Mention recent added ioctls in bpf man page.
[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 *
f03c9cf4 34 * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.25 2008/02/23 03:01:08 dillon Exp $
8750964d
MD
35 */
36
7dc57964
MD
37#ifndef VFS_HAMMER_DISK_H_
38#define VFS_HAMMER_DISK_H_
39
8750964d
MD
40#ifndef _SYS_UUID_H_
41#include <sys/uuid.h>
42#endif
43
44/*
45 * The structures below represent the on-disk format for a HAMMER
46 * filesystem. Note that all fields for on-disk structures are naturally
47 * aligned. The host endian format is used - compatibility is possible
48 * if the implementation detects reversed endian and adjusts data accordingly.
49 *
50 * Most of HAMMER revolves around the concept of an object identifier. An
51 * obj_id is a 64 bit quantity which uniquely identifies a filesystem object
52 * FOR THE ENTIRE LIFE OF THE FILESYSTEM. This uniqueness allows backups
53 * and mirrors to retain varying amounts of filesystem history by removing
54 * any possibility of conflict through identifier reuse.
55 *
56 * A HAMMER filesystem may spam multiple volumes.
57 *
58 * A HAMMER filesystem uses a 16K filesystem buffer size. All filesystem
c60bb2c5
MD
59 * I/O is done in multiples of 16K. Most buffer-sized headers such as those
60 * used by volumes, super-clusters, clusters, and basic filesystem buffers
61 * use fixed-sized A-lists which are heavily dependant on HAMMER_BUFSIZE.
40043e7f
MD
62 *
63 * Per-volume storage limit: 52 bits 4096 TB
64 * Per-Zone storage limit: 59 bits 512 KTB (due to blockmap)
65 * Per-filesystem storage limit: 60 bits 1 MTB
8750964d 66 */
40043e7f
MD
67#define HAMMER_BUFSIZE 16384
68#define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1)
69#define HAMMER_MAXDATA (256*1024)
70#define HAMMER_BUFFER_BITS 14
71
72#if (1 << HAMMER_BUFFER_BITS) != HAMMER_BUFSIZE
73#error "HAMMER_BUFFER_BITS BROKEN"
74#endif
8750964d 75
47197d71
MD
76#define HAMMER_BUFSIZE64 ((u_int64_t)HAMMER_BUFSIZE)
77#define HAMMER_BUFMASK64 ((u_int64_t)HAMMER_BUFMASK)
78
79#define HAMMER_OFF_ZONE_MASK 0xF000000000000000ULL /* zone portion */
80#define HAMMER_OFF_VOL_MASK 0x0FF0000000000000ULL /* volume portion */
81#define HAMMER_OFF_SHORT_MASK 0x000FFFFFFFFFFFFFULL /* offset portion */
82#define HAMMER_OFF_LONG_MASK 0x0FFFFFFFFFFFFFFFULL /* offset portion */
83#define HAMMER_OFF_SHORT_REC_MASK 0x000FFFFFFF000000ULL /* recovery boundary */
84#define HAMMER_OFF_LONG_REC_MASK 0x0FFFFFFFFF000000ULL /* recovery boundary */
85#define HAMMER_RECOVERY_BND 0x0000000001000000ULL
86
8750964d
MD
87/*
88 * Hammer transction ids are 64 bit unsigned integers and are usually
89 * synchronized with the time of day in nanoseconds.
47197d71
MD
90 *
91 * Hammer offsets are used for FIFO indexing and embed a cycle counter
92 * and volume number in addition to the offset. Most offsets are required
93 * to be 64-byte aligned.
