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 /* normal freemap rotation */ 275 #define HAMMER2_ZONE_FREEMAP_07 29 /* normal freemap rotation */ 276 #define HAMMER2_ZONE_FREEMAP_END 33 /* (non-inclusive) */ 277 278 #define HAMMER2_ZONE_UNUSED33 33 279 #define HAMMER2_ZONE_UNUSED34 34 280 #define HAMMER2_ZONE_UNUSED35 35 281 #define HAMMER2_ZONE_UNUSED36 36 282 #define HAMMER2_ZONE_UNUSED37 37 283 #define HAMMER2_ZONE_UNUSED38 38 284 #define HAMMER2_ZONE_UNUSED39 39 285 #define HAMMER2_ZONE_UNUSED40 40 286 #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 /* 2GB leafmap */ 315 #define HAMMER2_ZONEFM_LEVEL2 1 /* 2TB indmap */ 316 #define HAMMER2_ZONEFM_LEVEL3 2 /* 2PB indmap */ 317 #define HAMMER2_ZONEFM_LEVEL4 3 /* 2EB indmap */ 318 /* LEVEL5 is a set of 8 blockrefs in the volume header 16EB */ 319 320 /* 321 * Freemap radix. Note that the LEVEL 1 blockref points to a 64KB freemap 322 * block containing 1024 x LEVEL0 hammer2_bmap_data structures. LEVEL 0 323 * represents one structure. 324 */ 325 #define HAMMER2_FREEMAP_LEVEL5_RADIX 64 /* 16EB (end) */ 326 #define HAMMER2_FREEMAP_LEVEL4_RADIX 61 /* 2EB */ 327 #define HAMMER2_FREEMAP_LEVEL3_RADIX 51 /* 2PB */ 328 #define HAMMER2_FREEMAP_LEVEL2_RADIX 41 /* 2TB */ 329 #define HAMMER2_FREEMAP_LEVEL1_RADIX 31 /* 2GB */ 330 #define HAMMER2_FREEMAP_LEVEL0_RADIX 21 /* 2MB (entry in l-1 leaf) */ 331 332 #define HAMMER2_FREEMAP_LEVELN_PSIZE 65536 /* physical bytes */ 333 334 #define HAMMER2_FREEMAP_LEVEL4_SIZE ((hammer2_off_t)1 << \ 335 HAMMER2_FREEMAP_LEVEL4_RADIX) 336 #define HAMMER2_FREEMAP_LEVEL3_SIZE ((hammer2_off_t)1 << \ 337 HAMMER2_FREEMAP_LEVEL3_RADIX) 338 #define HAMMER2_FREEMAP_LEVEL2_SIZE ((hammer2_off_t)1 << \ 339 HAMMER2_FREEMAP_LEVEL2_RADIX) 340 #define HAMMER2_FREEMAP_LEVEL1_SIZE ((hammer2_off_t)1 << \ 341 HAMMER2_FREEMAP_LEVEL1_RADIX) 342 #define HAMMER2_FREEMAP_LEVEL0_SIZE ((hammer2_off_t)1 << \ 343 HAMMER2_FREEMAP_LEVEL0_RADIX) 344 345 #define HAMMER2_FREEMAP_LEVEL4_MASK (HAMMER2_FREEMAP_LEVEL4_SIZE - 1) 346 #define HAMMER2_FREEMAP_LEVEL3_MASK (HAMMER2_FREEMAP_LEVEL3_SIZE - 1) 347 #define HAMMER2_FREEMAP_LEVEL2_MASK (HAMMER2_FREEMAP_LEVEL2_SIZE - 1) 348 #define HAMMER2_FREEMAP_LEVEL1_MASK (HAMMER2_FREEMAP_LEVEL1_SIZE - 1) 349 #define HAMMER2_FREEMAP_LEVEL0_MASK (HAMMER2_FREEMAP_LEVEL0_SIZE - 1) 350 351 #define HAMMER2_FREEMAP_COUNT (int)(HAMMER2_FREEMAP_LEVELN_PSIZE / \ 352 sizeof(hammer2_bmap_data_t)) 353 #define HAMMER2_FREEMAP_BLOCK_RADIX 14 354 #define HAMMER2_FREEMAP_BLOCK_SIZE (1 << HAMMER2_FREEMAP_BLOCK_RADIX) 355 #define HAMMER2_FREEMAP_BLOCK_MASK (HAMMER2_FREEMAP_BLOCK_SIZE - 1) 356 357 /* 358 * bitmap[] structure. 2 bits per HAMMER2_FREEMAP_BLOCK_SIZE. Each bitmap[] 359 * element is 32 bits and thus represents 16 blocks (radix 4). 360 */ 361 #define HAMMER2_BMAP_INDEX_RADIX 4 362 #define HAMMER2_BMAP_INDEX_SIZE (HAMMER2_FREEMAP_BLOCK_SIZE * 16) 363 #define HAMMER2_BMAP_INDEX_MASK (HAMMER2_BMAP_INDEX_SIZE - 1) 364 365 /* 366 * Two linear areas can be reserved after the initial 2MB segment in the base 367 * zone (the one starting at offset 0). These areas are NOT managed by the 368 * block allocator and do not fall under HAMMER2 crc checking rules based 369 * at the volume header (but can be self-CRCd internally, depending). 370 */ 371 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN 372 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024) 373 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024) 374 375 #define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN 376 #define HAMMER2_REDO_NOM_BYTES (256*1024*1024) 377 #define HAMMER2_REDO_MAX_BYTES (1024*1024*1024) 378 379 /* 380 * Most HAMMER2 types are implemented as unsigned 64-bit integers. 381 * Transaction ids are monotonic. 382 * 383 * We utilize 32-bit iSCSI CRCs. 384 */ 385 typedef uint64_t hammer2_tid_t; 386 typedef uint64_t hammer2_off_t; 387 typedef uint64_t hammer2_key_t; 388 typedef uint32_t hammer2_crc32_t; 389 390 /* 391 * Miscellanious ranges (all are unsigned). 392 */ 393 #define HAMMER2_TID_MIN 1ULL 394 #define HAMMER2_TID_MAX 0xFFFFFFFFFFFFFFFFULL 395 #define HAMMER2_KEY_MIN 0ULL 396 #define HAMMER2_KEY_MAX 0xFFFFFFFFFFFFFFFFULL 397 #define HAMMER2_OFFSET_MIN 0ULL 398 #define HAMMER2_OFFSET_MAX 0xFFFFFFFFFFFFFFFFULL 399 400 /* 401 * HAMMER2 data offset special cases and masking. 402 * 403 * All HAMMER2 data offsets have to be broken down into a 64K buffer base 404 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO). 405 * 406 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits 407 * of the data offset field specifies how large the data chunk being pointed 408 * to as a power of 2. The theoretical minimum radix is thus 6 (The space 409 * needed in the low bits of the data offset field). However, the practical 410 * minimum allocation chunk size is 1KB (a radix of 10), so HAMMER2 sets 411 * HAMMER2_RADIX_MIN to 10. The maximum radix is currently 16 (64KB), but 412 * we fully intend to support larger extents in the future. 