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 * Translate a zone address to zone-2 address. 203 */ 204 #define hammer_xlate_to_zone2(offset) \ 205 (((offset) & ~HAMMER_OFF_ZONE_MASK) | HAMMER_ZONE_RAW_BUFFER) 206 207 /* 208 * Big-Block backing store 209 * 210 * A blockmap is a two-level map which translates a blockmap-backed zone 211 * offset into a raw zone 2 offset. The layer 1 handles 18 bits and the 212 * layer 2 handles 19 bits. The 8M big-block size is 23 bits so two 213 * layers gives us 18+19+23 = 60 bits of address space. 214 * 215 * When using hinting for a blockmap lookup, the hint is lost when the 216 * scan leaves the HINTBLOCK, which is typically several BIGBLOCK's. 217 * HINTBLOCK is a heuristic. 218 */ 219 #define HAMMER_HINTBLOCK_SIZE (HAMMER_BIGBLOCK_SIZE * 4) 220 #define HAMMER_HINTBLOCK_MASK64 ((u_int64_t)HAMMER_HINTBLOCK_SIZE - 1) 221 #define HAMMER_BIGBLOCK_SIZE (8192 * 1024) 222 #define HAMMER_BIGBLOCK_OVERFILL (6144 * 1024) 223 #define HAMMER_BIGBLOCK_SIZE64 ((u_int64_t)HAMMER_BIGBLOCK_SIZE) 224 #define HAMMER_BIGBLOCK_MASK (HAMMER_BIGBLOCK_SIZE - 1) 225 #define HAMMER_BIGBLOCK_MASK64 ((u_int64_t)HAMMER_BIGBLOCK_SIZE - 1) 226 #define HAMMER_BIGBLOCK_BITS 23 227 #if (1 << HAMMER_BIGBLOCK_BITS) != HAMMER_BIGBLOCK_SIZE 228 #error "HAMMER_BIGBLOCK_BITS BROKEN" 229 #endif 230 231 #define HAMMER_BUFFERS_PER_BIGBLOCK \ 232 (HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE) 233 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK \ 234 (HAMMER_BUFFERS_PER_BIGBLOCK - 1) 235 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK64 \ 236 ((hammer_off_t)HAMMER_BUFFERS_PER_BIGBLOCK_MASK) 237 238 /* 239 * Maximum number of mirrors operating in master mode (multi-master 240 * clustering and mirroring). 241 */ 242 #define HAMMER_MAX_MASTERS 16 243 244 /* 245 * The blockmap is somewhat of a degenerate structure. HAMMER only actually 246 * uses it in its original incarnation to implement the free-map. 247 * 248 * zone:1 raw volume (no blockmap) 249 * zone:2 raw buffer (no blockmap) 250 * zone:3 undo-map (direct layer2 array in volume header) 251 * zone:4 free-map (the only real blockmap) 252 * zone:8-15 zone id used to classify big-block only, address is actually 253 * a zone-2 address. 254 */ 255 struct hammer_blockmap { 256 hammer_off_t phys_offset; /* zone-2 physical offset */ 257 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */ 258 hammer_off_t next_offset; /* zone-X logical offset */ 259 hammer_off_t alloc_offset; /* zone-X logical offset */ 260 u_int32_t reserved01; 261 hammer_crc_t entry_crc; 262 }; 263 264 typedef struct hammer_blockmap *hammer_blockmap_t; 265 266 #define HAMMER_BLOCKMAP_CRCSIZE \ 267 offsetof(struct hammer_blockmap, entry_crc) 268 269 /* 270 * The blockmap is a 2-layer entity made up of big-blocks. The first layer 271 * contains 262144 32-byte entries (18 bits), the second layer contains 272 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps. 273 * 18+19+23 = 60 bits. The top four bits are the zone id. 274 * 275 * Currently only the freemap utilizes both layers in all their glory. 276 * All primary data/meta-data zones actually encode a zone-2 address 277 * requiring no real blockmap translation. 278 * 279 * The freemap uses the upper 8 bits of layer-1 to identify the volume, 280 * thus any space allocated via the freemap can be directly translated 281 * to a zone:2 (or zone:8-15) address. 282 * 283 * zone-X blockmap offset: [zone:4][layer1:18][layer2:19][bigblock:23] 284 */ 285 struct hammer_blockmap_layer1 { 286 hammer_off_t blocks_free; /* big-blocks free */ 287 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */ 288 hammer_off_t reserved01; 289 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */ 290 /* (not yet used) */ 291 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/ 292 }; 293 294 typedef struct hammer_blockmap_layer1 *hammer_blockmap_layer1_t; 295 296 #define HAMMER_LAYER1_CRCSIZE \ 297 offsetof(struct hammer_blockmap_layer1, layer1_crc) 298 299 /* 300 * layer2 entry for 8MB bigblock. 