8750964d
MD
94 */
95typedef u_int64_t hammer_tid_t;
47197d71
MD
96typedef u_int64_t hammer_off_t;
97
98#define HAMMER_MIN_TID 0ULL /* unsigned */
99#define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL /* unsigned */
100#define HAMMER_MIN_KEY -0x8000000000000000LL /* signed */
101#define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL /* signed */
102#define HAMMER_MIN_OBJID HAMMER_MIN_KEY /* signed */
103#define HAMMER_MAX_OBJID HAMMER_MAX_KEY /* signed */
104#define HAMMER_MIN_RECTYPE 0x0U /* unsigned */
105#define HAMMER_MAX_RECTYPE 0xFFFFU /* unsigned */
106#define HAMMER_MIN_OFFSET 0ULL /* unsigned */
107#define HAMMER_MAX_OFFSET 0xFFFFFFFFFFFFFFFFULL /* unsigned */
108
109/*
110 * hammer_off_t has several different encodings. Note that not all zones
111 * encode a vol_no.
112 *
113 * zone 0 (z,v,o): reserved (for sanity)
114 * zone 1 (z,v,o): raw volume relative (offset 0 is the volume header)
115 * zone 2 (z,v,o): raw buffer relative (offset 0 is the first buffer)
40043e7f 116 * zone 3 (z,o): undo fifo - blockmap backed
c3be93f2 117 * zone 4 (z,v,o): freemap - freemap-backed self-mapping
40043e7f
MD
118 *
119 * zone 8 (z,o): B-Tree - blkmap-backed
120 * zone 9 (z,o): Record - blkmap-backed
121 * zone 10 (z,o): Large-data - blkmap-backed
47197d71
MD
122 */
123
124#define HAMMER_ZONE_RAW_VOLUME 0x1000000000000000ULL
125#define HAMMER_ZONE_RAW_BUFFER 0x2000000000000000ULL
40043e7f 126#define HAMMER_ZONE_UNDO 0x3000000000000000ULL
c3be93f2 127#define HAMMER_ZONE_FREEMAP 0x4000000000000000ULL
40043e7f
MD
128#define HAMMER_ZONE_RESERVED05 0x5000000000000000ULL
129#define HAMMER_ZONE_RESERVED06 0x6000000000000000ULL
130#define HAMMER_ZONE_RESERVED07 0x7000000000000000ULL
131#define HAMMER_ZONE_BTREE 0x8000000000000000ULL
132#define HAMMER_ZONE_RECORD 0x9000000000000000ULL
133#define HAMMER_ZONE_LARGE_DATA 0xA000000000000000ULL
134#define HAMMER_ZONE_SMALL_DATA 0xB000000000000000ULL
135#define HAMMER_ZONE_RESERVED0C 0xC000000000000000ULL
136#define HAMMER_ZONE_RESERVED0D 0xD000000000000000ULL
137#define HAMMER_ZONE_RESERVED0E 0xE000000000000000ULL
138#define HAMMER_ZONE_RESERVED0F 0xF000000000000000ULL
139
140#define HAMMER_ZONE_RAW_VOLUME_INDEX 1
141#define HAMMER_ZONE_RAW_BUFFER_INDEX 2
142#define HAMMER_ZONE_UNDO_INDEX 3
c3be93f2 143#define HAMMER_ZONE_FREEMAP_INDEX 4
40043e7f
MD
144#define HAMMER_ZONE_BTREE_INDEX 8
145#define HAMMER_ZONE_RECORD_INDEX 9
146#define HAMMER_ZONE_LARGE_DATA_INDEX 10
147#define HAMMER_ZONE_SMALL_DATA_INDEX 11
148
f03c9cf4
MD
149/*
150 * Per-zone size limitation. This just makes the iterator easier
151 * to deal with by preventing an iterator overflow.
