1 /* 2 * Copyright (c) 2011-2014 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 36 #ifndef _VFS_HAMMER2_DISK_H_ 37 #define _VFS_HAMMER2_DISK_H_ 38 39 #ifndef _SYS_UUID_H_ 40 #include <sys/uuid.h> 41 #endif 42 #ifndef _SYS_DMSG_H_ 43 #include <sys/dmsg.h> 44 #endif 45 46 /* 47 * The structures below represent the on-disk media structures for the HAMMER2 48 * filesystem. Note that all fields for on-disk structures are naturally 49 * aligned. The host endian format is typically used - compatibility is 50 * possible if the implementation detects reversed endian and adjusts accesses 51 * accordingly. 52 * 53 * HAMMER2 primarily revolves around the directory topology: inodes, 54 * directory entries, and block tables. Block device buffer cache buffers 55 * are always 64KB. Logical file buffers are typically 16KB. All data 56 * references utilize 64-bit byte offsets. 57 * 58 * Free block management is handled independently using blocks reserved by 59 * the media topology. 60 */ 61 62 /* 63 * The data at the end of a file or directory may be a fragment in order 64 * to optimize storage efficiency. The minimum fragment size is 1KB. 65 * Since allocations are in powers of 2 fragments must also be sized in 66 * powers of 2 (1024, 2048, ... 65536). 67 * 68 * For the moment the maximum allocation size is HAMMER2_PBUFSIZE (64K), 69 * which is 2^16. Larger extents may be supported in the future. Smaller 70 * fragments might be supported in the future (down to 64 bytes is possible), 71 * but probably will not be. 72 * 73 * A full indirect block use supports 1024 x 64-byte blockrefs in a 64KB 74 * buffer. Indirect blocks down to 1KB are supported to keep small 75 * directories small. 76 * 77 * A maximally sized file (2^64-1 bytes) requires 5 indirect block levels. 78 * The hammer2_blockset in the volume header or file inode has another 8 79 * entries, giving us 66+3 = 69 bits of address space. However, some bits 80 * are taken up by (potentially) requests for redundant copies. HAMMER2 81 * currently supports up to 8 copies, which brings the address space down 82 * to 66 bits and gives us 2 bits of leeway. 83 */ 84 #define HAMMER2_ALLOC_MIN 1024 /* minimum allocation size */ 85 #define HAMMER2_RADIX_MIN 10 /* minimum allocation size 2^N */ 86 #define HAMMER2_ALLOC_MAX 65536 /* maximum allocation size */ 87 #define HAMMER2_RADIX_MAX 16 /* maximum allocation size 2^N */ 88 #define HAMMER2_RADIX_KEY 64 /* number of bits in key */ 89 90 /* 91 * MINALLOCSIZE - The minimum allocation size. This can be smaller 92 * or larger than the minimum physical IO size. 93 * 94 * NOTE: Should not be larger than 1K since inodes 95 * are 1K. 96 * 97 * MINIOSIZE - The minimum IO size. This must be less than 98 * or equal to HAMMER2_LBUFSIZE. 99 * 100 * HAMMER2_LBUFSIZE - Nominal buffer size for I/O rollups. 101 * 102 * HAMMER2_PBUFSIZE - Topological block size used by files for all 103 * blocks except the block straddling EOF. 104 * 105 * HAMMER2_SEGSIZE - Allocation map segment size, typically 2MB 106 * (space represented by a level0 bitmap). 107 */ 108 109 #define HAMMER2_SEGSIZE (1 << HAMMER2_FREEMAP_LEVEL0_RADIX) 110 #define HAMMER2_SEGRADIX HAMMER2_FREEMAP_LEVEL0_RADIX 111 112 #define HAMMER2_PBUFRADIX 16 /* physical buf (1<<16) bytes */ 113 #define HAMMER2_PBUFSIZE 65536 114 #define HAMMER2_LBUFRADIX 14 /* logical buf (1<<14) bytes */ 115 #define HAMMER2_LBUFSIZE 16384 116 117 /* 118 * Generally speaking we want to use 16K and 64K I/Os 119 */ 120 #define HAMMER2_MINIORADIX HAMMER2_LBUFRADIX 121 #define HAMMER2_MINIOSIZE HAMMER2_LBUFSIZE 122 123 #define HAMMER2_IND_BYTES_MIN HAMMER2_LBUFSIZE 124 #define HAMMER2_IND_BYTES_MAX HAMMER2_PBUFSIZE 125 #define HAMMER2_IND_COUNT_MIN (HAMMER2_IND_BYTES_MIN / \ 126 sizeof(hammer2_blockref_t)) 127 #define HAMMER2_IND_COUNT_MAX (HAMMER2_IND_BYTES_MAX / \ 128 sizeof(hammer2_blockref_t)) 129 130 /* 131 * In HAMMER2, arrays of blockrefs are fully set-associative, meaning that 132 * any element can occur at any index and holes can be anywhere. As a 133 * future optimization we will be able to flag that such arrays are sorted 134 * and thus optimize lookups, but for now we don't. 135 * 136 * Inodes embed either 512 bytes of direct data or an array of 8 blockrefs, 137 * resulting in highly efficient storage for files <= 512 bytes and for files 138 * <= 512KB. Up to 8 directory entries can be referenced from a directory 139 * without requiring an indirect block. 140 * 141 * Indirect blocks are typically either 4KB (64 blockrefs / ~4MB represented), 142 * or 64KB (1024 blockrefs / ~64MB represented). 143 */ 144 #define HAMMER2_SET_COUNT 8 /* direct entries */ 145 #define HAMMER2_SET_RADIX 3 146 #define HAMMER2_EMBEDDED_BYTES 512 /* inode blockset/dd size */ 147 #define HAMMER2_EMBEDDED_RADIX 9 148 149 #define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1) 150 #define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1) 151 #define HAMMER2_SEGMASK (HAMMER2_SEGSIZE - 1) 152 153 #define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK) 154 #define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE) 155 #define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK) 156 #define HAMMER2_SEGSIZE64 ((hammer2_off_t)HAMMER2_SEGSIZE) 157 #define HAMMER2_SEGMASK64 ((hammer2_off_t)HAMMER2_SEGMASK) 158 159 #define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5" 160 161 /* 162 * A HAMMER2 filesystem is always sized in multiples of 8MB. 163 * 164 * A 4MB segment is reserved at the beginning of each 2GB zone. This segment 165 * contains the volume header (or backup volume header), the free block 166 * table, and possibly other information in the future. 