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 1.27 2008/03/19 20:18:17 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 spam 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 * Per-volume storage limit: 52 bits 4096 TB 64 * Per-Zone storage limit: 59 bits 512 KTB (due to blockmap) 65 * Per-filesystem storage limit: 60 bits 1 MTB 66 */ 67 #define HAMMER_BUFSIZE 16384 68 #define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1) 69 #define HAMMER_MAXDATA (256*1024) 70 #define HAMMER_BUFFER_BITS 14 71 72 #if (1 << HAMMER_BUFFER_BITS) != HAMMER_BUFSIZE 73 #error "HAMMER_BUFFER_BITS BROKEN" 74 #endif 75 76 #define HAMMER_BUFSIZE64 ((u_int64_t)HAMMER_BUFSIZE) 77 #define HAMMER_BUFMASK64 ((u_int64_t)HAMMER_BUFMASK) 78 79 #define HAMMER_OFF_ZONE_MASK 0xF000000000000000ULL /* zone portion */ 80 #define HAMMER_OFF_VOL_MASK 0x0FF0000000000000ULL /* volume portion */ 81 #define HAMMER_OFF_SHORT_MASK 0x000FFFFFFFFFFFFFULL /* offset portion */ 82 #define HAMMER_OFF_LONG_MASK 0x0FFFFFFFFFFFFFFFULL /* offset portion */ 83 #define HAMMER_OFF_SHORT_REC_MASK 0x000FFFFFFF000000ULL /* recovery boundary */ 84 #define HAMMER_OFF_LONG_REC_MASK 0x0FFFFFFFFF000000ULL /* recovery boundary */ 85 #define HAMMER_RECOVERY_BND 0x0000000001000000ULL 86 87 /* 88 * Hammer transction ids are 64 bit unsigned integers and are usually 89 * synchronized with the time of day in nanoseconds. 90 * 91 * Hammer offsets are used for FIFO indexing and embed a cycle counter 92 * and volume number in addition to the offset. Most offsets are required 93 * to be 64-byte aligned. 94 */ 95 typedef u_int64_t hammer_tid_t; 96 typedef u_int64_t hammer_off_t; 97 98 #define HAMMER_MIN_TID 0ULL /* unsigned */ 99 #define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */ 100 #define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */ 101 #define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */ 102 #define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */ 103 #define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */ 104 #define HAMMER_MIN_RECTYPE 0x0U /* unsigned */ 105 #define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */ 106 #define HAMMER_MIN_OFFSET 0ULL /* unsigned */ 107 #define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */ 108 109 /* 110 * hammer_off_t has several different encodings. Note that not all zones 111 * encode a vol_no. 112 * 113 * zone 0 (z,v,o): reserved (for sanity) 114 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header) 115 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer) 116 * zone 3 (z,o): undo fifo - fixed layer2 array in root vol hdr 117 * zone 4 (z,v,o): freemap - freemap-backed self-mapping special 118 * cased layering. 119 * 120 * zone 8 (z,o): B-Tree - blkmap-backed 121 * zone 9 (z,o): Record - blkmap-backed 122 * zone 10 (z,o): Large-data - blkmap-backed 123 */ 124 125 #define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL 126 #define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL 127 #define HAMMER_ZONE_UNDO 0x3000000000000000ULL 128 #define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL 129 #define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL 130 #define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL 131 #define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL 132 #define HAMMER_ZONE_BTREE 0x8000000000000000ULL 133 #define HAMMER_ZONE_RECORD 0x9000000000000000ULL 134 #define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL 135 #define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL 136 #define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL 137 #define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL 138 #define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL 139 #define HAMMER_ZONE_RESERVED0F 0xF000000000000000ULL 140 141 #define HAMMER_ZONE_RAW_VOLUME_INDEX 1 142 #define HAMMER_ZONE_RAW_BUFFER_INDEX 2 143 #define HAMMER_ZONE_UNDO_INDEX 3 144 #define HAMMER_ZONE_FREEMAP_INDEX 4 145 #define HAMMER_ZONE_BTREE_INDEX 8 146 #define HAMMER_ZONE_RECORD_INDEX 9 147 #define HAMMER_ZONE_LARGE_DATA_INDEX 10 148 #define HAMMER_ZONE_SMALL_DATA_INDEX 11 149 150 /* 151 * Per-zone size limitation. This just makes the iterator easier 152 * to deal with by preventing an iterator overflow. 