1 /* 2 * Copyright (c) 2011-2012 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Matthew Dillon <dillon@dragonflybsd.org> 6 * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org> 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in 16 * the documentation and/or other materials provided with the 17 * distribution. 18 * 3. Neither the name of The DragonFly Project nor the names of its 19 * contributors may be used to endorse or promote products derived 20 * from this software without specific, prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 23 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 24 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 25 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 26 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 27 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 28 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 29 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 30 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 31 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 32 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 33 * SUCH DAMAGE. 34 */ 35 #ifndef VFS_HAMMER2_DISK_H_ 36 #define VFS_HAMMER2_DISK_H_ 37 38 #ifndef _SYS_UUID_H_ 39 #include <sys/uuid.h> 40 #endif 41 #ifndef _SYS_DMSG_H_ 42 #include <sys/dmsg.h> 43 #endif 44 45 /* 46 * The structures below represent the on-disk media structures for the HAMMER2 47 * filesystem. Note that all fields for on-disk structures are naturally 48 * aligned. The host endian format is typically used - compatibility is 49 * possible if the implementation detects reversed endian and adjusts accesses 50 * accordingly. 51 * 52 * HAMMER2 primarily revolves around the directory topology: inodes, 53 * directory entries, and block tables. Block device buffer cache buffers 54 * are always 64KB. Logical file buffers are typically 16KB. All data 55 * references utilize 64-bit byte offsets. 56 * 57 * Free block management is handled independently using blocks reserved by 58 * the media topology. 59 */ 60 61 /* 62 * The data at the end of a file or directory may be a fragment in order 63 * to optimize storage efficiency. The minimum fragment size is 64 bytes. 64 * Since allocations are in powers of 2 fragments must also be sized in 65 * powers of 2 (64, 128, 256, ... 65536). 66 * 67 * For the moment the maximum allocation size is HAMMER2_PBUFSIZE (64K), 68 * which is 2^16. Larger extents may be supported in the future. 69 * 70 * A full indirect block uses supports 1024 x 64-byte blockrefs. 71 * 72 * A maximally sized file (2^64-1 bytes) requires 5 indirect block levels. 73 * The hammer2_blockset in the volume header or file inode has another 8 74 * entries, giving us 66+3 = 69 bits of address space. However, some bits 75 * are taken up by (potentially) requests for redundant copies. HAMMER2 76 * currently supports up to 8 copies, which brings the address space down 77 * to 66 bits and gives us 2 bits of leeway. 78 */ 79 #define HAMMER2_MIN_ALLOC 64 /* minimum allocation size */ 80 #define HAMMER2_MIN_RADIX 6 /* minimum allocation size 2^N */ 81 #define HAMMER2_MAX_RADIX 16 /* maximum allocation size 2^N */ 82 #define HAMMER2_KEY_RADIX 64 /* number of bits in key */ 83 84 /* 85 * MINALLOCSIZE - The minimum allocation size. This can be smaller 86 * or larger than the minimum physical IO size. 87 * 88 * NOTE: Should not be larger than 1K since inodes 89 * are 1K. 90 * 91 * MINIOSIZE - The minimum IO size. This must be less than 92 * or equal to HAMMER2_PBUFSIZE. 93 * 94 * XXX currently must be set to MINALLOCSIZE until/if 95 * we deal with recursive buffer cache locks. 