167 * 168 * 4MB = 64 x 64K blocks. Each 4MB segment is broken down as follows: 169 * 170 * +-----------------------+ 171 * | Volume Hdr | block 0 volume header & alternates 172 * +-----------------------+ (first four zones only) 173 * | FreeBlk Section A | block 1-4 174 * +-----------------------+ 175 * | FreeBlk Section B | block 5-8 176 * +-----------------------+ 177 * | FreeBlk Section C | block 9-12 178 * +-----------------------+ 179 * | FreeBlk Section D | block 13-16 180 * +-----------------------+ 181 * | | block 17...63 182 * | reserved | 183 * | | 184 * +-----------------------+ 185 * 186 * The first few 2GB zones contain volume headers and volume header backups. 187 * After that the volume header block# is reserved for future use. Similarly, 188 * there are many blocks related to various Freemap levels which are not 189 * used in every segment and those are also reserved for future use. 190 * 191 * Freemap (see the FREEMAP document) 192 * 193 * The freemap utilizes blocks #1-16 in 8 sets of 4 blocks. Each block in 194 * a set represents a level of depth in the freemap topology. Eight sets 195 * exist to prevent live updates from disturbing the state of the freemap 196 * were a crash/reboot to occur. That is, a live update is not committed 197 * until the update's flush reaches the volume root. There are FOUR volume 198 * roots representing the last four synchronization points, so the freemap 199 * must be consistent no matter which volume root is chosen by the mount 200 * code. 201 * 202 * Each freemap set is 4 x 64K blocks and represents the 2GB, 2TB, 2PB, 203 * and 2EB indirect map. The volume header itself has a set of 8 freemap 204 * blockrefs representing another 3 bits, giving us a total 64 bits of 205 * representable address space. 206 * 207 * The Level 0 64KB block represents 2GB of storage represented by 208 * (64 x struct hammer2_bmap_data). Each structure represents 2MB of storage 209 * and has a 256 bit bitmap, using 2 bits to represent a 16KB chunk of 210 * storage. These 2 bits represent the following states: 211 * 212 * 00 Free 213 * 01 (reserved) (Possibly partially allocated) 214 * 10 Possibly free 215 * 11 Allocated 216 * 217 * One important thing to note here is that the freemap resolution is 16KB, 218 * but the minimum storage allocation size is 1KB. The hammer2 vfs keeps 219 * track of sub-allocations in memory, which means that on a unmount or reboot 220 * the entire 16KB of a partially allocated block will be considered fully 221 * allocated. It is possible for fragmentation to build up over time, but 222 * defragmentation is fairly easy to accomplish since all modifications 223 * allocate a new block. 224 * 225 * The Second thing to note is that due to the way snapshots and inode 226 * replication works, deleting a file cannot immediately free the related 227 * space. Furthermore, deletions often do not bother to traverse the 228 * block subhierarchy being deleted. And to go even further, whole 229 * sub-directory trees can be deleted simply by deleting the directory inode 230 * at the top. So even though we have a symbol to represent a 'possibly free' 231 * block (binary 10), only the bulk free scanning code can actually use it. 232 * Normal 'rm's or other deletions do not. 233 * 234 * WARNING! ZONE_SEG and VOLUME_ALIGN must be a multiple of 1<<LEVEL0_RADIX 235 * (i.e. a multiple of 2MB). VOLUME_ALIGN must be >= ZONE_SEG. 236 * 237 * In Summary: 238 * 239 * (1) Modifications to freemap blocks 'allocate' a new copy (aka use a block 240 * from the next set). The new copy is reused until a flush occurs at 241 * which point the next modification will then rotate to the next set. 242 * 243 * (2) A total of 10 freemap sets is required. 244 * 245 * - 8 sets - 2 sets per volume header backup x 4 volume header backups 246 * - 2 sets used as backing store for the bulk freemap scan. 247 * - The freemap recovery scan which runs on-mount just uses the inactive 248 * set for whichever volume header was selected by the mount code. 249 * 250 */ 251 #define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024) 252 #define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 253 #define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 254 #define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK) 255 256 #define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN) 257 #define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 258 #define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 259 #define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK) 260 261 #define HAMMER2_ZONE_BYTES64 (2LLU * 1024 * 1024 * 1024) 262 #define HAMMER2_ZONE_MASK64 (HAMMER2_ZONE_BYTES64 - 1) 263 #define HAMMER2_ZONE_SEG (4 * 1024 * 1024) 264 #define HAMMER2_ZONE_SEG64 ((hammer2_off_t)HAMMER2_ZONE_SEG) 265 #define HAMMER2_ZONE_BLOCKS_SEG (HAMMER2_ZONE_SEG / HAMMER2_PBUFSIZE) 266 267 #define HAMMER2_ZONE_VOLHDR 0 /* volume header or backup */ 268 #define HAMMER2_ZONE_FREEMAP_00 1 /* normal freemap rotation */ 269 #define HAMMER2_ZONE_FREEMAP_01 5 /* normal freemap rotation */ 270 #define HAMMER2_ZONE_FREEMAP_02 9 /* normal freemap rotation */ 271 #define HAMMER2_ZONE_FREEMAP_03 13 /* normal freemap rotation */ 272 #define HAMMER2_ZONE_FREEMAP_04 17 /* normal freemap rotation */ 273 #define HAMMER2_ZONE_FREEMAP_05 21 /* normal freemap rotation */ 274 #define HAMMER2_ZONE_FREEMAP_06 25 /* batch freeing code only */ 275 #define HAMMER2_ZONE_FREEMAP_07 29 /* batch freeing code only */ 276 #define HAMMER2_ZONE_FREEMAP_08 33 /* (non-inclusive) */ 277 #define HAMMER2_ZONE_UNUSED62 62 278 #define HAMMER2_ZONE_UNUSED63 63 279 280 /* relative to FREEMAP_x */ 281 #define HAMMER2_ZONEFM_LEVEL1 0 /* 2GB leafmap */ 282 #define HAMMER2_ZONEFM_LEVEL2 1 /* 2TB indmap */ 283 #define HAMMER2_ZONEFM_LEVEL3 2 /* 2PB indmap */ 284 #define HAMMER2_ZONEFM_LEVEL4 3 /* 2EB indmap */ 285 /* LEVEL5 is a set of 8 blockrefs in the volume header 16EB */ 286 287 288 /* 289 * Freemap radii. Please note that LEVEL 1 blockref array entries 290 * point to 256-byte sections of the bitmap representing 2MB of storage. 