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 91 #define HAMMER_OFF_BAD ((hammer_off_t)-1) 92 93 /* 94 * The current limit of volumes that can make up a HAMMER FS 95 */ 96 #define HAMMER_MAX_VOLUMES 256 97 98 /* 99 * Hammer transaction ids are 64 bit unsigned integers and are usually 100 * synchronized with the time of day in nanoseconds. 101 * 102 * Hammer offsets are used for FIFO indexing and embed a cycle counter 103 * and volume number in addition to the offset. Most offsets are required 104 * to be 16 KB aligned. 105 */ 106 typedef u_int64_t hammer_tid_t; 107 typedef u_int64_t hammer_off_t; 108 typedef u_int32_t hammer_crc_t; 109 110 #define HAMMER_MIN_TID 0ULL /* unsigned */ 111 #define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */ 112 #define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */ 113 #define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */ 114 #define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */ 115 #define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */ 116 #define HAMMER_MIN_RECTYPE 0x0U /* unsigned */ 117 #define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */ 118 #define HAMMER_MIN_OFFSET 0ULL /* unsigned */ 119 #define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */ 120 121 /* 122 * hammer_off_t has several different encodings. Note that not all zones 123 * encode a vol_no. 124 * 125 * zone 0: reserved for sanity 126 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header) 127 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer) 128 * zone 3 (z,o): undo fifo - actually zone-2 address, fixed phys array in vol hdr 129 * zone 4 (z,v,o): freemap - only real blockmap 130 * zone 8 (z,v,o): B-Tree - actually zone-2 address 131 * zone 9 (z,v,o): meta - actually zone-2 address 132 * zone 10 (z,v,o): large-data - actually zone-2 address 133 * zone 11 (z,v,o): small-data - actually zone-2 address 134 * zone 15: reserved for sanity 135 * 136 * layer1/layer2 direct map: 137 * zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo 138 * ----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo 139 */ 140 141 #define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL 142 #define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL 143 #define HAMMER_ZONE_UNDO 0x3000000000000000ULL 144 #define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL 145 #define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL 146 #define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL 147 #define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL 148 #define HAMMER_ZONE_BTREE 0x8000000000000000ULL 149 #define HAMMER_ZONE_META 0x9000000000000000ULL 150 #define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL 151 #define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL 152 #define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL 153 #define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL 154 #define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL 155 #define HAMMER_ZONE_UNAVAIL 0xF000000000000000ULL 156 157 #define HAMMER_ZONE_RAW_VOLUME_INDEX 1 158 #define HAMMER_ZONE_RAW_BUFFER_INDEX 2 159 #define HAMMER_ZONE_UNDO_INDEX 3 160 #define HAMMER_ZONE_FREEMAP_INDEX 4 161 #define HAMMER_ZONE_BTREE_INDEX 8 162 #define HAMMER_ZONE_META_INDEX 9 163 #define HAMMER_ZONE_LARGE_DATA_INDEX 10 164 #define HAMMER_ZONE_SMALL_DATA_INDEX 11 165 #define HAMMER_ZONE_UNAVAIL_INDEX 15 /* unavailable */ 166 167 #define HAMMER_MAX_ZONES 16 168 169 /* 170 * Backend zones that are mapped to zone-2 (except for zone-3) 171 * starts from this index which is 8. 