xref: /src/usr.sbin/fstyp/hammer_disk.h (revision 836ac989933bcd5b662979bfdec429a201c123a6)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 2007 The DragonFly Project.  All rights reserved.
5  *
6  * This code is derived from software contributed to The DragonFly Project
7  * by Matthew Dillon <dillon@backplane.com>
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in
17  *    the documentation and/or other materials provided with the
18  *    distribution.
19  * 3. Neither the name of The DragonFly Project nor the names of its
20  *    contributors may be used to endorse or promote products derived
21  *    from this software without specific, prior written permission.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
26  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
27  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
28  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
29  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
30  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
31  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
32  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
33  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  * $DragonFly: src/sys/vfs/hammer/hammer_disk.h,v 1.55 2008/11/13 02:18:43 dillon Exp $
37  */
38 
39 #ifndef VFS_HAMMER_DISK_H_
40 #define VFS_HAMMER_DISK_H_
41 
42 #include <sys/endian.h>
43 
44 #ifndef _SYS_UUID_H_
45 #include <sys/uuid.h>
46 #endif
47 
48 /*
49  * The structures below represent the on-disk format for a HAMMER
50  * filesystem.  Note that all fields for on-disk structures are naturally
51  * aligned.  HAMMER uses little endian for fields in on-disk structures.
52  * HAMMER doesn't support big endian arch, but is planned.
53  *
54  * Most of HAMMER revolves around the concept of an object identifier.  An
55  * obj_id is a 64 bit quantity which uniquely identifies a filesystem object
56  * FOR THE ENTIRE LIFE OF THE FILESYSTEM.  This uniqueness allows backups
57  * and mirrors to retain varying amounts of filesystem history by removing
58  * any possibility of conflict through identifier reuse.
59  *
60  * A HAMMER filesystem may span multiple volumes.
61  *
62  * A HAMMER filesystem uses a 16K filesystem buffer size.  All filesystem
63  * I/O is done in multiples of 16K.
64  *
65  * 64K X-bufs are used for blocks >= a file's 1MB mark.
66  *
67  * Per-volume storage limit: 52 bits		4096 TB
68  * Per-Zone storage limit: 60 bits		1 MTB
69  * Per-filesystem storage limit: 60 bits	1 MTB
70  */
71 #define HAMMER_BUFSIZE		16384
72 #define HAMMER_XBUFSIZE		65536
73 #define HAMMER_HBUFSIZE		(HAMMER_BUFSIZE / 2)
74 #define HAMMER_XDEMARC		(1024 * 1024)
75 #define HAMMER_BUFMASK		(HAMMER_BUFSIZE - 1)
76 #define HAMMER_XBUFMASK		(HAMMER_XBUFSIZE - 1)
77 
78 #define HAMMER_BUFSIZE64	((uint64_t)HAMMER_BUFSIZE)
79 #define HAMMER_BUFMASK64	((uint64_t)HAMMER_BUFMASK)
80 
81 #define HAMMER_XBUFSIZE64	((uint64_t)HAMMER_XBUFSIZE)
82 #define HAMMER_XBUFMASK64	((uint64_t)HAMMER_XBUFMASK)
83 
84 #define HAMMER_OFF_ZONE_MASK	0xF000000000000000ULL /* zone portion */
85 #define HAMMER_OFF_VOL_MASK	0x0FF0000000000000ULL /* volume portion */
86 #define HAMMER_OFF_SHORT_MASK	0x000FFFFFFFFFFFFFULL /* offset portion */
87 #define HAMMER_OFF_LONG_MASK	0x0FFFFFFFFFFFFFFFULL /* offset portion */
88 
89 #define HAMMER_OFF_BAD		((hammer_off_t)-1)
90 
91 #define HAMMER_BUFSIZE_DOALIGN(offset)				\
92 	(((offset) + HAMMER_BUFMASK) & ~HAMMER_BUFMASK)
93 #define HAMMER_BUFSIZE64_DOALIGN(offset)			\
94 	(((offset) + HAMMER_BUFMASK64) & ~HAMMER_BUFMASK64)
95 
96 #define HAMMER_XBUFSIZE_DOALIGN(offset)				\
97 	(((offset) + HAMMER_XBUFMASK) & ~HAMMER_XBUFMASK)
98 #define HAMMER_XBUFSIZE64_DOALIGN(offset)			\
99 	(((offset) + HAMMER_XBUFMASK64) & ~HAMMER_XBUFMASK64)
100 
101 /*
102  * The current limit of volumes that can make up a HAMMER FS
103  */
104 #define HAMMER_MAX_VOLUMES	256
105 
106 /*
107  * Reserved space for (future) header junk after the volume header.
108  */
109 #define HAMMER_MIN_VOL_JUNK	(HAMMER_BUFSIZE * 16)	/* 256 KB */
110 #define HAMMER_MAX_VOL_JUNK	HAMMER_MIN_VOL_JUNK
111 #define HAMMER_VOL_JUNK_SIZE	HAMMER_MIN_VOL_JUNK
112 
113 /*
114  * Hammer transaction ids are 64 bit unsigned integers and are usually
115  * synchronized with the time of day in nanoseconds.
116  *
117  * Hammer offsets are used for FIFO indexing and embed a cycle counter
118  * and volume number in addition to the offset.  Most offsets are required
119  * to be 16 KB aligned.
120  */
121 typedef uint64_t hammer_tid_t;
122 typedef uint64_t hammer_off_t;
123 typedef uint32_t hammer_crc_t;
124 typedef uuid_t hammer_uuid_t;
125 
126 #define HAMMER_MIN_TID		0ULL			/* unsigned */
127 #define HAMMER_MAX_TID		0xFFFFFFFFFFFFFFFFULL	/* unsigned */
128 #define HAMMER_MIN_KEY		-0x8000000000000000LL	/* signed */
129 #define HAMMER_MAX_KEY		0x7FFFFFFFFFFFFFFFLL	/* signed */
130 #define HAMMER_MIN_OBJID	HAMMER_MIN_KEY		/* signed */
131 #define HAMMER_MAX_OBJID	HAMMER_MAX_KEY		/* signed */
132 #define HAMMER_MIN_RECTYPE	0x0U			/* unsigned */
133 #define HAMMER_MAX_RECTYPE	0xFFFFU			/* unsigned */
134 #define HAMMER_MIN_OFFSET	0ULL			/* unsigned */
135 #define HAMMER_MAX_OFFSET	0xFFFFFFFFFFFFFFFFULL	/* unsigned */
136 
137 /*
138  * hammer_off_t has several different encodings.  Note that not all zones
139  * encode a vol_no.  Zone bits are not a part of filesystem capacity as
140  * the zone bits aren't directly or indirectly mapped to physical volumes.
141  *
142  * In other words, HAMMER's logical filesystem offset consists of 64 bits,
143  * but the filesystem is considered 60 bits filesystem, not 64 bits.
144  * The maximum filesystem capacity is 1EB, not 16EB.