153 */ 154 #define HAMMER_ZONE_LIMIT \ 155 (0x1000000000000000ULL - HAMMER_BLOCKMAP_LAYER2) 156 157 #define HAMMER_MAX_ZONES 16 158 159 #define HAMMER_VOL_ENCODE(vol_no) \ 160 ((hammer_off_t)((vol_no) & 255) << 52) 161 #define HAMMER_VOL_DECODE(ham_off) \ 162 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255) 163 #define HAMMER_ZONE_DECODE(ham_off) \ 164 (int32_t)(((hammer_off_t)(ham_off) >> 60)) 165 #define HAMMER_ZONE_ENCODE(zone, ham_off) \ 166 (((hammer_off_t)(zone) << 60) | (ham_off)) 167 #define HAMMER_SHORT_OFF_ENCODE(offset) \ 168 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK) 169 #define HAMMER_LONG_OFF_ENCODE(offset) \ 170 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK) 171 172 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \ 173 (HAMMER_ZONE_RAW_VOLUME | \ 174 HAMMER_VOL_ENCODE(vol_no) | \ 175 HAMMER_SHORT_OFF_ENCODE(offset)) 176 177 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \ 178 (HAMMER_ZONE_RAW_BUFFER | \ 179 HAMMER_VOL_ENCODE(vol_no) | \ 180 HAMMER_SHORT_OFF_ENCODE(offset)) 181 182 #define HAMMER_ENCODE_FREEMAP(vol_no, offset) \ 183 (HAMMER_ZONE_FREEMAP | \ 184 HAMMER_VOL_ENCODE(vol_no) | \ 185 HAMMER_SHORT_OFF_ENCODE(offset)) 186 187 /* 188 * Large-Block backing store 189 * 190 * A blockmap is a two-level map which translates a blockmap-backed zone 191 * offset into a raw zone 2 offset. Each layer handles 18 bits. The 8M 192 * large-block size is 23 bits so two layers gives us 23+18+18 = 59 bits 193 * of address space. 194 */ 195 #define HAMMER_LARGEBLOCK_SIZE (8192 * 1024) 196 #define HAMMER_LARGEBLOCK_SIZE64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE) 197 #define HAMMER_LARGEBLOCK_MASK (HAMMER_LARGEBLOCK_SIZE - 1) 198 #define HAMMER_LARGEBLOCK_MASK64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE - 1) 199 #define HAMMER_LARGEBLOCK_BITS 23 200 #if (1 << HAMMER_LARGEBLOCK_BITS) != HAMMER_LARGEBLOCK_SIZE 201 #error "HAMMER_LARGEBLOCK_BITS BROKEN" 202 #endif 203 204 #define HAMMER_BUFFERS_PER_LARGEBLOCK \ 205 (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE) 206 #define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK \ 207 (HAMMER_BUFFERS_PER_LARGEBLOCK - 1) 208 #define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK64 \ 209 ((hammer_off_t)HAMMER_BUFFERS_PER_LARGEBLOCK_MASK) 210 211 /* 212 * Every blockmap has this root structure in the root volume header. 213 * 214 * NOTE: zone 3 (the undo FIFO) does not use phys_offset. first and next 215 * offsets represent the FIFO. 216 */ 217 struct hammer_blockmap { 218 hammer_off_t phys_offset; /* zone-2 physical offset */ 219 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */ 220 hammer_off_t next_offset; /* zone-X logical offset */ 221 hammer_off_t alloc_offset; /* zone-X logical offset */ 222 u_int32_t entry_crc; 223 u_int32_t reserved01; 224 }; 225 226 typedef struct hammer_blockmap *hammer_blockmap_t; 227 228 /* 229 * The blockmap is a 2-layer entity made up of big-blocks. The first layer 230 * contains 262144 32-byte entries (18 bits), the second layer contains 231 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps. 232 * 18+19+23 = 60 bits. The top four bits are the zone id. 233 * 234 * Layer 2 encodes the physical bigblock mapping for a blockmap. The freemap 235 * uses this field to encode the virtual blockmap offset that allocated the 236 * physical block. 237 * 238 * NOTE: The freemap maps the vol_no in the upper 8 bits of layer1. 239 * 240 * zone-4 blockmap offset: [z:4][layer1:18][layer2:19][bigblock:23] 241 */ 242 struct hammer_blockmap_layer1 { 243 hammer_off_t blocks_free; /* big-blocks free */ 244 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */ 245 u_int32_t layer1_crc; /* crc of this entry */ 246 u_int32_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */ 247 hammer_off_t reserved01; 248 }; 249 250 struct hammer_blockmap_layer2 { 251 u_int32_t entry_crc; 252 u_int32_t bytes_free; /* bytes free within this bigblock */ 253 union { 254 hammer_off_t owner; /* used by freemap */ 255 hammer_off_t phys_offset; /* used by blockmap */ 256 } u; 257 }; 258 259 #define HAMMER_BLOCKMAP_FREE 0ULL 260 #define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL) 261 262 #define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \ 263 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1)) 264 #define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \ 265 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2)) 