152 */
153#define HAMMER_ZONE_LIMIT \
154 (0x1000000000000000ULL - HAMMER_BLOCKMAP_LAYER2)
155
40043e7f 156#define HAMMER_MAX_ZONES 16
47197d71
MD
157
158#define HAMMER_VOL_ENCODE(vol_no) \
159 ((hammer_off_t)((vol_no) & 255) << 52)
160#define HAMMER_VOL_DECODE(ham_off) \
161 (int32_t)(((hammer_off_t)(ham_off) >> 52) & 255)
40043e7f
MD
162#define HAMMER_ZONE_DECODE(ham_off) \
163 (int32_t)(((hammer_off_t)(ham_off) >> 60))
f03c9cf4
MD
164#define HAMMER_ZONE_ENCODE(zone, ham_off) \
165 (((hammer_off_t)(zone) << 60) | (ham_off))
47197d71
MD
166#define HAMMER_SHORT_OFF_ENCODE(offset) \
167 ((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
168#define HAMMER_LONG_OFF_ENCODE(offset) \
169 ((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
170
171#define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset) \
172 (HAMMER_ZONE_RAW_VOLUME | \
173 HAMMER_VOL_ENCODE(vol_no) | \
174 HAMMER_SHORT_OFF_ENCODE(offset))
175
176#define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset) \
177 (HAMMER_ZONE_RAW_BUFFER | \
178 HAMMER_VOL_ENCODE(vol_no) | \
179 HAMMER_SHORT_OFF_ENCODE(offset))
8750964d 180
c3be93f2
MD
181#define HAMMER_ENCODE_FREEMAP(vol_no, offset) \
182 (HAMMER_ZONE_FREEMAP | \
183 HAMMER_VOL_ENCODE(vol_no) | \
184 HAMMER_SHORT_OFF_ENCODE(offset))
185
40043e7f
MD
186/*
187 * Large-Block backing store
188 *
189 * A blockmap is a two-level map which translates a blockmap-backed zone
190 * offset into a raw zone 2 offset. Each layer handles 18 bits. The 8M
191 * large-block size is 23 bits so two layers gives us 23+18+18 = 59 bits
192 * of address space.
193 */
194#define HAMMER_LARGEBLOCK_SIZE (8192 * 1024)
c3be93f2 195#define HAMMER_LARGEBLOCK_SIZE64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE)
40043e7f
MD
196#define HAMMER_LARGEBLOCK_MASK (HAMMER_LARGEBLOCK_SIZE - 1)
197#define HAMMER_LARGEBLOCK_MASK64 ((u_int64_t)HAMMER_LARGEBLOCK_SIZE - 1)
198#define HAMMER_LARGEBLOCK_BITS 23
199#if (1 << HAMMER_LARGEBLOCK_BITS) != HAMMER_LARGEBLOCK_SIZE
200#error "HAMMER_LARGEBLOCK_BITS BROKEN"
201#endif
202
203#define HAMMER_BUFFERS_PER_LARGEBLOCK \
204 (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE)
205#define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK \
206 (HAMMER_BUFFERS_PER_LARGEBLOCK - 1)
207#define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK64 \
208 ((hammer_off_t)HAMMER_BUFFERS_PER_LARGEBLOCK_MASK)
209
c3be93f2
MD
210/*
211 * Every blockmap has this root structure in the root volume header.
212 */
213struct hammer_blockmap {
214 hammer_off_t phys_offset; /* zone-2 physical offset */
215 hammer_off_t next_offset; /* zone-X logical offset */
216 hammer_off_t alloc_offset; /* zone-X logical offset */
217 u_int32_t entry_crc;
218 u_int32_t reserved01;
219};
40043e7f 220
c3be93f2 221typedef struct hammer_blockmap *hammer_blockmap_t;
40043e7f 222
c3be93f2
MD
223/*
224 * The blockmap is a 2-layer entity made up of big-blocks. The first layer
225 * contains 262144 32-byte entries (18 bits), the second layer contains
226 * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
227 * 18+19+23 = 60 bits. The top four bits are the zone id.
228 *
229 * Layer 2 encodes the physical bigblock mapping for a blockmap. The freemap
230 * uses this field to encode the virtual blockmap offset that allocated the
231 * physical block.
232 *
233 * NOTE: The freemap maps the vol_no in the upper 8 bits of layer1.