301 * 302 * NOTE: bytes_free is signed and can legally go negative if/when data 303 * de-dup occurs. This field will never go higher than 304 * HAMMER_BIGBLOCK_SIZE. If exactly HAMMER_BIGBLOCK_SIZE 305 * the big-block is completely free. 306 */ 307 struct hammer_blockmap_layer2 { 308 u_int8_t zone; /* typed allocation zone */ 309 u_int8_t unused01; 310 u_int16_t unused02; 311 u_int32_t append_off; /* allocatable space index */ 312 int32_t bytes_free; /* bytes free within this bigblock */ 313 hammer_crc_t entry_crc; 314 }; 315 316 typedef struct hammer_blockmap_layer2 *hammer_blockmap_layer2_t; 317 318 #define HAMMER_LAYER2_CRCSIZE \ 319 offsetof(struct hammer_blockmap_layer2, entry_crc) 320 321 #define HAMMER_BLOCKMAP_FREE 0ULL 322 #define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL) 323 324 #define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \ 325 (HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1)) 326 #define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \ 327 (HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2)) 328 329 #define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \ 330 (HAMMER_BLOCKMAP_RADIX1 / (HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE)) 331 #define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \ 332 (HAMMER_BLOCKMAP_RADIX2 / (HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE)) 333 334 #define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 - 1EB */ \ 335 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2) 336 #define HAMMER_BLOCKMAP_LAYER2 /* 19+23 - 4TB */ \ 337 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_BIGBLOCK_SIZE64) 338 339 #define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1) 340 #define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1) 341 342 /* 343 * byte offset within layer1 or layer2 big-block for the entry representing 344 * a zone-2 physical offset. 345 */ 346 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \ 347 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \ 348 HAMMER_BLOCKMAP_LAYER2 * sizeof(struct hammer_blockmap_layer1)) 349 350 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \ 351 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \ 352 HAMMER_BIGBLOCK_SIZE64 * sizeof(struct hammer_blockmap_layer2)) 353 354 /* 355 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume 356 * header with an array of layer2 structures. A maximum of (128x8MB) = 1GB 357 * may be reserved. The size of the undo fifo is usually set a newfs time 358 * but can be adjusted if the filesystem is taken offline. 359 */ 360 #define HAMMER_UNDO_LAYER2 128 /* max layer2 undo mapping entries */ 361 362 /* 363 * All on-disk HAMMER structures which make up elements of the UNDO FIFO 364 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure 365 * contains all the information required to validate the fifo element 366 * and to scan the fifo in either direction. The head is typically embedded 367 * in higher level hammer on-disk structures while the tail is typically 368 * out-of-band. hdr_size is the size of the whole mess, including the tail. 369 * 370 * All undo structures are guaranteed to not cross a 16K filesystem 371 * buffer boundary. Most undo structures are fairly small. Data spaces 372 * are not immediately reused by HAMMER so file data is not usually recorded 373 * as part of an UNDO. 374 * 375 * PAD elements are allowed to take up only 8 bytes of space as a special 376 * case, containing only hdr_signature, hdr_type, and hdr_size fields, 377 * and with the tail overloaded onto the head structure for 8 bytes total. 378 * 379 * Every undo record has a sequence number. This number is unrelated to 380 * transaction ids and instead collects the undo transactions associated 381 * with a single atomic operation. A larger transactional operation, such 382 * as a remove(), may consist of several smaller atomic operations 383 * representing raw meta-data operations. 