xref: /freebsd/usr.sbin/fstyp/hammer_disk.h (revision 42b388439bd3795e09258c57a74ce9eec3651c7b)
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  */
36 
37 #ifndef VFS_HAMMER_DISK_H_
38 #define VFS_HAMMER_DISK_H_
39 
40 #include <sys/endian.h>
41 
42 #ifndef _SYS_UUID_H_
43 #include <sys/uuid.h>
44 #endif
45 
46 /*
47  * The structures below represent the on-disk format for a HAMMER
48  * filesystem.  Note that all fields for on-disk structures are naturally
49  * aligned.  HAMMER uses little endian for fields in on-disk structures.
50  * HAMMER doesn't support big endian arch, but is planned.
51  *
52  * Most of HAMMER revolves around the concept of an object identifier.  An
53  * obj_id is a 64 bit quantity which uniquely identifies a filesystem object
54  * FOR THE ENTIRE LIFE OF THE FILESYSTEM.  This uniqueness allows backups
55  * and mirrors to retain varying amounts of filesystem history by removing
56  * any possibility of conflict through identifier reuse.
57  *
58  * A HAMMER filesystem may span multiple volumes.
59  *
60  * A HAMMER filesystem uses a 16K filesystem buffer size.  All filesystem
61  * I/O is done in multiples of 16K.
62  *
63  * 64K X-bufs are used for blocks >= a file's 1MB mark.
64  *
65  * Per-volume storage limit: 52 bits		4096 TB
66  * Per-Zone storage limit: 60 bits		1 MTB
67  * Per-filesystem storage limit: 60 bits	1 MTB
68  */
69 #define HAMMER_BUFSIZE		16384
70 #define HAMMER_XBUFSIZE		65536
71 #define HAMMER_HBUFSIZE		(HAMMER_BUFSIZE / 2)
72 #define HAMMER_XDEMARC		(1024 * 1024)
73 #define HAMMER_BUFMASK		(HAMMER_BUFSIZE - 1)
74 #define HAMMER_XBUFMASK		(HAMMER_XBUFSIZE - 1)
75 
76 #define HAMMER_BUFSIZE64	((uint64_t)HAMMER_BUFSIZE)
77 #define HAMMER_BUFMASK64	((uint64_t)HAMMER_BUFMASK)
78 
79 #define HAMMER_XBUFSIZE64	((uint64_t)HAMMER_XBUFSIZE)
80 #define HAMMER_XBUFMASK64	((uint64_t)HAMMER_XBUFMASK)
81 
82 #define HAMMER_OFF_ZONE_MASK	0xF000000000000000ULL /* zone portion */
83 #define HAMMER_OFF_VOL_MASK	0x0FF0000000000000ULL /* volume portion */
84 #define HAMMER_OFF_SHORT_MASK	0x000FFFFFFFFFFFFFULL /* offset portion */
85 #define HAMMER_OFF_LONG_MASK	0x0FFFFFFFFFFFFFFFULL /* offset portion */
86 
87 #define HAMMER_OFF_BAD		((hammer_off_t)-1)
88 
89 #define HAMMER_BUFSIZE_DOALIGN(offset)				\
90 	(((offset) + HAMMER_BUFMASK) & ~HAMMER_BUFMASK)
91 #define HAMMER_BUFSIZE64_DOALIGN(offset)			\
92 	(((offset) + HAMMER_BUFMASK64) & ~HAMMER_BUFMASK64)
93 
94 #define HAMMER_XBUFSIZE_DOALIGN(offset)				\
95 	(((offset) + HAMMER_XBUFMASK) & ~HAMMER_XBUFMASK)
96 #define HAMMER_XBUFSIZE64_DOALIGN(offset)			\
97 	(((offset) + HAMMER_XBUFMASK64) & ~HAMMER_XBUFMASK64)
98 
99 /*
100  * The current limit of volumes that can make up a HAMMER FS
101  */
102 #define HAMMER_MAX_VOLUMES	256
103 
104 /*
105  * Reserved space for (future) header junk after the volume header.
106  */
107 #define HAMMER_MIN_VOL_JUNK	(HAMMER_BUFSIZE * 16)	/* 256 KB */
108 #define HAMMER_MAX_VOL_JUNK	HAMMER_MIN_VOL_JUNK
109 #define HAMMER_VOL_JUNK_SIZE	HAMMER_MIN_VOL_JUNK
110 
111 /*
112  * Hammer transaction ids are 64 bit unsigned integers and are usually
113  * synchronized with the time of day in nanoseconds.
114  *
115  * Hammer offsets are used for FIFO indexing and embed a cycle counter
116  * and volume number in addition to the offset.  Most offsets are required
117  * to be 16 KB aligned.
118  */
119 typedef uint64_t hammer_tid_t;
120 typedef uint64_t hammer_off_t;
121 typedef uint32_t hammer_crc_t;
122 typedef uuid_t hammer_uuid_t;
123 
124 #define HAMMER_MIN_TID		0ULL			/* unsigned */
125 #define HAMMER_MAX_TID		0xFFFFFFFFFFFFFFFFULL	/* unsigned */
126 #define HAMMER_MIN_KEY		-0x8000000000000000LL	/* signed */
127 #define HAMMER_MAX_KEY		0x7FFFFFFFFFFFFFFFLL	/* signed */
128 #define HAMMER_MIN_OBJID	HAMMER_MIN_KEY		/* signed */
129 #define HAMMER_MAX_OBJID	HAMMER_MAX_KEY		/* signed */
130 #define HAMMER_MIN_RECTYPE	0x0U			/* unsigned */
131 #define HAMMER_MAX_RECTYPE	0xFFFFU			/* unsigned */
132 #define HAMMER_MIN_OFFSET	0ULL			/* unsigned */
133 #define HAMMER_MAX_OFFSET	0xFFFFFFFFFFFFFFFFULL	/* unsigned */
134 
135 /*
136  * hammer_off_t has several different encodings.  Note that not all zones
137  * encode a vol_no.  Zone bits are not a part of filesystem capacity as
138  * the zone bits aren't directly or indirectly mapped to physical volumes.
139  *
140  * In other words, HAMMER's logical filesystem offset consists of 64 bits,
141  * but the filesystem is considered 60 bits filesystem, not 64 bits.
142  * The maximum filesystem capacity is 1EB, not 16EB.