291 * Even though the chain structures represent only 256 bytes, they are 292 * mapped using larger 16K or 64K buffer cache buffers. 293 */ 294 #define HAMMER2_FREEMAP_LEVEL5_RADIX 64 /* 16EB */ 295 #define HAMMER2_FREEMAP_LEVEL4_RADIX 61 /* 2EB */ 296 #define HAMMER2_FREEMAP_LEVEL3_RADIX 51 /* 2PB */ 297 #define HAMMER2_FREEMAP_LEVEL2_RADIX 41 /* 2TB */ 298 #define HAMMER2_FREEMAP_LEVEL1_RADIX 31 /* 2GB */ 299 #define HAMMER2_FREEMAP_LEVEL0_RADIX 21 /* 2MB (entry in l-1 leaf) */ 300 301 #define HAMMER2_FREEMAP_LEVELN_PSIZE 65536 /* physical bytes */ 302 303 #define HAMMER2_FREEMAP_COUNT (int)(HAMMER2_FREEMAP_LEVELN_PSIZE / \ 304 sizeof(hammer2_bmap_data_t)) 305 #define HAMMER2_FREEMAP_BLOCK_RADIX 14 306 #define HAMMER2_FREEMAP_BLOCK_SIZE (1 << HAMMER2_FREEMAP_BLOCK_RADIX) 307 #define HAMMER2_FREEMAP_BLOCK_MASK (HAMMER2_FREEMAP_BLOCK_SIZE - 1) 308 309 /* 310 * Two linear areas can be reserved after the initial 2MB segment in the base 311 * zone (the one starting at offset 0). These areas are NOT managed by the 312 * block allocator and do not fall under HAMMER2 crc checking rules based 313 * at the volume header (but can be self-CRCd internally, depending). 314 */ 315 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN 316 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024) 317 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024) 318 319 #define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN 320 #define HAMMER2_REDO_NOM_BYTES (256*1024*1024) 321 #define HAMMER2_REDO_MAX_BYTES (1024*1024*1024) 322 323 /* 324 * Most HAMMER2 types are implemented as unsigned 64-bit integers. 325 * Transaction ids are monotonic. 326 * 327 * We utilize 32-bit iSCSI CRCs. 328 */ 329 typedef uint64_t hammer2_tid_t; 330 typedef uint64_t hammer2_off_t; 331 typedef uint64_t hammer2_key_t; 332 typedef uint32_t hammer2_crc32_t; 333 334 /* 335 * Miscellanious ranges (all are unsigned). 336 */ 337 #define HAMMER2_TID_MIN 1ULL 338 #define HAMMER2_TID_MAX 0xFFFFFFFFFFFFFFFFULL 339 #define HAMMER2_KEY_MIN 0ULL 340 #define HAMMER2_KEY_MAX 0xFFFFFFFFFFFFFFFFULL 341 #define HAMMER2_OFFSET_MIN 0ULL 342 #define HAMMER2_OFFSET_MAX 0xFFFFFFFFFFFFFFFFULL 343 344 /* 345 * HAMMER2 data offset special cases and masking. 346 * 347 * All HAMMER2 data offsets have to be broken down into a 64K buffer base 348 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO). 349 * 350 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits 351 * of the data offset field specifies how large the data chunk being pointed 352 * to as a power of 2. The theoretical minimum radix is thus 6 (The space 353 * needed in the low bits of the data offset field). However, the practical 354 * minimum allocation chunk size is 1KB (a radix of 10), so HAMMER2 sets 355 * HAMMER2_RADIX_MIN to 10. The maximum radix is currently 16 (64KB), but 356 * we fully intend to support larger extents in the future. 357 */ 358 #define HAMMER2_OFF_BAD ((hammer2_off_t)-1) 359 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL 360 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64) 361 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64) 362 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL 363 #define HAMMER2_MAX_COPIES 6 364 365 /* 366 * HAMMER2 directory support and pre-defined keys 367 */ 368 #define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL 369 #define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL 370 #define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL 371 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL 372 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */ 373 374 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */ 375 376 /************************************************************************ 377 * DMSG SUPPORT * 378 ************************************************************************ 379 * LNK_VOLCONF 380 * 381 * All HAMMER2 directories directly under the super-root on your local 382 * media can be mounted separately, even if they share the same physical 383 * device. 384 * 385 * When you do a HAMMER2 mount you are effectively tying into a HAMMER2 386 * cluster via local media. The local media does not have to participate 387 * in the cluster, other than to provide the hammer2_volconf[] array and 388 * root inode for the mount. 389 * 390 * This is important: The mount device path you specify serves to bootstrap 391 * your entry into the cluster, but your mount will make active connections 392 * to ALL copy elements in the hammer2_volconf[] array which match the 393 * PFSID of the directory in the super-root that you specified. The local 394 * media path does not have to be mentioned in this array but becomes part 395 * of the cluster based on its type and access rights. ALL ELEMENTS ARE 396 * TREATED ACCORDING TO TYPE NO MATTER WHICH ONE YOU MOUNT FROM. 397 * 398 * The actual cluster may be far larger than the elements you list in the 399 * hammer2_volconf[] array. You list only the elements you wish to 400 * directly connect to and you are able to access the rest of the cluster 401 * indirectly through those connections. 402 * 403 * WARNING! This structure must be exactly 128 bytes long for its config 404 * array to fit in the volume header. 