384 * 385 * HAMMER VERSION 4 CHANGES 386 * 387 * In HAMMER version 4 the undo structure alignment is reduced from 16384 388 * to 512 bytes in order to ensure that each 512 byte sector begins with 389 * a header. The reserved01 field in the header is now a 32 bit sequence 390 * number. This allows the recovery code to detect missing sectors 391 * without relying on the 32-bit crc and to definitively identify the current 392 * undo sequence space without having to rely on information from the volume 393 * header. In addition, new REDO entries in the undo space are used to 394 * record write, write/extend, and transaction id updates. 395 * 396 * The grand result is: 397 * 398 * (1) The volume header no longer needs to be synchronized for most 399 * flush and fsync operations. 400 * 401 * (2) Most fsync operations need only lay down REDO records 402 * 403 * (3) Data overwrite for nohistory operations covered by REDO records 404 * can be supported (instead of rolling a new block allocation), 405 * by rolling UNDO for the prior contents of the data. 406 * 407 * HAMMER VERSION 5 CHANGES 408 * 409 * Hammer version 5 contains a minor adjustment making layer2's bytes_free 410 * field signed, allowing dedup to push it into the negative domain. 411 */ 412 #define HAMMER_HEAD_ONDISK_SIZE 32 413 #define HAMMER_HEAD_ALIGN 8 414 #define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1) 415 #define HAMMER_TAIL_ONDISK_SIZE 8 416 #define HAMMER_HEAD_DOALIGN(bytes) \ 417 (((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK) 418 419 #define HAMMER_UNDO_ALIGN 512 420 #define HAMMER_UNDO_ALIGN64 ((u_int64_t)512) 421 #define HAMMER_UNDO_MASK (HAMMER_UNDO_ALIGN - 1) 422 #define HAMMER_UNDO_MASK64 (HAMMER_UNDO_ALIGN64 - 1) 423 424 struct hammer_fifo_head { 425 u_int16_t hdr_signature; 426 u_int16_t hdr_type; 427 u_int32_t hdr_size; /* Aligned size of the whole mess */ 428 u_int32_t hdr_seq; /* Sequence number */ 429 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */ 430 }; 431 432 #define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc) 433 434 struct hammer_fifo_tail { 435 u_int16_t tail_signature; 436 u_int16_t tail_type; 437 u_int32_t tail_size; /* aligned size of the whole mess */ 438 }; 439 440 typedef struct hammer_fifo_head *hammer_fifo_head_t; 441 typedef struct hammer_fifo_tail *hammer_fifo_tail_t; 442 443 /* 444 * Fifo header types. 445 */ 446 #define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE) 447 #define HAMMER_HEAD_TYPE_DUMMY 0x0041U /* dummy entry w/seqno */ 448 #define HAMMER_HEAD_TYPE_42 0x0042U 449 #define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */ 450 #define HAMMER_HEAD_TYPE_REDO 0x0044U /* data REDO / fast fsync */ 451 #define HAMMER_HEAD_TYPE_45 0x0045U 452 453 #define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */ 454 455 #define HAMMER_HEAD_SIGNATURE 0xC84EU 456 #define HAMMER_TAIL_SIGNATURE 0xC74FU 457 458 /* 459 * Misc FIFO structures. 460 * 461 * UNDO - Raw meta-data media updates. 462 */ 463 struct hammer_fifo_undo { 464 struct hammer_fifo_head head; 465 hammer_off_t undo_offset; /* zone-1,2 offset */ 466 int32_t undo_data_bytes; 467 int32_t undo_reserved01; 468 /* followed by data */ 469 }; 470 471 /* 472 * REDO (HAMMER version 4+) - Logical file writes/truncates. 473 * 474 * REDOs contain information which will be duplicated in a later meta-data 475 * update, allowing fast write()+fsync() operations. REDOs can be ignored 476 * without harming filesystem integrity but must be processed if fsync() 477 * semantics are desired. 478 * 479 * Unlike UNDOs which are processed backwards within the recovery span, 480 * REDOs must be processed forwards starting further back (starting outside 481 * the recovery span). 482 * 483 * WRITE - Write logical file (with payload). Executed both 484 * out-of-span and in-span. Out-of-span WRITEs may be 485 * filtered out by TERMs. 486 * 487 * TRUNC - Truncate logical file (no payload). Executed both 488 * out-of-span and in-span. Out-of-span WRITEs may be 489 * filtered out by TERMs. 