Break out the scan info structure's support routines so external code
[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 *
a89aec1b 34 * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.7 2007/11/20 07:16:28 dillon Exp $
8750964d
MD
35 */
36
37#ifndef _SYS_UUID_H_
38#include <sys/uuid.h>
39#endif
40
41/*
42 * The structures below represent the on-disk format for a HAMMER
43 * filesystem. Note that all fields for on-disk structures are naturally
44 * aligned. The host endian format is used - compatibility is possible
45 * if the implementation detects reversed endian and adjusts data accordingly.
46 *
47 * Most of HAMMER revolves around the concept of an object identifier. An
48 * obj_id is a 64 bit quantity which uniquely identifies a filesystem object
49 * FOR THE ENTIRE LIFE OF THE FILESYSTEM. This uniqueness allows backups
50 * and mirrors to retain varying amounts of filesystem history by removing
51 * any possibility of conflict through identifier reuse.
52 *
53 * A HAMMER filesystem may spam multiple volumes.
54 *
55 * A HAMMER filesystem uses a 16K filesystem buffer size. All filesystem
c60bb2c5
MD
56 * I/O is done in multiples of 16K. Most buffer-sized headers such as those
57 * used by volumes, super-clusters, clusters, and basic filesystem buffers
58 * use fixed-sized A-lists which are heavily dependant on HAMMER_BUFSIZE.
8750964d
MD
59 */
60#define HAMMER_BUFSIZE 16384
61#define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1)
62
63/*
64 * Hammer transction ids are 64 bit unsigned integers and are usually
65 * synchronized with the time of day in nanoseconds.
66 */
67typedef u_int64_t hammer_tid_t;
68
66325755
MD
69#define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL
70
8750964d
MD
71/*
72 * Most HAMMER data structures are embedded in 16K filesystem buffers.
73 * All filesystem buffers except those designated as pure-data buffers
74 * contain this 128-byte header.
75 *
76 * This structure contains an embedded A-List used to manage space within
77 * the filesystem buffer. It is not used by volume or cluster header
78 * buffers, or by pure-data buffers. The granularity is variable and
79 * depends on the type of filesystem buffer. BLKSIZE is just a minimum.
80 */
81
82#define HAMMER_FSBUF_HEAD_SIZE 128
83#define HAMMER_FSBUF_MAXBLKS 256
9775c955 84#define HAMMER_FSBUF_BLKMASK (HAMMER_FSBUF_MAXBLKS - 1)
c60bb2c5 85#define HAMMER_FSBUF_METAELMS HAMMER_ALIST_METAELMS_256_1LYR /* 11 */
8750964d
MD
86
87struct hammer_fsbuf_head {
88 u_int64_t buf_type;
89 u_int32_t buf_crc;
90 u_int32_t buf_reserved07;
c60bb2c5 91 u_int32_t reserved[6];
8750964d
MD
92 struct hammer_almeta buf_almeta[HAMMER_FSBUF_METAELMS];
93};
94
95typedef struct hammer_fsbuf_head *hammer_fsbuf_head_t;
96
c60bb2c5
MD
97/*
98 * Note: Pure-data buffers contain pure-data and have no buf_type.
99 * Piecemeal data buffers do have a header and use HAMMER_FSBUF_DATA.
100 */
8750964d 101#define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
c60bb2c5 102#define HAMMER_FSBUF_SUPERCL 0xC8414D52C3555052ULL /* HAMRSUPR */
8750964d
MD
103#define HAMMER_FSBUF_CLUSTER 0xC8414D52C34C5553ULL /* HAMRCLUS */
104#define HAMMER_FSBUF_RECORDS 0xC8414D52D2454353ULL /* HAMRRECS */
105#define HAMMER_FSBUF_BTREE 0xC8414D52C2545245ULL /* HAMRBTRE */
106#define HAMMER_FSBUF_DATA 0xC8414D52C4415441ULL /* HAMRDATA */
107
108#define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
109
110/*
111 * The B-Tree structures need hammer_fsbuf_head.
