acxcontrol(8) is gone.
[dragonfly.git] / sys / vfs / hammer / hammer_disk.h
... / ...
CommitLineData
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 *
34 * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.21 2008/02/05 01:44:04 dillon Exp $
35 */
36
37#ifndef VFS_HAMMER_DISK_H_
38#define VFS_HAMMER_DISK_H_
39
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
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.
62 */
63#define HAMMER_BUFSIZE 16384
64#define HAMMER_BUFMASK (HAMMER_BUFSIZE - 1)
65#define HAMMER_MAXDATA (256*1024)
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
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
81
82/*
83 * Don't include the whole mess unless the caller also has included
84 * the hammer alist header. The ioctl code only needs hammer_tid_t.
85 */
86#ifdef HAMMER_ALIST_METAELMS_256_1LYR
87
88/*
89 * Most HAMMER data structures are embedded in 16K filesystem buffers.
90 * All filesystem buffers except those designated as pure-data buffers
91 * contain this 128-byte header.
92 *
93 * This structure contains an embedded A-List used to manage space within
94 * the filesystem buffer. It is not used by volume or cluster header
95 * buffers, or by pure-data buffers. The granularity is variable and
96 * depends on the type of filesystem buffer. BLKSIZE is just a minimum.
97 */
98
99#define HAMMER_FSBUF_HEAD_SIZE 128
100#define HAMMER_FSBUF_MAXBLKS 256
101#define HAMMER_FSBUF_BLKMASK (HAMMER_FSBUF_MAXBLKS - 1)
102#define HAMMER_FSBUF_METAELMS HAMMER_ALIST_METAELMS_256_1LYR /* 11 */
103
104struct hammer_fsbuf_head {
105 u_int64_t buf_type;
106 u_int32_t buf_crc;
107 u_int32_t buf_reserved07;
108 u_int32_t reserved[6];
109 struct hammer_almeta buf_almeta[HAMMER_FSBUF_METAELMS];
110};
111
112typedef struct hammer_fsbuf_head *hammer_fsbuf_head_t;
113
114/*
115 * Note: Pure-data buffers contain pure-data and have no buf_type.
116 * Piecemeal data buffers do have a header and use HAMMER_FSBUF_DATA.
117 */
118#define HAMMER_FSBUF_VOLUME 0xC8414D4DC5523031ULL /* HAMMER01 */
119#define HAMMER_FSBUF_SUPERCL 0xC8414D52C3555052ULL /* HAMRSUPR */
120#define HAMMER_FSBUF_CLUSTER 0xC8414D52C34C5553ULL /* HAMRCLUS */
121#define HAMMER_FSBUF_RECORDS 0xC8414D52D2454353ULL /* HAMRRECS */
122#define HAMMER_FSBUF_BTREE 0xC8414D52C2545245ULL /* HAMRBTRE */
123#define HAMMER_FSBUF_DATA 0xC8414D52C4415441ULL /* HAMRDATA */
124
125#define HAMMER_FSBUF_VOLUME_REV 0x313052C54D4D41C8ULL /* (reverse endian) */
126
127/*
128 * The B-Tree structures need hammer_fsbuf_head.
129 */
130#include "hammer_btree.h"
131
132/*
133 * HAMMER Volume header
134 *
135 * A HAMMER filesystem is built from any number of block devices, Each block
136 * device contains a volume header followed by however many super-clusters
137 * and clusters fit into the volume. Clusters cannot be migrated but the
138 * data they contain can, so HAMMER can use a truncated cluster for any
139 * extra space at the end of the volume.
140 *
141 * The volume containing the root cluster is designated as the master volume.
142 * The root cluster designation can be moved to any volume.
143 *
144 * The volume header takes up an entire 16K filesystem buffer and includes
145 * a one or two-layered A-list to manage the clusters making up the volume.
146 * A volume containing up to 32768 clusters (2TB) can be managed with a
147 * single-layered A-list. A two-layer A-list is capable of managing up
148 * to 4096 super-clusters with each super-cluster containing 32768 clusters
149 * (8192 TB per volume total). The number of volumes is limited to 32768
150 * but it only takes 512 to fill out a 64 bit address space so for all
151 * intents and purposes the filesystem has no limits.