145  *
146  * zone 0:		available, a big-block that contains the offset is unused
147  * zone 1 (z,v,o):	raw volume relative (offset 0 is the volume header)
148  * zone 2 (z,v,o):	raw buffer relative (offset 0 is the first buffer)
149  * zone 3 (z,o):	undo/redo fifo	- fixed zone-2 offset array in volume header
150  * zone 4 (z,v,o):	freemap		- only real blockmap
151  * zone 8 (z,v,o):	B-Tree		- actually zone-2 address
152  * zone 9 (z,v,o):	meta		- actually zone-2 address
153  * zone 10 (z,v,o):	large-data	- actually zone-2 address
154  * zone 11 (z,v,o):	small-data	- actually zone-2 address
155  * zone 15:		unavailable, usually the offset is beyond volume size
156  *
157  * layer1/layer2 direct map:
158  *	     Maximum HAMMER filesystem capacity from volume aspect
159  *	     2^8(max volumes) * 2^52(max volume size) = 2^60 = 1EB (long offset)
160  *	    <------------------------------------------------------------->
161  *	     8bits   52bits (short offset)
162  *	    <------><----------------------------------------------------->
163  *	zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo
164  *	----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo
165  *	    <-----------------><------------------><---------------------->
166  *	     18bits             19bits              23bits
167  *	    <------------------------------------------------------------->
168  *	     2^18(layer1) * 2^19(layer2) * 2^23(big-block) = 2^60 = 1EB
169  *	     Maximum HAMMER filesystem capacity from blockmap aspect
170  *
171  * volume#0 layout
172  *	+-------------------------> offset 0 of a device/partition
173  *	| volume header (1928 bytes)
174  *	| the rest of header junk space (HAMMER_BUFSIZE aligned)
175  *	+-------------------------> vol_bot_beg
176  *	| boot area (HAMMER_BUFSIZE aligned)
177  *	+-------------------------> vol_mem_beg
178  *	| memory log (HAMMER_BUFSIZE aligned)
179  *	+-------------------------> vol_buf_beg (physical offset of zone-2)
180  *	| zone-4 big-block for layer1
181  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE
182  *	| zone-4 big-blocks for layer2
183  *	| ... (1 big-block per 4TB space)
184  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
185  *	| zone-3 big-blocks for UNDO/REDO FIFO
186  *	| ... (max 128 big-blocks)
187  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
188  *	| zone-8 big-block for root B-Tree node/etc
189  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
190  *	| zone-9 big-block for root inode/PFS/etc
191  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
192  *	| zone-X big-blocks
193  *	| ... (big-blocks for new zones after newfs_hammer)
194  *	| ...
195  *	| ...
196  *	| ...
197  *	| ...
198  *	+-------------------------> vol_buf_end (HAMMER_BUFSIZE aligned)
199  *	+-------------------------> end of a device/partition
200  *
201  * volume#N layout (0<N<256)
202  *	+-------------------------> offset 0 of a device/partition
203  *	| volume header (1928 bytes)
204  *	| the rest of header junk space (HAMMER_BUFSIZE aligned)
205  *	+-------------------------> vol_bot_beg
206  *	| boot area (HAMMER_BUFSIZE aligned)
207  *	+-------------------------> vol_mem_beg
208  *	| memory log (HAMMER_BUFSIZE aligned)
209  *	+-------------------------> vol_buf_beg (physical offset of zone-2)
210  *	| zone-4 big-blocks for layer2
211  *	| ... (1 big-block per 4TB space)
212  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
213  *	| zone-X big-blocks
214  *	| ... (unused until volume#(N-1) runs out of space)
215  *	| ...
216  *	| ...
217  *	| ...
218  *	| ...
219  *	+-------------------------> vol_buf_end (HAMMER_BUFSIZE aligned)
220  *	+-------------------------> end of a device/partition
221  */
222 
223 #define HAMMER_ZONE_RAW_VOLUME		0x1000000000000000ULL
224 #define HAMMER_ZONE_RAW_BUFFER		0x2000000000000000ULL
225 #define HAMMER_ZONE_UNDO		0x3000000000000000ULL
226 #define HAMMER_ZONE_FREEMAP		0x4000000000000000ULL
227 #define HAMMER_ZONE_RESERVED05		0x5000000000000000ULL  /* not used */
228 #define HAMMER_ZONE_RESERVED06		0x6000000000000000ULL  /* not used */
229 #define HAMMER_ZONE_RESERVED07		0x7000000000000000ULL  /* not used */
230 #define HAMMER_ZONE_BTREE		0x8000000000000000ULL
231 #define HAMMER_ZONE_META		0x9000000000000000ULL
232 #define HAMMER_ZONE_LARGE_DATA		0xA000000000000000ULL
233 #define HAMMER_ZONE_SMALL_DATA		0xB000000000000000ULL
234 #define HAMMER_ZONE_RESERVED0C		0xC000000000000000ULL  /* not used */
235 #define HAMMER_ZONE_RESERVED0D		0xD000000000000000ULL  /* not used */
236 #define HAMMER_ZONE_RESERVED0E		0xE000000000000000ULL  /* not used */
237 #define HAMMER_ZONE_UNAVAIL		0xF000000000000000ULL
238 
239 #define HAMMER_ZONE_RAW_VOLUME_INDEX	1
240 #define HAMMER_ZONE_RAW_BUFFER_INDEX	2
241 #define HAMMER_ZONE_UNDO_INDEX		3
242 #define HAMMER_ZONE_FREEMAP_INDEX	4
243 #define HAMMER_ZONE_BTREE_INDEX		8
244 #define HAMMER_ZONE_META_INDEX		9
245 #define HAMMER_ZONE_LARGE_DATA_INDEX	10
246 #define HAMMER_ZONE_SMALL_DATA_INDEX	11
247 #define HAMMER_ZONE_UNAVAIL_INDEX	15
248 
249 #define HAMMER_MAX_ZONES		16
250 
251 #define HAMMER_ZONE(offset)		((offset) & HAMMER_OFF_ZONE_MASK)
252 
253 #define hammer_is_zone_raw_volume(offset)		\
254 	(HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_VOLUME)
255 #define hammer_is_zone_raw_buffer(offset)		\
256 	(HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_BUFFER)
257 #define hammer_is_zone_undo(offset)			\