490 * 491 * TERM_* - Indicates meta-data was committed (if out-of-span) or 492 * will be rolled-back (in-span). Any out-of-span TERMs 493 * matching earlier WRITEs remove those WRITEs from 494 * consideration as they might conflict with a later data 495 * commit (which is not being rolled-back). 496 * 497 * SYNC - The earliest in-span SYNC (the last one when scanning 498 * backwards) tells the recovery code how far out-of-span 499 * it must go to run REDOs. 500 * 501 * NOTE: WRITEs do not always have matching TERMs even under 502 * perfect conditions because truncations might remove the 503 * buffers from consideration. I/O problems can also remove 504 * buffers from consideration. 505 * 506 * TRUNCSs do not always have matching TERMs because several 507 * truncations may be aggregated together into a single TERM. 508 */ 509 struct hammer_fifo_redo { 510 struct hammer_fifo_head head; 511 int64_t redo_objid; /* file being written */ 512 hammer_off_t redo_offset; /* logical offset in file */ 513 int32_t redo_data_bytes; 514 u_int32_t redo_flags; 515 u_int32_t redo_localization; 516 u_int32_t redo_reserved; 517 u_int64_t redo_mtime; /* set mtime */ 518 }; 519 520 #define HAMMER_REDO_WRITE 0x00000001 521 #define HAMMER_REDO_TRUNC 0x00000002 522 #define HAMMER_REDO_TERM_WRITE 0x00000004 523 #define HAMMER_REDO_TERM_TRUNC 0x00000008 524 #define HAMMER_REDO_SYNC 0x00000010 525 526 union hammer_fifo_any { 527 struct hammer_fifo_head head; 528 struct hammer_fifo_undo undo; 529 struct hammer_fifo_redo redo; 530 }; 531 532 typedef struct hammer_fifo_redo *hammer_fifo_redo_t; 533 typedef struct hammer_fifo_undo *hammer_fifo_undo_t; 534 typedef union hammer_fifo_any *hammer_fifo_any_t; 535 536 /* 537 * Volume header types 538 */ 539 #define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */ 540 #define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */ 541 542 /* 543 * The B-Tree structures need hammer_fsbuf_head. 544 */ 545 #include "hammer_btree.h" 546 547 /* 548 * HAMMER Volume header 549 * 550 * A HAMMER filesystem is built from any number of block devices, Each block 551 * device contains a volume header followed by however many buffers fit 552 * into the volume. 553 * 554 * One of the volumes making up a HAMMER filesystem is the master, the 555 * rest are slaves. It does not have to be volume #0. 556 * 557 * The volume header takes up an entire 16K filesystem buffer and may 558 * represent up to 64KTB (65536 TB) of space. 559 * 560 * Special field notes: 561 * 562 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes) 563 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes) 564 * vol_buf_beg - offset of the first buffer. 565 * 566 * The memory log area allows a kernel to cache new records and data 567 * in memory without allocating space in the actual filesystem to hold 568 * the records and data. In the event that a filesystem becomes full, 569 * any records remaining in memory can be flushed to the memory log 570 * area. This allows the kernel to immediately return success. 571 */ 572 573 #define HAMMER_BOOT_MINBYTES (32*1024) 574 #define HAMMER_BOOT_NOMBYTES (64LL*1024*1024) 575 #define HAMMER_BOOT_MAXBYTES (256LL*1024*1024) 576 577 #define HAMMER_MEM_MINBYTES (256*1024) 578 #define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024) 579 #define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024) 580 581 struct hammer_volume_ondisk { 582 u_int64_t vol_signature;/* Signature */ 583 584 int64_t vol_bot_beg; /* byte offset of boot area or 0 */ 585 int64_t vol_mem_beg; /* byte offset of memory log or 0 */ 586 int64_t vol_buf_beg; /* byte offset of first buffer in volume */ 587 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */ 588 int64_t vol_locked; /* reserved clusters are >= this offset */ 589 590 uuid_t vol_fsid; /* identify filesystem */ 591 uuid_t vol_fstype; /* identify filesystem type */ 592 char vol_name[64]; /* Name of volume */ 593 594 int32_t vol_no; /* volume number within filesystem */ 595 int32_t vol_count; /* number of volumes making up FS */ 596 597 u_int32_t