2 * Copyright (c) 2011-2014 The DragonFly Project. All rights reserved.
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>
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
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
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.
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
36 #ifndef _VFS_HAMMER2_DISK_H_
37 #define _VFS_HAMMER2_DISK_H_
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
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.
58 * Free block management is handled independently using blocks reserved by
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).
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.
73 * A full indirect block use supports 512 x 128-byte blockrefs in a 64KB
74 * buffer. Indirect blocks down to 1KB are supported to keep small
77 * A maximally sized file (2^64-1 bytes) requires ~6 indirect block levels
78 * using 64KB indirect blocks (128 byte refs, 512 or radix 9 per indblk).
80 * 16(datablk) + 9 + 9 + 9 + 9 + 9 + 9 = ~70.
81 * 16(datablk) + 7 + 9 + 9 + 9 + 9 + 9 = ~68. (smaller top level indblk)
83 * The actual depth depends on copies redundancy and whether the filesystem
84 * has chosen to use a smaller indirect block size at the top level or not.
86 #define HAMMER2_ALLOC_MIN 1024 /* minimum allocation size */
87 #define HAMMER2_RADIX_MIN 10 /* minimum allocation size 2^N */
88 #define HAMMER2_ALLOC_MAX 65536 /* maximum allocation size */
89 #define HAMMER2_RADIX_MAX 16 /* maximum allocation size 2^N */
90 #define HAMMER2_RADIX_KEY 64 /* number of bits in key */
93 * MINALLOCSIZE - The minimum allocation size. This can be smaller
94 * or larger than the minimum physical IO size.
96 * NOTE: Should not be larger than 1K since inodes
99 * MINIOSIZE - The minimum IO size. This must be less than
100 * or equal to HAMMER2_LBUFSIZE.
102 * HAMMER2_LBUFSIZE - Nominal buffer size for I/O rollups.
104 * HAMMER2_PBUFSIZE - Topological block size used by files for all
105 * blocks except the block straddling EOF.
107 * HAMMER2_SEGSIZE - Allocation map segment size, typically 2MB
108 * (space represented by a level0 bitmap).
111 #define HAMMER2_SEGSIZE (1 << HAMMER2_FREEMAP_LEVEL0_RADIX)
112 #define HAMMER2_SEGRADIX HAMMER2_FREEMAP_LEVEL0_RADIX
114 #define HAMMER2_PBUFRADIX 16 /* physical buf (1<<16) bytes */
115 #define HAMMER2_PBUFSIZE 65536
116 #define HAMMER2_LBUFRADIX 14 /* logical buf (1<<14) bytes */
117 #define HAMMER2_LBUFSIZE 16384
120 * Generally speaking we want to use 16K and 64K I/Os
122 #define HAMMER2_MINIORADIX HAMMER2_LBUFRADIX
123 #define HAMMER2_MINIOSIZE HAMMER2_LBUFSIZE
125 #define HAMMER2_IND_BYTES_MIN HAMMER2_LBUFSIZE
126 #define HAMMER2_IND_BYTES_MAX HAMMER2_PBUFSIZE
127 #define HAMMER2_IND_COUNT_MIN (HAMMER2_IND_BYTES_MIN / \
128 sizeof(hammer2_blockref_t))
129 #define HAMMER2_IND_COUNT_MAX (HAMMER2_IND_BYTES_MAX / \
130 sizeof(hammer2_blockref_t))
133 * In HAMMER2, arrays of blockrefs are fully set-associative, meaning that
134 * any element can occur at any index and holes can be anywhere. As a
135 * future optimization we will be able to flag that such arrays are sorted
136 * and thus optimize lookups, but for now we don't.
138 * Inodes embed either 512 bytes of direct data or an array of 8 blockrefs,
139 * resulting in highly efficient storage for files <= 512 bytes and for files
140 * <= 512KB. Up to 8 directory entries can be referenced from a directory
141 * without requiring an indirect block.
143 * Indirect blocks are typically either 4KB (64 blockrefs / ~4MB represented),
144 * or 64KB (1024 blockrefs / ~64MB represented).
146 #define HAMMER2_SET_RADIX 2 /* radix 2 = 4 entries */
147 #define HAMMER2_SET_COUNT (1 << HAMMER2_SET_RADIX)
148 #define HAMMER2_EMBEDDED_BYTES 512 /* inode blockset/dd size */
149 #define HAMMER2_EMBEDDED_RADIX 9
151 #define HAMMER2_PBUFMASK (HAMMER2_PBUFSIZE - 1)
152 #define HAMMER2_LBUFMASK (HAMMER2_LBUFSIZE - 1)
153 #define HAMMER2_SEGMASK (HAMMER2_SEGSIZE - 1)
155 #define HAMMER2_LBUFMASK64 ((hammer2_off_t)HAMMER2_LBUFMASK)
156 #define HAMMER2_PBUFSIZE64 ((hammer2_off_t)HAMMER2_PBUFSIZE)
157 #define HAMMER2_PBUFMASK64 ((hammer2_off_t)HAMMER2_PBUFMASK)
158 #define HAMMER2_SEGSIZE64 ((hammer2_off_t)HAMMER2_SEGSIZE)
159 #define HAMMER2_SEGMASK64 ((hammer2_off_t)HAMMER2_SEGMASK)
161 #define HAMMER2_UUID_STRING "5cbb9ad1-862d-11dc-a94d-01301bb8a9f5"
164 * A HAMMER2 filesystem is always sized in multiples of 8MB.
166 * A 4MB segment is reserved at the beginning of each 2GB zone. This segment
167 * contains the volume header (or backup volume header), the free block
168 * table, and possibly other information in the future.
170 * 4MB = 64 x 64K blocks. Each 4MB segment is broken down as follows:
172 * +-----------------------+
173 * | Volume Hdr | block 0 volume header & alternates
174 * +-----------------------+ (first four zones only)
175 * | FreeBlk Section A | block 1-4
176 * +-----------------------+
177 * | FreeBlk Section B | block 5-8
178 * +-----------------------+
179 * | FreeBlk Section C | block 9-12
180 * +-----------------------+
181 * | FreeBlk Section D | block 13-16
182 * +-----------------------+
186 * +-----------------------+
188 * The first few 2GB zones contain volume headers and volume header backups.
189 * After that the volume header block# is reserved for future use. Similarly,
190 * there are many blocks related to various Freemap levels which are not
191 * used in every segment and those are also reserved for future use.
193 * Freemap (see the FREEMAP document)
195 * The freemap utilizes blocks #1-16 in 8 sets of 4 blocks. Each block in
196 * a set represents a level of depth in the freemap topology. Eight sets
197 * exist to prevent live updates from disturbing the state of the freemap
198 * were a crash/reboot to occur. That is, a live update is not committed
199 * until the update's flush reaches the volume root. There are FOUR volume
200 * roots representing the last four synchronization points, so the freemap
201 * must be consistent no matter which volume root is chosen by the mount
204 * Each freemap set is 4 x 64K blocks and represents the 2GB, 2TB, 2PB,
205 * and 2EB indirect map. The volume header itself has a set of 8 freemap
206 * blockrefs representing another 3 bits, giving us a total 64 bits of
207 * representable address space.
