xref: /dflybsd-src/sys/vfs/hammer2/hammer2_disk.h (revision e19e5bbc20dd1d64f1833c5d0ac7a605c8e9bfa0)
1 /*
2  * Copyright (c) 2011-2014 The DragonFly Project.  All rights reserved.
3  *
4  * This code is derived from software contributed to The DragonFly Project
5  * by Matthew Dillon <dillon@dragonflybsd.org>
6  * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  *
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in
16  *    the documentation and/or other materials provided with the
17  *    distribution.
18  * 3. Neither the name of The DragonFly Project nor the names of its
19  *    contributors may be used to endorse or promote products derived
20  *    from this software without specific, prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
23  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
24  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
25  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
26  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
27  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
28  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
29  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
30  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
31  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
32  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33  * SUCH DAMAGE.
34  */
35 
36 #ifndef _VFS_HAMMER2_DISK_H_
37 #define _VFS_HAMMER2_DISK_H_
38 
39 #ifndef _SYS_UUID_H_
40 #include <sys/uuid.h>
41 #endif
42 #ifndef _SYS_DMSG_H_
43 #include <sys/dmsg.h>
44 #endif
45 
46 /*
47  * The structures below represent the on-disk media structures for the HAMMER2
48  * filesystem.  Note that all fields for on-disk structures are naturally
49  * aligned.  The host endian format is typically used - compatibility is
50  * possible if the implementation detects reversed endian and adjusts accesses
51  * accordingly.
52  *
53  * HAMMER2 primarily revolves around the directory topology:  inodes,
54  * directory entries, and block tables.  Block device buffer cache buffers
55  * are always 64KB.  Logical file buffers are typically 16KB.  All data
56  * references utilize 64-bit byte offsets.
57  *
58  * Free block management is handled independently using blocks reserved by
59  * the media topology.
60  */
61 
62 /*
63  * The data at the end of a file or directory may be a fragment in order
64  * to optimize storage efficiency.  The minimum fragment size is 1KB.
65  * Since allocations are in powers of 2 fragments must also be sized in
66  * powers of 2 (1024, 2048, ... 65536).
67  *
68  * For the moment the maximum allocation size is HAMMER2_PBUFSIZE (64K),
69  * which is 2^16.  Larger extents may be supported in the future.  Smaller
70  * fragments might be supported in the future (down to 64 bytes is possible),
71  * but probably will not be.
72  *
73  * A full indirect block use supports 1024 x 64-byte blockrefs in a 64KB
74  * buffer.  Indirect blocks down to 1KB are supported to keep small
75  * directories small.
76  *
77  * A maximally sized file (2^64-1 bytes) requires 5 indirect block levels.
78  * The hammer2_blockset in the volume header or file inode has another 8
79  * entries, giving us 66+3 = 69 bits of address space.  However, some bits
80  * are taken up by (potentially) requests for redundant copies.  HAMMER2
81  * currently supports up to 8 copies, which brings the address space down
82  * to 66 bits and gives us 2 bits of leeway.
83  */
84 #define HAMMER2_ALLOC_MIN	1024	/* minimum allocation size */
85 #define HAMMER2_RADIX_MIN	10	/* minimum allocation size 2^N */
86 #define HAMMER2_ALLOC_MAX	65536	/* maximum allocation size */
87 #define HAMMER2_RADIX_MAX	16	/* maximum allocation size 2^N */
88 #define HAMMER2_RADIX_KEY	64	/* number of bits in key */
89 
90 /*
91  * MINALLOCSIZE		- The minimum allocation size.  This can be smaller
92  *		  	  or larger than the minimum physical IO size.
93  *
94  *			  NOTE: Should not be larger than 1K since inodes
95  *				are 1K.
96  *
97  * MINIOSIZE		- The minimum IO size.  This must be less than
98  *			  or equal to HAMMER2_LBUFSIZE.
99  *
100  * HAMMER2_LBUFSIZE	- Nominal buffer size for I/O rollups.
101  *
102  * HAMMER2_PBUFSIZE	- Topological block size used by files for all
103  *			  blocks except the block straddling EOF.
104  *
105  * HAMMER2_SEGSIZE	- Allocation map segment size, typically 2MB
106  *			  (space represented by a level0 bitmap).
107  */
108 
109 #define HAMMER2_SEGSIZE		(1 << HAMMER2_FREEMAP_LEVEL0_RADIX)
110 #define HAMMER2_SEGRADIX	HAMMER2_FREEMAP_LEVEL0_RADIX
111 
112 #define HAMMER2_PBUFRADIX	16	/* physical buf (1<<16) bytes */
113 #define HAMMER2_PBUFSIZE	65536
114 #define HAMMER2_LBUFRADIX	14	/* logical buf (1<<14) bytes */
115 #define HAMMER2_LBUFSIZE	16384
116 
117 /*
118  * Generally speaking we want to use 16K and 64K I/Os
119  */
120 #define HAMMER2_MINIORADIX	HAMMER2_LBUFRADIX
121 #define HAMMER2_MINIOSIZE	HAMMER2_LBUFSIZE
122 
123 #define HAMMER2_IND_BYTES_MIN	HAMMER2_LBUFSIZE
124 #define HAMMER2_IND_BYTES_MAX	HAMMER2_PBUFSIZE
125 #define HAMMER2_IND_COUNT_MIN	(HAMMER2_IND_BYTES_MIN / \
126 				 sizeof(hammer2_blockref_t))
127 #define HAMMER2_IND_COUNT_MAX	(HAMMER2_IND_BYTES_MAX / \
128 				 sizeof(hammer2_blockref_t))
129 
130 /*
131  * In HAMMER2, arrays of blockrefs are fully set-associative, meaning that
132  * any element can occur at any index and holes can be anywhere.  As a
133  * future optimization we will be able to flag that such arrays are sorted
134  * and thus optimize lookups, but for now we don't.
135  *
136  * Inodes embed either 512 bytes of direct data or an array of 8 blockrefs,
137  * resulting in highly efficient storage for files <= 512 bytes and for files
138  * <= 512KB.  Up to 8 directory entries can be referenced from a directory
139  * without requiring an indirect block.
140  *
141  * Indirect blocks are typically either 4KB (64 blockrefs / ~4MB represented),
142  * or 64KB (1024 blockrefs / ~64MB represented).
143  */
144 #define HAMMER2_SET_COUNT		8	/* direct entries */
145 #define HAMMER2_SET_RADIX		3
146 #define HAMMER2_EMBEDDED_BYTES		512	/* inode blockset/dd size */
147 #define HAMMER2_EMBEDDED_RADIX		9
148 
149 #define HAMMER2_PBUFMASK	(HAMMER2_PBUFSIZE - 1)
150 #define HAMMER2_LBUFMASK	(HAMMER2_LBUFSIZE - 1)
151 #define HAMMER2_SEGMASK		(HAMMER2_SEGSIZE - 1)
152 
153 #define HAMMER2_LBUFMASK64	((hammer2_off_t)HAMMER2_LBUFMASK)
154 #define HAMMER2_PBUFSIZE64	((hammer2_off_t)HAMMER2_PBUFSIZE)
155 #define HAMMER2_PBUFMASK64	((hammer2_off_t)HAMMER2_PBUFMASK)
156 #define HAMMER2_SEGSIZE64	((hammer2_off_t)HAMMER2_SEGSIZE)
157 #define HAMMER2_SEGMASK64	((hammer2_off_t)HAMMER2_SEGMASK)
158 
159 #define HAMMER2_UUID_STRING	"5cbb9ad1-862d-11dc-a94d-01301bb8a9f5"
160 
161 /*
162  * A HAMMER2 filesystem is always sized in multiples of 8MB.
