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: 60 bits 1 MTB 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 #define HAMMER_OFF_BAD ((hammer_off_t)-1) 95 96 /* 97 * The current limit of volumes that can make up a HAMMER FS 98 */ 99 #define HAMMER_MAX_VOLUMES 256 100 101 /* 102 * Hammer transction ids are 64 bit unsigned integers and are usually 103 * synchronized with the time of day in nanoseconds. 104 * 105 * Hammer offsets are used for FIFO indexing and embed a cycle counter 106 * and volume number in addition to the offset. Most offsets are required 107 * to be 16 KB aligned. 108 */ 109 typedef u_int64_t hammer_tid_t; 110 typedef u_int64_t hammer_off_t; 111 typedef u_int32_t hammer_seq_t; 112 typedef u_int32_t hammer_crc_t; 113 114 #define HAMMER_MIN_TID 0ULL /* unsigned */ 115 #define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */ 116 #define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */ 117 #define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */ 118 #define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */ 119 #define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */ 120 #define HAMMER_MIN_RECTYPE 0x0U /* unsigned */ 121 #define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */ 122 #define HAMMER_MIN_OFFSET 0ULL /* unsigned */ 123 #define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */ 124 125 /* 126 * hammer_off_t has several different encodings. Note that not all zones 127 * encode a vol_no. 128 * 129 * zone 0: reserved for sanity 130 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header) 131 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer) 132 * zone 3 (z,o): undo fifo - actually fixed phys array in vol hdr 133 * zone 4 (z,v,o): freemap - only real blockmap 134 * zone 8 (z,v,o): B-Tree - actually zone-2 address 135 * zone 9 (z,v,o): Record - actually zone-2 address 136 * zone 10 (z,v,o): Large-data - actually zone-2 address 137 * zone 11 (z,v,o): Small-data - actually zone-2 address 138 * zone 15: reserved for sanity 139 * 140 * layer1/layer2 direct map: 141 * zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo 142 * ----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo 143 */ 144 145 #define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL 146 #define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL 147 #define HAMMER_ZONE_UNDO 0x3000000000000000ULL 148 #define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL 149 #define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL 150 #define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL 151 #define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL 152 #define HAMMER_ZONE_BTREE 0x8000000000000000ULL 153 #define HAMMER_ZONE_META 0x9000000000000000ULL 154 #define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL 155 #define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL 156 #define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL 157 #define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL 158 #define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL 159 #define HAMMER_ZONE_UNAVAIL 0xF000000000000000ULL 160 161 #define HAMMER_ZONE_RAW_VOLUME_INDEX 1 162 #define HAMMER_ZONE_RAW_BUFFER_INDEX 2 163 #define HAMMER_ZONE_UNDO_INDEX 3 164 #define HAMMER_ZONE_FREEMAP_INDEX 4 165 #define HAMMER_ZONE_BTREE_INDEX 8 166 #define HAMMER_ZONE_META_INDEX 9 167 #define HAMMER_ZONE_LARGE_DATA_INDEX 10 168 #define HAMMER_ZONE_SMALL_DATA_INDEX 11 169 #define HAMMER_ZONE_UNAVAIL_INDEX 15 /* unavailable */ 170 171 #define HAMMER_MAX_ZONES 16 172 173 #define HAMMER_VOL_ENCODE(vol_no) \ 174 ((hammer_off_t)((vol_no) & 255) << 52) 175 #define HAMMER_VOL_DECODE(ham_off) \ 176 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255) 177 #define HAMMER_ZONE_DECODE(ham_off) \ 178 (int32_t)(((hammer_off_t)(ham_off) >> 60)) 179 #define HAMMER_ZONE_ENCODE(zone, ham_off) \ 180 (((hammer_off_t)(zone) << 60) | (ham_off)) 181 #define HAMMER_SHORT_OFF_ENCODE(offset) \ 182 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK) 183 #define HAMMER_LONG_OFF_ENCODE(offset) \ 