xref: /dflybsd-src/sys/vfs/hammer/hammer_disk.h (revision c4bf625e67439f34b29bfd33c4e2555ffea63ce9)
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 1.28 2008/04/22 19:00:15 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  * Per-volume storage limit: 52 bits		4096 TB
64  * Per-Zone storage limit: 59 bits		512 KTB (due to blockmap)
65  * Per-filesystem storage limit: 60 bits	1 MTB
66  */
67 #define HAMMER_BUFSIZE		16384
68 #define HAMMER_BUFMASK		(HAMMER_BUFSIZE - 1)
69 #define HAMMER_MAXDATA		(256*1024)
70 #define HAMMER_BUFFER_BITS	14
71 
72 #if (1 << HAMMER_BUFFER_BITS) != HAMMER_BUFSIZE
73 #error "HAMMER_BUFFER_BITS BROKEN"
74 #endif
75 
76 #define HAMMER_BUFSIZE64	((u_int64_t)HAMMER_BUFSIZE)
77 #define HAMMER_BUFMASK64	((u_int64_t)HAMMER_BUFMASK)
78 
79 #define HAMMER_OFF_ZONE_MASK	0xF000000000000000ULL /* zone portion */
80 #define HAMMER_OFF_VOL_MASK	0x0FF0000000000000ULL /* volume portion */
81 #define HAMMER_OFF_SHORT_MASK	0x000FFFFFFFFFFFFFULL /* offset portion */
82 #define HAMMER_OFF_LONG_MASK	0x0FFFFFFFFFFFFFFFULL /* offset portion */
83 #define HAMMER_OFF_SHORT_REC_MASK 0x000FFFFFFF000000ULL /* recovery boundary */
84 #define HAMMER_OFF_LONG_REC_MASK 0x0FFFFFFFFF000000ULL /* recovery boundary */
85 #define HAMMER_RECOVERY_BND	0x0000000001000000ULL
86 
87 /*
88  * Hammer transction ids are 64 bit unsigned integers and are usually
89  * synchronized with the time of day in nanoseconds.
90  *
91  * Hammer offsets are used for FIFO indexing and embed a cycle counter
92  * and volume number in addition to the offset.  Most offsets are required
93  * to be 64-byte aligned.
94  */
95 typedef u_int64_t hammer_tid_t;
96 typedef u_int64_t hammer_off_t;
97 typedef u_int32_t hammer_seq_t;
98 typedef u_int32_t hammer_crc_t;
99 
100 #define HAMMER_MIN_TID		0ULL			/* unsigned */
101 #define HAMMER_MAX_TID		0xFFFFFFFFFFFFFFFFULL	/* unsigned */
102 #define HAMMER_MIN_KEY		-0x8000000000000000LL	/* signed */
103 #define HAMMER_MAX_KEY		0x7FFFFFFFFFFFFFFFLL	/* signed */
104 #define HAMMER_MIN_OBJID	HAMMER_MIN_KEY		/* signed */
105 #define HAMMER_MAX_OBJID	HAMMER_MAX_KEY		/* signed */
106 #define HAMMER_MIN_RECTYPE	0x0U			/* unsigned */
107 #define HAMMER_MAX_RECTYPE	0xFFFFU			/* unsigned */
108 #define HAMMER_MIN_OFFSET	0ULL			/* unsigned */
109 #define HAMMER_MAX_OFFSET	0xFFFFFFFFFFFFFFFFULL	/* unsigned */
110 
111 /*
112  * hammer_off_t has several different encodings.  Note that not all zones
113  * encode a vol_no.
114  *
115  * zone 0 (z,v,o):	reserved (for sanity)
116  * zone 1 (z,v,o):	raw volume relative (offset 0 is the volume header)
117  * zone 2 (z,v,o):	raw buffer relative (offset 0 is the first buffer)
118  * zone 3 (z,o):	undo fifo	- fixed layer2 array in root vol hdr
119  * zone 4 (z,v,o):	freemap		- freemap-backed self-mapping special
120  *					  cased layering.