172 */ 173 #define HAMMER_ZONE2_MAPPED_INDEX HAMMER_ZONE_BTREE_INDEX 174 175 #define HAMMER_ZONE_ENCODE(zone, ham_off) \ 176 (((hammer_off_t)(zone) << 60) | (ham_off)) 177 #define HAMMER_ZONE_DECODE(ham_off) \ 178 (int32_t)(((hammer_off_t)(ham_off) >> 60)) 179 180 #define HAMMER_VOL_ENCODE(vol_no) \ 181 ((hammer_off_t)((vol_no) & 255) << 52) 182 #define HAMMER_VOL_DECODE(ham_off) \ 183 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255) 184 185 #define HAMMER_OFF_SHORT_ENCODE(offset) \ 186 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK) 187 #define HAMMER_OFF_LONG_ENCODE(offset) \ 188 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK) 189 190 #define HAMMER_ENCODE(zone, vol_no, offset) \ 191 (((hammer_off_t)(zone) << 60) | \ 192 HAMMER_VOL_ENCODE(vol_no) | \ 193 HAMMER_OFF_SHORT_ENCODE(offset)) 194 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \ 195 HAMMER_ENCODE(HAMMER_ZONE_RAW_VOLUME_INDEX, vol_no, offset) 196 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \ 197 HAMMER_ENCODE(HAMMER_ZONE_RAW_BUFFER_INDEX, vol_no, offset) 198 #define HAMMER_ENCODE_FREEMAP(vol_no, offset) \ 199 HAMMER_ENCODE(HAMMER_ZONE_FREEMAP_INDEX, vol_no, offset) 200 201 /* 202 * Translate a zone address to zone-X address. 203 */ 204 #define hammer_xlate_to_zoneX(zone, offset) \ 205 HAMMER_ZONE_ENCODE((zone), (offset) & ~HAMMER_OFF_ZONE_MASK) 206 #define hammer_xlate_to_zone2(offset) \ 207 hammer_xlate_to_zoneX(HAMMER_ZONE_RAW_BUFFER_INDEX, (offset)) 208 209 /* 210 * Big-Block backing store 211 * 212 * A blockmap is a two-level map which translates a blockmap-backed zone 213 * offset into a raw zone 2 offset. The layer 1 handles 18 bits and the 214 * layer 2 handles 19 bits. The 8M big-block size is 23 bits so two 215 * layers gives us 18+19+23 = 60 bits of address space. 216 * 217 * When using hinting for a blockmap lookup, the hint is lost when the 218 * scan leaves the HINTBLOCK, which is typically several BIGBLOCK's. 219 * HINTBLOCK is a heuristic. 220 */ 221 #define HAMMER_HINTBLOCK_SIZE (HAMMER_BIGBLOCK_SIZE * 4) 222 #define HAMMER_HINTBLOCK_MASK64 ((u_int64_t)HAMMER_HINTBLOCK_SIZE - 1) 223 #define HAMMER_BIGBLOCK_SIZE (8192 * 1024) 224 #define HAMMER_BIGBLOCK_OVERFILL (6144 * 1024) 225 #define HAMMER_BIGBLOCK_SIZE64 ((u_int64_t)HAMMER_BIGBLOCK_SIZE) 226 #define HAMMER_BIGBLOCK_MASK (HAMMER_BIGBLOCK_SIZE - 1) 227 #define HAMMER_BIGBLOCK_MASK64 ((u_int64_t)HAMMER_BIGBLOCK_SIZE - 1) 228 #define HAMMER_BIGBLOCK_BITS 23 229 #if (1 << HAMMER_BIGBLOCK_BITS) != HAMMER_BIGBLOCK_SIZE 230 #error "HAMMER_BIGBLOCK_BITS BROKEN" 231 #endif 232 233 #define HAMMER_BUFFERS_PER_BIGBLOCK \ 234 (HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE) 235 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK \ 236 (HAMMER_BUFFERS_PER_BIGBLOCK - 1) 237 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK64 \ 238 ((hammer_off_t)HAMMER_BUFFERS_PER_BIGBLOCK_MASK) 239 240 /* 241 * Maximum number of mirrors operating in master mode (multi-master 242 * clustering and mirroring). Note that HAMMER1 does not support 243 * multi-master clustering as of 2015. 244 */ 245 #define HAMMER_MAX_MASTERS 16 246 247 /* 248 * The blockmap is somewhat of a degenerate structure. HAMMER only actually 249 * uses it in its original incarnation to implement the freemap. 250 * 251 * zone:1 raw volume (no blockmap) 252 * zone:2 raw buffer (no blockmap) 253 * zone:3 undomap (direct layer2 array in volume header) 254 * zone:4 freemap (the only real blockmap) 255 * zone:8-15 zone id used to classify big-block only, address is actually 256 * a zone-2 address. 257 */ 258 struct hammer_blockmap { 259 hammer_off_t phys_offset; /* zone-2 physical offset */ 260 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */ 261 hammer_off_t next_offset; /* zone-X logical offset */ 262 hammer_off_t alloc_offset; /* zone-X logical offset */ 263 u_int32_t reserved01; 264 hammer_crc_t entry_crc; 265 }; 266 267 typedef struct hammer_blockmap *hammer_blockmap_t; 268 269 #define HAMMER_BLOCKMAP_CRCSIZE \ 270 offsetof(struct hammer_blockmap, entry_crc) 271 272 /* 273 * The blockmap is a 2-layer entity made up of big-blocks. The first layer 274 * contains 262144 32-byte entries (18 bits), the second layer contains 275 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps. 276 * 18+19+23 = 60 bits. The top four bits are the zone id. 277 * 278 * Currently only the freemap utilizes both layers in all their glory. 279 * All primary data/meta-data zones actually encode a zone-2 address 280 * requiring no real blockmap translation. 281 * 282 * The freemap uses the upper 8 bits of layer-1 to identify the volume, 283 * thus any space allocated via the freemap can be directly translated 284 * to a zone:2 (or zone:8-15) address. 