Strip "serialized" suffix
[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 *
47637bff 34 * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.35 2008/06/07 07:41:51 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 *
00e31441 56 * A HAMMER filesystem may span multiple volumes.
8750964d
MD
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 96typedef u_int64_t hammer_off_t;
059819e3
MD
97typedef u_int32_t hammer_seq_t;
98typedef u_int32_t hammer_crc_t;
47197d71
MD
99
100#define HAMMER_MIN_TID 0ULL /* unsigned */
101#define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */
102#define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */
103#define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */
104#define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */
105#define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */
106#define HAMMER_MIN_RECTYPE 0x0U /* unsigned */
107#define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */
108#define HAMMER_MIN_OFFSET 0ULL /* unsigned */
109#define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */
110
111/*
112 * hammer_off_t has several different encodings. Note that not all zones
113 * encode a vol_no.
114 *
115 * zone 0 (z,v,o): reserved (for sanity)
116 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header)
117 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer)
bf686dbe
MD
118 * zone 3 (z,o): undo fifo - fixed layer2 array in root vol hdr
119 * zone 4 (z,v,o): freemap - freemap-backed self-mapping special
120 * cased layering.
40043e7f
MD
121 *
122 * zone 8 (z,o): B-Tree - blkmap-backed
123 * zone 9 (z,o): Record - blkmap-backed
124 * zone 10 (z,o): Large-data - blkmap-backed
47197d71
MD
125 */
126
127#define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL
128#define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL
40043e7f 129#define HAMMER_ZONE_UNDO 0x3000000000000000ULL
c3be93f2 130#define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL
40043e7f
MD
131#define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL
132#define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL
133#define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL
134#define HAMMER_ZONE_BTREE 0x8000000000000000ULL
135#define HAMMER_ZONE_RECORD 0x9000000000000000ULL
136#define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL
137#define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL
138#define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL
139#define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL
140#define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL
141#define HAMMER_ZONE_RESERVED0F 0xF000000000000000ULL
142
143#define HAMMER_ZONE_RAW_VOLUME_INDEX 1
144#define HAMMER_ZONE_RAW_BUFFER_INDEX 2
145#define HAMMER_ZONE_UNDO_INDEX 3
c3be93f2 146#define HAMMER_ZONE_FREEMAP_INDEX 4
40043e7f
MD
147#define HAMMER_ZONE_BTREE_INDEX 8
148#define HAMMER_ZONE_RECORD_INDEX 9
149#define HAMMER_ZONE_LARGE_DATA_INDEX 10
150#define HAMMER_ZONE_SMALL_DATA_INDEX 11
151
f03c9cf4
MD
152/*
153 * Per-zone size limitation. This just makes the iterator easier
154 * to deal with by preventing an iterator overflow.
155 */
156#define HAMMER_ZONE_LIMIT \
0729c8c8 157 (0x1000000000000000ULL - HAMMER_BLOCKMAP_LAYER2 * 2)
f03c9cf4 158
40043e7f 159#define HAMMER_MAX_ZONES 16
47197d71
MD
160
161#define HAMMER_VOL_ENCODE(vol_no) \
162 ((hammer_off_t)((vol_no) & 255) << 52)
163#define HAMMER_VOL_DECODE(ham_off) \
164 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255)
40043e7f
MD
165#define HAMMER_ZONE_DECODE(ham_off) \
166 (int32_t)(((hammer_off_t)(ham_off) >> 60))
f03c9cf4
MD
167#define HAMMER_ZONE_ENCODE(zone, ham_off) \
168 (((hammer_off_t)(zone) << 60) | (ham_off))
47197d71
MD
169#define HAMMER_SHORT_OFF_ENCODE(offset) \
170 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
171#define HAMMER_LONG_OFF_ENCODE(offset) \
172 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
173
174#define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \
175 (HAMMER_ZONE_RAW_VOLUME | \
176 HAMMER_VOL_ENCODE(vol_no) | \
177 HAMMER_SHORT_OFF_ENCODE(offset))
178
179#define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \
180 (HAMMER_ZONE_RAW_BUFFER | \
181 HAMMER_VOL_ENCODE(vol_no) | \
182 HAMMER_SHORT_OFF_ENCODE(offset))
8750964d 183
c3be93f2
MD
184#define HAMMER_ENCODE_FREEMAP(vol_no, offset) \
185 (HAMMER_ZONE_FREEMAP | \
186 HAMMER_VOL_ENCODE(vol_no) | \
187 HAMMER_SHORT_OFF_ENCODE(offset))
188
40043e7f
MD
189/*
190 * Large-Block backing store
191 *
192 * A blockmap is a two-level map which translates a blockmap-backed zone
193 * offset into a raw zone 2 offset. Each layer handles 18 bits. The 8M
194 * large-block size is 23 bits so two layers gives us 23+18+18 = 59 bits
195 * of address space.
