2 * Copyright (c) 2007 The DragonFly Project. All rights reserved.
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.24 2008/02/20 00:55:51 dillon Exp $
37 #ifndef VFS_HAMMER_DISK_H_
38 #define VFS_HAMMER_DISK_H_
45 * The structures below represent the on-disk format for a HAMMER
46 * filesystem. Note that all fields for on-disk structures are naturally
47 * aligned. The host endian format is used - compatibility is possible
48 * if the implementation detects reversed endian and adjusts data accordingly.
50 * Most of HAMMER revolves around the concept of an object identifier. An
51 * obj_id is a 64 bit quantity which uniquely identifies a filesystem object
52 * FOR THE ENTIRE LIFE OF THE FILESYSTEM. This uniqueness allows backups
53 * and mirrors to retain varying amounts of filesystem history by removing
54 * any possibility of conflict through identifier reuse.
56 * A HAMMER filesystem may spam multiple volumes.
58 * A HAMMER filesystem uses a 16K filesystem buffer size. All filesystem
59 * I/O is done in multiples of 16K. Most buffer-sized headers such as those
60 * used by volumes, super-clusters, clusters, and basic filesystem buffers
61 * use fixed-sized A-lists which are heavily dependant on HAMMER_BUFSIZE.
63 * Per-volume storage limit: 52 bits 4096 TB
64 * Per-Zone storage limit: 59 bits 512 KTB (due to blockmap)
65 * Per-filesystem storage limit: 60 bits 1 MTB
67 #define HAMMER_BUFSIZE 16384
68 #define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1)
69 #define HAMMER_MAXDATA (256*1024)
70 #define HAMMER_BUFFER_BITS 14
72 #if (1 << HAMMER_BUFFER_BITS) != HAMMER_BUFSIZE
73 #error "HAMMER_BUFFER_BITS BROKEN"
76 #define HAMMER_BUFSIZE64 ((u_int64_t)HAMMER_BUFSIZE)
77 #define HAMMER_BUFMASK64 ((u_int64_t)HAMMER_BUFMASK)
79 #define HAMMER_OFF_ZONE_MASK 0xF000000000000000ULL /* zone portion */
80 #define HAMMER_OFF_VOL_MASK 0x0FF0000000000000ULL /* volume portion */
81 #define HAMMER_OFF_SHORT_MASK 0x000FFFFFFFFFFFFFULL /* offset portion */
82 #define HAMMER_OFF_LONG_MASK 0x0FFFFFFFFFFFFFFFULL /* offset portion */
83 #define HAMMER_OFF_SHORT_REC_MASK 0x000FFFFFFF000000ULL /* recovery boundary */
84 #define HAMMER_OFF_LONG_REC_MASK 0x0FFFFFFFFF000000ULL /* recovery boundary */
85 #define HAMMER_RECOVERY_BND 0x0000000001000000ULL
88 * Hammer transction ids are 64 bit unsigned integers and are usually
89 * synchronized with the time of day in nanoseconds.
91 * Hammer offsets are used for FIFO indexing and embed a cycle counter
92 * and volume number in addition to the offset. Most offsets are required
93 * to be 64-byte aligned.