405 */ 406 struct hammer2_volconf { 407 uint8_t copyid; /* 00 copyid 0-255 (must match slot) */ 408 uint8_t inprog; /* 01 operation in progress, or 0 */ 409 uint8_t chain_to; /* 02 operation chaining to, or 0 */ 410 uint8_t chain_from; /* 03 operation chaining from, or 0 */ 411 uint16_t flags; /* 04-05 flags field */ 412 uint8_t error; /* 06 last operational error */ 413 uint8_t priority; /* 07 priority and round-robin flag */ 414 uint8_t remote_pfs_type;/* 08 probed direct remote PFS type */ 415 uint8_t reserved08[23]; /* 09-1F */ 416 uuid_t pfs_clid; /* 20-2F copy target must match this uuid */ 417 uint8_t label[16]; /* 30-3F import/export label */ 418 uint8_t path[64]; /* 40-7F target specification string or key */ 419 }; 420 421 typedef struct hammer2_volconf hammer2_volconf_t; 422 423 #define DMSG_VOLF_ENABLED 0x0001 424 #define DMSG_VOLF_INPROG 0x0002 425 #define DMSG_VOLF_CONN_RR 0x80 /* round-robin at same priority */ 426 #define DMSG_VOLF_CONN_EF 0x40 /* media errors flagged */ 427 #define DMSG_VOLF_CONN_PRI 0x0F /* select priority 0-15 (15=best) */ 428 429 struct dmsg_lnk_hammer2_volconf { 430 dmsg_hdr_t head; 431 hammer2_volconf_t copy; /* copy spec */ 432 int32_t index; 433 int32_t unused01; 434 uuid_t mediaid; 435 int64_t reserved02[32]; 436 }; 437 438 typedef struct dmsg_lnk_hammer2_volconf dmsg_lnk_hammer2_volconf_t; 439 440 #define DMSG_LNK_HAMMER2_VOLCONF DMSG_LNK(DMSG_LNK_CMD_HAMMER2_VOLCONF, \ 441 dmsg_lnk_hammer2_volconf) 442 443 #define H2_LNK_VOLCONF(msg) ((dmsg_lnk_hammer2_volconf_t *)(msg)->any.buf) 444 445 /* 446 * The media block reference structure. This forms the core of the HAMMER2 447 * media topology recursion. This 64-byte data structure is embedded in the 448 * volume header, in inodes (which are also directory entries), and in 449 * indirect blocks. 450 * 451 * A blockref references a single media item, which typically can be a 452 * directory entry (aka inode), indirect block, or data block. 453 * 454 * The primary feature a blockref represents is the ability to validate 455 * the entire tree underneath it via its check code. Any modification to 456 * anything propagates up the blockref tree all the way to the root, replacing 457 * the related blocks. Propagations can shortcut to the volume root to 458 * implement the 'fast syncing' feature but this only delays the eventual 459 * propagation. 460 * 461 * The check code can be a simple 32-bit iscsi code, a 64-bit crc, 462 * or as complex as a 192 bit cryptographic hash. 192 bits is the maximum 463 * supported check code size, which is not sufficient for unverified dedup 464 * UNLESS one doesn't mind once-in-a-blue-moon data corruption (such as when 465 * farming web data). HAMMER2 has an unverified dedup feature for just this 466 * purpose. 467 * 468 * -- 469 * 470 * NOTE: The range of keys represented by the blockref is (key) to 471 * ((key) + (1LL << keybits) - 1). HAMMER2 usually populates 472 * blocks bottom-up, inserting a new root when radix expansion 473 * is required. 474 * 475 * -- 476 * FUTURE BLOCKREF EXPANSION 477 * 478 * In order to implement a 256-bit content addressable index we want to 479 * have a 256-bit key which essentially represents the cryptographic hash. 480 * (so, 64-bit key + 192-bit crypto-hash or 256-bit key-is-the-hash + 481 * 32-bit consistency check for indirect block layers). 482 * 483 * THIS IS POSSIBLE in a 64-byte blockref structure. Of course, any number 484 * of bits can be represented by sizing the blockref. For the purposes of 485 * HAMMER2 though my limit is 256 bits. Not only that, but it will be an 486 * optimal construction because H2 already uses a variably-sized radix to 487 * pack the blockrefs at each level. A 256-bit mechanic would allow us 488 * to implement a content-addressable index. 489 */ 490 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */ 491 uint8_t type; /* type of underlying item */ 492 uint8_t methods; /* check method & compression method */ 493 uint8_t copyid; /* specify which copy this is */ 494 uint8_t keybits; /* #of keybits masked off 0=leaf */ 495 uint8_t vradix; /* virtual data/meta-data size */ 496 uint8_t flags; /* blockref flags */ 497 uint8_t reserved06; 498 uint8_t reserved07; 499 hammer2_key_t key; /* key specification */ 500 hammer2_tid_t mirror_tid; /* propagate for mirror scan */ 501 hammer2_tid_t modify_tid; /* modifications sans propagation */ 502 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/ 503 union { /* check info */ 504 char buf[24]; 505 struct { 506 uint32_t value; 507 uint32_t unused[5]; 508 } iscsi32; 509 struct { 510 uint64_t value; 511 uint64_t unused[2]; 512 } crc64; 513 struct { 514 char data[24]; 515 } sha192; 516 517 /* 518 * Freemap hints are embedded in addition to the icrc32. 519 * 520 * bigmask - Radixes available for allocation (0-31). 521 * Heuristical (may be permissive but not 522 * restrictive). Typically only radix values 523 * 10-16 are used (i.e. (1<<10) through (1<<16)). 524 * 525 * avail - Total available space remaining, in bytes 526 */ 527 struct { 528 uint32_t icrc32; 529 uint32_t bigmask; /* available radixes */ 530 uint64_t avail; /* total available bytes */ 531 uint64_t unused; /* unused must be 0 */ 532 } freemap; 533 } check; 534 }; 535 536 typedef struct hammer2_blockref hammer2_blockref_t; 537 538 #define HAMMER2_BLOCKREF_BYTES 64 /* blockref struct in bytes */ 539 540 /* 541 * On-media and off-media blockref types. 542 */ 543 #define HAMMER2_BREF_TYPE_EMPTY 0 544 #define HAMMER2_BREF_TYPE_INODE 1 545 #define HAMMER2_BREF_TYPE_INDIRECT 2 546 #define HAMMER2_BREF_TYPE_DATA 3 547 #define HAMMER2_BREF_TYPE_UNUSED04 4 548 #define HAMMER2_BREF_TYPE_FREEMAP_NODE 5 549 #define HAMMER2_BREF_TYPE_FREEMAP_LEAF 6 550 #define HAMMER2_BREF_TYPE_FREEMAP 254 /* pseudo-type */ 551 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */ 552 553 #define HAMMER2_BREF_FLAG_PFSROOT 0x01 /* see also related opflag */ 554 #define HAMMER2_BREF_FLAG_ZERO 0x02 555 556 /* 557 * Encode/decode check mode and compression mode for 558 * bref.methods. The compression level is not encoded in 559 * bref.methods. 