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_
43 * The structures below represent the on-disk media structures for the HAMMER2
44 * filesystem. Note that all fields for on-disk structures are naturally
45 * aligned. The host endian format is typically used - compatibility is
46 * possible if the implementation detects reversed endian and adjusts accesses
49 * HAMMER2 primarily revolves around the directory topology: inodes,
50 * directory entries, and block tables. Block device buffer cache buffers
51 * are always 64KB. Logical file buffers are typically 16KB. All data
52 * references utilize 64-bit byte offsets.
54 * Free block management is handled independently using blocks reserved by
59 * The data at the end of a file or directory may be a fragment in order
60 * to optimize storage efficiency. The minimum fragment size is 64 bytes.
61 * Since allocations are in powers of 2 fragments must also be sized in
62 * powers of 2 (64, 128, 256, ... 65536).
64 * For the moment the maximum allocation size is HAMMER2_PBUFSIZE (64K),
65 * which is 2^16. Larger extents may be supported in the future.
67 * A full indirect block uses supports 1024 x 64-byte blockrefs.
69 * A maximally sized file (2^64-1 bytes) requires 5 indirect block levels.
70 * The hammer2_blockset in the volume header or file inode has another 8
71 * entries, giving us 66+3 = 69 bits of address space. However, some bits
72 * are taken up by (potentially) requests for redundant copies. HAMMER2
73 * currently supports up to 8 copies, which brings the address space down
74 * to 66 bits and gives us 2 bits of leeway.
76 #define HAMMER2_MIN_ALLOC 64 /* minimum allocation size */
77 #define HAMMER2_MIN_RADIX 6 /* minimum allocation size 2^N */
78 #define HAMMER2_MAX_RADIX 16 /* maximum allocation size 2^N */
79 #define HAMMER2_KEY_RADIX 64 /* number of bits in key */
82 * MINALLOCSIZE - The minimum allocation size. This can be smaller
83 * or larger than the minimum physical IO size.
85 * NOTE: Should not be larger than 1K since inodes
88 * MINIOSIZE - The minimum IO size. This must be less than
89 * or equal to HAMMER2_PBUFSIZE.
91 * XXX currently must be set to MINALLOCSIZE until/if
92 * we deal with recursive buffer cache locks.
94 * HAMMER2_PBUFSIZE - Topological block size used by files for all
95 * blocks except the block straddling EOF.
97 * HAMMER2_SEGSIZE - Allocation map segment size, typically 2MB
100 #define HAMMER2_SEGSIZE (65536 * 8)
102 #define HAMMER2_PBUFRADIX 16 /* physical buf (1<<16) bytes */
103 #define HAMMER2_PBUFSIZE 65536
104 #define HAMMER2_LBUFRADIX 14 /* logical buf (1<<14) bytes */
105 #define HAMMER2_LBUFSIZE 16384
108 #define HAMMER2_MINIORADIX 16 /* minimum phsical IO size */
109 #define HAMMER2_MINIOSIZE 65536
111 #define HAMMER2_MINIORADIX HAMMER2_MINALLOCRADIX
112 #define HAMMER2_MINIOSIZE HAMMER2_MINALLOCSIZE
114 #define HAMMER2_MINALLOCRADIX 10 /* minimum block allocation size */
115 #define HAMMER2_MINALLOCSIZE 1024
116 #define HAMMER2_IND_BYTES_MIN 4096 /* first indirect layer only */
117 #define HAMMER2_IND_BYTES_MAX HAMMER2_PBUFSIZE
118 #define HAMMER2_IND_COUNT_MIN (HAMMER2_IND_BYTES_MIN / \
119 sizeof(hammer2_blockref_t))
120 #define HAMMER2_IND_COUNT_MAX (HAMMER2_IND_BYTES_MAX / \
121 sizeof(hammer2_blockref_t))
124 * HAMMER2 processes blockrefs in sets of 8. The set is fully associative,
125 * is not sorted, and may contain holes.
127 * A full indirect block supports 1024 blockrefs.
129 * An inode embeds one set of blockrefs but may also use the data area for
130 * up to 512 bytes of direct data.
