Pre-2.0 release: Sync with HAMMER 64 - simplify PFS operations, fix pfs-upgrade
[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 *
ac14cb85 34 * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.50.2.1 2008/07/19 18:46:20 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
MD
94/*
95 * Hammer transction ids are 64 bit unsigned integers and are usually
96 * synchronized with the time of day in nanoseconds.
47197d71
MD
97 *
98 * Hammer offsets are used for FIFO indexing and embed a cycle counter
99 * and volume number in addition to the offset. Most offsets are required
100 * to be 64-byte aligned.
8750964d
MD
101 */
102typedef u_int64_t hammer_tid_t;
47197d71 103typedef u_int64_t hammer_off_t;
059819e3
MD
104typedef u_int32_t hammer_seq_t;
105typedef u_int32_t hammer_crc_t;
47197d71
MD
106
107#define HAMMER_MIN_TID 0ULL /* unsigned */
108#define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */
109#define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */
110#define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */
111#define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */
112#define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */
113#define HAMMER_MIN_RECTYPE 0x0U /* unsigned */
114#define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */
115#define HAMMER_MIN_OFFSET 0ULL /* unsigned */
116#define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */
117
118/*
119 * hammer_off_t has several different encodings. Note that not all zones
120 * encode a vol_no.
121 *
cb51be26 122 * zone 0: reserved for sanity
47197d71
MD
123 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header)
124 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer)
cb51be26
MD
125 * zone 3 (z,o): undo fifo - actually fixed phys array in vol hdr
126 * zone 4 (z,v,o): freemap - only real blockmap
127 * zone 8 (z,v,o): B-Tree - actually zone-2 address
128 * zone 9 (z,v,o): Record - actually zone-2 address
129 * zone 10 (z,v,o): Large-data - actually zone-2 address
130 * zone 15: reserved for sanity
47197d71
MD
131 */
132
133#define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL
134#define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL
40043e7f 135#define HAMMER_ZONE_UNDO 0x3000000000000000ULL
c3be93f2 136#define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL
40043e7f
MD
137#define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL
138#define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL
139#define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL
140#define HAMMER_ZONE_BTREE 0x8000000000000000ULL
bf3b416b 141#define HAMMER_ZONE_META 0x9000000000000000ULL
40043e7f
MD
142#define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL
143#define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL
144#define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL
145#define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL
146#define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL
cb51be26 147#define HAMMER_ZONE_UNAVAIL 0xF000000000000000ULL
40043e7f
MD
148
149#define HAMMER_ZONE_RAW_VOLUME_INDEX 1
150#define HAMMER_ZONE_RAW_BUFFER_INDEX 2
151#define HAMMER_ZONE_UNDO_INDEX 3
c3be93f2 152#define HAMMER_ZONE_FREEMAP_INDEX 4
40043e7f 153#define HAMMER_ZONE_BTREE_INDEX 8
bf3b416b 154#define HAMMER_ZONE_META_INDEX 9
40043e7f
MD
155#define HAMMER_ZONE_LARGE_DATA_INDEX 10
156#define HAMMER_ZONE_SMALL_DATA_INDEX 11
cb51be26 157#define HAMMER_ZONE_UNAVAIL_INDEX 15 /* unavailable */
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
5de0c0e5
MD
213/*
214 * Maximum number of mirrors operating in master mode (multi-master
215 * clustering and mirroring).
216 */
217#define HAMMER_MAX_MASTERS 16
218
c3be93f2 219/*
cb51be26
MD
220 * The blockmap is somewhat of a degenerate structure. HAMMER only actually
221 * uses it in its original incarnation to implement the free-map.
bf686dbe 222 *
cb51be26
MD
223 * zone:1 raw volume (no blockmap)
224 * zone:2 raw buffer (no blockmap)
225 * zone:3 undo-map (direct layer2 array in volume header)
226 * zone:4 free-map (the only real blockmap)
227 * zone:8-15 zone id used to classify big-block only, address is actually
228 * a zone-2 address.
c3be93f2
MD
229 */
230struct hammer_blockmap {
231 hammer_off_t phys_offset; /* zone-2 physical offset */
bf686dbe 232 hammer_off_t first_offset; /* zone-X logical offset (zone 3) */
c3be93f2
MD
233 hammer_off_t next_offset; /* zone-X logical offset */
234 hammer_off_t alloc_offset; /* zone-X logical offset */
c3be93f2 235 u_int32_t reserved01;
19619882 236 hammer_crc_t entry_crc;
c3be93f2 237};
40043e7f 238
c3be93f2 239typedef struct hammer_blockmap *hammer_blockmap_t;
40043e7f 240
19619882
MD
241#define HAMMER_BLOCKMAP_CRCSIZE \
242 offsetof(struct hammer_blockmap, entry_crc)
243
c3be93f2
MD
244/*
245 * The blockmap is a 2-layer entity made up of big-blocks. The first layer
246 * contains 262144 32-byte entries (18 bits), the second layer contains
247 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
248 * 18+19+23 = 60 bits. The top four bits are the zone id.
