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_KEY_RADIX 64 /* number of bits in key */
78 * MINALLOCSIZE - The minimum allocation size. This can be smaller
79 * or larger than the minimum physical IO size.
81 * NOTE: Should not be larger than 1K since inodes
84 * MINIOSIZE - The minimum IO size. This must be less than
85 * or equal to HAMMER2_PBUFSIZE.
87 * XXX currently must be set to MINALLOCSIZE until/if
88 * we deal with recursive buffer cache locks.
90 * HAMMER2_PBUFSIZE - Topological block size used by files for all
91 * blocks except the block straddling EOF.
93 * HAMMER2_SEGSIZE - Allocation map segment size, typically 2MB
96 #define HAMMER2_SEGSIZE (65536 * 8)
98 #define HAMMER2_PBUFRADIX 16 /* physical buf (1<<16) bytes */
99 #define HAMMER2_PBUFSIZE 65536
100 #define HAMMER2_LBUFRADIX 14 /* logical buf (1<<14) bytes */
101 #define HAMMER2_LBUFSIZE 16384
104 #define HAMMER2_MINIORADIX 16 /* minimum phsical IO size */
105 #define HAMMER2_MINIOSIZE 65536
107 #define HAMMER2_MINIORADIX HAMMER2_MINALLOCRADIX
108 #define HAMMER2_MINIOSIZE HAMMER2_MINALLOCSIZE
110 #define HAMMER2_MINALLOCRADIX 10 /* minimum block allocation size */
111 #define HAMMER2_MINALLOCSIZE 1024
112 #define HAMMER2_IND_BYTES_MIN 4096 /* first indirect layer only */
113 #define HAMMER2_IND_BYTES_MAX HAMMER2_PBUFSIZE
114 #define HAMMER2_IND_COUNT_MIN (HAMMER2_IND_BYTES_MIN / \
115 sizeof(hammer2_blockref_t))
116 #define HAMMER2_IND_COUNT_MAX (HAMMER2_IND_BYTES_MAX / \
117 sizeof(hammer2_blockref_t))
120 * HAMMER2 processes blockrefs in sets of 8. The set is fully associative,
121 * is not sorted, and may contain holes.
123 * A full indirect block supports 1024 blockrefs.
125 * An inode embeds one set of blockrefs but may also use the data area for
126 * up to 512 bytes of direct data.
128 #define HAMMER2_SET_COUNT 8 /* direct entries & associativity */
129 #define HAMMER2_SET_RADIX 3
130 #define HAMMER2_EMBEDDED_BYTES 512
131 #define HAMMER2_EMBEDDED_RADIX 9
133 #define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1)
134 #define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1)
135 #define HAMMER2_SEGMASK (HAMMER2_SEGSIZE - 1)
137 #define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK)
138 #define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE)
139 #define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK)
140 #define HAMMER2_SEGSIZE64 ((hammer2_off_t)HAMMER2_SEGSIZE)
141 #define HAMMER2_SEGMASK64 ((hammer2_off_t)HAMMER2_SEGMASK)
143 #define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5"
146 * A HAMMER2 filesystem is always sized in multiples of 8MB.
148 * A 4MB segment is reserved at the beginning of each 2GB zone. This segment
149 * contains the volume header, the free block table, and possibly other
150 * information in the future. 4MB = 64 x 64K blocks.
152 #define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024)
153 #define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN)
154 #define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1)
155 #define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK)
157 #define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN)
158 #define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN)
159 #define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1)
160 #define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK)
162 #define HAMMER2_RESERVE_BYTES64 (2LLU * 1024 * 1024 * 1024)
163 #define HAMMER2_RESERVE_MASK64 (HAMMER2_RESERVE_BYTES64 - 1)
164 #define HAMMER2_RESERVE_SEG (4 * 1024 * 1024)
165 #define HAMMER2_RESERVE_SEG64 ((hammer2_off_t)HAMMER2_RESERVE_SEG)
166 #define HAMMER2_RESERVE_BLOCKS (HAMMER2_RESERVE_SEG / HAMMER2_PBUFSIZE)
169 * Two linear areas can be reserved after the initial 2MB segment in the base
170 * zone (the one starting at offset 0). These areas are NOT managed by the
171 * block allocator and do not fall under HAMMER2 crc checking rules based
172 * at the volume header (but can be self-CRCd internally, depending).
