xref: /freebsd/usr.sbin/fstyp/hammer_disk.h (revision 349cc55c9796c4596a5b9904cd3281af295f878f)
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 1.55 2008/11/13 02:18:43 dillon Exp $
35  * $FreeBSD$
36  */
37 
38 #ifndef VFS_HAMMER_DISK_H_
39 #define VFS_HAMMER_DISK_H_
40 
41 #include <sys/endian.h>
42 
43 #ifndef _SYS_UUID_H_
44 #include <sys/uuid.h>
45 #endif
46 
47 /*
48  * The structures below represent the on-disk format for a HAMMER
49  * filesystem.  Note that all fields for on-disk structures are naturally
50  * aligned.  HAMMER uses little endian for fields in on-disk structures.
51  * HAMMER doesn't support big endian arch, but is planned.
52  *
53  * Most of HAMMER revolves around the concept of an object identifier.  An
54  * obj_id is a 64 bit quantity which uniquely identifies a filesystem object
55  * FOR THE ENTIRE LIFE OF THE FILESYSTEM.  This uniqueness allows backups
56  * and mirrors to retain varying amounts of filesystem history by removing
57  * any possibility of conflict through identifier reuse.
58  *
59  * A HAMMER filesystem may span multiple volumes.
60  *
61  * A HAMMER filesystem uses a 16K filesystem buffer size.  All filesystem
62  * I/O is done in multiples of 16K.
63  *
64  * 64K X-bufs are used for blocks >= a file's 1MB mark.
65  *
66  * Per-volume storage limit: 52 bits		4096 TB
67  * Per-Zone storage limit: 60 bits		1 MTB
68  * Per-filesystem storage limit: 60 bits	1 MTB
69  */
70 #define HAMMER_BUFSIZE		16384
71 #define HAMMER_XBUFSIZE		65536
72 #define HAMMER_HBUFSIZE		(HAMMER_BUFSIZE / 2)
73 #define HAMMER_XDEMARC		(1024 * 1024)
74 #define HAMMER_BUFMASK		(HAMMER_BUFSIZE - 1)
75 #define HAMMER_XBUFMASK		(HAMMER_XBUFSIZE - 1)
76 
77 #define HAMMER_BUFSIZE64	((uint64_t)HAMMER_BUFSIZE)
78 #define HAMMER_BUFMASK64	((uint64_t)HAMMER_BUFMASK)
79 
80 #define HAMMER_XBUFSIZE64	((uint64_t)HAMMER_XBUFSIZE)
81 #define HAMMER_XBUFMASK64	((uint64_t)HAMMER_XBUFMASK)
82 
83 #define HAMMER_OFF_ZONE_MASK	0xF000000000000000ULL /* zone portion */
84 #define HAMMER_OFF_VOL_MASK	0x0FF0000000000000ULL /* volume portion */
85 #define HAMMER_OFF_SHORT_MASK	0x000FFFFFFFFFFFFFULL /* offset portion */
86 #define HAMMER_OFF_LONG_MASK	0x0FFFFFFFFFFFFFFFULL /* offset portion */
87 
88 #define HAMMER_OFF_BAD		((hammer_off_t)-1)
89 
90 #define HAMMER_BUFSIZE_DOALIGN(offset)				\
91 	(((offset) + HAMMER_BUFMASK) & ~HAMMER_BUFMASK)
92 #define HAMMER_BUFSIZE64_DOALIGN(offset)			\
93 	(((offset) + HAMMER_BUFMASK64) & ~HAMMER_BUFMASK64)
94 
95 #define HAMMER_XBUFSIZE_DOALIGN(offset)				\
96 	(((offset) + HAMMER_XBUFMASK) & ~HAMMER_XBUFMASK)
97 #define HAMMER_XBUFSIZE64_DOALIGN(offset)			\
98 	(((offset) + HAMMER_XBUFMASK64) & ~HAMMER_XBUFMASK64)
99 
100 /*
101  * The current limit of volumes that can make up a HAMMER FS
102  */
103 #define HAMMER_MAX_VOLUMES	256
104 
105 /*
106  * Reserved space for (future) header junk after the volume header.
107  */
108 #define HAMMER_MIN_VOL_JUNK	(HAMMER_BUFSIZE * 16)	/* 256 KB */
109 #define HAMMER_MAX_VOL_JUNK	HAMMER_MIN_VOL_JUNK
110 #define HAMMER_VOL_JUNK_SIZE	HAMMER_MIN_VOL_JUNK
111 
112 /*
113  * Hammer transaction ids are 64 bit unsigned integers and are usually
114  * synchronized with the time of day in nanoseconds.
115  *
116  * Hammer offsets are used for FIFO indexing and embed a cycle counter
117  * and volume number in addition to the offset.  Most offsets are required
118  * to be 16 KB aligned.
119  */
120 typedef uint64_t hammer_tid_t;
121 typedef uint64_t hammer_off_t;
122 typedef uint32_t hammer_crc_t;
123 typedef uuid_t hammer_uuid_t;
124 
125 #define HAMMER_MIN_TID		0ULL			/* unsigned */
126 #define HAMMER_MAX_TID		0xFFFFFFFFFFFFFFFFULL	/* unsigned */
127 #define HAMMER_MIN_KEY		-0x8000000000000000LL	/* signed */
128 #define HAMMER_MAX_KEY		0x7FFFFFFFFFFFFFFFLL	/* signed */
129 #define HAMMER_MIN_OBJID	HAMMER_MIN_KEY		/* signed */
130 #define HAMMER_MAX_OBJID	HAMMER_MAX_KEY		/* signed */
131 #define HAMMER_MIN_RECTYPE	0x0U			/* unsigned */
132 #define HAMMER_MAX_RECTYPE	0xFFFFU			/* unsigned */
133 #define HAMMER_MIN_OFFSET	0ULL			/* unsigned */
134 #define HAMMER_MAX_OFFSET	0xFFFFFFFFFFFFFFFFULL	/* unsigned */
135 
136 /*
137  * hammer_off_t has several different encodings.  Note that not all zones
138  * encode a vol_no.  Zone bits are not a part of filesystem capacity as
139  * the zone bits aren't directly or indirectly mapped to physical volumes.
140  *
141  * In other words, HAMMER's logical filesystem offset consists of 64 bits,
142  * but the filesystem is considered 60 bits filesystem, not 64 bits.
143  * The maximum filesystem capacity is 1EB, not 16EB.
