xref: /netbsd-src/usr.sbin/fstyp/hammer_disk.h (revision d30e8975889765c89d2c3868fd91b7fc8bdbb712)
1 /*        $NetBSD: hammer_disk.h,v 1.6 2024/02/05 21:39:52 andvar Exp $      */
2 
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 #include <sys/cdefs.h>
39 __KERNEL_RCSID(0, "$NetBSD: hammer_disk.h,v 1.6 2024/02/05 21:39:52 andvar Exp $");
40 
41 #ifndef VFS_HAMMER_DISK_H_
42 #define VFS_HAMMER_DISK_H_
43 
44 #include <sys/endian.h>
45 
46 #ifndef _SYS_UUID_H_
47 #include <sys/uuid.h>
48 #endif
49 
50 /*
51  * The structures below represent the on-disk format for a HAMMER
52  * filesystem.  Note that all fields for on-disk structures are naturally
53  * aligned.  HAMMER uses little endian for fields in on-disk structures.
54  * HAMMER doesn't support big endian arch, but is planned.
55  *
56  * Most of HAMMER revolves around the concept of an object identifier.  An
57  * obj_id is a 64 bit quantity which uniquely identifies a filesystem object
58  * FOR THE ENTIRE LIFE OF THE FILESYSTEM.  This uniqueness allows backups
59  * and mirrors to retain varying amounts of filesystem history by removing
60  * any possibility of conflict through identifier reuse.
61  *
62  * A HAMMER filesystem may span multiple volumes.
63  *
64  * A HAMMER filesystem uses a 16K filesystem buffer size.  All filesystem
65  * I/O is done in multiples of 16K.
66  *
67  * 64K X-bufs are used for blocks >= a file's 1MB mark.
68  *
69  * Per-volume storage limit: 52 bits		4096 TB
70  * Per-Zone storage limit: 60 bits		1 MTB
71  * Per-filesystem storage limit: 60 bits	1 MTB
72  */
73 #define HAMMER_BUFSIZE		16384
74 #define HAMMER_XBUFSIZE		65536
75 #define HAMMER_HBUFSIZE		(HAMMER_BUFSIZE / 2)
76 #define HAMMER_XDEMARC		(1024 * 1024)
77 #define HAMMER_BUFMASK		(HAMMER_BUFSIZE - 1)
78 #define HAMMER_XBUFMASK		(HAMMER_XBUFSIZE - 1)
79 
80 #define HAMMER_BUFSIZE64	((uint64_t)HAMMER_BUFSIZE)
81 #define HAMMER_BUFMASK64	((uint64_t)HAMMER_BUFMASK)
82 
83 #define HAMMER_XBUFSIZE64	((uint64_t)HAMMER_XBUFSIZE)
84 #define HAMMER_XBUFMASK64	((uint64_t)HAMMER_XBUFMASK)
85 
86 #define HAMMER_OFF_ZONE_MASK	0xF000000000000000ULL /* zone portion */
87 #define HAMMER_OFF_VOL_MASK	0x0FF0000000000000ULL /* volume portion */
88 #define HAMMER_OFF_SHORT_MASK	0x000FFFFFFFFFFFFFULL /* offset portion */
89 #define HAMMER_OFF_LONG_MASK	0x0FFFFFFFFFFFFFFFULL /* offset portion */
90 
91 #define HAMMER_OFF_BAD		((hammer_off_t)-1)
92 
93 #define HAMMER_BUFSIZE_DOALIGN(offset)				\
94 	(((offset) + HAMMER_BUFMASK) & ~HAMMER_BUFMASK)
95 #define HAMMER_BUFSIZE64_DOALIGN(offset)			\
96 	(((offset) + HAMMER_BUFMASK64) & ~HAMMER_BUFMASK64)
97 
98 #define HAMMER_XBUFSIZE_DOALIGN(offset)				\
99 	(((offset) + HAMMER_XBUFMASK) & ~HAMMER_XBUFMASK)
100 #define HAMMER_XBUFSIZE64_DOALIGN(offset)			\
101 	(((offset) + HAMMER_XBUFMASK64) & ~HAMMER_XBUFMASK64)
102 
103 /*
104  * The current limit of volumes that can make up a HAMMER FS
105  */
106 #define HAMMER_MAX_VOLUMES	256
107 
108 /*
109  * Reserved space for (future) header junk after the volume header.
110  */
111 #define HAMMER_MIN_VOL_JUNK	(HAMMER_BUFSIZE * 16)	/* 256 KB */
112 #define HAMMER_MAX_VOL_JUNK	HAMMER_MIN_VOL_JUNK
113 #define HAMMER_VOL_JUNK_SIZE	HAMMER_MIN_VOL_JUNK
114 
115 /*
116  * Hammer transaction ids are 64 bit unsigned integers and are usually
117  * synchronized with the time of day in nanoseconds.
118  *
119  * Hammer offsets are used for FIFO indexing and embed a cycle counter
120  * and volume number in addition to the offset.  Most offsets are required
121  * to be 16 KB aligned.
122  */
123 typedef uint64_t hammer_tid_t;
124 typedef uint64_t hammer_off_t;
125 typedef uint32_t hammer_crc_t;
126 typedef uuid_t hammer_uuid_t;
127 
128 #define HAMMER_MIN_TID		0ULL			/* unsigned */
129 #define HAMMER_MAX_TID		0xFFFFFFFFFFFFFFFFULL	/* unsigned */
130 #define HAMMER_MIN_KEY		-0x8000000000000000LL	/* signed */
131 #define HAMMER_MAX_KEY		0x7FFFFFFFFFFFFFFFLL	/* signed */
132 #define HAMMER_MIN_OBJID	HAMMER_MIN_KEY		/* signed */
133 #define HAMMER_MAX_OBJID	HAMMER_MAX_KEY		/* signed */
134 #define HAMMER_MIN_RECTYPE	0x0U			/* unsigned */
135 #define HAMMER_MAX_RECTYPE	0xFFFFU			/* unsigned */
136 #define HAMMER_MIN_OFFSET	0ULL			/* unsigned */
137 #define HAMMER_MAX_OFFSET	0xFFFFFFFFFFFFFFFFULL	/* unsigned */
138 
139 /*
140  * hammer_off_t has several different encodings.  Note that not all zones
141  * encode a vol_no.  Zone bits are not a part of filesystem capacity as
142  * the zone bits aren't directly or indirectly mapped to physical volumes.
143  *
144  * In other words, HAMMER's logical filesystem offset consists of 64 bits,
145  * but the filesystem is considered 60 bits filesystem, not 64 bits.
146  * The maximum filesystem capacity is 1EB, not 16EB.