132 #define HAMMER2_SET_COUNT 8 /* direct entries & associativity */
133 #define HAMMER2_SET_RADIX 3
134 #define HAMMER2_EMBEDDED_BYTES 512
135 #define HAMMER2_EMBEDDED_RADIX 9
137 #define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1)
138 #define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1)
139 #define HAMMER2_SEGMASK (HAMMER2_SEGSIZE - 1)
141 #define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK)
142 #define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE)
143 #define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK)
144 #define HAMMER2_SEGSIZE64 ((hammer2_off_t)HAMMER2_SEGSIZE)
145 #define HAMMER2_SEGMASK64 ((hammer2_off_t)HAMMER2_SEGMASK)
147 #define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5"
150 * A HAMMER2 filesystem is always sized in multiples of 8MB.
152 * A 4MB segment is reserved at the beginning of each 2GB zone. This segment
153 * contains the volume header, the free block table, and possibly other
154 * information in the future. 4MB = 64 x 64K blocks.
156 #define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024)
157 #define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN)
158 #define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1)
159 #define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK)
161 #define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN)
162 #define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN)
163 #define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1)
164 #define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK)
166 #define HAMMER2_ZONE_BYTES64 (2LLU * 1024 * 1024 * 1024)
167 #define HAMMER2_ZONE_MASK64 (HAMMER2_ZONE_BYTES64 - 1)
168 #define HAMMER2_ZONE_SEG (4 * 1024 * 1024)
169 #define HAMMER2_ZONE_SEG64 ((hammer2_off_t)HAMMER2_ZONE_SEG)
170 #define HAMMER2_ZONE_BLOCKS_SEG (HAMMER2_ZONE_SEG / HAMMER2_PBUFSIZE)
173 * Two linear areas can be reserved after the initial 2MB segment in the base
174 * zone (the one starting at offset 0). These areas are NOT managed by the
175 * block allocator and do not fall under HAMMER2 crc checking rules based
176 * at the volume header (but can be self-CRCd internally, depending).
178 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN
179 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024)
180 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024)
182 #define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN
183 #define HAMMER2_REDO_NOM_BYTES (256*1024*1024)
184 #define HAMMER2_REDO_MAX_BYTES (1024*1024*1024)
187 * Most HAMMER2 types are implemented as unsigned 64-bit integers.
188 * Transaction ids are monotonic.
190 * We utilize 32-bit iSCSI CRCs.
192 typedef uint64_t hammer2_tid_t;
193 typedef uint64_t hammer2_off_t;
194 typedef uint64_t hammer2_key_t;
195 typedef uint32_t hammer2_crc32_t;
198 * Miscellanious ranges (all are unsigned).
200 #define HAMMER2_MIN_TID 1ULL
201 #define HAMMER2_MAX_TID 0xFFFFFFFFFFFFFFFFULL
202 #define HAMMER2_MIN_KEY 0ULL
203 #define HAMMER2_MAX_KEY 0xFFFFFFFFFFFFFFFFULL
204 #define HAMMER2_MIN_OFFSET 0ULL
205 #define HAMMER2_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL
208 * HAMMER2 data offset special cases and masking.
210 * All HAMMER2 data offsets have to be broken down into a 64K buffer base
211 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO).
213 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits
214 * of the data offset field specifies how large the data chunk being pointed
215 * to as a power of 2. This value typically ranges from HAMMER2_MIN_RADIX
216 * to HAMMER2_MAX_RADIX (6-16). Larger values may be supported in the future
217 * to support file extents.
219 #define HAMMER2_OFF_BAD ((hammer2_off_t)-1)
220 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL
221 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64)
222 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64)
223 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL
224 #define HAMMER2_MAX_COPIES 6
227 * HAMMER2 directory support and pre-defined keys
229 #define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL
230 #define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL
231 #define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL
232 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL
233 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */
235 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */
238 * The media block reference structure. This forms the core of the HAMMER2
239 * media topology recursion. This 64-byte data structure is embedded in the
240 * volume header, in inodes (which are also directory entries), and in
243 * A blockref references a single media item, which typically can be a
244 * directory entry (aka inode), indirect block, or data block.