184 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK) 185 186 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \ 187 (HAMMER_ZONE_RAW_VOLUME | \ 188 HAMMER_VOL_ENCODE(vol_no) | \ 189 HAMMER_SHORT_OFF_ENCODE(offset)) 190 191 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \ 192 (HAMMER_ZONE_RAW_BUFFER | \ 193 HAMMER_VOL_ENCODE(vol_no) | \ 194 HAMMER_SHORT_OFF_ENCODE(offset)) 195 196 #define HAMMER_ENCODE_FREEMAP(vol_no, offset) \ 197 (HAMMER_ZONE_FREEMAP | \ 198 HAMMER_VOL_ENCODE(vol_no) | \ 199 HAMMER_SHORT_OFF_ENCODE(offset)) 200 201 /* 202 * Big-Block backing store 203 * 204 * A blockmap is a two-level map which translates a blockmap-backed zone 205 * offset into a raw zone 2 offset. The layer 1 handles 18 bits and the 206 * layer 2 handles 19 bits. The 8M big-block size is 23 bits so two 207 * layers gives us 18+19+23 = 60 bits of address space. 208 * 209 * When using hinting for a blockmap lookup, the hint is lost when the 210 * scan leaves the HINTBLOCK, which is typically several BIGBLOCK's. 211 * HINTBLOCK is a heuristic. 212 */ 213 #define HAMMER_HINTBLOCK_SIZE (HAMMER_BIGBLOCK_SIZE * 4) 214 #define HAMMER_HINTBLOCK_MASK64 ((u_int64_t)HAMMER_HINTBLOCK_SIZE - 1) 215 #define HAMMER_BIGBLOCK_SIZE (8192 * 1024) 216 #define HAMMER_BIGBLOCK_OVERFILL (6144 * 1024) 217 #define HAMMER_BIGBLOCK_SIZE64 ((u_int64_t)HAMMER_BIGBLOCK_SIZE) 218 #define HAMMER_BIGBLOCK_MASK (HAMMER_BIGBLOCK_SIZE - 1) 219 #define HAMMER_BIGBLOCK_MASK64 ((u_int64_t)HAMMER_BIGBLOCK_SIZE - 1) 220 #define HAMMER_BIGBLOCK_BITS 23 221 #if (1 << HAMMER_BIGBLOCK_BITS) != HAMMER_BIGBLOCK_SIZE 222 #error "HAMMER_BIGBLOCK_BITS BROKEN" 223 #endif 224 225 #define HAMMER_BUFFERS_PER_BIGBLOCK \ 226 (HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE) 227 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK \ 228 (HAMMER_BUFFERS_PER_BIGBLOCK - 1) 229 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK64 \ 230 ((hammer_off_t)HAMMER_BUFFERS_PER_BIGBLOCK_MASK) 231 232 /* 233 * Maximum number of mirrors operating in master mode (multi-master 234 * clustering and mirroring). 235 */ 236 #define HAMMER_MAX_MASTERS 16 237 238 /* 239 * The blockmap is somewhat of a degenerate structure. HAMMER only actually 240 * uses it in its original incarnation to implement the free-map. 241 * 242 * zone:1 raw volume (no blockmap) 243 * zone:2 raw buffer (no blockmap) 244 * zone:3 undo-map (direct layer2 array in volume header) 245 * zone:4 free-map (the only real blockmap) 246 * zone:8-15 zone id used to classify big-block only, address is actually 247 * a zone-2 address. 248 */ 249 struct hammer_blockmap { 250 hammer_off_t phys_offset; /* zone-2 physical offset */ 251 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */ 252 hammer_off_t next_offset; /* zone-X logical offset */ 253 hammer_off_t alloc_offset; /* zone-X logical offset */ 254 u_int32_t reserved01; 255 hammer_crc_t entry_crc; 256 }; 257 258 typedef struct hammer_blockmap *hammer_blockmap_t; 259 260 #define HAMMER_BLOCKMAP_CRCSIZE \ 261 offsetof(struct hammer_blockmap, entry_crc) 262 263 /* 264 * The blockmap is a 2-layer entity made up of big-blocks. The first layer 265 * contains 262144 32-byte entries (18 bits), the second layer contains 266 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps. 267 * 18+19+23 = 60 bits. The top four bits are the zone id. 268 * 269 * Currently only the freemap utilizes both layers in all their glory. 270 * All primary data/meta-data zones actually encode a zone-2 address 271 * requiring no real blockmap translation. 272 * 273 * The freemap uses the upper 8 bits of layer-1 to identify the volume, 274 * thus any space allocated via the freemap can be directly translated 275 * to a zone:2 (or zone:8-15) address. 276 * 277 * zone-X blockmap offset: [zone:4][layer1:18][layer2:19][bigblock:23] 278 */ 279 struct hammer_blockmap_layer1 { 280 hammer_off_t blocks_free; /* big-blocks free */ 281 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */ 282 hammer_off_t reserved01; 283 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */ 284 /* (not yet used) */ 285 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/ 286 }; 287 288 typedef struct hammer_blockmap_layer1 *hammer_blockmap_layer1_t; 289 290 #define HAMMER_LAYER1_CRCSIZE \ 291 offsetof(struct hammer_blockmap_layer1, layer1_crc) 292 293 /* 294 * layer2 entry for 8MB bigblock. 