147  *
148  * zone 0:		available, a big-block that contains the offset is unused
149  * zone 1 (z,v,o):	raw volume relative (offset 0 is the volume header)
150  * zone 2 (z,v,o):	raw buffer relative (offset 0 is the first buffer)
151  * zone 3 (z,o):	undo/redo fifo	- fixed zone-2 offset array in volume header
152  * zone 4 (z,v,o):	freemap		- only real blockmap
153  * zone 8 (z,v,o):	B-Tree		- actually zone-2 address
154  * zone 9 (z,v,o):	meta		- actually zone-2 address
155  * zone 10 (z,v,o):	large-data	- actually zone-2 address
156  * zone 11 (z,v,o):	small-data	- actually zone-2 address
157  * zone 15:		unavailable, usually the offset is beyond volume size
158  *
159  * layer1/layer2 direct map:
160  *	     Maximum HAMMER filesystem capacity from volume aspect
161  *	     2^8(max volumes) * 2^52(max volume size) = 2^60 = 1EB (long offset)
162  *	    <------------------------------------------------------------->
163  *	     8bits   52bits (short offset)
164  *	    <------><----------------------------------------------------->
165  *	zzzzvvvvvvvvoooo oooooooooooooooo oooooooooooooooo oooooooooooooooo
166  *	----111111111111 1111112222222222 222222222ooooooo oooooooooooooooo
167  *	    <-----------------><------------------><---------------------->
168  *	     18bits             19bits              23bits
169  *	    <------------------------------------------------------------->
170  *	     2^18(layer1) * 2^19(layer2) * 2^23(big-block) = 2^60 = 1EB
171  *	     Maximum HAMMER filesystem capacity from blockmap aspect
172  *
173  * volume#0 layout
174  *	+-------------------------> offset 0 of a device/partition
175  *	| volume header (1928 bytes)
176  *	| the rest of header junk space (HAMMER_BUFSIZE aligned)
177  *	+-------------------------> vol_bot_beg
178  *	| boot area (HAMMER_BUFSIZE aligned)
179  *	+-------------------------> vol_mem_beg
180  *	| memory log (HAMMER_BUFSIZE aligned)
181  *	+-------------------------> vol_buf_beg (physical offset of zone-2)
182  *	| zone-4 big-block for layer1
183  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE
184  *	| zone-4 big-blocks for layer2
185  *	| ... (1 big-block per 4TB space)
186  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
187  *	| zone-3 big-blocks for UNDO/REDO FIFO
188  *	| ... (max 128 big-blocks)
189  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
190  *	| zone-8 big-block for root B-Tree node/etc
191  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
192  *	| zone-9 big-block for root inode/PFS/etc
193  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
194  *	| zone-X big-blocks
195  *	| ... (big-blocks for new zones after newfs_hammer)
196  *	| ...
197  *	| ...
198  *	| ...
199  *	| ...
200  *	+-------------------------> vol_buf_end (HAMMER_BUFSIZE aligned)
201  *	+-------------------------> end of a device/partition
202  *
203  * volume#N layout (0<N<256)
204  *	+-------------------------> offset 0 of a device/partition
205  *	| volume header (1928 bytes)
206  *	| the rest of header junk space (HAMMER_BUFSIZE aligned)
207  *	+-------------------------> vol_bot_beg
208  *	| boot area (HAMMER_BUFSIZE aligned)
209  *	+-------------------------> vol_mem_beg
210  *	| memory log (HAMMER_BUFSIZE aligned)
211  *	+-------------------------> vol_buf_beg (physical offset of zone-2)
212  *	| zone-4 big-blocks for layer2
213  *	| ... (1 big-block per 4TB space)
214  *	+-------------------------> vol_buf_beg + HAMMER_BIGBLOCK_SIZE * ...
215  *	| zone-X big-blocks
216  *	| ... (unused until volume#(N-1) runs out of space)
217  *	| ...
218  *	| ...
219  *	| ...
220  *	| ...
221  *	+-------------------------> vol_buf_end (HAMMER_BUFSIZE aligned)
222  *	+-------------------------> end of a device/partition
223  */
224 
225 #define HAMMER_ZONE_RAW_VOLUME		0x1000000000000000ULL
226 #define HAMMER_ZONE_RAW_BUFFER		0x2000000000000000ULL
227 #define HAMMER_ZONE_UNDO		0x3000000000000000ULL
228 #define HAMMER_ZONE_FREEMAP		0x4000000000000000ULL
229 #define HAMMER_ZONE_RESERVED05		0x5000000000000000ULL  /* not used */
230 #define HAMMER_ZONE_RESERVED06		0x6000000000000000ULL  /* not used */
231 #define HAMMER_ZONE_RESERVED07		0x7000000000000000ULL  /* not used */
232 #define HAMMER_ZONE_BTREE		0x8000000000000000ULL
233 #define HAMMER_ZONE_META		0x9000000000000000ULL
234 #define HAMMER_ZONE_LARGE_DATA		0xA000000000000000ULL
235 #define HAMMER_ZONE_SMALL_DATA		0xB000000000000000ULL
236 #define HAMMER_ZONE_RESERVED0C		0xC000000000000000ULL  /* not used */
237 #define HAMMER_ZONE_RESERVED0D		0xD000000000000000ULL  /* not used */
238 #define HAMMER_ZONE_RESERVED0E		0xE000000000000000ULL  /* not used */
239 #define HAMMER_ZONE_UNAVAIL		0xF000000000000000ULL
240 
241 #define HAMMER_ZONE_RAW_VOLUME_INDEX	1
242 #define HAMMER_ZONE_RAW_BUFFER_INDEX	2
243 #define HAMMER_ZONE_UNDO_INDEX		3
244 #define HAMMER_ZONE_FREEMAP_INDEX	4
245 #define HAMMER_ZONE_BTREE_INDEX		8
246 #define HAMMER_ZONE_META_INDEX		9
247 #define HAMMER_ZONE_LARGE_DATA_INDEX	10
248 #define HAMMER_ZONE_SMALL_DATA_INDEX	11
249 #define HAMMER_ZONE_UNAVAIL_INDEX	15
250 
251 #define HAMMER_MAX_ZONES		16
252 
253 #define HAMMER_ZONE(offset)		((offset) & HAMMER_OFF_ZONE_MASK)
254 
255 #define hammer_is_zone_raw_volume(offset)		\
256 	(HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_VOLUME)
257 #define hammer_is_zone_raw_buffer(offset)		\
258 	(HAMMER_ZONE(offset) == HAMMER_ZONE_RAW_BUFFER)
