1 /* 2 * Copyright (c) 2007 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@backplane.com> 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 3. Neither the name of The DragonFly Project nor the names of its 18 * contributors may be used to endorse or promote products derived 19 * from this software without specific, prior written permission. 20 * 21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.55 2008/11/13 02:18:43 dillon Exp $ 35 */ 36 37 #ifndef VFS_HAMMER_DISK_H_ 38 #define VFS_HAMMER_DISK_H_ 39 40 #ifndef _SYS_UUID_H_ 41 #include <sys/uuid.h> 42 #endif 43 44 /* 45 * The structures below represent the on-disk format for a HAMMER 46 * filesystem. Note that all fields for on-disk structures are naturally 47 * aligned. The host endian format is used - compatibility is possible 48 * if the implementation detects reversed endian and adjusts data accordingly. 49 * 50 * Most of HAMMER revolves around the concept of an object identifier. An 51 * obj_id is a 64 bit quantity which uniquely identifies a filesystem object 52 * FOR THE ENTIRE LIFE OF THE FILESYSTEM. This uniqueness allows backups 53 * and mirrors to retain varying amounts of filesystem history by removing 54 * any possibility of conflict through identifier reuse. 55 * 56 * A HAMMER filesystem may span multiple volumes. 57 * 58 * A HAMMER filesystem uses a 16K filesystem buffer size. All filesystem 59 * I/O is done in multiples of 16K. Most buffer-sized headers such as those 60 * used by volumes, super-clusters, clusters, and basic filesystem buffers 61 * use fixed-sized A-lists which are heavily dependant on HAMMER_BUFSIZE. 62 * 63 * 64K X-bufs are used for blocks >= a file's 1MB mark. 64 * 65 * Per-volume storage limit: 52 bits 4096 TB 66 * Per-Zone storage limit: 59 bits 512 KTB (due to blockmap) 67 * Per-filesystem storage limit: 60 bits 1 MTB 68 */ 69 #define HAMMER_BUFSIZE 16384 70 #define HAMMER_XBUFSIZE 65536 71 #define HAMMER_XDEMARC (1024 * 1024) 72 #define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1) 73 #define HAMMER_XBUFMASK (HAMMER_XBUFSIZE - 1) 74 #define HAMMER_BUFFER_BITS 14 75 76 #if (1 << HAMMER_BUFFER_BITS) != HAMMER_BUFSIZE 77 #error "HAMMER_BUFFER_BITS BROKEN" 78 #endif 79 80 #define HAMMER_BUFSIZE64 ((u_int64_t)HAMMER_BUFSIZE) 81 #define HAMMER_BUFMASK64 ((u_int64_t)HAMMER_BUFMASK) 82 83 #define HAMMER_XBUFSIZE64 ((u_int64_t)HAMMER_XBUFSIZE) 84 #define HAMMER_XBUFMASK64 ((u_int64_t)HAMMER_XBUFMASK) 85 86 #define HAMMER_OFF_ZONE_MASK 0xF000000000000000ULL /* zone portion */ 87 #define HAMMER_OFF_VOL_MASK 0x0FF0000000000000ULL /* volume portion */ 88 #define HAMMER_OFF_SHORT_MASK 0x000FFFFFFFFFFFFFULL /* offset portion */ 89 #define HAMMER_OFF_LONG_MASK 0x0FFFFFFFFFFFFFFFULL /* offset portion */ 90 #define HAMMER_OFF_SHORT_REC_MASK 0x000FFFFFFF000000ULL /* recovery boundary */ 91 #define HAMMER_OFF_LONG_REC_MASK 0x0FFFFFFFFF000000ULL /* recovery boundary */ 92 #define HAMMER_RECOVERY_BND 0x0000000001000000ULL 93 94 /* 95 * The current limit of volumes that can make up a HAMMER FS 96 */ 97 #define HAMMER_MAX_VOLUMES 256 98 99 /* 100 * Hammer transction ids are 64 bit unsigned integers and are usually 101 * synchronized with the time of day in nanoseconds. 102 * 103 * Hammer offsets are used for FIFO indexing and embed a cycle counter 104 * and volume number in addition to the offset. Most offsets are required 105 * to be 64-byte aligned. 106 */ 107 typedef u_int64_t hammer_tid_t; 108 typedef u_int64_t hammer_off_t; 109 typedef u_int32_t hammer_seq_t; 110 typedef u_int32_t hammer_crc_t; 111 112 #define HAMMER_MIN_TID 0ULL /* unsigned */ 113 #define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */ 114 #define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */ 115 #define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */ 116 #define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */ 117 #define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */ 118 #define HAMMER_MIN_RECTYPE 0x0U /* unsigned */ 119 #define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */ 120 #define HAMMER_MIN_OFFSET 0ULL /* unsigned */ 121 #define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */ 122 123 /* 124 * hammer_off_t has several different encodings. Note that not all zones 125 * encode a vol_no. 126 * 127 * zone 0: reserved for sanity 128 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header) 