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_
46 * The structures below represent the on-disk media structures for the HAMMER2
47 * filesystem. Note that all fields for on-disk structures are naturally
48 * aligned. The host endian format is typically used - compatibility is
49 * possible if the implementation detects reversed endian and adjusts accesses
52 * HAMMER2 primarily revolves around the directory topology: inodes,
53 * directory entries, and block tables. Block device buffer cache buffers
54 * are always 64KB. Logical file buffers are typically 16KB. All data
55 * references utilize 64-bit byte offsets.
57 * Free block management is handled independently using blocks reserved by
62 * The data at the end of a file or directory may be a fragment in order
63 * to optimize storage efficiency. The minimum fragment size is 1KB.
64 * Since allocations are in powers of 2 fragments must also be sized in
65 * powers of 2 (1024, 2048, ... 65536).
67 * For the moment the maximum allocation size is HAMMER2_PBUFSIZE (64K),
68 * which is 2^16. Larger extents may be supported in the future. Smaller
69 * fragments might be supported in the future (down to 64 bytes is possible),
70 * but probably will not be.
72 * A full indirect block use supports 1024 x 64-byte blockrefs in a 64KB
73 * buffer. Indirect blocks down to 1KB are supported to keep small
76 * A maximally sized file (2^64-1 bytes) requires 5 indirect block levels.
77 * The hammer2_blockset in the volume header or file inode has another 8
78 * entries, giving us 66+3 = 69 bits of address space. However, some bits
79 * are taken up by (potentially) requests for redundant copies. HAMMER2
80 * currently supports up to 8 copies, which brings the address space down
81 * to 66 bits and gives us 2 bits of leeway.
83 #define HAMMER2_MIN_ALLOC 1024 /* minimum allocation size */
84 #define HAMMER2_MIN_RADIX 10 /* minimum allocation size 2^N */
85 #define HAMMER2_MAX_ALLOC 65536 /* maximum allocation size */
86 #define HAMMER2_MAX_RADIX 16 /* maximum allocation size 2^N */
87 #define HAMMER2_KEY_RADIX 64 /* number of bits in key */
90 * MINALLOCSIZE - The minimum allocation size. This can be smaller
91 * or larger than the minimum physical IO size.
93 * NOTE: Should not be larger than 1K since inodes
96 * MINIOSIZE - The minimum IO size. This must be less than
97 * or equal to HAMMER2_PBUFSIZE.
99 * XXX currently must be set to MINALLOCSIZE until/if
100 * we deal with recursive buffer cache locks.
102 * HAMMER2_PBUFSIZE - Topological block size used by files for all
103 * blocks except the block straddling EOF.
105 * HAMMER2_SEGSIZE - Allocation map segment size, typically 2MB
108 #define HAMMER2_SEGSIZE (65536 * 8)
110 #define HAMMER2_PBUFRADIX 16 /* physical buf (1<<16) bytes */
111 #define HAMMER2_PBUFSIZE 65536
112 #define HAMMER2_LBUFRADIX 14 /* logical buf (1<<14) bytes */
113 #define HAMMER2_LBUFSIZE 16384
116 * XXX FIXME multiple locked inodes deadlock on the buffer cache
119 #define HAMMER2_MINIORADIX HAMMER2_LBUFRADIX
120 #define HAMMER2_MINIOSIZE HAMMER2_LBUFSIZE
122 #define HAMMER2_MINIORADIX 10
123 #define HAMMER2_MINIOSIZE 1024
126 #define HAMMER2_IND_BYTES_MIN 4096 /* first indirect layer only */
127 #define HAMMER2_IND_BYTES_MAX HAMMER2_PBUFSIZE
128 #define HAMMER2_IND_COUNT_MIN (HAMMER2_IND_BYTES_MIN / \
129 sizeof(hammer2_blockref_t))
130 #define HAMMER2_IND_COUNT_MAX (HAMMER2_IND_BYTES_MAX / \
131 sizeof(hammer2_blockref_t))
134 * In HAMMER2, arrays of blockrefs are fully set-associative, meaning that
135 * any element can occur at any index and holes can be anywhere. As a
136 * future optimization we will be able to flag that such arrays are sorted
137 * and thus optimize lookups, but for now we don't.
139 * Inodes embed either 512 bytes of direct data or an array of 8 blockrefs,
140 * resulting in highly efficient storage for files <= 512 bytes and for files
141 * <= 512KB. Up to 8 directory entries can be referenced from a directory
142 * without requiring an indirect block.
