Update diskless help:
[dragonfly.git] / sys / vfs / hammer / hammer_disk.h
CommitLineData
8750964d
MD
1/*
2 * Copyright (c) 2007 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 *
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
16 * distribution.
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.
20 *
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
32 * SUCH DAMAGE.
33 *
40043e7f 34 * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.23 2008/02/10 09:51:01 dillon Exp $
8750964d
MD
35 */
36
7dc57964
MD
37#ifndef VFS_HAMMER_DISK_H_
38#define VFS_HAMMER_DISK_H_
39
8750964d
MD
40#ifndef _SYS_UUID_H_
41#include <sys/uuid.h>
42#endif
43
44/*
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.
49 *
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.
55 *
56 * A HAMMER filesystem may spam multiple volumes.
57 *
58 * A HAMMER filesystem uses a 16K filesystem buffer size. All filesystem
c60bb2c5
MD
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.
40043e7f
MD
62 *
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
8750964d 66 */
40043e7f
MD
67#define HAMMER_BUFSIZE 16384
68#define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1)
69#define HAMMER_MAXDATA (256*1024)
70#define HAMMER_BUFFER_BITS 14
71
72#if (1 << HAMMER_BUFFER_BITS) != HAMMER_BUFSIZE
73#error "HAMMER_BUFFER_BITS BROKEN"
74#endif
8750964d 75
47197d71
MD
76#define HAMMER_BUFSIZE64 ((u_int64_t)HAMMER_BUFSIZE)
77#define HAMMER_BUFMASK64 ((u_int64_t)HAMMER_BUFMASK)
78
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
86
8750964d
MD
87/*
88 * Hammer transction ids are 64 bit unsigned integers and are usually
89 * synchronized with the time of day in nanoseconds.
47197d71
MD
90 *
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.
8750964d
MD
94 */
95typedef u_int64_t hammer_tid_t;
47197d71
MD
96typedef u_int64_t hammer_off_t;
97
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 */
108
109/*
110 * hammer_off_t has several different encodings. Note that not all zones
111 * encode a vol_no.
112 *
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)
40043e7f
MD
116 * zone 3 (z,o): undo fifo - blockmap backed
117 *
118 * zone 8 (z,o): B-Tree - blkmap-backed
119 * zone 9 (z,o): Record - blkmap-backed
120 * zone 10 (z,o): Large-data - blkmap-backed
47197d71
MD
121 */
122
123#define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL
124#define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL
40043e7f
MD
125#define HAMMER_ZONE_UNDO 0x3000000000000000ULL
126#define HAMMER_ZONE_RESERVED04 0x4000000000000000ULL
127#define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL
128#define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL
129#define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL
130#define HAMMER_ZONE_BTREE 0x8000000000000000ULL
131#define HAMMER_ZONE_RECORD 0x9000000000000000ULL
132#define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL
133#define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL
134#define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL
135#define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL
136#define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL
137#define HAMMER_ZONE_RESERVED0F 0xF000000000000000ULL
138
139#define HAMMER_ZONE_RAW_VOLUME_INDEX 1
140#define HAMMER_ZONE_RAW_BUFFER_INDEX 2
141#define HAMMER_ZONE_UNDO_INDEX 3
142#define HAMMER_ZONE_BTREE_INDEX 8
143#define HAMMER_ZONE_RECORD_INDEX 9
144#define HAMMER_ZONE_LARGE_DATA_INDEX 10
145#define HAMMER_ZONE_SMALL_DATA_INDEX 11
146
147#define HAMMER_MAX_ZONES 16
47197d71
MD
148
149#define HAMMER_VOL_ENCODE(vol_no) \
150 ((hammer_off_t)((vol_no) & 255) << 52)
151#define HAMMER_VOL_DECODE(ham_off) \
152 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255)
40043e7f
MD
153#define HAMMER_ZONE_DECODE(ham_off) \
154 (int32_t)(((hammer_off_t)(ham_off) >> 60))
47197d71
MD
155#define HAMMER_SHORT_OFF_ENCODE(offset) \
156 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
157#define HAMMER_LONG_OFF_ENCODE(offset) \
158 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
159
160#define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \
161 (HAMMER_ZONE_RAW_VOLUME | \
162 HAMMER_VOL_ENCODE(vol_no) | \
163 HAMMER_SHORT_OFF_ENCODE(offset))
164
165#define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \
166 (HAMMER_ZONE_RAW_BUFFER | \
167 HAMMER_VOL_ENCODE(vol_no) | \
168 HAMMER_SHORT_OFF_ENCODE(offset))
8750964d 169
40043e7f
MD
170/*
171 * Large-Block backing store
172 *
173 * A blockmap is a two-level map which translates a blockmap-backed zone
174 * offset into a raw zone 2 offset. Each layer handles 18 bits. The 8M
175 * large-block size is 23 bits so two layers gives us 23+18+18 = 59 bits
176 * of address space.
