Kind of sync makewhatis.local with FreeBSD. Ignore r/o directories.
[dragonfly.git] / sys / vfs / hammer / hammer_disk.h
CommitLineData
8750964d
MD
1/*
2 * Copyright (c) 2007 The DragonFly Project. All rights reserved.
745703c7 3 *
8750964d
MD
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
745703c7 6 *
8750964d
MD
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
745703c7 10 *
8750964d
MD
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.
745703c7 20 *
8750964d
MD
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.
745703c7 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
c655a349
TK
40#include <sys/endian.h>
41
8750964d
MD
42#ifndef _SYS_UUID_H_
43#include <sys/uuid.h>
44#endif
45
46/*
47 * The structures below represent the on-disk format for a HAMMER
48 * filesystem. Note that all fields for on-disk structures are naturally
c655a349
TK
49 * aligned. HAMMER uses little endian for fields in on-disk structures.
50 * HAMMER doesn't support big endian arch, but is planned.
8750964d
MD
51 *
52 * Most of HAMMER revolves around the concept of an object identifier. An
53 * obj_id is a 64 bit quantity which uniquely identifies a filesystem object
54 * FOR THE ENTIRE LIFE OF THE FILESYSTEM. This uniqueness allows backups
55 * and mirrors to retain varying amounts of filesystem history by removing
56 * any possibility of conflict through identifier reuse.
57 *
00e31441 58 * A HAMMER filesystem may span multiple volumes.
8750964d
MD
59 *
60 * A HAMMER filesystem uses a 16K filesystem buffer size. All filesystem
f176517c 61 * I/O is done in multiples of 16K.
40043e7f 62 *
4a2796f3
MD
63 * 64K X-bufs are used for blocks >= a file's 1MB mark.
64 *
40043e7f 65 * Per-volume storage limit: 52 bits 4096 TB
23234e40 66 * Per-Zone storage limit: 60 bits 1 MTB
40043e7f 67 * Per-filesystem storage limit: 60 bits 1 MTB
8750964d 68 */
40043e7f 69#define HAMMER_BUFSIZE 16384
4a2796f3 70#define HAMMER_XBUFSIZE 65536
72f2cb83 71#define HAMMER_HBUFSIZE (HAMMER_BUFSIZE / 2)
4a2796f3 72#define HAMMER_XDEMARC (1024 * 1024)
40043e7f 73#define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1)
4a2796f3 74#define HAMMER_XBUFMASK (HAMMER_XBUFSIZE - 1)
8750964d 75
46137e17
TK
76#define HAMMER_BUFSIZE64 ((uint64_t)HAMMER_BUFSIZE)
77#define HAMMER_BUFMASK64 ((uint64_t)HAMMER_BUFMASK)
47197d71 78
46137e17
TK
79#define HAMMER_XBUFSIZE64 ((uint64_t)HAMMER_XBUFSIZE)
80#define HAMMER_XBUFMASK64 ((uint64_t)HAMMER_XBUFMASK)
4a2796f3 81
47197d71
MD
82#define HAMMER_OFF_ZONE_MASK 0xF000000000000000ULL /* zone portion */
83#define HAMMER_OFF_VOL_MASK 0x0FF0000000000000ULL /* volume portion */
84#define HAMMER_OFF_SHORT_MASK 0x000FFFFFFFFFFFFFULL /* offset portion */
85#define HAMMER_OFF_LONG_MASK 0x0FFFFFFFFFFFFFFFULL /* offset portion */
47197d71 86
6b3804c9
MD
87#define HAMMER_OFF_BAD ((hammer_off_t)-1)
88
bbc5e78b
TK
89#define HAMMER_BUFSIZE_DOALIGN(offset) \
90 (((offset) + HAMMER_BUFMASK) & ~HAMMER_BUFMASK)
91#define HAMMER_BUFSIZE64_DOALIGN(offset) \
92 (((offset) + HAMMER_BUFMASK64) & ~HAMMER_BUFMASK64)
93
94#define HAMMER_XBUFSIZE_DOALIGN(offset) \
95 (((offset) + HAMMER_XBUFMASK) & ~HAMMER_XBUFMASK)
96#define HAMMER_XBUFSIZE64_DOALIGN(offset) \
97 (((offset) + HAMMER_XBUFMASK64) & ~HAMMER_XBUFMASK64)
98
94509284
MD
99/*
100 * The current limit of volumes that can make up a HAMMER FS
101 */
102#define HAMMER_MAX_VOLUMES 256
103
dee5b238
TK
104/*
105 * Reserved space for (future) header junk after the volume header.
106 */
d454cd69
TK
107#define HAMMER_MIN_VOL_JUNK (HAMMER_BUFSIZE * 16) /* 256 KB */
108#define HAMMER_MAX_VOL_JUNK HAMMER_MIN_VOL_JUNK
109#define HAMMER_VOL_JUNK_SIZE HAMMER_MIN_VOL_JUNK
dee5b238 110
8750964d 111/*
905f4108 112 * Hammer transaction ids are 64 bit unsigned integers and are usually
8750964d 113 * synchronized with the time of day in nanoseconds.
47197d71
MD
114 *
115 * Hammer offsets are used for FIFO indexing and embed a cycle counter
116 * and volume number in addition to the offset. Most offsets are required
23234e40 117 * to be 16 KB aligned.
8750964d 118 */
46137e17
TK
119typedef uint64_t hammer_tid_t;
120typedef uint64_t hammer_off_t;
121typedef uint32_t hammer_crc_t;
90da8fc8 122typedef uuid_t hammer_uuid_t;
47197d71
MD
123
124#define HAMMER_MIN_TID 0ULL /* unsigned */
125#define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */
126#define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */
127#define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */
128#define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */
129#define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */
130#define HAMMER_MIN_RECTYPE 0x0U /* unsigned */
131#define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */
132#define HAMMER_MIN_OFFSET 0ULL /* unsigned */
133#define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */
134
135/*
136 * hammer_off_t has several different encodings. Note that not all zones
a003db3d
TK
137 * encode a vol_no. Zone bits are not a part of filesystem capacity as
138 * the zone bits aren't directly or indirectly mapped to physical volumes.
139 *
140 * In other words, HAMMER's logical filesystem offset consists of 64 bits,
141 * but the filesystem is considered 60 bits filesystem, not 64 bits.
142 * The maximum filesystem capacity is 1EB, not 16EB.