258 	(HAMMER_ZONE(offset) == HAMMER_ZONE_UNDO)
259 #define hammer_is_zone_freemap(offset)			\
260 	(HAMMER_ZONE(offset) == HAMMER_ZONE_FREEMAP)
261 #define hammer_is_zone_btree(offset)			\
262 	(HAMMER_ZONE(offset) == HAMMER_ZONE_BTREE)
263 #define hammer_is_zone_meta(offset)			\
264 	(HAMMER_ZONE(offset) == HAMMER_ZONE_META)
265 #define hammer_is_zone_large_data(offset)		\
266 	(HAMMER_ZONE(offset) == HAMMER_ZONE_LARGE_DATA)
267 #define hammer_is_zone_small_data(offset)		\
268 	(HAMMER_ZONE(offset) == HAMMER_ZONE_SMALL_DATA)
269 #define hammer_is_zone_unavail(offset)			\
270 	(HAMMER_ZONE(offset) == HAMMER_ZONE_UNAVAIL)
271 #define hammer_is_zone_data(offset)			\
272 	(hammer_is_zone_large_data(offset) || hammer_is_zone_small_data(offset))
273 
274 #define hammer_is_index_record(zone)			\
275 	((zone) >= HAMMER_ZONE_BTREE_INDEX &&		\
276 	 (zone) < HAMMER_MAX_ZONES)
277 
278 #define hammer_is_zone_record(offset)			\
279 	hammer_is_index_record(HAMMER_ZONE_DECODE(offset))
280 
281 #define hammer_is_index_direct_xlated(zone)		\
282 	(((zone) == HAMMER_ZONE_RAW_BUFFER_INDEX) ||	\
283 	 ((zone) == HAMMER_ZONE_FREEMAP_INDEX) ||	\
284 	 hammer_is_index_record(zone))
285 
286 #define hammer_is_zone_direct_xlated(offset)		\
287 	hammer_is_index_direct_xlated(HAMMER_ZONE_DECODE(offset))
288 
289 #define HAMMER_ZONE_ENCODE(zone, ham_off)		\
290 	(((hammer_off_t)(zone) << 60) | (ham_off))
291 #define HAMMER_ZONE_DECODE(ham_off)			\
292 	((int)(((hammer_off_t)(ham_off) >> 60)))
293 
294 #define HAMMER_VOL_ENCODE(vol_no)			\
295 	((hammer_off_t)((vol_no) & 255) << 52)
296 #define HAMMER_VOL_DECODE(ham_off)			\
297 	((int)(((hammer_off_t)(ham_off) >> 52) & 255))
298 
299 #define HAMMER_OFF_SHORT_ENCODE(offset)			\
300 	((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
301 #define HAMMER_OFF_LONG_ENCODE(offset)			\
302 	((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
303 
304 #define HAMMER_ENCODE(zone, vol_no, offset)		\
305 	(((hammer_off_t)(zone) << 60) |			\
306 	HAMMER_VOL_ENCODE(vol_no) |			\
307 	HAMMER_OFF_SHORT_ENCODE(offset))
308 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset)	\
309 	HAMMER_ENCODE(HAMMER_ZONE_RAW_VOLUME_INDEX, vol_no, offset)
310 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset)	\
311 	HAMMER_ENCODE(HAMMER_ZONE_RAW_BUFFER_INDEX, vol_no, offset)
312 #define HAMMER_ENCODE_UNDO(offset)			\
313 	HAMMER_ENCODE(HAMMER_ZONE_UNDO_INDEX, HAMMER_ROOT_VOLNO, offset)
314 #define HAMMER_ENCODE_FREEMAP(vol_no, offset)		\
315 	HAMMER_ENCODE(HAMMER_ZONE_FREEMAP_INDEX, vol_no, offset)
316 
317 /*
318  * Translate a zone address to zone-X address.
319  */
320 #define hammer_xlate_to_zoneX(zone, offset)		\
321 	HAMMER_ZONE_ENCODE((zone), (offset) & ~HAMMER_OFF_ZONE_MASK)
322 #define hammer_xlate_to_zone2(offset)			\
323 	hammer_xlate_to_zoneX(HAMMER_ZONE_RAW_BUFFER_INDEX, (offset))
324 
325 #define hammer_data_zone(data_len)			\
326 	(((data_len) >= HAMMER_BUFSIZE) ?		\
327 	 HAMMER_ZONE_LARGE_DATA :			\
328 	 HAMMER_ZONE_SMALL_DATA)
329 #define hammer_data_zone_index(data_len)		\
330 	(((data_len) >= HAMMER_BUFSIZE) ?		\
331 	 HAMMER_ZONE_LARGE_DATA_INDEX :			\
332 	 HAMMER_ZONE_SMALL_DATA_INDEX)
333 
334 /*
335  * Big-Block backing store
336  *
337  * A blockmap is a two-level map which translates a blockmap-backed zone
338  * offset into a raw zone 2 offset.  The layer 1 handles 18 bits and the
339  * layer 2 handles 19 bits.  The 8M big-block size is 23 bits so two
340  * layers gives us 18+19+23 = 60 bits of address space.
341  *
342  * When using hinting for a blockmap lookup, the hint is lost when the
343  * scan leaves the HINTBLOCK, which is typically several BIGBLOCK's.
344  * HINTBLOCK is a heuristic.
345  */
346 #define HAMMER_HINTBLOCK_SIZE		(HAMMER_BIGBLOCK_SIZE * 4)
347 #define HAMMER_HINTBLOCK_MASK64		((uint64_t)HAMMER_HINTBLOCK_SIZE - 1)
348 #define HAMMER_BIGBLOCK_SIZE		(8192 * 1024)
349 #define HAMMER_BIGBLOCK_SIZE64		((uint64_t)HAMMER_BIGBLOCK_SIZE)
350 #define HAMMER_BIGBLOCK_MASK		(HAMMER_BIGBLOCK_SIZE - 1)
351 #define HAMMER_BIGBLOCK_MASK64		((uint64_t)HAMMER_BIGBLOCK_SIZE - 1)
352 #define HAMMER_BIGBLOCK_BITS		23
353 #if 0
354 #define HAMMER_BIGBLOCK_OVERFILL	(6144 * 1024)
355 #endif
356 #if (1 << HAMMER_BIGBLOCK_BITS) != HAMMER_BIGBLOCK_SIZE
357 #error "HAMMER_BIGBLOCK_BITS BROKEN"
358 #endif
359 
360 #define HAMMER_BUFFERS_PER_BIGBLOCK			\
361 	(HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE)
362 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK		\
363 	(HAMMER_BUFFERS_PER_BIGBLOCK - 1)
364 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK64		\
365 	((hammer_off_t)HAMMER_BUFFERS_PER_BIGBLOCK_MASK)
366 
367 #define HAMMER_BIGBLOCK_DOALIGN(offset)				\
368 	(((offset) + HAMMER_BIGBLOCK_MASK64) & ~HAMMER_BIGBLOCK_MASK64)
369 
370 /*
371  * Maximum number of mirrors operating in master mode (multi-master
372  * clustering and mirroring). Note that HAMMER1 does not support
373  * multi-master clustering as of 2015.
374  */
375 #define HAMMER_MAX_MASTERS		16
376 
377 /*
378  * The blockmap is somewhat of a degenerate structure.  HAMMER only actually
379  * uses it in its original incarnation to implement the freemap.
380  *
381  * zone:1	raw volume (no blockmap)
382  * zone:2	raw buffer (no blockmap)
383  * zone:3	undomap    (direct layer2 array in volume header)
384  * zone:4	freemap    (the only real blockmap)
385  * zone:8-15	zone id used to classify big-block only, address is actually
386  *		a zone-2 address.