295 * 296 * NOTE: bytes_free is signed and can legally go negative if/when data 297 * de-dup occurs. This field will never go higher than 298 * HAMMER_BIGBLOCK_SIZE. If exactly HAMMER_BIGBLOCK_SIZE 299 * the big-block is completely free. 300 */ 301 struct hammer_blockmap_layer2 { 302 u_int8_t zone; /* typed allocation zone */ 303 u_int8_t unused01; 304 u_int16_t unused02; 305 u_int32_t append_off; /* allocatable space index */ 306 int32_t bytes_free; /* bytes free within this bigblock */ 307 hammer_crc_t entry_crc; 308 }; 309 310 typedef struct hammer_blockmap_layer2 *hammer_blockmap_layer2_t; 311 312 #define HAMMER_LAYER2_CRCSIZE \ 313 offsetof(struct hammer_blockmap_layer2, entry_crc) 314 315 #define HAMMER_BLOCKMAP_FREE 0ULL 316 #define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL) 317 318 #define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \ 319 (HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1)) 320 #define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \ 321 (HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2)) 322 323 #define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \ 324 (HAMMER_BLOCKMAP_RADIX1 / (HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE)) 325 #define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \ 326 (HAMMER_BLOCKMAP_RADIX2 / (HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE)) 327 328 #define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 - 1EB */ \ 329 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2) 330 #define HAMMER_BLOCKMAP_LAYER2 /* 19+23 - 4TB */ \ 331 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_BIGBLOCK_SIZE64) 332 333 #define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1) 334 #define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1) 335 336 /* 337 * byte offset within layer1 or layer2 big-block for the entry representing 338 * a zone-2 physical offset. 339 */ 340 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \ 341 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \ 342 HAMMER_BLOCKMAP_LAYER2 * sizeof(struct hammer_blockmap_layer1)) 343 344 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \ 345 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \ 346 HAMMER_BIGBLOCK_SIZE64 * sizeof(struct hammer_blockmap_layer2)) 347 348 /* 349 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume 350 * header with an array of layer2 structures. A maximum of (128x8MB) = 1GB 351 * may be reserved. The size of the undo fifo is usually set a newfs time 352 * but can be adjusted if the filesystem is taken offline. 353 */ 354 #define HAMMER_UNDO_LAYER2 128 /* max layer2 undo mapping entries */ 355 356 /* 357 * All on-disk HAMMER structures which make up elements of the UNDO FIFO 358 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure 359 * contains all the information required to validate the fifo element 360 * and to scan the fifo in either direction. The head is typically embedded 361 * in higher level hammer on-disk structures while the tail is typically 362 * out-of-band. hdr_size is the size of the whole mess, including the tail. 363 * 364 * All undo structures are guaranteed to not cross a 16K filesystem 365 * buffer boundary. Most undo structures are fairly small. Data spaces 366 * are not immediately reused by HAMMER so file data is not usually recorded 367 * as part of an UNDO. 368 * 369 * PAD elements are allowed to take up only 8 bytes of space as a special 370 * case, containing only hdr_signature, hdr_type, and hdr_size fields, 371 * and with the tail overloaded onto the head structure for 8 bytes total. 372 * 373 * Every undo record has a sequence number. This number is unrelated to 374 * transaction ids and instead collects the undo transactions associated 375 * with a single atomic operation. A larger transactional operation, such 376 * as a remove(), may consist of several smaller atomic operations 377 * representing raw meta-data operations. 