95 typedef u_int64_t hammer_tid_t;
96 typedef u_int64_t hammer_off_t;
98 #define HAMMER_MIN_TID 0ULL /* unsigned */
99 #define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */
100 #define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */
101 #define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */
102 #define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */
103 #define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */
104 #define HAMMER_MIN_RECTYPE 0x0U /* unsigned */
105 #define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */
106 #define HAMMER_MIN_OFFSET 0ULL /* unsigned */
107 #define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */
110 * hammer_off_t has several different encodings. Note that not all zones
113 * zone 0 (z,v,o): reserved (for sanity)
114 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header)
115 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer)
116 * zone 3 (z,o): undo fifo - blockmap backed
117 * zone 4 (z,v,o): freemap - freemap-backed self-mapping
119 * zone 8 (z,o): B-Tree - blkmap-backed
120 * zone 9 (z,o): Record - blkmap-backed
121 * zone 10 (z,o): Large-data - blkmap-backed
124 #define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL
125 #define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL
126 #define HAMMER_ZONE_UNDO 0x3000000000000000ULL
127 #define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL
128 #define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL
129 #define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL
130 #define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL
131 #define HAMMER_ZONE_BTREE 0x8000000000000000ULL
132 #define HAMMER_ZONE_RECORD 0x9000000000000000ULL
133 #define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL
134 #define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL
135 #define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL
136 #define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL
137 #define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL
138 #define HAMMER_ZONE_RESERVED0F 0xF000000000000000ULL
140 #define HAMMER_ZONE_RAW_VOLUME_INDEX 1
141 #define HAMMER_ZONE_RAW_BUFFER_INDEX 2
142 #define HAMMER_ZONE_UNDO_INDEX 3
143 #define HAMMER_ZONE_FREEMAP_INDEX 4
144 #define HAMMER_ZONE_BTREE_INDEX 8
145 #define HAMMER_ZONE_RECORD_INDEX 9
146 #define HAMMER_ZONE_LARGE_DATA_INDEX 10
147 #define HAMMER_ZONE_SMALL_DATA_INDEX 11
149 #define HAMMER_MAX_ZONES 16
151 #define HAMMER_VOL_ENCODE(vol_no) \
152 ((hammer_off_t)((vol_no) & 255) << 52)
153 #define HAMMER_VOL_DECODE(ham_off) \
154 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255)
155 #define HAMMER_ZONE_DECODE(ham_off) \
156 (int32_t)(((hammer_off_t)(ham_off) >> 60))
157 #define HAMMER_SHORT_OFF_ENCODE(offset) \
158 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
159 #define HAMMER_LONG_OFF_ENCODE(offset) \
160 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
162 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \
163 (HAMMER_ZONE_RAW_VOLUME | \
164 HAMMER_VOL_ENCODE(vol_no) | \
165 HAMMER_SHORT_OFF_ENCODE(offset))
167 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \
168 (HAMMER_ZONE_RAW_BUFFER | \
169 HAMMER_VOL_ENCODE(vol_no) | \
170 HAMMER_SHORT_OFF_ENCODE(offset))
172 #define HAMMER_ENCODE_FREEMAP(vol_no, offset) \
173 (HAMMER_ZONE_FREEMAP | \
174 HAMMER_VOL_ENCODE(vol_no) | \
175 HAMMER_SHORT_OFF_ENCODE(offset))
178 * Large-Block backing store
180 * A blockmap is a two-level map which translates a blockmap-backed zone
181 * offset into a raw zone 2 offset. Each layer handles 18 bits. The 8M
182 * large-block size is 23 bits so two layers gives us 23+18+18 = 59 bits
185 #define HAMMER_LARGEBLOCK_SIZE (8192 * 1024)
186 #define HAMMER_LARGEBLOCK_SIZE64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE)
187 #define HAMMER_LARGEBLOCK_MASK (HAMMER_LARGEBLOCK_SIZE - 1)
188 #define HAMMER_LARGEBLOCK_MASK64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE - 1)
189 #define HAMMER_LARGEBLOCK_BITS 23
190 #if (1 << HAMMER_LARGEBLOCK_BITS) != HAMMER_LARGEBLOCK_SIZE
191 #error "HAMMER_LARGEBLOCK_BITS BROKEN"
194 #define HAMMER_BUFFERS_PER_LARGEBLOCK \
195 (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE)
196 #define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK \
197 (HAMMER_BUFFERS_PER_LARGEBLOCK - 1)
198 #define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK64 \
199 ((hammer_off_t)HAMMER_BUFFERS_PER_LARGEBLOCK_MASK)
202 * Every blockmap has this root structure in the root volume header.
