Use correct extension.
[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 *
b33e2cc0 34 * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.19 2008/01/24 02:14:45 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)
4d75d829 62#define HAMMER_MAXDATA (256*1024)
8750964d
MD
63
64/*
65 * Hammer transction ids are 64 bit unsigned integers and are usually
66 * synchronized with the time of day in nanoseconds.
67 */
68typedef u_int64_t hammer_tid_t;
69
b33e2cc0
MD
70#define HAMMER_MIN_TID 0ULL
71#define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL
72#define HAMMER_MIN_KEY -0x8000000000000000LL
73#define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL
74#define HAMMER_MIN_OBJID HAMMER_MIN_KEY
75#define HAMMER_MAX_OBJID HAMMER_MAX_KEY
76#define HAMMER_MIN_RECTYPE 0x0U
77#define HAMMER_MAX_RECTYPE 0xFFFFU
66325755 78
8750964d
MD
79/*
80 * Most HAMMER data structures are embedded in 16K filesystem buffers.
81 * All filesystem buffers except those designated as pure-data buffers
82 * contain this 128-byte header.
83 *
84 * This structure contains an embedded A-List used to manage space within
85 * the filesystem buffer. It is not used by volume or cluster header
86 * buffers, or by pure-data buffers. The granularity is variable and
87 * depends on the type of filesystem buffer. BLKSIZE is just a minimum.
88 */
89
90#define HAMMER_FSBUF_HEAD_SIZE 128
91#define HAMMER_FSBUF_MAXBLKS 256
9775c955 92#define HAMMER_FSBUF_BLKMASK (HAMMER_FSBUF_MAXBLKS - 1)
c60bb2c5 93#define HAMMER_FSBUF_METAELMS HAMMER_ALIST_METAELMS_256_1LYR /* 11 */
8750964d
MD
94
95struct hammer_fsbuf_head {
96 u_int64_t buf_type;
97 u_int32_t buf_crc;
98 u_int32_t buf_reserved07;
c60bb2c5 99 u_int32_t reserved[6];
8750964d
MD
100 struct hammer_almeta buf_almeta[HAMMER_FSBUF_METAELMS];
101};
102
103typedef struct hammer_fsbuf_head *hammer_fsbuf_head_t;
104
c60bb2c5
MD
105/*
106 * Note: Pure-data buffers contain pure-data and have no buf_type.
107 * Piecemeal data buffers do have a header and use HAMMER_FSBUF_DATA.
108 */
8750964d 109#define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
c60bb2c5 110#define HAMMER_FSBUF_SUPERCL 0xC8414D52C3555052ULL /* HAMRSUPR */
8750964d
MD
111#define HAMMER_FSBUF_CLUSTER 0xC8414D52C34C5553ULL /* HAMRCLUS */
112#define HAMMER_FSBUF_RECORDS 0xC8414D52D2454353ULL /* HAMRRECS */
113#define HAMMER_FSBUF_BTREE 0xC8414D52C2545245ULL /* HAMRBTRE */
114#define HAMMER_FSBUF_DATA 0xC8414D52C4415441ULL /* HAMRDATA */
115
116#define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
117
118/*
119 * The B-Tree structures need hammer_fsbuf_head.
120 */
121#include "hammer_btree.h"
122
123/*
124 * HAMMER Volume header
125 *
126 * A HAMMER filesystem is built from any number of block devices, Each block
c60bb2c5
MD
127 * device contains a volume header followed by however many super-clusters
128 * and clusters fit into the volume. Clusters cannot be migrated but the
129 * data they contain can, so HAMMER can use a truncated cluster for any
130 * extra space at the end of the volume.
8750964d
MD
131 *
132 * The volume containing the root cluster is designated as the master volume.
133 * The root cluster designation can be moved to any volume.
134 *
135 * The volume header takes up an entire 16K filesystem buffer and includes
c60bb2c5
MD
136 * a one or two-layered A-list to manage the clusters making up the volume.