152 *
153 * cluster addressing within a volume depends on whether a single or
154 * duel-layer A-list is used. If a duel-layer A-list is used a 16K
155 * super-cluster buffer is needed for every 32768 clusters in the volume.
156 * However, because the A-list's hinting is grouped in multiples of 16
157 * we group 16 super-cluster buffers together (starting just after the
158 * volume header), followed by 16384x16 clusters, and repeat.
159 *
160 * The number of super-clusters is limited to 4096 because the A-list's
161 * master radix is stored as a 32 bit signed quantity which will overflow
162 * if more then 4096*32768 elements is specified. XXX
163 *
164 * NOTE: A 32768-element single-layer and 16384-element duel-layer A-list
165 * is the same size.
166 *
167 * Special field notes:
168 *
169 * vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
170 * vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
171 * vol_clo_beg - offset of cluster #0 in volume
172 *
173 * The memory log area allows a kernel to cache new records and data
174 * in memory without allocating space in the actual filesystem to hold
175 * the records and data. In the event that a filesystem becomes full,
176 * any records remaining in memory can be flushed to the memory log
177 * area. This allows the kernel to immediately return success.
178 */
179#define HAMMER_VOL_MAXCLUSTERS 32768 /* 1-layer */
180#define HAMMER_VOL_MAXSUPERCLUSTERS 4096 /* 2-layer */
181#define HAMMER_VOL_SUPERCLUSTER_GROUP 16
182#define HAMMER_VOL_METAELMS_1LYR HAMMER_ALIST_METAELMS_32K_1LYR
183#define HAMMER_VOL_METAELMS_2LYR HAMMER_ALIST_METAELMS_16K_2LYR
184
185#define HAMMER_BOOT_MINBYTES (32*1024)
186#define HAMMER_BOOT_NOMBYTES (64LL*1024*1024)
187#define HAMMER_BOOT_MAXBYTES (256LL*1024*1024)
188
189#define HAMMER_MEM_MINBYTES (256*1024)
190#define HAMMER_MEM_NOMBYTES (1LL*1024*1024*1024)
191#define HAMMER_MEM_MAXBYTES (64LL*1024*1024*1024)
192
193struct hammer_volume_ondisk {
194 struct hammer_fsbuf_head head;
195 int64_t vol_bot_beg; /* byte offset of boot area or 0 */
196 int64_t vol_mem_beg; /* byte offset of memory log or 0 */
197 int64_t vol_clo_beg; /* byte offset of first cl/supercl in volume */
198 int64_t vol_clo_end; /* byte offset of volume EOF */
199 int64_t vol_locked; /* reserved clusters are >= this offset */
200
201 uuid_t vol_fsid; /* identify filesystem */
202 uuid_t vol_fstype; /* identify filesystem type */
203 char vol_name[64]; /* Name of volume */
204
205 int32_t vol_no; /* volume number within filesystem */
206 int32_t vol_count; /* number of volumes making up FS */
207
208 u_int32_t vol_version; /* version control information */
209 u_int32_t vol_reserved01;
210 u_int32_t vol_flags; /* volume flags */
211 u_int32_t vol_rootvol; /* which volume is the root volume? */
212
213 int32_t vol_clsize; /* cluster size (same for all volumes) */
214 int32_t vol_nclusters;
215 u_int32_t vol_reserved06;
216 u_int32_t vol_reserved07;
217
218 int32_t vol_blocksize; /* for statfs only */
219 int64_t vol_nblocks; /* total allocatable hammer bufs */
220
221 /*
222 * This statistical information can get out of sync after a crash
223 * and is recovered slowly.
224 */
225 int64_t vol_stat_bytes; /* for statfs only */
226 int64_t unused08; /* for statfs only */
227 int64_t vol_stat_data_bufs; /* hammer bufs allocated to data */
228 int64_t vol_stat_rec_bufs; /* hammer bufs allocated to records */
229 int64_t vol_stat_idx_bufs; /* hammer bufs allocated to B-Tree */
230
231 /*
232 * These fields are initialized and space is reserved in every
233 * volume making up a HAMMER filesytem, but only the master volume
234 * contains valid data.