vol_version; /* version control information */ 598 hammer_crc_t vol_crc; /* header crc */ 599 u_int32_t vol_flags; /* volume flags */ 600 u_int32_t vol_rootvol; /* which volume is the root volume? */ 601 602 int32_t vol_reserved04; 603 int32_t vol_reserved05; 604 u_int32_t vol_reserved06; 605 u_int32_t vol_reserved07; 606 607 int32_t vol_blocksize; /* for statfs only */ 608 int32_t vol_reserved08; 609 int64_t vol_nblocks; /* total allocatable hammer bufs */ 610 611 /* 612 * These fields are initialized and space is reserved in every 613 * volume making up a HAMMER filesytem, but only the master volume 614 * contains valid data. 615 */ 616 int64_t vol0_stat_bigblocks; /* total bigblocks when fs is empty */ 617 int64_t vol0_stat_freebigblocks;/* number of free bigblocks */ 618 int64_t vol0_stat_bytes; /* for statfs only */ 619 int64_t vol0_stat_inodes; /* for statfs only */ 620 int64_t vol0_stat_records; /* total records in filesystem */ 621 hammer_off_t vol0_btree_root; /* B-Tree root */ 622 hammer_tid_t vol0_next_tid; /* highest partially synchronized TID */ 623 hammer_off_t vol0_unused03; 624 625 /* 626 * Blockmaps for zones. Not all zones use a blockmap. Note that 627 * the entire root blockmap is cached in the hammer_mount structure. 628 */ 629 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES]; 630 631 /* 632 * Array of zone-2 addresses for undo FIFO. 633 */ 634 hammer_off_t vol0_undo_array[HAMMER_UNDO_LAYER2]; 635 636 }; 637 638 typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t; 639 640 #define HAMMER_VOLF_VALID 0x0001 /* valid entry */ 641 #define HAMMER_VOLF_OPEN 0x0002 /* volume is open */ 642 #define HAMMER_VOLF_NEEDFLUSH 0x0004 /* volume needs flush */ 643 644 #define HAMMER_VOL_CRCSIZE1 \ 645 offsetof(struct hammer_volume_ondisk, vol_crc) 646 #define HAMMER_VOL_CRCSIZE2 \ 647 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \ 648 sizeof(hammer_crc_t)) 649 650 #define HAMMER_VOL_VERSION_MIN 1 /* minimum supported version */ 651 #define HAMMER_VOL_VERSION_DEFAULT 6 /* newfs default version */ 652 #define HAMMER_VOL_VERSION_WIP 7 /* version >= this is WIP */ 653 #define HAMMER_VOL_VERSION_MAX 6 /* maximum supported version */ 654 655 #define HAMMER_VOL_VERSION_ONE 1 656 #define HAMMER_VOL_VERSION_TWO 2 /* new dirent layout (2.3+) */ 657 #define HAMMER_VOL_VERSION_THREE 3 /* new snapshot layout (2.5+) */ 658 #define HAMMER_VOL_VERSION_FOUR 4 /* new undo/flush (2.5+) */ 659 #define HAMMER_VOL_VERSION_FIVE 5 /* dedup (2.9+) */ 660 #define HAMMER_VOL_VERSION_SIX 6 /* DIRHASH_ALG1 */ 661 662 /* 663 * Record types are fairly straightforward. The B-Tree includes the record 664 * type in its index sort. 665 */ 666 #define HAMMER_RECTYPE_UNKNOWN 0 667 #define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */ 668 #define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */ 669 #define HAMMER_RECTYPE_UNUSED02 2 670 #define HAMMER_RECTYPE_UNUSED03 3 671 #define HAMMER_RECTYPE_DATA 0x0010 672 #define HAMMER_RECTYPE_DIRENTRY 0x0011 673 #define HAMMER_RECTYPE_DB 0x0012 674 #define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */ 675 #define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */ 676 #define HAMMER_RECTYPE_PFS 0x0015 /* PFS management */ 677 #define HAMMER_RECTYPE_SNAPSHOT 0x0016 /* Snapshot management */ 678 #define HAMMER_RECTYPE_CONFIG 0x0017 /* hammer cleanup config */ 679 #define HAMMER_RECTYPE_MOVED 0x8000 /* special recovery flag */ 680 #define HAMMER_RECTYPE_MAX 0xFFFF 681 682 #define HAMMER_RECTYPE_ENTRY_START (HAMMER_RECTYPE_INODE + 1) 683 #define HAMMER_RECTYPE_CLEAN_START HAMMER_RECTYPE_EXT 684 685 #define HAMMER_FIXKEY_SYMLINK 1 686 687 #define HAMMER_OBJTYPE_UNKNOWN 0 /* never exists on-disk as unknown */ 688 #define HAMMER_OBJTYPE_DIRECTORY 1 689 #define HAMMER_OBJTYPE_REGFILE 2 690 #define HAMMER_OBJTYPE_DBFILE 3 691 #define HAMMER_OBJTYPE_FIFO 4 692 #define HAMMER_OBJTYPE_CDEV 5 693 #define HAMMER_OBJTYPE_BDEV 6 694 #define HAMMER_OBJTYPE_SOFTLINK 7 695 #define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */ 696 #define HAMMER_OBJTYPE_SOCKET 9 697 698 /* 699 * HAMMER inode attribute data 700 * 701 * The data reference for a HAMMER inode points to this structure. Any 702 * modifications to the contents of this structure will result in a 703 * replacement operation. 