163  *
164  * A 4MB segment is reserved at the beginning of each 2GB zone.  This segment
165  * contains the volume header (or backup volume header), the free block
166  * table, and possibly other information in the future.
167  *
168  * 4MB = 64 x 64K blocks.  Each 4MB segment is broken down as follows:
169  *
170  *	+-----------------------+
171  *      |	Volume Hdr	| block 0	volume header & alternates
172  *	+-----------------------+		(first four zones only)
173  *	|   FreeBlk Section A   | block 1-4
174  *	+-----------------------+
175  *	|   FreeBlk Section B   | block 5-8
176  *	+-----------------------+
177  *	|   FreeBlk Section C   | block 9-12
178  *	+-----------------------+
179  *	|   FreeBlk Section D   | block 13-16
180  *	+-----------------------+
181  *      |			| block 17...63
182  *      |	reserved	|
183  *      |			|
184  *	+-----------------------+
185  *
186  * The first few 2GB zones contain volume headers and volume header backups.
187  * After that the volume header block# is reserved.
188  *
189  *			Freemap (see the FREEMAP document)
190  *
191  * The freemap utilizes blocks #1-16 for now, see the FREEMAP document.
192  * The filesystems rotations through the sections to avoid disturbing the
193  * 'previous' version of the freemap during a flush.
194  *
195  * Each freemap section is 4 x 64K blocks and represents 2GB, 2TB, 2PB,
196  * and 2EB indirect map, plus the volume header has a set of 8 blockrefs
197  * for another 3 bits for a total of 64 bits of address space.  The Level 0
198  * 64KB block representing 2GB of storage is a hammer2_bmap_data[1024].
199  * Each element contains a 128x2 bit bitmap representing 16KB per chunk for
200  * 2MB of storage (x1024 elements = 2GB).  2 bits per chunk:
201  *
202  *	00	Free
203  *	01	(reserved)
204  *	10	Possibly free
205  *	11	Allocated
206  *
207  * One important thing to note here is that the freemap resolution is 16KB,
208  * but the minimuim storage allocation size is 1KB.  The hammer2 vfs keeps
209  * track of sub-allocations in memory (on umount or reboot obvious the whole
210  * 16KB will be considered allocated even if only 1KB is allocated).  It is
211  * possible for fragmentation to build up over time.
212  *
213  * The Second thing to note is that due to the way snapshots and inode
214  * replication works, deleting a file cannot immediately free the related
215  * space.  Instead, the freemap elements transition from 11->10.  The bulk
216  * freeing code which does a complete scan is then responsible for
217  * transitioning the elements to 00 or back to 11 or to 01 for that matter.
218  *
219  * WARNING!  ZONE_SEG and VOLUME_ALIGN must be a multiple of 1<<LEVEL0_RADIX
220  *	     (i.e. a multiple of 2MB).  VOLUME_ALIGN must be >= ZONE_SEG.
221  */
222 #define HAMMER2_VOLUME_ALIGN		(8 * 1024 * 1024)
223 #define HAMMER2_VOLUME_ALIGN64		((hammer2_off_t)HAMMER2_VOLUME_ALIGN)
224 #define HAMMER2_VOLUME_ALIGNMASK	(HAMMER2_VOLUME_ALIGN - 1)
225 #define HAMMER2_VOLUME_ALIGNMASK64     ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK)
226 
227 #define HAMMER2_NEWFS_ALIGN		(HAMMER2_VOLUME_ALIGN)
228 #define HAMMER2_NEWFS_ALIGN64		((hammer2_off_t)HAMMER2_VOLUME_ALIGN)
229 #define HAMMER2_NEWFS_ALIGNMASK		(HAMMER2_VOLUME_ALIGN - 1)
230 #define HAMMER2_NEWFS_ALIGNMASK64	((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK)
231 
232 #define HAMMER2_ZONE_BYTES64		(2LLU * 1024 * 1024 * 1024)
233 #define HAMMER2_ZONE_MASK64		(HAMMER2_ZONE_BYTES64 - 1)
234 #define HAMMER2_ZONE_SEG		(4 * 1024 * 1024)
235 #define HAMMER2_ZONE_SEG64		((hammer2_off_t)HAMMER2_ZONE_SEG)
236 #define HAMMER2_ZONE_BLOCKS_SEG		(HAMMER2_ZONE_SEG / HAMMER2_PBUFSIZE)
237 
238 /*
239  * 64 x 64KB blocks are reserved at the base of each 2GB zone.  These blocks
240  * are used to store the volume header or volume header backups, allocation
241  * tree, and other information in the future.
242  *
243  * All specified blocks are not necessarily used in all 2GB zones.  However,
244  * dead areas are reserved for future use and MUST NOT BE USED for other
245  * purposes.
246  *
247  * The freemap is arranged into 15 groups of 4x64KB each.  The 4 sub-groups
248  * are labeled ZONEFM1..4 and representing HAMMER2_FREEMAP_LEVEL{1-4}_RADIX,
249  * for the up to 4 levels of radix tree representing the freemap.  For
250  * simplicity we are reserving all four radix tree layers even though the
251  * higher layers do not require teh reservation at each 2GB mark.  That
252  * space is reserved for future use.
253  *
254  * Freemap blocks are not allocated dynamically but instead rotate through
255  * one of 15 possible copies.  We require 15 copies for several reasons:
256  *
257  * (1) For distinguishing freemap 'allocations' made by the current flush
258  *     verses the concurrently running front-end (at flush_tid + 1).  This
259  *     theoretically requires two copies but the algorithm is greatly
260  *     simplified if we use three.
261  *
262  * (2) There are up to 4 copies of the volume header (iterated on each flush),
263  *     and if the mount code is forced to use an older copy due to corruption
264  *     we must be sure that the state of the freemap AS-OF the earlier copy
265  *     remains valid.
266  *
267  *     This means 3 copies x 4 flushes = 12 copies to be able to mount any
268  *     of the four volume header backups after on boot or after a crash.
269  *
270  * (3) Freemap recovery on-mount eats a copy.  We don't want freemap recovery
271  *     to blow away the copy used by some other volume header in case H2
272  *     crashes during the recovery.  Total is now 13.
273  *
274  * (4) And I want some breathing room to ensure that complex flushes do not
275  *     cause problems.  Also note that bulk block freeing itself must be
276  *     careful so even on a live system, post-mount, the four volume header
277  *     backups effectively represent short-lived snapshots.  And I only
278  *     have room for 15 copies so it works out.