378 * 379 * HAMMER VERSION 4 CHANGES 380 * 381 * In HAMMER version 4 the undo structure alignment is reduced from 16384 382 * to 512 bytes in order to ensure that each 512 byte sector begins with 383 * a header. The reserved01 field in the header is now a 32 bit sequence 384 * number. This allows the recovery code to detect missing sectors 385 * without relying on the 32-bit crc and to definitively identify the current 386 * undo sequence space without having to rely on information from the volume 387 * header. In addition, new REDO entries in the undo space are used to 388 * record write, write/extend, and transaction id updates. 389 * 390 * The grand result is: 391 * 392 * (1) The volume header no longer needs to be synchronized for most 393 * flush and fsync operations. 394 * 395 * (2) Most fsync operations need only lay down REDO records 396 * 397 * (3) Data overwrite for nohistory operations covered by REDO records 398 * can be supported (instead of rolling a new block allocation), 399 * by rolling UNDO for the prior contents of the data. 400 * 401 * HAMMER VERSION 5 CHANGES 402 * 403 * Hammer version 5 contains a minor adjustment making layer2's bytes_free 404 * field signed, allowing dedup to push it into the negative domain. 405 */ 406 #define HAMMER_HEAD_ONDISK_SIZE 32 407 #define HAMMER_HEAD_ALIGN 8 408 #define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1) 409 #define HAMMER_TAIL_ONDISK_SIZE 8 410 #define HAMMER_HEAD_DOALIGN(bytes) \ 411 (((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK) 412 413 #define HAMMER_UNDO_ALIGN 512 414 #define HAMMER_UNDO_ALIGN64 ((u_int64_t)512) 415 #define HAMMER_UNDO_MASK (HAMMER_UNDO_ALIGN - 1) 416 #define HAMMER_UNDO_MASK64 (HAMMER_UNDO_ALIGN64 - 1) 417 418 struct hammer_fifo_head { 419 u_int16_t hdr_signature; 420 u_int16_t hdr_type; 421 u_int32_t hdr_size; /* Aligned size of the whole mess */ 422 u_int32_t hdr_seq; /* Sequence number */ 423 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */ 424 }; 425 426 #define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc) 427 428 struct hammer_fifo_tail { 429 u_int16_t tail_signature; 430 u_int16_t tail_type; 431 u_int32_t tail_size; /* aligned size of the whole mess */ 432 }; 433 434 typedef struct hammer_fifo_head *hammer_fifo_head_t; 435 typedef struct hammer_fifo_tail *hammer_fifo_tail_t; 436 437 /* 438 * Fifo header types. 439 */ 440 #define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE) 441 #define HAMMER_HEAD_TYPE_DUMMY 0x0041U /* dummy entry w/seqno */ 442 #define HAMMER_HEAD_TYPE_42 0x0042U 443 #define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */ 444 #define HAMMER_HEAD_TYPE_REDO 0x0044U /* data REDO / fast fsync */ 445 #define HAMMER_HEAD_TYPE_45 0x0045U 446 447 #define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */ 448 449 #define HAMMER_HEAD_SIGNATURE 0xC84EU 450 #define HAMMER_TAIL_SIGNATURE 0xC74FU 451 452 #define HAMMER_HEAD_SEQ_BEG 0x80000000U 453 #define HAMMER_HEAD_SEQ_END 0x40000000U 454 #define HAMMER_HEAD_SEQ_MASK 0x3FFFFFFFU 455 456 /* 457 * Misc FIFO structures. 458 * 459 * UNDO - Raw meta-data media updates. 460 */ 461 struct hammer_fifo_undo { 462 struct hammer_fifo_head head; 463 hammer_off_t undo_offset; /* zone-1 offset */ 464 int32_t undo_data_bytes; 465 int32_t undo_reserved01; 466 /* followed by data */ 467 }; 468 469 /* 470 * REDO (HAMMER version 4+) - Logical file writes/truncates. 471 * 472 * REDOs contain information which will be duplicated in a later meta-data 473 * update, allowing fast write()+fsync() operations. REDOs can be ignored 474 * without harming filesystem integrity but must be processed if fsync() 475 * semantics are desired. 476 * 477 * Unlike UNDOs which are processed backwards within the recovery span, 478 * REDOs must be processed forwards starting further back (starting outside 479 * the recovery span). 