121  *
122  * zone 8 (z,o):	B-Tree		- blkmap-backed
123  * zone 9 (z,o):	Record		- blkmap-backed
124  * zone 10 (z,o):	Large-data	- blkmap-backed
125  */
126 
127 #define HAMMER_ZONE_RAW_VOLUME		0x1000000000000000ULL
128 #define HAMMER_ZONE_RAW_BUFFER		0x2000000000000000ULL
129 #define HAMMER_ZONE_UNDO		0x3000000000000000ULL
130 #define HAMMER_ZONE_FREEMAP		0x4000000000000000ULL
131 #define HAMMER_ZONE_RESERVED05		0x5000000000000000ULL
132 #define HAMMER_ZONE_RESERVED06		0x6000000000000000ULL
133 #define HAMMER_ZONE_RESERVED07		0x7000000000000000ULL
134 #define HAMMER_ZONE_BTREE		0x8000000000000000ULL
135 #define HAMMER_ZONE_RECORD		0x9000000000000000ULL
136 #define HAMMER_ZONE_LARGE_DATA		0xA000000000000000ULL
137 #define HAMMER_ZONE_SMALL_DATA		0xB000000000000000ULL
138 #define HAMMER_ZONE_RESERVED0C		0xC000000000000000ULL
139 #define HAMMER_ZONE_RESERVED0D		0xD000000000000000ULL
140 #define HAMMER_ZONE_RESERVED0E		0xE000000000000000ULL
141 #define HAMMER_ZONE_RESERVED0F		0xF000000000000000ULL
142 
143 #define HAMMER_ZONE_RAW_VOLUME_INDEX	1
144 #define HAMMER_ZONE_RAW_BUFFER_INDEX	2
145 #define HAMMER_ZONE_UNDO_INDEX		3
146 #define HAMMER_ZONE_FREEMAP_INDEX	4
147 #define HAMMER_ZONE_BTREE_INDEX		8
148 #define HAMMER_ZONE_RECORD_INDEX	9
149 #define HAMMER_ZONE_LARGE_DATA_INDEX	10
150 #define HAMMER_ZONE_SMALL_DATA_INDEX	11
151 
152 /*
153  * Per-zone size limitation.  This just makes the iterator easier
154  * to deal with by preventing an iterator overflow.
155  */
156 #define HAMMER_ZONE_LIMIT		\
157 	(0x1000000000000000ULL - HAMMER_BLOCKMAP_LAYER2)
158 
159 #define HAMMER_MAX_ZONES		16
160 
161 #define HAMMER_VOL_ENCODE(vol_no)			\
162 	((hammer_off_t)((vol_no) & 255) << 52)
163 #define HAMMER_VOL_DECODE(ham_off)			\
164 	(int32_t)(((hammer_off_t)(ham_off) >> 52) & 255)
165 #define HAMMER_ZONE_DECODE(ham_off)			\
166 	(int32_t)(((hammer_off_t)(ham_off) >> 60))
167 #define HAMMER_ZONE_ENCODE(zone, ham_off)		\
168 	(((hammer_off_t)(zone) << 60) | (ham_off))
169 #define HAMMER_SHORT_OFF_ENCODE(offset)			\
170 	((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
171 #define HAMMER_LONG_OFF_ENCODE(offset)			\
172 	((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
173 
174 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset)	\
175 	(HAMMER_ZONE_RAW_VOLUME |			\
176 	HAMMER_VOL_ENCODE(vol_no) |			\
177 	HAMMER_SHORT_OFF_ENCODE(offset))
178 
179 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset)	\
180 	(HAMMER_ZONE_RAW_BUFFER |			\
181 	HAMMER_VOL_ENCODE(vol_no) |			\
182 	HAMMER_SHORT_OFF_ENCODE(offset))
183 
184 #define HAMMER_ENCODE_FREEMAP(vol_no, offset)		\
185 	(HAMMER_ZONE_FREEMAP |				\
186 	HAMMER_VOL_ENCODE(vol_no) |			\
187 	HAMMER_SHORT_OFF_ENCODE(offset))
188 
189 /*
190  * Large-Block backing store
191  *
192  * A blockmap is a two-level map which translates a blockmap-backed zone
193  * offset into a raw zone 2 offset.  Each layer handles 18 bits.  The 8M
194  * large-block size is 23 bits so two layers gives us 23+18+18 = 59 bits
195  * of address space.