209 * The Level 0 64KB block represents 2GB of storage represented by
210 * (64 x struct hammer2_bmap_data). Each structure represents 2MB of storage
211 * and has a 256 bit bitmap, using 2 bits to represent a 16KB chunk of
212 * storage. These 2 bits represent the following states:
215 * 01 (reserved) (Possibly partially allocated)
219 * One important thing to note here is that the freemap resolution is 16KB,
220 * but the minimum storage allocation size is 1KB. The hammer2 vfs keeps
221 * track of sub-allocations in memory, which means that on a unmount or reboot
222 * the entire 16KB of a partially allocated block will be considered fully
223 * allocated. It is possible for fragmentation to build up over time, but
224 * defragmentation is fairly easy to accomplish since all modifications
225 * allocate a new block.
227 * The Second thing to note is that due to the way snapshots and inode
228 * replication works, deleting a file cannot immediately free the related
229 * space. Furthermore, deletions often do not bother to traverse the
230 * block subhierarchy being deleted. And to go even further, whole
231 * sub-directory trees can be deleted simply by deleting the directory inode
232 * at the top. So even though we have a symbol to represent a 'possibly free'
233 * block (binary 10), only the bulk free scanning code can actually use it.
234 * Normal 'rm's or other deletions do not.
236 * WARNING! ZONE_SEG and VOLUME_ALIGN must be a multiple of 1<<LEVEL0_RADIX
237 * (i.e. a multiple of 2MB). VOLUME_ALIGN must be >= ZONE_SEG.
241 * (1) Modifications to freemap blocks 'allocate' a new copy (aka use a block
242 * from the next set). The new copy is reused until a flush occurs at
243 * which point the next modification will then rotate to the next set.
245 * (2) A total of 10 freemap sets is required.
247 * - 8 sets - 2 sets per volume header backup x 4 volume header backups
248 * - 2 sets used as backing store for the bulk freemap scan.
249 * - The freemap recovery scan which runs on-mount just uses the inactive
250 * set for whichever volume header was selected by the mount code.
253 #define HAMMER2_VOLUME_ALIGN (8 * 1024 * 1024)
254 #define HAMMER2_VOLUME_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN)
255 #define HAMMER2_VOLUME_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1)
256 #define HAMMER2_VOLUME_ALIGNMASK64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGNMASK)
258 #define HAMMER2_NEWFS_ALIGN (HAMMER2_VOLUME_ALIGN)
259 #define HAMMER2_NEWFS_ALIGN64 ((hammer2_off_t)HAMMER2_VOLUME_ALIGN)
260 #define HAMMER2_NEWFS_ALIGNMASK (HAMMER2_VOLUME_ALIGN - 1)
261 #define HAMMER2_NEWFS_ALIGNMASK64 ((hammer2_off_t)HAMMER2_NEWFS_ALIGNMASK)
263 #define HAMMER2_ZONE_BYTES64 (2LLU * 1024 * 1024 * 1024)
264 #define HAMMER2_ZONE_MASK64 (HAMMER2_ZONE_BYTES64 - 1)
265 #define HAMMER2_ZONE_SEG (4 * 1024 * 1024)
266 #define HAMMER2_ZONE_SEG64 ((hammer2_off_t)HAMMER2_ZONE_SEG)
267 #define HAMMER2_ZONE_BLOCKS_SEG (HAMMER2_ZONE_SEG / HAMMER2_PBUFSIZE)
269 #define HAMMER2_ZONE_FREEMAP_INC 5 /* 5 deep */
271 #define HAMMER2_ZONE_VOLHDR 0 /* volume header or backup */
272 #define HAMMER2_ZONE_FREEMAP_00 1 /* normal freemap rotation */
273 #define HAMMER2_ZONE_FREEMAP_01 6 /* normal freemap rotation */
274 #define HAMMER2_ZONE_FREEMAP_02 11 /* normal freemap rotation */
275 #define HAMMER2_ZONE_FREEMAP_03 16 /* normal freemap rotation */
276 #define HAMMER2_ZONE_FREEMAP_04 21 /* normal freemap rotation */
277 #define HAMMER2_ZONE_FREEMAP_05 26 /* normal freemap rotation */
278 #define HAMMER2_ZONE_FREEMAP_06 31 /* normal freemap rotation */
279 #define HAMMER2_ZONE_FREEMAP_07 36 /* normal freemap rotation */
280 #define HAMMER2_ZONE_FREEMAP_END 41 /* (non-inclusive) */
282 #define HAMMER2_ZONE_UNUSED41 41
283 #define HAMMER2_ZONE_UNUSED42 42
284 #define HAMMER2_ZONE_UNUSED43 43
285 #define HAMMER2_ZONE_UNUSED44 44
286 #define HAMMER2_ZONE_UNUSED45 45
287 #define HAMMER2_ZONE_UNUSED46 46
288 #define HAMMER2_ZONE_UNUSED47 47
289 #define HAMMER2_ZONE_UNUSED48 48
290 #define HAMMER2_ZONE_UNUSED49 49
291 #define HAMMER2_ZONE_UNUSED50 50
292 #define HAMMER2_ZONE_UNUSED51 51
293 #define HAMMER2_ZONE_UNUSED52 52
294 #define HAMMER2_ZONE_UNUSED53 53
295 #define HAMMER2_ZONE_UNUSED54 54
296 #define HAMMER2_ZONE_UNUSED55 55
297 #define HAMMER2_ZONE_UNUSED56 56
298 #define HAMMER2_ZONE_UNUSED57 57
299 #define HAMMER2_ZONE_UNUSED58 58
300 #define HAMMER2_ZONE_UNUSED59 59
301 #define HAMMER2_ZONE_UNUSED60 60
302 #define HAMMER2_ZONE_UNUSED61 61
303 #define HAMMER2_ZONE_UNUSED62 62
304 #define HAMMER2_ZONE_UNUSED63 63
305 #define HAMMER2_ZONE_END 64 /* non-inclusive */
307 #define HAMMER2_NFREEMAPS 8 /* FREEMAP_00 - FREEMAP_07 */
309 /* relative to FREEMAP_x */
310 #define HAMMER2_ZONEFM_LEVEL1 0 /* 1GB leafmap */
311 #define HAMMER2_ZONEFM_LEVEL2 1 /* 256GB indmap */
312 #define HAMMER2_ZONEFM_LEVEL3 2 /* 64TB indmap */
313 #define HAMMER2_ZONEFM_LEVEL4 3 /* 16PB indmap */
314 #define HAMMER2_ZONEFM_LEVEL5 4 /* 4EB indmap */
315 /* LEVEL6 is a set of 4 blockrefs in the volume header 16EB */
318 * Freemap radix. Assumes a set-count of 4, 128-byte blockrefs,
319 * 32KB indirect block for freemap (LEVELN_PSIZE below).
321 * Leaf entry represents 4MB of storage broken down into a 512-bit
322 * bitmap, 2-bits per entry. So course bitmap item represents 16KB.