480 * 481 * WRITE - Write logical file (with payload). Executed both 482 * out-of-span and in-span. Out-of-span WRITEs may be 483 * filtered out by TERMs. 484 * 485 * TRUNC - Truncate logical file (no payload). Executed both 486 * out-of-span and in-span. Out-of-span WRITEs may be 487 * filtered out by TERMs. 488 * 489 * TERM_* - Indicates meta-data was committed (if out-of-span) or 490 * will be rolled-back (in-span). Any out-of-span TERMs 491 * matching earlier WRITEs remove those WRITEs from 492 * consideration as they might conflict with a later data 493 * commit (which is not being rolled-back). 494 * 495 * SYNC - The earliest in-span SYNC (the last one when scanning 496 * backwards) tells the recovery code how far out-of-span 497 * it must go to run REDOs. 498 * 499 * NOTE: WRITEs do not always have matching TERMs even under 500 * perfect conditions because truncations might remove the 501 * buffers from consideration. I/O problems can also remove 502 * buffers from consideration. 503 * 504 * TRUNCSs do not always have matching TERMs because several 505 * truncations may be aggregated together into a single TERM. 506 */ 507 struct hammer_fifo_redo { 508 struct hammer_fifo_head head; 509 int64_t redo_objid; /* file being written */ 510 hammer_off_t redo_offset; /* logical offset in file */ 511 int32_t redo_data_bytes; 512 u_int32_t redo_flags; 513 u_int32_t redo_localization; 514 u_int32_t redo_reserved; 515 u_int64_t redo_mtime; /* set mtime */ 516 }; 517 518 #define HAMMER_REDO_WRITE 0x00000001 519 #define HAMMER_REDO_TRUNC 0x00000002 520 #define HAMMER_REDO_TERM_WRITE 0x00000004 521 #define HAMMER_REDO_TERM_TRUNC 0x00000008 522 #define HAMMER_REDO_SYNC 0x00000010 523 524 union hammer_fifo_any { 525 struct hammer_fifo_head head; 526 struct hammer_fifo_undo undo; 527 struct hammer_fifo_redo redo; 528 }; 529 530 typedef struct hammer_fifo_redo *hammer_fifo_redo_t; 531 typedef struct hammer_fifo_undo *hammer_fifo_undo_t; 532 typedef union hammer_fifo_any *hammer_fifo_any_t; 533 534 /* 535 * Volume header types 536 */ 537 #define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */ 538 #define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */ 539 540 /* 541 * The B-Tree structures need hammer_fsbuf_head. 542 */ 543 #include "hammer_btree.h" 544 545 /* 546 * HAMMER Volume header 547 * 548 * A HAMMER filesystem is built from any number of block devices, Each block 549 * device contains a volume header followed by however many buffers fit 550 * into the volume. 551 * 552 * One of the volumes making up a HAMMER filesystem is the master, the 553 * rest are slaves. It does not have to be volume #0. 554 * 555 * The volume header takes up an entire 16K filesystem buffer and may 556 * represent up to 64KTB (65536 TB) of space. 557 * 558 * Special field notes: 559 * 560 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes) 561 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes) 562 * vol_buf_beg - offset of the first buffer. 563 * 564 * The memory log area allows a kernel to cache new records and data 565 * in memory without allocating space in the actual filesystem to hold 566 * the records and data. In the event that a filesystem becomes full, 567 * any records remaining in memory can be flushed to the memory log 568 * area. This allows the kernel to immediately return success. 569 */ 570 571 #define HAMMER_BOOT_MINBYTES (32*1024) 572 #define HAMMER_BOOT_NOMBYTES (64LL*1024*1024) 573 #define HAMMER_BOOT_MAXBYTES (256LL*1024*1024) 574 575 #define HAMMER_MEM_MINBYTES (256*1024) 576 #define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024) 577 #define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024) 578 579 struct hammer_volume_ondisk { 580 u_int64_t vol_signature;/* Signature */ 581 582 int64_t vol_bot_beg; /* byte offset of boot area or 0 */ 583 int64_t vol_mem_beg; /* byte offset of memory log or 0 */ 584 int64_t vol_buf_beg; /* byte offset of first buffer in volume */ 585 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */ 586 int64_t vol_locked; /* reserved clusters are >= this offset */ 587 588 uuid_t vol_fsid; /* identify filesystem */ 589 uuid_t