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 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #ifndef _SYS_ZIL_H 27 #define _SYS_ZIL_H 28 29 #include <sys/types.h> 30 #include <sys/spa.h> 31 #include <sys/zio.h> 32 #include <sys/dmu.h> 33 34 #ifdef __cplusplus 35 extern "C" { 36 #endif 37 38 /* 39 * Intent log format: 40 * 41 * Each objset has its own intent log. The log header (zil_header_t) 42 * for objset N's intent log is kept in the Nth object of the SPA's 43 * intent_log objset. The log header points to a chain of log blocks, 44 * each of which contains log records (i.e., transactions) followed by 45 * a log block trailer (zil_trailer_t). The format of a log record 46 * depends on the record (or transaction) type, but all records begin 47 * with a common structure that defines the type, length, and txg. 48 */ 49 50 /* 51 * Intent log header - this on disk structure holds fields to manage 52 * the log. All fields are 64 bit to easily handle cross architectures. 53 */ 54 typedef struct zil_header { 55 uint64_t zh_claim_txg; /* txg in which log blocks were claimed */ 56 uint64_t zh_replay_seq; /* highest replayed sequence number */ 57 blkptr_t zh_log; /* log chain */ 58 uint64_t zh_claim_seq; /* highest claimed sequence number */ 59 uint64_t zh_flags; /* header flags */ 60 uint64_t zh_pad[4]; 61 } zil_header_t; 62 63 /* 64 * zh_flags bit settings 65 */ 66 #define ZIL_REPLAY_NEEDED 0x1 /* replay needed - internal only */ 67 68 /* 69 * Log block trailer - structure at the end of the header and each log block 70 * 71 * The zit_bt contains a zbt_cksum which for the intent log is 72 * the sequence number of this log block. A seq of 0 is invalid. 73 * The zbt_cksum is checked by the SPA against the sequence 74 * number passed in the blk_cksum field of the blkptr_t 75 */ 76 typedef struct zil_trailer { 77 uint64_t zit_pad; 78 blkptr_t zit_next_blk; /* next block in chain */ 79 uint64_t zit_nused; /* bytes in log block used */ 80 zio_block_tail_t zit_bt; /* block trailer */ 81 } zil_trailer_t; 82 83 #define ZIL_MIN_BLKSZ 4096ULL 84 #define ZIL_MAX_BLKSZ SPA_MAXBLOCKSIZE 85 #define ZIL_BLK_DATA_SZ(lwb) ((lwb)->lwb_sz - sizeof (zil_trailer_t)) 86 87 /* 88 * The words of a log block checksum. 89 */ 90 #define ZIL_ZC_GUID_0 0 91 #define ZIL_ZC_GUID_1 1 92 #define ZIL_ZC_OBJSET 2 93 #define ZIL_ZC_SEQ 3 94 95 typedef enum zil_create { 96 Z_FILE, 97 Z_DIR, 98 Z_XATTRDIR, 99 } zil_create_t; 100 101 /* 102 * size of xvattr log section. 103 * its composed of lr_attr_t + xvattr bitmap + 2 64 bit timestamps 104 * for create time and a single 64 bit integer for all of the attributes, 105 * and 4 64 bit integers (32 bytes) for the scanstamp. 106 * 107 */ 108 109 #define ZIL_XVAT_SIZE(mapsize) \ 110 sizeof (lr_attr_t) + (sizeof (uint32_t) * (mapsize - 1)) + \ 111 (sizeof (uint64_t) * 7) 112 113 /* 114 * Size of ACL in log. The ACE data is padded out to properly align 115 * on 8 byte boundary. 116 */ 117 118 #define ZIL_ACE_LENGTH(x) (roundup(x, sizeof (uint64_t))) 119 120 /* 121 * Intent log transaction types and record structures 122 */ 123 #define TX_CREATE 1 /* Create file */ 124 #define TX_MKDIR 2 /* Make directory */ 125 #define TX_MKXATTR 3 /* Make XATTR directory */ 126 #define TX_SYMLINK 4 /* Create symbolic link to a file */ 127 #define TX_REMOVE 5 /* Remove file */ 128 #define TX_RMDIR 6 /* Remove directory */ 129 #define TX_LINK 7 /* Create hard link to a file */ 130 #define TX_RENAME 8 /* Rename a file */ 131 #define TX_WRITE 9 /* File write */ 132 #define TX_TRUNCATE 10 /* Truncate a file */ 133 #define TX_SETATTR 11 /* Set file attributes */ 134 #define TX_ACL_V0 12 /* Set old formatted ACL */ 135 #define TX_ACL 13 /* Set ACL */ 136 #define TX_CREATE_ACL 14 /* create with ACL */ 137 #define TX_CREATE_ATTR 15 /* create + attrs */ 138 #define TX_CREATE_ACL_ATTR 16 /* create with ACL + attrs */ 139 #define TX_MKDIR_ACL 17 /* mkdir with ACL */ 140 #define TX_MKDIR_ATTR 18 /* mkdir with attr */ 141 #define TX_MKDIR_ACL_ATTR 19 /* mkdir with ACL + attrs */ 142 #define TX_WRITE2 20 /* dmu_sync EALREADY write */ 143 #define TX_MAX_TYPE 21 /* Max transaction type */ 144 145 /* 146 * The transactions for mkdir, symlink, remove, rmdir, link, and rename 147 * may have the following bit set, indicating the original request 148 * specified case-insensitive handling of names. 149 */ 150 #define TX_CI ((uint64_t)0x1 << 63) /* case-insensitive behavior requested */ 151 152 /* 153 * Format of log records. 154 * The fields are carefully defined to allow them to be aligned 155 * and sized the same on sparc & intel architectures. 156 * Each log record has a common structure at the beginning. 157 * 158 * Note, lrc_seq holds two different sequence numbers. Whilst in memory 159 * it contains the transaction sequence number. The log record on 160 * disk holds the sequence number of all log records which is used to 161 * ensure we don't replay the same record. The two sequence numbers are 162 * different because the transactions can now be pushed out of order. 163 */ 164 typedef struct { /* common log record header */ 165 uint64_t lrc_txtype; /* intent log transaction type */ 166 uint64_t lrc_reclen; /* transaction record length */ 167 uint64_t lrc_txg; /* dmu transaction group number */ 168 uint64_t lrc_seq; /* see comment above */ 169 } lr_t; 170 171 /* 172 * Handle option extended vattr attributes. 173 * 174 * Whenever new attributes are added the version number 175 * will need to be updated as will code in 176 * zfs_log.c and zfs_replay.c 177 */ 178 typedef struct { 179 uint32_t lr_attr_masksize; /* number of elements in array */ 180 uint32_t lr_attr_bitmap; /* First entry of array */ 181 /* remainder of array and any additional fields */ 182 } lr_attr_t; 183 184 /* 185 * log record for creates without optional ACL. 186 * This log record does support optional xvattr_t attributes. 187 */ 188 typedef struct { 189 lr_t lr_common; /* common portion of log record */ 190 uint64_t lr_doid; /* object id of directory */ 191 uint64_t lr_foid; /* object id of created file object */ 192 uint64_t lr_mode; /* mode of object */ 193 uint64_t lr_uid; /* uid of object */ 194 uint64_t lr_gid; /* gid of object */ 195 uint64_t lr_gen; /* generation (txg of creation) */ 196 uint64_t lr_crtime[2]; /* creation time */ 197 uint64_t lr_rdev; /* rdev of object to create */ 198 /* name of object to create follows this */ 199 /* for symlinks, link content follows name */ 200 /* for creates with xvattr data, the name follows the xvattr info */ 201 } lr_create_t; 202 203 /* 204 * FUID ACL record will be an array of ACEs from the original ACL. 205 * If this array includes ephemeral IDs, the record will also include 206 * an array of log-specific FUIDs to replace the ephemeral IDs. 207 * Only one copy of each unique domain will be present, so the log-specific 208 * FUIDs will use an index into a compressed domain table. On replay this 209 * information will be used to construct real FUIDs (and bypass idmap, 210 * since it may not be available). 