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 512 x 128-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 ~6 indirect block levels 78 * using 64KB indirect blocks (128 byte refs, 512 or radix 9 per indblk). 79 * 80 * 16(datablk) + 9 + 9 + 9 + 9 + 9 + 9 = ~70. 81 * 16(datablk) + 7 + 9 + 9 + 9 + 9 + 9 = ~68. (smaller top level indblk) 82 * 83 * The actual depth depends on copies redundancy and whether the filesystem 84 * has chosen to use a smaller indirect block size at the top level or not. 85 */ 86 #define HAMMER2_ALLOC_MIN 1024 /* minimum allocation size */ 87 #define HAMMER2_RADIX_MIN 10 /* minimum allocation size 2^N */ 88 #define HAMMER2_ALLOC_MAX 65536 /* maximum allocation size */ 89 #define HAMMER2_RADIX_MAX 16 /* maximum allocation size 2^N */ 90 #define HAMMER2_RADIX_KEY 64 /* number of bits in key */ 91 92 /* 93 * MINALLOCSIZE - The minimum allocation size. This can be smaller 94 * or larger than the minimum physical IO size. 95 * 96 * NOTE: Should not be larger than 1K since inodes 97 * are 1K. 98 * 99 * MINIOSIZE - The minimum IO size. This must be less than 100 * or equal to HAMMER2_LBUFSIZE. 101 * 102 * HAMMER2_LBUFSIZE - Nominal buffer size for I/O rollups. 103 * 104 * HAMMER2_PBUFSIZE - Topological block size used by files for all 105 * blocks except the block straddling EOF. 106 * 107 * HAMMER2_SEGSIZE - Allocation map segment size, typically 2MB 108 * (space represented by a level0 bitmap). 109 */ 110 111 #define HAMMER2_SEGSIZE (1 << HAMMER2_FREEMAP_LEVEL0_RADIX) 112 #define HAMMER2_SEGRADIX HAMMER2_FREEMAP_LEVEL0_RADIX 113 114 #define HAMMER2_PBUFRADIX 16 /* physical buf (1<<16) bytes */ 115 #define HAMMER2_PBUFSIZE 65536 116 #define HAMMER2_LBUFRADIX 14 /* logical buf (1<<14) bytes */ 117 #define HAMMER2_LBUFSIZE 16384 118 119 /* 120 * Generally speaking we want to use 16K and 64K I/Os 121 */ 122 #define HAMMER2_MINIORADIX HAMMER2_LBUFRADIX 123 #define HAMMER2_MINIOSIZE HAMMER2_LBUFSIZE 124 125 #define HAMMER2_IND_BYTES_MIN 4096 126 #define HAMMER2_IND_BYTES_NOM HAMMER2_LBUFSIZE 127 #define HAMMER2_IND_BYTES_MAX HAMMER2_PBUFSIZE 128 #define HAMMER2_IND_RADIX_MIN 12 129 #define HAMMER2_IND_RADIX_NOM HAMMER2_LBUFRADIX 130 #define HAMMER2_IND_RADIX_MAX HAMMER2_PBUFRADIX 131 #define HAMMER2_IND_COUNT_MIN (HAMMER2_IND_BYTES_MIN / \ 132 sizeof(hammer2_blockref_t)) 133 #define HAMMER2_IND_COUNT_MAX (HAMMER2_IND_BYTES_MAX / \ 134 sizeof(hammer2_blockref_t)) 135 136 /* 137 * In HAMMER2, arrays of blockrefs are fully set-associative, meaning that 138 * any element can occur at any index and holes can be anywhere. As a 139 * future optimization we will be able to flag that such arrays are sorted 140 * and thus optimize lookups, but for now we don't. 141 * 142 * Inodes embed either 512 bytes of direct data or an array of 8 blockrefs, 143 * resulting in highly efficient storage for files <= 512 bytes and for files 144 * <= 512KB. Up to 8 directory entries can be referenced from a directory 145 * without requiring an indirect block. 146 * 147 * Indirect blocks are typically either 4KB (64 blockrefs / ~4MB represented), 148 * or 64KB (1024 blockrefs / ~64MB represented). 149 */ 150 #define HAMMER2_SET_RADIX 2 /* radix 2 = 4 entries */ 151 #define HAMMER2_SET_COUNT (1 << HAMMER2_SET_RADIX) 152 #define HAMMER2_EMBEDDED_BYTES 512 /* inode blockset/dd size */ 153 #define HAMMER2_EMBEDDED_RADIX 9 154 155 #define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1) 156 #define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1) 157 #define HAMMER2_SEGMASK (HAMMER2_SEGSIZE - 1) 158 159 #define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK) 160 #define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE) 161 #define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK) 162 #define HAMMER2_SEGSIZE64 ((hammer2_off_t)HAMMER2_SEGSIZE) 163 #define HAMMER2_SEGMASK64 ((hammer2_off_t)HAMMER2_SEGMASK) 164 165 #define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5" 166 167 /* 168 * A HAMMER2 filesystem is always sized in multiples of 8MB. 169 * 170 * A 4MB segment is reserved at the beginning of each 2GB zone. This segment 171 * contains the volume header (or backup volume header), the free block 172 * table, and possibly other information in the future. 173 * 174 * 4MB = 64 x 64K blocks. Each 4MB segment is broken down as follows: 175 * 176 * +-----------------------+ 177 * | Volume Hdr | block 0 volume header & alternates 178 * +-----------------------+ (first four zones only) 179 * | FreeBlk Section A | block 1-4 180 * +-----------------------+ 181 * | FreeBlk Section B | block 5-8 182 * +-----------------------+ 183 * | FreeBlk Section C | block 9-12 184 * +-----------------------+ 185 * | FreeBlk Section D | block 13-16 186 * +-----------------------+ 187 * | | block 17...63 188 * | reserved | 189 * | | 190 * +-----------------------+ 191 * 192 * The first few 2GB zones contain volume headers and volume header backups. 193 * After that the volume header block# is reserved for future use. Similarly, 194 * there are many blocks related to various Freemap levels which are not 195 * used in every segment and those are also reserved for future use. 196 * 197 * Freemap (see the FREEMAP document) 198 * 199 * The freemap utilizes blocks #1-16 in 8 sets of 4 blocks. Each block in 200 * a set represents a level of depth in the freemap topology. Eight sets 201 * exist to prevent live updates from disturbing the state of the freemap 202 * were a crash/reboot to occur. That is, a live update is not committed 203 * until the update's flush reaches the volume root. There are FOUR volume 204 * roots representing the last four synchronization points, so the freemap 205 * must be consistent no matter which volume root is chosen by the mount 206 * code. 207 * 208 * Each freemap set is 4 x 64K blocks and represents the 2GB, 2TB, 2PB, 209 * and 2EB indirect map. The volume header itself has a set of 8 freemap 210 * blockrefs representing another 3 bits, giving us a total 64 bits of 211 * representable address space. 212 * 213 * The Level 0 64KB block represents 2GB of storage represented by 214 * (64 x struct hammer2_bmap_data). Each structure represents 2MB of storage 215 * and has a 256 bit bitmap, using 2 bits to represent a 16KB chunk of 216 * storage. These 2 bits represent the following states: 217 * 218 * 00 Free 219 * 01 (reserved) (Possibly partially allocated) 220 * 10 Possibly free 221 * 11 Allocated 222 * 223 * One important thing to note here is that the freemap resolution is 16KB, 224 * but the minimum storage allocation size is 1KB. The hammer2 vfs keeps 225 * track of sub-allocations in memory, which means that on a unmount or reboot 226 * the entire 16KB of a partially allocated block will be considered fully 227 * allocated. It is possible for fragmentation to build up over time, but 228 * defragmentation is fairly easy to accomplish since all modifications 229 * allocate a new block. 