96 * 97 * HAMMER2_PBUFSIZE - Topological block size used by files for all 98 * blocks except the block straddling EOF. 99 * 100 * HAMMER2_SEGSIZE - Allocation map segment size, typically 2MB 101 */ 102 103 #define HAMMER2_SEGSIZE (65536 * 8) 104 105 #define HAMMER2_PBUFRADIX 16 /* physical buf (1<<16) bytes */ 106 #define HAMMER2_PBUFSIZE 65536 107 #define HAMMER2_LBUFRADIX 14 /* logical buf (1<<14) bytes */ 108 #define HAMMER2_LBUFSIZE 16384 109 110 #if 0 111 #define HAMMER2_MINIORADIX 16 /* minimum phsical IO size */ 112 #define HAMMER2_MINIOSIZE 65536 113 #endif 114 #define HAMMER2_MINIORADIX HAMMER2_MINALLOCRADIX 115 #define HAMMER2_MINIOSIZE HAMMER2_MINALLOCSIZE 116 117 #define HAMMER2_MINALLOCRADIX 10 /* minimum block allocation size */ 118 #define HAMMER2_MINALLOCSIZE 1024 119 #define HAMMER2_IND_BYTES_MIN 4096 /* first indirect layer only */ 120 #define HAMMER2_IND_BYTES_MAX HAMMER2_PBUFSIZE 121 #define HAMMER2_IND_COUNT_MIN (HAMMER2_IND_BYTES_MIN / \ 122 sizeof(hammer2_blockref_t)) 123 #define HAMMER2_IND_COUNT_MAX (HAMMER2_IND_BYTES_MAX / \ 124 sizeof(hammer2_blockref_t)) 125 126 /* 127 * HAMMER2 processes blockrefs in sets of 8. The set is fully associative, 128 * is not sorted, and may contain holes. 129 * 130 * A full indirect block supports 1024 blockrefs. 131 * 132 * An inode embeds one set of blockrefs but may also use the data area for 133 * up to 512 bytes of direct data. 134 */ 135 #define HAMMER2_SET_COUNT 8 /* direct entries & associativity */ 136 #define HAMMER2_SET_RADIX 3 137 #define HAMMER2_EMBEDDED_BYTES 512 138 #define HAMMER2_EMBEDDED_RADIX 9 139 140 #define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1) 141 #define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1) 142 #define HAMMER2_SEGMASK (HAMMER2_SEGSIZE - 1) 143 144 #define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK) 145 #define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE) 146 #define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK) 147 #define HAMMER2_SEGSIZE64 ((hammer2_off_t)HAMMER2_SEGSIZE) 148 #define HAMMER2_SEGMASK64 ((hammer2_off_t)HAMMER2_SEGMASK) 149 150 #define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5" 151 152 /* 153 * A HAMMER2 filesystem is always sized in multiples of 8MB. 154 * 155 * A 4MB segment is reserved at the beginning of each 2GB zone. This segment 156 * contains the volume header, the free block table, and possibly other 157 * information in the future. 4MB = 64 x 64K blocks. 158 */ 159 #define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024) 160 #define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 161 #define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 162 #define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK) 163 164 #define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN) 165 #define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 166 #define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 167 #define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK) 168 169 #define HAMMER2_ZONE_BYTES64 (2LLU * 1024 * 1024 * 1024) 170 #define HAMMER2_ZONE_MASK64 (HAMMER2_ZONE_BYTES64 - 1) 171 #define HAMMER2_ZONE_SEG (4 * 1024 * 1024) 172 #define HAMMER2_ZONE_SEG64 ((hammer2_off_t)HAMMER2_ZONE_SEG) 173 #define HAMMER2_ZONE_BLOCKS_SEG (HAMMER2_ZONE_SEG / HAMMER2_PBUFSIZE) 174 175 /* 176 * Two linear areas can be reserved after the initial 2MB segment in the base 177 * zone (the one starting at offset 0). These areas are NOT managed by the 178 * block allocator and do not fall under HAMMER2 crc checking rules based 179 * at the volume header (but can be self-CRCd internally, depending). 180 */ 181 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN 182 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024) 183 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024) 184 185 #define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN 186 #define HAMMER2_REDO_NOM_BYTES (256*1024*1024) 187 #define HAMMER2_REDO_MAX_BYTES (1024*1024*1024) 188 189 /* 190 * Most HAMMER2 types are implemented as unsigned 64-bit integers. 191 * Transaction ids are monotonic. 192 * 193 * We utilize 32-bit iSCSI CRCs. 194 */ 195 typedef uint64_t hammer2_tid_t; 196 typedef uint64_t hammer2_off_t; 197 typedef uint64_t hammer2_key_t; 198 typedef uint32_t hammer2_crc32_t; 199 200 /* 201 * Miscellanious ranges (all are unsigned). 202 */ 203 #define HAMMER2_MIN_TID 1ULL 204 #define HAMMER2_MAX_TID 0xFFFFFFFFFFFFFFFFULL 205 #define HAMMER2_MIN_KEY 0ULL 206 #define HAMMER2_MAX_KEY 0xFFFFFFFFFFFFFFFFULL 207 #define HAMMER2_MIN_OFFSET 0ULL 208 #define HAMMER2_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL 209 210 /* 211 * HAMMER2 data offset special cases and masking. 