324 #if HAMMER2_SET_COUNT != 4
325 #error "hammer2_disk.h - freemap assumes SET_COUNT is 4"
327 #define HAMMER2_FREEMAP_LEVEL6_RADIX 64 /* 16EB (end) */
328 #define HAMMER2_FREEMAP_LEVEL5_RADIX 62 /* 4EB */
329 #define HAMMER2_FREEMAP_LEVEL4_RADIX 54 /* 16PB */
330 #define HAMMER2_FREEMAP_LEVEL3_RADIX 46 /* 64TB */
331 #define HAMMER2_FREEMAP_LEVEL2_RADIX 38 /* 256GB */
332 #define HAMMER2_FREEMAP_LEVEL1_RADIX 30 /* 1GB */
333 #define HAMMER2_FREEMAP_LEVEL0_RADIX 22 /* 4MB (128by in l-1 leaf) */
335 #define HAMMER2_FREEMAP_LEVELN_PSIZE 32768 /* physical bytes */
337 #define HAMMER2_FREEMAP_LEVEL5_SIZE ((hammer2_off_t)1 << \
338 HAMMER2_FREEMAP_LEVEL5_RADIX)
339 #define HAMMER2_FREEMAP_LEVEL4_SIZE ((hammer2_off_t)1 << \
340 HAMMER2_FREEMAP_LEVEL4_RADIX)
341 #define HAMMER2_FREEMAP_LEVEL3_SIZE ((hammer2_off_t)1 << \
342 HAMMER2_FREEMAP_LEVEL3_RADIX)
343 #define HAMMER2_FREEMAP_LEVEL2_SIZE ((hammer2_off_t)1 << \
344 HAMMER2_FREEMAP_LEVEL2_RADIX)
345 #define HAMMER2_FREEMAP_LEVEL1_SIZE ((hammer2_off_t)1 << \
346 HAMMER2_FREEMAP_LEVEL1_RADIX)
347 #define HAMMER2_FREEMAP_LEVEL0_SIZE ((hammer2_off_t)1 << \
348 HAMMER2_FREEMAP_LEVEL0_RADIX)
350 #define HAMMER2_FREEMAP_LEVEL5_MASK (HAMMER2_FREEMAP_LEVEL5_SIZE - 1)
351 #define HAMMER2_FREEMAP_LEVEL4_MASK (HAMMER2_FREEMAP_LEVEL4_SIZE - 1)
352 #define HAMMER2_FREEMAP_LEVEL3_MASK (HAMMER2_FREEMAP_LEVEL3_SIZE - 1)
353 #define HAMMER2_FREEMAP_LEVEL2_MASK (HAMMER2_FREEMAP_LEVEL2_SIZE - 1)
354 #define HAMMER2_FREEMAP_LEVEL1_MASK (HAMMER2_FREEMAP_LEVEL1_SIZE - 1)
355 #define HAMMER2_FREEMAP_LEVEL0_MASK (HAMMER2_FREEMAP_LEVEL0_SIZE - 1)
357 #define HAMMER2_FREEMAP_COUNT (int)(HAMMER2_FREEMAP_LEVELN_PSIZE / \
358 sizeof(hammer2_bmap_data_t))
361 * 16KB bitmap granularity (x2 bits per entry).
363 #define HAMMER2_FREEMAP_BLOCK_RADIX 14
364 #define HAMMER2_FREEMAP_BLOCK_SIZE (1 << HAMMER2_FREEMAP_BLOCK_RADIX)
365 #define HAMMER2_FREEMAP_BLOCK_MASK (HAMMER2_FREEMAP_BLOCK_SIZE - 1)
368 * bitmap[] structure. 2 bits per HAMMER2_FREEMAP_BLOCK_SIZE.
370 * 8 x 64-bit elements, 2 bits per block.
371 * 32 blocks (radix 5) per element.
372 * representing INDEX_SIZE bytes worth of storage per element.
375 typedef uint64_t hammer2_bitmap_t;
377 #define HAMMER2_BMAP_ALLONES ((hammer2_bitmap_t)-1)
378 #define HAMMER2_BMAP_ELEMENTS 8
379 #define HAMMER2_BMAP_BITS_PER_ELEMENT 64
380 #define HAMMER2_BMAP_INDEX_RADIX 5 /* 32 blocks per element */
381 #define HAMMER2_BMAP_BLOCKS_PER_ELEMENT (1 << HAMMER2_BMAP_INDEX_RADIX)
383 #define HAMMER2_BMAP_INDEX_SIZE (HAMMER2_FREEMAP_BLOCK_SIZE * \
384 HAMMER2_BMAP_BLOCKS_PER_ELEMENT)
385 #define HAMMER2_BMAP_INDEX_MASK (HAMMER2_BMAP_INDEX_SIZE - 1)
388 * Two linear areas can be reserved after the initial 2MB segment in the base
389 * zone (the one starting at offset 0). These areas are NOT managed by the
390 * block allocator and do not fall under HAMMER2 crc checking rules based
391 * at the volume header (but can be self-CRCd internally, depending).
393 #define HAMMER2_BOOT_MIN_BYTES HAMMER2_VOLUME_ALIGN
394 #define HAMMER2_BOOT_NOM_BYTES (64*1024*1024)
395 #define HAMMER2_BOOT_MAX_BYTES (256*1024*1024)
397 #define HAMMER2_REDO_MIN_BYTES HAMMER2_VOLUME_ALIGN
398 #define HAMMER2_REDO_NOM_BYTES (256*1024*1024)
399 #define HAMMER2_REDO_MAX_BYTES (1024*1024*1024)
402 * Most HAMMER2 types are implemented as unsigned 64-bit integers.
403 * Transaction ids are monotonic.
405 * We utilize 32-bit iSCSI CRCs.
407 typedef uint64_t hammer2_tid_t;
408 typedef uint64_t hammer2_off_t;
409 typedef uint64_t hammer2_key_t;
410 typedef uint32_t hammer2_crc32_t;
413 * Miscellanious ranges (all are unsigned).
415 #define HAMMER2_TID_MIN 1ULL
416 #define HAMMER2_TID_MAX 0xFFFFFFFFFFFFFFFFULL
417 #define HAMMER2_KEY_MIN 0ULL
418 #define HAMMER2_KEY_MAX 0xFFFFFFFFFFFFFFFFULL
419 #define HAMMER2_OFFSET_MIN 0ULL
420 #define HAMMER2_OFFSET_MAX 0xFFFFFFFFFFFFFFFFULL
423 * HAMMER2 data offset special cases and masking.
425 * All HAMMER2 data offsets have to be broken down into a 64K buffer base
426 * offset (HAMMER2_OFF_MASK_HI) and a 64K buffer index (HAMMER2_OFF_MASK_LO).
428 * Indexes into physical buffers are always 64-byte aligned. The low 6 bits
429 * of the data offset field specifies how large the data chunk being pointed
430 * to as a power of 2. The theoretical minimum radix is thus 6 (The space
431 * needed in the low bits of the data offset field). However, the practical
432 * minimum allocation chunk size is 1KB (a radix of 10), so HAMMER2 sets
433 * HAMMER2_RADIX_MIN to 10. The maximum radix is currently 16 (64KB), but
434 * we fully intend to support larger extents in the future.
