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 1.25 2008/02/23 03:01:08 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
150 * Per-zone size limitation. This just makes the iterator easier
151 * to deal with by preventing an iterator overflow.
153 #define HAMMER_ZONE_LIMIT \
154 (0x1000000000000000ULL - HAMMER_BLOCKMAP_LAYER2)
156 #define HAMMER_MAX_ZONES 16
158 #define HAMMER_VOL_ENCODE(vol_no) \
159 ((hammer_off_t)((vol_no) & 255) << 52)
160 #define HAMMER_VOL_DECODE(ham_off) \
161 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255)
162 #define HAMMER_ZONE_DECODE(ham_off) \
163 (int32_t)(((hammer_off_t)(ham_off) >> 60))
164 #define HAMMER_ZONE_ENCODE(zone, ham_off) \
165 (((hammer_off_t)(zone) << 60) | (ham_off))
166 #define HAMMER_SHORT_OFF_ENCODE(offset) \
167 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
168 #define HAMMER_LONG_OFF_ENCODE(offset) \
169 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
171 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \
172 (HAMMER_ZONE_RAW_VOLUME | \
173 HAMMER_VOL_ENCODE(vol_no) | \
174 HAMMER_SHORT_OFF_ENCODE(offset))
176 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \
177 (HAMMER_ZONE_RAW_BUFFER | \
178 HAMMER_VOL_ENCODE(vol_no) | \
179 HAMMER_SHORT_OFF_ENCODE(offset))
181 #define HAMMER_ENCODE_FREEMAP(vol_no, offset) \
182 (HAMMER_ZONE_FREEMAP | \
183 HAMMER_VOL_ENCODE(vol_no) | \
184 HAMMER_SHORT_OFF_ENCODE(offset))
187 * Large-Block backing store
189 * A blockmap is a two-level map which translates a blockmap-backed zone
190 * offset into a raw zone 2 offset. Each layer handles 18 bits. The 8M
191 * large-block size is 23 bits so two layers gives us 23+18+18 = 59 bits
194 #define HAMMER_LARGEBLOCK_SIZE (8192 * 1024)
195 #define HAMMER_LARGEBLOCK_SIZE64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE)
196 #define HAMMER_LARGEBLOCK_MASK (HAMMER_LARGEBLOCK_SIZE - 1)
197 #define HAMMER_LARGEBLOCK_MASK64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE - 1)
198 #define HAMMER_LARGEBLOCK_BITS 23
199 #if (1 << HAMMER_LARGEBLOCK_BITS) != HAMMER_LARGEBLOCK_SIZE
200 #error "HAMMER_LARGEBLOCK_BITS BROKEN"
203 #define HAMMER_BUFFERS_PER_LARGEBLOCK \
204 (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE)
205 #define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK \
206 (HAMMER_BUFFERS_PER_LARGEBLOCK - 1)
207 #define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK64 \
208 ((hammer_off_t)HAMMER_BUFFERS_PER_LARGEBLOCK_MASK)
211 * Every blockmap has this root structure in the root volume header.
213 struct hammer_blockmap {
214 hammer_off_t phys_offset; /* zone-2 physical offset */
215 hammer_off_t next_offset; /* zone-X logical offset */
216 hammer_off_t alloc_offset; /* zone-X logical offset */
218 u_int32_t reserved01;
221 typedef struct hammer_blockmap *hammer_blockmap_t;
224 * The blockmap is a 2-layer entity made up of big-blocks. The first layer
225 * contains 262144 32-byte entries (18 bits), the second layer contains
226 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
227 * 18+19+23 = 60 bits. The top four bits are the zone id.
229 * Layer 2 encodes the physical bigblock mapping for a blockmap. The freemap
230 * uses this field to encode the virtual blockmap offset that allocated the
233 * NOTE: The freemap maps the vol_no in the upper 8 bits of layer1.
235 * zone-4 blockmap offset: [z:4][layer1:18][layer2:19][bigblock:23]
237 struct hammer_blockmap_layer1 {
238 hammer_off_t blocks_free; /* big-blocks free */
239 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */
240 u_int32_t layer1_crc; /* crc of this entry */
241 u_int32_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */
242 hammer_off_t reserved01;
245 struct hammer_blockmap_layer2 {
247 u_int32_t bytes_free; /* bytes free within this bigblock */
249 hammer_off_t owner; /* used by freemap */
250 hammer_off_t phys_offset; /* used by blockmap */
254 #define HAMMER_BLOCKMAP_FREE 0ULL
255 #define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL)
257 #define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \
258 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1))
259 #define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \
260 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2))
262 #define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \
263 (HAMMER_BLOCKMAP_RADIX1 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
264 #define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \
265 (HAMMER_BLOCKMAP_RADIX2 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
267 #define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 */ \
268 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
269 #define HAMMER_BLOCKMAP_LAYER2 /* 19+23 */ \
270 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_LARGEBLOCK_SIZE64)
272 #define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1)
273 #define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1)
276 * byte offset within layer1 or layer2 big-block for the entry representing
277 * a zone-2 physical offset.