144  *
145  * zone 0:		available, a big-block that contains the offset is unused
146  * zone 1 (z,v,o):	raw volume relative (offset 0 is the volume header)
147  * zone 2 (z,v,o):	raw buffer relative (offset 0 is the first buffer)
148  * zone 3 (z,o):	undo/redo fifo	- fixed zone-2 offset array in volume header
149  * zone 4 (z,v,o):	freemap		- only real blockmap
150  * zone 8 (z,v,o):	B-Tree		- actually zone-2 address
151  * zone 9 (z,v,o):	meta		- actually zone-2 address
152  * zone 10 (z,v,o):	large-data	- actually zone-2 address
153  * zone 11 (z,v,o):	small-data	- actually zone-2 address
154  * zone 15:		unavailable, usually the offset is beyond volume size
155  *
156  * layer1/layer2 direct map:
157  *	     Maximum HAMMER filesystem capacity from volume aspect
158  *	     2^8(max volumes) * 2^52(max volume size) = 2^60 = 1EB (long offset)
159  *	    <------------------------------------------------------------->
160  *	     8bits   52bits (short offset)
161  *	    <------><----------------------------------------------------->
162  *	zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo
163  *	----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo
164  *	    <-----------------><------------------><---------------------->
165  *	     18bits             19bits              23bits
166  *	    <------------------------------------------------------------->
167  *	     2^18(layer1) * 2^19(layer2) * 2^23(big-block) = 2^60 = 1EB
168  *	     Maximum HAMMER filesystem capacity from blockmap aspect
169  *
170  * volume#0 layout
171  *	+-------------------------> offset 0 of a device/partition
172  *	| volume header (1928 bytes)
173  *	| the rest of header junk space (HAMMER_BUFSIZE aligned)
174  *	+-------------------------> vol_bot_beg
175  *	| boot area (HAMMER_BUFSIZE aligned)
176  *	+-------------------------> vol_mem_beg
177  *	| memory log (HAMMER_BUFSIZE aligned)
178  *	+-------------------------> vol_buf_beg (physical offset of zone-2)
179  *	| zone-4 big-block for layer1
180  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE
181  *	| zone-4 big-blocks for layer2
182  *	| ... (1 big-block per 4TB space)
183  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
184  *	| zone-3 big-blocks for UNDO/REDO FIFO
185  *	| ... (max 128 big-blocks)
186  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
187  *	| zone-8 big-block for root B-Tree node/etc
188  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
189  *	| zone-9 big-block for root inode/PFS/etc
190  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
191  *	| zone-X big-blocks
192  *	| ... (big-blocks for new zones after newfs_hammer)
193  *	| ...
194  *	| ...
195  *	| ...
196  *	| ...
197  *	+-------------------------> vol_buf_end (HAMMER_BUFSIZE aligned)
198  *	+-------------------------> end of a device/partition
199  *
200  * volume#N layout (0<N<256)
201  *	+-------------------------> offset 0 of a device/partition
202  *	| volume header (1928 bytes)
203  *	| the rest of header junk space (HAMMER_BUFSIZE aligned)
204  *	+-------------------------> vol_bot_beg
205  *	| boot area (HAMMER_BUFSIZE aligned)
206  *	+-------------------------> vol_mem_beg
207  *	| memory log (HAMMER_BUFSIZE aligned)
208  *	+-------------------------> vol_buf_beg (physical offset of zone-2)
209  *	| zone-4 big-blocks for layer2
210  *	| ... (1 big-block per 4TB space)
211  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
212  *	| zone-X big-blocks
213  *	| ... (unused until volume#(N-1) runs out of space)
214  *	| ...
215  *	| ...
216  *	| ...
217  *	| ...
218  *	+-------------------------> vol_buf_end (HAMMER_BUFSIZE aligned)
219  *	+-------------------------> end of a device/partition
220  */
221 
222 #define HAMMER_ZONE_RAW_VOLUME		0x1000000000000000ULL
223 #define HAMMER_ZONE_RAW_BUFFER		0x2000000000000000ULL
224 #define HAMMER_ZONE_UNDO		0x3000000000000000ULL
225 #define HAMMER_ZONE_FREEMAP		0x4000000000000000ULL
226 #define HAMMER_ZONE_RESERVED05		0x5000000000000000ULL  /* not used */
227 #define HAMMER_ZONE_RESERVED06		0x6000000000000000ULL  /* not used */
228 #define HAMMER_ZONE_RESERVED07		0x7000000000000000ULL  /* not used */
229 #define HAMMER_ZONE_BTREE		0x8000000000000000ULL
230 #define HAMMER_ZONE_META		0x9000000000000000ULL
231 #define HAMMER_ZONE_LARGE_DATA		0xA000000000000000ULL
232 #define HAMMER_ZONE_SMALL_DATA		0xB000000000000000ULL
233 #define HAMMER_ZONE_RESERVED0C		0xC000000000000000ULL  /* not used */
234 #define HAMMER_ZONE_RESERVED0D		0xD000000000000000ULL  /* not used */
235 #define HAMMER_ZONE_RESERVED0E		0xE000000000000000ULL  /* not used */
236 #define HAMMER_ZONE_UNAVAIL		0xF000000000000000ULL
237 
238 #define HAMMER_ZONE_RAW_VOLUME_INDEX	1
239 #define HAMMER_ZONE_RAW_BUFFER_INDEX	2
240 #define HAMMER_ZONE_UNDO_INDEX		3
241 #define HAMMER_ZONE_FREEMAP_INDEX	4
242 #define HAMMER_ZONE_BTREE_INDEX		8
243 #define HAMMER_ZONE_META_INDEX		9
244 #define HAMMER_ZONE_LARGE_DATA_INDEX	10
245 #define HAMMER_ZONE_SMALL_DATA_INDEX	11
246 #define HAMMER_ZONE_UNAVAIL_INDEX	15
247 
248 #define HAMMER_MAX_ZONES		16
249 
250 #define HAMMER_ZONE(offset)		((offset) & HAMMER_OFF_ZONE_MASK)
251 
252 #define hammer_is_zone_raw_volume(offset)		\
253 	(HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_VOLUME)
254 #define hammer_is_zone_raw_buffer(offset)		\
255 	(HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_BUFFER)
256 #define hammer_is_zone_undo(offset)			\