436 #define HAMMER2_OFF_BAD ((hammer2_off_t)-1)
437 #define HAMMER2_OFF_MASK 0xFFFFFFFFFFFFFFC0ULL
438 #define HAMMER2_OFF_MASK_LO (HAMMER2_OFF_MASK & HAMMER2_PBUFMASK64)
439 #define HAMMER2_OFF_MASK_HI (~HAMMER2_PBUFMASK64)
440 #define HAMMER2_OFF_MASK_RADIX 0x000000000000003FULL
441 #define HAMMER2_MAX_COPIES 6
444 * HAMMER2 directory support and pre-defined keys
446 #define HAMMER2_DIRHASH_VISIBLE 0x8000000000000000ULL
447 #define HAMMER2_DIRHASH_USERMSK 0x7FFFFFFFFFFFFFFFULL
448 #define HAMMER2_DIRHASH_LOMASK 0x0000000000007FFFULL
449 #define HAMMER2_DIRHASH_HIMASK 0xFFFFFFFFFFFF0000ULL
450 #define HAMMER2_DIRHASH_FORCED 0x0000000000008000ULL /* bit forced on */
452 #define HAMMER2_SROOT_KEY 0x0000000000000000ULL /* volume to sroot */
453 #define HAMMER2_BOOT_KEY 0xd9b36ce135528000ULL /* sroot to BOOT PFS */
455 /************************************************************************
457 ************************************************************************
460 * All HAMMER2 directories directly under the super-root on your local
461 * media can be mounted separately, even if they share the same physical
464 * When you do a HAMMER2 mount you are effectively tying into a HAMMER2
465 * cluster via local media. The local media does not have to participate
466 * in the cluster, other than to provide the hammer2_volconf[] array and
467 * root inode for the mount.
469 * This is important: The mount device path you specify serves to bootstrap
470 * your entry into the cluster, but your mount will make active connections
471 * to ALL copy elements in the hammer2_volconf[] array which match the
472 * PFSID of the directory in the super-root that you specified. The local
473 * media path does not have to be mentioned in this array but becomes part
474 * of the cluster based on its type and access rights. ALL ELEMENTS ARE
475 * TREATED ACCORDING TO TYPE NO MATTER WHICH ONE YOU MOUNT FROM.
477 * The actual cluster may be far larger than the elements you list in the
478 * hammer2_volconf[] array. You list only the elements you wish to
479 * directly connect to and you are able to access the rest of the cluster
480 * indirectly through those connections.
482 * WARNING! This structure must be exactly 128 bytes long for its config
483 * array to fit in the volume header.
485 struct hammer2_volconf {
486 uint8_t copyid; /* 00 copyid 0-255 (must match slot) */
487 uint8_t inprog; /* 01 operation in progress, or 0 */
488 uint8_t chain_to; /* 02 operation chaining to, or 0 */
489 uint8_t chain_from; /* 03 operation chaining from, or 0 */
490 uint16_t flags; /* 04-05 flags field */
491 uint8_t error; /* 06 last operational error */
492 uint8_t priority; /* 07 priority and round-robin flag */
493 uint8_t remote_pfs_type;/* 08 probed direct remote PFS type */
494 uint8_t reserved08[23]; /* 09-1F */
495 uuid_t pfs_clid; /* 20-2F copy target must match this uuid */
496 uint8_t label[16]; /* 30-3F import/export label */
497 uint8_t path[64]; /* 40-7F target specification string or key */
500 typedef struct hammer2_volconf hammer2_volconf_t;
502 #define DMSG_VOLF_ENABLED 0x0001
503 #define DMSG_VOLF_INPROG 0x0002
504 #define DMSG_VOLF_CONN_RR 0x80 /* round-robin at same priority */
505 #define DMSG_VOLF_CONN_EF 0x40 /* media errors flagged */
506 #define DMSG_VOLF_CONN_PRI 0x0F /* select priority 0-15 (15=best) */
508 struct dmsg_lnk_hammer2_volconf {
510 hammer2_volconf_t copy; /* copy spec */
514 int64_t reserved02[32];
517 typedef struct dmsg_lnk_hammer2_volconf dmsg_lnk_hammer2_volconf_t;
519 #define DMSG_LNK_HAMMER2_VOLCONF DMSG_LNK(DMSG_LNK_CMD_HAMMER2_VOLCONF, \
520 dmsg_lnk_hammer2_volconf)
522 #define H2_LNK_VOLCONF(msg) ((dmsg_lnk_hammer2_volconf_t *)(msg)->any.buf)
525 * The media block reference structure. This forms the core of the HAMMER2
526 * media topology recursion. This 128-byte data structure is embedded in the
527 * volume header, in inodes (which are also directory entries), and in
530 * A blockref references a single media item, which typically can be a
531 * directory entry (aka inode), indirect block, or data block.
533 * The primary feature a blockref represents is the ability to validate
534 * the entire tree underneath it via its check code. Any modification to
535 * anything propagates up the blockref tree all the way to the root, replacing
536 * the related blocks and compounding the generated check code.
538 * The check code can be a simple 32-bit iscsi code, a 64-bit crc, or as
539 * complex as a 512 bit cryptographic hash. I originally used a 64-byte
540 * blockref but later expanded it to 128 bytes to be able to support the
541 * larger check code as well as to embed statistics for quota operation.
543 * Simple check codes are not sufficient for unverified dedup. Even with
544 * a maximally-sized check code unverified dedup should only be used in
545 * in subdirectory trees where you do not need 100% data integrity.
547 * Unverified dedup is deduping based on meta-data only without verifying
548 * that the data blocks are actually identical. Verified dedup guarantees
549 * integrity but is a far more I/O-expensive operation.
553 * mirror_tid - per cluster node modified (propagated upward by flush)
554 * modify_tid - clc record modified (not propagated).
555 * update_tid - clc record updated (propagated upward on verification)
557 * CLC - Stands for 'Cluster Level Change', identifiers which are identical
558 * within the topology across all cluster nodes (when fully
561 * NOTE: The range of keys represented by the blockref is (key) to
562 * ((key) + (1LL << keybits) - 1). HAMMER2 usually populates
563 * blocks bottom-up, inserting a new root when radix expansion
568 * A number of blockref fields are reserved and should generally be set to
569 * 0 for future compatibility.
571 * FUTURE BLOCKREF EXPANSION
573 * CONTENT ADDRESSABLE INDEXING (future) - Using a 256 or 512-bit check code.
575 struct hammer2_blockref { /* MUST BE EXACTLY 64 BYTES */
576 uint8_t type; /* type of underlying item */
577 uint8_t methods; /* check method & compression method */
578 uint8_t copyid; /* specify which copy this is */
579 uint8_t keybits; /* #of keybits masked off 0=leaf */
580 uint8_t vradix; /* virtual data/meta-data size */
581 uint8_t flags; /* blockref flags */
584 hammer2_key_t key; /* key specification */
585 hammer2_tid_t mirror_tid; /* media flush topology & freemap */
586 hammer2_tid_t modify_tid; /* clc modify (not propagated) */
587 hammer2_off_t data_off; /* low 6 bits is phys size (radix)*/
588 hammer2_key_t data_count; /* statistics aggregation */
589 hammer2_key_t inode_count; /* statistics aggregation */
590 hammer2_tid_t update_tid; /* clc modify (propagated upward) */
591 union { /* check info */
595 uint32_t reserved[15];
599 uint64_t reserved[7];
614 * Freemap hints are embedded in addition to the icrc32.
616 * bigmask - Radixes available for allocation (0-31).
617 * Heuristical (may be permissive but not
618 * restrictive). Typically only radix values
619 * 10-16 are used (i.e. (1<<10) through (1<<16)).
621 * avail - Total available space remaining, in bytes
625 uint32_t bigmask; /* available radixes */
626 uint64_t avail; /* total available bytes */
632 typedef struct hammer2_blockref hammer2_blockref_t;
634 #define HAMMER2_BLOCKREF_BYTES 128 /* blockref struct in bytes */
637 * On-media and off-media blockref types.