47197d71 143 *
6fa871ed 144 * zone 0: available, a big-block that contains the offset is unused
47197d71
MD
145 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header)
146 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer)
50d83294 147 * zone 3 (z,o): undo/redo fifo - fixed zone-2 offset array in volume header
cb51be26
MD
148 * zone 4 (z,v,o): freemap - only real blockmap
149 * zone 8 (z,v,o): B-Tree - actually zone-2 address
f97726ae
TK
150 * zone 9 (z,v,o): meta - actually zone-2 address
151 * zone 10 (z,v,o): large-data - actually zone-2 address
152 * zone 11 (z,v,o): small-data - actually zone-2 address
6fa871ed 153 * zone 15: unavailable, usually the offset is beyond volume size
d40bfeca
MD
154 *
155 * layer1/layer2 direct map:
55e3f6f3 156 * Maximum HAMMER filesystem capacity from volume aspect
7f1e2b68 157 * 2^8(max volumes) * 2^52(max volume size) = 2^60 = 1EB (long offset)
55e3f6f3 158 * <------------------------------------------------------------->
7f1e2b68 159 * 8bits 52bits (short offset)
55e3f6f3 160 * <------><----------------------------------------------------->
d40bfeca
MD
161 * zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo
162 * ----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo
89660e86
TK
163 * <-----------------><------------------><---------------------->
164 * 18bits 19bits 23bits
165 * <------------------------------------------------------------->
89660e86 166 * 2^18(layer1) * 2^19(layer2) * 2^23(big-block) = 2^60 = 1EB
55e3f6f3 167 * Maximum HAMMER filesystem capacity from blockmap aspect
7f1e2b68
TK
168 *
169 * volume#0 layout
170 * +-------------------------> offset 0 of a device/partition
171 * | volume header (1928 bytes)
172 * | the rest of header junk space (HAMMER_BUFSIZE aligned)
173 * +-------------------------> vol_bot_beg
174 * | boot area (HAMMER_BUFSIZE aligned)
175 * +-------------------------> vol_mem_beg
176 * | memory log (HAMMER_BUFSIZE aligned)
177 * +-------------------------> vol_buf_beg (physical offset of zone-2)
178 * | zone-4 big-block for layer1
179 * +-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE
180 * | zone-4 big-blocks for layer2
d7d5ced7 181 * | ... (1 big-block per 4TB space)
7f1e2b68
TK
182 * +-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
183 * | zone-3 big-blocks for UNDO/REDO FIFO
184 * | ... (max 128 big-blocks)
185 * +-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
186 * | zone-8 big-block for root B-Tree node/etc
187 * +-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
a5ff7917 188 * | zone-9 big-block for root inode/PFS/etc
7f1e2b68
TK
189 * +-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
190 * | zone-X big-blocks
191 * | ... (big-blocks for new zones after newfs_hammer)
192 * | ...
193 * | ...
194 * | ...
195 * | ...
196 * +-------------------------> vol_buf_end (HAMMER_BUFSIZE aligned)
197 * +-------------------------> end of a device/partition
198 *
d7d5ced7 199 * volume#N layout (0<N<256)
7f1e2b68
TK
200 * +-------------------------> offset 0 of a device/partition
201 * | volume header (1928 bytes)
202 * | the rest of header junk space (HAMMER_BUFSIZE aligned)
203 * +-------------------------> vol_bot_beg
204 * | boot area (HAMMER_BUFSIZE aligned)
205 * +-------------------------> vol_mem_beg
206 * | memory log (HAMMER_BUFSIZE aligned)
207 * +-------------------------> vol_buf_beg (physical offset of zone-2)
208 * | zone-4 big-blocks for layer2
d7d5ced7 209 * | ... (1 big-block per 4TB space)
7f1e2b68
TK
210 * +-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
211 * | zone-X big-blocks
212 * | ... (unused until volume#(N-1) runs out of space)
213 * | ...
214 * | ...
215 * | ...
216 * | ...
217 * +-------------------------> vol_buf_end (HAMMER_BUFSIZE aligned)
218 * +-------------------------> end of a device/partition
47197d71
MD
219 */
220
221#define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL
222#define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL
40043e7f 223#define HAMMER_ZONE_UNDO 0x3000000000000000ULL
c3be93f2 224#define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL
6fa871ed
TK
225#define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL /* not used */
226#define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL /* not used */
227#define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL /* not used */
40043e7f 228#define HAMMER_ZONE_BTREE 0x8000000000000000ULL
bf3b416b 229#define HAMMER_ZONE_META 0x9000000000000000ULL
40043e7f
MD
230#define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL
231#define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL
6fa871ed
TK
232#define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL /* not used */
233#define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL /* not used */
234#define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL /* not used */
cb51be26 235#define HAMMER_ZONE_UNAVAIL 0xF000000000000000ULL
40043e7f
MD
236
237#define HAMMER_ZONE_RAW_VOLUME_INDEX 1
238#define HAMMER_ZONE_RAW_BUFFER_INDEX 2
239#define HAMMER_ZONE_UNDO_INDEX 3
c3be93f2 240#define HAMMER_ZONE_FREEMAP_INDEX 4
40043e7f 241#define HAMMER_ZONE_BTREE_INDEX 8
bf3b416b 242#define HAMMER_ZONE_META_INDEX 9
40043e7f
MD
243#define HAMMER_ZONE_LARGE_DATA_INDEX 10
244#define HAMMER_ZONE_SMALL_DATA_INDEX 11
6fa871ed 245#define HAMMER_ZONE_UNAVAIL_INDEX 15
f03c9cf4 246
40043e7f 247#define HAMMER_MAX_ZONES 16
47197d71 248
5bce7157
TK
249#define HAMMER_ZONE(offset) ((offset) & HAMMER_OFF_ZONE_MASK)
250
e1545c47
TK
251#define hammer_is_zone_raw_volume(offset) \
252 (HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_VOLUME)
253#define hammer_is_zone_raw_buffer(offset) \
254 (HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_BUFFER)
255#define hammer_is_zone_undo(offset) \
256 (HAMMER_ZONE(offset) == HAMMER_ZONE_UNDO)
257#define hammer_is_zone_freemap(offset) \
258 (HAMMER_ZONE(offset) == HAMMER_ZONE_FREEMAP)
259#define hammer_is_zone_btree(offset) \
260 (HAMMER_ZONE(offset) == HAMMER_ZONE_BTREE)
261#define hammer_is_zone_meta(offset) \
262 (HAMMER_ZONE(offset) == HAMMER_ZONE_META)
263#define hammer_is_zone_large_data(offset) \
264 (HAMMER_ZONE(offset) == HAMMER_ZONE_LARGE_DATA)
265#define hammer_is_zone_small_data(offset) \
266 (HAMMER_ZONE(offset) == HAMMER_ZONE_SMALL_DATA)
267#define hammer_is_zone_unavail(offset) \
268 (HAMMER_ZONE(offset) == HAMMER_ZONE_UNAVAIL)
66d704af
TK
269#define hammer_is_zone_data(offset) \
270 (hammer_is_zone_large_data(offset) || hammer_is_zone_small_data(offset))
e1545c47 271
f6d29b27 272#define hammer_is_index_record(zone) \
f43eff17
TK
273 ((zone) >= HAMMER_ZONE_BTREE_INDEX && \
274 (zone) < HAMMER_MAX_ZONES)
f6d29b27
TK
275
276#define hammer_is_zone_record(offset) \
277 hammer_is_index_record(HAMMER_ZONE_DECODE(offset))
278
279#define hammer_is_index_direct_xlated(zone) \
f43eff17
TK
280 (((zone) == HAMMER_ZONE_RAW_BUFFER_INDEX) || \
281 ((zone) == HAMMER_ZONE_FREEMAP_INDEX) || \
f6d29b27
TK
282 hammer_is_index_record(zone))
283
284#define hammer_is_zone_direct_xlated(offset) \
285 hammer_is_index_direct_xlated(HAMMER_ZONE_DECODE(offset))
ccdea233 286
0ac41b31
TK
287#define HAMMER_ZONE_ENCODE(zone, ham_off) \
288 (((hammer_off_t)(zone) << 60) | (ham_off))
289#define HAMMER_ZONE_DECODE(ham_off) \
afb71584 290 ((int)(((hammer_off_t)(ham_off) >> 60)))
0ac41b31 291
47197d71
MD
292#define HAMMER_VOL_ENCODE(vol_no) \
293 ((hammer_off_t)((vol_no) & 255) << 52)
294#define HAMMER_VOL_DECODE(ham_off) \
afb71584 295 ((int)(((hammer_off_t)(ham_off) >> 52) & 255))
0ac41b31 296
ac7efd27 297#define HAMMER_OFF_SHORT_ENCODE(offset) \
47197d71 298 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
ac7efd27 299#define HAMMER_OFF_LONG_ENCODE(offset) \
47197d71
MD
300 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
301
6fdd989a
TK
302#define HAMMER_ENCODE(zone, vol_no, offset) \
303 (((hammer_off_t)(zone) << 60) | \
cbf2551d 304 HAMMER_VOL_ENCODE(vol_no) | \
ac7efd27 305 HAMMER_OFF_SHORT_ENCODE(offset))
47197d71 306#define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \
6fdd989a 307 HAMMER_ENCODE(HAMMER_ZONE_RAW_VOLUME_INDEX, vol_no, offset)
47197d71 308#define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \
6fdd989a 309 HAMMER_ENCODE(HAMMER_ZONE_RAW_BUFFER_INDEX, vol_no, offset)
459a6a40
TK
310#define HAMMER_ENCODE_UNDO(offset) \
311 HAMMER_ENCODE(HAMMER_ZONE_UNDO_INDEX, HAMMER_ROOT_VOLNO, offset)
c3be93f2 312#define HAMMER_ENCODE_FREEMAP(vol_no, offset) \
6fdd989a 313 HAMMER_ENCODE(HAMMER_ZONE_FREEMAP_INDEX, vol_no, offset)
c3be93f2 314
32a31210 315/*
d0ea2047 316 * Translate a zone address to zone-X address.
