Don't use special CWARNFLAGS. Our default is good enough.
[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 *
7dc57964 34 * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.20 2008/02/04 08:33:17 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.
8750964d
MD
62 */
63#define HAMMER_BUFSIZE 16384
64#define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1)
4d75d829 65#define HAMMER_MAXDATA (256*1024)
8750964d
MD
66
67/*
68 * Hammer transction ids are 64 bit unsigned integers and are usually
69 * synchronized with the time of day in nanoseconds.
70 */
71typedef u_int64_t hammer_tid_t;
72
b33e2cc0
MD
73#define HAMMER_MIN_TID 0ULL
74#define HAMMER_MAX_TID 0xFFFFFFFFFFFFFFFFULL
75#define HAMMER_MIN_KEY -0x8000000000000000LL
76#define HAMMER_MAX_KEY 0x7FFFFFFFFFFFFFFFLL
77#define HAMMER_MIN_OBJID HAMMER_MIN_KEY
78#define HAMMER_MAX_OBJID HAMMER_MAX_KEY
79#define HAMMER_MIN_RECTYPE 0x0U
80#define HAMMER_MAX_RECTYPE 0xFFFFU
66325755 81
8750964d
MD
82/*
83 * Most HAMMER data structures are embedded in 16K filesystem buffers.
84 * All filesystem buffers except those designated as pure-data buffers
85 * contain this 128-byte header.
86 *
87 * This structure contains an embedded A-List used to manage space within
88 * the filesystem buffer. It is not used by volume or cluster header
89 * buffers, or by pure-data buffers. The granularity is variable and
90 * depends on the type of filesystem buffer. BLKSIZE is just a minimum.
91 */
92
93#define HAMMER_FSBUF_HEAD_SIZE 128
94#define HAMMER_FSBUF_MAXBLKS 256
9775c955 95#define HAMMER_FSBUF_BLKMASK (HAMMER_FSBUF_MAXBLKS - 1)
c60bb2c5 96#define HAMMER_FSBUF_METAELMS HAMMER_ALIST_METAELMS_256_1LYR /* 11 */
8750964d
MD
97
98struct hammer_fsbuf_head {
99 u_int64_t buf_type;
100 u_int32_t buf_crc;
101 u_int32_t buf_reserved07;
c60bb2c5 102 u_int32_t reserved[6];
8750964d
MD
103 struct hammer_almeta buf_almeta[HAMMER_FSBUF_METAELMS];
104};
105
106typedef struct hammer_fsbuf_head *hammer_fsbuf_head_t;
107
c60bb2c5
MD
108/*
109 * Note: Pure-data buffers contain pure-data and have no buf_type.
110 * Piecemeal data buffers do have a header and use HAMMER_FSBUF_DATA.
111 */
8750964d 112#define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
c60bb2c5 113#define HAMMER_FSBUF_SUPERCL 0xC8414D52C3555052ULL /* HAMRSUPR */
8750964d
MD
114#define HAMMER_FSBUF_CLUSTER 0xC8414D52C34C5553ULL /* HAMRCLUS */
115#define HAMMER_FSBUF_RECORDS 0xC8414D52D2454353ULL /* HAMRRECS */
116#define HAMMER_FSBUF_BTREE 0xC8414D52C2545245ULL /* HAMRBTRE */
117#define HAMMER_FSBUF_DATA 0xC8414D52C4415441ULL /* HAMRDATA */
118
119#define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
120
121/*
122 * The B-Tree structures need hammer_fsbuf_head.
123 */
124#include "hammer_btree.h"
125
126/*
127 * HAMMER Volume header
128 *
129 * A HAMMER filesystem is built from any number of block devices, Each block
c60bb2c5
MD
130 * device contains a volume header followed by however many super-clusters
131 * and clusters fit into the volume. Clusters cannot be migrated but the
132 * data they contain can, so HAMMER can use a truncated cluster for any
133 * extra space at the end of the volume.
8750964d
MD
134 *
135 * The volume containing the root cluster is designated as the master volume.