112 */
113#include "hammer_btree.h"
114
115/*
116 * HAMMER Volume header
117 *
118 * A HAMMER filesystem is built from any number of block devices, Each block
c60bb2c5
MD
119 * device contains a volume header followed by however many super-clusters
120 * and clusters fit into the volume. Clusters cannot be migrated but the
121 * data they contain can, so HAMMER can use a truncated cluster for any
122 * extra space at the end of the volume.
8750964d
MD
123 *
124 * The volume containing the root cluster is designated as the master volume.
125 * The root cluster designation can be moved to any volume.
126 *
127 * The volume header takes up an entire 16K filesystem buffer and includes
c60bb2c5
MD
128 * a one or two-layered A-list to manage the clusters making up the volume.
129 * A volume containing up to 32768 clusters (2TB) can be managed with a
130 * single-layered A-list. A two-layer A-list is capable of managing up
131 * to 16384 super-clusters with each super-cluster containing 32768 clusters
132 * (32768 TB per volume total). The number of volumes is limited to 32768
133 * but it only takes 512 to fill out a 64 bit address space so for all
134 * intents and purposes the filesystem has no limits.
135 *
136 * cluster addressing within a volume depends on whether a single or
137 * duel-layer A-list is used. If a duel-layer A-list is used a 16K
138 * super-cluster buffer is needed for every 16384 clusters in the volume.
139 * However, because the A-list's hinting is grouped in multiples of 16
140 * we group 16 super-cluster buffers together (starting just after the
141 * volume header), followed by 16384x16 clusters, and repeat.
142 *
143 * NOTE: A 32768-element single-layer and 16384-element duel-layer A-list
144 * is the same size.
a89aec1b
MD
145 *
146 * Special field notes:
147 *
148 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
149 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
150 * vol_clo_beg - offset of cluster #0 in volume
151 *
152 * The memory log area allows a kernel to cache new records and data
153 * in memory without allocating space in the actual filesystem to hold
154 * the records and data. In the event that a filesystem becomes full,
155 * any records remaining in memory can be flushed to the memory log
156 * area. This allows the kernel to immediately return success.
8750964d 157 */
c60bb2c5
MD
158#define HAMMER_VOL_MAXCLUSTERS 32768 /* 1-layer */
159#define HAMMER_VOL_MAXSUPERCLUSTERS 16384 /* 2-layer */
160#define HAMMER_VOL_SUPERCLUSTER_GROUP 16
161#define HAMMER_VOL_METAELMS_1LYR HAMMER_ALIST_METAELMS_32K_1LYR
162#define HAMMER_VOL_METAELMS_2LYR HAMMER_ALIST_METAELMS_16K_2LYR
8750964d 163
a89aec1b
MD
164#define HAMMER_BOOT_MINBYTES (32*1024)
165#define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
166#define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
167
168#define HAMMER_MEM_MINBYTES (256*1024)
169#define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
170#define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
171
8750964d
MD
172struct hammer_volume_ondisk {
173 struct hammer_fsbuf_head head;
a89aec1b
MD
174 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
175 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
176 int64_t vol_clo_beg; /* byte offset of first cl/supercl in volume */
177 int64_t vol_clo_end; /* byte offset of volume EOF */
8750964d
MD
178 int64_t vol_locked; /* reserved clusters are >= this offset */
179
180 uuid_t vol_fsid; /* identify filesystem */
181 uuid_t vol_fstype; /* identify filesystem type */
182 char vol_name[64]; /* Name of volume */
183
184 int32_t vol_no; /* volume number within filesystem */
185 int32_t vol_count; /* number of volumes making up FS */
186
187 u_int32_t vol_version; /* version control information */
9775c955 188 u_int32_t vol_reserved01;
8750964d
MD
189 u_int32_t vol_flags; /* volume flags */
190 u_int32_t vol_rootvol; /* which volume is the root volume? */
191
192 int32_t vol_clsize; /* cluster size (same for all volumes) */
9775c955 193 int32_t vol_nclusters;
8750964d
MD
194 u_int32_t vol_reserved06;
195 u_int32_t vol_reserved07;
196
27ea2398
MD
197 int32_t vol_stat_blocksize; /* for statfs only */
198 int64_t vol_stat_bytes; /* for statfs only */
199 int64_t vol_stat_inodes; /* for statfs only */
200
8750964d
MD
201 /*
202 * These fields are initialized and space is reserved in every
203 * volume making up a HAMMER filesytem, but only the master volume
204 * contains valid data.