177 */
178#define HAMMER_LARGEBLOCK_SIZE (8192 * 1024)
179#define HAMMER_LARGEBLOCK_MASK (HAMMER_LARGEBLOCK_SIZE - 1)
180#define HAMMER_LARGEBLOCK_MASK64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE - 1)
181#define HAMMER_LARGEBLOCK_BITS 23
182#if (1 << HAMMER_LARGEBLOCK_BITS) != HAMMER_LARGEBLOCK_SIZE
183#error "HAMMER_LARGEBLOCK_BITS BROKEN"
184#endif
185
186#define HAMMER_BUFFERS_PER_LARGEBLOCK \
187 (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE)
188#define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK \
189 (HAMMER_BUFFERS_PER_LARGEBLOCK - 1)
190#define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK64 \
191 ((hammer_off_t)HAMMER_BUFFERS_PER_LARGEBLOCK_MASK)
192
193#define HAMMER_BLOCKMAP_RADIX \
194 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_entry))
195#define HAMMER_BLOCKMAP_RADIX_MASK \
196 (HAMMER_BLOCKMAP_RADIX - 1)
197#define HAMMER_BLOCKMAP_BITS 18
198#if (1 << HAMMER_BLOCKMAP_BITS) != (HAMMER_LARGEBLOCK_SIZE / 32)
199#error "HAMMER_BLOCKMAP_BITS BROKEN"
200#endif
201
202#define HAMMER_LARGEBLOCK_LAYER1 \
203 ((hammer_off_t)HAMMER_LARGEBLOCK_SIZE * HAMMER_BLOCKMAP_RADIX)
204#define HAMMER_LARGEBLOCK_LAYER2 \
205 (HAMMER_LARGEBLOCK_LAYER1 * HAMMER_BLOCKMAP_RADIX)
206
207#define HAMMER_LARGEBLOCK_LAYER1_MASK (HAMMER_LARGEBLOCK_LAYER1 - 1)
208#define HAMMER_LARGEBLOCK_LAYER2_MASK (HAMMER_LARGEBLOCK_LAYER2 - 1)
209
210struct hammer_blockmap_entry {
211 hammer_off_t phys_offset; /* zone-2 physical offset */
212 int32_t bytes_free; /* bytes free within the big-block */
213 u_int32_t entry_crc;
214 u_int32_t reserved01;
215 u_int32_t reserved02;
216 hammer_off_t alloc_offset; /* zone-X logical offset */
217};
218
219typedef struct hammer_blockmap_entry *hammer_blockmap_entry_t;
66325755 220
9380c752 221/*
47197d71 222 * All on-disk HAMMER structures which make up elements of the FIFO contain
40043e7f
MD
223 * a hammer_fifo_head and hammer_fifo_tail structure. This structure
224 * contains all the information required to validate the fifo element
225 * and to scan the fifo in either direction. The head is typically embedded
226 * in higher level hammer on-disk structures while the tail is typically
227 * out-of-band. hdr_size is the size of the whole mess, including the tail.
47197d71
MD
228 *
229 * Nearly all such structures are guaranteed to not cross a 16K filesystem
230 * buffer boundary. The one exception is a record, whos related data may
231 * cross a buffer boundary.
232 *
233 * HAMMER guarantees alignment with a fifo head structure at 16MB intervals
234 * (i.e. the base of the buffer will not be in the middle of a data record).
