zil.h revision 332525
1280933Sdteske/* 2280933Sdteske * CDDL HEADER START 3280933Sdteske * 4280933Sdteske * The contents of this file are subject to the terms of the 5280933Sdteske * Common Development and Distribution License (the "License"). 6280933Sdteske * You may not use this file except in compliance with the License. 7280933Sdteske * 8280933Sdteske * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9280933Sdteske * or http://www.opensolaris.org/os/licensing. 10280933Sdteske * See the License for the specific language governing permissions 11280933Sdteske * and limitations under the License. 12280933Sdteske * 13280933Sdteske * When distributing Covered Code, include this CDDL HEADER in each 14280933Sdteske * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15280933Sdteske * If applicable, add the following below this CDDL HEADER, with the 16280933Sdteske * fields enclosed by brackets "[]" replaced with your own identifying 17280933Sdteske * information: Portions Copyright [yyyy] [name of copyright owner] 18280933Sdteske * 19280933Sdteske * CDDL HEADER END 20280933Sdteske */ 21280933Sdteske/* 22280933Sdteske * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 23280933Sdteske * Copyright (c) 2012, 2017 by Delphix. All rights reserved. 24280933Sdteske * Copyright (c) 2014 Integros [integros.com] 25280933Sdteske */ 26280933Sdteske 27280933Sdteske/* Portions Copyright 2010 Robert Milkowski */ 28280933Sdteske 29280933Sdteske#ifndef _SYS_ZIL_H 30281843Sdteske#define _SYS_ZIL_H 31281843Sdteske 32281843Sdteske#include <sys/types.h> 33281843Sdteske#include <sys/spa.h> 34281843Sdteske#include <sys/zio.h> 35281843Sdteske#include <sys/dmu.h> 36280933Sdteske 37280933Sdteske#ifdef __cplusplus 38281843Sdteskeextern "C" { 39281843Sdteske#endif 40281843Sdteske 41281843Sdteskestruct dsl_pool; 42281843Sdteskestruct dsl_dataset; 43281843Sdteskestruct lwb; 44281843Sdteske 45281843Sdteske/* 46281843Sdteske * Intent log format: 47281843Sdteske * 48281843Sdteske * Each objset has its own intent log. The log header (zil_header_t) 49281843Sdteske * for objset N's intent log is kept in the Nth object of the SPA's 50281843Sdteske * intent_log objset. The log header points to a chain of log blocks, 51281843Sdteske * each of which contains log records (i.e., transactions) followed by 52281843Sdteske * a log block trailer (zil_trailer_t). The format of a log record 53281843Sdteske * depends on the record (or transaction) type, but all records begin 54281843Sdteske * with a common structure that defines the type, length, and txg. 55281843Sdteske */ 56281843Sdteske 57281843Sdteske/* 58281843Sdteske * Intent log header - this on disk structure holds fields to manage 59280933Sdteske * the log. All fields are 64 bit to easily handle cross architectures. 60 */ 61typedef 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 compatability. 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 */ 90typedef 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 116typedef 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 * Format of log records. 189 * The fields are carefully defined to allow them to be aligned 190 * and sized the same on sparc & intel architectures. 191 * Each log record has a common structure at the beginning. 192 * 193 * The log record on disk (lrc_seq) holds the sequence number of all log 194 * records which is used to ensure we don't replay the same record. 195 */ 196typedef struct { /* common log record header */ 197 uint64_t lrc_txtype; /* intent log transaction type */ 198 uint64_t lrc_reclen; /* transaction record length */ 199 uint64_t lrc_txg; /* dmu transaction group number */ 200 uint64_t lrc_seq; /* see comment above */ 201} lr_t; 202 203/* 204 * Common start of all out-of-order record types (TX_OOO() above). 205 */ 206typedef struct { 207 lr_t lr_common; /* common portion of log record */ 208 uint64_t lr_foid; /* object id */ 209} lr_ooo_t; 210 211/* 212 * Handle option extended vattr attributes. 213 * 214 * Whenever new attributes are added the version number 215 * will need to be updated as will code in 216 * zfs_log.c and zfs_replay.c 217 */ 218typedef struct { 219 uint32_t lr_attr_masksize; /* number of elements in array */ 220 uint32_t lr_attr_bitmap; /* First entry of array */ 221 /* remainder of array and any additional fields */ 222} lr_attr_t; 223 224/* 225 * log record for creates without optional ACL. 226 * This log record does support optional xvattr_t attributes. 