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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23 * Copyright (c) 2012, 2018 by Delphix. All rights reserved. 24 */ 25 26 /* Portions Copyright 2010 Robert Milkowski */ 27 28 #ifndef _SYS_ZIL_H 29 #define _SYS_ZIL_H 30 31 #include <sys/types.h> 32 #include <sys/spa.h> 33 #include <sys/zio.h> 34 #include <sys/dmu.h> 35 #include <sys/zio_crypt.h> 36 37 #ifdef __cplusplus 38 extern "C" { 39 #endif 40 41 struct dsl_pool; 42 struct dsl_dataset; 43 struct lwb; 44 45 /* 46 * Intent log format: 47 * 48 * Each objset has its own intent log. The log header (zil_header_t) 49 * for objset N's intent log is kept in the Nth object of the SPA's 50 * intent_log objset. The log header points to a chain of log blocks, 51 * each of which contains log records (i.e., transactions) followed by 52 * a log block trailer (zil_trailer_t). The format of a log record 53 * depends on the record (or transaction) type, but all records begin 54 * with a common structure that defines the type, length, and txg. 55 */ 56 57 /* 58 * Intent log header - this on disk structure holds fields to manage 59 * the log. All fields are 64 bit to easily handle cross architectures. 60 */ 61 typedef struct zil_header { 62 uint64_t zh_claim_txg; /* txg in which log blocks were claimed */ 63 uint64_t zh_replay_seq; /* highest replayed sequence number */ 64 blkptr_t zh_log; /* log chain */ 65 uint64_t zh_claim_blk_seq; /* highest claimed block sequence number */ 66 uint64_t zh_flags; /* header flags */ 67 uint64_t zh_claim_lr_seq; /* highest claimed lr sequence number */ 68 uint64_t zh_pad[3]; 69 } zil_header_t; 70 71 /* 72 * zh_flags bit settings 73 */ 74 #define ZIL_REPLAY_NEEDED 0x1 /* replay needed - internal only */ 75 #define ZIL_CLAIM_LR_SEQ_VALID 0x2 /* zh_claim_lr_seq field is valid */ 76 77 /* 78 * Log block chaining. 79 * 80 * Log blocks are chained together. Originally they were chained at the 81 * end of the block. For performance reasons the chain was moved to the 82 * beginning of the block which allows writes for only the data being used. 83 * The older position is supported for backwards compatibility. 84 * 85 * The zio_eck_t contains a zec_cksum which for the intent log is 86 * the sequence number of this log block. A seq of 0 is invalid. 87 * The zec_cksum is checked by the SPA against the sequence 88 * number passed in the blk_cksum field of the blkptr_t 89 */ 90 typedef struct zil_chain { 91 uint64_t zc_pad; 92 blkptr_t zc_next_blk; /* next block in chain */ 93 uint64_t zc_nused; /* bytes in log block used */ 94 zio_eck_t zc_eck; /* block trailer */ 95 } zil_chain_t; 96 97 #define ZIL_MIN_BLKSZ 4096ULL 98 99 /* 100 * ziltest is by and large an ugly hack, but very useful in 101 * checking replay without tedious work. 102 * When running ziltest we want to keep all itx's and so maintain 103 * a single list in the zl_itxg[] that uses a high txg: ZILTEST_TXG 104 * We subtract TXG_CONCURRENT_STATES to allow for common code. 105 */ 106 #define ZILTEST_TXG (UINT64_MAX - TXG_CONCURRENT_STATES) 107 108 /* 109 * The words of a log block checksum. 110 */ 111 #define ZIL_ZC_GUID_0 0 112 #define ZIL_ZC_GUID_1 1 113 #define ZIL_ZC_OBJSET 2 114 #define ZIL_ZC_SEQ 3 115 116 typedef enum zil_create { 117 Z_FILE, 118 Z_DIR, 119 Z_XATTRDIR, 120 } zil_create_t; 121 122 /* 123 * size of xvattr log section. 124 * its composed of lr_attr_t + xvattr bitmap + 2 64 bit timestamps 125 * for create time and a single 64 bit integer for all of the attributes, 126 * and 4 64 bit integers (32 bytes) for the scanstamp. 127 * 128 */ 129 130 #define ZIL_XVAT_SIZE(mapsize) \ 131 sizeof (lr_attr_t) + (sizeof (uint32_t) * (mapsize - 1)) + \ 132 (sizeof (uint64_t) * 7) 133 134 /* 135 * Size of ACL in log. The ACE data is padded out to properly align 136 * on 8 byte boundary. 