639 * types >= 128 are pseudo values that should never be present on-media.
641 #define HAMMER2_BREF_TYPE_EMPTY 0
642 #define HAMMER2_BREF_TYPE_INODE 1
643 #define HAMMER2_BREF_TYPE_INDIRECT 2
644 #define HAMMER2_BREF_TYPE_DATA 3
645 #define HAMMER2_BREF_TYPE_UNUSED04 4
646 #define HAMMER2_BREF_TYPE_FREEMAP_NODE 5
647 #define HAMMER2_BREF_TYPE_FREEMAP_LEAF 6
648 #define HAMMER2_BREF_TYPE_FREEMAP 254 /* pseudo-type */
649 #define HAMMER2_BREF_TYPE_VOLUME 255 /* pseudo-type */
651 #define HAMMER2_BREF_FLAG_PFSROOT 0x01 /* see also related opflag */
652 #define HAMMER2_BREF_FLAG_ZERO 0x02
655 * Encode/decode check mode and compression mode for
656 * bref.methods. The compression level is not encoded in
659 #define HAMMER2_ENC_CHECK(n) (((n) & 15) << 4)
660 #define HAMMER2_DEC_CHECK(n) (((n) >> 4) & 15)
661 #define HAMMER2_ENC_COMP(n) ((n) & 15)
662 #define HAMMER2_DEC_COMP(n) ((n) & 15)
664 #define HAMMER2_CHECK_NONE 0
665 #define HAMMER2_CHECK_DISABLED 1
666 #define HAMMER2_CHECK_ISCSI32 2
667 #define HAMMER2_CHECK_CRC64 3
668 #define HAMMER2_CHECK_SHA192 4
669 #define HAMMER2_CHECK_FREEMAP 5
671 /* user-specifiable check modes only */
672 #define HAMMER2_CHECK_STRINGS { "none", "disabled", "crc32", \
674 #define HAMMER2_CHECK_STRINGS_COUNT 5
677 * Encode/decode check or compression algorithm request in
678 * ipdata->meta.check_algo and ipdata->meta.comp_algo.
680 #define HAMMER2_ENC_ALGO(n) (n)
681 #define HAMMER2_DEC_ALGO(n) ((n) & 15)
682 #define HAMMER2_ENC_LEVEL(n) ((n) << 4)
683 #define HAMMER2_DEC_LEVEL(n) (((n) >> 4) & 15)
685 #define HAMMER2_COMP_NONE 0
686 #define HAMMER2_COMP_AUTOZERO 1
687 #define HAMMER2_COMP_LZ4 2
688 #define HAMMER2_COMP_ZLIB 3
690 #define HAMMER2_COMP_NEWFS_DEFAULT HAMMER2_COMP_LZ4
691 #define HAMMER2_COMP_STRINGS { "none", "autozero", "lz4", "zlib" }
692 #define HAMMER2_COMP_STRINGS_COUNT 4
696 * HAMMER2 block references are collected into sets of 4 blockrefs. These
697 * sets are fully associative, meaning the elements making up a set are
698 * not sorted in any way and may contain duplicate entries, holes, or
699 * entries which shortcut multiple levels of indirection. Sets are used
702 * (1) When redundancy is desired a set may contain several duplicate
703 * entries pointing to different copies of the same data. Up to 4 copies
706 * (2) The blockrefs in a set can shortcut multiple levels of indirections
707 * within the bounds imposed by the parent of set.
709 * When a set fills up another level of indirection is inserted, moving
710 * some or all of the set's contents into indirect blocks placed under the
711 * set. This is a top-down approach in that indirect blocks are not created
712 * until the set actually becomes full (that is, the entries in the set can
713 * shortcut the indirect blocks when the set is not full). Depending on how
714 * things are filled multiple indirect blocks will eventually be created.
716 * Indirect blocks are typically 4KB (64 entres) or 64KB (1024 entries) and
717 * are also treated as fully set-associative.
719 struct hammer2_blockset {
720 hammer2_blockref_t blockref[HAMMER2_SET_COUNT];
723 typedef struct hammer2_blockset hammer2_blockset_t;
726 * Catch programmer snafus
728 #if (1 << HAMMER2_SET_RADIX) != HAMMER2_SET_COUNT
729 #error "hammer2 direct radix is incorrect"
731 #if (1 << HAMMER2_PBUFRADIX) != HAMMER2_PBUFSIZE
732 #error "HAMMER2_PBUFRADIX and HAMMER2_PBUFSIZE are inconsistent"
734 #if (1 << HAMMER2_RADIX_MIN) != HAMMER2_ALLOC_MIN
735 #error "HAMMER2_RADIX_MIN and HAMMER2_ALLOC_MIN are inconsistent"
739 * hammer2_bmap_data - A freemap entry in the LEVEL1 block.
741 * Each 128-byte entry contains the bitmap and meta-data required to manage
742 * a LEVEL0 (128KB) block of storage. The storage is managed in 128 x 1KB
745 * A smaller allocation granularity is supported via a linear iterator and/or
746 * must otherwise be tracked in ram.
748 * (data structure must be 128 bytes exactly)
750 * linear - A BYTE linear allocation offset used for sub-16KB allocations
751 * only. May contain values between 0 and 2MB. Must be ignored
752 * if 16KB-aligned (i.e. force bitmap scan), otherwise may be
753 * used to sub-allocate within the 16KB block (which is already
754 * marked as allocated in the bitmap).
756 * Sub-allocations need only be 1KB-aligned and do not have to be
757 * size-aligned, and 16KB or larger allocations do not update this
758 * field, resulting in pretty good packing.
760 * Please note that file data granularity may be limited by
761 * other issues such as buffer cache direct-mapping and the
762 * desire to support sector sizes up to 16KB (so H2 only issues
763 * I/O's in multiples of 16KB anyway).
765 * class - Clustering class. Cleared to 0 only if the entire leaf becomes
766 * free. Used to cluster device buffers so all elements must have
767 * the same device block size, but may mix logical sizes.
769 * Typically integrated with the blockref type in the upper 8 bits
770 * to localize inodes and indrect blocks, improving bulk free scans
771 * and directory scans.
773 * bitmap - Two bits per 16KB allocation block arranged in arrays of
774 * 32-bit elements, 256x2 bits representing ~4MB worth of media
775 * storage. Bit patterns are as follows:
782 struct hammer2_bmap_data {
783 int32_t linear; /* 00 linear sub-granular allocation offset */
784 uint16_t class; /* 04-05 clustering class ((type<<8)|radix) */
785 uint8_t reserved06; /* 06 */
786 uint8_t reserved07; /* 07 */
787 uint32_t reserved08; /* 08 */
788 uint32_t reserved0C; /* 0C */
789 uint32_t reserved10; /* 10 */
790 uint32_t reserved14; /* 14 */
791 uint32_t reserved18; /* 18 */
792 uint32_t avail; /* 1C */
793 uint32_t reserved20[8]; /* 20-3F 256 bits manages 128K/1KB/2-bits */
794 /* 40-7F 512 bits manages 4MB of storage */
795 hammer2_bitmap_t bitmapq[HAMMER2_BMAP_ELEMENTS];
798 typedef struct hammer2_bmap_data hammer2_bmap_data_t;
801 * In HAMMER2 inodes ARE directory entries, with a special exception for
802 * hardlinks. The inode number is stored in the inode rather than being
803 * based on the location of the inode (since the location moves every time
804 * the inode or anything underneath the inode is modified).