136 * The root cluster designation can be moved to any volume.
137 *
138 * The volume header takes up an entire 16K filesystem buffer and includes
c60bb2c5
MD
139 * a one or two-layered A-list to manage the clusters making up the volume.
140 * A volume containing up to 32768 clusters (2TB) can be managed with a
141 * single-layered A-list. A two-layer A-list is capable of managing up
7f7c1f84
MD
142 * to 4096 super-clusters with each super-cluster containing 32768 clusters
143 * (8192 TB per volume total). The number of volumes is limited to 32768
c60bb2c5
MD
144 * but it only takes 512 to fill out a 64 bit address space so for all
145 * intents and purposes the filesystem has no limits.
146 *
147 * cluster addressing within a volume depends on whether a single or
148 * duel-layer A-list is used. If a duel-layer A-list is used a 16K
7f7c1f84 149 * super-cluster buffer is needed for every 32768 clusters in the volume.
c60bb2c5
MD
150 * However, because the A-list's hinting is grouped in multiples of 16
151 * we group 16 super-cluster buffers together (starting just after the
152 * volume header), followed by 16384x16 clusters, and repeat.
153 *
7f7c1f84
MD
154 * The number of super-clusters is limited to 4096 because the A-list's
155 * master radix is stored as a 32 bit signed quantity which will overflow
156 * if more then 4096*32768 elements is specified. XXX
157 *
c60bb2c5
MD
158 * NOTE: A 32768-element single-layer and 16384-element duel-layer A-list
159 * is the same size.
a89aec1b
MD
160 *
161 * Special field notes:
162 *
163 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
164 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
165 * vol_clo_beg - offset of cluster #0 in volume
166 *
167 * The memory log area allows a kernel to cache new records and data
168 * in memory without allocating space in the actual filesystem to hold
169 * the records and data. In the event that a filesystem becomes full,
170 * any records remaining in memory can be flushed to the memory log
171 * area. This allows the kernel to immediately return success.
8750964d 172 */
c60bb2c5 173#define HAMMER_VOL_MAXCLUSTERS 32768 /* 1-layer */
7f7c1f84 174#define HAMMER_VOL_MAXSUPERCLUSTERS 4096 /* 2-layer */
c60bb2c5
MD
175#define HAMMER_VOL_SUPERCLUSTER_GROUP 16
176#define HAMMER_VOL_METAELMS_1LYR HAMMER_ALIST_METAELMS_32K_1LYR
177#define HAMMER_VOL_METAELMS_2LYR HAMMER_ALIST_METAELMS_16K_2LYR
8750964d 178
a89aec1b
MD
179#define HAMMER_BOOT_MINBYTES (32*1024)
180#define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
181#define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
182
183#define HAMMER_MEM_MINBYTES (256*1024)
184#define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
185#define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
186
8750964d
MD
187struct hammer_volume_ondisk {
188 struct hammer_fsbuf_head head;
a89aec1b
MD
189 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
190 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
191 int64_t vol_clo_beg; /* byte offset of first cl/supercl in volume */
192 int64_t vol_clo_end; /* byte offset of volume EOF */
8750964d
MD
193 int64_t vol_locked; /* reserved clusters are >= this offset */
194
195 uuid_t vol_fsid; /* identify filesystem */
196 uuid_t vol_fstype; /* identify filesystem type */
197 char vol_name[64]; /* Name of volume */
198
199 int32_t vol_no; /* volume number within filesystem */
200 int32_t vol_count; /* number of volumes making up FS */
201
202 u_int32_t vol_version; /* version control information */
9775c955 203 u_int32_t vol_reserved01;
8750964d
MD
204 u_int32_t vol_flags; /* volume flags */
205 u_int32_t vol_rootvol; /* which volume is the root volume? */
206
207 int32_t vol_clsize; /* cluster size (same for all volumes) */
9775c955 208 int32_t vol_nclusters;
8750964d
MD
209 u_int32_t vol_reserved06;
210 u_int32_t vol_reserved07;
211
fbc6e32a
MD
212 int32_t vol_blocksize; /* for statfs only */
213 int64_t vol_nblocks; /* total allocatable hammer bufs */
214
215 /*
216 * This statistical information can get out of sync after a crash
217 * and is recovered slowly.