413 */ 414 #define HAMMER2_OFF_BAD ((hammer2_off_t)-1) 415 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL 416 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64) 417 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64) 418 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL 419 #define HAMMER2_MAX_COPIES 6 420 421 /* 422 * HAMMER2 directory support and pre-defined keys 423 */ 424 #define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL 425 #define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL 426 #define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL 427 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL 428 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */ 429 430 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */ 431 432 /************************************************************************ 433 * DMSG SUPPORT * 434 ************************************************************************ 435 * LNK_VOLCONF 436 * 437 * All HAMMER2 directories directly under the super-root on your local 438 * media can be mounted separately, even if they share the same physical 439 * device. 440 * 441 * When you do a HAMMER2 mount you are effectively tying into a HAMMER2 442 * cluster via local media. The local media does not have to participate 443 * in the cluster, other than to provide the hammer2_volconf[] array and 444 * root inode for the mount. 445 * 446 * This is important: The mount device path you specify serves to bootstrap 447 * your entry into the cluster, but your mount will make active connections 448 * to ALL copy elements in the hammer2_volconf[] array which match the 449 * PFSID of the directory in the super-root that you specified. The local 450 * media path does not have to be mentioned in this array but becomes part 451 * of the cluster based on its type and access rights. ALL ELEMENTS ARE 452 * TREATED ACCORDING TO TYPE NO MATTER WHICH ONE YOU MOUNT FROM. 453 * 454 * The actual cluster may be far larger than the elements you list in the 455 * hammer2_volconf[] array. You list only the elements you wish to 456 * directly connect to and you are able to access the rest of the cluster 457 * indirectly through those connections. 458 * 459 * WARNING! This structure must be exactly 128 bytes long for its config 460 * array to fit in the volume header. 461 */ 462 struct hammer2_volconf { 463 uint8_t copyid; /* 00 copyid 0-255 (must match slot) */ 464 uint8_t inprog; /* 01 operation in progress, or 0 */ 465 uint8_t chain_to; /* 02 operation chaining to, or 0 */ 466 uint8_t chain_from; /* 03 operation chaining from, or 0 */ 467 uint16_t flags; /* 04-05 flags field */ 468 uint8_t error; /* 06 last operational error */ 469 uint8_t priority; /* 07 priority and round-robin flag */ 470 uint8_t remote_pfs_type;/* 08 probed direct remote PFS type */ 471 uint8_t reserved08[23]; /* 09-1F */ 472 uuid_t pfs_clid; /* 20-2F copy target must match this uuid */ 473 uint8_t label[16]; /* 30-3F import/export label */ 474 uint8_t path[64]; /* 40-7F target specification string or key */ 475 }; 476 477 typedef struct hammer2_volconf hammer2_volconf_t; 478 479 #define DMSG_VOLF_ENABLED 0x0001 480 #define DMSG_VOLF_INPROG 0x0002 481 #define DMSG_VOLF_CONN_RR 0x80 /* round-robin at same priority */ 482 #define DMSG_VOLF_CONN_EF 0x40 /* media errors flagged */ 483 #define DMSG_VOLF_CONN_PRI 0x0F /* select priority 0-15 (15=best) */ 484 485 struct dmsg_lnk_hammer2_volconf { 486 dmsg_hdr_t head; 487 hammer2_volconf_t copy; /* copy spec */ 488 int32_t index; 489 int32_t unused01; 490 uuid_t mediaid; 491 int64_t reserved02[32]; 492 }; 493 494 typedef struct dmsg_lnk_hammer2_volconf dmsg_lnk_hammer2_volconf_t; 495 496 #define DMSG_LNK_HAMMER2_VOLCONF DMSG_LNK(DMSG_LNK_CMD_HAMMER2_VOLCONF, \ 497 dmsg_lnk_hammer2_volconf) 498 499 #define H2_LNK_VOLCONF(msg) ((dmsg_lnk_hammer2_volconf_t *)(msg)->any.buf) 500 501 /* 502 * The media block reference structure. This forms the core of the HAMMER2 503 * media topology recursion. This 64-byte data structure is embedded in the 504 * volume header, in inodes (which are also directory entries), and in 505 * indirect blocks. 506 * 507 * A blockref references a single media item, which typically can be a 508 * directory entry (aka inode), indirect block, or data block. 509 * 510 * The primary feature a blockref represents is the ability to validate 511 * the entire tree underneath it via its check code. Any modification to 512 * anything propagates up the blockref tree all the way to the root, replacing 513 * the related blocks. Propagations can shortcut to the volume root to 514 * implement the 'fast syncing' feature but this only delays the eventual 515 * propagation. 