387  */
388 typedef struct hammer_blockmap {
389 	hammer_off_t	phys_offset;  /* zone-2 offset only used by zone-4 */
390 	hammer_off_t	first_offset; /* zone-X offset only used by zone-3 */
391 	hammer_off_t	next_offset;  /* zone-X offset for allocation */
392 	hammer_off_t	alloc_offset; /* zone-X offset only used by zone-3 */
393 	uint32_t	reserved01;
394 	hammer_crc_t	entry_crc;
395 } *hammer_blockmap_t;
396 
397 #define HAMMER_BLOCKMAP_CRCSIZE	\
398 	offsetof(struct hammer_blockmap, entry_crc)
399 
400 /*
401  * The blockmap is a 2-layer entity made up of big-blocks.  The first layer
402  * contains 262144 32-byte entries (18 bits), the second layer contains
403  * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
404  * 18+19+23 = 60 bits.  The top four bits are the zone id.
405  *
406  * Currently only the freemap utilizes both layers in all their glory.
407  * All primary data/meta-data zones actually encode a zone-2 address
408  * requiring no real blockmap translation.
409  *
410  * The freemap uses the upper 8 bits of layer-1 to identify the volume,
411  * thus any space allocated via the freemap can be directly translated
412  * to a zone:2 (or zone:8-15) address.
413  *
414  * zone-X blockmap offset: [zone:4][layer1:18][layer2:19][big-block:23]
415  */
416 
417 /*
418  * 32 bytes layer1 entry for 8MB big-block.
419  * A big-block can hold 2^23 / 2^5 = 2^18 layer1 entries,
420  * which equals bits assigned for layer1 in zone-2 address.
421  */
422 typedef struct hammer_blockmap_layer1 {
423 	hammer_off_t	blocks_free;	/* big-blocks free */
424 	hammer_off_t	phys_offset;	/* UNAVAIL or zone-2 */
425 	hammer_off_t	reserved01;
426 	hammer_crc_t	layer2_crc;	/* xor'd crc's of HAMMER_BLOCKSIZE */
427 					/* (not yet used) */
428 	hammer_crc_t	layer1_crc;	/* MUST BE LAST FIELD OF STRUCTURE*/
429 } *hammer_blockmap_layer1_t;
430 
431 #define HAMMER_LAYER1_CRCSIZE	\
432 	offsetof(struct hammer_blockmap_layer1, layer1_crc)
433 
434 /*
435  * 16 bytes layer2 entry for 8MB big-blocks.
436  * A big-block can hold 2^23 / 2^4 = 2^19 layer2 entries,
437  * which equals bits assigned for layer2 in zone-2 address.
438  *
439  * NOTE: bytes_free is signed and can legally go negative if/when data
440  *	 de-dup occurs.  This field will never go higher than
441  *	 HAMMER_BIGBLOCK_SIZE.  If exactly HAMMER_BIGBLOCK_SIZE
442  *	 the big-block is completely free.
443  */
444 typedef struct hammer_blockmap_layer2 {
445 	uint8_t		zone;		/* typed allocation zone */
446 	uint8_t		reserved01;
447 	uint16_t	reserved02;
448 	uint32_t	append_off;	/* allocatable space index */
449 	int32_t		bytes_free;	/* bytes free within this big-block */
450 	hammer_crc_t	entry_crc;
451 } *hammer_blockmap_layer2_t;
452 
453 #define HAMMER_LAYER2_CRCSIZE	\
454 	offsetof(struct hammer_blockmap_layer2, entry_crc)
455 
456 #define HAMMER_BLOCKMAP_UNAVAIL	((hammer_off_t)-1LL)
457 
458 #define HAMMER_BLOCKMAP_RADIX1	/* 2^18 = 262144 */	\
459 	((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1)))
460 #define HAMMER_BLOCKMAP_RADIX2	/* 2^19 = 524288 */	\
461 	((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2)))
462 
463 #define HAMMER_BLOCKMAP_LAYER1	/* 2^(18+19+23) = 1EB */	\
464 	(HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
465 #define HAMMER_BLOCKMAP_LAYER2	/* 2^(19+23) = 4TB */		\
466 	(HAMMER_BLOCKMAP_RADIX2 * HAMMER_BIGBLOCK_SIZE64)
467 
468 #define HAMMER_BLOCKMAP_LAYER1_MASK	(HAMMER_BLOCKMAP_LAYER1 - 1)
469 #define HAMMER_BLOCKMAP_LAYER2_MASK	(HAMMER_BLOCKMAP_LAYER2 - 1)
470 
471 #define HAMMER_BLOCKMAP_LAYER2_DOALIGN(offset)			\
472 	(((offset) + HAMMER_BLOCKMAP_LAYER2_MASK) &		\
473 	 ~HAMMER_BLOCKMAP_LAYER2_MASK)
474 
475 /*
476  * Index within layer1 or layer2 big-block for the entry representing
477  * a zone-2 physical offset.
478  */
479 #define HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset)		\
480 	((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) /	\
481 	 HAMMER_BLOCKMAP_LAYER2))
482 
483 #define HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset)		\
484 	((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) /	\
485 	HAMMER_BIGBLOCK_SIZE64))
486 
487 /*
488  * Byte offset within layer1 or layer2 big-block for the entry representing
489  * a zone-2 physical offset.  Multiply the index by sizeof(blockmap_layer).
490  */
491 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset)		\
492 	(HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) *		\
493 	 sizeof(struct hammer_blockmap_layer1))
494 
495 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset)		\
496 	(HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) *		\
497 	 sizeof(struct hammer_blockmap_layer2))
498 
499 /*
500  * Move on to offset 0 of the next layer1 or layer2.
501  */
502 #define HAMMER_ZONE_LAYER1_NEXT_OFFSET(offset)			\
503 	(((offset) + HAMMER_BLOCKMAP_LAYER2) & ~HAMMER_BLOCKMAP_LAYER2_MASK)
504 
505 #define HAMMER_ZONE_LAYER2_NEXT_OFFSET(offset)			\
506 	(((offset) + HAMMER_BIGBLOCK_SIZE) & ~HAMMER_BIGBLOCK_MASK64)
507 
508 /*
509  * HAMMER UNDO parameters.  The UNDO fifo is mapped directly in the volume
510  * header with an array of zone-2 offsets.  A maximum of (128x8MB) = 1GB,
511  * and minimum of (64x8MB) = 512MB may be reserved.  The size of the undo
512  * fifo is usually set a newfs time.
513  */
514 #define HAMMER_MIN_UNDO_BIGBLOCKS		64
515 #define HAMMER_MAX_UNDO_BIGBLOCKS		128
516 
517 /*
518  * All on-disk HAMMER structures which make up elements of the UNDO FIFO
519  * contain a hammer_fifo_head and hammer_fifo_tail structure.  This structure
520  * contains all the information required to validate the fifo element
521  * and to scan the fifo in either direction.  The head is typically embedded
522  * in higher level hammer on-disk structures while the tail is typically
523  * out-of-band.  hdr_size is the size of the whole mess, including the tail.