196 */
197#define HAMMER_LARGEBLOCK_SIZE (8192 * 1024)
c3be93f2 198#define HAMMER_LARGEBLOCK_SIZE64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE)
40043e7f
MD
199#define HAMMER_LARGEBLOCK_MASK (HAMMER_LARGEBLOCK_SIZE - 1)
200#define HAMMER_LARGEBLOCK_MASK64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE - 1)
201#define HAMMER_LARGEBLOCK_BITS 23
202#if (1 << HAMMER_LARGEBLOCK_BITS) != HAMMER_LARGEBLOCK_SIZE
203#error "HAMMER_LARGEBLOCK_BITS BROKEN"
204#endif
205
206#define HAMMER_BUFFERS_PER_LARGEBLOCK \
207 (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE)
208#define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK \
209 (HAMMER_BUFFERS_PER_LARGEBLOCK - 1)
210#define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK64 \
211 ((hammer_off_t)HAMMER_BUFFERS_PER_LARGEBLOCK_MASK)
212
c3be93f2
MD
213/*
214 * Every blockmap has this root structure in the root volume header.
bf686dbe
MD
215 *
216 * NOTE: zone 3 (the undo FIFO) does not use phys_offset. first and next
217 * offsets represent the FIFO.
c3be93f2
MD
218 */
219struct hammer_blockmap {
220 hammer_off_t phys_offset; /* zone-2 physical offset */
bf686dbe 221 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */
c3be93f2
MD
222 hammer_off_t next_offset; /* zone-X logical offset */
223 hammer_off_t alloc_offset; /* zone-X logical offset */
c3be93f2 224 u_int32_t reserved01;
19619882 225 hammer_crc_t entry_crc;
c3be93f2 226};
40043e7f 227
c3be93f2 228typedef struct hammer_blockmap *hammer_blockmap_t;
40043e7f 229
19619882
MD
230#define HAMMER_BLOCKMAP_CRCSIZE \
231 offsetof(struct hammer_blockmap, entry_crc)
232
c3be93f2
MD
233/*
234 * The blockmap is a 2-layer entity made up of big-blocks. The first layer
235 * contains 262144 32-byte entries (18 bits), the second layer contains
236 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
237 * 18+19+23 = 60 bits. The top four bits are the zone id.
238 *
239 * Layer 2 encodes the physical bigblock mapping for a blockmap. The freemap
240 * uses this field to encode the virtual blockmap offset that allocated the
241 * physical block.
242 *
243 * NOTE: The freemap maps the vol_no in the upper 8 bits of layer1.