196  */
197 #define HAMMER_LARGEBLOCK_SIZE		(8192 * 1024)
198 #define HAMMER_LARGEBLOCK_SIZE64	((u_int64_t)HAMMER_LARGEBLOCK_SIZE)
199 #define HAMMER_LARGEBLOCK_MASK		(HAMMER_LARGEBLOCK_SIZE - 1)
200 #define HAMMER_LARGEBLOCK_MASK64	((u_int64_t)HAMMER_LARGEBLOCK_SIZE - 1)
201 #define HAMMER_LARGEBLOCK_BITS		23
202 #if (1 << HAMMER_LARGEBLOCK_BITS) != HAMMER_LARGEBLOCK_SIZE
203 #error "HAMMER_LARGEBLOCK_BITS BROKEN"
204 #endif
205 
206 #define HAMMER_BUFFERS_PER_LARGEBLOCK			\
207 	(HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE)
208 #define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK		\
209 	(HAMMER_BUFFERS_PER_LARGEBLOCK - 1)
210 #define HAMMER_BUFFERS_PER_LARGEBLOCK_MASK64		\
211 	((hammer_off_t)HAMMER_BUFFERS_PER_LARGEBLOCK_MASK)
212 
213 /*
214  * Every blockmap has this root structure in the root volume header.
215  *
216  * NOTE: zone 3 (the undo FIFO) does not use phys_offset.  first and next
217  * offsets represent the FIFO.
218  */
219 struct hammer_blockmap {
220 	hammer_off_t	phys_offset;    /* zone-2 physical offset */
221 	hammer_off_t	first_offset;	/* zone-X logical offset (zone 3) */
222 	hammer_off_t	next_offset;	/* zone-X logical offset */
223 	hammer_off_t	alloc_offset;	/* zone-X logical offset */
224 	hammer_crc_t	entry_crc;
225 	u_int32_t	reserved01;
226 };
227 
228 typedef struct hammer_blockmap *hammer_blockmap_t;
229 
230 /*
231  * The blockmap is a 2-layer entity made up of big-blocks.  The first layer
232  * contains 262144 32-byte entries (18 bits), the second layer contains
233  * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
234  * 18+19+23 = 60 bits.  The top four bits are the zone id.
235  *
236  * Layer 2 encodes the physical bigblock mapping for a blockmap.  The freemap
237  * uses this field to encode the virtual blockmap offset that allocated the
238  * physical block.
239  *
240  * NOTE:  The freemap maps the vol_no in the upper 8 bits of layer1.
241  *
242  * zone-4 blockmap offset: [z:4][layer1:18][layer2:19][bigblock:23]
243  */
244 struct hammer_blockmap_layer1 {
245 	hammer_off_t	blocks_free;	/* big-blocks free */
246 	hammer_off_t	phys_offset;	/* UNAVAIL or zone-2 */
247 	hammer_crc_t	layer1_crc;	/* crc of this entry */
248 	hammer_crc_t	layer2_crc;	/* xor'd crc's of HAMMER_BLOCKSIZE */
249 	hammer_off_t	reserved01;
250 };
251 
252 struct hammer_blockmap_layer2 {
253 	hammer_crc_t	entry_crc;
254 	u_int32_t	bytes_free;	/* bytes free within this bigblock */
255 	union {
256 		hammer_off_t	owner;		/* used by freemap */
257 		hammer_off_t	phys_offset;	/* used by blockmap */
258 	} u;
259 };
260 
261 #define HAMMER_BLOCKMAP_FREE	0ULL
262 #define HAMMER_BLOCKMAP_UNAVAIL	((hammer_off_t)-1LL)
263 
264 #define HAMMER_BLOCKMAP_RADIX1	/* 262144 (18) */	\
265 	(HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1))
266 #define HAMMER_BLOCKMAP_RADIX2	/* 524288 (19) */	\
267 	(HAMMER_LARGEBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2))
268 
269 #define HAMMER_BLOCKMAP_RADIX1_PERBUFFER	\
270 	(HAMMER_BLOCKMAP_RADIX1 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
271 #define HAMMER_BLOCKMAP_RADIX2_PERBUFFER	\
272 	(HAMMER_BLOCKMAP_RADIX2 / (HAMMER_LARGEBLOCK_SIZE / HAMMER_BUFSIZE))
273 
274 #define HAMMER_BLOCKMAP_LAYER1	/* 18+19+23 */		\
275 	(HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