279  *
280  * Preferably I would like to improve the algorithm to only use 2 copies per
281  * volume header (which would be a total of 2 x 4 = 8 + 1 for freemap recovery
282  * + 1 for breathing room = 10 total instead of 15).  For now we use 15.
283  */
284 #define HAMMER2_ZONE_VOLHDR		0	/* volume header or backup */
285 #define HAMMER2_ZONE_FREEMAP_00		1
286 #define HAMMER2_ZONE_FREEMAP_01		5
287 #define HAMMER2_ZONE_FREEMAP_02		9
288 #define HAMMER2_ZONE_FREEMAP_03		13
289 #define HAMMER2_ZONE_FREEMAP_04		17
290 #define HAMMER2_ZONE_FREEMAP_05		21
291 #define HAMMER2_ZONE_FREEMAP_06		25
292 #define HAMMER2_ZONE_FREEMAP_07		29
293 #define HAMMER2_ZONE_FREEMAP_08		33
294 #define HAMMER2_ZONE_FREEMAP_09		37
295 #define HAMMER2_ZONE_FREEMAP_10		41
296 #define HAMMER2_ZONE_FREEMAP_11		45
297 #define HAMMER2_ZONE_FREEMAP_12		49
298 #define HAMMER2_ZONE_FREEMAP_13		53
299 #define HAMMER2_ZONE_FREEMAP_14		57
300 #define HAMMER2_ZONE_FREEMAP_END	61	/* (non-inclusive) */
301 #define HAMMER2_ZONE_UNUSED62		62
302 #define HAMMER2_ZONE_UNUSED63		63
303 
304 #define HAMMER2_ZONE_FREEMAP_COPIES	15
305 						/* relative to FREEMAP_x */
306 #define HAMMER2_ZONEFM_LEVEL1		0	/* 2GB leafmap */
307 #define HAMMER2_ZONEFM_LEVEL2		1	/* 2TB indmap */
308 #define HAMMER2_ZONEFM_LEVEL3		2	/* 2PB indmap */
309 #define HAMMER2_ZONEFM_LEVEL4		3	/* 2EB indmap */
310 /* LEVEL5 is a set of 8 blockrefs in the volume header 16EB */
311 
312 
313 /*
314  * Freemap radii.  Please note that LEVEL 1 blockref array entries
315  * point to 256-byte sections of the bitmap representing 2MB of storage.
316  * Even though the chain structures represent only 256 bytes, they are
317  * mapped using larger 16K or 64K buffer cache buffers.
318  */
319 #define HAMMER2_FREEMAP_LEVEL5_RADIX	64	/* 16EB */
320 #define HAMMER2_FREEMAP_LEVEL4_RADIX	61	/* 2EB */
321 #define HAMMER2_FREEMAP_LEVEL3_RADIX	51	/* 2PB */
322 #define HAMMER2_FREEMAP_LEVEL2_RADIX	41	/* 2TB */
323 #define HAMMER2_FREEMAP_LEVEL1_RADIX	31	/* 2GB */
324 #define HAMMER2_FREEMAP_LEVEL0_RADIX	21	/* 2MB (entry in l-1 leaf) */
325 
326 #define HAMMER2_FREEMAP_LEVELN_PSIZE	65536	/* physical bytes */
327 
328 #define HAMMER2_FREEMAP_COUNT		(int)(HAMMER2_FREEMAP_LEVELN_PSIZE / \
329 					 sizeof(hammer2_bmap_data_t))
330 #define HAMMER2_FREEMAP_BLOCK_RADIX	14
331 #define HAMMER2_FREEMAP_BLOCK_SIZE	(1 << HAMMER2_FREEMAP_BLOCK_RADIX)
332 #define HAMMER2_FREEMAP_BLOCK_MASK	(HAMMER2_FREEMAP_BLOCK_SIZE - 1)
333 
334 /*
335  * Two linear areas can be reserved after the initial 2MB segment in the base
336  * zone (the one starting at offset 0).  These areas are NOT managed by the
337  * block allocator and do not fall under HAMMER2 crc checking rules based
338  * at the volume header (but can be self-CRCd internally, depending).
339  */
340 #define HAMMER2_BOOT_MIN_BYTES		HAMMER2_VOLUME_ALIGN
341 #define HAMMER2_BOOT_NOM_BYTES		(64*1024*1024)
342 #define HAMMER2_BOOT_MAX_BYTES		(256*1024*1024)
343 
344 #define HAMMER2_REDO_MIN_BYTES		HAMMER2_VOLUME_ALIGN
345 #define HAMMER2_REDO_NOM_BYTES		(256*1024*1024)
346 #define HAMMER2_REDO_MAX_BYTES		(1024*1024*1024)
347 
348 /*
349  * Most HAMMER2 types are implemented as unsigned 64-bit integers.
350  * Transaction ids are monotonic.
351  *
352  * We utilize 32-bit iSCSI CRCs.
353  */
354 typedef uint64_t hammer2_tid_t;
355 typedef uint64_t hammer2_off_t;
356 typedef uint64_t hammer2_key_t;
357 typedef uint32_t hammer2_crc32_t;
358 
359 /*
360  * Miscellanious ranges (all are unsigned).
361  */
362 #define HAMMER2_TID_MIN		1ULL
363 #define HAMMER2_TID_MAX		0xFFFFFFFFFFFFFFFFULL
364 #define HAMMER2_KEY_MIN		0ULL
365 #define HAMMER2_KEY_MAX		0xFFFFFFFFFFFFFFFFULL
366 #define HAMMER2_OFFSET_MIN	0ULL
367 #define HAMMER2_OFFSET_MAX	0xFFFFFFFFFFFFFFFFULL
368 
369 /*
370  * HAMMER2 data offset special cases and masking.
371  *
372  * All HAMMER2 data offsets have to be broken down into a 64K buffer base
373  * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO).
374  *
375  * Indexes into physical buffers are always 64-byte aligned.  The low 6 bits
376  * of the data offset field specifies how large the data chunk being pointed
377  * to as a power of 2.  The theoretical minimum radix is thus 6 (The space
378  * needed in the low bits of the data offset field).  However, the practical
379  * minimum allocation chunk size is 1KB (a radix of 10), so HAMMER2 sets
380  * HAMMER2_RADIX_MIN to 10.  The maximum radix is currently 16 (64KB), but
381  * we fully intend to support larger extents in the future.