249 *
cb51be26
MD
250 * Currently only the freemap utilizes both layers in all their glory.
251 * All primary data/meta-data zones actually encode a zone-2 address
252 * requiring no real blockmap translation.
c3be93f2 253 *
cb51be26
MD
254 * The freemap uses the upper 8 bits of layer-1 to identify the volume,
255 * thus any space allocated via the freemap can be directly translated
256 * to a zone:2 (or zone:8-15) address.
c3be93f2 257 *
cb51be26 258 * zone-X blockmap offset: [z:4][layer1:18][layer2:19][bigblock:23]
c3be93f2
MD
259 */
260struct hammer_blockmap_layer1 {
261 hammer_off_t blocks_free; /* big-blocks free */
262 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */
c3be93f2 263 hammer_off_t reserved01;
19619882
MD
264 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */
265 /* (not yet used) */
266 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/
c3be93f2 267};
40043e7f 268
47637bff
MD
269typedef struct hammer_blockmap_layer1 *hammer_blockmap_layer1_t;
270
19619882
MD
271#define HAMMER_LAYER1_CRCSIZE \
272 offsetof(struct hammer_blockmap_layer1, layer1_crc)
273
c3be93f2 274struct hammer_blockmap_layer2 {
cb51be26
MD
275 u_int8_t zone; /* typed allocation zone */
276 u_int8_t unused01;
277 u_int16_t unused02;
278 u_int32_t append_off; /* allocatable space index */
19619882
MD
279 u_int32_t bytes_free; /* bytes free within this bigblock */
280 hammer_crc_t entry_crc;
40043e7f
MD
281};
282
47637bff
MD
283typedef struct hammer_blockmap_layer2 *hammer_blockmap_layer2_t;
284
19619882
MD
285#define HAMMER_LAYER2_CRCSIZE \
286 offsetof(struct hammer_blockmap_layer2, entry_crc)
287
c3be93f2
MD
288#define HAMMER_BLOCKMAP_FREE 0ULL
289#define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL)
290
291#define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \
292 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1))
293#define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \
294 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2))
295
296#define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \
297 (HAMMER_BLOCKMAP_RADIX1 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
298#define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \
299 (HAMMER_BLOCKMAP_RADIX2 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
300
301#define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 */ \
302 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
6f97fce3 303#define HAMMER_BLOCKMAP_LAYER2 /* 19+23 - 4TB */ \
c3be93f2
MD
304 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_LARGEBLOCK_SIZE64)
305
306#define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1)
307#define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1)
308
309/*
310 * byte offset within layer1 or layer2 big-block for the entry representing
311 * a zone-2 physical offset.
312 */
313#define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \
314 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \
315 HAMMER_BLOCKMAP_LAYER2 * sizeof(struct hammer_blockmap_layer1))
316
317#define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \
318 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \
319 HAMMER_LARGEBLOCK_SIZE64 * sizeof(struct hammer_blockmap_layer2))
66325755 320
bf686dbe
MD
321/*
322 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume
cb51be26 323 * header with an array of layer2 structures. A maximum of (128x8MB) = 1GB
bf686dbe
MD
324 * may be reserved. The size of the undo fifo is usually set a newfs time
325 * but can be adjusted if the filesystem is taken offline.
326 */
327
cb51be26 328#define HAMMER_UNDO_LAYER2 128 /* max layer2 undo mapping entries */
bf686dbe 329
9380c752 330/*
36f82b23
MD
331 * All on-disk HAMMER structures which make up elements of the UNDO FIFO
332 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure
40043e7f
MD
333 * contains all the information required to validate the fifo element
334 * and to scan the fifo in either direction. The head is typically embedded
335 * in higher level hammer on-disk structures while the tail is typically
336 * out-of-band. hdr_size is the size of the whole mess, including the tail.
47197d71 337 *
36f82b23
MD
338 * All undo structures are guaranteed to not cross a 16K filesystem
339 * buffer boundary. Most undo structures are fairly small. Data spaces
340 * are not immediately reused by HAMMER so file data is not usually recorded
341 * as part of an UNDO.