704 * 705 * parent_obj_id is only valid for directories (which cannot be hard-linked), 706 * and specifies the parent directory obj_id. This field will also be set 707 * for non-directory inodes as a recovery aid, but can wind up holding 708 * stale information. However, since object id's are not reused, the worse 709 * that happens is that the recovery code is unable to use it. 710 * 711 * NOTE: Future note on directory hardlinks. We can implement a record type 712 * which allows us to point to multiple parent directories. 713 */ 714 struct hammer_inode_data { 715 u_int16_t version; /* inode data version */ 716 u_int16_t mode; /* basic unix permissions */ 717 u_int32_t uflags; /* chflags */ 718 u_int32_t rmajor; /* used by device nodes */ 719 u_int32_t rminor; /* used by device nodes */ 720 u_int64_t ctime; 721 int64_t parent_obj_id; /* parent directory obj_id */ 722 uuid_t uid; 723 uuid_t gid; 724 725 u_int8_t obj_type; 726 u_int8_t cap_flags; /* capability support flags (extension) */ 727 u_int16_t reserved02; 728 u_int32_t reserved03; /* RESERVED FOR POSSIBLE FUTURE BIRTHTIME */ 729 u_int64_t nlinks; /* hard links */ 730 u_int64_t size; /* filesystem object size */ 731 union { 732 struct { 733 char reserved06[16]; 734 u_int32_t parent_obj_localization; 735 u_int32_t integrity_crc; 736 } obj; 737 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */ 738 } ext; 739 u_int64_t mtime; /* mtime must be second-to-last */ 740 u_int64_t atime; /* atime must be last */ 741 }; 742 743 /* 744 * Neither mtime nor atime upates are CRCd by the B-Tree element. 745 * mtime updates have UNDO, atime updates do not. 746 */ 747 #define HAMMER_ITIMES_BASE(ino_data) (&(ino_data)->mtime) 748 #define HAMMER_ITIMES_BYTES (sizeof(u_int64_t) * 2) 749 750 #define HAMMER_INODE_CRCSIZE \ 751 offsetof(struct hammer_inode_data, mtime) 752 753 #define HAMMER_INODE_DATA_VERSION 1 754 #define HAMMER_OBJID_ROOT 1 /* root inodes # */ 755 #define HAMMER_INODE_BASESYMLEN 24 /* see ext.symlink */ 756 757 /* 758 * Capability & implementation flags. 759 * 760 * DIR_LOCAL_INO - Use inode B-Tree localization for directory entries. 761 */ 762 #define HAMMER_INODE_CAP_DIRHASH_MASK 0x03 /* directory: hash algorithm */ 763 #define HAMMER_INODE_CAP_DIRHASH_ALG0 0x00 764 #define HAMMER_INODE_CAP_DIRHASH_ALG1 0x01 765 #define HAMMER_INODE_CAP_DIRHASH_ALG2 0x02 766 #define HAMMER_INODE_CAP_DIRHASH_ALG3 0x03 767 #define HAMMER_INODE_CAP_DIR_LOCAL_INO 0x04 /* use inode localization */ 768 769 /* 770 * A HAMMER directory entry associates a HAMMER filesystem object with a 771 * namespace. It is possible to hook into a pseudo-filesystem (with its 772 * own inode numbering space) in the filesystem by setting the high 773 * 16 bits of the localization field. The low 16 bits must be 0 and 774 * are reserved for future use. 775 * 776 * Directory entries are indexed with a 128 bit namekey rather then an 777 * offset. A portion of the namekey is an iterator/randomizer to deal 778 * with collisions. 779 * 780 * NOTE: leaf.base.obj_type from the related B-Tree leaf entry holds 781 * the filesystem object type of obj_id, e.g. a den_type equivalent. 782 * It is not stored in hammer_entry_data. 783 * 784 * NOTE: name field / the filename data reference is NOT terminated with \0. 785 */ 786 struct hammer_entry_data { 787 int64_t obj_id; /* object being referenced */ 788 u_int32_t localization; /* identify pseudo-filesystem */ 789 u_int32_t reserved02; 790 char name[16]; /* name (extended) */ 791 }; 792 793 #define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_entry_data, name[0]) 794 #define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_entry_data, name[nlen]) 795 796 /* 797 * Symlink data which does not fit in the inode is stored in a separte 798 * FIX type record. 