212 * 213 * All HAMMER2 data offsets have to be broken down into a 64K buffer base 214 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO). 215 * 216 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits 217 * of the data offset field specifies how large the data chunk being pointed 218 * to as a power of 2. This value typically ranges from HAMMER2_MIN_RADIX 219 * to HAMMER2_MAX_RADIX (6-16). Larger values may be supported in the future 220 * to support file extents. 221 */ 222 #define HAMMER2_OFF_BAD ((hammer2_off_t)-1) 223 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL 224 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64) 225 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64) 226 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL 227 #define HAMMER2_MAX_COPIES 6 228 229 /* 230 * HAMMER2 directory support and pre-defined keys 231 */ 232 #define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL 233 #define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL 234 #define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL 235 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL 236 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */ 237 238 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */ 239 240 /* 241 * The media block reference structure. This forms the core of the HAMMER2 242 * media topology recursion. This 64-byte data structure is embedded in the 243 * volume header, in inodes (which are also directory entries), and in 244 * indirect blocks. 245 * 246 * A blockref references a single media item, which typically can be a 247 * directory entry (aka inode), indirect block, or data block. 248 * 249 * The primary feature a blockref represents is the ability to validate 250 * the entire tree underneath it via its check code. Any modification to 251 * anything propagates up the blockref tree all the way to the root, replacing 252 * the related blocks. Propagations can shortcut to the volume root to 253 * implement the 'fast syncing' feature but this only delays the eventual 254 * propagation. 255 * 256 * The check code can be a simple 32-bit iscsi code, a 64-bit crc, 257 * or as complex as a 192 bit cryptographic hash. 192 bits is the maximum 258 * supported check code size, which is not sufficient for unverified dedup 259 * UNLESS one doesn't mind once-in-a-blue-moon data corruption (such as when 260 * farming web data). HAMMER2 has an unverified dedup feature for just this 261 * purpose. 262 */ 263 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */ 264 uint8_t type; /* type of underlying item */ 265 uint8_t methods; /* check method & compression method */ 266 uint8_t copyid; /* specify which copy this is */ 267 uint8_t keybits; /* #of keybits masked off 0=leaf */ 268 uint8_t vradix; /* virtual data/meta-data size */ 269 uint8_t flags; /* blockref flags */ 270 uint8_t reserved06; 271 uint8_t reserved07; 272 hammer2_key_t key; /* key specification */ 273 hammer2_tid_t mirror_tid; /* propagate for mirror scan */ 274 hammer2_tid_t modify_tid; /* modifications sans propagation */ 275 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/ 276 union { /* check info */ 277 char buf[24]; 278 struct { 279 uint32_t value; 280 uint32_t unused[5]; 281 } iscsi32; 282 struct { 283 uint64_t value; 284 uint64_t unused[2]; 285 } crc64; 286 struct { 287 char data[24]; 288 } sha192; 289 } check; 290 }; 291 292 typedef struct hammer2_blockref hammer2_blockref_t; 293 294 #define HAMMER2_BREF_SYNC1 0x01 /* modification synchronized */ 295 #define HAMMER2_BREF_SYNC2 0x02 /* modification committed */ 296 #define HAMMER2_BREF_DESYNCCHLD 0x04 /* desynchronize children */ 297 #define HAMMER2_BREF_DELETED 0x80 /* indicates a deletion */ 298 299 #define HAMMER2_BLOCKREF_BYTES 64 /* blockref struct in bytes */ 300 301 #define HAMMER2_BREF_TYPE_EMPTY 0 302 #define HAMMER2_BREF_TYPE_INODE 1 303 #define HAMMER2_BREF_TYPE_INDIRECT 2 304 #define HAMMER2_BREF_TYPE_DATA 3 305 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */ 306 307 #define HAMMER2_ENC_COMPMETHOD(n) (n) 308 #define HAMMER2_ENC_CHECKMETHOD(n) ((n) << 4) 309 #define HAMMER2_DEC_COMPMETHOD(n) ((n) & 