218 */
27ea2398 219 int64_t vol_stat_bytes; /* for statfs only */
fbc6e32a
MD
220 int64_t unused08; /* for statfs only */
221 int64_t vol_stat_data_bufs; /* hammer bufs allocated to data */
222 int64_t vol_stat_rec_bufs; /* hammer bufs allocated to records */
223 int64_t vol_stat_idx_bufs; /* hammer bufs allocated to B-Tree */
27ea2398 224
8750964d
MD
225 /*
226 * These fields are initialized and space is reserved in every
227 * volume making up a HAMMER filesytem, but only the master volume
228 * contains valid data.
229 */
fbc6e32a
MD
230 int64_t vol0_stat_bytes; /* for statfs only */
231 int64_t vol0_stat_inodes; /* for statfs only */
b33e2cc0 232 int64_t vol0_stat_records; /* total records in filesystem */
fbc6e32a
MD
233 int64_t vol0_stat_data_bufs; /* hammer bufs allocated to data */
234 int64_t vol0_stat_rec_bufs; /* hammer bufs allocated to records */
235 int64_t vol0_stat_idx_bufs; /* hammer bufs allocated to B-Tree */
236
427e5fc6
MD
237 int32_t vol0_root_clu_no; /* root cluster no (index) in rootvol */
238 hammer_tid_t vol0_root_clu_id; /* root cluster id */
8750964d
MD
239 hammer_tid_t vol0_nexttid; /* next TID */
240 u_int64_t vol0_recid; /* fs-wide record id allocator */
66325755 241 u_int64_t vol0_synchronized_rec_id; /* XXX */
8750964d
MD
242
243 char reserved[1024];
244
c60bb2c5
MD
245 /*
246 * Meta elements for the volume header's A-list, which is either a
247 * 1-layer A-list capable of managing 32768 clusters, or a 2-layer
248 * A-list capable of managing 16384 super-clusters (each of which
249 * can handle 32768 clusters).
250 */
251 union {
9775c955
MD
252 struct hammer_almeta super[HAMMER_VOL_METAELMS_2LYR];
253 struct hammer_almeta normal[HAMMER_VOL_METAELMS_1LYR];
c60bb2c5 254 } vol_almeta;
8750964d
MD
255 u_int32_t vol0_bitmap[1024];
256};
257
8cd0a023
MD
258typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
259
c60bb2c5
MD
260#define HAMMER_VOLF_VALID 0x0001 /* valid entry */
261#define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
427e5fc6 262#define HAMMER_VOLF_USINGSUPERCL 0x0004 /* using superclusters */
c60bb2c5
MD
263
264/*
265 * HAMMER Super-cluster header
266 *
267 * A super-cluster is used to increase the maximum size of a volume.
268 * HAMMER's volume header can manage up to 32768 direct clusters or
269 * 16384 super-clusters. Each super-cluster (which is basically just
270 * a 16K filesystem buffer) can manage up to 32768 clusters. So adding
271 * a super-cluster layer allows a HAMMER volume to be sized upwards of
272 * around 32768TB instead of 2TB.
273 *
274 * Any volume initially formatted to be over 32G reserves space for the layer
275 * but the layer is only enabled if the volume exceeds 2TB.
276 */
277#define HAMMER_SUPERCL_METAELMS HAMMER_ALIST_METAELMS_32K_1LYR
9775c955 278#define HAMMER_SCL_MAXCLUSTERS HAMMER_VOL_MAXCLUSTERS
c60bb2c5
MD
279
280struct hammer_supercl_ondisk {
281 struct hammer_fsbuf_head head;
282 uuid_t vol_fsid; /* identify filesystem - sanity check */
283 uuid_t vol_fstype; /* identify filesystem type - sanity check */
284 int32_t reserved[1024];
285
9775c955 286 struct hammer_almeta scl_meta[HAMMER_SUPERCL_METAELMS];
c60bb2c5 287};
8750964d 288
8cd0a023
MD
289typedef struct hammer_supercl_ondisk *hammer_supercl_ondisk_t;
290
8750964d
MD
291/*
292 * HAMMER Cluster header
293 *
c60bb2c5
MD
294 * A cluster is limited to 64MB and is made up of 4096 16K filesystem
295 * buffers. The cluster header contains four A-lists to manage these
296 * buffers.