137 * A volume containing up to 32768 clusters (2TB) can be managed with a
138 * single-layered A-list. A two-layer A-list is capable of managing up
7f7c1f84
MD
139 * to 4096 super-clusters with each super-cluster containing 32768 clusters
140 * (8192 TB per volume total). The number of volumes is limited to 32768
c60bb2c5
MD
141 * but it only takes 512 to fill out a 64 bit address space so for all
142 * intents and purposes the filesystem has no limits.
143 *
144 * cluster addressing within a volume depends on whether a single or
145 * duel-layer A-list is used. If a duel-layer A-list is used a 16K
7f7c1f84 146 * super-cluster buffer is needed for every 32768 clusters in the volume.
c60bb2c5
MD
147 * However, because the A-list's hinting is grouped in multiples of 16
148 * we group 16 super-cluster buffers together (starting just after the
149 * volume header), followed by 16384x16 clusters, and repeat.
150 *
7f7c1f84
MD
151 * The number of super-clusters is limited to 4096 because the A-list's
152 * master radix is stored as a 32 bit signed quantity which will overflow
153 * if more then 4096*32768 elements is specified. XXX
154 *
c60bb2c5
MD
155 * NOTE: A 32768-element single-layer and 16384-element duel-layer A-list
156 * is the same size.
a89aec1b
MD
157 *
158 * Special field notes:
159 *
160 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
161 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
162 * vol_clo_beg - offset of cluster #0 in volume
163 *
164 * The memory log area allows a kernel to cache new records and data
165 * in memory without allocating space in the actual filesystem to hold
166 * the records and data. In the event that a filesystem becomes full,
167 * any records remaining in memory can be flushed to the memory log
168 * area. This allows the kernel to immediately return success.
8750964d 169 */
c60bb2c5 170#define HAMMER_VOL_MAXCLUSTERS 32768 /* 1-layer */
7f7c1f84 171#define HAMMER_VOL_MAXSUPERCLUSTERS 4096 /* 2-layer */
c60bb2c5
MD
172#define HAMMER_VOL_SUPERCLUSTER_GROUP 16
173#define HAMMER_VOL_METAELMS_1LYR HAMMER_ALIST_METAELMS_32K_1LYR
174#define HAMMER_VOL_METAELMS_2LYR HAMMER_ALIST_METAELMS_16K_2LYR
8750964d 175
a89aec1b
MD
176#define HAMMER_BOOT_MINBYTES (32*1024)
177#define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
178#define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
179
180#define HAMMER_MEM_MINBYTES (256*1024)
181#define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
182#define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
183
8750964d
MD
184struct hammer_volume_ondisk {
185 struct hammer_fsbuf_head head;
a89aec1b
MD
186 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
187 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
188 int64_t vol_clo_beg; /* byte offset of first cl/supercl in volume */
189 int64_t vol_clo_end; /* byte offset of volume EOF */
8750964d
MD
190 int64_t vol_locked; /* reserved clusters are >= this offset */
191
192 uuid_t vol_fsid; /* identify filesystem */
193 uuid_t vol_fstype; /* identify filesystem type */
194 char vol_name[64]; /* Name of volume */
195
196 int32_t vol_no; /* volume number within filesystem */
197 int32_t vol_count; /* number of volumes making up FS */
198
199 u_int32_t vol_version; /* version control information */
9775c955 200 u_int32_t vol_reserved01;
8750964d
MD
201 u_int32_t vol_flags; /* volume flags */
202 u_int32_t vol_rootvol; /* which volume is the root volume? */
203
204 int32_t vol_clsize; /* cluster size (same for all volumes) */
9775c955 205 int32_t vol_nclusters;
8750964d
MD
206 u_int32_t vol_reserved06;
207 u_int32_t vol_reserved07;
208
fbc6e32a
MD
209 int32_t vol_blocksize; /* for statfs only */
210 int64_t vol_nblocks; /* total allocatable hammer bufs */
211
212 /*
213 * This statistical information can get out of sync after a crash
214 * and is recovered slowly.
