218 uint64_t io_txg; 219 blkptr_t *io_bp; 220 blkptr_t io_bp_copy; 221 zio_t *io_child; 222 zio_t *io_sibling_prev; 223 zio_t *io_sibling_next; 224 zio_transform_t *io_transform_stack; 225 zio_t *io_logical; 226 list_node_t zio_link_node; 227 228 /* Callback info */ 229 zio_done_func_t *io_ready; 230 zio_done_func_t *io_done; 231 void *io_private; 232 blkptr_t io_bp_orig; 233 234 /* Data represented by this I/O */ 235 void *io_data; 236 uint64_t io_size; 237 238 /* Stuff for the vdev stack */ 239 vdev_t *io_vd; 240 void *io_vsd; 241 uint64_t io_offset; 242 uint64_t io_deadline; 243 uint64_t io_timestamp; 244 avl_node_t io_offset_node; 245 avl_node_t io_deadline_node; 246 avl_tree_t *io_vdev_tree; 247 zio_t *io_delegate_list; 248 zio_t *io_delegate_next; 249 250 /* Internal pipeline state */ 251 int io_flags; 252 int io_orig_flags; 253 enum zio_type io_type; 254 enum zio_stage io_stage; 255 enum zio_stage io_orig_stage; 256 uint8_t io_stalled; 257 uint8_t io_priority; 258 struct dk_callback io_dk_callback; 259 int io_cmd; 260 int io_retries; 261 int io_error; 262 uint32_t io_numerrors; 263 uint32_t io_pipeline; 264 uint32_t io_orig_pipeline; 265 uint32_t io_async_stages; 266 uint64_t io_children_notready; 267 uint64_t io_children_notdone; 268 void *io_waiter; 269 kmutex_t io_lock; 270 kcondvar_t io_cv; 271 272 /* FMA state */ 273 uint64_t io_ena; 274 }; 275 276 extern zio_t *zio_null(zio_t *pio, spa_t *spa, 277 zio_done_func_t *done, void *private, int flags); 278 279 extern zio_t *zio_root(spa_t *spa, 280 zio_done_func_t *done, void *private, int flags); 281 282 extern zio_t *zio_read(zio_t *pio, spa_t *spa, blkptr_t *bp, void *data, 283 uint64_t size, zio_done_func_t *done, void *private, 284 int priority, int flags, zbookmark_t *zb); 285 286 extern zio_t *zio_write(zio_t *pio, spa_t *spa, int checksum, int compress, 287 int ncopies, uint64_t txg, blkptr_t *bp, void *data, uint64_t size, 288 zio_done_func_t *ready, zio_done_func_t *done, void *private, int priority, 289 int flags, zbookmark_t *zb); 290 291 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, int checksum, 292 uint64_t txg, blkptr_t *bp, void *data, uint64_t size, 293 zio_done_func_t *done, void *private, int priority, int flags, 294 zbookmark_t *zb); 295 296 extern zio_t *zio_free(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, 297 zio_done_func_t *done, void *private); 298 299 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, 300 zio_done_func_t *done, void *private); 301 302 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, 303 zio_done_func_t *done, void *private, int priority, int flags); 304 305 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, 306 uint64_t size, void *data, int checksum, 307 zio_done_func_t *done, void *private, int priority, int flags); 308 309 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, 310 uint64_t size, void *data, int checksum, 311 zio_done_func_t *done, void *private, int priority, int flags); 312 313 extern int zio_alloc_blk(spa_t *spa, uint64_t size, blkptr_t *new_bp, 314 blkptr_t *old_bp, uint64_t txg); 315 extern void zio_free_blk(spa_t *spa, blkptr_t *bp, uint64_t txg); 316 extern void zio_flush_vdev(spa_t *spa, uint64_t vdev, zio_t **zio); 317 318 extern int zio_wait(zio_t *zio); 319 extern void zio_nowait(zio_t *zio); 320 321 extern void *zio_buf_alloc(size_t size); 322 extern void zio_buf_free(void *buf, size_t size); 323 extern void *zio_data_buf_alloc(size_t size); 324 extern void zio_data_buf_free(void *buf, size_t size); 325 326 /* 327 * Move an I/O to the next stage of the pipeline and execute that stage. 328 * There's no locking on io_stage because there's no legitimate way for 329 * multiple threads to be attempting to process the same I/O. 330 */ 331 extern void zio_next_stage(zio_t *zio); 332 extern void zio_next_stage_async(zio_t *zio); 333 extern void zio_resubmit_stage_async(void *); 334 extern void zio_wait_children_done(zio_t *zio); 335 336 /* 337 * Delegate I/O to a child vdev. 338 */ 339 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd, 340 uint64_t offset, void *data, uint64_t size, int type, int priority, 341 int flags, zio_done_func_t *done, void *private); 342 343 extern void zio_vdev_io_bypass(zio_t *zio); 344 extern void zio_vdev_io_reissue(zio_t *zio); 345 extern void zio_vdev_io_redone(zio_t *zio); 346 347 extern void zio_checksum_verified(zio_t *zio); 348 extern void zio_set_gang_verifier(zio_t *zio, zio_cksum_t *zcp); 349 350 extern uint8_t zio_checksum_select(uint8_t child, uint8_t parent); 351 extern uint8_t zio_compress_select(uint8_t child, uint8_t parent); 352 353 extern boolean_t zio_should_retry(zio_t *zio); 354 extern int zio_vdev_resume_io(spa_t *); 355 356 /* 357 * Initial setup and teardown. 358 */ 359 extern void zio_init(void); 360 extern void zio_fini(void); 361 362 /* 363 * Fault injection 364 */ 365 struct zinject_record; 366 extern uint32_t zio_injection_enabled; 367 extern int zio_inject_fault(char *name, int flags, int *id, 368 struct zinject_record *record); 369 extern int zio_inject_list_next(int *id, char *name, size_t buflen, 370 struct zinject_record *record); 371 extern int zio_clear_fault(int id); 372 extern int zio_handle_fault_injection(zio_t *zio, int error); 373 extern int zio_handle_device_injection(vdev_t *vd, int error); 374 375 #ifdef __cplusplus 376 } 377 #endif 378 379 #endif /* _ZIO_H */ 380