382  */
383 #define HAMMER2_OFF_BAD		((hammer2_off_t)-1)
384 #define HAMMER2_OFF_MASK	0xFFFFFFFFFFFFFFC0ULL
385 #define HAMMER2_OFF_MASK_LO	(HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64)
386 #define HAMMER2_OFF_MASK_HI	(~HAMMER2_PBUFMASK64)
387 #define HAMMER2_OFF_MASK_RADIX	0x000000000000003FULL
388 #define HAMMER2_MAX_COPIES	6
389 
390 /*
391  * HAMMER2 directory support and pre-defined keys
392  */
393 #define HAMMER2_DIRHASH_VISIBLE	0x8000000000000000ULL
394 #define HAMMER2_DIRHASH_USERMSK	0x7FFFFFFFFFFFFFFFULL
395 #define HAMMER2_DIRHASH_LOMASK	0x0000000000007FFFULL
396 #define HAMMER2_DIRHASH_HIMASK	0xFFFFFFFFFFFF0000ULL
397 #define HAMMER2_DIRHASH_FORCED	0x0000000000008000ULL	/* bit forced on */
398 
399 #define HAMMER2_SROOT_KEY	0x0000000000000000ULL	/* volume to sroot */
400 
401 /************************************************************************
402  *				DMSG SUPPORT				*
403  ************************************************************************
404  * LNK_VOLCONF
405  *
406  * All HAMMER2 directories directly under the super-root on your local
407  * media can be mounted separately, even if they share the same physical
408  * device.
409  *
410  * When you do a HAMMER2 mount you are effectively tying into a HAMMER2
411  * cluster via local media.  The local media does not have to participate
412  * in the cluster, other than to provide the hammer2_volconf[] array and
413  * root inode for the mount.
414  *
415  * This is important: The mount device path you specify serves to bootstrap
416  * your entry into the cluster, but your mount will make active connections
417  * to ALL copy elements in the hammer2_volconf[] array which match the
418  * PFSID of the directory in the super-root that you specified.  The local
419  * media path does not have to be mentioned in this array but becomes part
420  * of the cluster based on its type and access rights.  ALL ELEMENTS ARE
421  * TREATED ACCORDING TO TYPE NO MATTER WHICH ONE YOU MOUNT FROM.
422  *
423  * The actual cluster may be far larger than the elements you list in the
424  * hammer2_volconf[] array.  You list only the elements you wish to
425  * directly connect to and you are able to access the rest of the cluster
426  * indirectly through those connections.
427  *
428  * WARNING!  This structure must be exactly 128 bytes long for its config
429  *	     array to fit in the volume header.
430  */
431 struct hammer2_volconf {
432 	uint8_t	copyid;		/* 00	 copyid 0-255 (must match slot) */
433 	uint8_t inprog;		/* 01	 operation in progress, or 0 */
434 	uint8_t chain_to;	/* 02	 operation chaining to, or 0 */
435 	uint8_t chain_from;	/* 03	 operation chaining from, or 0 */
436 	uint16_t flags;		/* 04-05 flags field */
437 	uint8_t error;		/* 06	 last operational error */
438 	uint8_t priority;	/* 07	 priority and round-robin flag */
439 	uint8_t remote_pfs_type;/* 08	 probed direct remote PFS type */
440 	uint8_t reserved08[23];	/* 09-1F */
441 	uuid_t	pfs_clid;	/* 20-2F copy target must match this uuid */
442 	uint8_t label[16];	/* 30-3F import/export label */
443 	uint8_t path[64];	/* 40-7F target specification string or key */
444 };
445 
446 typedef struct hammer2_volconf hammer2_volconf_t;
447 
448 #define DMSG_VOLF_ENABLED	0x0001
449 #define DMSG_VOLF_INPROG	0x0002
450 #define DMSG_VOLF_CONN_RR	0x80	/* round-robin at same priority */
451 #define DMSG_VOLF_CONN_EF	0x40	/* media errors flagged */
452 #define DMSG_VOLF_CONN_PRI	0x0F	/* select priority 0-15 (15=best) */
453 
454 struct dmsg_lnk_hammer2_volconf {
455 	dmsg_hdr_t		head;
456 	hammer2_volconf_t	copy;	/* copy spec */
457 	int32_t			index;
458 	int32_t			unused01;
459 	uuid_t			mediaid;
460 	int64_t			reserved02[32];
461 };
462 
463 typedef struct dmsg_lnk_hammer2_volconf dmsg_lnk_hammer2_volconf_t;
464 
465 #define DMSG_LNK_HAMMER2_VOLCONF DMSG_LNK(DMSG_LNK_CMD_HAMMER2_VOLCONF, \
466 					  dmsg_lnk_hammer2_volconf)
467 
468 #define H2_LNK_VOLCONF(msg)	((dmsg_lnk_hammer2_volconf_t *)(msg)->any.buf)
469 
470 /*
471  * The media block reference structure.  This forms the core of the HAMMER2
472  * media topology recursion.  This 64-byte data structure is embedded in the
473  * volume header, in inodes (which are also directory entries), and in
474  * indirect blocks.
475  *
476  * A blockref references a single media item, which typically can be a
477  * directory entry (aka inode), indirect block, or data block.
478  *
479  * The primary feature a blockref represents is the ability to validate
480  * the entire tree underneath it via its check code.  Any modification to
481  * anything propagates up the blockref tree all the way to the root, replacing
482  * the related blocks.  Propagations can shortcut to the volume root to
483  * implement the 'fast syncing' feature but this only delays the eventual
484  * propagation.
485  *
486  * The check code can be a simple 32-bit iscsi code, a 64-bit crc,
487  * or as complex as a 192 bit cryptographic hash.  192 bits is the maximum
488  * supported check code size, which is not sufficient for unverified dedup
489  * UNLESS one doesn't mind once-in-a-blue-moon data corruption (such as when
490  * farming web data).  HAMMER2 has an unverified dedup feature for just this
491  * purpose.
492  *
493  * --
494  *
495  * NOTE: The range of keys represented by the blockref is (key) to
496  *	 ((key) + (1LL << keybits) - 1).  HAMMER2 usually populates
497  *	 blocks bottom-up, inserting a new root when radix expansion
498  *	 is required.
499  */
500 struct hammer2_blockref {		/* MUST BE EXACTLY 64 BYTES */
501 	uint8_t		type;		/* type of underlying item */
502 	uint8_t		methods;	/* check method & compression method */
503 	uint8_t		copyid;		/* specify which copy this is */
504 	uint8_t		keybits;	/* #of keybits masked off 0=leaf */
505 	uint8_t		vradix;		/* virtual data/meta-data size */
506 	uint8_t		flags;		/* blockref flags */
507 	uint8_t		reserved06;
508 	uint8_t		reserved07;
509 	hammer2_key_t	key;		/* key specification */
510 	hammer2_tid_t	mirror_tid;	/* propagate for mirror scan */
511 	hammer2_tid_t	modify_tid;	/* modifications sans propagation */
512 	hammer2_off_t	data_off;	/* low 6 bits is phys size (radix)*/
513 	union {				/* check info */
514 		char	buf[24];
515 		struct {
516 			uint32_t value;
517 			uint32_t unused[5];
518 		} iscsi32;
519 		struct {
520 			uint64_t value;
521 			uint64_t unused[2];
522 		} crc64;
523 		struct {
524 			char data[24];
525 		} sha192;
526 
527 		/*
528 		 * Freemap hints are embedded in addition to the icrc32.
529 		 *
530 		 * bigmask - Radixes available for allocation (0-31).
531 		 *	     Heuristical (may be permissive but not
532 		 *	     restrictive).  Typically only radix values
533 		 *	     10-16 are used (i.e. (1<<10) through (1<<16)).
