HAMMER VFS - Version 4 part 1/many - UNDO FIFO layout work.
[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 *
44a83111 34 * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.55 2008/11/13 02:18:43 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 62 *
4a2796f3
MD
63 * 64K X-bufs are used for blocks >= a file's 1MB mark.
64 *
40043e7f
MD
65 * Per-volume storage limit: 52 bits 4096 TB
66 * Per-Zone storage limit: 59 bits 512 KTB (due to blockmap)
67 * Per-filesystem storage limit: 60 bits 1 MTB
8750964d 68 */
40043e7f 69#define HAMMER_BUFSIZE 16384
4a2796f3
MD
70#define HAMMER_XBUFSIZE 65536
71#define HAMMER_XDEMARC (1024 * 1024)
40043e7f 72#define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1)
4a2796f3 73#define HAMMER_XBUFMASK (HAMMER_XBUFSIZE - 1)
40043e7f
MD
74#define HAMMER_BUFFER_BITS 14
75
76#if (1 << HAMMER_BUFFER_BITS) != HAMMER_BUFSIZE
77#error "HAMMER_BUFFER_BITS BROKEN"
78#endif
8750964d 79
47197d71
MD
80#define HAMMER_BUFSIZE64 ((u_int64_t)HAMMER_BUFSIZE)
81#define HAMMER_BUFMASK64 ((u_int64_t)HAMMER_BUFMASK)
82
4a2796f3
MD
83#define HAMMER_XBUFSIZE64 ((u_int64_t)HAMMER_XBUFSIZE)
84#define HAMMER_XBUFMASK64 ((u_int64_t)HAMMER_XBUFMASK)
85
47197d71
MD
86#define HAMMER_OFF_ZONE_MASK 0xF000000000000000ULL /* zone portion */
87#define HAMMER_OFF_VOL_MASK 0x0FF0000000000000ULL /* volume portion */
88#define HAMMER_OFF_SHORT_MASK 0x000FFFFFFFFFFFFFULL /* offset portion */
89#define HAMMER_OFF_LONG_MASK 0x0FFFFFFFFFFFFFFFULL /* offset portion */
90#define HAMMER_OFF_SHORT_REC_MASK 0x000FFFFFFF000000ULL /* recovery boundary */
91#define HAMMER_OFF_LONG_REC_MASK 0x0FFFFFFFFF000000ULL /* recovery boundary */
92#define HAMMER_RECOVERY_BND 0x0000000001000000ULL
93
8750964d 94/*
94509284
MD
95 * The current limit of volumes that can make up a HAMMER FS
96 */
97#define HAMMER_MAX_VOLUMES 256
98
99/*
8750964d
MD
100 * Hammer transction ids are 64 bit unsigned integers and are usually
101 * synchronized with the time of day in nanoseconds.
47197d71
MD
102 *
103 * Hammer offsets are used for FIFO indexing and embed a cycle counter
104 * and volume number in addition to the offset. Most offsets are required
105 * to be 64-byte aligned.
8750964d
MD
106 */
107typedef u_int64_t hammer_tid_t;
47197d71 108typedef u_int64_t hammer_off_t;
059819e3
MD
109typedef u_int32_t hammer_seq_t;
110typedef u_int32_t hammer_crc_t;
47197d71
MD
111
112#define HAMMER_MIN_TID 0ULL /* unsigned */
113#define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */
114#define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */
115#define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */
116#define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */
117#define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */
118#define HAMMER_MIN_RECTYPE 0x0U /* unsigned */
119#define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */
120#define HAMMER_MIN_OFFSET 0ULL /* unsigned */
121#define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */
122
123/*
124 * hammer_off_t has several different encodings. Note that not all zones
125 * encode a vol_no.