230 * 231 * The Second thing to note is that due to the way snapshots and inode 232 * replication works, deleting a file cannot immediately free the related 233 * space. Furthermore, deletions often do not bother to traverse the 234 * block subhierarchy being deleted. And to go even further, whole 235 * sub-directory trees can be deleted simply by deleting the directory inode 236 * at the top. So even though we have a symbol to represent a 'possibly free' 237 * block (binary 10), only the bulk free scanning code can actually use it. 238 * Normal 'rm's or other deletions do not. 239 * 240 * WARNING! ZONE_SEG and VOLUME_ALIGN must be a multiple of 1<<LEVEL0_RADIX 241 * (i.e. a multiple of 2MB). VOLUME_ALIGN must be >= ZONE_SEG. 242 * 243 * In Summary: 244 * 245 * (1) Modifications to freemap blocks 'allocate' a new copy (aka use a block 246 * from the next set). The new copy is reused until a flush occurs at 247 * which point the next modification will then rotate to the next set. 248 * 249 * (2) A total of 10 freemap sets is required. 250 * 251 * - 8 sets - 2 sets per volume header backup x 4 volume header backups 252 * - 2 sets used as backing store for the bulk freemap scan. 253 * - The freemap recovery scan which runs on-mount just uses the inactive 254 * set for whichever volume header was selected by the mount code. 255 * 256 */ 257 #define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024) 258 #define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 259 #define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 260 #define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK) 261 262 #define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN) 263 #define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN) 264 #define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1) 265 #define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK) 266 267 #define HAMMER2_ZONE_BYTES64 (2LLU * 1024 * 1024 * 1024) 268 #define HAMMER2_ZONE_MASK64 (HAMMER2_ZONE_BYTES64 - 1) 269 #define HAMMER2_ZONE_SEG (4 * 1024 * 1024) 270 #define HAMMER2_ZONE_SEG64 ((hammer2_off_t)HAMMER2_ZONE_SEG) 271 #define HAMMER2_ZONE_BLOCKS_SEG (HAMMER2_ZONE_SEG / HAMMER2_PBUFSIZE) 272 273 #define HAMMER2_ZONE_FREEMAP_INC 5 /* 5 deep */ 274 275 #define HAMMER2_ZONE_VOLHDR 0 /* volume header or backup */ 276 #define HAMMER2_ZONE_FREEMAP_00 1 /* normal freemap rotation */ 277 #define HAMMER2_ZONE_FREEMAP_01 6 /* normal freemap rotation */ 278 #define HAMMER2_ZONE_FREEMAP_02 11 /* normal freemap rotation */ 279 #define HAMMER2_ZONE_FREEMAP_03 16 /* normal freemap rotation */ 280 #define HAMMER2_ZONE_FREEMAP_04 21 /* normal freemap rotation */ 281 #define HAMMER2_ZONE_FREEMAP_05 26 /* normal freemap rotation */ 282 #define HAMMER2_ZONE_FREEMAP_06 31 /* normal freemap rotation */ 283 #define HAMMER2_ZONE_FREEMAP_07 36 /* normal freemap rotation */ 284 #define HAMMER2_ZONE_FREEMAP_END 41 /* (non-inclusive) */ 285 286 #define HAMMER2_ZONE_UNUSED41 41 287 #define HAMMER2_ZONE_UNUSED42 42 288 #define HAMMER2_ZONE_UNUSED43 43 289 #define HAMMER2_ZONE_UNUSED44 44 290 #define HAMMER2_ZONE_UNUSED45 45 291 #define HAMMER2_ZONE_UNUSED46 46 292 #define HAMMER2_ZONE_UNUSED47 47 293 #define HAMMER2_ZONE_UNUSED48 48 294 #define HAMMER2_ZONE_UNUSED49 49 295 #define HAMMER2_ZONE_UNUSED50 50 296 #define HAMMER2_ZONE_UNUSED51 51 297 #define HAMMER2_ZONE_UNUSED52 52 298 #define HAMMER2_ZONE_UNUSED53 53 299 #define HAMMER2_ZONE_UNUSED54 54 300 #define HAMMER2_ZONE_UNUSED55 55 301 #define HAMMER2_ZONE_UNUSED56 56 302 #define HAMMER2_ZONE_UNUSED57 57 303 #define HAMMER2_ZONE_UNUSED58 58 304 #define HAMMER2_ZONE_UNUSED59 59 305 #define HAMMER2_ZONE_UNUSED60 60 306 #define HAMMER2_ZONE_UNUSED61 61 307 #define HAMMER2_ZONE_UNUSED62 62 308 #define HAMMER2_ZONE_UNUSED63 63 309 #define HAMMER2_ZONE_END 64 /* non-inclusive */ 310 311 #define HAMMER2_NFREEMAPS 8 /* FREEMAP_00 - FREEMAP_07 */ 312 313 /* relative to FREEMAP_x */ 314 #define HAMMER2_ZONEFM_LEVEL1 0 /* 1GB leafmap */ 315 #define HAMMER2_ZONEFM_LEVEL2 1 /* 256GB indmap */ 316 #define HAMMER2_ZONEFM_LEVEL3 2 /* 64TB indmap */ 317 #define HAMMER2_ZONEFM_LEVEL4 3 /* 16PB indmap */ 318 #define HAMMER2_ZONEFM_LEVEL5 4 /* 4EB indmap */ 319 /* LEVEL6 is a set of 4 blockrefs in the volume header 16EB */ 320 321 /* 322 * Freemap radix. Assumes a set-count of 4, 128-byte blockrefs, 323 * 32KB indirect block for freemap (LEVELN_PSIZE below). 324 * 325 * Leaf entry represents 4MB of storage broken down into a 512-bit 326 * bitmap, 2-bits per entry. So course bitmap item represents 16KB. 327 */ 328 #if HAMMER2_SET_COUNT != 4 329 #error "hammer2_disk.h - freemap assumes SET_COUNT is 4" 330 #endif 331 #define HAMMER2_FREEMAP_LEVEL6_RADIX 64 /* 16EB (end) */ 332 #define HAMMER2_FREEMAP_LEVEL5_RADIX 62 /* 4EB */ 333 #define HAMMER2_FREEMAP_LEVEL4_RADIX 54 /* 16PB */ 334 #define HAMMER2_FREEMAP_LEVEL3_RADIX 46 /* 64TB */ 335 #define HAMMER2_FREEMAP_LEVEL2_RADIX 38 /* 256GB */ 336 #define HAMMER2_FREEMAP_LEVEL1_RADIX 30 /* 1GB */ 337 #define HAMMER2_FREEMAP_LEVEL0_RADIX 22 /* 4MB (128by in l-1 leaf) */ 338 339 #define HAMMER2_FREEMAP_LEVELN_PSIZE 32768 /* physical bytes */ 340 341 #define HAMMER2_FREEMAP_LEVEL5_SIZE ((hammer2_off_t)1 << \ 342 HAMMER2_FREEMAP_LEVEL5_RADIX) 343 #define HAMMER2_FREEMAP_LEVEL4_SIZE ((hammer2_off_t)1 << \ 344 HAMMER2_FREEMAP_LEVEL4_RADIX) 345 #define HAMMER2_FREEMAP_LEVEL3_SIZE ((hammer2_off_t)1 << \ 346 HAMMER2_FREEMAP_LEVEL3_RADIX) 347 #define HAMMER2_FREEMAP_LEVEL2_SIZE ((hammer2_off_t)1 << \ 348 HAMMER2_FREEMAP_LEVEL2_RADIX) 349 #define HAMMER2_FREEMAP_LEVEL1_SIZE ((hammer2_off_t)1 << \ 350 HAMMER2_FREEMAP_LEVEL1_RADIX) 351 #define HAMMER2_FREEMAP_LEVEL0_SIZE ((hammer2_off_t)1 << \ 352 HAMMER2_FREEMAP_LEVEL0_RADIX) 353 354 #define HAMMER2_FREEMAP_LEVEL5_MASK (HAMMER2_FREEMAP_LEVEL5_SIZE - 1) 355 #define HAMMER2_FREEMAP_LEVEL4_MASK (HAMMER2_FREEMAP_LEVEL4_SIZE - 1) 356 #define HAMMER2_FREEMAP_LEVEL3_MASK (HAMMER2_FREEMAP_LEVEL3_SIZE - 1) 357 #define HAMMER2_FREEMAP_LEVEL2_MASK (HAMMER2_FREEMAP_LEVEL2_SIZE - 1) 358 #define HAMMER2_FREEMAP_LEVEL1_MASK (HAMMER2_FREEMAP_LEVEL1_SIZE - 1) 359 #define HAMMER2_FREEMAP_LEVEL0_MASK (HAMMER2_FREEMAP_LEVEL0_SIZE - 1) 360 361 #define HAMMER2_FREEMAP_COUNT (int)(HAMMER2_FREEMAP_LEVELN_PSIZE / \ 362 sizeof(hammer2_bmap_data_t)) 363 364 /* 365 * 16KB bitmap granularity (x2 bits per entry). 