174 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN
175 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024)
176 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024)
178 #define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN
179 #define HAMMER2_REDO_NOM_BYTES (256*1024*1024)
180 #define HAMMER2_REDO_MAX_BYTES (1024*1024*1024)
183 * Most HAMMER2 types are implemented as unsigned 64-bit integers.
184 * Transaction ids are monotonic.
186 * We utilize 32-bit iSCSI CRCs.
188 typedef uint64_t hammer2_tid_t;
189 typedef uint64_t hammer2_off_t;
190 typedef uint64_t hammer2_key_t;
191 typedef uint32_t hammer2_crc32_t;
194 * Miscellanious ranges (all are unsigned).
196 #define HAMMER2_MIN_TID 1ULL
197 #define HAMMER2_MAX_TID 0xFFFFFFFFFFFFFFFFULL
198 #define HAMMER2_MIN_KEY 0ULL
199 #define HAMMER2_MAX_KEY 0xFFFFFFFFFFFFFFFFULL
200 #define HAMMER2_MIN_OFFSET 0ULL
201 #define HAMMER2_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL
204 * HAMMER2 data offset special cases and masking.
206 * All HAMMER2 data offsets have to be broken down into a 64K buffer base
207 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO).
209 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits
210 * of the data offset field specifies how large the data chunk being pointed
211 * to as a power of 2. This value typically ranges from HAMMER2_MIN_RADIX
212 * to HAMMER2_MAX_RADIX (6-16). Larger values may be supported in the future
213 * to support file extents.
215 #define HAMMER2_OFF_BAD ((hammer2_off_t)-1)
216 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL
217 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64)
218 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64)
219 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL
220 #define HAMMER2_MAX_COPIES 6
223 * HAMMER2 directory support and pre-defined keys
225 #define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL
226 #define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL
227 #define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL
228 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL
229 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */
231 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */
234 * The media block reference structure. This forms the core of the HAMMER2
235 * media topology recursion. This 64-byte data structure is embedded in the
236 * volume header, in inodes (which are also directory entries), and in
239 * A blockref references a single media item, which typically can be a
240 * directory entry (aka inode), indirect block, or data block.
242 * The primary feature a blockref represents is the ability to validate
243 * the entire tree underneath it via its check code. Any modification to
244 * anything propagates up the blockref tree all the way to the root, replacing
245 * the related blocks. Propagations can shortcut to the volume root to
246 * implement the 'fast syncing' feature but this only delays the eventual
249 * The check code can be a simple 32-bit iscsi code, a 64-bit crc,
250 * or as complex as a 192 bit cryptographic hash. 192 bits is the maximum
251 * supported check code size, which is not sufficient for unverified dedup
252 * UNLESS one doesn't mind once-in-a-blue-moon data corruption (such as when
253 * farming web data). HAMMER2 has an unverified dedup feature for just this
256 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */
257 uint8_t type; /* type of underlying item */
258 uint8_t methods; /* check method & compression method */
259 uint8_t copyid; /* specify which copy this is */
260 uint8_t keybits; /* #of keybits masked off 0=leaf */
261 uint8_t vradix; /* virtual data/meta-data size */
262 uint8_t flags; /* blockref flags */
265 hammer2_key_t key; /* key specification */
266 hammer2_tid_t mirror_tid; /* propagate for mirror scan */
267 hammer2_tid_t modify_tid; /* modifications sans propagation */