524  *
525  * All undo structures are guaranteed to not cross a 16K filesystem
526  * buffer boundary.  Most undo structures are fairly small.  Data spaces
527  * are not immediately reused by HAMMER so file data is not usually recorded
528  * as part of an UNDO.
529  *
530  * PAD elements are allowed to take up only 8 bytes of space as a special
531  * case, containing only hdr_signature, hdr_type, and hdr_size fields,
532  * and with the tail overloaded onto the head structure for 8 bytes total.
533  *
534  * Every undo record has a sequence number.  This number is unrelated to
535  * transaction ids and instead collects the undo transactions associated
536  * with a single atomic operation.  A larger transactional operation, such
537  * as a remove(), may consist of several smaller atomic operations
538  * representing raw meta-data operations.
539  *
540  *				HAMMER VERSION 4 CHANGES
541  *
542  * In HAMMER version 4 the undo structure alignment is reduced from 16384
543  * to 512 bytes in order to ensure that each 512 byte sector begins with
544  * a header.  The hdr_seq field in the header is a 32 bit sequence number
545  * which allows the recovery code to detect missing sectors
546  * without relying on the 32-bit crc and to definitively identify the current
547  * undo sequence space without having to rely on information from the volume
548  * header.  In addition, new REDO entries in the undo space are used to
549  * record write, write/extend, and transaction id updates.
550  *
551  * The grand result is:
552  *
553  * (1) The volume header no longer needs to be synchronized for most
554  *     flush and fsync operations.
555  *
556  * (2) Most fsync operations need only lay down REDO records
557  *
558  * (3) Data overwrite for nohistory operations covered by REDO records
559  *     can be supported (instead of rolling a new block allocation),
560  *     by rolling UNDO for the prior contents of the data.
561  *
562  *				HAMMER VERSION 5 CHANGES
563  *
564  * Hammer version 5 contains a minor adjustment making layer2's bytes_free
565  * field signed, allowing dedup to push it into the negative domain.
566  */
567 #define HAMMER_HEAD_ALIGN		8
568 #define HAMMER_HEAD_ALIGN_MASK		(HAMMER_HEAD_ALIGN - 1)
569 #define HAMMER_HEAD_DOALIGN(bytes)	\
570 	(((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK)
571 
572 #define HAMMER_UNDO_ALIGN		512
573 #define HAMMER_UNDO_ALIGN64		((uint64_t)512)
574 #define HAMMER_UNDO_MASK		(HAMMER_UNDO_ALIGN - 1)
575 #define HAMMER_UNDO_MASK64		(HAMMER_UNDO_ALIGN64 - 1)
576 #define HAMMER_UNDO_DOALIGN(offset)	\
577 	(((offset) + HAMMER_UNDO_MASK) & ~HAMMER_UNDO_MASK64)
578 
579 typedef struct hammer_fifo_head {
580 	uint16_t hdr_signature;
581 	uint16_t hdr_type;
582 	uint32_t hdr_size;	/* Aligned size of the whole mess */
583 	uint32_t hdr_seq;	/* Sequence number */
584 	hammer_crc_t hdr_crc;	/* XOR crc up to field w/ crc after field */
585 } *hammer_fifo_head_t;
586 
587 #define HAMMER_FIFO_HEAD_CRCOFF	offsetof(struct hammer_fifo_head, hdr_crc)
588 
589 typedef struct hammer_fifo_tail {
590 	uint16_t tail_signature;
591 	uint16_t tail_type;
592 	uint32_t tail_size;	/* aligned size of the whole mess */
593 } *hammer_fifo_tail_t;
594 
595 /*
596  * Fifo header types.
597  *
598  * NOTE: 0x8000U part of HAMMER_HEAD_TYPE_PAD can be removed if the HAMMER
599  * version ever gets bumped again. It exists only to keep compatibility with
600  * older versions.
601  */
602 #define HAMMER_HEAD_TYPE_PAD	(0x0040U | 0x8000U)
603 #define HAMMER_HEAD_TYPE_DUMMY	0x0041U		/* dummy entry w/seqno */
604 #define HAMMER_HEAD_TYPE_UNDO	0x0043U		/* random UNDO information */
605 #define HAMMER_HEAD_TYPE_REDO	0x0044U		/* data REDO / fast fsync */
606 
607 #define HAMMER_HEAD_SIGNATURE	0xC84EU
608 #define HAMMER_TAIL_SIGNATURE	0xC74FU
609 
610 /*
611  * Misc FIFO structures.
612  *
613  * UNDO - Raw meta-data media updates.
614  */
615 typedef struct hammer_fifo_undo {
616 	struct hammer_fifo_head	head;
617 	hammer_off_t		undo_offset;	/* zone-1,2 offset */
618 	int32_t			undo_data_bytes;
619 	int32_t			undo_reserved01;
620 	/* followed by data */
621 } *hammer_fifo_undo_t;
622 
623 /*
624  * REDO (HAMMER version 4+) - Logical file writes/truncates.
625  *
626  * REDOs contain information which will be duplicated in a later meta-data
627  * update, allowing fast write()+fsync() operations.  REDOs can be ignored
628  * without harming filesystem integrity but must be processed if fsync()
629  * semantics are desired.
630  *
631  * Unlike UNDOs which are processed backwards within the recovery span,
632  * REDOs must be processed forwards starting further back (starting outside
633  * the recovery span).
634  *
635  *	WRITE	- Write logical file (with payload).  Executed both
636  *		  out-of-span and in-span.  Out-of-span WRITEs may be
637  *		  filtered out by TERMs.
638  *
639  *	TRUNC	- Truncate logical file (no payload).  Executed both
640  *		  out-of-span and in-span.  Out-of-span WRITEs may be
641  *		  filtered out by TERMs.
642  *
643  *	TERM_*	- Indicates meta-data was committed (if out-of-span) or
644  *		  will be rolled-back (in-span).  Any out-of-span TERMs
645  *		  matching earlier WRITEs remove those WRITEs from
646  *		  consideration as they might conflict with a later data
647  *		  commit (which is not being rolled-back).
648  *
649  *	SYNC	- The earliest in-span SYNC (the last one when scanning
650  *		  backwards) tells the recovery code how far out-of-span
651  *		  it must go to run REDOs.
652  *
653  * NOTE: WRITEs do not always have matching TERMs even under
654  *	 perfect conditions because truncations might remove the
655  *	 buffers from consideration.  I/O problems can also remove
656  *	 buffers from consideration.
657  *
658  *	 TRUNCSs do not always have matching TERMs because several
659  *	 truncations may be aggregated together into a single TERM.
