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 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #ifndef _ZIO_H 27 #define _ZIO_H 28 29 #pragma ident "%Z%%M% %I% %E% SMI" 30 31 #include <sys/zfs_context.h> 32 #include <sys/spa.h> 33 #include <sys/txg.h> 34 #include <sys/avl.h> 35 #include <sys/dkio.h> 36 #include <sys/fs/zfs.h> 37 38 #ifdef __cplusplus 39 extern "C" { 40 #endif 41 42 #define ZBT_MAGIC 0x210da7ab10c7a11ULL /* zio data bloc tail */ 43 44 typedef struct zio_block_tail { 45 uint64_t zbt_magic; /* for validation, endianness */ 46 zio_cksum_t zbt_cksum; /* 256-bit checksum */ 47 } zio_block_tail_t; 48 49 /* 50 * Gang block headers are self-checksumming and contain an array 51 * of block pointers. 52 */ 53 #define SPA_GANGBLOCKSIZE SPA_MINBLOCKSIZE 54 #define SPA_GBH_NBLKPTRS ((SPA_GANGBLOCKSIZE - \ 55 sizeof (zio_block_tail_t)) / sizeof (blkptr_t)) 56 #define SPA_GBH_FILLER ((SPA_GANGBLOCKSIZE - \ 57 sizeof (zio_block_tail_t) - \ 58 (SPA_GBH_NBLKPTRS * sizeof (blkptr_t))) /\ 59 sizeof (uint64_t)) 60 61 #define ZIO_GET_DVA(zio) (&(zio)->io_bp->blk_dva[(zio)->io_dva_index]) 62 #define ZIO_GET_IOSIZE(zio) \ 63 (DVA_GET_GANG(ZIO_GET_DVA(zio)) ? \ 64 SPA_GANGBLOCKSIZE : BP_GET_PSIZE((zio)->io_bp)) 65 66 typedef struct zio_gbh { 67 blkptr_t zg_blkptr[SPA_GBH_NBLKPTRS]; 68 uint64_t zg_filler[SPA_GBH_FILLER]; 69 zio_block_tail_t zg_tail; 70 } zio_gbh_phys_t; 71 72 enum zio_checksum { 73 ZIO_CHECKSUM_INHERIT = 0, 74 ZIO_CHECKSUM_ON, 75 ZIO_CHECKSUM_OFF, 76 ZIO_CHECKSUM_LABEL, 77 ZIO_CHECKSUM_GANG_HEADER, 78 ZIO_CHECKSUM_ZILOG, 79 ZIO_CHECKSUM_FLETCHER_2, 80 ZIO_CHECKSUM_FLETCHER_4, 81 ZIO_CHECKSUM_SHA256, 82 ZIO_CHECKSUM_FUNCTIONS 83 }; 84 85 #define ZIO_CHECKSUM_ON_VALUE ZIO_CHECKSUM_FLETCHER_2 86 #define ZIO_CHECKSUM_DEFAULT ZIO_CHECKSUM_ON 87 88 enum zio_compress { 89 ZIO_COMPRESS_INHERIT = 0, 90 ZIO_COMPRESS_ON, 91 ZIO_COMPRESS_OFF, 92 ZIO_COMPRESS_LZJB, 93 ZIO_COMPRESS_FUNCTIONS 94 }; 95 96 #define ZIO_COMPRESS_ON_VALUE ZIO_COMPRESS_LZJB 97 #define ZIO_COMPRESS_DEFAULT ZIO_COMPRESS_OFF 98 99 #define ZIO_PRIORITY_NOW (zio_priority_table[0]) 100 #define ZIO_PRIORITY_SYNC_READ (zio_priority_table[1]) 101 #define ZIO_PRIORITY_SYNC_WRITE (zio_priority_table[2]) 102 #define ZIO_PRIORITY_ASYNC_READ (zio_priority_table[3]) 103 #define ZIO_PRIORITY_ASYNC_WRITE (zio_priority_table[4]) 104 #define ZIO_PRIORITY_FREE (zio_priority_table[5]) 105 #define ZIO_PRIORITY_CACHE_FILL (zio_priority_table[6]) 106 #define ZIO_PRIORITY_LOG_WRITE (zio_priority_table[7]) 107 #define ZIO_PRIORITY_RESILVER (zio_priority_table[8]) 108 #define ZIO_PRIORITY_SCRUB (zio_priority_table[9]) 109 #define ZIO_PRIORITY_TABLE_SIZE 10 110 111 #define ZIO_FLAG_MUSTSUCCEED 0x00000 112 #define ZIO_FLAG_CANFAIL 0x00001 113 #define ZIO_FLAG_FAILFAST 0x00002 114 #define ZIO_FLAG_CONFIG_HELD 0x00004 115 116 #define ZIO_FLAG_DONT_CACHE 0x00010 117 #define ZIO_FLAG_DONT_QUEUE 0x00020 118 #define ZIO_FLAG_DONT_PROPAGATE 0x00040 119 #define ZIO_FLAG_DONT_RETRY 0x00080 120 121 #define ZIO_FLAG_PHYSICAL 0x00100 122 #define ZIO_FLAG_IO_BYPASS 0x00200 123 #define ZIO_FLAG_IO_REPAIR 0x00400 124 #define ZIO_FLAG_SPECULATIVE 0x00800 125 126 #define ZIO_FLAG_RESILVER 0x01000 127 #define ZIO_FLAG_SCRUB 0x02000 128 #define ZIO_FLAG_SUBBLOCK 0x04000 129 130 #define ZIO_FLAG_NOBOOKMARK 0x10000 131 132 #define ZIO_FLAG_GANG_INHERIT \ 133 (ZIO_FLAG_CANFAIL | \ 134 ZIO_FLAG_FAILFAST | \ 135 ZIO_FLAG_CONFIG_HELD | \ 136 ZIO_FLAG_DONT_RETRY | \ 137 ZIO_FLAG_IO_REPAIR | \ 138 ZIO_FLAG_SPECULATIVE | \ 139 ZIO_FLAG_RESILVER | \ 140 ZIO_FLAG_SCRUB) 141 142 #define ZIO_FLAG_VDEV_INHERIT \ 143 (ZIO_FLAG_GANG_INHERIT | \ 144 ZIO_FLAG_DONT_CACHE | \ 145 ZIO_FLAG_PHYSICAL) 146 147 /* 148 * We'll take the unused errno 'EBADE' (from the Convergent graveyard) 149 * to indicate checksum errors. 