266 267 #define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \ 268 (HAMMER_BLOCKMAP_RADIX1 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE)) 269 #define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \ 270 (HAMMER_BLOCKMAP_RADIX2 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE)) 271 272 #define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 */ \ 273 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2) 274 #define HAMMER_BLOCKMAP_LAYER2 /* 19+23 */ \ 275 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_LARGEBLOCK_SIZE64) 276 277 #define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1) 278 #define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1) 279 280 /* 281 * byte offset within layer1 or layer2 big-block for the entry representing 282 * a zone-2 physical offset. 283 */ 284 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \ 285 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \ 286 HAMMER_BLOCKMAP_LAYER2 * sizeof(struct hammer_blockmap_layer1)) 287 288 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \ 289 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \ 290 HAMMER_LARGEBLOCK_SIZE64 * sizeof(struct hammer_blockmap_layer2)) 291 292 /* 293 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume 294 * header with an array of layer2 structures. A maximum of (64x8MB) = 512MB 295 * may be reserved. The size of the undo fifo is usually set a newfs time 296 * but can be adjusted if the filesystem is taken offline. 297 */ 298 299 #define HAMMER_UNDO_LAYER2 64 /* max layer2 undo mapping entries */ 300 301 /* 302 * All on-disk HAMMER structures which make up elements of the UNDO FIFO 303 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure 304 * contains all the information required to validate the fifo element 305 * and to scan the fifo in either direction. The head is typically embedded 306 * in higher level hammer on-disk structures while the tail is typically 307 * out-of-band. hdr_size is the size of the whole mess, including the tail. 308 * 309 * All undo structures are guaranteed to not cross a 16K filesystem 310 * buffer boundary. Most undo structures are fairly small. Data spaces 311 * are not immediately reused by HAMMER so file data is not usually recorded 312 * as part of an UNDO. 313 * 314 * PAD elements are allowed to take up only 8 bytes of space as a special 315 * case, containing only hdr_signature, hdr_type, and hdr_size fields, 316 * and with the tail overloaded onto the head structure for 8 bytes total. 317 * 318 * Every undo record has a sequence number. This number is unrelated to 319 * transaction ids and instead collects the undo transactions associated 320 * with a single atomic operation. A larger transactional operation, such 321 * as a remove(), may consist of several smaller atomic operations 322 * representing raw meta-data operations. 323 */ 324 #define HAMMER_HEAD_ONDISK_SIZE 32 325 #define HAMMER_HEAD_ALIGN 8 326 #define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1) 327 #define HAMMER_TAIL_ONDISK_SIZE 8 328 329 struct hammer_fifo_head { 330 u_int16_t hdr_signature; 331 u_int16_t hdr_type; 332 u_int32_t hdr_size; /* aligned size of the whole mess */ 333 u_int32_t hdr_crc; 334 u_int32_t hdr_seq; 335 u_int64_t hdr_tid; /* related TID */ 336 }; 337 338 struct hammer_fifo_tail { 339 u_int16_t tail_signature; 340 u_int16_t tail_type; 341 u_int32_t tail_size; /* aligned size of the whole mess */ 342 }; 343 344 typedef struct hammer_fifo_head *hammer_fifo_head_t; 345 typedef struct hammer_fifo_tail *hammer_fifo_tail_t; 346 347 /* 348 * Fifo header types. 349 */ 350 #define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE) 351 #define HAMMER_HEAD_TYPE_VOL 0x0041U /* Volume (dummy header) */ 352 #define HAMMER_HEAD_TYPE_BTREE 0x0042U /* B-Tree node */ 353 #define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */ 354 #define HAMMER_HEAD_TYPE_DELETE 0x0044U /* record deletion */ 355 #define HAMMER_HEAD_TYPE_RECORD 0x0045U /* Filesystem record */ 356 357 #define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */ 358 359 #define HAMMER_HEAD_SIGNATURE 0xC84EU 360 #define HAMMER_TAIL_SIGNATURE 0xC74FU 361 362 #define HAMMER_HEAD_SEQ_BEG 0x80000000U 363 #define HAMMER_HEAD_SEQ_END 0x40000000U 364 #define HAMMER_HEAD_SEQ_MASK 0x3FFFFFFFU 365 366 /* 367 * Misc FIFO structures. 