215 */
27ea2398 216 int64_t vol_stat_bytes; /* for statfs only */
fbc6e32a
MD
217 int64_t unused08; /* for statfs only */
218 int64_t vol_stat_data_bufs; /* hammer bufs allocated to data */
219 int64_t vol_stat_rec_bufs; /* hammer bufs allocated to records */
220 int64_t vol_stat_idx_bufs; /* hammer bufs allocated to B-Tree */
27ea2398 221
8750964d
MD
222 /*
223 * These fields are initialized and space is reserved in every
224 * volume making up a HAMMER filesytem, but only the master volume
225 * contains valid data.
226 */
fbc6e32a
MD
227 int64_t vol0_stat_bytes; /* for statfs only */
228 int64_t vol0_stat_inodes; /* for statfs only */
b33e2cc0 229 int64_t vol0_stat_records; /* total records in filesystem */
fbc6e32a
MD
230 int64_t vol0_stat_data_bufs; /* hammer bufs allocated to data */
231 int64_t vol0_stat_rec_bufs; /* hammer bufs allocated to records */
232 int64_t vol0_stat_idx_bufs; /* hammer bufs allocated to B-Tree */
233
427e5fc6
MD
234 int32_t vol0_root_clu_no; /* root cluster no (index) in rootvol */
235 hammer_tid_t vol0_root_clu_id; /* root cluster id */
8750964d
MD
236 hammer_tid_t vol0_nexttid; /* next TID */
237 u_int64_t vol0_recid; /* fs-wide record id allocator */
66325755 238 u_int64_t vol0_synchronized_rec_id; /* XXX */
8750964d
MD
239
240 char reserved[1024];
241
c60bb2c5
MD
242 /*
243 * Meta elements for the volume header's A-list, which is either a
244 * 1-layer A-list capable of managing 32768 clusters, or a 2-layer
245 * A-list capable of managing 16384 super-clusters (each of which
246 * can handle 32768 clusters).
247 */
248 union {
9775c955
MD
249 struct hammer_almeta super[HAMMER_VOL_METAELMS_2LYR];
250 struct hammer_almeta normal[HAMMER_VOL_METAELMS_1LYR];
c60bb2c5 251 } vol_almeta;
8750964d
MD
252 u_int32_t vol0_bitmap[1024];
253};
254
8cd0a023
MD
255typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
256
c60bb2c5
MD
257#define HAMMER_VOLF_VALID 0x0001 /* valid entry */
258#define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
427e5fc6 259#define HAMMER_VOLF_USINGSUPERCL 0x0004 /* using superclusters */
c60bb2c5
MD
260
261/*
262 * HAMMER Super-cluster header
263 *
264 * A super-cluster is used to increase the maximum size of a volume.
265 * HAMMER's volume header can manage up to 32768 direct clusters or
266 * 16384 super-clusters. Each super-cluster (which is basically just
267 * a 16K filesystem buffer) can manage up to 32768 clusters. So adding
268 * a super-cluster layer allows a HAMMER volume to be sized upwards of
269 * around 32768TB instead of 2TB.
270 *
271 * Any volume initially formatted to be over 32G reserves space for the layer
272 * but the layer is only enabled if the volume exceeds 2TB.
273 */
274#define HAMMER_SUPERCL_METAELMS HAMMER_ALIST_METAELMS_32K_1LYR
9775c955 275#define HAMMER_SCL_MAXCLUSTERS HAMMER_VOL_MAXCLUSTERS
c60bb2c5
MD
276
277struct hammer_supercl_ondisk {
278 struct hammer_fsbuf_head head;
279 uuid_t vol_fsid; /* identify filesystem - sanity check */
280 uuid_t vol_fstype; /* identify filesystem type - sanity check */
281 int32_t reserved[1024];
282
9775c955 283 struct hammer_almeta scl_meta[HAMMER_SUPERCL_METAELMS];
c60bb2c5 284};
8750964d 285
8cd0a023
MD
286typedef struct hammer_supercl_ondisk *hammer_supercl_ondisk_t;
287
8750964d
MD
288/*
289 * HAMMER Cluster header
290 *
c60bb2c5
MD
291 * A cluster is limited to 64MB and is made up of 4096 16K filesystem
292 * buffers. The cluster header contains four A-lists to manage these
293 * buffers.