660  */
661 typedef struct hammer_fifo_redo {
662 	struct hammer_fifo_head	head;
663 	int64_t			redo_objid;	/* file being written */
664 	hammer_off_t		redo_offset;	/* logical offset in file */
665 	int32_t			redo_data_bytes;
666 	uint32_t		redo_flags;
667 	uint32_t		redo_localization;
668 	uint32_t		redo_reserved01;
669 	uint64_t		redo_reserved02;
670 	/* followed by data */
671 } *hammer_fifo_redo_t;
672 
673 #define HAMMER_REDO_WRITE	0x00000001
674 #define HAMMER_REDO_TRUNC	0x00000002
675 #define HAMMER_REDO_TERM_WRITE	0x00000004
676 #define HAMMER_REDO_TERM_TRUNC	0x00000008
677 #define HAMMER_REDO_SYNC	0x00000010
678 
679 typedef union hammer_fifo_any {
680 	struct hammer_fifo_head	head;
681 	struct hammer_fifo_undo	undo;
682 	struct hammer_fifo_redo	redo;
683 } *hammer_fifo_any_t;
684 
685 /*
686  * Volume header types
687  */
688 #define HAMMER_FSBUF_VOLUME	0xC8414D4DC5523031ULL	/* HAMMER01 */
689 #define HAMMER_FSBUF_VOLUME_REV	0x313052C54D4D41C8ULL	/* (reverse endian) */
690 
691 /*
692  * HAMMER Volume header
693  *
694  * A HAMMER filesystem can be built from 1-256 block devices, each block
695  * device contains a volume header followed by however many buffers fit
696  * into the volume.
697  *
698  * One of the volumes making up a HAMMER filesystem is the root volume.
699  * The root volume is always volume #0 which is the first block device path
700  * specified by newfs_hammer(8).  All HAMMER volumes have a volume header,
701  * however the root volume may be the only volume that has valid values for
702  * some fields in the header.
703  *
704  * Special field notes:
705  *
706  *	vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
707  *	vol_mem_beg - offset of memory log (buf_beg - mem_beg bytes)
708  *	vol_buf_beg - offset of the first buffer in volume
709  *	vol_buf_end - offset of volume EOF (on buffer boundary)
710  *
711  *	The memory log area allows a kernel to cache new records and data
712  *	in memory without allocating space in the actual filesystem to hold
713  *	the records and data.  In the event that a filesystem becomes full,
714  *	any records remaining in memory can be flushed to the memory log
715  *	area.  This allows the kernel to immediately return success.
716  *
717  *	The buffer offset is a physical offset of zone-2 offset. The lower
718  *	52 bits of the zone-2 offset is added to the buffer offset of each
719  *	volume to generate an actual I/O offset within the block device.
720  *
721  *	NOTE: boot area and memory log are currently not used.
722  */
723 
724 /*
725  * Filesystem type string
726  */
727 #define HAMMER_FSTYPE_STRING		"DragonFly HAMMER"
728 
729 /*
730  * These macros are only used by userspace when userspace commands either
731  * initialize or add a new HAMMER volume.
732  */
733 #define HAMMER_BOOT_MINBYTES		(32*1024)
734 #define HAMMER_BOOT_NOMBYTES		(64LL*1024*1024)
735 #define HAMMER_BOOT_MAXBYTES		(256LL*1024*1024)
736 
737 #define HAMMER_MEM_MINBYTES		(256*1024)
738 #define HAMMER_MEM_NOMBYTES		(1LL*1024*1024*1024)
739 #define HAMMER_MEM_MAXBYTES		(64LL*1024*1024*1024)
740 
741 typedef struct hammer_volume_ondisk {
742 	uint64_t vol_signature;	/* HAMMER_FSBUF_VOLUME for a valid header */
743 
744 	/*
745 	 * These are relative to block device offset, not zone offsets.
746 	 */
747 	int64_t vol_bot_beg;	/* offset of boot area */
748 	int64_t vol_mem_beg;	/* offset of memory log */
749 	int64_t vol_buf_beg;	/* offset of the first buffer in volume */
750 	int64_t vol_buf_end;	/* offset of volume EOF (on buffer boundary) */
751 	int64_t vol_reserved01;
752 
753 	hammer_uuid_t vol_fsid;	/* identify filesystem */
754 	hammer_uuid_t vol_fstype; /* identify filesystem type */
755 	char vol_label[64];	/* filesystem label */
756 
757 	int32_t vol_no;		/* volume number within filesystem */
758 	int32_t vol_count;	/* number of volumes making up filesystem */
759 
760 	uint32_t vol_version;	/* version control information */
761 	hammer_crc_t vol_crc;	/* header crc */
762 	uint32_t vol_flags;	/* volume flags */
763 	uint32_t vol_rootvol;	/* the root volume number (must be 0) */
764 
765 	uint32_t vol_reserved[8];
766 
767 	/*
768 	 * These fields are initialized and space is reserved in every
769 	 * volume making up a HAMMER filesystem, but only the root volume
770 	 * contains valid data.  Note that vol0_stat_bigblocks does not
771 	 * include big-blocks for freemap and undomap initially allocated
772 	 * by newfs_hammer(8).
773 	 */
774 	int64_t vol0_stat_bigblocks;	/* total big-blocks when fs is empty */
775 	int64_t vol0_stat_freebigblocks;/* number of free big-blocks */
776 	int64_t	vol0_reserved01;
777 	int64_t vol0_stat_inodes;	/* for statfs only */
778 	int64_t vol0_reserved02;
779 	hammer_off_t vol0_btree_root;	/* B-Tree root offset in zone-8 */
780 	hammer_tid_t vol0_next_tid;	/* highest partially synchronized TID */
781 	hammer_off_t vol0_reserved03;
782 
783 	/*
784 	 * Blockmaps for zones.  Not all zones use a blockmap.  Note that
785 	 * the entire root blockmap is cached in the hammer_mount structure.
786 	 */
787 	struct hammer_blockmap	vol0_blockmap[HAMMER_MAX_ZONES];
788 
789 	/*
790 	 * Array of zone-2 addresses for undo FIFO.