126 *
cb51be26 127 * zone 0: reserved for sanity
47197d71
MD
128 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header)
129 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer)
cb51be26
MD
130 * zone 3 (z,o): undo fifo - actually fixed phys array in vol hdr
131 * zone 4 (z,v,o): freemap - only real blockmap
132 * zone 8 (z,v,o): B-Tree - actually zone-2 address
133 * zone 9 (z,v,o): Record - actually zone-2 address
134 * zone 10 (z,v,o): Large-data - actually zone-2 address
135 * zone 15: reserved for sanity
d40bfeca
MD
136 *
137 * layer1/layer2 direct map:
138 * zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo
139 * ----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo
47197d71
MD
140 */
141
142#define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL
143#define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL
40043e7f 144#define HAMMER_ZONE_UNDO 0x3000000000000000ULL
c3be93f2 145#define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL
40043e7f
MD
146#define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL
147#define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL
148#define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL
149#define HAMMER_ZONE_BTREE 0x8000000000000000ULL
bf3b416b 150#define HAMMER_ZONE_META 0x9000000000000000ULL
40043e7f
MD
151#define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL
152#define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL
153#define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL
154#define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL
155#define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL
cb51be26 156#define HAMMER_ZONE_UNAVAIL 0xF000000000000000ULL
40043e7f
MD
157
158#define HAMMER_ZONE_RAW_VOLUME_INDEX 1
159#define HAMMER_ZONE_RAW_BUFFER_INDEX 2
160#define HAMMER_ZONE_UNDO_INDEX 3
c3be93f2 161#define HAMMER_ZONE_FREEMAP_INDEX 4
40043e7f 162#define HAMMER_ZONE_BTREE_INDEX 8
bf3b416b 163#define HAMMER_ZONE_META_INDEX 9
40043e7f
MD
164#define HAMMER_ZONE_LARGE_DATA_INDEX 10
165#define HAMMER_ZONE_SMALL_DATA_INDEX 11
cb51be26 166#define HAMMER_ZONE_UNAVAIL_INDEX 15 /* unavailable */
f03c9cf4 167
40043e7f 168#define HAMMER_MAX_ZONES 16
47197d71
MD
169
170#define HAMMER_VOL_ENCODE(vol_no) \
171 ((hammer_off_t)((vol_no) & 255) << 52)
172#define HAMMER_VOL_DECODE(ham_off) \
173 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255)
40043e7f
MD
174#define HAMMER_ZONE_DECODE(ham_off) \
175 (int32_t)(((hammer_off_t)(ham_off) >> 60))
f03c9cf4
MD
176#define HAMMER_ZONE_ENCODE(zone, ham_off) \
177 (((hammer_off_t)(zone) << 60) | (ham_off))
47197d71
MD
178#define HAMMER_SHORT_OFF_ENCODE(offset) \
179 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
180#define HAMMER_LONG_OFF_ENCODE(offset) \
181 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
182
183#define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \
184 (HAMMER_ZONE_RAW_VOLUME | \
185 HAMMER_VOL_ENCODE(vol_no) | \
186 HAMMER_SHORT_OFF_ENCODE(offset))
187
188#define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \
189 (HAMMER_ZONE_RAW_BUFFER | \
190 HAMMER_VOL_ENCODE(vol_no) | \
191 HAMMER_SHORT_OFF_ENCODE(offset))
8750964d 192
c3be93f2
MD
193#define HAMMER_ENCODE_FREEMAP(vol_no, offset) \
194 (HAMMER_ZONE_FREEMAP | \
195 HAMMER_VOL_ENCODE(vol_no) | \
196 HAMMER_SHORT_OFF_ENCODE(offset))
197
40043e7f
MD
198/*
199 * Large-Block backing store
200 *
201 * A blockmap is a two-level map which translates a blockmap-backed zone
202 * offset into a raw zone 2 offset. Each layer handles 18 bits. The 8M
203 * large-block size is 23 bits so two layers gives us 23+18+18 = 59 bits
204 * of address space.
df2ccbac
MD
205 *
206 * When using hinting for a blockmap lookup, the hint is lost when the
207 * scan leaves the HINTBLOCK, which is typically several LARGEBLOCK's.
208 * HINTBLOCK is a heuristic.
40043e7f 209 */
df2ccbac
MD
210#define HAMMER_HINTBLOCK_SIZE (HAMMER_LARGEBLOCK_SIZE * 4)
211#define HAMMER_HINTBLOCK_MASK64 ((u_int64_t)HAMMER_HINTBLOCK_SIZE - 1)
40043e7f 212#define HAMMER_LARGEBLOCK_SIZE (8192 * 1024)
df2ccbac 213#define HAMMER_LARGEBLOCK_OVERFILL (6144 * 1024)
c3be93f2 214#define HAMMER_LARGEBLOCK_SIZE64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE)
40043e7f
MD
215#define HAMMER_LARGEBLOCK_MASK (HAMMER_LARGEBLOCK_SIZE - 1)
216#define HAMMER_LARGEBLOCK_MASK64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE - 1)
217#define HAMMER_LARGEBLOCK_BITS 23
218#if (1 << HAMMER_LARGEBLOCK_BITS) != HAMMER_LARGEBLOCK_SIZE
219#error "HAMMER_LARGEBLOCK_BITS BROKEN"
220#endif
221
222#define HAMMER_BUFFERS_PER_LARGEBLOCK \
223 (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE)
224#define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK \
225 (HAMMER_BUFFERS_PER_LARGEBLOCK - 1)
226#define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK64 \
227 ((hammer_off_t)HAMMER_BUFFERS_PER_LARGEBLOCK_MASK)
228
c3be93f2 229/*
5de0c0e5
MD
230 * Maximum number of mirrors operating in master mode (multi-master
231 * clustering and mirroring).