234 *
235 * zone-4 blockmap offset: [z:4][layer1:18][layer2:19][bigblock:23]
236 */
237struct hammer_blockmap_layer1 {
238 hammer_off_t blocks_free; /* big-blocks free */
239 hammer_off_t phys_offset; /* UNAVAIL or zone-2 */
240 u_int32_t layer1_crc; /* crc of this entry */
241 u_int32_t layer2_crc; /* xor'd crc's of HAMMER_BLOCKSIZE */
242 hammer_off_t reserved01;
243};
40043e7f 244
c3be93f2 245struct hammer_blockmap_layer2 {
40043e7f 246 u_int32_t entry_crc;
c3be93f2
MD
247 u_int32_t bytes_free; /* bytes free within this bigblock */
248 union {
249 hammer_off_t owner; /* used by freemap */
250 hammer_off_t phys_offset; /* used by blockmap */
251 } u;
40043e7f
MD
252};
253
c3be93f2
MD
254#define HAMMER_BLOCKMAP_FREE 0ULL
255#define HAMMER_BLOCKMAP_UNAVAIL ((hammer_off_t)-1LL)
256
257#define HAMMER_BLOCKMAP_RADIX1 /* 262144 (18) */ \
258 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1))
259#define HAMMER_BLOCKMAP_RADIX2 /* 524288 (19) */ \
260 (HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2))
261
262#define HAMMER_BLOCKMAP_RADIX1_PERBUFFER \
263 (HAMMER_BLOCKMAP_RADIX1 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
264#define HAMMER_BLOCKMAP_RADIX2_PERBUFFER \
265 (HAMMER_BLOCKMAP_RADIX2 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
266
267#define HAMMER_BLOCKMAP_LAYER1 /* 18+19+23 */ \
268 (HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
269#define HAMMER_BLOCKMAP_LAYER2 /* 19+23 */ \
270 (HAMMER_BLOCKMAP_RADIX2 * HAMMER_LARGEBLOCK_SIZE64)
271
272#define HAMMER_BLOCKMAP_LAYER1_MASK (HAMMER_BLOCKMAP_LAYER1 - 1)
273#define HAMMER_BLOCKMAP_LAYER2_MASK (HAMMER_BLOCKMAP_LAYER2 - 1)
274
275/*
276 * byte offset within layer1 or layer2 big-block for the entry representing
277 * a zone-2 physical offset.
278 */
279#define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset) \
280 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / \
281 HAMMER_BLOCKMAP_LAYER2 * sizeof(struct hammer_blockmap_layer1))
282
283#define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset) \
284 (((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) / \
285 HAMMER_LARGEBLOCK_SIZE64 * sizeof(struct hammer_blockmap_layer2))
66325755 286
9380c752 287/*
47197d71 288 * All on-disk HAMMER structures which make up elements of the FIFO contain
40043e7f
MD
289 * a hammer_fifo_head and hammer_fifo_tail structure. This structure
290 * contains all the information required to validate the fifo element
291 * and to scan the fifo in either direction. The head is typically embedded
292 * in higher level hammer on-disk structures while the tail is typically
293 * out-of-band. hdr_size is the size of the whole mess, including the tail.
47197d71
MD
294 *
295 * Nearly all such structures are guaranteed to not cross a 16K filesystem
296 * buffer boundary. The one exception is a record, whos related data may
297 * cross a buffer boundary.
298 *
299 * HAMMER guarantees alignment with a fifo head structure at 16MB intervals
300 * (i.e. the base of the buffer will not be in the middle of a data record).
301 * This is used to allow the recovery code to re-sync after hitting corrupted
302 * data.
40043e7f
MD
303 *
304 * PAD elements are allowed to take up only 8 bytes of space as a special
305 * case, containing only hdr_signature, hdr_type, and hdr_size fields,
306 * and with the tail overloaded onto the head structure for 8 bytes total.
