2 * Copyright (c) 2011-2012 The DragonFly Project. All rights reserved.
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>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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
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.
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
35 #ifndef VFS_HAMMER2_DISK_H_
36 #define VFS_HAMMER2_DISK_H_
39 * The structures below represent the on-disk media structures for the HAMMER2
40 * filesystem. Note that all fields for on-disk structures are naturally
41 * aligned. The host endian format is typically used - compatibility is
42 * possible if the implementation detects reversed endian and adjusts accesses
45 * HAMMER2 primarily revolves around the directory topology: inodes,
46 * directory entries, and block tables. Block device buffer cache buffers
47 * are always 64KB. Logical file buffers are typically 16KB. All data
48 * references utilize 64-bit byte offsets.
50 * Free block management is handled independently using blocks reserved by
55 * The data at the end of a file or directory may be a fragment in order
56 * to optimize storage efficiency. The minimum fragment size is 64 bytes.
57 * Since allocations are in powers of 2 fragments must also be sized in
58 * powers of 2 (64, 128, 256, ... 65536).
60 * For the moment the maximum allocation size is HAMMER2_PBUFSIZE (64K),
61 * which is 2^16. Larger extents may be supported in the future.
63 * A full indirect block uses supports 1024 x 64-byte blockrefs.
65 * A maximally sized file (2^64-1 bytes) requires 5 indirect block levels.
66 * The hammer2_blockset in the volume header or file inode has another 8
67 * entries, giving us 66+3 = 69 bits of address space. However, some bits
68 * are taken up by (potentially) requests for redundant copies. HAMMER2
69 * currently supports up to 8 copies, which brings the address space down
70 * to 66 bits and gives us 2 bits of leeway.
72 #define HAMMER2_MIN_ALLOC 64 /* minimum allocation size */
73 #define HAMMER2_MIN_RADIX 6 /* minimum allocation size 2^N */
74 #define HAMMER2_MAX_RADIX 16 /* maximum allocation size 2^N */
75 #define HAMMER2_IND1_RADIX 26 /* lowest full indirect block radix */
76 #define HAMMER2_IND2_RADIX 36
77 #define HAMMER2_IND3_RADIX 46
78 #define HAMMER2_IND4_RADIX 56
79 #define HAMMER2_IND5_RADIX 66 /* highest full indirect block radix */
82 * HAMMER2 utilizes 64K physical buffers and 16K logical filesystem buffers.
83 * The smaller logical filesystem buffers reduce ram waste when the OS is
84 * caching lots of small files.
86 #define HAMMER2_PBUFRADIX 16 /* physical buf (1<<16) bytes */
87 #define HAMMER2_PBUFSIZE 65536 /* fixed physical device buffer size */
88 #define HAMMER2_LBUFSIZE 16384 /* vnode/logical file buffer size */
91 * HAMMER2 processes blockrefs in sets of 8. The set is fully associative,
92 * is not sorted, and may contain holes.
94 * A full indirect block supports 1024 blockrefs.
96 * An inode embeds one set of blockrefs but may also use the data area for
97 * up to 512 bytes of direct data.
99 #define HAMMER2_SET_COUNT 8 /* direct entries & associativity */
100 #define HAMMER2_SET_RADIX 3
101 #define HAMMER2_IND_COUNT 1024 /* 1 << HAMMER2_IND_RADIX */
102 #define HAMMER2_IND_RADIX 10
103 #define HAMMER2_EMBEDDED_BYTES 512
104 #define HAMMER2_EMBEDDED_RADIX 9
106 #define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1)
107 #define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1)
109 #define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE)
110 #define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK)
111 #define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK)
113 #define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5"
116 * A HAMMER2 filesystem is always sized in multiples of 8MB.
118 * A 2MB segment is reserved at the beginning of each 2GB zone. This segment
119 * contains the volume header and the free block table.
