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 1.6 2007/11/19 00:53:40 dillon Exp $ 35 */ 36 37 #ifndef _SYS_UUID_H_ 38 #include <sys/uuid.h> 39 #endif 40 41 /* 42 * The structures below represent the on-disk format for a HAMMER 43 * filesystem. Note that all fields for on-disk structures are naturally 44 * aligned. The host endian format is used - compatibility is possible 45 * if the implementation detects reversed endian and adjusts data accordingly. 46 * 47 * Most of HAMMER revolves around the concept of an object identifier. An 48 * obj_id is a 64 bit quantity which uniquely identifies a filesystem object 49 * FOR THE ENTIRE LIFE OF THE FILESYSTEM. This uniqueness allows backups 50 * and mirrors to retain varying amounts of filesystem history by removing 51 * any possibility of conflict through identifier reuse. 52 * 53 * A HAMMER filesystem may spam multiple volumes. 54 * 55 * A HAMMER filesystem uses a 16K filesystem buffer size. All filesystem 56 * I/O is done in multiples of 16K. Most buffer-sized headers such as those 57 * used by volumes, super-clusters, clusters, and basic filesystem buffers 58 * use fixed-sized A-lists which are heavily dependant on HAMMER_BUFSIZE. 59 */ 60 #define HAMMER_BUFSIZE 16384 61 #define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1) 62 63 /* 64 * Hammer transction ids are 64 bit unsigned integers and are usually 65 * synchronized with the time of day in nanoseconds. 66 */ 67 typedef u_int64_t hammer_tid_t; 68 69 #define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL 70 71 /* 72 * Most HAMMER data structures are embedded in 16K filesystem buffers. 73 * All filesystem buffers except those designated as pure-data buffers 74 * contain this 128-byte header. 75 * 76 * This structure contains an embedded A-List used to manage space within 77 * the filesystem buffer. It is not used by volume or cluster header 78 * buffers, or by pure-data buffers. The granularity is variable and 79 * depends on the type of filesystem buffer. BLKSIZE is just a minimum. 80 */ 81 82 #define HAMMER_FSBUF_HEAD_SIZE 128 83 #define HAMMER_FSBUF_MAXBLKS 256 84 #define HAMMER_FSBUF_BLKMASK (HAMMER_FSBUF_MAXBLKS - 1) 85 #define HAMMER_FSBUF_METAELMS HAMMER_ALIST_METAELMS_256_1LYR /* 11 */ 86 87 struct hammer_fsbuf_head { 88 u_int64_t buf_type; 89 u_int32_t buf_crc; 90 u_int32_t buf_reserved07; 91 u_int32_t reserved[6]; 92 struct hammer_almeta buf_almeta[HAMMER_FSBUF_METAELMS]; 93 }; 94 95 typedef struct hammer_fsbuf_head *hammer_fsbuf_head_t; 96 97 /* 98 * Note: Pure-data buffers contain pure-data and have no buf_type. 99 * Piecemeal data buffers do have a header and use HAMMER_FSBUF_DATA. 100 */ 101 #define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */ 102 #define HAMMER_FSBUF_SUPERCL 0xC8414D52C3555052ULL /* HAMRSUPR */ 103 #define HAMMER_FSBUF_CLUSTER 0xC8414D52C34C5553ULL /* HAMRCLUS */ 104 #define HAMMER_FSBUF_RECORDS 0xC8414D52D2454353ULL /* HAMRRECS */ 105 #define HAMMER_FSBUF_BTREE 0xC8414D52C2545245ULL /* HAMRBTRE */ 106 #define HAMMER_FSBUF_DATA 0xC8414D52C4415441ULL /* HAMRDATA */ 107 108 #define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */ 109 110 /* 111 * The B-Tree structures need hammer_fsbuf_head. 112 */ 113 #include "hammer_btree.h" 114 115 /* 116 * HAMMER Volume header 117 * 118 * A HAMMER filesystem is built from any number of block devices, Each block 119 * device contains a volume header followed by however many super-clusters 120 * and clusters fit into the volume. Clusters cannot be migrated but the 121 * data they contain can, so HAMMER can use a truncated cluster for any 122 * extra space at the end of the volume. 123 * 124 * The volume containing the root cluster is designated as the master volume. 125 * The root cluster designation can be moved to any volume. 126 * 127 * The volume header takes up an entire 16K filesystem buffer and includes 128 * a one or two-layered A-list to manage the clusters making up the volume. 