143  *
144  * zone 0:		available, a big-block that contains the offset is unused
145  * zone 1 (z,v,o):	raw volume relative (offset 0 is the volume header)
146  * zone 2 (z,v,o):	raw buffer relative (offset 0 is the first buffer)
147  * zone 3 (z,o):	undo/redo fifo	- fixed zone-2 offset array in volume header
148  * zone 4 (z,v,o):	freemap		- only real blockmap
149  * zone 8 (z,v,o):	B-Tree		- actually zone-2 address
150  * zone 9 (z,v,o):	meta		- actually zone-2 address
151  * zone 10 (z,v,o):	large-data	- actually zone-2 address
152  * zone 11 (z,v,o):	small-data	- actually zone-2 address
153  * zone 15:		unavailable, usually the offset is beyond volume size
154  *
155  * layer1/layer2 direct map:
156  *	     Maximum HAMMER filesystem capacity from volume aspect
157  *	     2^8(max volumes) * 2^52(max volume size) = 2^60 = 1EB (long offset)
158  *	    <------------------------------------------------------------->
159  *	     8bits   52bits (short offset)
160  *	    <------><----------------------------------------------------->
161  *	zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo
162  *	----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo
163  *	    <-----------------><------------------><---------------------->
164  *	     18bits             19bits              23bits
165  *	    <------------------------------------------------------------->
166  *	     2^18(layer1) * 2^19(layer2) * 2^23(big-block) = 2^60 = 1EB
167  *	     Maximum HAMMER filesystem capacity from blockmap aspect
168  *
169  * volume#0 layout
170  *	+-------------------------> offset 0 of a device/partition
171  *	| volume header (1928 bytes)
172  *	| the rest of header junk space (HAMMER_BUFSIZE aligned)
173  *	+-------------------------> vol_bot_beg
174  *	| boot area (HAMMER_BUFSIZE aligned)
175  *	+-------------------------> vol_mem_beg
176  *	| memory log (HAMMER_BUFSIZE aligned)
177  *	+-------------------------> vol_buf_beg (physical offset of zone-2)
178  *	| zone-4 big-block for layer1
179  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE
180  *	| zone-4 big-blocks for layer2
181  *	| ... (1 big-block per 4TB space)
182  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
183  *	| zone-3 big-blocks for UNDO/REDO FIFO
184  *	| ... (max 128 big-blocks)
185  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
186  *	| zone-8 big-block for root B-Tree node/etc
187  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
188  *	| zone-9 big-block for root inode/PFS/etc
189  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
190  *	| zone-X big-blocks
191  *	| ... (big-blocks for new zones after newfs_hammer)
192  *	| ...
193  *	| ...
194  *	| ...
195  *	| ...
196  *	+-------------------------> vol_buf_end (HAMMER_BUFSIZE aligned)
197  *	+-------------------------> end of a device/partition
198  *
199  * volume#N layout (0<N<256)
200  *	+-------------------------> offset 0 of a device/partition
201  *	| volume header (1928 bytes)
202  *	| the rest of header junk space (HAMMER_BUFSIZE aligned)
203  *	+-------------------------> vol_bot_beg
204  *	| boot area (HAMMER_BUFSIZE aligned)
205  *	+-------------------------> vol_mem_beg
206  *	| memory log (HAMMER_BUFSIZE aligned)
207  *	+-------------------------> vol_buf_beg (physical offset of zone-2)
208  *	| zone-4 big-blocks for layer2
209  *	| ... (1 big-block per 4TB space)
210  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
211  *	| zone-X big-blocks
212  *	| ... (unused until volume#(N-1) runs out of space)
213  *	| ...
214  *	| ...
215  *	| ...
216  *	| ...
217  *	+-------------------------> vol_buf_end (HAMMER_BUFSIZE aligned)
218  *	+-------------------------> end of a device/partition
219  */
220 
221 #define HAMMER_ZONE_RAW_VOLUME		0x1000000000000000ULL
222 #define HAMMER_ZONE_RAW_BUFFER		0x2000000000000000ULL
223 #define HAMMER_ZONE_UNDO		0x3000000000000000ULL
224 #define HAMMER_ZONE_FREEMAP		0x4000000000000000ULL
225 #define HAMMER_ZONE_RESERVED05		0x5000000000000000ULL  /* not used */
226 #define HAMMER_ZONE_RESERVED06		0x6000000000000000ULL  /* not used */
227 #define HAMMER_ZONE_RESERVED07		0x7000000000000000ULL  /* not used */
228 #define HAMMER_ZONE_BTREE		0x8000000000000000ULL
229 #define HAMMER_ZONE_META		0x9000000000000000ULL
230 #define HAMMER_ZONE_LARGE_DATA		0xA000000000000000ULL
231 #define HAMMER_ZONE_SMALL_DATA		0xB000000000000000ULL
232 #define HAMMER_ZONE_RESERVED0C		0xC000000000000000ULL  /* not used */
233 #define HAMMER_ZONE_RESERVED0D		0xD000000000000000ULL  /* not used */
234 #define HAMMER_ZONE_RESERVED0E		0xE000000000000000ULL  /* not used */
235 #define HAMMER_ZONE_UNAVAIL		0xF000000000000000ULL
236 
237 #define HAMMER_ZONE_RAW_VOLUME_INDEX	1
238 #define HAMMER_ZONE_RAW_BUFFER_INDEX	2
239 #define HAMMER_ZONE_UNDO_INDEX		3
240 #define HAMMER_ZONE_FREEMAP_INDEX	4
241 #define HAMMER_ZONE_BTREE_INDEX		8
242 #define HAMMER_ZONE_META_INDEX		9
243 #define HAMMER_ZONE_LARGE_DATA_INDEX	10
244 #define HAMMER_ZONE_SMALL_DATA_INDEX	11
245 #define HAMMER_ZONE_UNAVAIL_INDEX	15
246 
247 #define HAMMER_MAX_ZONES		16
248 
249 #define HAMMER_ZONE(offset)		((offset) & HAMMER_OFF_ZONE_MASK)
250 
251 #define hammer_is_zone_raw_volume(offset)		\
252 	(HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_VOLUME)
253 #define hammer_is_zone_raw_buffer(offset)		\
254 	(HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_BUFFER)
255 #define hammer_is_zone_undo(offset)			\