276 #define HAMMER_BLOCKMAP_LAYER2	/* 19+23 */		\
277 	(HAMMER_BLOCKMAP_RADIX2 * HAMMER_LARGEBLOCK_SIZE64)
278 
279 #define HAMMER_BLOCKMAP_LAYER1_MASK	(HAMMER_BLOCKMAP_LAYER1 - 1)
280 #define HAMMER_BLOCKMAP_LAYER2_MASK	(HAMMER_BLOCKMAP_LAYER2 - 1)
281 
282 /*
283  * byte offset within layer1 or layer2 big-block for the entry representing
284  * a zone-2 physical offset.
285  */
286 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset)	\
287 	(((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) / 	\
288 	 HAMMER_BLOCKMAP_LAYER2 * sizeof(struct hammer_blockmap_layer1))
289 
290 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset)	\
291 	(((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) /	\
292 	HAMMER_LARGEBLOCK_SIZE64 * sizeof(struct hammer_blockmap_layer2))
293 
294 /*
295  * HAMMER UNDO parameters.  The UNDO fifo is mapped directly in the volume
296  * header with an array of layer2 structures.  A maximum of (64x8MB) = 512MB
297  * may be reserved.  The size of the undo fifo is usually set a newfs time
298  * but can be adjusted if the filesystem is taken offline.
299  */
300 
301 #define HAMMER_UNDO_LAYER2	64	/* max layer2 undo mapping entries */
302 
303 /*
304  * All on-disk HAMMER structures which make up elements of the UNDO FIFO
305  * contain a hammer_fifo_head and hammer_fifo_tail structure.  This structure
306  * contains all the information required to validate the fifo element
307  * and to scan the fifo in either direction.  The head is typically embedded
308  * in higher level hammer on-disk structures while the tail is typically
309  * out-of-band.  hdr_size is the size of the whole mess, including the tail.
310  *
311  * All undo structures are guaranteed to not cross a 16K filesystem
312  * buffer boundary.  Most undo structures are fairly small.  Data spaces
313  * are not immediately reused by HAMMER so file data is not usually recorded
314  * as part of an UNDO.
315  *
316  * PAD elements are allowed to take up only 8 bytes of space as a special
317  * case, containing only hdr_signature, hdr_type, and hdr_size fields,
318  * and with the tail overloaded onto the head structure for 8 bytes total.
319  *
320  * Every undo record has a sequence number.  This number is unrelated to
321  * transaction ids and instead collects the undo transactions associated
322  * with a single atomic operation.  A larger transactional operation, such
323  * as a remove(), may consist of several smaller atomic operations
324  * representing raw meta-data operations.
325  */
326 #define HAMMER_HEAD_ONDISK_SIZE		32
327 #define HAMMER_HEAD_ALIGN		8
328 #define HAMMER_HEAD_ALIGN_MASK		(HAMMER_HEAD_ALIGN - 1)
329 #define HAMMER_TAIL_ONDISK_SIZE		8
330 
331 struct hammer_fifo_head {
332 	u_int16_t hdr_signature;
333 	u_int16_t hdr_type;
334 	u_int32_t hdr_size;	/* aligned size of the whole mess */
335 	u_int32_t reserved01;	/* (0) reserved for future use */
336 	hammer_crc_t hdr_crc;
337 };
338 
339 struct hammer_fifo_tail {
340 	u_int16_t tail_signature;
341 	u_int16_t tail_type;
342 	u_int32_t tail_size;	/* aligned size of the whole mess */
343 };
344 
345 typedef struct hammer_fifo_head *hammer_fifo_head_t;
346 typedef struct hammer_fifo_tail *hammer_fifo_tail_t;
347 
348 /*
349  * Fifo header types.