205 */
427e5fc6
MD
206 int32_t vol0_root_clu_no; /* root cluster no (index) in rootvol */
207 hammer_tid_t vol0_root_clu_id; /* root cluster id */
8750964d
MD
208 hammer_tid_t vol0_nexttid; /* next TID */
209 u_int64_t vol0_recid; /* fs-wide record id allocator */
66325755 210 u_int64_t vol0_synchronized_rec_id; /* XXX */
8750964d
MD
211
212 char reserved[1024];
213
c60bb2c5
MD
214 /*
215 * Meta elements for the volume header's A-list, which is either a
216 * 1-layer A-list capable of managing 32768 clusters, or a 2-layer
217 * A-list capable of managing 16384 super-clusters (each of which
218 * can handle 32768 clusters).
219 */
220 union {
9775c955
MD
221 struct hammer_almeta super[HAMMER_VOL_METAELMS_2LYR];
222 struct hammer_almeta normal[HAMMER_VOL_METAELMS_1LYR];
c60bb2c5 223 } vol_almeta;
8750964d
MD
224 u_int32_t vol0_bitmap[1024];
225};
226
8cd0a023
MD
227typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
228
c60bb2c5
MD
229#define HAMMER_VOLF_VALID 0x0001 /* valid entry */
230#define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
427e5fc6 231#define HAMMER_VOLF_USINGSUPERCL 0x0004 /* using superclusters */
c60bb2c5
MD
232
233/*
234 * HAMMER Super-cluster header
235 *
236 * A super-cluster is used to increase the maximum size of a volume.
237 * HAMMER's volume header can manage up to 32768 direct clusters or
238 * 16384 super-clusters. Each super-cluster (which is basically just
239 * a 16K filesystem buffer) can manage up to 32768 clusters. So adding
240 * a super-cluster layer allows a HAMMER volume to be sized upwards of
241 * around 32768TB instead of 2TB.
242 *
243 * Any volume initially formatted to be over 32G reserves space for the layer
244 * but the layer is only enabled if the volume exceeds 2TB.
245 */
246#define HAMMER_SUPERCL_METAELMS HAMMER_ALIST_METAELMS_32K_1LYR
9775c955 247#define HAMMER_SCL_MAXCLUSTERS HAMMER_VOL_MAXCLUSTERS
c60bb2c5
MD
248
249struct hammer_supercl_ondisk {
250 struct hammer_fsbuf_head head;
251 uuid_t vol_fsid; /* identify filesystem - sanity check */
252 uuid_t vol_fstype; /* identify filesystem type - sanity check */
253 int32_t reserved[1024];
254
9775c955 255 struct hammer_almeta scl_meta[HAMMER_SUPERCL_METAELMS];
c60bb2c5 256};
8750964d 257
8cd0a023
MD
258typedef struct hammer_supercl_ondisk *hammer_supercl_ondisk_t;
259
8750964d
MD
260/*
261 * HAMMER Cluster header
262 *
c60bb2c5
MD
263 * A cluster is limited to 64MB and is made up of 4096 16K filesystem
264 * buffers. The cluster header contains four A-lists to manage these
265 * buffers.
266 *
267 * master_alist - This is a non-layered A-list which manages pure-data
268 * allocations and allocations on behalf of other A-lists.
269 *
270 * btree_alist - This is a layered A-list which manages filesystem buffers
271 * containing B-Tree nodes.
8750964d 272 *
c60bb2c5
MD
273 * record_alist - This is a layered A-list which manages filesystem buffers
274 * containing records.
275 *
276 * mdata_alist - This is a layered A-list which manages filesystem buffers
277 * containing piecemeal record data.