285 * 286 * zone-X blockmap offset: [zone:4][layer1:18][layer2:19][big-block:23] 287 */ 288 struct hammer_blockmap_layer1 { 289 hammer_off_t blocks_free; /* big-blocks free */ 290 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */ 291 hammer_off_t reserved01; 292 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */ 293 /* (not yet used) */ 294 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/ 295 }; 296 297 typedef struct hammer_blockmap_layer1 *hammer_blockmap_layer1_t; 298 299 #define HAMMER_LAYER1_CRCSIZE \ 300 offsetof(struct hammer_blockmap_layer1, layer1_crc) 301 302 /* 303 * layer2 entry for 8MB big-block. 304 * 305 * NOTE: bytes_free is signed and can legally go negative if/when data 306 * de-dup occurs. This field will never go higher than 307 * HAMMER_BIGBLOCK_SIZE. If exactly HAMMER_BIGBLOCK_SIZE 308 * the big-block is completely free. 309 */ 310 struct hammer_blockmap_layer2 { 311 u_int8_t zone; /* typed allocation zone */ 312 u_int8_t unused01; 313 u_int16_t unused02; 314 u_int32_t append_off; /* allocatable space index */ 315 int32_t bytes_free; /* bytes free within this big-block */ 316 hammer_crc_t entry_crc; 317 }; 318 319 typedef struct hammer_blockmap_layer2 *hammer_blockmap_layer2_t; 320 321 #define HAMMER_LAYER2_CRCSIZE \ 322 offsetof(struct hammer_blockmap_layer2, entry_crc) 323 324 #define HAMMER_BLOCKMAP_FREE 0ULL 325 #define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL) 326 327 #define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \ 328 (HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1)) 329 #define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \ 330 (HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2)) 331 332 #define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \ 333 (HAMMER_BLOCKMAP_RADIX1 / HAMMER_BUFFERS_PER_BIGBLOCK) 334 #define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \ 335 (HAMMER_BLOCKMAP_RADIX2 / HAMMER_BUFFERS_PER_BIGBLOCK) 336 337 #define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 - 1EB */ \ 338 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2) 339 #define HAMMER_BLOCKMAP_LAYER2 /* 19+23 - 4TB */ \ 340 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_BIGBLOCK_SIZE64) 341 342 #define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1) 343 #define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1) 344 345 /* 346 * Index within layer1 or layer2 big-block for the entry representing 347 * a zone-2 physical offset. 348 */ 349 #define HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) \ 350 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \ 351 HAMMER_BLOCKMAP_LAYER2) 352 353 #define HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) \ 354 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \ 355 HAMMER_BIGBLOCK_SIZE64) 356 357 /* 358 * Byte offset within layer1 or layer2 big-block for the entry representing 359 * a zone-2 physical offset. Multiply the index by sizeof(blockmap_layer). 360 */ 361 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \ 362 (HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) * \ 363 sizeof(struct hammer_blockmap_layer1)) 364 365 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \ 366 (HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) * \ 367 sizeof(struct hammer_blockmap_layer2)) 368 369 /* 370 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume 371 * header with an array of layer2 structures. A maximum of (128x8MB) = 1GB 372 * may be reserved. The size of the undo fifo is usually set a newfs time 373 * but can be adjusted if the filesystem is taken offline. 374 */ 375 #define HAMMER_UNDO_LAYER2 128 /* max layer2 undo mapping entries */ 376 377 /* 378 * All on-disk HAMMER structures which make up elements of the UNDO FIFO 379 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure 380 * contains all the information required to validate the fifo element 381 * and to scan the fifo in either direction. The head is typically embedded 382 * in higher level hammer on-disk structures while the tail is typically 383 * out-of-band. hdr_size is the size of the whole mess, including the tail. 384 * 385 * All undo structures are guaranteed to not cross a 16K filesystem 386 * buffer boundary. Most undo structures are fairly small. Data spaces 387 * are not immediately reused by HAMMER so file data is not usually recorded 388 * as part of an UNDO. 389 * 390 * PAD elements are allowed to take up only 8 bytes of space as a special 391 * case, containing only hdr_signature, hdr_type, and hdr_size fields, 392 * and with the tail overloaded onto the head structure for 8 bytes total. 