257 	(HAMMER_ZONE(offset) == HAMMER_ZONE_UNDO)
258 #define hammer_is_zone_freemap(offset)			\
259 	(HAMMER_ZONE(offset) == HAMMER_ZONE_FREEMAP)
260 #define hammer_is_zone_btree(offset)			\
261 	(HAMMER_ZONE(offset) == HAMMER_ZONE_BTREE)
262 #define hammer_is_zone_meta(offset)			\
263 	(HAMMER_ZONE(offset) == HAMMER_ZONE_META)
264 #define hammer_is_zone_large_data(offset)		\
265 	(HAMMER_ZONE(offset) == HAMMER_ZONE_LARGE_DATA)
266 #define hammer_is_zone_small_data(offset)		\
267 	(HAMMER_ZONE(offset) == HAMMER_ZONE_SMALL_DATA)
268 #define hammer_is_zone_unavail(offset)			\
269 	(HAMMER_ZONE(offset) == HAMMER_ZONE_UNAVAIL)
270 #define hammer_is_zone_data(offset)			\
271 	(hammer_is_zone_large_data(offset) || hammer_is_zone_small_data(offset))
272 
273 #define hammer_is_index_record(zone)			\
274 	((zone) >= HAMMER_ZONE_BTREE_INDEX &&		\
275 	 (zone) < HAMMER_MAX_ZONES)
276 
277 #define hammer_is_zone_record(offset)			\
278 	hammer_is_index_record(HAMMER_ZONE_DECODE(offset))
279 
280 #define hammer_is_index_direct_xlated(zone)		\
281 	(((zone) == HAMMER_ZONE_RAW_BUFFER_INDEX) ||	\
282 	 ((zone) == HAMMER_ZONE_FREEMAP_INDEX) ||	\
283 	 hammer_is_index_record(zone))
284 
285 #define hammer_is_zone_direct_xlated(offset)		\
286 	hammer_is_index_direct_xlated(HAMMER_ZONE_DECODE(offset))
287 
288 #define HAMMER_ZONE_ENCODE(zone, ham_off)		\
289 	(((hammer_off_t)(zone) << 60) | (ham_off))
290 #define HAMMER_ZONE_DECODE(ham_off)			\
291 	((int)(((hammer_off_t)(ham_off) >> 60)))
292 
293 #define HAMMER_VOL_ENCODE(vol_no)			\
294 	((hammer_off_t)((vol_no) & 255) << 52)
295 #define HAMMER_VOL_DECODE(ham_off)			\
296 	((int)(((hammer_off_t)(ham_off) >> 52) & 255))
297 
298 #define HAMMER_OFF_SHORT_ENCODE(offset)			\
299 	((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
300 #define HAMMER_OFF_LONG_ENCODE(offset)			\
301 	((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
302 
303 #define HAMMER_ENCODE(zone, vol_no, offset)		\
304 	(((hammer_off_t)(zone) << 60) |			\
305 	HAMMER_VOL_ENCODE(vol_no) |			\
306 	HAMMER_OFF_SHORT_ENCODE(offset))
307 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset)	\
308 	HAMMER_ENCODE(HAMMER_ZONE_RAW_VOLUME_INDEX, vol_no, offset)
309 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset)	\
310 	HAMMER_ENCODE(HAMMER_ZONE_RAW_BUFFER_INDEX, vol_no, offset)
311 #define HAMMER_ENCODE_UNDO(offset)			\
312 	HAMMER_ENCODE(HAMMER_ZONE_UNDO_INDEX, HAMMER_ROOT_VOLNO, offset)
313 #define HAMMER_ENCODE_FREEMAP(vol_no, offset)		\
314 	HAMMER_ENCODE(HAMMER_ZONE_FREEMAP_INDEX, vol_no, offset)
315 
316 /*
317  * Translate a zone address to zone-X address.
318  */
319 #define hammer_xlate_to_zoneX(zone, offset)		\
320 	HAMMER_ZONE_ENCODE((zone), (offset) & ~HAMMER_OFF_ZONE_MASK)
321 #define hammer_xlate_to_zone2(offset)			\
322 	hammer_xlate_to_zoneX(HAMMER_ZONE_RAW_BUFFER_INDEX, (offset))
323 
324 #define hammer_data_zone(data_len)			\
325 	(((data_len) >= HAMMER_BUFSIZE) ?		\
326 	 HAMMER_ZONE_LARGE_DATA :			\
327 	 HAMMER_ZONE_SMALL_DATA)
328 #define hammer_data_zone_index(data_len)		\
329 	(((data_len) >= HAMMER_BUFSIZE) ?		\
330 	 HAMMER_ZONE_LARGE_DATA_INDEX :			\
331 	 HAMMER_ZONE_SMALL_DATA_INDEX)
332 
333 /*
334  * Big-Block backing store
335  *
336  * A blockmap is a two-level map which translates a blockmap-backed zone
337  * offset into a raw zone 2 offset.  The layer 1 handles 18 bits and the
338  * layer 2 handles 19 bits.  The 8M big-block size is 23 bits so two
339  * layers gives us 18+19+23 = 60 bits of address space.
340  *
341  * When using hinting for a blockmap lookup, the hint is lost when the
342  * scan leaves the HINTBLOCK, which is typically several BIGBLOCK's.
343  * HINTBLOCK is a heuristic.
344  */
345 #define HAMMER_HINTBLOCK_SIZE		(HAMMER_BIGBLOCK_SIZE * 4)
346 #define HAMMER_HINTBLOCK_MASK64		((uint64_t)HAMMER_HINTBLOCK_SIZE - 1)
347 #define HAMMER_BIGBLOCK_SIZE		(8192 * 1024)
348 #define HAMMER_BIGBLOCK_SIZE64		((uint64_t)HAMMER_BIGBLOCK_SIZE)
349 #define HAMMER_BIGBLOCK_MASK		(HAMMER_BIGBLOCK_SIZE - 1)
350 #define HAMMER_BIGBLOCK_MASK64		((uint64_t)HAMMER_BIGBLOCK_SIZE - 1)
351 #define HAMMER_BIGBLOCK_BITS		23
352 #if 0
353 #define HAMMER_BIGBLOCK_OVERFILL	(6144 * 1024)
354 #endif
355 #if (1 << HAMMER_BIGBLOCK_BITS) != HAMMER_BIGBLOCK_SIZE
356 #error "HAMMER_BIGBLOCK_BITS BROKEN"
357 #endif
358 
359 #define HAMMER_BUFFERS_PER_BIGBLOCK			\
360 	(HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE)
361 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK		\
362 	(HAMMER_BUFFERS_PER_BIGBLOCK - 1)
363 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK64		\
364 	((hammer_off_t)HAMMER_BUFFERS_PER_BIGBLOCK_MASK)
365 
366 #define HAMMER_BIGBLOCK_DOALIGN(offset)				\
367 	(((offset) + HAMMER_BIGBLOCK_MASK64) & ~HAMMER_BIGBLOCK_MASK64)
368 
369 /*
370  * Maximum number of mirrors operating in master mode (multi-master
371  * clustering and mirroring). Note that HAMMER1 does not support
372  * multi-master clustering as of 2015.
373  */
374 #define HAMMER_MAX_MASTERS		16
375 
376 /*
377  * The blockmap is somewhat of a degenerate structure.  HAMMER only actually
378  * uses it in its original incarnation to implement the freemap.
379  *
380  * zone:1	raw volume (no blockmap)
381  * zone:2	raw buffer (no blockmap)
382  * zone:3	undomap    (direct layer2 array in volume header)
383  * zone:4	freemap    (the only real blockmap)
384  * zone:8-15	zone id used to classify big-block only, address is actually
385  *		a zone-2 address.