516 * 517 * The check code can be a simple 32-bit iscsi code, a 64-bit crc, 518 * or as complex as a 192 bit cryptographic hash. 192 bits is the maximum 519 * supported check code size, which is not sufficient for unverified dedup 520 * UNLESS one doesn't mind once-in-a-blue-moon data corruption (such as when 521 * farming web data). HAMMER2 has an unverified dedup feature for just this 522 * purpose. 523 * 524 * -- 525 * 526 * NOTE: The range of keys represented by the blockref is (key) to 527 * ((key) + (1LL << keybits) - 1). HAMMER2 usually populates 528 * blocks bottom-up, inserting a new root when radix expansion 529 * is required. 530 * 531 * -- 532 * FUTURE BLOCKREF EXPANSION 533 * 534 * In order to implement a 256-bit content addressable index we want to 535 * have a 256-bit key which essentially represents the cryptographic hash. 536 * (so, 64-bit key + 192-bit crypto-hash or 256-bit key-is-the-hash + 537 * 32-bit consistency check for indirect block layers). 538 * 539 * THIS IS POSSIBLE in a 64-byte blockref structure. Of course, any number 540 * of bits can be represented by sizing the blockref. For the purposes of 541 * HAMMER2 though my limit is 256 bits. Not only that, but it will be an 542 * optimal construction because H2 already uses a variably-sized radix to 543 * pack the blockrefs at each level. A 256-bit mechanic would allow us 544 * to implement a content-addressable index. 545 */ 546 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */ 547 uint8_t type; /* type of underlying item */ 548 uint8_t methods; /* check method & compression method */ 549 uint8_t copyid; /* specify which copy this is */ 550 uint8_t keybits; /* #of keybits masked off 0=leaf */ 551 uint8_t vradix; /* virtual data/meta-data size */ 552 uint8_t flags; /* blockref flags */ 553 uint8_t reserved06; 554 uint8_t reserved07; 555 hammer2_key_t key; /* key specification */ 556 hammer2_tid_t mirror_tid; /* media flush topology & freemap */ 557 hammer2_tid_t modify_tid; /* cluster level change / flush */ 558 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/ 559 union { /* check info */ 560 char buf[24]; 561 struct { 562 uint32_t value; 563 uint32_t unused[5]; 564 } iscsi32; 565 struct { 566 uint64_t value; 567 uint64_t unused[2]; 568 } crc64; 569 struct { 570 char data[24]; 571 } sha192; 572 573 /* 574 * Freemap hints are embedded in addition to the icrc32. 575 * 576 * bigmask - Radixes available for allocation (0-31). 577 * Heuristical (may be permissive but not 578 * restrictive). Typically only radix values 579 * 10-16 are used (i.e. (1<<10) through (1<<16)). 580 * 581 * avail - Total available space remaining, in bytes 582 */ 583 struct { 584 uint32_t icrc32; 585 uint32_t bigmask; /* available radixes */ 586 uint64_t avail; /* total available bytes */ 587 uint64_t unused; /* unused must be 0 */ 588 } freemap; 589 } check; 590 }; 591 592 typedef struct hammer2_blockref hammer2_blockref_t; 593 594 #define HAMMER2_BLOCKREF_BYTES 64 /* blockref struct in bytes */ 595 596 /* 597 * On-media and off-media blockref types. 598 * 599 * types >= 128 are pseudo values that should never be present on-media. 600 */ 601 #define HAMMER2_BREF_TYPE_EMPTY 0 602 #define HAMMER2_BREF_TYPE_INODE 1 603 #define HAMMER2_BREF_TYPE_INDIRECT 2 604 #define HAMMER2_BREF_TYPE_DATA 3 605 #define HAMMER2_BREF_TYPE_UNUSED04 4 606 #define HAMMER2_BREF_TYPE_FREEMAP_NODE 5 607 #define HAMMER2_BREF_TYPE_FREEMAP_LEAF 6 608 #define HAMMER2_BREF_TYPE_FREEMAP 254 /* pseudo-type */ 609 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */ 610 611 #define HAMMER2_BREF_FLAG_PFSROOT 0x01 /* see also related opflag */ 612 #define HAMMER2_BREF_FLAG_ZERO 0x02 613 614 /* 615 * Encode/decode check mode and compression mode for 616 * bref.methods. The compression level is not encoded in 617 * bref.methods. 618 */ 619 #define HAMMER2_ENC_CHECK(n) (((n) & 15) << 4) 620 #define HAMMER2_DEC_CHECK(n) (((n) >> 4) & 15) 621 #define HAMMER2_ENC_COMP(n) ((n) & 15) 622 #define HAMMER2_DEC_COMP(n) ((n) & 15) 623 624 #define HAMMER2_CHECK_NONE 0 625 #define HAMMER2_CHECK_DISABLED 1 626 #define HAMMER2_CHECK_ISCSI32 2 627 #define HAMMER2_CHECK_CRC64 3 628 #define HAMMER2_CHECK_SHA192 4 629 #define HAMMER2_CHECK_FREEMAP 5 630 631 /* user-specifiable check modes only */ 632 #define HAMMER2_CHECK_STRINGS { "none", "disabled", "crc32", \ 633 "crc64", "sha192" } 634 #define HAMMER2_CHECK_STRINGS_COUNT 5 635 636 /* 637 * Encode/decode check or compression algorithm request in 638 * ipdata->check_algo and ipdata->comp_algo. 639 */ 640 #define HAMMER2_ENC_ALGO(n) (n) 641 #define HAMMER2_DEC_ALGO(n) ((n) & 15) 642 #define HAMMER2_ENC_LEVEL(n) ((n) << 4) 643 #define HAMMER2_DEC_LEVEL(n) (((n) >> 