32a31210 317 */
d0ea2047
TK
318#define hammer_xlate_to_zoneX(zone, offset) \
319 HAMMER_ZONE_ENCODE((zone), (offset) & ~HAMMER_OFF_ZONE_MASK)
320#define hammer_xlate_to_zone2(offset) \
321 hammer_xlate_to_zoneX(HAMMER_ZONE_RAW_BUFFER_INDEX, (offset))
32a31210 322
22c06dfe
TK
323#define hammer_data_zone(data_len) \
324 (((data_len) >= HAMMER_BUFSIZE) ? \
325 HAMMER_ZONE_LARGE_DATA : \
326 HAMMER_ZONE_SMALL_DATA)
327#define hammer_data_zone_index(data_len) \
328 (((data_len) >= HAMMER_BUFSIZE) ? \
329 HAMMER_ZONE_LARGE_DATA_INDEX : \
330 HAMMER_ZONE_SMALL_DATA_INDEX)
331
40043e7f 332/*
e04ee2de 333 * Big-Block backing store
40043e7f
MD
334 *
335 * A blockmap is a two-level map which translates a blockmap-backed zone
23234e40 336 * offset into a raw zone 2 offset. The layer 1 handles 18 bits and the
e04ee2de 337 * layer 2 handles 19 bits. The 8M big-block size is 23 bits so two
23234e40 338 * layers gives us 18+19+23 = 60 bits of address space.
df2ccbac
MD
339 *
340 * When using hinting for a blockmap lookup, the hint is lost when the
e04ee2de 341 * scan leaves the HINTBLOCK, which is typically several BIGBLOCK's.
df2ccbac 342 * HINTBLOCK is a heuristic.
40043e7f 343 */
e04ee2de 344#define HAMMER_HINTBLOCK_SIZE (HAMMER_BIGBLOCK_SIZE * 4)
46137e17 345#define HAMMER_HINTBLOCK_MASK64 ((uint64_t)HAMMER_HINTBLOCK_SIZE - 1)
e04ee2de 346#define HAMMER_BIGBLOCK_SIZE (8192 * 1024)
46137e17 347#define HAMMER_BIGBLOCK_SIZE64 ((uint64_t)HAMMER_BIGBLOCK_SIZE)
e04ee2de 348#define HAMMER_BIGBLOCK_MASK (HAMMER_BIGBLOCK_SIZE - 1)
46137e17 349#define HAMMER_BIGBLOCK_MASK64 ((uint64_t)HAMMER_BIGBLOCK_SIZE - 1)
e04ee2de 350#define HAMMER_BIGBLOCK_BITS 23
077bd8a4
TK
351#if 0
352#define HAMMER_BIGBLOCK_OVERFILL (6144 * 1024)
353#endif
e04ee2de
TK
354#if (1 << HAMMER_BIGBLOCK_BITS) != HAMMER_BIGBLOCK_SIZE
355#error "HAMMER_BIGBLOCK_BITS BROKEN"
40043e7f
MD
356#endif
357
e04ee2de
TK
358#define HAMMER_BUFFERS_PER_BIGBLOCK \
359 (HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE)
360#define HAMMER_BUFFERS_PER_BIGBLOCK_MASK \
361 (HAMMER_BUFFERS_PER_BIGBLOCK - 1)
362#define HAMMER_BUFFERS_PER_BIGBLOCK_MASK64 \
363 ((hammer_off_t)HAMMER_BUFFERS_PER_BIGBLOCK_MASK)
40043e7f 364
4eb78dd2
TK
365#define HAMMER_BIGBLOCK_DOALIGN(offset) \
366 (((offset) + HAMMER_BIGBLOCK_MASK64) & ~HAMMER_BIGBLOCK_MASK64)
367
5de0c0e5
MD
368/*
369 * Maximum number of mirrors operating in master mode (multi-master
c204ebbe
TK
370 * clustering and mirroring). Note that HAMMER1 does not support
371 * multi-master clustering as of 2015.
5de0c0e5
MD
372 */
373#define HAMMER_MAX_MASTERS 16
374
c3be93f2 375/*
cb51be26 376 * The blockmap is somewhat of a degenerate structure. HAMMER only actually
4437887b 377 * uses it in its original incarnation to implement the freemap.
bf686dbe 378 *
cb51be26
MD
379 * zone:1 raw volume (no blockmap)
380 * zone:2 raw buffer (no blockmap)
4437887b
TK
381 * zone:3 undomap (direct layer2 array in volume header)
382 * zone:4 freemap (the only real blockmap)
cb51be26
MD
383 * zone:8-15 zone id used to classify big-block only, address is actually
384 * a zone-2 address.
c3be93f2 385 */
22a0040d 386typedef struct hammer_blockmap {
f43e824c
TK
387 hammer_off_t phys_offset; /* zone-2 offset only used by zone-4 */
388 hammer_off_t first_offset; /* zone-X offset only used by zone-3 */
389 hammer_off_t next_offset; /* zone-X offset for allocation */
390 hammer_off_t alloc_offset; /* zone-X offset only used by zone-3 */
46137e17 391 uint32_t reserved01;
19619882 392 hammer_crc_t entry_crc;
22a0040d 393} *hammer_blockmap_t;
40043e7f 394
19619882
MD
395#define HAMMER_BLOCKMAP_CRCSIZE \
396 offsetof(struct hammer_blockmap, entry_crc)
397
c3be93f2
MD
398/*
399 * The blockmap is a 2-layer entity made up of big-blocks. The first layer
400 * contains 262144 32-byte entries (18 bits), the second layer contains
401 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
402 * 18+19+23 = 60 bits. The top four bits are the zone id.