129 * A volume containing up to 32768 clusters (2TB) can be managed with a 130 * single-layered A-list. A two-layer A-list is capable of managing up 131 * to 16384 super-clusters with each super-cluster containing 32768 clusters 132 * (32768 TB per volume total). The number of volumes is limited to 32768 133 * but it only takes 512 to fill out a 64 bit address space so for all 134 * intents and purposes the filesystem has no limits. 135 * 136 * cluster addressing within a volume depends on whether a single or 137 * duel-layer A-list is used. If a duel-layer A-list is used a 16K 138 * super-cluster buffer is needed for every 16384 clusters in the volume. 139 * However, because the A-list's hinting is grouped in multiples of 16 140 * we group 16 super-cluster buffers together (starting just after the 141 * volume header), followed by 16384x16 clusters, and repeat. 142 * 143 * NOTE: A 32768-element single-layer and 16384-element duel-layer A-list 144 * is the same size. 145 */ 146 #define HAMMER_VOL_MAXCLUSTERS 32768 /* 1-layer */ 147 #define HAMMER_VOL_MAXSUPERCLUSTERS 16384 /* 2-layer */ 148 #define HAMMER_VOL_SUPERCLUSTER_GROUP 16 149 #define HAMMER_VOL_METAELMS_1LYR HAMMER_ALIST_METAELMS_32K_1LYR 150 #define HAMMER_VOL_METAELMS_2LYR HAMMER_ALIST_METAELMS_16K_2LYR 151 152 struct hammer_volume_ondisk { 153 struct hammer_fsbuf_head head; 154 int64_t vol_beg; /* byte offset of first cl/supercl in volume */ 155 int64_t vol_end; /* byte offset of volume EOF */ 156 int64_t vol_locked; /* reserved clusters are >= this offset */ 157 158 uuid_t vol_fsid; /* identify filesystem */ 159 uuid_t vol_fstype; /* identify filesystem type */ 160 char vol_name[64]; /* Name of volume */ 161 162 int32_t vol_no; /* volume number within filesystem */ 163 int32_t vol_count; /* number of volumes making up FS */ 164 165 u_int32_t vol_version; /* version control information */ 166 u_int32_t vol_reserved01; 167 u_int32_t vol_flags; /* volume flags */ 168 u_int32_t vol_rootvol; /* which volume is the root volume? */ 169 170 int32_t vol_clsize; /* cluster size (same for all volumes) */ 171 int32_t vol_nclusters; 172 u_int32_t vol_reserved06; 173 u_int32_t vol_reserved07; 174 175 int32_t vol_stat_blocksize; /* for statfs only */ 176 int64_t vol_stat_bytes; /* for statfs only */ 177 int64_t vol_stat_inodes; /* for statfs only */ 178 179 /* 180 * These fields are initialized and space is reserved in every 181 * volume making up a HAMMER filesytem, but only the master volume 182 * contains valid data. 183 */ 184 int32_t vol0_root_clu_no; /* root cluster no (index) in rootvol */ 185 hammer_tid_t vol0_root_clu_id; /* root cluster id */ 186 hammer_tid_t vol0_nexttid; /* next TID */ 187 u_int64_t vol0_recid; /* fs-wide record id allocator */ 188 u_int64_t vol0_synchronized_rec_id; /* XXX */ 189 190 char reserved[1024]; 191 192 /* 193 * Meta elements for the volume header's A-list, which is either a 194 * 1-layer A-list capable of managing 32768 clusters, or a 2-layer 195 * A-list capable of managing 16384 super-clusters (each of which 196 * can handle 32768 clusters). 197 */ 198 union { 199 struct hammer_almeta super[HAMMER_VOL_METAELMS_2LYR]; 200 struct hammer_almeta normal[HAMMER_VOL_METAELMS_1LYR]; 201 } vol_almeta; 202 u_int32_t vol0_bitmap[1024]; 203 }; 204 205 typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t; 206 207 #define HAMMER_VOLF_VALID 0x0001 /* valid entry */ 208 #define HAMMER_VOLF_OPEN 0x0002 /* volume is open */ 209 #define HAMMER_VOLF_USINGSUPERCL 0x0004 /* using superclusters */ 210 211 /* 212 * HAMMER Super-cluster header 213 * 214 * A super-cluster is used to increase the maximum size of a volume. 