vol_fstype; /* identify filesystem type */ 590 char vol_name[64]; /* Name of volume */ 591 592 int32_t vol_no; /* volume number within filesystem */ 593 int32_t vol_count; /* number of volumes making up FS */ 594 595 u_int32_t vol_version; /* version control information */ 596 hammer_crc_t vol_crc; /* header crc */ 597 u_int32_t vol_flags; /* volume flags */ 598 u_int32_t vol_rootvol; /* which volume is the root volume? */ 599 600 int32_t vol_reserved04; 601 int32_t vol_reserved05; 602 u_int32_t vol_reserved06; 603 u_int32_t vol_reserved07; 604 605 int32_t vol_blocksize; /* for statfs only */ 606 int32_t vol_reserved08; 607 int64_t vol_nblocks; /* total allocatable hammer bufs */ 608 609 /* 610 * These fields are initialized and space is reserved in every 611 * volume making up a HAMMER filesytem, but only the master volume 612 * contains valid data. 613 */ 614 int64_t vol0_stat_bigblocks; /* total bigblocks when fs is empty */ 615 int64_t vol0_stat_freebigblocks;/* number of free bigblocks */ 616 int64_t vol0_stat_bytes; /* for statfs only */ 617 int64_t vol0_stat_inodes; /* for statfs only */ 618 int64_t vol0_stat_records; /* total records in filesystem */ 619 hammer_off_t vol0_btree_root; /* B-Tree root */ 620 hammer_tid_t vol0_next_tid; /* highest partially synchronized TID */ 621 hammer_off_t vol0_unused03; 622 623 /* 624 * Blockmaps for zones. Not all zones use a blockmap. Note that 625 * the entire root blockmap is cached in the hammer_mount structure. 626 */ 627 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES]; 628 629 /* 630 * Array of zone-2 addresses for undo FIFO. 631 */ 632 hammer_off_t vol0_undo_array[HAMMER_UNDO_LAYER2]; 633 634 }; 635 636 typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t; 637 638 #define HAMMER_VOLF_VALID 0x0001 /* valid entry */ 639 #define HAMMER_VOLF_OPEN 0x0002 /* volume is open */ 640 #define HAMMER_VOLF_NEEDFLUSH 0x0004 /* volume needs flush */ 641 642 #define HAMMER_VOL_CRCSIZE1 \ 643 offsetof(struct hammer_volume_ondisk, vol_crc) 644 #define HAMMER_VOL_CRCSIZE2 \ 645 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \ 646 sizeof(hammer_crc_t)) 647 648 #define HAMMER_VOL_VERSION_MIN 1 /* minimum supported version */ 649 #define HAMMER_VOL_VERSION_DEFAULT 6 /* newfs default version */ 650 #define HAMMER_VOL_VERSION_WIP 7 /* version >= this is WIP */ 651 #define HAMMER_VOL_VERSION_MAX 6 /* maximum supported version */ 652 653 #define HAMMER_VOL_VERSION_ONE 1 654 #define HAMMER_VOL_VERSION_TWO 2 /* new dirent layout (2.3+) */ 655 #define HAMMER_VOL_VERSION_THREE 3 /* new snapshot layout (2.5+) */ 656 #define HAMMER_VOL_VERSION_FOUR 4 /* new undo/flush (2.5+) */ 657 #define HAMMER_VOL_VERSION_FIVE 5 /* dedup (2.9+) */ 658 #define HAMMER_VOL_VERSION_SIX 6 /* DIRHASH_ALG1 */ 659 660 /* 661 * Record types are fairly straightforward. The B-Tree includes the record 662 * type in its index sort. 663 */ 664 #define HAMMER_RECTYPE_UNKNOWN 0 665 #define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */ 666 #define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */ 667 #define HAMMER_RECTYPE_UNUSED02 2 668 #define HAMMER_RECTYPE_UNUSED03 3 669 #define HAMMER_RECTYPE_DATA 0x0010 670 #define HAMMER_RECTYPE_DIRENTRY 0x0011 671 #define HAMMER_RECTYPE_DB 0x0012 672 #define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */ 673 #define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */ 674 #define HAMMER_RECTYPE_PFS 0x0015 /* PFS management */ 675 #define HAMMER_RECTYPE_SNAPSHOT 0x0016 /* Snapshot management */ 676 #define HAMMER_RECTYPE_CONFIG 0x0017 /* hammer cleanup config */ 677 #define HAMMER_RECTYPE_MOVED 0x8000 /* special recovery flag */ 678 #define HAMMER_RECTYPE_MAX 0xFFFF 679 680 #define HAMMER_RECTYPE_ENTRY_START (HAMMER_RECTYPE_INODE + 1) 681 #define HAMMER_RECTYPE_CLEAN_START HAMMER_RECTYPE_EXT 682 683 #define HAMMER_FIXKEY_SYMLINK 1 684 685 #define HAMMER_OBJTYPE_UNKNOWN 0 /* never exists on-disk as unknown */ 686 #define HAMMER_OBJTYPE_DIRECTORY 1 687 #define HAMMER_OBJTYPE_REGFILE 2 688 #define HAMMER_OBJTYPE_DBFILE 3 689 #define HAMMER_OBJTYPE_FIFO 4 690 #define HAMMER_OBJTYPE_CDEV 5 691 #define HAMMER_OBJTYPE_BDEV 6 692 #define HAMMER_OBJTYPE_SOFTLINK 7 693 #define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */ 694 #define HAMMER_OBJTYPE_SOCKET 9 695 696 /* 697 * HAMMER inode attribute data 698 * 699 * The data reference for a HAMMER inode points to this structure. Any 700 * modifications to the contents of this structure will result in a 701 * replacement operation. 