232 */
233#define HAMMER_MAX_MASTERS 16
234
235/*
cb51be26
MD
236 * The blockmap is somewhat of a degenerate structure. HAMMER only actually
237 * uses it in its original incarnation to implement the free-map.
bf686dbe 238 *
cb51be26
MD
239 * zone:1 raw volume (no blockmap)
240 * zone:2 raw buffer (no blockmap)
241 * zone:3 undo-map (direct layer2 array in volume header)
242 * zone:4 free-map (the only real blockmap)
243 * zone:8-15 zone id used to classify big-block only, address is actually
244 * a zone-2 address.
c3be93f2
MD
245 */
246struct hammer_blockmap {
247 hammer_off_t phys_offset; /* zone-2 physical offset */
bf686dbe 248 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */
c3be93f2
MD
249 hammer_off_t next_offset; /* zone-X logical offset */
250 hammer_off_t alloc_offset; /* zone-X logical offset */
c3be93f2 251 u_int32_t reserved01;
19619882 252 hammer_crc_t entry_crc;
c3be93f2 253};
40043e7f 254
c3be93f2 255typedef struct hammer_blockmap *hammer_blockmap_t;
40043e7f 256
19619882
MD
257#define HAMMER_BLOCKMAP_CRCSIZE \
258 offsetof(struct hammer_blockmap, entry_crc)
259
c3be93f2
MD
260/*
261 * The blockmap is a 2-layer entity made up of big-blocks. The first layer
262 * contains 262144 32-byte entries (18 bits), the second layer contains
263 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
264 * 18+19+23 = 60 bits. The top four bits are the zone id.
265 *
cb51be26
MD
266 * Currently only the freemap utilizes both layers in all their glory.
267 * All primary data/meta-data zones actually encode a zone-2 address
268 * requiring no real blockmap translation.
c3be93f2 269 *
cb51be26
MD
270 * The freemap uses the upper 8 bits of layer-1 to identify the volume,
271 * thus any space allocated via the freemap can be directly translated
272 * to a zone:2 (or zone:8-15) address.
c3be93f2 273 *
cb51be26 274 * zone-X blockmap offset: [z:4][layer1:18][layer2:19][bigblock:23]
c3be93f2
MD
275 */
276struct hammer_blockmap_layer1 {
277 hammer_off_t blocks_free; /* big-blocks free */
278 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */
c3be93f2 279 hammer_off_t reserved01;
19619882
MD
280 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */
281 /* (not yet used) */
282 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/
c3be93f2 283};
40043e7f 284
47637bff
MD
285typedef struct hammer_blockmap_layer1 *hammer_blockmap_layer1_t;
286
19619882
MD
287#define HAMMER_LAYER1_CRCSIZE \
288 offsetof(struct hammer_blockmap_layer1, layer1_crc)
289
c3be93f2 290struct hammer_blockmap_layer2 {
cb51be26
MD
291 u_int8_t zone; /* typed allocation zone */
292 u_int8_t unused01;
293 u_int16_t unused02;
294 u_int32_t append_off; /* allocatable space index */
19619882
MD
295 u_int32_t bytes_free; /* bytes free within this bigblock */
296 hammer_crc_t entry_crc;
40043e7f
MD
297};
298
47637bff
MD
299typedef struct hammer_blockmap_layer2 *hammer_blockmap_layer2_t;
300
19619882
MD
301#define HAMMER_LAYER2_CRCSIZE \
302 offsetof(struct hammer_blockmap_layer2, entry_crc)
303
c3be93f2
MD
304#define HAMMER_BLOCKMAP_FREE 0ULL
305#define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL)
306
307#define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \
308 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1))
309#define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \
310 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2))
311
312#define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \
313 (HAMMER_BLOCKMAP_RADIX1 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
314#define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \
315 (HAMMER_BLOCKMAP_RADIX2 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
316
317#define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 */ \
318 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
6f97fce3 319#define HAMMER_BLOCKMAP_LAYER2 /* 19+23 - 4TB */ \
c3be93f2
MD
320 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_LARGEBLOCK_SIZE64)
321
322#define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1)
323#define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1)
324
325/*
326 * byte offset within layer1 or layer2 big-block for the entry representing
327 * a zone-2 physical offset.
328 */
329#define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \
330 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \
331 HAMMER_BLOCKMAP_LAYER2 * sizeof(struct hammer_blockmap_layer1))
332
333#define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \
334 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \
335 HAMMER_LARGEBLOCK_SIZE64 * sizeof(struct hammer_blockmap_layer2))
66325755 336
8750964d 337/*
bf686dbe 338 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume
cb51be26 339 * header with an array of layer2 structures. A maximum of (128x8MB) = 1GB
bf686dbe
MD
340 * may be reserved. The size of the undo fifo is usually set a newfs time
341 * but can be adjusted if the filesystem is taken offline.