121 #define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024)
122 #define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN)
123 #define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1)
124 #define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK)
126 #define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN)
127 #define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN)
128 #define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1)
129 #define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK)
131 #define HAMMER2_RESERVE_BYTES64 (2LLU * 1024 * 1024 * 1024)
132 #define HAMMER2_RESERVE_MASK64 (HAMMER2_RESERVE_BYTES64 - 1)
133 #define HAMMER2_RESERVE_SEG (2 * 1024 * 1024)
134 #define HAMMER2_RESERVE_SEG64 ((hammer2_off_t)HAMMER2_RESERVE_SEG)
135 #define HAMMER2_RESERVE_SEG_ENTRIES (HAMMER2_RESERVE_SEG/HAMMER2_BUFSIZE)
138 * Two linear areas can be reserved after the initial 2MB segment in the base
139 * zone (the one starting at offset 0). These areas are NOT managed by the
140 * block allocator and do not fall under HAMMER2 crc checking rules based
141 * at the volume header (but can be self-CRCd internally, depending).
143 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN
144 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024)
145 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024)
147 #define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN
148 #define HAMMER2_REDO_NOM_BYTES (256*1024*1024)
149 #define HAMMER2_REDO_MAX_BYTES (1024*1024*1024)
152 * Most HAMMER2 types are implemented as unsigned 64-bit integers.
153 * Transaction ids are monotonic.
155 * We utilize 32-bit iSCSI CRCs.
157 typedef uint64_t hammer2_tid_t;
158 typedef uint64_t hammer2_off_t;
159 typedef uint64_t hammer2_key_t;
160 typedef uint32_t hammer2_crc32_t;
163 * Miscellanious ranges (all are unsigned).
165 #define HAMMER2_MIN_TID 1ULL
166 #define HAMMER2_MAX_TID 0xFFFFFFFFFFFFFFFFULL
167 #define HAMMER2_MIN_KEY 0ULL
168 #define HAMMER2_MAX_KEY 0xFFFFFFFFFFFFFFFFULL
169 #define HAMMER2_MIN_OFFSET 0ULL
170 #define HAMMER2_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL
173 * HAMMER2 data offset special cases and masking.
175 * All HAMMER2 data offsets have to be broken down into a 64K buffer base
176 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO).
178 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits
179 * of the data offset field specifies how large the data chunk being pointed
180 * to as a power of 2. This value typically ranges from HAMMER2_MIN_RADIX
181 * to HAMMER2_MAX_RADIX (6-16). Larger values may be supported in the future
182 * to support file extents.
184 #define HAMMER2_OFF_BAD ((hammer2_off_t)-1)
185 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL
186 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64)
187 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64)
188 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL
189 #define HAMMER2_MAX_COPIES 6
192 * The media block reference structure. This forms the core of the HAMMER2
193 * media topology recursion. This 64-byte data structure is embedded in the
194 * volume header, in inodes (which are also directory entries), and in
197 * A blockref references a single media item, which typically can be a
198 * directory entry (aka inode), indirect block, or data block.