279 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \
280 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \
281 HAMMER_BLOCKMAP_LAYER2 * sizeof(struct hammer_blockmap_layer1))
283 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \
284 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \
285 HAMMER_LARGEBLOCK_SIZE64 * sizeof(struct hammer_blockmap_layer2))
288 * All on-disk HAMMER structures which make up elements of the FIFO contain
289 * a hammer_fifo_head and hammer_fifo_tail structure. This structure
290 * contains all the information required to validate the fifo element
291 * and to scan the fifo in either direction. The head is typically embedded
292 * in higher level hammer on-disk structures while the tail is typically
293 * out-of-band. hdr_size is the size of the whole mess, including the tail.
295 * Nearly all such structures are guaranteed to not cross a 16K filesystem
296 * buffer boundary. The one exception is a record, whos related data may
297 * cross a buffer boundary.
299 * HAMMER guarantees alignment with a fifo head structure at 16MB intervals
300 * (i.e. the base of the buffer will not be in the middle of a data record).
301 * This is used to allow the recovery code to re-sync after hitting corrupted
304 * PAD elements are allowed to take up only 8 bytes of space as a special
305 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
306 * and with the tail overloaded onto the head structure for 8 bytes total.
308 #define HAMMER_HEAD_ONDISK_SIZE 24
309 #define HAMMER_HEAD_RECOVERY_ALIGNMENT (16 * 1024 * 1024)
310 #define HAMMER_HEAD_ALIGN 8
311 #define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
312 #define HAMMER_TAIL_ONDISK_SIZE 8
314 struct hammer_fifo_head {
315 u_int16_t hdr_signature;
317 u_int32_t hdr_size; /* aligned size of the whole mess */
319 u_int32_t hdr_reserved02;
320 hammer_tid_t hdr_seq; /* related sequence number */
323 struct hammer_fifo_tail {
324 u_int16_t tail_signature;
326 u_int32_t tail_size; /* aligned size of the whole mess */
329 typedef struct hammer_fifo_head *hammer_fifo_head_t;
330 typedef struct hammer_fifo_tail *hammer_fifo_tail_t;
335 #define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE)
336 #define HAMMER_HEAD_TYPE_VOL 0x0041U /* Volume (dummy header) */
337 #define HAMMER_HEAD_TYPE_BTREE 0x0042U /* B-Tree node */
338 #define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
339 #define HAMMER_HEAD_TYPE_DELETE 0x0044U /* record deletion */
340 #define HAMMER_HEAD_TYPE_RECORD 0x0045U /* Filesystem record */
342 #define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */
344 #define HAMMER_HEAD_SIGNATURE 0xC84EU
345 #define HAMMER_TAIL_SIGNATURE 0xC74FU
348 * Misc FIFO structures (except for the B-Tree node and hammer record)
350 struct hammer_fifo_undo {
351 struct hammer_fifo_head head;
352 hammer_off_t undo_offset;
353 /* followed by data */
356 typedef struct hammer_fifo_undo *hammer_fifo_undo_t;
359 * Volume header types
361 #define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
362 #define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
365 * The B-Tree structures need hammer_fsbuf_head.
367 #include "hammer_btree.h"
370 * HAMMER Volume header
372 * A HAMMER filesystem is built from any number of block devices, Each block
373 * device contains a volume header followed by however many buffers fit
376 * One of the volumes making up a HAMMER filesystem is the master, the
377 * rest are slaves. It does not have to be volume #0.
379 * The volume header takes up an entire 16K filesystem buffer and may
380 * represent up to 64KTB (65536 TB) of space.
382 * Special field notes:
384 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
385 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
386 * vol_buf_beg - offset of the first buffer.
388 * The memory log area allows a kernel to cache new records and data
389 * in memory without allocating space in the actual filesystem to hold
390 * the records and data. In the event that a filesystem becomes full,
391 * any records remaining in memory can be flushed to the memory log
392 * area. This allows the kernel to immediately return success.