235 * This is used to allow the recovery code to re-sync after hitting corrupted
236 * data.
40043e7f
MD
237 *
238 * PAD elements are allowed to take up only 8 bytes of space as a special
239 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
240 * and with the tail overloaded onto the head structure for 8 bytes total.
9380c752 241 */
40043e7f 242#define HAMMER_HEAD_ONDISK_SIZE 24
47197d71 243#define HAMMER_HEAD_RECOVERY_ALIGNMENT (16 * 1024 * 1024)
40043e7f 244#define HAMMER_HEAD_ALIGN 8
47197d71 245#define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
40043e7f 246#define HAMMER_TAIL_ONDISK_SIZE 8
47197d71
MD
247
248struct hammer_fifo_head {
249 u_int16_t hdr_signature;
250 u_int16_t hdr_type;
40043e7f 251 u_int32_t hdr_size; /* aligned size of the whole mess */
47197d71 252 u_int32_t hdr_crc;
40043e7f
MD
253 u_int32_t hdr_reserved02;
254 hammer_tid_t hdr_seq; /* related sequence number */
255};
256
257struct hammer_fifo_tail {
258 u_int16_t tail_signature;
259 u_int16_t tail_type;
260 u_int32_t tail_size; /* aligned size of the whole mess */
47197d71
MD
261};
262
263typedef struct hammer_fifo_head *hammer_fifo_head_t;
40043e7f 264typedef struct hammer_fifo_tail *hammer_fifo_tail_t;
9380c752 265
8750964d 266/*
47197d71 267 * Fifo header types.
8750964d 268 */
40043e7f
MD
269#define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE)
270#define HAMMER_HEAD_TYPE_VOL 0x0041U /* Volume (dummy header) */
271#define HAMMER_HEAD_TYPE_BTREE 0x0042U /* B-Tree node */
272#define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
273#define HAMMER_HEAD_TYPE_DELETE 0x0044U /* record deletion */
274#define HAMMER_HEAD_TYPE_RECORD 0x0045U /* Filesystem record */
47197d71 275
40043e7f 276#define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */
8750964d 277
47197d71 278#define HAMMER_HEAD_SIGNATURE 0xC84EU
40043e7f 279#define HAMMER_TAIL_SIGNATURE 0xC74FU
47197d71
MD
280
281/*
282 * Misc FIFO structures (except for the B-Tree node and hammer record)
283 */
284struct hammer_fifo_undo {
285 struct hammer_fifo_head head;
286 hammer_off_t undo_offset;
287 /* followed by data */
8750964d
MD
288};
289
47197d71 290typedef struct hammer_fifo_undo *hammer_fifo_undo_t;
8750964d 291
c60bb2c5 292/*
47197d71 293 * Volume header types
c60bb2c5 294 */
8750964d 295#define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
8750964d
MD
296#define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
297
298/*
299 * The B-Tree structures need hammer_fsbuf_head.
300 */
301#include "hammer_btree.h"
302
303/*
304 * HAMMER Volume header
305 *
306 * A HAMMER filesystem is built from any number of block devices, Each block
47197d71
MD
307 * device contains a volume header followed by however many buffers fit
308 * into the volume.
8750964d 309 *
47197d71
MD
310 * One of the volumes making up a HAMMER filesystem is the master, the
311 * rest are slaves. It does not have to be volume #0.
c60bb2c5 312 *
47197d71
MD
313 * The volume header takes up an entire 16K filesystem buffer and may
314 * represent up to 64KTB (65536 TB) of space.
a89aec1b
MD
315 *
316 * Special field notes:
317 *
318 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
319 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
47197d71 320 * vol_buf_beg - offset of the first buffer.
a89aec1b
MD
321 *
322 * The memory log area allows a kernel to cache new records and data
323 * in memory without allocating space in the actual filesystem to hold
324 * the records and data. In the event that a filesystem becomes full,
325 * any records remaining in memory can be flushed to the memory log
326 * area. This allows the kernel to immediately return success.