200 * The primary feature a blockref represents is the ability to validate
201 * the entire tree underneath it via its check code. Any modification to
202 * anything propagates up the blockref tree all the way to the root, replacing
203 * the related blocks. Propagations can shortcut to the volume root to
204 * implement the 'fast syncing' feature but this only delays the eventual
207 * The check code can be a simple 32-bit iscsi code, a 64-bit crc,
208 * or as complex as a 192 bit cryptographic hash. 192 bits is the maximum
209 * supported check code size, which is not sufficient for unverified dedup
210 * UNLESS one doesn't mind once-in-a-blue-moon data corruption (such as when
211 * farming web data). HAMMER2 has an unverified dedup feature for just this
214 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */
215 uint8_t type; /* type of underlying item */
216 uint8_t methods; /* check method & compression method */
217 uint8_t copyid; /* specify which copy this is */
218 uint8_t keybits; /* key mask bits for recursion */
219 uint8_t vradix; /* virtual data/meta-data size */
220 uint8_t flags; /* blockref flags */
223 hammer2_key_t key; /* key specification */
224 hammer2_tid_t mirror_tid; /* propagate for mirror scan */
225 hammer2_tid_t modify_tid; /* modifications sans propagation */
226 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/
227 union { /* check info */
243 typedef struct hammer2_blockref hammer2_blockref_t;
245 #define HAMMER2_BREF_SYNC1 0x01 /* modification synchronized */
246 #define HAMMER2_BREF_SYNC2 0x02 /* modification committed */
247 #define HAMMER2_BREF_DESYNCCHLD 0x04 /* children must be desynchronized */
248 #define HAMMER2_BREF_DELETED 0x80 /* indicates a deletion */
250 #define HAMMER2_BLOCKREF_BYTES 64 /* blockref struct in bytes */
251 #define HAMMER2_ENC_COMPMETHOD(n) (n)
252 #define HAMMER2_ENC_CHECKMETHOD(n) ((n) << 4)
253 #define HAMMER2_DEC_COMPMETHOD(n) ((n) & 15)
254 #define HAMMER2_DEC_CHECKMETHOD(n) (((n) >> 4) & 15)
257 * HAMMER2 block references are collected into sets of 8 blockrefs. These
258 * sets are fully associative, meaning the elements making up a set are
259 * not sorted in any way and may contain duplicate entries, holes, or
260 * entries which shortcut multiple levels of indirection. Sets are used
263 * (1) When redundancy is desired a set may contain several duplicate
264 * entries pointing to different copies of the same data. Up to 8 copies
265 * are supported but the set structure becomes a bit inefficient once
268 * (2) The blockrefs in a set can shortcut multiple levels of indirections
269 * within the bounds imposed by the parent of set.
271 * When a set fills up another level of indirection is inserted, moving
272 * some or all of the set's contents into indirect blocks placed under the
273 * set. This is a top-down approach in that indirect blocks are not created
274 * until the set actually becomes full (that is, the entries in the set can
275 * shortcut the indirect blocks when the set is not full). Depending on how
276 * things are filled multiple indirect blocks will eventually be created.
278 struct hammer2_blockset {
279 hammer2_blockref_t refs[HAMMER2_SET_COUNT];
283 * Catch programmer snafus
285 #if (1 << HAMMER2_IND_RADIX) != HAMMER2_IND_COUNT
286 #error "hammer2 indirect radix is incorrect"
288 #if (HAMMER2_IND_COUNT * 64) != HAMMER2_BUFSIZE
289 #error "hammer2 indirect entries is incorrect"
291 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT
292 #error "hammer2 direct radix is incorrect"
294 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE
295 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent"
297 #if (1 << HAMMER2_MIN_RADIX) != HAMMER2_MIN_ALLOC
298 #error "HAMMER2_MIN_RADIX and HAMMER2_MIN_ALLOC are inconsistent"
302 * The media indirect block structure.
304 struct hammer2_indblock {
305 hammer2_blockref_t blocks[HAMMER2_IND_COUNT];
308 typedef struct hammer2_indblock hammer2_indblock_t;
311 * In HAMMER2 inodes ARE directory entries, with a special exception for
312 * hardlinks. The inode number is stored in the inode rather than being
313 * based on the location of the inode (since the location moves every time
314 * the inode or anything underneath the inode is modified).
316 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes
317 * for the filename, and 512 bytes worth of direct file data OR an embedded
320 * Directories represent one inode per blockref. Inodes are not laid out
321 * as a file but instead are represented by the related blockrefs. The
322 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember
323 * that blocksets are fully associative, so a certain degree efficiency is
324 * achieved just from that.
326 * Up to 512 bytes of direct data can be embedded in an inode, and since
327 * inodes are essentially directory entries this also means that small data
328 * files end up simply being laid out linearly in the directory, resulting
329 * in fewer seeks and highly optimal access.
331 * The compression mode can be changed at any time in the inode and is
332 * recorded on a blockref-by-blockref basis.