395 #define HAMMER_BOOT_MINBYTES (32*1024)
396 #define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
397 #define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
399 #define HAMMER_MEM_MINBYTES (256*1024)
400 #define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
401 #define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
403 struct hammer_volume_ondisk {
404 u_int64_t vol_signature;/* Signature */
406 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
407 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
408 int64_t vol_buf_beg; /* byte offset of first buffer in volume */
409 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */
410 int64_t vol_locked; /* reserved clusters are >= this offset */
412 uuid_t vol_fsid; /* identify filesystem */
413 uuid_t vol_fstype; /* identify filesystem type */
414 char vol_name[64]; /* Name of volume */
416 int32_t vol_no; /* volume number within filesystem */
417 int32_t vol_count; /* number of volumes making up FS */
419 u_int32_t vol_version; /* version control information */
420 u_int32_t vol_reserved01;
421 u_int32_t vol_flags; /* volume flags */
422 u_int32_t vol_rootvol; /* which volume is the root volume? */
424 int32_t vol_reserved04;
425 int32_t vol_reserved05;
426 u_int32_t vol_reserved06;
427 u_int32_t vol_reserved07;
429 int32_t vol_blocksize; /* for statfs only */
430 int32_t vol_reserved08;
431 int64_t vol_nblocks; /* total allocatable hammer bufs */
434 * These fields are initialized and space is reserved in every
435 * volume making up a HAMMER filesytem, but only the master volume
436 * contains valid data.
438 int64_t vol0_stat_bigblocks; /* total bigblocks when fs is empty */
439 int64_t vol0_stat_freebigblocks;/* number of free bigblocks */
440 int64_t vol0_stat_bytes; /* for statfs only */
441 int64_t vol0_stat_inodes; /* for statfs only */
442 int64_t vol0_stat_records; /* total records in filesystem */
443 hammer_off_t vol0_btree_root; /* B-Tree root */
444 hammer_tid_t vol0_next_tid; /* highest synchronized TID */
445 hammer_tid_t vol0_next_seq; /* next SEQ no for undo */
448 * Blockmaps for zones. Not all zones use a blockmap.
450 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES];
454 typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
456 #define HAMMER_VOLF_VALID 0x0001 /* valid entry */
457 #define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
460 * All HAMMER records have a common 64-byte base and a 32 byte extension,
461 * plus a possible data reference. The data reference can be in-band or
465 #define HAMMER_RECORD_SIZE (64+32)
467 struct hammer_base_record {
468 u_int32_t signature; /* record signature */
469 u_int32_t data_crc; /* data crc */
470 struct hammer_base_elm base; /* 40 byte base element */
471 hammer_off_t data_off; /* in-band or out-of-band */
472 int32_t data_len; /* size of data in bytes */
473 u_int32_t reserved02;
477 * Record types are fairly straightforward. The B-Tree includes the record
478 * type in its index sort.
480 * In particular please note that it is possible to create a pseudo-
481 * filesystem within a HAMMER filesystem by creating a special object
482 * type within a directory. Pseudo-filesystems are used as replication
483 * targets and even though they are built within a HAMMER filesystem they
484 * get their own obj_id space (and thus can serve as a replication target)
485 * and look like a mount point to the system.
487 * Inter-cluster records are special-cased in the B-Tree. These records
488 * are referenced from a B-Tree INTERNAL node, NOT A LEAF. This means
489 * that the element in the B-Tree node is actually a boundary element whos
490 * base element fields, including rec_type, reflect the boundary, NOT
491 * the inter-cluster record type.
493 * HAMMER_RECTYPE_CLUSTER - only set in the actual inter-cluster record,
494 * not set in the left or right boundary elements around the inter-cluster
495 * reference of an internal node in the B-Tree (because doing so would
496 * interfere with the boundary tests).
498 * NOTE: hammer_ip_delete_range_all() deletes all record types greater
499 * then HAMMER_RECTYPE_INODE.
501 #define HAMMER_RECTYPE_UNKNOWN 0
502 #define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
503 #define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
504 #define HAMMER_RECTYPE_PSEUDO_INODE 2 /* pseudo filesysem */
505 #define HAMMER_RECTYPE_CLUSTER 3 /* inter-cluster reference */
506 #define HAMMER_RECTYPE_DATA 0x10
507 #define HAMMER_RECTYPE_DIRENTRY 0x11
508 #define HAMMER_RECTYPE_DB 0x12
509 #define HAMMER_RECTYPE_EXT 0x13 /* ext attributes */
510 #define HAMMER_RECTYPE_FIX 0x14 /* fixed attribute */
512 #define HAMMER_FIXKEY_SYMLINK 1
514 #define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */
515 #define HAMMER_OBJTYPE_DIRECTORY 1
516 #define HAMMER_OBJTYPE_REGFILE 2
517 #define HAMMER_OBJTYPE_DBFILE 3
518 #define HAMMER_OBJTYPE_FIFO 4
519 #define HAMMER_OBJTYPE_CDEV 5
520 #define HAMMER_OBJTYPE_BDEV 6
521 #define HAMMER_OBJTYPE_SOFTLINK 7
522 #define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
525 * A HAMMER inode record.
527 * This forms the basis for a filesystem object. obj_id is the inode number,
528 * key1 represents the pseudo filesystem id for security partitioning
529 * (preventing cross-links and/or restricting a NFS export and specifying the
530 * security policy), and key2 represents the data retention policy id.