342 */
cb51be26 343#define HAMMER_UNDO_LAYER2 128 /* max layer2 undo mapping entries */
bf686dbe
MD
344
345/*
36f82b23
MD
346 * All on-disk HAMMER structures which make up elements of the UNDO FIFO
347 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure
40043e7f
MD
348 * contains all the information required to validate the fifo element
349 * and to scan the fifo in either direction. The head is typically embedded
350 * in higher level hammer on-disk structures while the tail is typically
351 * out-of-band. hdr_size is the size of the whole mess, including the tail.
47197d71 352 *
36f82b23
MD
353 * All undo structures are guaranteed to not cross a 16K filesystem
354 * buffer boundary. Most undo structures are fairly small. Data spaces
355 * are not immediately reused by HAMMER so file data is not usually recorded
356 * as part of an UNDO.
40043e7f
MD
357 *
358 * PAD elements are allowed to take up only 8 bytes of space as a special
359 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
360 * and with the tail overloaded onto the head structure for 8 bytes total.
36f82b23
MD
361 *
362 * Every undo record has a sequence number. This number is unrelated to
363 * transaction ids and instead collects the undo transactions associated
364 * with a single atomic operation. A larger transactional operation, such
365 * as a remove(), may consist of several smaller atomic operations
366 * representing raw meta-data operations.
02428fb6
MD
367 *
368 * HAMMER VERSION 4 CHANGES
369 *
370 * In HAMMER version 4 the undo structure alignment is reduced from 16384
371 * to 512 bytes in order to ensure that each 512 byte sector begins with
372 * a header. The reserved01 field in the header is now a 32 bit sequence
373 * number. This allows the recovery code to detect missing sectors
374 * without relying on the 32-bit crc and to definitively identify the current
375 * undo sequence space without having to rely on information from the volume
376 * header. In addition, new REDO entries in the undo space are used to
377 * record write, write/extend, and transaction id updates.
378 *
379 * The grand result is:
380 *
381 * (1) The volume header no longer needs to be synchronized for most
382 * flush and fsync operations.
383 *
384 * (2) Most fsync operations need only lay down REDO records
385 *
386 * (3) Data overwrite for nohistory operations covered by REDO records
387 * can be supported (instead of rolling a new block allocation),
388 * by rolling UNDO for the prior contents of the data.
9380c752 389 */
36f82b23 390#define HAMMER_HEAD_ONDISK_SIZE 32
40043e7f 391#define HAMMER_HEAD_ALIGN 8
47197d71 392#define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
40043e7f 393#define HAMMER_TAIL_ONDISK_SIZE 8
4c038e17
MD
394#define HAMMER_HEAD_DOALIGN(bytes) \
395 (((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK)
47197d71 396
02428fb6
MD
397#define HAMMER_UNDO_ALIGN 512
398#define HAMMER_UNDO_ALIGN64 ((u_int64_t)512)
399#define HAMMER_UNDO_MASK (HAMMER_UNDO_ALIGN - 1)
400#define HAMMER_UNDO_MASK64 (HAMMER_UNDO_ALIGN64 - 1)
401
47197d71
MD
402struct hammer_fifo_head {
403 u_int16_t hdr_signature;
404 u_int16_t hdr_type;
02428fb6
MD
405 u_int32_t hdr_size; /* Aligned size of the whole mess */
406 u_int32_t hdr_seq; /* Sequence number */
19619882 407 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */
40043e7f
MD
408};
409
19619882
MD
410#define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc)
411
40043e7f
MD
412struct hammer_fifo_tail {
413 u_int16_t tail_signature;
414 u_int16_t tail_type;
415 u_int32_t tail_size; /* aligned size of the whole mess */
47197d71
MD
416};
417
418typedef struct hammer_fifo_head *hammer_fifo_head_t;
40043e7f 419typedef struct hammer_fifo_tail *hammer_fifo_tail_t;
9380c752
MD
420
421/*
47197d71 422 * Fifo header types.
8750964d 423 */
40043e7f 424#define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE)
02428fb6
MD
425#define HAMMER_HEAD_TYPE_DUMMY 0x0041U /* dummy entry w/seqno */
426#define HAMMER_HEAD_TYPE_42 0x0042U
40043e7f 427#define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
02428fb6
MD
428#define HAMMER_HEAD_TYPE_REDO 0x0044U /* data REDO / fast fsync */
429#define HAMMER_HEAD_TYPE_45 0x0045U
47197d71 430
40043e7f 431#define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */
8750964d 432
47197d71 433#define HAMMER_HEAD_SIGNATURE 0xC84EU
40043e7f 434#define HAMMER_TAIL_SIGNATURE 0xC74FU
47197d71 435
36f82b23
MD
436#define HAMMER_HEAD_SEQ_BEG 0x80000000U
437#define HAMMER_HEAD_SEQ_END 0x40000000U
438#define HAMMER_HEAD_SEQ_MASK 0x3FFFFFFFU
439
47197d71 440/*
bf686dbe 441 * Misc FIFO structures.