403 *
cb51be26
MD
404 * Currently only the freemap utilizes both layers in all their glory.
405 * All primary data/meta-data zones actually encode a zone-2 address
406 * requiring no real blockmap translation.
c3be93f2 407 *
cb51be26
MD
408 * The freemap uses the upper 8 bits of layer-1 to identify the volume,
409 * thus any space allocated via the freemap can be directly translated
410 * to a zone:2 (or zone:8-15) address.
c3be93f2 411 *
a981af19 412 * zone-X blockmap offset: [zone:4][layer1:18][layer2:19][big-block:23]
c3be93f2 413 */
89660e86
TK
414
415/*
416 * 32 bytes layer1 entry for 8MB big-block.
417 * A big-block can hold 2^23 / 2^5 = 2^18 layer1 entries,
418 * which equals bits assigned for layer1 in zone-2 address.
419 */
22a0040d 420typedef struct hammer_blockmap_layer1 {
c3be93f2
MD
421 hammer_off_t blocks_free; /* big-blocks free */
422 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */
c3be93f2 423 hammer_off_t reserved01;
19619882
MD
424 hammer_crc_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */
425 /* (not yet used) */
426 hammer_crc_t layer1_crc; /* MUST BE LAST FIELD OF STRUCTURE*/
22a0040d 427} *hammer_blockmap_layer1_t;
47637bff 428
19619882
MD
429#define HAMMER_LAYER1_CRCSIZE \
430 offsetof(struct hammer_blockmap_layer1, layer1_crc)
431
320a5c59 432/*
89660e86
TK
433 * 16 bytes layer2 entry for 8MB big-blocks.
434 * A big-block can hold 2^23 / 2^4 = 2^19 layer2 entries,
435 * which equals bits assigned for layer2 in zone-2 address.
320a5c59
MD
436 *
437 * NOTE: bytes_free is signed and can legally go negative if/when data
438 * de-dup occurs. This field will never go higher than
e04ee2de 439 * HAMMER_BIGBLOCK_SIZE. If exactly HAMMER_BIGBLOCK_SIZE
320a5c59
MD
440 * the big-block is completely free.
441 */
22a0040d 442typedef struct hammer_blockmap_layer2 {
46137e17 443 uint8_t zone; /* typed allocation zone */
77f7bc0a
TK
444 uint8_t reserved01;
445 uint16_t reserved02;
46137e17 446 uint32_t append_off; /* allocatable space index */
a981af19 447 int32_t bytes_free; /* bytes free within this big-block */
19619882 448 hammer_crc_t entry_crc;
22a0040d 449} *hammer_blockmap_layer2_t;
47637bff 450
19619882
MD
451#define HAMMER_LAYER2_CRCSIZE \
452 offsetof(struct hammer_blockmap_layer2, entry_crc)
453
c3be93f2
MD
454#define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL)
455
89660e86 456#define HAMMER_BLOCKMAP_RADIX1 /* 2^18 = 262144 */ \
37fa1bc7 457 ((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1)))
89660e86 458#define HAMMER_BLOCKMAP_RADIX2 /* 2^19 = 524288 */ \
37fa1bc7 459 ((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2)))
c3be93f2 460
89660e86 461#define HAMMER_BLOCKMAP_LAYER1 /* 2^(18+19+23) = 1EB */ \
c3be93f2 462 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
89660e86 463#define HAMMER_BLOCKMAP_LAYER2 /* 2^(19+23) = 4TB */ \
e04ee2de 464 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_BIGBLOCK_SIZE64)
c3be93f2
MD
465
466#define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1)
467#define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1)
468
1ea93af3
TK
469#define HAMMER_BLOCKMAP_LAYER2_DOALIGN(offset) \
470 (((offset) + HAMMER_BLOCKMAP_LAYER2_MASK) & \
471 ~HAMMER_BLOCKMAP_LAYER2_MASK)
472
c3be93f2 473/*
f5b0d00f 474 * Index within layer1 or layer2 big-block for the entry representing
745703c7 475 * a zone-2 physical offset.
c3be93f2 476 */
f5b0d00f 477#define HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) \
e2124556
TK
478 ((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \
479 HAMMER_BLOCKMAP_LAYER2))
c3be93f2 480
f5b0d00f 481#define HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) \
e2124556
TK
482 ((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \
483 HAMMER_BIGBLOCK_SIZE64))
f5b0d00f
TK
484
485/*
486 * Byte offset within layer1 or layer2 big-block for the entry representing
487 * a zone-2 physical offset. Multiply the index by sizeof(blockmap_layer).
488 */
489#define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \
490 (HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) * \
491 sizeof(struct hammer_blockmap_layer1))
492
493#define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \
494 (HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) * \
495 sizeof(struct hammer_blockmap_layer2))
66325755 496
b7f51e3b
TK
497/*
498 * Move on to offset 0 of the next layer1 or layer2.
499 */
500#define HAMMER_ZONE_LAYER1_NEXT_OFFSET(offset) \
501 (((offset) + HAMMER_BLOCKMAP_LAYER2) & ~HAMMER_BLOCKMAP_LAYER2_MASK)
502
c4e15ea8
TK
503#define HAMMER_ZONE_LAYER2_NEXT_OFFSET(offset) \
504 (((offset) + HAMMER_BIGBLOCK_SIZE) & ~HAMMER_BIGBLOCK_MASK64)
505
bf686dbe
MD
506/*
507 * HAMMER UNDO parameters. The UNDO fifo is mapped directly in the volume
668f11d8
TK
508 * header with an array of zone-2 offsets. A maximum of (128x8MB) = 1GB,
509 * and minimum of (64x8MB) = 512MB may be reserved. The size of the undo
510 * fifo is usually set a newfs time.
bf686dbe 511 */
668f11d8 512#define HAMMER_MIN_UNDO_BIGBLOCKS 64
f04399b1 513#define HAMMER_MAX_UNDO_BIGBLOCKS 128
bf686dbe 514
9380c752 515/*
36f82b23
MD
516 * All on-disk HAMMER structures which make up elements of the UNDO FIFO
517 * contain a hammer_fifo_head and hammer_fifo_tail structure. This structure
40043e7f
MD
518 * contains all the information required to validate the fifo element
519 * and to scan the fifo in either direction. The head is typically embedded
520 * in higher level hammer on-disk structures while the tail is typically
521 * out-of-band. hdr_size is the size of the whole mess, including the tail.
47197d71 522 *
36f82b23
MD
523 * All undo structures are guaranteed to not cross a 16K filesystem
524 * buffer boundary. Most undo structures are fairly small. Data spaces
525 * are not immediately reused by HAMMER so file data is not usually recorded
526 * as part of an UNDO.
40043e7f
MD
527 *
528 * PAD elements are allowed to take up only 8 bytes of space as a special
529 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
530 * and with the tail overloaded onto the head structure for 8 bytes total.
36f82b23
MD
531 *
532 * Every undo record has a sequence number. This number is unrelated to
533 * transaction ids and instead collects the undo transactions associated
534 * with a single atomic operation. A larger transactional operation, such
535 * as a remove(), may consist of several smaller atomic operations
536 * representing raw meta-data operations.