246 * The primary feature a blockref represents is the ability to validate
247 * the entire tree underneath it via its check code. Any modification to
248 * anything propagates up the blockref tree all the way to the root, replacing
249 * the related blocks. Propagations can shortcut to the volume root to
250 * implement the 'fast syncing' feature but this only delays the eventual
253 * The check code can be a simple 32-bit iscsi code, a 64-bit crc,
254 * or as complex as a 192 bit cryptographic hash. 192 bits is the maximum
255 * supported check code size, which is not sufficient for unverified dedup
256 * UNLESS one doesn't mind once-in-a-blue-moon data corruption (such as when
257 * farming web data). HAMMER2 has an unverified dedup feature for just this
260 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */
261 uint8_t type; /* type of underlying item */
262 uint8_t methods; /* check method & compression method */
263 uint8_t copyid; /* specify which copy this is */
264 uint8_t keybits; /* #of keybits masked off 0=leaf */
265 uint8_t vradix; /* virtual data/meta-data size */
266 uint8_t flags; /* blockref flags */
269 hammer2_key_t key; /* key specification */
270 hammer2_tid_t mirror_tid; /* propagate for mirror scan */
271 hammer2_tid_t modify_tid; /* modifications sans propagation */
272 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/
273 union { /* check info */
289 typedef struct hammer2_blockref hammer2_blockref_t;
291 #define HAMMER2_BREF_SYNC1 0x01 /* modification synchronized */
292 #define HAMMER2_BREF_SYNC2 0x02 /* modification committed */
293 #define HAMMER2_BREF_DESYNCCHLD 0x04 /* desynchronize children */
294 #define HAMMER2_BREF_DELETED 0x80 /* indicates a deletion */
296 #define HAMMER2_BLOCKREF_BYTES 64 /* blockref struct in bytes */
298 #define HAMMER2_BREF_TYPE_EMPTY 0
299 #define HAMMER2_BREF_TYPE_INODE 1
300 #define HAMMER2_BREF_TYPE_INDIRECT 2
301 #define HAMMER2_BREF_TYPE_DATA 3
302 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */
304 #define HAMMER2_ENC_COMPMETHOD(n) (n)
305 #define HAMMER2_ENC_CHECKMETHOD(n) ((n) << 4)
306 #define HAMMER2_DEC_COMPMETHOD(n) ((n) & 15)
307 #define HAMMER2_DEC_CHECKMETHOD(n) (((n) >> 4) & 15)
310 * HAMMER2 block references are collected into sets of 8 blockrefs. These
311 * sets are fully associative, meaning the elements making up a set are
312 * not sorted in any way and may contain duplicate entries, holes, or
313 * entries which shortcut multiple levels of indirection. Sets are used
316 * (1) When redundancy is desired a set may contain several duplicate
317 * entries pointing to different copies of the same data. Up to 8 copies
318 * are supported but the set structure becomes a bit inefficient once
321 * (2) The blockrefs in a set can shortcut multiple levels of indirections
322 * within the bounds imposed by the parent of set.
324 * When a set fills up another level of indirection is inserted, moving
325 * some or all of the set's contents into indirect blocks placed under the
326 * set. This is a top-down approach in that indirect blocks are not created
327 * until the set actually becomes full (that is, the entries in the set can
328 * shortcut the indirect blocks when the set is not full). Depending on how
329 * things are filled multiple indirect blocks will eventually be created.
331 struct hammer2_blockset {
332 hammer2_blockref_t blockref[HAMMER2_SET_COUNT];
335 typedef struct hammer2_blockset hammer2_blockset_t;
338 * Catch programmer snafus
340 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT
341 #error "hammer2 direct radix is incorrect"
343 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE
344 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent"
346 #if (1 << HAMMER2_MIN_RADIX) != HAMMER2_MIN_ALLOC
347 #error "HAMMER2_MIN_RADIX and HAMMER2_MIN_ALLOC are inconsistent"
351 * The media indirect block structure.
353 struct hammer2_indblock_data {
354 hammer2_blockref_t blockref[HAMMER2_IND_COUNT_MAX];
357 typedef struct hammer2_indblock_data hammer2_indblock_data_t;
360 * In HAMMER2 inodes ARE directory entries, with a special exception for
361 * hardlinks. The inode number is stored in the inode rather than being
362 * based on the location of the inode (since the location moves every time
363 * the inode or anything underneath the inode is modified).