386  */
387 typedef struct hammer_blockmap {
388 	hammer_off_t	phys_offset;  /* zone-2 offset only used by zone-4 */
389 	hammer_off_t	first_offset; /* zone-X offset only used by zone-3 */
390 	hammer_off_t	next_offset;  /* zone-X offset for allocation */
391 	hammer_off_t	alloc_offset; /* zone-X offset only used by zone-3 */
392 	uint32_t	reserved01;
393 	hammer_crc_t	entry_crc;
394 } *hammer_blockmap_t;
395 
396 #define HAMMER_BLOCKMAP_CRCSIZE	\
397 	offsetof(struct hammer_blockmap, entry_crc)
398 
399 /*
400  * The blockmap is a 2-layer entity made up of big-blocks.  The first layer
401  * contains 262144 32-byte entries (18 bits), the second layer contains
402  * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
403  * 18+19+23 = 60 bits.  The top four bits are the zone id.
404  *
405  * Currently only the freemap utilizes both layers in all their glory.
406  * All primary data/meta-data zones actually encode a zone-2 address
407  * requiring no real blockmap translation.
408  *
409  * The freemap uses the upper 8 bits of layer-1 to identify the volume,
410  * thus any space allocated via the freemap can be directly translated
411  * to a zone:2 (or zone:8-15) address.
412  *
413  * zone-X blockmap offset: [zone:4][layer1:18][layer2:19][big-block:23]
414  */
415 
416 /*
417  * 32 bytes layer1 entry for 8MB big-block.
418  * A big-block can hold 2^23 / 2^5 = 2^18 layer1 entries,
419  * which equals bits assigned for layer1 in zone-2 address.
420  */
421 typedef struct hammer_blockmap_layer1 {
422 	hammer_off_t	blocks_free;	/* big-blocks free */
423 	hammer_off_t	phys_offset;	/* UNAVAIL or zone-2 */
424 	hammer_off_t	reserved01;
425 	hammer_crc_t	layer2_crc;	/* xor'd crc's of HAMMER_BLOCKSIZE */
426 					/* (not yet used) */
427 	hammer_crc_t	layer1_crc;	/* MUST BE LAST FIELD OF STRUCTURE*/
428 } *hammer_blockmap_layer1_t;
429 
430 #define HAMMER_LAYER1_CRCSIZE	\
431 	offsetof(struct hammer_blockmap_layer1, layer1_crc)
432 
433 /*
434  * 16 bytes layer2 entry for 8MB big-blocks.
435  * A big-block can hold 2^23 / 2^4 = 2^19 layer2 entries,
436  * which equals bits assigned for layer2 in zone-2 address.
437  *
438  * NOTE: bytes_free is signed and can legally go negative if/when data
439  *	 de-dup occurs.  This field will never go higher than
440  *	 HAMMER_BIGBLOCK_SIZE.  If exactly HAMMER_BIGBLOCK_SIZE
441  *	 the big-block is completely free.
442  */
443 typedef struct hammer_blockmap_layer2 {
444 	uint8_t		zone;		/* typed allocation zone */
445 	uint8_t		reserved01;
446 	uint16_t	reserved02;
447 	uint32_t	append_off;	/* allocatable space index */
448 	int32_t		bytes_free;	/* bytes free within this big-block */
449 	hammer_crc_t	entry_crc;
450 } *hammer_blockmap_layer2_t;
451 
452 #define HAMMER_LAYER2_CRCSIZE	\
453 	offsetof(struct hammer_blockmap_layer2, entry_crc)
454 
455 #define HAMMER_BLOCKMAP_UNAVAIL	((hammer_off_t)-1LL)
456 
457 #define HAMMER_BLOCKMAP_RADIX1	/* 2^18 = 262144 */	\
458 	((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1)))
459 #define HAMMER_BLOCKMAP_RADIX2	/* 2^19 = 524288 */	\
460 	((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2)))
461 
462 #define HAMMER_BLOCKMAP_LAYER1	/* 2^(18+19+23) = 1EB */	\
463 	(HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
464 #define HAMMER_BLOCKMAP_LAYER2	/* 2^(19+23) = 4TB */		\
465 	(HAMMER_BLOCKMAP_RADIX2 * HAMMER_BIGBLOCK_SIZE64)
466 
467 #define HAMMER_BLOCKMAP_LAYER1_MASK	(HAMMER_BLOCKMAP_LAYER1 - 1)
468 #define HAMMER_BLOCKMAP_LAYER2_MASK	(HAMMER_BLOCKMAP_LAYER2 - 1)
469 
470 #define HAMMER_BLOCKMAP_LAYER2_DOALIGN(offset)			\
471 	(((offset) + HAMMER_BLOCKMAP_LAYER2_MASK) &		\
472 	 ~HAMMER_BLOCKMAP_LAYER2_MASK)
473 
474 /*
475  * Index within layer1 or layer2 big-block for the entry representing
476  * a zone-2 physical offset.
477  */
478 #define HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset)		\
479 	((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) /	\
480 	 HAMMER_BLOCKMAP_LAYER2))
481 
482 #define HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset)		\
483 	((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) /	\
484 	HAMMER_BIGBLOCK_SIZE64))
485 
486 /*
487  * Byte offset within layer1 or layer2 big-block for the entry representing
488  * a zone-2 physical offset.  Multiply the index by sizeof(blockmap_layer).
489  */
490 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset)		\
491 	(HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) *		\
492 	 sizeof(struct hammer_blockmap_layer1))
493 
494 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset)		\
495 	(HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) *		\
496 	 sizeof(struct hammer_blockmap_layer2))
497 
498 /*
499  * Move on to offset 0 of the next layer1 or layer2.
500  */
501 #define HAMMER_ZONE_LAYER1_NEXT_OFFSET(offset)			\
502 	(((offset) + HAMMER_BLOCKMAP_LAYER2) & ~HAMMER_BLOCKMAP_LAYER2_MASK)
503 
504 #define HAMMER_ZONE_LAYER2_NEXT_OFFSET(offset)			\
505 	(((offset) + HAMMER_BIGBLOCK_SIZE) & ~HAMMER_BIGBLOCK_MASK64)
506 
507 /*
508  * HAMMER UNDO parameters.  The UNDO fifo is mapped directly in the volume
509  * header with an array of zone-2 offsets.  A maximum of (128x8MB) = 1GB,
510  * and minimum of (64x8MB) = 512MB may be reserved.  The size of the undo
511  * fifo is usually set a newfs time.
512  */
513 #define HAMMER_MIN_UNDO_BIGBLOCKS		64
514 #define HAMMER_MAX_UNDO_BIGBLOCKS		128
515 
516 /*
517  * All on-disk HAMMER structures which make up elements of the UNDO FIFO
518  * contain a hammer_fifo_head and hammer_fifo_tail structure.  This structure
519  * contains all the information required to validate the fifo element
520  * and to scan the fifo in either direction.  The head is typically embedded
521  * in higher level hammer on-disk structures while the tail is typically
522  * out-of-band.  hdr_size is the size of the whole mess, including the tail.
523  *
524  * All undo structures are guaranteed to not cross a 16K filesystem
525  * buffer boundary.  Most undo structures are fairly small.  Data spaces
526  * are not immediately reused by HAMMER so file data is not usually recorded
527  * as part of an UNDO.