137 */ 138 139 #define ZIL_ACE_LENGTH(x) (roundup(x, sizeof (uint64_t))) 140 141 /* 142 * Intent log transaction types and record structures 143 */ 144 #define TX_COMMIT 0 /* Commit marker (no on-disk state) */ 145 #define TX_CREATE 1 /* Create file */ 146 #define TX_MKDIR 2 /* Make directory */ 147 #define TX_MKXATTR 3 /* Make XATTR directory */ 148 #define TX_SYMLINK 4 /* Create symbolic link to a file */ 149 #define TX_REMOVE 5 /* Remove file */ 150 #define TX_RMDIR 6 /* Remove directory */ 151 #define TX_LINK 7 /* Create hard link to a file */ 152 #define TX_RENAME 8 /* Rename a file */ 153 #define TX_WRITE 9 /* File write */ 154 #define TX_TRUNCATE 10 /* Truncate a file */ 155 #define TX_SETATTR 11 /* Set file attributes */ 156 #define TX_ACL_V0 12 /* Set old formatted ACL */ 157 #define TX_ACL 13 /* Set ACL */ 158 #define TX_CREATE_ACL 14 /* create with ACL */ 159 #define TX_CREATE_ATTR 15 /* create + attrs */ 160 #define TX_CREATE_ACL_ATTR 16 /* create with ACL + attrs */ 161 #define TX_MKDIR_ACL 17 /* mkdir with ACL */ 162 #define TX_MKDIR_ATTR 18 /* mkdir with attr */ 163 #define TX_MKDIR_ACL_ATTR 19 /* mkdir with ACL + attrs */ 164 #define TX_WRITE2 20 /* dmu_sync EALREADY write */ 165 #define TX_MAX_TYPE 21 /* Max transaction type */ 166 167 /* 168 * The transactions for mkdir, symlink, remove, rmdir, link, and rename 169 * may have the following bit set, indicating the original request 170 * specified case-insensitive handling of names. 171 */ 172 #define TX_CI ((uint64_t)0x1 << 63) /* case-insensitive behavior requested */ 173 174 /* 175 * Transactions for write, truncate, setattr, acl_v0, and acl can be logged 176 * out of order. For convenience in the code, all such records must have 177 * lr_foid at the same offset. 178 */ 179 #define TX_OOO(txtype) \ 180 ((txtype) == TX_WRITE || \ 181 (txtype) == TX_TRUNCATE || \ 182 (txtype) == TX_SETATTR || \ 183 (txtype) == TX_ACL_V0 || \ 184 (txtype) == TX_ACL || \ 185 (txtype) == TX_WRITE2) 186 187 /* 188 * The number of dnode slots consumed by the object is stored in the 8 189 * unused upper bits of the object ID. We subtract 1 from the value 190 * stored on disk for compatibility with implementations that don't 191 * support large dnodes. The slot count for a single-slot dnode will 192 * contain 0 for those bits to preserve the log record format for 193 * "small" dnodes. 194 */ 195 #define LR_FOID_GET_SLOTS(oid) (BF64_GET((oid), 56, 8) + 1) 196 #define LR_FOID_SET_SLOTS(oid, x) BF64_SET((oid), 56, 8, (x) - 1) 197 #define LR_FOID_GET_OBJ(oid) BF64_GET((oid), 0, DN_MAX_OBJECT_SHIFT) 198 #define LR_FOID_SET_OBJ(oid, x) BF64_SET((oid), 0, DN_MAX_OBJECT_SHIFT, (x)) 199 200 /* 201 * Format of log records. 202 * The fields are carefully defined to allow them to be aligned 203 * and sized the same on sparc & intel architectures. 204 * Each log record has a common structure at the beginning. 205 * 206 * The log record on disk (lrc_seq) holds the sequence number of all log 207 * records which is used to ensure we don't replay the same record. 208 */ 209 typedef struct { /* common log record header */ 210 uint64_t lrc_txtype; /* intent log transaction type */ 211 uint64_t lrc_reclen; /* transaction record length */ 212 uint64_t lrc_txg; /* dmu transaction group number */ 213 uint64_t lrc_seq; /* see comment above */ 214 } lr_t; 215 216 /* 217 * Common start of all out-of-order record types (TX_OOO() above). 218 */ 219 typedef struct { 220 lr_t lr_common; /* common portion of log record */ 221 uint64_t lr_foid; /* object id */ 222 } lr_ooo_t; 223 224 /* 225 * Additional lr_attr_t fields. 226 */ 227 typedef struct { 228 uint64_t lr_attr_attrs; /* all of the attributes */ 229 uint64_t lr_attr_crtime[2]; /* create time */ 230 uint8_t lr_attr_scanstamp[32]; 231 } lr_attr_end_t; 232 233 /* 234 * Handle option extended vattr attributes. 