294 *
295 * master_alist - This is a non-layered A-list which manages pure-data
296 * allocations and allocations on behalf of other A-lists.
297 *
298 * btree_alist - This is a layered A-list which manages filesystem buffers
299 * containing B-Tree nodes.
8750964d 300 *
c60bb2c5
MD
301 * record_alist - This is a layered A-list which manages filesystem buffers
302 * containing records.
303 *
304 * mdata_alist - This is a layered A-list which manages filesystem buffers
305 * containing piecemeal record data.
306 *
307 * General storage management works like this: All the A-lists except the
308 * master start in an all-allocated state. Now lets say you wish to allocate
309 * a B-Tree node out the btree_alist. If the allocation fails you allocate
310 * a pure data block out of master_alist and then free that block in
311 * btree_alist, thereby assigning more space to the btree_alist, and then
312 * retry your allocation out of the btree_alist. In the reverse direction,
313 * filesystem buffers can be garbage collected back to master_alist simply
314 * by doing whole-buffer allocations in btree_alist and then freeing the
315 * space in master_alist. The whole-buffer-allocation approach to garbage
316 * collection works because A-list allocations are always power-of-2 sized
317 * and aligned.
8750964d 318 */
c60bb2c5
MD
319#define HAMMER_CLU_MAXBUFFERS 4096
320#define HAMMER_CLU_MASTER_METAELMS HAMMER_ALIST_METAELMS_4K_1LYR
321#define HAMMER_CLU_SLAVE_METAELMS HAMMER_ALIST_METAELMS_4K_2LYR
9775c955 322#define HAMMER_CLU_MAXBYTES (HAMMER_CLU_MAXBUFFERS * HAMMER_BUFSIZE)
8750964d
MD
323
324struct hammer_cluster_ondisk {
325 struct hammer_fsbuf_head head;
326 uuid_t vol_fsid; /* identify filesystem - sanity check */
327 uuid_t vol_fstype; /* identify filesystem type - sanity check */
328
8750964d 329 hammer_tid_t clu_id; /* unique cluster self identification */
8cd0a023 330 hammer_tid_t clu_gen; /* generation number */
8750964d
MD
331 int32_t vol_no; /* cluster contained in volume (sanity) */
332 u_int32_t clu_flags; /* cluster flags */
333
334 int32_t clu_start; /* start of data (byte offset) */
335 int32_t clu_limit; /* end of data (byte offset) */
336 int32_t clu_no; /* cluster index in volume (sanity) */
337 u_int32_t clu_reserved03;
338
339 u_int32_t clu_reserved04;
340 u_int32_t clu_reserved05;
341 u_int32_t clu_reserved06;
342 u_int32_t clu_reserved07;
343
fbc6e32a 344 /*
4d75d829
MD
345 * These fields are mostly heuristics to aid in locality of
346 * reference allocations.
fbc6e32a 347 */
9775c955
MD
348 int32_t idx_data; /* data append point (element no) */
349 int32_t idx_index; /* index append point (element no) */
350 int32_t idx_record; /* record prepend point (element no) */
c0ade690 351 int32_t idx_ldata; /* large block data append pt (buf_no) */
8750964d 352
fbc6e32a
MD
353 /*
354 * These fields can become out of sync after a filesystem crash
355 * and are cleaned up in the background. They are used for
356 * reporting only.
b33e2cc0
MD
357 *
358 * NOTE: stat_records counts a spike as two records even though there
359 * is only one record. This is done so we can properly calculate
360 * the worst-case space needed to hold the B-Tree.