278 *
279 * General storage management works like this: All the A-lists except the
280 * master start in an all-allocated state. Now lets say you wish to allocate
281 * a B-Tree node out the btree_alist. If the allocation fails you allocate
282 * a pure data block out of master_alist and then free that block in
283 * btree_alist, thereby assigning more space to the btree_alist, and then
284 * retry your allocation out of the btree_alist. In the reverse direction,
285 * filesystem buffers can be garbage collected back to master_alist simply
286 * by doing whole-buffer allocations in btree_alist and then freeing the
287 * space in master_alist. The whole-buffer-allocation approach to garbage
288 * collection works because A-list allocations are always power-of-2 sized
289 * and aligned.
8750964d 290 */
c60bb2c5
MD
291#define HAMMER_CLU_MAXBUFFERS 4096
292#define HAMMER_CLU_MASTER_METAELMS HAMMER_ALIST_METAELMS_4K_1LYR
293#define HAMMER_CLU_SLAVE_METAELMS HAMMER_ALIST_METAELMS_4K_2LYR
9775c955 294#define HAMMER_CLU_MAXBYTES (HAMMER_CLU_MAXBUFFERS * HAMMER_BUFSIZE)
8750964d
MD
295
296struct hammer_cluster_ondisk {
297 struct hammer_fsbuf_head head;
298 uuid_t vol_fsid; /* identify filesystem - sanity check */
299 uuid_t vol_fstype; /* identify filesystem type - sanity check */
300
8750964d 301 hammer_tid_t clu_id; /* unique cluster self identification */
8cd0a023 302 hammer_tid_t clu_gen; /* generation number */
8750964d
MD
303 int32_t vol_no; /* cluster contained in volume (sanity) */
304 u_int32_t clu_flags; /* cluster flags */
305
306 int32_t clu_start; /* start of data (byte offset) */
307 int32_t clu_limit; /* end of data (byte offset) */
308 int32_t clu_no; /* cluster index in volume (sanity) */
309 u_int32_t clu_reserved03;
310
311 u_int32_t clu_reserved04;
312 u_int32_t clu_reserved05;
313 u_int32_t clu_reserved06;
314 u_int32_t clu_reserved07;
315
9775c955
MD
316 int32_t idx_data; /* data append point (element no) */
317 int32_t idx_index; /* index append point (element no) */
318 int32_t idx_record; /* record prepend point (element no) */
8750964d
MD
319 u_int32_t idx_reserved03;
320
321 /*
c60bb2c5 322 * Specify the range of information stored in this cluster as two
8cd0a023
MD
323 * btree elements. These elements match the left and right
324 * boundary elements in the internal B-Tree node of the parent
325 * cluster that points to the root of our cluster. Because these
326 * are boundary elements, the right boundary is range-NONinclusive.
8750964d 327 */
c60bb2c5
MD
328 struct hammer_base_elm clu_btree_beg;
329 struct hammer_base_elm clu_btree_end;
8750964d
MD
330
331 /*
c60bb2c5
MD
332 * The cluster's B-Tree root can change as a side effect of insertion
333 * and deletion operations so store an offset instead of embedding
8cd0a023
MD
334 * the root node. The parent_offset is stale if the generation number
335 * does not match.
336 *
337 * Parent linkages are explicit.
8750964d 338 */
c60bb2c5
MD
339 int32_t clu_btree_root;
340 int32_t clu_btree_parent_vol_no;
341 int32_t clu_btree_parent_clu_no;
8cd0a023
MD
342 int32_t clu_btree_parent_offset;
343 hammer_tid_t clu_btree_parent_clu_gen;
8750964d 344
8750964d 345 u_int64_t synchronized_rec_id;
8750964d 346
9775c955
MD
347 struct hammer_almeta clu_master_meta[HAMMER_CLU_MASTER_METAELMS];
348 struct hammer_almeta clu_btree_meta[HAMMER_CLU_SLAVE_METAELMS];
349 struct hammer_almeta clu_record_meta[HAMMER_CLU_SLAVE_METAELMS];
350 struct hammer_almeta clu_mdata_meta[HAMMER_CLU_SLAVE_METAELMS];
8750964d
MD
351};
352
8cd0a023
MD
353typedef struct hammer_cluster_ondisk *hammer_cluster_ondisk_t;
354
8750964d
MD
355/*
356 * HAMMER records are 96 byte entities encoded into 16K filesystem buffers.