215 * HAMMER's volume header can manage up to 32768 direct clusters or 216 * 16384 super-clusters. Each super-cluster (which is basically just 217 * a 16K filesystem buffer) can manage up to 32768 clusters. So adding 218 * a super-cluster layer allows a HAMMER volume to be sized upwards of 219 * around 32768TB instead of 2TB. 220 * 221 * Any volume initially formatted to be over 32G reserves space for the layer 222 * but the layer is only enabled if the volume exceeds 2TB. 223 */ 224 #define HAMMER_SUPERCL_METAELMS HAMMER_ALIST_METAELMS_32K_1LYR 225 #define HAMMER_SCL_MAXCLUSTERS HAMMER_VOL_MAXCLUSTERS 226 227 struct hammer_supercl_ondisk { 228 struct hammer_fsbuf_head head; 229 uuid_t vol_fsid; /* identify filesystem - sanity check */ 230 uuid_t vol_fstype; /* identify filesystem type - sanity check */ 231 int32_t reserved[1024]; 232 233 struct hammer_almeta scl_meta[HAMMER_SUPERCL_METAELMS]; 234 }; 235 236 typedef struct hammer_supercl_ondisk *hammer_supercl_ondisk_t; 237 238 /* 239 * HAMMER Cluster header 240 * 241 * A cluster is limited to 64MB and is made up of 4096 16K filesystem 242 * buffers. The cluster header contains four A-lists to manage these 243 * buffers. 244 * 245 * master_alist - This is a non-layered A-list which manages pure-data 246 * allocations and allocations on behalf of other A-lists. 247 * 248 * btree_alist - This is a layered A-list which manages filesystem buffers 249 * containing B-Tree nodes. 250 * 251 * record_alist - This is a layered A-list which manages filesystem buffers 252 * containing records. 253 * 254 * mdata_alist - This is a layered A-list which manages filesystem buffers 255 * containing piecemeal record data. 256 * 257 * General storage management works like this: All the A-lists except the 258 * master start in an all-allocated state. Now lets say you wish to allocate 259 * a B-Tree node out the btree_alist. If the allocation fails you allocate 260 * a pure data block out of master_alist and then free that block in 261 * btree_alist, thereby assigning more space to the btree_alist, and then 262 * retry your allocation out of the btree_alist. In the reverse direction, 263 * filesystem buffers can be garbage collected back to master_alist simply 264 * by doing whole-buffer allocations in btree_alist and then freeing the 265 * space in master_alist. The whole-buffer-allocation approach to garbage 266 * collection works because A-list allocations are always power-of-2 sized 267 * and aligned. 268 */ 269 #define HAMMER_CLU_MAXBUFFERS 4096 270 #define HAMMER_CLU_MASTER_METAELMS HAMMER_ALIST_METAELMS_4K_1LYR 271 #define HAMMER_CLU_SLAVE_METAELMS HAMMER_ALIST_METAELMS_4K_2LYR 272 #define HAMMER_CLU_MAXBYTES (HAMMER_CLU_MAXBUFFERS * HAMMER_BUFSIZE) 273 274 struct hammer_cluster_ondisk { 275 struct hammer_fsbuf_head head; 276 uuid_t vol_fsid; /* identify filesystem - sanity check */ 277 uuid_t vol_fstype; /* identify filesystem type - sanity check */ 278 279 hammer_tid_t clu_id; /* unique cluster self identification */ 280 hammer_tid_t clu_gen; /* generation number */ 281 int32_t vol_no; /* cluster contained in volume (sanity) */ 282 u_int32_t clu_flags; /* cluster flags */ 283 284 int32_t clu_start; /* start of data (byte offset) */ 285 int32_t clu_limit; /* end of data (byte offset) */ 286 int32_t clu_no; /* cluster index in volume (sanity) */ 287 u_int32_t clu_reserved03; 288 289 u_int32_t clu_reserved04; 290 u_int32_t clu_reserved05; 291 u_int32_t clu_reserved06; 292 u_int32_t clu_reserved07; 293 294 int32_t idx_data; /* data append point (element no) */ 295 int32_t idx_index; /* index append point (element no) */ 296 int32_t idx_record; /* record prepend point (element no) */ 297 u_int32_t idx_reserved03; 298 299 /* 300 * Specify the range of information stored in this cluster as two 301 * btree elements. These elements match the left and right 302 * boundary elements in the internal B-Tree node of the parent 303 * cluster that points to the root of our cluster. Because these 304 * are boundary elements, the right boundary is range-NONinclusive. 