02428fb6
MD
442 *
443 * NOTE: redo records are for version 4+ filesystems.
47197d71
MD
444 */
445struct hammer_fifo_undo {
446 struct hammer_fifo_head head;
bf686dbe
MD
447 hammer_off_t undo_offset; /* zone-1 offset */
448 int32_t undo_data_bytes;
449 int32_t undo_reserved01;
47197d71 450 /* followed by data */
8750964d
MD
451};
452
02428fb6
MD
453struct hammer_fifo_redo {
454 struct hammer_fifo_head head;
455 int64_t redo_objid; /* file being written */
456 hammer_off_t redo_offset; /* logical offset in file */
457 int32_t redo_data_bytes;
458 int32_t redo_reserved01;
459};
8750964d 460
02428fb6
MD
461union hammer_fifo_any {
462 struct hammer_fifo_head head;
463 struct hammer_fifo_undo undo;
464 struct hammer_fifo_redo redo;
bf686dbe
MD
465};
466
02428fb6
MD
467typedef struct hammer_fifo_redo *hammer_fifo_redo_t;
468typedef struct hammer_fifo_undo *hammer_fifo_undo_t;
469typedef union hammer_fifo_any *hammer_fifo_any_t;
470
c60bb2c5 471/*
47197d71 472 * Volume header types
c60bb2c5 473 */
8750964d 474#define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
8750964d
MD
475#define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
476
477/*
478 * The B-Tree structures need hammer_fsbuf_head.
479 */
480#include "hammer_btree.h"
481
482/*
483 * HAMMER Volume header
484 *
485 * A HAMMER filesystem is built from any number of block devices, Each block
47197d71
MD
486 * device contains a volume header followed by however many buffers fit
487 * into the volume.
8750964d 488 *
47197d71
MD
489 * One of the volumes making up a HAMMER filesystem is the master, the
490 * rest are slaves. It does not have to be volume #0.
c60bb2c5 491 *
47197d71
MD
492 * The volume header takes up an entire 16K filesystem buffer and may
493 * represent up to 64KTB (65536 TB) of space.
a89aec1b
MD
494 *
495 * Special field notes:
496 *
497 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
498 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
47197d71 499 * vol_buf_beg - offset of the first buffer.
a89aec1b
MD
500 *
501 * The memory log area allows a kernel to cache new records and data
502 * in memory without allocating space in the actual filesystem to hold
503 * the records and data. In the event that a filesystem becomes full,
504 * any records remaining in memory can be flushed to the memory log
505 * area. This allows the kernel to immediately return success.
8750964d 506 */
8750964d 507
a89aec1b
MD
508#define HAMMER_BOOT_MINBYTES (32*1024)
509#define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
510#define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
511
512#define HAMMER_MEM_MINBYTES (256*1024)
513#define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
514#define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
515
8750964d 516struct hammer_volume_ondisk {
40043e7f
MD
517 u_int64_t vol_signature;/* Signature */
518
a89aec1b
MD
519 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
520 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
47197d71
MD
521 int64_t vol_buf_beg; /* byte offset of first buffer in volume */
522 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */
8750964d
MD
523 int64_t vol_locked; /* reserved clusters are >= this offset */
524
525 uuid_t vol_fsid; /* identify filesystem */
526 uuid_t vol_fstype; /* identify filesystem type */
527 char vol_name[64]; /* Name of volume */
528
529 int32_t vol_no; /* volume number within filesystem */
530 int32_t vol_count; /* number of volumes making up FS */
531
532 u_int32_t vol_version; /* version control information */
19619882 533 hammer_crc_t vol_crc; /* header crc */
8750964d
MD
534 u_int32_t vol_flags; /* volume flags */
535 u_int32_t vol_rootvol; /* which volume is the root volume? */
536
40043e7f 537 int32_t vol_reserved04;
47197d71 538 int32_t vol_reserved05;
8750964d
MD
539 u_int32_t vol_reserved06;
540 u_int32_t vol_reserved07;
541
fbc6e32a 542 int32_t vol_blocksize; /* for statfs only */
47197d71 543 int32_t vol_reserved08;
fbc6e32a
MD
544 int64_t vol_nblocks; /* total allocatable hammer bufs */
545
546 /*
8750964d
MD
547 * These fields are initialized and space is reserved in every
548 * volume making up a HAMMER filesytem, but only the master volume
549 * contains valid data.