368 */ 369 struct hammer_fifo_undo { 370 struct hammer_fifo_head head; 371 hammer_off_t undo_offset; /* zone-1 offset */ 372 int32_t undo_data_bytes; 373 int32_t undo_reserved01; 374 /* followed by data */ 375 }; 376 377 typedef struct hammer_fifo_undo *hammer_fifo_undo_t; 378 379 struct hammer_fifo_buf_commit { 380 hammer_off_t undo_offset; 381 }; 382 383 /* 384 * Volume header types 385 */ 386 #define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */ 387 #define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */ 388 389 /* 390 * The B-Tree structures need hammer_fsbuf_head. 391 */ 392 #include "hammer_btree.h" 393 394 /* 395 * HAMMER Volume header 396 * 397 * A HAMMER filesystem is built from any number of block devices, Each block 398 * device contains a volume header followed by however many buffers fit 399 * into the volume. 400 * 401 * One of the volumes making up a HAMMER filesystem is the master, the 402 * rest are slaves. It does not have to be volume #0. 403 * 404 * The volume header takes up an entire 16K filesystem buffer and may 405 * represent up to 64KTB (65536 TB) of space. 406 * 407 * Special field notes: 408 * 409 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes) 410 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes) 411 * vol_buf_beg - offset of the first buffer. 412 * 413 * The memory log area allows a kernel to cache new records and data 414 * in memory without allocating space in the actual filesystem to hold 415 * the records and data. In the event that a filesystem becomes full, 416 * any records remaining in memory can be flushed to the memory log 417 * area. This allows the kernel to immediately return success. 418 */ 419 420 #define HAMMER_BOOT_MINBYTES (32*1024) 421 #define HAMMER_BOOT_NOMBYTES (64LL*1024*1024) 422 #define HAMMER_BOOT_MAXBYTES (256LL*1024*1024) 423 424 #define HAMMER_MEM_MINBYTES (256*1024) 425 #define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024) 426 #define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024) 427 428 struct hammer_volume_ondisk { 429 u_int64_t vol_signature;/* Signature */ 430 431 int64_t vol_bot_beg; /* byte offset of boot area or 0 */ 432 int64_t vol_mem_beg; /* byte offset of memory log or 0 */ 433 int64_t vol_buf_beg; /* byte offset of first buffer in volume */ 434 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */ 435 int64_t vol_locked; /* reserved clusters are >= this offset */ 436 437 uuid_t vol_fsid; /* identify filesystem */ 438 uuid_t vol_fstype; /* identify filesystem type */ 439 char vol_name[64]; /* Name of volume */ 440 441 int32_t vol_no; /* volume number within filesystem */ 442 int32_t vol_count; /* number of volumes making up FS */ 443 444 u_int32_t vol_version; /* version control information */ 445 u_int32_t vol_reserved01; 446 u_int32_t vol_flags; /* volume flags */ 447 u_int32_t vol_rootvol; /* which volume is the root volume? */ 448 449 int32_t vol_reserved04; 450 int32_t vol_reserved05; 451 u_int32_t vol_reserved06; 452 u_int32_t vol_reserved07; 453 454 int32_t vol_blocksize; /* for statfs only */ 455 int32_t vol_reserved08; 456 int64_t vol_nblocks; /* total allocatable hammer bufs */ 457 458 /* 459 * These fields are initialized and space is reserved in every 460 * volume making up a HAMMER filesytem, but only the master volume 461 * contains valid data. 462 */ 463 int64_t vol0_stat_bigblocks; /* total bigblocks when fs is empty */ 464 int64_t vol0_stat_freebigblocks;/* number of free bigblocks */ 465 int64_t vol0_stat_bytes; /* for statfs only */ 466 int64_t vol0_stat_inodes; /* for statfs only */ 467 int64_t vol0_stat_records; /* total records in filesystem */ 468 hammer_off_t vol0_btree_root; /* B-Tree root */ 469 hammer_tid_t vol0_next_tid; /* highest synchronized TID */ 470 u_int32_t vol0_reserved00; 471 u_int32_t vol0_reserved01; 472 473 /* 474 * Blockmaps for zones. Not all zones use a blockmap. 