534 		 *
535 		 * avail   - Total available space remaining, in bytes
536 		 */
537 		struct {
538 			uint32_t icrc32;
539 			uint32_t bigmask;	/* available radixes */
540 			uint64_t avail;		/* total available bytes */
541 			uint64_t unused;	/* unused must be 0 */
542 		} freemap;
543 
544 		/*
545 		 * Debugging
546 		 */
547 		struct {
548 			hammer2_tid_t sync_tid;
549 		} debug;
550 	} check;
551 };
552 
553 typedef struct hammer2_blockref hammer2_blockref_t;
554 
555 #define HAMMER2_BLOCKREF_BYTES		64	/* blockref struct in bytes */
556 
557 /*
558  * On-media and off-media blockref types.
559  */
560 #define HAMMER2_BREF_TYPE_EMPTY		0
561 #define HAMMER2_BREF_TYPE_INODE		1
562 #define HAMMER2_BREF_TYPE_INDIRECT	2
563 #define HAMMER2_BREF_TYPE_DATA		3
564 #define HAMMER2_BREF_TYPE_UNUSED04	4
565 #define HAMMER2_BREF_TYPE_FREEMAP_NODE	5
566 #define HAMMER2_BREF_TYPE_FREEMAP_LEAF	6
567 #define HAMMER2_BREF_TYPE_FREEMAP	254	/* pseudo-type */
568 #define HAMMER2_BREF_TYPE_VOLUME	255	/* pseudo-type */
569 
570 #define HAMMER2_BREF_FLAG_PFSROOT	0x01	/* see also related opflag */
571 
572 #define HAMMER2_ENC_CHECK(n)		((n) << 4)
573 #define HAMMER2_DEC_CHECK(n)		(((n) >> 4) & 15)
574 
575 #define HAMMER2_CHECK_NONE		0
576 #define HAMMER2_CHECK_ISCSI32		1
577 #define HAMMER2_CHECK_CRC64		2
578 #define HAMMER2_CHECK_SHA192		3
579 #define HAMMER2_CHECK_FREEMAP		4
580 
581 #define HAMMER2_ENC_COMP(n)		(n)
582 #define HAMMER2_ENC_LEVEL(n)		((n) << 4)
583 #define HAMMER2_DEC_COMP(n)		((n) & 15)
584 #define HAMMER2_DEC_LEVEL(n)		(((n) >> 4) & 15)
585 
586 #define HAMMER2_COMP_NONE		0
587 #define HAMMER2_COMP_AUTOZERO		1
588 #define HAMMER2_COMP_LZ4		2
589 #define HAMMER2_COMP_ZLIB		3
590 
591 #define HAMMER2_COMP_NEWFS_DEFAULT	HAMMER2_COMP_LZ4
592 #define HAMMER2_COMP_STRINGS		{ "none", "autozero", "lz4", "zlib" }
593 #define HAMMER2_COMP_STRINGS_COUNT	4
594 
595 
596 /*
597  * HAMMER2 block references are collected into sets of 8 blockrefs.  These
598  * sets are fully associative, meaning the elements making up a set are
599  * not sorted in any way and may contain duplicate entries, holes, or
600  * entries which shortcut multiple levels of indirection.  Sets are used
601  * in various ways:
602  *
603  * (1) When redundancy is desired a set may contain several duplicate
604  *     entries pointing to different copies of the same data.  Up to 8 copies
605  *     are supported but the set structure becomes a bit inefficient once
606  *     you go over 4.
607  *
608  * (2) The blockrefs in a set can shortcut multiple levels of indirections
609  *     within the bounds imposed by the parent of set.
610  *
611  * When a set fills up another level of indirection is inserted, moving
612  * some or all of the set's contents into indirect blocks placed under the
613  * set.  This is a top-down approach in that indirect blocks are not created
614  * until the set actually becomes full (that is, the entries in the set can
615  * shortcut the indirect blocks when the set is not full).  Depending on how
616  * things are filled multiple indirect blocks will eventually be created.
617  *
618  * Indirect blocks are typically 4KB (64 entres) or 64KB (1024 entries) and
619  * are also treated as fully set-associative.
620  */
621 struct hammer2_blockset {
622 	hammer2_blockref_t	blockref[HAMMER2_SET_COUNT];
623 };
624 
625 typedef struct hammer2_blockset hammer2_blockset_t;
626 
627 /*
628  * Catch programmer snafus
629  */
630 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT
631 #error "hammer2 direct radix is incorrect"
632 #endif
633 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE
634 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent"
635 #endif
636 #if (1 << HAMMER2_RADIX_MIN) != HAMMER2_ALLOC_MIN
637 #error "HAMMER2_RADIX_MIN and HAMMER2_ALLOC_MIN are inconsistent"
638 #endif
639 
640 /*
641  * hammer2_bmap_data - A freemap entry in the LEVEL1 block.
642  *
643  * Each 64-byte entry contains the bitmap and meta-data required to manage
644  * a LEVEL0 (2MB) block of storage.  The storage is managed in 128 x 16KB
645  * chunks.  Smaller allocation granularity is supported via a linear iterator
646  * and/or must otherwise be tracked in ram.
647  *
648  * (data structure must be 64 bytes exactly)
649  *
650  * linear  - A BYTE linear allocation offset used for sub-16KB allocations
651  *	     only.  May contain values between 0 and 2MB.  Must be ignored
652  *	     if 16KB-aligned (i.e. force bitmap scan), otherwise may be
653  *	     used to sub-allocate within the 16KB block (which is already
654  *	     marked as allocated in the bitmap).
655  *
656  *	     Sub-allocations need only be 1KB-aligned and do not have to be
657  *	     size-aligned, and 16KB or larger allocations do not update this
658  *	     field, resulting in pretty good packing.
659  *
660  *	     Please note that file data granularity may be limited by
661  *	     other issues such as buffer cache direct-mapping and the
662  *	     desire to support sector sizes up to 16KB (so H2 only issues
663  *	     I/O's in multiples of 16KB anyway).
664  *
665  * class   - Clustering class.  Cleared to 0 only if the entire leaf becomes
666  *	     free.  Used to cluster device buffers so all elements must have
667  *	     the same device block size, but may mix logical sizes.
668  *
669  *	     Typically integrated with the blockref type in the upper 8 bits
670  *	     to localize inodes and indrect blocks, improving bulk free scans
671  *	     and directory scans.