528  *
529  * PAD elements are allowed to take up only 8 bytes of space as a special
530  * case, containing only hdr_signature, hdr_type, and hdr_size fields,
531  * and with the tail overloaded onto the head structure for 8 bytes total.
532  *
533  * Every undo record has a sequence number.  This number is unrelated to
534  * transaction ids and instead collects the undo transactions associated
535  * with a single atomic operation.  A larger transactional operation, such
536  * as a remove(), may consist of several smaller atomic operations
537  * representing raw meta-data operations.
538  *
539  *				HAMMER VERSION 4 CHANGES
540  *
541  * In HAMMER version 4 the undo structure alignment is reduced from 16384
542  * to 512 bytes in order to ensure that each 512 byte sector begins with
543  * a header.  The hdr_seq field in the header is a 32 bit sequence number
544  * which allows the recovery code to detect missing sectors
545  * without relying on the 32-bit crc and to definitively identify the current
546  * undo sequence space without having to rely on information from the volume
547  * header.  In addition, new REDO entries in the undo space are used to
548  * record write, write/extend, and transaction id updates.
549  *
550  * The grand result is:
551  *
552  * (1) The volume header no longer needs to be synchronized for most
553  *     flush and fsync operations.
554  *
555  * (2) Most fsync operations need only lay down REDO records
556  *
557  * (3) Data overwrite for nohistory operations covered by REDO records
558  *     can be supported (instead of rolling a new block allocation),
559  *     by rolling UNDO for the prior contents of the data.
560  *
561  *				HAMMER VERSION 5 CHANGES
562  *
563  * Hammer version 5 contains a minor adjustment making layer2's bytes_free
564  * field signed, allowing dedup to push it into the negative domain.
565  */
566 #define HAMMER_HEAD_ALIGN		8
567 #define HAMMER_HEAD_ALIGN_MASK		(HAMMER_HEAD_ALIGN - 1)
568 #define HAMMER_HEAD_DOALIGN(bytes)	\
569 	(((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK)
570 
571 #define HAMMER_UNDO_ALIGN		512
572 #define HAMMER_UNDO_ALIGN64		((uint64_t)512)
573 #define HAMMER_UNDO_MASK		(HAMMER_UNDO_ALIGN - 1)
574 #define HAMMER_UNDO_MASK64		(HAMMER_UNDO_ALIGN64 - 1)
575 #define HAMMER_UNDO_DOALIGN(offset)	\
576 	(((offset) + HAMMER_UNDO_MASK) & ~HAMMER_UNDO_MASK64)
577 
578 typedef struct hammer_fifo_head {
579 	uint16_t hdr_signature;
580 	uint16_t hdr_type;
581 	uint32_t hdr_size;	/* Aligned size of the whole mess */
582 	uint32_t hdr_seq;	/* Sequence number */
583 	hammer_crc_t hdr_crc;	/* XOR crc up to field w/ crc after field */
584 } *hammer_fifo_head_t;
585 
586 #define HAMMER_FIFO_HEAD_CRCOFF	offsetof(struct hammer_fifo_head, hdr_crc)
587 
588 typedef struct hammer_fifo_tail {
589 	uint16_t tail_signature;
590 	uint16_t tail_type;
591 	uint32_t tail_size;	/* aligned size of the whole mess */
592 } *hammer_fifo_tail_t;
593 
594 /*
595  * Fifo header types.
596  *
597  * NOTE: 0x8000U part of HAMMER_HEAD_TYPE_PAD can be removed if the HAMMER
598  * version ever gets bumped again. It exists only to keep compatibility with
599  * older versions.
600  */
601 #define HAMMER_HEAD_TYPE_PAD	(0x0040U | 0x8000U)
602 #define HAMMER_HEAD_TYPE_DUMMY	0x0041U		/* dummy entry w/seqno */
603 #define HAMMER_HEAD_TYPE_UNDO	0x0043U		/* random UNDO information */
604 #define HAMMER_HEAD_TYPE_REDO	0x0044U		/* data REDO / fast fsync */
605 
606 #define HAMMER_HEAD_SIGNATURE	0xC84EU
607 #define HAMMER_TAIL_SIGNATURE	0xC74FU
608 
609 /*
610  * Misc FIFO structures.
611  *
612  * UNDO - Raw meta-data media updates.
613  */
614 typedef struct hammer_fifo_undo {
615 	struct hammer_fifo_head	head;
616 	hammer_off_t		undo_offset;	/* zone-1,2 offset */
617 	int32_t			undo_data_bytes;
618 	int32_t			undo_reserved01;
619 	/* followed by data */
620 } *hammer_fifo_undo_t;
621 
622 /*
623  * REDO (HAMMER version 4+) - Logical file writes/truncates.
624  *
625  * REDOs contain information which will be duplicated in a later meta-data
626  * update, allowing fast write()+fsync() operations.  REDOs can be ignored
627  * without harming filesystem integrity but must be processed if fsync()
628  * semantics are desired.
629  *
630  * Unlike UNDOs which are processed backwards within the recovery span,
631  * REDOs must be processed forwards starting further back (starting outside
632  * the recovery span).
633  *
634  *	WRITE	- Write logical file (with payload).  Executed both
635  *		  out-of-span and in-span.  Out-of-span WRITEs may be
636  *		  filtered out by TERMs.
637  *
638  *	TRUNC	- Truncate logical file (no payload).  Executed both
639  *		  out-of-span and in-span.  Out-of-span WRITEs may be
640  *		  filtered out by TERMs.
641  *
642  *	TERM_*	- Indicates meta-data was committed (if out-of-span) or
643  *		  will be rolled-back (in-span).  Any out-of-span TERMs
644  *		  matching earlier WRITEs remove those WRITEs from
645  *		  consideration as they might conflict with a later data
646  *		  commit (which is not being rolled-back).
647  *
648  *	SYNC	- The earliest in-span SYNC (the last one when scanning
649  *		  backwards) tells the recovery code how far out-of-span
650  *		  it must go to run REDOs.
651  *
652  * NOTE: WRITEs do not always have matching TERMs even under
653  *	 perfect conditions because truncations might remove the
654  *	 buffers from consideration.  I/O problems can also remove
655  *	 buffers from consideration.
656  *
657  *	 TRUNCSs do not always have matching TERMs because several
658  *	 truncations may be aggregated together into a single TERM.