40043e7f
MD
342 *
343 * PAD elements are allowed to take up only 8 bytes of space as a special
344 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
345 * and with the tail overloaded onto the head structure for 8 bytes total.
36f82b23
MD
346 *
347 * Every undo record has a sequence number. This number is unrelated to
348 * transaction ids and instead collects the undo transactions associated
349 * with a single atomic operation. A larger transactional operation, such
350 * as a remove(), may consist of several smaller atomic operations
351 * representing raw meta-data operations.
9380c752 352 */
36f82b23 353#define HAMMER_HEAD_ONDISK_SIZE 32
40043e7f 354#define HAMMER_HEAD_ALIGN 8
47197d71 355#define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
40043e7f 356#define HAMMER_TAIL_ONDISK_SIZE 8
4c038e17
MD
357#define HAMMER_HEAD_DOALIGN(bytes) \
358 (((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK)
47197d71
MD
359
360struct hammer_fifo_head {
361 u_int16_t hdr_signature;
362 u_int16_t hdr_type;
40043e7f 363 u_int32_t hdr_size; /* aligned size of the whole mess */
059819e3 364 u_int32_t reserved01; /* (0) reserved for future use */
19619882 365 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */
40043e7f
MD
366};
367
19619882
MD
368#define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc)
369
40043e7f
MD
370struct hammer_fifo_tail {
371 u_int16_t tail_signature;
372 u_int16_t tail_type;
373 u_int32_t tail_size; /* aligned size of the whole mess */
47197d71
MD
374};
375
376typedef struct hammer_fifo_head *hammer_fifo_head_t;
40043e7f 377typedef struct hammer_fifo_tail *hammer_fifo_tail_t;
9380c752 378
8750964d 379/*
47197d71 380 * Fifo header types.
8750964d 381 */
40043e7f
MD
382#define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE)
383#define HAMMER_HEAD_TYPE_VOL 0x0041U /* Volume (dummy header) */
384#define HAMMER_HEAD_TYPE_BTREE 0x0042U /* B-Tree node */
385#define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
386#define HAMMER_HEAD_TYPE_DELETE 0x0044U /* record deletion */
387#define HAMMER_HEAD_TYPE_RECORD 0x0045U /* Filesystem record */
47197d71 388
40043e7f 389#define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */
8750964d 390
47197d71 391#define HAMMER_HEAD_SIGNATURE 0xC84EU
40043e7f 392#define HAMMER_TAIL_SIGNATURE 0xC74FU
47197d71 393
36f82b23
MD
394#define HAMMER_HEAD_SEQ_BEG 0x80000000U
395#define HAMMER_HEAD_SEQ_END 0x40000000U
396#define HAMMER_HEAD_SEQ_MASK 0x3FFFFFFFU
397
47197d71 398/*
bf686dbe 399 * Misc FIFO structures.
47197d71
MD
400 */
401struct hammer_fifo_undo {
402 struct hammer_fifo_head head;
bf686dbe
MD
403 hammer_off_t undo_offset; /* zone-1 offset */
404 int32_t undo_data_bytes;
405 int32_t undo_reserved01;
47197d71 406 /* followed by data */
8750964d
MD
407};
408
47197d71 409typedef struct hammer_fifo_undo *hammer_fifo_undo_t;
8750964d 410
bf686dbe
MD
411struct hammer_fifo_buf_commit {
412 hammer_off_t undo_offset;
413};
414
c60bb2c5 415/*
47197d71 416 * Volume header types
c60bb2c5 417 */
8750964d 418#define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
8750964d
MD
419#define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
420
421/*
422 * The B-Tree structures need hammer_fsbuf_head.
423 */
424#include "hammer_btree.h"
425
426/*
427 * HAMMER Volume header
428 *
429 * A HAMMER filesystem is built from any number of block devices, Each block
47197d71
MD
430 * device contains a volume header followed by however many buffers fit
431 * into the volume.
8750964d 432 *
47197d71
MD
433 * One of the volumes making up a HAMMER filesystem is the master, the
434 * rest are slaves. It does not have to be volume #0.
c60bb2c5 435 *
47197d71
MD
436 * The volume header takes up an entire 16K filesystem buffer and may
437 * represent up to 64KTB (65536 TB) of space.
a89aec1b
MD
438 *
439 * Special field notes:
440 *
441 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
442 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
47197d71 443 * vol_buf_beg - offset of the first buffer.
a89aec1b
MD
444 *
445 * The memory log area allows a kernel to cache new records and data
446 * in memory without allocating space in the actual filesystem to hold
447 * the records and data. In the event that a filesystem becomes full,
448 * any records remaining in memory can be flushed to the memory log
449 * area. This allows the kernel to immediately return success.