144 * Indirect blocks are typically either 4KB (64 blockrefs / ~4MB represented),
145 * or 64KB (1024 blockrefs / ~64MB represented).
147 #define HAMMER2_SET_COUNT 8 /* direct entries */
148 #define HAMMER2_SET_RADIX 3
149 #define HAMMER2_EMBEDDED_BYTES 512 /* inode blockset/dd size */
150 #define HAMMER2_EMBEDDED_RADIX 9
152 #define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1)
153 #define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1)
154 #define HAMMER2_SEGMASK (HAMMER2_SEGSIZE - 1)
156 #define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK)
157 #define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE)
158 #define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK)
159 #define HAMMER2_SEGSIZE64 ((hammer2_off_t)HAMMER2_SEGSIZE)
160 #define HAMMER2_SEGMASK64 ((hammer2_off_t)HAMMER2_SEGMASK)
162 #define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5"
165 * A HAMMER2 filesystem is always sized in multiples of 8MB.
167 * A 4MB segment is reserved at the beginning of each 2GB zone. This segment
168 * contains the volume header (or backup volume header), the free block
169 * table, and possibly other information in the future.
171 * 4MB = 64 x 64K blocks. Each 4MB segment is broken down as follows:
173 * +-----------------------+
174 * | Volume Hdr | block 0 volume header & alternates
175 * +-----------------------+ (first four zones only)
176 * | FreeBlk Section A | block 1-8
177 * +-----------------------+
178 * | FreeBlk Section B | block 9-16
179 * +-----------------------+
180 * | FreeBlk Section C | block 17-24
181 * +-----------------------+
182 * | FreeBlk Section D | block 25-32
183 * +-----------------------+
187 * +-----------------------+
189 * The first few 2GB zones contain volume headers and volume header backups.
190 * After that the volume header block# is reserved.
192 * The freemap utilizes blocks #1-32 for now, see the FREEMAP document.
193 * The Free block table has a resolution of 1KB
195 * WARNING! ZONE_SEG and VOLUME_ALIGN must be a multiple of 1<<LEVEL0_RADIX
196 * (i.e. a multiple of 2MB). VOLUME_ALIGN must be >= ZONE_SEG.
198 #define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024)
199 #define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN)
200 #define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1)
201 #define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK)
203 #define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN)
204 #define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN)
205 #define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1)
206 #define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK)
208 #define HAMMER2_ZONE_BYTES64 (2LLU * 1024 * 1024 * 1024)
209 #define HAMMER2_ZONE_MASK64 (HAMMER2_ZONE_BYTES64 - 1)
210 #define HAMMER2_ZONE_SEG (4 * 1024 * 1024)
211 #define HAMMER2_ZONE_SEG64 ((hammer2_off_t)HAMMER2_ZONE_SEG)
212 #define HAMMER2_ZONE_BLOCKS_SEG (HAMMER2_ZONE_SEG / HAMMER2_PBUFSIZE)
215 * 64 x 64KB blocks are reserved at the base of each 2GB zone. These blocks
216 * are used to store the volume header or volume header backups, allocation
217 * tree, and other information in the future.
219 * All specified blocks are not necessarily used in all 2GB zones. However,
220 * dead areas are reserved for future use and MUST NOT BE USED for other
223 * The freemap is arranged into four groups. Modifications rotate through
224 * the groups on a block by block basis (so all the blocks are not necessarily
225 * synchronized to the same group). Because the freemap is flushed
226 * independent of the main filesystem, the freemap only really needs two
227 * groups to operate efficiently.