256 	(HAMMER_ZONE(offset) == HAMMER_ZONE_UNDO)
257 #define hammer_is_zone_freemap(offset)			\
258 	(HAMMER_ZONE(offset) == HAMMER_ZONE_FREEMAP)
259 #define hammer_is_zone_btree(offset)			\
260 	(HAMMER_ZONE(offset) == HAMMER_ZONE_BTREE)
261 #define hammer_is_zone_meta(offset)			\
262 	(HAMMER_ZONE(offset) == HAMMER_ZONE_META)
263 #define hammer_is_zone_large_data(offset)		\
264 	(HAMMER_ZONE(offset) == HAMMER_ZONE_LARGE_DATA)
265 #define hammer_is_zone_small_data(offset)		\
266 	(HAMMER_ZONE(offset) == HAMMER_ZONE_SMALL_DATA)
267 #define hammer_is_zone_unavail(offset)			\
268 	(HAMMER_ZONE(offset) == HAMMER_ZONE_UNAVAIL)
269 #define hammer_is_zone_data(offset)			\
270 	(hammer_is_zone_large_data(offset) || hammer_is_zone_small_data(offset))
271 
272 #define hammer_is_index_record(zone)			\
273 	((zone) >= HAMMER_ZONE_BTREE_INDEX &&		\
274 	 (zone) < HAMMER_MAX_ZONES)
275 
276 #define hammer_is_zone_record(offset)			\
277 	hammer_is_index_record(HAMMER_ZONE_DECODE(offset))
278 
279 #define hammer_is_index_direct_xlated(zone)		\
280 	(((zone) == HAMMER_ZONE_RAW_BUFFER_INDEX) ||	\
281 	 ((zone) == HAMMER_ZONE_FREEMAP_INDEX) ||	\
282 	 hammer_is_index_record(zone))
283 
284 #define hammer_is_zone_direct_xlated(offset)		\
285 	hammer_is_index_direct_xlated(HAMMER_ZONE_DECODE(offset))
286 
287 #define HAMMER_ZONE_ENCODE(zone, ham_off)		\
288 	(((hammer_off_t)(zone) << 60) | (ham_off))
289 #define HAMMER_ZONE_DECODE(ham_off)			\
290 	((int)(((hammer_off_t)(ham_off) >> 60)))
291 
292 #define HAMMER_VOL_ENCODE(vol_no)			\
293 	((hammer_off_t)((vol_no) & 255) << 52)
294 #define HAMMER_VOL_DECODE(ham_off)			\
295 	((int)(((hammer_off_t)(ham_off) >> 52) & 255))
296 
297 #define HAMMER_OFF_SHORT_ENCODE(offset)			\
298 	((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
299 #define HAMMER_OFF_LONG_ENCODE(offset)			\
300 	((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
301 
302 #define HAMMER_ENCODE(zone, vol_no, offset)		\
303 	(((hammer_off_t)(zone) << 60) |			\
304 	HAMMER_VOL_ENCODE(vol_no) |			\
305 	HAMMER_OFF_SHORT_ENCODE(offset))
306 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset)	\
307 	HAMMER_ENCODE(HAMMER_ZONE_RAW_VOLUME_INDEX, vol_no, offset)
308 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset)	\
309 	HAMMER_ENCODE(HAMMER_ZONE_RAW_BUFFER_INDEX, vol_no, offset)
310 #define HAMMER_ENCODE_UNDO(offset)			\
311 	HAMMER_ENCODE(HAMMER_ZONE_UNDO_INDEX, HAMMER_ROOT_VOLNO, offset)
312 #define HAMMER_ENCODE_FREEMAP(vol_no, offset)		\
313 	HAMMER_ENCODE(HAMMER_ZONE_FREEMAP_INDEX, vol_no, offset)
314 
315 /*
316  * Translate a zone address to zone-X address.
317  */
318 #define hammer_xlate_to_zoneX(zone, offset)		\
319 	HAMMER_ZONE_ENCODE((zone), (offset) & ~HAMMER_OFF_ZONE_MASK)
320 #define hammer_xlate_to_zone2(offset)			\
321 	hammer_xlate_to_zoneX(HAMMER_ZONE_RAW_BUFFER_INDEX, (offset))
322 
323 #define hammer_data_zone(data_len)			\
324 	(((data_len) >= HAMMER_BUFSIZE) ?		\
325 	 HAMMER_ZONE_LARGE_DATA :			\
326 	 HAMMER_ZONE_SMALL_DATA)
327 #define hammer_data_zone_index(data_len)		\
328 	(((data_len) >= HAMMER_BUFSIZE) ?		\
329 	 HAMMER_ZONE_LARGE_DATA_INDEX :			\
330 	 HAMMER_ZONE_SMALL_DATA_INDEX)
331 
332 /*
333  * Big-Block backing store
334  *
335  * A blockmap is a two-level map which translates a blockmap-backed zone
336  * offset into a raw zone 2 offset.  The layer 1 handles 18 bits and the
337  * layer 2 handles 19 bits.  The 8M big-block size is 23 bits so two
338  * layers gives us 18+19+23 = 60 bits of address space.
339  *
340  * When using hinting for a blockmap lookup, the hint is lost when the
341  * scan leaves the HINTBLOCK, which is typically several BIGBLOCK's.
342  * HINTBLOCK is a heuristic.
343  */
344 #define HAMMER_HINTBLOCK_SIZE		(HAMMER_BIGBLOCK_SIZE * 4)
345 #define HAMMER_HINTBLOCK_MASK64		((uint64_t)HAMMER_HINTBLOCK_SIZE - 1)
346 #define HAMMER_BIGBLOCK_SIZE		(8192 * 1024)
347 #define HAMMER_BIGBLOCK_SIZE64		((uint64_t)HAMMER_BIGBLOCK_SIZE)
348 #define HAMMER_BIGBLOCK_MASK		(HAMMER_BIGBLOCK_SIZE - 1)
349 #define HAMMER_BIGBLOCK_MASK64		((uint64_t)HAMMER_BIGBLOCK_SIZE - 1)
350 #define HAMMER_BIGBLOCK_BITS		23
351 #if 0
352 #define HAMMER_BIGBLOCK_OVERFILL	(6144 * 1024)
353 #endif
354 #if (1 << HAMMER_BIGBLOCK_BITS) != HAMMER_BIGBLOCK_SIZE
355 #error "HAMMER_BIGBLOCK_BITS BROKEN"
356 #endif
357 
358 #define HAMMER_BUFFERS_PER_BIGBLOCK			\
359 	(HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE)
360 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK		\
361 	(HAMMER_BUFFERS_PER_BIGBLOCK - 1)
362 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK64		\
363 	((hammer_off_t)HAMMER_BUFFERS_PER_BIGBLOCK_MASK)
364 
365 #define HAMMER_BIGBLOCK_DOALIGN(offset)				\
366 	(((offset) + HAMMER_BIGBLOCK_MASK64) & ~HAMMER_BIGBLOCK_MASK64)
367 
368 /*
369  * Maximum number of mirrors operating in master mode (multi-master
370  * clustering and mirroring). Note that HAMMER1 does not support
371  * multi-master clustering as of 2015.
372  */
373 #define HAMMER_MAX_MASTERS		16
374 
375 /*
376  * The blockmap is somewhat of a degenerate structure.  HAMMER only actually
377  * uses it in its original incarnation to implement the freemap.
378  *
379  * zone:1	raw volume (no blockmap)
380  * zone:2	raw buffer (no blockmap)
381  * zone:3	undomap    (direct layer2 array in volume header)
382  * zone:4	freemap    (the only real blockmap)
383  * zone:8-15	zone id used to classify big-block only, address is actually
384  *		a zone-2 address.