204 struct hammer_blockmap {
205 hammer_off_t phys_offset; /* zone-2 physical offset */
206 hammer_off_t next_offset; /* zone-X logical offset */
207 hammer_off_t alloc_offset; /* zone-X logical offset */
209 u_int32_t reserved01;
212 typedef struct hammer_blockmap *hammer_blockmap_t;
215 * The blockmap is a 2-layer entity made up of big-blocks. The first layer
216 * contains 262144 32-byte entries (18 bits), the second layer contains
217 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
218 * 18+19+23 = 60 bits. The top four bits are the zone id.
220 * Layer 2 encodes the physical bigblock mapping for a blockmap. The freemap
221 * uses this field to encode the virtual blockmap offset that allocated the
224 * NOTE: The freemap maps the vol_no in the upper 8 bits of layer1.
226 * zone-4 blockmap offset: [z:4][layer1:18][layer2:19][bigblock:23]
228 struct hammer_blockmap_layer1 {
229 hammer_off_t blocks_free; /* big-blocks free */
230 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */
231 u_int32_t layer1_crc; /* crc of this entry */
232 u_int32_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */
233 hammer_off_t reserved01;
236 struct hammer_blockmap_layer2 {
238 u_int32_t bytes_free; /* bytes free within this bigblock */
240 hammer_off_t owner; /* used by freemap */
241 hammer_off_t phys_offset; /* used by blockmap */
245 #define HAMMER_BLOCKMAP_FREE 0ULL
246 #define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL)
248 #define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \
249 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1))
250 #define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \
251 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2))
253 #define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \
254 (HAMMER_BLOCKMAP_RADIX1 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
255 #define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \
256 (HAMMER_BLOCKMAP_RADIX2 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
258 #define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 */ \
259 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
260 #define HAMMER_BLOCKMAP_LAYER2 /* 19+23 */ \
261 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_LARGEBLOCK_SIZE64)
263 #define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1)
264 #define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1)
267 * byte offset within layer1 or layer2 big-block for the entry representing
268 * a zone-2 physical offset.
270 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \
271 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \
272 HAMMER_BLOCKMAP_LAYER2 * sizeof(struct hammer_blockmap_layer1))
274 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \
275 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \
276 HAMMER_LARGEBLOCK_SIZE64 * sizeof(struct hammer_blockmap_layer2))
279 * All on-disk HAMMER structures which make up elements of the FIFO contain
280 * a hammer_fifo_head and hammer_fifo_tail structure. This structure
281 * contains all the information required to validate the fifo element
282 * and to scan the fifo in either direction. The head is typically embedded
283 * in higher level hammer on-disk structures while the tail is typically
284 * out-of-band. hdr_size is the size of the whole mess, including the tail.
286 * Nearly all such structures are guaranteed to not cross a 16K filesystem
287 * buffer boundary. The one exception is a record, whos related data may
288 * cross a buffer boundary.
290 * HAMMER guarantees alignment with a fifo head structure at 16MB intervals
291 * (i.e. the base of the buffer will not be in the middle of a data record).
292 * This is used to allow the recovery code to re-sync after hitting corrupted
295 * PAD elements are allowed to take up only 8 bytes of space as a special
296 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
297 * and with the tail overloaded onto the head structure for 8 bytes total.