659  */
660 typedef struct hammer_fifo_redo {
661 	struct hammer_fifo_head	head;
662 	int64_t			redo_objid;	/* file being written */
663 	hammer_off_t		redo_offset;	/* logical offset in file */
664 	int32_t			redo_data_bytes;
665 	uint32_t		redo_flags;
666 	uint32_t		redo_localization;
667 	uint32_t		redo_reserved01;
668 	uint64_t		redo_reserved02;
669 	/* followed by data */
670 } *hammer_fifo_redo_t;
671 
672 #define HAMMER_REDO_WRITE	0x00000001
673 #define HAMMER_REDO_TRUNC	0x00000002
674 #define HAMMER_REDO_TERM_WRITE	0x00000004
675 #define HAMMER_REDO_TERM_TRUNC	0x00000008
676 #define HAMMER_REDO_SYNC	0x00000010
677 
678 typedef union hammer_fifo_any {
679 	struct hammer_fifo_head	head;
680 	struct hammer_fifo_undo	undo;
681 	struct hammer_fifo_redo	redo;
682 } *hammer_fifo_any_t;
683 
684 /*
685  * Volume header types
686  */
687 #define HAMMER_FSBUF_VOLUME	0xC8414D4DC5523031ULL	/* HAMMER01 */
688 #define HAMMER_FSBUF_VOLUME_REV	0x313052C54D4D41C8ULL	/* (reverse endian) */
689 
690 /*
691  * HAMMER Volume header
692  *
693  * A HAMMER filesystem can be built from 1-256 block devices, each block
694  * device contains a volume header followed by however many buffers fit
695  * into the volume.
696  *
697  * One of the volumes making up a HAMMER filesystem is the root volume.
698  * The root volume is always volume #0 which is the first block device path
699  * specified by newfs_hammer(8).  All HAMMER volumes have a volume header,
700  * however the root volume may be the only volume that has valid values for
701  * some fields in the header.
702  *
703  * Special field notes:
704  *
705  *	vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
706  *	vol_mem_beg - offset of memory log (buf_beg - mem_beg bytes)
707  *	vol_buf_beg - offset of the first buffer in volume
708  *	vol_buf_end - offset of volume EOF (on buffer boundary)
709  *
710  *	The memory log area allows a kernel to cache new records and data
711  *	in memory without allocating space in the actual filesystem to hold
712  *	the records and data.  In the event that a filesystem becomes full,
713  *	any records remaining in memory can be flushed to the memory log
714  *	area.  This allows the kernel to immediately return success.
715  *
716  *	The buffer offset is a physical offset of zone-2 offset. The lower
717  *	52 bits of the zone-2 offset is added to the buffer offset of each
718  *	volume to generate an actual I/O offset within the block device.
719  *
720  *	NOTE: boot area and memory log are currently not used.
721  */
722 
723 /*
724  * Filesystem type string
725  */
726 #define HAMMER_FSTYPE_STRING		"DragonFly HAMMER"
727 
728 /*
729  * These macros are only used by userspace when userspace commands either
730  * initialize or add a new HAMMER volume.
731  */
732 #define HAMMER_BOOT_MINBYTES		(32*1024)
733 #define HAMMER_BOOT_NOMBYTES		(64LL*1024*1024)
734 #define HAMMER_BOOT_MAXBYTES		(256LL*1024*1024)
735 
736 #define HAMMER_MEM_MINBYTES		(256*1024)
737 #define HAMMER_MEM_NOMBYTES		(1LL*1024*1024*1024)
738 #define HAMMER_MEM_MAXBYTES		(64LL*1024*1024*1024)
739 
740 typedef struct hammer_volume_ondisk {
741 	uint64_t vol_signature;	/* HAMMER_FSBUF_VOLUME for a valid header */
742 
743 	/*
744 	 * These are relative to block device offset, not zone offsets.
745 	 */
746 	int64_t vol_bot_beg;	/* offset of boot area */
747 	int64_t vol_mem_beg;	/* offset of memory log */
748 	int64_t vol_buf_beg;	/* offset of the first buffer in volume */
749 	int64_t vol_buf_end;	/* offset of volume EOF (on buffer boundary) */
750 	int64_t vol_reserved01;
751 
752 	hammer_uuid_t vol_fsid;	/* identify filesystem */
753 	hammer_uuid_t vol_fstype; /* identify filesystem type */
754 	char vol_label[64];	/* filesystem label */
755 
756 	int32_t vol_no;		/* volume number within filesystem */
757 	int32_t vol_count;	/* number of volumes making up filesystem */
758 
759 	uint32_t vol_version;	/* version control information */
760 	hammer_crc_t vol_crc;	/* header crc */
761 	uint32_t vol_flags;	/* volume flags */
762 	uint32_t vol_rootvol;	/* the root volume number (must be 0) */
763 
764 	uint32_t vol_reserved[8];
765 
766 	/*
767 	 * These fields are initialized and space is reserved in every
768 	 * volume making up a HAMMER filesystem, but only the root volume
769 	 * contains valid data.  Note that vol0_stat_bigblocks does not
770 	 * include big-blocks for freemap and undomap initially allocated
771 	 * by newfs_hammer(8).
772 	 */
773 	int64_t vol0_stat_bigblocks;	/* total big-blocks when fs is empty */
774 	int64_t vol0_stat_freebigblocks;/* number of free big-blocks */
775 	int64_t	vol0_reserved01;
776 	int64_t vol0_stat_inodes;	/* for statfs only */
777 	int64_t vol0_reserved02;
778 	hammer_off_t vol0_btree_root;	/* B-Tree root offset in zone-8 */
779 	hammer_tid_t vol0_next_tid;	/* highest partially synchronized TID */
780 	hammer_off_t vol0_reserved03;
781 
782 	/*
783 	 * Blockmaps for zones.  Not all zones use a blockmap.  Note that
784 	 * the entire root blockmap is cached in the hammer_mount structure.
785 	 */
786 	struct hammer_blockmap	vol0_blockmap[HAMMER_MAX_ZONES];
787 
788 	/*
789 	 * Array of zone-2 addresses for undo FIFO.