560 */ 561 #define HAMMER2_ENC_CHECK(n) (((n) & 15) << 4) 562 #define HAMMER2_DEC_CHECK(n) (((n) >> 4) & 15) 563 #define HAMMER2_ENC_COMP(n) ((n) & 15) 564 #define HAMMER2_DEC_COMP(n) ((n) & 15) 565 566 #define HAMMER2_CHECK_NONE 0 567 #define HAMMER2_CHECK_DISABLED 1 568 #define HAMMER2_CHECK_ISCSI32 2 569 #define HAMMER2_CHECK_CRC64 3 570 #define HAMMER2_CHECK_SHA192 4 571 #define HAMMER2_CHECK_FREEMAP 5 572 573 /* user-specifiable check modes only */ 574 #define HAMMER2_CHECK_STRINGS { "none", "disabled", "crc32", \ 575 "crc64", "sha192" } 576 #define HAMMER2_CHECK_STRINGS_COUNT 5 577 578 /* 579 * Encode/decode check or compression algorithm request in 580 * ipdata->check_algo and ipdata->comp_algo. 581 */ 582 #define HAMMER2_ENC_ALGO(n) (n) 583 #define HAMMER2_DEC_ALGO(n) ((n) & 15) 584 #define HAMMER2_ENC_LEVEL(n) ((n) << 4) 585 #define HAMMER2_DEC_LEVEL(n) (((n) >> 4) & 15) 586 587 #define HAMMER2_COMP_NONE 0 588 #define HAMMER2_COMP_AUTOZERO 1 589 #define HAMMER2_COMP_LZ4 2 590 #define HAMMER2_COMP_ZLIB 3 591 592 #define HAMMER2_COMP_NEWFS_DEFAULT HAMMER2_COMP_LZ4 593 #define HAMMER2_COMP_STRINGS { "none", "autozero", "lz4", "zlib" } 594 #define HAMMER2_COMP_STRINGS_COUNT 4 595 596 597 /* 598 * HAMMER2 block references are collected into sets of 8 blockrefs. These 599 * sets are fully associative, meaning the elements making up a set are 600 * not sorted in any way and may contain duplicate entries, holes, or 601 * entries which shortcut multiple levels of indirection. Sets are used 602 * in various ways: 603 * 604 * (1) When redundancy is desired a set may contain several duplicate 605 * entries pointing to different copies of the same data. Up to 8 copies 606 * are supported but the set structure becomes a bit inefficient once 607 * you go over 4. 608 * 609 * (2) The blockrefs in a set can shortcut multiple levels of indirections 610 * within the bounds imposed by the parent of set. 611 * 612 * When a set fills up another level of indirection is inserted, moving 613 * some or all of the set's contents into indirect blocks placed under the 614 * set. This is a top-down approach in that indirect blocks are not created 615 * until the set actually becomes full (that is, the entries in the set can 616 * shortcut the indirect blocks when the set is not full). Depending on how 617 * things are filled multiple indirect blocks will eventually be created. 618 * 619 * Indirect blocks are typically 4KB (64 entres) or 64KB (1024 entries) and 620 * are also treated as fully set-associative. 621 */ 622 struct hammer2_blockset { 623 hammer2_blockref_t blockref[HAMMER2_SET_COUNT]; 624 }; 625 626 typedef struct hammer2_blockset hammer2_blockset_t; 627 628 /* 629 * Catch programmer snafus 630 */ 631 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT 632 #error "hammer2 direct radix is incorrect" 633 #endif 634 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE 635 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent" 636 #endif 637 #if (1 << HAMMER2_RADIX_MIN) != HAMMER2_ALLOC_MIN 638 #error "HAMMER2_RADIX_MIN and HAMMER2_ALLOC_MIN are inconsistent" 639 #endif 640 641 /* 642 * hammer2_bmap_data - A freemap entry in the LEVEL1 block. 643 * 644 * Each 64-byte entry contains the bitmap and meta-data required to manage 645 * a LEVEL0 (2MB) block of storage. The storage is managed in 128 x 16KB 646 * chunks. Smaller allocation granularity is supported via a linear iterator 647 * and/or must otherwise be tracked in ram. 648 * 649 * (data structure must be 64 bytes exactly) 650 * 651 * linear - A BYTE linear allocation offset used for sub-16KB allocations 652 * only. May contain values between 0 and 2MB. Must be ignored 653 * if 16KB-aligned (i.e. force bitmap scan), otherwise may be 654 * used to sub-allocate within the 16KB block (which is already 655 * marked as allocated in the bitmap). 656 * 657 * Sub-allocations need only be 1KB-aligned and do not have to be 658 * size-aligned, and 16KB or larger allocations do not update this 659 * field, resulting in pretty good packing. 660 * 661 * Please note that file data granularity may be limited by 662 * other issues such as buffer cache direct-mapping and the 663 * desire to support sector sizes up to 16KB (so H2 only issues 664 * I/O's in multiples of 16KB anyway). 665 * 666 * class - Clustering class. Cleared to 0 only if the entire leaf becomes 667 * free. Used to cluster device buffers so all elements must have 668 * the same device block size, but may mix logical sizes. 669 * 670 * Typically integrated with the blockref type in the upper 8 bits 671 * to localize inodes and indrect blocks, improving bulk free scans 672 * and directory scans. 673 * 674 * bitmap - Two bits per 16KB allocation block arranged in arrays of 675 * 32-bit elements, 128x2 bits representing ~2MB worth of media 676 * storage. Bit patterns are as follows: 677 * 678 * 00 Unallocated 679 * 01 (reserved) 680 * 10 Possibly free 681 * 11 Allocated 682 */ 683 struct hammer2_bmap_data { 684 int32_t linear; /* 00 linear sub-granular allocation offset */ 685 uint16_t class; /* 04-05 clustering class ((type<<8)|radix) */ 686 uint8_t reserved06; /* 06 */ 687 uint8_t reserved07; /* 07 */ 688 uint32_t reserved08; /* 08 */ 689 uint32_t reserved0C; /* 0C */ 690 uint32_t reserved10; /* 10 */ 691 uint32_t reserved14; /* 14 */ 692 uint32_t reserved18; /* 18 */ 693 uint32_t avail; /* 1C */ 694 uint32_t bitmap[8]; /* 20-3F 256 bits manages 2MB/16KB/2-bits */ 695 }; 696 697 typedef struct hammer2_bmap_data hammer2_bmap_data_t; 698 699 /* 700 * In HAMMER2 inodes ARE directory entries, with a special exception for 701 * hardlinks. The inode number is stored in the inode rather than being 702 * based on the location of the inode (since the location moves every time 703 * the inode or anything underneath the inode is modified). 