350  */
351 #define HAMMER_HEAD_TYPE_PAD	(0x0040U|HAMMER_HEAD_FLAG_FREE)
352 #define HAMMER_HEAD_TYPE_VOL	0x0041U		/* Volume (dummy header) */
353 #define HAMMER_HEAD_TYPE_BTREE	0x0042U		/* B-Tree node */
354 #define HAMMER_HEAD_TYPE_UNDO	0x0043U		/* random UNDO information */
355 #define HAMMER_HEAD_TYPE_DELETE	0x0044U		/* record deletion */
356 #define HAMMER_HEAD_TYPE_RECORD	0x0045U		/* Filesystem record */
357 
358 #define HAMMER_HEAD_FLAG_FREE	0x8000U		/* Indicates object freed */
359 
360 #define HAMMER_HEAD_SIGNATURE	0xC84EU
361 #define HAMMER_TAIL_SIGNATURE	0xC74FU
362 
363 #define HAMMER_HEAD_SEQ_BEG	0x80000000U
364 #define HAMMER_HEAD_SEQ_END	0x40000000U
365 #define HAMMER_HEAD_SEQ_MASK	0x3FFFFFFFU
366 
367 /*
368  * Misc FIFO structures.
369  */
370 struct hammer_fifo_undo {
371 	struct hammer_fifo_head	head;
372 	hammer_off_t		undo_offset;	/* zone-1 offset */
373 	int32_t			undo_data_bytes;
374 	int32_t			undo_reserved01;
375 	/* followed by data */
376 };
377 
378 typedef struct hammer_fifo_undo *hammer_fifo_undo_t;
379 
380 struct hammer_fifo_buf_commit {
381 	hammer_off_t		undo_offset;
382 };
383 
384 /*
385  * Volume header types
386  */
387 #define HAMMER_FSBUF_VOLUME	0xC8414D4DC5523031ULL	/* HAMMER01 */
388 #define HAMMER_FSBUF_VOLUME_REV	0x313052C54D4D41C8ULL	/* (reverse endian) */
389 
390 /*
391  * The B-Tree structures need hammer_fsbuf_head.
392  */
393 #include "hammer_btree.h"
394 
395 /*
396  * HAMMER Volume header
397  *
398  * A HAMMER filesystem is built from any number of block devices,  Each block
399  * device contains a volume header followed by however many buffers fit
400  * into the volume.
401  *
402  * One of the volumes making up a HAMMER filesystem is the master, the
403  * rest are slaves.  It does not have to be volume #0.
404  *
405  * The volume header takes up an entire 16K filesystem buffer and may
406  * represent up to 64KTB (65536 TB) of space.
407  *
408  * Special field notes:
409  *
410  *	vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
411  *	vol_mem_beg - offset of memory log (clu_beg - mem_beg bytes)
412  *	vol_buf_beg - offset of the first buffer.
413  *
414  *	The memory log area allows a kernel to cache new records and data
415  *	in memory without allocating space in the actual filesystem to hold
416  *	the records and data.  In the event that a filesystem becomes full,
417  *	any records remaining in memory can be flushed to the memory log
418  *	area.  This allows the kernel to immediately return success.