268 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/
269 union { /* check info */
285 typedef struct hammer2_blockref hammer2_blockref_t;
287 #define HAMMER2_BREF_SYNC1 0x01 /* modification synchronized */
288 #define HAMMER2_BREF_SYNC2 0x02 /* modification committed */
289 #define HAMMER2_BREF_DESYNCCHLD 0x04 /* desynchronize children */
290 #define HAMMER2_BREF_DELETED 0x80 /* indicates a deletion */
292 #define HAMMER2_BLOCKREF_BYTES 64 /* blockref struct in bytes */
294 #define HAMMER2_BREF_TYPE_EMPTY 0
295 #define HAMMER2_BREF_TYPE_INODE 1
296 #define HAMMER2_BREF_TYPE_INDIRECT 2
297 #define HAMMER2_BREF_TYPE_DATA 3
298 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */
300 #define HAMMER2_ENC_COMPMETHOD(n) (n)
301 #define HAMMER2_ENC_CHECKMETHOD(n) ((n) << 4)
302 #define HAMMER2_DEC_COMPMETHOD(n) ((n) & 15)
303 #define HAMMER2_DEC_CHECKMETHOD(n) (((n) >> 4) & 15)
306 * HAMMER2 block references are collected into sets of 8 blockrefs. These
307 * sets are fully associative, meaning the elements making up a set are
308 * not sorted in any way and may contain duplicate entries, holes, or
309 * entries which shortcut multiple levels of indirection. Sets are used
312 * (1) When redundancy is desired a set may contain several duplicate
313 * entries pointing to different copies of the same data. Up to 8 copies
314 * are supported but the set structure becomes a bit inefficient once
317 * (2) The blockrefs in a set can shortcut multiple levels of indirections
318 * within the bounds imposed by the parent of set.
320 * When a set fills up another level of indirection is inserted, moving
321 * some or all of the set's contents into indirect blocks placed under the
322 * set. This is a top-down approach in that indirect blocks are not created
323 * until the set actually becomes full (that is, the entries in the set can
324 * shortcut the indirect blocks when the set is not full). Depending on how
325 * things are filled multiple indirect blocks will eventually be created.
327 struct hammer2_blockset {
328 hammer2_blockref_t blockref[HAMMER2_SET_COUNT];
331 typedef struct hammer2_blockset hammer2_blockset_t;
334 * Catch programmer snafus
336 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT
337 #error "hammer2 direct radix is incorrect"
339 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE
340 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent"
342 #if (1 << HAMMER2_MIN_RADIX) != HAMMER2_MIN_ALLOC
343 #error "HAMMER2_MIN_RADIX and HAMMER2_MIN_ALLOC are inconsistent"
347 * The media indirect block structure.
349 struct hammer2_indblock_data {
350 hammer2_blockref_t blockref[HAMMER2_IND_COUNT_MAX];
353 typedef struct hammer2_indblock_data hammer2_indblock_data_t;
356 * In HAMMER2 inodes ARE directory entries, with a special exception for
357 * hardlinks. The inode number is stored in the inode rather than being
358 * based on the location of the inode (since the location moves every time
359 * the inode or anything underneath the inode is modified).
361 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes
362 * for the filename, and 512 bytes worth of direct file data OR an embedded
365 * Directories represent one inode per blockref. Inodes are not laid out
366 * as a file but instead are represented by the related blockrefs. The
367 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember
368 * that blocksets are fully associative, so a certain degree efficiency is
369 * achieved just from that.
371 * Up to 512 bytes of direct data can be embedded in an inode, and since
372 * inodes are essentially directory entries this also means that small data
373 * files end up simply being laid out linearly in the directory, resulting
374 * in fewer seeks and highly optimal access.
376 * The compression mode can be changed at any time in the inode and is
377 * recorded on a blockref-by-blockref basis.