299 #define HAMMER_HEAD_ONDISK_SIZE 24
300 #define HAMMER_HEAD_RECOVERY_ALIGNMENT (16 * 1024 * 1024)
301 #define HAMMER_HEAD_ALIGN 8
302 #define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
303 #define HAMMER_TAIL_ONDISK_SIZE 8
305 struct hammer_fifo_head {
306 u_int16_t hdr_signature;
308 u_int32_t hdr_size; /* aligned size of the whole mess */
310 u_int32_t hdr_reserved02;
311 hammer_tid_t hdr_seq; /* related sequence number */
314 struct hammer_fifo_tail {
315 u_int16_t tail_signature;
317 u_int32_t tail_size; /* aligned size of the whole mess */
320 typedef struct hammer_fifo_head *hammer_fifo_head_t;
321 typedef struct hammer_fifo_tail *hammer_fifo_tail_t;
326 #define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE)
327 #define HAMMER_HEAD_TYPE_VOL 0x0041U /* Volume (dummy header) */
328 #define HAMMER_HEAD_TYPE_BTREE 0x0042U /* B-Tree node */
329 #define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
330 #define HAMMER_HEAD_TYPE_DELETE 0x0044U /* record deletion */
331 #define HAMMER_HEAD_TYPE_RECORD 0x0045U /* Filesystem record */
333 #define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */
335 #define HAMMER_HEAD_SIGNATURE 0xC84EU
336 #define HAMMER_TAIL_SIGNATURE 0xC74FU
339 * Misc FIFO structures (except for the B-Tree node and hammer record)
341 struct hammer_fifo_undo {
342 struct hammer_fifo_head head;
343 hammer_off_t undo_offset;
344 /* followed by data */
347 typedef struct hammer_fifo_undo *hammer_fifo_undo_t;
350 * Volume header types
352 #define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
353 #define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
356 * The B-Tree structures need hammer_fsbuf_head.
358 #include "hammer_btree.h"
361 * HAMMER Volume header
363 * A HAMMER filesystem is built from any number of block devices, Each block
364 * device contains a volume header followed by however many buffers fit
367 * One of the volumes making up a HAMMER filesystem is the master, the
368 * rest are slaves. It does not have to be volume #0.
370 * The volume header takes up an entire 16K filesystem buffer and may
371 * represent up to 64KTB (65536 TB) of space.
373 * Special field notes:
375 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
376 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
377 * vol_buf_beg - offset of the first buffer.
379 * The memory log area allows a kernel to cache new records and data
380 * in memory without allocating space in the actual filesystem to hold
381 * the records and data. In the event that a filesystem becomes full,
382 * any records remaining in memory can be flushed to the memory log
383 * area. This allows the kernel to immediately return success.
386 #define HAMMER_BOOT_MINBYTES (32*1024)
387 #define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
388 #define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
390 #define HAMMER_MEM_MINBYTES (256*1024)
391 #define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
392 #define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
394 struct hammer_volume_ondisk {
395 u_int64_t vol_signature;/* Signature */
397 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
398 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
399 int64_t vol_buf_beg; /* byte offset of first buffer in volume */
400 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */
401 int64_t vol_locked; /* reserved clusters are >= this offset */
403 uuid_t vol_fsid; /* identify filesystem */
404 uuid_t vol_fstype; /* identify filesystem type */
405 char vol_name[64]; /* Name of volume */
407 int32_t vol_no; /* volume number within filesystem */
408 int32_t vol_count; /* number of volumes making up FS */
410 u_int32_t vol_version; /* version control information */
411 u_int32_t vol_reserved01;
412 u_int32_t vol_flags; /* volume flags */
413 u_int32_t vol_rootvol; /* which volume is the root volume? */
415 int32_t vol_reserved04;
416 int32_t vol_reserved05;
417 u_int32_t vol_reserved06;
418 u_int32_t vol_reserved07;
420 int32_t vol_blocksize; /* for statfs only */
421 int32_t vol_reserved08;
422 int64_t vol_nblocks; /* total allocatable hammer bufs */
425 * These fields are initialized and space is reserved in every
426 * volume making up a HAMMER filesytem, but only the master volume
427 * contains valid data.
429 int64_t vol0_stat_bigblocks; /* total bigblocks when fs is empty */
430 int64_t vol0_stat_freebigblocks;/* number of free bigblocks */
431 int64_t vol0_stat_bytes; /* for statfs only */
432 int64_t vol0_stat_inodes; /* for statfs only */
433 int64_t vol0_stat_records; /* total records in filesystem */
434 hammer_off_t vol0_btree_root; /* B-Tree root */
435 hammer_tid_t vol0_next_tid; /* highest synchronized TID */
436 hammer_tid_t vol0_next_seq; /* next SEQ no for undo */
439 * Blockmaps for zones. Not all zones use a blockmap.