232 #define HAMMER2_ZONE_VOLHDR 0 /* volume header or backup */
233 #define HAMMER2_ZONE_FREEMAP_A 1 /* freemap layer group A */
234 #define HAMMER2_ZONE_FREEMAP_B 9 /* freemap layer group B */
235 #define HAMMER2_ZONE_FREEMAP_C 17 /* freemap layer group C */
236 #define HAMMER2_ZONE_FREEMAP_D 25 /* freemap layer group D */
238 /* relative to FREEMAP_x */
239 #define HAMMER2_ZONEFM_LEVEL0 0 /* 256KB bitmap (4 blks) */
240 #define HAMMER2_ZONEFM_LEVEL1 4 /* 2GB indmap */
241 #define HAMMER2_ZONEFM_LEVEL2 5 /* 2TB indmap */
242 #define HAMMER2_ZONEFM_LEVEL3 6 /* 2PB indmap */
243 #define HAMMER2_ZONEFM_LEVEL4 7 /* 2EB indmap */
244 /* LEVEL5 is a set of 8 blockrefs in the volume header 16EB */
246 #define HAMMER2_ZONE_BLOCK49 49 /* future */
247 #define HAMMER2_ZONE_BLOCK50 50 /* future */
248 #define HAMMER2_ZONE_BLOCK51 51 /* future */
249 #define HAMMER2_ZONE_BLOCK52 52 /* future */
250 #define HAMMER2_ZONE_BLOCK53 53 /* future */
251 #define HAMMER2_ZONE_BLOCK54 54 /* future */
252 #define HAMMER2_ZONE_BLOCK55 55 /* future */
253 #define HAMMER2_ZONE_BLOCK56 56 /* future */
254 #define HAMMER2_ZONE_BLOCK57 57 /* future */
255 #define HAMMER2_ZONE_BLOCK58 58 /* future */
256 #define HAMMER2_ZONE_BLOCK59 59 /* future */
258 #define HAMMER2_ZONE_BLOCK60 60 /* future */
259 #define HAMMER2_ZONE_BLOCK61 61 /* future */
260 #define HAMMER2_ZONE_BLOCK62 62 /* future */
261 #define HAMMER2_ZONE_BLOCK63 63 /* future */
264 * Freemap radii. Please note that LEVEL 1 blockref array entries
265 * point to 256-byte sections of the bitmap representing 2MB of storage.
266 * Even though the chain structures represent only 256 bytes, they are
267 * mapped using larger 16K or 64K buffer cache buffers.
269 #define HAMMER2_FREEMAP_LEVEL5_RADIX 64 /* 16EB */
270 #define HAMMER2_FREEMAP_LEVEL4_RADIX 61 /* 2EB */
271 #define HAMMER2_FREEMAP_LEVEL3_RADIX 51 /* 2PB */
272 #define HAMMER2_FREEMAP_LEVEL2_RADIX 41 /* 2TB */
273 #define HAMMER2_FREEMAP_LEVEL1_RADIX 31 /* 2GB (256KB of bitmap) */
274 #define HAMMER2_FREEMAP_LEVEL0_RADIX 21 /* 2MB (256 bytes of bitmap) */
276 #define HAMMER2_FREEMAP_LEVELN_PSIZE 65536 /* physical bytes */
277 #define HAMMER2_FREEMAP_LEVEL0_PSIZE 256 /* physical bytes */
281 * Two linear areas can be reserved after the initial 2MB segment in the base
282 * zone (the one starting at offset 0). These areas are NOT managed by the
283 * block allocator and do not fall under HAMMER2 crc checking rules based
284 * at the volume header (but can be self-CRCd internally, depending).
286 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN
287 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024)
288 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024)
290 #define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN
291 #define HAMMER2_REDO_NOM_BYTES (256*1024*1024)
292 #define HAMMER2_REDO_MAX_BYTES (1024*1024*1024)
295 * Most HAMMER2 types are implemented as unsigned 64-bit integers.
296 * Transaction ids are monotonic.
298 * We utilize 32-bit iSCSI CRCs.
300 typedef uint64_t hammer2_tid_t;
301 typedef uint64_t hammer2_off_t;
302 typedef uint64_t hammer2_key_t;
303 typedef uint32_t hammer2_crc32_t;
306 * Miscellanious ranges (all are unsigned).
308 #define HAMMER2_MIN_TID 1ULL
309 #define HAMMER2_MAX_TID 0xFFFFFFFFFFFFFFFFULL
310 #define HAMMER2_MIN_KEY 0ULL
311 #define HAMMER2_MAX_KEY 0xFFFFFFFFFFFFFFFFULL
312 #define HAMMER2_MIN_OFFSET 0ULL
313 #define HAMMER2_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL
316 * HAMMER2 data offset special cases and masking.
318 * All HAMMER2 data offsets have to be broken down into a 64K buffer base
319 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO).
321 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits
322 * of the data offset field specifies how large the data chunk being pointed
323 * to as a power of 2. The theoretical minimum radix is thus 6 (The space
324 * needed in the low bits of the data offset field). However, the practical
325 * minimum allocation chunk size is 1KB (a radix of 10), so HAMMER2 sets
326 * HAMMER2_MIN_RADIX to 10. The maximum radix is currently 16 (64KB), but
327 * we fully intend to support larger extents in the future.