235 */
236 int64_t vol0_stat_bytes; /* for statfs only */
237 int64_t vol0_stat_inodes; /* for statfs only */
238 int64_t vol0_stat_records; /* total records in filesystem */
239 int64_t vol0_stat_data_bufs; /* hammer bufs allocated to data */
240 int64_t vol0_stat_rec_bufs; /* hammer bufs allocated to records */
241 int64_t vol0_stat_idx_bufs; /* hammer bufs allocated to B-Tree */
242
243 int32_t vol0_root_clu_no; /* root cluster no (index) in rootvol */
244 hammer_tid_t vol0_root_clu_id; /* root cluster id */
245 hammer_tid_t vol0_nexttid; /* next TID */
246 u_int64_t vol0_recid; /* fs-wide record id allocator */
247 u_int64_t vol0_synchronized_rec_id; /* XXX */
248
249 char reserved[1024];
250
251 /*
252 * Meta elements for the volume header's A-list, which is either a
253 * 1-layer A-list capable of managing 32768 clusters, or a 2-layer
254 * A-list capable of managing 16384 super-clusters (each of which
255 * can handle 32768 clusters).
256 */
257 union {
258 struct hammer_almeta super[HAMMER_VOL_METAELMS_2LYR];
259 struct hammer_almeta normal[HAMMER_VOL_METAELMS_1LYR];
260 } vol_almeta;
261 u_int32_t vol0_bitmap[1024];
262};
263
264typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
265
266#define HAMMER_VOLF_VALID 0x0001 /* valid entry */
267#define HAMMER_VOLF_OPEN 0x0002 /* volume is open */
268#define HAMMER_VOLF_USINGSUPERCL 0x0004 /* using superclusters */
269
270/*
271 * HAMMER Super-cluster header
272 *
273 * A super-cluster is used to increase the maximum size of a volume.
274 * HAMMER's volume header can manage up to 32768 direct clusters or
275 * 16384 super-clusters. Each super-cluster (which is basically just
276 * a 16K filesystem buffer) can manage up to 32768 clusters. So adding
277 * a super-cluster layer allows a HAMMER volume to be sized upwards of
278 * around 32768TB instead of 2TB.
279 *
280 * Any volume initially formatted to be over 32G reserves space for the layer
281 * but the layer is only enabled if the volume exceeds 2TB.
282 */
283#define HAMMER_SUPERCL_METAELMS HAMMER_ALIST_METAELMS_32K_1LYR
284#define HAMMER_SCL_MAXCLUSTERS HAMMER_VOL_MAXCLUSTERS
285
286struct hammer_supercl_ondisk {
287 struct hammer_fsbuf_head head;
288 uuid_t vol_fsid; /* identify filesystem - sanity check */
289 uuid_t vol_fstype; /* identify filesystem type - sanity check */
290 int32_t reserved[1024];
291
292 struct hammer_almeta scl_meta[HAMMER_SUPERCL_METAELMS];
293};
294
295typedef struct hammer_supercl_ondisk *hammer_supercl_ondisk_t;
296
297/*
298 * HAMMER Cluster header
299 *
300 * A cluster is limited to 64MB and is made up of 4096 16K filesystem
301 * buffers. The cluster header contains four A-lists to manage these
302 * buffers.
303 *
304 * master_alist - This is a non-layered A-list which manages pure-data
305 * allocations and allocations on behalf of other A-lists.
306 *
307 * btree_alist - This is a layered A-list which manages filesystem buffers
308 * containing B-Tree nodes.
309 *
310 * record_alist - This is a layered A-list which manages filesystem buffers
311 * containing records.
312 *
313 * mdata_alist - This is a layered A-list which manages filesystem buffers
314 * containing piecemeal record data.
315 *
316 * General storage management works like this: All the A-lists except the
317 * master start in an all-allocated state. Now lets say you wish to allocate
318 * a B-Tree node out the btree_alist. If the allocation fails you allocate
319 * a pure data block out of master_alist and then free that block in
320 * btree_alist, thereby assigning more space to the btree_alist, and then
321 * retry your allocation out of the btree_alist. In the reverse direction,
322 * filesystem buffers can be garbage collected back to master_alist simply
323 * by doing whole-buffer allocations in btree_alist and then freeing the
324 * space in master_alist. The whole-buffer-allocation approach to garbage
325 * collection works because A-list allocations are always power-of-2 sized
326 * and aligned.