15) 310 #define HAMMER2_DEC_CHECKMETHOD(n) (((n) >> 4) & 15) 311 312 /* 313 * HAMMER2 block references are collected into sets of 8 blockrefs. These 314 * sets are fully associative, meaning the elements making up a set are 315 * not sorted in any way and may contain duplicate entries, holes, or 316 * entries which shortcut multiple levels of indirection. Sets are used 317 * in various ways: 318 * 319 * (1) When redundancy is desired a set may contain several duplicate 320 * entries pointing to different copies of the same data. Up to 8 copies 321 * are supported but the set structure becomes a bit inefficient once 322 * you go over 4. 323 * 324 * (2) The blockrefs in a set can shortcut multiple levels of indirections 325 * within the bounds imposed by the parent of set. 326 * 327 * When a set fills up another level of indirection is inserted, moving 328 * some or all of the set's contents into indirect blocks placed under the 329 * set. This is a top-down approach in that indirect blocks are not created 330 * until the set actually becomes full (that is, the entries in the set can 331 * shortcut the indirect blocks when the set is not full). Depending on how 332 * things are filled multiple indirect blocks will eventually be created. 333 */ 334 struct hammer2_blockset { 335 hammer2_blockref_t blockref[HAMMER2_SET_COUNT]; 336 }; 337 338 typedef struct hammer2_blockset hammer2_blockset_t; 339 340 /* 341 * Catch programmer snafus 342 */ 343 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT 344 #error "hammer2 direct radix is incorrect" 345 #endif 346 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE 347 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent" 348 #endif 349 #if (1 << HAMMER2_MIN_RADIX) != HAMMER2_MIN_ALLOC 350 #error "HAMMER2_MIN_RADIX and HAMMER2_MIN_ALLOC are inconsistent" 351 #endif 352 353 /* 354 * The media indirect block structure. 355 */ 356 struct hammer2_indblock_data { 357 hammer2_blockref_t blockref[HAMMER2_IND_COUNT_MAX]; 358 }; 359 360 typedef struct hammer2_indblock_data hammer2_indblock_data_t; 361 362 /* 363 * In HAMMER2 inodes ARE directory entries, with a special exception for 364 * hardlinks. The inode number is stored in the inode rather than being 365 * based on the location of the inode (since the location moves every time 366 * the inode or anything underneath the inode is modified). 367 * 368 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes 369 * for the filename, and 512 bytes worth of direct file data OR an embedded 370 * blockset. 371 * 372 * Directories represent one inode per blockref. Inodes are not laid out 373 * as a file but instead are represented by the related blockrefs. The 374 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember 375 * that blocksets are fully associative, so a certain degree efficiency is 376 * achieved just from that. 377 * 378 * Up to 512 bytes of direct data can be embedded in an inode, and since 379 * inodes are essentially directory entries this also means that small data 380 * files end up simply being laid out linearly in the directory, resulting 381 * in fewer seeks and highly optimal access. 382 * 383 * The compression mode can be changed at any time in the inode and is 384 * recorded on a blockref-by-blockref basis. 385 * 386 * Hardlinks are supported via the inode map. Essentially the way a hardlink 387 * works is that all individual directory entries representing the same file 388 * are special cased and specify the same inode number. The actual file 389 * is placed in the nearest parent directory that is parent to all instances 390 * of the hardlink. If all hardlinks to a file are in the same directory 391 * the actual file will also be placed in that directory. This file uses 392 * the inode number as the directory entry key and is invisible to normal 393 * directory scans. Real directory entry keys are differentiated from the 394 * inode number key via bit 63. Access to the hardlink silently looks up 395 * the real file and forwards all operations to that file. Removal of the 396 * last hardlink also removes the real file. 397 * 398 * (attr_tid) is only updated when the inode's specific attributes