365 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes
366 * for the filename, and 512 bytes worth of direct file data OR an embedded
369 * Directories represent one inode per blockref. Inodes are not laid out
370 * as a file but instead are represented by the related blockrefs. The
371 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember
372 * that blocksets are fully associative, so a certain degree efficiency is
373 * achieved just from that.
375 * Up to 512 bytes of direct data can be embedded in an inode, and since
376 * inodes are essentially directory entries this also means that small data
377 * files end up simply being laid out linearly in the directory, resulting
378 * in fewer seeks and highly optimal access.
380 * The compression mode can be changed at any time in the inode and is
381 * recorded on a blockref-by-blockref basis.
383 * Hardlinks are supported via the inode map. Essentially the way a hardlink
384 * works is that all individual directory entries representing the same file
385 * are special cased and specify the same inode number. The actual file
386 * is placed in the nearest parent directory that is parent to all instances
387 * of the hardlink. If all hardlinks to a file are in the same directory
388 * the actual file will also be placed in that directory. This file uses
389 * the inode number as the directory entry key and is invisible to normal
390 * directory scans. Real directory entry keys are differentiated from the
391 * inode number key via bit 63. Access to the hardlink silently looks up
392 * the real file and forwards all operations to that file. Removal of the
393 * last hardlink also removes the real file.
395 * (attr_tid) is only updated when the inode's specific attributes or regular
396 * file size has changed, and affects path lookups and stat. (attr_tid)
397 * represents a special cache coherency lock under the inode. The inode
398 * blockref's modify_tid will always cover it.
400 * (dirent_tid) is only updated when an entry under a directory inode has
401 * been created, deleted, renamed, or had its attributes change, and affects
402 * directory lookups and scans. (dirent_tid) represents another special cache
403 * coherency lock under the inode. The inode blockref's modify_tid will
406 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */
407 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */
408 #define HAMMER2_INODE_VERSION_ONE 1
410 struct hammer2_inode_data {
411 uint16_t version; /* 0000 inode data version */
412 uint16_t reserved02; /* 0002 */
415 * core inode attributes, inode type, misc flags
417 uint32_t uflags; /* 0004 chflags */
418 uint32_t rmajor; /* 0008 available for device nodes */
419 uint32_t rminor; /* 000C available for device nodes */
420 uint64_t ctime; /* 0010 inode change time */
421 uint64_t mtime; /* 0018 modified time */
422 uint64_t atime; /* 0020 access time (unsupported) */
423 uint64_t btime; /* 0028 birth time */
424 uuid_t uid; /* 0030 uid / degenerate unix uid */
425 uuid_t gid; /* 0040 gid / degenerate unix gid */
427 uint8_t type; /* 0050 object type */
428 uint8_t op_flags; /* 0051 operational flags */
429 uint16_t cap_flags; /* 0052 capability flags */
430 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */
433 * inode size, identification, localized recursive configuration
434 * for compression and backup copies.
436 hammer2_tid_t inum; /* 0058 inode number */
437 hammer2_off_t size; /* 0060 size of file */
438 uint64_t nlinks; /* 0068 hard links (typ only dirs) */
439 hammer2_tid_t iparent; /* 0070 parent inum (recovery only) */
440 hammer2_key_t name_key; /* 0078 full filename key */
441 uint16_t name_len; /* 0080 filename length */
442 uint8_t ncopies; /* 0082 ncopies to local media */
443 uint8_t comp_algo; /* 0083 compression request & algo */
446 * These fields are currently only applicable to PFSROOTs.
448 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely
449 * identify an instance of a PFS in the cluster because
450 * a mount may contain more than one copy of the PFS as
451 * a separate node. {pfs_clid, pfs_fsid} must be used for
452 * registration in the cluster.
454 uint8_t target_type; /* 0084 hardlink target type */
455 uint8_t reserved85; /* 0085 */
456 uint8_t reserved86; /* 0086 */
457 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */
458 uint64_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */
459 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */
460 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */
463 * Quotas and cumulative sub-tree counters.