235 * 236 * Whenever new attributes are added the version number 237 * will need to be updated as will code in 238 * zfs_log.c and zfs_replay.c 239 */ 240 typedef struct { 241 uint32_t lr_attr_masksize; /* number of elements in array */ 242 uint32_t lr_attr_bitmap; /* First entry of array */ 243 /* remainder of array and additional lr_attr_end_t fields */ 244 } lr_attr_t; 245 246 /* 247 * log record for creates without optional ACL. 248 * This log record does support optional xvattr_t attributes. 249 */ 250 typedef struct { 251 lr_t lr_common; /* common portion of log record */ 252 uint64_t lr_doid; /* object id of directory */ 253 uint64_t lr_foid; /* object id of created file object */ 254 uint64_t lr_mode; /* mode of object */ 255 uint64_t lr_uid; /* uid of object */ 256 uint64_t lr_gid; /* gid of object */ 257 uint64_t lr_gen; /* generation (txg of creation) */ 258 uint64_t lr_crtime[2]; /* creation time */ 259 uint64_t lr_rdev; /* rdev of object to create */ 260 /* name of object to create follows this */ 261 /* for symlinks, link content follows name */ 262 /* for creates with xvattr data, the name follows the xvattr info */ 263 } lr_create_t; 264 265 /* 266 * FUID ACL record will be an array of ACEs from the original ACL. 267 * If this array includes ephemeral IDs, the record will also include 268 * an array of log-specific FUIDs to replace the ephemeral IDs. 269 * Only one copy of each unique domain will be present, so the log-specific 270 * FUIDs will use an index into a compressed domain table. On replay this 271 * information will be used to construct real FUIDs (and bypass idmap, 272 * since it may not be available). 273 */ 274 275 /* 276 * Log record for creates with optional ACL 277 * This log record is also used for recording any FUID 278 * information needed for replaying the create. If the 279 * file doesn't have any actual ACEs then the lr_aclcnt 280 * would be zero. 281 * 282 * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's. 283 * If create is also setting xvattr's, then acl data follows xvattr. 284 * If ACE FUIDs are needed then they will follow the xvattr_t. Following 285 * the FUIDs will be the domain table information. The FUIDs for the owner 286 * and group will be in lr_create. Name follows ACL data. 287 */ 288 typedef struct { 289 lr_create_t lr_create; /* common create portion */ 290 uint64_t lr_aclcnt; /* number of ACEs in ACL */ 291 uint64_t lr_domcnt; /* number of unique domains */ 292 uint64_t lr_fuidcnt; /* number of real fuids */ 293 uint64_t lr_acl_bytes; /* number of bytes in ACL */ 294 uint64_t lr_acl_flags; /* ACL flags */ 295 } lr_acl_create_t; 296 297 typedef struct { 298 lr_t lr_common; /* common portion of log record */ 299 uint64_t lr_doid; /* obj id of directory */ 300 /* name of object to remove follows this */ 301 } lr_remove_t; 302 303 typedef struct { 304 lr_t lr_common; /* common portion of log record */ 305 uint64_t lr_doid; /* obj id of directory */ 306 uint64_t lr_link_obj; /* obj id of link */ 307 /* name of object to link follows this */ 308 } lr_link_t; 309 310 typedef struct { 311 lr_t lr_common; /* common portion of log record */ 312 uint64_t lr_sdoid; /* obj id of source directory */ 313 uint64_t lr_tdoid; /* obj id of target directory */ 314 /* 2 strings: names of source and destination follow this */ 315 } lr_rename_t; 316 317 typedef struct { 318 lr_t lr_common; /* common portion of log record */ 319 uint64_t lr_foid; /* file object to write */ 320 uint64_t lr_offset; /* offset to write to */ 321 uint64_t lr_length; /* user data length to write */ 322 uint64_t lr_blkoff; /* no longer used */ 323 blkptr_t lr_blkptr; /* spa block pointer for replay */ 324 /* write data will follow for small writes */ 325 } lr_write_t; 326 327 typedef struct { 328 lr_t lr_common; /* common portion of log record */ 329 uint64_t lr_foid; /* object id of file to truncate */ 330 uint64_t lr_offset; /* offset to truncate from */ 331 uint64_t lr_length; /* length to truncate */ 332 } lr_truncate_t; 333 334 typedef struct { 335 lr_t lr_common; /* common portion of log record */ 336 uint64_t