672  *
673  * bitmap  - Two bits per 16KB allocation block arranged in arrays of
674  *	     32-bit elements, 128x2 bits representing ~2MB worth of media
675  *	     storage.  Bit patterns are as follows:
676  *
677  *	     00	Unallocated
678  *	     01 (reserved)
679  *	     10 Possibly free
680  *           11 Allocated
681  */
682 struct hammer2_bmap_data {
683 	int32_t linear;		/* 00 linear sub-granular allocation offset */
684 	uint16_t class;		/* 04-05 clustering class ((type<<8)|radix) */
685 	uint8_t reserved06;	/* 06 */
686 	uint8_t reserved07;	/* 07 */
687 	uint32_t reserved08;	/* 08 */
688 	uint32_t reserved0C;	/* 0C */
689 	uint32_t reserved10;	/* 10 */
690 	uint32_t reserved14;	/* 14 */
691 	uint32_t reserved18;	/* 18 */
692 	uint32_t avail;		/* 1C */
693 	uint32_t bitmap[8];	/* 20-3F 256 bits manages 2MB/16KB/2-bits */
694 };
695 
696 typedef struct hammer2_bmap_data hammer2_bmap_data_t;
697 
698 /*
699  * In HAMMER2 inodes ARE directory entries, with a special exception for
700  * hardlinks.  The inode number is stored in the inode rather than being
701  * based on the location of the inode (since the location moves every time
702  * the inode or anything underneath the inode is modified).
703  *
704  * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes
705  * for the filename, and 512 bytes worth of direct file data OR an embedded
706  * blockset.
707  *
708  * Directories represent one inode per blockref.  Inodes are not laid out
709  * as a file but instead are represented by the related blockrefs.  The
710  * blockrefs, in turn, are indexed by the 64-bit directory hash key.  Remember
711  * that blocksets are fully associative, so a certain degree efficiency is
712  * achieved just from that.
713  *
714  * Up to 512 bytes of direct data can be embedded in an inode, and since
715  * inodes are essentially directory entries this also means that small data
716  * files end up simply being laid out linearly in the directory, resulting
717  * in fewer seeks and highly optimal access.
718  *
719  * The compression mode can be changed at any time in the inode and is
720  * recorded on a blockref-by-blockref basis.
721  *
722  * Hardlinks are supported via the inode map.  Essentially the way a hardlink
723  * works is that all individual directory entries representing the same file
724  * are special cased and specify the same inode number.  The actual file
725  * is placed in the nearest parent directory that is parent to all instances
726  * of the hardlink.  If all hardlinks to a file are in the same directory
727  * the actual file will also be placed in that directory.  This file uses
728  * the inode number as the directory entry key and is invisible to normal
729  * directory scans.  Real directory entry keys are differentiated from the
730  * inode number key via bit 63.  Access to the hardlink silently looks up
731  * the real file and forwards all operations to that file.  Removal of the
732  * last hardlink also removes the real file.
733  *
734  * (attr_tid) is only updated when the inode's specific attributes or regular
735  * file size has changed, and affects path lookups and stat.  (attr_tid)
736  * represents a special cache coherency lock under the inode.  The inode
737  * blockref's modify_tid will always cover it.
738  *
739  * (dirent_tid) is only updated when an entry under a directory inode has
740  * been created, deleted, renamed, or had its attributes change, and affects
741  * directory lookups and scans.  (dirent_tid) represents another special cache
742  * coherency lock under the inode.  The inode blockref's modify_tid will
743  * always cover it.
744  */
745 #define HAMMER2_INODE_BYTES		1024	/* (asserted by code) */
746 #define HAMMER2_INODE_MAXNAME		256	/* maximum name in bytes */
747 #define HAMMER2_INODE_VERSION_ONE	1
748 
749 #define HAMMER2_INODE_HIDDENDIR		16	/* special inode */
750 #define HAMMER2_INODE_START		1024	/* dynamically allocated */
751 
752 struct hammer2_inode_data {
753 	uint16_t	version;	/* 0000 inode data version */
754 	uint16_t	reserved02;	/* 0002 */
755 
756 	/*
757 	 * core inode attributes, inode type, misc flags
758 	 */
759 	uint32_t	uflags;		/* 0004 chflags */
760 	uint32_t	rmajor;		/* 0008 available for device nodes */
761 	uint32_t	rminor;		/* 000C available for device nodes */
762 	uint64_t	ctime;		/* 0010 inode change time */
763 	uint64_t	mtime;		/* 0018 modified time */
764 	uint64_t	atime;		/* 0020 access time (unsupported) */
765 	uint64_t	btime;		/* 0028 birth time */
766 	uuid_t		uid;		/* 0030 uid / degenerate unix uid */
767 	uuid_t		gid;		/* 0040 gid / degenerate unix gid */
768 
769 	uint8_t		type;		/* 0050 object type */
770 	uint8_t		op_flags;	/* 0051 operational flags */
771 	uint16_t	cap_flags;	/* 0052 capability flags */
772 	uint32_t	mode;		/* 0054 unix modes (typ low 16 bits) */
773 
774 	/*
775 	 * inode size, identification, localized recursive configuration
776 	 * for compression and backup copies.
777 	 */
778 	hammer2_tid_t	inum;		/* 0058 inode number */
779 	hammer2_off_t	size;		/* 0060 size of file */
780 	uint64_t	nlinks;		/* 0068 hard links (typ only dirs) */
781 	hammer2_tid_t	iparent;	/* 0070 parent inum (recovery only) */
782 	hammer2_key_t	name_key;	/* 0078 full filename key */
783 	uint16_t	name_len;	/* 0080 filename length */
784 	uint8_t		ncopies;	/* 0082 ncopies to local media */
785 	uint8_t		comp_algo;	/* 0083 compression request & algo */
786 
787 	/*
788 	 * These fields are currently only applicable to PFSROOTs.
789 	 *
790 	 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely
791 	 *	 identify an instance of a PFS in the cluster because
792 	 *	 a mount may contain more than one copy of the PFS as
793 	 *	 a separate node.  {pfs_clid, pfs_fsid} must be used for
794 	 *	 registration in the cluster.
795 	 */
796 	uint8_t		target_type;	/* 0084 hardlink target type */
797 	uint8_t		reserved85;	/* 0085 */
798 	uint8_t		pfs_nmasters;	/* 0086 (if PFSROOT) if multi-master */
799 	uint8_t		pfs_type;	/* 0087 (if PFSROOT) node type */
800 	uint64_t	pfs_inum;	/* 0088 (if PFSROOT) inum allocator */
801 	uuid_t		pfs_clid;	/* 0090 (if PFSROOT) cluster uuid */
802 	uuid_t		pfs_fsid;	/* 00A0 (if PFSROOT) unique uuid */
803 
804 	/*
805 	 * Quotas and cumulative sub-tree counters.
806 	 */
807 	hammer2_key_t	data_quota;	/* 00B0 subtree quota in bytes */
808 	hammer2_key_t	data_count;	/* 00B8 subtree byte count */
809 	hammer2_key_t	inode_quota;	/* 00C0 subtree quota inode count */
810 	hammer2_key_t	inode_count;	/* 00C8 subtree inode count */
811 	hammer2_tid_t	attr_tid;	/* 00D0 attributes changed */
812 	hammer2_tid_t	dirent_tid;	/* 00D8 directory/attr changed */
813 
814 	/*
815 	 * Tracks (possibly degenerate) free areas covering all sub-tree
816 	 * allocations under inode, not counting the inode itself.
817 	 * 0/0 indicates empty entry.  fully set-associative.