366 */ 367 #define HAMMER2_FREEMAP_BLOCK_RADIX 14 368 #define HAMMER2_FREEMAP_BLOCK_SIZE (1 << HAMMER2_FREEMAP_BLOCK_RADIX) 369 #define HAMMER2_FREEMAP_BLOCK_MASK (HAMMER2_FREEMAP_BLOCK_SIZE - 1) 370 371 /* 372 * bitmap[] structure. 2 bits per HAMMER2_FREEMAP_BLOCK_SIZE. 373 * 374 * 8 x 64-bit elements, 2 bits per block. 375 * 32 blocks (radix 5) per element. 376 * representing INDEX_SIZE bytes worth of storage per element. 377 */ 378 379 typedef uint64_t hammer2_bitmap_t; 380 381 #define HAMMER2_BMAP_ALLONES ((hammer2_bitmap_t)-1) 382 #define HAMMER2_BMAP_ELEMENTS 8 383 #define HAMMER2_BMAP_BITS_PER_ELEMENT 64 384 #define HAMMER2_BMAP_INDEX_RADIX 5 /* 32 blocks per element */ 385 #define HAMMER2_BMAP_BLOCKS_PER_ELEMENT (1 << HAMMER2_BMAP_INDEX_RADIX) 386 387 #define HAMMER2_BMAP_INDEX_SIZE (HAMMER2_FREEMAP_BLOCK_SIZE * \ 388 HAMMER2_BMAP_BLOCKS_PER_ELEMENT) 389 #define HAMMER2_BMAP_INDEX_MASK (HAMMER2_BMAP_INDEX_SIZE - 1) 390 391 /* 392 * Two linear areas can be reserved after the initial 2MB segment in the base 393 * zone (the one starting at offset 0). These areas are NOT managed by the 394 * block allocator and do not fall under HAMMER2 crc checking rules based 395 * at the volume header (but can be self-CRCd internally, depending). 396 */ 397 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN 398 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024) 399 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024) 400 401 #define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN 402 #define HAMMER2_REDO_NOM_BYTES (256*1024*1024) 403 #define HAMMER2_REDO_MAX_BYTES (1024*1024*1024) 404 405 /* 406 * Most HAMMER2 types are implemented as unsigned 64-bit integers. 407 * Transaction ids are monotonic. 408 * 409 * We utilize 32-bit iSCSI CRCs. 410 */ 411 typedef uint64_t hammer2_tid_t; 412 typedef uint64_t hammer2_off_t; 413 typedef uint64_t hammer2_key_t; 414 typedef uint32_t hammer2_crc32_t; 415 416 /* 417 * Miscellanious ranges (all are unsigned). 418 */ 419 #define HAMMER2_TID_MIN 1ULL 420 #define HAMMER2_TID_MAX 0xFFFFFFFFFFFFFFFFULL 421 #define HAMMER2_KEY_MIN 0ULL 422 #define HAMMER2_KEY_MAX 0xFFFFFFFFFFFFFFFFULL 423 #define HAMMER2_OFFSET_MIN 0ULL 424 #define HAMMER2_OFFSET_MAX 0xFFFFFFFFFFFFFFFFULL 425 426 /* 427 * HAMMER2 data offset special cases and masking. 428 * 429 * All HAMMER2 data offsets have to be broken down into a 64K buffer base 430 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO). 431 * 432 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits 433 * of the data offset field specifies how large the data chunk being pointed 434 * to as a power of 2. The theoretical minimum radix is thus 6 (The space 435 * needed in the low bits of the data offset field). However, the practical 436 * minimum allocation chunk size is 1KB (a radix of 10), so HAMMER2 sets 437 * HAMMER2_RADIX_MIN to 10. The maximum radix is currently 16 (64KB), but 438 * we fully intend to support larger extents in the future. 439 * 440 * WARNING! A radix of 0 (such as when data_off is all 0's) is a special 441 * case which means no data associated with the blockref, and 442 * not the '1 byte' it would otherwise calculate to. 443 */ 444 #define HAMMER2_OFF_BAD ((hammer2_off_t)-1) 445 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL 446 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64) 447 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64) 448 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL 449 #define HAMMER2_MAX_COPIES 6 450 451 /* 452 * HAMMER2 directory support and pre-defined keys 453 */ 454 #define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL 455 #define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL 456 #define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL 457 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL 458 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */ 459 460 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */ 461 #define HAMMER2_BOOT_KEY 0xd9b36ce135528000ULL /* sroot to BOOT PFS */ 462 463 /************************************************************************ 464 * DMSG SUPPORT * 465 ************************************************************************ 466 * LNK_VOLCONF 467 * 468 * All HAMMER2 directories directly under the super-root on your local 469 * media can be mounted separately, even if they share the same physical 470 * device. 471 * 472 * When you do a HAMMER2 mount you are effectively tying into a HAMMER2 473 * cluster via local media. The local media does not have to participate 474 * in the cluster, other than to provide the hammer2_volconf[] array and 475 * root inode for the mount. 476 * 477 * This is important: The mount device path you specify serves to bootstrap 478 * your entry into the cluster, but your mount will make active connections 479 * to ALL copy elements in the hammer2_volconf[] array which match the 480 * PFSID of the directory in the super-root that you specified. The local 481 * media path does not have to be mentioned in this array but becomes part 482 * of the cluster based on its type and access rights. ALL ELEMENTS ARE 483 * TREATED ACCORDING TO TYPE NO MATTER WHICH ONE YOU MOUNT FROM. 484 * 485 * The actual cluster may be far larger than the elements you list in the 486 * hammer2_volconf[] array. You list only the elements you wish to 487 * directly connect to and you are able to access the rest of the cluster 488 * indirectly through those connections. 489 * 490 * WARNING! This structure must be exactly 128 bytes long for its config 491 * array to fit in the volume header. 