393 * 394 * Every undo record has a sequence number. This number is unrelated to 395 * transaction ids and instead collects the undo transactions associated 396 * with a single atomic operation. A larger transactional operation, such 397 * as a remove(), may consist of several smaller atomic operations 398 * representing raw meta-data operations. 399 * 400 * HAMMER VERSION 4 CHANGES 401 * 402 * In HAMMER version 4 the undo structure alignment is reduced from 16384 403 * to 512 bytes in order to ensure that each 512 byte sector begins with 404 * a header. The reserved01 field in the header is now a 32 bit sequence 405 * number. This allows the recovery code to detect missing sectors 406 * without relying on the 32-bit crc and to definitively identify the current 407 * undo sequence space without having to rely on information from the volume 408 * header. In addition, new REDO entries in the undo space are used to 409 * record write, write/extend, and transaction id updates. 410 * 411 * The grand result is: 412 * 413 * (1) The volume header no longer needs to be synchronized for most 414 * flush and fsync operations. 415 * 416 * (2) Most fsync operations need only lay down REDO records 417 * 418 * (3) Data overwrite for nohistory operations covered by REDO records 419 * can be supported (instead of rolling a new block allocation), 420 * by rolling UNDO for the prior contents of the data. 421 * 422 * HAMMER VERSION 5 CHANGES 423 * 424 * Hammer version 5 contains a minor adjustment making layer2's bytes_free 425 * field signed, allowing dedup to push it into the negative domain. 426 */ 427 #define HAMMER_HEAD_ONDISK_SIZE 32 428 #define HAMMER_HEAD_ALIGN 8 429 #define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1) 430 #define HAMMER_TAIL_ONDISK_SIZE 8 431 #define HAMMER_HEAD_DOALIGN(bytes) \ 432 (((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK) 433 434 #define HAMMER_UNDO_ALIGN 512 435 #define HAMMER_UNDO_ALIGN64 ((u_int64_t)512) 436 #define HAMMER_UNDO_MASK (HAMMER_UNDO_ALIGN - 1) 437 #define HAMMER_UNDO_MASK64 (HAMMER_UNDO_ALIGN64 - 1) 438 439 struct hammer_fifo_head { 440 u_int16_t hdr_signature; 441 u_int16_t hdr_type; 442 u_int32_t hdr_size; /* Aligned size of the whole mess */ 443 u_int32_t hdr_seq; /* Sequence number */ 444 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */ 445 }; 446 447 #define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc) 448 449 struct hammer_fifo_tail { 450 u_int16_t tail_signature; 451 u_int16_t tail_type; 452 u_int32_t tail_size; /* aligned size of the whole mess */ 453 }; 454 455 typedef struct hammer_fifo_head *hammer_fifo_head_t; 456 typedef struct hammer_fifo_tail *hammer_fifo_tail_t; 457 458 /* 459 * Fifo header types. 460 */ 461 #define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE) 462 #define HAMMER_HEAD_TYPE_DUMMY 0x0041U /* dummy entry w/seqno */ 463 #define HAMMER_HEAD_TYPE_42 0x0042U 464 #define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */ 465 #define HAMMER_HEAD_TYPE_REDO 0x0044U /* data REDO / fast fsync */ 466 #define HAMMER_HEAD_TYPE_45 0x0045U 467 468 #define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */ 469 470 #define HAMMER_HEAD_SIGNATURE 0xC84EU 471 #define HAMMER_TAIL_SIGNATURE 0xC74FU 472 473 /* 474 * Misc FIFO structures. 475 * 476 * UNDO - Raw meta-data media updates. 477 */ 478 struct hammer_fifo_undo { 479 struct hammer_fifo_head head; 480 hammer_off_t undo_offset; /* zone-1,2 offset */ 481 int32_t undo_data_bytes; 482 int32_t undo_reserved01; 483 /* followed by data */ 484 }; 485 486 /* 487 * REDO (HAMMER version 4+) - Logical file writes/truncates. 488 * 489 * REDOs contain information which will be duplicated in a later meta-data 490 * update, allowing fast write()+fsync() operations. REDOs can be ignored 491 * without harming filesystem integrity but must be processed if fsync() 492 * semantics are desired. 