259 #define hammer_is_zone_undo(offset)			\
260 	(HAMMER_ZONE(offset) == HAMMER_ZONE_UNDO)
261 #define hammer_is_zone_freemap(offset)			\
262 	(HAMMER_ZONE(offset) == HAMMER_ZONE_FREEMAP)
263 #define hammer_is_zone_btree(offset)			\
264 	(HAMMER_ZONE(offset) == HAMMER_ZONE_BTREE)
265 #define hammer_is_zone_meta(offset)			\
266 	(HAMMER_ZONE(offset) == HAMMER_ZONE_META)
267 #define hammer_is_zone_large_data(offset)		\
268 	(HAMMER_ZONE(offset) == HAMMER_ZONE_LARGE_DATA)
269 #define hammer_is_zone_small_data(offset)		\
270 	(HAMMER_ZONE(offset) == HAMMER_ZONE_SMALL_DATA)
271 #define hammer_is_zone_unavail(offset)			\
272 	(HAMMER_ZONE(offset) == HAMMER_ZONE_UNAVAIL)
273 #define hammer_is_zone_data(offset)			\
274 	(hammer_is_zone_large_data(offset) || hammer_is_zone_small_data(offset))
275 
276 #define hammer_is_index_record(zone)			\
277 	((zone) >= HAMMER_ZONE_BTREE_INDEX &&		\
278 	 (zone) < HAMMER_MAX_ZONES)
279 
280 #define hammer_is_zone_record(offset)			\
281 	hammer_is_index_record(HAMMER_ZONE_DECODE(offset))
282 
283 #define hammer_is_index_direct_xlated(zone)		\
284 	(((zone) == HAMMER_ZONE_RAW_BUFFER_INDEX) ||	\
285 	 ((zone) == HAMMER_ZONE_FREEMAP_INDEX) ||	\
286 	 hammer_is_index_record(zone))
287 
288 #define hammer_is_zone_direct_xlated(offset)		\
289 	hammer_is_index_direct_xlated(HAMMER_ZONE_DECODE(offset))
290 
291 #define HAMMER_ZONE_ENCODE(zone, ham_off)		\
292 	(((hammer_off_t)(zone) << 60) | (ham_off))
293 #define HAMMER_ZONE_DECODE(ham_off)			\
294 	((int)(((hammer_off_t)(ham_off) >> 60)))
295 
296 #define HAMMER_VOL_ENCODE(vol_no)			\
297 	((hammer_off_t)((vol_no) & 255) << 52)
298 #define HAMMER_VOL_DECODE(ham_off)			\
299 	((int)(((hammer_off_t)(ham_off) >> 52) & 255))
300 
301 #define HAMMER_OFF_SHORT_ENCODE(offset)			\
302 	((hammer_off_t)(offset) & HAMMER_OFF_SHORT_MASK)
303 #define HAMMER_OFF_LONG_ENCODE(offset)			\
304 	((hammer_off_t)(offset) & HAMMER_OFF_LONG_MASK)
305 
306 #define HAMMER_ENCODE(zone, vol_no, offset)		\
307 	(((hammer_off_t)(zone) << 60) |			\
308 	HAMMER_VOL_ENCODE(vol_no) |			\
309 	HAMMER_OFF_SHORT_ENCODE(offset))
310 #define HAMMER_ENCODE_RAW_VOLUME(vol_no, offset)	\
311 	HAMMER_ENCODE(HAMMER_ZONE_RAW_VOLUME_INDEX, vol_no, offset)
312 #define HAMMER_ENCODE_RAW_BUFFER(vol_no, offset)	\
313 	HAMMER_ENCODE(HAMMER_ZONE_RAW_BUFFER_INDEX, vol_no, offset)
314 #define HAMMER_ENCODE_UNDO(offset)			\
315 	HAMMER_ENCODE(HAMMER_ZONE_UNDO_INDEX, HAMMER_ROOT_VOLNO, offset)
316 #define HAMMER_ENCODE_FREEMAP(vol_no, offset)		\
317 	HAMMER_ENCODE(HAMMER_ZONE_FREEMAP_INDEX, vol_no, offset)
318 
319 /*
320  * Translate a zone address to zone-X address.
321  */
322 #define hammer_xlate_to_zoneX(zone, offset)		\
323 	HAMMER_ZONE_ENCODE((zone), (offset) & ~HAMMER_OFF_ZONE_MASK)
324 #define hammer_xlate_to_zone2(offset)			\
325 	hammer_xlate_to_zoneX(HAMMER_ZONE_RAW_BUFFER_INDEX, (offset))
326 
327 #define hammer_data_zone(data_len)			\
328 	(((data_len) >= HAMMER_BUFSIZE) ?		\
329 	 HAMMER_ZONE_LARGE_DATA :			\
330 	 HAMMER_ZONE_SMALL_DATA)
331 #define hammer_data_zone_index(data_len)		\
332 	(((data_len) >= HAMMER_BUFSIZE) ?		\
333 	 HAMMER_ZONE_LARGE_DATA_INDEX :			\
334 	 HAMMER_ZONE_SMALL_DATA_INDEX)
335 
336 /*
337  * Big-Block backing store
338  *
339  * A blockmap is a two-level map which translates a blockmap-backed zone
340  * offset into a raw zone 2 offset.  The layer 1 handles 18 bits and the
341  * layer 2 handles 19 bits.  The 8M big-block size is 23 bits so two
342  * layers gives us 18+19+23 = 60 bits of address space.
343  *
344  * When using hinting for a blockmap lookup, the hint is lost when the
345  * scan leaves the HINTBLOCK, which is typically several BIGBLOCK's.
346  * HINTBLOCK is a heuristic.
347  */
348 #define HAMMER_HINTBLOCK_SIZE		(HAMMER_BIGBLOCK_SIZE * 4)
349 #define HAMMER_HINTBLOCK_MASK64		((uint64_t)HAMMER_HINTBLOCK_SIZE - 1)
350 #define HAMMER_BIGBLOCK_SIZE		(8192 * 1024)
351 #define HAMMER_BIGBLOCK_SIZE64		((uint64_t)HAMMER_BIGBLOCK_SIZE)
352 #define HAMMER_BIGBLOCK_MASK		(HAMMER_BIGBLOCK_SIZE - 1)
353 #define HAMMER_BIGBLOCK_MASK64		((uint64_t)HAMMER_BIGBLOCK_SIZE - 1)
354 #define HAMMER_BIGBLOCK_BITS		23
355 #if 0
356 #define HAMMER_BIGBLOCK_OVERFILL	(6144 * 1024)
357 #endif
358 #if (1 << HAMMER_BIGBLOCK_BITS) != HAMMER_BIGBLOCK_SIZE
359 #error "HAMMER_BIGBLOCK_BITS BROKEN"
360 #endif
361 
362 #define HAMMER_BUFFERS_PER_BIGBLOCK			\
363 	(HAMMER_BIGBLOCK_SIZE / HAMMER_BUFSIZE)
364 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK		\
365 	(HAMMER_BUFFERS_PER_BIGBLOCK - 1)
366 #define HAMMER_BUFFERS_PER_BIGBLOCK_MASK64		\
367 	((hammer_off_t)HAMMER_BUFFERS_PER_BIGBLOCK_MASK)
368 
369 #define HAMMER_BIGBLOCK_DOALIGN(offset)				\
370 	(((offset) + HAMMER_BIGBLOCK_MASK64) & ~HAMMER_BIGBLOCK_MASK64)
371 
372 /*
373  * Maximum number of mirrors operating in master mode (multi-master
374  * clustering and mirroring). Note that HAMMER1 does not support
375  * multi-master clustering as of 2015.
376  */
377 #define HAMMER_MAX_MASTERS		16
378 
379 /*
380  * The blockmap is somewhat of a degenerate structure.  HAMMER only actually
381  * uses it in its original incarnation to implement the freemap.
382  *
383  * zone:1	raw volume (no blockmap)
384  * zone:2	raw buffer (no blockmap)
385  * zone:3	undomap    (direct layer2 array in volume header)
386  * zone:4	freemap    (the only real blockmap)
387  * zone:8-15	zone id used to classify big-block only, address is actually
388  *		a zone-2 address.