379 * Hardlinks are supported via the inode map. Essentially the way a hardlink
380 * works is that all individual directory entries representing the same file
381 * are special cased and specify the same inode number. The actual file
382 * is placed in the nearest parent directory that is parent to all instances
383 * of the hardlink. If all hardlinks to a file are in the same directory
384 * the actual file will also be placed in that directory. This file uses
385 * the inode number as the directory entry key and is invisible to normal
386 * directory scans. Real directory entry keys are differentiated from the
387 * inode number key via bit 63. Access to the hardlink silently looks up
388 * the real file and forwards all operations to that file. Removal of the
389 * last hardlink also removes the real file.
391 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */
392 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */
393 #define HAMMER2_INODE_VERSION_ONE 1
395 struct hammer2_inode_data {
396 uint16_t version; /* 0000 inode data version */
397 uint16_t reserved02; /* 0002 */
398 uint32_t uflags; /* 0004 chflags */
399 uint32_t rmajor; /* 0008 available for device nodes */
400 uint32_t rminor; /* 000C available for device nodes */
401 uint64_t ctime; /* 0010 inode change time */
402 uint64_t mtime; /* 0018 modified time */
403 uint64_t atime; /* 0020 access time (unsupported) */
404 uint64_t btime; /* 0028 birth time */
405 uuid_t uid; /* 0030 uid / degenerate unix uid */
406 uuid_t gid; /* 0040 gid / degenerate unix gid */
408 uint8_t type; /* 0050 object type */
409 uint8_t op_flags; /* 0051 operational flags */
410 uint16_t cap_flags; /* 0052 capability flags */
411 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */
413 hammer2_tid_t inum; /* 0058 inode number */
414 hammer2_off_t size; /* 0060 size of file */
415 uint64_t nlinks; /* 0068 hard links (typ only dirs) */
416 hammer2_tid_t iparent; /* 0070 parent inum (recovery only) */
417 uint64_t reserved78; /* 0078 */
419 hammer2_off_t data_quota; /* 0080 subtree quota in bytes */
420 hammer2_off_t data_count; /* 0088 subtree byte count */
421 hammer2_off_t inode_quota; /* 0090 subtree quota inode count */
422 hammer2_off_t inode_count; /* 0098 subtree inode count */
423 uint16_t name_len; /* 00A0 filename length */
424 uint8_t comp_algo; /* 00A2 compression request & algo */
425 uint8_t reservedA3; /* 00A3 */
426 uint32_t reservedA4; /* 00A4 */
427 hammer2_key_t name_key; /* 00A8 full filename key */
428 uint8_t copyids[8]; /* 00B0 request copies to (up to 8) */
429 uuid_t pfsid; /* 00B8 pfs uuid if PFSROOT */
430 uint64_t pfsinum; /* 00C8 pfs inum allocator */
431 uint64_t reservedD0; /* 00D0 */
432 uint64_t reservedD8; /* 00D8 */
433 uint64_t reservedE0; /* 00E0 */
434 uint64_t reservedE8; /* 00E8 */
435 uint64_t reservedF0; /* 00F0 */
436 uint64_t reservedF8; /* 00F8 */
438 unsigned char filename[HAMMER2_INODE_MAXNAME];
439 /* 0100-01FF (256 char, unterminated) */
440 union { /* 0200-03FF (64x8 = 512 bytes) */
441 struct hammer2_blockset blockset;
442 char data[HAMMER2_EMBEDDED_BYTES];
446 typedef struct hammer2_inode_data hammer2_inode_data_t;
448 #define HAMMER2_OPFLAG_DIRECTDATA 0x01
449 #define HAMMER2_OPFLAG_PFSROOT 0x02
451 #define HAMMER2_OBJTYPE_UNKNOWN 0
452 #define HAMMER2_OBJTYPE_DIRECTORY 1
453 #define HAMMER2_OBJTYPE_REGFILE 2
454 #define HAMMER2_OBJTYPE_FIFO 4
455 #define HAMMER2_OBJTYPE_CDEV 5
456 #define HAMMER2_OBJTYPE_BDEV 6
457 #define HAMMER2_OBJTYPE_SOFTLINK 7
458 #define HAMMER2_OBJTYPE_HARDLINK 8 /* dummy entry for hardlink */
459 #define HAMMER2_OBJTYPE_SOCKET 9
460 #define HAMMER2_OBJTYPE_WHITEOUT 10
462 #define HAMMER2_COPYID_NONE 0
463 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1)
465 #define HAMMER2_COMP_NONE 0
466 #define HAMMER2_COMP_AUTOZERO 1
468 #define HAMMER2_CHECK_NONE 0
469 #define HAMMER2_CHECK_ICRC 1
472 * The allocref structure represents the allocation table. One 64K block
473 * is broken down into 4096 x 16 byte entries. Each indirect block chops
474 * 11 bits off the 64-bit storage space, with leaf entries representing
475 * 64KB blocks. So: (12, 12, 12, 12, 16) = 64 bit storage space.
