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