305 */ 306 struct hammer_base_elm clu_btree_beg; 307 struct hammer_base_elm clu_btree_end; 308 309 /* 310 * The cluster's B-Tree root can change as a side effect of insertion 311 * and deletion operations so store an offset instead of embedding 312 * the root node. The parent_offset is stale if the generation number 313 * does not match. 314 * 315 * Parent linkages are explicit. 316 */ 317 int32_t clu_btree_root; 318 int32_t clu_btree_parent_vol_no; 319 int32_t clu_btree_parent_clu_no; 320 int32_t clu_btree_parent_offset; 321 hammer_tid_t clu_btree_parent_clu_gen; 322 323 u_int64_t synchronized_rec_id; 324 325 struct hammer_almeta clu_master_meta[HAMMER_CLU_MASTER_METAELMS]; 326 struct hammer_almeta clu_btree_meta[HAMMER_CLU_SLAVE_METAELMS]; 327 struct hammer_almeta clu_record_meta[HAMMER_CLU_SLAVE_METAELMS]; 328 struct hammer_almeta clu_mdata_meta[HAMMER_CLU_SLAVE_METAELMS]; 329 }; 330 331 typedef struct hammer_cluster_ondisk *hammer_cluster_ondisk_t; 332 333 /* 334 * HAMMER records are 96 byte entities encoded into 16K filesystem buffers. 335 * Each record has a 64 byte header and a 32 byte extension. 170 records 336 * fit into each buffer. Storage is managed by the buffer's A-List. 337 * 338 * Each record may have an explicit data reference to a block of data up 339 * to 2^31-1 bytes in size within the current cluster. Note that multiple 340 * records may share the same or overlapping data references. 341 */ 342 343 /* 344 * All HAMMER records have a common 64-byte base and a 32-byte extension. 345 * 346 * Many HAMMER record types reference out-of-band data within the cluster. 347 * This data can also be stored in-band in the record itself if it is small 348 * enough. Either way, (data_offset, data_len) points to it. 349 * 350 * Key comparison order: obj_id, rec_type, key, create_tid 351 */ 352 struct hammer_base_record { 353 /* 354 * 40 byte base element info - same base as used in B-Tree internal 355 * and leaf node element arrays. 356 * 357 * Fields: obj_id, key, create_tid, delete_tid, rec_type, obj_type, 358 * reserved07. 359 */ 360 struct hammer_base_elm base; /* 00 base element info */ 361 362 int32_t data_len; /* 28 size of data (remainder zero-fill) */ 363 u_int32_t data_crc; /* 2C data sanity check */ 364 u_int64_t rec_id; /* 30 record id (iterator for recovery) */ 365 int32_t data_offset; /* 38 cluster-relative data reference or 0 */ 366 u_int32_t reserved07; /* 3C */ 367 /* 40 */ 368 }; 369 370 /* 371 * Record types are fairly straightforward. The B-Tree includes the record 372 * type in its index sort. 373 * 374 * In particular please note that it is possible to create a pseudo- 375 * filesystem within a HAMMER filesystem by creating a special object 376 * type within a directory. Pseudo-filesystems are used as replication 377 * targets and even though they are built within a HAMMER filesystem they 378 * get their own obj_id space (and thus can serve as a replication target) 379 * and look like a mount point to the system. 380 * 381 * Inter-cluster records are special-cased in the B-Tree. These records 382 * are referenced from a B-Tree INTERNAL node, NOT A LEAF. This means 383 * that the element in the B-Tree node is actually a boundary element whos 384 * base element fields, including rec_type, reflect the boundary, NOT 385 * the inter-cluster record type. 386 * 387 * HAMMER_RECTYPE_CLUSTER - only set in the actual inter-cluster record, 388 * not set in the left or right boundary elements around the inter-cluster 389 * reference of an internal node in the B-Tree (because doing so would 390 * interfere with the boundary tests). 