02428fb6
MD
537 *
538 * HAMMER VERSION 4 CHANGES
539 *
540 * In HAMMER version 4 the undo structure alignment is reduced from 16384
541 * to 512 bytes in order to ensure that each 512 byte sector begins with
8cf688a7
TK
542 * a header. The hdr_seq field in the header is a 32 bit sequence number
543 * which allows the recovery code to detect missing sectors
02428fb6
MD
544 * without relying on the 32-bit crc and to definitively identify the current
545 * undo sequence space without having to rely on information from the volume
546 * header. In addition, new REDO entries in the undo space are used to
547 * record write, write/extend, and transaction id updates.
548 *
549 * The grand result is:
550 *
551 * (1) The volume header no longer needs to be synchronized for most
552 * flush and fsync operations.
553 *
554 * (2) Most fsync operations need only lay down REDO records
555 *
556 * (3) Data overwrite for nohistory operations covered by REDO records
557 * can be supported (instead of rolling a new block allocation),
558 * by rolling UNDO for the prior contents of the data.
3045a179
MD
559 *
560 * HAMMER VERSION 5 CHANGES
561 *
562 * Hammer version 5 contains a minor adjustment making layer2's bytes_free
563 * field signed, allowing dedup to push it into the negative domain.
9380c752 564 */
40043e7f 565#define HAMMER_HEAD_ALIGN 8
47197d71 566#define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
4c038e17
MD
567#define HAMMER_HEAD_DOALIGN(bytes) \
568 (((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK)
47197d71 569
02428fb6 570#define HAMMER_UNDO_ALIGN 512
46137e17 571#define HAMMER_UNDO_ALIGN64 ((uint64_t)512)
02428fb6
MD
572#define HAMMER_UNDO_MASK (HAMMER_UNDO_ALIGN - 1)
573#define HAMMER_UNDO_MASK64 (HAMMER_UNDO_ALIGN64 - 1)
19836c70
TK
574#define HAMMER_UNDO_DOALIGN(offset) \
575 (((offset) + HAMMER_UNDO_MASK) & ~HAMMER_UNDO_MASK64)
02428fb6 576
22a0040d 577typedef struct hammer_fifo_head {
46137e17
TK
578 uint16_t hdr_signature;
579 uint16_t hdr_type;
580 uint32_t hdr_size; /* Aligned size of the whole mess */
581 uint32_t hdr_seq; /* Sequence number */
19619882 582 hammer_crc_t hdr_crc; /* XOR crc up to field w/ crc after field */
22a0040d 583} *hammer_fifo_head_t;
40043e7f 584
19619882
MD
585#define HAMMER_FIFO_HEAD_CRCOFF offsetof(struct hammer_fifo_head, hdr_crc)
586
22a0040d 587typedef struct hammer_fifo_tail {
46137e17
TK
588 uint16_t tail_signature;
589 uint16_t tail_type;
590 uint32_t tail_size; /* aligned size of the whole mess */
22a0040d 591} *hammer_fifo_tail_t;
9380c752 592
8750964d 593/*
47197d71 594 * Fifo header types.
89660e86
TK
595 *
596 * NOTE: 0x8000U part of HAMMER_HEAD_TYPE_PAD can be removed if the HAMMER
597 * version ever gets bumped again. It exists only to keep compatibility with
598 * older versions.
8750964d 599 */
bd129eaa 600#define HAMMER_HEAD_TYPE_PAD (0x0040U | 0x8000U)
02428fb6 601#define HAMMER_HEAD_TYPE_DUMMY 0x0041U /* dummy entry w/seqno */
40043e7f 602#define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
02428fb6 603#define HAMMER_HEAD_TYPE_REDO 0x0044U /* data REDO / fast fsync */
47197d71 604
47197d71 605#define HAMMER_HEAD_SIGNATURE 0xC84EU
40043e7f 606#define HAMMER_TAIL_SIGNATURE 0xC74FU
47197d71
MD
607
608/*
bf686dbe 609 * Misc FIFO structures.
02428fb6 610 *
47f363f1 611 * UNDO - Raw meta-data media updates.
47197d71 612 */
22a0040d 613typedef struct hammer_fifo_undo {
47197d71 614 struct hammer_fifo_head head;
66086a39 615 hammer_off_t undo_offset; /* zone-1,2 offset */
bf686dbe
MD
616 int32_t undo_data_bytes;
617 int32_t undo_reserved01;
47197d71 618 /* followed by data */
22a0040d 619} *hammer_fifo_undo_t;
8750964d 620
47f363f1
MD
621/*
622 * REDO (HAMMER version 4+) - Logical file writes/truncates.
623 *
624 * REDOs contain information which will be duplicated in a later meta-data
625 * update, allowing fast write()+fsync() operations. REDOs can be ignored
626 * without harming filesystem integrity but must be processed if fsync()
627 * semantics are desired.
628 *
629 * Unlike UNDOs which are processed backwards within the recovery span,
630 * REDOs must be processed forwards starting further back (starting outside
631 * the recovery span).
632 *
633 * WRITE - Write logical file (with payload). Executed both
634 * out-of-span and in-span. Out-of-span WRITEs may be
635 * filtered out by TERMs.
636 *
637 * TRUNC - Truncate logical file (no payload). Executed both
638 * out-of-span and in-span. Out-of-span WRITEs may be
639 * filtered out by TERMs.
640 *
641 * TERM_* - Indicates meta-data was committed (if out-of-span) or
642 * will be rolled-back (in-span). Any out-of-span TERMs
643 * matching earlier WRITEs remove those WRITEs from
644 * consideration as they might conflict with a later data
645 * commit (which is not being rolled-back).
646 *
647 * SYNC - The earliest in-span SYNC (the last one when scanning
648 * backwards) tells the recovery code how far out-of-span
649 * it must go to run REDOs.
650 *
651 * NOTE: WRITEs do not always have matching TERMs even under
652 * perfect conditions because truncations might remove the
653 * buffers from consideration. I/O problems can also remove
654 * buffers from consideration.
655 *
656 * TRUNCSs do not always have matching TERMs because several
657 * truncations may be aggregated together into a single TERM.
658 */
22a0040d 659typedef struct hammer_fifo_redo {
02428fb6
MD
660 struct hammer_fifo_head head;
661 int64_t redo_objid; /* file being written */
662 hammer_off_t redo_offset; /* logical offset in file */
663 int32_t redo_data_bytes;
46137e17
TK
664 uint32_t redo_flags;
665 uint32_t redo_localization;
1b0a8c58
TK
666 uint32_t redo_reserved01;
667 uint64_t redo_reserved02;
668 /* followed by data */
22a0040d 669} *hammer_fifo_redo_t;
8750964d 670
47f363f1
MD
671#define HAMMER_REDO_WRITE 0x00000001
672#define HAMMER_REDO_TRUNC 0x00000002
673#define HAMMER_REDO_TERM_WRITE 0x00000004
674#define HAMMER_REDO_TERM_TRUNC 0x00000008
675#define HAMMER_REDO_SYNC 0x00000010
676
22a0040d 677typedef union hammer_fifo_any {
02428fb6
MD
678 struct hammer_fifo_head head;
679 struct hammer_fifo_undo undo;
680 struct hammer_fifo_redo redo;
22a0040d 681} *hammer_fifo_any_t;
02428fb6 682
c60bb2c5 683/*
47197d71 684 * Volume header types
c60bb2c5 685 */
8750964d 686#define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
8750964d
MD
687#define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
688
8750964d
MD
689/*
690 * HAMMER Volume header
691 *
89660e86 692 * A HAMMER filesystem can be built from 1-256 block devices, each block
47197d71
MD
693 * device contains a volume header followed by however many buffers fit
694 * into the volume.