or regular 399 * file size has changed, and affects path lookups and stat. (attr_tid) 400 * represents a special cache coherency lock under the inode. The inode 401 * blockref's modify_tid will always cover it. 402 * 403 * (dirent_tid) is only updated when an entry under a directory inode has 404 * been created, deleted, renamed, or had its attributes change, and affects 405 * directory lookups and scans. (dirent_tid) represents another special cache 406 * coherency lock under the inode. The inode blockref's modify_tid will 407 * always cover it. 408 */ 409 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */ 410 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */ 411 #define HAMMER2_INODE_VERSION_ONE 1 412 413 struct hammer2_inode_data { 414 uint16_t version; /* 0000 inode data version */ 415 uint16_t reserved02; /* 0002 */ 416 417 /* 418 * core inode attributes, inode type, misc flags 419 */ 420 uint32_t uflags; /* 0004 chflags */ 421 uint32_t rmajor; /* 0008 available for device nodes */ 422 uint32_t rminor; /* 000C available for device nodes */ 423 uint64_t ctime; /* 0010 inode change time */ 424 uint64_t mtime; /* 0018 modified time */ 425 uint64_t atime; /* 0020 access time (unsupported) */ 426 uint64_t btime; /* 0028 birth time */ 427 uuid_t uid; /* 0030 uid / degenerate unix uid */ 428 uuid_t gid; /* 0040 gid / degenerate unix gid */ 429 430 uint8_t type; /* 0050 object type */ 431 uint8_t op_flags; /* 0051 operational flags */ 432 uint16_t cap_flags; /* 0052 capability flags */ 433 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */ 434 435 /* 436 * inode size, identification, localized recursive configuration 437 * for compression and backup copies. 438 */ 439 hammer2_tid_t inum; /* 0058 inode number */ 440 hammer2_off_t size; /* 0060 size of file */ 441 uint64_t nlinks; /* 0068 hard links (typ only dirs) */ 442 hammer2_tid_t iparent; /* 0070 parent inum (recovery only) */ 443 hammer2_key_t name_key; /* 0078 full filename key */ 444 uint16_t name_len; /* 0080 filename length */ 445 uint8_t ncopies; /* 0082 ncopies to local media */ 446 uint8_t comp_algo; /* 0083 compression request & algo */ 447 448 /* 449 * These fields are currently only applicable to PFSROOTs. 450 * 451 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely 452 * identify an instance of a PFS in the cluster because 453 * a mount may contain more than one copy of the PFS as 454 * a separate node. {pfs_clid, pfs_fsid} must be used for 455 * registration in the cluster. 456 */ 457 uint8_t target_type; /* 0084 hardlink target type */ 458 uint8_t reserved85; /* 0085 */ 459 uint8_t reserved86; /* 0086 */ 460 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */ 461 uint64_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */ 462 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */ 463 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */ 464 465 /* 466 * Quotas and cumulative sub-tree counters. 467 */ 468 hammer2_off_t data_quota; /* 00B0 subtree quota in bytes */ 469 hammer2_off_t data_count; /* 00B8 subtree byte count */ 470 hammer2_off_t inode_quota; /* 00C0 subtree quota inode count */ 471 hammer2_off_t inode_count; /* 00C8 subtree inode count */ 472 hammer2_tid_t attr_tid; /* 00D0 attributes changed */ 473 hammer2_tid_t dirent_tid; /* 00D8 directory/attr changed */ 474 uint64_t reservedE0; /* 00E0 */ 475 uint64_t reservedE8; /* 00E8 */ 476 uint64_t reservedF0; /* 00F0 */ 477 uint64_t reservedF8; /* 00F8 */ 478 479 unsigned char filename[HAMMER2_INODE_MAXNAME]; 480 /* 0100-01FF (256 char, unterminated) */ 481 union { /* 0200-03FF (64x8 = 512 bytes) */ 482 struct hammer2_blockset blockset; 483 char data[HAMMER2_EMBEDDED_BYTES]; 484 } u; 485 }; 486 487 typedef struct hammer2_inode_data hammer2_inode_data_t; 488 489 #define HAMMER2_OPFLAG_DIRECTDATA 0x01 490 #define HAMMER2_OPFLAG_PFSROOT 0x02 491 #define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */ 492 493 #define HAMMER2_OBJTYPE_UNKNOWN 0 494 #define HAMMER2_OBJTYPE_DIRECTORY 1 495 #define HAMMER2_OBJTYPE_REGFILE 2 496 #define HAMMER2_OBJTYPE_FIFO 4 497 #define HAMMER2_OBJTYPE_CDEV 5 498 #define HAMMER2_OBJTYPE_BDEV 6 499 #define HAMMER2_OBJTYPE_SOFTLINK 7 500 #define HAMMER2_OBJTYPE_HARDLINK 8 /* dummy entry for hardlink */ 501 #define HAMMER2_OBJTYPE_SOCKET 9 502 #define HAMMER2_OBJTYPE_WHITEOUT 10 503 504 #define HAMMER2_COPYID_NONE 0 505 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1) 506 507 #define HAMMER2_COMP_NONE 0 508 #define HAMMER2_COMP_AUTOZERO 1 509 510 #define HAMMER2_CHECK_NONE 0 511 #define HAMMER2_CHECK_ICRC 1 512 513 /* 514 * PEER types identify connections and help cluster controller filter 515 * out unwanted SPANs. 