465 hammer2_off_t data_quota; /* 00B0 subtree quota in bytes */
466 hammer2_off_t data_count; /* 00B8 subtree byte count */
467 hammer2_off_t inode_quota; /* 00C0 subtree quota inode count */
468 hammer2_off_t inode_count; /* 00C8 subtree inode count */
469 hammer2_tid_t attr_tid; /* 00D0 attributes changed */
470 hammer2_tid_t dirent_tid; /* 00D8 directory/attr changed */
471 uint64_t reservedE0; /* 00E0 */
472 uint64_t reservedE8; /* 00E8 */
473 uint64_t reservedF0; /* 00F0 */
474 uint64_t reservedF8; /* 00F8 */
476 unsigned char filename[HAMMER2_INODE_MAXNAME];
477 /* 0100-01FF (256 char, unterminated) */
478 union { /* 0200-03FF (64x8 = 512 bytes) */
479 struct hammer2_blockset blockset;
480 char data[HAMMER2_EMBEDDED_BYTES];
484 typedef struct hammer2_inode_data hammer2_inode_data_t;
486 #define HAMMER2_OPFLAG_DIRECTDATA 0x01
487 #define HAMMER2_OPFLAG_PFSROOT 0x02
488 #define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */
490 #define HAMMER2_OBJTYPE_UNKNOWN 0
491 #define HAMMER2_OBJTYPE_DIRECTORY 1
492 #define HAMMER2_OBJTYPE_REGFILE 2
493 #define HAMMER2_OBJTYPE_FIFO 4
494 #define HAMMER2_OBJTYPE_CDEV 5
495 #define HAMMER2_OBJTYPE_BDEV 6
496 #define HAMMER2_OBJTYPE_SOFTLINK 7
497 #define HAMMER2_OBJTYPE_HARDLINK 8 /* dummy entry for hardlink */
498 #define HAMMER2_OBJTYPE_SOCKET 9
499 #define HAMMER2_OBJTYPE_WHITEOUT 10
501 #define HAMMER2_COPYID_NONE 0
502 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1)
504 #define HAMMER2_COMP_NONE 0
505 #define HAMMER2_COMP_AUTOZERO 1
507 #define HAMMER2_CHECK_NONE 0
508 #define HAMMER2_CHECK_ICRC 1
511 * PEER types identify connections and help cluster controller filter
512 * out unwanted SPANs.
514 #define HAMMER2_PEER_NONE 0
515 #define HAMMER2_PEER_CLUSTER 1 /* a cluster controller */
516 #define HAMMER2_PEER_BLOCK 2 /* block devices */
517 #define HAMMER2_PEER_HAMMER2 3 /* hammer2-mounted volumes */
520 * PFS types identify a PFS on media and in LNK_SPAN messages.
522 #define HAMMER2_PFSTYPE_NONE 0
523 #define HAMMER2_PFSTYPE_ADMIN 1
524 #define HAMMER2_PFSTYPE_CLIENT 2
525 #define HAMMER2_PFSTYPE_CACHE 3
526 #define HAMMER2_PFSTYPE_COPY 4
527 #define HAMMER2_PFSTYPE_SLAVE 5
528 #define HAMMER2_PFSTYPE_SOFT_SLAVE 6
529 #define HAMMER2_PFSTYPE_SOFT_MASTER 7
530 #define HAMMER2_PFSTYPE_MASTER 8
531 #define HAMMER2_PFSTYPE_MAX 9 /* 0-8 */
534 * The allocref structure represents the allocation table. One 64K block
535 * is broken down into 4096 x 16 byte entries. Each indirect block chops
536 * 11 bits off the 64-bit storage space, with leaf entries representing
537 * 64KB blocks. So: (12, 12, 12, 12, 16) = 64 bit storage space.
539 * Each 64K freemap block breaks the 4096 entries into a 64x64 tree with
540 * big_hint1 representing the top level every 64th entry and big_hint2
541 * representing the lower level in each entry. These fields specify the
542 * largest contiguous radix (1-63) available for allocation in the related
543 * sub-tree. The largest contiguous radix available for the entire block
544 * is saved in the parent (for the root this will be alloc_blockref in the
545 * volume header). The hints may be larger than actual and will be corrected
546 * on the fly but must not be smaller. The allocator uses the hints to
547 * very quickly locate nearby blocks of the desired size.