492 */ 493 struct hammer2_volconf { 494 uint8_t copyid; /* 00 copyid 0-255 (must match slot) */ 495 uint8_t inprog; /* 01 operation in progress, or 0 */ 496 uint8_t chain_to; /* 02 operation chaining to, or 0 */ 497 uint8_t chain_from; /* 03 operation chaining from, or 0 */ 498 uint16_t flags; /* 04-05 flags field */ 499 uint8_t error; /* 06 last operational error */ 500 uint8_t priority; /* 07 priority and round-robin flag */ 501 uint8_t remote_pfs_type;/* 08 probed direct remote PFS type */ 502 uint8_t reserved08[23]; /* 09-1F */ 503 uuid_t pfs_clid; /* 20-2F copy target must match this uuid */ 504 uint8_t label[16]; /* 30-3F import/export label */ 505 uint8_t path[64]; /* 40-7F target specification string or key */ 506 } __packed; 507 508 typedef struct hammer2_volconf hammer2_volconf_t; 509 510 #define DMSG_VOLF_ENABLED 0x0001 511 #define DMSG_VOLF_INPROG 0x0002 512 #define DMSG_VOLF_CONN_RR 0x80 /* round-robin at same priority */ 513 #define DMSG_VOLF_CONN_EF 0x40 /* media errors flagged */ 514 #define DMSG_VOLF_CONN_PRI 0x0F /* select priority 0-15 (15=best) */ 515 516 struct dmsg_lnk_hammer2_volconf { 517 dmsg_hdr_t head; 518 hammer2_volconf_t copy; /* copy spec */ 519 int32_t index; 520 int32_t unused01; 521 uuid_t mediaid; 522 int64_t reserved02[32]; 523 } __packed; 524 525 typedef struct dmsg_lnk_hammer2_volconf dmsg_lnk_hammer2_volconf_t; 526 527 #define DMSG_LNK_HAMMER2_VOLCONF DMSG_LNK(DMSG_LNK_CMD_HAMMER2_VOLCONF, \ 528 dmsg_lnk_hammer2_volconf) 529 530 #define H2_LNK_VOLCONF(msg) ((dmsg_lnk_hammer2_volconf_t *)(msg)->any.buf) 531 532 /* 533 * HAMMER2 directory entry header (embedded in blockref) exactly 16 bytes 534 */ 535 struct hammer2_dirent_head { 536 hammer2_tid_t inum; /* inode number */ 537 uint16_t namlen; /* name length */ 538 uint8_t type; /* OBJTYPE_* */ 539 uint8_t unused0B; 540 uint8_t unused0C[4]; 541 } __packed; 542 543 typedef struct hammer2_dirent_head hammer2_dirent_head_t; 544 545 /* 546 * The media block reference structure. This forms the core of the HAMMER2 547 * media topology recursion. This 128-byte data structure is embedded in the 548 * volume header, in inodes (which are also directory entries), and in 549 * indirect blocks. 550 * 551 * A blockref references a single media item, which typically can be a 552 * directory entry (aka inode), indirect block, or data block. 553 * 554 * The primary feature a blockref represents is the ability to validate 555 * the entire tree underneath it via its check code. Any modification to 556 * anything propagates up the blockref tree all the way to the root, replacing 557 * the related blocks and compounding the generated check code. 558 * 559 * The check code can be a simple 32-bit iscsi code, a 64-bit crc, or as 560 * complex as a 512 bit cryptographic hash. I originally used a 64-byte 561 * blockref but later expanded it to 128 bytes to be able to support the 562 * larger check code as well as to embed statistics for quota operation. 563 * 564 * Simple check codes are not sufficient for unverified dedup. Even with 565 * a maximally-sized check code unverified dedup should only be used in 566 * in subdirectory trees where you do not need 100% data integrity. 567 * 568 * Unverified dedup is deduping based on meta-data only without verifying 569 * that the data blocks are actually identical. Verified dedup guarantees 570 * integrity but is a far more I/O-expensive operation. 571 * 572 * -- 573 * 574 * mirror_tid - per cluster node modified (propagated upward by flush) 575 * modify_tid - clc record modified (not propagated). 576 * update_tid - clc record updated (propagated upward on verification) 577 * 578 * CLC - Stands for 'Cluster Level Change', identifiers which are identical 579 * within the topology across all cluster nodes (when fully 580 * synchronized). 581 * 582 * NOTE: The range of keys represented by the blockref is (key) to 583 * ((key) + (1LL << keybits) - 1). HAMMER2 usually populates 584 * blocks bottom-up, inserting a new root when radix expansion 585 * is required. 586 * 587 * RESERVED FIELDS 588 * 589 * A number of blockref fields are reserved and should generally be set to 590 * 0 for future compatibility. 591 * 592 * FUTURE BLOCKREF EXPANSION 593 * 594 * CONTENT ADDRESSABLE INDEXING (future) - Using a 256 or 512-bit check code. 595 */ 596 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */ 597 uint8_t type; /* type of underlying item */ 598 uint8_t methods; /* check method & compression method */ 599 uint8_t copyid; /* specify which copy this is */ 600 uint8_t keybits; /* #of keybits masked off 0=leaf */ 601 uint8_t vradix; /* virtual data/meta-data size */ 602 uint8_t flags; /* blockref flags */ 603 uint8_t reserved06; 604 uint8_t reserved07; 605 hammer2_key_t key; /* key specification */ 606 hammer2_tid_t mirror_tid; /* media flush topology & freemap */ 607 hammer2_tid_t modify_tid; /* clc modify (not propagated) */ 608 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/ 609 hammer2_tid_t update_tid; /* clc modify (propagated upward) */ 610 union { 611 char buf[16]; 612 613 /* 614 * Directory entry header (BREF_TYPE_DIRENT) 615 * 616 * NOTE: check.buf contains filename if <= 64 bytes. Longer 617 * filenames are stored in a data reference of size 618 * HAMMER2_ALLOC_MIN (at least 256, typically 1024). 619 * 620 * NOTE: inode structure may contain a copy of a recently 621 * associated filename, for recovery purposes. 622 * 623 * NOTE: Superroot entries are INODEs, not DIRENTs. Code 624 * allows both cases. 625 */ 626 hammer2_dirent_head_t dirent; 627 628 /* 629 * Statistics aggregation (BREF_TYPE_INODE, BREF_TYPE_INDIRECT) 630 */ 631 struct { 632 hammer2_key_t data_count; 633 hammer2_key_t inode_count; 634 } stats; 635 } embed; 636 union { /* check info */ 637 char buf[64]; 638 struct { 639 uint32_t value; 640 uint32_t reserved[15]; 641 } iscsi32; 642 struct { 643 uint64_t value; 644 uint64_t reserved[7]; 645 } xxhash64; 646 struct { 647 char data[24]; 648 char reserved[40]; 649 } sha192; 650 struct { 651 char data[32]; 652 char reserved[32]; 653 } sha256; 654 struct { 655 char data[64]; 656 } sha512; 657 658 /* 659 * Freemap hints are embedded in addition to the icrc32. 660 * 661 * bigmask - Radixes available for allocation (0-31). 662 * Heuristical (may be permissive but not 663 * restrictive). Typically only radix values 664 * 10-16 are used (i.e. (1<<10) through (1<<16)). 665 * 666 * avail - Total available space remaining, in bytes 667 */ 668 struct { 669 uint32_t icrc32; 670 uint32_t bigmask; /* available radixes */ 671 uint64_t avail; /* total available bytes */ 672 char reserved[48]; 673 } freemap; 674 } check; 675 } __packed; 676 677 typedef struct hammer2_blockref hammer2_blockref_t; 678 679 #define HAMMER2_BLOCKREF_BYTES 128 /* blockref struct in bytes */ 680 #define HAMMER2_BLOCKREF_RADIX 7 681 682 /* 683 * On-media and off-media blockref types. 684 * 685 * types >= 128 are pseudo values that should never be present on-media. 686 */ 687 #define HAMMER2_BREF_TYPE_EMPTY 0 688 #define HAMMER2_BREF_TYPE_INODE 1 689 #define HAMMER2_BREF_TYPE_INDIRECT 2 690 #define HAMMER2_BREF_TYPE_DATA 3 691 #define HAMMER2_BREF_TYPE_DIRENT 4 692 #define HAMMER2_BREF_TYPE_FREEMAP_NODE 5 693 #define HAMMER2_BREF_TYPE_FREEMAP_LEAF 6 694 #define HAMMER2_BREF_TYPE_FREEMAP 254 /* pseudo-type */ 695 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */ 696 697 #define HAMMER2_BREF_FLAG_PFSROOT 0x01 /* see also related opflag */ 698 #define HAMMER2_BREF_FLAG_ZERO 0x02 699 700 /* 701 * Encode/decode check mode and compression mode for 702 * bref.methods. The compression level is not encoded in 703 * bref.methods. 