516 */ 517 #define HAMMER2_PEER_NONE DMSG_PEER_NONE 518 #define HAMMER2_PEER_CLUSTER DMSG_PEER_CLUSTER 519 #define HAMMER2_PEER_BLOCK DMSG_PEER_BLOCK 520 #define HAMMER2_PEER_HAMMER2 DMSG_PEER_HAMMER2 521 522 #define HAMMER2_COPYID_COUNT DMSG_COPYID_COUNT 523 524 /* 525 * PFS types identify a PFS on media and in LNK_SPAN messages. 526 */ 527 #define HAMMER2_PFSTYPE_NONE DMSG_PFSTYPE_NONE 528 #define HAMMER2_PFSTYPE_ADMIN DMSG_PFSTYPE_ADMIN 529 #define HAMMER2_PFSTYPE_CLIENT DMSG_PFSTYPE_CLIENT 530 #define HAMMER2_PFSTYPE_CACHE DMSG_PFSTYPE_CACHE 531 #define HAMMER2_PFSTYPE_COPY DMSG_PFSTYPE_COPY 532 #define HAMMER2_PFSTYPE_SLAVE DMSG_PFSTYPE_SLAVE 533 #define HAMMER2_PFSTYPE_SOFT_SLAVE DMSG_PFSTYPE_SOFT_SLAVE 534 #define HAMMER2_PFSTYPE_SOFT_MASTER DMSG_PFSTYPE_SOFT_MASTER 535 #define HAMMER2_PFSTYPE_MASTER DMSG_PFSTYPE_MASTER 536 #define HAMMER2_PFSTYPE_MAX DMSG_PFSTYPE_MAX 537 538 /* 539 * The allocref structure represents the allocation table. One 64K block 540 * is broken down into 4096 x 16 byte entries. Each indirect block chops 541 * 11 bits off the 64-bit storage space, with leaf entries representing 542 * 64KB blocks. So: (12, 12, 12, 12, 16) = 64 bit storage space. 543 * 544 * Each 64K freemap block breaks the 4096 entries into a 64x64 tree with 545 * big_hint1 representing the top level every 64th entry and big_hint2 546 * representing the lower level in each entry. These fields specify the 547 * largest contiguous radix (1-63) available for allocation in the related 548 * sub-tree. The largest contiguous radix available for the entire block 549 * is saved in the parent (for the root this will be alloc_blockref in the 550 * volume header). The hints may be larger than actual and will be corrected 551 * on the fly but must not be smaller. The allocator uses the hints to 552 * very quickly locate nearby blocks of the desired size. 553 * 554 * In indirect blocks the 64-bit free[_or_mask] field stores the total free 555 * space for each of the 4096 sub-nodes in bytes. The total free space 556 * represented by the indirect block is stored in its parent. 557 * 558 * Each leaf element represents a 64K block. A bitmap replaces the free space 559 * count, giving us a 1KB allocation resolution. A micro-allocation append 560 * offset replaces the icrc field. The micro-allocation feature is not 561 * currently implemented and the field will be set to 65536. 562 * 563 * The allocation map uses reserved blocks so no data block reference is 564 * required, only a bit in the flags field to specify which of two possible 565 * reserved blocks to use. This allows the allocation map to be flushed to 566 * disk with minimal synchronization. 567 */ 568 struct hammer2_allocref { 569 uint32_t icrc_or_app; /* node: icrc, leaf: append offset */ 570 uint16_t flags; 571 uint8_t big_hint1; /* upper level hint */ 572 uint8_t big_hint2; /* lower level hint */ 573 uint64_t free_or_mask; /* node: free bytes, leaf: bitmask */ 574 }; 575 576 typedef struct hammer2_allocref hammer2_allocref_t; 577 578 /* 579 * WARNING - allocref size x entries must equate to the hammer buffer size, 580 * and 12 bits per recursion is assumed by the allocator. 581 * 582 * ALTA-D Since no data_offset is specified flags are needed to select 583 * which sub-block to recurse down into for root & internal nodes. 584 * (only ALTA and ALTB is currently supported). 585 * 586 * LEAF Terminal entry, always set for leafs. May be used to support 587 * 4MB extent allocations and early termination in the future. 588 * (not required to shortcut allocation scans as the big_hint1/2 589 * fields are used for this). 