389  */
390 typedef struct hammer_blockmap {
391 	hammer_off_t	phys_offset;  /* zone-2 offset only used by zone-4 */
392 	hammer_off_t	first_offset; /* zone-X offset only used by zone-3 */
393 	hammer_off_t	next_offset;  /* zone-X offset for allocation */
394 	hammer_off_t	alloc_offset; /* zone-X offset only used by zone-3 */
395 	uint32_t	reserved01;
396 	hammer_crc_t	entry_crc;
397 } *hammer_blockmap_t;
398 
399 #define HAMMER_BLOCKMAP_CRCSIZE	\
400 	offsetof(struct hammer_blockmap, entry_crc)
401 
402 /*
403  * The blockmap is a 2-layer entity made up of big-blocks.  The first layer
404  * contains 262144 32-byte entries (18 bits), the second layer contains
405  * 524288 16-byte entries (19 bits), representing 8MB (23 bit) blockmaps.
406  * 18+19+23 = 60 bits.  The top four bits are the zone id.
407  *
408  * Currently only the freemap utilizes both layers in all their glory.
409  * All primary data/meta-data zones actually encode a zone-2 address
410  * requiring no real blockmap translation.
411  *
412  * The freemap uses the upper 8 bits of layer-1 to identify the volume,
413  * thus any space allocated via the freemap can be directly translated
414  * to a zone:2 (or zone:8-15) address.
415  *
416  * zone-X blockmap offset: [zone:4][layer1:18][layer2:19][big-block:23]
417  */
418 
419 /*
420  * 32 bytes layer1 entry for 8MB big-block.
421  * A big-block can hold 2^23 / 2^5 = 2^18 layer1 entries,
422  * which equals bits assigned for layer1 in zone-2 address.
423  */
424 typedef struct hammer_blockmap_layer1 {
425 	hammer_off_t	blocks_free;	/* big-blocks free */
426 	hammer_off_t	phys_offset;	/* UNAVAIL or zone-2 */
427 	hammer_off_t	reserved01;
428 	hammer_crc_t	layer2_crc;	/* xor'd crc's of HAMMER_BLOCKSIZE */
429 					/* (not yet used) */
430 	hammer_crc_t	layer1_crc;	/* MUST BE LAST FIELD OF STRUCTURE*/
431 } *hammer_blockmap_layer1_t;
432 
433 #define HAMMER_LAYER1_CRCSIZE	\
434 	offsetof(struct hammer_blockmap_layer1, layer1_crc)
435 
436 /*
437  * 16 bytes layer2 entry for 8MB big-blocks.
438  * A big-block can hold 2^23 / 2^4 = 2^19 layer2 entries,
439  * which equals bits assigned for layer2 in zone-2 address.
440  *
441  * NOTE: bytes_free is signed and can legally go negative if/when data
442  *	 de-dup occurs.  This field will never go higher than
443  *	 HAMMER_BIGBLOCK_SIZE.  If exactly HAMMER_BIGBLOCK_SIZE
444  *	 the big-block is completely free.
445  */
446 typedef struct hammer_blockmap_layer2 {
447 	uint8_t		zone;		/* typed allocation zone */
448 	uint8_t		reserved01;
449 	uint16_t	reserved02;
450 	uint32_t	append_off;	/* allocatable space index */
451 	int32_t		bytes_free;	/* bytes free within this big-block */
452 	hammer_crc_t	entry_crc;
453 } *hammer_blockmap_layer2_t;
454 
455 #define HAMMER_LAYER2_CRCSIZE	\
456 	offsetof(struct hammer_blockmap_layer2, entry_crc)
457 
458 #define HAMMER_BLOCKMAP_UNAVAIL	((hammer_off_t)-1LL)
459 
460 #define HAMMER_BLOCKMAP_RADIX1	/* 2^18 = 262144 */	\
461 	((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer1)))
462 #define HAMMER_BLOCKMAP_RADIX2	/* 2^19 = 524288 */	\
463 	((int)(HAMMER_BIGBLOCK_SIZE / sizeof(struct hammer_blockmap_layer2)))
464 
465 #define HAMMER_BLOCKMAP_LAYER1	/* 2^(18+19+23) = 1EB */	\
466 	(HAMMER_BLOCKMAP_RADIX1 * HAMMER_BLOCKMAP_LAYER2)
467 #define HAMMER_BLOCKMAP_LAYER2	/* 2^(19+23) = 4TB */		\
468 	(HAMMER_BLOCKMAP_RADIX2 * HAMMER_BIGBLOCK_SIZE64)
469 
470 #define HAMMER_BLOCKMAP_LAYER1_MASK	(HAMMER_BLOCKMAP_LAYER1 - 1)
471 #define HAMMER_BLOCKMAP_LAYER2_MASK	(HAMMER_BLOCKMAP_LAYER2 - 1)
472 
473 #define HAMMER_BLOCKMAP_LAYER2_DOALIGN(offset)			\
474 	(((offset) + HAMMER_BLOCKMAP_LAYER2_MASK) &		\
475 	 ~HAMMER_BLOCKMAP_LAYER2_MASK)
476 
477 /*
478  * Index within layer1 or layer2 big-block for the entry representing
479  * a zone-2 physical offset.
480  */
481 #define HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset)		\
482 	((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER1_MASK) /	\
483 	 HAMMER_BLOCKMAP_LAYER2))
484 
485 #define HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset)		\
486 	((int)(((zone2_offset) & HAMMER_BLOCKMAP_LAYER2_MASK) /	\
487 	HAMMER_BIGBLOCK_SIZE64))
488 
489 /*
490  * Byte offset within layer1 or layer2 big-block for the entry representing
491  * a zone-2 physical offset.  Multiply the index by sizeof(blockmap_layer).
492  */
493 #define HAMMER_BLOCKMAP_LAYER1_OFFSET(zone2_offset)		\
494 	(HAMMER_BLOCKMAP_LAYER1_INDEX(zone2_offset) *		\
495 	 sizeof(struct hammer_blockmap_layer1))
496 
497 #define HAMMER_BLOCKMAP_LAYER2_OFFSET(zone2_offset)		\
498 	(HAMMER_BLOCKMAP_LAYER2_INDEX(zone2_offset) *		\
499 	 sizeof(struct hammer_blockmap_layer2))
500 
501 /*
502  * Move on to offset 0 of the next layer1 or layer2.
503  */
504 #define HAMMER_ZONE_LAYER1_NEXT_OFFSET(offset)			\
505 	(((offset) + HAMMER_BLOCKMAP_LAYER2) & ~HAMMER_BLOCKMAP_LAYER2_MASK)
506 
507 #define HAMMER_ZONE_LAYER2_NEXT_OFFSET(offset)			\
508 	(((offset) + HAMMER_BIGBLOCK_SIZE) & ~HAMMER_BIGBLOCK_MASK64)
509 
510 /*
511  * HAMMER UNDO parameters.  The UNDO fifo is mapped directly in the volume
512  * header with an array of zone-2 offsets.  A maximum of (128x8MB) = 1GB,
513  * and minimum of (64x8MB) = 512MB may be reserved.  The size of the undo
514  * fifo is usually set a newfs time.