704 */ 705 #define HAMMER2_ENC_CHECK(n) (((n) & 15) << 4) 706 #define HAMMER2_DEC_CHECK(n) (((n) >> 4) & 15) 707 #define HAMMER2_ENC_COMP(n) ((n) & 15) 708 #define HAMMER2_DEC_COMP(n) ((n) & 15) 709 710 #define HAMMER2_CHECK_NONE 0 711 #define HAMMER2_CHECK_DISABLED 1 712 #define HAMMER2_CHECK_ISCSI32 2 713 #define HAMMER2_CHECK_XXHASH64 3 714 #define HAMMER2_CHECK_SHA192 4 715 #define HAMMER2_CHECK_FREEMAP 5 716 717 #define HAMMER2_CHECK_DEFAULT HAMMER2_CHECK_XXHASH64 718 719 /* user-specifiable check modes only */ 720 #define HAMMER2_CHECK_STRINGS { "none", "disabled", "crc32", \ 721 "xxhash64", "sha192" } 722 #define HAMMER2_CHECK_STRINGS_COUNT 5 723 724 /* 725 * Encode/decode check or compression algorithm request in 726 * ipdata->meta.check_algo and ipdata->meta.comp_algo. 727 */ 728 #define HAMMER2_ENC_ALGO(n) (n) 729 #define HAMMER2_DEC_ALGO(n) ((n) & 15) 730 #define HAMMER2_ENC_LEVEL(n) ((n) << 4) 731 #define HAMMER2_DEC_LEVEL(n) (((n) >> 4) & 15) 732 733 #define HAMMER2_COMP_NONE 0 734 #define HAMMER2_COMP_AUTOZERO 1 735 #define HAMMER2_COMP_LZ4 2 736 #define HAMMER2_COMP_ZLIB 3 737 738 #define HAMMER2_COMP_NEWFS_DEFAULT HAMMER2_COMP_LZ4 739 #define HAMMER2_COMP_STRINGS { "none", "autozero", "lz4", "zlib" } 740 #define HAMMER2_COMP_STRINGS_COUNT 4 741 742 /* 743 * Passed to hammer2_chain_create(), causes methods to be inherited from 744 * parent. 745 */ 746 #define HAMMER2_METH_DEFAULT -1 747 748 /* 749 * HAMMER2 block references are collected into sets of 4 blockrefs. These 750 * sets are fully associative, meaning the elements making up a set are 751 * not sorted in any way and may contain duplicate entries, holes, or 752 * entries which shortcut multiple levels of indirection. Sets are used 753 * in various ways: 754 * 755 * (1) When redundancy is desired a set may contain several duplicate 756 * entries pointing to different copies of the same data. Up to 4 copies 757 * are supported. 758 * 759 * (2) The blockrefs in a set can shortcut multiple levels of indirections 760 * within the bounds imposed by the parent of set. 761 * 762 * When a set fills up another level of indirection is inserted, moving 763 * some or all of the set's contents into indirect blocks placed under the 764 * set. This is a top-down approach in that indirect blocks are not created 765 * until the set actually becomes full (that is, the entries in the set can 766 * shortcut the indirect blocks when the set is not full). Depending on how 767 * things are filled multiple indirect blocks will eventually be created. 768 * 769 * Indirect blocks are typically 4KB (64 entres) or 64KB (1024 entries) and 770 * are also treated as fully set-associative. 771 */ 772 struct hammer2_blockset { 773 hammer2_blockref_t blockref[HAMMER2_SET_COUNT]; 774 }; 775 776 typedef struct hammer2_blockset hammer2_blockset_t; 777 778 /* 779 * Catch programmer snafus 780 */ 781 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT 782 #error "hammer2 direct radix is incorrect" 783 #endif 784 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE 785 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent" 786 #endif 787 #if (1 << HAMMER2_RADIX_MIN) != HAMMER2_ALLOC_MIN 788 #error "HAMMER2_RADIX_MIN and HAMMER2_ALLOC_MIN are inconsistent" 789 #endif 790 791 /* 792 * hammer2_bmap_data - A freemap entry in the LEVEL1 block. 793 * 794 * Each 128-byte entry contains the bitmap and meta-data required to manage 795 * a LEVEL0 (128KB) block of storage. The storage is managed in 128 x 1KB 796 * chunks. 797 * 798 * A smaller allocation granularity is supported via a linear iterator and/or 799 * must otherwise be tracked in ram. 800 * 801 * (data structure must be 128 bytes exactly) 802 * 803 * linear - A BYTE linear allocation offset used for sub-16KB allocations 804 * only. May contain values between 0 and 2MB. Must be ignored 805 * if 16KB-aligned (i.e. force bitmap scan), otherwise may be 806 * used to sub-allocate within the 16KB block (which is already 807 * marked as allocated in the bitmap). 808 * 809 * Sub-allocations need only be 1KB-aligned and do not have to be 810 * size-aligned, and 16KB or larger allocations do not update this 811 * field, resulting in pretty good packing. 812 * 813 * Please note that file data granularity may be limited by 814 * other issues such as buffer cache direct-mapping and the 815 * desire to support sector sizes up to 16KB (so H2 only issues 816 * I/O's in multiples of 16KB anyway). 817 * 818 * class - Clustering class. Cleared to 0 only if the entire leaf becomes 819 * free. Used to cluster device buffers so all elements must have 820 * the same device block size, but may mix logical sizes. 821 * 822 * Typically integrated with the blockref type in the upper 8 bits 823 * to localize inodes and indrect blocks, improving bulk free scans 824 * and directory scans. 825 * 826 * bitmap - Two bits per 16KB allocation block arranged in arrays of 827 * 32-bit elements, 256x2 bits representing ~4MB worth of media 828 * storage. Bit patterns are as follows: 829 * 830 * 00 Unallocated 831 * 01 (reserved) 832 * 10 Possibly free 833 * 11 Allocated 834 */ 835 struct hammer2_bmap_data { 836 int32_t linear; /* 00 linear sub-granular allocation offset */ 837 uint16_t class; /* 04-05 clustering class ((type<<8)|radix) */ 838 uint8_t reserved06; /* 06 */ 839 uint8_t reserved07; /* 07 */ 840 uint32_t reserved08; /* 08 */ 841 uint32_t reserved0C; /* 0C */ 842 uint32_t reserved10; /* 10 */ 843 uint32_t reserved14; /* 14 */ 844 uint32_t reserved18; /* 18 */ 845 uint32_t avail; /* 1C */ 846 uint32_t reserved20[8]; /* 20-3F 256 bits manages 128K/1KB/2-bits */ 847 /* 40-7F 512 bits manages 4MB of storage */ 848 hammer2_bitmap_t bitmapq[HAMMER2_BMAP_ELEMENTS]; 849 } __packed; 850 851 typedef struct hammer2_bmap_data hammer2_bmap_data_t; 852 853 /* 854 * In HAMMER2 inodes ARE directory entries, with a special exception for 855 * hardlinks. The inode number is stored in the inode rather than being 856 * based on the location of the inode (since the location moves every time 857 * the inode or anything underneath the inode is modified). 