327 */
328#define HAMMER_CLU_MAXBUFFERS 4096
329#define HAMMER_CLU_MASTER_METAELMS HAMMER_ALIST_METAELMS_4K_1LYR
330#define HAMMER_CLU_SLAVE_METAELMS HAMMER_ALIST_METAELMS_4K_2LYR
331#define HAMMER_CLU_MAXBYTES (HAMMER_CLU_MAXBUFFERS * HAMMER_BUFSIZE)
332
333struct hammer_cluster_ondisk {
334 struct hammer_fsbuf_head head;
335 uuid_t vol_fsid; /* identify filesystem - sanity check */
336 uuid_t vol_fstype; /* identify filesystem type - sanity check */
337
338 hammer_tid_t clu_id; /* unique cluster self identification */
339 hammer_tid_t clu_gen; /* generation number */
340 int32_t vol_no; /* cluster contained in volume (sanity) */
341 u_int32_t clu_flags; /* cluster flags */
342
343 int32_t clu_start; /* start of data (byte offset) */
344 int32_t clu_limit; /* end of data (byte offset) */
345 int32_t clu_no; /* cluster index in volume (sanity) */
346 u_int32_t clu_reserved03;
347
348 u_int32_t clu_reserved04;
349 u_int32_t clu_reserved05;
350 u_int32_t clu_reserved06;
351 u_int32_t clu_reserved07;
352
353 /*
354 * These fields are mostly heuristics to aid in locality of
355 * reference allocations.
356 */
357 int32_t idx_data; /* data append point (element no) */
358 int32_t idx_index; /* index append point (element no) */
359 int32_t idx_record; /* record prepend point (element no) */
360 int32_t idx_ldata; /* large block data append pt (buf_no) */
361
362 /*
363 * These fields can become out of sync after a filesystem crash
364 * and are cleaned up in the background. They are used for
365 * reporting only.
366 *
367 * NOTE: stat_records counts a spike as two records even though there
368 * is only one record. This is done so we can properly calculate
369 * the worst-case space needed to hold the B-Tree.
370 */
371 int32_t stat_inodes; /* number of inodes in cluster */
372 int32_t stat_records; /* number of records in cluster */
373 int32_t stat_data_bufs; /* hammer bufs allocated to data */
374 int32_t stat_rec_bufs; /* hammer bufs allocated to records */
375 int32_t stat_idx_bufs; /* hammer bufs allocated to B-Tree */
376
377 /*
378 * Specify the range of information stored in this cluster as two
379 * btree elements. These elements match the left and right
380 * boundary elements in the internal B-Tree node of the parent
381 * cluster that points to the root of our cluster. Because these
382 * are boundary elements, the right boundary is range-NONinclusive.
383 */
384 struct hammer_base_elm clu_btree_beg;
385 struct hammer_base_elm clu_btree_end;
386
387 /*
388 * The cluster's B-Tree root can change as a side effect of insertion
389 * and deletion operations so store an offset instead of embedding
390 * the root node. The parent_offset is stale if the generation number
391 * does not match.
392 *
393 * Parent linkages are explicit.
394 */
395 int32_t clu_btree_root;
396 int32_t clu_btree_parent_vol_no;
397 int32_t clu_btree_parent_clu_no;
398 int32_t clu_btree_parent_offset;
399 hammer_tid_t clu_btree_parent_clu_gen;
400
401 /*
402 * The synchronized record id is used for recovery purposes.
403 *
404 * For recovery purposes, only clu_record_meta[] is recovered.
405 * The remaining a-list's are regenerated based on the records
406 * found.
407 */
408 u_int64_t synchronized_rec_id;
409 u_int32_t reserved16[510];
410
411 struct hammer_almeta clu_master_meta[HAMMER_CLU_MASTER_METAELMS];
412 struct hammer_almeta clu_btree_meta[HAMMER_CLU_SLAVE_METAELMS];
413 struct hammer_almeta clu_record_meta[HAMMER_CLU_SLAVE_METAELMS];
414 struct hammer_almeta clu_mdata_meta[HAMMER_CLU_SLAVE_METAELMS];
415};
416
417typedef struct hammer_cluster_ondisk *hammer_cluster_ondisk_t;
418
419/*
420 * Cluster clu_flags
421 *
422 * OPEN - A cluster is marked open and synchronized to disk prior to any
423 * modifications being made to either the cluster header or any cluster
424 * buffers. If initial access to a cluster finds this flag set, the
425 * cluster is recovered before any further operations are performed on it.