329 #define HAMMER2_OFF_BAD ((hammer2_off_t)-1)
330 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL
331 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64)
332 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64)
333 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL
334 #define HAMMER2_MAX_COPIES 6
337 * HAMMER2 directory support and pre-defined keys
339 #define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL
340 #define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL
341 #define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL
342 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL
343 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */
345 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */
348 * The media block reference structure. This forms the core of the HAMMER2
349 * media topology recursion. This 64-byte data structure is embedded in the
350 * volume header, in inodes (which are also directory entries), and in
353 * A blockref references a single media item, which typically can be a
354 * directory entry (aka inode), indirect block, or data block.
356 * The primary feature a blockref represents is the ability to validate
357 * the entire tree underneath it via its check code. Any modification to
358 * anything propagates up the blockref tree all the way to the root, replacing
359 * the related blocks. Propagations can shortcut to the volume root to
360 * implement the 'fast syncing' feature but this only delays the eventual
363 * The check code can be a simple 32-bit iscsi code, a 64-bit crc,
364 * or as complex as a 192 bit cryptographic hash. 192 bits is the maximum
365 * supported check code size, which is not sufficient for unverified dedup
366 * UNLESS one doesn't mind once-in-a-blue-moon data corruption (such as when
367 * farming web data). HAMMER2 has an unverified dedup feature for just this
372 * NOTE: The range of keys represented by the blockref is (key) to
373 * ((key) + (1LL << keybits) - 1). HAMMER2 usually populates
374 * blocks bottom-up, inserting a new root when radix expansion
377 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */
378 uint8_t type; /* type of underlying item */
379 uint8_t methods; /* check method & compression method */
380 uint8_t copyid; /* specify which copy this is */
381 uint8_t keybits; /* #of keybits masked off 0=leaf */
382 uint8_t vradix; /* virtual data/meta-data size */
383 uint8_t flags; /* blockref flags */
386 hammer2_key_t key; /* key specification */
387 hammer2_tid_t mirror_tid; /* propagate for mirror scan */
388 hammer2_tid_t modify_tid; /* modifications sans propagation */
389 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/
390 union { /* check info */
405 * Freemap hints are embedded in addition to the icrc32.
407 * biggest - largest possible allocation 2^N within sub-tree.
408 * typically initialized to 64 in freemap_blockref
409 * and reduced as-needed when a request fails.
411 * An allocation > 2^N is guaranteed to fail. An
412 * allocation <= 2^N MAY fail, and if it does the
413 * biggest hint will be adjusted downward.
415 * Used when allocating space.
423 uint64_t avail; /* total available bytes */
424 uint64_t unused; /* unused must be 0 */
429 typedef struct hammer2_blockref hammer2_blockref_t;
432 #define HAMMER2_BREF_SYNC1 0x01 /* modification synchronized */
433 #define HAMMER2_BREF_SYNC2 0x02 /* modification committed */
434 #define HAMMER2_BREF_DESYNCCHLD 0x04 /* desynchronize children */
435 #define HAMMER2_BREF_DELETED 0x80 /* indicates a deletion */
438 #define HAMMER2_BLOCKREF_BYTES 64 /* blockref struct in bytes */
440 #define HAMMER2_BREF_TYPE_EMPTY 0
441 #define HAMMER2_BREF_TYPE_INODE 1
442 #define HAMMER2_BREF_TYPE_INDIRECT 2
443 #define HAMMER2_BREF_TYPE_DATA 3
444 #define HAMMER2_BREF_TYPE_UNUSED04 4
445 #define HAMMER2_BREF_TYPE_FREEMAP_NODE 5
446 #define HAMMER2_BREF_TYPE_FREEMAP_LEAF 6
447 #define HAMMER2_BREF_TYPE_FREEMAP 254 /* pseudo-type */
448 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */
450 #define HAMMER2_ENC_CHECK(n) ((n) << 4)
451 #define HAMMER2_DEC_CHECK(n) (((n) >> 4) & 15)
453 #define HAMMER2_CHECK_NONE 0
454 #define HAMMER2_CHECK_ISCSI32 1
455 #define HAMMER2_CHECK_CRC64 2
456 #define HAMMER2_CHECK_SHA192 3
457 #define HAMMER2_CHECK_FREEMAP 4
459 #define HAMMER2_ENC_COMP(n) (n)
460 #define HAMMER2_DEC_COMP(n) ((n) & 15)
462 #define HAMMER2_COMP_NONE 0
463 #define HAMMER2_COMP_AUTOZERO 1
467 * HAMMER2 block references are collected into sets of 8 blockrefs. These
468 * sets are fully associative, meaning the elements making up a set are
469 * not sorted in any way and may contain duplicate entries, holes, or
470 * entries which shortcut multiple levels of indirection. Sets are used
473 * (1) When redundancy is desired a set may contain several duplicate
474 * entries pointing to different copies of the same data. Up to 8 copies
475 * are supported but the set structure becomes a bit inefficient once
478 * (2) The blockrefs in a set can shortcut multiple levels of indirections
479 * within the bounds imposed by the parent of set.