858 * 859 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes 860 * for the filename, and 512 bytes worth of direct file data OR an embedded 861 * blockset. The in-memory hammer2_inode structure contains only the mostly- 862 * node-independent meta-data portion (some flags are node-specific and will 863 * not be synchronized). The rest of the inode is node-specific and chain I/O 864 * is required to obtain it. 865 * 866 * Directories represent one inode per blockref. Inodes are not laid out 867 * as a file but instead are represented by the related blockrefs. The 868 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember 869 * that blocksets are fully associative, so a certain degree efficiency is 870 * achieved just from that. 871 * 872 * Up to 512 bytes of direct data can be embedded in an inode, and since 873 * inodes are essentially directory entries this also means that small data 874 * files end up simply being laid out linearly in the directory, resulting 875 * in fewer seeks and highly optimal access. 876 * 877 * The compression mode can be changed at any time in the inode and is 878 * recorded on a blockref-by-blockref basis. 879 * 880 * Hardlinks are supported via the inode map. Essentially the way a hardlink 881 * works is that all individual directory entries representing the same file 882 * are special cased and specify the same inode number. The actual file 883 * is placed in the nearest parent directory that is parent to all instances 884 * of the hardlink. If all hardlinks to a file are in the same directory 885 * the actual file will also be placed in that directory. This file uses 886 * the inode number as the directory entry key and is invisible to normal 887 * directory scans. Real directory entry keys are differentiated from the 888 * inode number key via bit 63. Access to the hardlink silently looks up 889 * the real file and forwards all operations to that file. Removal of the 890 * last hardlink also removes the real file. 891 * 892 * (attr_tid) is only updated when the inode's specific attributes or regular 893 * file size has changed, and affects path lookups and stat. (attr_tid) 894 * represents a special cache coherency lock under the inode. The inode 895 * blockref's modify_tid will always cover it. 896 * 897 * (dirent_tid) is only updated when an entry under a directory inode has 898 * been created, deleted, renamed, or had its attributes change, and affects 899 * directory lookups and scans. (dirent_tid) represents another special cache 900 * coherency lock under the inode. The inode blockref's modify_tid will 901 * always cover it. 902 */ 903 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */ 904 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */ 905 #define HAMMER2_INODE_VERSION_ONE 1 906 907 #define HAMMER2_INODE_START 1024 /* dynamically allocated */ 908 909 struct hammer2_inode_meta { 910 uint16_t version; /* 0000 inode data version */ 911 uint8_t reserved02; /* 0002 */ 912 uint8_t pfs_subtype; /* 0003 pfs sub-type */ 913 914 /* 915 * core inode attributes, inode type, misc flags 916 */ 917 uint32_t uflags; /* 0004 chflags */ 918 uint32_t rmajor; /* 0008 available for device nodes */ 919 uint32_t rminor; /* 000C available for device nodes */ 920 uint64_t ctime; /* 0010 inode change time */ 921 uint64_t mtime; /* 0018 modified time */ 922 uint64_t atime; /* 0020 access time (unsupported) */ 923 uint64_t btime; /* 0028 birth time */ 924 uuid_t uid; /* 0030 uid / degenerate unix uid */ 925 uuid_t gid; /* 0040 gid / degenerate unix gid */ 926 927 uint8_t type; /* 0050 object type */ 928 uint8_t op_flags; /* 0051 operational flags */ 929 uint16_t cap_flags; /* 0052 capability flags */ 930 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */ 931 932 /* 933 * inode size, identification, localized recursive configuration 934 * for compression and backup copies. 935 * 936 * NOTE: Nominal parent inode number (iparent) is only applicable 937 * for directories but can also help for files during 938 * catastrophic recovery. 939 */ 940 hammer2_tid_t inum; /* 0058 inode number */ 941 hammer2_off_t size; /* 0060 size of file */ 942 uint64_t nlinks; /* 0068 hard links (typ only dirs) */ 943 hammer2_tid_t iparent; /* 0070 nominal parent inum */ 944 hammer2_key_t name_key; /* 0078 full filename key */ 945 uint16_t name_len; /* 0080 filename length */ 946 uint8_t ncopies; /* 0082 ncopies to local media */ 947 uint8_t comp_algo; /* 0083 compression request & algo */ 948 949 /* 950 * These fields are currently only applicable to PFSROOTs. 951 * 952 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely 953 * identify an instance of a PFS in the cluster because 954 * a mount may contain more than one copy of the PFS as 955 * a separate node. {pfs_clid, pfs_fsid} must be used for 956 * registration in the cluster. 957 */ 958 uint8_t target_type; /* 0084 hardlink target type */ 959 uint8_t check_algo; /* 0085 check code request & algo */ 960 uint8_t pfs_nmasters; /* 0086 (if PFSROOT) if multi-master */ 961 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */ 962 uint64_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */ 963 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */ 964 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */ 965 966 /* 967 * Quotas and aggregate sub-tree inode and data counters. Note that 968 * quotas are not replicated downward, they are explicitly set by 969 * the sysop and in-memory structures keep track of inheritence. 970 */ 971 hammer2_key_t data_quota; /* 00B0 subtree quota in bytes */ 972 hammer2_key_t unusedB8; /* 00B8 subtree byte count */ 973 hammer2_key_t inode_quota; /* 00C0 subtree quota inode count */ 974 hammer2_key_t unusedC8; /* 00C8 subtree inode count */ 975 976 /* 977 * The last snapshot tid is tested against modify_tid to determine 978 * when a copy must be made of a data block whos check mode has been 979 * disabled (a disabled check mode allows data blocks to be updated 980 * in place instead of copy-on-write). 981 */ 982 hammer2_tid_t pfs_lsnap_tid; /* 00D0 last snapshot tid */ 983 hammer2_tid_t reservedD8; /* 00D8 (avail) */ 984 985 /* 986 * Tracks (possibly degenerate) free areas covering all sub-tree 987 * allocations under inode, not counting the inode itself. 988 * 0/0 indicates empty entry. fully set-associative. 