227 */ 228typedef struct { 229 lr_t lr_common; /* common portion of log record */ 230 uint64_t lr_doid; /* object id of directory */ 231 uint64_t lr_foid; /* object id of created file object */ 232 uint64_t lr_mode; /* mode of object */ 233 uint64_t lr_uid; /* uid of object */ 234 uint64_t lr_gid; /* gid of object */ 235 uint64_t lr_gen; /* generation (txg of creation) */ 236 uint64_t lr_crtime[2]; /* creation time */ 237 uint64_t lr_rdev; /* rdev of object to create */ 238 /* name of object to create follows this */ 239 /* for symlinks, link content follows name */ 240 /* for creates with xvattr data, the name follows the xvattr info */ 241} lr_create_t; 242 243/* 244 * FUID ACL record will be an array of ACEs from the original ACL. 245 * If this array includes ephemeral IDs, the record will also include 246 * an array of log-specific FUIDs to replace the ephemeral IDs. 247 * Only one copy of each unique domain will be present, so the log-specific 248 * FUIDs will use an index into a compressed domain table. On replay this 249 * information will be used to construct real FUIDs (and bypass idmap, 250 * since it may not be available). 251 */ 252 253/* 254 * Log record for creates with optional ACL 255 * This log record is also used for recording any FUID 256 * information needed for replaying the create. If the 257 * file doesn't have any actual ACEs then the lr_aclcnt 258 * would be zero. 259 * 260 * After lr_acl_flags, there are a lr_acl_bytes number of variable sized ace's. 261 * If create is also setting xvattr's, then acl data follows xvattr. 262 * If ACE FUIDs are needed then they will follow the xvattr_t. Following 263 * the FUIDs will be the domain table information. The FUIDs for the owner 264 * and group will be in lr_create. Name follows ACL data. 265 */ 266typedef struct { 267 lr_create_t lr_create; /* common create portion */ 268 uint64_t lr_aclcnt; /* number of ACEs in ACL */ 269 uint64_t lr_domcnt; /* number of unique domains */ 270 uint64_t lr_fuidcnt; /* number of real fuids */ 271 uint64_t lr_acl_bytes; /* number of bytes in ACL */ 272 uint64_t lr_acl_flags; /* ACL flags */ 273} lr_acl_create_t; 274 275typedef struct { 276 lr_t lr_common; /* common portion of log record */ 277 uint64_t lr_doid; /* obj id of directory */ 278 /* name of object to remove follows this */ 279} lr_remove_t; 280 281typedef struct { 282 lr_t lr_common; /* common portion of log record */ 283 uint64_t lr_doid; /* obj id of directory */ 284 uint64_t lr_link_obj; /* obj id of link */ 285 /* name of object to link follows this */ 286} lr_link_t; 287 288typedef struct { 289 lr_t lr_common; /* common portion of log record */ 290 uint64_t lr_sdoid; /* obj id of source directory */ 291 uint64_t lr_tdoid; /* obj id of target directory */ 292 /* 2 strings: names of source and destination follow this */ 293} lr_rename_t; 294 295typedef struct { 296 lr_t lr_common; /* common portion of log record */ 297 uint64_t lr_foid; /* file object to write */ 298 uint64_t lr_offset; /* offset to write to */ 299 uint64_t lr_length; /* user data length to write */ 300 uint64_t lr_blkoff; /* no longer used */ 301 blkptr_t lr_blkptr; /* spa block pointer for replay */ 302 /* write data will follow for small writes */ 303} lr_write_t; 304 305typedef struct { 306 lr_t lr_common; /* common portion of log record */ 307 uint64_t lr_foid; /* object id of file to truncate */ 308 uint64_t lr_offset; /* offset to truncate from */ 309 uint64_t lr_length; /* length to truncate */ 310} lr_truncate_t; 311 312typedef struct { 313 lr_t lr_common; /* common portion of log record */ 314 uint64_t lr_foid; /* file object to change attributes */ 315 uint64_t lr_mask; /* mask of attributes to set */ 316 uint64_t lr_mode; /* mode to set */ 317 uint64_t lr_uid; /* uid to set */ 318 uint64_t lr_gid; /* gid to set */ 319 uint64_t lr_size; /* size to set */ 320 uint64_t lr_atime[2]; /* access time */ 321 uint64_t lr_mtime[2]; /* modification time */ 322 /* optional attribute lr_attr_t may be here */ 323} lr_setattr_t; 324 325typedef struct { 326 lr_t lr_common; /* common portion of log record */ 327 uint64_t lr_foid; /* obj id of file */ 328 uint64_t lr_aclcnt; /* number of acl entries */ 329 /* lr_aclcnt number of ace_t entries follow this */ 330} lr_acl_v0_t; 331 332typedef struct { 333 lr_t lr_common; /* common portion of log record */ 334 uint64_t lr_foid; /* obj id of file */ 335 uint64_t lr_aclcnt; /* number of ACEs in ACL */ 336 uint64_t lr_domcnt; /* number of unique domains */ 337 uint64_t lr_fuidcnt; /* number of real fuids */ 338 uint64_t lr_acl_bytes; /* number of bytes in ACL */ 339 uint64_t lr_acl_flags; /* ACL flags */ 340 /* lr_acl_bytes number of variable sized ace's follows */ 341} lr_acl_t; 342 343/* 344 * ZIL structure definitions, interface function prototype and globals. 