385  */
386 typedef struct hammer_blockmap {
387 	hammer_off_t	phys_offset;  /* zone-2 offset only used by zone-4 */
388 	hammer_off_t	first_offset; /* zone-X offset only used by zone-3 */
389 	hammer_off_t	next_offset;  /* zone-X offset for allocation */
390 	hammer_off_t	alloc_offset; /* zone-X offset only used by zone-3 */
391 	uint32_t	reserved01;
392 	hammer_crc_t	entry_crc;
393 } *hammer_blockmap_t;
394 
395 #define HAMMER_BLOCKMAP_CRCSIZE	\
396 	offsetof(struct hammer_blockmap, entry_crc)
397 
398 /*
399  * The blockmap is a 2-layer entity made up of big-blocks.  The first layer
400  * contains 262144 32-byte entries (18 bits), the second layer contains
401  * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
402  * 18+19+23 = 60 bits.  The top four bits are the zone id.
403  *
404  * Currently only the freemap utilizes both layers in all their glory.
405  * All primary data/meta-data zones actually encode a zone-2 address
406  * requiring no real blockmap translation.
407  *
408  * The freemap uses the upper 8 bits of layer-1 to identify the volume,
409  * thus any space allocated via the freemap can be directly translated
410  * to a zone:2 (or zone:8-15) address.
411  *
412  * zone-X blockmap offset: [zone:4][layer1:18][layer2:19][big-block:23]
413  */
414 
415 /*
416  * 32 bytes layer1 entry for 8MB big-block.
417  * A big-block can hold 2^23 / 2^5 = 2^18 layer1 entries,
418  * which equals bits assigned for layer1 in zone-2 address.
419  */
420 typedef struct hammer_blockmap_layer1 {
421 	hammer_off_t	blocks_free;	/* big-blocks free */
422 	hammer_off_t	phys_offset;	/* UNAVAIL or zone-2 */
423 	hammer_off_t	reserved01;
424 	hammer_crc_t	layer2_crc;	/* xor'd crc's of HAMMER_BLOCKSIZE */
425 					/* (not yet used) */
426 	hammer_crc_t	layer1_crc;	/* MUST BE LAST FIELD OF STRUCTURE*/
427 } *hammer_blockmap_layer1_t;
428 
429 #define HAMMER_LAYER1_CRCSIZE	\
430 	offsetof(struct hammer_blockmap_layer1, layer1_crc)
431 
432 /*
433  * 16 bytes layer2 entry for 8MB big-blocks.
434  * A big-block can hold 2^23 / 2^4 = 2^19 layer2 entries,
435  * which equals bits assigned for layer2 in zone-2 address.
436  *
437  * NOTE: bytes_free is signed and can legally go negative if/when data
438  *	 de-dup occurs.  This field will never go higher than
439  *	 HAMMER_BIGBLOCK_SIZE.  If exactly HAMMER_BIGBLOCK_SIZE
440  *	 the big-block is completely free.
441  */
442 typedef struct hammer_blockmap_layer2 {
443 	uint8_t		zone;		/* typed allocation zone */
444 	uint8_t		reserved01;
445 	uint16_t	reserved02;
446 	uint32_t	append_off;	/* allocatable space index */
447 	int32_t		bytes_free;	/* bytes free within this big-block */
448 	hammer_crc_t	entry_crc;
449 } *hammer_blockmap_layer2_t;
450 
451 #define HAMMER_LAYER2_CRCSIZE	\
452 	offsetof(struct hammer_blockmap_layer2, entry_crc)
453 
454 #define HAMMER_BLOCKMAP_UNAVAIL	((hammer_off_t)-1LL)
455 
456 #define HAMMER_BLOCKMAP_RADIX1	/* 2^18 = 262144 */	\
457 	((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1)))
458 #define HAMMER_BLOCKMAP_RADIX2	/* 2^19 = 524288 */	\
459 	((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2)))
460 
461 #define HAMMER_BLOCKMAP_LAYER1	/* 2^(18+19+23) = 1EB */	\
462 	(HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
463 #define HAMMER_BLOCKMAP_LAYER2	/* 2^(19+23) = 4TB */		\
464 	(HAMMER_BLOCKMAP_RADIX2 * HAMMER_BIGBLOCK_SIZE64)
465 
466 #define HAMMER_BLOCKMAP_LAYER1_MASK	(HAMMER_BLOCKMAP_LAYER1 - 1)
467 #define HAMMER_BLOCKMAP_LAYER2_MASK	(HAMMER_BLOCKMAP_LAYER2 - 1)
468 
469 #define HAMMER_BLOCKMAP_LAYER2_DOALIGN(offset)			\
470 	(((offset) + HAMMER_BLOCKMAP_LAYER2_MASK) &		\
471 	 ~HAMMER_BLOCKMAP_LAYER2_MASK)
472 
473 /*
474  * Index within layer1 or layer2 big-block for the entry representing
475  * a zone-2 physical offset.
476  */
477 #define HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset)		\
478 	((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) /	\
479 	 HAMMER_BLOCKMAP_LAYER2))
480 
481 #define HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset)		\
482 	((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) /	\
483 	HAMMER_BIGBLOCK_SIZE64))
484 
485 /*
486  * Byte offset within layer1 or layer2 big-block for the entry representing
487  * a zone-2 physical offset.  Multiply the index by sizeof(blockmap_layer).
488  */
489 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset)		\
490 	(HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) *		\
491 	 sizeof(struct hammer_blockmap_layer1))
492 
493 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset)		\
494 	(HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) *		\
495 	 sizeof(struct hammer_blockmap_layer2))
496 
497 /*
498  * Move on to offset 0 of the next layer1 or layer2.
499  */
500 #define HAMMER_ZONE_LAYER1_NEXT_OFFSET(offset)			\
501 	(((offset) + HAMMER_BLOCKMAP_LAYER2) & ~HAMMER_BLOCKMAP_LAYER2_MASK)
502 
503 #define HAMMER_ZONE_LAYER2_NEXT_OFFSET(offset)			\
504 	(((offset) + HAMMER_BIGBLOCK_SIZE) & ~HAMMER_BIGBLOCK_MASK64)
505 
506 /*
507  * HAMMER UNDO parameters.  The UNDO fifo is mapped directly in the volume
508  * header with an array of zone-2 offsets.  A maximum of (128x8MB) = 1GB,
509  * and minimum of (64x8MB) = 512MB may be reserved.  The size of the undo
510  * fifo is usually set a newfs time.
511  */
512 #define HAMMER_MIN_UNDO_BIGBLOCKS		64
513 #define HAMMER_MAX_UNDO_BIGBLOCKS		128
514 
515 /*
516  * All on-disk HAMMER structures which make up elements of the UNDO FIFO
517  * contain a hammer_fifo_head and hammer_fifo_tail structure.  This structure
518  * contains all the information required to validate the fifo element
519  * and to scan the fifo in either direction.  The head is typically embedded
520  * in higher level hammer on-disk structures while the tail is typically
521  * out-of-band.  hdr_size is the size of the whole mess, including the tail.
522  *
523  * All undo structures are guaranteed to not cross a 16K filesystem
524  * buffer boundary.  Most undo structures are fairly small.  Data spaces
525  * are not immediately reused by HAMMER so file data is not usually recorded
526  * as part of an UNDO.