989 * 990 * (not yet implemented) 991 */ 992 uint64_t decrypt_check; /* 00E0 decryption validator */ 993 hammer2_off_t reservedE0[3]; /* 00E8/F0/F8 */ 994 } __packed; 995 996 typedef struct hammer2_inode_meta hammer2_inode_meta_t; 997 998 struct hammer2_inode_data { 999 hammer2_inode_meta_t meta; /* 0000-00FF */ 1000 unsigned char filename[HAMMER2_INODE_MAXNAME]; 1001 /* 0100-01FF (256 char, unterminated) */ 1002 union { /* 0200-03FF (64x8 = 512 bytes) */ 1003 struct hammer2_blockset blockset; 1004 char data[HAMMER2_EMBEDDED_BYTES]; 1005 } u; 1006 } __packed; 1007 1008 typedef struct hammer2_inode_data hammer2_inode_data_t; 1009 1010 #define HAMMER2_OPFLAG_DIRECTDATA 0x01 1011 #define HAMMER2_OPFLAG_PFSROOT 0x02 /* (see also bref flag) */ 1012 #define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */ 1013 1014 #define HAMMER2_OBJTYPE_UNKNOWN 0 1015 #define HAMMER2_OBJTYPE_DIRECTORY 1 1016 #define HAMMER2_OBJTYPE_REGFILE 2 1017 #define HAMMER2_OBJTYPE_FIFO 4 1018 #define HAMMER2_OBJTYPE_CDEV 5 1019 #define HAMMER2_OBJTYPE_BDEV 6 1020 #define HAMMER2_OBJTYPE_SOFTLINK 7 1021 #define HAMMER2_OBJTYPE_UNUSED08 8 1022 #define HAMMER2_OBJTYPE_SOCKET 9 1023 #define HAMMER2_OBJTYPE_WHITEOUT 10 1024 1025 #define HAMMER2_COPYID_NONE 0 1026 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1) 1027 1028 #define HAMMER2_COPYID_COUNT 256 1029 1030 /* 1031 * PFS types identify the role of a PFS within a cluster. The PFS types 1032 * is stored on media and in LNK_SPAN messages and used in other places. 1033 * 1034 * The low 4 bits specify the current active type while the high 4 bits 1035 * specify the transition target if the PFS is being upgraded or downgraded, 1036 * If the upper 4 bits are not zero it may effect how a PFS is used during 1037 * the transition. 1038 * 1039 * Generally speaking, downgrading a MASTER to a SLAVE cannot complete until 1040 * at least all MASTERs have updated their pfs_nmasters field. And upgrading 1041 * a SLAVE to a MASTER cannot complete until the new prospective master has 1042 * been fully synchronized (though theoretically full synchronization is 1043 * not required if a (new) quorum of other masters are fully synchronized). 1044 * 1045 * It generally does not matter which PFS element you actually mount, you 1046 * are mounting 'the cluster'. So, for example, a network mount will mount 1047 * a DUMMY PFS type on a memory filesystem. However, there are two exceptions. 1048 * In order to gain the benefits of a SOFT_MASTER or SOFT_SLAVE, those PFSs 1049 * must be directly mounted. 1050 */ 1051 #define HAMMER2_PFSTYPE_NONE 0x00 1052 #define HAMMER2_PFSTYPE_CACHE 0x01 1053 #define HAMMER2_PFSTYPE_UNUSED02 0x02 1054 #define HAMMER2_PFSTYPE_SLAVE 0x03 1055 #define HAMMER2_PFSTYPE_SOFT_SLAVE 0x04 1056 #define HAMMER2_PFSTYPE_SOFT_MASTER 0x05 1057 #define HAMMER2_PFSTYPE_MASTER 0x06 1058 #define HAMMER2_PFSTYPE_UNUSED07 0x07 1059 #define HAMMER2_PFSTYPE_SUPROOT 0x08 1060 #define HAMMER2_PFSTYPE_DUMMY 0x09 1061 #define HAMMER2_PFSTYPE_MAX 16 1062 1063 #define HAMMER2_PFSTRAN_NONE 0x00 /* no transition in progress */ 1064 #define HAMMER2_PFSTRAN_CACHE 0x10 1065 #define HAMMER2_PFSTRAN_UNMUSED20 0x20 1066 #define HAMMER2_PFSTRAN_SLAVE 0x30 1067 #define HAMMER2_PFSTRAN_SOFT_SLAVE 0x40 1068 #define HAMMER2_PFSTRAN_SOFT_MASTER 0x50 1069 #define HAMMER2_PFSTRAN_MASTER 0x60 1070 #define HAMMER2_PFSTRAN_UNUSED70 0x70 1071 #define HAMMER2_PFSTRAN_SUPROOT 0x80 1072 #define HAMMER2_PFSTRAN_DUMMY 0x90 1073 1074 #define HAMMER2_PFS_DEC(n) ((n) & 0x0F) 1075 #define HAMMER2_PFS_DEC_TRANSITION(n) (((n) >> 4) & 0x0F) 1076 #define HAMMER2_PFS_ENC_TRANSITION(n) (((n) & 0x0F) << 4) 1077 1078 #define HAMMER2_PFSSUBTYPE_NONE 0 1079 #define HAMMER2_PFSSUBTYPE_SNAPSHOT 1 /* manual/managed snapshot */ 1080 #define HAMMER2_PFSSUBTYPE_AUTOSNAP 2 /* automatic snapshot */ 1081 1082 /* 1083 * PFS mode of operation is a bitmask. This is typically not stored 1084 * on-media, but defined here because the field may be used in dmsgs. 1085 */ 1086 #define HAMMER2_PFSMODE_QUORUM 0x01 1087 #define HAMMER2_PFSMODE_RW 0x02 1088 1089 /* 1090 * Allocation Table 1091 * 1092 */ 1093 1094 1095 /* 1096 * Flags (8 bits) - blockref, for freemap only 1097 * 1098 * Note that the minimum chunk size is 1KB so we could theoretically have 1099 * 10 bits here, but we might have some future extension that allows a 1100 * chunk size down to 256 bytes and if so we will need bits 8 and 9. 1101 */ 1102 #define HAMMER2_AVF_SELMASK 0x03 /* select group */ 1103 #define HAMMER2_AVF_ALL_ALLOC 0x04 /* indicate all allocated */ 1104 #define HAMMER2_AVF_ALL_FREE 0x08 /* indicate all free */ 1105 #define HAMMER2_AVF_RESERVED10 0x10 1106 #define HAMMER2_AVF_RESERVED20 0x20 1107 #define HAMMER2_AVF_RESERVED40 0x40 1108 #define HAMMER2_AVF_RESERVED80 0x80 1109 #define HAMMER2_AVF_AVMASK32 ((uint32_t)0xFFFFFF00LU) 1110 #define HAMMER2_AVF_AVMASK64 ((uint64_t)0xFFFFFFFFFFFFFF00LLU) 1111 1112 #define HAMMER2_AV_SELECT_A 0x00 1113 #define HAMMER2_AV_SELECT_B 0x01 1114 #define HAMMER2_AV_SELECT_C 0x02 1115 #define HAMMER2_AV_SELECT_D 0x03 1116 1117 /* 1118 * The volume header eats a 64K block. There is currently an issue where 1119 * we want to try to fit all nominal filesystem updates in a 512-byte section 1120 * but it may be a lost cause due to the need for a blockset. 1121 * 1122 * All information is stored in host byte order. The volume header's magic 1123 * number may be checked to determine the byte order. If you wish to mount 1124 * between machines w/ different endian modes you'll need filesystem code 1125 * which acts on the media data consistently (either all one way or all the 1126 * other). Our code currently does not do that. 1127 * 1128 * A read-write mount may have to recover missing allocations by doing an 1129 * incremental mirror scan looking for modifications made after alloc_tid. 1130 * If alloc_tid == last_tid then no recovery operation is needed. Recovery 1131 * operations are usually very, very fast. 1132 * 1133 * Read-only mounts do not need to do any recovery, access to the filesystem 1134 * topology is always consistent after a crash (is always consistent, period). 1135 * However, there may be shortcutted blockref updates present from deep in 1136 * the tree which are stored in the volumeh eader and must be tracked on 1137 * the fly. 1138 * 1139 * NOTE: The copyinfo[] array contains the configuration for both the 1140 * cluster connections and any local media copies. The volume 1141 * header will be replicated for each local media copy. 1142 * 1143 * The mount command may specify multiple medias or just one and 1144 * allow HAMMER2 to pick up the others when it checks the copyinfo[] 1145 * array on mount. 