fbc6e32a
MD
361 */
362 int32_t stat_inodes; /* number of inodes in cluster */
b33e2cc0 363 int32_t stat_records; /* number of records in cluster */
fbc6e32a
MD
364 int32_t stat_data_bufs; /* hammer bufs allocated to data */
365 int32_t stat_rec_bufs; /* hammer bufs allocated to records */
366 int32_t stat_idx_bufs; /* hammer bufs allocated to B-Tree */
367
8750964d 368 /*
c60bb2c5 369 * Specify the range of information stored in this cluster as two
8cd0a023
MD
370 * btree elements. These elements match the left and right
371 * boundary elements in the internal B-Tree node of the parent
372 * cluster that points to the root of our cluster. Because these
373 * are boundary elements, the right boundary is range-NONinclusive.
8750964d 374 */
c60bb2c5
MD
375 struct hammer_base_elm clu_btree_beg;
376 struct hammer_base_elm clu_btree_end;
8750964d
MD
377
378 /*
c60bb2c5
MD
379 * The cluster's B-Tree root can change as a side effect of insertion
380 * and deletion operations so store an offset instead of embedding
8cd0a023
MD
381 * the root node. The parent_offset is stale if the generation number
382 * does not match.
383 *
384 * Parent linkages are explicit.
8750964d 385 */
c60bb2c5
MD
386 int32_t clu_btree_root;
387 int32_t clu_btree_parent_vol_no;
388 int32_t clu_btree_parent_clu_no;
8cd0a023
MD
389 int32_t clu_btree_parent_offset;
390 hammer_tid_t clu_btree_parent_clu_gen;
8750964d 391
4d75d829
MD
392 /*
393 * The synchronized record id is used for recovery purposes.
b33e2cc0
MD
394 *
395 * For recovery purposes, only clu_record_meta[] is recovered.
396 * The remaining a-list's are regenerated based on the records
397 * found.
4d75d829 398 */
eaeff70d 399 u_int64_t synchronized_rec_id;
4d75d829 400 u_int32_t reserved16[510];
8750964d 401
9775c955
MD
402 struct hammer_almeta clu_master_meta[HAMMER_CLU_MASTER_METAELMS];
403 struct hammer_almeta clu_btree_meta[HAMMER_CLU_SLAVE_METAELMS];
404 struct hammer_almeta clu_record_meta[HAMMER_CLU_SLAVE_METAELMS];
405 struct hammer_almeta clu_mdata_meta[HAMMER_CLU_SLAVE_METAELMS];
8750964d
MD
406};
407
8cd0a023
MD
408typedef struct hammer_cluster_ondisk *hammer_cluster_ondisk_t;
409
4d75d829
MD
410/*
411 * Cluster clu_flags
412 *
413 * OPEN - A cluster is marked open and synchronized to disk prior to any
414 * modifications being made to either the cluster header or any cluster
415 * buffers. If initial access to a cluster finds this flag set, the
416 * cluster is recovered before any further operations are performed on it.
417 */
fbc6e32a
MD
418#define HAMMER_CLUF_OPEN 0x0001 /* cluster is dirty */
419
8750964d
MD
420/*
421 * HAMMER records are 96 byte entities encoded into 16K filesystem buffers.
422 * Each record has a 64 byte header and a 32 byte extension. 170 records
423 * fit into each buffer. Storage is managed by the buffer's A-List.
424 *
425 * Each record may have an explicit data reference to a block of data up
426 * to 2^31-1 bytes in size within the current cluster. Note that multiple
427 * records may share the same or overlapping data references.
428 */
429
430/*
431 * All HAMMER records have a common 64-byte base and a 32-byte extension.
432 *
433 * Many HAMMER record types reference out-of-band data within the cluster.
434 * This data can also be stored in-band in the record itself if it is small
435 * enough. Either way, (data_offset, data_len) points to it.
436 *
d5530d22 437 * Key comparison order: obj_id, rec_type, key, delete_tid
8750964d
MD
438 */
439struct hammer_base_record {
427e5fc6
MD
440 /*
441 * 40 byte base element info - same base as used in B-Tree internal
442 * and leaf node element arrays.