475 */ 476 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES]; 477 478 /* 479 * Layer-2 array for undo fifo 480 */ 481 struct hammer_blockmap_layer2 vol0_undo_array[HAMMER_UNDO_LAYER2]; 482 483 }; 484 485 typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t; 486 487 #define HAMMER_VOLF_VALID 0x0001 /* valid entry */ 488 #define HAMMER_VOLF_OPEN 0x0002 /* volume is open */ 489 490 /* 491 * All HAMMER records have a common 64-byte base and a 32 byte extension, 492 * plus a possible data reference. The data reference can be in-band or 493 * out-of-band. 494 */ 495 496 #define HAMMER_RECORD_SIZE (64+32) 497 498 struct hammer_base_record { 499 u_int32_t signature; /* record signature */ 500 u_int32_t data_crc; /* data crc */ 501 struct hammer_base_elm base; /* 40 byte base element */ 502 hammer_off_t data_off; /* in-band or out-of-band */ 503 int32_t data_len; /* size of data in bytes */ 504 u_int32_t reserved02; 505 }; 506 507 /* 508 * Record types are fairly straightforward. The B-Tree includes the record 509 * type in its index sort. 510 * 511 * In particular please note that it is possible to create a pseudo- 512 * filesystem within a HAMMER filesystem by creating a special object 513 * type within a directory. Pseudo-filesystems are used as replication 514 * targets and even though they are built within a HAMMER filesystem they 515 * get their own obj_id space (and thus can serve as a replication target) 516 * and look like a mount point to the system. 517 * 518 * Inter-cluster records are special-cased in the B-Tree. These records 519 * are referenced from a B-Tree INTERNAL node, NOT A LEAF. This means 520 * that the element in the B-Tree node is actually a boundary element whos 521 * base element fields, including rec_type, reflect the boundary, NOT 522 * the inter-cluster record type. 523 * 524 * HAMMER_RECTYPE_CLUSTER - only set in the actual inter-cluster record, 525 * not set in the left or right boundary elements around the inter-cluster 526 * reference of an internal node in the B-Tree (because doing so would 527 * interfere with the boundary tests). 528 * 529 * NOTE: hammer_ip_delete_range_all() deletes all record types greater 530 * then HAMMER_RECTYPE_INODE. 531 */ 532 #define HAMMER_RECTYPE_UNKNOWN 0 533 #define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */ 534 #define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */ 535 #define HAMMER_RECTYPE_PSEUDO_INODE 2 /* pseudo filesysem */ 536 #define HAMMER_RECTYPE_CLUSTER 3 /* inter-cluster reference */ 537 #define HAMMER_RECTYPE_DATA 0x0010 538 #define HAMMER_RECTYPE_DIRENTRY 0x0011 539 #define HAMMER_RECTYPE_DB 0x0012 540 #define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */ 541 #define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */ 542 #define HAMMER_RECTYPE_MOVED 0x8000 /* special recovery flag */ 543 544 #define HAMMER_FIXKEY_SYMLINK 1 545 546 #define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */ 547 #define HAMMER_OBJTYPE_DIRECTORY 1 548 #define HAMMER_OBJTYPE_REGFILE 2 549 #define HAMMER_OBJTYPE_DBFILE 3 550 #define HAMMER_OBJTYPE_FIFO 4 551 #define HAMMER_OBJTYPE_CDEV 5 552 #define HAMMER_OBJTYPE_BDEV 6 553 #define HAMMER_OBJTYPE_SOFTLINK 7 554 #define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */ 555 556 /* 557 * A HAMMER inode record. 558 * 559 * This forms the basis for a filesystem object. obj_id is the inode number, 560 * key1 represents the pseudo filesystem id for security partitioning 561 * (preventing cross-links and/or restricting a NFS export and specifying the 562 * security policy), and key2 represents the data retention policy id. 563 * 564 * Inode numbers are 64 bit quantities which uniquely identify a filesystem 565 * object for the ENTIRE life of the filesystem, even after the object has 566 * been deleted. For all intents and purposes inode numbers are simply 567 * allocated by incrementing a sequence space. 568 * 569 * There is an important distinction between the data stored in the inode 570 * record and the record's data reference. The record references a 571 * hammer_inode_data structure but the filesystem object size and hard link 572 * count is stored in the inode record itself. This allows multiple inodes 573 * to share the same hammer_inode_data structure. This is possible because 574 * any modifications will lay out new data. The HAMMER implementation need 575 * not use the data-sharing ability when laying down new records. 576 * 577 * A HAMMER inode is subject to the same historical storage requirements 578 * as any other record. In particular any change in filesystem or hard link 579 * count will lay down a new inode record when the filesystem is synced to 580 * disk. This can lead to a lot of junk records which get cleaned up by 581 * the data retention policy. 