345 */ 346 347/* 348 * Writes are handled in three different ways: 349 * 350 * WR_INDIRECT: 351 * In this mode, if we need to commit the write later, then the block 352 * is immediately written into the file system (using dmu_sync), 353 * and a pointer to the block is put into the log record. 354 * When the txg commits the block is linked in. 355 * This saves additionally writing the data into the log record. 356 * There are a few requirements for this to occur: 357 * - write is greater than zfs/zvol_immediate_write_sz 358 * - not using slogs (as slogs are assumed to always be faster 359 * than writing into the main pool) 360 * - the write occupies only one block 361 * WR_COPIED: 362 * If we know we'll immediately be committing the 363 * transaction (FSYNC or FDSYNC), the we allocate a larger 364 * log record here for the data and copy the data in. 365 * WR_NEED_COPY: 366 * Otherwise we don't allocate a buffer, and *if* we need to 367 * flush the write later then a buffer is allocated and 368 * we retrieve the data using the dmu. 369 */ 370typedef enum { 371 WR_INDIRECT, /* indirect - a large write (dmu_sync() data */ 372 /* and put blkptr in log, rather than actual data) */ 373 WR_COPIED, /* immediate - data is copied into lr_write_t */ 374 WR_NEED_COPY, /* immediate - data needs to be copied if pushed */ 375 WR_NUM_STATES /* number of states */ 376} itx_wr_state_t; 377 378typedef struct itx { 379 list_node_t itx_node; /* linkage on zl_itx_list */ 380 void *itx_private; /* type-specific opaque data */ 381 itx_wr_state_t itx_wr_state; /* write state */ 382 uint8_t itx_sync; /* synchronous transaction */ 383 uint64_t itx_oid; /* object id */ 384 lr_t itx_lr; /* common part of log record */ 385 /* followed by type-specific part of lr_xx_t and its immediate data */ 386} itx_t; 387 388typedef int zil_parse_blk_func_t(zilog_t *zilog, blkptr_t *bp, void *arg, 389 uint64_t txg); 390typedef int zil_parse_lr_func_t(zilog_t *zilog, lr_t *lr, void *arg, 391 uint64_t txg); 392typedef int zil_replay_func_t(void *arg1, void *arg2, boolean_t byteswap); 393typedef int zil_get_data_t(void *arg, lr_write_t *lr, char *dbuf, 394 struct lwb *lwb, zio_t *zio); 395 396extern int zil_parse(zilog_t *zilog, zil_parse_blk_func_t *parse_blk_func, 397 zil_parse_lr_func_t *parse_lr_func, void *arg, uint64_t txg); 398 399extern void zil_init(void); 400extern void zil_fini(void); 401 402extern zilog_t *zil_alloc(objset_t *os, zil_header_t *zh_phys); 403extern void zil_free(zilog_t *zilog); 404 405extern zilog_t *zil_open(objset_t *os, zil_get_data_t *get_data); 406extern void zil_close(zilog_t *zilog); 407 408extern void zil_replay(objset_t *os, void *arg, 409 zil_replay_func_t *replay_func[TX_MAX_TYPE]); 410extern boolean_t zil_replaying(zilog_t *zilog, dmu_tx_t *tx); 411extern void zil_destroy(zilog_t *zilog, boolean_t keep_first); 412extern void zil_destroy_sync(zilog_t *zilog, dmu_tx_t *tx); 413extern void zil_rollback_destroy(zilog_t *zilog, dmu_tx_t *tx); 414 415extern itx_t *zil_itx_create(uint64_t txtype, size_t lrsize); 416extern void zil_itx_destroy(itx_t *itx); 417extern void zil_itx_assign(zilog_t *zilog, itx_t *itx, dmu_tx_t *tx); 418 419extern void zil_async_to_sync(zilog_t *zilog, uint64_t oid); 420extern void zil_commit(zilog_t *zilog, uint64_t oid); 421extern void zil_commit_impl(zilog_t *zilog, uint64_t oid); 422 423extern int zil_reset(const char *osname, void *txarg); 424extern int zil_claim(struct dsl_pool *dp, 425 struct dsl_dataset *ds, void *txarg); 426extern int zil_check_log_chain(struct dsl_pool *dp, 427 struct dsl_dataset *ds, void *tx); 428extern void zil_sync(zilog_t *zilog, dmu_tx_t *tx); 429extern void zil_clean(zilog_t *zilog, uint64_t synced_txg); 430 431extern int zil_suspend(const char *osname, void **cookiep); 432extern void zil_resume(void *cookie); 433 434extern void zil_lwb_add_block(struct lwb *lwb, const blkptr_t *bp); 435extern void zil_lwb_add_txg(struct lwb *lwb, uint64_t txg); 436extern int zil_bp_tree_add(zilog_t *zilog, const blkptr_t *bp); 437 438extern void zil_set_sync(zilog_t *zilog, uint64_t syncval); 439 440extern void zil_set_logbias(zilog_t *zilog, uint64_t slogval); 441 442extern int zil_replay_disable; 443 444#ifdef __cplusplus 445} 446#endif 447 448#endif /* _SYS_ZIL_H */ 449