799 */ 800 struct hammer_symlink_data { 801 char name[16]; /* name (extended) */ 802 }; 803 804 #define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0]) 805 806 /* 807 * The root inode for the primary filesystem and root inode for any 808 * pseudo-fs may be tagged with an optional data structure using 809 * HAMMER_RECTYPE_PFS and localization id. This structure allows 810 * the node to be used as a mirroring master or slave. 811 * 812 * When operating as a slave CD's into the node automatically become read-only 813 * and as-of sync_end_tid. 814 * 815 * When operating as a master the read PFSD info sets sync_end_tid to 816 * the most recently flushed TID. 817 * 818 * sync_low_tid is not yet used but will represent the highest pruning 819 * end-point, after which full history is available. 820 * 821 * We need to pack this structure making it equally sized on both 32-bit and 822 * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is 823 * send over the wire in hammer mirror operations. Only on 64-bit machines 824 * the size of this struct differ when packed or not. This leads us to the 825 * situation where old 64-bit systems (using the non-packed structure), 826 * which were never able to mirror to/from 32-bit systems, are now no longer 827 * able to mirror to/from newer 64-bit systems (using the packed structure). 828 */ 829 struct hammer_pseudofs_data { 830 hammer_tid_t sync_low_tid; /* full history beyond this point */ 831 hammer_tid_t sync_beg_tid; /* earliest tid w/ full history avail */ 832 hammer_tid_t sync_end_tid; /* current synchronizatoin point */ 833 u_int64_t sync_beg_ts; /* real-time of last completed sync */ 834 u_int64_t sync_end_ts; /* initiation of current sync cycle */ 835 uuid_t shared_uuid; /* shared uuid (match required) */ 836 uuid_t unique_uuid; /* unique uuid of this master/slave */ 837 int32_t reserved01; /* reserved for future master_id */ 838 int32_t mirror_flags; /* misc flags */ 839 char label[64]; /* filesystem space label */ 840 char snapshots[64]; /* softlink dir for pruning */ 841 int16_t prune_time; /* how long to spend pruning */ 842 int16_t prune_freq; /* how often we prune */ 843 int16_t reblock_time; /* how long to spend reblocking */ 844 int16_t reblock_freq; /* how often we reblock */ 845 int32_t snapshot_freq; /* how often we create a snapshot */ 846 int32_t prune_min; /* do not prune recent history */ 847 int32_t prune_max; /* do not retain history beyond here */ 848 int32_t reserved[16]; 849 } __packed; 850 851 typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t; 852 853 #define HAMMER_PFSD_SLAVE 0x00000001 854 #define HAMMER_PFSD_DELETED 0x80000000 855 856 /* 857 * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }. 858 * 859 * Snapshot records replace the old <fs>/snapshots/<softlink> methodology. Snapshot 860 * records are mirrored but may be independantly managed once they are laid down on 861 * a slave. 862 * 863 * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the 864 * results. 865 * 866 * NOTE: Reserved fields must be zero (as usual) 867 */ 868 struct hammer_snapshot_data { 869 hammer_tid_t tid; /* the snapshot TID itself (== key) */ 870 u_int64_t ts; /* real-time when snapshot was made */ 871 u_int64_t reserved01; 872 u_int64_t reserved02; 873 char label[64]; /* user-supplied description */ 874 u_int64_t reserved03[4]; 875 }; 876 877 /* 878 * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }. 879 * 880 * Used to store the hammer cleanup config. This data is not mirrored. 881 */ 882 struct hammer_config_data { 883 char text[1024]; 884 }; 885 886 /* 887 * Rollup various structures embedded as record data 888 */ 889 union hammer_data_ondisk { 890 struct hammer_entry_data entry; 891 struct hammer_inode_data inode; 892 struct hammer_symlink_data symlink; 893 struct hammer_pseudofs_data pfsd; 894 struct hammer_snapshot_data snap; 895 struct hammer_config_data config; 896 }; 897 898 typedef union hammer_data_ondisk *hammer_data_ondisk_t; 899 900 #endif 901