426 */
427#define HAMMER_CLUF_OPEN 0x0001 /* cluster is dirty */
428
429/*
430 * HAMMER records are 96 byte entities encoded into 16K filesystem buffers.
431 * Each record has a 64 byte header and a 32 byte extension. 170 records
432 * fit into each buffer. Storage is managed by the buffer's A-List.
433 *
434 * Each record may have an explicit data reference to a block of data up
435 * to 2^31-1 bytes in size within the current cluster. Note that multiple
436 * records may share the same or overlapping data references.
437 */
438
439/*
440 * All HAMMER records have a common 64-byte base and a 32-byte extension.
441 *
442 * Many HAMMER record types reference out-of-band data within the cluster.
443 * This data can also be stored in-band in the record itself if it is small
444 * enough. Either way, (data_offset, data_len) points to it.
445 *
446 * Key comparison order: obj_id, rec_type, key, delete_tid
447 */
448struct hammer_base_record {
449 /*
450 * 40 byte base element info - same base as used in B-Tree internal
451 * and leaf node element arrays.
452 *
453 * Fields: obj_id, key, create_tid, delete_tid, rec_type, obj_type,
454 * reserved07.
455 */
456 struct hammer_base_elm base; /* 00 base element info */
457
458 int32_t data_len; /* 28 size of data (remainder zero-fill) */
459 u_int32_t data_crc; /* 2C data sanity check */
460 u_int64_t rec_id; /* 30 record id (iterator for recovery) */
461 int32_t data_offset; /* 38 cluster-relative data reference or 0 */
462 u_int32_t reserved07; /* 3C */
463 /* 40 */
464};
465
466/*
467 * Record types are fairly straightforward. The B-Tree includes the record
468 * type in its index sort.
469 *
470 * In particular please note that it is possible to create a pseudo-
471 * filesystem within a HAMMER filesystem by creating a special object
472 * type within a directory. Pseudo-filesystems are used as replication
473 * targets and even though they are built within a HAMMER filesystem they
474 * get their own obj_id space (and thus can serve as a replication target)
475 * and look like a mount point to the system.
476 *
477 * Inter-cluster records are special-cased in the B-Tree. These records
478 * are referenced from a B-Tree INTERNAL node, NOT A LEAF. This means
479 * that the element in the B-Tree node is actually a boundary element whos
480 * base element fields, including rec_type, reflect the boundary, NOT
481 * the inter-cluster record type.
482 *
483 * HAMMER_RECTYPE_CLUSTER - only set in the actual inter-cluster record,
484 * not set in the left or right boundary elements around the inter-cluster
485 * reference of an internal node in the B-Tree (because doing so would
486 * interfere with the boundary tests).
487 *
488 * NOTE: hammer_ip_delete_range_all() deletes all record types greater
489 * then HAMMER_RECTYPE_INODE.
490 */
491#define HAMMER_RECTYPE_UNKNOWN 0
492#define HAMMER_RECTYPE_LOWEST 1 /* lowest record type avail */
493#define HAMMER_RECTYPE_INODE 1 /* inode in obj_id space */
494#define HAMMER_RECTYPE_PSEUDO_INODE 2 /* pseudo filesysem */
495#define HAMMER_RECTYPE_CLUSTER 3 /* inter-cluster reference */
496#define HAMMER_RECTYPE_DATA 0x10
497#define HAMMER_RECTYPE_DIRENTRY 0x11
498#define HAMMER_RECTYPE_DB 0x12
499#define HAMMER_RECTYPE_EXT 0x13 /* ext attributes */
500#define HAMMER_RECTYPE_FIX 0x14 /* fixed attribute */
501
502#define HAMMER_FIXKEY_SYMLINK 1
503
504#define HAMMER_OBJTYPE_UNKNOWN 0 /* (never exists on-disk) */
505#define HAMMER_OBJTYPE_DIRECTORY 1
506#define HAMMER_OBJTYPE_REGFILE 2
507#define HAMMER_OBJTYPE_DBFILE 3
508#define HAMMER_OBJTYPE_FIFO 4
509#define HAMMER_OBJTYPE_CDEV 5
510#define HAMMER_OBJTYPE_BDEV 6
511#define HAMMER_OBJTYPE_SOFTLINK 7
512#define HAMMER_OBJTYPE_PSEUDOFS 8 /* pseudo filesystem obj */
513
514/*
515 * Generic full-sized record
516 */
517struct hammer_generic_record {
518 struct hammer_base_record base;
519 char filler[32];
520};
521
522/*
523 * A HAMMER inode record.