8750964d 450 */
8750964d 451
a89aec1b
MD
452#define HAMMER_BOOT_MINBYTES (32*1024)
453#define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
454#define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
455
456#define HAMMER_MEM_MINBYTES (256*1024)
457#define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
458#define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
459
8750964d 460struct hammer_volume_ondisk {
40043e7f
MD
461 u_int64_t vol_signature;/* Signature */
462
a89aec1b
MD
463 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
464 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
47197d71
MD
465 int64_t vol_buf_beg; /* byte offset of first buffer in volume */
466 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */
8750964d
MD
467 int64_t vol_locked; /* reserved clusters are >= this offset */
468
469 uuid_t vol_fsid; /* identify filesystem */
470 uuid_t vol_fstype; /* identify filesystem type */
471 char vol_name[64]; /* Name of volume */
472
473 int32_t vol_no; /* volume number within filesystem */
474 int32_t vol_count; /* number of volumes making up FS */
475
476 u_int32_t vol_version; /* version control information */
19619882 477 hammer_crc_t vol_crc; /* header crc */
8750964d
MD
478 u_int32_t vol_flags; /* volume flags */
479 u_int32_t vol_rootvol; /* which volume is the root volume? */
480
40043e7f 481 int32_t vol_reserved04;
47197d71 482 int32_t vol_reserved05;
8750964d
MD
483 u_int32_t vol_reserved06;
484 u_int32_t vol_reserved07;
485
fbc6e32a 486 int32_t vol_blocksize; /* for statfs only */
47197d71 487 int32_t vol_reserved08;
fbc6e32a
MD
488 int64_t vol_nblocks; /* total allocatable hammer bufs */
489
8750964d
MD
490 /*
491 * These fields are initialized and space is reserved in every
492 * volume making up a HAMMER filesytem, but only the master volume
493 * contains valid data.
494 */
c3be93f2
MD
495 int64_t vol0_stat_bigblocks; /* total bigblocks when fs is empty */
496 int64_t vol0_stat_freebigblocks;/* number of free bigblocks */
fbc6e32a
MD
497 int64_t vol0_stat_bytes; /* for statfs only */
498 int64_t vol0_stat_inodes; /* for statfs only */
b33e2cc0 499 int64_t vol0_stat_records; /* total records in filesystem */
47197d71
MD
500 hammer_off_t vol0_btree_root; /* B-Tree root */
501 hammer_tid_t vol0_next_tid; /* highest synchronized TID */
5fa5c92f 502 hammer_off_t vol0_unused03;
40043e7f
MD
503
504 /*
0729c8c8
MD
505 * Blockmaps for zones. Not all zones use a blockmap. Note that
506 * the entire root blockmap is cached in the hammer_mount structure.
40043e7f 507 */
c3be93f2 508 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES];
40043e7f 509
bf686dbe 510 /*
cb51be26 511 * Array of zone-2 addresses for undo FIFO.
bf686dbe 512 */
cb51be26 513 hammer_off_t vol0_undo_array[HAMMER_UNDO_LAYER2];
bf686dbe 514
8750964d
MD
515};
516
8cd0a023
MD
517typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
518
c60bb2c5
MD
519#define HAMMER_VOLF_VALID 0x0001 /* valid entry */
520#define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
8cd0a023 521
19619882
MD
522#define HAMMER_VOL_CRCSIZE1 \
523 offsetof(struct hammer_volume_ondisk, vol_crc)
524#define HAMMER_VOL_CRCSIZE2 \
525 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \
526 sizeof(hammer_crc_t))
527
c60bb2c5
MD
528/*
529 * Record types are fairly straightforward. The B-Tree includes the record
530 * type in its index sort.