532 * Inode numbers are 64 bit quantities which uniquely identify a filesystem
533 * object for the ENTIRE life of the filesystem, even after the object has
534 * been deleted. For all intents and purposes inode numbers are simply
535 * allocated by incrementing a sequence space.
537 * There is an important distinction between the data stored in the inode
538 * record and the record's data reference. The record references a
539 * hammer_inode_data structure but the filesystem object size and hard link
540 * count is stored in the inode record itself. This allows multiple inodes
541 * to share the same hammer_inode_data structure. This is possible because
542 * any modifications will lay out new data. The HAMMER implementation need
543 * not use the data-sharing ability when laying down new records.
545 * A HAMMER inode is subject to the same historical storage requirements
546 * as any other record. In particular any change in filesystem or hard link
547 * count will lay down a new inode record when the filesystem is synced to
548 * disk. This can lead to a lot of junk records which get cleaned up by
549 * the data retention policy.
551 * The ino_atime and ino_mtime fields are a special case. Modifications to
552 * these fields do NOT lay down a new record by default, though the values
553 * are effectively frozen for snapshots which access historical versions
554 * of the inode record due to other operations. This means that atime will
555 * not necessarily be accurate in snapshots, backups, or mirrors. mtime
556 * will be accurate in backups and mirrors since it can be regenerated from
557 * the mirroring stream.
559 * Because nlinks is historically retained the hardlink count will be
560 * accurate when accessing a HAMMER filesystem snapshot.
562 struct hammer_inode_record {
563 struct hammer_base_record base;
564 u_int64_t ino_atime; /* last access time (not historical) */
565 u_int64_t ino_mtime; /* last modified time (not historical) */
566 u_int64_t ino_size; /* filesystem object size */
567 u_int64_t ino_nlinks; /* hard links */
571 * Data records specify the entire contents of a regular file object,
572 * including attributes. Small amounts of data can theoretically be
573 * embedded in the record itself but the use of this ability verses using
574 * an out-of-band data reference depends on the implementation.
576 struct hammer_data_record {
577 struct hammer_base_record base;
582 * A directory entry specifies the HAMMER filesystem object id, a copy of
583 * the file type, and file name (either embedded or as out-of-band data).
584 * If the file name is short enough to fit into den_name[] (including a
585 * terminating nul) then it will be embedded in the record, otherwise it
586 * is stored out-of-band. The base record's data reference always points
587 * to the nul-terminated filename regardless.
589 * Directory entries are indexed with a 128 bit namekey rather then an
590 * offset. A portion of the namekey is an iterator or randomizer to deal
593 * NOTE: base.base.obj_type holds the filesystem object type of obj_id,
594 * e.g. a den_type equivalent.
596 * NOTE: den_name / the filename data reference is NOT terminated with \0.
599 struct hammer_entry_record {
600 struct hammer_base_record base;
601 u_int64_t obj_id; /* object being referenced */
602 u_int64_t reserved01;
607 * Hammer rollup record
609 union hammer_record_ondisk {
610 struct hammer_base_record base;
611 struct hammer_inode_record inode;
612 struct hammer_data_record data;
613 struct hammer_entry_record entry;
616 typedef union hammer_record_ondisk *hammer_record_ondisk_t;
619 * HAMMER UNIX Attribute data
621 * The data reference in a HAMMER inode record points to this structure. Any
622 * modifications to the contents of this structure will result in a record
623 * replacement operation.
625 * short_data_off allows a small amount of data to be embedded in the
626 * hammer_inode_data structure. HAMMER typically uses this to represent
627 * up to 64 bytes of data, or to hold symlinks. Remember that allocations
628 * are in powers of 2 so 64, 192, 448, or 960 bytes of embedded data is
629 * support (64+64, 64+192, 64+448 64+960).
631 * parent_obj_id is only valid for directories (which cannot be hard-linked),
632 * and specifies the parent directory obj_id. This field will also be set
633 * for non-directory inodes as a recovery aid, but can wind up specifying
634 * stale information. However, since object id's are not reused, the worse
635 * that happens is that the recovery code is unable to use it.
637 struct hammer_inode_data {
638 u_int16_t version; /* inode data version */
639 u_int16_t mode; /* basic unix permissions */
640 u_int32_t uflags; /* chflags */
641 u_int32_t rmajor; /* used by device nodes */
642 u_int32_t rminor; /* used by device nodes */
644 u_int64_t parent_obj_id;/* parent directory obj_id */
647 /* XXX device, softlink extension */
650 #define HAMMER_INODE_DATA_VERSION 1
652 #define HAMMER_OBJID_ROOT 1
655 * Rollup various structures embedded as record data
657 union hammer_data_ondisk {
658 struct hammer_inode_data inode;