129 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer) 130 * zone 3 (z,o): undo fifo - actually fixed phys array in vol hdr 131 * zone 4 (z,v,o): freemap - only real blockmap 132 * zone 8 (z,v,o): B-Tree - actually zone-2 address 133 * zone 9 (z,v,o): Record - actually zone-2 address 134 * zone 10 (z,v,o): Large-data - actually zone-2 address 135 * zone 15: reserved for sanity 136 * 137 * layer1/layer2 direct map: 138 * zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo 139 * ----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo 140 */ 141 142 #define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL 143 #define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL 144 #define HAMMER_ZONE_UNDO 0x3000000000000000ULL 145 #define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL 146 #define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL 147 #define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL 148 #define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL 149 #define HAMMER_ZONE_BTREE 0x8000000000000000ULL 150 #define HAMMER_ZONE_META 0x9000000000000000ULL 151 #define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL 152 #define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL 153 #define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL 154 #define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL 155 #define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL 156 #define HAMMER_ZONE_UNAVAIL 0xF000000000000000ULL 157 158 #define HAMMER_ZONE_RAW_VOLUME_INDEX 1 159 #define HAMMER_ZONE_RAW_BUFFER_INDEX 2 160 #define HAMMER_ZONE_UNDO_INDEX 3 161 #define HAMMER_ZONE_FREEMAP_INDEX 4 162 #define HAMMER_ZONE_BTREE_INDEX 8 163 #define HAMMER_ZONE_META_INDEX 9 164 #define HAMMER_ZONE_LARGE_DATA_INDEX 10 165 #define HAMMER_ZONE_SMALL_DATA_INDEX 11 166 #define HAMMER_ZONE_UNAVAIL_INDEX 15 /* unavailable */ 167 168 #define HAMMER_MAX_ZONES 16 169 170 #define HAMMER_VOL_ENCODE(vol_no) \ 171 ((hammer_off_t)((vol_no) & 255) << 52) 172 #define HAMMER_VOL_DECODE(ham_off) \ 173 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255) 174 #define HAMMER_ZONE_DECODE(ham_off) \ 175 (int32_t)(((hammer_off_t)(ham_off) >> 60)) 176 #define HAMMER_ZONE_ENCODE(zone, ham_off) \ 177 (((hammer_off_t)(zone) << 60) | (ham_off)) 178 #define HAMMER_SHORT_OFF_ENCODE(offset) \ 179 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK) 180 #define HAMMER_LONG_OFF_ENCODE(offset) \ 181 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK) 182 183 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \ 184 (HAMMER_ZONE_RAW_VOLUME | \ 185 HAMMER_VOL_ENCODE(vol_no) | \ 186 HAMMER_SHORT_OFF_ENCODE(offset)) 187 188 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \ 189 (HAMMER_ZONE_RAW_BUFFER | \ 190 HAMMER_VOL_ENCODE(vol_no) | \ 191 HAMMER_SHORT_OFF_ENCODE(offset)) 192 193 #define HAMMER_ENCODE_FREEMAP(vol_no, offset) \ 194 (HAMMER_ZONE_FREEMAP | \ 195 HAMMER_VOL_ENCODE(vol_no) | \ 196 HAMMER_SHORT_OFF_ENCODE(offset)) 197 198 /* 199 * Large-Block backing store 200 * 201 * A blockmap is a two-level map which translates a blockmap-backed zone 202 * offset into a raw zone 2 offset. Each layer handles 18 bits. The 8M 203 * large-block size is 23 bits so two layers gives us 23+18+18 = 59 bits 204 * of address space. 205 * 206 * When using hinting for a blockmap lookup, the hint is lost when the 207 * scan leaves the HINTBLOCK, which is typically several LARGEBLOCK's. 208 * HINTBLOCK is a heuristic. 209 */ 210 #define HAMMER_HINTBLOCK_SIZE (HAMMER_LARGEBLOCK_SIZE * 4) 211 #define HAMMER_HINTBLOCK_MASK64 ((u_int64_t)HAMMER_HINTBLOCK_SIZE - 1) 212 #define HAMMER_LARGEBLOCK_SIZE (8192 * 1024) 213 #define HAMMER_LARGEBLOCK_OVERFILL (6144 * 1024) 214 #define HAMMER_LARGEBLOCK_SIZE64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE) 215 #define HAMMER_LARGEBLOCK_MASK (HAMMER_LARGEBLOCK_SIZE - 1) 216 #define HAMMER_LARGEBLOCK_MASK64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE - 1) 217 #define HAMMER_LARGEBLOCK_BITS 23 218 #if (1 << HAMMER_LARGEBLOCK_BITS) != HAMMER_LARGEBLOCK_SIZE 219 #error "HAMMER_LARGEBLOCK_BITS BROKEN" 220 #endif 221 222 #define HAMMER_BUFFERS_PER_LARGEBLOCK \ 223 (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE) 224 #define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK \ 225 (HAMMER_BUFFERS_PER_LARGEBLOCK - 1) 226 #define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK64 \ 227 ((hammer_off_t)HAMMER_BUFFERS_PER_LARGEBLOCK_MASK) 228 229 /* 230 * Maximum number of mirrors operating in master mode (multi-master 231 * clustering and mirroring). 