441 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES];
445 typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
447 #define HAMMER_VOLF_VALID 0x0001 /* valid entry */
448 #define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
451 * All HAMMER records have a common 64-byte base and a 32 byte extension,
452 * plus a possible data reference. The data reference can be in-band or
456 #define HAMMER_RECORD_SIZE (64+32)
458 struct hammer_base_record {
459 u_int32_t signature; /* record signature */
460 u_int32_t data_crc; /* data crc */
461 struct hammer_base_elm base; /* 40 byte base element */
462 hammer_off_t data_off; /* in-band or out-of-band */
463 int32_t data_len; /* size of data in bytes */
464 u_int32_t reserved02;
468 * Record types are fairly straightforward. The B-Tree includes the record
469 * type in its index sort.
471 * In particular please note that it is possible to create a pseudo-
472 * filesystem within a HAMMER filesystem by creating a special object
473 * type within a directory. Pseudo-filesystems are used as replication
474 * targets and even though they are built within a HAMMER filesystem they
475 * get their own obj_id space (and thus can serve as a replication target)
476 * and look like a mount point to the system.
478 * Inter-cluster records are special-cased in the B-Tree. These records
479 * are referenced from a B-Tree INTERNAL node, NOT A LEAF. This means
480 * that the element in the B-Tree node is actually a boundary element whos
481 * base element fields, including rec_type, reflect the boundary, NOT
482 * the inter-cluster record type.
484 * HAMMER_RECTYPE_CLUSTER - only set in the actual inter-cluster record,
485 * not set in the left or right boundary elements around the inter-cluster
486 * reference of an internal node in the B-Tree (because doing so would
487 * interfere with the boundary tests).
489 * NOTE: hammer_ip_delete_range_all() deletes all record types greater
490 * then HAMMER_RECTYPE_INODE.
492 #define HAMMER_RECTYPE_UNKNOWN 0
493 #define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
494 #define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
495 #define HAMMER_RECTYPE_PSEUDO_INODE 2 /* pseudo filesysem */
496 #define HAMMER_RECTYPE_CLUSTER 3 /* inter-cluster reference */
497 #define HAMMER_RECTYPE_DATA 0x10
498 #define HAMMER_RECTYPE_DIRENTRY 0x11
499 #define HAMMER_RECTYPE_DB 0x12
500 #define HAMMER_RECTYPE_EXT 0x13 /* ext attributes */
501 #define HAMMER_RECTYPE_FIX 0x14 /* fixed attribute */
503 #define HAMMER_FIXKEY_SYMLINK 1
505 #define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */
506 #define HAMMER_OBJTYPE_DIRECTORY 1
507 #define HAMMER_OBJTYPE_REGFILE 2
508 #define HAMMER_OBJTYPE_DBFILE 3
509 #define HAMMER_OBJTYPE_FIFO 4
510 #define HAMMER_OBJTYPE_CDEV 5
511 #define HAMMER_OBJTYPE_BDEV 6
512 #define HAMMER_OBJTYPE_SOFTLINK 7
513 #define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
516 * A HAMMER inode record.
518 * This forms the basis for a filesystem object. obj_id is the inode number,
519 * key1 represents the pseudo filesystem id for security partitioning
520 * (preventing cross-links and/or restricting a NFS export and specifying the
521 * security policy), and key2 represents the data retention policy id.
523 * Inode numbers are 64 bit quantities which uniquely identify a filesystem
524 * object for the ENTIRE life of the filesystem, even after the object has
525 * been deleted. For all intents and purposes inode numbers are simply
526 * allocated by incrementing a sequence space.
528 * There is an important distinction between the data stored in the inode
529 * record and the record's data reference. The record references a
530 * hammer_inode_data structure but the filesystem object size and hard link
531 * count is stored in the inode record itself. This allows multiple inodes
532 * to share the same hammer_inode_data structure. This is possible because
533 * any modifications will lay out new data. The HAMMER implementation need
534 * not use the data-sharing ability when laying down new records.