704 * 705 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes 706 * for the filename, and 512 bytes worth of direct file data OR an embedded 707 * blockset. 708 * 709 * Directories represent one inode per blockref. Inodes are not laid out 710 * as a file but instead are represented by the related blockrefs. The 711 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember 712 * that blocksets are fully associative, so a certain degree efficiency is 713 * achieved just from that. 714 * 715 * Up to 512 bytes of direct data can be embedded in an inode, and since 716 * inodes are essentially directory entries this also means that small data 717 * files end up simply being laid out linearly in the directory, resulting 718 * in fewer seeks and highly optimal access. 719 * 720 * The compression mode can be changed at any time in the inode and is 721 * recorded on a blockref-by-blockref basis. 722 * 723 * Hardlinks are supported via the inode map. Essentially the way a hardlink 724 * works is that all individual directory entries representing the same file 725 * are special cased and specify the same inode number. The actual file 726 * is placed in the nearest parent directory that is parent to all instances 727 * of the hardlink. If all hardlinks to a file are in the same directory 728 * the actual file will also be placed in that directory. This file uses 729 * the inode number as the directory entry key and is invisible to normal 730 * directory scans. Real directory entry keys are differentiated from the 731 * inode number key via bit 63. Access to the hardlink silently looks up 732 * the real file and forwards all operations to that file. Removal of the 733 * last hardlink also removes the real file. 734 * 735 * (attr_tid) is only updated when the inode's specific attributes or regular 736 * file size has changed, and affects path lookups and stat. (attr_tid) 737 * represents a special cache coherency lock under the inode. The inode 738 * blockref's modify_tid will always cover it. 739 * 740 * (dirent_tid) is only updated when an entry under a directory inode has 741 * been created, deleted, renamed, or had its attributes change, and affects 742 * directory lookups and scans. (dirent_tid) represents another special cache 743 * coherency lock under the inode. The inode blockref's modify_tid will 744 * always cover it. 745 */ 746 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */ 747 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */ 748 #define HAMMER2_INODE_VERSION_ONE 1 749 750 #define HAMMER2_INODE_HIDDENDIR 16 /* special inode */ 751 #define HAMMER2_INODE_START 1024 /* dynamically allocated */ 752 753 struct hammer2_inode_data { 754 uint16_t version; /* 0000 inode data version */ 755 uint16_t reserved02; /* 0002 */ 756 757 /* 758 * core inode attributes, inode type, misc flags 759 */ 760 uint32_t uflags; /* 0004 chflags */ 761 uint32_t rmajor; /* 0008 available for device nodes */ 762 uint32_t rminor; /* 000C available for device nodes */ 763 uint64_t ctime; /* 0010 inode change time */ 764 uint64_t mtime; /* 0018 modified time */ 765 uint64_t atime; /* 0020 access time (unsupported) */ 766 uint64_t btime; /* 0028 birth time */ 767 uuid_t uid; /* 0030 uid / degenerate unix uid */ 768 uuid_t gid; /* 0040 gid / degenerate unix gid */ 769 770 uint8_t type; /* 0050 object type */ 771 uint8_t op_flags; /* 0051 operational flags */ 772 uint16_t cap_flags; /* 0052 capability flags */ 773 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */ 774 775 /* 776 * inode size, identification, localized recursive configuration 777 * for compression and backup copies. 778 */ 779 hammer2_tid_t inum; /* 0058 inode number */ 780 hammer2_off_t size; /* 0060 size of file */ 781 uint64_t nlinks; /* 0068 hard links (typ only dirs) */ 782 hammer2_tid_t iparent; /* 0070 parent inum (recovery only) */ 783 hammer2_key_t name_key; /* 0078 full filename key */ 784 uint16_t name_len; /* 0080 filename length */ 785 uint8_t ncopies; /* 0082 ncopies to local media */ 786 uint8_t comp_algo; /* 0083 compression request & algo */ 787 788 /* 789 * These fields are currently only applicable to PFSROOTs. 790 * 791 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely 792 * identify an instance of a PFS in the cluster because 793 * a mount may contain more than one copy of the PFS as 794 * a separate node. {pfs_clid, pfs_fsid} must be used for 795 * registration in the cluster. 796 */ 797 uint8_t target_type; /* 0084 hardlink target type */ 798 uint8_t check_algo; /* 0085 check code request & algo */ 799 uint8_t pfs_nmasters; /* 0086 (if PFSROOT) if multi-master */ 800 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */ 801 uint64_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */ 802 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */ 803 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */ 804 805 /* 806 * Quotas and aggregate sub-tree inode and data counters. Note that 807 * quotas are not replicated downward, they are explicitly set by 808 * the sysop and in-memory structures keep track of inheritence. 