515  */
516 #define HAMMER_MIN_UNDO_BIGBLOCKS		64
517 #define HAMMER_MAX_UNDO_BIGBLOCKS		128
518 
519 /*
520  * All on-disk HAMMER structures which make up elements of the UNDO FIFO
521  * contain a hammer_fifo_head and hammer_fifo_tail structure.  This structure
522  * contains all the information required to validate the fifo element
523  * and to scan the fifo in either direction.  The head is typically embedded
524  * in higher level hammer on-disk structures while the tail is typically
525  * out-of-band.  hdr_size is the size of the whole mess, including the tail.
526  *
527  * All undo structures are guaranteed to not cross a 16K filesystem
528  * buffer boundary.  Most undo structures are fairly small.  Data spaces
529  * are not immediately reused by HAMMER so file data is not usually recorded
530  * as part of an UNDO.
531  *
532  * PAD elements are allowed to take up only 8 bytes of space as a special
533  * case, containing only hdr_signature, hdr_type, and hdr_size fields,
534  * and with the tail overloaded onto the head structure for 8 bytes total.
535  *
536  * Every undo record has a sequence number.  This number is unrelated to
537  * transaction ids and instead collects the undo transactions associated
538  * with a single atomic operation.  A larger transactional operation, such
539  * as a remove(), may consist of several smaller atomic operations
540  * representing raw meta-data operations.
541  *
542  *				HAMMER VERSION 4 CHANGES
543  *
544  * In HAMMER version 4 the undo structure alignment is reduced from 16384
545  * to 512 bytes in order to ensure that each 512 byte sector begins with
546  * a header.  The hdr_seq field in the header is a 32 bit sequence number
547  * which allows the recovery code to detect missing sectors
548  * without relying on the 32-bit crc and to definitively identify the current
549  * undo sequence space without having to rely on information from the volume
550  * header.  In addition, new REDO entries in the undo space are used to
551  * record write, write/extend, and transaction id updates.
552  *
553  * The grand result is:
554  *
555  * (1) The volume header no longer needs to be synchronized for most
556  *     flush and fsync operations.
557  *
558  * (2) Most fsync operations need only lay down REDO records
559  *
560  * (3) Data overwrite for nohistory operations covered by REDO records
561  *     can be supported (instead of rolling a new block allocation),
562  *     by rolling UNDO for the prior contents of the data.
563  *
564  *				HAMMER VERSION 5 CHANGES
565  *
566  * Hammer version 5 contains a minor adjustment making layer2's bytes_free
567  * field signed, allowing dedup to push it into the negative domain.
568  */
569 #define HAMMER_HEAD_ALIGN		8
570 #define HAMMER_HEAD_ALIGN_MASK		(HAMMER_HEAD_ALIGN - 1)
571 #define HAMMER_HEAD_DOALIGN(bytes)	\
572 	(((bytes) + HAMMER_HEAD_ALIGN_MASK) & ~HAMMER_HEAD_ALIGN_MASK)
573 
574 #define HAMMER_UNDO_ALIGN		512
575 #define HAMMER_UNDO_ALIGN64		((uint64_t)512)
576 #define HAMMER_UNDO_MASK		(HAMMER_UNDO_ALIGN - 1)
577 #define HAMMER_UNDO_MASK64		(HAMMER_UNDO_ALIGN64 - 1)
578 #define HAMMER_UNDO_DOALIGN(offset)	\
579 	(((offset) + HAMMER_UNDO_MASK) & ~HAMMER_UNDO_MASK64)
580 
581 typedef struct hammer_fifo_head {
582 	uint16_t hdr_signature;
583 	uint16_t hdr_type;
584 	uint32_t hdr_size;	/* Aligned size of the whole mess */
585 	uint32_t hdr_seq;	/* Sequence number */
586 	hammer_crc_t hdr_crc;	/* XOR crc up to field w/ crc after field */
587 } *hammer_fifo_head_t;
588 
589 #define HAMMER_FIFO_HEAD_CRCOFF	offsetof(struct hammer_fifo_head, hdr_crc)
590 
591 typedef struct hammer_fifo_tail {
592 	uint16_t tail_signature;
593 	uint16_t tail_type;
594 	uint32_t tail_size;	/* aligned size of the whole mess */
595 } *hammer_fifo_tail_t;
596 
597 /*
598  * Fifo header types.
599  *
600  * NOTE: 0x8000U part of HAMMER_HEAD_TYPE_PAD can be removed if the HAMMER
601  * version ever gets bumped again. It exists only to keep compatibility with
602  * older versions.
603  */
604 #define HAMMER_HEAD_TYPE_PAD	(0x0040U | 0x8000U)
605 #define HAMMER_HEAD_TYPE_DUMMY	0x0041U		/* dummy entry w/seqno */
606 #define HAMMER_HEAD_TYPE_UNDO	0x0043U		/* random UNDO information */
607 #define HAMMER_HEAD_TYPE_REDO	0x0044U		/* data REDO / fast fsync */
608 
609 #define HAMMER_HEAD_SIGNATURE	0xC84EU
610 #define HAMMER_TAIL_SIGNATURE	0xC74FU
611 
612 /*
613  * Misc FIFO structures.
614  *
615  * UNDO - Raw meta-data media updates.
616  */
617 typedef struct hammer_fifo_undo {
618 	struct hammer_fifo_head	head;
619 	hammer_off_t		undo_offset;	/* zone-1,2 offset */
620 	int32_t			undo_data_bytes;
621 	int32_t			undo_reserved01;
622 	/* followed by data */
623 } *hammer_fifo_undo_t;
624 
625 /*
626  * REDO (HAMMER version 4+) - Logical file writes/truncates.
627  *
628  * REDOs contain information which will be duplicated in a later meta-data
629  * update, allowing fast write()+fsync() operations.  REDOs can be ignored
630  * without harming filesystem integrity but must be processed if fsync()
631  * semantics are desired.
632  *
633  * Unlike UNDOs which are processed backwards within the recovery span,
634  * REDOs must be processed forwards starting further back (starting outside
635  * the recovery span).
636  *
637  *	WRITE	- Write logical file (with payload).  Executed both
638  *		  out-of-span and in-span.  Out-of-span WRITEs may be
639  *		  filtered out by TERMs.
640  *
641  *	TRUNC	- Truncate logical file (no payload).  Executed both
642  *		  out-of-span and in-span.  Out-of-span WRITEs may be
643  *		  filtered out by TERMs.
644  *
645  *	TERM_*	- Indicates meta-data was committed (if out-of-span) or
646  *		  will be rolled-back (in-span).  Any out-of-span TERMs
647  *		  matching earlier WRITEs remove those WRITEs from
648  *		  consideration as they might conflict with a later data
649  *		  commit (which is not being rolled-back).
650  *
651  *	SYNC	- The earliest in-span SYNC (the last one when scanning
652  *		  backwards) tells the recovery code how far out-of-span
653  *		  it must go to run REDOs.
654  *
655  * NOTE: WRITEs do not always have matching TERMs even under
656  *	 perfect conditions because truncations might remove the
657  *	 buffers from consideration.  I/O problems can also remove
658  *	 buffers from consideration.
659  *
660  *	 TRUNCSs do not always have matching TERMs because several
661  *	 truncations may be aggregated together into a single TERM.