357 * Each record has a 64 byte header and a 32 byte extension. 170 records
358 * fit into each buffer. Storage is managed by the buffer's A-List.
359 *
360 * Each record may have an explicit data reference to a block of data up
361 * to 2^31-1 bytes in size within the current cluster. Note that multiple
362 * records may share the same or overlapping data references.
363 */
364
365/*
366 * All HAMMER records have a common 64-byte base and a 32-byte extension.
367 *
368 * Many HAMMER record types reference out-of-band data within the cluster.
369 * This data can also be stored in-band in the record itself if it is small
370 * enough. Either way, (data_offset, data_len) points to it.
371 *
372 * Key comparison order: obj_id, rec_type, key, create_tid
373 */
374struct hammer_base_record {
427e5fc6
MD
375 /*
376 * 40 byte base element info - same base as used in B-Tree internal
377 * and leaf node element arrays.
378 *
379 * Fields: obj_id, key, create_tid, delete_tid, rec_type, obj_type,
380 * reserved07.
381 */
382 struct hammer_base_elm base; /* 00 base element info */
8750964d 383
8750964d
MD
384 int32_t data_len; /* 28 size of data (remainder zero-fill) */
385 u_int32_t data_crc; /* 2C data sanity check */
386 u_int64_t rec_id; /* 30 record id (iterator for recovery) */
427e5fc6
MD
387 int32_t data_offset; /* 38 cluster-relative data reference or 0 */
388 u_int32_t reserved07; /* 3C */
8750964d
MD
389 /* 40 */
390};
391
c60bb2c5
MD
392/*
393 * Record types are fairly straightforward. The B-Tree includes the record
394 * type in its index sort.
395 *
396 * In particular please note that it is possible to create a pseudo-
397 * filesystem within a HAMMER filesystem by creating a special object
398 * type within a directory. Pseudo-filesystems are used as replication
399 * targets and even though they are built within a HAMMER filesystem they
400 * get their own obj_id space (and thus can serve as a replication target)
401 * and look like a mount point to the system.
8cd0a023
MD
402 *
403 * Inter-cluster records are special-cased in the B-Tree. These records
404 * are referenced from a B-Tree INTERNAL node, NOT A LEAF. This means
405 * that the element in the B-Tree node is actually a boundary element whos
406 * base element fields, including rec_type, reflect the boundary, NOT
407 * the inter-cluster record type.
408 *
409 * HAMMER_RECTYPE_CLUSTER - only set in the actual inter-cluster record,
410 * not set in the left or right boundary elements around the inter-cluster
411 * reference of an internal node in the B-Tree (because doing so would
412 * interfere with the boundary tests).
c60bb2c5 413 */
8750964d 414#define HAMMER_RECTYPE_UNKNOWN 0
66325755 415#define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
8750964d 416#define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
c60bb2c5 417#define HAMMER_RECTYPE_PSEUDO_INODE 2 /* pseudo filesysem */
8cd0a023 418#define HAMMER_RECTYPE_CLUSTER 3 /* inter-cluster reference */
66325755
MD
419#define HAMMER_RECTYPE_DATA 0x10
420#define HAMMER_RECTYPE_DIRENTRY 0x11
421#define HAMMER_RECTYPE_DB 0x12
422#define HAMMER_RECTYPE_EXT 0x13 /* ext attributes */
8750964d 423
66325755 424#define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */
8750964d
MD
425#define HAMMER_OBJTYPE_DIRECTORY 1
426#define HAMMER_OBJTYPE_REGFILE 2
427#define HAMMER_OBJTYPE_DBFILE 3
428#define HAMMER_OBJTYPE_FIFO 4
c60bb2c5
MD
429#define HAMMER_OBJTYPE_CDEV 5
430#define HAMMER_OBJTYPE_BDEV 6
431#define HAMMER_OBJTYPE_SOFTLINK 7
432#define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
433
8750964d
MD
434/*
435 * Generic full-sized record
436 */
437struct hammer_generic_record {
438 struct hammer_base_record base;
439 char filler[32];
440};
441
442/*
443 * A HAMMER inode record.