c60bb2c5 531 */
8750964d 532#define HAMMER_RECTYPE_UNKNOWN 0
66325755 533#define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
8750964d 534#define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
ddfdf542
MD
535#define HAMMER_RECTYPE_UNUSED02 2
536#define HAMMER_RECTYPE_UNUSED03 3
bf686dbe
MD
537#define HAMMER_RECTYPE_DATA 0x0010
538#define HAMMER_RECTYPE_DIRENTRY 0x0011
539#define HAMMER_RECTYPE_DB 0x0012
540#define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */
541#define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */
ea434b6f 542#define HAMMER_RECTYPE_PFS 0x0015 /* PFS management */
bf686dbe 543#define HAMMER_RECTYPE_MOVED 0x8000 /* special recovery flag */
a9d52b76
MD
544#define HAMMER_RECTYPE_MAX 0xFFFF
545
546#define HAMMER_RECTYPE_CLEAN_START HAMMER_RECTYPE_EXT
7a04d74f
MD
547
548#define HAMMER_FIXKEY_SYMLINK 1
8750964d 549
66325755 550#define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */
8750964d
MD
551#define HAMMER_OBJTYPE_DIRECTORY 1
552#define HAMMER_OBJTYPE_REGFILE 2
553#define HAMMER_OBJTYPE_DBFILE 3
554#define HAMMER_OBJTYPE_FIFO 4
c60bb2c5
MD
555#define HAMMER_OBJTYPE_CDEV 5
556#define HAMMER_OBJTYPE_BDEV 6
557#define HAMMER_OBJTYPE_SOFTLINK 7
558#define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
b3bad96f 559#define HAMMER_OBJTYPE_SOCKET 9
c60bb2c5 560
8750964d 561/*
11ad5ade 562 * HAMMER inode attribute data
8750964d 563 *
11ad5ade
MD
564 * The data reference for a HAMMER inode points to this structure. Any
565 * modifications to the contents of this structure will result in a
566 * replacement operation.
8750964d 567 *
5a930e66
MD
568 * parent_obj_id is only valid for directories (which cannot be hard-linked),
569 * and specifies the parent directory obj_id. This field will also be set
570 * for non-directory inodes as a recovery aid, but can wind up holding
571 * stale information. However, since object id's are not reused, the worse
572 * that happens is that the recovery code is unable to use it.
ddfdf542
MD
573 *
574 * NOTE: Future note on directory hardlinks. We can implement a record type
575 * which allows us to point to multiple parent directories.
8750964d 576 *
11ad5ade
MD
577 * NOTE: atime is stored in the inode's B-Tree element and not in the inode
578 * data. This allows the atime to be updated without having to lay down a
579 * new record.
8750964d 580 */
11ad5ade
MD
581struct hammer_inode_data {
582 u_int16_t version; /* inode data version */
583 u_int16_t mode; /* basic unix permissions */
584 u_int32_t uflags; /* chflags */
585 u_int32_t rmajor; /* used by device nodes */
586 u_int32_t rminor; /* used by device nodes */
587 u_int64_t ctime;
adf01747 588 int64_t parent_obj_id; /* parent directory obj_id */
11ad5ade
MD
589 uuid_t uid;
590 uuid_t gid;
591
592 u_int8_t obj_type;
5a930e66
MD
593 u_int8_t cap_flags; /* capability support flags (extension) */
594 u_int16_t reserved02;
ddfdf542 595 u_int32_t reserved03; /* RESERVED FOR POSSIBLE FUTURE BIRTHTIME */
11ad5ade 596 u_int64_t nlinks; /* hard links */
bcac4bbb 597 u_int64_t size; /* filesystem object size */
2f85fa4d 598 union {
5a930e66
MD
599 struct {
600 char reserved06[16];
601 u_int32_t parent_obj_localization;
602 u_int32_t integrity_crc;
603 } obj;
2f85fa4d
MD
604 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */
605 } ext;
ddfdf542
MD
606 u_int64_t mtime; /* mtime must be second-to-last */
607 u_int64_t atime; /* atime must be last */
8750964d
MD
608};
609
ddfdf542
MD
610/*
611 * Neither mtime nor atime upates are CRCd by the B-Tree element.
612 * mtime updates have UNDO, atime updates do not.
613 */
bcac4bbb
MD
614#define HAMMER_ITIMES_BASE(ino_data) (&(ino_data)->mtime)
615#define HAMMER_ITIMES_BYTES (sizeof(u_int64_t) * 2)
616
ddfdf542
MD
617#define HAMMER_INODE_CRCSIZE \
618 offsetof(struct hammer_inode_data, mtime)
619
11ad5ade
MD
620#define HAMMER_INODE_DATA_VERSION 1
621#define HAMMER_OBJID_ROOT 1
ddfdf542 622#define HAMMER_INODE_BASESYMLEN 24 /* see ext.symlink */
8750964d
MD
623
624/*
ddfdf542
MD
625 * A HAMMER directory entry associates a HAMMER filesystem object with a
626 * namespace. It is possible to hook into a pseudo-filesystem (with its
627 * own inode numbering space) in the filesystem by setting the high
628 * 16 bits of the localization field. The low 16 bits must be 0 and
629 * are reserved for future use.