9380c752 307 */
40043e7f 308#define HAMMER_HEAD_ONDISK_SIZE 24
47197d71 309#define HAMMER_HEAD_RECOVERY_ALIGNMENT (16 * 1024 * 1024)
40043e7f 310#define HAMMER_HEAD_ALIGN 8
47197d71 311#define HAMMER_HEAD_ALIGN_MASK (HAMMER_HEAD_ALIGN - 1)
40043e7f 312#define HAMMER_TAIL_ONDISK_SIZE 8
47197d71
MD
313
314struct hammer_fifo_head {
315 u_int16_t hdr_signature;
316 u_int16_t hdr_type;
40043e7f 317 u_int32_t hdr_size; /* aligned size of the whole mess */
47197d71 318 u_int32_t hdr_crc;
40043e7f
MD
319 u_int32_t hdr_reserved02;
320 hammer_tid_t hdr_seq; /* related sequence number */
321};
322
323struct hammer_fifo_tail {
324 u_int16_t tail_signature;
325 u_int16_t tail_type;
326 u_int32_t tail_size; /* aligned size of the whole mess */
47197d71
MD
327};
328
329typedef struct hammer_fifo_head *hammer_fifo_head_t;
40043e7f 330typedef struct hammer_fifo_tail *hammer_fifo_tail_t;
9380c752 331
8750964d 332/*
47197d71 333 * Fifo header types.
8750964d 334 */
40043e7f
MD
335#define HAMMER_HEAD_TYPE_PAD (0x0040U|HAMMER_HEAD_FLAG_FREE)
336#define HAMMER_HEAD_TYPE_VOL 0x0041U /* Volume (dummy header) */
337#define HAMMER_HEAD_TYPE_BTREE 0x0042U /* B-Tree node */
338#define HAMMER_HEAD_TYPE_UNDO 0x0043U /* random UNDO information */
339#define HAMMER_HEAD_TYPE_DELETE 0x0044U /* record deletion */
340#define HAMMER_HEAD_TYPE_RECORD 0x0045U /* Filesystem record */
47197d71 341
40043e7f 342#define HAMMER_HEAD_FLAG_FREE 0x8000U /* Indicates object freed */
8750964d 343
47197d71 344#define HAMMER_HEAD_SIGNATURE 0xC84EU
40043e7f 345#define HAMMER_TAIL_SIGNATURE 0xC74FU
47197d71
MD
346
347/*
348 * Misc FIFO structures (except for the B-Tree node and hammer record)
349 */
350struct hammer_fifo_undo {
351 struct hammer_fifo_head head;
352 hammer_off_t undo_offset;
353 /* followed by data */
8750964d
MD
354};
355
47197d71 356typedef struct hammer_fifo_undo *hammer_fifo_undo_t;
8750964d 357
c60bb2c5 358/*
47197d71 359 * Volume header types
c60bb2c5 360 */
8750964d 361#define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
8750964d
MD
362#define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
363
364/*
365 * The B-Tree structures need hammer_fsbuf_head.
366 */
367#include "hammer_btree.h"
368
369/*
370 * HAMMER Volume header
371 *
372 * A HAMMER filesystem is built from any number of block devices, Each block
47197d71
MD
373 * device contains a volume header followed by however many buffers fit
374 * into the volume.
8750964d 375 *
47197d71
MD
376 * One of the volumes making up a HAMMER filesystem is the master, the
377 * rest are slaves. It does not have to be volume #0.
c60bb2c5 378 *
47197d71
MD
379 * The volume header takes up an entire 16K filesystem buffer and may
380 * represent up to 64KTB (65536 TB) of space.
a89aec1b
MD
381 *
382 * Special field notes:
383 *
384 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
385 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
47197d71 386 * vol_buf_beg - offset of the first buffer.
a89aec1b
MD
387 *
388 * The memory log area allows a kernel to cache new records and data
389 * in memory without allocating space in the actual filesystem to hold
390 * the records and data. In the event that a filesystem becomes full,
391 * any records remaining in memory can be flushed to the memory log
392 * area. This allows the kernel to immediately return success.