232 */ 233 #define HAMMER_MAX_MASTERS 16 234 235 /* 236 * The blockmap is somewhat of a degenerate structure. HAMMER only actually 237 * uses it in its original incarnation to implement the free-map. 238 * 239 * zone:1 raw volume (no blockmap) 240 * zone:2 raw buffer (no blockmap) 241 * zone:3 undo-map (direct layer2 array in volume header) 242 * zone:4 free-map (the only real blockmap) 243 * zone:8-15 zone id used to classify big-block only, address is actually 244 * a zone-2 address. 245 */ 246 struct hammer_blockmap { 247 hammer_off_t phys_offset; /* zone-2 physical offset */ 248 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */ 249 hammer_off_t next_offset; /* zone-X logical offset */ 250 hammer_off_t alloc_offset; /* zone-X logical offset */ 251 u_int32_t reserved01; 252 hammer_crc_t entry_crc; 253 }; 254 255 typedef struct hammer_blockmap *hammer_blockmap_t; 256 257 #define HAMMER_BLOCKMAP_CRCSIZE \ 258 offsetof(struct hammer_blockmap, entry_crc) 259 260 /* 261 * The blockmap is a 2-layer entity made up of big-blocks. The first layer 262 * contains 262144 32-byte entries (18 bits), the second layer contains 263 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps. 264 * 18+19+23 = 60 bits. The top four bits are the zone id. 265 * 266 * Currently only the freemap utilizes both layers in all their glory. 267 * All primary data/meta-data zones actually encode a zone-2 address 268 * requiring no real blockmap translation. 269 * 270 * The freemap uses the upper 8 bits of layer-1 to identify the volume, 271 * thus any space allocated via the freemap can be directly translated 272 * to a zone:2 (or zone:8-15) address. 273 * 274 * zone-X blockmap offset: [z:4][layer1:18][layer2:19][bigblock:23] 275 */ 276 struct hammer_blockmap_layer1 { 277 hammer_off_t blocks_free; /* big-blocks free */ 278 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */ 279 hammer_off_t reserved01; 280 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */ 281 /* (not yet used) */ 282 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/ 283 }; 284 285 typedef struct hammer_blockmap_layer1 *hammer_blockmap_layer1_t; 286 287 #define HAMMER_LAYER1_CRCSIZE \ 288 offsetof(struct hammer_blockmap_layer1, layer1_crc) 289 290 struct hammer_blockmap_layer2 { 291 u_int8_t zone; /* typed allocation zone */ 292 u_int8_t unused01; 293 u_int16_t unused02; 294 u_int32_t append_off; /* allocatable space index */ 295 u_int32_t bytes_free; /* bytes free within this bigblock */ 296 hammer_crc_t entry_crc; 297 }; 298 299 typedef struct hammer_blockmap_layer2 *hammer_blockmap_layer2_t; 300 301 #define HAMMER_LAYER2_CRCSIZE \ 302 offsetof(struct hammer_blockmap_layer2, entry_crc) 303 304 #define HAMMER_BLOCKMAP_FREE 0ULL 305 #define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL) 306 307 #define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \ 308 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1)) 309 #define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \ 310 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2)) 311 312 #define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \ 313 (HAMMER_BLOCKMAP_RADIX1 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE)) 314 #define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \ 315 (HAMMER_BLOCKMAP_RADIX2 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE)) 316 317 #define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 */ \ 318 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2) 319 #define HAMMER_BLOCKMAP_LAYER2 /* 19+23 - 4TB */ \ 320 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_LARGEBLOCK_SIZE64) 321 322 #define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1) 323 #define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1) 324 325 /* 326 * byte offset within layer1 or layer2 big-block for the entry representing 327 * a zone-2 physical offset. 