481 * When a set fills up another level of indirection is inserted, moving
482 * some or all of the set's contents into indirect blocks placed under the
483 * set. This is a top-down approach in that indirect blocks are not created
484 * until the set actually becomes full (that is, the entries in the set can
485 * shortcut the indirect blocks when the set is not full). Depending on how
486 * things are filled multiple indirect blocks will eventually be created.
488 * Indirect blocks are typically 4KB (64 entres) or 64KB (1024 entries) and
489 * are also treated as fully set-associative.
491 struct hammer2_blockset {
492 hammer2_blockref_t blockref[HAMMER2_SET_COUNT];
495 typedef struct hammer2_blockset hammer2_blockset_t;
498 * Catch programmer snafus
500 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT
501 #error "hammer2 direct radix is incorrect"
503 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE
504 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent"
506 #if (1 << HAMMER2_MIN_RADIX) != HAMMER2_MIN_ALLOC
507 #error "HAMMER2_MIN_RADIX and HAMMER2_MIN_ALLOC are inconsistent"
511 * The media indirect block structure.
513 struct hammer2_indblock_data {
514 hammer2_blockref_t blockref[HAMMER2_IND_COUNT_MAX];
517 typedef struct hammer2_indblock_data hammer2_indblock_data_t;
520 * In HAMMER2 inodes ARE directory entries, with a special exception for
521 * hardlinks. The inode number is stored in the inode rather than being
522 * based on the location of the inode (since the location moves every time
523 * the inode or anything underneath the inode is modified).
525 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes
526 * for the filename, and 512 bytes worth of direct file data OR an embedded
529 * Directories represent one inode per blockref. Inodes are not laid out
530 * as a file but instead are represented by the related blockrefs. The
531 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember
532 * that blocksets are fully associative, so a certain degree efficiency is
533 * achieved just from that.
535 * Up to 512 bytes of direct data can be embedded in an inode, and since
536 * inodes are essentially directory entries this also means that small data
537 * files end up simply being laid out linearly in the directory, resulting
538 * in fewer seeks and highly optimal access.
540 * The compression mode can be changed at any time in the inode and is
541 * recorded on a blockref-by-blockref basis.
543 * Hardlinks are supported via the inode map. Essentially the way a hardlink
544 * works is that all individual directory entries representing the same file
545 * are special cased and specify the same inode number. The actual file
546 * is placed in the nearest parent directory that is parent to all instances
547 * of the hardlink. If all hardlinks to a file are in the same directory
548 * the actual file will also be placed in that directory. This file uses
549 * the inode number as the directory entry key and is invisible to normal
550 * directory scans. Real directory entry keys are differentiated from the
551 * inode number key via bit 63. Access to the hardlink silently looks up
552 * the real file and forwards all operations to that file. Removal of the
553 * last hardlink also removes the real file.
555 * (attr_tid) is only updated when the inode's specific attributes or regular
556 * file size has changed, and affects path lookups and stat. (attr_tid)
557 * represents a special cache coherency lock under the inode. The inode
558 * blockref's modify_tid will always cover it.
560 * (dirent_tid) is only updated when an entry under a directory inode has
561 * been created, deleted, renamed, or had its attributes change, and affects
562 * directory lookups and scans. (dirent_tid) represents another special cache
563 * coherency lock under the inode. The inode blockref's modify_tid will
566 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */
567 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */
568 #define HAMMER2_INODE_VERSION_ONE 1
570 struct hammer2_inode_data {
571 uint16_t version; /* 0000 inode data version */
572 uint16_t reserved02; /* 0002 */
575 * core inode attributes, inode type, misc flags
577 uint32_t uflags; /* 0004 chflags */
578 uint32_t rmajor; /* 0008 available for device nodes */
579 uint32_t rminor; /* 000C available for device nodes */
580 uint64_t ctime; /* 0010 inode change time */
581 uint64_t mtime; /* 0018 modified time */
582 uint64_t atime; /* 0020 access time (unsupported) */
583 uint64_t btime; /* 0028 birth time */
584 uuid_t uid; /* 0030 uid / degenerate unix uid */
585 uuid_t gid; /* 0040 gid / degenerate unix gid */
587 uint8_t type; /* 0050 object type */
588 uint8_t op_flags; /* 0051 operational flags */
589 uint16_t cap_flags; /* 0052 capability flags */
590 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */
593 * inode size, identification, localized recursive configuration
594 * for compression and backup copies.