8750964d 393 */
8750964d 394
a89aec1b
MD
395#define HAMMER_BOOT_MINBYTES (32*1024)
396#define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
397#define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
398
399#define HAMMER_MEM_MINBYTES (256*1024)
400#define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
401#define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
402
8750964d 403struct hammer_volume_ondisk {
40043e7f
MD
404 u_int64_t vol_signature;/* Signature */
405
a89aec1b
MD
406 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
407 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
47197d71
MD
408 int64_t vol_buf_beg; /* byte offset of first buffer in volume */
409 int64_t vol_buf_end; /* byte offset of volume EOF (on buf bndry) */
8750964d
MD
410 int64_t vol_locked; /* reserved clusters are >= this offset */
411
412 uuid_t vol_fsid; /* identify filesystem */
413 uuid_t vol_fstype; /* identify filesystem type */
414 char vol_name[64]; /* Name of volume */
415
416 int32_t vol_no; /* volume number within filesystem */
417 int32_t vol_count; /* number of volumes making up FS */
418
419 u_int32_t vol_version; /* version control information */
9775c955 420 u_int32_t vol_reserved01;
8750964d
MD
421 u_int32_t vol_flags; /* volume flags */
422 u_int32_t vol_rootvol; /* which volume is the root volume? */
423
40043e7f 424 int32_t vol_reserved04;
47197d71 425 int32_t vol_reserved05;
8750964d
MD
426 u_int32_t vol_reserved06;
427 u_int32_t vol_reserved07;
428
fbc6e32a 429 int32_t vol_blocksize; /* for statfs only */
47197d71 430 int32_t vol_reserved08;
fbc6e32a
MD
431 int64_t vol_nblocks; /* total allocatable hammer bufs */
432
8750964d
MD
433 /*
434 * These fields are initialized and space is reserved in every
435 * volume making up a HAMMER filesytem, but only the master volume
436 * contains valid data.
437 */
c3be93f2
MD
438 int64_t vol0_stat_bigblocks; /* total bigblocks when fs is empty */
439 int64_t vol0_stat_freebigblocks;/* number of free bigblocks */
fbc6e32a
MD
440 int64_t vol0_stat_bytes; /* for statfs only */
441 int64_t vol0_stat_inodes; /* for statfs only */
b33e2cc0 442 int64_t vol0_stat_records; /* total records in filesystem */
47197d71
MD
443 hammer_off_t vol0_btree_root; /* B-Tree root */
444 hammer_tid_t vol0_next_tid; /* highest synchronized TID */
40043e7f
MD
445 hammer_tid_t vol0_next_seq; /* next SEQ no for undo */
446
447 /*
448 * Blockmaps for zones. Not all zones use a blockmap.
449 */
c3be93f2 450 struct hammer_blockmap vol0_blockmap[HAMMER_MAX_ZONES];
40043e7f 451
8750964d
MD
452};
453
8cd0a023
MD
454typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
455
c60bb2c5
MD
456#define HAMMER_VOLF_VALID 0x0001 /* valid entry */
457#define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
8cd0a023 458
8750964d 459/*
40043e7f
MD
460 * All HAMMER records have a common 64-byte base and a 32 byte extension,
461 * plus a possible data reference. The data reference can be in-band or
462 * out-of-band.
8750964d 463 */
40043e7f
MD
464
465#define HAMMER_RECORD_SIZE (64+32)
466
8750964d 467struct hammer_base_record {
40043e7f
MD
468 u_int32_t signature; /* record signature */
469 u_int32_t data_crc; /* data crc */
47197d71
MD
470 struct hammer_base_elm base; /* 40 byte base element */
471 hammer_off_t data_off; /* in-band or out-of-band */
472 int32_t data_len; /* size of data in bytes */
40043e7f 473 u_int32_t reserved02;
8750964d
MD
474};
475
c60bb2c5
MD
476/*
477 * Record types are fairly straightforward. The B-Tree includes the record
478 * type in its index sort.
479 *
480 * In particular please note that it is possible to create a pseudo-
481 * filesystem within a HAMMER filesystem by creating a special object
482 * type within a directory. Pseudo-filesystems are used as replication
483 * targets and even though they are built within a HAMMER filesystem they
484 * get their own obj_id space (and thus can serve as a replication target)
485 * and look like a mount point to the system.
8cd0a023
MD
486 *
487 * Inter-cluster records are special-cased in the B-Tree. These records
488 * are referenced from a B-Tree INTERNAL node, NOT A LEAF. This means
489 * that the element in the B-Tree node is actually a boundary element whos
490 * base element fields, including rec_type, reflect the boundary, NOT
491 * the inter-cluster record type.