527  *
528  * PAD elements are allowed to take up only 8 bytes of space as a special
529  * case, containing only hdr_signature, hdr_type, and hdr_size fields,
530  * and with the tail overloaded onto the head structure for 8 bytes total.
531  *
532  * Every undo record has a sequence number.  This number is unrelated to
533  * transaction ids and instead collects the undo transactions associated
534  * with a single atomic operation.  A larger transactional operation, such
535  * as a remove(), may consist of several smaller atomic operations
536  * representing raw meta-data operations.
537  *
538  *				HAMMER VERSION 4 CHANGES
539  *
540  * In HAMMER version 4 the undo structure alignment is reduced from 16384
541  * to 512 bytes in order to ensure that each 512 byte sector begins with
542  * a header.  The hdr_seq field in the header is a 32 bit sequence number
543  * which allows the recovery code to detect missing sectors
544  * without relying on the 32-bit crc and to definitively identify the current
545  * undo sequence space without having to rely on information from the volume
546  * header.  In addition, new REDO entries in the undo space are used to
547  * record write, write/extend, and transaction id updates.
548  *
549  * The grand result is:
550  *
551  * (1) The volume header no longer needs to be synchronized for most
552  *     flush and fsync operations.
553  *
554  * (2) Most fsync operations need only lay down REDO records
555  *
556  * (3) Data overwrite for nohistory operations covered by REDO records
557  *     can be supported (instead of rolling a new block allocation),
558  *     by rolling UNDO for the prior contents of the data.
559  *
560  *				HAMMER VERSION 5 CHANGES
561  *
562  * Hammer version 5 contains a minor adjustment making layer2's bytes_free
563  * field signed, allowing dedup to push it into the negative domain.
564  */
565 #define HAMMER_HEAD_ALIGN		8
566 #define HAMMER_HEAD_ALIGN_MASK		(HAMMER_HEAD_ALIGN - 1)
567 #define HAMMER_HEAD_DOALIGN(bytes)	\
568 	(((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK)
569 
570 #define HAMMER_UNDO_ALIGN		512
571 #define HAMMER_UNDO_ALIGN64		((uint64_t)512)
572 #define HAMMER_UNDO_MASK		(HAMMER_UNDO_ALIGN - 1)
573 #define HAMMER_UNDO_MASK64		(HAMMER_UNDO_ALIGN64 - 1)
574 #define HAMMER_UNDO_DOALIGN(offset)	\
575 	(((offset) + HAMMER_UNDO_MASK) & ~HAMMER_UNDO_MASK64)
576 
577 typedef struct hammer_fifo_head {
578 	uint16_t hdr_signature;
579 	uint16_t hdr_type;
580 	uint32_t hdr_size;	/* Aligned size of the whole mess */
581 	uint32_t hdr_seq;	/* Sequence number */
582 	hammer_crc_t hdr_crc;	/* XOR crc up to field w/ crc after field */
583 } *hammer_fifo_head_t;
584 
585 #define HAMMER_FIFO_HEAD_CRCOFF	offsetof(struct hammer_fifo_head, hdr_crc)
586 
587 typedef struct hammer_fifo_tail {
588 	uint16_t tail_signature;
589 	uint16_t tail_type;
590 	uint32_t tail_size;	/* aligned size of the whole mess */
591 } *hammer_fifo_tail_t;
592 
593 /*
594  * Fifo header types.
595  *
596  * NOTE: 0x8000U part of HAMMER_HEAD_TYPE_PAD can be removed if the HAMMER
597  * version ever gets bumped again. It exists only to keep compatibility with
598  * older versions.
599  */
600 #define HAMMER_HEAD_TYPE_PAD	(0x0040U | 0x8000U)
601 #define HAMMER_HEAD_TYPE_DUMMY	0x0041U		/* dummy entry w/seqno */
602 #define HAMMER_HEAD_TYPE_UNDO	0x0043U		/* random UNDO information */
603 #define HAMMER_HEAD_TYPE_REDO	0x0044U		/* data REDO / fast fsync */
604 
605 #define HAMMER_HEAD_SIGNATURE	0xC84EU
606 #define HAMMER_TAIL_SIGNATURE	0xC74FU
607 
608 /*
609  * Misc FIFO structures.
610  *
611  * UNDO - Raw meta-data media updates.
612  */
613 typedef struct hammer_fifo_undo {
614 	struct hammer_fifo_head	head;
615 	hammer_off_t		undo_offset;	/* zone-1,2 offset */
616 	int32_t			undo_data_bytes;
617 	int32_t			undo_reserved01;
618 	/* followed by data */
619 } *hammer_fifo_undo_t;
620 
621 /*
622  * REDO (HAMMER version 4+) - Logical file writes/truncates.
623  *
624  * REDOs contain information which will be duplicated in a later meta-data
625  * update, allowing fast write()+fsync() operations.  REDOs can be ignored
626  * without harming filesystem integrity but must be processed if fsync()
627  * semantics are desired.
628  *
629  * Unlike UNDOs which are processed backwards within the recovery span,
630  * REDOs must be processed forwards starting further back (starting outside
631  * the recovery span).
632  *
633  *	WRITE	- Write logical file (with payload).  Executed both
634  *		  out-of-span and in-span.  Out-of-span WRITEs may be
635  *		  filtered out by TERMs.
636  *
637  *	TRUNC	- Truncate logical file (no payload).  Executed both
638  *		  out-of-span and in-span.  Out-of-span WRITEs may be
639  *		  filtered out by TERMs.
640  *
641  *	TERM_*	- Indicates meta-data was committed (if out-of-span) or
642  *		  will be rolled-back (in-span).  Any out-of-span TERMs
643  *		  matching earlier WRITEs remove those WRITEs from
644  *		  consideration as they might conflict with a later data
645  *		  commit (which is not being rolled-back).
646  *
647  *	SYNC	- The earliest in-span SYNC (the last one when scanning
648  *		  backwards) tells the recovery code how far out-of-span
649  *		  it must go to run REDOs.
650  *
651  * NOTE: WRITEs do not always have matching TERMs even under
652  *	 perfect conditions because truncations might remove the
653  *	 buffers from consideration.  I/O problems can also remove
654  *	 buffers from consideration.
655  *
656  *	 TRUNCSs do not always have matching TERMs because several
657  *	 truncations may be aggregated together into a single TERM.