524 *
525 * This forms the basis for a filesystem object. obj_id is the inode number,
526 * key1 represents the pseudo filesystem id for security partitioning
527 * (preventing cross-links and/or restricting a NFS export and specifying the
528 * security policy), and key2 represents the data retention policy id.
529 *
530 * Inode numbers are 64 bit quantities which uniquely identify a filesystem
531 * object for the ENTIRE life of the filesystem, even after the object has
532 * been deleted. For all intents and purposes inode numbers are simply
533 * allocated by incrementing a sequence space.
534 *
535 * There is an important distinction between the data stored in the inode
536 * record and the record's data reference. The record references a
537 * hammer_inode_data structure but the filesystem object size and hard link
538 * count is stored in the inode record itself. This allows multiple inodes
539 * to share the same hammer_inode_data structure. This is possible because
540 * any modifications will lay out new data. The HAMMER implementation need
541 * not use the data-sharing ability when laying down new records.
542 *
543 * A HAMMER inode is subject to the same historical storage requirements
544 * as any other record. In particular any change in filesystem or hard link
545 * count will lay down a new inode record when the filesystem is synced to
546 * disk. This can lead to a lot of junk records which get cleaned up by
547 * the data retention policy.
548 *
549 * The ino_atime and ino_mtime fields are a special case. Modifications to
550 * these fields do NOT lay down a new record by default, though the values
551 * are effectively frozen for snapshots which access historical versions
552 * of the inode record due to other operations. This means that atime will
553 * not necessarily be accurate in snapshots, backups, or mirrors. mtime
554 * will be accurate in backups and mirrors since it can be regenerated from
555 * the mirroring stream.
556 *
557 * Because nlinks is historically retained the hardlink count will be
558 * accurate when accessing a HAMMER filesystem snapshot.
559 */
560struct hammer_inode_record {
561 struct hammer_base_record base;
562 u_int64_t ino_atime; /* last access time (not historical) */
563 u_int64_t ino_mtime; /* last modified time (not historical) */
564 u_int64_t ino_size; /* filesystem object size */
565 u_int64_t ino_nlinks; /* hard links */
566};
567
568/*
569 * Data records specify the entire contents of a regular file object,
570 * including attributes. Small amounts of data can theoretically be
571 * embedded in the record itself but the use of this ability verses using
572 * an out-of-band data reference depends on the implementation.
573 */
574struct hammer_data_record {
575 struct hammer_base_record base;
576 char filler[32];
577};
578
579/*
580 * A directory entry specifies the HAMMER filesystem object id, a copy of
581 * the file type, and file name (either embedded or as out-of-band data).
582 * If the file name is short enough to fit into den_name[] (including a
583 * terminating nul) then it will be embedded in the record, otherwise it
584 * is stored out-of-band. The base record's data reference always points
585 * to the nul-terminated filename regardless.
586 *
587 * Directory entries are indexed with a 128 bit namekey rather then an
588 * offset. A portion of the namekey is an iterator or randomizer to deal
589 * with collisions.
590 *
591 * NOTE: base.base.obj_type holds the filesystem object type of obj_id,
592 * e.g. a den_type equivalent.
593 *
594 * NOTE: den_name / the filename data reference is NOT terminated with \0.