297 *
298 * master_alist - This is a non-layered A-list which manages pure-data
299 * allocations and allocations on behalf of other A-lists.
300 *
301 * btree_alist - This is a layered A-list which manages filesystem buffers
302 * containing B-Tree nodes.
8750964d 303 *
c60bb2c5
MD
304 * record_alist - This is a layered A-list which manages filesystem buffers
305 * containing records.
306 *
307 * mdata_alist - This is a layered A-list which manages filesystem buffers
308 * containing piecemeal record data.
309 *
310 * General storage management works like this: All the A-lists except the
311 * master start in an all-allocated state. Now lets say you wish to allocate
312 * a B-Tree node out the btree_alist. If the allocation fails you allocate
313 * a pure data block out of master_alist and then free that block in
314 * btree_alist, thereby assigning more space to the btree_alist, and then
315 * retry your allocation out of the btree_alist. In the reverse direction,
316 * filesystem buffers can be garbage collected back to master_alist simply
317 * by doing whole-buffer allocations in btree_alist and then freeing the
318 * space in master_alist. The whole-buffer-allocation approach to garbage
319 * collection works because A-list allocations are always power-of-2 sized
320 * and aligned.
8750964d 321 */
c60bb2c5
MD
322#define HAMMER_CLU_MAXBUFFERS 4096
323#define HAMMER_CLU_MASTER_METAELMS HAMMER_ALIST_METAELMS_4K_1LYR
324#define HAMMER_CLU_SLAVE_METAELMS HAMMER_ALIST_METAELMS_4K_2LYR
9775c955 325#define HAMMER_CLU_MAXBYTES (HAMMER_CLU_MAXBUFFERS * HAMMER_BUFSIZE)
8750964d
MD
326
327struct hammer_cluster_ondisk {
328 struct hammer_fsbuf_head head;
329 uuid_t vol_fsid; /* identify filesystem - sanity check */
330 uuid_t vol_fstype; /* identify filesystem type - sanity check */
331
8750964d 332 hammer_tid_t clu_id; /* unique cluster self identification */
8cd0a023 333 hammer_tid_t clu_gen; /* generation number */
8750964d
MD
334 int32_t vol_no; /* cluster contained in volume (sanity) */
335 u_int32_t clu_flags; /* cluster flags */
336
337 int32_t clu_start; /* start of data (byte offset) */
338 int32_t clu_limit; /* end of data (byte offset) */
339 int32_t clu_no; /* cluster index in volume (sanity) */
340 u_int32_t clu_reserved03;
341
342 u_int32_t clu_reserved04;
343 u_int32_t clu_reserved05;
344 u_int32_t clu_reserved06;
345 u_int32_t clu_reserved07;
346
fbc6e32a 347 /*
4d75d829
MD
348 * These fields are mostly heuristics to aid in locality of
349 * reference allocations.
fbc6e32a 350 */
9775c955
MD
351 int32_t idx_data; /* data append point (element no) */
352 int32_t idx_index; /* index append point (element no) */
353 int32_t idx_record; /* record prepend point (element no) */
c0ade690 354 int32_t idx_ldata; /* large block data append pt (buf_no) */
8750964d 355
fbc6e32a
MD
356 /*
357 * These fields can become out of sync after a filesystem crash
358 * and are cleaned up in the background. They are used for
359 * reporting only.
b33e2cc0
MD
360 *
361 * NOTE: stat_records counts a spike as two records even though there
362 * is only one record. This is done so we can properly calculate
363 * the worst-case space needed to hold the B-Tree.