419  */
420 
421 #define HAMMER_BOOT_MINBYTES		(32*1024)
422 #define HAMMER_BOOT_NOMBYTES		(64LL*1024*1024)
423 #define HAMMER_BOOT_MAXBYTES		(256LL*1024*1024)
424 
425 #define HAMMER_MEM_MINBYTES		(256*1024)
426 #define HAMMER_MEM_NOMBYTES		(1LL*1024*1024*1024)
427 #define HAMMER_MEM_MAXBYTES		(64LL*1024*1024*1024)
428 
429 struct hammer_volume_ondisk {
430 	u_int64_t vol_signature;/* Signature */
431 
432 	int64_t vol_bot_beg;	/* byte offset of boot area or 0 */
433 	int64_t vol_mem_beg;	/* byte offset of memory log or 0 */
434 	int64_t vol_buf_beg;	/* byte offset of first buffer in volume */
435 	int64_t vol_buf_end;	/* byte offset of volume EOF (on buf bndry) */
436 	int64_t vol_locked;	/* reserved clusters are >= this offset */
437 
438 	uuid_t    vol_fsid;	/* identify filesystem */
439 	uuid_t    vol_fstype;	/* identify filesystem type */
440 	char	  vol_name[64];	/* Name of volume */
441 
442 	int32_t vol_no;		/* volume number within filesystem */
443 	int32_t vol_count;	/* number of volumes making up FS */
444 
445 	u_int32_t vol_version;	/* version control information */
446 	u_int32_t vol_reserved01;
447 	u_int32_t vol_flags;	/* volume flags */
448 	u_int32_t vol_rootvol;	/* which volume is the root volume? */
449 
450 	int32_t vol_reserved04;
451 	int32_t vol_reserved05;
452 	u_int32_t vol_reserved06;
453 	u_int32_t vol_reserved07;
454 
455 	int32_t vol_blocksize;		/* for statfs only */
456 	int32_t vol_reserved08;
457 	int64_t vol_nblocks;		/* total allocatable hammer bufs */
458 
459 	/*
460 	 * These fields are initialized and space is reserved in every
461 	 * volume making up a HAMMER filesytem, but only the master volume
462 	 * contains valid data.
463 	 */
464 	int64_t vol0_stat_bigblocks;	/* total bigblocks when fs is empty */
465 	int64_t vol0_stat_freebigblocks;/* number of free bigblocks */
466 	int64_t	vol0_stat_bytes;	/* for statfs only */
467 	int64_t vol0_stat_inodes;	/* for statfs only */
468 	int64_t vol0_stat_records;	/* total records in filesystem */
469 	hammer_off_t vol0_btree_root;	/* B-Tree root */
470 	hammer_tid_t vol0_next_tid;	/* highest synchronized TID */
471 	u_int32_t vol0_reserved00;
472 	u_int32_t vol0_reserved01;
473 
474 	/*
475 	 * Blockmaps for zones.  Not all zones use a blockmap.
476 	 */
477 	struct hammer_blockmap	vol0_blockmap[HAMMER_MAX_ZONES];
478 
479 	/*
480 	 * Layer-2 array for undo fifo
481 	 */
482 	struct hammer_blockmap_layer2 vol0_undo_array[HAMMER_UNDO_LAYER2];
483 
484 };
485 
486 typedef struct hammer_volume_ondisk *hammer_volume_ondisk_t;
487 
488 #define HAMMER_VOLF_VALID		0x0001	/* valid entry */
489 #define HAMMER_VOLF_OPEN		0x0002	/* volume is open */
490 
491 /*
492  * All HAMMER records have a common 64-byte base and a 32 byte extension,
493  * plus a possible data reference.  The data reference can be in-band or
494  * out-of-band.
495  */
496 
497 #define HAMMER_RECORD_SIZE		(64+32)
498 
499 struct hammer_base_record {
500 	u_int32_t	signature;	/* record signature */
501 	hammer_crc_t	data_crc;	/* data crc */
502 	struct hammer_base_elm base;	/* 40 byte base element */
503 	hammer_off_t	data_off;	/* in-band or out-of-band */
504 	int32_t		data_len;	/* size of data in bytes */
505 	u_int32_t	reserved02;
506 };
507 
508 /*
509  * Record types are fairly straightforward.  The B-Tree includes the record
510  * type in its index sort.
511  *
512  * In particular please note that it is possible to create a pseudo-
513  * filesystem within a HAMMER filesystem by creating a special object
514  * type within a directory.  Pseudo-filesystems are used as replication
515  * targets and even though they are built within a HAMMER filesystem they
516  * get their own obj_id space (and thus can serve as a replication target)
517  * and look like a mount point to the system.