4) & 15) 644 645 #define HAMMER2_COMP_NONE 0 646 #define HAMMER2_COMP_AUTOZERO 1 647 #define HAMMER2_COMP_LZ4 2 648 #define HAMMER2_COMP_ZLIB 3 649 650 #define HAMMER2_COMP_NEWFS_DEFAULT HAMMER2_COMP_LZ4 651 #define HAMMER2_COMP_STRINGS { "none", "autozero", "lz4", "zlib" } 652 #define HAMMER2_COMP_STRINGS_COUNT 4 653 654 655 /* 656 * HAMMER2 block references are collected into sets of 8 blockrefs. These 657 * sets are fully associative, meaning the elements making up a set are 658 * not sorted in any way and may contain duplicate entries, holes, or 659 * entries which shortcut multiple levels of indirection. Sets are used 660 * in various ways: 661 * 662 * (1) When redundancy is desired a set may contain several duplicate 663 * entries pointing to different copies of the same data. Up to 8 copies 664 * are supported but the set structure becomes a bit inefficient once 665 * you go over 4. 666 * 667 * (2) The blockrefs in a set can shortcut multiple levels of indirections 668 * within the bounds imposed by the parent of set. 669 * 670 * When a set fills up another level of indirection is inserted, moving 671 * some or all of the set's contents into indirect blocks placed under the 672 * set. This is a top-down approach in that indirect blocks are not created 673 * until the set actually becomes full (that is, the entries in the set can 674 * shortcut the indirect blocks when the set is not full). Depending on how 675 * things are filled multiple indirect blocks will eventually be created. 676 * 677 * Indirect blocks are typically 4KB (64 entres) or 64KB (1024 entries) and 678 * are also treated as fully set-associative. 679 */ 680 struct hammer2_blockset { 681 hammer2_blockref_t blockref[HAMMER2_SET_COUNT]; 682 }; 683 684 typedef struct hammer2_blockset hammer2_blockset_t; 685 686 /* 687 * Catch programmer snafus 688 */ 689 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT 690 #error "hammer2 direct radix is incorrect" 691 #endif 692 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE 693 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent" 694 #endif 695 #if (1 << HAMMER2_RADIX_MIN) != HAMMER2_ALLOC_MIN 696 #error "HAMMER2_RADIX_MIN and HAMMER2_ALLOC_MIN are inconsistent" 697 #endif 698 699 /* 700 * hammer2_bmap_data - A freemap entry in the LEVEL1 block. 701 * 702 * Each 64-byte entry contains the bitmap and meta-data required to manage 703 * a LEVEL0 (2MB) block of storage. The storage is managed in 128 x 16KB 704 * chunks. Smaller allocation granularity is supported via a linear iterator 705 * and/or must otherwise be tracked in ram. 706 * 707 * (data structure must be 64 bytes exactly) 708 * 709 * linear - A BYTE linear allocation offset used for sub-16KB allocations 710 * only. May contain values between 0 and 2MB. Must be ignored 711 * if 16KB-aligned (i.e. force bitmap scan), otherwise may be 712 * used to sub-allocate within the 16KB block (which is already 713 * marked as allocated in the bitmap). 714 * 715 * Sub-allocations need only be 1KB-aligned and do not have to be 716 * size-aligned, and 16KB or larger allocations do not update this 717 * field, resulting in pretty good packing. 718 * 719 * Please note that file data granularity may be limited by 720 * other issues such as buffer cache direct-mapping and the 721 * desire to support sector sizes up to 16KB (so H2 only issues 722 * I/O's in multiples of 16KB anyway). 723 * 724 * class - Clustering class. Cleared to 0 only if the entire leaf becomes 725 * free. Used to cluster device buffers so all elements must have 726 * the same device block size, but may mix logical sizes. 727 * 728 * Typically integrated with the blockref type in the upper 8 bits 729 * to localize inodes and indrect blocks, improving bulk free scans 730 * and directory scans. 731 * 732 * bitmap - Two bits per 16KB allocation block arranged in arrays of 733 * 32-bit elements, 128x2 bits representing ~2MB worth of media 734 * storage. Bit patterns are as follows: 735 * 736 * 00 Unallocated 737 * 01 (reserved) 738 * 10 Possibly free 739 * 11 Allocated 740 */ 741 struct hammer2_bmap_data { 742 int32_t linear; /* 00 linear sub-granular allocation offset */ 743 uint16_t class; /* 04-05 clustering class ((type<<8)|radix) */ 744 uint8_t reserved06; /* 06 */ 745 uint8_t reserved07; /* 07 */ 746 uint32_t reserved08; /* 08 */ 747 uint32_t reserved0C; /* 0C */ 748 uint32_t reserved10; /* 10 */ 749 uint32_t reserved14; /* 14 */ 750 uint32_t reserved18; /* 18 */ 751 uint32_t avail; /* 1C */ 752 uint32_t bitmap[8]; /* 20-3F 256 bits manages 2MB/16KB/2-bits */ 753 }; 754 755 typedef struct hammer2_bmap_data hammer2_bmap_data_t; 756 757 /* 758 * In HAMMER2 inodes ARE directory entries, with a special exception for 759 * hardlinks. The inode number is stored in the inode rather than being 760 * based on the location of the inode (since the location moves every time 761 * the inode or anything underneath the inode is modified). 