328 */ 329 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \ 330 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \ 331 HAMMER_BLOCKMAP_LAYER2 * sizeof(struct hammer_blockmap_layer1)) 332 333 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \ 334 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \ 335 HAMMER_LARGEBLOCK_SIZE64 * sizeof(struct hammer_blockmap_layer2)) 336 337 /* 338 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume 339 * header with an array of layer2 structures. A maximum of (128x8MB) = 1GB 340 * may be reserved. The size of the undo fifo is usually set a newfs time 341 * but can be adjusted if the filesystem is taken offline. 342 */ 343 #define HAMMER_UNDO_LAYER2 128 /* max layer2 undo mapping entries */ 344 345 /* 346 * All on-disk HAMMER structures which make up elements of the UNDO FIFO 347 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure 348 * contains all the information required to validate the fifo element 349 * and to scan the fifo in either direction. The head is typically embedded 350 * in higher level hammer on-disk structures while the tail is typically 351 * out-of-band. hdr_size is the size of the whole mess, including the tail. 352 * 353 * All undo structures are guaranteed to not cross a 16K filesystem 354 * buffer boundary. Most undo structures are fairly small. Data spaces 355 * are not immediately reused by HAMMER so file data is not usually recorded 356 * as part of an UNDO. 357 * 358 * PAD elements are allowed to take up only 8 bytes of space as a special 359 * case, containing only hdr_signature, hdr_type, and hdr_size fields, 360 * and with the tail overloaded onto the head structure for 8 bytes total. 361 * 362 * Every undo record has a sequence number. This number is unrelated to 363 * transaction ids and instead collects the undo transactions associated 364 * with a single atomic operation. A larger transactional operation, such 365 * as a remove(), may consist of several smaller atomic operations 366 * representing raw meta-data operations. 367 * 368 * HAMMER VERSION 4 CHANGES 369 * 370 * In HAMMER version 4 the undo structure alignment is reduced from 16384 371 * to 512 bytes in order to ensure that each 512 byte sector begins with 372 * a header. The reserved01 field in the header is now a 32 bit sequence 373 * number. This allows the recovery code to detect missing sectors 374 * without relying on the 32-bit crc and to definitively identify the current 375 * undo sequence space without having to rely on information from the volume 376 * header. In addition, new REDO entries in the undo space are used to 377 * record write, write/extend, and transaction id updates. 378 * 379 * The grand result is: 380 * 381 * (1) The volume header no longer needs to be synchronized for most 382 * flush and fsync operations. 383 * 384 * (2) Most fsync operations need only lay down REDO records 385 * 386 * (3) Data overwrite for nohistory operations covered by REDO records 387 * can be supported (instead of rolling a new block allocation), 388 * by rolling UNDO for the prior contents of the data. 389 */ 390 #define HAMMER_HEAD_ONDISK_SIZE 32 391 #define HAMMER_HEAD_ALIGN 8 392 #define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1) 393 #define HAMMER_TAIL_ONDISK_SIZE 8 394 #define HAMMER_HEAD_DOALIGN(bytes) \ 395 (((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK) 396 397 #define HAMMER_UNDO_ALIGN 512 398 #define HAMMER_UNDO_ALIGN64 ((u_int64_t)512) 399 #define HAMMER_UNDO_MASK (HAMMER_UNDO_ALIGN - 1) 400 #define HAMMER_UNDO_MASK64 (HAMMER_UNDO_ALIGN64 - 1) 401 402 struct hammer_fifo_head { 403 u_int16_t hdr_signature; 404 u_int16_t hdr_type; 405 u_int32_t hdr_size; /* Aligned size of the whole mess */ 406 u_int32_t hdr_seq; /* Sequence number */ 407 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */ 408 }; 409 410 #define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc) 411 412 struct hammer_fifo_tail { 413 u_int16_t tail_signature; 414 u_int16_t tail_type; 415 u_int32_t tail_size; /* aligned size of the whole mess */ 416 }; 417 418 typedef struct hammer_fifo_head *hammer_fifo_head_t; 419 typedef struct hammer_fifo_tail *hammer_fifo_tail_t; 420 421 /* 422 * Fifo header types. 423 */ 424 #define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE) 425 #define HAMMER_HEAD_TYPE_DUMMY 0x0041U /* dummy entry w/seqno */ 426 #define HAMMER_HEAD_TYPE_42 0x0042U 427 #define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */ 428 #define HAMMER_HEAD_TYPE_REDO 0x0044U /* data REDO / fast fsync */ 429 #define HAMMER_HEAD_TYPE_45 0x0045U 430 431 #define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */ 432 433 #define HAMMER_HEAD_SIGNATURE 0xC84EU 434 #define HAMMER_TAIL_SIGNATURE 0xC74FU 435 436 #define HAMMER_HEAD_SEQ_BEG 0x80000000U 437 #define HAMMER_HEAD_SEQ_END 0x40000000U 438 #define HAMMER_HEAD_SEQ_MASK 0x3FFFFFFFU 439 440 /* 441 * Misc FIFO structures. 