596 hammer2_tid_t inum; /* 0058 inode number */
597 hammer2_off_t size; /* 0060 size of file */
598 uint64_t nlinks; /* 0068 hard links (typ only dirs) */
599 hammer2_tid_t iparent; /* 0070 parent inum (recovery only) */
600 hammer2_key_t name_key; /* 0078 full filename key */
601 uint16_t name_len; /* 0080 filename length */
602 uint8_t ncopies; /* 0082 ncopies to local media */
603 uint8_t comp_algo; /* 0083 compression request & algo */
606 * These fields are currently only applicable to PFSROOTs.
608 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely
609 * identify an instance of a PFS in the cluster because
610 * a mount may contain more than one copy of the PFS as
611 * a separate node. {pfs_clid, pfs_fsid} must be used for
612 * registration in the cluster.
614 uint8_t target_type; /* 0084 hardlink target type */
615 uint8_t reserved85; /* 0085 */
616 uint8_t reserved86; /* 0086 */
617 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */
618 uint64_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */
619 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */
620 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */
623 * Quotas and cumulative sub-tree counters.
625 hammer2_off_t data_quota; /* 00B0 subtree quota in bytes */
626 hammer2_off_t data_count; /* 00B8 subtree byte count */
627 hammer2_off_t inode_quota; /* 00C0 subtree quota inode count */
628 hammer2_off_t inode_count; /* 00C8 subtree inode count */
629 hammer2_tid_t attr_tid; /* 00D0 attributes changed */
630 hammer2_tid_t dirent_tid; /* 00D8 directory/attr changed */
631 uint64_t reservedE0; /* 00E0 */
632 uint64_t reservedE8; /* 00E8 */
633 uint64_t reservedF0; /* 00F0 */
634 uint64_t reservedF8; /* 00F8 */
636 unsigned char filename[HAMMER2_INODE_MAXNAME];
637 /* 0100-01FF (256 char, unterminated) */
638 union { /* 0200-03FF (64x8 = 512 bytes) */
639 struct hammer2_blockset blockset;
640 char data[HAMMER2_EMBEDDED_BYTES];
644 typedef struct hammer2_inode_data hammer2_inode_data_t;
646 #define HAMMER2_OPFLAG_DIRECTDATA 0x01
647 #define HAMMER2_OPFLAG_PFSROOT 0x02
648 #define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */
650 #define HAMMER2_OBJTYPE_UNKNOWN 0
651 #define HAMMER2_OBJTYPE_DIRECTORY 1
652 #define HAMMER2_OBJTYPE_REGFILE 2
653 #define HAMMER2_OBJTYPE_FIFO 4
654 #define HAMMER2_OBJTYPE_CDEV 5
655 #define HAMMER2_OBJTYPE_BDEV 6
656 #define HAMMER2_OBJTYPE_SOFTLINK 7
657 #define HAMMER2_OBJTYPE_HARDLINK 8 /* dummy entry for hardlink */
658 #define HAMMER2_OBJTYPE_SOCKET 9
659 #define HAMMER2_OBJTYPE_WHITEOUT 10
661 #define HAMMER2_COPYID_NONE 0
662 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1)
665 * PEER types identify connections and help cluster controller filter
666 * out unwanted SPANs.
668 #define HAMMER2_PEER_NONE DMSG_PEER_NONE
669 #define HAMMER2_PEER_CLUSTER DMSG_PEER_CLUSTER
670 #define HAMMER2_PEER_BLOCK DMSG_PEER_BLOCK
671 #define HAMMER2_PEER_HAMMER2 DMSG_PEER_HAMMER2
673 #define HAMMER2_COPYID_COUNT DMSG_COPYID_COUNT
676 * PFS types identify a PFS on media and in LNK_SPAN messages.