762 * 763 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes 764 * for the filename, and 512 bytes worth of direct file data OR an embedded 765 * blockset. 766 * 767 * Directories represent one inode per blockref. Inodes are not laid out 768 * as a file but instead are represented by the related blockrefs. The 769 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember 770 * that blocksets are fully associative, so a certain degree efficiency is 771 * achieved just from that. 772 * 773 * Up to 512 bytes of direct data can be embedded in an inode, and since 774 * inodes are essentially directory entries this also means that small data 775 * files end up simply being laid out linearly in the directory, resulting 776 * in fewer seeks and highly optimal access. 777 * 778 * The compression mode can be changed at any time in the inode and is 779 * recorded on a blockref-by-blockref basis. 780 * 781 * Hardlinks are supported via the inode map. Essentially the way a hardlink 782 * works is that all individual directory entries representing the same file 783 * are special cased and specify the same inode number. The actual file 784 * is placed in the nearest parent directory that is parent to all instances 785 * of the hardlink. If all hardlinks to a file are in the same directory 786 * the actual file will also be placed in that directory. This file uses 787 * the inode number as the directory entry key and is invisible to normal 788 * directory scans. Real directory entry keys are differentiated from the 789 * inode number key via bit 63. Access to the hardlink silently looks up 790 * the real file and forwards all operations to that file. Removal of the 791 * last hardlink also removes the real file. 792 * 793 * (attr_tid) is only updated when the inode's specific attributes or regular 794 * file size has changed, and affects path lookups and stat. (attr_tid) 795 * represents a special cache coherency lock under the inode. The inode 796 * blockref's modify_tid will always cover it. 797 * 798 * (dirent_tid) is only updated when an entry under a directory inode has 799 * been created, deleted, renamed, or had its attributes change, and affects 800 * directory lookups and scans. (dirent_tid) represents another special cache 801 * coherency lock under the inode. The inode blockref's modify_tid will 802 * always cover it. 803 */ 804 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */ 805 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */ 806 #define HAMMER2_INODE_VERSION_ONE 1 807 808 #define HAMMER2_INODE_HIDDENDIR 16 /* special inode */ 809 #define HAMMER2_INODE_START 1024 /* dynamically allocated */ 810 811 struct hammer2_inode_data { 812 uint16_t version; /* 0000 inode data version */ 813 uint8_t reserved02; /* 0002 */ 814 uint8_t pfs_subtype; /* 0003 pfs sub-type */ 815 816 /* 817 * core inode attributes, inode type, misc flags 818 */ 819 uint32_t uflags; /* 0004 chflags */ 820 uint32_t rmajor; /* 0008 available for device nodes */ 821 uint32_t rminor; /* 000C available for device nodes */ 822 uint64_t ctime; /* 0010 inode change time */ 823 uint64_t mtime; /* 0018 modified time */ 824 uint64_t atime; /* 0020 access time (unsupported) */ 825 uint64_t btime; /* 0028 birth time */ 826 uuid_t uid; /* 0030 uid / degenerate unix uid */ 827 uuid_t gid; /* 0040 gid / degenerate unix gid */ 828 829 uint8_t type; /* 0050 object type */ 830 uint8_t op_flags; /* 0051 operational flags */ 831 uint16_t cap_flags; /* 0052 capability flags */ 832 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */ 833 834 /* 835 * inode size, identification, localized recursive configuration 836 * for compression and backup copies. 837 */ 838 hammer2_tid_t inum; /* 0058 inode number */ 839 hammer2_off_t size; /* 0060 size of file */ 840 uint64_t nlinks; /* 0068 hard links (typ only dirs) */ 841 hammer2_tid_t iparent; /* 0070 parent inum (recovery only) */ 842 hammer2_key_t name_key; /* 0078 full filename key */ 843 uint16_t name_len; /* 0080 filename length */ 844 uint8_t ncopies; /* 0082 ncopies to local media */ 845 uint8_t comp_algo; /* 0083 compression request & algo */ 846 847 /* 848 * These fields are currently only applicable to PFSROOTs. 849 * 850 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely 851 * identify an instance of a PFS in the cluster because 852 * a mount may contain more than one copy of the PFS as 853 * a separate node. {pfs_clid, pfs_fsid} must be used for 854 * registration in the cluster. 