806 * The inode is 1024 bytes, made up of 256 bytes of meta-data, 256 bytes
807 * for the filename, and 512 bytes worth of direct file data OR an embedded
808 * blockset. The in-memory hammer2_inode structure contains only the mostly-
809 * node-independent meta-data portion (some flags are node-specific and will
810 * not be synchronized). The rest of the inode is node-specific and chain I/O
811 * is required to obtain it.
813 * Directories represent one inode per blockref. Inodes are not laid out
814 * as a file but instead are represented by the related blockrefs. The
815 * blockrefs, in turn, are indexed by the 64-bit directory hash key. Remember
816 * that blocksets are fully associative, so a certain degree efficiency is
817 * achieved just from that.
819 * Up to 512 bytes of direct data can be embedded in an inode, and since
820 * inodes are essentially directory entries this also means that small data
821 * files end up simply being laid out linearly in the directory, resulting
822 * in fewer seeks and highly optimal access.
824 * The compression mode can be changed at any time in the inode and is
825 * recorded on a blockref-by-blockref basis.
827 * Hardlinks are supported via the inode map. Essentially the way a hardlink
828 * works is that all individual directory entries representing the same file
829 * are special cased and specify the same inode number. The actual file
830 * is placed in the nearest parent directory that is parent to all instances
831 * of the hardlink. If all hardlinks to a file are in the same directory
832 * the actual file will also be placed in that directory. This file uses
833 * the inode number as the directory entry key and is invisible to normal
834 * directory scans. Real directory entry keys are differentiated from the
835 * inode number key via bit 63. Access to the hardlink silently looks up
836 * the real file and forwards all operations to that file. Removal of the
837 * last hardlink also removes the real file.
839 * (attr_tid) is only updated when the inode's specific attributes or regular
840 * file size has changed, and affects path lookups and stat. (attr_tid)
841 * represents a special cache coherency lock under the inode. The inode
842 * blockref's modify_tid will always cover it.
844 * (dirent_tid) is only updated when an entry under a directory inode has
845 * been created, deleted, renamed, or had its attributes change, and affects
846 * directory lookups and scans. (dirent_tid) represents another special cache
847 * coherency lock under the inode. The inode blockref's modify_tid will
850 #define HAMMER2_INODE_BYTES 1024 /* (asserted by code) */
851 #define HAMMER2_INODE_MAXNAME 256 /* maximum name in bytes */
852 #define HAMMER2_INODE_VERSION_ONE 1
854 #define HAMMER2_INODE_HIDDENDIR 16 /* special inode */
855 #define HAMMER2_INODE_START 1024 /* dynamically allocated */
857 struct hammer2_inode_meta {
858 uint16_t version; /* 0000 inode data version */
859 uint8_t reserved02; /* 0002 */
860 uint8_t pfs_subtype; /* 0003 pfs sub-type */
863 * core inode attributes, inode type, misc flags
865 uint32_t uflags; /* 0004 chflags */
866 uint32_t rmajor; /* 0008 available for device nodes */
867 uint32_t rminor; /* 000C available for device nodes */
868 uint64_t ctime; /* 0010 inode change time */
869 uint64_t mtime; /* 0018 modified time */
870 uint64_t atime; /* 0020 access time (unsupported) */
871 uint64_t btime; /* 0028 birth time */
872 uuid_t uid; /* 0030 uid / degenerate unix uid */
873 uuid_t gid; /* 0040 gid / degenerate unix gid */
875 uint8_t type; /* 0050 object type */
876 uint8_t op_flags; /* 0051 operational flags */
877 uint16_t cap_flags; /* 0052 capability flags */
878 uint32_t mode; /* 0054 unix modes (typ low 16 bits) */
881 * inode size, identification, localized recursive configuration
882 * for compression and backup copies.
884 hammer2_tid_t inum; /* 0058 inode number */
885 hammer2_off_t size; /* 0060 size of file */
886 uint64_t nlinks; /* 0068 hard links (typ only dirs) */
887 hammer2_tid_t iparent; /* 0070 parent inum (recovery only) */
888 hammer2_key_t name_key; /* 0078 full filename key */
889 uint16_t name_len; /* 0080 filename length */
890 uint8_t ncopies; /* 0082 ncopies to local media */
891 uint8_t comp_algo; /* 0083 compression request & algo */
894 * These fields are currently only applicable to PFSROOTs.
896 * NOTE: We can't use {volume_data->fsid, pfs_clid} to uniquely
897 * identify an instance of a PFS in the cluster because
898 * a mount may contain more than one copy of the PFS as
899 * a separate node. {pfs_clid, pfs_fsid} must be used for
900 * registration in the cluster.
902 uint8_t target_type; /* 0084 hardlink target type */
903 uint8_t check_algo; /* 0085 check code request & algo */
904 uint8_t pfs_nmasters; /* 0086 (if PFSROOT) if multi-master */
905 uint8_t pfs_type; /* 0087 (if PFSROOT) node type */
906 uint64_t pfs_inum; /* 0088 (if PFSROOT) inum allocator */
907 uuid_t pfs_clid; /* 0090 (if PFSROOT) cluster uuid */
908 uuid_t pfs_fsid; /* 00A0 (if PFSROOT) unique uuid */
911 * Quotas and aggregate sub-tree inode and data counters. Note that
912 * quotas are not replicated downward, they are explicitly set by
913 * the sysop and in-memory structures keep track of inheritence.
915 hammer2_key_t data_quota; /* 00B0 subtree quota in bytes */
916 hammer2_key_t unusedB8; /* 00B8 subtree byte count */
917 hammer2_key_t inode_quota; /* 00C0 subtree quota inode count */
918 hammer2_key_t unusedC8; /* 00C8 subtree inode count */
919 hammer2_tid_t attr_tid; /* 00D0 attributes changed */
920 hammer2_tid_t dirent_tid; /* 00D8 directory/attr changed */
923 * Tracks (possibly degenerate) free areas covering all sub-tree
924 * allocations under inode, not counting the inode itself.
925 * 0/0 indicates empty entry. fully set-associative.
927 * (not yet implemented)
929 uint64_t decrypt_check; /* 00E0 decryption validator */
930 hammer2_off_t reservedE0[3]; /* 00E8/F0/F8 */
933 typedef struct hammer2_inode_meta hammer2_inode_meta_t;
935 struct hammer2_inode_data {
936 hammer2_inode_meta_t meta; /* 0000-00FF */
937 unsigned char filename[HAMMER2_INODE_MAXNAME];
938 /* 0100-01FF (256 char, unterminated) */
939 union { /* 0200-03FF (64x8 = 512 bytes) */
940 struct hammer2_blockset blockset;
941 char data[HAMMER2_EMBEDDED_BYTES];
945 typedef struct hammer2_inode_data hammer2_inode_data_t;
947 #define HAMMER2_OPFLAG_DIRECTDATA 0x01
948 #define HAMMER2_OPFLAG_PFSROOT 0x02 /* (see also bref flag) */
949 #define HAMMER2_OPFLAG_COPYIDS 0x04 /* copyids override parent */
951 #define HAMMER2_OBJTYPE_UNKNOWN 0
952 #define HAMMER2_OBJTYPE_DIRECTORY 1
953 #define HAMMER2_OBJTYPE_REGFILE 2
954 #define HAMMER2_OBJTYPE_FIFO 4
955 #define HAMMER2_OBJTYPE_CDEV 5
956 #define HAMMER2_OBJTYPE_BDEV 6
957 #define HAMMER2_OBJTYPE_SOFTLINK 7
958 #define HAMMER2_OBJTYPE_HARDLINK 8 /* dummy entry for hardlink */
959 #define HAMMER2_OBJTYPE_SOCKET 9
960 #define HAMMER2_OBJTYPE_WHITEOUT 10
962 #define HAMMER2_COPYID_NONE 0
963 #define HAMMER2_COPYID_LOCAL ((uint8_t)-1)
965 #define HAMMER2_COPYID_COUNT 256
968 * PFS types identify the role of a PFS within a cluster. The PFS types
969 * is stored on media and in LNK_SPAN messages and used in other places.