334 * Hardlinks are supported via the inode map. Essentially the way a hardlink
335 * works is that all individual directory entries representing the same file
336 * are special cased and specify the same inode number. The actual file
337 * is placed in the nearest parent directory that is parent to all instances
338 * of the hardlink. If all hardlinks to a file are in the same directory
339 * the actual file will also be placed in that directory. This file uses
340 * the inode number as the directory entry key and is invisible to normal
341 * directory scans. Real directory entry keys are differentiated from the
342 * inode number key via bit 63. Access to the hardlink silently looks up
343 * the real file and forwards all operations to that file. Removal of the
344 * last hardlink also removes the real file.
346 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */
347 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */
348 #define HAMMER2_INODE_VERSION_ONE 1
350 struct hammer2_inode_data {
351 uint16_t version; /* 0000 inode data version */
352 uint16_t reserved02; /* 0002 */
353 uint32_t uflags; /* 0004 chflags */
354 uint32_t rmajor; /* 0008 available for device nodes */
355 uint32_t rminor; /* 000C available for device nodes */
356 uint64_t ctime; /* 0010 inode change time */
357 uint64_t mtime; /* 0018 modified time */
358 uint64_t atime; /* 0020 access time (unsupported) */
359 uint64_t btime; /* 0028 birth time */
360 uuid_t uid; /* 0030 uid / degenerate unix uid */
361 uuid_t gid; /* 0040 gid / degenerate unix gid */
363 uint8_t type; /* 0050 object type */
364 uint8_t op_flags; /* 0051 operational flags */
365 uint16_t cap_flags; /* 0052 capability flags */
366 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */
368 hammer2_tid_t inum; /* 0058 inode number */
369 hammer2_off_t size; /* 0060 size of file */
370 uint64_t nlinks; /* 0068 hard links (typ only dirs) */
371 hammer2_tid_t iparent; /* 0070 parent inum (recovery only) */
372 uint64_t reserved78; /* 0078 */
374 hammer2_off_t data_quota; /* 0080 subtree quota in bytes */
375 hammer2_off_t data_count; /* 0088 subtree byte count */
376 hammer2_off_t inode_quota; /* 0090 subtree quota inode count */
377 hammer2_off_t inode_count; /* 0098 subtree inode count */
378 uint16_t name_len; /* 00A0 filename length */
379 uint8_t comp_algo; /* 00A2 compression request & algo */
380 uint8_t reservedA3; /* 00A3 */
381 uint32_t reservedA4; /* 00A4 */
382 hammer2_key_t name_key; /* 00A8 full filename key */
383 uint8_t copyids[8]; /* 00B0 request copies to (up to 8) */
384 uint64_t reservedB8; /* 00B8 */
385 uint64_t reservedC0; /* 00C0 */
386 uint64_t reservedC8; /* 00C8 */
387 uint64_t reservedD0; /* 00D0 */
388 uint64_t reservedD8; /* 00D8 */
389 uint64_t reservedE0; /* 00E0 */
390 uint64_t reservedE8; /* 00E8 */
391 uint64_t reservedF0; /* 00F0 */
392 uint64_t reservedF8; /* 00F8 */
394 char filename[HAMMER_INODE_MAXNAME];
395 /* 0100-01FF (256 char, unterminated) */
396 union { /* 0200-03FF (64x8 = 512 bytes) */
397 struct hammer2_blockset blockset;
398 char data[HAMMER2_EMBEDDED_BYTES];
402 #define HAMMER2_OPFLAG_DIRECTDATA 0x01
404 #define HAMMER2_OBJTYPE_UNKNOWN 0
405 #define HAMMER2_OBJTYPE_DIRECTORY 1
406 #define HAMMER2_OBJTYPE_REGFILE 2
407 #define HAMMER2_OBJTYPE_FIFO 4
408 #define HAMMER2_OBJTYPE_CDEV 5
409 #define HAMMER2_OBJTYPE_BDEV 6
410 #define HAMMER2_OBJTYPE_SOFTLINK 7
411 #define HAMMER2_OBJTYPE_HARDLINK 8
412 #define HAMMER2_OBJTYPE_SOCKET 9
413 #define HAMMER2_OBJTYPE_WHITEOUT 10
417 * HAMMER2 special blocks, 128 64K buffers at the beginning of each 2GB segment.