lr_foid; /* file object to change attributes */ 337 uint64_t lr_mask; /* mask of attributes to set */ 338 uint64_t lr_mode; /* mode to set */ 339 uint64_t lr_uid; /* uid to set */ 340 uint64_t lr_gid; /* gid to set */ 341 uint64_t lr_size; /* size to set */ 342 uint64_t lr_atime[2]; /* access time */ 343 uint64_t lr_mtime[2]; /* modification time */ 344 /* optional attribute lr_attr_t may be here */ 345 } lr_setattr_t; 346 347 typedef struct { 348 lr_t lr_common; /* common portion of log record */ 349 uint64_t lr_foid; /* obj id of file */ 350 uint64_t lr_aclcnt; /* number of acl entries */ 351 /* lr_aclcnt number of ace_t entries follow this */ 352 } lr_acl_v0_t; 353 354 typedef struct { 355 lr_t lr_common; /* common portion of log record */ 356 uint64_t lr_foid; /* obj id of file */ 357 uint64_t lr_aclcnt; /* number of ACEs in ACL */ 358 uint64_t lr_domcnt; /* number of unique domains */ 359 uint64_t lr_fuidcnt; /* number of real fuids */ 360 uint64_t lr_acl_bytes; /* number of bytes in ACL */ 361 uint64_t lr_acl_flags; /* ACL flags */ 362 /* lr_acl_bytes number of variable sized ace's follows */ 363 } lr_acl_t; 364 365 /* 366 * ZIL structure definitions, interface function prototype and globals. 367 */ 368 369 /* 370 * Writes are handled in three different ways: 371 * 372 * WR_INDIRECT: 373 * In this mode, if we need to commit the write later, then the block 374 * is immediately written into the file system (using dmu_sync), 375 * and a pointer to the block is put into the log record. 376 * When the txg commits the block is linked in. 377 * This saves additionally writing the data into the log record. 378 * There are a few requirements for this to occur: 379 * - write is greater than zfs/zvol_immediate_write_sz 380 * - not using slogs (as slogs are assumed to always be faster 381 * than writing into the main pool) 382 * - the write occupies only one block 383 * WR_COPIED: 384 * If we know we'll immediately be committing the 385 * transaction (O_SYNC or O_DSYNC), then we allocate a larger 386 * log record here for the data and copy the data in. 387 * WR_NEED_COPY: 388 * Otherwise we don't allocate a buffer, and *if* we need to 389 * flush the write later then a buffer is allocated and 390 * we retrieve the data using the dmu. 391 */ 392 typedef enum { 393 WR_INDIRECT, /* indirect - a large write (dmu_sync() data */ 394 /* and put blkptr in log, rather than actual data) */ 395 WR_COPIED, /* immediate - data is copied into lr_write_t */ 396 WR_NEED_COPY, /* immediate - data needs to be copied if pushed */ 397 WR_NUM_STATES /* number of states */ 398 } itx_wr_state_t; 399 400 typedef void (*zil_callback_t)(void *data); 401 402 typedef struct itx { 403 list_node_t itx_node; /* linkage on zl_itx_list */ 404 void *itx_private; /* type-specific opaque data */ 405 itx_wr_state_t itx_wr_state; /* write state */ 406 uint8_t itx_sync; /* synchronous transaction */ 407 zil_callback_t itx_callback; /* Called when the itx is persistent */ 408 void *itx_callback_data; /* User data for the callback */ 409 size_t itx_size; /* allocated itx structure size */ 410 uint64_t itx_oid; /* object id */ 411 uint64_t itx_gen; /* gen number for zfs_get_data */ 412 lr_t itx_lr; /* common part of log record */ 413 /* followed by type-specific part of lr_xx_t and its immediate data */ 414 } itx_t; 415 416 /* 417 * Used for zil kstat. 418 */ 419 typedef struct zil_stats { 420 /* 421 * Number of times a ZIL commit (e.g. fsync) has been requested. 422 */ 423 kstat_named_t zil_commit_count; 424 425 /* 426 * Number of times the ZIL has been flushed to stable storage. 427 * This is less than zil_commit_count when commits are "merged" 428 * (see the documentation above zil_commit()). 429 */ 430 kstat_named_t zil_commit_writer_count; 431 432 /* 433 * Number of transactions (reads, writes, renames, etc.) 434 * that have been committed. 435 */ 436 kstat_named_t zil_itx_count; 437 438 /* 439 * See the documentation for itx_wr_state_t above. 