550 */
c3be93f2
MD
551 int64_t vol0_stat_bigblocks; /* total bigblocks when fs is empty */
552 int64_t vol0_stat_freebigblocks;/* number of free bigblocks */
fbc6e32a
MD
553 int64_t vol0_stat_bytes; /* for statfs only */
554 int64_t vol0_stat_inodes; /* for statfs only */
b33e2cc0 555 int64_t vol0_stat_records; /* total records in filesystem */
47197d71 556 hammer_off_t vol0_btree_root; /* B-Tree root */
4889cbd4 557 hammer_tid_t vol0_next_tid; /* highest partially synchronized TID */
5fa5c92f 558 hammer_off_t vol0_unused03;
40043e7f
MD
559
560 /*
0729c8c8
MD
561 * Blockmaps for zones. Not all zones use a blockmap. Note that
562 * the entire root blockmap is cached in the hammer_mount structure.
40043e7f 563 */
c3be93f2 564 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES];
40043e7f 565
bf686dbe 566 /*
cb51be26 567 * Array of zone-2 addresses for undo FIFO.
bf686dbe 568 */
cb51be26 569 hammer_off_t vol0_undo_array[HAMMER_UNDO_LAYER2];
bf686dbe 570
8750964d
MD
571};
572
8cd0a023
MD
573typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
574
c60bb2c5
MD
575#define HAMMER_VOLF_VALID 0x0001 /* valid entry */
576#define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
748efb59 577#define HAMMER_VOLF_NEEDFLUSH 0x0004 /* volume needs flush */
8cd0a023 578
19619882
MD
579#define HAMMER_VOL_CRCSIZE1 \
580 offsetof(struct hammer_volume_ondisk, vol_crc)
581#define HAMMER_VOL_CRCSIZE2 \
582 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \
583 sizeof(hammer_crc_t))
584
44a83111 585#define HAMMER_VOL_VERSION_MIN 1 /* minimum supported version */
02428fb6
MD
586#define HAMMER_VOL_VERSION_DEFAULT 3 /* newfs default version */
587#define HAMMER_VOL_VERSION_WIP 4 /* version >= this is WIP */
588#define HAMMER_VOL_VERSION_MAX 4 /* maximum supported version */
44a83111 589
5e435c92 590#define HAMMER_VOL_VERSION_ONE 1
beec5dc4 591#define HAMMER_VOL_VERSION_TWO 2 /* new dirent layout (2.3+) */
83f2a3aa 592#define HAMMER_VOL_VERSION_THREE 3 /* new snapshot layout (2.5+) */
02428fb6 593#define HAMMER_VOL_VERSION_FOUR 4 /* new undo/redo/flush (2.5+) */
83f2a3aa 594
8750964d 595/*
c60bb2c5
MD
596 * Record types are fairly straightforward. The B-Tree includes the record
597 * type in its index sort.
c60bb2c5 598 */
8750964d 599#define HAMMER_RECTYPE_UNKNOWN 0
66325755 600#define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
8750964d 601#define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
ddfdf542
MD
602#define HAMMER_RECTYPE_UNUSED02 2
603#define HAMMER_RECTYPE_UNUSED03 3
bf686dbe
MD
604#define HAMMER_RECTYPE_DATA 0x0010
605#define HAMMER_RECTYPE_DIRENTRY 0x0011
606#define HAMMER_RECTYPE_DB 0x0012
607#define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */
608#define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */
ea434b6f 609#define HAMMER_RECTYPE_PFS 0x0015 /* PFS management */
83f2a3aa
MD
610#define HAMMER_RECTYPE_SNAPSHOT 0x0016 /* Snapshot management */
611#define HAMMER_RECTYPE_CONFIG 0x0017 /* hammer cleanup config */
bf686dbe 612#define HAMMER_RECTYPE_MOVED 0x8000 /* special recovery flag */
a9d52b76
MD
613#define HAMMER_RECTYPE_MAX 0xFFFF
614
615#define HAMMER_RECTYPE_CLEAN_START HAMMER_RECTYPE_EXT
7a04d74f
MD
616
617#define HAMMER_FIXKEY_SYMLINK 1
8750964d 618
66325755 619#define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */
8750964d
MD
620#define HAMMER_OBJTYPE_DIRECTORY 1
621#define HAMMER_OBJTYPE_REGFILE 2
622#define HAMMER_OBJTYPE_DBFILE 3
623#define HAMMER_OBJTYPE_FIFO 4
c60bb2c5
MD
624#define HAMMER_OBJTYPE_CDEV 5
625#define HAMMER_OBJTYPE_BDEV 6
626#define HAMMER_OBJTYPE_SOFTLINK 7
627#define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
b3bad96f 628#define HAMMER_OBJTYPE_SOCKET 9
c60bb2c5 629
8750964d 630/*
11ad5ade 631 * HAMMER inode attribute data
8750964d 632 *
11ad5ade
MD
633 * The data reference for a HAMMER inode points to this structure. Any
634 * modifications to the contents of this structure will result in a
635 * replacement operation.