855 */ 856 uint8_t target_type; /* 0084 hardlink target type */ 857 uint8_t check_algo; /* 0085 check code request & algo */ 858 uint8_t pfs_nmasters; /* 0086 (if PFSROOT) if multi-master */ 859 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */ 860 uint64_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */ 861 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */ 862 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */ 863 864 /* 865 * Quotas and aggregate sub-tree inode and data counters. Note that 866 * quotas are not replicated downward, they are explicitly set by 867 * the sysop and in-memory structures keep track of inheritence. 868 */ 869 hammer2_key_t data_quota; /* 00B0 subtree quota in bytes */ 870 hammer2_key_t data_count; /* 00B8 subtree byte count */ 871 hammer2_key_t inode_quota; /* 00C0 subtree quota inode count */ 872 hammer2_key_t inode_count; /* 00C8 subtree inode count */ 873 hammer2_tid_t attr_tid; /* 00D0 attributes changed */ 874 hammer2_tid_t dirent_tid; /* 00D8 directory/attr changed */ 875 876 /* 877 * Tracks (possibly degenerate) free areas covering all sub-tree 878 * allocations under inode, not counting the inode itself. 879 * 0/0 indicates empty entry. fully set-associative. 880 * 881 * (not yet implemented) 882 */ 883 uint64_t decrypt_check; /* 00E0 decryption validator */ 884 hammer2_off_t reservedE0[3]; /* 00E8/F0/F8 */ 885 886 unsigned char filename[HAMMER2_INODE_MAXNAME]; 887 /* 0100-01FF (256 char, unterminated) */ 888 union { /* 0200-03FF (64x8 = 512 bytes) */ 889 struct hammer2_blockset blockset; 890 char data[HAMMER2_EMBEDDED_BYTES]; 891 } u; 892 }; 893 894 typedef struct hammer2_inode_data hammer2_inode_data_t; 895 896 #define HAMMER2_OPFLAG_DIRECTDATA 0x01 897 #define HAMMER2_OPFLAG_PFSROOT 0x02 /* (see also bref flag) */ 898 #define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */ 899 900 #define HAMMER2_OBJTYPE_UNKNOWN 0 901 #define HAMMER2_OBJTYPE_DIRECTORY 1 902 #define HAMMER2_OBJTYPE_REGFILE 2 903 #define HAMMER2_OBJTYPE_FIFO 4 904 #define HAMMER2_OBJTYPE_CDEV 5 905 #define HAMMER2_OBJTYPE_BDEV 6 906 #define HAMMER2_OBJTYPE_SOFTLINK 7 907 #define HAMMER2_OBJTYPE_HARDLINK 8 /* dummy entry for hardlink */ 908 #define HAMMER2_OBJTYPE_SOCKET 9 909 #define HAMMER2_OBJTYPE_WHITEOUT 10 910 911 #define HAMMER2_COPYID_NONE 0 912 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1) 913 914 #define HAMMER2_COPYID_COUNT 256 915 916 /* 917 * PFS types identify the role of a PFS within a cluster. The PFS types 918 * is stored on media and in LNK_SPAN messages and used in other places. 919 * 920 * The low 4 bits specify the current active type while the high 4 bits 921 * specify the transition target if the PFS is being upgraded or downgraded, 922 * If the upper 4 bits are not zero it may effect how a PFS is used during 923 * the transition. 924 * 925 * Generally speaking, downgrading a MASTER to a SLAVE cannot complete until 926 * at least all MASTERs have updated their pfs_nmasters field. And upgrading 927 * a SLAVE to a MASTER cannot complete until the new prospective master has 928 * been fully synchronized (though theoretically full synchronization is 929 * not required if a (new) quorum of other masters are fully synchronized). 930 * 931 * It generally does not matter which PFS element you actually mount, you 932 * are mounting 'the cluster'. So, for example, a network mount will mount 933 * a DUMMY PFS type on a memory filesystem. However, there are two exceptions. 934 * In order to gain the benefits of a SOFT_MASTER or SOFT_SLAVE, those PFSs 935 * must be directly mounted. 936 */ 937 #define HAMMER2_PFSTYPE_NONE 0x00 938 #define HAMMER2_PFSTYPE_CACHE 0x01 939 #define HAMMER2_PFSTYPE_UNUSED02 0x02 940 #define HAMMER2_PFSTYPE_SLAVE 0x03 941 #define HAMMER2_PFSTYPE_SOFT_SLAVE 0x04 942 #define HAMMER2_PFSTYPE_SOFT_MASTER 0x05 943 #define HAMMER2_PFSTYPE_MASTER 0x06 944 #define HAMMER2_PFSTYPE_UNUSED07 0x07 945 #define HAMMER2_PFSTYPE_SUPROOT 0x08 946 #define HAMMER2_PFSTYPE_DUMMY 0x09 947 #define HAMMER2_PFSTYPE_MAX 16 948 949 #define HAMMER2_PFSTRAN_NONE 0x00 /* no transition in progress */ 950 #define HAMMER2_PFSTRAN_CACHE 0x10 951 #define HAMMER2_PFSTRAN_UNMUSED20 0x20 952 #define HAMMER2_PFSTRAN_SLAVE 0x30 953 #define HAMMER2_PFSTRAN_SOFT_SLAVE 0x40 954 #define HAMMER2_PFSTRAN_SOFT_MASTER 0x50 955 #define HAMMER2_PFSTRAN_MASTER 0x60 956 #define HAMMER2_PFSTRAN_UNUSED70 0x70 957 #define HAMMER2_PFSTRAN_SUPROOT 0x80 958 #define HAMMER2_PFSTRAN_DUMMY 0x90 959 960 #define HAMMER2_PFS_DEC(n) ((n) & 0x0F) 961 #define HAMMER2_PFS_DEC_TRANSITION(n) (((n) >> 4) & 0x0F) 962 #define HAMMER2_PFS_ENC_TRANSITION(n) (((n) & 0x0F) << 4) 963 964 #define HAMMER2_PFSSUBTYPE_NONE 0 965 #define HAMMER2_PFSSUBTYPE_SNAPSHOT 1 /* manual/managed snapshot */ 966 #define HAMMER2_PFSSUBTYPE_AUTOSNAP 2 /* automatic snapshot */ 967 968 /* 969 * PFS mode of operation is a bitmask. This is typically not stored 970 * on-media, but defined here because the field may be used in dmsgs. 