150 */ 151 #define ECKSUM EBADE 152 153 typedef struct zio zio_t; 154 typedef void zio_done_func_t(zio_t *zio); 155 typedef struct zio_transform zio_transform_t; 156 157 extern uint8_t zio_priority_table[ZIO_PRIORITY_TABLE_SIZE]; 158 extern char *zio_type_name[ZIO_TYPES]; 159 160 /* 161 * A bookmark is a four-tuple <objset, object, level, blkid> that uniquely 162 * identifies any block in the pool. By convention, the meta-objset (MOS) 163 * is objset 0, the meta-dnode is object 0, the root block (osphys_t) is 164 * level -1 of the meta-dnode, and intent log blocks (which are chained 165 * off the root block) have blkid == sequence number. In summary: 166 * 167 * mos is objset 0 168 * meta-dnode is object 0 169 * root block is <objset, 0, -1, 0> 170 * intent log is <objset, 0, -1, ZIL sequence number> 171 * 172 * Note: this structure is called a bookmark because its first purpose was 173 * to remember where to resume a pool-wide traverse. The absolute ordering 174 * for block visitation during traversal is defined in compare_bookmark(). 175 * 176 * Note: this structure is passed between userland and the kernel. 177 * Therefore it must not change size or alignment between 32/64 bit 178 * compilation options. 179 */ 180 typedef struct zbookmark { 181 uint64_t zb_objset; 182 uint64_t zb_object; 183 int64_t zb_level; 184 uint64_t zb_blkid; 185 } zbookmark_t; 186 187 struct zio { 188 /* Core information about this I/O */ 189 zio_t *io_parent; 190 zio_t *io_root; 191 spa_t *io_spa; 192 zbookmark_t io_bookmark; 193 int io_checksum; 194 int io_compress; 195 int io_dva_index; 196 uint64_t io_txg; 197 blkptr_t *io_bp; 198 blkptr_t io_bp_copy; 199 zio_t *io_child; 200 zio_t *io_sibling_prev; 201 zio_t *io_sibling_next; 202 zio_transform_t *io_transform_stack; 203 zio_t *io_logical; 204 205 /* Callback info */ 206 zio_done_func_t *io_done; 207 void *io_private; 208 blkptr_t io_bp_orig; 209 210 /* Data represented by this I/O */ 211 void *io_data; 212 uint64_t io_size; 213 214 /* Stuff for the vdev stack */ 215 vdev_t *io_vd; 216 void *io_vsd; 217 uint64_t io_offset; 218 uint64_t io_deadline; 219 uint64_t io_timestamp; 220 avl_node_t io_offset_node; 221 avl_node_t io_deadline_node; 222 avl_tree_t *io_vdev_tree; 223 zio_t *io_delegate_list; 224 zio_t *io_delegate_next; 225 226 /* Internal pipeline state */ 227 int io_flags; 228 uint8_t io_type; 229 uint8_t io_stage; 230 uint8_t io_stalled; 231 uint8_t io_priority; 232 struct dk_callback io_dk_callback; 233 int io_cmd; 234 int io_retries; 235 int io_error; 236 uint32_t io_numerrors; 237 uint32_t io_pipeline; 238 uint32_t io_async_stages; 239 uint64_t io_children_notready; 240 uint64_t io_children_notdone; 241 void *io_waiter; 242 kmutex_t io_lock; 243 kcondvar_t io_cv; 244 245 /* FMA state */ 246 uint64_t io_ena; 247 }; 248 249 extern zio_t *zio_null(zio_t *pio, spa_t *spa, 250 zio_done_func_t *done, void *private, int flags); 251 252 extern zio_t *zio_root(spa_t *spa, 253 zio_done_func_t *done, void *private, int flags); 254 255 extern zio_t *zio_read(zio_t *pio, spa_t *spa, blkptr_t *bp, void *data, 256 uint64_t