244 *
245 * zone-4 blockmap offset: [z:4][layer1:18][layer2:19][bigblock:23]
246 */
247struct hammer_blockmap_layer1 {
248 hammer_off_t blocks_free; /* big-blocks free */
249 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */
c3be93f2 250 hammer_off_t reserved01;
19619882
MD
251 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */
252 /* (not yet used) */
253 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/
c3be93f2 254};
40043e7f 255
47637bff
MD
256typedef struct hammer_blockmap_layer1 *hammer_blockmap_layer1_t;
257
19619882
MD
258#define HAMMER_LAYER1_CRCSIZE \
259 offsetof(struct hammer_blockmap_layer1, layer1_crc)
260
c3be93f2 261struct hammer_blockmap_layer2 {
c3be93f2
MD
262 union {
263 hammer_off_t owner; /* used by freemap */
264 hammer_off_t phys_offset; /* used by blockmap */
265 } u;
19619882
MD
266 u_int32_t bytes_free; /* bytes free within this bigblock */
267 hammer_crc_t entry_crc;
40043e7f
MD
268};
269
47637bff
MD
270typedef struct hammer_blockmap_layer2 *hammer_blockmap_layer2_t;
271
19619882
MD
272#define HAMMER_LAYER2_CRCSIZE \
273 offsetof(struct hammer_blockmap_layer2, entry_crc)
274
c3be93f2
MD
275#define HAMMER_BLOCKMAP_FREE 0ULL
276#define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL)
277
278#define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \
279 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1))
280#define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \
281 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2))
282
283#define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \
284 (HAMMER_BLOCKMAP_RADIX1 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
285#define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \
286 (HAMMER_BLOCKMAP_RADIX2 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
287
288#define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 */ \
289 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
6f97fce3 290#define HAMMER_BLOCKMAP_LAYER2 /* 19+23 - 4TB */ \
c3be93f2
MD
291 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_LARGEBLOCK_SIZE64)
292
293#define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1)
294#define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1)
295
296/*
297 * byte offset within layer1 or layer2 big-block for the entry representing
298 * a zone-2 physical offset.
299 */
300#define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \
301 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \
302 HAMMER_BLOCKMAP_LAYER2 * sizeof(struct hammer_blockmap_layer1))
303
304#define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \
305 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \
306 HAMMER_LARGEBLOCK_SIZE64 * sizeof(struct hammer_blockmap_layer2))
66325755 307
bf686dbe
MD
308/*
309 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume
310 * header with an array of layer2 structures. A maximum of (64x8MB) = 512MB
311 * may be reserved. The size of the undo fifo is usually set a newfs time
312 * but can be adjusted if the filesystem is taken offline.
313 */
314
315#define HAMMER_UNDO_LAYER2 64 /* max layer2 undo mapping entries */
316
9380c752 317/*
36f82b23
MD
318 * All on-disk HAMMER structures which make up elements of the UNDO FIFO
319 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure
40043e7f
MD
320 * contains all the information required to validate the fifo element
321 * and to scan the fifo in either direction. The head is typically embedded
322 * in higher level hammer on-disk structures while the tail is typically
323 * out-of-band. hdr_size is the size of the whole mess, including the tail.
47197d71 324 *
36f82b23
MD
325 * All undo structures are guaranteed to not cross a 16K filesystem
326 * buffer boundary. Most undo structures are fairly small. Data spaces
327 * are not immediately reused by HAMMER so file data is not usually recorded
328 * as part of an UNDO.
40043e7f
MD
329 *
330 * PAD elements are allowed to take up only 8 bytes of space as a special
331 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
332 * and with the tail overloaded onto the head structure for 8 bytes total.
36f82b23
MD
333 *
334 * Every undo record has a sequence number. This number is unrelated to
335 * transaction ids and instead collects the undo transactions associated
336 * with a single atomic operation. A larger transactional operation, such
337 * as a remove(), may consist of several smaller atomic operations
338 * representing raw meta-data operations.
9380c752 339 */
36f82b23 340#define HAMMER_HEAD_ONDISK_SIZE 32
40043e7f 341#define HAMMER_HEAD_ALIGN 8
47197d71 342#define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
40043e7f 343#define HAMMER_TAIL_ONDISK_SIZE 8
47197d71
MD
344
345struct hammer_fifo_head {
346 u_int16_t hdr_signature;
347 u_int16_t hdr_type;
40043e7f 348 u_int32_t hdr_size; /* aligned size of the whole mess */
059819e3 349 u_int32_t reserved01; /* (0) reserved for future use */
19619882 350 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */
40043e7f
MD
351};
352
19619882
MD
353#define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc)
354
40043e7f
MD
355struct hammer_fifo_tail {
356 u_int16_t tail_signature;
357 u_int16_t tail_type;
358 u_int32_t tail_size; /* aligned size of the whole mess */
47197d71
MD
359};
360
361typedef struct hammer_fifo_head *hammer_fifo_head_t;
40043e7f 362typedef struct hammer_fifo_tail *hammer_fifo_tail_t;
9380c752 363
8750964d 364/*
47197d71 365 * Fifo header types.