658  */
659 typedef struct hammer_fifo_redo {
660 	struct hammer_fifo_head	head;
661 	int64_t			redo_objid;	/* file being written */
662 	hammer_off_t		redo_offset;	/* logical offset in file */
663 	int32_t			redo_data_bytes;
664 	uint32_t		redo_flags;
665 	uint32_t		redo_localization;
666 	uint32_t		redo_reserved01;
667 	uint64_t		redo_reserved02;
668 	/* followed by data */
669 } *hammer_fifo_redo_t;
670 
671 #define HAMMER_REDO_WRITE	0x00000001
672 #define HAMMER_REDO_TRUNC	0x00000002
673 #define HAMMER_REDO_TERM_WRITE	0x00000004
674 #define HAMMER_REDO_TERM_TRUNC	0x00000008
675 #define HAMMER_REDO_SYNC	0x00000010
676 
677 typedef union hammer_fifo_any {
678 	struct hammer_fifo_head	head;
679 	struct hammer_fifo_undo	undo;
680 	struct hammer_fifo_redo	redo;
681 } *hammer_fifo_any_t;
682 
683 /*
684  * Volume header types
685  */
686 #define HAMMER_FSBUF_VOLUME	0xC8414D4DC5523031ULL	/* HAMMER01 */
687 #define HAMMER_FSBUF_VOLUME_REV	0x313052C54D4D41C8ULL	/* (reverse endian) */
688 
689 /*
690  * HAMMER Volume header
691  *
692  * A HAMMER filesystem can be built from 1-256 block devices, each block
693  * device contains a volume header followed by however many buffers fit
694  * into the volume.
695  *
696  * One of the volumes making up a HAMMER filesystem is the root volume.
697  * The root volume is always volume #0 which is the first block device path
698  * specified by newfs_hammer(8).  All HAMMER volumes have a volume header,
699  * however the root volume may be the only volume that has valid values for
700  * some fields in the header.
701  *
702  * Special field notes:
703  *
704  *	vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
705  *	vol_mem_beg - offset of memory log (buf_beg - mem_beg bytes)
706  *	vol_buf_beg - offset of the first buffer in volume
707  *	vol_buf_end - offset of volume EOF (on buffer boundary)
708  *
709  *	The memory log area allows a kernel to cache new records and data
710  *	in memory without allocating space in the actual filesystem to hold
711  *	the records and data.  In the event that a filesystem becomes full,
712  *	any records remaining in memory can be flushed to the memory log
713  *	area.  This allows the kernel to immediately return success.
714  *
715  *	The buffer offset is a physical offset of zone-2 offset. The lower
716  *	52 bits of the zone-2 offset is added to the buffer offset of each
717  *	volume to generate an actual I/O offset within the block device.
718  *
719  *	NOTE: boot area and memory log are currently not used.
720  */
721 
722 /*
723  * Filesystem type string
724  */
725 #define HAMMER_FSTYPE_STRING		"DragonFly HAMMER"
726 
727 /*
728  * These macros are only used by userspace when userspace commands either
729  * initialize or add a new HAMMER volume.
730  */
731 #define HAMMER_BOOT_MINBYTES		(32*1024)
732 #define HAMMER_BOOT_NOMBYTES		(64LL*1024*1024)
733 #define HAMMER_BOOT_MAXBYTES		(256LL*1024*1024)
734 
735 #define HAMMER_MEM_MINBYTES		(256*1024)
736 #define HAMMER_MEM_NOMBYTES		(1LL*1024*1024*1024)
737 #define HAMMER_MEM_MAXBYTES		(64LL*1024*1024*1024)
738 
739 typedef struct hammer_volume_ondisk {
740 	uint64_t vol_signature;	/* HAMMER_FSBUF_VOLUME for a valid header */
741 
742 	/*
743 	 * These are relative to block device offset, not zone offsets.
744 	 */
745 	int64_t vol_bot_beg;	/* offset of boot area */
746 	int64_t vol_mem_beg;	/* offset of memory log */
747 	int64_t vol_buf_beg;	/* offset of the first buffer in volume */
748 	int64_t vol_buf_end;	/* offset of volume EOF (on buffer boundary) */
749 	int64_t vol_reserved01;
750 
751 	hammer_uuid_t vol_fsid;	/* identify filesystem */
752 	hammer_uuid_t vol_fstype; /* identify filesystem type */
753 	char vol_label[64];	/* filesystem label */
754 
755 	int32_t vol_no;		/* volume number within filesystem */
756 	int32_t vol_count;	/* number of volumes making up filesystem */
757 
758 	uint32_t vol_version;	/* version control information */
759 	hammer_crc_t vol_crc;	/* header crc */
760 	uint32_t vol_flags;	/* volume flags */
761 	uint32_t vol_rootvol;	/* the root volume number (must be 0) */
762 
763 	uint32_t vol_reserved[8];
764 
765 	/*
766 	 * These fields are initialized and space is reserved in every
767 	 * volume making up a HAMMER filesystem, but only the root volume
768 	 * contains valid data.  Note that vol0_stat_bigblocks does not
769 	 * include big-blocks for freemap and undomap initially allocated
770 	 * by newfs_hammer(8).
771 	 */
772 	int64_t vol0_stat_bigblocks;	/* total big-blocks when fs is empty */
773 	int64_t vol0_stat_freebigblocks;/* number of free big-blocks */
774 	int64_t	vol0_reserved01;
775 	int64_t vol0_stat_inodes;	/* for statfs only */
776 	int64_t vol0_reserved02;
777 	hammer_off_t vol0_btree_root;	/* B-Tree root offset in zone-8 */
778 	hammer_tid_t vol0_next_tid;	/* highest partially synchronized TID */
779 	hammer_off_t vol0_reserved03;
780 
781 	/*
782 	 * Blockmaps for zones.  Not all zones use a blockmap.  Note that
783 	 * the entire root blockmap is cached in the hammer_mount structure.
784 	 */
785 	struct hammer_blockmap	vol0_blockmap[HAMMER_MAX_ZONES];
786 
787 	/*
788 	 * Array of zone-2 addresses for undo FIFO.
789 	 */
790 	hammer_off_t		vol0_undo_array[HAMMER_MAX_UNDO_BIGBLOCKS];
791 } *hammer_volume_ondisk_t;
792 
793 #define HAMMER_ROOT_VOLNO		0
794 
795 #define HAMMER_VOLF_NEEDFLUSH		0x0004	/* volume needs flush */
796 
797 #define HAMMER_VOL_CRCSIZE1	\
798 	offsetof(struct hammer_volume_ondisk, vol_crc)
799 #define HAMMER_VOL_CRCSIZE2	\
800 	(sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 -	\
801 	 sizeof(hammer_crc_t))
802 
803 #define HAMMER_VOL_VERSION_MIN		1	/* minimum supported version */
804 #define HAMMER_VOL_VERSION_DEFAULT	7	/* newfs default version */
805 #define HAMMER_VOL_VERSION_WIP		8	/* version >= this is WIP */
806 #define HAMMER_VOL_VERSION_MAX		7	/* maximum supported version */
807 
808 #define HAMMER_VOL_VERSION_ONE		1
809 #define HAMMER_VOL_VERSION_TWO		2	/* new dirent layout (2.3+) */
810 #define HAMMER_VOL_VERSION_THREE	3	/* new snapshot layout (2.5+) */
811 #define HAMMER_VOL_VERSION_FOUR		4	/* new undo/flush (2.5+) */
812 #define HAMMER_VOL_VERSION_FIVE		5	/* dedup (2.9+) */
813 #define HAMMER_VOL_VERSION_SIX		6	/* DIRHASH_ALG1 */
814 #define HAMMER_VOL_VERSION_SEVEN	7	/* use the faster iscsi_crc */
815 
816 /*
817  * Translate a zone-2 address to physical address
818  */
819 #define hammer_xlate_to_phys(volume, zone2_offset)	\
820 	((volume)->vol_buf_beg + HAMMER_OFF_SHORT_ENCODE(zone2_offset))
821 
822 /*
823  * Translate a zone-3 address to zone-2 address
824  */
825 #define HAMMER_UNDO_INDEX(zone3_offset)			\
826 	(HAMMER_OFF_SHORT_ENCODE(zone3_offset) / HAMMER_BIGBLOCK_SIZE)
827 
828 #define hammer_xlate_to_undo(volume, zone3_offset)			\
829 	((volume)->vol0_undo_array[HAMMER_UNDO_INDEX(zone3_offset)] +	\
830 	 (zone3_offset & HAMMER_BIGBLOCK_MASK64))
831 
832 /*
833  * Effective per-volume filesystem capacity including big-blocks for layer1/2
834  */
835 #define HAMMER_VOL_BUF_SIZE(volume)			\
836 	((volume)->vol_buf_end - (volume)->vol_buf_beg)
837 
838 /*
839  * Record types are fairly straightforward.  The B-Tree includes the record
840  * type in its index sort.