size, zio_done_func_t *done, void *private, 257 int priority, int flags, zbookmark_t *zb); 258 259 extern zio_t *zio_write(zio_t *pio, spa_t *spa, int checksum, int compress, 260 uint64_t txg, blkptr_t *bp, void *data, uint64_t size, 261 zio_done_func_t *done, void *private, int priority, int flags, 262 zbookmark_t *zb); 263 264 extern zio_t *zio_rewrite(zio_t *pio, spa_t *spa, int checksum, 265 uint64_t txg, blkptr_t *bp, void *data, uint64_t size, 266 zio_done_func_t *done, void *private, int priority, int flags, 267 zbookmark_t *zb); 268 269 extern zio_t *zio_free(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, 270 zio_done_func_t *done, void *private); 271 272 extern zio_t *zio_claim(zio_t *pio, spa_t *spa, uint64_t txg, blkptr_t *bp, 273 zio_done_func_t *done, void *private); 274 275 extern zio_t *zio_ioctl(zio_t *pio, spa_t *spa, vdev_t *vd, int cmd, 276 zio_done_func_t *done, void *private, int priority, int flags); 277 278 extern zio_t *zio_read_phys(zio_t *pio, vdev_t *vd, uint64_t offset, 279 uint64_t size, void *data, int checksum, 280 zio_done_func_t *done, void *private, int priority, int flags); 281 282 extern zio_t *zio_write_phys(zio_t *pio, vdev_t *vd, uint64_t offset, 283 uint64_t size, void *data, int checksum, 284 zio_done_func_t *done, void *private, int priority, int flags); 285 286 extern int zio_alloc_blk(spa_t *spa, int checksum, uint64_t size, 287 blkptr_t *bp, uint64_t txg); 288 extern void zio_free_blk(spa_t *spa, blkptr_t *bp, uint64_t txg); 289 290 extern int zio_wait(zio_t *zio); 291 extern void zio_nowait(zio_t *zio); 292 293 extern void *zio_buf_alloc(size_t size); 294 extern void zio_buf_free(void *buf, size_t size); 295 296 /* 297 * Move an I/O to the next stage of the pipeline and execute that stage. 298 * There's no locking on io_stage because there's no legitimate way for 299 * multiple threads to be attempting to process the same I/O. 300 */ 301 extern void zio_next_stage(zio_t *zio); 302 extern void zio_next_stage_async(zio_t *zio); 303 extern void zio_wait_children_done(zio_t *zio); 304 305 /* 306 * Delegate I/O to a child vdev. 307 */ 308 extern zio_t *zio_vdev_child_io(zio_t *zio, blkptr_t *bp, vdev_t *vd, 309 uint64_t offset, void *data, uint64_t size, int type, int priority, 310 int flags, zio_done_func_t *done, void *private); 311 312 extern void zio_vdev_io_bypass(zio_t *zio); 313 extern void zio_vdev_io_reissue(zio_t *zio); 314 extern void zio_vdev_io_redone(zio_t *zio); 315 316 extern void zio_checksum_verified(zio_t *zio); 317 extern void zio_set_gang_verifier(zio_t *zio, zio_cksum_t *zcp); 318 319 extern uint8_t zio_checksum_select(uint8_t child, uint8_t parent); 320 extern uint8_t zio_compress_select(uint8_t child, uint8_t parent); 321 322 boolean_t zio_should_retry(zio_t *zio); 323 324 /* 325 * Initial setup and teardown. 326 */ 327 extern void zio_init(void); 328 extern void zio_fini(void); 329 330 /* 331 * Fault injection 332 */ 333 struct zinject_record; 334 extern uint32_t zio_injection_enabled; 335 extern int zio_inject_fault(char *name, int flags, int *id, 336 struct zinject_record *record); 337 extern int zio_inject_list_next(int *id, char *name, size_t buflen, 338 struct zinject_record *record); 339 extern int zio_clear_fault(int id); 340 extern int zio_handle_fault_injection(zio_t *zio, int error); 341 extern int zio_handle_device_injection(vdev_t *vd, int error); 342 343 #ifdef __cplusplus 344 } 345 #endif 346 347 #endif /* _ZIO_H */ 348