8750964d 636 *
5a930e66
MD
637 * parent_obj_id is only valid for directories (which cannot be hard-linked),
638 * and specifies the parent directory obj_id. This field will also be set
639 * for non-directory inodes as a recovery aid, but can wind up holding
640 * stale information. However, since object id's are not reused, the worse
641 * that happens is that the recovery code is unable to use it.
ddfdf542
MD
642 *
643 * NOTE: Future note on directory hardlinks. We can implement a record type
644 * which allows us to point to multiple parent directories.
8750964d 645 *
11ad5ade
MD
646 * NOTE: atime is stored in the inode's B-Tree element and not in the inode
647 * data. This allows the atime to be updated without having to lay down a
648 * new record.
8750964d 649 */
11ad5ade
MD
650struct hammer_inode_data {
651 u_int16_t version; /* inode data version */
652 u_int16_t mode; /* basic unix permissions */
653 u_int32_t uflags; /* chflags */
654 u_int32_t rmajor; /* used by device nodes */
655 u_int32_t rminor; /* used by device nodes */
656 u_int64_t ctime;
adf01747 657 int64_t parent_obj_id; /* parent directory obj_id */
11ad5ade
MD
658 uuid_t uid;
659 uuid_t gid;
660
661 u_int8_t obj_type;
5a930e66
MD
662 u_int8_t cap_flags; /* capability support flags (extension) */
663 u_int16_t reserved02;
ddfdf542 664 u_int32_t reserved03; /* RESERVED FOR POSSIBLE FUTURE BIRTHTIME */
11ad5ade 665 u_int64_t nlinks; /* hard links */
bcac4bbb 666 u_int64_t size; /* filesystem object size */
2f85fa4d 667 union {
5a930e66
MD
668 struct {
669 char reserved06[16];
670 u_int32_t parent_obj_localization;
671 u_int32_t integrity_crc;
672 } obj;
2f85fa4d
MD
673 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */
674 } ext;
ddfdf542
MD
675 u_int64_t mtime; /* mtime must be second-to-last */
676 u_int64_t atime; /* atime must be last */
8750964d
MD
677};
678
ddfdf542
MD
679/*
680 * Neither mtime nor atime upates are CRCd by the B-Tree element.
681 * mtime updates have UNDO, atime updates do not.
682 */
bcac4bbb
MD
683#define HAMMER_ITIMES_BASE(ino_data) (&(ino_data)->mtime)
684#define HAMMER_ITIMES_BYTES (sizeof(u_int64_t) * 2)
685
ddfdf542
MD
686#define HAMMER_INODE_CRCSIZE \
687 offsetof(struct hammer_inode_data, mtime)
688
11ad5ade
MD
689#define HAMMER_INODE_DATA_VERSION 1
690#define HAMMER_OBJID_ROOT 1
ddfdf542 691#define HAMMER_INODE_BASESYMLEN 24 /* see ext.symlink */
8750964d
MD
692
693/*
5e435c92 694 * Capability & implementation flags.
beec5dc4
MD
695 *
696 * DIR_LOCAL_INO - Use inode B-Tree localization for directory entries.
5e435c92
MD
697 */
698#define HAMMER_INODE_CAP_DIRHASH_MASK 0x03 /* directory: hash algorithm */
699#define HAMMER_INODE_CAP_DIRHASH_ALG0 0x00
700#define HAMMER_INODE_CAP_DIRHASH_ALG1 0x01
701#define HAMMER_INODE_CAP_DIRHASH_ALG2 0x02
702#define HAMMER_INODE_CAP_DIRHASH_ALG3 0x03
beec5dc4 703#define HAMMER_INODE_CAP_DIR_LOCAL_INO 0x04 /* use inode localization */
5e435c92
MD
704
705/*
ddfdf542
MD
706 * A HAMMER directory entry associates a HAMMER filesystem object with a
707 * namespace. It is possible to hook into a pseudo-filesystem (with its
708 * own inode numbering space) in the filesystem by setting the high
709 * 16 bits of the localization field. The low 16 bits must be 0 and
710 * are reserved for future use.
8750964d
MD
711 *
712 * Directory entries are indexed with a 128 bit namekey rather then an
ddfdf542 713 * offset. A portion of the namekey is an iterator/randomizer to deal
8750964d 714 * with collisions.
66325755 715 *
ddfdf542
MD
716 * NOTE: base.base.obj_type from the related B-Tree leaf entry holds
717 * the filesystem object type of obj_id, e.g. a den_type equivalent.
718 * It is not stored in hammer_entry_data.