809 */ 810 hammer2_key_t data_quota; /* 00B0 subtree quota in bytes */ 811 hammer2_key_t data_count; /* 00B8 subtree byte count */ 812 hammer2_key_t inode_quota; /* 00C0 subtree quota inode count */ 813 hammer2_key_t inode_count; /* 00C8 subtree inode count */ 814 hammer2_tid_t attr_tid; /* 00D0 attributes changed */ 815 hammer2_tid_t dirent_tid; /* 00D8 directory/attr changed */ 816 817 /* 818 * Tracks (possibly degenerate) free areas covering all sub-tree 819 * allocations under inode, not counting the inode itself. 820 * 0/0 indicates empty entry. fully set-associative. 821 * 822 * (not yet implemented) 823 */ 824 hammer2_off_t reservedE0[4]; /* 00E0/E8/F0/F8 */ 825 826 unsigned char filename[HAMMER2_INODE_MAXNAME]; 827 /* 0100-01FF (256 char, unterminated) */ 828 union { /* 0200-03FF (64x8 = 512 bytes) */ 829 struct hammer2_blockset blockset; 830 char data[HAMMER2_EMBEDDED_BYTES]; 831 } u; 832 }; 833 834 typedef struct hammer2_inode_data hammer2_inode_data_t; 835 836 #define HAMMER2_OPFLAG_DIRECTDATA 0x01 837 #define HAMMER2_OPFLAG_PFSROOT 0x02 /* (see also bref flag) */ 838 #define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */ 839 840 #define HAMMER2_OBJTYPE_UNKNOWN 0 841 #define HAMMER2_OBJTYPE_DIRECTORY 1 842 #define HAMMER2_OBJTYPE_REGFILE 2 843 #define HAMMER2_OBJTYPE_FIFO 4 844 #define HAMMER2_OBJTYPE_CDEV 5 845 #define HAMMER2_OBJTYPE_BDEV 6 846 #define HAMMER2_OBJTYPE_SOFTLINK 7 847 #define HAMMER2_OBJTYPE_HARDLINK 8 /* dummy entry for hardlink */ 848 #define HAMMER2_OBJTYPE_SOCKET 9 849 #define HAMMER2_OBJTYPE_WHITEOUT 10 850 851 #define HAMMER2_COPYID_NONE 0 852 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1) 853 854 #define HAMMER2_COPYID_COUNT 256 855 856 /* 857 * PFS types identify a PFS on media and in LNK_SPAN messages. 858 * 859 * PFS types >= 16 belong to HAMMER, 0-15 are defined in sys/dmsg.h 860 */ 861 /* 0-15 reserved by sys/dmsg.h */ 862 #define HAMMER2_PFSTYPE_CACHE 16 863 #define HAMMER2_PFSTYPE_COPY 17 864 #define HAMMER2_PFSTYPE_SLAVE 18 865 #define HAMMER2_PFSTYPE_SOFT_SLAVE 19 866 #define HAMMER2_PFSTYPE_SOFT_MASTER 20 867 #define HAMMER2_PFSTYPE_MASTER 21 868 #define HAMMER2_PFSTYPE_SNAPSHOT 22 869 #define HAMMER2_PFSTYPE_SUPROOT 23 870 #define HAMMER2_PFSTYPE_MAX 32 871 872 /* 873 * Allocation Table 874 * 875 */ 876 877 878 /* 879 * Flags (8 bits) - blockref, for freemap only 880 * 881 * Note that the minimum chunk size is 1KB so we could theoretically have 882 * 10 bits here, but we might have some future extension that allows a 883 * chunk size down to 256 bytes and if so we will need bits 8 and 9. 884 */ 885 #define HAMMER2_AVF_SELMASK 0x03 /* select group */ 886 #define HAMMER2_AVF_ALL_ALLOC 0x04 /* indicate all allocated */ 887 #define HAMMER2_AVF_ALL_FREE 0x08 /* indicate all free */ 888 #define HAMMER2_AVF_RESERVED10 0x10 889 #define HAMMER2_AVF_RESERVED20 0x20 890 #define HAMMER2_AVF_RESERVED40 0x40 891 #define HAMMER2_AVF_RESERVED80 0x80 892 #define HAMMER2_AVF_AVMASK32 ((uint32_t)0xFFFFFF00LU) 893 #define HAMMER2_AVF_AVMASK64 ((uint64_t)0xFFFFFFFFFFFFFF00LLU) 894 895 #define HAMMER2_AV_SELECT_A 0x00 896 #define HAMMER2_AV_SELECT_B 0x01 897 #define HAMMER2_AV_SELECT_C 0x02 898 #define HAMMER2_AV_SELECT_D 0x03 899 900 /* 901 * The volume header eats a 64K block. There is currently an issue where 902 * we want to try to fit all nominal filesystem updates in a 512-byte section 903 * but it may be a lost cause due to the need for a blockset. 904 * 905 * All information is stored in host byte order. The volume header's magic 906 * number may be checked to determine the byte order. If you wish to mount 907 * between machines w/ different endian modes you'll need filesystem code 908 * which acts on the media data consistently (either all one way or all the 909 * other). Our code currently does not do that. 910 * 911 * A read-write mount may have to recover missing allocations by doing an 912 * incremental mirror scan looking for modifications made after alloc_tid. 913 * If alloc_tid == last_tid then no recovery operation is needed. Recovery 914 * operations are usually very, very fast. 915 * 916 * Read-only mounts do not need to do any recovery, access to the filesystem 917 * topology is always consistent after a crash (is always consistent, period). 918 * However, there may be shortcutted blockref updates present from deep in 919 * the tree which are stored in the volumeh eader and must be tracked on 920 * the fly. 921 * 922 * NOTE: The copyinfo[] array contains the configuration for both the 923 * cluster connections and any local media copies. The volume 924 * header will be replicated for each local media copy. 925 * 926 * The mount command may specify multiple medias or just one and 927 * allow HAMMER2 to pick up the others when it checks the copyinfo[] 928 * array on mount. 929 * 930 * NOTE: root_blockref points to the super-root directory, not the root 931 * directory. The root directory will be a subdirectory under the 932 * super-root. 933 * 934 * The super-root directory contains all root directories and all 935 * snapshots (readonly or writable). It is possible to do a 936 * null-mount of the super-root using special path constructions 937 * relative to your mounted root. 938 * 939 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were 940 * a PFS, including mirroring and storage quota operations, and this is 941 * prefered over creating discrete PFSs in the super-root. Instead 942 * the super-root is most typically used to create writable snapshots, 943 * alternative roots, and so forth. The super-root is also used by 944 * the automatic snapshotting mechanism. 