391 */ 392 #define HAMMER_RECTYPE_UNKNOWN 0 393 #define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */ 394 #define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */ 395 #define HAMMER_RECTYPE_PSEUDO_INODE 2 /* pseudo filesysem */ 396 #define HAMMER_RECTYPE_CLUSTER 3 /* inter-cluster reference */ 397 #define HAMMER_RECTYPE_DATA 0x10 398 #define HAMMER_RECTYPE_DIRENTRY 0x11 399 #define HAMMER_RECTYPE_DB 0x12 400 #define HAMMER_RECTYPE_EXT 0x13 /* ext attributes */ 401 402 #define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */ 403 #define HAMMER_OBJTYPE_DIRECTORY 1 404 #define HAMMER_OBJTYPE_REGFILE 2 405 #define HAMMER_OBJTYPE_DBFILE 3 406 #define HAMMER_OBJTYPE_FIFO 4 407 #define HAMMER_OBJTYPE_CDEV 5 408 #define HAMMER_OBJTYPE_BDEV 6 409 #define HAMMER_OBJTYPE_SOFTLINK 7 410 #define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */ 411 412 /* 413 * Generic full-sized record 414 */ 415 struct hammer_generic_record { 416 struct hammer_base_record base; 417 char filler[32]; 418 }; 419 420 /* 421 * A HAMMER inode record. 422 * 423 * This forms the basis for a filesystem object. obj_id is the inode number, 424 * key1 represents the pseudo filesystem id for security partitioning 425 * (preventing cross-links and/or restricting a NFS export and specifying the 426 * security policy), and key2 represents the data retention policy id. 427 * 428 * Inode numbers are 64 bit quantities which uniquely identify a filesystem 429 * object for the ENTIRE life of the filesystem, even after the object has 430 * been deleted. For all intents and purposes inode numbers are simply 431 * allocated by incrementing a sequence space. 432 * 433 * There is an important distinction between the data stored in the inode 434 * record and the record's data reference. The record references a 435 * hammer_inode_data structure but the filesystem object size and hard link 436 * count is stored in the inode record itself. This allows multiple inodes 437 * to share the same hammer_inode_data structure. This is possible because 438 * any modifications will lay out new data. The HAMMER implementation need 439 * not use the data-sharing ability when laying down new records. 440 * 441 * A HAMMER inode is subject to the same historical storage requirements 442 * as any other record. In particular any change in filesystem or hard link 443 * count will lay down a new inode record when the filesystem is synced to 444 * disk. This can lead to a lot of junk records which get cleaned up by 445 * the data retention policy. 446 * 447 * The ino_atime and ino_mtime fields are a special case. Modifications to 448 * these fields do NOT lay down a new record by default, though the values 449 * are effectively frozen for snapshots which access historical versions 450 * of the inode record due to other operations. This means that atime will 451 * not necessarily be accurate in snapshots, backups, or mirrors. mtime 452 * will be accurate in backups and mirrors since it can be regenerated from 453 * the mirroring stream. 454 * 455 * Because nlinks is historically retained the hardlink count will be 456 * accurate when accessing a HAMMER filesystem snapshot. 457 */ 458 struct hammer_inode_record { 459 struct hammer_base_record base; 460 u_int64_t ino_atime; /* last access time (not historical) */ 461 u_int64_t ino_mtime; /* last modified time (not historical) */ 462 u_int64_t ino_size; /* filesystem object size */ 463 u_int64_t ino_nlinks; /* hard links */ 464 }; 465 466 /* 467 * Data records specify the entire contents of a regular file object, 468 * including attributes. Small amounts of data can theoretically be 469 * embedded in the record itself but the use of this ability verses using 470 * an out-of-band data reference depends on the implementation. 471 */ 472 struct hammer_data_record { 473 struct hammer_base_record base; 474 char filler[32]; 475 }; 476 477 /* 478 * A directory entry specifies the HAMMER filesystem object id, a copy of 479 * the file type, and file name (either embedded or as out-of-band data). 480 * If the file name is short enough to fit into den_name[] (including a 481 * terminating nul) then it will be embedded in the record, otherwise it 482 * is stored out-of-band. The base record's data reference always points 483 * to the nul-terminated filename regardless. 