971 * The low 4 bits specify the current active type while the high 4 bits
972 * specify the transition target if the PFS is being upgraded or downgraded,
973 * If the upper 4 bits are not zero it may effect how a PFS is used during
976 * Generally speaking, downgrading a MASTER to a SLAVE cannot complete until
977 * at least all MASTERs have updated their pfs_nmasters field. And upgrading
978 * a SLAVE to a MASTER cannot complete until the new prospective master has
979 * been fully synchronized (though theoretically full synchronization is
980 * not required if a (new) quorum of other masters are fully synchronized).
982 * It generally does not matter which PFS element you actually mount, you
983 * are mounting 'the cluster'. So, for example, a network mount will mount
984 * a DUMMY PFS type on a memory filesystem. However, there are two exceptions.
985 * In order to gain the benefits of a SOFT_MASTER or SOFT_SLAVE, those PFSs
986 * must be directly mounted.
988 #define HAMMER2_PFSTYPE_NONE 0x00
989 #define HAMMER2_PFSTYPE_CACHE 0x01
990 #define HAMMER2_PFSTYPE_UNUSED02 0x02
991 #define HAMMER2_PFSTYPE_SLAVE 0x03
992 #define HAMMER2_PFSTYPE_SOFT_SLAVE 0x04
993 #define HAMMER2_PFSTYPE_SOFT_MASTER 0x05
994 #define HAMMER2_PFSTYPE_MASTER 0x06
995 #define HAMMER2_PFSTYPE_UNUSED07 0x07
996 #define HAMMER2_PFSTYPE_SUPROOT 0x08
997 #define HAMMER2_PFSTYPE_DUMMY 0x09
998 #define HAMMER2_PFSTYPE_MAX 16
1000 #define HAMMER2_PFSTRAN_NONE 0x00 /* no transition in progress */
1001 #define HAMMER2_PFSTRAN_CACHE 0x10
1002 #define HAMMER2_PFSTRAN_UNMUSED20 0x20
1003 #define HAMMER2_PFSTRAN_SLAVE 0x30
1004 #define HAMMER2_PFSTRAN_SOFT_SLAVE 0x40
1005 #define HAMMER2_PFSTRAN_SOFT_MASTER 0x50
1006 #define HAMMER2_PFSTRAN_MASTER 0x60
1007 #define HAMMER2_PFSTRAN_UNUSED70 0x70
1008 #define HAMMER2_PFSTRAN_SUPROOT 0x80
1009 #define HAMMER2_PFSTRAN_DUMMY 0x90
1011 #define HAMMER2_PFS_DEC(n) ((n) & 0x0F)
1012 #define HAMMER2_PFS_DEC_TRANSITION(n) (((n) >> 4) & 0x0F)
1013 #define HAMMER2_PFS_ENC_TRANSITION(n) (((n) & 0x0F) << 4)
1015 #define HAMMER2_PFSSUBTYPE_NONE 0
1016 #define HAMMER2_PFSSUBTYPE_SNAPSHOT 1 /* manual/managed snapshot */
1017 #define HAMMER2_PFSSUBTYPE_AUTOSNAP 2 /* automatic snapshot */
1020 * PFS mode of operation is a bitmask. This is typically not stored
1021 * on-media, but defined here because the field may be used in dmsgs.
1023 #define HAMMER2_PFSMODE_QUORUM 0x01
1024 #define HAMMER2_PFSMODE_RW 0x02
1033 * Flags (8 bits) - blockref, for freemap only
1035 * Note that the minimum chunk size is 1KB so we could theoretically have
1036 * 10 bits here, but we might have some future extension that allows a
1037 * chunk size down to 256 bytes and if so we will need bits 8 and 9.
1039 #define HAMMER2_AVF_SELMASK 0x03 /* select group */
1040 #define HAMMER2_AVF_ALL_ALLOC 0x04 /* indicate all allocated */
1041 #define HAMMER2_AVF_ALL_FREE 0x08 /* indicate all free */
1042 #define HAMMER2_AVF_RESERVED10 0x10
1043 #define HAMMER2_AVF_RESERVED20 0x20
1044 #define HAMMER2_AVF_RESERVED40 0x40
1045 #define HAMMER2_AVF_RESERVED80 0x80
1046 #define HAMMER2_AVF_AVMASK32 ((uint32_t)0xFFFFFF00LU)
1047 #define HAMMER2_AVF_AVMASK64 ((uint64_t)0xFFFFFFFFFFFFFF00LLU)
1049 #define HAMMER2_AV_SELECT_A 0x00
1050 #define HAMMER2_AV_SELECT_B 0x01
1051 #define HAMMER2_AV_SELECT_C 0x02
1052 #define HAMMER2_AV_SELECT_D 0x03
1055 * The volume header eats a 64K block. There is currently an issue where
1056 * we want to try to fit all nominal filesystem updates in a 512-byte section
1057 * but it may be a lost cause due to the need for a blockset.
1059 * All information is stored in host byte order. The volume header's magic
1060 * number may be checked to determine the byte order. If you wish to mount
1061 * between machines w/ different endian modes you'll need filesystem code
1062 * which acts on the media data consistently (either all one way or all the
1063 * other). Our code currently does not do that.
1065 * A read-write mount may have to recover missing allocations by doing an
1066 * incremental mirror scan looking for modifications made after alloc_tid.
1067 * If alloc_tid == last_tid then no recovery operation is needed. Recovery
1068 * operations are usually very, very fast.
1070 * Read-only mounts do not need to do any recovery, access to the filesystem
1071 * topology is always consistent after a crash (is always consistent, period).
1072 * However, there may be shortcutted blockref updates present from deep in
1073 * the tree which are stored in the volumeh eader and must be tracked on
1076 * NOTE: The copyinfo[] array contains the configuration for both the
1077 * cluster connections and any local media copies. The volume
1078 * header will be replicated for each local media copy.
1080 * The mount command may specify multiple medias or just one and
1081 * allow HAMMER2 to pick up the others when it checks the copyinfo[]
1084 * NOTE: root_blockref points to the super-root directory, not the root
1085 * directory. The root directory will be a subdirectory under the
1088 * The super-root directory contains all root directories and all
1089 * snapshots (readonly or writable). It is possible to do a
1090 * null-mount of the super-root using special path constructions
1091 * relative to your mounted root.
1093 * NOTE: HAMMER2 allows any subdirectory tree to be managed as if it were
1094 * a PFS, including mirroring and storage quota operations, and this is
1095 * prefered over creating discrete PFSs in the super-root. Instead
1096 * the super-root is most typically used to create writable snapshots,
1097 * alternative roots, and so forth. The super-root is also used by
1098 * the automatic snapshotting mechanism.