493 * 494 * Unlike UNDOs which are processed backwards within the recovery span, 495 * REDOs must be processed forwards starting further back (starting outside 496 * the recovery span). 497 * 498 * WRITE - Write logical file (with payload). Executed both 499 * out-of-span and in-span. Out-of-span WRITEs may be 500 * filtered out by TERMs. 501 * 502 * TRUNC - Truncate logical file (no payload). Executed both 503 * out-of-span and in-span. Out-of-span WRITEs may be 504 * filtered out by TERMs. 505 * 506 * TERM_* - Indicates meta-data was committed (if out-of-span) or 507 * will be rolled-back (in-span). Any out-of-span TERMs 508 * matching earlier WRITEs remove those WRITEs from 509 * consideration as they might conflict with a later data 510 * commit (which is not being rolled-back). 511 * 512 * SYNC - The earliest in-span SYNC (the last one when scanning 513 * backwards) tells the recovery code how far out-of-span 514 * it must go to run REDOs. 515 * 516 * NOTE: WRITEs do not always have matching TERMs even under 517 * perfect conditions because truncations might remove the 518 * buffers from consideration. I/O problems can also remove 519 * buffers from consideration. 520 * 521 * TRUNCSs do not always have matching TERMs because several 522 * truncations may be aggregated together into a single TERM. 523 */ 524 struct hammer_fifo_redo { 525 struct hammer_fifo_head head; 526 int64_t redo_objid; /* file being written */ 527 hammer_off_t redo_offset; /* logical offset in file */ 528 int32_t redo_data_bytes; 529 u_int32_t redo_flags; 530 u_int32_t redo_localization; 531 u_int32_t redo_reserved; 532 u_int64_t redo_mtime; /* set mtime */ 533 }; 534 535 #define HAMMER_REDO_WRITE 0x00000001 536 #define HAMMER_REDO_TRUNC 0x00000002 537 #define HAMMER_REDO_TERM_WRITE 0x00000004 538 #define HAMMER_REDO_TERM_TRUNC 0x00000008 539 #define HAMMER_REDO_SYNC 0x00000010 540 541 union hammer_fifo_any { 542 struct hammer_fifo_head head; 543 struct hammer_fifo_undo undo; 544 struct hammer_fifo_redo redo; 545 }; 546 547 typedef struct hammer_fifo_redo *hammer_fifo_redo_t; 548 typedef struct hammer_fifo_undo *hammer_fifo_undo_t; 549 typedef union hammer_fifo_any *hammer_fifo_any_t; 550 551 /* 552 * Volume header types 553 */ 554 #define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */ 555 #define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */ 556 557 /* 558 * HAMMER Volume header 559 * 560 * A HAMMER filesystem is built from any number of block devices, Each block 561 * device contains a volume header followed by however many buffers fit 562 * into the volume. 563 * 564 * One of the volumes making up a HAMMER filesystem is the master, the 565 * rest are slaves. It does not have to be volume #0. 566 * 567 * The volume header takes up an entire 16K filesystem buffer and may 568 * represent up to 64KTB (65536 TB) of space. 569 * 570 * Special field notes: 571 * 572 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes) 573 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes) 574 * vol_buf_beg - offset of the first buffer. 575 * 576 * The memory log area allows a kernel to cache new records and data 577 * in memory without allocating space in the actual filesystem to hold 578 * the records and data. In the event that a filesystem becomes full, 579 * any records remaining in memory can be flushed to the memory log 580 * area. This allows the kernel to immediately return success. 581 */ 582 583 #define HAMMER_BOOT_MINBYTES (32*1024) 584 #define HAMMER_BOOT_NOMBYTES (64LL*1024*1024) 585 #define HAMMER_BOOT_MAXBYTES (256LL*1024*1024) 586 587 #define HAMMER_MEM_MINBYTES (256*1024) 588 #define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024) 589 #define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024) 590 591 struct hammer_volume_ondisk { 592 u_int64_t vol_signature;/* Signature */ 593 594 int64_t vol_bot_beg; /* byte offset of boot area or 0 */ 595 int64_t vol_mem_beg; /* byte offset of memory log or 0 */ 596 int64_t vol_buf_beg; /* byte offset of first buffer in volume */ 597 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */ 598 int64_t vol_locked; /* reserved clusters are >= this offset */ 599 600 uuid_t vol_fsid; /* identify filesystem */ 601 uuid_t vol_fstype; /* identify filesystem type */ 602 char vol_name[64]; /* Name of volume */ 603 604 int32_t vol_no; /* volume number within filesystem */ 605 int32_t vol_count; /* number of volumes making up FS */ 606 607 u_int32_t vol_version; /* version control information */ 608 hammer_crc_t vol_crc; /* header crc */ 609 u_int32_t vol_flags; /* volume flags */ 610 u_int32_t vol_rootvol; /* which volume is the root volume? */ 611 612 int32_t vol_reserved04; 613 int32_t vol_reserved05; 614 u_int32_t vol_reserved06; 615 u_int32_t vol_reserved07; 616 617 int32_t vol_blocksize; /* for statfs only */ 618 int32_t vol_reserved08; 619 int64_t vol_nblocks; /* total allocatable hammer bufs */ 620 621 /* 622 * These fields are initialized and space is reserved in every 623 * volume making up a HAMMER filesytem, but only the master volume 624 * contains valid data. Note that vol0_stat_bigblocks does not 625 * include big-blocks for freemap and undomap initially allocated 626 * by newfs_hammer. 627 */ 628 int64_t vol0_stat_bigblocks; /* total big-blocks when fs is empty */ 629 int64_t vol0_stat_freebigblocks;/* number of free big-blocks */ 630 int64_t vol0_stat_bytes; /* for statfs only */ 631 int64_t vol0_stat_inodes; /* for statfs only */ 632 int64_t vol0_stat_records; /* total records in filesystem */ 633 hammer_off_t vol0_btree_root; /* B-Tree root */ 634 hammer_tid_t vol0_next_tid; /* highest partially synchronized TID */ 635 hammer_off_t vol0_unused03; 636 637 /* 638 * Blockmaps for zones. Not all zones use a blockmap. Note that 639 * the entire root blockmap is cached in the hammer_mount structure. 640 */ 641 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES]; 642 643 /* 644 * Array of zone-2 addresses for undo FIFO. 645 */ 646 hammer_off_t vol0_undo_array[HAMMER_UNDO_LAYER2]; 647 }; 648 649 typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t; 650 651 #define HAMMER_VOLF_VALID 0x0001 /* valid entry */ 652 #define HAMMER_VOLF_OPEN 0x0002 /* volume is open */ 653 #define HAMMER_VOLF_NEEDFLUSH 0x0004 /* volume needs flush */ 654 655 #define HAMMER_VOL_CRCSIZE1 \ 656 offsetof(struct hammer_volume_ondisk, vol_crc) 657 #define HAMMER_VOL_CRCSIZE2 \ 658 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \ 659 sizeof(hammer_crc_t)) 660 661 #define HAMMER_VOL_VERSION_MIN 1 /* minimum supported version */ 662 #define HAMMER_VOL_VERSION_DEFAULT 6 /* newfs default version */ 663 #define HAMMER_VOL_VERSION_WIP 7 /* version >= this is WIP */ 664 #define HAMMER_VOL_VERSION_MAX 6 /* maximum supported version */ 665 666 #define HAMMER_VOL_VERSION_ONE 1 667 #define HAMMER_VOL_VERSION_TWO 2 /* new dirent layout (2.3+) */ 668 #define HAMMER_VOL_VERSION_THREE 3 /* new snapshot layout (2.5+) */ 669 #define HAMMER_VOL_VERSION_FOUR 4 /* new undo/flush (2.5+) */ 670 #define HAMMER_VOL_VERSION_FIVE 5 /* dedup (2.9+) */ 671 #define HAMMER_VOL_VERSION_SIX 6 /* DIRHASH_ALG1 */ 672 673 /* 674 * Record types are fairly straightforward. The B-Tree includes the record 675 * type in its index sort. 676 */ 677 #define HAMMER_RECTYPE_UNKNOWN 0 678 #define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */ 679 #define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */ 680 #define HAMMER_RECTYPE_UNUSED02 2 681 #define HAMMER_RECTYPE_UNUSED03 3 682 #define HAMMER_RECTYPE_DATA 0x0010 683 #define HAMMER_RECTYPE_DIRENTRY 0x0011 684 #define HAMMER_RECTYPE_DB 0x0012 685 #define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */ 686 #define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */ 687 #define HAMMER_RECTYPE_PFS 0x0015 /* PFS management */ 688 #define HAMMER_RECTYPE_SNAPSHOT 0x0016 /* Snapshot management */ 689 #define HAMMER_RECTYPE_CONFIG 0x0017 /* hammer cleanup config */ 690 #define HAMMER_RECTYPE_MOVED 0x8000 /* special recovery flag */ 691 #define HAMMER_RECTYPE_MAX 0xFFFF 692 693 #define HAMMER_RECTYPE_ENTRY_START (HAMMER_RECTYPE_INODE + 1) 694 #define HAMMER_RECTYPE_CLEAN_START HAMMER_RECTYPE_EXT 695 696 #define HAMMER_FIXKEY_SYMLINK 1 697 698 #define HAMMER_OBJTYPE_UNKNOWN 0 /* never exists on-disk as unknown */ 699 #define HAMMER_OBJTYPE_DIRECTORY 1 700 #define HAMMER_OBJTYPE_REGFILE 2 701 #define HAMMER_OBJTYPE_DBFILE 3 702 #define HAMMER_OBJTYPE_FIFO 4 703 #define HAMMER_OBJTYPE_CDEV 5 704 #define HAMMER_OBJTYPE_BDEV 6 705 #define HAMMER_OBJTYPE_SOFTLINK 7 706 #define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */ 707 #define HAMMER_OBJTYPE_SOCKET 9 708 709 /* 710 * HAMMER inode attribute data 711 * 712 * The data reference for a HAMMER inode points to this structure. Any 713 * modifications to the contents of this structure will result in a 714 * replacement operation. 