662  */
663 typedef struct hammer_fifo_redo {
664 	struct hammer_fifo_head	head;
665 	int64_t			redo_objid;	/* file being written */
666 	hammer_off_t		redo_offset;	/* logical offset in file */
667 	int32_t			redo_data_bytes;
668 	uint32_t		redo_flags;
669 	uint32_t		redo_localization;
670 	uint32_t		redo_reserved01;
671 	uint64_t		redo_reserved02;
672 	/* followed by data */
673 } *hammer_fifo_redo_t;
674 
675 #define HAMMER_REDO_WRITE	0x00000001
676 #define HAMMER_REDO_TRUNC	0x00000002
677 #define HAMMER_REDO_TERM_WRITE	0x00000004
678 #define HAMMER_REDO_TERM_TRUNC	0x00000008
679 #define HAMMER_REDO_SYNC	0x00000010
680 
681 typedef union hammer_fifo_any {
682 	struct hammer_fifo_head	head;
683 	struct hammer_fifo_undo	undo;
684 	struct hammer_fifo_redo	redo;
685 } *hammer_fifo_any_t;
686 
687 /*
688  * Volume header types
689  */
690 #define HAMMER_FSBUF_VOLUME	0xC8414D4DC5523031ULL	/* HAMMER01 */
691 #define HAMMER_FSBUF_VOLUME_REV	0x313052C54D4D41C8ULL	/* (reverse endian) */
692 
693 /*
694  * HAMMER Volume header
695  *
696  * A HAMMER filesystem can be built from 1-256 block devices, each block
697  * device contains a volume header followed by however many buffers fit
698  * into the volume.
699  *
700  * One of the volumes making up a HAMMER filesystem is the root volume.
701  * The root volume is always volume #0 which is the first block device path
702  * specified by newfs_hammer(8).  All HAMMER volumes have a volume header,
703  * however the root volume may be the only volume that has valid values for
704  * some fields in the header.
705  *
706  * Special field notes:
707  *
708  *	vol_bot_beg - offset of boot area (mem_beg - bot_beg bytes)
709  *	vol_mem_beg - offset of memory log (buf_beg - mem_beg bytes)
710  *	vol_buf_beg - offset of the first buffer in volume
711  *	vol_buf_end - offset of volume EOF (on buffer boundary)
712  *
713  *	The memory log area allows a kernel to cache new records and data
714  *	in memory without allocating space in the actual filesystem to hold
715  *	the records and data.  In the event that a filesystem becomes full,
716  *	any records remaining in memory can be flushed to the memory log
717  *	area.  This allows the kernel to immediately return success.
718  *
719  *	The buffer offset is a physical offset of zone-2 offset. The lower
720  *	52 bits of the zone-2 offset is added to the buffer offset of each
721  *	volume to generate an actual I/O offset within the block device.
722  *
723  *	NOTE: boot area and memory log are currently not used.
724  */
725 
726 /*
727  * Filesystem type string
728  */
729 #define HAMMER_FSTYPE_STRING		"DragonFly HAMMER"
730 
731 /*
732  * These macros are only used by userspace when userspace commands either
733  * initialize or add a new HAMMER volume.
734  */
735 #define HAMMER_BOOT_MINBYTES		(32*1024)
736 #define HAMMER_BOOT_NOMBYTES		(64LL*1024*1024)
737 #define HAMMER_BOOT_MAXBYTES		(256LL*1024*1024)
738 
739 #define HAMMER_MEM_MINBYTES		(256*1024)
740 #define HAMMER_MEM_NOMBYTES		(1LL*1024*1024*1024)
741 #define HAMMER_MEM_MAXBYTES		(64LL*1024*1024*1024)
742 
743 typedef struct hammer_volume_ondisk {
744 	uint64_t vol_signature;	/* HAMMER_FSBUF_VOLUME for a valid header */
745 
746 	/*
747 	 * These are relative to block device offset, not zone offsets.
748 	 */
749 	int64_t vol_bot_beg;	/* offset of boot area */
750 	int64_t vol_mem_beg;	/* offset of memory log */
751 	int64_t vol_buf_beg;	/* offset of the first buffer in volume */
752 	int64_t vol_buf_end;	/* offset of volume EOF (on buffer boundary) */
753 	int64_t vol_reserved01;
754 
755 	hammer_uuid_t vol_fsid;	/* identify filesystem */
756 	hammer_uuid_t vol_fstype; /* identify filesystem type */
757 	char vol_label[64];	/* filesystem label */
758 
759 	int32_t vol_no;		/* volume number within filesystem */
760 	int32_t vol_count;	/* number of volumes making up filesystem */
761 
762 	uint32_t vol_version;	/* version control information */
763 	hammer_crc_t vol_crc;	/* header crc */
764 	uint32_t vol_flags;	/* volume flags */
765 	uint32_t vol_rootvol;	/* the root volume number (must be 0) */
766 
767 	uint32_t vol_reserved[8];
768 
769 	/*
770 	 * These fields are initialized and space is reserved in every
771 	 * volume making up a HAMMER filesystem, but only the root volume
772 	 * contains valid data.  Note that vol0_stat_bigblocks does not
773 	 * include big-blocks for freemap and undomap initially allocated
774 	 * by newfs_hammer(8).
775 	 */
776 	int64_t vol0_stat_bigblocks;	/* total big-blocks when fs is empty */
777 	int64_t vol0_stat_freebigblocks;/* number of free big-blocks */
778 	int64_t	vol0_reserved01;
779 	int64_t vol0_stat_inodes;	/* for statfs only */
780 	int64_t vol0_reserved02;
781 	hammer_off_t vol0_btree_root;	/* B-Tree root offset in zone-8 */
782 	hammer_tid_t vol0_next_tid;	/* highest partially synchronized TID */
783 	hammer_off_t vol0_reserved03;
784 
785 	/*
786 	 * Blockmaps for zones.  Not all zones use a blockmap.  Note that
787 	 * the entire root blockmap is cached in the hammer_mount structure.
788 	 */
789 	struct hammer_blockmap	vol0_blockmap[HAMMER_MAX_ZONES];
790 
791 	/*
792 	 * Array of zone-2 addresses for undo FIFO.