6b4f890b
MD
719 *
720 * NOTE: den_name / the filename data reference is NOT terminated with \0.
8750964d 721 */
11ad5ade 722struct hammer_entry_data {
adf01747 723 int64_t obj_id; /* object being referenced */
ddfdf542
MD
724 u_int32_t localization; /* identify pseudo-filesystem */
725 u_int32_t reserved02;
11ad5ade 726 char name[16]; /* name (extended) */
8750964d
MD
727};
728
11ad5ade
MD
729#define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_entry_data, name[0])
730#define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_entry_data, name[nlen])
19619882 731
ddfdf542
MD
732/*
733 * Symlink data which does not fit in the inode is stored in a separte
734 * FIX type record.
735 */
11ad5ade
MD
736struct hammer_symlink_data {
737 char name[16];
8750964d
MD
738};
739
11ad5ade 740#define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0])
7a04d74f 741
8750964d 742/*
5fa5c92f
MD
743 * The root inode for the primary filesystem and root inode for any
744 * pseudo-fs may be tagged with an optional data structure using
745 * HAMMER_RECTYPE_FIX/HAMMER_FIXKEY_PSEUDOFS. This structure allows
746 * the node to be used as a mirroring master or slave.
747 *
748 * When operating as a slave CD's into the node automatically become read-only
a56cb012 749 * and as-of sync_end_tid.
5fa5c92f 750 *
a56cb012 751 * When operating as a master the read PFSD info sets sync_end_tid to
5fa5c92f
MD
752 * the most recently flushed TID.
753 *
754 * sync_low_tid is not yet used but will represent the highest pruning
755 * end-point, after which full history is available.
756 */
757struct hammer_pseudofs_data {
758 hammer_tid_t sync_low_tid; /* full history beyond this point */
a56cb012
MD
759 hammer_tid_t sync_beg_tid; /* earliest tid w/ full history avail */
760 hammer_tid_t sync_end_tid; /* current synchronizatoin point */
5fa5c92f
MD
761 u_int64_t sync_beg_ts; /* real-time of last completed sync */
762 u_int64_t sync_end_ts; /* initiation of current sync cycle */
763 uuid_t shared_uuid; /* shared uuid (match required) */
764 uuid_t unique_uuid; /* unique uuid of this master/slave */
732a1697 765 int32_t reserved01; /* reserved for future master_id */
4c038e17 766 int32_t mirror_flags; /* misc flags */
5fa5c92f 767 char label[64]; /* filesystem space label */
260b4b9e 768 char snapshots[64]; /* softlink dir for pruning */
4c038e17
MD
769 int16_t prune_time; /* how long to spend pruning */
770 int16_t prune_freq; /* how often we prune */
771 int16_t reblock_time; /* how long to spend reblocking */
772 int16_t reblock_freq; /* how often we reblock */
773 int32_t snapshot_freq; /* how often we create a snapshot */
774 int32_t prune_min; /* do not prune recent history */
775 int32_t prune_max; /* do not retain history beyond here */
776 int32_t reserved[16];
5fa5c92f
MD
777};
778
779typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t;
780
781#define HAMMER_PFSD_SLAVE 0x00000001
842e7a70 782#define HAMMER_PFSD_DELETED 0x80000000
5fa5c92f
MD
783
784/*
83f2a3aa
MD
785 * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }.
786 *
787 * Snapshot records replace the old <fs>/snapshots/<softlink> methodology. Snapshot
788 * records are mirrored but may be independantly managed once they are laid down on
789 * a slave.
790 *
791 * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the
792 * results.
793 *
794 * NOTE: Reserved fields must be zero (as usual)
795 */
796struct hammer_snapshot_data {
797 hammer_tid_t tid; /* the snapshot TID itself (== key) */
798 u_int64_t ts; /* real-time when snapshot was made */
799 u_int64_t reserved01;
800 u_int64_t reserved02;
801 char label[64]; /* user-supplied description */
802 u_int64_t reserved03[4];
803};
804
805/*
806 * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }.
807 *
808 * Used to store the hammer cleanup config. This data is not mirrored.
809 */
810struct hammer_config_data {
811 char text[1024];
812};
813
814/*
c60bb2c5
MD
815 * Rollup various structures embedded as record data
816 */
427e5fc6 817union hammer_data_ondisk {
11ad5ade 818 struct hammer_entry_data entry;
c60bb2c5 819 struct hammer_inode_data inode;
11ad5ade 820 struct hammer_symlink_data symlink;
842e7a70 821 struct hammer_pseudofs_data pfsd;
83f2a3aa
MD
822 struct hammer_snapshot_data snap;
823 struct hammer_config_data config;
c60bb2c5
MD
824};
825
11ad5ade
MD
826typedef union hammer_data_ondisk *hammer_data_ondisk_t;
827
7dc57964 828#endif