791 	 */
792 	hammer_off_t		vol0_undo_array[HAMMER_MAX_UNDO_BIGBLOCKS];
793 } *hammer_volume_ondisk_t;
794 
795 #define HAMMER_ROOT_VOLNO		0
796 
797 #define HAMMER_VOLF_NEEDFLUSH		0x0004	/* volume needs flush */
798 
799 #define HAMMER_VOL_CRCSIZE1	\
800 	offsetof(struct hammer_volume_ondisk, vol_crc)
801 #define HAMMER_VOL_CRCSIZE2	\
802 	(sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 -	\
803 	 sizeof(hammer_crc_t))
804 
805 #define HAMMER_VOL_VERSION_MIN		1	/* minimum supported version */
806 #define HAMMER_VOL_VERSION_DEFAULT	7	/* newfs default version */
807 #define HAMMER_VOL_VERSION_WIP		8	/* version >= this is WIP */
808 #define HAMMER_VOL_VERSION_MAX		7	/* maximum supported version */
809 
810 #define HAMMER_VOL_VERSION_ONE		1
811 #define HAMMER_VOL_VERSION_TWO		2	/* new dirent layout (2.3+) */
812 #define HAMMER_VOL_VERSION_THREE	3	/* new snapshot layout (2.5+) */
813 #define HAMMER_VOL_VERSION_FOUR		4	/* new undo/flush (2.5+) */
814 #define HAMMER_VOL_VERSION_FIVE		5	/* dedup (2.9+) */
815 #define HAMMER_VOL_VERSION_SIX		6	/* DIRHASH_ALG1 */
816 #define HAMMER_VOL_VERSION_SEVEN	7	/* use the faster iscsi_crc */
817 
818 /*
819  * Translate a zone-2 address to physical address
820  */
821 #define hammer_xlate_to_phys(volume, zone2_offset)	\
822 	((volume)->vol_buf_beg + HAMMER_OFF_SHORT_ENCODE(zone2_offset))
823 
824 /*
825  * Translate a zone-3 address to zone-2 address
826  */
827 #define HAMMER_UNDO_INDEX(zone3_offset)			\
828 	(HAMMER_OFF_SHORT_ENCODE(zone3_offset) / HAMMER_BIGBLOCK_SIZE)
829 
830 #define hammer_xlate_to_undo(volume, zone3_offset)			\
831 	((volume)->vol0_undo_array[HAMMER_UNDO_INDEX(zone3_offset)] +	\
832 	 (zone3_offset & HAMMER_BIGBLOCK_MASK64))
833 
834 /*
835  * Effective per-volume filesystem capacity including big-blocks for layer1/2
836  */
837 #define HAMMER_VOL_BUF_SIZE(volume)			\
838 	((volume)->vol_buf_end - (volume)->vol_buf_beg)
839 
840 /*
841  * Record types are fairly straightforward.  The B-Tree includes the record
842  * type in its index sort.
843  */
844 #define HAMMER_RECTYPE_UNKNOWN		0x0000
845 #define HAMMER_RECTYPE_INODE		0x0001	/* inode in obj_id space */
846 #define HAMMER_RECTYPE_DATA		0x0010
847 #define HAMMER_RECTYPE_DIRENTRY		0x0011
848 #define HAMMER_RECTYPE_DB		0x0012
849 #define HAMMER_RECTYPE_EXT		0x0013	/* ext attributes */
850 #define HAMMER_RECTYPE_FIX		0x0014	/* fixed attribute */
851 #define HAMMER_RECTYPE_PFS		0x0015	/* PFS management */
852 #define HAMMER_RECTYPE_SNAPSHOT		0x0016	/* Snapshot management */
853 #define HAMMER_RECTYPE_CONFIG		0x0017	/* hammer cleanup config */
854 #define HAMMER_RECTYPE_MAX		0xFFFF
855 
856 #define HAMMER_RECTYPE_ENTRY_START	(HAMMER_RECTYPE_INODE + 1)
857 #define HAMMER_RECTYPE_CLEAN_START	HAMMER_RECTYPE_EXT
858 
859 #define HAMMER_FIXKEY_SYMLINK		1
860 
861 #define HAMMER_OBJTYPE_UNKNOWN		0	/* never exists on-disk as unknown */
862 #define HAMMER_OBJTYPE_DIRECTORY	1
863 #define HAMMER_OBJTYPE_REGFILE		2
864 #define HAMMER_OBJTYPE_DBFILE		3
865 #define HAMMER_OBJTYPE_FIFO		4
866 #define HAMMER_OBJTYPE_CDEV		5
867 #define HAMMER_OBJTYPE_BDEV		6
868 #define HAMMER_OBJTYPE_SOFTLINK		7
869 #define HAMMER_OBJTYPE_PSEUDOFS		8	/* pseudo filesystem obj */
870 #define HAMMER_OBJTYPE_SOCKET		9
871 
872 /*
873  * HAMMER inode attribute data
874  *
875  * The data reference for a HAMMER inode points to this structure.  Any
876  * modifications to the contents of this structure will result in a
877  * replacement operation.
878  *
879  * parent_obj_id is only valid for directories (which cannot be hard-linked),
880  * and specifies the parent directory obj_id.  This field will also be set
881  * for non-directory inodes as a recovery aid, but can wind up holding
882  * stale information.  However, since object id's are not reused, the worse
883  * that happens is that the recovery code is unable to use it.
884  * A parent_obj_id of 0 means it's a root inode of root or non-root PFS.
885  *
886  * NOTE: Future note on directory hardlinks.  We can implement a record type
887  * which allows us to point to multiple parent directories.
888  */
889 typedef struct hammer_inode_data {
890 	uint16_t version;	/* inode data version */
891 	uint16_t mode;		/* basic unix permissions */
892 	uint32_t uflags;	/* chflags */
893 	uint32_t rmajor;	/* used by device nodes */
894 	uint32_t rminor;	/* used by device nodes */
895 	uint64_t ctime;
896 	int64_t parent_obj_id;	/* parent directory obj_id */
897 	hammer_uuid_t uid;
898 	hammer_uuid_t gid;
899 
900 	uint8_t obj_type;
901 	uint8_t cap_flags;	/* capability support flags (extension) */
902 	uint16_t reserved01;
903 	uint32_t reserved02;
904 	uint64_t nlinks;	/* hard links */
905 	uint64_t size;		/* filesystem object size */
906 	union {
907 		char	symlink[24];	/* HAMMER_INODE_BASESYMLEN */
908 	} ext;
909 	uint64_t mtime;	/* mtime must be second-to-last */
910 	uint64_t atime;	/* atime must be last */
911 } *hammer_inode_data_t;
912 
913 /*
914  * Neither mtime nor atime updates are CRCd by the B-Tree element.
915  * mtime updates have UNDO, atime updates do not.
916  */
917 #define HAMMER_INODE_CRCSIZE	\
918 	offsetof(struct hammer_inode_data, mtime)
919 
920 #define HAMMER_INODE_DATA_VERSION	1
921 #define HAMMER_OBJID_ROOT		1	/* root inodes # */
922 #define HAMMER_INODE_BASESYMLEN		24	/* see ext.symlink */
923 
924 /*
925  * Capability & implementation flags.
926  *
927  * HAMMER_INODE_CAP_DIR_LOCAL_INO - Use inode B-Tree localization
928  * for directory entries.  Also see HAMMER_DIR_INODE_LOCALIZATION().
929  */
930 #define HAMMER_INODE_CAP_DIRHASH_MASK	0x03	/* directory: hash algorithm */
931 #define HAMMER_INODE_CAP_DIRHASH_ALG0	0x00
932 #define HAMMER_INODE_CAP_DIRHASH_ALG1	0x01
933 #define HAMMER_INODE_CAP_DIRHASH_ALG2	0x02
934 #define HAMMER_INODE_CAP_DIRHASH_ALG3	0x03
935 #define HAMMER_INODE_CAP_DIR_LOCAL_INO	0x04	/* use inode localization */
936 
937 #define HAMMER_DATA_DOALIGN(offset)				\
938 	(((offset) + 15) & ~15)
939 #define HAMMER_DATA_DOALIGN_WITH(type, offset)			\
940 	(((type)(offset) + 15) & (~(type)15))
941 
942 /*
943  * A HAMMER directory entry associates a HAMMER filesystem object with a
944  * namespace.  It is hooked into a pseudo-filesystem (with its own inode
945  * numbering space) in the filesystem by setting the high 16 bits of the
946  * localization field.  The low 16 bits must be 0 and are reserved for
947  * future use.