484 * 485 * Directory entries are indexed with a 128 bit namekey rather then an 486 * offset. A portion of the namekey is an iterator or randomizer to deal 487 * with collisions. 488 * 489 * Note that base.base.obj_type holds the filesystem object type of obj_id, 490 * e.g. a den_type equivalent. 491 * 492 */ 493 struct hammer_entry_record { 494 struct hammer_base_record base; 495 u_int64_t obj_id; /* object being referenced */ 496 u_int64_t reserved01; 497 char den_name[16]; /* short file names fit in record */ 498 }; 499 500 /* 501 * Hammer rollup record 502 */ 503 union hammer_record_ondisk { 504 struct hammer_base_record base; 505 struct hammer_generic_record generic; 506 struct hammer_inode_record inode; 507 struct hammer_data_record data; 508 struct hammer_entry_record entry; 509 }; 510 511 typedef union hammer_record_ondisk *hammer_record_ondisk_t; 512 513 /* 514 * Filesystem buffer for records 515 */ 516 #define HAMMER_RECORD_NODES \ 517 ((HAMMER_BUFSIZE - sizeof(struct hammer_fsbuf_head)) / \ 518 sizeof(union hammer_record_ondisk)) 519 520 struct hammer_fsbuf_recs { 521 struct hammer_fsbuf_head head; 522 char unused[32]; 523 union hammer_record_ondisk recs[HAMMER_RECORD_NODES]; 524 }; 525 526 /* 527 * Filesystem buffer for piecemeal data. Note that this does not apply 528 * to dedicated pure-data buffers as such buffers do not have a header. 529 */ 530 531 #define HAMMER_DATA_SIZE (HAMMER_BUFSIZE - sizeof(struct hammer_fsbuf_head)) 532 #define HAMMER_DATA_BLKSIZE 64 533 #define HAMMER_DATA_BLKMASK (HAMMER_DATA_BLKSIZE-1) 534 #define HAMMER_DATA_NODES (HAMMER_DATA_SIZE / HAMMER_DATA_BLKSIZE) 535 536 struct hammer_fsbuf_data { 537 struct hammer_fsbuf_head head; 538 u_int8_t data[HAMMER_DATA_NODES][HAMMER_DATA_BLKSIZE]; 539 }; 540 541 /* 542 * Filesystem buffer rollup 543 */ 544 union hammer_fsbuf_ondisk { 545 struct hammer_fsbuf_head head; 546 struct hammer_fsbuf_btree btree; 547 struct hammer_fsbuf_recs record; 548 struct hammer_fsbuf_data data; 549 }; 550 551 typedef union hammer_fsbuf_ondisk *hammer_fsbuf_ondisk_t; 552 553 /* 554 * HAMMER UNIX Attribute data 555 * 556 * The data reference in a HAMMER inode record points to this structure. Any 557 * modifications to the contents of this structure will result in a record 558 * replacement operation. 559 * 560 * state_sum allows a filesystem object to be validated to a degree by 561 * generating a checksum of all of its pieces (in no particular order) and 562 * checking it against this field. 563 * 564 * short_data_off allows a small amount of data to be embedded in the 565 * hammer_inode_data structure. HAMMER typically uses this to represent 566 * up to 64 bytes of data, or to hold symlinks. Remember that allocations 567 * are in powers of 2 so 64, 192, 448, or 960 bytes of embedded data is 568 * support (64+64, 64+192, 64+448 64+960). 569 * 570 * parent_obj_id is only valid for directories (which cannot be hard-linked), 571 * and specifies the parent directory obj_id. This field will also be set 572 * for non-directory inodes as a recovery aid, but can wind up specifying 573 * stale information. However, since object id's are not reused, the worse 574 * that happens is that the recovery code is unable to use it. 575 */ 576 struct hammer_inode_data { 577 u_int16_t version; /* inode data version */ 578 u_int16_t mode; /* basic unix permissions */ 579 u_int32_t uflags; /* chflags */ 580 u_int16_t short_data_off; /* degenerate data case */ 581 u_int16_t short_data_len; 582 u_int32_t state_sum; 583 u_int64_t ctime; 584 u_int64_t parent_obj_id;/* parent directory obj_id */ 585 uuid_t uid; 586 uuid_t gid; 587 /* XXX device, softlink extension */ 588 }; 589 590 #define HAMMER_INODE_DATA_VERSION 1 591 592 /* 593 * Rollup various structures embedded as record data 594 */ 595 union hammer_data_ondisk { 596 struct hammer_inode_data inode; 597 }; 598 599