211 */ 212 213 /* 214 * Log record for creates with optional ACL 215 * This log record is also used for recording any FUID 216 * information needed for replaying the create. If the 217 * file doesn't have any actual ACEs then the lr_aclcnt 218 * would be zero. 219 */ 220 typedef struct { 221 lr_create_t lr_create; /* common create portion */ 222 uint64_t lr_aclcnt; /* number of ACEs in ACL */ 223 uint64_t lr_domcnt; /* number of unique domains */ 224 uint64_t lr_fuidcnt; /* number of real fuids */ 225 uint64_t lr_acl_bytes; /* number of bytes in ACL */ 226 uint64_t lr_acl_flags; /* ACL flags */ 227 /* lr_acl_bytes number of variable sized ace's follows */ 228 /* if create is also setting xvattr's, then acl data follows xvattr */ 229 /* if ACE FUIDs are needed then they will follow the xvattr_t */ 230 /* Following the FUIDs will be the domain table information. */ 231 /* The FUIDs for the owner and group will be in the lr_create */ 232 /* portion of the record. */ 233 /* name follows ACL data */ 234 } lr_acl_create_t; 235 236 typedef struct { 237 lr_t lr_common; /* common portion of log record */ 238 uint64_t lr_doid; /* obj id of directory */ 239 /* name of object to remove follows this */ 240 } lr_remove_t; 241 242 typedef struct { 243 lr_t lr_common; /* common portion of log record */ 244 uint64_t lr_doid; /* obj id of directory */ 245 uint64_t lr_link_obj; /* obj id of link */ 246 /* name of object to link follows this */ 247 } lr_link_t; 248 249 typedef struct { 250 lr_t lr_common; /* common portion of log record */ 251 uint64_t lr_sdoid; /* obj id of source directory */ 252 uint64_t lr_tdoid; /* obj id of target directory */ 253 /* 2 strings: names of source and destination follow this */ 254 } lr_rename_t; 255 256 typedef struct { 257 lr_t lr_common; /* common portion of log record */ 258 uint64_t lr_foid; /* file object to write */ 259 uint64_t lr_offset; /* offset to write to */ 260 uint64_t lr_length; /* user data length to write */ 261 uint64_t lr_blkoff; /* offset represented by lr_blkptr */ 262 blkptr_t lr_blkptr; /* spa block pointer for replay */ 263 /* write data will follow for small writes */ 264 } lr_write_t; 265 266 typedef struct { 267 lr_t lr_common; /* common portion of log record */ 268 uint64_t lr_foid; /* object id of file to truncate */ 269 uint64_t lr_offset; /* offset to truncate from */ 270 uint64_t lr_length; /* length to truncate */ 271 } lr_truncate_t; 272 273 typedef struct { 274 lr_t lr_common; /* common portion of log record */ 275 uint64_t lr_foid; /* file object to change attributes */ 276 uint64_t lr_mask; /* mask of attributes to set */ 277 uint64_t lr_mode; /* mode to set */ 278 uint64_t lr_uid; /* uid to set */ 279 uint64_t lr_gid; /* gid to set */ 280 uint64_t lr_size; /* size to set */ 281 uint64_t lr_atime[2]; /* access time */ 282 uint64_t lr_mtime[2]; /* modification time */ 283 /* optional attribute lr_attr_t may be here */ 284 } lr_setattr_t; 285 286 typedef struct { 287 lr_t lr_common; /* common portion of log record */ 288 uint64_t lr_foid; /* obj id of file */ 289 uint64_t lr_aclcnt; /* number of acl entries */ 290 /* lr_aclcnt number of ace_t entries follow this */ 291 } lr_acl_v0_t; 292 293 typedef struct { 294 lr_t lr_common; /* common portion of log record */ 295 uint64_t lr_foid; /* obj id of file */ 296 uint64_t lr_aclcnt; /* number of ACEs in ACL */ 297 uint64_t lr_domcnt; /* number of unique domains */ 298 uint64_t lr_fuidcnt; /* number of real fuids */ 299 uint64_t lr_acl_bytes; /* number of bytes in ACL */ 300 uint64_t lr_acl_flags; /* ACL flags */ 301 /* lr_acl_bytes number of variable sized ace's follows */ 302 } lr_acl_t; 303 304 /* 305 * ZIL structure definitions, interface function prototype and globals. 306 */ 307 308 /* 309 * Writes are handled in three different ways: 310 * 311 * WR_INDIRECT: 312 * In this mode, if we need to commit the write later, then the block 313 * is immediately written into the file system (using dmu_sync), 314 * and a pointer to the block is put into the log record. 