582 * 583 * The ino_atime and ino_mtime fields are a special case. Modifications to 584 * these fields do NOT lay down a new record by default, though the values 585 * are effectively frozen for snapshots which access historical versions 586 * of the inode record due to other operations. This means that atime will 587 * not necessarily be accurate in snapshots, backups, or mirrors. mtime 588 * will be accurate in backups and mirrors since it can be regenerated from 589 * the mirroring stream. 590 * 591 * Because nlinks is historically retained the hardlink count will be 592 * accurate when accessing a HAMMER filesystem snapshot. 593 */ 594 struct hammer_inode_record { 595 struct hammer_base_record base; 596 u_int64_t ino_atime; /* last access time (not historical) */ 597 u_int64_t ino_mtime; /* last modified time (not historical) */ 598 u_int64_t ino_size; /* filesystem object size */ 599 u_int64_t ino_nlinks; /* hard links */ 600 }; 601 602 /* 603 * Data records specify the entire contents of a regular file object, 604 * including attributes. Small amounts of data can theoretically be 605 * embedded in the record itself but the use of this ability verses using 606 * an out-of-band data reference depends on the implementation. 607 */ 608 struct hammer_data_record { 609 struct hammer_base_record base; 610 char data[32]; 611 }; 612 613 /* 614 * A directory entry specifies the HAMMER filesystem object id, a copy of 615 * the file type, and file name (either embedded or as out-of-band data). 616 * If the file name is short enough to fit into den_name[] (including a 617 * terminating nul) then it will be embedded in the record, otherwise it 618 * is stored out-of-band. The base record's data reference always points 619 * to the nul-terminated filename regardless. 620 * 621 * Directory entries are indexed with a 128 bit namekey rather then an 622 * offset. A portion of the namekey is an iterator or randomizer to deal 623 * with collisions. 624 * 625 * NOTE: base.base.obj_type holds the filesystem object type of obj_id, 626 * e.g. a den_type equivalent. 627 * 628 * NOTE: den_name / the filename data reference is NOT terminated with \0. 629 * 630 */ 631 struct hammer_entry_record { 632 struct hammer_base_record base; 633 u_int64_t obj_id; /* object being referenced */ 634 u_int64_t reserved01; 635 char name[16]; 636 }; 637 638 /* 639 * Hammer rollup record 640 */ 641 union hammer_record_ondisk { 642 struct hammer_base_record base; 643 struct hammer_inode_record inode; 644 struct hammer_data_record data; 645 struct hammer_entry_record entry; 646 }; 647 648 typedef union hammer_record_ondisk *hammer_record_ondisk_t; 649 650 /* 651 * HAMMER UNIX Attribute data 652 * 653 * The data reference in a HAMMER inode record points to this structure. Any 654 * modifications to the contents of this structure will result in a record 655 * replacement operation. 656 * 657 * short_data_off allows a small amount of data to be embedded in the 658 * hammer_inode_data structure. HAMMER typically uses this to represent 659 * up to 64 bytes of data, or to hold symlinks. Remember that allocations 660 * are in powers of 2 so 64, 192, 448, or 960 bytes of embedded data is 661 * support (64+64, 64+192, 64+448 64+960). 662 * 663 * parent_obj_id is only valid for directories (which cannot be hard-linked), 664 * and specifies the parent directory obj_id. This field will also be set 665 * for non-directory inodes as a recovery aid, but can wind up specifying 666 * stale information. However, since object id's are not reused, the worse 667 * that happens is that the recovery code is unable to use it. 668 */ 669 struct hammer_inode_data { 670 u_int16_t version; /* inode data version */ 671 u_int16_t mode; /* basic unix permissions */ 672 u_int32_t uflags; /* chflags */ 673 u_int32_t rmajor; /* used by device nodes */ 674 u_int32_t rminor; /* used by device nodes */ 675 u_int64_t ctime; 676 u_int64_t parent_obj_id;/* parent directory obj_id */ 677 uuid_t uid; 678 uuid_t gid; 679 /* XXX device, softlink extension */ 680 }; 681 682 #define HAMMER_INODE_DATA_VERSION 1 683 684 #define HAMMER_OBJID_ROOT 1 685 686 /* 687 * Rollup various structures embedded as record data 688 */ 689 union hammer_data_ondisk { 690 struct hammer_inode_data inode; 691 }; 692 693 #endif 694