443 *
444 * Fields: obj_id, key, create_tid, delete_tid, rec_type, obj_type,
445 * reserved07.
446 */
447 struct hammer_base_elm base; /* 00 base element info */
8750964d 448
8750964d
MD
449 int32_t data_len; /* 28 size of data (remainder zero-fill) */
450 u_int32_t data_crc; /* 2C data sanity check */
451 u_int64_t rec_id; /* 30 record id (iterator for recovery) */
427e5fc6
MD
452 int32_t data_offset; /* 38 cluster-relative data reference or 0 */
453 u_int32_t reserved07; /* 3C */
8750964d
MD
454 /* 40 */
455};
456
c60bb2c5
MD
457/*
458 * Record types are fairly straightforward. The B-Tree includes the record
459 * type in its index sort.
460 *
461 * In particular please note that it is possible to create a pseudo-
462 * filesystem within a HAMMER filesystem by creating a special object
463 * type within a directory. Pseudo-filesystems are used as replication
464 * targets and even though they are built within a HAMMER filesystem they
465 * get their own obj_id space (and thus can serve as a replication target)
466 * and look like a mount point to the system.
8cd0a023
MD
467 *
468 * Inter-cluster records are special-cased in the B-Tree. These records
469 * are referenced from a B-Tree INTERNAL node, NOT A LEAF. This means
470 * that the element in the B-Tree node is actually a boundary element whos
471 * base element fields, including rec_type, reflect the boundary, NOT
472 * the inter-cluster record type.
473 *
474 * HAMMER_RECTYPE_CLUSTER - only set in the actual inter-cluster record,
475 * not set in the left or right boundary elements around the inter-cluster
476 * reference of an internal node in the B-Tree (because doing so would
477 * interfere with the boundary tests).
b3deaf57
MD
478 *
479 * NOTE: hammer_ip_delete_range_all() deletes all record types greater
480 * then HAMMER_RECTYPE_INODE.
c60bb2c5 481 */
8750964d 482#define HAMMER_RECTYPE_UNKNOWN 0
66325755 483#define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
8750964d 484#define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
c60bb2c5 485#define HAMMER_RECTYPE_PSEUDO_INODE 2 /* pseudo filesysem */
8cd0a023 486#define HAMMER_RECTYPE_CLUSTER 3 /* inter-cluster reference */
66325755
MD
487#define HAMMER_RECTYPE_DATA 0x10
488#define HAMMER_RECTYPE_DIRENTRY 0x11
489#define HAMMER_RECTYPE_DB 0x12
490#define HAMMER_RECTYPE_EXT 0x13 /* ext attributes */
7a04d74f
MD
491#define HAMMER_RECTYPE_FIX 0x14 /* fixed attribute */
492
493#define HAMMER_FIXKEY_SYMLINK 1
8750964d 494
66325755 495#define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */
8750964d
MD
496#define HAMMER_OBJTYPE_DIRECTORY 1
497#define HAMMER_OBJTYPE_REGFILE 2
498#define HAMMER_OBJTYPE_DBFILE 3
499#define HAMMER_OBJTYPE_FIFO 4
c60bb2c5
MD
500#define HAMMER_OBJTYPE_CDEV 5
501#define HAMMER_OBJTYPE_BDEV 6
502#define HAMMER_OBJTYPE_SOFTLINK 7
503#define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
504
8750964d
MD
505/*
506 * Generic full-sized record
507 */
508struct hammer_generic_record {
509 struct hammer_base_record base;
510 char filler[32];
511};
512
513/*
514 * A HAMMER inode record.
515 *
516 * This forms the basis for a filesystem object. obj_id is the inode number,
517 * key1 represents the pseudo filesystem id for security partitioning
518 * (preventing cross-links and/or restricting a NFS export and specifying the
519 * security policy), and key2 represents the data retention policy id.