715 * 716 * parent_obj_id is only valid for directories (which cannot be hard-linked), 717 * and specifies the parent directory obj_id. This field will also be set 718 * for non-directory inodes as a recovery aid, but can wind up holding 719 * stale information. However, since object id's are not reused, the worse 720 * that happens is that the recovery code is unable to use it. 721 * 722 * NOTE: Future note on directory hardlinks. We can implement a record type 723 * which allows us to point to multiple parent directories. 724 */ 725 struct hammer_inode_data { 726 u_int16_t version; /* inode data version */ 727 u_int16_t mode; /* basic unix permissions */ 728 u_int32_t uflags; /* chflags */ 729 u_int32_t rmajor; /* used by device nodes */ 730 u_int32_t rminor; /* used by device nodes */ 731 u_int64_t ctime; 732 int64_t parent_obj_id; /* parent directory obj_id */ 733 uuid_t uid; 734 uuid_t gid; 735 736 u_int8_t obj_type; 737 u_int8_t cap_flags; /* capability support flags (extension) */ 738 u_int16_t reserved02; 739 u_int32_t reserved03; /* RESERVED FOR POSSIBLE FUTURE BIRTHTIME */ 740 u_int64_t nlinks; /* hard links */ 741 u_int64_t size; /* filesystem object size */ 742 union { 743 struct { 744 char reserved06[16]; 745 u_int32_t parent_obj_localization; 746 u_int32_t integrity_crc; 747 } obj; 748 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */ 749 } ext; 750 u_int64_t mtime; /* mtime must be second-to-last */ 751 u_int64_t atime; /* atime must be last */ 752 }; 753 754 /* 755 * Neither mtime nor atime upates are CRCd by the B-Tree element. 756 * mtime updates have UNDO, atime updates do not. 757 */ 758 #define HAMMER_ITIMES_BASE(ino_data) (&(ino_data)->mtime) 759 #define HAMMER_ITIMES_BYTES (sizeof(u_int64_t) * 2) 760 761 #define HAMMER_INODE_CRCSIZE \ 762 offsetof(struct hammer_inode_data, mtime) 763 764 #define HAMMER_INODE_DATA_VERSION 1 765 #define HAMMER_OBJID_ROOT 1 /* root inodes # */ 766 #define HAMMER_INODE_BASESYMLEN 24 /* see ext.symlink */ 767 768 /* 769 * Capability & implementation flags. 770 * 771 * DIR_LOCAL_INO - Use inode B-Tree localization for directory entries. 772 */ 773 #define HAMMER_INODE_CAP_DIRHASH_MASK 0x03 /* directory: hash algorithm */ 774 #define HAMMER_INODE_CAP_DIRHASH_ALG0 0x00 775 #define HAMMER_INODE_CAP_DIRHASH_ALG1 0x01 776 #define HAMMER_INODE_CAP_DIRHASH_ALG2 0x02 777 #define HAMMER_INODE_CAP_DIRHASH_ALG3 0x03 778 #define HAMMER_INODE_CAP_DIR_LOCAL_INO 0x04 /* use inode localization */ 779 780 /* 781 * A HAMMER directory entry associates a HAMMER filesystem object with a 782 * namespace. It is possible to hook into a pseudo-filesystem (with its 783 * own inode numbering space) in the filesystem by setting the high 784 * 16 bits of the localization field. The low 16 bits must be 0 and 785 * are reserved for future use. 786 * 787 * Directory entries are indexed with a 128 bit namekey rather then an 788 * offset. A portion of the namekey is an iterator/randomizer to deal 789 * with collisions. 790 * 791 * NOTE: leaf.base.obj_type from the related B-Tree leaf entry holds 792 * the filesystem object type of obj_id, e.g. a den_type equivalent. 793 * It is not stored in hammer_entry_data. 794 * 795 * NOTE: name field / the filename data reference is NOT terminated with \0. 796 */ 797 struct hammer_entry_data { 798 int64_t obj_id; /* object being referenced */ 799 u_int32_t localization; /* identify pseudo-filesystem */ 800 u_int32_t reserved02; 801 char name[16]; /* name (extended) */ 802 }; 803 804 #define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_entry_data, name[0]) 805 #define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_entry_data, name[nlen]) 806 807 /* 808 * Symlink data which does not fit in the inode is stored in a separate 809 * FIX type record. 