442 * 443 * UNDO - Raw meta-data media updates. 444 */ 445 struct hammer_fifo_undo { 446 struct hammer_fifo_head head; 447 hammer_off_t undo_offset; /* zone-1 offset */ 448 int32_t undo_data_bytes; 449 int32_t undo_reserved01; 450 /* followed by data */ 451 }; 452 453 /* 454 * REDO (HAMMER version 4+) - Logical file writes/truncates. 455 * 456 * REDOs contain information which will be duplicated in a later meta-data 457 * update, allowing fast write()+fsync() operations. REDOs can be ignored 458 * without harming filesystem integrity but must be processed if fsync() 459 * semantics are desired. 460 * 461 * Unlike UNDOs which are processed backwards within the recovery span, 462 * REDOs must be processed forwards starting further back (starting outside 463 * the recovery span). 464 * 465 * WRITE - Write logical file (with payload). Executed both 466 * out-of-span and in-span. Out-of-span WRITEs may be 467 * filtered out by TERMs. 468 * 469 * TRUNC - Truncate logical file (no payload). Executed both 470 * out-of-span and in-span. Out-of-span WRITEs may be 471 * filtered out by TERMs. 472 * 473 * TERM_* - Indicates meta-data was committed (if out-of-span) or 474 * will be rolled-back (in-span). Any out-of-span TERMs 475 * matching earlier WRITEs remove those WRITEs from 476 * consideration as they might conflict with a later data 477 * commit (which is not being rolled-back). 478 * 479 * SYNC - The earliest in-span SYNC (the last one when scanning 480 * backwards) tells the recovery code how far out-of-span 481 * it must go to run REDOs. 482 * 483 * NOTE: WRITEs do not always have matching TERMs even under 484 * perfect conditions because truncations might remove the 485 * buffers from consideration. I/O problems can also remove 486 * buffers from consideration. 487 * 488 * TRUNCSs do not always have matching TERMs because several 489 * truncations may be aggregated together into a single TERM. 490 */ 491 struct hammer_fifo_redo { 492 struct hammer_fifo_head head; 493 int64_t redo_objid; /* file being written */ 494 hammer_off_t redo_offset; /* logical offset in file */ 495 int32_t redo_data_bytes; 496 u_int32_t redo_flags; 497 u_int32_t redo_localization; 498 u_int32_t redo_reserved; 499 u_int64_t redo_mtime; /* set mtime */ 500 }; 501 502 #define HAMMER_REDO_WRITE 0x00000001 503 #define HAMMER_REDO_TRUNC 0x00000002 504 #define HAMMER_REDO_TERM_WRITE 0x00000004 505 #define HAMMER_REDO_TERM_TRUNC 0x00000008 506 #define HAMMER_REDO_SYNC 0x00000010 507 508 union hammer_fifo_any { 509 struct hammer_fifo_head head; 510 struct hammer_fifo_undo undo; 511 struct hammer_fifo_redo redo; 512 }; 513 514 typedef struct hammer_fifo_redo *hammer_fifo_redo_t; 515 typedef struct hammer_fifo_undo *hammer_fifo_undo_t; 516 typedef union hammer_fifo_any *hammer_fifo_any_t; 517 518 /* 519 * Volume header types 520 */ 521 #define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */ 522 #define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */ 523 524 /* 525 * The B-Tree structures need hammer_fsbuf_head. 526 */ 527 #include "hammer_btree.h" 528 529 /* 530 * HAMMER Volume header 531 * 532 * A HAMMER filesystem is built from any number of block devices, Each block 533 * device contains a volume header followed by however many buffers fit 534 * into the volume. 535 * 536 * One of the volumes making up a HAMMER filesystem is the master, the 537 * rest are slaves. It does not have to be volume #0. 538 * 539 * The volume header takes up an entire 16K filesystem buffer and may 540 * represent up to 64KTB (65536 TB) of space. 541 * 542 * Special field notes: 543 * 544 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes) 545 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes) 546 * vol_buf_beg - offset of the first buffer. 547 * 548 * The memory log area allows a kernel to cache new records and data 549 * in memory without allocating space in the actual filesystem to hold 550 * the records and data. In the event that a filesystem becomes full, 551 * any records remaining in memory can be flushed to the memory log 552 * area. This allows the kernel to immediately return success. 