841  */
842 #define HAMMER_RECTYPE_UNKNOWN		0x0000
843 #define HAMMER_RECTYPE_INODE		0x0001	/* inode in obj_id space */
844 #define HAMMER_RECTYPE_DATA		0x0010
845 #define HAMMER_RECTYPE_DIRENTRY		0x0011
846 #define HAMMER_RECTYPE_DB		0x0012
847 #define HAMMER_RECTYPE_EXT		0x0013	/* ext attributes */
848 #define HAMMER_RECTYPE_FIX		0x0014	/* fixed attribute */
849 #define HAMMER_RECTYPE_PFS		0x0015	/* PFS management */
850 #define HAMMER_RECTYPE_SNAPSHOT		0x0016	/* Snapshot management */
851 #define HAMMER_RECTYPE_CONFIG		0x0017	/* hammer cleanup config */
852 #define HAMMER_RECTYPE_MAX		0xFFFF
853 
854 #define HAMMER_RECTYPE_ENTRY_START	(HAMMER_RECTYPE_INODE + 1)
855 #define HAMMER_RECTYPE_CLEAN_START	HAMMER_RECTYPE_EXT
856 
857 #define HAMMER_FIXKEY_SYMLINK		1
858 
859 #define HAMMER_OBJTYPE_UNKNOWN		0	/* never exists on-disk as unknown */
860 #define HAMMER_OBJTYPE_DIRECTORY	1
861 #define HAMMER_OBJTYPE_REGFILE		2
862 #define HAMMER_OBJTYPE_DBFILE		3
863 #define HAMMER_OBJTYPE_FIFO		4
864 #define HAMMER_OBJTYPE_CDEV		5
865 #define HAMMER_OBJTYPE_BDEV		6
866 #define HAMMER_OBJTYPE_SOFTLINK		7
867 #define HAMMER_OBJTYPE_PSEUDOFS		8	/* pseudo filesystem obj */
868 #define HAMMER_OBJTYPE_SOCKET		9
869 
870 /*
871  * HAMMER inode attribute data
872  *
873  * The data reference for a HAMMER inode points to this structure.  Any
874  * modifications to the contents of this structure will result in a
875  * replacement operation.
876  *
877  * parent_obj_id is only valid for directories (which cannot be hard-linked),
878  * and specifies the parent directory obj_id.  This field will also be set
879  * for non-directory inodes as a recovery aid, but can wind up holding
880  * stale information.  However, since object id's are not reused, the worse
881  * that happens is that the recovery code is unable to use it.
882  * A parent_obj_id of 0 means it's a root inode of root or non-root PFS.
883  *
884  * NOTE: Future note on directory hardlinks.  We can implement a record type
885  * which allows us to point to multiple parent directories.
886  */
887 typedef struct hammer_inode_data {
888 	uint16_t version;	/* inode data version */
889 	uint16_t mode;		/* basic unix permissions */
890 	uint32_t uflags;	/* chflags */
891 	uint32_t rmajor;	/* used by device nodes */
892 	uint32_t rminor;	/* used by device nodes */
893 	uint64_t ctime;
894 	int64_t parent_obj_id;	/* parent directory obj_id */
895 	hammer_uuid_t uid;
896 	hammer_uuid_t gid;
897 
898 	uint8_t obj_type;
899 	uint8_t cap_flags;	/* capability support flags (extension) */
900 	uint16_t reserved01;
901 	uint32_t reserved02;
902 	uint64_t nlinks;	/* hard links */
903 	uint64_t size;		/* filesystem object size */
904 	union {
905 		char	symlink[24];	/* HAMMER_INODE_BASESYMLEN */
906 	} ext;
907 	uint64_t mtime;	/* mtime must be second-to-last */
908 	uint64_t atime;	/* atime must be last */
909 } *hammer_inode_data_t;
910 
911 /*
912  * Neither mtime nor atime updates are CRCd by the B-Tree element.
913  * mtime updates have UNDO, atime updates do not.
914  */
915 #define HAMMER_INODE_CRCSIZE	\
916 	offsetof(struct hammer_inode_data, mtime)
917 
918 #define HAMMER_INODE_DATA_VERSION	1
919 #define HAMMER_OBJID_ROOT		1	/* root inodes # */
920 #define HAMMER_INODE_BASESYMLEN		24	/* see ext.symlink */
921 
922 /*
923  * Capability & implementation flags.
924  *
925  * HAMMER_INODE_CAP_DIR_LOCAL_INO - Use inode B-Tree localization
926  * for directory entries.  Also see HAMMER_DIR_INODE_LOCALIZATION().
927  */
928 #define HAMMER_INODE_CAP_DIRHASH_MASK	0x03	/* directory: hash algorithm */
929 #define HAMMER_INODE_CAP_DIRHASH_ALG0	0x00
930 #define HAMMER_INODE_CAP_DIRHASH_ALG1	0x01
931 #define HAMMER_INODE_CAP_DIRHASH_ALG2	0x02
932 #define HAMMER_INODE_CAP_DIRHASH_ALG3	0x03
933 #define HAMMER_INODE_CAP_DIR_LOCAL_INO	0x04	/* use inode localization */
934 
935 #define HAMMER_DATA_DOALIGN(offset)				\
936 	(((offset) + 15) & ~15)
937 #define HAMMER_DATA_DOALIGN_WITH(type, offset)			\
938 	(((type)(offset) + 15) & (~(type)15))
939 
940 /*
941  * A HAMMER directory entry associates a HAMMER filesystem object with a
942  * namespace.  It is hooked into a pseudo-filesystem (with its own inode
943  * numbering space) in the filesystem by setting the high 16 bits of the
944  * localization field.  The low 16 bits must be 0 and are reserved for
945  * future use.