818 	 */
819 	hammer2_off_t	reservedE0[4];	/* 00E0/E8/F0/F8 */
820 
821 	unsigned char	filename[HAMMER2_INODE_MAXNAME];
822 					/* 0100-01FF (256 char, unterminated) */
823 	union {				/* 0200-03FF (64x8 = 512 bytes) */
824 		struct hammer2_blockset blockset;
825 		char data[HAMMER2_EMBEDDED_BYTES];
826 	} u;
827 };
828 
829 typedef struct hammer2_inode_data hammer2_inode_data_t;
830 
831 #define HAMMER2_OPFLAG_DIRECTDATA	0x01
832 #define HAMMER2_OPFLAG_PFSROOT		0x02	/* (see also bref flag) */
833 #define HAMMER2_OPFLAG_COPYIDS		0x04	/* copyids override parent */
834 
835 #define HAMMER2_OBJTYPE_UNKNOWN		0
836 #define HAMMER2_OBJTYPE_DIRECTORY	1
837 #define HAMMER2_OBJTYPE_REGFILE		2
838 #define HAMMER2_OBJTYPE_FIFO		4
839 #define HAMMER2_OBJTYPE_CDEV		5
840 #define HAMMER2_OBJTYPE_BDEV		6
841 #define HAMMER2_OBJTYPE_SOFTLINK	7
842 #define HAMMER2_OBJTYPE_HARDLINK	8	/* dummy entry for hardlink */
843 #define HAMMER2_OBJTYPE_SOCKET		9
844 #define HAMMER2_OBJTYPE_WHITEOUT	10
845 
846 #define HAMMER2_COPYID_NONE		0
847 #define HAMMER2_COPYID_LOCAL		((uint8_t)-1)
848 
849 #define HAMMER2_COPYID_COUNT		256
850 
851 /*
852  * PFS types identify a PFS on media and in LNK_SPAN messages.
853  *
854  * PFS types >= 16 belong to HAMMER, 0-15 are defined in sys/dmsg.h
855  */
856 /* 0-15 reserved by sys/dmsg.h */
857 #define HAMMER2_PFSTYPE_CACHE		16
858 #define HAMMER2_PFSTYPE_COPY		17
859 #define HAMMER2_PFSTYPE_SLAVE		18
860 #define HAMMER2_PFSTYPE_SOFT_SLAVE	19
861 #define HAMMER2_PFSTYPE_SOFT_MASTER	20
862 #define HAMMER2_PFSTYPE_MASTER		21
863 #define HAMMER2_PFSTYPE_SNAPSHOT	22
864 #define HAMMER2_PFSTYPE_SUPROOT		23
865 #define HAMMER2_PFSTYPE_MAX		32
866 
867 /*
868  *				Allocation Table
869  *
870  */
871 
872 
873 /*
874  * Flags (8 bits) - blockref, for freemap only
875  *
876  * Note that the minimum chunk size is 1KB so we could theoretically have
877  * 10 bits here, but we might have some future extension that allows a
878  * chunk size down to 256 bytes and if so we will need bits 8 and 9.
879  */
880 #define HAMMER2_AVF_SELMASK		0x03	/* select group */
881 #define HAMMER2_AVF_ALL_ALLOC		0x04	/* indicate all allocated */
882 #define HAMMER2_AVF_ALL_FREE		0x08	/* indicate all free */
883 #define HAMMER2_AVF_RESERVED10		0x10
884 #define HAMMER2_AVF_RESERVED20		0x20
885 #define HAMMER2_AVF_RESERVED40		0x40
886 #define HAMMER2_AVF_RESERVED80		0x80
887 #define HAMMER2_AVF_AVMASK32		((uint32_t)0xFFFFFF00LU)
888 #define HAMMER2_AVF_AVMASK64		((uint64_t)0xFFFFFFFFFFFFFF00LLU)
889 
890 #define HAMMER2_AV_SELECT_A		0x00
891 #define HAMMER2_AV_SELECT_B		0x01
892 #define HAMMER2_AV_SELECT_C		0x02
893 #define HAMMER2_AV_SELECT_D		0x03
894 
895 /*
896  * The volume header eats a 64K block.  There is currently an issue where
897  * we want to try to fit all nominal filesystem updates in a 512-byte section
898  * but it may be a lost cause due to the need for a blockset.
899  *
900  * All information is stored in host byte order.  The volume header's magic
901  * number may be checked to determine the byte order.  If you wish to mount
902  * between machines w/ different endian modes you'll need filesystem code
903  * which acts on the media data consistently (either all one way or all the
904  * other).  Our code currently does not do that.
905  *
906  * A read-write mount may have to recover missing allocations by doing an
907  * incremental mirror scan looking for modifications made after alloc_tid.
908  * If alloc_tid == last_tid then no recovery operation is needed.  Recovery
909  * operations are usually very, very fast.
910  *
911  * Read-only mounts do not need to do any recovery, access to the filesystem
912  * topology is always consistent after a crash (is always consistent, period).
913  * However, there may be shortcutted blockref updates present from deep in
914  * the tree which are stored in the volumeh eader and must be tracked on
915  * the fly.
916  *
917  * NOTE: The copyinfo[] array contains the configuration for both the
918  *	 cluster connections and any local media copies.  The volume
919  *	 header will be replicated for each local media copy.
920  *
921  *	 The mount command may specify multiple medias or just one and
922  *	 allow HAMMER2 to pick up the others when it checks the copyinfo[]
923  *	 array on mount.
924  *
925  * NOTE: root_blockref points to the super-root directory, not the root
926  *	 directory.  The root directory will be a subdirectory under the
927  *	 super-root.
928  *
929  *	 The super-root directory contains all root directories and all
930  *	 snapshots (readonly or writable).  It is possible to do a
931  *	 null-mount of the super-root using special path constructions
932  *	 relative to your mounted root.
933  *
934  * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were
935  *	 a PFS, including mirroring and storage quota operations, and this is
936  *	 prefered over creating discrete PFSs in the super-root.  Instead
937  *	 the super-root is most typically used to create writable snapshots,
938  *	 alternative roots, and so forth.  The super-root is also used by
939  *	 the automatic snapshotting mechanism.
940  */
941 #define HAMMER2_VOLUME_ID_HBO	0x48414d3205172011LLU
942 #define HAMMER2_VOLUME_ID_ABO	0x11201705324d4148LLU
943 
944 struct hammer2_volume_data {
945 	/*
946 	 * sector #0 - 512 bytes
947 	 */
948 	uint64_t	magic;			/* 0000 Signature */
949 	hammer2_off_t	boot_beg;		/* 0008 Boot area (future) */
950 	hammer2_off_t	boot_end;		/* 0010 (size = end - beg) */
951 	hammer2_off_t	aux_beg;		/* 0018 Aux area (future) */
952 	hammer2_off_t	aux_end;		/* 0020 (size = end - beg) */
953 	hammer2_off_t	volu_size;		/* 0028 Volume size, bytes */
954 
955 	uint32_t	version;		/* 0030 */
956 	uint32_t	flags;			/* 0034 */
957 	uint8_t		copyid;			/* 0038 copyid of phys vol */
958 	uint8_t		freemap_version;	/* 0039 freemap algorithm */
959 	uint8_t		peer_type;		/* 003A HAMMER2_PEER_xxx */
960 	uint8_t		reserved003B;		/* 003B */
961 	uint32_t	reserved003C;		/* 003C */
962 
963 	uuid_t		fsid;			/* 0040 */
964 	uuid_t		fstype;			/* 0050 */
965 
966 	/*
967 	 * allocator_size is precalculated at newfs time and does not include
968 	 * reserved blocks, boot, or redo areas.