419 #define HAMMER2_SPECBLOCK(n) (HAMMER2_PBUFSIZE64 * (n))
421 #define HAMMER2_SBLOCK_VOLHDR (0)
422 #define HAMMER2_SBLOCK_FREEMAP_ROOT(side) (1 + (8 * (side)))
423 #define HAMMER2_SBLOCK_FREEMAP_L1(side) (2 + (8 * (side)))
424 #define HAMMER2_SBLOCK_FREEMAP_L2(side) (3 + (8 * (side)))
425 #define HAMMER2_SBLOCK_FREEMAP_LEAF(side, n) (4 + (8 * (side)) + (n))
428 * The allocref structure represents the allocation table. One 64K block
429 * is broken down into 4096 x 16 byte entries. Each indirect block chops
430 * 11 bits off the 64-bit storage space, with leaf entries representing
431 * 64KB blocks. So: (12, 12, 12, 12, 16) = 64 bit storage space.
433 * Each 64K freemap block breaks the 4096 entries into a 64x64 tree with
434 * big_hint1 representing the top level every 64th entry and big_hint2
435 * representing the lower level in each entry. These fields specify the
436 * largest contiguous radix (1-63) available for allocation in the related
437 * sub-tree. The largest contiguous radix available for the entire block
438 * is saved in the parent (for the root this will be alloc_blockref in the
439 * volume header). The hints may be larger than actual and will be corrected
440 * on the fly but must not be smaller. The allocator uses the hints to
441 * very quickly locate nearby blocks of the desired size.
443 * In indirect blocks the 64-bit free[_or_mask] field stores the total free
444 * space for each of the 4096 sub-nodes in bytes. The total free space
445 * represented by the indirect block is stored in its parent.
447 * Each leaf element represents a 64K block. A bitmap replaces the free space
448 * count, giving us a 1KB allocation resolution. A micro-allocation append
449 * offset replaces the icrc field. The micro-allocation feature is not
450 * currently implemented and the field will be set to 65536.
452 * The allocation map uses reserved blocks so no data block reference is
453 * required, only a bit in the flags field to specify which of two possible
454 * reserved blocks to use. This allows the allocation map to be flushed to
455 * disk with minimal synchronization.
457 struct hammer2_allocref {
458 uint32_t icrc_or_app; /* node: icrc, leaf: append offset */
460 uint8_t big_hint1; /* upper level hint */
461 uint8_t big_hint2; /* lower level hint */
462 uint64_t free_or_mask; /* node: free bytes, leaf: bitmask */
465 typedef struct hammer2_allocref hammer2_allocref_t;
468 * WARNING - allocref size x entries must equate to the hammer buffer size,
469 * and 12 bits per recursion is assumed by the allocator.
471 * ALTA-D Since no data_offset is specified flags are needed to select
472 * which sub-block to recurse down into for root & internal nodes.
473 * (only ALTA and ALTB is currently supported).
475 * LEAF Terminal entry, always set for leafs. May be used to support
476 * 4MB extent allocations and early termination in the future.
477 * (not required to shortcut allocation scans as the big_hint1/2
478 * fields are used for this).