945 */ 946 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU 947 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU 948 949 struct hammer2_volume_data { 950 /* 951 * sector #0 - 512 bytes 952 */ 953 uint64_t magic; /* 0000 Signature */ 954 hammer2_off_t boot_beg; /* 0008 Boot area (future) */ 955 hammer2_off_t boot_end; /* 0010 (size = end - beg) */ 956 hammer2_off_t aux_beg; /* 0018 Aux area (future) */ 957 hammer2_off_t aux_end; /* 0020 (size = end - beg) */ 958 hammer2_off_t volu_size; /* 0028 Volume size, bytes */ 959 960 uint32_t version; /* 0030 */ 961 uint32_t flags; /* 0034 */ 962 uint8_t copyid; /* 0038 copyid of phys vol */ 963 uint8_t freemap_version; /* 0039 freemap algorithm */ 964 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */ 965 uint8_t reserved003B; /* 003B */ 966 uint32_t reserved003C; /* 003C */ 967 968 uuid_t fsid; /* 0040 */ 969 uuid_t fstype; /* 0050 */ 970 971 /* 972 * allocator_size is precalculated at newfs time and does not include 973 * reserved blocks, boot, or redo areas. 974 * 975 * Initial non-reserved-area allocations do not use the freemap 976 * but instead adjust alloc_iterator. Dynamic allocations take 977 * over starting at (allocator_beg). This makes newfs_hammer2's 978 * job a lot easier and can also serve as a testing jig. 979 */ 980 hammer2_off_t allocator_size; /* 0060 Total data space */ 981 hammer2_off_t allocator_free; /* 0068 Free space */ 982 hammer2_off_t allocator_beg; /* 0070 Initial allocations */ 983 984 /* 985 * mirror_tid reflects the highest committed super-root change 986 * freemap_tid reflects the highest committed freemap change 987 * 988 * NOTE: mirror_tid does not track (and should not track) changes 989 * made to or under PFS roots. 990 */ 991 hammer2_tid_t mirror_tid; /* 0078 committed tid (vol) */ 992 hammer2_tid_t reserved0080; /* 0080 */ 993 hammer2_tid_t reserved0088; /* 0088 */ 994 hammer2_tid_t freemap_tid; /* 0090 committed tid (fmap) */ 995 hammer2_tid_t bulkfree_tid; /* 0098 bulkfree incremental */ 996 hammer2_tid_t reserved00A0[5]; /* 00A0-00C7 */ 997 998 /* 999 * Copyids are allocated dynamically from the copyexists bitmap. 1000 * An id from the active copies set (up to 8, see copyinfo later on) 1001 * may still exist after the copy set has been removed from the 1002 * volume header and its bit will remain active in the bitmap and 1003 * cannot be reused until it is 100% removed from the hierarchy. 1004 */ 1005 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */ 1006 char reserved0140[248]; /* 00E8-01DF */ 1007 1008 /* 1009 * 32 bit CRC array at the end of the first 512 byte sector. 1010 * 1011 * icrc_sects[7] - First 512-4 bytes of volume header (including all 1012 * the other icrc's except this one). 1013 * 1014 * icrc_sects[6] - Sector 1 (512 bytes) of volume header, which is 1015 * the blockset for the root. 1016 * 1017 * icrc_sects[5] - Sector 2 1018 * icrc_sects[4] - Sector 3 1019 * icrc_sects[3] - Sector 4 (the freemap blockset) 1020 */ 1021 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */ 1022 1023 /* 1024 * sector #1 - 512 bytes 1025 * 1026 * The entire sector is used by a blockset. 1027 */ 1028 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */ 1029 1030 /* 1031 * sector #2-7 1032 */ 1033 char sector2[512]; /* 0400-05FF reserved */ 1034 char sector3[512]; /* 0600-07FF reserved */ 1035 hammer2_blockset_t freemap_blockset; /* 0800-09FF freemap */ 1036 char sector5[512]; /* 0A00-0BFF reserved */ 1037 char sector6[512]; /* 0C00-0DFF reserved */ 1038 char sector7[512]; /* 0E00-0FFF reserved */ 1039 1040 /* 1041 * sector #8-71 - 32768 bytes 1042 * 1043 * Contains the configuration for up to 256 copyinfo targets. These 1044 * specify local and remote copies operating as masters or slaves. 1045 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255 1046 * indicates the local media). 1047 * 1048 * Each inode contains a set of up to 8 copyids, either inherited 1049 * from its parent or explicitly specified in the inode, which 1050 * indexes into this array. 1051 */ 1052 /* 1000-8FFF copyinfo config */ 1053 hammer2_volconf_t copyinfo[HAMMER2_COPYID_COUNT]; 1054 1055 /* 1056 * Remaining sections are reserved for future use. 1057 */ 1058 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */ 1059 1060 /* 1061 * icrc on entire volume header 1062 */ 1063 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/ 1064 }; 1065 1066 typedef struct hammer2_volume_data hammer2_volume_data_t; 1067 1068 /* 1069 * Various parts of the volume header have their own iCRCs. 1070 * 1071 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes 1072 * and not included the icrc calculation. 1073 * 1074 * The second 512 bytes also has its own iCRC but it is stored in the first 1075 * 512 bytes so it covers the entire second 512 bytes. 1076 * 1077 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes, 1078 * which is where the iCRC for the whole volume is stored. This is currently 1079 * a catch-all for anything not individually iCRCd. 1080 */ 1081 #define HAMMER2_VOL_ICRC_SECT0 7 1082 #define HAMMER2_VOL_ICRC_SECT1 6 1083 1084 #define HAMMER2_VOLUME_BYTES 65536 1085 1086 #define HAMMER2_VOLUME_ICRC0_OFF 0 1087 #define HAMMER2_VOLUME_ICRC1_OFF 512 1088 #define HAMMER2_VOLUME_ICRCVH_OFF 0 1089 1090 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4) 1091 #define HAMMER2_VOLUME_ICRC1_SIZE (512) 1092 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4) 1093 1094 #define HAMMER2_VOL_VERSION_MIN 1 1095 #define HAMMER2_VOL_VERSION_DEFAULT 1 1096 #define HAMMER2_VOL_VERSION_WIP 2 1097 1098 #define HAMMER2_NUM_VOLHDRS 4 1099 1100 union hammer2_media_data { 1101 hammer2_volume_data_t voldata; 1102 hammer2_inode_data_t ipdata; 1103 hammer2_blockref_t npdata[HAMMER2_IND_COUNT_MAX]; 1104 hammer2_bmap_data_t bmdata[HAMMER2_FREEMAP_COUNT]; 1105 char buf[HAMMER2_PBUFSIZE]; 1106 }; 1107 1108 typedef union hammer2_media_data hammer2_media_data_t; 1109 1110 #endif /* !_VFS_HAMMER2_DISK_H_ */ 1111