702 * 703 * parent_obj_id is only valid for directories (which cannot be hard-linked), 704 * and specifies the parent directory obj_id. This field will also be set 705 * for non-directory inodes as a recovery aid, but can wind up holding 706 * stale information. However, since object id's are not reused, the worse 707 * that happens is that the recovery code is unable to use it. 708 * 709 * NOTE: Future note on directory hardlinks. We can implement a record type 710 * which allows us to point to multiple parent directories. 711 */ 712 struct hammer_inode_data { 713 u_int16_t version; /* inode data version */ 714 u_int16_t mode; /* basic unix permissions */ 715 u_int32_t uflags; /* chflags */ 716 u_int32_t rmajor; /* used by device nodes */ 717 u_int32_t rminor; /* used by device nodes */ 718 u_int64_t ctime; 719 int64_t parent_obj_id; /* parent directory obj_id */ 720 uuid_t uid; 721 uuid_t gid; 722 723 u_int8_t obj_type; 724 u_int8_t cap_flags; /* capability support flags (extension) */ 725 u_int16_t reserved02; 726 u_int32_t reserved03; /* RESERVED FOR POSSIBLE FUTURE BIRTHTIME */ 727 u_int64_t nlinks; /* hard links */ 728 u_int64_t size; /* filesystem object size */ 729 union { 730 struct { 731 char reserved06[16]; 732 u_int32_t parent_obj_localization; 733 u_int32_t integrity_crc; 734 } obj; 735 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */ 736 } ext; 737 u_int64_t mtime; /* mtime must be second-to-last */ 738 u_int64_t atime; /* atime must be last */ 739 }; 740 741 /* 742 * Neither mtime nor atime upates are CRCd by the B-Tree element. 743 * mtime updates have UNDO, atime updates do not. 744 */ 745 #define HAMMER_ITIMES_BASE(ino_data) (&(ino_data)->mtime) 746 #define HAMMER_ITIMES_BYTES (sizeof(u_int64_t) * 2) 747 748 #define HAMMER_INODE_CRCSIZE \ 749 offsetof(struct hammer_inode_data, mtime) 750 751 #define HAMMER_INODE_DATA_VERSION 1 752 #define HAMMER_OBJID_ROOT 1 753 #define HAMMER_INODE_BASESYMLEN 24 /* see ext.symlink */ 754 755 /* 756 * Capability & implementation flags. 757 * 758 * DIR_LOCAL_INO - Use inode B-Tree localization for directory entries. 759 */ 760 #define HAMMER_INODE_CAP_DIRHASH_MASK 0x03 /* directory: hash algorithm */ 761 #define HAMMER_INODE_CAP_DIRHASH_ALG0 0x00 762 #define HAMMER_INODE_CAP_DIRHASH_ALG1 0x01 763 #define HAMMER_INODE_CAP_DIRHASH_ALG2 0x02 764 #define HAMMER_INODE_CAP_DIRHASH_ALG3 0x03 765 #define HAMMER_INODE_CAP_DIR_LOCAL_INO 0x04 /* use inode localization */ 766 767 /* 768 * A HAMMER directory entry associates a HAMMER filesystem object with a 769 * namespace. It is possible to hook into a pseudo-filesystem (with its 770 * own inode numbering space) in the filesystem by setting the high 771 * 16 bits of the localization field. The low 16 bits must be 0 and 772 * are reserved for future use. 773 * 774 * Directory entries are indexed with a 128 bit namekey rather then an 775 * offset. A portion of the namekey is an iterator/randomizer to deal 776 * with collisions. 777 * 778 * NOTE: base.base.obj_type from the related B-Tree leaf entry holds 779 * the filesystem object type of obj_id, e.g. a den_type equivalent. 780 * It is not stored in hammer_entry_data. 781 * 782 * NOTE: den_name / the filename data reference is NOT terminated with \0. 783 */ 784 struct hammer_entry_data { 785 int64_t obj_id; /* object being referenced */ 786 u_int32_t localization; /* identify pseudo-filesystem */ 787 u_int32_t reserved02; 788 char name[16]; /* name (extended) */ 789 }; 790 791 #define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_entry_data, name[0]) 792 #define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_entry_data, name[nlen]) 793 794 /* 795 * Symlink data which does not fit in the inode is stored in a separte 796 * FIX type record. 