790 	 */
791 	hammer_off_t		vol0_undo_array[HAMMER_MAX_UNDO_BIGBLOCKS];
792 } *hammer_volume_ondisk_t;
793 
794 #define HAMMER_ROOT_VOLNO		0
795 
796 #define HAMMER_VOLF_NEEDFLUSH		0x0004	/* volume needs flush */
797 
798 #define HAMMER_VOL_CRCSIZE1	\
799 	offsetof(struct hammer_volume_ondisk, vol_crc)
800 #define HAMMER_VOL_CRCSIZE2	\
801 	(sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 -	\
802 	 sizeof(hammer_crc_t))
803 
804 #define HAMMER_VOL_VERSION_MIN		1	/* minimum supported version */
805 #define HAMMER_VOL_VERSION_DEFAULT	7	/* newfs default version */
806 #define HAMMER_VOL_VERSION_WIP		8	/* version >= this is WIP */
807 #define HAMMER_VOL_VERSION_MAX		7	/* maximum supported version */
808 
809 #define HAMMER_VOL_VERSION_ONE		1
810 #define HAMMER_VOL_VERSION_TWO		2	/* new dirent layout (2.3+) */
811 #define HAMMER_VOL_VERSION_THREE	3	/* new snapshot layout (2.5+) */
812 #define HAMMER_VOL_VERSION_FOUR		4	/* new undo/flush (2.5+) */
813 #define HAMMER_VOL_VERSION_FIVE		5	/* dedup (2.9+) */
814 #define HAMMER_VOL_VERSION_SIX		6	/* DIRHASH_ALG1 */
815 #define HAMMER_VOL_VERSION_SEVEN	7	/* use the faster iscsi_crc */
816 
817 /*
818  * Translate a zone-2 address to physical address
819  */
820 #define hammer_xlate_to_phys(volume, zone2_offset)	\
821 	((volume)->vol_buf_beg + HAMMER_OFF_SHORT_ENCODE(zone2_offset))
822 
823 /*
824  * Translate a zone-3 address to zone-2 address
825  */
826 #define HAMMER_UNDO_INDEX(zone3_offset)			\
827 	(HAMMER_OFF_SHORT_ENCODE(zone3_offset) / HAMMER_BIGBLOCK_SIZE)
828 
829 #define hammer_xlate_to_undo(volume, zone3_offset)			\
830 	((volume)->vol0_undo_array[HAMMER_UNDO_INDEX(zone3_offset)] +	\
831 	 (zone3_offset & HAMMER_BIGBLOCK_MASK64))
832 
833 /*
834  * Effective per-volume filesystem capacity including big-blocks for layer1/2
835  */
836 #define HAMMER_VOL_BUF_SIZE(volume)			\
837 	((volume)->vol_buf_end - (volume)->vol_buf_beg)
838 
839 /*
840  * Record types are fairly straightforward.  The B-Tree includes the record
841  * type in its index sort.
842  */
843 #define HAMMER_RECTYPE_UNKNOWN		0x0000
844 #define HAMMER_RECTYPE_INODE		0x0001	/* inode in obj_id space */
845 #define HAMMER_RECTYPE_DATA		0x0010
846 #define HAMMER_RECTYPE_DIRENTRY		0x0011
847 #define HAMMER_RECTYPE_DB		0x0012
848 #define HAMMER_RECTYPE_EXT		0x0013	/* ext attributes */
849 #define HAMMER_RECTYPE_FIX		0x0014	/* fixed attribute */
850 #define HAMMER_RECTYPE_PFS		0x0015	/* PFS management */
851 #define HAMMER_RECTYPE_SNAPSHOT		0x0016	/* Snapshot management */
852 #define HAMMER_RECTYPE_CONFIG		0x0017	/* hammer cleanup config */
853 #define HAMMER_RECTYPE_MAX		0xFFFF
854 
855 #define HAMMER_RECTYPE_ENTRY_START	(HAMMER_RECTYPE_INODE + 1)
856 #define HAMMER_RECTYPE_CLEAN_START	HAMMER_RECTYPE_EXT
857 
858 #define HAMMER_FIXKEY_SYMLINK		1
859 
860 #define HAMMER_OBJTYPE_UNKNOWN		0	/* never exists on-disk as unknown */
861 #define HAMMER_OBJTYPE_DIRECTORY	1
862 #define HAMMER_OBJTYPE_REGFILE		2
863 #define HAMMER_OBJTYPE_DBFILE		3
864 #define HAMMER_OBJTYPE_FIFO		4
865 #define HAMMER_OBJTYPE_CDEV		5
866 #define HAMMER_OBJTYPE_BDEV		6
867 #define HAMMER_OBJTYPE_SOFTLINK		7
868 #define HAMMER_OBJTYPE_PSEUDOFS		8	/* pseudo filesystem obj */
869 #define HAMMER_OBJTYPE_SOCKET		9
870 
871 /*
872  * HAMMER inode attribute data
873  *
874  * The data reference for a HAMMER inode points to this structure.  Any
875  * modifications to the contents of this structure will result in a
876  * replacement operation.
877  *
878  * parent_obj_id is only valid for directories (which cannot be hard-linked),
879  * and specifies the parent directory obj_id.  This field will also be set
880  * for non-directory inodes as a recovery aid, but can wind up holding
881  * stale information.  However, since object id's are not reused, the worse
882  * that happens is that the recovery code is unable to use it.
883  * A parent_obj_id of 0 means it's a root inode of root or non-root PFS.
884  *
885  * NOTE: Future note on directory hardlinks.  We can implement a record type
886  * which allows us to point to multiple parent directories.
887  */
888 typedef struct hammer_inode_data {
889 	uint16_t version;	/* inode data version */
890 	uint16_t mode;		/* basic unix permissions */
891 	uint32_t uflags;	/* chflags */
892 	uint32_t rmajor;	/* used by device nodes */
893 	uint32_t rminor;	/* used by device nodes */
894 	uint64_t ctime;
895 	int64_t parent_obj_id;	/* parent directory obj_id */
896 	hammer_uuid_t uid;
897 	hammer_uuid_t gid;
898 
899 	uint8_t obj_type;
900 	uint8_t cap_flags;	/* capability support flags (extension) */
901 	uint16_t reserved01;
902 	uint32_t reserved02;
903 	uint64_t nlinks;	/* hard links */
904 	uint64_t size;		/* filesystem object size */
905 	union {
906 		char	symlink[24];	/* HAMMER_INODE_BASESYMLEN */
907 	} ext;
908 	uint64_t mtime;	/* mtime must be second-to-last */
909 	uint64_t atime;	/* atime must be last */
910 } *hammer_inode_data_t;
911 
912 /*
913  * Neither mtime nor atime upates are CRCd by the B-Tree element.
914  * mtime updates have UNDO, atime updates do not.
915  */
916 #define HAMMER_INODE_CRCSIZE	\
917 	offsetof(struct hammer_inode_data, mtime)
918 
919 #define HAMMER_INODE_DATA_VERSION	1
920 #define HAMMER_OBJID_ROOT		1	/* root inodes # */
921 #define HAMMER_INODE_BASESYMLEN		24	/* see ext.symlink */
922 
923 /*
924  * Capability & implementation flags.
925  *
926  * HAMMER_INODE_CAP_DIR_LOCAL_INO - Use inode B-Tree localization
927  * for directory entries.  Also see HAMMER_DIR_INODE_LOCALIZATION().
928  */
929 #define HAMMER_INODE_CAP_DIRHASH_MASK	0x03	/* directory: hash algorithm */
930 #define HAMMER_INODE_CAP_DIRHASH_ALG0	0x00
931 #define HAMMER_INODE_CAP_DIRHASH_ALG1	0x01
932 #define HAMMER_INODE_CAP_DIRHASH_ALG2	0x02
933 #define HAMMER_INODE_CAP_DIRHASH_ALG3	0x03
934 #define HAMMER_INODE_CAP_DIR_LOCAL_INO	0x04	/* use inode localization */
935 
936 #define HAMMER_DATA_DOALIGN(offset)				\
937 	(((offset) + 15) & ~15)
938 #define HAMMER_DATA_DOALIGN_WITH(type, offset)			\
939 	(((type)(offset) + 15) & (~(type)15))
940 
941 /*
942  * A HAMMER directory entry associates a HAMMER filesystem object with a
943  * namespace.  It is hooked into a pseudo-filesystem (with its own inode
944  * numbering space) in the filesystem by setting the high 16 bits of the
945  * localization field.  The low 16 bits must be 0 and are reserved for
946  * future use.