fbc6e32a
MD
364 */
365 int32_t stat_inodes; /* number of inodes in cluster */
b33e2cc0 366 int32_t stat_records; /* number of records in cluster */
fbc6e32a
MD
367 int32_t stat_data_bufs; /* hammer bufs allocated to data */
368 int32_t stat_rec_bufs; /* hammer bufs allocated to records */
369 int32_t stat_idx_bufs; /* hammer bufs allocated to B-Tree */
370
8750964d 371 /*
c60bb2c5 372 * Specify the range of information stored in this cluster as two
8cd0a023
MD
373 * btree elements. These elements match the left and right
374 * boundary elements in the internal B-Tree node of the parent
375 * cluster that points to the root of our cluster. Because these
376 * are boundary elements, the right boundary is range-NONinclusive.
8750964d 377 */
c60bb2c5
MD
378 struct hammer_base_elm clu_btree_beg;
379 struct hammer_base_elm clu_btree_end;
8750964d
MD
380
381 /*
c60bb2c5
MD
382 * The cluster's B-Tree root can change as a side effect of insertion
383 * and deletion operations so store an offset instead of embedding
8cd0a023
MD
384 * the root node. The parent_offset is stale if the generation number
385 * does not match.
386 *
387 * Parent linkages are explicit.
8750964d 388 */
c60bb2c5
MD
389 int32_t clu_btree_root;
390 int32_t clu_btree_parent_vol_no;
391 int32_t clu_btree_parent_clu_no;
8cd0a023
MD
392 int32_t clu_btree_parent_offset;
393 hammer_tid_t clu_btree_parent_clu_gen;
8750964d 394
4d75d829
MD
395 /*
396 * The synchronized record id is used for recovery purposes.
b33e2cc0
MD
397 *
398 * For recovery purposes, only clu_record_meta[] is recovered.
399 * The remaining a-list's are regenerated based on the records
400 * found.
4d75d829 401 */
eaeff70d 402 u_int64_t synchronized_rec_id;
4d75d829 403 u_int32_t reserved16[510];
8750964d 404
9775c955
MD
405 struct hammer_almeta clu_master_meta[HAMMER_CLU_MASTER_METAELMS];
406 struct hammer_almeta clu_btree_meta[HAMMER_CLU_SLAVE_METAELMS];
407 struct hammer_almeta clu_record_meta[HAMMER_CLU_SLAVE_METAELMS];
408 struct hammer_almeta clu_mdata_meta[HAMMER_CLU_SLAVE_METAELMS];
8750964d
MD
409};
410
8cd0a023
MD
411typedef struct hammer_cluster_ondisk *hammer_cluster_ondisk_t;
412
4d75d829
MD
413/*
414 * Cluster clu_flags
415 *
416 * OPEN - A cluster is marked open and synchronized to disk prior to any
417 * modifications being made to either the cluster header or any cluster
418 * buffers. If initial access to a cluster finds this flag set, the
419 * cluster is recovered before any further operations are performed on it.
420 */
fbc6e32a
MD
421#define HAMMER_CLUF_OPEN 0x0001 /* cluster is dirty */
422
8750964d
MD
423/*
424 * HAMMER records are 96 byte entities encoded into 16K filesystem buffers.
425 * Each record has a 64 byte header and a 32 byte extension. 170 records
426 * fit into each buffer. Storage is managed by the buffer's A-List.
427 *
428 * Each record may have an explicit data reference to a block of data up
429 * to 2^31-1 bytes in size within the current cluster. Note that multiple
430 * records may share the same or overlapping data references.
431 */
432
433/*
434 * All HAMMER records have a common 64-byte base and a 32-byte extension.
435 *
436 * Many HAMMER record types reference out-of-band data within the cluster.
437 * This data can also be stored in-band in the record itself if it is small
438 * enough. Either way, (data_offset, data_len) points to it.