444 *
445 * This forms the basis for a filesystem object. obj_id is the inode number,
446 * key1 represents the pseudo filesystem id for security partitioning
447 * (preventing cross-links and/or restricting a NFS export and specifying the
448 * security policy), and key2 represents the data retention policy id.
449 *
450 * Inode numbers are 64 bit quantities which uniquely identify a filesystem
451 * object for the ENTIRE life of the filesystem, even after the object has
452 * been deleted. For all intents and purposes inode numbers are simply
453 * allocated by incrementing a sequence space.
454 *
455 * There is an important distinction between the data stored in the inode
456 * record and the record's data reference. The record references a
457 * hammer_inode_data structure but the filesystem object size and hard link
458 * count is stored in the inode record itself. This allows multiple inodes
459 * to share the same hammer_inode_data structure. This is possible because
460 * any modifications will lay out new data. The HAMMER implementation need
461 * not use the data-sharing ability when laying down new records.
462 *
463 * A HAMMER inode is subject to the same historical storage requirements
464 * as any other record. In particular any change in filesystem or hard link
465 * count will lay down a new inode record when the filesystem is synced to
466 * disk. This can lead to a lot of junk records which get cleaned up by
467 * the data retention policy.
468 *
469 * The ino_atime and ino_mtime fields are a special case. Modifications to
470 * these fields do NOT lay down a new record by default, though the values
471 * are effectively frozen for snapshots which access historical versions
472 * of the inode record due to other operations. This means that atime will
473 * not necessarily be accurate in snapshots, backups, or mirrors. mtime
474 * will be accurate in backups and mirrors since it can be regenerated from
475 * the mirroring stream.
476 *
477 * Because nlinks is historically retained the hardlink count will be
478 * accurate when accessing a HAMMER filesystem snapshot.
479 */
480struct hammer_inode_record {
481 struct hammer_base_record base;
482 u_int64_t ino_atime; /* last access time (not historical) */
483 u_int64_t ino_mtime; /* last modified time (not historical) */
484 u_int64_t ino_size; /* filesystem object size */
485 u_int64_t ino_nlinks; /* hard links */
486};
487
488/*
489 * Data records specify the entire contents of a regular file object,
490 * including attributes. Small amounts of data can theoretically be
491 * embedded in the record itself but the use of this ability verses using
492 * an out-of-band data reference depends on the implementation.
493 */
494struct hammer_data_record {
495 struct hammer_base_record base;
496 char filler[32];
497};
498
499/*
500 * A directory entry specifies the HAMMER filesystem object id, a copy of
501 * the file type, and file name (either embedded or as out-of-band data).
502 * If the file name is short enough to fit into den_name[] (including a
503 * terminating nul) then it will be embedded in the record, otherwise it
504 * is stored out-of-band. The base record's data reference always points
505 * to the nul-terminated filename regardless.
506 *
507 * Directory entries are indexed with a 128 bit namekey rather then an
508 * offset. A portion of the namekey is an iterator or randomizer to deal
509 * with collisions.
66325755
MD
510 *
511 * Note that base.base.obj_type holds the filesystem object type of obj_id,
512 * e.g. a den_type equivalent.