947  *
948  * Directory entries are indexed with a 128 bit namekey rather then an
949  * offset.  A portion of the namekey is an iterator/randomizer to deal
950  * with collisions.
951  *
952  * NOTE: leaf.base.obj_type from the related B-Tree leaf entry holds
953  * the filesystem object type of obj_id, e.g. a den_type equivalent.
954  * It is not stored in hammer_direntry_data.
955  *
956  * NOTE: name field / the filename data reference is NOT terminated with \0.
957  */
958 typedef struct hammer_direntry_data {
959 	int64_t obj_id;			/* object being referenced */
960 	uint32_t localization;		/* identify pseudo-filesystem */
961 	uint32_t reserved01;
962 	char	name[16];		/* name (extended) */
963 } *hammer_direntry_data_t;
964 
965 #define HAMMER_ENTRY_NAME_OFF	offsetof(struct hammer_direntry_data, name[0])
966 #define HAMMER_ENTRY_SIZE(nlen)	offsetof(struct hammer_direntry_data, name[nlen])
967 
968 /*
969  * Symlink data which does not fit in the inode is stored in a separate
970  * FIX type record.
971  */
972 typedef struct hammer_symlink_data {
973 	char	name[16];		/* name (extended) */
974 } *hammer_symlink_data_t;
975 
976 #define HAMMER_SYMLINK_NAME_OFF	offsetof(struct hammer_symlink_data, name[0])
977 
978 /*
979  * The root inode for the primary filesystem and root inode for any
980  * pseudo-fs may be tagged with an optional data structure using
981  * HAMMER_RECTYPE_PFS and localization id.  This structure allows
982  * the node to be used as a mirroring master or slave.
983  *
984  * When operating as a slave CD's into the node automatically become read-only
985  * and as-of sync_end_tid.
986  *
987  * When operating as a master the read PFSD info sets sync_end_tid to
988  * the most recently flushed TID.
989  *
990  * sync_low_tid is not yet used but will represent the highest pruning
991  * end-point, after which full history is available.
992  *
993  * We need to pack this structure making it equally sized on both 32-bit and
994  * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is
995  * send over the wire in hammer mirror operations. Only on 64-bit machines
996  * the size of this struct differ when packed or not. This leads us to the
997  * situation where old 64-bit systems (using the non-packed structure),
998  * which were never able to mirror to/from 32-bit systems, are now no longer
999  * able to mirror to/from newer 64-bit systems (using the packed structure).
1000  */
1001 struct hammer_pseudofs_data {
1002 	hammer_tid_t	sync_low_tid;	/* full history beyond this point */
1003 	hammer_tid_t	sync_beg_tid;	/* earliest tid w/ full history avail */
1004 	hammer_tid_t	sync_end_tid;	/* current synchronizatoin point */
1005 	uint64_t	sync_beg_ts;	/* real-time of last completed sync */
1006 	uint64_t	sync_end_ts;	/* initiation of current sync cycle */
1007 	hammer_uuid_t	shared_uuid;	/* shared uuid (match required) */
1008 	hammer_uuid_t	unique_uuid;	/* unique uuid of this master/slave */
1009 	int32_t		reserved01;	/* reserved for future master_id */
1010 	int32_t		mirror_flags;	/* misc flags */
1011 	char		label[64];	/* filesystem space label */
1012 	char		snapshots[64];	/* softlink dir for pruning */
1013 	int32_t		reserved02;	/* was prune_{time,freq} */
1014 	int32_t		reserved03;	/* was reblock_{time,freq} */
1015 	int32_t		reserved04;	/* was snapshot_freq */
1016 	int32_t		prune_min;	/* do not prune recent history */
1017 	int32_t		prune_max;	/* do not retain history beyond here */
1018 	int32_t		reserved[16];
1019 } __packed;
1020 
1021 typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t;
1022 
1023 #define HAMMER_PFSD_SLAVE	0x00000001
1024 #define HAMMER_PFSD_DELETED	0x80000000
1025 
1026 #define hammer_is_pfs_slave(pfsd)			\
1027 	(((pfsd)->mirror_flags & HAMMER_PFSD_SLAVE) != 0)
1028 #define hammer_is_pfs_master(pfsd)			\
1029 	(!hammer_is_pfs_slave(pfsd))
1030 #define hammer_is_pfs_deleted(pfsd)			\
1031 	(((pfsd)->mirror_flags & HAMMER_PFSD_DELETED) != 0)
1032 
1033 #define HAMMER_MAX_PFS		65536
1034 #define HAMMER_MAX_PFSID	(HAMMER_MAX_PFS - 1)
1035 #define HAMMER_ROOT_PFSID	0
1036 
1037 /*
1038  * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }.
1039  *
1040  * Snapshot records replace the old <fs>/snapshots/<softlink> methodology.  Snapshot
1041  * records are mirrored but may be independantly managed once they are laid down on
1042  * a slave.
1043  *
1044  * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the
1045  *	 results.
1046  *
1047  * NOTE: Reserved fields must be zero (as usual)
1048  */
1049 typedef struct hammer_snapshot_data {
1050 	hammer_tid_t	tid;		/* the snapshot TID itself (== key) */
1051 	uint64_t	ts;		/* real-time when snapshot was made */
1052 	uint64_t	reserved01;
1053 	uint64_t	reserved02;
1054 	char		label[64];	/* user-supplied description */
1055 	uint64_t	reserved03[4];
1056 } *hammer_snapshot_data_t;
1057 
1058 /*
1059  * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }.
1060  *
1061  * Used to store the hammer cleanup config.  This data is not mirrored.
1062  */
1063 typedef struct hammer_config_data {
1064 	char		text[1024];
1065 } *hammer_config_data_t;
1066 
1067 /*
1068  * Rollup various structures embedded as record data
1069  */
1070 typedef union hammer_data_ondisk {
1071 	struct hammer_direntry_data entry;
1072 	struct hammer_inode_data inode;
1073 	struct hammer_symlink_data symlink;
1074 	struct hammer_pseudofs_data pfsd;
1075 	struct hammer_snapshot_data snap;
1076 	struct hammer_config_data config;
1077 } *hammer_data_ondisk_t;
1078 
1079 /*
1080  * Ondisk layout of B-Tree related structures
1081  */
1082 #if 0	 /* Not needed for fstype(8) */
1083 #include "hammer_btree.h"
1084 #endif
1085 
1086 #define HAMMER_DIR_INODE_LOCALIZATION(ino_data)				\
1087 	(((ino_data)->cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO) ?	\
1088 	 HAMMER_LOCALIZE_INODE :					\
1089 	 HAMMER_LOCALIZE_MISC)
1090 
1091 #endif /* !VFS_HAMMER_DISK_H_ */
1092