8750964d 327 */
8750964d 328
a89aec1b
MD
329#define HAMMER_BOOT_MINBYTES (32*1024)
330#define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
331#define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
332
333#define HAMMER_MEM_MINBYTES (256*1024)
334#define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
335#define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
336
8750964d 337struct hammer_volume_ondisk {
40043e7f
MD
338 u_int64_t vol_signature;/* Signature */
339
a89aec1b
MD
340 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
341 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
47197d71
MD
342 int64_t vol_buf_beg; /* byte offset of first buffer in volume */
343 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */
8750964d
MD
344 int64_t vol_locked; /* reserved clusters are >= this offset */
345
346 uuid_t vol_fsid; /* identify filesystem */
347 uuid_t vol_fstype; /* identify filesystem type */
348 char vol_name[64]; /* Name of volume */
349
350 int32_t vol_no; /* volume number within filesystem */
351 int32_t vol_count; /* number of volumes making up FS */
352
353 u_int32_t vol_version; /* version control information */
9775c955 354 u_int32_t vol_reserved01;
8750964d
MD
355 u_int32_t vol_flags; /* volume flags */
356 u_int32_t vol_rootvol; /* which volume is the root volume? */
357
40043e7f 358 int32_t vol_reserved04;
47197d71 359 int32_t vol_reserved05;
8750964d
MD
360 u_int32_t vol_reserved06;
361 u_int32_t vol_reserved07;
362
fbc6e32a 363 int32_t vol_blocksize; /* for statfs only */
47197d71 364 int32_t vol_reserved08;
fbc6e32a
MD
365 int64_t vol_nblocks; /* total allocatable hammer bufs */
366
40043e7f
MD
367 /*
368 * bigblock freemap.
369 *
370 * XXX not implemented yet, just use a sequential index at
371 * the moment.
372 */
373 hammer_off_t vol0_free_off;
374
8750964d
MD
375 /*
376 * These fields are initialized and space is reserved in every
377 * volume making up a HAMMER filesytem, but only the master volume
378 * contains valid data.
379 */
fbc6e32a
MD
380 int64_t vol0_stat_bytes; /* for statfs only */
381 int64_t vol0_stat_inodes; /* for statfs only */
b33e2cc0 382 int64_t vol0_stat_records; /* total records in filesystem */
47197d71
MD
383 hammer_off_t vol0_btree_root; /* B-Tree root */
384 hammer_tid_t vol0_next_tid; /* highest synchronized TID */
40043e7f
MD
385 hammer_tid_t vol0_next_seq; /* next SEQ no for undo */
386
387 /*
388 * Blockmaps for zones. Not all zones use a blockmap.
389 */
390 struct hammer_blockmap_entry vol0_blockmap[HAMMER_MAX_ZONES];
391
8750964d
MD
392};
393
8cd0a023
MD
394typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
395
c60bb2c5
MD
396#define HAMMER_VOLF_VALID 0x0001 /* valid entry */
397#define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
8cd0a023 398
8750964d 399/*
40043e7f
MD
400 * All HAMMER records have a common 64-byte base and a 32 byte extension,
401 * plus a possible data reference. The data reference can be in-band or
402 * out-of-band.
8750964d 403 */
40043e7f
MD
404
405#define HAMMER_RECORD_SIZE (64+32)
406
8750964d 407struct hammer_base_record {
40043e7f
MD
408 u_int32_t signature; /* record signature */
409 u_int32_t data_crc; /* data crc */
47197d71
MD
410 struct hammer_base_elm base; /* 40 byte base element */
411 hammer_off_t data_off; /* in-band or out-of-band */
412 int32_t data_len; /* size of data in bytes */
40043e7f 413 u_int32_t reserved02;
8750964d
MD
414};
415
c60bb2c5
MD
416/*
417 * Record types are fairly straightforward. The B-Tree includes the record
418 * type in its index sort.
419 *
420 * In particular please note that it is possible to create a pseudo-
421 * filesystem within a HAMMER filesystem by creating a special object
422 * type within a directory. Pseudo-filesystems are used as replication
423 * targets and even though they are built within a HAMMER filesystem they
424 * get their own obj_id space (and thus can serve as a replication target)
425 * and look like a mount point to the system.