793 	 */
794 	hammer_off_t		vol0_undo_array[HAMMER_MAX_UNDO_BIGBLOCKS];
795 } *hammer_volume_ondisk_t;
796 
797 #define HAMMER_ROOT_VOLNO		0
798 
799 #define HAMMER_VOLF_NEEDFLUSH		0x0004	/* volume needs flush */
800 
801 #define HAMMER_VOL_CRCSIZE1	\
802 	offsetof(struct hammer_volume_ondisk, vol_crc)
803 #define HAMMER_VOL_CRCSIZE2	\
804 	(sizeof(struct hammer_volume_ondisk) - HAMMER_VOL_CRCSIZE1 -	\
805 	 sizeof(hammer_crc_t))
806 
807 #define HAMMER_VOL_VERSION_MIN		1	/* minimum supported version */
808 #define HAMMER_VOL_VERSION_DEFAULT	7	/* newfs default version */
809 #define HAMMER_VOL_VERSION_WIP		8	/* version >= this is WIP */
810 #define HAMMER_VOL_VERSION_MAX		7	/* maximum supported version */
811 
812 #define HAMMER_VOL_VERSION_ONE		1
813 #define HAMMER_VOL_VERSION_TWO		2	/* new dirent layout (2.3+) */
814 #define HAMMER_VOL_VERSION_THREE	3	/* new snapshot layout (2.5+) */
815 #define HAMMER_VOL_VERSION_FOUR		4	/* new undo/flush (2.5+) */
816 #define HAMMER_VOL_VERSION_FIVE		5	/* dedup (2.9+) */
817 #define HAMMER_VOL_VERSION_SIX		6	/* DIRHASH_ALG1 */
818 #define HAMMER_VOL_VERSION_SEVEN	7	/* use the faster iscsi_crc */
819 
820 /*
821  * Translate a zone-2 address to physical address
822  */
823 #define hammer_xlate_to_phys(volume, zone2_offset)	\
824 	((volume)->vol_buf_beg + HAMMER_OFF_SHORT_ENCODE(zone2_offset))
825 
826 /*
827  * Translate a zone-3 address to zone-2 address
828  */
829 #define HAMMER_UNDO_INDEX(zone3_offset)			\
830 	(HAMMER_OFF_SHORT_ENCODE(zone3_offset) / HAMMER_BIGBLOCK_SIZE)
831 
832 #define hammer_xlate_to_undo(volume, zone3_offset)			\
833 	((volume)->vol0_undo_array[HAMMER_UNDO_INDEX(zone3_offset)] +	\
834 	 (zone3_offset & HAMMER_BIGBLOCK_MASK64))
835 
836 /*
837  * Effective per-volume filesystem capacity including big-blocks for layer1/2
838  */
839 #define HAMMER_VOL_BUF_SIZE(volume)			\
840 	((volume)->vol_buf_end - (volume)->vol_buf_beg)
841 
842 /*
843  * Record types are fairly straightforward.  The B-Tree includes the record
844  * type in its index sort.
845  */
846 #define HAMMER_RECTYPE_UNKNOWN		0x0000
847 #define HAMMER_RECTYPE_INODE		0x0001	/* inode in obj_id space */
848 #define HAMMER_RECTYPE_DATA		0x0010
849 #define HAMMER_RECTYPE_DIRENTRY		0x0011
850 #define HAMMER_RECTYPE_DB		0x0012
851 #define HAMMER_RECTYPE_EXT		0x0013	/* ext attributes */
852 #define HAMMER_RECTYPE_FIX		0x0014	/* fixed attribute */
853 #define HAMMER_RECTYPE_PFS		0x0015	/* PFS management */
854 #define HAMMER_RECTYPE_SNAPSHOT		0x0016	/* Snapshot management */
855 #define HAMMER_RECTYPE_CONFIG		0x0017	/* hammer cleanup config */
856 #define HAMMER_RECTYPE_MAX		0xFFFF
857 
858 #define HAMMER_RECTYPE_ENTRY_START	(HAMMER_RECTYPE_INODE + 1)
859 #define HAMMER_RECTYPE_CLEAN_START	HAMMER_RECTYPE_EXT
860 
861 #define HAMMER_FIXKEY_SYMLINK		1
862 
863 #define HAMMER_OBJTYPE_UNKNOWN		0	/* never exists on-disk as unknown */
864 #define HAMMER_OBJTYPE_DIRECTORY	1
865 #define HAMMER_OBJTYPE_REGFILE		2
866 #define HAMMER_OBJTYPE_DBFILE		3
867 #define HAMMER_OBJTYPE_FIFO		4
868 #define HAMMER_OBJTYPE_CDEV		5
869 #define HAMMER_OBJTYPE_BDEV		6
870 #define HAMMER_OBJTYPE_SOFTLINK		7
871 #define HAMMER_OBJTYPE_PSEUDOFS		8	/* pseudo filesystem obj */
872 #define HAMMER_OBJTYPE_SOCKET		9
873 
874 /*
875  * HAMMER inode attribute data
876  *
877  * The data reference for a HAMMER inode points to this structure.  Any
878  * modifications to the contents of this structure will result in a
879  * replacement operation.
880  *
881  * parent_obj_id is only valid for directories (which cannot be hard-linked),
882  * and specifies the parent directory obj_id.  This field will also be set
883  * for non-directory inodes as a recovery aid, but can wind up holding
884  * stale information.  However, since object id's are not reused, the worse
885  * that happens is that the recovery code is unable to use it.
886  * A parent_obj_id of 0 means it's a root inode of root or non-root PFS.
887  *
888  * NOTE: Future note on directory hardlinks.  We can implement a record type
889  * which allows us to point to multiple parent directories.
890  */
891 typedef struct hammer_inode_data {
892 	uint16_t version;	/* inode data version */
893 	uint16_t mode;		/* basic unix permissions */
894 	uint32_t uflags;	/* chflags */
895 	uint32_t rmajor;	/* used by device nodes */
896 	uint32_t rminor;	/* used by device nodes */
897 	uint64_t ctime;
898 	int64_t parent_obj_id;	/* parent directory obj_id */
899 	hammer_uuid_t uid;
900 	hammer_uuid_t gid;
901 
902 	uint8_t obj_type;
903 	uint8_t cap_flags;	/* capability support flags (extension) */
904 	uint16_t reserved01;
905 	uint32_t reserved02;
906 	uint64_t nlinks;	/* hard links */
907 	uint64_t size;		/* filesystem object size */
908 	union {
909 		char	symlink[24];	/* HAMMER_INODE_BASESYMLEN */
910 	} ext;
911 	uint64_t mtime;	/* mtime must be second-to-last */
912 	uint64_t atime;	/* atime must be last */
913 } *hammer_inode_data_t;
914 
915 /*
916  * Neither mtime nor atime updates are CRCd by the B-Tree element.
917  * mtime updates have UNDO, atime updates do not.
918  */
919 #define HAMMER_INODE_CRCSIZE	\
920 	offsetof(struct hammer_inode_data, mtime)
921 
922 #define HAMMER_INODE_DATA_VERSION	1
923 #define HAMMER_OBJID_ROOT		1	/* root inodes # */
924 #define HAMMER_INODE_BASESYMLEN		24	/* see ext.symlink */
925 
926 /*
927  * Capability & implementation flags.
928  *
929  * HAMMER_INODE_CAP_DIR_LOCAL_INO - Use inode B-Tree localization
930  * for directory entries.  Also see HAMMER_DIR_INODE_LOCALIZATION().
931  */
932 #define HAMMER_INODE_CAP_DIRHASH_MASK	0x03	/* directory: hash algorithm */
933 #define HAMMER_INODE_CAP_DIRHASH_ALG0	0x00
934 #define HAMMER_INODE_CAP_DIRHASH_ALG1	0x01
935 #define HAMMER_INODE_CAP_DIRHASH_ALG2	0x02
936 #define HAMMER_INODE_CAP_DIRHASH_ALG3	0x03
937 #define HAMMER_INODE_CAP_DIR_LOCAL_INO	0x04	/* use inode localization */
938 
939 #define HAMMER_DATA_DOALIGN(offset)				\
940 	(((offset) + 15) & ~15)
941 #define HAMMER_DATA_DOALIGN_WITH(type, offset)			\
942 	(((type)(offset) + 15) & (~(type)15))
943 
944 /*
945  * A HAMMER directory entry associates a HAMMER filesystem object with a
946  * namespace.  It is hooked into a pseudo-filesystem (with its own inode
947  * numbering space) in the filesystem by setting the high 16 bits of the
948  * localization field.  The low 16 bits must be 0 and are reserved for
949  * future use.