678 #define HAMMER2_PFSTYPE_NONE DMSG_PFSTYPE_NONE
679 #define HAMMER2_PFSTYPE_ADMIN DMSG_PFSTYPE_ADMIN
680 #define HAMMER2_PFSTYPE_CLIENT DMSG_PFSTYPE_CLIENT
681 #define HAMMER2_PFSTYPE_CACHE DMSG_PFSTYPE_CACHE
682 #define HAMMER2_PFSTYPE_COPY DMSG_PFSTYPE_COPY
683 #define HAMMER2_PFSTYPE_SLAVE DMSG_PFSTYPE_SLAVE
684 #define HAMMER2_PFSTYPE_SOFT_SLAVE DMSG_PFSTYPE_SOFT_SLAVE
685 #define HAMMER2_PFSTYPE_SOFT_MASTER DMSG_PFSTYPE_SOFT_MASTER
686 #define HAMMER2_PFSTYPE_MASTER DMSG_PFSTYPE_MASTER
687 #define HAMMER2_PFSTYPE_SNAPSHOT DMSG_PFSTYPE_SNAPSHOT
688 #define HAMMER2_PFSTYPE_MAX DMSG_PFSTYPE_MAX
697 * Flags (8 bits) - blockref, for freemap only
699 * Note that the minimum chunk size is 1KB so we could theoretically have
700 * 10 bits here, but we might have some future extension that allows a
701 * chunk size down to 256 bytes and if so we will need bits 8 and 9.
703 #define HAMMER2_AVF_SELMASK 0x03 /* select group */
704 #define HAMMER2_AVF_ALL_ALLOC 0x04 /* indicate all allocated */
705 #define HAMMER2_AVF_ALL_FREE 0x08 /* indicate all free */
706 #define HAMMER2_AVF_RESERVED10 0x10
707 #define HAMMER2_AVF_RESERVED20 0x20
708 #define HAMMER2_AVF_RESERVED40 0x40
709 #define HAMMER2_AVF_RESERVED80 0x80
710 #define HAMMER2_AVF_AVMASK32 ((uint32_t)0xFFFFFF00LU)
711 #define HAMMER2_AVF_AVMASK64 ((uint64_t)0xFFFFFFFFFFFFFF00LLU)
713 #define HAMMER2_AV_SELECT_A 0x00
714 #define HAMMER2_AV_SELECT_B 0x01
715 #define HAMMER2_AV_SELECT_C 0x02
716 #define HAMMER2_AV_SELECT_D 0x03
719 * The volume header eats a 64K block. There is currently an issue where
720 * we want to try to fit all nominal filesystem updates in a 512-byte section
721 * but it may be a lost cause due to the need for a blockset.
723 * All information is stored in host byte order. The volume header's magic
724 * number may be checked to determine the byte order. If you wish to mount
725 * between machines w/ different endian modes you'll need filesystem code
726 * which acts on the media data consistently (either all one way or all the
727 * other). Our code currently does not do that.
729 * A read-write mount may have to recover missing allocations by doing an
730 * incremental mirror scan looking for modifications made after alloc_tid.
731 * If alloc_tid == last_tid then no recovery operation is needed. Recovery
732 * operations are usually very, very fast.
734 * Read-only mounts do not need to do any recovery, access to the filesystem
735 * topology is always consistent after a crash (is always consistent, period).
736 * However, there may be shortcutted blockref updates present from deep in
737 * the tree which are stored in the volumeh eader and must be tracked on
740 * NOTE: The copyinfo[] array contains the configuration for both the
741 * cluster connections and any local media copies. The volume
742 * header will be replicated for each local media copy.
744 * The mount command may specify multiple medias or just one and
745 * allow HAMMER2 to pick up the others when it checks the copyinfo[]
748 * NOTE: root_blockref points to the super-root directory, not the root
749 * directory. The root directory will be a subdirectory under the
752 * The super-root directory contains all root directories and all
753 * snapshots (readonly or writable). It is possible to do a
754 * null-mount of the super-root using special path constructions
755 * relative to your mounted root.
757 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were
758 * a PFS, including mirroring and storage quota operations, and this is
759 * prefered over creating discrete PFSs in the super-root. Instead
760 * the super-root is most typically used to create writable snapshots,
761 * alternative roots, and so forth. The super-root is also used by
762 * the automatic snapshotting mechanism.
764 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU
765 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU
767 struct hammer2_volume_data {
769 * sector #0 - 512 bytes
771 uint64_t magic; /* 0000 Signature */
772 hammer2_off_t boot_beg; /* 0008 Boot area (future) */
773 hammer2_off_t boot_end; /* 0010 (size = end - beg) */
774 hammer2_off_t aux_beg; /* 0018 Aux area (future) */
775 hammer2_off_t aux_end; /* 0020 (size = end - beg) */
776 hammer2_off_t volu_size; /* 0028 Volume size, bytes */
778 uint32_t version; /* 0030 */
779 uint32_t flags; /* 0034 */
780 uint8_t copyid; /* 0038 copyid of phys vol */
781 uint8_t freemap_version; /* 0039 freemap algorithm */
782 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */
783 uint8_t reserved003B; /* 003B */
784 uint32_t reserved003C; /* 003C */
786 uuid_t fsid; /* 0040 */
787 uuid_t fstype; /* 0050 */
790 * allocator_size is precalculated at newfs time and does not include
791 * reserved blocks, boot, or redo areas.