8cd0a023
MD
426 *
427 * Inter-cluster records are special-cased in the B-Tree. These records
428 * are referenced from a B-Tree INTERNAL node, NOT A LEAF. This means
429 * that the element in the B-Tree node is actually a boundary element whos
430 * base element fields, including rec_type, reflect the boundary, NOT
431 * the inter-cluster record type.
432 *
433 * HAMMER_RECTYPE_CLUSTER - only set in the actual inter-cluster record,
434 * not set in the left or right boundary elements around the inter-cluster
435 * reference of an internal node in the B-Tree (because doing so would
436 * interfere with the boundary tests).
b3deaf57
MD
437 *
438 * NOTE: hammer_ip_delete_range_all() deletes all record types greater
439 * then HAMMER_RECTYPE_INODE.
c60bb2c5 440 */
8750964d 441#define HAMMER_RECTYPE_UNKNOWN 0
66325755 442#define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
8750964d 443#define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
c60bb2c5 444#define HAMMER_RECTYPE_PSEUDO_INODE 2 /* pseudo filesysem */
8cd0a023 445#define HAMMER_RECTYPE_CLUSTER 3 /* inter-cluster reference */
66325755
MD
446#define HAMMER_RECTYPE_DATA 0x10
447#define HAMMER_RECTYPE_DIRENTRY 0x11
448#define HAMMER_RECTYPE_DB 0x12
449#define HAMMER_RECTYPE_EXT 0x13 /* ext attributes */
7a04d74f
MD
450#define HAMMER_RECTYPE_FIX 0x14 /* fixed attribute */
451
452#define HAMMER_FIXKEY_SYMLINK 1
8750964d 453
66325755 454#define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */
8750964d
MD
455#define HAMMER_OBJTYPE_DIRECTORY 1
456#define HAMMER_OBJTYPE_REGFILE 2
457#define HAMMER_OBJTYPE_DBFILE 3
458#define HAMMER_OBJTYPE_FIFO 4
c60bb2c5
MD
459#define HAMMER_OBJTYPE_CDEV 5
460#define HAMMER_OBJTYPE_BDEV 6
461#define HAMMER_OBJTYPE_SOFTLINK 7
462#define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
463
8750964d
MD
464/*
465 * A HAMMER inode record.
466 *
467 * This forms the basis for a filesystem object. obj_id is the inode number,
468 * key1 represents the pseudo filesystem id for security partitioning
469 * (preventing cross-links and/or restricting a NFS export and specifying the
470 * security policy), and key2 represents the data retention policy id.
471 *
472 * Inode numbers are 64 bit quantities which uniquely identify a filesystem
473 * object for the ENTIRE life of the filesystem, even after the object has
474 * been deleted. For all intents and purposes inode numbers are simply
475 * allocated by incrementing a sequence space.
476 *
477 * There is an important distinction between the data stored in the inode
478 * record and the record's data reference. The record references a
479 * hammer_inode_data structure but the filesystem object size and hard link
480 * count is stored in the inode record itself. This allows multiple inodes
481 * to share the same hammer_inode_data structure. This is possible because
482 * any modifications will lay out new data. The HAMMER implementation need
483 * not use the data-sharing ability when laying down new records.
484 *
485 * A HAMMER inode is subject to the same historical storage requirements
486 * as any other record. In particular any change in filesystem or hard link
487 * count will lay down a new inode record when the filesystem is synced to
488 * disk. This can lead to a lot of junk records which get cleaned up by
489 * the data retention policy.
490 *
491 * The ino_atime and ino_mtime fields are a special case. Modifications to
492 * these fields do NOT lay down a new record by default, though the values
493 * are effectively frozen for snapshots which access historical versions
494 * of the inode record due to other operations. This means that atime will
495 * not necessarily be accurate in snapshots, backups, or mirrors. mtime
496 * will be accurate in backups and mirrors since it can be regenerated from
497 * the mirroring stream.
498 *
499 * Because nlinks is historically retained the hardlink count will be
500 * accurate when accessing a HAMMER filesystem snapshot.