480 #define HAMMER2_ALLOCREF_BYTES 16 /* structure size */
481 #define HAMMER2_ALLOCREF_ENTRIES 4096 /* entries */
482 #define HAMMER2_ALLOCREF_RADIX 12 /* log2(entries) */
484 #if (HAMMER2_ALLOCREF_BYTES * HAMMER2_ALLOCREF_ENTRIES) != HAMMER2_BUFSIZE
485 #error "allocref parameters do not fit in hammer buffer"
487 #if (1 << HAMMER2_ALLOCREF_RADIX) != HAMMER2_ALLOCREF_ENTRIES
488 #error "allocref parameters are inconsistent"
491 #define HAMMER2_ALLOCREF_ALTMASK 0x0003 /* select block for recurse */
492 #define HAMMER2_ALLOCREF_ALTA 0x0000
493 #define HAMMER2_ALLOCREF_ALTB 0x0001
494 #define HAMMER2_ALLOCREF_ALTC 0x0002 /* unsupported */
495 #define HAMMER2_ALLOCREF_ALTD 0x0003 /* unsupported */
496 #define HAMMER2_ALLOCREF_LEAF 0x0004
501 * Copies information stored in the volume header. Typically formatted
502 * e.g. like 'serno/A21343249.s1d'
504 * There are 8 copy_data[]'s in the volume header but up to 256 copyid's.
505 * When a copy is removed its copyid remains reserved in the copyid bitmap
506 * (copyexists[] bitmap in volume_data) until the copy references have
507 * been removed from the entire filesystem and cannot be reused until the
508 * removal is complete. However, new copy entries with other ids can be
509 * instantly added, replacing the original copy_data[]... which is fine as
510 * long as the copyid does not conflict.
512 * This structure must be exactly 64 bytes long.
514 struct hammer2_copy_data {
515 uint8_t copyid; /* 0-255 */
519 uint8_t path[60]; /* up to 59-char string, nul-terminated */
522 typedef struct hammer2_copy_data hammer2_copy_data_t;
524 #define COPYDATAF_OUTOFSYNC 0x0001
527 * The volume header eats a 64K block. There is currently an issue where
528 * we want to try to fit all nominal filesystem updates in a 512-byte section
529 * but it may be a lost cause due to the need for a blockset.
531 * All information is stored in host byte order. The volume header's magic
532 * number may be checked to determine the byte order. If you wish to mount
533 * between machines w/ different endian modes you'll need filesystem code
534 * which acts on the media data consistently (either all one way or all the
535 * other). Our code currently does not do that.
537 * A read-write mount may have to recover missing allocations by doing an
538 * incremental mirror scan looking for modifications made after alloc_tid.
539 * If alloc_tid == last_tid then no recovery operation is needed. Recovery
540 * operations are usually very, very fast.
542 * Read-only mounts do not need to do any recovery, access to the filesystem
543 * topology is always consistent after a crash (is always consistent, period).
544 * However, there may be shortcutted blockref updates present from deep in
545 * the tree which are stored in the volumeh eader and must be tracked on
548 * icrc_sect0 only applies to the first 512-4 bytes in the volume header.
550 * COPIES: Multiple copies may be specified on the mount line AND/OR you
551 * just specify one and the mount code tries to pick up the others
552 * from copyinfo[]. The copyid field in the volume header along
553 * with the fsid validates the copies.
555 * NOTE: root_blockref points to the super-root directory, not the root
556 * directory. The root directory will be a subdirectory under the
559 * The super-root directory contains all root directories and all
560 * snapshots (readonly or writable). It is possible to do a
561 * null-mount of the super-root using special path constructions
562 * relative to your mounted root.
564 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were
565 * a PFS, including mirroring and storage quota operations, and this is
566 * prefered over creating discrete PFSs in the super-root. Instead
567 * the super-root is most typically used to create writable snapshots,
568 * alternative roots, and so forth. The super-root is also used by
569 * the automatic snapshotting mechanism.