439 *
d5530d22 440 * Key comparison order: obj_id, rec_type, key, delete_tid
8750964d
MD
441 */
442struct hammer_base_record {
427e5fc6
MD
443 /*
444 * 40 byte base element info - same base as used in B-Tree internal
445 * and leaf node element arrays.
446 *
447 * Fields: obj_id, key, create_tid, delete_tid, rec_type, obj_type,
448 * reserved07.
449 */
450 struct hammer_base_elm base; /* 00 base element info */
8750964d 451
8750964d
MD
452 int32_t data_len; /* 28 size of data (remainder zero-fill) */
453 u_int32_t data_crc; /* 2C data sanity check */
454 u_int64_t rec_id; /* 30 record id (iterator for recovery) */
427e5fc6
MD
455 int32_t data_offset; /* 38 cluster-relative data reference or 0 */
456 u_int32_t reserved07; /* 3C */
8750964d
MD
457 /* 40 */
458};
459
c60bb2c5
MD
460/*
461 * Record types are fairly straightforward. The B-Tree includes the record
462 * type in its index sort.
463 *
464 * In particular please note that it is possible to create a pseudo-
465 * filesystem within a HAMMER filesystem by creating a special object
466 * type within a directory. Pseudo-filesystems are used as replication
467 * targets and even though they are built within a HAMMER filesystem they
468 * get their own obj_id space (and thus can serve as a replication target)
469 * and look like a mount point to the system.
8cd0a023
MD
470 *
471 * Inter-cluster records are special-cased in the B-Tree. These records
472 * are referenced from a B-Tree INTERNAL node, NOT A LEAF. This means
473 * that the element in the B-Tree node is actually a boundary element whos
474 * base element fields, including rec_type, reflect the boundary, NOT
475 * the inter-cluster record type.
476 *
477 * HAMMER_RECTYPE_CLUSTER - only set in the actual inter-cluster record,
478 * not set in the left or right boundary elements around the inter-cluster
479 * reference of an internal node in the B-Tree (because doing so would
480 * interfere with the boundary tests).
b3deaf57
MD
481 *
482 * NOTE: hammer_ip_delete_range_all() deletes all record types greater
483 * then HAMMER_RECTYPE_INODE.
c60bb2c5 484 */
8750964d 485#define HAMMER_RECTYPE_UNKNOWN 0
66325755 486#define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
8750964d 487#define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
c60bb2c5 488#define HAMMER_RECTYPE_PSEUDO_INODE 2 /* pseudo filesysem */
8cd0a023 489#define HAMMER_RECTYPE_CLUSTER 3 /* inter-cluster reference */
66325755
MD
490#define HAMMER_RECTYPE_DATA 0x10
491#define HAMMER_RECTYPE_DIRENTRY 0x11
492#define HAMMER_RECTYPE_DB 0x12
493#define HAMMER_RECTYPE_EXT 0x13 /* ext attributes */
7a04d74f
MD
494#define HAMMER_RECTYPE_FIX 0x14 /* fixed attribute */
495
496#define HAMMER_FIXKEY_SYMLINK 1
8750964d 497
66325755 498#define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */
8750964d
MD
499#define HAMMER_OBJTYPE_DIRECTORY 1
500#define HAMMER_OBJTYPE_REGFILE 2
501#define HAMMER_OBJTYPE_DBFILE 3
502#define HAMMER_OBJTYPE_FIFO 4
c60bb2c5
MD
503#define HAMMER_OBJTYPE_CDEV 5
504#define HAMMER_OBJTYPE_BDEV 6
505#define HAMMER_OBJTYPE_SOFTLINK 7
506#define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
507
8750964d
MD
508/*
509 * Generic full-sized record
510 */
511struct hammer_generic_record {
512 struct hammer_base_record base;
513 char filler[32];
514};
515
516/*
517 * A HAMMER inode record.
518 *
519 * This forms the basis for a filesystem object. obj_id is the inode number,
520 * key1 represents the pseudo filesystem id for security partitioning
521 * (preventing cross-links and/or restricting a NFS export and specifying the
522 * security policy), and key2 represents the data retention policy id.