477 * Each 64K freemap block breaks the 4096 entries into a 64x64 tree with
478 * big_hint1 representing the top level every 64th entry and big_hint2
479 * representing the lower level in each entry. These fields specify the
480 * largest contiguous radix (1-63) available for allocation in the related
481 * sub-tree. The largest contiguous radix available for the entire block
482 * is saved in the parent (for the root this will be alloc_blockref in the
483 * volume header). The hints may be larger than actual and will be corrected
484 * on the fly but must not be smaller. The allocator uses the hints to
485 * very quickly locate nearby blocks of the desired size.
487 * In indirect blocks the 64-bit free[_or_mask] field stores the total free
488 * space for each of the 4096 sub-nodes in bytes. The total free space
489 * represented by the indirect block is stored in its parent.
491 * Each leaf element represents a 64K block. A bitmap replaces the free space
492 * count, giving us a 1KB allocation resolution. A micro-allocation append
493 * offset replaces the icrc field. The micro-allocation feature is not
494 * currently implemented and the field will be set to 65536.
496 * The allocation map uses reserved blocks so no data block reference is
497 * required, only a bit in the flags field to specify which of two possible
498 * reserved blocks to use. This allows the allocation map to be flushed to
499 * disk with minimal synchronization.
501 struct hammer2_allocref {
502 uint32_t icrc_or_app; /* node: icrc, leaf: append offset */
504 uint8_t big_hint1; /* upper level hint */
505 uint8_t big_hint2; /* lower level hint */
506 uint64_t free_or_mask; /* node: free bytes, leaf: bitmask */
509 typedef struct hammer2_allocref hammer2_allocref_t;
512 * WARNING - allocref size x entries must equate to the hammer buffer size,
513 * and 12 bits per recursion is assumed by the allocator.
515 * ALTA-D Since no data_offset is specified flags are needed to select
516 * which sub-block to recurse down into for root & internal nodes.
517 * (only ALTA and ALTB is currently supported).
519 * LEAF Terminal entry, always set for leafs. May be used to support
520 * 4MB extent allocations and early termination in the future.
521 * (not required to shortcut allocation scans as the big_hint1/2
522 * fields are used for this).
524 #define HAMMER2_ALLOCREF_BYTES 16 /* structure size */
525 #define HAMMER2_ALLOCREF_ENTRIES 4096 /* entries */
526 #define HAMMER2_ALLOCREF_RADIX 12 /* log2(entries) */
528 #if (HAMMER2_ALLOCREF_BYTES * HAMMER2_ALLOCREF_ENTRIES) != HAMMER2_PBUFSIZE
529 #error "allocref parameters do not fit in hammer buffer"
531 #if (1 << HAMMER2_ALLOCREF_RADIX) != HAMMER2_ALLOCREF_ENTRIES
532 #error "allocref parameters are inconsistent"
535 #define HAMMER2_ALLOCREF_ALTMASK 0x0003 /* select block for recurse */
536 #define HAMMER2_ALLOCREF_ALTA 0x0000
537 #define HAMMER2_ALLOCREF_ALTB 0x0001
538 #define HAMMER2_ALLOCREF_ALTC 0x0002 /* unsupported */
539 #define HAMMER2_ALLOCREF_ALTD 0x0003 /* unsupported */
540 #define HAMMER2_ALLOCREF_LEAF 0x0004
543 * Copies information stored in the volume header. Typically formatted
544 * e.g. like 'serno/A21343249.s1d'
546 * There are 8 copy_data[]'s in the volume header but up to 256 copyid's.