948  *
949  * Directory entries are indexed with a 128 bit namekey rather then an
950  * offset.  A portion of the namekey is an iterator/randomizer to deal
951  * with collisions.
952  *
953  * NOTE: leaf.base.obj_type from the related B-Tree leaf entry holds
954  * the filesystem object type of obj_id, e.g. a den_type equivalent.
955  * It is not stored in hammer_direntry_data.
956  *
957  * NOTE: name field / the filename data reference is NOT terminated with \0.
958  */
959 typedef struct hammer_direntry_data {
960 	int64_t obj_id;			/* object being referenced */
961 	uint32_t localization;		/* identify pseudo-filesystem */
962 	uint32_t reserved01;
963 	char	name[16];		/* name (extended) */
964 } *hammer_direntry_data_t;
965 
966 #define HAMMER_ENTRY_NAME_OFF	offsetof(struct hammer_direntry_data, name[0])
967 #define HAMMER_ENTRY_SIZE(nlen)	offsetof(struct hammer_direntry_data, name[nlen])
968 
969 /*
970  * Symlink data which does not fit in the inode is stored in a separate
971  * FIX type record.
972  */
973 typedef struct hammer_symlink_data {
974 	char	name[16];		/* name (extended) */
975 } *hammer_symlink_data_t;
976 
977 #define HAMMER_SYMLINK_NAME_OFF	offsetof(struct hammer_symlink_data, name[0])
978 
979 /*
980  * The root inode for the primary filesystem and root inode for any
981  * pseudo-fs may be tagged with an optional data structure using
982  * HAMMER_RECTYPE_PFS and localization id.  This structure allows
983  * the node to be used as a mirroring master or slave.
984  *
985  * When operating as a slave CD's into the node automatically become read-only
986  * and as-of sync_end_tid.
987  *
988  * When operating as a master the read PFSD info sets sync_end_tid to
989  * the most recently flushed TID.
990  *
991  * sync_low_tid is not yet used but will represent the highest pruning
992  * end-point, after which full history is available.
993  *
994  * We need to pack this structure making it equally sized on both 32-bit and
995  * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is
996  * send over the wire in hammer mirror operations. Only on 64-bit machines
997  * the size of this struct differ when packed or not. This leads us to the
998  * situation where old 64-bit systems (using the non-packed structure),
999  * which were never able to mirror to/from 32-bit systems, are now no longer
1000  * able to mirror to/from newer 64-bit systems (using the packed structure).
1001  */
1002 struct hammer_pseudofs_data {
1003 	hammer_tid_t	sync_low_tid;	/* full history beyond this point */
1004 	hammer_tid_t	sync_beg_tid;	/* earliest tid w/ full history avail */
1005 	hammer_tid_t	sync_end_tid;	/* current synchronizatoin point */
1006 	uint64_t	sync_beg_ts;	/* real-time of last completed sync */
1007 	uint64_t	sync_end_ts;	/* initiation of current sync cycle */
1008 	hammer_uuid_t	shared_uuid;	/* shared uuid (match required) */
1009 	hammer_uuid_t	unique_uuid;	/* unique uuid of this master/slave */
1010 	int32_t		reserved01;	/* reserved for future master_id */
1011 	int32_t		mirror_flags;	/* misc flags */
1012 	char		label[64];	/* filesystem space label */
1013 	char		snapshots[64];	/* softlink dir for pruning */
1014 	int32_t		reserved02;	/* was prune_{time,freq} */
1015 	int32_t		reserved03;	/* was reblock_{time,freq} */
1016 	int32_t		reserved04;	/* was snapshot_freq */
1017 	int32_t		prune_min;	/* do not prune recent history */
1018 	int32_t		prune_max;	/* do not retain history beyond here */
1019 	int32_t		reserved[16];
1020 } __packed;
1021 
1022 typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t;
1023 
1024 #define HAMMER_PFSD_SLAVE	0x00000001
1025 #define HAMMER_PFSD_DELETED	0x80000000
1026 
1027 #define hammer_is_pfs_slave(pfsd)			\
1028 	(((pfsd)->mirror_flags & HAMMER_PFSD_SLAVE) != 0)
1029 #define hammer_is_pfs_master(pfsd)			\
1030 	(!hammer_is_pfs_slave(pfsd))
1031 #define hammer_is_pfs_deleted(pfsd)			\
1032 	(((pfsd)->mirror_flags & HAMMER_PFSD_DELETED) != 0)
1033 
1034 #define HAMMER_MAX_PFS		65536
1035 #define HAMMER_MAX_PFSID	(HAMMER_MAX_PFS - 1)
1036 #define HAMMER_ROOT_PFSID	0
1037 
1038 /*
1039  * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }.
1040  *
1041  * Snapshot records replace the old <fs>/snapshots/<softlink> methodology.  Snapshot
1042  * records are mirrored but may be independently managed once they are laid down on
1043  * a slave.
1044  *
1045  * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the
1046  *	 results.
1047  *
1048  * NOTE: Reserved fields must be zero (as usual)
1049  */
1050 typedef struct hammer_snapshot_data {
1051 	hammer_tid_t	tid;		/* the snapshot TID itself (== key) */
1052 	uint64_t	ts;		/* real-time when snapshot was made */
1053 	uint64_t	reserved01;
1054 	uint64_t	reserved02;
1055 	char		label[64];	/* user-supplied description */
1056 	uint64_t	reserved03[4];
1057 } *hammer_snapshot_data_t;
1058 
1059 /*
1060  * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }.
1061  *
1062  * Used to store the hammer cleanup config.  This data is not mirrored.
1063  */
1064 typedef struct hammer_config_data {
1065 	char		text[1024];
1066 } *hammer_config_data_t;
1067 
1068 /*
1069  * Rollup various structures embedded as record data
1070  */
1071 typedef union hammer_data_ondisk {
1072 	struct hammer_direntry_data entry;
1073 	struct hammer_inode_data inode;
1074 	struct hammer_symlink_data symlink;
1075 	struct hammer_pseudofs_data pfsd;
1076 	struct hammer_snapshot_data snap;
1077 	struct hammer_config_data config;
1078 } *hammer_data_ondisk_t;
1079 
1080 /*
1081  * Ondisk layout of B-Tree related structures
1082  */
1083 #if 0	 /* Not needed for fstype(8) */
1084 #include "hammer_btree.h"
1085 #endif
1086 
1087 #define HAMMER_DIR_INODE_LOCALIZATION(ino_data)				\
1088 	(((ino_data)->cap_flags & HAMMER_INODE_CAP_DIR_LOCAL_INO) ?	\
1089 	 HAMMER_LOCALIZE_INODE :					\
1090 	 HAMMER_LOCALIZE_MISC)
1091 
1092 #endif /* !VFS_HAMMER_DISK_H_ */
1093