523 *
524 * Inode numbers are 64 bit quantities which uniquely identify a filesystem
525 * object for the ENTIRE life of the filesystem, even after the object has
526 * been deleted. For all intents and purposes inode numbers are simply
527 * allocated by incrementing a sequence space.
528 *
529 * There is an important distinction between the data stored in the inode
530 * record and the record's data reference. The record references a
531 * hammer_inode_data structure but the filesystem object size and hard link
532 * count is stored in the inode record itself. This allows multiple inodes
533 * to share the same hammer_inode_data structure. This is possible because
534 * any modifications will lay out new data. The HAMMER implementation need
535 * not use the data-sharing ability when laying down new records.
536 *
537 * A HAMMER inode is subject to the same historical storage requirements
538 * as any other record. In particular any change in filesystem or hard link
539 * count will lay down a new inode record when the filesystem is synced to
540 * disk. This can lead to a lot of junk records which get cleaned up by
541 * the data retention policy.
542 *
543 * The ino_atime and ino_mtime fields are a special case. Modifications to
544 * these fields do NOT lay down a new record by default, though the values
545 * are effectively frozen for snapshots which access historical versions
546 * of the inode record due to other operations. This means that atime will
547 * not necessarily be accurate in snapshots, backups, or mirrors. mtime
548 * will be accurate in backups and mirrors since it can be regenerated from
549 * the mirroring stream.
550 *
551 * Because nlinks is historically retained the hardlink count will be
552 * accurate when accessing a HAMMER filesystem snapshot.
553 */
554struct hammer_inode_record {
555 struct hammer_base_record base;
556 u_int64_t ino_atime; /* last access time (not historical) */
557 u_int64_t ino_mtime; /* last modified time (not historical) */
558 u_int64_t ino_size; /* filesystem object size */
559 u_int64_t ino_nlinks; /* hard links */
560};
561
562/*
563 * Data records specify the entire contents of a regular file object,
564 * including attributes. Small amounts of data can theoretically be
565 * embedded in the record itself but the use of this ability verses using
566 * an out-of-band data reference depends on the implementation.
567 */
568struct hammer_data_record {
569 struct hammer_base_record base;
570 char filler[32];
571};
572
573/*
574 * A directory entry specifies the HAMMER filesystem object id, a copy of
575 * the file type, and file name (either embedded or as out-of-band data).
576 * If the file name is short enough to fit into den_name[] (including a
577 * terminating nul) then it will be embedded in the record, otherwise it
578 * is stored out-of-band. The base record's data reference always points
579 * to the nul-terminated filename regardless.
580 *
581 * Directory entries are indexed with a 128 bit namekey rather then an
582 * offset. A portion of the namekey is an iterator or randomizer to deal
583 * with collisions.
66325755 584 *
6b4f890b
MD
585 * NOTE: base.base.obj_type holds the filesystem object type of obj_id,
586 * e.g. a den_type equivalent.
587 *
588 * NOTE: den_name / the filename data reference is NOT terminated with \0.
66325755 589 *
8750964d
MD
590 */
591struct hammer_entry_record {
592 struct hammer_base_record base;
593 u_int64_t obj_id; /* object being referenced */
594 u_int64_t reserved01;
66325755 595 char den_name[16]; /* short file names fit in record */
8750964d
MD
596};
597
f3b0f382
MD
598/*
599 * Spike record
600 */
601struct hammer_spike_record {
602 struct hammer_base_record base;
603 int32_t clu_no;
604 int32_t vol_no;
605 hammer_tid_t clu_id;
606 char reserved[16];
607};
608
8750964d
MD
609/*
610 * Hammer rollup record
611 */
c60bb2c5 612union hammer_record_ondisk {
8750964d
MD
613 struct hammer_base_record base;
614 struct hammer_generic_record generic;
f3b0f382 615 struct hammer_spike_record spike;
8750964d
MD
616 struct hammer_inode_record inode;
617 struct hammer_data_record data;
618 struct hammer_entry_record entry;
619};
620
c60bb2c5 621typedef union hammer_record_ondisk *hammer_record_ondisk_t;
8750964d
MD
622
623/*
624 * Filesystem buffer for records
625 */
626#define HAMMER_RECORD_NODES \
7f7c1f84 627 ((HAMMER_BUFSIZE - sizeof(struct hammer_fsbuf_head) - 32) / \
c60bb2c5 628 sizeof(union hammer_record_ondisk))
8750964d 629
195c19a1
MD
630#define HAMMER_RECORD_SIZE (64+32)
631
8750964d
MD
632struct hammer_fsbuf_recs {
633 struct hammer_fsbuf_head head;
634 char unused[32];
c60bb2c5 635 union hammer_record_ondisk recs[HAMMER_RECORD_NODES];
8750964d
MD
636};
637
638/*
639 * Filesystem buffer for piecemeal data. Note that this does not apply
640 * to dedicated pure-data buffers as such buffers do not have a header.