520 *
521 * Inode numbers are 64 bit quantities which uniquely identify a filesystem
522 * object for the ENTIRE life of the filesystem, even after the object has
523 * been deleted. For all intents and purposes inode numbers are simply
524 * allocated by incrementing a sequence space.
525 *
526 * There is an important distinction between the data stored in the inode
527 * record and the record's data reference. The record references a
528 * hammer_inode_data structure but the filesystem object size and hard link
529 * count is stored in the inode record itself. This allows multiple inodes
530 * to share the same hammer_inode_data structure. This is possible because
531 * any modifications will lay out new data. The HAMMER implementation need
532 * not use the data-sharing ability when laying down new records.
533 *
534 * A HAMMER inode is subject to the same historical storage requirements
535 * as any other record. In particular any change in filesystem or hard link
536 * count will lay down a new inode record when the filesystem is synced to
537 * disk. This can lead to a lot of junk records which get cleaned up by
538 * the data retention policy.
539 *
540 * The ino_atime and ino_mtime fields are a special case. Modifications to
541 * these fields do NOT lay down a new record by default, though the values
542 * are effectively frozen for snapshots which access historical versions
543 * of the inode record due to other operations. This means that atime will
544 * not necessarily be accurate in snapshots, backups, or mirrors. mtime
545 * will be accurate in backups and mirrors since it can be regenerated from
546 * the mirroring stream.
547 *
548 * Because nlinks is historically retained the hardlink count will be
549 * accurate when accessing a HAMMER filesystem snapshot.
550 */
551struct hammer_inode_record {
552 struct hammer_base_record base;
553 u_int64_t ino_atime; /* last access time (not historical) */
554 u_int64_t ino_mtime; /* last modified time (not historical) */
555 u_int64_t ino_size; /* filesystem object size */
556 u_int64_t ino_nlinks; /* hard links */
557};
558
559/*
560 * Data records specify the entire contents of a regular file object,
561 * including attributes. Small amounts of data can theoretically be
562 * embedded in the record itself but the use of this ability verses using
563 * an out-of-band data reference depends on the implementation.
564 */
565struct hammer_data_record {
566 struct hammer_base_record base;
567 char filler[32];
568};
569
570/*
571 * A directory entry specifies the HAMMER filesystem object id, a copy of
572 * the file type, and file name (either embedded or as out-of-band data).
573 * If the file name is short enough to fit into den_name[] (including a
574 * terminating nul) then it will be embedded in the record, otherwise it
575 * is stored out-of-band. The base record's data reference always points
576 * to the nul-terminated filename regardless.
577 *
578 * Directory entries are indexed with a 128 bit namekey rather then an
579 * offset. A portion of the namekey is an iterator or randomizer to deal
580 * with collisions.
66325755 581 *
6b4f890b
MD
582 * NOTE: base.base.obj_type holds the filesystem object type of obj_id,
583 * e.g. a den_type equivalent.
584 *
585 * NOTE: den_name / the filename data reference is NOT terminated with \0.
66325755 586 *
8750964d
MD
587 */
588struct hammer_entry_record {
589 struct hammer_base_record base;
590 u_int64_t obj_id; /* object being referenced */
591 u_int64_t reserved01;
66325755 592 char den_name[16]; /* short file names fit in record */
8750964d
MD
593};
594
f3b0f382
MD
595/*
596 * Spike record
597 */
598struct hammer_spike_record {
599 struct hammer_base_record base;
600 int32_t clu_no;
601 int32_t vol_no;
602 hammer_tid_t clu_id;
603 char reserved[16];
604};
605
8750964d
MD
606/*
607 * Hammer rollup record
608 */
c60bb2c5 609union hammer_record_ondisk {
8750964d
MD
610 struct hammer_base_record base;
611 struct hammer_generic_record generic;
f3b0f382 612 struct hammer_spike_record spike;
8750964d
MD
613 struct hammer_inode_record inode;
614 struct hammer_data_record data;
615 struct hammer_entry_record entry;
616};
617
c60bb2c5 618typedef union hammer_record_ondisk *hammer_record_ondisk_t;
8750964d
MD
619
620/*
621 * Filesystem buffer for records
622 */
623#define HAMMER_RECORD_NODES \
7f7c1f84 624 ((HAMMER_BUFSIZE - sizeof(struct hammer_fsbuf_head) - 32) / \
c60bb2c5 625 sizeof(union hammer_record_ondisk))
8750964d 626
195c19a1
MD
627#define HAMMER_RECORD_SIZE (64+32)
628
8750964d
MD
629struct hammer_fsbuf_recs {
630 struct hammer_fsbuf_head head;
631 char unused[32];
c60bb2c5 632 union hammer_record_ondisk recs[HAMMER_RECORD_NODES];
8750964d
MD
633};
634
635/*
636 * Filesystem buffer for piecemeal data. Note that this does not apply
637 * to dedicated pure-data buffers as such buffers do not have a header.