969 	 *
970 	 * Initial non-reserved-area allocations do not use the freemap
971 	 * but instead adjust alloc_iterator.  Dynamic allocations take
972 	 * over starting at (allocator_beg).  This makes newfs_hammer2's
973 	 * job a lot easier and can also serve as a testing jig.
974 	 */
975 	hammer2_off_t	allocator_size;		/* 0060 Total data space */
976 	hammer2_off_t   allocator_free;		/* 0068	Free space */
977 	hammer2_off_t	allocator_beg;		/* 0070 Initial allocations */
978 
979 	/*
980 	 * mirror_tid reflects the highest committed super-root change
981 	 * freemap_tid reflects the highest committed freemap change
982 	 *
983 	 * NOTE: mirror_tid does not track (and should not track) changes
984 	 *	 made to or under PFS roots.
985 	 */
986 	hammer2_tid_t	mirror_tid;		/* 0078 committed tid (vol) */
987 	hammer2_tid_t	reserved0080;		/* 0080 */
988 	hammer2_tid_t	reserved0088;		/* 0088 */
989 	hammer2_tid_t	freemap_tid;		/* 0090 committed tid (fmap) */
990 	hammer2_tid_t	bulkfree_tid;		/* 0098 bulkfree incremental */
991 	hammer2_tid_t	reserved00A0[5];	/* 00A0-00C7 */
992 
993 	/*
994 	 * Copyids are allocated dynamically from the copyexists bitmap.
995 	 * An id from the active copies set (up to 8, see copyinfo later on)
996 	 * may still exist after the copy set has been removed from the
997 	 * volume header and its bit will remain active in the bitmap and
998 	 * cannot be reused until it is 100% removed from the hierarchy.
999 	 */
1000 	uint32_t	copyexists[8];		/* 00C8-00E7 copy exists bmap */
1001 	char		reserved0140[248];	/* 00E8-01DF */
1002 
1003 	/*
1004 	 * 32 bit CRC array at the end of the first 512 byte sector.
1005 	 *
1006 	 * icrc_sects[7] - First 512-4 bytes of volume header (including all
1007 	 *		   the other icrc's except this one).
1008 	 *
1009 	 * icrc_sects[6] - Sector 1 (512 bytes) of volume header, which is
1010 	 *		   the blockset for the root.
1011 	 *
1012 	 * icrc_sects[5] - Sector 2
1013 	 * icrc_sects[4] - Sector 3
1014 	 * icrc_sects[3] - Sector 4 (the freemap blockset)
1015 	 */
1016 	hammer2_crc32_t	icrc_sects[8];		/* 01E0-01FF */
1017 
1018 	/*
1019 	 * sector #1 - 512 bytes
1020 	 *
1021 	 * The entire sector is used by a blockset.
1022 	 */
1023 	hammer2_blockset_t sroot_blockset;	/* 0200-03FF Superroot dir */
1024 
1025 	/*
1026 	 * sector #2-7
1027 	 */
1028 	char	sector2[512];			/* 0400-05FF reserved */
1029 	char	sector3[512];			/* 0600-07FF reserved */
1030 	hammer2_blockset_t freemap_blockset;	/* 0800-09FF freemap  */
1031 	char	sector5[512];			/* 0A00-0BFF reserved */
1032 	char	sector6[512];			/* 0C00-0DFF reserved */
1033 	char	sector7[512];			/* 0E00-0FFF reserved */
1034 
1035 	/*
1036 	 * sector #8-71	- 32768 bytes
1037 	 *
1038 	 * Contains the configuration for up to 256 copyinfo targets.  These
1039 	 * specify local and remote copies operating as masters or slaves.
1040 	 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255
1041 	 * indicates the local media).
1042 	 *
1043 	 * Each inode contains a set of up to 8 copyids, either inherited
1044 	 * from its parent or explicitly specified in the inode, which
1045 	 * indexes into this array.
1046 	 */
1047 						/* 1000-8FFF copyinfo config */
1048 	hammer2_volconf_t copyinfo[HAMMER2_COPYID_COUNT];
1049 
1050 	/*
1051 	 * Remaining sections are reserved for future use.
1052 	 */
1053 	char		reserved0400[0x6FFC];	/* 9000-FFFB reserved */
1054 
1055 	/*
1056 	 * icrc on entire volume header
1057 	 */
1058 	hammer2_crc32_t	icrc_volheader;		/* FFFC-FFFF full volume icrc*/
1059 };
1060 
1061 typedef struct hammer2_volume_data hammer2_volume_data_t;
1062 
1063 /*
1064  * Various parts of the volume header have their own iCRCs.
1065  *
1066  * The first 512 bytes has its own iCRC stored at the end of the 512 bytes
1067  * and not included the icrc calculation.
1068  *
1069  * The second 512 bytes also has its own iCRC but it is stored in the first
1070  * 512 bytes so it covers the entire second 512 bytes.
1071  *
1072  * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes,
1073  * which is where the iCRC for the whole volume is stored.  This is currently
1074  * a catch-all for anything not individually iCRCd.
1075  */
1076 #define HAMMER2_VOL_ICRC_SECT0		7
1077 #define HAMMER2_VOL_ICRC_SECT1		6
1078 
1079 #define HAMMER2_VOLUME_BYTES		65536
1080 
1081 #define HAMMER2_VOLUME_ICRC0_OFF	0
1082 #define HAMMER2_VOLUME_ICRC1_OFF	512
1083 #define HAMMER2_VOLUME_ICRCVH_OFF	0
1084 
1085 #define HAMMER2_VOLUME_ICRC0_SIZE	(512 - 4)
1086 #define HAMMER2_VOLUME_ICRC1_SIZE	(512)
1087 #define HAMMER2_VOLUME_ICRCVH_SIZE	(65536 - 4)
1088 
1089 #define HAMMER2_VOL_VERSION_MIN		1
1090 #define HAMMER2_VOL_VERSION_DEFAULT	1
1091 #define HAMMER2_VOL_VERSION_WIP 	2
1092 
1093 #define HAMMER2_NUM_VOLHDRS		4
1094 
1095 union hammer2_media_data {
1096 	hammer2_volume_data_t	voldata;
1097         hammer2_inode_data_t    ipdata;
1098 	hammer2_blockref_t	npdata[HAMMER2_IND_COUNT_MAX];
1099 	hammer2_bmap_data_t	bmdata[HAMMER2_FREEMAP_COUNT];
1100 	char			buf[HAMMER2_PBUFSIZE];
1101 };
1102 
1103 typedef union hammer2_media_data hammer2_media_data_t;
1104 
1105 #endif /* !_VFS_HAMMER2_DISK_H_ */
1106