440 * Note that "bytes" accumulates the length of the transactions 441 * (i.e. data), not the actual log record sizes. 442 */ 443 kstat_named_t zil_itx_indirect_count; 444 kstat_named_t zil_itx_indirect_bytes; 445 kstat_named_t zil_itx_copied_count; 446 kstat_named_t zil_itx_copied_bytes; 447 kstat_named_t zil_itx_needcopy_count; 448 kstat_named_t zil_itx_needcopy_bytes; 449 450 /* 451 * Transactions which have been allocated to the "normal" 452 * (i.e. not slog) storage pool. Note that "bytes" accumulate 453 * the actual log record sizes - which do not include the actual 454 * data in case of indirect writes. 455 */ 456 kstat_named_t zil_itx_metaslab_normal_count; 457 kstat_named_t zil_itx_metaslab_normal_bytes; 458 459 /* 460 * Transactions which have been allocated to the "slog" storage pool. 461 * If there are no separate log devices, this is the same as the 462 * "normal" pool. 463 */ 464 kstat_named_t zil_itx_metaslab_slog_count; 465 kstat_named_t zil_itx_metaslab_slog_bytes; 466 } zil_stats_t; 467 468 extern zil_stats_t zil_stats; 469 470 #define ZIL_STAT_INCR(stat, val) \ 471 atomic_add_64(&zil_stats.stat.value.ui64, (val)); 472 #define ZIL_STAT_BUMP(stat) \ 473 ZIL_STAT_INCR(stat, 1); 474 475 typedef int zil_parse_blk_func_t(zilog_t *zilog, const blkptr_t *bp, void *arg, 476 uint64_t txg); 477 typedef int zil_parse_lr_func_t(zilog_t *zilog, const lr_t *lr, void *arg, 478 uint64_t txg); 479 typedef int zil_replay_func_t(void *arg1, void *arg2, boolean_t byteswap); 480 typedef int zil_get_data_t(void *arg, uint64_t arg2, lr_write_t *lr, char *dbuf, 481 struct lwb *lwb, zio_t *zio); 482 483 extern int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, 484 zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg, 485 boolean_t decrypt); 486 487 extern void zil_init(void); 488 extern void zil_fini(void); 489 490 extern zilog_t *zil_alloc(objset_t *os, zil_header_t *zh_phys); 491 extern void zil_free(zilog_t *zilog); 492 493 extern zilog_t *zil_open(objset_t *os, zil_get_data_t *get_data); 494 extern void zil_close(zilog_t *zilog); 495 496 extern void zil_replay(objset_t *os, void *arg, 497 zil_replay_func_t *replay_func[TX_MAX_TYPE]); 498 extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx); 499 extern void zil_destroy(zilog_t *zilog, boolean_t keep_first); 500 extern void zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx); 501 502 extern itx_t *zil_itx_create(uint64_t txtype, size_t lrsize); 503 extern void zil_itx_destroy(itx_t *itx); 504 extern void zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx); 505 506 extern void zil_async_to_sync(zilog_t *zilog, uint64_t oid); 507 extern void zil_commit(zilog_t *zilog, uint64_t oid); 508 extern void zil_commit_impl(zilog_t *zilog, uint64_t oid); 509 extern void zil_remove_async(zilog_t *zilog, uint64_t oid); 510 511 extern int zil_reset(const char *osname, void *txarg); 512 extern int zil_claim(struct dsl_pool *dp, 513 struct dsl_dataset *ds, void *txarg); 514 extern int zil_check_log_chain(struct dsl_pool *dp, 515 struct dsl_dataset *ds, void *tx); 516 extern void zil_sync(zilog_t *zilog, dmu_tx_t *tx); 517 extern void zil_clean(zilog_t *zilog, uint64_t synced_txg); 518 519 extern int zil_suspend(const char *osname, void **cookiep); 520 extern void zil_resume(void *cookie); 521 522 extern void zil_lwb_add_block(struct lwb *lwb, const blkptr_t *bp); 523 extern void zil_lwb_add_txg(struct lwb *lwb, uint64_t txg); 524 extern int zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp); 525 526 extern void zil_set_sync(zilog_t *zilog, uint64_t syncval); 527 528 extern void zil_set_logbias(zilog_t *zilog, uint64_t slogval); 529 530 extern uint64_t zil_max_copied_data(zilog_t *zilog); 531 extern uint64_t zil_max_log_data(zilog_t *zilog); 532 533 extern int zil_replay_disable; 534 535 #ifdef __cplusplus 536 } 537 #endif 538 539 #endif /* _SYS_ZIL_H */ 540