549 * In indirect blocks the 64-bit free[_or_mask] field stores the total free
550 * space for each of the 4096 sub-nodes in bytes. The total free space
551 * represented by the indirect block is stored in its parent.
553 * Each leaf element represents a 64K block. A bitmap replaces the free space
554 * count, giving us a 1KB allocation resolution. A micro-allocation append
555 * offset replaces the icrc field. The micro-allocation feature is not
556 * currently implemented and the field will be set to 65536.
558 * The allocation map uses reserved blocks so no data block reference is
559 * required, only a bit in the flags field to specify which of two possible
560 * reserved blocks to use. This allows the allocation map to be flushed to
561 * disk with minimal synchronization.
563 struct hammer2_allocref {
564 uint32_t icrc_or_app; /* node: icrc, leaf: append offset */
566 uint8_t big_hint1; /* upper level hint */
567 uint8_t big_hint2; /* lower level hint */
568 uint64_t free_or_mask; /* node: free bytes, leaf: bitmask */
571 typedef struct hammer2_allocref hammer2_allocref_t;
574 * WARNING - allocref size x entries must equate to the hammer buffer size,
575 * and 12 bits per recursion is assumed by the allocator.
577 * ALTA-D Since no data_offset is specified flags are needed to select
578 * which sub-block to recurse down into for root & internal nodes.
579 * (only ALTA and ALTB is currently supported).
581 * LEAF Terminal entry, always set for leafs. May be used to support
582 * 4MB extent allocations and early termination in the future.
583 * (not required to shortcut allocation scans as the big_hint1/2
584 * fields are used for this).
586 #define HAMMER2_ALLOCREF_BYTES 16 /* structure size */
587 #define HAMMER2_ALLOCREF_ENTRIES 4096 /* entries */
588 #define HAMMER2_ALLOCREF_RADIX 12 /* log2(entries) */
590 #if (HAMMER2_ALLOCREF_BYTES * HAMMER2_ALLOCREF_ENTRIES) != HAMMER2_PBUFSIZE
591 #error "allocref parameters do not fit in hammer buffer"
593 #if (1 << HAMMER2_ALLOCREF_RADIX) != HAMMER2_ALLOCREF_ENTRIES
594 #error "allocref parameters are inconsistent"
597 #define HAMMER2_ALLOCREF_ALTMASK 0x0003 /* select block for recurse */
598 #define HAMMER2_ALLOCREF_ALTA 0x0000
599 #define HAMMER2_ALLOCREF_ALTB 0x0001
600 #define HAMMER2_ALLOCREF_ALTC 0x0002 /* unsupported */
601 #define HAMMER2_ALLOCREF_ALTD 0x0003 /* unsupported */
602 #define HAMMER2_ALLOCREF_LEAF 0x0004
605 * All HAMMER2 directories directly under the super-root on your local
606 * media can be mounted separately, even if they share the same physical
609 * When you do a HAMMER2 mount you are effectively tying into a HAMMER2
610 * cluster via local media. The local media does not have to participate
611 * in the cluster, other than to provide the hammer2_copy_data[] array and
612 * root inode for the mount.
614 * This is important: The mount device path you specify serves to bootstrap
615 * your entry into the cluster, but your mount will make active connections
616 * to ALL copy elements in the hammer2_copy_data[] array which match the
617 * PFSID of the directory in the super-root that you specified. The local
618 * media path does not have to be mentioned in this array but becomes part
619 * of the cluster based on its type and access rights. ALL ELEMENTS ARE
620 * TREATED ACCORDING TO TYPE NO MATTER WHICH ONE YOU MOUNT FROM.
622 * The actual cluster may be far larger than the elements you list in the
623 * hammer2_copy_data[] array. You list only the elements you wish to
624 * directly connect to and you are able to access the rest of the cluster
625 * indirectly through those connections.
627 * This structure must be exactly 128 bytes long.