638 */
639
640#define HAMMER_DATA_SIZE (HAMMER_BUFSIZE - sizeof(struct hammer_fsbuf_head))
641#define HAMMER_DATA_BLKSIZE 64
9775c955 642#define HAMMER_DATA_BLKMASK (HAMMER_DATA_BLKSIZE-1)
8750964d
MD
643#define HAMMER_DATA_NODES (HAMMER_DATA_SIZE / HAMMER_DATA_BLKSIZE)
644
645struct hammer_fsbuf_data {
646 struct hammer_fsbuf_head head;
647 u_int8_t data[HAMMER_DATA_NODES][HAMMER_DATA_BLKSIZE];
648};
649
9775c955
MD
650/*
651 * Filesystem buffer rollup
652 */
653union hammer_fsbuf_ondisk {
654 struct hammer_fsbuf_head head;
655 struct hammer_fsbuf_btree btree;
656 struct hammer_fsbuf_recs record;
657 struct hammer_fsbuf_data data;
658};
659
660typedef union hammer_fsbuf_ondisk *hammer_fsbuf_ondisk_t;
8750964d
MD
661
662/*
663 * HAMMER UNIX Attribute data
664 *
665 * The data reference in a HAMMER inode record points to this structure. Any
666 * modifications to the contents of this structure will result in a record
667 * replacement operation.
668 *
66325755
MD
669 * short_data_off allows a small amount of data to be embedded in the
670 * hammer_inode_data structure. HAMMER typically uses this to represent
671 * up to 64 bytes of data, or to hold symlinks. Remember that allocations
672 * are in powers of 2 so 64, 192, 448, or 960 bytes of embedded data is
673 * support (64+64, 64+192, 64+448 64+960).
674 *
675 * parent_obj_id is only valid for directories (which cannot be hard-linked),
676 * and specifies the parent directory obj_id. This field will also be set
677 * for non-directory inodes as a recovery aid, but can wind up specifying
678 * stale information. However, since object id's are not reused, the worse
679 * that happens is that the recovery code is unable to use it.
8750964d
MD
680 */
681struct hammer_inode_data {
682 u_int16_t version; /* inode data version */
683 u_int16_t mode; /* basic unix permissions */
684 u_int32_t uflags; /* chflags */
7a04d74f
MD
685 u_int32_t rmajor; /* used by device nodes */
686 u_int32_t rminor; /* used by device nodes */
66325755
MD
687 u_int64_t ctime;
688 u_int64_t parent_obj_id;/* parent directory obj_id */
8750964d
MD
689 uuid_t uid;
690 uuid_t gid;
8cd0a023 691 /* XXX device, softlink extension */
8750964d
MD
692};
693
694#define HAMMER_INODE_DATA_VERSION 1
695
7a04d74f
MD
696#define HAMMER_OBJID_ROOT 1
697
c60bb2c5
MD
698/*
699 * Rollup various structures embedded as record data
700 */
427e5fc6 701union hammer_data_ondisk {
c60bb2c5
MD
702 struct hammer_inode_data inode;
703};
704