501 */
502struct hammer_inode_record {
503 struct hammer_base_record base;
504 u_int64_t ino_atime; /* last access time (not historical) */
505 u_int64_t ino_mtime; /* last modified time (not historical) */
506 u_int64_t ino_size; /* filesystem object size */
507 u_int64_t ino_nlinks; /* hard links */
508};
509
510/*
511 * Data records specify the entire contents of a regular file object,
512 * including attributes. Small amounts of data can theoretically be
513 * embedded in the record itself but the use of this ability verses using
514 * an out-of-band data reference depends on the implementation.
515 */
516struct hammer_data_record {
517 struct hammer_base_record base;
40043e7f 518 char data[32];
8750964d
MD
519};
520
521/*
522 * A directory entry specifies the HAMMER filesystem object id, a copy of
523 * the file type, and file name (either embedded or as out-of-band data).
524 * If the file name is short enough to fit into den_name[] (including a
525 * terminating nul) then it will be embedded in the record, otherwise it
526 * is stored out-of-band. The base record's data reference always points
527 * to the nul-terminated filename regardless.
528 *
529 * Directory entries are indexed with a 128 bit namekey rather then an
530 * offset. A portion of the namekey is an iterator or randomizer to deal
531 * with collisions.
66325755 532 *
6b4f890b
MD
533 * NOTE: base.base.obj_type holds the filesystem object type of obj_id,
534 * e.g. a den_type equivalent.
535 *
536 * NOTE: den_name / the filename data reference is NOT terminated with \0.
66325755 537 *
8750964d
MD
538 */
539struct hammer_entry_record {
540 struct hammer_base_record base;
541 u_int64_t obj_id; /* object being referenced */
542 u_int64_t reserved01;
40043e7f 543 char name[16];
f3b0f382
MD
544};
545
8750964d
MD
546/*
547 * Hammer rollup record
548 */
c60bb2c5 549union hammer_record_ondisk {
8750964d 550 struct hammer_base_record base;
8750964d
MD
551 struct hammer_inode_record inode;
552 struct hammer_data_record data;
553 struct hammer_entry_record entry;
554};
555
c60bb2c5 556typedef union hammer_record_ondisk *hammer_record_ondisk_t;
8750964d 557
8750964d
MD
558/*
559 * HAMMER UNIX Attribute data
560 *
561 * The data reference in a HAMMER inode record points to this structure. Any
562 * modifications to the contents of this structure will result in a record
563 * replacement operation.
564 *
66325755
MD
565 * short_data_off allows a small amount of data to be embedded in the
566 * hammer_inode_data structure. HAMMER typically uses this to represent
567 * up to 64 bytes of data, or to hold symlinks. Remember that allocations
568 * are in powers of 2 so 64, 192, 448, or 960 bytes of embedded data is
569 * support (64+64, 64+192, 64+448 64+960).
570 *
571 * parent_obj_id is only valid for directories (which cannot be hard-linked),
572 * and specifies the parent directory obj_id. This field will also be set
573 * for non-directory inodes as a recovery aid, but can wind up specifying
574 * stale information. However, since object id's are not reused, the worse
575 * that happens is that the recovery code is unable to use it.
8750964d
MD
576 */
577struct hammer_inode_data {
578 u_int16_t version; /* inode data version */
579 u_int16_t mode; /* basic unix permissions */
580 u_int32_t uflags; /* chflags */
7a04d74f
MD
581 u_int32_t rmajor; /* used by device nodes */
582 u_int32_t rminor; /* used by device nodes */
66325755
MD
583 u_int64_t ctime;
584 u_int64_t parent_obj_id;/* parent directory obj_id */
8750964d
MD
585 uuid_t uid;
586 uuid_t gid;
8cd0a023 587 /* XXX device, softlink extension */
8750964d
MD
588};
589
590#define HAMMER_INODE_DATA_VERSION 1
591
7a04d74f
MD
592#define HAMMER_OBJID_ROOT 1
593
c60bb2c5
MD
594/*
595 * Rollup various structures embedded as record data
596 */
427e5fc6 597union hammer_data_ondisk {
c60bb2c5
MD
598 struct hammer_inode_data inode;
599};
600
7dc57964 601#endif