536 * A HAMMER inode is subject to the same historical storage requirements
537 * as any other record. In particular any change in filesystem or hard link
538 * count will lay down a new inode record when the filesystem is synced to
539 * disk. This can lead to a lot of junk records which get cleaned up by
540 * the data retention policy.
542 * The ino_atime and ino_mtime fields are a special case. Modifications to
543 * these fields do NOT lay down a new record by default, though the values
544 * are effectively frozen for snapshots which access historical versions
545 * of the inode record due to other operations. This means that atime will
546 * not necessarily be accurate in snapshots, backups, or mirrors. mtime
547 * will be accurate in backups and mirrors since it can be regenerated from
548 * the mirroring stream.
550 * Because nlinks is historically retained the hardlink count will be
551 * accurate when accessing a HAMMER filesystem snapshot.
553 struct hammer_inode_record {
554 struct hammer_base_record base;
555 u_int64_t ino_atime; /* last access time (not historical) */
556 u_int64_t ino_mtime; /* last modified time (not historical) */
557 u_int64_t ino_size; /* filesystem object size */
558 u_int64_t ino_nlinks; /* hard links */
562 * Data records specify the entire contents of a regular file object,
563 * including attributes. Small amounts of data can theoretically be
564 * embedded in the record itself but the use of this ability verses using
565 * an out-of-band data reference depends on the implementation.
567 struct hammer_data_record {
568 struct hammer_base_record base;
573 * A directory entry specifies the HAMMER filesystem object id, a copy of
574 * the file type, and file name (either embedded or as out-of-band data).
575 * If the file name is short enough to fit into den_name[] (including a
576 * terminating nul) then it will be embedded in the record, otherwise it
577 * is stored out-of-band. The base record's data reference always points
578 * to the nul-terminated filename regardless.
580 * Directory entries are indexed with a 128 bit namekey rather then an
581 * offset. A portion of the namekey is an iterator or randomizer to deal
584 * NOTE: base.base.obj_type holds the filesystem object type of obj_id,
585 * e.g. a den_type equivalent.
587 * NOTE: den_name / the filename data reference is NOT terminated with \0.
590 struct hammer_entry_record {
591 struct hammer_base_record base;
592 u_int64_t obj_id; /* object being referenced */
593 u_int64_t reserved01;
598 * Hammer rollup record
600 union hammer_record_ondisk {
601 struct hammer_base_record base;
602 struct hammer_inode_record inode;
603 struct hammer_data_record data;
604 struct hammer_entry_record entry;
607 typedef union hammer_record_ondisk *hammer_record_ondisk_t;
610 * HAMMER UNIX Attribute data
612 * The data reference in a HAMMER inode record points to this structure. Any
613 * modifications to the contents of this structure will result in a record
614 * replacement operation.
616 * short_data_off allows a small amount of data to be embedded in the
617 * hammer_inode_data structure. HAMMER typically uses this to represent
618 * up to 64 bytes of data, or to hold symlinks. Remember that allocations
619 * are in powers of 2 so 64, 192, 448, or 960 bytes of embedded data is
620 * support (64+64, 64+192, 64+448 64+960).
622 * parent_obj_id is only valid for directories (which cannot be hard-linked),
623 * and specifies the parent directory obj_id. This field will also be set
624 * for non-directory inodes as a recovery aid, but can wind up specifying
625 * stale information. However, since object id's are not reused, the worse
626 * that happens is that the recovery code is unable to use it.
628 struct hammer_inode_data {
629 u_int16_t version; /* inode data version */
630 u_int16_t mode; /* basic unix permissions */
631 u_int32_t uflags; /* chflags */
632 u_int32_t rmajor; /* used by device nodes */
633 u_int32_t rminor; /* used by device nodes */
635 u_int64_t parent_obj_id;/* parent directory obj_id */
638 /* XXX device, softlink extension */
641 #define HAMMER_INODE_DATA_VERSION 1
643 #define HAMMER_OBJID_ROOT 1
646 * Rollup various structures embedded as record data
648 union hammer_data_ondisk {
649 struct hammer_inode_data inode;