492 *
493 * HAMMER_RECTYPE_CLUSTER - only set in the actual inter-cluster record,
494 * not set in the left or right boundary elements around the inter-cluster
495 * reference of an internal node in the B-Tree (because doing so would
496 * interfere with the boundary tests).
b3deaf57
MD
497 *
498 * NOTE: hammer_ip_delete_range_all() deletes all record types greater
499 * then HAMMER_RECTYPE_INODE.
c60bb2c5 500 */
8750964d 501#define HAMMER_RECTYPE_UNKNOWN 0
66325755 502#define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
8750964d 503#define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
c60bb2c5 504#define HAMMER_RECTYPE_PSEUDO_INODE 2 /* pseudo filesysem */
8cd0a023 505#define HAMMER_RECTYPE_CLUSTER 3 /* inter-cluster reference */
66325755
MD
506#define HAMMER_RECTYPE_DATA 0x10
507#define HAMMER_RECTYPE_DIRENTRY 0x11
508#define HAMMER_RECTYPE_DB 0x12
509#define HAMMER_RECTYPE_EXT 0x13 /* ext attributes */
7a04d74f
MD
510#define HAMMER_RECTYPE_FIX 0x14 /* fixed attribute */
511
512#define HAMMER_FIXKEY_SYMLINK 1
8750964d 513
66325755 514#define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */
8750964d
MD
515#define HAMMER_OBJTYPE_DIRECTORY 1
516#define HAMMER_OBJTYPE_REGFILE 2
517#define HAMMER_OBJTYPE_DBFILE 3
518#define HAMMER_OBJTYPE_FIFO 4
c60bb2c5
MD
519#define HAMMER_OBJTYPE_CDEV 5
520#define HAMMER_OBJTYPE_BDEV 6
521#define HAMMER_OBJTYPE_SOFTLINK 7
522#define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
523
8750964d
MD
524/*
525 * A HAMMER inode record.
526 *
527 * This forms the basis for a filesystem object. obj_id is the inode number,
528 * key1 represents the pseudo filesystem id for security partitioning
529 * (preventing cross-links and/or restricting a NFS export and specifying the
530 * security policy), and key2 represents the data retention policy id.
531 *
532 * Inode numbers are 64 bit quantities which uniquely identify a filesystem
533 * object for the ENTIRE life of the filesystem, even after the object has
534 * been deleted. For all intents and purposes inode numbers are simply
535 * allocated by incrementing a sequence space.
536 *
537 * There is an important distinction between the data stored in the inode
538 * record and the record's data reference. The record references a
539 * hammer_inode_data structure but the filesystem object size and hard link
540 * count is stored in the inode record itself. This allows multiple inodes
541 * to share the same hammer_inode_data structure. This is possible because
542 * any modifications will lay out new data. The HAMMER implementation need
543 * not use the data-sharing ability when laying down new records.
544 *
545 * A HAMMER inode is subject to the same historical storage requirements
546 * as any other record. In particular any change in filesystem or hard link
547 * count will lay down a new inode record when the filesystem is synced to
548 * disk. This can lead to a lot of junk records which get cleaned up by
549 * the data retention policy.
550 *
551 * The ino_atime and ino_mtime fields are a special case. Modifications to
552 * these fields do NOT lay down a new record by default, though the values
553 * are effectively frozen for snapshots which access historical versions
554 * of the inode record due to other operations. This means that atime will
555 * not necessarily be accurate in snapshots, backups, or mirrors. mtime
556 * will be accurate in backups and mirrors since it can be regenerated from
557 * the mirroring stream.
558 *
559 * Because nlinks is historically retained the hardlink count will be
560 * accurate when accessing a HAMMER filesystem snapshot.