590 */ 591 #define HAMMER2_ALLOCREF_BYTES 16 /* structure size */ 592 #define HAMMER2_ALLOCREF_ENTRIES 4096 /* entries */ 593 #define HAMMER2_ALLOCREF_RADIX 12 /* log2(entries) */ 594 595 #if (HAMMER2_ALLOCREF_BYTES * HAMMER2_ALLOCREF_ENTRIES) != HAMMER2_PBUFSIZE 596 #error "allocref parameters do not fit in hammer buffer" 597 #endif 598 #if (1 << HAMMER2_ALLOCREF_RADIX) != HAMMER2_ALLOCREF_ENTRIES 599 #error "allocref parameters are inconsistent" 600 #endif 601 602 #define HAMMER2_ALLOCREF_ALTMASK 0x0003 /* select block for recurse */ 603 #define HAMMER2_ALLOCREF_ALTA 0x0000 604 #define HAMMER2_ALLOCREF_ALTB 0x0001 605 #define HAMMER2_ALLOCREF_ALTC 0x0002 /* unsupported */ 606 #define HAMMER2_ALLOCREF_ALTD 0x0003 /* unsupported */ 607 #define HAMMER2_ALLOCREF_LEAF 0x0004 608 609 /* 610 * The volume header eats a 64K block. There is currently an issue where 611 * we want to try to fit all nominal filesystem updates in a 512-byte section 612 * but it may be a lost cause due to the need for a blockset. 613 * 614 * All information is stored in host byte order. The volume header's magic 615 * number may be checked to determine the byte order. If you wish to mount 616 * between machines w/ different endian modes you'll need filesystem code 617 * which acts on the media data consistently (either all one way or all the 618 * other). Our code currently does not do that. 619 * 620 * A read-write mount may have to recover missing allocations by doing an 621 * incremental mirror scan looking for modifications made after alloc_tid. 622 * If alloc_tid == last_tid then no recovery operation is needed. Recovery 623 * operations are usually very, very fast. 624 * 625 * Read-only mounts do not need to do any recovery, access to the filesystem 626 * topology is always consistent after a crash (is always consistent, period). 627 * However, there may be shortcutted blockref updates present from deep in 628 * the tree which are stored in the volumeh eader and must be tracked on 629 * the fly. 630 * 631 * NOTE: The copyinfo[] array contains the configuration for both the 632 * cluster connections and any local media copies. The volume 633 * header will be replicated for each local media copy. 634 * 635 * The mount command may specify multiple medias or just one and 636 * allow HAMMER2 to pick up the others when it checks the copyinfo[] 637 * array on mount. 638 * 639 * NOTE: root_blockref points to the super-root directory, not the root 640 * directory. The root directory will be a subdirectory under the 641 * super-root. 642 * 643 * The super-root directory contains all root directories and all 644 * snapshots (readonly or writable). It is possible to do a 645 * null-mount of the super-root using special path constructions 646 * relative to your mounted root. 647 * 648 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were 649 * a PFS, including mirroring and storage quota operations, and this is 650 * prefered over creating discrete PFSs in the super-root. Instead 651 * the super-root is most typically used to create writable snapshots, 652 * alternative roots, and so forth. The super-root is also used by 653 * the automatic snapshotting mechanism. 654 */ 655 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU 656 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU 657 658 struct hammer2_volume_data { 659 /* 660 * sector #0 - 512 bytes 661 */ 662 uint64_t magic; /* 0000 Signature */ 663 hammer2_off_t boot_beg; /* 0008 Boot area (future) */ 664 hammer2_off_t boot_end; /* 0010 (size = end - beg) */ 665 hammer2_off_t aux_beg; /* 0018 Aux area (future) */ 666 hammer2_off_t aux_end; /* 0020 (size = end - beg) */ 667 hammer2_off_t volu_size; /* 0028 Volume size, bytes */ 668 669 uint32_t version; /* 0030 */ 670 uint32_t flags; /* 0034 */ 671 uint8_t copyid; /* 0038 copyid of phys vol */ 672 uint8_t freemap_version; /* 0039 freemap algorithm */ 673 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */ 674 uint8_t reserved003B; /* 003B */ 675 uint32_t reserved003C; /* 003C */ 676 677 uuid_t fsid; /* 0040 */ 678 uuid_t fstype; /* 0050 */ 679 680 /* 681 * allocator_size is precalculated at newfs time and does not include 682 * reserved blocks, boot, or redo areas. 