8750964d 366 */
40043e7f
MD
367#define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE)
368#define HAMMER_HEAD_TYPE_VOL 0x0041U /* Volume (dummy header) */
369#define HAMMER_HEAD_TYPE_BTREE 0x0042U /* B-Tree node */
370#define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
371#define HAMMER_HEAD_TYPE_DELETE 0x0044U /* record deletion */
372#define HAMMER_HEAD_TYPE_RECORD 0x0045U /* Filesystem record */
47197d71 373
40043e7f 374#define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */
8750964d 375
47197d71 376#define HAMMER_HEAD_SIGNATURE 0xC84EU
40043e7f 377#define HAMMER_TAIL_SIGNATURE 0xC74FU
47197d71 378
36f82b23
MD
379#define HAMMER_HEAD_SEQ_BEG 0x80000000U
380#define HAMMER_HEAD_SEQ_END 0x40000000U
381#define HAMMER_HEAD_SEQ_MASK 0x3FFFFFFFU
382
47197d71 383/*
bf686dbe 384 * Misc FIFO structures.
47197d71
MD
385 */
386struct hammer_fifo_undo {
387 struct hammer_fifo_head head;
bf686dbe
MD
388 hammer_off_t undo_offset; /* zone-1 offset */
389 int32_t undo_data_bytes;
390 int32_t undo_reserved01;
47197d71 391 /* followed by data */
8750964d
MD
392};
393
47197d71 394typedef struct hammer_fifo_undo *hammer_fifo_undo_t;
8750964d 395
bf686dbe
MD
396struct hammer_fifo_buf_commit {
397 hammer_off_t undo_offset;
398};
399
c60bb2c5 400/*
47197d71 401 * Volume header types
c60bb2c5 402 */
8750964d 403#define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
8750964d
MD
404#define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
405
406/*
407 * The B-Tree structures need hammer_fsbuf_head.
408 */
409#include "hammer_btree.h"
410
411/*
412 * HAMMER Volume header
413 *
414 * A HAMMER filesystem is built from any number of block devices, Each block
47197d71
MD
415 * device contains a volume header followed by however many buffers fit
416 * into the volume.
8750964d 417 *
47197d71
MD
418 * One of the volumes making up a HAMMER filesystem is the master, the
419 * rest are slaves. It does not have to be volume #0.
c60bb2c5 420 *
47197d71
MD
421 * The volume header takes up an entire 16K filesystem buffer and may
422 * represent up to 64KTB (65536 TB) of space.
a89aec1b
MD
423 *
424 * Special field notes:
425 *
426 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
427 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
47197d71 428 * vol_buf_beg - offset of the first buffer.
a89aec1b
MD
429 *
430 * The memory log area allows a kernel to cache new records and data
431 * in memory without allocating space in the actual filesystem to hold
432 * the records and data. In the event that a filesystem becomes full,
433 * any records remaining in memory can be flushed to the memory log
434 * area. This allows the kernel to immediately return success.
8750964d 435 */
8750964d 436
a89aec1b
MD
437#define HAMMER_BOOT_MINBYTES (32*1024)
438#define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
439#define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
440
441#define HAMMER_MEM_MINBYTES (256*1024)
442#define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
443#define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
444
8750964d 445struct hammer_volume_ondisk {
40043e7f
MD
446 u_int64_t vol_signature;/* Signature */
447
a89aec1b
MD
448 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
449 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
47197d71
MD
450 int64_t vol_buf_beg; /* byte offset of first buffer in volume */
451 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */
8750964d
MD
452 int64_t vol_locked; /* reserved clusters are >= this offset */
453
454 uuid_t vol_fsid; /* identify filesystem */
455 uuid_t vol_fstype; /* identify filesystem type */
456 char vol_name[64]; /* Name of volume */
457
458 int32_t vol_no; /* volume number within filesystem */
459 int32_t vol_count; /* number of volumes making up FS */
460
461 u_int32_t vol_version; /* version control information */
19619882 462 hammer_crc_t vol_crc; /* header crc */
8750964d
MD
463 u_int32_t vol_flags; /* volume flags */
464 u_int32_t vol_rootvol; /* which volume is the root volume? */
465
40043e7f 466 int32_t vol_reserved04;
47197d71 467 int32_t vol_reserved05;
8750964d
MD
468 u_int32_t vol_reserved06;
469 u_int32_t vol_reserved07;
470
fbc6e32a 471 int32_t vol_blocksize; /* for statfs only */
47197d71 472 int32_t vol_reserved08;
fbc6e32a
MD
473 int64_t vol_nblocks; /* total allocatable hammer bufs */
474
8750964d
MD
475 /*
476 * These fields are initialized and space is reserved in every
477 * volume making up a HAMMER filesytem, but only the master volume
478 * contains valid data.