971 */ 972 #define HAMMER2_PFSMODE_QUORUM 0x01 973 #define HAMMER2_PFSMODE_RW 0x02 974 975 /* 976 * Allocation Table 977 * 978 */ 979 980 981 /* 982 * Flags (8 bits) - blockref, for freemap only 983 * 984 * Note that the minimum chunk size is 1KB so we could theoretically have 985 * 10 bits here, but we might have some future extension that allows a 986 * chunk size down to 256 bytes and if so we will need bits 8 and 9. 987 */ 988 #define HAMMER2_AVF_SELMASK 0x03 /* select group */ 989 #define HAMMER2_AVF_ALL_ALLOC 0x04 /* indicate all allocated */ 990 #define HAMMER2_AVF_ALL_FREE 0x08 /* indicate all free */ 991 #define HAMMER2_AVF_RESERVED10 0x10 992 #define HAMMER2_AVF_RESERVED20 0x20 993 #define HAMMER2_AVF_RESERVED40 0x40 994 #define HAMMER2_AVF_RESERVED80 0x80 995 #define HAMMER2_AVF_AVMASK32 ((uint32_t)0xFFFFFF00LU) 996 #define HAMMER2_AVF_AVMASK64 ((uint64_t)0xFFFFFFFFFFFFFF00LLU) 997 998 #define HAMMER2_AV_SELECT_A 0x00 999 #define HAMMER2_AV_SELECT_B 0x01 1000 #define HAMMER2_AV_SELECT_C 0x02 1001 #define HAMMER2_AV_SELECT_D 0x03 1002 1003 /* 1004 * The volume header eats a 64K block. There is currently an issue where 1005 * we want to try to fit all nominal filesystem updates in a 512-byte section 1006 * but it may be a lost cause due to the need for a blockset. 1007 * 1008 * All information is stored in host byte order. The volume header's magic 1009 * number may be checked to determine the byte order. If you wish to mount 1010 * between machines w/ different endian modes you'll need filesystem code 1011 * which acts on the media data consistently (either all one way or all the 1012 * other). Our code currently does not do that. 1013 * 1014 * A read-write mount may have to recover missing allocations by doing an 1015 * incremental mirror scan looking for modifications made after alloc_tid. 1016 * If alloc_tid == last_tid then no recovery operation is needed. Recovery 1017 * operations are usually very, very fast. 1018 * 1019 * Read-only mounts do not need to do any recovery, access to the filesystem 1020 * topology is always consistent after a crash (is always consistent, period). 1021 * However, there may be shortcutted blockref updates present from deep in 1022 * the tree which are stored in the volumeh eader and must be tracked on 1023 * the fly. 1024 * 1025 * NOTE: The copyinfo[] array contains the configuration for both the 1026 * cluster connections and any local media copies. The volume 1027 * header will be replicated for each local media copy. 1028 * 1029 * The mount command may specify multiple medias or just one and 1030 * allow HAMMER2 to pick up the others when it checks the copyinfo[] 1031 * array on mount. 1032 * 1033 * NOTE: root_blockref points to the super-root directory, not the root 1034 * directory. The root directory will be a subdirectory under the 1035 * super-root. 1036 * 1037 * The super-root directory contains all root directories and all 1038 * snapshots (readonly or writable). It is possible to do a 1039 * null-mount of the super-root using special path constructions 1040 * relative to your mounted root. 1041 * 1042 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were 1043 * a PFS, including mirroring and storage quota operations, and this is 1044 * prefered over creating discrete PFSs in the super-root. Instead 1045 * the super-root is most typically used to create writable snapshots, 1046 * alternative roots, and so forth. The super-root is also used by 1047 * the automatic snapshotting mechanism. 1048 */ 1049 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU 1050 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU 1051 1052 struct hammer2_volume_data { 1053 /* 1054 * sector #0 - 512 bytes 1055 */ 1056 uint64_t magic; /* 0000 Signature */ 1057 hammer2_off_t boot_beg; /* 0008 Boot area (future) */ 1058 hammer2_off_t boot_end; /* 0010 (size = end - beg) */ 1059 hammer2_off_t aux_beg; /* 0018 Aux area (future) */ 1060 hammer2_off_t aux_end; /* 0020 (size = end - beg) */ 1061 hammer2_off_t volu_size; /* 0028 Volume size, bytes */ 1062 1063 uint32_t version; /* 0030 */ 1064 uint32_t flags; /* 0034 */ 1065 uint8_t copyid; /* 0038 copyid of phys vol */ 1066 uint8_t freemap_version; /* 0039 freemap algorithm */ 1067 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */ 1068 uint8_t reserved003B; /* 003B */ 1069 uint32_t reserved003C; /* 003C */ 1070 1071 uuid_t fsid; /* 0040 */ 1072 uuid_t fstype; /* 0050 */ 1073 1074 /* 1075 * allocator_size is precalculated at newfs time and does not include 1076 * reserved blocks, boot, or redo areas. 