946  *
947  * Directory entries are indexed with a 128 bit namekey rather then an
948  * offset.  A portion of the namekey is an iterator/randomizer to deal
949  * with collisions.
950  *
951  * NOTE: leaf.base.obj_type from the related B-Tree leaf entry holds
952  * the filesystem object type of obj_id, e.g. a den_type equivalent.
953  * It is not stored in hammer_direntry_data.
954  *
955  * NOTE: name field / the filename data reference is NOT terminated with \0.
956  */
957 typedef struct hammer_direntry_data {
958 	int64_t obj_id;			/* object being referenced */
959 	uint32_t localization;		/* identify pseudo-filesystem */
960 	uint32_t reserved01;
961 	char	name[16];		/* name (extended) */
962 } *hammer_direntry_data_t;
963 
964 #define HAMMER_ENTRY_NAME_OFF	offsetof(struct hammer_direntry_data, name[0])
965 #define HAMMER_ENTRY_SIZE(nlen)	offsetof(struct hammer_direntry_data, name[nlen])
966 
967 /*
968  * Symlink data which does not fit in the inode is stored in a separate
969  * FIX type record.
970  */
971 typedef struct hammer_symlink_data {
972 	char	name[16];		/* name (extended) */
973 } *hammer_symlink_data_t;
974 
975 #define HAMMER_SYMLINK_NAME_OFF	offsetof(struct hammer_symlink_data, name[0])
976 
977 /*
978  * The root inode for the primary filesystem and root inode for any
979  * pseudo-fs may be tagged with an optional data structure using
980  * HAMMER_RECTYPE_PFS and localization id.  This structure allows
981  * the node to be used as a mirroring master or slave.
982  *
983  * When operating as a slave CD's into the node automatically become read-only
984  * and as-of sync_end_tid.
985  *
986  * When operating as a master the read PFSD info sets sync_end_tid to
987  * the most recently flushed TID.
988  *
989  * sync_low_tid is not yet used but will represent the highest pruning
990  * end-point, after which full history is available.
991  *
992  * We need to pack this structure making it equally sized on both 32-bit and
993  * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is
994  * send over the wire in hammer mirror operations. Only on 64-bit machines
995  * the size of this struct differ when packed or not. This leads us to the
996  * situation where old 64-bit systems (using the non-packed structure),
997  * which were never able to mirror to/from 32-bit systems, are now no longer
998  * able to mirror to/from newer 64-bit systems (using the packed structure).
999  */
1000 struct hammer_pseudofs_data {
1001 	hammer_tid_t	sync_low_tid;	/* full history beyond this point */
1002 	hammer_tid_t	sync_beg_tid;	/* earliest tid w/ full history avail */
1003 	hammer_tid_t	sync_end_tid;	/* current synchronizatoin point */
1004 	uint64_t	sync_beg_ts;	/* real-time of last completed sync */
1005 	uint64_t	sync_end_ts;	/* initiation of current sync cycle */
1006 	hammer_uuid_t	shared_uuid;	/* shared uuid (match required) */
1007 	hammer_uuid_t	unique_uuid;	/* unique uuid of this master/slave */
1008 	int32_t		reserved01;	/* reserved for future master_id */
1009 	int32_t		mirror_flags;	/* misc flags */
1010 	char		label[64];	/* filesystem space label */
1011 	char		snapshots[64];	/* softlink dir for pruning */
1012 	int32_t		reserved02;	/* was prune_{time,freq} */
1013 	int32_t		reserved03;	/* was reblock_{time,freq} */
1014 	int32_t		reserved04;	/* was snapshot_freq */
1015 	int32_t		prune_min;	/* do not prune recent history */
1016 	int32_t		prune_max;	/* do not retain history beyond here */
1017 	int32_t		reserved[16];
1018 } __packed;
1019 
1020 typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t;
1021 
1022 #define HAMMER_PFSD_SLAVE	0x00000001
1023 #define HAMMER_PFSD_DELETED	0x80000000
1024 
1025 #define hammer_is_pfs_slave(pfsd)			\
1026 	(((pfsd)->mirror_flags & HAMMER_PFSD_SLAVE) != 0)
1027 #define hammer_is_pfs_master(pfsd)			\
1028 	(!hammer_is_pfs_slave(pfsd))
1029 #define hammer_is_pfs_deleted(pfsd)			\
1030 	(((pfsd)->mirror_flags & HAMMER_PFSD_DELETED) != 0)
1031 
1032 #define HAMMER_MAX_PFS		65536
1033 #define HAMMER_MAX_PFSID	(HAMMER_MAX_PFS - 1)
1034 #define HAMMER_ROOT_PFSID	0
1035 
1036 /*
1037  * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }.
1038  *
1039  * Snapshot records replace the old <fs>/snapshots/<softlink> methodology.  Snapshot
1040  * records are mirrored but may be independently managed once they are laid down on
1041  * a slave.
1042  *
1043  * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the
1044  *	 results.
1045  *
1046  * NOTE: Reserved fields must be zero (as usual)
1047  */
1048 typedef struct hammer_snapshot_data {
1049 	hammer_tid_t	tid;		/* the snapshot TID itself (== key) */
1050 	uint64_t	ts;		/* real-time when snapshot was made */
1051 	uint64_t	reserved01;
1052 	uint64_t	reserved02;
1053 	char		label[64];	/* user-supplied description */
1054 	uint64_t	reserved03[4];
1055 } *hammer_snapshot_data_t;
1056 
1057 /*
1058  * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }.
1059  *
1060  * Used to store the hammer cleanup config.  This data is not mirrored.
1061  */
1062 typedef struct hammer_config_data {
1063 	char		text[1024];
1064 } *hammer_config_data_t;
1065 
1066 /*
1067  * Rollup various structures embedded as record data
1068  */
1069 typedef union hammer_data_ondisk {
1070 	struct hammer_direntry_data entry;
1071 	struct hammer_inode_data inode;
1072 	struct hammer_symlink_data symlink;
1073 	struct hammer_pseudofs_data pfsd;
1074 	struct hammer_snapshot_data snap;
1075 	struct hammer_config_data config;
1076 } *hammer_data_ondisk_t;
1077 
1078 /*
1079  * Ondisk layout of B-Tree related structures
1080  */
1081 #if 0	 /* Not needed for fstype(8) */
1082 #include "hammer_btree.h"
1083 #endif
1084 
1085 #define HAMMER_DIR_INODE_LOCALIZATION(ino_data)				\
1086 	(((ino_data)->cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO) ?	\
1087 	 HAMMER_LOCALIZE_INODE :					\
1088 	 HAMMER_LOCALIZE_MISC)
1089 
1090 #endif /* !VFS_HAMMER_DISK_H_ */
1091