479 */
c3be93f2
MD
480 int64_t vol0_stat_bigblocks; /* total bigblocks when fs is empty */
481 int64_t vol0_stat_freebigblocks;/* number of free bigblocks */
fbc6e32a
MD
482 int64_t vol0_stat_bytes; /* for statfs only */
483 int64_t vol0_stat_inodes; /* for statfs only */
b33e2cc0 484 int64_t vol0_stat_records; /* total records in filesystem */
47197d71
MD
485 hammer_off_t vol0_btree_root; /* B-Tree root */
486 hammer_tid_t vol0_next_tid; /* highest synchronized TID */
0729c8c8 487 hammer_off_t vol0_zone_limit; /* limit the zone size */
40043e7f
MD
488
489 /*
0729c8c8
MD
490 * Blockmaps for zones. Not all zones use a blockmap. Note that
491 * the entire root blockmap is cached in the hammer_mount structure.
40043e7f 492 */
c3be93f2 493 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES];
40043e7f 494
bf686dbe
MD
495 /*
496 * Layer-2 array for undo fifo
497 */
498 struct hammer_blockmap_layer2 vol0_undo_array[HAMMER_UNDO_LAYER2];
499
8750964d
MD
500};
501
8cd0a023
MD
502typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
503
c60bb2c5
MD
504#define HAMMER_VOLF_VALID 0x0001 /* valid entry */
505#define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
8cd0a023 506
19619882
MD
507#define HAMMER_VOL_CRCSIZE1 \
508 offsetof(struct hammer_volume_ondisk, vol_crc)
509#define HAMMER_VOL_CRCSIZE2 \
510 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \
511 sizeof(hammer_crc_t))
512
c60bb2c5
MD
513/*
514 * Record types are fairly straightforward. The B-Tree includes the record
515 * type in its index sort.
516 *
517 * In particular please note that it is possible to create a pseudo-
518 * filesystem within a HAMMER filesystem by creating a special object
519 * type within a directory. Pseudo-filesystems are used as replication
520 * targets and even though they are built within a HAMMER filesystem they
521 * get their own obj_id space (and thus can serve as a replication target)
522 * and look like a mount point to the system.
8cd0a023 523 *
b3deaf57
MD
524 * NOTE: hammer_ip_delete_range_all() deletes all record types greater
525 * then HAMMER_RECTYPE_INODE.
c60bb2c5 526 */
8750964d 527#define HAMMER_RECTYPE_UNKNOWN 0
66325755 528#define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
8750964d 529#define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
c60bb2c5 530#define HAMMER_RECTYPE_PSEUDO_INODE 2 /* pseudo filesysem */
8cd0a023 531#define HAMMER_RECTYPE_CLUSTER 3 /* inter-cluster reference */
bf686dbe
MD
532#define HAMMER_RECTYPE_DATA 0x0010
533#define HAMMER_RECTYPE_DIRENTRY 0x0011
534#define HAMMER_RECTYPE_DB 0x0012
535#define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */
536#define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */
537#define HAMMER_RECTYPE_MOVED 0x8000 /* special recovery flag */
7a04d74f
MD
538
539#define HAMMER_FIXKEY_SYMLINK 1
8750964d 540
66325755 541#define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */
8750964d
MD
542#define HAMMER_OBJTYPE_DIRECTORY 1
543#define HAMMER_OBJTYPE_REGFILE 2
544#define HAMMER_OBJTYPE_DBFILE 3
545#define HAMMER_OBJTYPE_FIFO 4
c60bb2c5
MD
546#define HAMMER_OBJTYPE_CDEV 5
547#define HAMMER_OBJTYPE_BDEV 6
548#define HAMMER_OBJTYPE_SOFTLINK 7
549#define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
550
8750964d 551/*
11ad5ade 552 * HAMMER inode attribute data
8750964d 553 *
11ad5ade
MD
554 * The data reference for a HAMMER inode points to this structure. Any
555 * modifications to the contents of this structure will result in a
556 * replacement operation.