547 * When a copy is removed its copyid remains reserved in the copyid bitmap
548 * (copyexists[] bitmap in volume_data) until the copy references have
549 * been removed from the entire filesystem and cannot be reused until the
550 * removal is complete. However, new copy entries with other ids can be
551 * instantly added, replacing the original copy_data[]... which is fine as
552 * long as the copyid does not conflict.
554 * This structure must be exactly 64 bytes long.
556 struct hammer2_copy_data {
557 uint8_t copyid; /* 0-255 */
561 uint8_t path[60]; /* up to 59-char string, nul-terminated */
564 typedef struct hammer2_copy_data hammer2_copy_data_t;
566 #define COPYDATAF_OUTOFSYNC 0x0001
569 * The volume header eats a 64K block. There is currently an issue where
570 * we want to try to fit all nominal filesystem updates in a 512-byte section
571 * but it may be a lost cause due to the need for a blockset.
573 * All information is stored in host byte order. The volume header's magic
574 * number may be checked to determine the byte order. If you wish to mount
575 * between machines w/ different endian modes you'll need filesystem code
576 * which acts on the media data consistently (either all one way or all the
577 * other). Our code currently does not do that.
579 * A read-write mount may have to recover missing allocations by doing an
580 * incremental mirror scan looking for modifications made after alloc_tid.
581 * If alloc_tid == last_tid then no recovery operation is needed. Recovery
582 * operations are usually very, very fast.
584 * Read-only mounts do not need to do any recovery, access to the filesystem
585 * topology is always consistent after a crash (is always consistent, period).
586 * However, there may be shortcutted blockref updates present from deep in
587 * the tree which are stored in the volumeh eader and must be tracked on
590 * COPIES: Multiple copies may be specified on the mount line AND/OR you
591 * just specify one and the mount code tries to pick up the others
592 * from copyinfo[]. The copyid field in the volume header along
593 * with the fsid validates the copies.
595 * NOTE: root_blockref points to the super-root directory, not the root
596 * directory. The root directory will be a subdirectory under the
599 * The super-root directory contains all root directories and all
600 * snapshots (readonly or writable). It is possible to do a
601 * null-mount of the super-root using special path constructions
602 * relative to your mounted root.
604 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were
605 * a PFS, including mirroring and storage quota operations, and this is
606 * prefered over creating discrete PFSs in the super-root. Instead
607 * the super-root is most typically used to create writable snapshots,
608 * alternative roots, and so forth. The super-root is also used by
609 * the automatic snapshotting mechanism.
611 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU
612 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU
614 struct hammer2_volume_data {
618 uint64_t magic; /* 0000 Signature */
619 hammer2_off_t boot_beg; /* 0008 Boot area (future) */
620 hammer2_off_t boot_end; /* 0010 (size = end - beg) */
621 hammer2_off_t aux_beg; /* 0018 Aux area (future) */
622 hammer2_off_t aux_end; /* 0020 (size = end - beg) */
623 hammer2_off_t volu_size; /* 0028 Volume size, bytes */
625 uint32_t version; /* 0030 */
626 uint32_t flags; /* 0034 */
627 uint8_t copyid; /* 0038 copyid of phys vol */
628 uint8_t freemap_version; /* 0039 freemap algorithm */
629 uint8_t reserved003A; /* 003A */
630 uint8_t reserved003B; /* 003B */
631 uint32_t reserved003C; /* 003C */
633 uuid_t fsid; /* 0040 */
634 uuid_t fstype; /* 0050 */
637 * allocator_size is precalculated at newfs time and does not include
638 * reserved blocks, boot, or redo areas.