513 *
8750964d
MD
514 */
515struct hammer_entry_record {
516 struct hammer_base_record base;
517 u_int64_t obj_id; /* object being referenced */
518 u_int64_t reserved01;
66325755 519 char den_name[16]; /* short file names fit in record */
8750964d
MD
520};
521
522/*
523 * Hammer rollup record
524 */
c60bb2c5 525union hammer_record_ondisk {
8750964d
MD
526 struct hammer_base_record base;
527 struct hammer_generic_record generic;
528 struct hammer_inode_record inode;
529 struct hammer_data_record data;
530 struct hammer_entry_record entry;
531};
532
c60bb2c5 533typedef union hammer_record_ondisk *hammer_record_ondisk_t;
8750964d
MD
534
535/*
536 * Filesystem buffer for records
537 */
538#define HAMMER_RECORD_NODES \
539 ((HAMMER_BUFSIZE - sizeof(struct hammer_fsbuf_head)) / \
c60bb2c5 540 sizeof(union hammer_record_ondisk))
8750964d
MD
541
542struct hammer_fsbuf_recs {
543 struct hammer_fsbuf_head head;
544 char unused[32];
c60bb2c5 545 union hammer_record_ondisk recs[HAMMER_RECORD_NODES];
8750964d
MD
546};
547
548/*
549 * Filesystem buffer for piecemeal data. Note that this does not apply
550 * to dedicated pure-data buffers as such buffers do not have a header.
551 */
552
553#define HAMMER_DATA_SIZE (HAMMER_BUFSIZE - sizeof(struct hammer_fsbuf_head))
554#define HAMMER_DATA_BLKSIZE 64
9775c955 555#define HAMMER_DATA_BLKMASK (HAMMER_DATA_BLKSIZE-1)
8750964d
MD
556#define HAMMER_DATA_NODES (HAMMER_DATA_SIZE / HAMMER_DATA_BLKSIZE)
557
558struct hammer_fsbuf_data {
559 struct hammer_fsbuf_head head;
560 u_int8_t data[HAMMER_DATA_NODES][HAMMER_DATA_BLKSIZE];
561};
562
9775c955
MD
563/*
564 * Filesystem buffer rollup
565 */
566union hammer_fsbuf_ondisk {
567 struct hammer_fsbuf_head head;
568 struct hammer_fsbuf_btree btree;
569 struct hammer_fsbuf_recs record;
570 struct hammer_fsbuf_data data;
571};
572
573typedef union hammer_fsbuf_ondisk *hammer_fsbuf_ondisk_t;
8750964d
MD
574
575/*
576 * HAMMER UNIX Attribute data
577 *
578 * The data reference in a HAMMER inode record points to this structure. Any
579 * modifications to the contents of this structure will result in a record
580 * replacement operation.
581 *
582 * state_sum allows a filesystem object to be validated to a degree by
583 * generating a checksum of all of its pieces (in no particular order) and
584 * checking it against this field.
66325755
MD
585 *
586 * short_data_off allows a small amount of data to be embedded in the
587 * hammer_inode_data structure. HAMMER typically uses this to represent
588 * up to 64 bytes of data, or to hold symlinks. Remember that allocations
589 * are in powers of 2 so 64, 192, 448, or 960 bytes of embedded data is
590 * support (64+64, 64+192, 64+448 64+960).
591 *
592 * parent_obj_id is only valid for directories (which cannot be hard-linked),
593 * and specifies the parent directory obj_id. This field will also be set
594 * for non-directory inodes as a recovery aid, but can wind up specifying
595 * stale information. However, since object id's are not reused, the worse
596 * that happens is that the recovery code is unable to use it.
8750964d
MD
597 */
598struct hammer_inode_data {
599 u_int16_t version; /* inode data version */
600 u_int16_t mode; /* basic unix permissions */
601 u_int32_t uflags; /* chflags */
66325755
MD
602 u_int16_t short_data_off; /* degenerate data case */
603 u_int16_t short_data_len;
604 u_int32_t state_sum;
605 u_int64_t ctime;
606 u_int64_t parent_obj_id;/* parent directory obj_id */
8750964d
MD
607 uuid_t uid;
608 uuid_t gid;
8cd0a023 609 /* XXX device, softlink extension */
8750964d
MD
610};
611
612#define HAMMER_INODE_DATA_VERSION 1
613
c60bb2c5
MD
614/*
615 * Rollup various structures embedded as record data
616 */
427e5fc6 617union hammer_data_ondisk {
c60bb2c5
MD
618 struct hammer_inode_data inode;
619};
620