797 */ 798 struct hammer_symlink_data { 799 char name[16]; 800 }; 801 802 #define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0]) 803 804 /* 805 * The root inode for the primary filesystem and root inode for any 806 * pseudo-fs may be tagged with an optional data structure using 807 * HAMMER_RECTYPE_PFS and localization id. This structure allows 808 * the node to be used as a mirroring master or slave. 809 * 810 * When operating as a slave CD's into the node automatically become read-only 811 * and as-of sync_end_tid. 812 * 813 * When operating as a master the read PFSD info sets sync_end_tid to 814 * the most recently flushed TID. 815 * 816 * sync_low_tid is not yet used but will represent the highest pruning 817 * end-point, after which full history is available. 818 * 819 * We need to pack this structure making it equally sized on both 32-bit and 820 * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is 821 * send over the wire in hammer mirror operations. Only on 64-bit machines 822 * the size of this struct differ when packed or not. This leads us to the 823 * situation where old 64-bit systems (using the non-packed structure), 824 * which were never able to mirror to/from 32-bit systems, are now no longer 825 * able to mirror to/from newer 64-bit systems (using the packed structure). 826 */ 827 struct hammer_pseudofs_data { 828 hammer_tid_t sync_low_tid; /* full history beyond this point */ 829 hammer_tid_t sync_beg_tid; /* earliest tid w/ full history avail */ 830 hammer_tid_t sync_end_tid; /* current synchronizatoin point */ 831 u_int64_t sync_beg_ts; /* real-time of last completed sync */ 832 u_int64_t sync_end_ts; /* initiation of current sync cycle */ 833 uuid_t shared_uuid; /* shared uuid (match required) */ 834 uuid_t unique_uuid; /* unique uuid of this master/slave */ 835 int32_t reserved01; /* reserved for future master_id */ 836 int32_t mirror_flags; /* misc flags */ 837 char label[64]; /* filesystem space label */ 838 char snapshots[64]; /* softlink dir for pruning */ 839 int16_t prune_time; /* how long to spend pruning */ 840 int16_t prune_freq; /* how often we prune */ 841 int16_t reblock_time; /* how long to spend reblocking */ 842 int16_t reblock_freq; /* how often we reblock */ 843 int32_t snapshot_freq; /* how often we create a snapshot */ 844 int32_t prune_min; /* do not prune recent history */ 845 int32_t prune_max; /* do not retain history beyond here */ 846 int32_t reserved[16]; 847 } __packed; 848 849 typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t; 850 851 #define HAMMER_PFSD_SLAVE 0x00000001 852 #define HAMMER_PFSD_DELETED 0x80000000 853 854 /* 855 * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }. 856 * 857 * Snapshot records replace the old <fs>/snapshots/<softlink> methodology. Snapshot 858 * records are mirrored but may be independantly managed once they are laid down on 859 * a slave. 860 * 861 * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the 862 * results. 863 * 864 * NOTE: Reserved fields must be zero (as usual) 865 */ 866 struct hammer_snapshot_data { 867 hammer_tid_t tid; /* the snapshot TID itself (== key) */ 868 u_int64_t ts; /* real-time when snapshot was made */ 869 u_int64_t reserved01; 870 u_int64_t reserved02; 871 char label[64]; /* user-supplied description */ 872 u_int64_t reserved03[4]; 873 }; 874 875 /* 876 * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }. 877 * 878 * Used to store the hammer cleanup config. This data is not mirrored. 879 */ 880 struct hammer_config_data { 881 char text[1024]; 882 }; 883 884 /* 885 * Rollup various structures embedded as record data 886 */ 887 union hammer_data_ondisk { 888 struct hammer_entry_data entry; 889 struct hammer_inode_data inode; 890 struct hammer_symlink_data symlink; 891 struct hammer_pseudofs_data pfsd; 892 struct hammer_snapshot_data snap; 893 struct hammer_config_data config; 894 }; 895 896 typedef union hammer_data_ondisk *hammer_data_ondisk_t; 897 898 #endif 899