553 */ 554 555 #define HAMMER_BOOT_MINBYTES (32*1024) 556 #define HAMMER_BOOT_NOMBYTES (64LL*1024*1024) 557 #define HAMMER_BOOT_MAXBYTES (256LL*1024*1024) 558 559 #define HAMMER_MEM_MINBYTES (256*1024) 560 #define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024) 561 #define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024) 562 563 struct hammer_volume_ondisk { 564 u_int64_t vol_signature;/* Signature */ 565 566 int64_t vol_bot_beg; /* byte offset of boot area or 0 */ 567 int64_t vol_mem_beg; /* byte offset of memory log or 0 */ 568 int64_t vol_buf_beg; /* byte offset of first buffer in volume */ 569 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */ 570 int64_t vol_locked; /* reserved clusters are >= this offset */ 571 572 uuid_t vol_fsid; /* identify filesystem */ 573 uuid_t vol_fstype; /* identify filesystem type */ 574 char vol_name[64]; /* Name of volume */ 575 576 int32_t vol_no; /* volume number within filesystem */ 577 int32_t vol_count; /* number of volumes making up FS */ 578 579 u_int32_t vol_version; /* version control information */ 580 hammer_crc_t vol_crc; /* header crc */ 581 u_int32_t vol_flags; /* volume flags */ 582 u_int32_t vol_rootvol; /* which volume is the root volume? */ 583 584 int32_t vol_reserved04; 585 int32_t vol_reserved05; 586 u_int32_t vol_reserved06; 587 u_int32_t vol_reserved07; 588 589 int32_t vol_blocksize; /* for statfs only */ 590 int32_t vol_reserved08; 591 int64_t vol_nblocks; /* total allocatable hammer bufs */ 592 593 /* 594 * These fields are initialized and space is reserved in every 595 * volume making up a HAMMER filesytem, but only the master volume 596 * contains valid data. 597 */ 598 int64_t vol0_stat_bigblocks; /* total bigblocks when fs is empty */ 599 int64_t vol0_stat_freebigblocks;/* number of free bigblocks */ 600 int64_t vol0_stat_bytes; /* for statfs only */ 601 int64_t vol0_stat_inodes; /* for statfs only */ 602 int64_t vol0_stat_records; /* total records in filesystem */ 603 hammer_off_t vol0_btree_root; /* B-Tree root */ 604 hammer_tid_t vol0_next_tid; /* highest partially synchronized TID */ 605 hammer_off_t vol0_unused03; 606 607 /* 608 * Blockmaps for zones. Not all zones use a blockmap. Note that 609 * the entire root blockmap is cached in the hammer_mount structure. 610 */ 611 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES]; 612 613 /* 614 * Array of zone-2 addresses for undo FIFO. 615 */ 616 hammer_off_t vol0_undo_array[HAMMER_UNDO_LAYER2]; 617 618 }; 619 620 typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t; 621 622 #define HAMMER_VOLF_VALID 0x0001 /* valid entry */ 623 #define HAMMER_VOLF_OPEN 0x0002 /* volume is open */ 624 #define HAMMER_VOLF_NEEDFLUSH 0x0004 /* volume needs flush */ 625 626 #define HAMMER_VOL_CRCSIZE1 \ 627 offsetof(struct hammer_volume_ondisk, vol_crc) 628 #define HAMMER_VOL_CRCSIZE2 \ 629 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \ 630 sizeof(hammer_crc_t)) 631 632 #define HAMMER_VOL_VERSION_MIN 1 /* minimum supported version */ 633 #define HAMMER_VOL_VERSION_DEFAULT 4 /* newfs default version */ 634 #define HAMMER_VOL_VERSION_WIP 5 /* version >= this is WIP */ 635 #define HAMMER_VOL_VERSION_MAX 4 /* maximum supported version */ 636 637 #define HAMMER_VOL_VERSION_ONE 1 638 #define HAMMER_VOL_VERSION_TWO 2 /* new dirent layout (2.3+) */ 639 #define HAMMER_VOL_VERSION_THREE 3 /* new snapshot layout (2.5+) */ 640 #define HAMMER_VOL_VERSION_FOUR 4 /* new undo/flush (2.5+) */ 641 642 /* 643 * Record types are fairly straightforward. The B-Tree includes the record 644 * type in its index sort. 645 */ 646 #define HAMMER_RECTYPE_UNKNOWN 0 647 #define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */ 648 #define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */ 649 #define HAMMER_RECTYPE_UNUSED02 2 650 #define HAMMER_RECTYPE_UNUSED03 3 651 #define HAMMER_RECTYPE_DATA 0x0010 652 #define HAMMER_RECTYPE_DIRENTRY 0x0011 653 #define HAMMER_RECTYPE_DB 0x0012 654 #define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */ 655 #define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */ 656 #define HAMMER_RECTYPE_PFS 0x0015 /* PFS management */ 657 #define HAMMER_RECTYPE_SNAPSHOT 0x0016 /* Snapshot management */ 658 #define HAMMER_RECTYPE_CONFIG 0x0017 /* hammer cleanup config */ 659 #define HAMMER_RECTYPE_MOVED 