315 * When the txg commits the block is linked in. 316 * This saves additionally writing the data into the log record. 317 * There are a few requirements for this to occur: 318 * - write is greater than zfs/zvol_immediate_write_sz 319 * - not using slogs (as slogs are assumed to always be faster 320 * than writing into the main pool) 321 * - the write occupies only one block 322 * WR_COPIED: 323 * If we know we'll immediately be committing the 324 * transaction (FSYNC or FDSYNC), the we allocate a larger 325 * log record here for the data and copy the data in. 326 * WR_NEED_COPY: 327 * Otherwise we don't allocate a buffer, and *if* we need to 328 * flush the write later then a buffer is allocated and 329 * we retrieve the data using the dmu. 330 */ 331 typedef enum { 332 WR_INDIRECT, /* indirect - a large write (dmu_sync() data */ 333 /* and put blkptr in log, rather than actual data) */ 334 WR_COPIED, /* immediate - data is copied into lr_write_t */ 335 WR_NEED_COPY, /* immediate - data needs to be copied if pushed */ 336 } itx_wr_state_t; 337 338 typedef struct itx { 339 list_node_t itx_node; /* linkage on zl_itx_list */ 340 void *itx_private; /* type-specific opaque data */ 341 itx_wr_state_t itx_wr_state; /* write state */ 342 uint8_t itx_sync; /* synchronous transaction */ 343 uint64_t itx_sod; /* record size on disk */ 344 lr_t itx_lr; /* common part of log record */ 345 /* followed by type-specific part of lr_xx_t and its immediate data */ 346 } itx_t; 347 348 349 /* 350 * zgd_t is passed through dmu_sync() to the callback routine zfs_get_done() 351 * to handle the cleanup of the dmu_sync() buffer write 352 */ 353 typedef struct { 354 zilog_t *zgd_zilog; /* zilog */ 355 blkptr_t *zgd_bp; /* block pointer */ 356 struct rl *zgd_rl; /* range lock */ 357 } zgd_t; 358 359 360 typedef void zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg, 361 uint64_t txg); 362 typedef void zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg, 363 uint64_t txg); 364 typedef int zil_replay_func_t(); 365 typedef int zil_get_data_t(void *arg, lr_write_t *lr, char *dbuf, zio_t *zio); 366 367 extern uint64_t zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, 368 zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg); 369 370 extern void zil_init(void); 371 extern void zil_fini(void); 372 373 extern zilog_t *zil_alloc(objset_t *os, zil_header_t *zh_phys); 374 extern void zil_free(zilog_t *zilog); 375 376 extern zilog_t *zil_open(objset_t *os, zil_get_data_t *get_data); 377 extern void zil_close(zilog_t *zilog); 378 379 extern void zil_replay(objset_t *os, void *arg, 380 zil_replay_func_t *replay_func[TX_MAX_TYPE]); 381 extern void zil_destroy(zilog_t *zilog, boolean_t keep_first); 382 extern void zil_rollback_destroy(zilog_t *zilog, dmu_tx_t *tx); 383 384 extern itx_t *zil_itx_create(uint64_t txtype, size_t lrsize); 385 extern uint64_t zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx); 386 387 extern void zil_commit(zilog_t *zilog, uint64_t seq, uint64_t oid); 388 389 extern int zil_vdev_offline(char *osname, void *txarg); 390 extern int zil_claim(char *osname, void *txarg); 391 extern int zil_check_log_chain(char *osname, void *txarg); 392 extern void zil_sync(zilog_t *zilog, dmu_tx_t *tx); 393 extern void zil_clean(zilog_t *zilog); 394 extern int zil_is_committed(zilog_t *zilog); 395 396 extern int zil_suspend(zilog_t *zilog); 397 extern void zil_resume(zilog_t *zilog); 398 399 extern void zil_add_block(zilog_t *zilog, blkptr_t *bp); 400 extern void zil_set_logbias(zilog_t *zilog, uint64_t slogval); 401 extern void zil_get_replay_data(zilog_t *zilog, lr_write_t *lr); 402 403 extern int zil_disable; 404 405 #ifdef __cplusplus 406 } 407 #endif 408 409 #endif /* _SYS_ZIL_H */ 410