8750964d 695 *
89660e86
TK
696 * One of the volumes making up a HAMMER filesystem is the root volume.
697 * The root volume is always volume #0 which is the first block device path
698 * specified by newfs_hammer(8). All HAMMER volumes have a volume header,
699 * however the root volume may be the only volume that has valid values for
700 * some fields in the header.
a89aec1b
MD
701 *
702 * Special field notes:
703 *
704 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
55e3f6f3
TK
705 * vol_mem_beg - offset of memory log (buf_beg - mem_beg bytes)
706 * vol_buf_beg - offset of the first buffer in volume
707 * vol_buf_end - offset of volume EOF (on buffer boundary)
a89aec1b
MD
708 *
709 * The memory log area allows a kernel to cache new records and data
710 * in memory without allocating space in the actual filesystem to hold
711 * the records and data. In the event that a filesystem becomes full,
712 * any records remaining in memory can be flushed to the memory log
713 * area. This allows the kernel to immediately return success.
55e3f6f3
TK
714 *
715 * The buffer offset is a physical offset of zone-2 offset. The lower
716 * 52 bits of the zone-2 offset is added to the buffer offset of each
717 * volume to generate an actual I/O offset within the block device.
718 *
719 * NOTE: boot area and memory log are currently not used.
8750964d 720 */
8750964d 721
4e8098b2
TK
722/*
723 * Filesystem type string
724 */
725#define HAMMER_FSTYPE_STRING "DragonFly HAMMER"
726
89660e86
TK
727/*
728 * These macros are only used by userspace when userspace commands either
729 * initialize or add a new HAMMER volume.
730 */
a89aec1b
MD
731#define HAMMER_BOOT_MINBYTES (32*1024)
732#define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
733#define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
734
735#define HAMMER_MEM_MINBYTES (256*1024)
736#define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
737#define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
738
22a0040d 739typedef struct hammer_volume_ondisk {
55e3f6f3 740 uint64_t vol_signature; /* HAMMER_FSBUF_VOLUME for a valid header */
40043e7f 741
52b12366
TK
742 /*
743 * These are relative to block device offset, not zone offsets.
744 */
55e3f6f3
TK
745 int64_t vol_bot_beg; /* offset of boot area */
746 int64_t vol_mem_beg; /* offset of memory log */
747 int64_t vol_buf_beg; /* offset of the first buffer in volume */
748 int64_t vol_buf_end; /* offset of volume EOF (on buffer boundary) */
77f7bc0a 749 int64_t vol_reserved01;
8750964d 750
90da8fc8
TK
751 hammer_uuid_t vol_fsid; /* identify filesystem */
752 hammer_uuid_t vol_fstype; /* identify filesystem type */
6c39d27a 753 char vol_label[64]; /* filesystem label */
8750964d
MD
754
755 int32_t vol_no; /* volume number within filesystem */
52b12366 756 int32_t vol_count; /* number of volumes making up filesystem */
8750964d 757
46137e17 758 uint32_t vol_version; /* version control information */
19619882 759 hammer_crc_t vol_crc; /* header crc */
46137e17 760 uint32_t vol_flags; /* volume flags */
52b12366 761 uint32_t vol_rootvol; /* the root volume number (must be 0) */
8750964d 762
cbccd318 763 uint32_t vol_reserved[8];
fbc6e32a 764
8750964d
MD
765 /*
766 * These fields are initialized and space is reserved in every
89660e86 767 * volume making up a HAMMER filesytem, but only the root volume
042505d7
TK
768 * contains valid data. Note that vol0_stat_bigblocks does not
769 * include big-blocks for freemap and undomap initially allocated
89660e86 770 * by newfs_hammer(8).
8750964d 771 */
a981af19
TK
772 int64_t vol0_stat_bigblocks; /* total big-blocks when fs is empty */
773 int64_t vol0_stat_freebigblocks;/* number of free big-blocks */
77f7bc0a 774 int64_t vol0_reserved01;
fbc6e32a 775 int64_t vol0_stat_inodes; /* for statfs only */
77f7bc0a 776 int64_t vol0_reserved02;
52b12366 777 hammer_off_t vol0_btree_root; /* B-Tree root offset in zone-8 */
4889cbd4 778 hammer_tid_t vol0_next_tid; /* highest partially synchronized TID */
77f7bc0a 779 hammer_off_t vol0_reserved03;
40043e7f
MD
780
781 /*
0729c8c8
MD
782 * Blockmaps for zones. Not all zones use a blockmap. Note that
783 * the entire root blockmap is cached in the hammer_mount structure.
40043e7f 784 */
c3be93f2 785 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES];
40043e7f 786
bf686dbe 787 /*
cb51be26 788 * Array of zone-2 addresses for undo FIFO.
bf686dbe 789 */
f04399b1 790 hammer_off_t vol0_undo_array[HAMMER_MAX_UNDO_BIGBLOCKS];
22a0040d 791} *hammer_volume_ondisk_t;
8cd0a023 792
7a39af79
TK
793#define HAMMER_ROOT_VOLNO 0
794
748efb59 795#define HAMMER_VOLF_NEEDFLUSH 0x0004 /* volume needs flush */
8cd0a023 796
19619882
MD
797#define HAMMER_VOL_CRCSIZE1 \
798 offsetof(struct hammer_volume_ondisk, vol_crc)
799#define HAMMER_VOL_CRCSIZE2 \
800 (sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 - \
801 sizeof(hammer_crc_t))
802
44a83111 803#define HAMMER_VOL_VERSION_MIN 1 /* minimum supported version */
4c09d9c4
MD
804#define HAMMER_VOL_VERSION_DEFAULT 7 /* newfs default version */
805#define HAMMER_VOL_VERSION_WIP 8 /* version >= this is WIP */
806#define HAMMER_VOL_VERSION_MAX 7 /* maximum supported version */
44a83111 807
5e435c92 808#define HAMMER_VOL_VERSION_ONE 1
beec5dc4 809#define HAMMER_VOL_VERSION_TWO 2 /* new dirent layout (2.3+) */
83f2a3aa 810#define HAMMER_VOL_VERSION_THREE 3 /* new snapshot layout (2.5+) */
fd74884e 811#define HAMMER_VOL_VERSION_FOUR 4 /* new undo/flush (2.5+) */
3045a179 812#define HAMMER_VOL_VERSION_FIVE 5 /* dedup (2.9+) */
89cba4df 813#define HAMMER_VOL_VERSION_SIX 6 /* DIRHASH_ALG1 */
4c09d9c4 814#define HAMMER_VOL_VERSION_SEVEN 7 /* use the faster iscsi_crc */
83f2a3aa 815
516655e8
TK
816/*
817 * Translate a zone-2 address to physical address
818 */
819#define hammer_xlate_to_phys(volume, zone2_offset) \
eee3f66c 820 ((volume)->vol_buf_beg + HAMMER_OFF_SHORT_ENCODE(zone2_offset))
516655e8 821
6831865c
TK
822/*
823 * Translate a zone-3 address to zone-2 address
824 */
825#define HAMMER_UNDO_INDEX(zone3_offset) \
826 (HAMMER_OFF_SHORT_ENCODE(zone3_offset) / HAMMER_BIGBLOCK_SIZE)
827
828#define hammer_xlate_to_undo(volume, zone3_offset) \
829 ((volume)->vol0_undo_array[HAMMER_UNDO_INDEX(zone3_offset)] + \
830 (zone3_offset & HAMMER_BIGBLOCK_MASK64))
831
0dd9c298
TK
832/*
833 * Effective per-volume filesystem capacity including big-blocks for layer1/2
834 */
835#define HAMMER_VOL_BUF_SIZE(volume) \
836 ((volume)->vol_buf_end - (volume)->vol_buf_beg)
837
c60bb2c5
MD
838/*
839 * Record types are fairly straightforward. The B-Tree includes the record
840 * type in its index sort.