0x8000 /* special recovery flag */ 660 #define HAMMER_RECTYPE_MAX 0xFFFF 661 662 #define HAMMER_RECTYPE_CLEAN_START HAMMER_RECTYPE_EXT 663 664 #define HAMMER_FIXKEY_SYMLINK 1 665 666 #define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */ 667 #define HAMMER_OBJTYPE_DIRECTORY 1 668 #define HAMMER_OBJTYPE_REGFILE 2 669 #define HAMMER_OBJTYPE_DBFILE 3 670 #define HAMMER_OBJTYPE_FIFO 4 671 #define HAMMER_OBJTYPE_CDEV 5 672 #define HAMMER_OBJTYPE_BDEV 6 673 #define HAMMER_OBJTYPE_SOFTLINK 7 674 #define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */ 675 #define HAMMER_OBJTYPE_SOCKET 9 676 677 /* 678 * HAMMER inode attribute data 679 * 680 * The data reference for a HAMMER inode points to this structure. Any 681 * modifications to the contents of this structure will result in a 682 * replacement operation. 683 * 684 * parent_obj_id is only valid for directories (which cannot be hard-linked), 685 * and specifies the parent directory obj_id. This field will also be set 686 * for non-directory inodes as a recovery aid, but can wind up holding 687 * stale information. However, since object id's are not reused, the worse 688 * that happens is that the recovery code is unable to use it. 689 * 690 * NOTE: Future note on directory hardlinks. We can implement a record type 691 * which allows us to point to multiple parent directories. 692 * 693 * NOTE: atime is stored in the inode's B-Tree element and not in the inode 694 * data. This allows the atime to be updated without having to lay down a 695 * new record. 696 */ 697 struct hammer_inode_data { 698 u_int16_t version; /* inode data version */ 699 u_int16_t mode; /* basic unix permissions */ 700 u_int32_t uflags; /* chflags */ 701 u_int32_t rmajor; /* used by device nodes */ 702 u_int32_t rminor; /* used by device nodes */ 703 u_int64_t ctime; 704 int64_t parent_obj_id; /* parent directory obj_id */ 705 uuid_t uid; 706 uuid_t gid; 707 708 u_int8_t obj_type; 709 u_int8_t cap_flags; /* capability support flags (extension) */ 710 u_int16_t reserved02; 711 u_int32_t reserved03; /* RESERVED FOR POSSIBLE FUTURE BIRTHTIME */ 712 u_int64_t nlinks; /* hard links */ 713 u_int64_t size; /* filesystem object size */ 714 union { 715 struct { 716 char reserved06[16]; 717 u_int32_t parent_obj_localization; 718 u_int32_t integrity_crc; 719 } obj; 720 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */ 721 } ext; 722 u_int64_t mtime; /* mtime must be second-to-last */ 723 u_int64_t atime; /* atime must be last */ 724 }; 725 726 /* 727 * Neither mtime nor atime upates are CRCd by the B-Tree element. 728 * mtime updates have UNDO, atime updates do not. 729 */ 730 #define HAMMER_ITIMES_BASE(ino_data) (&(ino_data)->mtime) 731 #define HAMMER_ITIMES_BYTES (sizeof(u_int64_t) * 2) 732 733 #define HAMMER_INODE_CRCSIZE \ 734 offsetof(struct hammer_inode_data, mtime) 735 736 #define HAMMER_INODE_DATA_VERSION 1 737 #define HAMMER_OBJID_ROOT 1 738 #define HAMMER_INODE_BASESYMLEN 24 /* see ext.symlink */ 739 740 /* 741 * Capability & implementation flags. 742 * 743 * DIR_LOCAL_INO - Use inode B-Tree localization for directory entries. 744 */ 745 #define HAMMER_INODE_CAP_DIRHASH_MASK 0x03 /* directory: hash algorithm */ 746 #define HAMMER_INODE_CAP_DIRHASH_ALG0 0x00 747 #define HAMMER_INODE_CAP_DIRHASH_ALG1 0x01 748 #define HAMMER_INODE_CAP_DIRHASH_ALG2 0x02 749 #define HAMMER_INODE_CAP_DIRHASH_ALG3 0x03 750 #define HAMMER_INODE_CAP_DIR_LOCAL_INO 0x04 /* use inode localization */ 751 752 /* 753 * A HAMMER directory entry associates a HAMMER filesystem object with a 754 * namespace. It is possible to hook into a pseudo-filesystem (with its 755 * own inode numbering space) in the filesystem by setting the high 756 * 16 bits of the localization field. The low 16 bits must be 0 and 757 * are reserved for future use. 758 * 759 * Directory entries are indexed with a 128 bit namekey rather then an 760 * offset. A portion of the namekey is an iterator/randomizer to deal 761 * with collisions. 762 * 763 * NOTE: base.base.obj_type from the related B-Tree leaf entry holds 764 * the filesystem object type of obj_id, e.g. a den_type equivalent. 765 * It is not stored in hammer_entry_data. 766 * 767 * NOTE: den_name / the filename data reference is NOT terminated with \0. 