1146 * 1147 * NOTE: root_blockref points to the super-root directory, not the root 1148 * directory. The root directory will be a subdirectory under the 1149 * super-root. 1150 * 1151 * The super-root directory contains all root directories and all 1152 * snapshots (readonly or writable). It is possible to do a 1153 * null-mount of the super-root using special path constructions 1154 * relative to your mounted root. 1155 * 1156 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were 1157 * a PFS, including mirroring and storage quota operations, and this is 1158 * prefered over creating discrete PFSs in the super-root. Instead 1159 * the super-root is most typically used to create writable snapshots, 1160 * alternative roots, and so forth. The super-root is also used by 1161 * the automatic snapshotting mechanism. 1162 */ 1163 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU 1164 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU 1165 1166 struct hammer2_volume_data { 1167 /* 1168 * sector #0 - 512 bytes 1169 */ 1170 uint64_t magic; /* 0000 Signature */ 1171 hammer2_off_t boot_beg; /* 0008 Boot area (future) */ 1172 hammer2_off_t boot_end; /* 0010 (size = end - beg) */ 1173 hammer2_off_t aux_beg; /* 0018 Aux area (future) */ 1174 hammer2_off_t aux_end; /* 0020 (size = end - beg) */ 1175 hammer2_off_t volu_size; /* 0028 Volume size, bytes */ 1176 1177 uint32_t version; /* 0030 */ 1178 uint32_t flags; /* 0034 */ 1179 uint8_t copyid; /* 0038 copyid of phys vol */ 1180 uint8_t freemap_version; /* 0039 freemap algorithm */ 1181 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */ 1182 uint8_t reserved003B; /* 003B */ 1183 uint32_t reserved003C; /* 003C */ 1184 1185 uuid_t fsid; /* 0040 */ 1186 uuid_t fstype; /* 0050 */ 1187 1188 /* 1189 * allocator_size is precalculated at newfs time and does not include 1190 * reserved blocks, boot, or redo areas. 1191 * 1192 * Initial non-reserved-area allocations do not use the freemap 1193 * but instead adjust alloc_iterator. Dynamic allocations take 1194 * over starting at (allocator_beg). This makes newfs_hammer2's 1195 * job a lot easier and can also serve as a testing jig. 1196 */ 1197 hammer2_off_t allocator_size; /* 0060 Total data space */ 1198 hammer2_off_t allocator_free; /* 0068 Free space */ 1199 hammer2_off_t allocator_beg; /* 0070 Initial allocations */ 1200 1201 /* 1202 * mirror_tid reflects the highest committed change for this 1203 * block device regardless of whether it is to the super-root 1204 * or to a PFS or whatever. 1205 * 1206 * freemap_tid reflects the highest committed freemap change for 1207 * this block device. 1208 */ 1209 hammer2_tid_t mirror_tid; /* 0078 committed tid (vol) */ 1210 hammer2_tid_t reserved0080; /* 0080 */ 1211 hammer2_tid_t reserved0088; /* 0088 */ 1212 hammer2_tid_t freemap_tid; /* 0090 committed tid (fmap) */ 1213 hammer2_tid_t bulkfree_tid; /* 0098 bulkfree incremental */ 1214 hammer2_tid_t reserved00A0[5]; /* 00A0-00C7 */ 1215 1216 /* 1217 * Copyids are allocated dynamically from the copyexists bitmap. 1218 * An id from the active copies set (up to 8, see copyinfo later on) 1219 * may still exist after the copy set has been removed from the 1220 * volume header and its bit will remain active in the bitmap and 1221 * cannot be reused until it is 100% removed from the hierarchy. 1222 */ 1223 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */ 1224 char reserved0140[248]; /* 00E8-01DF */ 1225 1226 /* 1227 * 32 bit CRC array at the end of the first 512 byte sector. 1228 * 1229 * icrc_sects[7] - First 512-4 bytes of volume header (including all 1230 * the other icrc's except this one). 1231 * 1232 * icrc_sects[6] - Sector 1 (512 bytes) of volume header, which is 1233 * the blockset for the root. 1234 * 1235 * icrc_sects[5] - Sector 2 1236 * icrc_sects[4] - Sector 3 1237 * icrc_sects[3] - Sector 4 (the freemap blockset) 1238 */ 1239 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */ 1240 1241 /* 1242 * sector #1 - 512 bytes 1243 * 1244 * The entire sector is used by a blockset. 1245 */ 1246 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */ 1247 1248 /* 1249 * sector #2-7 1250 */ 1251 char sector2[512]; /* 0400-05FF reserved */ 1252 char sector3[512]; /* 0600-07FF reserved */ 1253 hammer2_blockset_t freemap_blockset; /* 0800-09FF freemap */ 1254 char sector5[512]; /* 0A00-0BFF reserved */ 1255 char sector6[512]; /* 0C00-0DFF reserved */ 1256 char sector7[512]; /* 0E00-0FFF reserved */ 1257 1258 /* 1259 * sector #8-71 - 32768 bytes 1260 * 1261 * Contains the configuration for up to 256 copyinfo targets. These 1262 * specify local and remote copies operating as masters or slaves. 1263 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255 1264 * indicates the local media). 1265 * 1266 * Each inode contains a set of up to 8 copyids, either inherited 1267 * from its parent or explicitly specified in the inode, which 1268 * indexes into this array. 1269 */ 1270 /* 1000-8FFF copyinfo config */ 1271 hammer2_volconf_t copyinfo[HAMMER2_COPYID_COUNT]; 1272 1273 /* 1274 * Remaining sections are reserved for future use. 1275 */ 1276 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */ 1277 1278 /* 1279 * icrc on entire volume header 1280 */ 1281 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/ 1282 } __packed; 1283 1284 typedef struct hammer2_volume_data hammer2_volume_data_t; 1285 1286 /* 1287 * Various parts of the volume header have their own iCRCs. 1288 * 1289 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes 1290 * and not included the icrc calculation. 1291 * 1292 * The second 512 bytes also has its own iCRC but it is stored in the first 1293 * 512 bytes so it covers the entire second 512 bytes. 1294 * 1295 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes, 1296 * which is where the iCRC for the whole volume is stored. This is currently 1297 * a catch-all for anything not individually iCRCd. 1298 */ 1299 #define HAMMER2_VOL_ICRC_SECT0 7 1300 #define HAMMER2_VOL_ICRC_SECT1 6 1301 1302 #define HAMMER2_VOLUME_BYTES 65536 1303 1304 #define HAMMER2_VOLUME_ICRC0_OFF 0 1305 #define HAMMER2_VOLUME_ICRC1_OFF 512 1306 #define HAMMER2_VOLUME_ICRCVH_OFF 0 1307 1308 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4) 1309 #define HAMMER2_VOLUME_ICRC1_SIZE (512) 1310 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4) 1311 1312 #define HAMMER2_VOL_VERSION_MIN 1 1313 #define HAMMER2_VOL_VERSION_DEFAULT 1 1314 #define HAMMER2_VOL_VERSION_WIP 2 1315 1316 #define HAMMER2_NUM_VOLHDRS 4 1317 1318 union hammer2_media_data { 1319 hammer2_volume_data_t voldata; 1320 hammer2_inode_data_t ipdata; 1321 hammer2_blockset_t blkset; 1322 hammer2_blockref_t npdata[HAMMER2_IND_COUNT_MAX]; 1323 hammer2_bmap_data_t bmdata[HAMMER2_FREEMAP_COUNT]; 1324 char buf[HAMMER2_PBUFSIZE]; 1325 } __packed; 1326 1327 typedef union hammer2_media_data hammer2_media_data_t; 1328 1329 #endif /* !_VFS_HAMMER2_DISK_H_ */ 1330