518  *
519  * Inter-cluster records are special-cased in the B-Tree.  These records
520  * are referenced from a B-Tree INTERNAL node, NOT A LEAF.  This means
521  * that the element in the B-Tree node is actually a boundary element whos
522  * base element fields, including rec_type, reflect the boundary, NOT
523  * the inter-cluster record type.
524  *
525  * HAMMER_RECTYPE_CLUSTER - only set in the actual inter-cluster record,
526  * not set in the left or right boundary elements around the inter-cluster
527  * reference of an internal node in the B-Tree (because doing so would
528  * interfere with the boundary tests).
529  *
530  * NOTE: hammer_ip_delete_range_all() deletes all record types greater
531  * then HAMMER_RECTYPE_INODE.
532  */
533 #define HAMMER_RECTYPE_UNKNOWN		0
534 #define HAMMER_RECTYPE_LOWEST		1	/* lowest record type avail */
535 #define HAMMER_RECTYPE_INODE		1	/* inode in obj_id space */
536 #define HAMMER_RECTYPE_PSEUDO_INODE	2	/* pseudo filesysem */
537 #define HAMMER_RECTYPE_CLUSTER		3	/* inter-cluster reference */
538 #define HAMMER_RECTYPE_DATA		0x0010
539 #define HAMMER_RECTYPE_DIRENTRY		0x0011
540 #define HAMMER_RECTYPE_DB		0x0012
541 #define HAMMER_RECTYPE_EXT		0x0013	/* ext attributes */
542 #define HAMMER_RECTYPE_FIX		0x0014	/* fixed attribute */
543 #define HAMMER_RECTYPE_MOVED		0x8000	/* special recovery flag */
544 
545 #define HAMMER_FIXKEY_SYMLINK		1
546 
547 #define HAMMER_OBJTYPE_UNKNOWN		0	/* (never exists on-disk) */
548 #define HAMMER_OBJTYPE_DIRECTORY	1
549 #define HAMMER_OBJTYPE_REGFILE		2
550 #define HAMMER_OBJTYPE_DBFILE		3
551 #define HAMMER_OBJTYPE_FIFO		4
552 #define HAMMER_OBJTYPE_CDEV		5
553 #define HAMMER_OBJTYPE_BDEV		6
554 #define HAMMER_OBJTYPE_SOFTLINK		7
555 #define HAMMER_OBJTYPE_PSEUDOFS		8	/* pseudo filesystem obj */
556 
557 /*
558  * A HAMMER inode record.
559  *
560  * This forms the basis for a filesystem object.  obj_id is the inode number,
561  * key1 represents the pseudo filesystem id for security partitioning
562  * (preventing cross-links and/or restricting a NFS export and specifying the
563  * security policy), and key2 represents the data retention policy id.
564  *
565  * Inode numbers are 64 bit quantities which uniquely identify a filesystem
566  * object for the ENTIRE life of the filesystem, even after the object has
567  * been deleted.  For all intents and purposes inode numbers are simply
568  * allocated by incrementing a sequence space.
569  *
570  * There is an important distinction between the data stored in the inode
571  * record and the record's data reference.  The record references a
572  * hammer_inode_data structure but the filesystem object size and hard link
573  * count is stored in the inode record itself.  This allows multiple inodes
574  * to share the same hammer_inode_data structure.  This is possible because
575  * any modifications will lay out new data.  The HAMMER implementation need
576  * not use the data-sharing ability when laying down new records.
577  *
578  * A HAMMER inode is subject to the same historical storage requirements
579  * as any other record.  In particular any change in filesystem or hard link
580  * count will lay down a new inode record when the filesystem is synced to
581  * disk.  This can lead to a lot of junk records which get cleaned up by
582  * the data retention policy.
583  *
584  * The ino_atime and ino_mtime fields are a special case.  Modifications to
585  * these fields do NOT lay down a new record by default, though the values
586  * are effectively frozen for snapshots which access historical versions
587  * of the inode record due to other operations.  This means that atime will
588  * not necessarily be accurate in snapshots, backups, or mirrors.  mtime
589  * will be accurate in backups and mirrors since it can be regenerated from
590  * the mirroring stream.