1100 #define HAMMER2_VOLUME_ID_HBO 0x48414d3205172011LLU
1101 #define HAMMER2_VOLUME_ID_ABO 0x11201705324d4148LLU
1103 struct hammer2_volume_data {
1105 * sector #0 - 512 bytes
1107 uint64_t magic; /* 0000 Signature */
1108 hammer2_off_t boot_beg; /* 0008 Boot area (future) */
1109 hammer2_off_t boot_end; /* 0010 (size = end - beg) */
1110 hammer2_off_t aux_beg; /* 0018 Aux area (future) */
1111 hammer2_off_t aux_end; /* 0020 (size = end - beg) */
1112 hammer2_off_t volu_size; /* 0028 Volume size, bytes */
1114 uint32_t version; /* 0030 */
1115 uint32_t flags; /* 0034 */
1116 uint8_t copyid; /* 0038 copyid of phys vol */
1117 uint8_t freemap_version; /* 0039 freemap algorithm */
1118 uint8_t peer_type; /* 003A HAMMER2_PEER_xxx */
1119 uint8_t reserved003B; /* 003B */
1120 uint32_t reserved003C; /* 003C */
1122 uuid_t fsid; /* 0040 */
1123 uuid_t fstype; /* 0050 */
1126 * allocator_size is precalculated at newfs time and does not include
1127 * reserved blocks, boot, or redo areas.
1129 * Initial non-reserved-area allocations do not use the freemap
1130 * but instead adjust alloc_iterator. Dynamic allocations take
1131 * over starting at (allocator_beg). This makes newfs_hammer2's
1132 * job a lot easier and can also serve as a testing jig.
1134 hammer2_off_t allocator_size; /* 0060 Total data space */
1135 hammer2_off_t allocator_free; /* 0068 Free space */
1136 hammer2_off_t allocator_beg; /* 0070 Initial allocations */
1139 * mirror_tid reflects the highest committed change for this
1140 * block device regardless of whether it is to the super-root
1141 * or to a PFS or whatever.
1143 * freemap_tid reflects the highest committed freemap change for
1144 * this block device.
1146 hammer2_tid_t mirror_tid; /* 0078 committed tid (vol) */
1147 hammer2_tid_t reserved0080; /* 0080 */
1148 hammer2_tid_t reserved0088; /* 0088 */
1149 hammer2_tid_t freemap_tid; /* 0090 committed tid (fmap) */
1150 hammer2_tid_t bulkfree_tid; /* 0098 bulkfree incremental */
1151 hammer2_tid_t reserved00A0[5]; /* 00A0-00C7 */
1154 * Copyids are allocated dynamically from the copyexists bitmap.
1155 * An id from the active copies set (up to 8, see copyinfo later on)
1156 * may still exist after the copy set has been removed from the
1157 * volume header and its bit will remain active in the bitmap and
1158 * cannot be reused until it is 100% removed from the hierarchy.
1160 uint32_t copyexists[8]; /* 00C8-00E7 copy exists bmap */
1161 char reserved0140[248]; /* 00E8-01DF */
1164 * 32 bit CRC array at the end of the first 512 byte sector.
1166 * icrc_sects[7] - First 512-4 bytes of volume header (including all
1167 * the other icrc's except this one).
1169 * icrc_sects[6] - Sector 1 (512 bytes) of volume header, which is
1170 * the blockset for the root.
1172 * icrc_sects[5] - Sector 2
1173 * icrc_sects[4] - Sector 3
1174 * icrc_sects[3] - Sector 4 (the freemap blockset)
1176 hammer2_crc32_t icrc_sects[8]; /* 01E0-01FF */
1179 * sector #1 - 512 bytes
1181 * The entire sector is used by a blockset.
1183 hammer2_blockset_t sroot_blockset; /* 0200-03FF Superroot dir */
1188 char sector2[512]; /* 0400-05FF reserved */
1189 char sector3[512]; /* 0600-07FF reserved */
1190 hammer2_blockset_t freemap_blockset; /* 0800-09FF freemap */
1191 char sector5[512]; /* 0A00-0BFF reserved */
1192 char sector6[512]; /* 0C00-0DFF reserved */
1193 char sector7[512]; /* 0E00-0FFF reserved */
1196 * sector #8-71 - 32768 bytes
1198 * Contains the configuration for up to 256 copyinfo targets. These
1199 * specify local and remote copies operating as masters or slaves.
1200 * copyid's 0 and 255 are reserved (0 indicates an empty slot and 255
1201 * indicates the local media).
1203 * Each inode contains a set of up to 8 copyids, either inherited
1204 * from its parent or explicitly specified in the inode, which
1205 * indexes into this array.
1207 /* 1000-8FFF copyinfo config */
1208 hammer2_volconf_t copyinfo[HAMMER2_COPYID_COUNT];
1211 * Remaining sections are reserved for future use.
1213 char reserved0400[0x6FFC]; /* 9000-FFFB reserved */
1216 * icrc on entire volume header
1218 hammer2_crc32_t icrc_volheader; /* FFFC-FFFF full volume icrc*/
1221 typedef struct hammer2_volume_data hammer2_volume_data_t;
1224 * Various parts of the volume header have their own iCRCs.
1226 * The first 512 bytes has its own iCRC stored at the end of the 512 bytes
1227 * and not included the icrc calculation.
1229 * The second 512 bytes also has its own iCRC but it is stored in the first
1230 * 512 bytes so it covers the entire second 512 bytes.
1232 * The whole volume block (64KB) has an iCRC covering all but the last 4 bytes,
1233 * which is where the iCRC for the whole volume is stored. This is currently
1234 * a catch-all for anything not individually iCRCd.
1236 #define HAMMER2_VOL_ICRC_SECT0 7
1237 #define HAMMER2_VOL_ICRC_SECT1 6
1239 #define HAMMER2_VOLUME_BYTES 65536
1241 #define HAMMER2_VOLUME_ICRC0_OFF 0
1242 #define HAMMER2_VOLUME_ICRC1_OFF 512
1243 #define HAMMER2_VOLUME_ICRCVH_OFF 0
1245 #define HAMMER2_VOLUME_ICRC0_SIZE (512 - 4)
1246 #define HAMMER2_VOLUME_ICRC1_SIZE (512)
1247 #define HAMMER2_VOLUME_ICRCVH_SIZE (65536 - 4)
1249 #define HAMMER2_VOL_VERSION_MIN 1
1250 #define HAMMER2_VOL_VERSION_DEFAULT 1
1251 #define HAMMER2_VOL_VERSION_WIP 2
1253 #define HAMMER2_NUM_VOLHDRS 4
1255 union hammer2_media_data {
1256 hammer2_volume_data_t voldata;
1257 hammer2_inode_data_t ipdata;
1258 hammer2_blockset_t blkset;
1259 hammer2_blockref_t npdata[HAMMER2_IND_COUNT_MAX];
1260 hammer2_bmap_data_t bmdata[HAMMER2_FREEMAP_COUNT];
1261 char buf[HAMMER2_PBUFSIZE];
1264 typedef union hammer2_media_data hammer2_media_data_t;
1266 #endif /* !_VFS_HAMMER2_DISK_H_ */