810 */ 811 struct hammer_symlink_data { 812 char name[16]; /* name (extended) */ 813 }; 814 815 #define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0]) 816 817 /* 818 * The root inode for the primary filesystem and root inode for any 819 * pseudo-fs may be tagged with an optional data structure using 820 * HAMMER_RECTYPE_PFS and localization id. This structure allows 821 * the node to be used as a mirroring master or slave. 822 * 823 * When operating as a slave CD's into the node automatically become read-only 824 * and as-of sync_end_tid. 825 * 826 * When operating as a master the read PFSD info sets sync_end_tid to 827 * the most recently flushed TID. 828 * 829 * sync_low_tid is not yet used but will represent the highest pruning 830 * end-point, after which full history is available. 831 * 832 * We need to pack this structure making it equally sized on both 32-bit and 833 * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is 834 * send over the wire in hammer mirror operations. Only on 64-bit machines 835 * the size of this struct differ when packed or not. This leads us to the 836 * situation where old 64-bit systems (using the non-packed structure), 837 * which were never able to mirror to/from 32-bit systems, are now no longer 838 * able to mirror to/from newer 64-bit systems (using the packed structure). 839 */ 840 struct hammer_pseudofs_data { 841 hammer_tid_t sync_low_tid; /* full history beyond this point */ 842 hammer_tid_t sync_beg_tid; /* earliest tid w/ full history avail */ 843 hammer_tid_t sync_end_tid; /* current synchronizatoin point */ 844 u_int64_t sync_beg_ts; /* real-time of last completed sync */ 845 u_int64_t sync_end_ts; /* initiation of current sync cycle */ 846 uuid_t shared_uuid; /* shared uuid (match required) */ 847 uuid_t unique_uuid; /* unique uuid of this master/slave */ 848 int32_t reserved01; /* reserved for future master_id */ 849 int32_t mirror_flags; /* misc flags */ 850 char label[64]; /* filesystem space label */ 851 char snapshots[64]; /* softlink dir for pruning */ 852 int16_t prune_time; /* how long to spend pruning */ 853 int16_t prune_freq; /* how often we prune */ 854 int16_t reblock_time; /* how long to spend reblocking */ 855 int16_t reblock_freq; /* how often we reblock */ 856 int32_t snapshot_freq; /* how often we create a snapshot */ 857 int32_t prune_min; /* do not prune recent history */ 858 int32_t prune_max; /* do not retain history beyond here */ 859 int32_t reserved[16]; 860 } __packed; 861 862 typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t; 863 864 #define HAMMER_PFSD_SLAVE 0x00000001 865 #define HAMMER_PFSD_DELETED 0x80000000 866 867 /* 868 * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }. 869 * 870 * Snapshot records replace the old <fs>/snapshots/<softlink> methodology. Snapshot 871 * records are mirrored but may be independantly managed once they are laid down on 872 * a slave. 873 * 874 * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the 875 * results. 876 * 877 * NOTE: Reserved fields must be zero (as usual) 878 */ 879 struct hammer_snapshot_data { 880 hammer_tid_t tid; /* the snapshot TID itself (== key) */ 881 u_int64_t ts; /* real-time when snapshot was made */ 882 u_int64_t reserved01; 883 u_int64_t reserved02; 884 char label[64]; /* user-supplied description */ 885 u_int64_t reserved03[4]; 886 }; 887 888 /* 889 * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }. 890 * 891 * Used to store the hammer cleanup config. This data is not mirrored. 892 */ 893 struct hammer_config_data { 894 char text[1024]; 895 }; 896 897 /* 898 * Rollup various structures embedded as record data 899 */ 900 union hammer_data_ondisk { 901 struct hammer_entry_data entry; 902 struct hammer_inode_data inode; 903 struct hammer_symlink_data symlink; 904 struct hammer_pseudofs_data pfsd; 905 struct hammer_snapshot_data snap; 906 struct hammer_config_data config; 907 }; 908 909 typedef union hammer_data_ondisk *hammer_data_ondisk_t; 910 911 /* 912 * Ondisk layout of B-Tree related structures 913 */ 914 #include "hammer_btree.h" 915 #endif 916