c60bb2c5 841 */
2b519166 842#define HAMMER_RECTYPE_UNKNOWN 0x0000
2b519166 843#define HAMMER_RECTYPE_INODE 0x0001 /* inode in obj_id space */
bf686dbe
MD
844#define HAMMER_RECTYPE_DATA 0x0010
845#define HAMMER_RECTYPE_DIRENTRY 0x0011
846#define HAMMER_RECTYPE_DB 0x0012
847#define HAMMER_RECTYPE_EXT 0x0013 /* ext attributes */
848#define HAMMER_RECTYPE_FIX 0x0014 /* fixed attribute */
ea434b6f 849#define HAMMER_RECTYPE_PFS 0x0015 /* PFS management */
83f2a3aa
MD
850#define HAMMER_RECTYPE_SNAPSHOT 0x0016 /* Snapshot management */
851#define HAMMER_RECTYPE_CONFIG 0x0017 /* hammer cleanup config */
a9d52b76
MD
852#define HAMMER_RECTYPE_MAX 0xFFFF
853
b0037924 854#define HAMMER_RECTYPE_ENTRY_START (HAMMER_RECTYPE_INODE + 1)
a9d52b76 855#define HAMMER_RECTYPE_CLEAN_START HAMMER_RECTYPE_EXT
7a04d74f
MD
856
857#define HAMMER_FIXKEY_SYMLINK 1
8750964d 858
098f3e0c 859#define HAMMER_OBJTYPE_UNKNOWN 0 /* never exists on-disk as unknown */
8750964d
MD
860#define HAMMER_OBJTYPE_DIRECTORY 1
861#define HAMMER_OBJTYPE_REGFILE 2
862#define HAMMER_OBJTYPE_DBFILE 3
863#define HAMMER_OBJTYPE_FIFO 4
c60bb2c5
MD
864#define HAMMER_OBJTYPE_CDEV 5
865#define HAMMER_OBJTYPE_BDEV 6
866#define HAMMER_OBJTYPE_SOFTLINK 7
867#define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
b3bad96f 868#define HAMMER_OBJTYPE_SOCKET 9
c60bb2c5 869
8750964d 870/*
11ad5ade 871 * HAMMER inode attribute data
8750964d 872 *
11ad5ade
MD
873 * The data reference for a HAMMER inode points to this structure. Any
874 * modifications to the contents of this structure will result in a
875 * replacement operation.
8750964d 876 *
5a930e66
MD
877 * parent_obj_id is only valid for directories (which cannot be hard-linked),
878 * and specifies the parent directory obj_id. This field will also be set
879 * for non-directory inodes as a recovery aid, but can wind up holding
880 * stale information. However, since object id's are not reused, the worse
881 * that happens is that the recovery code is unable to use it.
d9ecde89 882 * A parent_obj_id of 0 means it's a root inode of root or non-root PFS.
ddfdf542
MD
883 *
884 * NOTE: Future note on directory hardlinks. We can implement a record type
885 * which allows us to point to multiple parent directories.
8750964d 886 */
1743d7ba 887typedef struct hammer_inode_data {
46137e17
TK
888 uint16_t version; /* inode data version */
889 uint16_t mode; /* basic unix permissions */
890 uint32_t uflags; /* chflags */
891 uint32_t rmajor; /* used by device nodes */
892 uint32_t rminor; /* used by device nodes */
893 uint64_t ctime;
adf01747 894 int64_t parent_obj_id; /* parent directory obj_id */
90da8fc8
TK
895 hammer_uuid_t uid;
896 hammer_uuid_t gid;
11ad5ade 897
46137e17
TK
898 uint8_t obj_type;
899 uint8_t cap_flags; /* capability support flags (extension) */
9d5dfd72 900 uint16_t reserved01;
cbccd318 901 uint32_t reserved02;
46137e17
TK
902 uint64_t nlinks; /* hard links */
903 uint64_t size; /* filesystem object size */
2f85fa4d 904 union {
2f85fa4d
MD
905 char symlink[24]; /* HAMMER_INODE_BASESYMLEN */
906 } ext;
46137e17
TK
907 uint64_t mtime; /* mtime must be second-to-last */
908 uint64_t atime; /* atime must be last */
1743d7ba 909} *hammer_inode_data_t;
8750964d 910
ddfdf542
MD
911/*
912 * Neither mtime nor atime upates are CRCd by the B-Tree element.
913 * mtime updates have UNDO, atime updates do not.
914 */
ddfdf542
MD
915#define HAMMER_INODE_CRCSIZE \
916 offsetof(struct hammer_inode_data, mtime)
917
11ad5ade 918#define HAMMER_INODE_DATA_VERSION 1
dc241916 919#define HAMMER_OBJID_ROOT 1 /* root inodes # */
ddfdf542 920#define HAMMER_INODE_BASESYMLEN 24 /* see ext.symlink */
8750964d 921
5e435c92
MD
922/*
923 * Capability & implementation flags.
beec5dc4 924 *
ed959c01
TK
925 * HAMMER_INODE_CAP_DIR_LOCAL_INO - Use inode B-Tree localization
926 * for directory entries. Also see HAMMER_DIR_INODE_LOCALIZATION().
5e435c92
MD
927 */
928#define HAMMER_INODE_CAP_DIRHASH_MASK 0x03 /* directory: hash algorithm */
929#define HAMMER_INODE_CAP_DIRHASH_ALG0 0x00
930#define HAMMER_INODE_CAP_DIRHASH_ALG1 0x01
931#define HAMMER_INODE_CAP_DIRHASH_ALG2 0x02
932#define HAMMER_INODE_CAP_DIRHASH_ALG3 0x03
beec5dc4 933#define HAMMER_INODE_CAP_DIR_LOCAL_INO 0x04 /* use inode localization */
5e435c92 934
f097bffe
TK
935#define HAMMER_DATA_DOALIGN(offset) \
936 (((offset) + 15) & ~15)
937#define HAMMER_DATA_DOALIGN_WITH(type, offset) \
938 (((type)(offset) + 15) & (~(type)15))
939
8750964d 940/*
ddfdf542 941 * A HAMMER directory entry associates a HAMMER filesystem object with a
8ee21624
TK
942 * namespace. It is hooked into a pseudo-filesystem (with its own inode
943 * numbering space) in the filesystem by setting the high 16 bits of the
944 * localization field. The low 16 bits must be 0 and are reserved for
945 * future use.
8750964d
MD
946 *
947 * Directory entries are indexed with a 128 bit namekey rather then an
ddfdf542 948 * offset. A portion of the namekey is an iterator/randomizer to deal
8750964d 949 * with collisions.
66325755 950 *
b5d9eb6d 951 * NOTE: leaf.base.obj_type from the related B-Tree leaf entry holds
ddfdf542 952 * the filesystem object type of obj_id, e.g. a den_type equivalent.