591  *
592  * Because nlinks is historically retained the hardlink count will be
593  * accurate when accessing a HAMMER filesystem snapshot.
594  */
595 struct hammer_inode_record {
596 	struct hammer_base_record base;
597 	u_int64_t ino_atime;	/* last access time (not historical) */
598 	u_int64_t ino_mtime;	/* last modified time (not historical) */
599 	u_int64_t ino_size;	/* filesystem object size */
600 	u_int64_t ino_nlinks;	/* hard links */
601 };
602 
603 /*
604  * Data records specify the entire contents of a regular file object,
605  * including attributes.  Small amounts of data can theoretically be
606  * embedded in the record itself but the use of this ability verses using
607  * an out-of-band data reference depends on the implementation.
608  */
609 struct hammer_data_record {
610 	struct hammer_base_record base;
611 	char	data[32];
612 };
613 
614 /*
615  * A directory entry specifies the HAMMER filesystem object id, a copy of
616  * the file type, and file name (either embedded or as out-of-band data).
617  * If the file name is short enough to fit into den_name[] (including a
618  * terminating nul) then it will be embedded in the record, otherwise it
619  * is stored out-of-band.  The base record's data reference always points
620  * to the nul-terminated filename regardless.
621  *
622  * Directory entries are indexed with a 128 bit namekey rather then an
623  * offset.  A portion of the namekey is an iterator or randomizer to deal
624  * with collisions.
625  *
626  * NOTE: base.base.obj_type holds the filesystem object type of obj_id,
627  *	 e.g. a den_type equivalent.
628  *
629  * NOTE: den_name / the filename data reference is NOT terminated with \0.
630  *
631  */
632 struct hammer_entry_record {
633 	struct hammer_base_record base;
634 	u_int64_t obj_id;		/* object being referenced */
635 	u_int64_t reserved01;
636 	char	name[16];
637 };
638 
639 /*
640  * Hammer rollup record
641  */
642 union hammer_record_ondisk {
643 	struct hammer_base_record	base;
644 	struct hammer_inode_record	inode;
645 	struct hammer_data_record	data;
646 	struct hammer_entry_record	entry;
647 };
648 
649 typedef union hammer_record_ondisk *hammer_record_ondisk_t;
650 
651 /*
652  * HAMMER UNIX Attribute data
653  *
654  * The data reference in a HAMMER inode record points to this structure.  Any
655  * modifications to the contents of this structure will result in a record
656  * replacement operation.
657  *
658  * short_data_off allows a small amount of data to be embedded in the
659  * hammer_inode_data structure.  HAMMER typically uses this to represent
660  * up to 64 bytes of data, or to hold symlinks.  Remember that allocations
661  * are in powers of 2 so 64, 192, 448, or 960 bytes of embedded data is
662  * support (64+64, 64+192, 64+448 64+960).
663  *
664  * parent_obj_id is only valid for directories (which cannot be hard-linked),
665  * and specifies the parent directory obj_id.  This field will also be set
666  * for non-directory inodes as a recovery aid, but can wind up specifying
667  * stale information.  However, since object id's are not reused, the worse
668  * that happens is that the recovery code is unable to use it.
669  */
670 struct hammer_inode_data {
671 	u_int16_t version;	/* inode data version */
672 	u_int16_t mode;		/* basic unix permissions */
673 	u_int32_t uflags;	/* chflags */
674 	u_int32_t rmajor;	/* used by device nodes */
675 	u_int32_t rminor;	/* used by device nodes */
676 	u_int64_t ctime;
677 	u_int64_t parent_obj_id;/* parent directory obj_id */
678 	uuid_t	uid;
679 	uuid_t	gid;
680 	/* XXX device, softlink extension */
681 };
682 
683 #define HAMMER_INODE_DATA_VERSION	1
684 
685 #define HAMMER_OBJID_ROOT		1
686 
687 /*
688  * Rollup various structures embedded as record data
689  */
690 union hammer_data_ondisk {
691 	struct hammer_inode_data inode;
692 };
693 
694 #endif
695