629 struct hammer2_copy_data {
630 uint8_t copyid; /* 00 copyid 0-255 (must match slot) */
631 uint8_t inprog; /* 01 operation in progress, or 0 */
632 uint8_t chain_to; /* 02 operation chaining to, or 0 */
633 uint8_t chain_from; /* 03 operation chaining from, or 0 */
634 uint16_t flags; /* 04-05 flags field */
635 uint8_t error; /* 06 last operational error */
636 uint8_t priority; /* 07 priority and round-robin flag */
637 uint8_t remote_pfs_type;/* 08 probed direct remote PFS type */
638 uint8_t reserved08[23]; /* 09-1F */
639 uuid_t pfs_clid; /* 20-2F copy target must match this uuid */
640 uint8_t label[16]; /* 30-3F import/export label */
641 uint8_t path[64]; /* 40-7F target specification string or key */
644 typedef struct hammer2_copy_data hammer2_copy_data_t;
646 #define COPYDATAF_ENABLED 0x0001
647 #define COPYDATAF_INPROG 0x0002
648 #define COPYDATAF_CONN_RR 0x80 /* round-robin at same priority */
649 #define COPYDATAF_CONN_EF 0x40 /* media errors flagged */
650 #define COPYDATAF_CONN_PRI 0x0F /* select priority 0-15 (15=best) */
653 * The volume header eats a 64K block. There is currently an issue where
654 * we want to try to fit all nominal filesystem updates in a 512-byte section
655 * but it may be a lost cause due to the need for a blockset.
657 * All information is stored in host byte order. The volume header's magic
658 * number may be checked to determine the byte order. If you wish to mount
659 * between machines w/ different endian modes you'll need filesystem code
660 * which acts on the media data consistently (either all one way or all the
661 * other). Our code currently does not do that.
663 * A read-write mount may have to recover missing allocations by doing an
664 * incremental mirror scan looking for modifications made after alloc_tid.
665 * If alloc_tid == last_tid then no recovery operation is needed. Recovery
666 * operations are usually very, very fast.
668 * Read-only mounts do not need to do any recovery, access to the filesystem
669 * topology is always consistent after a crash (is always consistent, period).
670 * However, there may be shortcutted blockref updates present from deep in
671 * the tree which are stored in the volumeh eader and must be tracked on
674 * NOTE: The copyinfo[] array contains the configuration for both the
675 * cluster connections and any local media copies. The volume
676 * header will be replicated for each local media copy.
678 * The mount command may specify multiple medias or just one and
679 * allow HAMMER2 to pick up the others when it checks the copyinfo[]
682 * NOTE: root_blockref points to the super-root directory, not the root
683 * directory. The root directory will be a subdirectory under the
686 * The super-root directory contains all root directories and all
687 * snapshots (readonly or writable). It is possible to do a
688 * null-mount of the super-root using special path constructions
689 * relative to your mounted root.
691 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were
692 * a PFS, including mirroring and storage quota operations, and this is
693 * prefered over creating discrete PFSs in the super-root. Instead
694 * the super-root is most typically used to create writable snapshots,
695 * alternative roots, and so forth. The super-root is also used by
696 * the automatic snapshotting mechanism.
698 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU
699 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU
701 #define HAMMER2_COPYID_COUNT 256
703 struct hammer2_volume_data {
705 * sector #0 - 512 bytes
707 uint64_t magic; /* 0000 Signature */
708 hammer2_off_t boot_beg; /* 0008 Boot area (future) */
709 hammer2_off_t boot_end; /* 0010 (size = end - beg) */
710 hammer2_off_t aux_beg; /* 0018 Aux area (future) */
711 hammer2_off_t aux_end; /* 0020 (size = end - beg) */
712 hammer2_off_t volu_size; /* 0028 Volume size, bytes */
714 uint32_t version; /* 0030 */
715 uint32_t flags; /* 0034 */
716 uint8_t copyid; /* 0038 copyid of phys vol */
717 uint8_t freemap_version; /* 0039 freemap algorithm */
718 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */
719 uint8_t reserved003B; /* 003B */
720 uint32_t reserved003C; /* 003C */
722 uuid_t fsid; /* 0040 */
723 uuid_t fstype; /* 0050 */
726 * allocator_size is precalculated at newfs time and does not include
727 * reserved blocks, boot, or redo areas.