8750964d
MD
630 *
631 * Directory entries are indexed with a 128 bit namekey rather then an
ddfdf542 632 * offset. A portion of the namekey is an iterator/randomizer to deal
8750964d 633 * with collisions.
66325755 634 *
ddfdf542
MD
635 * NOTE: base.base.obj_type from the related B-Tree leaf entry holds
636 * the filesystem object type of obj_id, e.g. a den_type equivalent.
637 * It is not stored in hammer_entry_data.
6b4f890b
MD
638 *
639 * NOTE: den_name / the filename data reference is NOT terminated with \0.
8750964d 640 */
11ad5ade 641struct hammer_entry_data {
adf01747 642 int64_t obj_id; /* object being referenced */
ddfdf542
MD
643 u_int32_t localization; /* identify pseudo-filesystem */
644 u_int32_t reserved02;
11ad5ade 645 char name[16]; /* name (extended) */
8750964d
MD
646};
647
11ad5ade
MD
648#define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_entry_data, name[0])
649#define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_entry_data, name[nlen])
19619882 650
ddfdf542
MD
651/*
652 * Symlink data which does not fit in the inode is stored in a separte
653 * FIX type record.
654 */
11ad5ade
MD
655struct hammer_symlink_data {
656 char name[16];
8750964d
MD
657};
658
11ad5ade 659#define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0])
7a04d74f 660
5fa5c92f
MD
661/*
662 * The root inode for the primary filesystem and root inode for any
663 * pseudo-fs may be tagged with an optional data structure using
664 * HAMMER_RECTYPE_FIX/HAMMER_FIXKEY_PSEUDOFS. This structure allows
665 * the node to be used as a mirroring master or slave.
666 *
667 * When operating as a slave CD's into the node automatically become read-only
a56cb012 668 * and as-of sync_end_tid.
5fa5c92f 669 *
a56cb012 670 * When operating as a master the read PFSD info sets sync_end_tid to
5fa5c92f
MD
671 * the most recently flushed TID.
672 *
673 * sync_low_tid is not yet used but will represent the highest pruning
674 * end-point, after which full history is available.
675 */
676struct hammer_pseudofs_data {
677 hammer_tid_t sync_low_tid; /* full history beyond this point */
a56cb012
MD
678 hammer_tid_t sync_beg_tid; /* earliest tid w/ full history avail */
679 hammer_tid_t sync_end_tid; /* current synchronizatoin point */
5fa5c92f
MD
680 u_int64_t sync_beg_ts; /* real-time of last completed sync */
681 u_int64_t sync_end_ts; /* initiation of current sync cycle */
682 uuid_t shared_uuid; /* shared uuid (match required) */
683 uuid_t unique_uuid; /* unique uuid of this master/slave */
ac14cb85 684 int32_t reserved01; /* reserved for future master_id */
4c038e17 685 int32_t mirror_flags; /* misc flags */
5fa5c92f 686 char label[64]; /* filesystem space label */
4c038e17
MD
687 char prune_path[64]; /* softlink dir for pruning */
688 int16_t prune_time; /* how long to spend pruning */
689 int16_t prune_freq; /* how often we prune */
690 int16_t reblock_time; /* how long to spend reblocking */
691 int16_t reblock_freq; /* how often we reblock */
692 int32_t snapshot_freq; /* how often we create a snapshot */
693 int32_t prune_min; /* do not prune recent history */
694 int32_t prune_max; /* do not retain history beyond here */
695 int32_t reserved[16];
5fa5c92f
MD
696};
697
698typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t;
699
700#define HAMMER_PFSD_SLAVE 0x00000001
842e7a70 701#define HAMMER_PFSD_DELETED 0x80000000
5fa5c92f 702
c60bb2c5
MD
703/*
704 * Rollup various structures embedded as record data
705 */
427e5fc6 706union hammer_data_ondisk {
11ad5ade 707 struct hammer_entry_data entry;
c60bb2c5 708 struct hammer_inode_data inode;
11ad5ade 709 struct hammer_symlink_data symlink;
842e7a70 710 struct hammer_pseudofs_data pfsd;
c60bb2c5
MD
711};
712
11ad5ade
MD
713typedef union hammer_data_ondisk *hammer_data_ondisk_t;
714
7dc57964 715#endif