640 * Initial non-reserved-area allocations do not use the allocation
641 * map but instead adjust alloc_iterator. Dynamic allocations take
642 * over starting at (allocator_beg). This makes newfs_hammer2's
643 * job a lot easier and can also serve as a testing jig.
645 hammer2_off_t allocator_size; /* 0060 Total data space */
646 hammer2_off_t allocator_free; /* 0068 Free space */
647 hammer2_tid_t allocator_beg; /* 0070 Initial allocations */
648 hammer2_tid_t last_tid; /* 0078 Last transaction id */
649 hammer2_tid_t alloc_tid; /* 0080 Alloctable modify tid */
650 hammer2_blockref_t alloc_blockref; /* 0088-00C7 */
653 * Copyids are allocated dynamically from the copyexists bitmap.
654 * An id from the active copies set (up to 8, see copyinfo later on)
655 * may still exist after the copy set has been removed from the
656 * volume header and its bit will remain active in the bitmap and
657 * cannot be reused until it is 100% removed from the hierarchy.
659 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */
660 char reserved0140[248]; /* 00E8-01DF */
663 * 32 bit CRC array at the end of the first 512 byte sector.
665 * icrc_sects[7] - First 512-4 bytes of volume header (including all
666 * the other icrc's except the last one).
668 * icrc_sects[6] - Second 512-4 bytes of volume header, which is
669 * the blockset for the root.
671 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */
676 * The entire sector is used by a blockset.
678 hammer2_blockset_t sroot_blockset; /* 0200 Superroot directory */
681 * 512-byte sector #2-33
683 * Up to 256 copyinfo specifications can be configured. Note that
684 * any given subdirectory tree can only use 8 of the 256. Having
685 * up to 256 configurable in the volume header allows
687 * A specification takes 64 bytes. Each specification typically
688 * configures a device path such as 'serno/<serial>.s1d'.
690 struct hammer2_copy_data copyinfo[256]; /* 0400-43FF copyinfo config */
693 * Remaining sections are reserved for future use.
695 char reserved0400[0xBBFC]; /* 4400-FFFB reserved */
698 * icrc on entire volume header
700 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/
703 typedef struct hammer2_volume_data hammer2_volume_data_t;
706 * Various parts of the volume header have their own iCRCs.
708 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes
709 * and not included the icrc calculation.
711 * The second 512 bytes also has its own iCRC but it is stored in the first
712 * 512 bytes so it covers the entire second 512 bytes.
714 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes,
715 * which is where the iCRC for the whole volume is stored. This is currently
716 * a catch-all for anything not individually iCRCd.
718 #define HAMMER2_VOL_ICRC_SECT0 7
719 #define HAMMER2_VOL_ICRC_SECT1 6
721 #define HAMMER2_VOLUME_BYTES 65536
723 #define HAMMER2_VOLUME_ICRC0_OFF 0
724 #define HAMMER2_VOLUME_ICRC1_OFF 512
725 #define HAMMER2_VOLUME_ICRCVH_OFF 0
727 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4)
728 #define HAMMER2_VOLUME_ICRC1_SIZE (512)
729 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4)
731 #define HAMMER2_VOL_VERSION_MIN 1
732 #define HAMMER2_VOL_VERSION_DEFAULT 1
733 #define HAMMER2_VOL_VERSION_WIP 2
735 #define HAMMER2_NUM_VOLHDRS 4
737 union hammer2_media_data {
738 hammer2_inode_data_t ipdata;
739 hammer2_indblock_data_t npdata;
740 char buf[HAMMER2_PBUFSIZE];
743 typedef union hammer2_media_data hammer2_media_data_t;
746 * Prototypes for user & kernel functions. Kernel-only prototypes are
749 uint32_t hammer2_icrc32(const void *buf, size_t size);
750 uint32_t hammer2_icrc32c(const void *buf, size_t size, uint32_t crc);