768 */ 769 struct hammer_entry_data { 770 int64_t obj_id; /* object being referenced */ 771 u_int32_t localization; /* identify pseudo-filesystem */ 772 u_int32_t reserved02; 773 char name[16]; /* name (extended) */ 774 }; 775 776 #define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_entry_data, name[0]) 777 #define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_entry_data, name[nlen]) 778 779 /* 780 * Symlink data which does not fit in the inode is stored in a separte 781 * FIX type record. 782 */ 783 struct hammer_symlink_data { 784 char name[16]; 785 }; 786 787 #define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0]) 788 789 /* 790 * The root inode for the primary filesystem and root inode for any 791 * pseudo-fs may be tagged with an optional data structure using 792 * HAMMER_RECTYPE_FIX/HAMMER_FIXKEY_PSEUDOFS. This structure allows 793 * the node to be used as a mirroring master or slave. 794 * 795 * When operating as a slave CD's into the node automatically become read-only 796 * and as-of sync_end_tid. 797 * 798 * When operating as a master the read PFSD info sets sync_end_tid to 799 * the most recently flushed TID. 800 * 801 * sync_low_tid is not yet used but will represent the highest pruning 802 * end-point, after which full history is available. 803 * 804 * We need to pack this structure making it equally sized on both 32-bit and 805 * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is 806 * send over the wire in hammer mirror operations. Only on 64-bit machines 807 * the size of this struct differ when packed or not. This leads us to the 808 * situation where old 64-bit systems (using the non-packed structure), 809 * which were never able to mirror to/from 32-bit systems, are now no longer 810 * able to mirror to/from newer 64-bit systems (using the packed structure). 811 */ 812 struct hammer_pseudofs_data { 813 hammer_tid_t sync_low_tid; /* full history beyond this point */ 814 hammer_tid_t sync_beg_tid; /* earliest tid w/ full history avail */ 815 hammer_tid_t sync_end_tid; /* current synchronizatoin point */ 816 u_int64_t sync_beg_ts; /* real-time of last completed sync */ 817 u_int64_t sync_end_ts; /* initiation of current sync cycle */ 818 uuid_t shared_uuid; /* shared uuid (match required) */ 819 uuid_t unique_uuid; /* unique uuid of this master/slave */ 820 int32_t reserved01; /* reserved for future master_id */ 821 int32_t mirror_flags; /* misc flags */ 822 char label[64]; /* filesystem space label */ 823 char snapshots[64]; /* softlink dir for pruning */ 824 int16_t prune_time; /* how long to spend pruning */ 825 int16_t prune_freq; /* how often we prune */ 826 int16_t reblock_time; /* how long to spend reblocking */ 827 int16_t reblock_freq; /* how often we reblock */ 828 int32_t snapshot_freq; /* how often we create a snapshot */ 829 int32_t prune_min; /* do not prune recent history */ 830 int32_t prune_max; /* do not retain history beyond here */ 831 int32_t reserved[16]; 832 } __packed; 833 834 typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t; 835 836 #define HAMMER_PFSD_SLAVE 0x00000001 837 #define HAMMER_PFSD_DELETED 0x80000000 838 839 /* 840 * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }. 841 * 842 * Snapshot records replace the old <fs>/snapshots/<softlink> methodology. Snapshot 843 * records are mirrored but may be independantly managed once they are laid down on 844 * a slave. 845 * 846 * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the 847 * results. 848 * 849 * NOTE: Reserved fields must be zero (as usual) 850 */ 851 struct hammer_snapshot_data { 852 hammer_tid_t tid; /* the snapshot TID itself (== key) */ 853 u_int64_t ts; /* real-time when snapshot was made */ 854 u_int64_t reserved01; 855 u_int64_t reserved02; 856 char label[64]; /* user-supplied description */ 857 u_int64_t reserved03[4]; 858 }; 859 860 /* 861 * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }. 862 * 863 * Used to store the hammer cleanup config. This data is not mirrored. 864 */ 865 struct hammer_config_data { 866 char text[1024]; 867 }; 868 869 /* 870 * Rollup various structures embedded as record data 871 */ 872 union hammer_data_ondisk { 873 struct hammer_entry_data entry; 874 struct hammer_inode_data inode; 875 struct hammer_symlink_data symlink; 876 struct hammer_pseudofs_data pfsd; 877 struct hammer_snapshot_data snap; 878 struct hammer_config_data config; 879 }; 880 881 typedef union hammer_data_ondisk *hammer_data_ondisk_t; 882 883 #endif 884