595 *
596 */
597struct hammer_entry_record {
598 struct hammer_base_record base;
599 u_int64_t obj_id; /* object being referenced */
600 u_int64_t reserved01;
601 char den_name[16]; /* short file names fit in record */
602};
603
604/*
605 * Spike record
606 */
607struct hammer_spike_record {
608 struct hammer_base_record base;
609 int32_t clu_no;
610 int32_t vol_no;
611 hammer_tid_t clu_id;
612 char reserved[16];
613};
614
615/*
616 * Hammer rollup record
617 */
618union hammer_record_ondisk {
619 struct hammer_base_record base;
620 struct hammer_generic_record generic;
621 struct hammer_spike_record spike;
622 struct hammer_inode_record inode;
623 struct hammer_data_record data;
624 struct hammer_entry_record entry;
625};
626
627typedef union hammer_record_ondisk *hammer_record_ondisk_t;
628
629/*
630 * Filesystem buffer for records
631 */
632#define HAMMER_RECORD_NODES \
633 ((HAMMER_BUFSIZE - sizeof(struct hammer_fsbuf_head) - 32) / \
634 sizeof(union hammer_record_ondisk))
635
636#define HAMMER_RECORD_SIZE (64+32)
637
638struct hammer_fsbuf_recs {
639 struct hammer_fsbuf_head head;
640 char unused[32];
641 union hammer_record_ondisk recs[HAMMER_RECORD_NODES];
642};
643
644/*
645 * Filesystem buffer for piecemeal data. Note that this does not apply
646 * to dedicated pure-data buffers as such buffers do not have a header.
647 */
648
649#define HAMMER_DATA_SIZE (HAMMER_BUFSIZE - sizeof(struct hammer_fsbuf_head))
650#define HAMMER_DATA_BLKSIZE 64
651#define HAMMER_DATA_BLKMASK (HAMMER_DATA_BLKSIZE-1)
652#define HAMMER_DATA_NODES (HAMMER_DATA_SIZE / HAMMER_DATA_BLKSIZE)
653
654struct hammer_fsbuf_data {
655 struct hammer_fsbuf_head head;
656 u_int8_t data[HAMMER_DATA_NODES][HAMMER_DATA_BLKSIZE];
657};
658
659/*
660 * Filesystem buffer rollup
661 */
662union hammer_fsbuf_ondisk {
663 struct hammer_fsbuf_head head;
664 struct hammer_fsbuf_btree btree;
665 struct hammer_fsbuf_recs record;
666 struct hammer_fsbuf_data data;
667};
668
669typedef union hammer_fsbuf_ondisk *hammer_fsbuf_ondisk_t;
670
671/*
672 * HAMMER UNIX Attribute data
673 *
674 * The data reference in a HAMMER inode record points to this structure. Any
675 * modifications to the contents of this structure will result in a record
676 * replacement operation.
677 *
678 * short_data_off allows a small amount of data to be embedded in the
679 * hammer_inode_data structure. HAMMER typically uses this to represent
680 * up to 64 bytes of data, or to hold symlinks. Remember that allocations
681 * are in powers of 2 so 64, 192, 448, or 960 bytes of embedded data is
682 * support (64+64, 64+192, 64+448 64+960).
683 *
684 * parent_obj_id is only valid for directories (which cannot be hard-linked),
685 * and specifies the parent directory obj_id. This field will also be set
686 * for non-directory inodes as a recovery aid, but can wind up specifying
687 * stale information. However, since object id's are not reused, the worse
688 * that happens is that the recovery code is unable to use it.
689 */
690struct hammer_inode_data {
691 u_int16_t version; /* inode data version */
692 u_int16_t mode; /* basic unix permissions */
693 u_int32_t uflags; /* chflags */
694 u_int32_t rmajor; /* used by device nodes */
695 u_int32_t rminor; /* used by device nodes */
696 u_int64_t ctime;
697 u_int64_t parent_obj_id;/* parent directory obj_id */
698 uuid_t uid;
699 uuid_t gid;
700 /* XXX device, softlink extension */
701};
702
703#define HAMMER_INODE_DATA_VERSION 1
704
705#define HAMMER_OBJID_ROOT 1
706
707/*
708 * Rollup various structures embedded as record data
709 */
710union hammer_data_ondisk {
711 struct hammer_inode_data inode;
712};
713
714#endif /* HAMMER_ALIST_METAELMS_256_1LYR */
715
716#endif