1077 * 1078 * Initial non-reserved-area allocations do not use the freemap 1079 * but instead adjust alloc_iterator. Dynamic allocations take 1080 * over starting at (allocator_beg). This makes newfs_hammer2's 1081 * job a lot easier and can also serve as a testing jig. 1082 */ 1083 hammer2_off_t allocator_size; /* 0060 Total data space */ 1084 hammer2_off_t allocator_free; /* 0068 Free space */ 1085 hammer2_off_t allocator_beg; /* 0070 Initial allocations */ 1086 1087 /* 1088 * mirror_tid reflects the highest committed change for this 1089 * block device regardless of whether it is to the super-root 1090 * or to a PFS or whatever. 1091 * 1092 * freemap_tid reflects the highest committed freemap change for 1093 * this block device. 1094 */ 1095 hammer2_tid_t mirror_tid; /* 0078 committed tid (vol) */ 1096 hammer2_tid_t reserved0080; /* 0080 */ 1097 hammer2_tid_t reserved0088; /* 0088 */ 1098 hammer2_tid_t freemap_tid; /* 0090 committed tid (fmap) */ 1099 hammer2_tid_t bulkfree_tid; /* 0098 bulkfree incremental */ 1100 hammer2_tid_t reserved00A0[5]; /* 00A0-00C7 */ 1101 1102 /* 1103 * Copyids are allocated dynamically from the copyexists bitmap. 1104 * An id from the active copies set (up to 8, see copyinfo later on) 1105 * may still exist after the copy set has been removed from the 1106 * volume header and its bit will remain active in the bitmap and 1107 * cannot be reused until it is 100% removed from the hierarchy. 1108 */ 1109 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */ 1110 char reserved0140[248]; /* 00E8-01DF */ 1111 1112 /* 1113 * 32 bit CRC array at the end of the first 512 byte sector. 1114 * 1115 * icrc_sects[7] - First 512-4 bytes of volume header (including all 1116 * the other icrc's except this one). 1117 * 1118 * icrc_sects[6] - Sector 1 (512 bytes) of volume header, which is 1119 * the blockset for the root. 1120 * 1121 * icrc_sects[5] - Sector 2 1122 * icrc_sects[4] - Sector 3 1123 * icrc_sects[3] - Sector 4 (the freemap blockset) 1124 */ 1125 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */ 1126 1127 /* 1128 * sector #1 - 512 bytes 1129 * 1130 * The entire sector is used by a blockset. 1131 */ 1132 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */ 1133 1134 /* 1135 * sector #2-7 1136 */ 1137 char sector2[512]; /* 0400-05FF reserved */ 1138 char sector3[512]; /* 0600-07FF reserved */ 1139 hammer2_blockset_t freemap_blockset; /* 0800-09FF freemap */ 1140 char sector5[512]; /* 0A00-0BFF reserved */ 1141 char sector6[512]; /* 0C00-0DFF reserved */ 1142 char sector7[512]; /* 0E00-0FFF reserved */ 1143 1144 /* 1145 * sector #8-71 - 32768 bytes 1146 * 1147 * Contains the configuration for up to 256 copyinfo targets. These 1148 * specify local and remote copies operating as masters or slaves. 1149 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255 1150 * indicates the local media). 1151 * 1152 * Each inode contains a set of up to 8 copyids, either inherited 1153 * from its parent or explicitly specified in the inode, which 1154 * indexes into this array. 1155 */ 1156 /* 1000-8FFF copyinfo config */ 1157 hammer2_volconf_t copyinfo[HAMMER2_COPYID_COUNT]; 1158 1159 /* 1160 * Remaining sections are reserved for future use. 1161 */ 1162 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */ 1163 1164 /* 1165 * icrc on entire volume header 1166 */ 1167 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/ 1168 }; 1169 1170 typedef struct hammer2_volume_data hammer2_volume_data_t; 1171 1172 /* 1173 * Various parts of the volume header have their own iCRCs. 1174 * 1175 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes 1176 * and not included the icrc calculation. 1177 * 1178 * The second 512 bytes also has its own iCRC but it is stored in the first 1179 * 512 bytes so it covers the entire second 512 bytes. 1180 * 1181 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes, 1182 * which is where the iCRC for the whole volume is stored. This is currently 1183 * a catch-all for anything not individually iCRCd. 1184 */ 1185 #define HAMMER2_VOL_ICRC_SECT0 7 1186 #define HAMMER2_VOL_ICRC_SECT1 6 1187 1188 #define HAMMER2_VOLUME_BYTES 65536 1189 1190 #define HAMMER2_VOLUME_ICRC0_OFF 0 1191 #define HAMMER2_VOLUME_ICRC1_OFF 512 1192 #define HAMMER2_VOLUME_ICRCVH_OFF 0 1193 1194 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4) 1195 #define HAMMER2_VOLUME_ICRC1_SIZE (512) 1196 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4) 1197 1198 #define HAMMER2_VOL_VERSION_MIN 1 1199 #define HAMMER2_VOL_VERSION_DEFAULT 1 1200 #define HAMMER2_VOL_VERSION_WIP 2 1201 1202 #define HAMMER2_NUM_VOLHDRS 4 1203 1204 union hammer2_media_data { 1205 hammer2_volume_data_t voldata; 1206 hammer2_inode_data_t ipdata; 1207 hammer2_blockref_t npdata[HAMMER2_IND_COUNT_MAX]; 1208 hammer2_bmap_data_t bmdata[HAMMER2_FREEMAP_COUNT]; 1209 char buf[HAMMER2_PBUFSIZE]; 1210 }; 1211 1212 typedef union hammer2_media_data hammer2_media_data_t; 1213 1214 #endif /* !_VFS_HAMMER2_DISK_H_ */ 1215