641 */
642
643#define HAMMER_DATA_SIZE (HAMMER_BUFSIZE - sizeof(struct hammer_fsbuf_head))
644#define HAMMER_DATA_BLKSIZE 64
9775c955 645#define HAMMER_DATA_BLKMASK (HAMMER_DATA_BLKSIZE-1)
8750964d
MD
646#define HAMMER_DATA_NODES (HAMMER_DATA_SIZE / HAMMER_DATA_BLKSIZE)
647
648struct hammer_fsbuf_data {
649 struct hammer_fsbuf_head head;
650 u_int8_t data[HAMMER_DATA_NODES][HAMMER_DATA_BLKSIZE];
651};
652
9775c955
MD
653/*
654 * Filesystem buffer rollup
655 */
656union hammer_fsbuf_ondisk {
657 struct hammer_fsbuf_head head;
658 struct hammer_fsbuf_btree btree;
659 struct hammer_fsbuf_recs record;
660 struct hammer_fsbuf_data data;
661};
662
663typedef union hammer_fsbuf_ondisk *hammer_fsbuf_ondisk_t;
8750964d
MD
664
665/*
666 * HAMMER UNIX Attribute data
667 *
668 * The data reference in a HAMMER inode record points to this structure. Any
669 * modifications to the contents of this structure will result in a record
670 * replacement operation.
671 *
66325755
MD
672 * short_data_off allows a small amount of data to be embedded in the
673 * hammer_inode_data structure. HAMMER typically uses this to represent
674 * up to 64 bytes of data, or to hold symlinks. Remember that allocations
675 * are in powers of 2 so 64, 192, 448, or 960 bytes of embedded data is
676 * support (64+64, 64+192, 64+448 64+960).
677 *
678 * parent_obj_id is only valid for directories (which cannot be hard-linked),
679 * and specifies the parent directory obj_id. This field will also be set
680 * for non-directory inodes as a recovery aid, but can wind up specifying
681 * stale information. However, since object id's are not reused, the worse
682 * that happens is that the recovery code is unable to use it.
8750964d
MD
683 */
684struct hammer_inode_data {
685 u_int16_t version; /* inode data version */
686 u_int16_t mode; /* basic unix permissions */
687 u_int32_t uflags; /* chflags */
7a04d74f
MD
688 u_int32_t rmajor; /* used by device nodes */
689 u_int32_t rminor; /* used by device nodes */
66325755
MD
690 u_int64_t ctime;
691 u_int64_t parent_obj_id;/* parent directory obj_id */
8750964d
MD
692 uuid_t uid;
693 uuid_t gid;
8cd0a023 694 /* XXX device, softlink extension */
8750964d
MD
695};
696
697#define HAMMER_INODE_DATA_VERSION 1
698
7a04d74f
MD
699#define HAMMER_OBJID_ROOT 1
700
c60bb2c5
MD
701/*
702 * Rollup various structures embedded as record data
703 */
427e5fc6 704union hammer_data_ondisk {
c60bb2c5
MD
705 struct hammer_inode_data inode;
706};
707
7dc57964 708#endif