9a46be39 953 * It is not stored in hammer_direntry_data.
6b4f890b 954 *
b5d9eb6d 955 * NOTE: name field / the filename data reference is NOT terminated with \0.
8750964d 956 */
1743d7ba 957typedef struct hammer_direntry_data {
adf01747 958 int64_t obj_id; /* object being referenced */
46137e17 959 uint32_t localization; /* identify pseudo-filesystem */
77f7bc0a 960 uint32_t reserved01;
11ad5ade 961 char name[16]; /* name (extended) */
1743d7ba 962} *hammer_direntry_data_t;
8750964d 963
9a46be39
TK
964#define HAMMER_ENTRY_NAME_OFF offsetof(struct hammer_direntry_data, name[0])
965#define HAMMER_ENTRY_SIZE(nlen) offsetof(struct hammer_direntry_data, name[nlen])
19619882 966
ddfdf542 967/*
905f4108 968 * Symlink data which does not fit in the inode is stored in a separate
ddfdf542
MD
969 * FIX type record.
970 */
1743d7ba 971typedef struct hammer_symlink_data {
b5d9eb6d 972 char name[16]; /* name (extended) */
1743d7ba 973} *hammer_symlink_data_t;
8750964d 974
11ad5ade 975#define HAMMER_SYMLINK_NAME_OFF offsetof(struct hammer_symlink_data, name[0])
7a04d74f 976
5fa5c92f
MD
977/*
978 * The root inode for the primary filesystem and root inode for any
979 * pseudo-fs may be tagged with an optional data structure using
cf620e4b 980 * HAMMER_RECTYPE_PFS and localization id. This structure allows
5fa5c92f
MD
981 * the node to be used as a mirroring master or slave.
982 *
983 * When operating as a slave CD's into the node automatically become read-only
a56cb012 984 * and as-of sync_end_tid.
5fa5c92f 985 *
a56cb012 986 * When operating as a master the read PFSD info sets sync_end_tid to
5fa5c92f
MD
987 * the most recently flushed TID.
988 *
989 * sync_low_tid is not yet used but will represent the highest pruning
990 * end-point, after which full history is available.
c7380858
MN
991 *
992 * We need to pack this structure making it equally sized on both 32-bit and
993 * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is
994 * send over the wire in hammer mirror operations. Only on 64-bit machines
995 * the size of this struct differ when packed or not. This leads us to the
996 * situation where old 64-bit systems (using the non-packed structure),
997 * which were never able to mirror to/from 32-bit systems, are now no longer
998 * able to mirror to/from newer 64-bit systems (using the packed structure).
5fa5c92f
MD
999 */
1000struct hammer_pseudofs_data {
1001 hammer_tid_t sync_low_tid; /* full history beyond this point */
a56cb012
MD
1002 hammer_tid_t sync_beg_tid; /* earliest tid w/ full history avail */
1003 hammer_tid_t sync_end_tid; /* current synchronizatoin point */
46137e17
TK
1004 uint64_t sync_beg_ts; /* real-time of last completed sync */
1005 uint64_t sync_end_ts; /* initiation of current sync cycle */
90da8fc8
TK
1006 hammer_uuid_t shared_uuid; /* shared uuid (match required) */
1007 hammer_uuid_t unique_uuid; /* unique uuid of this master/slave */
732a1697 1008 int32_t reserved01; /* reserved for future master_id */
4c038e17 1009 int32_t mirror_flags; /* misc flags */
5fa5c92f 1010 char label[64]; /* filesystem space label */
260b4b9e 1011 char snapshots[64]; /* softlink dir for pruning */
f2c87c8e
TK
1012 int32_t reserved02; /* was prune_{time,freq} */
1013 int32_t reserved03; /* was reblock_{time,freq} */
1014 int32_t reserved04; /* was snapshot_freq */
4c038e17
MD
1015 int32_t prune_min; /* do not prune recent history */
1016 int32_t prune_max; /* do not retain history beyond here */
1017 int32_t reserved[16];
c7380858 1018} __packed;
5fa5c92f
MD
1019
1020typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t;
1021
1022#define HAMMER_PFSD_SLAVE 0x00000001
842e7a70 1023#define HAMMER_PFSD_DELETED 0x80000000
5fa5c92f 1024
2bacc088
TK
1025#define hammer_is_pfs_slave(pfsd) \
1026 (((pfsd)->mirror_flags & HAMMER_PFSD_SLAVE) != 0)
1027#define hammer_is_pfs_master(pfsd) \
1028 (!hammer_is_pfs_slave(pfsd))
1029#define hammer_is_pfs_deleted(pfsd) \
1030 (((pfsd)->mirror_flags & HAMMER_PFSD_DELETED) != 0)
1031
d43a0fe3
TK
1032#define HAMMER_MAX_PFS 65536
1033#define HAMMER_MAX_PFSID (HAMMER_MAX_PFS - 1)
753daa18 1034#define HAMMER_ROOT_PFSID 0
d43a0fe3 1035
83f2a3aa
MD
1036/*
1037 * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }.
1038 *
1039 * Snapshot records replace the old <fs>/snapshots/<softlink> methodology. Snapshot
1040 * records are mirrored but may be independantly managed once they are laid down on
1041 * a slave.
1042 *
1043 * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the
1044 * results.
1045 *
1046 * NOTE: Reserved fields must be zero (as usual)
1047 */
1743d7ba 1048typedef struct hammer_snapshot_data {
83f2a3aa 1049 hammer_tid_t tid; /* the snapshot TID itself (== key) */
46137e17
TK
1050 uint64_t ts; /* real-time when snapshot was made */
1051 uint64_t reserved01;
1052 uint64_t reserved02;
83f2a3aa 1053 char label[64]; /* user-supplied description */
46137e17 1054 uint64_t reserved03[4];
1743d7ba 1055} *hammer_snapshot_data_t;
83f2a3aa
MD
1056
1057/*
1058 * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }.
1059 *
1060 * Used to store the hammer cleanup config. This data is not mirrored.
1061 */
1743d7ba 1062typedef struct hammer_config_data {
83f2a3aa 1063 char text[1024];
1743d7ba 1064} *hammer_config_data_t;
83f2a3aa 1065
c60bb2c5
MD
1066/*
1067 * Rollup various structures embedded as record data
1068 */
22a0040d 1069typedef union hammer_data_ondisk {
9a46be39 1070 struct hammer_direntry_data entry;
c60bb2c5 1071 struct hammer_inode_data inode;
11ad5ade 1072 struct hammer_symlink_data symlink;
842e7a70 1073 struct hammer_pseudofs_data pfsd;
83f2a3aa
MD
1074 struct hammer_snapshot_data snap;
1075 struct hammer_config_data config;
22a0040d 1076} *hammer_data_ondisk_t;
11ad5ade 1077
13c9bf36
TK
1078/*
1079 * Ondisk layout of B-Tree related structures
1080 */
1081#include "hammer_btree.h"
ed959c01
TK
1082
1083#define HAMMER_DIR_INODE_LOCALIZATION(ino_data) \
1084 (((ino_data)->cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO) ? \
1085 HAMMER_LOCALIZE_INODE : \
1086 HAMMER_LOCALIZE_MISC)
1087
964cb30d 1088#endif /* !VFS_HAMMER_DISK_H_ */