683 * 684 * Initial non-reserved-area allocations do not use the allocation 685 * map but instead adjust alloc_iterator. Dynamic allocations take 686 * over starting at (allocator_beg). This makes newfs_hammer2's 687 * job a lot easier and can also serve as a testing jig. 688 */ 689 hammer2_off_t allocator_size; /* 0060 Total data space */ 690 hammer2_off_t allocator_free; /* 0068 Free space */ 691 hammer2_off_t allocator_beg; /* 0070 Initial allocations */ 692 hammer2_tid_t mirror_tid; /* 0078 best committed tid */ 693 hammer2_tid_t alloc_tid; /* 0080 Alloctable modify tid */ 694 hammer2_blockref_t alloc_blockref; /* 0088-00C7 */ 695 696 /* 697 * Copyids are allocated dynamically from the copyexists bitmap. 698 * An id from the active copies set (up to 8, see copyinfo later on) 699 * may still exist after the copy set has been removed from the 700 * volume header and its bit will remain active in the bitmap and 701 * cannot be reused until it is 100% removed from the hierarchy. 702 */ 703 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */ 704 char reserved0140[248]; /* 00E8-01DF */ 705 706 /* 707 * 32 bit CRC array at the end of the first 512 byte sector. 708 * 709 * icrc_sects[7] - First 512-4 bytes of volume header (including all 710 * the other icrc's except the last one). 711 * 712 * icrc_sects[6] - Second 512-4 bytes of volume header, which is 713 * the blockset for the root. 714 */ 715 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */ 716 717 /* 718 * sector #1 - 512 bytes 719 * 720 * The entire sector is used by a blockset. 721 */ 722 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */ 723 724 /* 725 * sector #2-7 726 */ 727 char sector2[512]; /* 0400-05FF reserved */ 728 char sector3[512]; /* 0600-07FF reserved */ 729 char sector4[512]; /* 0800-09FF reserved */ 730 char sector5[512]; /* 0A00-0BFF reserved */ 731 char sector6[512]; /* 0C00-0DFF reserved */ 732 char sector7[512]; /* 0E00-0FFF reserved */ 733 734 /* 735 * sector #8-71 - 32768 bytes 736 * 737 * Contains the configuration for up to 256 copyinfo targets. These 738 * specify local and remote copies operating as masters or slaves. 739 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255 740 * indicates the local media). 741 * 742 * Each inode contains a set of up to 8 copyids, either inherited 743 * from its parent or explicitly specified in the inode, which 744 * indexes into this array. 745 */ 746 /* 1000-8FFF copyinfo config */ 747 dmsg_vol_data_t copyinfo[HAMMER2_COPYID_COUNT]; 748 749 /* 750 * Remaining sections are reserved for future use. 751 */ 752 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */ 753 754 /* 755 * icrc on entire volume header 756 */ 757 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/ 758 }; 759 760 typedef struct hammer2_volume_data hammer2_volume_data_t; 761 762 /* 763 * Various parts of the volume header have their own iCRCs. 764 * 765 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes 766 * and not included the icrc calculation. 767 * 768 * The second 512 bytes also has its own iCRC but it is stored in the first 769 * 512 bytes so it covers the entire second 512 bytes. 770 * 771 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes, 772 * which is where the iCRC for the whole volume is stored. This is currently 773 * a catch-all for anything not individually iCRCd. 774 */ 775 #define HAMMER2_VOL_ICRC_SECT0 7 776 #define HAMMER2_VOL_ICRC_SECT1 6 777 778 #define HAMMER2_VOLUME_BYTES 65536 779 780 #define HAMMER2_VOLUME_ICRC0_OFF 0 781 #define HAMMER2_VOLUME_ICRC1_OFF 512 782 #define HAMMER2_VOLUME_ICRCVH_OFF 0 783 784 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4) 785 #define HAMMER2_VOLUME_ICRC1_SIZE (512) 786 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4) 787 788 #define HAMMER2_VOL_VERSION_MIN 1 789 #define HAMMER2_VOL_VERSION_DEFAULT 1 790 #define HAMMER2_VOL_VERSION_WIP 2 791 792 #define HAMMER2_NUM_VOLHDRS 4 793 794 union hammer2_media_data { 795 hammer2_volume_data_t voldata; 796 hammer2_inode_data_t ipdata; 797 hammer2_indblock_data_t npdata; 798 char buf[HAMMER2_PBUFSIZE]; 799 }; 800 801 typedef union hammer2_media_data hammer2_media_data_t; 802 803 #endif 804