793 * Initial non-reserved-area allocations do not use the freemap
794 * but instead adjust alloc_iterator. Dynamic allocations take
795 * over starting at (allocator_beg). This makes newfs_hammer2's
796 * job a lot easier and can also serve as a testing jig.
798 hammer2_off_t allocator_size; /* 0060 Total data space */
799 hammer2_off_t allocator_free; /* 0068 Free space */
800 hammer2_off_t allocator_beg; /* 0070 Initial allocations */
801 hammer2_tid_t mirror_tid; /* 0078 best committed tid */
802 hammer2_tid_t alloc_tid; /* 0080 Alloctable modify tid */
803 hammer2_blockref_t reserved0088; /* 0088-00C7 */
806 * Copyids are allocated dynamically from the copyexists bitmap.
807 * An id from the active copies set (up to 8, see copyinfo later on)
808 * may still exist after the copy set has been removed from the
809 * volume header and its bit will remain active in the bitmap and
810 * cannot be reused until it is 100% removed from the hierarchy.
812 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */
813 char reserved0140[248]; /* 00E8-01DF */
816 * 32 bit CRC array at the end of the first 512 byte sector.
818 * icrc_sects[7] - First 512-4 bytes of volume header (including all
819 * the other icrc's except this one).
821 * icrc_sects[6] - Sector 1 (512 bytes) of volume header, which is
822 * the blockset for the root.
824 * icrc_sects[5] - Sector 2
825 * icrc_sects[4] - Sector 3
826 * icrc_sects[3] - Sector 4 (the freemap blockset)
828 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */
831 * sector #1 - 512 bytes
833 * The entire sector is used by a blockset.
835 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */
840 char sector2[512]; /* 0400-05FF reserved */
841 char sector3[512]; /* 0600-07FF reserved */
842 hammer2_blockset_t freemap_blockset; /* 0800-09FF freemap */
843 char sector5[512]; /* 0A00-0BFF reserved */
844 char sector6[512]; /* 0C00-0DFF reserved */
845 char sector7[512]; /* 0E00-0FFF reserved */
848 * sector #8-71 - 32768 bytes
850 * Contains the configuration for up to 256 copyinfo targets. These
851 * specify local and remote copies operating as masters or slaves.
852 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255
853 * indicates the local media).
855 * Each inode contains a set of up to 8 copyids, either inherited
856 * from its parent or explicitly specified in the inode, which
857 * indexes into this array.
859 /* 1000-8FFF copyinfo config */
860 dmsg_vol_data_t copyinfo[HAMMER2_COPYID_COUNT];
863 * Remaining sections are reserved for future use.
865 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */
868 * icrc on entire volume header
870 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/
873 typedef struct hammer2_volume_data hammer2_volume_data_t;
876 * Various parts of the volume header have their own iCRCs.
878 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes
879 * and not included the icrc calculation.
881 * The second 512 bytes also has its own iCRC but it is stored in the first
882 * 512 bytes so it covers the entire second 512 bytes.
884 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes,
885 * which is where the iCRC for the whole volume is stored. This is currently
886 * a catch-all for anything not individually iCRCd.
888 #define HAMMER2_VOL_ICRC_SECT0 7
889 #define HAMMER2_VOL_ICRC_SECT1 6
891 #define HAMMER2_VOLUME_BYTES 65536
893 #define HAMMER2_VOLUME_ICRC0_OFF 0
894 #define HAMMER2_VOLUME_ICRC1_OFF 512
895 #define HAMMER2_VOLUME_ICRCVH_OFF 0
897 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4)
898 #define HAMMER2_VOLUME_ICRC1_SIZE (512)
899 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4)
901 #define HAMMER2_VOL_VERSION_MIN 1
902 #define HAMMER2_VOL_VERSION_DEFAULT 1
903 #define HAMMER2_VOL_VERSION_WIP 2
905 #define HAMMER2_NUM_VOLHDRS 4
907 union hammer2_media_data {
908 hammer2_volume_data_t voldata;
909 hammer2_inode_data_t ipdata;
910 hammer2_indblock_data_t npdata;
911 char buf[HAMMER2_PBUFSIZE];
912 uint64_t bitmap[HAMMER2_FREEMAP_LEVEL0_PSIZE /
916 typedef union hammer2_media_data hammer2_media_data_t;