950  *
951  * Directory entries are indexed with a 128 bit namekey rather than an
952  * offset.  A portion of the namekey is an iterator/randomizer to deal
953  * with collisions.
954  *
955  * NOTE: leaf.base.obj_type from the related B-Tree leaf entry holds
956  * the filesystem object type of obj_id, e.g. a den_type equivalent.
957  * It is not stored in hammer_direntry_data.
958  *
959  * NOTE: name field / the filename data reference is NOT terminated with \0.
960  */
961 typedef struct hammer_direntry_data {
962 	int64_t obj_id;			/* object being referenced */
963 	uint32_t localization;		/* identify pseudo-filesystem */
964 	uint32_t reserved01;
965 	char	name[16];		/* name (extended) */
966 } *hammer_direntry_data_t;
967 
968 #define HAMMER_ENTRY_NAME_OFF	offsetof(struct hammer_direntry_data, name[0])
969 #define HAMMER_ENTRY_SIZE(nlen)	offsetof(struct hammer_direntry_data, name[nlen])
970 
971 /*
972  * Symlink data which does not fit in the inode is stored in a separate
973  * FIX type record.
974  */
975 typedef struct hammer_symlink_data {
976 	char	name[16];		/* name (extended) */
977 } *hammer_symlink_data_t;
978 
979 #define HAMMER_SYMLINK_NAME_OFF	offsetof(struct hammer_symlink_data, name[0])
980 
981 /*
982  * The root inode for the primary filesystem and root inode for any
983  * pseudo-fs may be tagged with an optional data structure using
984  * HAMMER_RECTYPE_PFS and localization id.  This structure allows
985  * the node to be used as a mirroring master or slave.
986  *
987  * When operating as a slave CD's into the node automatically become read-only
988  * and as-of sync_end_tid.
989  *
990  * When operating as a master the read PFSD info sets sync_end_tid to
991  * the most recently flushed TID.
992  *
993  * sync_low_tid is not yet used but will represent the highest pruning
994  * end-point, after which full history is available.
995  *
996  * We need to pack this structure making it equally sized on both 32-bit and
997  * 64-bit machines as it is part of struct hammer_ioc_mrecord_pfs which is
998  * send over the wire in hammer mirror operations. Only on 64-bit machines
999  * the size of this struct differ when packed or not. This leads us to the
1000  * situation where old 64-bit systems (using the non-packed structure),
1001  * which were never able to mirror to/from 32-bit systems, are now no longer
1002  * able to mirror to/from newer 64-bit systems (using the packed structure).
1003  */
1004 struct hammer_pseudofs_data {
1005 	hammer_tid_t	sync_low_tid;	/* full history beyond this point */
1006 	hammer_tid_t	sync_beg_tid;	/* earliest tid w/ full history avail */
1007 	hammer_tid_t	sync_end_tid;	/* current synchronization point */
1008 	uint64_t	sync_beg_ts;	/* real-time of last completed sync */
1009 	uint64_t	sync_end_ts;	/* initiation of current sync cycle */
1010 	hammer_uuid_t	shared_uuid;	/* shared uuid (match required) */
1011 	hammer_uuid_t	unique_uuid;	/* unique uuid of this master/slave */
1012 	int32_t		reserved01;	/* reserved for future master_id */
1013 	int32_t		mirror_flags;	/* misc flags */
1014 	char		label[64];	/* filesystem space label */
1015 	char		snapshots[64];	/* softlink dir for pruning */
1016 	int32_t		reserved02;	/* was prune_{time,freq} */
1017 	int32_t		reserved03;	/* was reblock_{time,freq} */
1018 	int32_t		reserved04;	/* was snapshot_freq */
1019 	int32_t		prune_min;	/* do not prune recent history */
1020 	int32_t		prune_max;	/* do not retain history beyond here */
1021 	int32_t		reserved[16];
1022 } __packed;
1023 
1024 typedef struct hammer_pseudofs_data *hammer_pseudofs_data_t;
1025 
1026 #define HAMMER_PFSD_SLAVE	0x00000001
1027 #define HAMMER_PFSD_DELETED	0x80000000
1028 
1029 #define hammer_is_pfs_slave(pfsd)			\
1030 	(((pfsd)->mirror_flags & HAMMER_PFSD_SLAVE) != 0)
1031 #define hammer_is_pfs_master(pfsd)			\
1032 	(!hammer_is_pfs_slave(pfsd))
1033 #define hammer_is_pfs_deleted(pfsd)			\
1034 	(((pfsd)->mirror_flags & HAMMER_PFSD_DELETED) != 0)
1035 
1036 #define HAMMER_MAX_PFS		65536
1037 #define HAMMER_MAX_PFSID	(HAMMER_MAX_PFS - 1)
1038 #define HAMMER_ROOT_PFSID	0
1039 
1040 /*
1041  * Snapshot meta-data { Objid = HAMMER_OBJID_ROOT, Key = tid, rectype = SNAPSHOT }.
1042  *
1043  * Snapshot records replace the old <fs>/snapshots/<softlink> methodology.  Snapshot
1044  * records are mirrored but may be independently managed once they are laid down on
1045  * a slave.
1046  *
1047  * NOTE: The b-tree key is signed, the tid is not, so callers must still sort the
1048  *	 results.
1049  *
1050  * NOTE: Reserved fields must be zero (as usual)
1051  */
1052 typedef struct hammer_snapshot_data {
1053 	hammer_tid_t	tid;		/* the snapshot TID itself (== key) */
1054 	uint64_t	ts;		/* real-time when snapshot was made */
1055 	uint64_t	reserved01;
1056 	uint64_t	reserved02;
1057 	char		label[64];	/* user-supplied description */
1058 	uint64_t	reserved03[4];
1059 } *hammer_snapshot_data_t;
1060 
1061 /*
1062  * Config meta-data { ObjId = HAMMER_OBJID_ROOT, Key = 0, rectype = CONFIG }.
1063  *
1064  * Used to store the hammer cleanup config.  This data is not mirrored.
1065  */
1066 typedef struct hammer_config_data {
1067 	char		text[1024];
1068 } *hammer_config_data_t;
1069 
1070 /*
1071  * Rollup various structures embedded as record data
1072  */
1073 typedef union hammer_data_ondisk {
1074 	struct hammer_direntry_data entry;
1075 	struct hammer_inode_data inode;
1076 	struct hammer_symlink_data symlink;
1077 	struct hammer_pseudofs_data pfsd;
1078 	struct hammer_snapshot_data snap;
1079 	struct hammer_config_data config;
1080 } *hammer_data_ondisk_t;
1081 
1082 /*
1083  * Ondisk layout of B-Tree related structures
1084  */
1085 //#include "hammer_btree.h"
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