8750964d 557 *
11ad5ade
MD
558 * parent_obj_id is only valid for directories (which cannot be hard-linked),
559 * and specifies the parent directory obj_id. This field will also be set
560 * for non-directory inodes as a recovery aid, but can wind up specifying
561 * stale information. However, since object id's are not reused, the worse
562 * that happens is that the recovery code is unable to use it.
8750964d 563 *
11ad5ade
MD
564 * NOTE: atime is stored in the inode's B-Tree element and not in the inode
565 * data. This allows the atime to be updated without having to lay down a
566 * new record.
8750964d 567 */
11ad5ade
MD
568struct hammer_inode_data {
569 u_int16_t version; /* inode data version */
570 u_int16_t mode; /* basic unix permissions */
571 u_int32_t uflags; /* chflags */
572 u_int32_t rmajor; /* used by device nodes */
573 u_int32_t rminor; /* used by device nodes */
574 u_int64_t ctime;
575 u_int64_t parent_obj_id;/* parent directory obj_id */
576 uuid_t uid;
577 uuid_t gid;
578
579 u_int8_t obj_type;
580 u_int8_t reserved01;
581 u_int16_t reserved02;
582 u_int32_t reserved03;
583 u_int64_t mtime;
584 u_int64_t size; /* filesystem object size */
585 u_int64_t nlinks; /* hard links */
2f85fa4d
MD
586 u_int64_t reserved04;
587 union {
588 char reserved06[24];
589 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */
590 } ext;
8750964d
MD
591};
592
11ad5ade
MD
593#define HAMMER_INODE_DATA_VERSION 1
594#define HAMMER_OBJID_ROOT 1
2f85fa4d 595#define HAMMER_INODE_BASESYMLEN 24
8750964d
MD
596
597/*
598 * A directory entry specifies the HAMMER filesystem object id, a copy of
599 * the file type, and file name (either embedded or as out-of-band data).
600 * If the file name is short enough to fit into den_name[] (including a
601 * terminating nul) then it will be embedded in the record, otherwise it
602 * is stored out-of-band. The base record's data reference always points
603 * to the nul-terminated filename regardless.
604 *
605 * Directory entries are indexed with a 128 bit namekey rather then an
606 * offset. A portion of the namekey is an iterator or randomizer to deal
607 * with collisions.
66325755 608 *
6b4f890b
MD
609 * NOTE: base.base.obj_type holds the filesystem object type of obj_id,
610 * e.g. a den_type equivalent.
611 *
612 * NOTE: den_name / the filename data reference is NOT terminated with \0.
66325755 613 *
8750964d 614 */
11ad5ade 615struct hammer_entry_data {
8750964d
MD
616 u_int64_t obj_id; /* object being referenced */
617 u_int64_t reserved01;
11ad5ade 618 char name[16]; /* name (extended) */
8750964d
MD
619};
620
11ad5ade
MD
621#define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_entry_data, name[0])
622#define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_entry_data, name[nlen])
19619882 623
11ad5ade
MD
624struct hammer_symlink_data {
625 char name[16];
8750964d
MD
626};
627
11ad5ade 628#define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0])
7a04d74f 629
c60bb2c5
MD
630/*
631 * Rollup various structures embedded as record data
632 */
427e5fc6 633union hammer_data_ondisk {
11ad5ade 634 struct hammer_entry_data entry;
c60bb2c5 635 struct hammer_inode_data inode;
11ad5ade 636 struct hammer_symlink_data symlink;
c60bb2c5
MD
637};
638
11ad5ade
MD
639typedef union hammer_data_ondisk *hammer_data_ondisk_t;
640
7dc57964 641#endif