571 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU
572 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU
574 struct hammer2_volume_data {
578 uint64_t magic; /* 0000 Signature */
579 hammer2_off_t boot_beg; /* 0008 Boot area (future) */
580 hammer2_off_t boot_end; /* 0010 (size = end - beg) */
581 hammer2_off_t redo_beg; /* 0018 Redo area (future) */
582 hammer2_off_t redo_end; /* 0020 (size = end - beg) */
583 hammer2_off_t volu_size; /* 0028 Volume size, bytes */
585 uint32_t version; /* 0030 */
586 uint32_t flags; /* 0034 */
587 uint8_t copyid; /* 0038 copyid of phys vol */
588 uint8_t freemap_version; /* 0039 freemap algorithm */
589 uint8_t reserved003A; /* 003A */
590 uint8_t reserved003B; /* 003B */
591 uint32_t reserved003C; /* 003C */
593 uuid_t fsid; /* 0040 */
594 uuid_t fstype; /* 0050 */
597 * allocator_size is precalculated at newfs time and does not include
598 * reserved blocks, boot, or redo areas.
600 * Initial non-reserved-area allocations do not use the allocation
601 * map but instead adjust alloc_iterator. Dynamic allocations take
602 * over starting at (allocator_beg). This makes newfs_hammer2's
603 * job a lot easier and can also serve as a testing jig.
605 hammer2_off_t allocator_size; /* 0060 Total data space */
606 hammer2_off_t allocator_free; /* 0068 Free space */
607 hammer2_tid_t allocator_beg; /* 0070 Initial allocations */
608 hammer2_tid_t last_tid; /* 0078 Last transaction id */
609 hammer2_tid_t alloc_tid; /* 0080 Alloctable modify tid */
610 hammer2_blockref_t alloc_blockref; /* 0088-00C7 */
613 * Copyids are allocated dynamically from the copyexists bitmap.
614 * An id from the active copies set (up to 8, see copyinfo later on)
615 * may still exist after the copy set has been removed from the
616 * volume header and its bit will remain active in the bitmap and
617 * cannot be reused until it is 100% removed from the hierarchy.
619 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */
620 char reserved0140[248]; /* 00E8-01DF */
623 * 32 bit CRC array at the end of the first 512 byte sector.
625 * icrc_sects[7] - First 512-4 bytes of volume header (including all
626 * the other icrc's except the last one).
628 * icrc_sects[6] - Second 512-4 bytes of volume header, which is
629 * the blockset for the root.
631 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */
636 * The entire sector is used by a blockset.
638 hammer2_blockset_t sroot_blockset; /* 0200 Superroot directory */
641 * 512-byte sector #2-33
643 * Up to 256 copyinfo specifications can be configured. Note that
644 * any given subdirectory tree can only use 8 of the 256. Having
645 * up to 256 configurable in the volume header allows
647 * A specification takes 64 bytes. Each specification typically
648 * configures a device path such as 'serno/<serial>.s1d'.
650 struct hammer2_copy_data copyinfo[256]; /* 0400-43FF copyinfo config */
653 * Remaining sections are reserved for future use.
655 char reserved0400[0xBBFC]; /* 4400-FFFB reserved */
658 * icrc on entire volume header
660 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/
664 * Section 0 and section 1 have their own iCRCs. Remaining icrc_sets[]
665 * entries are reserved for future use.
667 * icrc_volheader iCRCs the whole 64K volume header block and is catch-all
668 * for anything not individually iCRCd.
670 #define HAMMER2_VOL_ICRC_SECT0 7
671 #define HAMMER2_VOL_ICRC_SECT1 6
674 #define HAMMER2_VOLUME_BYTES 65536
675 #define HAMMER2_VOLUME_ICRCSIZE offsetof(hammer2_volume_data_t, icrc_sect0)
677 #define HAMMER2_VOL_VERSION_MIN 1
678 #define HAMMER2_VOL_VERSION_DEFAULT 1
679 #define HAMMER2_VOL_VERSION_WIP 2
681 #define HAMMER2_NUM_VOLHDRS 4
684 * Prototypes for user & kernel functions. Kernel-only prototypes are
687 uint32_t hammer2_icrc32(const void *buf, size_t size);
688 uint32_t hammer2_icrc32c(const void *buf, size_t size, uint32_t crc);