561 */
562struct hammer_inode_record {
563 struct hammer_base_record base;
564 u_int64_t ino_atime; /* last access time (not historical) */
565 u_int64_t ino_mtime; /* last modified time (not historical) */
566 u_int64_t ino_size; /* filesystem object size */
567 u_int64_t ino_nlinks; /* hard links */
568};
569
570/*
571 * Data records specify the entire contents of a regular file object,
572 * including attributes. Small amounts of data can theoretically be
573 * embedded in the record itself but the use of this ability verses using
574 * an out-of-band data reference depends on the implementation.
575 */
576struct hammer_data_record {
577 struct hammer_base_record base;
40043e7f 578 char data[32];
8750964d
MD
579};
580
581/*
582 * A directory entry specifies the HAMMER filesystem object id, a copy of
583 * the file type, and file name (either embedded or as out-of-band data).
584 * If the file name is short enough to fit into den_name[] (including a
585 * terminating nul) then it will be embedded in the record, otherwise it
586 * is stored out-of-band. The base record's data reference always points
587 * to the nul-terminated filename regardless.
588 *
589 * Directory entries are indexed with a 128 bit namekey rather then an
590 * offset. A portion of the namekey is an iterator or randomizer to deal
591 * with collisions.
66325755 592 *
6b4f890b
MD
593 * NOTE: base.base.obj_type holds the filesystem object type of obj_id,
594 * e.g. a den_type equivalent.
595 *
596 * NOTE: den_name / the filename data reference is NOT terminated with \0.
66325755 597 *
8750964d
MD
598 */
599struct hammer_entry_record {
600 struct hammer_base_record base;
601 u_int64_t obj_id; /* object being referenced */
602 u_int64_t reserved01;
40043e7f 603 char name[16];
f3b0f382
MD
604};
605
8750964d
MD
606/*
607 * Hammer rollup record
608 */
c60bb2c5 609union hammer_record_ondisk {
8750964d 610 struct hammer_base_record base;
8750964d
MD
611 struct hammer_inode_record inode;
612 struct hammer_data_record data;
613 struct hammer_entry_record entry;
614};
615
c60bb2c5 616typedef union hammer_record_ondisk *hammer_record_ondisk_t;
8750964d 617
8750964d
MD
618/*
619 * HAMMER UNIX Attribute data
620 *
621 * The data reference in a HAMMER inode record points to this structure. Any
622 * modifications to the contents of this structure will result in a record
623 * replacement operation.
624 *
66325755
MD
625 * short_data_off allows a small amount of data to be embedded in the
626 * hammer_inode_data structure. HAMMER typically uses this to represent
627 * up to 64 bytes of data, or to hold symlinks. Remember that allocations
628 * are in powers of 2 so 64, 192, 448, or 960 bytes of embedded data is
629 * support (64+64, 64+192, 64+448 64+960).
630 *
631 * parent_obj_id is only valid for directories (which cannot be hard-linked),
632 * and specifies the parent directory obj_id. This field will also be set
633 * for non-directory inodes as a recovery aid, but can wind up specifying
634 * stale information. However, since object id's are not reused, the worse
635 * that happens is that the recovery code is unable to use it.
8750964d
MD
636 */
637struct hammer_inode_data {
638 u_int16_t version; /* inode data version */
639 u_int16_t mode; /* basic unix permissions */
640 u_int32_t uflags; /* chflags */
7a04d74f
MD
641 u_int32_t rmajor; /* used by device nodes */
642 u_int32_t rminor; /* used by device nodes */
66325755
MD
643 u_int64_t ctime;
644 u_int64_t parent_obj_id;/* parent directory obj_id */
8750964d
MD
645 uuid_t uid;
646 uuid_t gid;
8cd0a023 647 /* XXX device, softlink extension */
8750964d
MD
648};
649
650#define HAMMER_INODE_DATA_VERSION 1
651
7a04d74f
MD
652#define HAMMER_OBJID_ROOT 1
653
c60bb2c5
MD
654/*
655 * Rollup various structures embedded as record data
656 */
427e5fc6 657union hammer_data_ondisk {
c60bb2c5
MD
658 struct hammer_inode_data inode;
659};
660
7dc57964 661#endif