729 * Initial non-reserved-area allocations do not use the allocation
730 * map but instead adjust alloc_iterator. Dynamic allocations take
731 * over starting at (allocator_beg). This makes newfs_hammer2's
732 * job a lot easier and can also serve as a testing jig.
734 hammer2_off_t allocator_size; /* 0060 Total data space */
735 hammer2_off_t allocator_free; /* 0068 Free space */
736 hammer2_off_t allocator_beg; /* 0070 Initial allocations */
737 hammer2_tid_t mirror_tid; /* 0078 best committed tid */
738 hammer2_tid_t alloc_tid; /* 0080 Alloctable modify tid */
739 hammer2_blockref_t alloc_blockref; /* 0088-00C7 */
742 * Copyids are allocated dynamically from the copyexists bitmap.
743 * An id from the active copies set (up to 8, see copyinfo later on)
744 * may still exist after the copy set has been removed from the
745 * volume header and its bit will remain active in the bitmap and
746 * cannot be reused until it is 100% removed from the hierarchy.
748 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */
749 char reserved0140[248]; /* 00E8-01DF */
752 * 32 bit CRC array at the end of the first 512 byte sector.
754 * icrc_sects[7] - First 512-4 bytes of volume header (including all
755 * the other icrc's except the last one).
757 * icrc_sects[6] - Second 512-4 bytes of volume header, which is
758 * the blockset for the root.
760 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */
763 * sector #1 - 512 bytes
765 * The entire sector is used by a blockset.
767 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */
772 char sector2[512]; /* 0400-05FF reserved */
773 char sector3[512]; /* 0600-07FF reserved */
774 char sector4[512]; /* 0800-09FF reserved */
775 char sector5[512]; /* 0A00-0BFF reserved */
776 char sector6[512]; /* 0C00-0DFF reserved */
777 char sector7[512]; /* 0E00-0FFF reserved */
780 * sector #8-71 - 32768 bytes
782 * Contains the configuration for up to 256 copyinfo targets. These
783 * specify local and remote copies operating as masters or slaves.
784 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255
785 * indicates the local media).
787 * Each inode contains a set of up to 8 copyids, either inherited
788 * from its parent or explicitly specified in the inode, which
789 * indexes into this array.
791 /* 1000-8FFF copyinfo config */
792 struct hammer2_copy_data copyinfo[HAMMER2_COPYID_COUNT];
799 * Remaining sections are reserved for future use.
801 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */
804 * icrc on entire volume header
806 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/
809 typedef struct hammer2_volume_data hammer2_volume_data_t;
812 * Various parts of the volume header have their own iCRCs.
814 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes
815 * and not included the icrc calculation.
817 * The second 512 bytes also has its own iCRC but it is stored in the first
818 * 512 bytes so it covers the entire second 512 bytes.
820 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes,
821 * which is where the iCRC for the whole volume is stored. This is currently
822 * a catch-all for anything not individually iCRCd.
824 #define HAMMER2_VOL_ICRC_SECT0 7
825 #define HAMMER2_VOL_ICRC_SECT1 6
827 #define HAMMER2_VOLUME_BYTES 65536
829 #define HAMMER2_VOLUME_ICRC0_OFF 0
830 #define HAMMER2_VOLUME_ICRC1_OFF 512
831 #define HAMMER2_VOLUME_ICRCVH_OFF 0
833 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4)
834 #define HAMMER2_VOLUME_ICRC1_SIZE (512)
835 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4)
837 #define HAMMER2_VOL_VERSION_MIN 1
838 #define HAMMER2_VOL_VERSION_DEFAULT 1
839 #define HAMMER2_VOL_VERSION_WIP 2
841 #define HAMMER2_NUM_VOLHDRS 4
843 union hammer2_media_data {
844 hammer2_volume_data_t voldata;
845 hammer2_inode_data_t ipdata;
846 hammer2_indblock_data_t npdata;
847 char buf[HAMMER2_PBUFSIZE];
850 typedef union hammer2_media_data hammer2_media_data_t;
853 * Prototypes for user & kernel functions. Kernel-only prototypes are
856 uint32_t hammer2_icrc32(const void *buf, size_t size);
857 uint32_t hammer2_icrc32c(const void *buf, size_t size, uint32_t crc);